Datasets:

mlcore

/

arxiv-classifier

like 1

Follow

Kilian's Group 9

1207.3281

1

1207

Thermal Spin-Accumulation in Electric Conductors and Insulators

The interpretation of some recent measurements of spin-dependent voltage for which the electric conduction does not play a role rises some new fundamental questions about the effects of spin-dependent heat currents. A two spin-channel model is proposed in order to describe the effect of out-of-equilibrium spin-dependent heat carriers in electric conductors and insulators. It is shown that thermal spin-accumulation can be generated by the heat currents only over an arbitrarily long distance for both electric conductors or electric insulators. The diffusion equations for thermal spin-accumulation are derived in both cases, and the principle of its detection based on Spin-Nernst effect is described.

cond-mat.mes-hall

cond-mat

Thermal Spin-Accumulation in Electric Conductors and Insulators J.-E. Wegrowe∗ and H.-J. Drouhin Ecole Polytechnique, LSI, CNRS and CEA/DSM/IRAMIS, Palaiseau F-91128, France D. Lacour Institut Jean Lamour, UMR CNRS 7198, Universit´e H. Poincarr´e, F -5406 Vandoeuvre les Nancy, France (Dated: April 17, 2019) Abstract The interpretation of some recent measurements of spin-dependent voltage for which the electric conduction does not play a role rises some new fundamental questions about the effects of spin- dependent heat currents. A two spin-channel model is proposed in order to describe the effect of out-of-equilibrium spin-dependent heat carriers in electric conductors and insulators. It is shown that thermal spin-accumulation can be generated by the heat currents only over an arbitrarily long distance for both electric conductors or electric insulators. The diffusion equations for thermal spin- accumulation are derived in both cases, and the principle of its detection based on Spin-Nernst effect is described. PACS numbers: 72.25.Mk, 85.75.-d 2 1 0 2 l u J 3 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 8 2 3 . 7 0 2 1 : v i X r a ∗Electronic address: jean-eric.wegrowe@polytechnique.fr 1 The role played by thermoelectric effects in the context of giant magnetoresistance (GMR) has long been recognized [1 -- 6] and exploited for possible applications in spintronics. Indeed, spin-dependent Seebeck and spin-dependent Peltier measurements are direct generalizations of giant magnetoresistance experiments performed in the presence of temperature gradi- ent, that can be described in the framework of the two spin-channel model for conduction electrons [7 -- 10]. Starting from the coupled transport equations of thermoelectricity, the in- troduction of the spin-dependent conduction coefficients σ(cid:108) and the spin-dependent Seebeck cross-coefficients S(cid:108) suffices to describe most of the spin-dependent thermoelectric effects in electric conductors [11 -- 18]. However, recent experiments pioneered by the group of Uchida et al. [19 -- 24] and still not well understood [12 -- 14] opened the way to a new class of spin- tronics effects in which no electric current is flowing and only spin-dependent heat currents are present in the sample. Such experiments are performed in electric conductors in the so-called non-local geometry (Fig. 1) at a long distance from the current injection, or in electrical insulators (Fig. 2). The goal of this work is to propose a phenomenological description of spin-accumulation produced by spin-dependent heat currents. The model first describes the spin-thermocouple effect that allows spin-accumulation to be generated by a temperature gradient in conduc- tors over arbitrarily large distances. This spin-dependent version of the thermocouple effect (that allows a temperature difference to be measured with a voltmeter in an open circuit), is corollary of the Seebeck effect (that allows an electric current to be generated with a temper- ature difference in a closed circuit). Both effects are deduced from the same thermoelectric transport equations with different boundary conditions and different power dissipation. The model is then applied to the case of electric insulators within the two spin-channel model of heat conduction. Finally, the transverse spin-dependent thermocouple effect occurring in the spin- Hall or spin-Nernst detector is also described. In all cases, the corresponding diffusion equations for the spin-accumulation are derived. For a non-isothermal ferromagnetic material - that can be an electric conductor or an electric insulator - the system under interest is composed of a statistical ensemble of out- of-equilibrium heat carriers of two different species: the heat carriers that carry a up spin ↑ and the heat carriers that carry a down spin ↓. In the case of electric conductors, the spin-dependent heat carriers are also electric carriers, while for electric insulators, the spin- dependent heat carriers are spin-dependent excitation modes (e.g. magnons). In both cases 2 FIG. 1: Schematic view of the spintronics non-local device with temperature gradient. Branch (II) under interest constitutes an open circuit from the point of view of the electric charges (J e = 0) but not from the point of view of the heat currents. The voltage is measured between points A and B in branch (II), which contains ferromagnetic/non-ferromagnetic interfaces, or between the points C and D of the lateral electrode. FIG. 2: Schematic view of non-local device with ferromagnetic insulator. Heat current is injected through the device and spin-accumulation can be produced in the lateral electrode. The potential due to spin-dependent Nernst effect is measured transversally along the y direction. however, the two sub-systems of ↑ and ↓ heat carriers are described by the corresponding local chemical potential µ↑ and µ↓, assuming the hypothesis of local equilibrium [25]. In line with the description of the giant magnetoresistance [7 -- 9], we define the thermal spin- accumulation by the difference of the two chemical potentials ∆µ = µ↑ − µ↓ of the heat carriers. In the framework of the non-equilibrium thermodynamic theory, the bivaluated spin 3 variable (cid:108) defines an internal degree of freedom [10, 25, 26]. The spin-relaxation occurring during the transport process is then treated as a chemical reaction that transforms heat carriers from the up channel to the down channel or inversely. The rate ψth of this reaction is a flux of spin variable in the spin space. Since this reaction changes the density of the heat carriers of the spin channels, the corresponding chemical affinity is ∆µ. The Onsager relation that links the flux ψth to the force ∆µ takes the same form as for electric spin-accumulation [10, 17, 26]: ψth = Lth∆µ, (1) where we have introduced the phenomenological Onsager coefficient Lth that describes the spin-dependent relaxation. At stationary regime (i.e. for the state of minimum entropy production of the system), the internal energy densities u↑ and u↓ obey the following con- servation equations: ∂u↑ ∂t = −(cid:126)∇. (cid:126)J q↑ − ψth + R/2 = 0 ∂t = −(cid:126)∇. (cid:126)J q↓ + ψth + R/2 = 0, ∂u↓ where R describes the dissipation out of the system (e.g. by radiation) and ψth describes the spin relaxation from one heat conduction channel to the other. (2) (3) In the case of electric conductors and in a first approximation, the heat carriers and the electric carriers are the same, so that the chemical potentials of the heat conduction channels is also that of the electric conduction channels. The well-known thermoelectric transport equations relate the electric currents (cid:126)J e(cid:108) and the heat currents (cid:126)J q(cid:108) to the corresponding forces (cid:126)∇µ(cid:108) and (cid:126)∇T [16, 17, 27]: (cid:126)∇µ(cid:108) − σ(cid:108)S(cid:108) (cid:126)J e(cid:108) = −σ(cid:108) e e (cid:126)J q(cid:108) = λ(cid:108) (cid:126)∇T − Π(cid:108) (cid:126)∇µ(cid:108) (cid:126)∇T, (4) where λ(cid:108) is the spin-dependent Fourier coefficient and Π(cid:108) = σ(cid:108)S(cid:108)T is the spin-dependent Peltier coefficient. The electrochemical potential reads µ(cid:108) = µch,(cid:108) + eV where e is the e charge of the electric carriers and V the electric voltage. We assume that the temperature dependence of the transport coefficients is negligible. The condition for the partial equilibrium (cid:126)J(cid:108) = 0 is then given by (cid:126)∇µ(cid:108) = −S(cid:108) (cid:126)∇T 4 (5) Eqs. (5) can be reformulated with introducing the total Seebeck coefficient S0 = S↑ + S↓: (cid:126)∇ (µ↑ + µ↓) = −S0 (cid:126)∇T and the Seebeck asymmetry coefficient ∆S = S↑ − S↓: (cid:126)∇(∆µ) = −∆S (cid:126)∇T. (6) (7) Eq. (6) is nothing but the equation of the thermocouple. Equation (7) is, in contrast, a direct consequence of the presence of spin-dependent Seebeck coefficients, and describes the spin- dependent thermocouple effect. Accordingly, it is expected that a gradient of temperature produces a spin-accumulation gradient without electric current and without non-local spin injection. The usual thermocouple effect allows the spin-accumulation generated by the temperature gradient to be measured: inserting Eq. (7) in to Eq. (6) we obtain: (cid:126)∇ (µ↑ + µ↓) = S0 ∆S (cid:126)∇(∆µ). (8) Eq. (8) is able to account for the observed Spin- Seebeck effects measured without electric current ( (cid:126)J e↑ = (cid:126)J e↓ = 0) at a distance arbitrarily large in the configuration of Fig. 1. The integration between the points A and B leads to the spin-dependent potential measured: (cid:90) B A VB − VA = S0 ∆S ∂∆µ ∂x dx. (9) On the other hand, Eqs. (3) to Eq. (8) can be generalized with the introduction of ten- sorial transport coefficients {σ(cid:108)}ij, {S(cid:108)}ij (i, j = {x, y, z} ) where the non-diagonal elements i (cid:54)= j account respectively for the Hall effect [28] and the corresponding Nernst effect in ho- mogeneous isotropic materials [25]. The detection of the transverse potential difference VDC performed on the right electrode in Fig. 1 and Fig. 2 is described by the transverse spin- dependent thermocouple effect, or spin-dependent Nernst effect. If ∆Sxy (cid:54)= 0, the projection of Eq. (7) over the Oy axis gives, after integration: (cid:90) D C (cid:90) D C VD − VC = ∂∆µ ∂y dy = ∆Sxy ∂T ∂x dy. (10) Eq. (9) and Eq. (10) are analogous to that of the usual spin-accumulation calculation for the giant magnetoresistance effect, with the difference that the spin-accumulation is pro- duced by the heat current instead of the electric current, and that the spin-accumulation 5 ∆µ(x) and ∆µ(y) are no longer governed by the usual diffusion equation of the spin- accumulation generated by electric current injection [7 -- 10]. Defining the total heat flux along the x axis (see Fig. 1 and Fig. 2) J q 0 = J q↑ + J q↓ and he spin-dependent heat flux ∆J q = J q↑ − J q↓ , we have from Eq. (4): (cid:1) (cid:1) , (cid:0) ∂ µ0 (cid:0) ∂∆µ ∆J q = ∆λ ∂T J q 0 = λ0 ∂x − Π0 ∂x − Π0 ∂x + β ∂∆µ ∂x + β ∂ µ0 ∂x ∂x ∂T 2e 2e Note that for J q where µ0 = µ↑+µ↓, λ0 = λ↑+λ↓, ∆λ = λ↑−λ↓, Π0 = Π↑+Π↓, and β = (Π↑ − Π↓) / (Π↑ + Π↓). 0 = 0, the flow ∆J q is a "pure spin-current" [12, 13] that cannot be distinguished from its electric homologue ∆J e = J e↑ − J e↓ with J e 0 = J e↑ + J e↓ = 0 (and more generally, a "pure spin-current" is independent of the nature of the spin carriers involved). Inserting Eq. (11) and Eq. (1) into Eq. (2), we deduce the diffusion equation for ∆µ: (11) (12) (13) where we have introduced the thermal spin-diffusion length for conducting materials = C1 ∂2T ∂x2 + C2, ∂2∆µ l2 c ∂x2 − ∆µ (cid:115) lc = Π0 (1 − β2) 4eLth and the constants C1 = 2e (∆λ − λ0β) / (Π0(1 − β2)), C2 = 2βRe/ (Π0(1 − β2)). Assuming a uniform temperature gradient and negligible external dissipation R ≈ 0, the solutions of Eq. (12) have the form ∆µ(x) = ∆µ(0)(cid:0)aex/lth + be−x/lth(cid:1) where a and b are constants imposed by the boundary conditions. It is the same solutions as that of the usual spin-accumulation responsible for GMR effects, with the difference that the diffusion length is defined with the help of the Peltier asymmetry coefficient β = ∆Π/Π0 instead of the asymmetry of the electric conductivity, and Lth plays the role of the usual spin-flip Onsager coefficient [10, 11]. In the case of spintronic devices with electric insulators (Fig. 2), there are no electric carriers and the chemical potentials µth(cid:108) are that of the corresponding heat carriers only (magnon, spin-dependent phonon, etc). The heat current is composed of a drift part (cid:126)J q drif t (cid:108) = λ(cid:108) (cid:126)∇T and a diffusion part (cid:126)J q dif f (cid:108) = −L(cid:108) (cid:126)∇µth(cid:108) , where L(cid:108) is the corresponding Onsager coefficient related to the diffusion coefficient D(cid:108) = kTL(cid:108)/nth, where nth is the density of the 6 heat carriers (for the sake of simplicity, we assume isotropic diffusion). In the configuration of Fig. 2, the transport equations reads: J q↑ = λ↑ ∂T J q↓ = λ↓ ∂T ∂x − L↑ ∂µth↑ ∂x ∂x − L↓ ∂µth↓ ∂x . (14) The diffusion equation for the chemical potential difference (i.e. spin accumulation) is de- duced by inserting Eq. (14) and Eq. (1) into Eqs. (2): ∂2∆µth ∂x2 − ∆µth l2 in = C1 (cid:16) (cid:118)(cid:117)(cid:117)(cid:116)L0 lin = ∂2T ∂x2 + C2, (cid:17) 1 − β2 4Lth , (15) (16) where we have introduced the thermal diffusion length for electric insulators (cid:16)L0(1 − β2) (cid:17) where L0 = L↑ + L↓ and β = (L↑ − L↓) /L0. On the other hand, C1 = 2(∆λ − λ0 and C2 = 2 βR/((1 − β2)L0). β)/ This spin-accumulation in the electrical insulator is transferred to a conducting electrodes through the condition of continuity of the heat currents at the interface. In that case also, the final profile of the spin-accumulation throughout the insulator/conductor interface is similar to that of the usual GMR spin-accumulation under electric current. In conclusion, the two spin-channel model has been applied to spin-dependent heat trans- port, in a circuit that is closed from the point of view of heat currents, and open for the point of view of electric currents. The result of the model shows that the heat current is able to play the same role as the electric current in a usual "local" spin-valve system, whatever the spin-dependent heat carriers considered. In particular, spin-injection and spin-accumulation can be performed with spin-dependent heat currents only at an interface. The model ex- plains the recent observations of spin-dependent voltage measured in electric conductors of electric insulators, in the configurations depicted in Fig. 1 and Fig. 2. [1] M. J. Conover, M. B. Brodsky, J. E. Mattson, C. H. Sowers, S. D. Bader, J. Mag. Mag. Magn. 102, L5 (1991), 7 [2] J. Sakurai, M. Horie, S. Araki, H. Yamamoto, and T, Shinijo, J. Phys. Sot. Jpu. 60, 2522 (1991). [3] J. Shi, R. C. Yu, S. S. P. Parkin, and M. B. Salamon, J. Appl. Phys. 73, 5524 (1993). [4] L. Piraux, M. Cassart, E. Grivei, M. Kinanyalaoui, J. S. Jiang, J. Q. Xiao, C. L. Chien, J. Mag. Mag. Magn. 136, 221 (1994). [5] L. Gravier, A. F`abi`an, A. Rudolf, A. Cachin, J.-E. Wegrowe, J.-Ph. Ansermet, J. Mag. Magn. Mat. 271, 153 (2004). [6] A. Fukushima, H. Kubota, A. Yamamoto, Y. Suzuki, and S. Yuasa, IEEE Trans. Mag. 41, 2571 (2005). [7] M. Johnson, R. H. Silsbee, Phys. Rev. Lett. 55, 1790 (1985) [8] P. C. van Son, H. van Kempen, P. Wyder, Phys. Rev. Lett. 58, 2271 (1998) [9] T. Valet, A. Fert, Phys. Rev. B 48, 7099 (1993) . [10] J.-E. Wegrowe, Phys. Rev. B 62, 1067 (2000). [11] A. A. Tulapurkar and Y. Suzuki, Phys. Rev. B 83 (2011). [12] J. Sinova, Nature Mater. 9, 880 (2010). [13] G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nature Mat. 11, 391 (2012). [14] B. Scharf, A. Matos-Abiague, I. Zuti´c, and J. Fabian, Phys. Rev. B 85, 085208 (2012). [15] K. Uchida, H. Adachi, T. Ota, H. Nakayama, S. Maekawa, J. Ieda, K. Harii, K. Ikeda, W. Koshibae, K. Ando, S. Maekawa, and E. Saitoh, J. Appl. Phys. 105, 07C908 (2009). [16] T. S. Nunner anf F. von Oppen, Phys. Rev. B 84, 020405(R), 2011. [17] J.-E. Wegrowe, Q. Anh Nguyen, M. Al-Barki, J.-F. Dayen, T. L. Wade, and H.-J. Drouhin, Phys. Rev. B, 73, 134422 (2006). [18] L. Gravier, S. Serrano-Guisan, F. Reuse, and J.-Ph. Ansermet, Phys. Rev B 73, 024419 (2006). [19] K. Uchida, S. Takahashi, K. Harii, J. Ieda, W. Koshibae, K. Ando, S. Maekawa, and E. Saitoh, Nature 455, 778 -- 781 (2008), [20] C. M. Jaworsky, J. Yang, S. Mack, D. D. Awschalom, J. P. Heremans, and R. C. Myers, Nature Mater. 9, 898 (2010). [21] K. Uchida, H. Adachi, T. Ota, H. Nakayama, S. Maekawa, Appl. Phys. Lett. 97, 172505 (2010). [22] M. Erekhinsky, F. Casanova, I. K. Schuller, and A. Sharoni, Appl. Phys. Lett. 100, 212401 (2012). 8 [23] E. Padr´on-Hern´andez, A. Azevedo, and S. M. Rezende, Phys. Rev. Lett. 104 197203 (2011). [24] G. L. da Silva, L. H. Viela-Leao, S. M. Rezende, and A. Azevedo, J. Appl. Phys. 111, 07C513 (2012). [25] S. R. De Groot, P. Mazur, Non-equilibrium Thermodynamics; North-Holland: Amsterdam, The Netherlands, 1962. [26] J.-E. Wegrowe and H.-J. Drouhin, Entropy 13, 316 (2012). [27] Without the spin indices, the microscopic derivation of the coupled Eqs. (3) and (4) (with explicit expressions of the phenomenological transport coefficients σ, λ, S and Π) can be found e.g. in A. C. Smith, J. Janak, and R. B. Adler, Electronic Conduction in Solids, Physical and Quantum Electronics Series McGraw-Hill, New York, 1967, page 4, Eqs. (1.3a) and (1.3b), and page 169, Eqs. (7.56) and following. [28] T. Jungwirth, J. Wunderlich and K. Olejnk, Nature Mat. 11, 382 (2012). 9 

1211.5055

2

1211

Kinetic theory of surface plasmon polariton in semiconductor nanowires

Based on the semiclassical model Hamiltonian of the surface plasmon polariton and the nonequilibrium Green-function approach, we present a microscopic kinetic theory to study the influence of the electron scattering on the dynamics of the surface plasmon polariton in semiconductor nanowires. The damping of the surface plasmon polariton originates from the resonant absorption by the electrons (Landau damping), and the corresponding damping exhibits size-dependent oscillations and distinct temperature dependence without any scattering. The scattering influences the damping by introducing a broadening and a shifting to the resonance. To demonstrate this, we investigate the damping of the surface plasmon polariton in InAs nanowires in the presence of the electron-impurity, electron-phonon and electron-electron Coulomb scatterings. The main effect of the electron-impurity and electron-phonon scatterings is to introduce a broadening, whereas the electron-electron Coulomb scattering can not only cause a broadening, but also introduce a shifting to the resonance. For InAs nanowires under investigation, the broadening due to the electron-phonon scattering dominates. As a result, the scattering has a pronounced influence on the damping of the surface plasmon polariton: The size-dependent oscillations are smeared out and the temperature dependence is also suppressed in the presence of the scattering. These results demonstrate the the important role of the scattering on the surface plasmon polariton damping in semiconductor nanowires.

cond-mat.mes-hall

cond-mat

Kinetic theory of surface plasmon polariton in semiconductor nanowires Y. Yin∗ and M. W. Wu† Hefei National Laboratory for Physical Sciences at Microscale and Department of Physics, University of Science and Technology of China, Hefei, Anhui, 230026, China (Dated: June 29, 2018) Based on the semiclassical model Hamiltonian of the surface plasmon polariton and the nonequi- librium Green-function approach, we present a microscopic kinetic theory to study the influence of the electron scattering on the dynamics of the surface plasmon polariton in semiconductor nanowires. The damping of the surface plasmon polariton originates from the resonant absorption by the elec- trons (Landau damping), and the corresponding damping exhibits size-dependent oscillations and distinct temperature dependence without any scattering. The scattering influences the damping by introducing a broadening and a shifting to the resonance. To demonstrate this, we investigate the damping of the surface plasmon polariton in InAs nanowires in the presence of the electron-impurity, electron-phonon and electron-electron Coulomb scatterings. The main effect of the electron-impurity and electron-phonon scatterings is to introduce a broadening, whereas the electron-electron Coulomb scattering can not only cause a broadening, but also introduce a shifting to the resonance. For InAs nanowires under investigation, the broadening due to the electron-phonon scattering dominates. As a result, the scattering has a pronounced influence on the damping of the surface plasmon polariton: The size-dependent oscillations are smeared out and the temperature dependence is also suppressed in the presence of the scattering. These results demonstrate the important role of the scattering on the surface plasmon polariton damping in semiconductor nanowires. PACS numbers: 73.20.Mf, 73.22.Lp, 72.30.+q, 71.10.-w I. INTRODUCTION Since the pioneering theoretical work by Ritchie1 and the electron-loss spectroscopy measurements by Pow- ell and Swan,2 physics of surface plasmon polariton (SPP) has been extensively studied for more than five decades.3 -- 9 SPPs are electromagnetic (EM) waves cou- pled to the collective excitations of electrons on the sur- face of a conductor.10,11 In this coupling, the electrons oscillate collectively in resonance with the EM waves and hence trap the EM waves on the surface. The resonant coupling leads to the SPPs and gives rise to their unique properties, such as the enhancement of the surface elec- tric fields and the slowing down of the group velocity of the EM waves.12 -- 20 Applications exploiting these prop- erties have been widely studied in biosensing,21 solar cells,22 quantum information processing,23 -- 25 subwave- length optical imaging and waveguiding devices.11,26 -- 31 The SPPs often suffer from dissipative losses.32 Over- coming the losses is crucial for the improvement of perfor- mance of many SPP-based devices, such as the fidelity of the waveguide and the sensitivity of the single-molecule sensor.21,33 An effective modulation of the losses is also highly desirable for active plasmonic devices proposed in recent years.34 -- 36 Thus, a thorough understanding of the damping processes responsible for the dissipative losses is essential. Since the SPPs in metals have been the major focus for many decades, previous studies on the damping processes have traditionally been focused on metals. It is found that the dominant damping process is the decay of the SPPs into electrons, i.e., the Landau damping.37 -- 41 The interband transitions and the many-body exchange- correlations can have pronounced influences on the Lan- dau damping. These influences can be incorporated into microscopic models based on time-dependent density- functional theory or semiclassical model Hamiltonian of the SPP.42 -- 44 Calculations based on these models have shown good agreement with experiments.45 -- 47 In recent years, doped semiconductors, such as SiC, GaAs, InAs, Cu2S and Cu2Se, are suggested as promis- ing candidates to replace metals in SPP applications.48 -- 52 The SPPs in doped semiconductors are characterized by their substantial low losses and tunable frequencies.53 -- 56 They are also easier to be manipulated via doping or ex- ternal electric/magnetic fields and to be integrated into complex, functional circuits.57 -- 59 The further develop- ment of the SPPs in doped semiconductors requires a bet- ter understanding of their damping processes. However, the physics involved in doped semiconductors can be quite different from that in metals. For doped semicon- ductors, although the Landau damping is still believed to be the leading damping process at large wavevectors,60,61 the charge depletion/accumulation layer62,63 and the electron scattering64,65 are found to have important in- fluence on the damping. Of particular importance is the effect of the electron scattering, since the typical scatter- ing rate can be comparable to the SPP frequency in semi- conductors. However, to the best of our knowledge, this effect has only been discussed by using phenomenologi- cal relaxation times.64,65 A microscopic theory exploiting this effect has not been established yet. In this paper, by combining the semiclassical model Hamiltonian of the SPP42 -- 45 and the nonequilibrium Green-function approach,66,67 we present a microscopic kinetic theory to study the damping of the SPP in doped semiconductors, within which the relevant elec- tron scatterings are treated fully microscopically. The main purpose of this work is to understand the influ- ence of these scatterings on the Landau damping of the SPP. To demonstrate this, we focus here on the SPPs in InAs nanowires23,28,29,51 and concentrate on the electron- impurity (ei), electron-phonon (ep) and electron-electron (ee) Coulomb scatterings. We find that the scattering can have pronounced influences on the Landau damp- ing of the SPP by modulating the resonance between the electrons and the SPPs. Different scattering has different effect on the resonance. The main effect of the ei and ep scatterings is to introduce a broadening to the resonance, whereas the ee scattering can not only case a broaden- ing, but also introduce a shifting to the resonance. For InAs nanowires, the ep-scattering -- induced broadening is found to be the dominant effect. These effects can lead to a pronounced influence on the damping of the SPP, which can be seen from both the size and temperature dependence of the SPP damping: (1) The size-dependent oscillations of the SPP damping are smeared out, and (2) the temperature dependence of the SPP damping is suppressed by the scattering. These results demonstrate the important role of the electron scattering on the SPP damping. This paper is organized as follows. In Sec. II, we intro- duce the semiclassical model Hamiltonian of the SPP for semiconductor nanowires and briefly outline the deriva- tion of the kinetic equations. We also present an analytic solution for the SPP damping which provides a simple and physically transparent picture to understand the in- fluence of the scattering on the SPP damping process. In Sec. III, we discuss in detail the influence of the electron scattering on the SPP damping by numerical solving the kinetic equations. The importance of the scattering is demonstrated by studying its influences on the tempera- ture and size dependence of the SPP damping. The an- alytic solution is also compared with the numerical ones in this section. We summarize and discuss in Sec. IV. II. MODEL AND FORMALISM A. Semiclassical model for SPP-electron system We consider an n-type free-standing cylindrical nanowire with radius R as illustrated in Fig. 1(a). The z-axis is chosen to be along the wire. Following the semi- classical approach developed in previous works,42 -- 45 we decompose the Hamiltonian into H = HSPP + Hel + HSPP-el, (1) where HSPP, Hel and HSPP-el are Hamiltonians for the SPP, electrons and the SPP-electron coupling, respec- tively. Here we only present the Hamiltonian, leaving the details to Appendix A. For nanowires, there exists one fundamental SPP mode with axial symmetry, which has no cutoff at 2 R = 32 nm 46 nm 28 nm 40 nm 18 (b) 50 40 -- n0 = 5.0 × 10 17 -3 cm 30 ) V e m ( ω 20 10 0 0 -- n0 = 1.5 × 10 17 -3 cm 6 -3 q (10 nm 12 -1 ) FIG. 1: (Color online) (a) Schematic of the structure of the electron field E and corresponding surface charge of the axial symmetric SPP mode in a cylindrical nanowire. ǫ∞ (ǫ2) is the dielectric constant inside (outside) the nanowire. The red curves with arrows represent the electric fields of the SPP mode. (b) The dispersive relation of the axial symmetric SPP mode for nanowires with different electron density ¯n0 and wire radius R. The thin grey line marks the energy of the LO phonon. 1 low frequency and has been found to be important for both the terahertz emission and quantum subwavelength optics.23,28,29,51 In this paper, we focus on the dynamics of this SPP mode. The corresponding Hamiltonian takes the form HSPP = Xq Ωqb† qbq, (2) where bq(b† q) represents the annihilation(creation) boson operator for the SPP, with Ωq being the corresponding dispersive relation illustrated in Fig. 1(b). Note that we set  = 1 throughout this paper. The Hamiltonian of electrons can be written as Hel = H 0 el + Hei + Hep + Hee, (3) where the free-electron Hamiltonian H 0 el is modelled by a mean-field potential. The interaction Hamiltonians Hei, Hep and Hee represent the ei, ep and ee interaction, re- spectively. By choosing the mean-field potential to be an infinite cylindrical potential well with radius R, the free-electron Hamiltonian H 0 el can be written as H 0 el = Xnkσ k c† εn nkσcnkσ, (4) 2m∗ n /R)2 k = k2+(λ m is the eigenenergy with m∗ in which εn representing the electron effective mass. The composed index n = ( m, n) labels the electron subband with m and n representing the angular and radial quantum numbers, respectively. λ m n denotes the n-th zero of the Bessel func- tion of the first kind J m(x). The corresponding eigen- states read ψnk(ρ, ϕ, z) = J m(λ m n ρ/R) √πRJ m+1(λ m n ) ei mϕeikz. (5) The ei interaction Hamiltonian can be written as Hei = Ni Xi Xkqσ vnn′ q ρi(q)c† n′k+qσcnkσ, (6) with Ni being the total impurity number and ρi(q) = e−iqzi . Here we have assumed that the impurities are distributed on the surface of the nanowire with an axially symmetric distribution. vnn′ is the matrix element for the ei interaction. The ep interaction Hamiltonian can be written as q Hep = XQq Xnn′kσ M nn′ Qq (cid:16)aQq + a† −Q−q(cid:17)c† nkσcn′k−qσ,(7) where aQq(a† Qq) represents the annihilation(creation) op- erator for the LO phonons, with Q and q representing the components of the phonon momentum perpendic- ular and parallel to the nanowire. Here we use bulk phonons in the present investigation, which is valid for nanowires with large diameters.68,69 M nn′ Qq is the matrix element for the ep interaction. It is noticed that although surface-optical (SO) phonons69 also exist in nanowires, they are of marginal importance since the correspond- ing electron-SO-phonon interaction is rather weak com- pared with the electron-LO-phonon interaction for the nanowires we consider here. The influence of the SO phonons will be further addressed in Sec. III B. The ee interaction Hamiltonian can be written as Hee = Xkk′qXnn′ Xσσ′ V nn′ q c† nkσc† n′k′σ′ cn′k′+qσ′ cnk−qσ,(8) q where V nn′ is the matrix element for the ee interaction. The SPP-electron coupling Hamiltonian HSPP-el can be written as HSPP-el =XnkσXn′k′ gnn′ k−k′(cid:16)bk−k′ + b† k′−k(cid:17)c† nkσcn′k′σ, q where gnn′ In these equations, matrix elements vnn′ and gnn′ is the SPP-electron coupling matrix element. Qq , V nn′ are given in detail in Appendix A. , M nn′ q q q B. Kinetic equations In this section, we briefly outline the derivation of the kinetic equations for the SPP-electron system. The de- tails can be found in Appendix B. with the detuning (9) 3 2 Bq2 ebqb† q e− 1 Bsi =Pq pQs function of the wave packet pQs sin[(Qs−q)L/2] The damping of the SPP is obtained by studying the temporal evolution of a coherent SPP wave packet with central wavevector Qs, which can be expressed as q0i (Ref. 70). The line-shape q = , where L is the wave packet length. Such wave packet is typical in a Fabry-Perot SPP resonator, which has been observed in various SPP systems.25,71 -- 73 The amplitude of the wave packet can be described by q Bq. The kinetic equation of Bs is obtained from the Heisenberg equation of the SPP annihilation operator bq, which has the form Bs =Pq pQs is chosen to be pQs (Qs−q)L/2 q ∂tBs(t) = Xnn′,kk′,σ k−k′ gnn′ pQs k−k′ G< σ (n′k′, nk; tt), (10) is the "lesser" function σ (nk, n′k′; tt′) where G< Green defined ihc† n′k′σ(t′)cnkσ(t)i (Ref. 66). The kinetic equation of the electron Green function σ (nk, n′k′; tt) is derived by using the nonequilibrium G< Green-function approach, which can be written as σ (nk, n′k′; tt′) electron = as G< σ (nk, n′k′; tt) k )iG< k′ − εn (B−q + B† q G< σ (nk, ¯nk′ − q; tt) q )hg ¯nn′ h − i∂t − (εn′ =X¯nq − gn¯n q G< σ (¯nk + q, n′k′; tt)i + I <σ nk,n′k′ (t), (11) where the first term in the right hand side of the equa- tions is the coherent driving term of the SPP, while the second term is the scattering term consisting the ei, ep and ee scatterings. Within the rotating wave approximation relative to the SPP central frequency Ωs,66,67 we obtain the kinetic equations for the SPP-electron system ∂t ¯Bs(t) = −i Xnn′σXkk′ ∂tPσ(nk, n′k′; t) = iωnn′ k−k′ gnn′ pQs k−k′hPσ(nk, n′k′; t)i† , (12) kk′ Pσ(nk, n′k′; t) + ignn′ k−k′ pQs k−k′ ¯B† shfnσ(k) − fn′σ(k′)i + ¯I σ nk,n′k′ ,(13) ωnn′ kk′ = εn′ k′ − εn k − Ωs. (14) ¯Bs(t) = Bs(t)eiΩst and In the above equations, σ (nk, n′k′; tt)eiΩst represent the Pσ(nk, n′k′; t) = −iG< SPP amplitude and electron polarization, respectively. fnσ(k) represents the equilibrium electron distribution which is conventionally chosen to be the spin-unpolarized Fermi-Dirac distribution at temperature T . It should be emphasized that in the derivation, we have treated the SPP-electron coupling gnn′ perturba- tively and linearized the equations by keeping only terms up to the linear order of gnn′ . Thus the SPP couples q q 4 only to the electron polarization corresponding to the off-diagonal electron Green function with respect to the electron momentum k and subband index n. The scattering term within the rotating wave approx- imation can be expressed as ¯I = ¯I ei + ¯I ep + ¯I ee, where ¯I ei, ¯I ep and ¯I ee are contributions from the ei, ep and ee scatterings, respectively. Their expressions can be found in Appendix B. C. Landau damping process In the kinetic equations, Eq. (13) describes the res- onant excitation of the electron polarizations, while Eq. (12) describes the back-action of the polarizations to the SPP. The summation in Eq. (12) implies that even without any scattering, the phase-mixing between polar- izations with different frequencies can also lead to the damping of the SPP, which is the origin of the Landau damping. To further clarify the Landau damping process de- scribed by the kinetic equations, we solve the equations without the scattering term ¯I. From Eq. (13), the corre- sponding polarization can be solved as ¯B† s(t) Pσ(nk, n′k′; t) = −δk′−k−qgnn′ × [fnσ(k) − fn′σ(k′)][ei(ωnn′ kk′ + i0+).(15) Note that in the derivation, we have assumed that the SPP amplitude ¯Bs varies slowly compared to the polar- ization P . q pQs q kk′ +i0+)t − 1]/(ωnn′ By substituting Eq. (15) into Eq. (12) and taking the long time limit t → ∞, one gets ∂t ¯Bs/ ¯Bs = −(τ −1 + iωs), (16) where ωs represents the frequency shifting, expressed as ωs = Xnk,n′k′,qσ q gnn′ pQs q 2[fn′σ(k′) − fnσ(k)] 1 ωnn′ kk′ δk′−k−q, (17) in which the summation is understood as a principal value integral. τ −1 is the damping rate of the SPP, which has the form 1 τ 2[fn′σ(k′)−fnσ(k)]δ(ωnn′ q gnn′ pQs kk′ )δk′−k−q. = π Xnk,n′k′,qσ q (18) Note that Eq. (18) agrees with the Landau damping rate derived from the Fermi golden rule in the literature.44,45 The above solution suggests that the Landau damping process can be understood as the resonant absorption of the SPP by electrons. The two δ-functions in Eq. (18) in- dicate that for a monochromatic SPP wave with wavevec- tor q, the absorption occurs between pairs of states nki and n′k′i satisfying the energy and momentum conser- vations ωnn′ kk′ = 0, k′ − k = q. (19) (20) FIG. 2: (Color online) Schematic of the resonant pairs cor- responding to the case of (i) strong Landau damping and (ii) weak Landau damping in the electron spectrum (a) and in the electron distribution (b). The two wavevector re- gions of the resonant pairs are illustrated by thick green/blue curves corresponding to the strong/weak Landau damping regime. The resonant pairs are centralized around the res- onance corresponding to the SPP central wavevector Qs as indicated by the vertical black dotted curves. The horizon- tal dashed line marks the chemical potential of the electrons µ. (c) The electron polarization and electron population dif- ference as function of the center wavevector K = k+k′ 2 with k′ − k = Qs. (i)/(ii) corresponds to the strong/weak Landau damping regime at low temperatures and (iii) corresponds to both the strong and the weak Landau damping regimes at high temperatures. The vertical solid line marks the reso- nance corresponding to Qs. Each pair of the states nki and n′k′i consist a resonant pair (nk, n′k′) relevant for the SPP damping. For the multi-subband system, there usually exist several such resonant pairs, laying between different subbands n and n′ and being well-separated from each other. For a non- monochromatic SPP wave packet with sufficiently narrow spectrum in q, the two states nki and n′k′i of each res- onant pair become two wavevector regions. Note that in the following discussion, we shall focus on the non- monochromatic SPP wave packet, and the resonant pair is referred to as the wavevector region unless otherwise specified. The resonant pairs are illustrated in Figs. 2(a) and (b). By using the resonant pairs, one can rewrite Eq. (18) into , τ −1 i τ −1 = Xi i = πXqσ X(nk,n′k′)∈i τ −1 × δ(ωnn′ kk′ )δk′−k−q, (21) q gnn′ pQs q 2[fn′σ(k′) − fnσ(k)] (22) with i being the index for the resonant pair corresponding to the SPP wave packet, whose spectrum is decided by q . (nk, n′k′) ∈ i means that the line-shape function pQs the two states nki and n′k′i belong to the i-th resonant pair. Thus one can see that the damping rate of the SPP wave packet is the sum over the absorption rates of all the relevant resonant pairs. Note that the electron population difference δf = fn′σ(k′)−fnσ(k) of the corresponding resonant pair plays an important role on its contribution to the SPP damp- ing. For the degenerate electrons where a well-defined Fermi surface exists around the chemical potential, there exist two regimes of the SPP damping: (i) a strong Lan- dau damping regime where states nki and nk′i of a resonant pair lay in each side of the chemical potential in the electron spectrum, leading to a large population dif- ference δf and hence a strong SPP damping; (ii) a weak Landau damping regime where the chemical potential lies outside all the resonant pairs, leading to a small δf and a weak SPP damping. This is illustrated in Figs. 2(a) and (b). Note that at high enough temperatures, the Fermi surface can be smeared out and such difference vanishes. It should be emphasized that according to Eq. (15), the resonant pair can be visualized as the resonant peak in the polarizations between the two subbands n and n′. In Fig. 2(c), we illustrate the polarizations P (nk, n′k′) cor- responding to the resonant pairs shown in Figs. 2(a) and (b) as function of the center wavevector K = (k + k′)/2 with k′−k = Qs. One finds that the polarization exhibits a Lorentzian peak around the resonance corresponding to the central wavevector Qs, as indicated by Eq. (15). Note that such resonant peak can show different fea- tures in the strong and weak Landau damping regimes at low temperatures. In the strong Landau damping regime, the corresponding polarization P exhibits a strong reso- nant peak concentrated in the region with large δf [(i) in Fig. 2(c)], indicating a large SPP absorption by the elec- 5 trons. In contrast, in the weak Landau damping regime, the corresponding resonant peak is weak and lies outside the large δf region [(ii) in Fig. 2(c)], indicating a small SPP absorption. In addition to the resonant peak, side peaks can also exist in the off-resonant regime due to the corresponding large δf . At high temperatures where the Fermi surface is smeared out, the population difference is rather flat for both the strong and weak Landau damp- ing regimes and the corresponding polarization exhibits a strong peak around the resonance for both regimes [(iii) in Fig. 2(c)]. D. Influence of the scattering The scattering influences the Landau damping by changing the resonant excitation of the polarizations. Specifically, (1) the scattering can introduce dissipative channels, inducing a decay of the polarization; (2) the scattering between polarizations with different preces- sion frequencies induces a frequency-mixing, leading to a modification of the polarization precession frequency. These two effects can be further clarified by assuming that for each resonant pair, the scattering term has the form ¯I iσ nk,n′k′ =Xq Γi[Pσ(nk−q, n′k′−q)−Pσ(nk, n′k′)], (23) where Γi stands for the phenomenological relaxation rate for the polarization of the i-th resonant pair. Note that (nk, n′k′) and (nk−q, n′k′−q) belong to the i-th resonant pair. For each resonant pair, the polarization P in the scat- tering term ¯I given above can be obtained by treating Γi perturbatively and solving Eq. (13) order by order, yielding Pσ(nk, n′k′; t) = δk′−k−qgnn′ ¯B† s(t)[fnσ(k) − fn′σ(k′)] q pQs q i )t − 1]/(ωnn′ kk′ − ¯Γa i + i¯Γb (24) i ), × [ei(ωnn′ kk′ −¯Γa i +i¯Γb where ωnn′ kk′ ), ¯Γb ¯Γa i = ΓiXq (1 − ωnn′ (ωnn′ kk′ − ωnn′ i = πΓiXq k−q,k′−q k−q,k′−q)δ(ωnn′ k−q,k′ −q). (25) (26) i The summation in Eq. (25) is understood as a princi- pal value integral. Note that we have omitted the k, k′- dependence of ¯Γa(b) inside each resonant pair for simplic- ity. The detail of the derivation is given in Appendix C. By comparing Eq. (24) to Eq. (15), one can see that the detuning ωnn′ in the resonant denominator is modified kk′ into ωnn′ kk′ − ¯Γa i , indicating that the polarization precession frequency is shifted by the scattering. The scattering also induces a finite imaginary part ¯Γb i to the resonant 6 the energy shift modifies the energy conservation Eq. (19) into ωnn′ kk′ − ¯Γa i = 0. (27) Thus the corresponding resonant pairs are shifted by the scattering. On the other hand, the energy broad- ening loosens the energy conservation constraint given by Eq. (19). Thus the corresponding resonant pairs are broadened. Note that the broadening and shifting are usually small and cannot induce overlaps between differ- ent resonant pairs. Accordingly, the broadening ¯Γb i also manifest themselves in the SPP damping rate. Following the same procedure of the derivation of Eq. (18), the SPP damping rate in the presence of the scattering can be written as i and shifting ¯Γa τ −1 i , τ −1 τ −1 = Xi q gnn′ i = Xqσ X(nk,n′k′)∈i pQs ¯Γb i i )2 + (¯Γb kk′ − ¯Γa (ωnn′ × (28) 2[fn′σ(k′) − fnσ(k)] q δk′−k−q. (29) i )2 By comparing the above equations to Eq. (18), one ob- serves that the δ-function corresponding to the energy conservation Eq. (19) is broadened into a Lorentzian with width ¯Γb i and shift ¯Γa i . It should be emphasized that the broadening and shift- ing of the resonance pair can also be visualized as the broadening and shifting of the corresponding resonant peak in the polarizations as illustrated in Fig. 3. This of- fers a simply way to interpret the influence of the broad- ening and shifting on the SPP damping rate. The in- fluence of the shifting depends on the direction of the shift. From the figure, one can see that the shifting re- duces the resonant peak in the polarization if the peak is shifted towards the region with smaller δf , thus the ab- sorption of the SPP by the corresponding resonant pairs is reduced [(iii) in Figs. 3(a) and (c)] and the correspond- ing SPP damping rate is suppressed. Otherwise, if the peak is shifted towards the region with larger δf [(iii) in Fig. 3(b)], the absorption is enhanced and the SPP damping rate is enhanced. The influence of the broadening can be different in the strong and weak Landau damping regimes at low tem- peratures as illustrated in Figs. 3(a) and (b). In the strong Landau damping regime, the broadening can re- duce the sharp resonance peak of the polarization [(ii) in Fig. 3(a)] and suppress the absorption of the SPP. Thus the corresponding SPP damping rate is suppressed in this regime. In contrast, in the weak Landau damping regime, the broadening of the resonance increases the absorption from the region with larger δf [(ii) in Fig. 3(b)], thus the absorption is enhanced and the corresponding SPP damping rate is enhanced. Note that at high tempera- tures, as the polarization exhibits a sharp peak around FIG. 3: (Color online) Schematic of the effect of the broad- ening and shifting on the electron polarization for (a) strong Landau damping regime at low temperatures; (b) weak Lan- dau damping regime at low temperatures and (c) both strong and weak Landau damping regimes at high temperatures. The thin solid vertical lines represent the resonance without broadening and shifting. The resonances with the broadening and shifting are represented by the thick solid and thin dotted vertical lines, respectively. For clarification, only the shifting towards the small K direction is illustrated. Note that the population differences δf become flat in both regimes at high temperatures. denominator, representing the decay of the polarization due to the scattering. The above solution indicates that the scattering modi- fies the resonance between the polarization and the SPP by introducing both an energy shift and an energy broad- ening to the corresponding resonance pairs. On one hand, the resonance in both the strong and weak Landau damp- ing regimes, the scattering tends to suppress the SPP damping rate in both regimes [(ii) in Fig. 3(c)]. E. Broadening and shifting from a simplified model In order to gain a further understanding of the micro- scopic origin of the broadening and shifting, we discuss the contributions of the ei, ep, and ee scatterings to the broadening and shifting within a simplified model in this section. Within this model, we assume that the scattering only occurs between the polarizations inside each resonant pair. Under such assumption, all the scattering terms can be written in an unified form for the i-th resonant pair ¯I iσ nk,n′k′ = Xq nΓa − Γb nk,n′k′,i(q)Pσ(nk − q, n′k′ − q) nk,n′k′,i(q)Pσ(nk, n′k′)o, (30) where both (nk, n′k′) and (nk− q, n′k′ − q) belong to the i-th resonant pair. Note that the corresponding Γa/b for each scattering mechanism can be obtained by comparing the scattering terms [Eqs. (B11-B13)] to ¯I iσ nk,n′k′ given in the above equation. Note that Eq. (30) has a similar structure as Eq. (23), thus one can derive the corresponding broadening and shifting following the similar procedure, yielding ¯Γb ¯Γa i =Xq (cid:16)Γb i = πXq nk,n′k′,i(q) − Γa nk,n′k′,i(q)(ωnn′ Γa nk,n′k′,i(q) (31) ωnn′ kk′ k−q,k′−q(cid:17), ωnn′ kk′ − ωnn′ k−q,k′−q)δ(ωnn′ k−q,k′ −q).(32) i and shifting ¯Γa The broadening ¯Γb i due to each scattering can be evaluated by substituting the corresponding scat- tering term into Eqs. (30-32). Note that we have omitted the k, k′-dependence of ¯Γa(b) inside each resonant pair to make the equation simple and physically transparent. For each resonant pair, ¯Γa(b) is evaluated by choosing (nk, n′k′) corresponding to the SPP central wave vector Qs (i.e., k′ − k = Qs) since the line-shape function pQs q is peaked at q = Qs. The shifting ¯Γa i can be written as i i ¯Γa(ei) i ¯Γa(ep) i ¯Γa(ee) i = 0, = 0, = 2π Xj=1,2X¯n Πn n′ (¯n¯kj, 0), (33) (34) (35) where Πn n′ (¯n¯k, q) =Xσ V n¯n q V ¯nn′ q ¯nσ(¯k + q)f < f > ¯nσ(¯k), (36) 7 k−Qs−εn′ with Vq being the screened ee interaction. k′ +Ωs)/Qs−Qs/2 and ¯k2 = 2(εn −2(εn Ωs)/Qs+Qs/2. ¯Γa(ei) correspond to the contributions from the ei, ep and ee scatterings, respec- tively. The corresponding broadenings have the forms ¯k1 = + and ¯Γa(ee) k−εn′ , ¯Γa(ep) k′−Qs i i i )(vnn vn′n′ 0 k′ Q,−qM nn′ M nn 0 − vn′n′ Q,−q) 0 ), (37) + nσ(k − q)]/k − q(cid:12)(cid:12)(cid:12)q=q(k) nσ(k − q)]/k − q(cid:12)(cid:12)(cid:12)q=q(k) − o n′ (¯n¯k, 0)]/k − ¯ko (38) (39) ¯Γb(ei) i ¯Γb(ep) i vnn 0 = m∗πni( k − = Xσ nπm∗(XQ × [N > + πm∗(XQ × [N < LOf > LOf > LOf < nσ(k − q) + N < Q,−qM nn′ Q,−q) M nn nσ(k − q) + N > LOf < ¯Γb(ee) i n(¯n¯k, 0) − Πn [Πn + {nk ←→ n′k′}, = n2πX¯n¯k + {nk ←→ n′k′}. ± satisfies [q(k) q where q(k) ± ]2− 2kq±± 2m∗ΩLO = 0 and vnn′ represents the screened ei interaction. {nk ←→ n′k′} stands for the same term as in the previous {} but with the interchange of indices nk ←→ n′k′. From Eqs. (33-39), one finds that different scattering has different contribution to the broadening and shift- ing. Only the ee scattering contributes to the shifting, while the contributions from the ei and ep scattering van- ish. Although all the scatterings can contribution to the broadening, their relative importance can be quite differ- ent. This is because for the nanowires considered here, the ei, ep and ee matrix elements vnn′ Qq and V nn′ are not sensitive to the subband index n. Thus, for the ei and ee scatterings, according to Eqs. (37) and (39), the terms in the bracket can largely cancel each other, leading to small contributions left. In contrast, such cancellation is absent for the ep scattering, thus its contribution to the broadening is expected to be larger than the ones from the ei and ee scatterings. , M nn′ q q Equations (28) and (29) with the broadening and shift- ing given in Eqs. (33-39) consist the analytic solution for the SPP damping rate. The analytic solution in this section offers a simple picture to understand the influ- ence of the scattering on the SPP damping: The SPP damping comes from the resonant absorption of the SPP by electrons, while the scattering can introduce a broad- ening and a shifting to the resonance and hence affects the damping process. At low temperatures, the broad- ening tends to suppress the SPP damping rate in the strong Landau damping regime. While in the weak Lan- dau damping regime, the broadening tends to enhance the SPP damping. At high temperatures, such difference vanishes and the broadening tends to suppress the damp- ing in both regimes. The shifting can suppress the SPP damping if the resonance is shifted towards the region with smaller δf , but boost it if the resonance is shifted toward the region with larger δf . Different scatterings can have different contributions to the broadening and shifting. From the simplified model in this section, one finds that the shifting is determined by the ee scattering, and the broadening is mainly decided by the ep scattering for the typical nanowires considered here. III. NUMERICAL RESULTS In the numerical investigation, we choose nanowires to be free-standing InAs nanowires. The typical elec- tron density ¯n0 is in the range of 1017 ∼ 1018 cm−3 and the wire radius R is around 25 ∼ 75 nm.74 The LO phonon energy ΩLO = 29 meV and the electron effective mass m∗ = 0.023m0 with m0 representing the free elec- tron mass. The dielectric constants of the nanowires are ǫ∞ 1 = 12.3 for high frequency and ǫ0 1 = 15.5 for low fre- quency. The dielectric constant outside the nanowire is ǫ2 = 1.0. Note that for such nanowires, there are 10-20 electron subbands relevant to the SPP damping. We set the impurity line density ni = 0.5ne with ne = πR2¯n0 being the electron line density. By numerical solving the kinetic equations Eqs. (12) and (13), one obtains the temporal evolution of the SPP amplitude ¯Bs. The SPP damping rate τ −1 can be extracted by fitting the real part of ¯Bs with a sin- gle exponential decay of the cosine oscillation: Re[ ¯Bs] = ¯B0 s exp(−t/τ ) cos(ωst), where the initial value of the SPP amplitude ¯B0 s is chosen to be real. We set the wave packet length L = 100R (Ref. 75). A. Landau damping: Size and temperature dependence Before we discuss the influence of the scattering, it is helpful to first obtain an understanding of the SPP damping without scattering. In Fig. 4(a), we show a typical behavior of the SPP damping rate as function of the SPP central wavevector Qs and wire radius R for the nanowire with electron density ¯n0 = 1.5 × 1017 cm−3. The temperature is chosen to be 100 K. One finds that the SPP damping rate oscillates with the radius R. Note that similar size-dependent oscillations have also been reported in metal nanoparticles and thin films.43 -- 45,76 -- 78 Such oscillations are usually attributed to the quan- tized electron states in the nanostructures,43 -- 45 which can be understood in terms of the resonant pairs in the nanowires we studied here. To illustrate this, we con- centrate on the damping rate corresponding to a typical SPP central wavevector Qs = 1.66× 10−2 nm−1 [skyblue curves in Fig. 4(a)] and show the corresponding resonant pairs for radii R = 34, 38.5, 40 and 43 nm in Fig. 4(b). Each resonant pair can be represented by the resonance corresponding to the central wave vector Qs, since the 8 line-shape function of the SPP wave packet is peaked at Qs. There are four resonant pairs in Fig. 4(b) which lay between different subbands: pair (i) is between the sub- bands 1 and 4, pair (ii) is between the subbands 2 and 6, pair (iii) is between the subbands 3 and 9 and pair (iv) is between the subbands 4 and 8. In the figure, the four res- onant pairs (i-iv) are denoted by the skyblue dots, green squares, brown open circles and yellow triangles, respec- tively. The subbands corresponding to each resonant pair are also plotted with solid curves in the same color. Note that for R = 34 nm, only the resonant pair (i) is relevant for the damping. For R = 38.5 and 40 nm, both the resonant pairs (i) and (ii) are relevant. For R = 43 nm, all the four resonant pairs (i-iv) contribute to the SPP damping. From Fig. 4(b), one can see that as the radius R in- creases, the resonant pairs move from left to right in the electron spectrum. When a resonant pair moves across the chemical potential marked by the horizontal black dashed lines, a crossover between the strong and weak Landau damping regimes occurs, which induces the size- dependent oscillations shown in Fig. 4(a). Note that the peaks/valleys correspond to the strong/weak Lan- dau damping regimes. For example, the oscillation from R = 32 to 38.5 nm is due to the crossover induced by the resonant pair (i). While the crossover induced by the resonant pair (ii) induces the oscillation from R = 38.5 to 43 nm. One finds from the figure that R = 34 and 40 nm correspond to the strong Landau damping regime whereas R = 38.5 and 43 nm correspond to the weak Landau damping regime. One also observes from Fig. 4(b) that due to the many subbands in the nanowires, there usually exist multi- resonant pairs relevant for the SPP damping. For a given Qs, more and more resonant pairs become involved as the radius R increases. The number of relevant reso- nant pairs are labeled in the R-Qs plane in Fig. 4(a). Note that as the number of the resonant pairs increases, the magnitude of the size-oscillations become less pro- nounced. This is mainly because the oscillations are usu- ally induced by the crossover due to one resonant pair as R varies. For the system with many resonant pairs, the contribution from one resonant pair becomes less signif- icant. Thus the size-dependent oscillations can be sup- pressed for nanowires with large R. It should be emphasized that the size-dependent os- cillations can also be suppressed by increasing temper- ature T . This is because the crossover is more pro- nounced for strongly degenerate electrons where a clear Fermi surface exists around the chemical potential. In high-temperature regime, the crossover is largely sup- pressed. To show this, we plot the damping rate τ −1 as function of the radius R and temperature T for Qs = 1.66 × 10−2 nm−1 in Fig. 4(c). One sees that as temper- ature increases, the damping rate τ −1 corresponding to the strong Landau damping regime decreases, while τ −1 corresponding to the weak Landau damping regime in- creases. This leads to a suppression of the size-dependent R=38.5 nm R=38.5 nm R=34 nm R=34 nm R=40 nm R=40 nm R=43 nm R=43 nm T=100K T=100K 10 10 τ-1 τ-1 τ-1 (ps (ps (ps 0.1 0.1 -1 -1 -1 ) ) ) 10 10 -3 -3 -5 -5 10 10 1 2 4 32 34 36 38 40 42 44 46 32 34 36 38 40 42 44 46 Qs=1.66 × 10 Qs=1.66 × 10 18 18 -2 -2 nm nm -1 -1 16.5 16.5 -1) -1) -3 nm -3 nm (10 (10 Q s Q s 15 15 R (nm) R (nm) Qs=1.66 × 10 -2 -1 nm τ-1 τ-1 -1 -1 (ps (ps ) ) 0 10 -1 10 -2 10 -3 10 46 44 42 40 38 R (nm) 36 34 32 300 100 200 T ( K ) 9 (b) R=34 nm R=40 nm (a) 0.15 0.1 ) V e ( ω 0.05 0 0.15 R=38.5 nm R=43 nm 0.1 ) V e ( ω 0.05 0 (c) -0.15 0 k (nm -1 ) 0.15 -0.15 0 k (nm -1 ) 0.15 (d) 0 10 R=34 nm R=40 nm ) 1 - s p ( 1 - τ ) 1 - s p ( 1 - τ (i) (ii) (iii) (iv) nosc(N) nosc(A) R=38.5 nm R=43 nm -1 10 -1 10 -2 10 50 100 150 200 T (K) 250 300 50 100 200 150 T (K) 250 300 FIG. 4: (Color online) (a) The SPP damping rate as function of SPP central wave vector Qs and wire radius R without scattering for ¯n0 = 1.5 × 1017 cm−3 and T = 100 K. The skyblue curve represents the damping rate corresponding to the SPP central wave vector Qs = 1.66 × 10−2 nm−1. The number of the relevant resonant pairs are labelled in the R-Qs plane. (b) The resonant pairs for R = 34, 38.5, 40 and 43 nm corresponding to Qs = 1.66 × 10−2 nm−1 in the electron spectrum. For clarification, only the lowest 10 subbands are plotted. (c) The SPP damping rate without scattering as function of radius R and temperature T for Qs = 1.66 × 10−2 nm−1. (d) Temperature dependence of the damping rate τ −1 without scattering for R = 34, 38.5, 40 and 43 nm with Qs = 1.66 × 10−2 nm−1. Symbols correspond to the numerical results. Red curves represent the results from the analytic solution. The contributions of each resonant pair from the analytic solution are also plotted as curves with different colors. The skyblue double-dotted chain, green chain, brown dotted and yellow solid curves correspond to the contribution from the resonant pair (i-iv), respectively. oscillations in high-temperature regime. One can also obtain the above results from the ana- lytic solution of the kinetic equations without scattering [Eqs. (21) and (22)]. To show this, we compare the tem- perature dependence of τ −1 from both the numerical (red squares) and analytic (red solid curves) solutions for the nanowires with R = 34, 38.5, 40 and 43 nm in Fig. 4(d). One finds good agreement between each other, indicat- ing that the analytic solution without scattering offers a good estimation to the numerical results. Note that according to the analytic solution, the temperature de- pendence of the SPP damping rate originates from the population difference of the resonant pairs. From the analytic solution, one can also identify con- tributions from different resonant pairs, which are plot- ted as curves with different colors and line shapes in Fig. 4(d). The skyblue double-dotted chain, green chain, brown dotted and yellow solid curves correspond to the contributions from the resonant pair (i-iv), respectively. It is clear that the relative importance of the resonant pairs can be quite different. For the strong Landau damp- ing regime, there usually exists one resonant pair whose contribution is much larger than the other pairs. For ex- ample, the damping rate τ −1 is mainly determined by the resonant pairs (i) and (ii) for R = 34 nm and R = 40 nm, respectively. For the weak Landau damping regime, the contributions from different resonant pairs can be com- parable. For example, for R = 38.5 and R = 43 nm, al- though the resonant pair (ii) has a large contribution to the damping rate τ −1, the other resonant pairs can also play important roles, especially at high temperatures. From the above results, one finds that the SPP damp- ing exhibits size-dependent oscillations and distinct tem- perature dependence without scattering, which can be explained by the analytic solution. τ-1 τ-1 (ps (ps -1 -1 ) ) 0 10 -1 10 -2 10 -3 10 46 44 42 40 38 R (nm) 36 34 32 0 10 ) 1 - s p ( 1 - τ -1 10 -1 10 ) 1 - s p ( 1 - τ R=34 nm R=40 nm -2 10 R=38.5 nm R=43 nm (a) 300 100 200 T ( K ) (b) nosc ep ee all ei 50 100 150 200 T (K) 250 300 50 100 200 150 T (K) 250 300 FIG. 5: (Color online) (a) SPP damping rate in the presence of all the scattering as function of radius R and temperature T for Qs = 1.66 × 10−2 nm−1. (b) SPP damping rate τ −1 cal- culated from the numerical results for Qs = 1.66 ×10−2 nm−1 with different scattering for R = 34, 38.5, 40 and 43 nm. Sym- bols with big blue dots, small green squares, small olive dots represent τ −1 calculated with the ep, ee and ei scatterings, respectively. The brown triangles represent τ −1 calculated in the presence of all the three scatterings and the red squares represent τ −1 without scattering. B. Influence of scattering Now we discuss the influence of the scattering on the SPP damping. In Fig. 5(a), we plot the damping rate τ −1 as function of the radius R and temperature T for Qs = 1.66× 10−2 nm−1 in the presence of all the scatter- ing. Comparing to the case without scattering [Fig. 4(c)], one finds that the scattering has pronounced influence on the SPP damping: (1) The size-dependent oscilla- tions are effectively smeared out, and (2) the temper- ature dependence also becomes weaker compared to the case without scattering. To gain a better understanding of the influence of the scattering, in Fig. 5(b), we compare the temperature de- pendence of the damping rate τ −1 with and without scat- tering for nanowires with four typical radii R = 34, 38.5, 40 and 43 nm. The brown triangles represent τ −1 cal- culated in the presence of all the three scatterings, while τ −1 without scattering are plotted with red squares for comparison. Note that R = 34 and 40 nm correspond to the strong Landau damping regime, whereas R = 38.5 and 43 nm correspond to the weak one. 10 nosc(N) ep(N) nosc(A) ep(A) R=34 nm i(A) ii(A) iii(A) iv(A) R=40 nm 0 10 ) 1 - s p ( 1 - τ -1 10 -1 10 ) 1 - s p ( 1 - τ -2 10 R=38.5 nm R=43 nm 50 100 150 200 T (K) 250 300 50 100 200 150 T (K) 250 300 FIG. 6: (Color online) Comparison between the numerical and analytical results of the SPP damping rate for R = 34, 38.5, 40 and 43 nm with the ep scattering. Blue dots repre- sent the numerical results, while the blue dashed curves show the results from the analytic solution. The contribution of each resonant pair from the analytic solution is also plotted. The skyblue double-dotted chain, green chain, brown dotted and yellow solid curves correspond to the contribution from the resonant pair (i-iv) , respectively. For comparison, the nu- merical and analytical results for the damping rate without scattering are also plotted with red squares and solid curves, respectively. In the presence of the scattering, it is seen that the damping rate τ −1 is markedly suppressed in the strong Landau damping regime (R = 34 and 40 nm). In con- trast, for the weak Landau damping regime (R = 38.5 and 43 nm), the scattering plays different roles in differ- ent temperature regimes: The damping rate is markedly enhanced in the low-temperature regime (T . 150 K), but is largely suppressed in the high-temperature regime (T & 150 K). A crossover exists at the intermediate tem- perature regime. It is also noted that the damping rate can be enhanced/suppressed by almost one order of mag- nitude by the scattering. To understand these influences, we first identify the dominant scattering mechanism. To do so, we calcu- late the damping rates τ −1 with the ep, ee or ei scatter- ing only, and plot them with big blue dots, small green squares and small olive dots in Fig. 5(b), respectively. It is clear to see from the figure that the damping rate τ −1 is dominated by the ep scattering.79 We first concentrate on the ep scattering. From the analytic solution within the simplified model, we have attributed the effect of the ep scattering to the broaden- ing of the resonant pairs. To see if this picture gives a proper description of the influence of the ep scattering in general case, we calculate the temperature dependence of the damping rate τ −1 by using the analytic solution Eqs. (28) and (29) with the ep-scattering -- induced broad- ening given in Eq. (38). The calculated analytic results are compared to the numerical ones in Fig. 6. In the figure, the blue dots represent the damping rate τ −1 from the numerical results, while the blue dashed curves represent τ −1 from the analytic results. For com- parison, we also plot the numerical and analytic results without scattering with red squares and solid curves, respectively. One finds good agreement between each other, indicating that the broadening can give a proper description of the effect of the ep scattering. Note that at low temperatures, the broadening tends to suppress the SPP damping rate in the strong Landau damping regime (R = 34 and 40 nm). While in the weak Landau damping regime (R = 38.5 and 43 nm), the broadening tends to enhance the SPP damping rate. At high tem- peratures, such difference vanishes and the SPP damping rate is suppressed in both the strong and weak Landau damping regimes. This leads to a crossover between the suppression and enhancement for R = 38.5 and 43 nm corresponding to the weak Landau damping regime as shown in Fig. 6. This also agrees with the conclusion from the analytic solution. It is worth noting that the scattering can also suppress the temperature dependence of the SPP damping rate. This is because the temperature dependence originates from the population difference of the resonant pairs. For the resonant pairs with the broadening, the correspond- ing population difference is less sensitive to the temper- ature, leading to the suppression of the temperature de- pendence of the corresponding SPP damping rate. We further point out that the scattering can also change the relative importance of different resonant pairs. To see this, we identify contributions of different resonant pairs from the analytic solution, as applied in Sec. III A for the case without scattering. In Fig. 6, the skyblue double-dotted chain, green chain, brown dotted and yel- low solid curves correspond to the contributions from the resonant pairs (i-iv), respectively. One finds that the res- onant pair (i) dominates the damping for R = 34 nm, whereas the resonant pair (ii) plays the most important role for R = 38.5 and 40 nm. For R = 43 nm, all the four resonant pairs have comparable contributions to the damping, with the largest contribution coming from the resonant pair (iii). Comparing to the case without scat- tering [Fig. 4(d)], one also finds that the relative im- portance of different resonant pairs is modified by the broadening, especially in weak Landau damping regime. From the above discussion, one comes to the conclu- sion that the influence of the ep scattering on the SPP damping rate can be understood as the broadening of the resonant pairs. The analytic solution incorporating such broadening shows good agreement with the numerical re- sult. Note that in the above results, the contribution of the SO phonons is omitted since it is much smaller than that from the LO phonons. This is shown in detail in Appendix D. Now we briefly address the ee and ei scatterings. From the analytic solution within the simplified model, the ef- fect of the ei scattering is attributed to the broadening of the resonant pairs. While for the ee scattering, the main effect is due to the shifting. The SPP damping rate 11 0 10 ) 1 - s p ( 1 - τ -1 10 nosc(N) ee(N) ei(N) ) 1 - s p ( 1 - τ -1 10 -2 10 nosc(A) ee(A) ei(A) R=34 nm R=40 nm R=38.5 nm R=43 nm 50 100 150 200 T (K) 250 300 50 100 200 150 T (K) 250 300 FIG. 7: (Color online) Comparison between the numerical and analytical results of the SPP damping rate for the ee and ei scatterings for R = 34, 38.5, 40 and 43 nm. Small green squares and green dotted curves represent the numer- ical and analytical results for the ee scattering, respectively. Small olive dots and olive double-dotted chain represent the numerical and analytical results for the ei scattering, respec- tively. For comparison, the numerical and analytical results for the damping rate without scattering are also plotted with red squares and solid curves, respectively. with the ee/ei scattering can also be calculated from the analytic solution [Eqs. (28) and (29) with the broaden- ing and shifting given in Eqs. (33) and (37) for the ei scattering and Eqs. (35) and (39) for the ee scattering, respectively]. The analytic results are compared to the numerical ones in Fig. 7. From the figure, one observes qualitatively good agreement between each other, indi- cating that the ee and ei scattering can also be under- stood as the broadening and/or shifting of the resonant pairs. It is pointed out that the effect of the broadening and shifting can be visualized from the behavior of the po- larization, which gives a more intuitive picture for the effect of the scattering. This is discussed in detail in Appendix E. From the above results, one finds that the scatter- ing tends to smear out the size-dependent oscillations of the SPP damping rate. The temperature depen- dence can also be suppressed by the scattering. Note that this effect is quite general and can be seen for nanowires with different electron densities. To demon- strate this, we show the size and temperature depen- dence of the SPP damping rates for nanowires with elec- tron density ¯n0 = 5.0 × 1017 cm−3 without and with scattering in Figs. 8(a) and (b), respectively. The cen- tral wavevector of the SPP wave packet is chosen to be Qs = 1.3×10−2 nm−1. From the figure, it is seen that the size-dependent oscillations are smeared out by the scat- tering. The temperature dependence is also suppressed. These effects are similar to the ones for nanowires with ¯n0 = 1.5× 1017 cm−3 investigated above, indicating that these effects are quite general for typical InAs nanowires. τ-1 τ-1 (ps) (ps) 0 10 -3 10 -6 10 40 38 34 36 R (nm) 32 30 28 τ-1 τ-1 (ps) (ps) 0 10 -3 10 -6 10 40 38 34 36 R (nm) 32 30 28 (a) 300 100 200 T ( K ) (b) 300 100 200 T ( K ) FIG. 8: (Color online) Size and temperature dependence of the SPP damping rate for nanowires with electron density ¯n0 = 5.0 × 1017 cm−3 (a) without any scattering and (b) with all the scattering. The central wavevector of the SPP wave packet is chosen to be Qs = 1.3 × 10−2 nm−1. IV. CONCLUSION AND DISCUSSION In conclusion, we present a microscopic kinetic theory to study the effect of the electron scattering on the Lan- dau damping of the SPP in semiconductor nanowires. Based on the semiclassical model Hamiltonian of the SPP-electron system and the nonequilibrium Green- function approach, we derive the kinetic equations of the SPP-electron system, with all the scattering explicitly included. Within this model, the SPP damping is un- derstood as the absorption of the SPP by the electron polarization of the resonant pairs. The population dif- ference of the resonant pairs δf plays an important role on the SPP damping, leading to a strong and a weak Landau damping regimes for degenerate electrons. The scattering influences the SPP damping via the broaden- ing and shifting of the resonant pairs, which have dif- ferent effects on the strong and weak Landau damping regimes. At low temperatures, the broadening tends to suppress the SPP damping in the strong Landau damping regime. Whereas in the weak Landau damping regime, the broadening tends to enhance the SPP damping. At high temperatures, this difference tends to be vanished. The shifting can suppress the SPP damping if the reso- nance is shifted towards the region with smaller δf , but boost it if the resonance is shifted toward the region with larger δf . The broadening and shifting can be visualized from the corresponding electron polarization of the res- 12 onant pairs. Moreover, different scattering has different contribution to the broadening and shifting. The main effect of the ei/ep scattering is to cause a broadening, whereas the main effect of the ee scattering is to intro- duce a shifting. The effect of the broadening and shift- ing can be incorporated into an analytic solution, which shows good agreement with the numerical result. To demonstrate the effect of the scattering, we inves- tigate the damping of the axial symmetric SPP mode in InAs nanowires in the presence of the ei, ee and ep scatterings. Without any scattering, the SPP damping exhibits size-dependent oscillations and a distinct tem- perature dependence. In the presence of the scattering, the size-dependent oscillations are markedly smeared out and the temperature dependence is also suppressed. The damping rate can be enhanced/suppressed by almost one order of magnitude. For InAs nanowires investigated here, the ep scattering is found to be dominant. These results are found to be general for typical InAs nanowires, which demonstrate the importance of the scattering on the SPP damping for semiconductor nanowires. It is further pointed out that our model can be applied to nanowires made of other semiconductors, offering a sys- tematic way to investigate the effect of electron scattering on the SPP damping in such systems. Acknowledgments This work was supported by the National Ba- sic Research Program of China under Grant No. 2012CB922002 and the Strategic Priority Research Pro- gram of the Chinese Academy of Sciences under Grant No. XDB01000000. One of the authors (YY) was also partially supported by the China Postdoctoral Science Foundation. Appendix A: Semiclassical model for SPP Our derivation of the quantized SPP Hamiltonian HSPP and its coupling to electrons HSPP-el follows the procedure used in Ref. 80, in which the SPPs are ob- tained from quantization of the classical plasmon field. The classical plasmon field is derived within the hydro- dynamical model, which treats plasmons as irrotational deformations of the conduction electrons.80 -- 84 From this model, the Hamiltonian of the plasmon can be written as80 n(r)∇η(r)2 2 Z dr′n(r)n(r′)Vee(r, r′) e2 (A1) HSPP = Z dr m∗ 2 e2 + 2 Z dr +Z drG[n(r)], where the irrotational flow has been assumed, i.e., v(r) = ∇η(r) with v(r) being the velocity field. n(r) is the electron density and m∗ represents the electron effective mass. For both r and r′ inside the nanowire, the Coulomb interaction Vee(r, r′) has the form85 Vee(r, r′) = e 1 h 1 r − r′ 4πǫ∞ + 2 π ( dk cos[k(z − z′)]Cm(kR, +∞ ǫ∞ Xm=−∞ 1 ǫ2 − 1) )Im(kρ)Im(kρ′)i, ǫ∞ 1 ǫ2 eim(ϕ−ϕ′) (A2) ×Z ∞ 0 where Cm(kR, ǫ∞ 1 ǫ2 ) = Im(kR)K ′ Km(kR)K ′ m(kR) − ǫ∞ 1 ǫ2 m(kR) I ′ m(kR)Km(kR) . the distortion of (A3) Note that the Coulomb interac- tion Vee(r, r′) by the dielectric mismatch between the nanowire and the environment has been taken into ac- count, which is not addressed in previous work. G[n(r)] represents the exchange, correlation and inter- nal energy of the electrons, which is approximated by the Thomas-Fermi functional By substituting the ansatz Eq. (A7) into Eqs. (A5) and (A6), one obtain the equation for the parameter κ 13 β2 ω2 p 1 − ×nC′ (κ2 − q2) − (qR)Im+1(qR)Km(qR) qm(cid:16) κR +h κR Im+1(κR) Im+1(qR) − Im+1(κR) Im+1(qR) Im(qR)(cid:17) A(1 + C′ Im(κR) qm) qR qR Im+1(κR) Im+1(qR) + Im(κR) Im+1(qR) Km+1(qR) Km(qR) (cid:17)io = 0, (A8) being the bulk plasma frequency and . The parameter A can be expressed as Im+1(qR)(qR)(1 + C′ qm) Im+1(κR)(κR)(cid:16)1 + C′ qm (qR)Im+1(qR)Im (κR) (κR)Im+1(κR)Im(qR)(cid:17) (A9) , qR − A(cid:16) κR with ωp =q e2 ¯n0 β =q 5γ 3m∗ ¯n2/3 A = − 0 m∗ǫ∞ 1 where G[n(r)] = 3 10m∗ (3π2)2/3n5/3(r), (A4) C′ m(qR) = 1 + ǫ∞ 1 ǫ2 ǫ∞ 1 ǫ2 − 1 Im+1(qR) Km(qR) . (A10) Im(qR) Km+1(qR) with the exchange-correlation contributions neglected. From the above Hamiltonian, up to the first order of the perturbation, one can derive linearized hydrodynamic equations as ∂tn1(r) = ∇ · [n0(r)∇η1(r)], ∂tη1(r) = m∗ Z dr′Vee(r, r′)n1(r′) e + 5γ 3m∗ n1(r) [n0(r)]1/3 , (A5) (A6) where n1 and η1 are perturbations around the equilib- rium. n0(r) = ¯n0Θ(R − ρ) is the electron density in the equilibrium with ¯n0 being the average electron density. γ = (3π)2/3 5m∗ . The normal modes from the above equations include both the surface plasmon and volume plasmon modes. The surface modes, which we focus on in this paper, can be represented by the ansatz83 ηqm(r) = RXq Qqmeimϕ+iqzhIm(qρ) + AIm(κρ)i, (A7) where Im(ρ) is the modified Bessel function of the first kind. A and κ are parameters depending on m and q, with m being the angular quantum number of the SPP mode and q representing the SPP wave vector along the wire axis. Here the first term in the square bracket rep- resents the incompressible deformation of the electrons while the second term represents the correction due to the finite compressibility. Once κ is obtained, the corresponding eigen-frequency Ωqm can be calculated from the equation ω2 p Ω2 qm − Ω2 1 + β2 κ2 − q2 ×h(κR)Im+1(κR)Im(qR) − (qR)Im(κR)Im+1(qR)io ǫ∞ 1 ǫ2 − 1)Cm(qR) qmn1 − ( = 0. (A11) Note that without the dielectric mismatch (ǫ∞ 1 = ǫ2), Eqs. (A8-A11) agree with the results for the surface mode from Ref. 83. The Hamiltonian of the SPP can be obtained by sub- stituting Eq. (A7) into Eq. (A1) and keeping only terms up to the linear order, which can be written as H = 2πR3m∗¯n0 2 Xqm h Q′ qm2 + Ω2 qmQ′ qm2i, where the canonical coordinate Q′ qm has the form Q′ qm = rhAIm(κR)κ + Im(qR)qi ×rhAIm(κR) + Im(qR)iQqm. Following the canonical quantization procedure, the col- lective coordinate Q′ qm can be transformed into Q′ qm = s 1 2πR3m∗¯n0Ωqm (bqm + b† −qm), (A14) (A12) (A13) where the annihilation(creation) operator bqm(b† qm) sat- isfies the boson commutation relation [bqm, b† q′m′] = δq,q′ δmm′ . Substituting Eq. (A14) into Eq. (A12), the Hamiltonian of the SPP is quantized as HSPP = Xqm Ωqm 2 (2b† qmbqm + 1). (A15) By choosing m = 0, one obtains the Hamiltonian of the axial symmetric SPP mode under investigation HSPP = Xq Ωqb† qbq, (A16) without taking into account the zero-point energy which is irrelevant for our study. The angular quantum number m is also omitted without confusion. Following the similar procedure, the induced potential of the SPP mode can be quantized as V in(ρ, ϕ, z) =Z dq V in q (ρ, ϕ, z)(bq + b† −q), (A17) where V in q (ρ, ϕ, z) = − e2 1 s ¯n0 2πǫ∞ I0(qρ)C′′ 2πm∗ωp q + AI0(κρ) eiqz pΩq p[AI1(κR)κR + I1(qR)qR][AI0(κR) + I0(qR)] (A18) , × with q = −h(κR)AI1(κR) + (qR)I1(qR)i C′′ ×hK0(qR) + C′ + Ah(κR)I1(κR)K0(qR) + (qR)I0(κR)K1(qR) 0(qR)K0(qR)(cid:16)(κR)I1(κR) − (qR)I0(κR) 0(qR)K0(qR)i + C′ I1(qR) I0(qR)(cid:17)i. (A19) The SPP-electron coupling Hamiltonian HSPP-el in Eq. (9) can be obtained by combining the single electron states in Eq. (5) and the induced potential of the SPP given above. The corresponding matrix element reads gnn′ k−k′ =Z ρdρdϕdzψ∗ nk(ρ, ϕ, z) V in k−k′ (ρ, ϕ, z)ψn′k′ (ρ, ϕ, z), (A20) with ψnk given in Eq. (5). Now let us discuss the matrix elements for the inter- action Hamiltonians for electrons. For the ei interaction, we have assumed that the impurities are distributed on the surface of the nanowire with an axially symmetric distribution.74 Given the electron wavefunction ψnk in Eq. (5), the corresponding matrix element vnn′ calculated as q can be 14 = vnn′ q e2δ m m′ 2π2ǫ∞ 1 RZ 1 ×hI0(qR¯ρ)K0(qR) + C′ 0 d¯ρ¯ρ J m(λ m J m+1(λ m+1 n n ¯ρ)J m′ (λ m′ n′ ¯ρ) )J m′+1(λ m′+1 n′ ) 0(qR)I0(qR¯ρ)K0(qR)i. (A21) Note that the dielectric mismatch effect has been taken into account in the above expression. For the ep inter- action, we here consider the polar interaction between the electrons and the LO phonons. The corresponding matrix element reads 1 − 1/ǫ0 1) n )J m′+1(λ m′ n′ ) M nn′ J m+1(λ m Qq = p2πe2ΩLO(1/ǫ∞ ×Z 1 n ¯ρ)J m′ (λ m′ d¯ρ¯ρJ m(λ m 0 2 q n′ ¯ρ)ei¯ρQ·eρRδ m m′, (A22) where q = (Q, q) represents the LO phonon wave vector and ΩLO is the LO phonon energy. eρ is the unit vector along the radial direction of the wire and ǫ0 1 is the dielec- tric constant of the nanowire in the static limit. Note that here we model the LO phonons by 3D modes, which is adequate for nanowires with large diameter.68,69 For the ee Coulomb interaction, the corresponding matrix el- ement reads V nn′ q = 2e2 πǫ∞ 1 Z 1 0 d¯ρ1 ¯ρ1Z 1 0 d¯ρ2 ¯ρ2hI0(qR¯ρ<)K0(qR¯ρ>) + ( ǫ∞ 1 ǫ2 − 1)I0(qR¯ρ1)I0(qR¯ρ2)C0(qR, ǫ∞ 1 ǫ2 )i (A23) n ¯ρ1) J m+1(λ m n ) ×h J m(λ m J m′ (λ m′ n′ ¯ρ2) J m′+1(λ m′ n′ )i2 , where ρ> = max(ρ1, ρ2) and ρ< = min(ρ1, ρ2). Here we have omitted the interband term of the Coulomb in- teraction which can be small for nanostructures.86 Note that the dielectric mismatch effect has been taken into account in the ee Coulomb interaction. Appendix B: Derivation of kinetic equations The time evolution of Bq(t) = hbq(t)i can be derived from the Heisenberg equation of the annihilation oepra- tor bq, which reads ∂tBq(t) = −iΩqBq(t)+ Xnn′,kk′,σ k−k′ gnn′ pQs k−k′ G< σ (n′k′, nk; tt). (B1) By using the definitions of ¯Bs and P given below Eq. (14), one has ∂t ¯Bs(t) = −iXq − i Xnn′σXkk′ (ΩQs+q − Ωs)pQs q BQs+qeiΩst k−k′ gnn′ pQs k−k′hPσ(nk, n′k′; t)i† , (B2) where the first term in the right hand side of the equation describes the effect of the SPP dispersion on the dynam- ics of the SPP wave packet and the second term describes the dissipative effect due to the coupling to electrons. For the SPP wave packet with sufficiently large length L, the line-shape function pQs exhibits a sharp symmetric peak around the central wave vector Qs. In this case, one can perform the Taylor expansion of the dispersive relation ΩQs+q around Qs up to the linear order of q to elimi- nate the first term. In doing so, we omit the effect of the SPP dispersion on the wave packet dynamics and consider only the dissipative effect. q Within the nonequilibrium Green-function approach, the quantum kinetic equation of electrons can be derived following the Kadanoff-Baym method,66,67 yielding σ (nk, n′k′; tt) k )iG≷ k′ − εn (B−q + B† q G≷ σ (nk, ¯nk′ − q; tt) q )hg ¯nn′ h − i∂t − (εn′ =X¯nq − gn¯n q G≷ σ (¯nk + q, n′k′; tt)i + I ≷σ nk,n′k′ (t), (B3) where the first term in the right hand side of the equation is the coherent driving term of electrons by the SPP and the second term describes the scattering I ≷σ nk,n′k′ (t) = Z t −∞ d¯tX¯k nhΣ≶(k¯k; t¯t)G≷ σ (¯kk′; ¯tt) − G≷ σ (k¯k; t¯t)Σ≶(¯kk′; ¯tt))i −h >←→<io, (B4) where [>←→<] stands for the same term as in the pre- vious [ ] but with the interchange of >←→<. It should be emphasized that due to the driving of the SPP, the dressed Green function G≷ σ has off-diagonal components with respect to the momentum k and subband index n. By treating the driving term as a perturbation, the kinetic equation can be linearized by keeping only terms up to the linear order, k′ − εn h − i∂t − (εn′ = (Bk−k′ + B† − gnn′ σ (nk, n′k′; tt) k )iG≷ k′−k)hgnn′ 0σ(n′k′, n′k′; tt)i + I ≷σ k′−kG≷ k′−kG≷ 0σ(nk, nk; tt) nk,n′k′ (t). (B5) The first term in the right hand side of the equation is the linearized driving term. Note that within the lineariza- tion, the dressed Green function G≷ σ in the driving term is replaced by the free electron Green function G≷ 0σ, which is diagonal with respect to the momentum k and sub- band index n. By using the definition Pσ(nk, n′k′; t) = σ (nk, n′k′; tt)eiΩst and fnσ(k) = −iG< −iG< 0σ(nk, nk; tt), one obtains Eq. (13) from the above equation. Now we turn to the scattering term. We first discuss the ei scattering. Following the standard approximation,66 the corresponding scattering term can be expressed as 15 I ei,σ nk,n′k′ = −hSei,σ(nk, n′k′; >, <) − S† +h >←→<i, Sei,σ(nk, n′k′; >, <) =Z t d¯t X¯n¯n′ nXkq −∞ ei,σ(n′k′, nk; <, >)i (B6) nivn¯n q v n¯n′ q σ (nk, n′k′; ¯tt), (B7) × G> σ (¯nk − q, ¯n′k − q; t¯t)G< = vnn′ q q /ǫnn′ where vnn′ (q) is the screened electron- impurity interaction. The screening ǫnn′ (q) is evaluated following Ref. 87 in the static limit, which can be written as ǫnn′ (q) = 1 −Xlσ q V ln′ V nl q V nn′ q Xk flσ(k + q) − flσ(k) εl k+q − εl k − i0+ .(B8) Note that Sei,σ contains the product of two dressed Green functions. By keeping only terms up to the linear order, Sei,σ can be linearized as Sei,σ(nk, n′k′; >, <) =Z t × G> −∞ d¯tX¯nn Xkq 0σ(¯nk − q, ¯nk − q; t¯t)G< d¯tX¯n¯n′ Xkq σ (¯nk − q, ¯n′k′ − q; t¯t)G< +Z t −∞ nivn¯n q v n¯n q σ (nk, n′k′; ¯tt) nivn¯n q vn′ ¯n′ q × G> 0σ(n′k′, n′k′; ¯tt). (B9) Note that within the linearization, Sei,σ is separated into two terms. Each term contains only one dressed Green function G≷ σ , while the other one is replaced by the free Green function G≷ 0σ. By applying the rotating wave approximation with respect to the SPP frequency Ωs, one can remove the non-frequency-matched terms in the above equation, yielding Sei,σ(nk, n′k′; >, <)e−iΩst = πni X¯n¯n′q δ(ε¯n k−q − εn′ k′ + Ωs) q ×hvn¯n q v ¯n¯n′ q v ¯n′n′ − vn¯n q ¯nσ(k − q)Pσ(¯n′k, n′k′) f > n′σ(k′)Pσ(¯nk − q, ¯n′k′ − q)i(B10) f < in the Markovian limit. Substituting Eq. (B10) into Eqs. (B6) and (B7), and using the definition ¯I σ nk,n′k′ = I σ nk,n′k′ eiΩst, one gets the final expression of the impurity scattering term ¯I ei,σ k′ + Ωs) q v ¯n′n′ k−q − εn′ nk,n′k′ = πni X¯n¯n′qnδ(ε¯n ×hvn¯n ×hvn¯n q Pσ(¯nk − q, ¯n′k′ − q) − vn¯n k − Ωs) q Pσ(¯nk − q, ¯n′k′ − q) − v ¯n¯n′ k′−q − εn q v ¯n′n′ + δ(ε¯n′ q q v ¯n¯n′ q Pσ(¯n′k, n′k′)i q Pσ(nk, ¯nk′)io. v ¯n′n′ (B11) The ep scattering term can be derived following the sim- ilar procedure, which reads ¯I ph,σ QqM ¯n′n′ nk,n′k′ = πX¯n¯n′ XqQ nδ(ε¯n ×hM n¯n − M n¯n Qq (cid:16)N > Qq (cid:16)N > QqM ¯nn′ LOf > ×hM n¯n − M n¯n QqM ¯n′n′ Qq (cid:16)N < Qq (cid:16)N < QqM ¯nn′ LOf > QqM ¯n′n′ ×hM n¯n − M n′ ¯n′ Qq M ¯n¯n′ Qq (cid:16)N > Qq (cid:16)N > QqM ¯n′n′ ×hM n¯n − M n′ ¯n′ Qq M ¯n¯n′ + δ(ε¯n′ Qq (cid:16)N < Qq (cid:16)N < k+q − εn′ k′ + Ωs + ΩLO) LOf < LOf > + δ(ε¯n n′σ(k′)(cid:17) n′σ(k′) + N < LOf < × Pσ(¯nk + q, ¯n′k′ + q) ¯nσ(k + q)(cid:17) ¯nσ(k + q) + N < × Pσ(¯n′k, n′k′)i k+q − εn′ k′ + Ωs − ΩLO) n′σ(k′)(cid:17) LOf > n′σ(k′) + N > LOf < × Pσ(¯nk + q, ¯n′k′ + q) ¯nσ(k + q)(cid:17) ¯nσ(k + q) + N > × Pσ(¯n′k, n′k′)i k′+q − εn k − Ωs + ΩLO) nσ(k)(cid:17) LOf > LOf < nσ(k) + N < × Pσ(¯nk + q, ¯n′k′ + q) LOf < ¯n′σ(k′ + q) + N < + δ(ε¯n′ LOf < LOf > ¯n′σ(k′ + q)(cid:17) ¯n′σ(k′ + q)(cid:17) (B12) × Pσ(nk, ¯nk′)i k′+q − εn k − Ωs − ΩLO) nσ(k)(cid:17) LOf < nσ(k) + N > × Pσ(¯nk + q, ¯n′k′ + q) LOf < ¯n′σ(k′ + q) + N > LOf > LOf > × Pσ(nk, ¯nk′)io, in the similar way, 16 ¯I ee,σ nk,n′k′ = 2πXq¯n¯knδ(εn k−q − εn′ k′ + Ωs + ε¯n ¯k+q − ε¯n ¯k ) ×hf < n′σ(k′)Πn − f > ×hf > n′σ(k′)Πn′ − f < ×hf < nσ(k)Πn − f > ×hf > nσ(k)Πn′ − f < ¯k−q) k′ + Ωs + ε¯n nσ(k − q)Πn nσ(k − q)Πn n′ (¯n¯k, q)Pσ(nk − q, n ′k ′ − q) n(¯n¯k, q)Pσ(nk , n ′k ′)i k−q − εn′ ¯k − ε¯n + δ(εn n (¯n¯k,−q)Pσ(nk − q, n ′k ′ − q) n(¯n¯k,−q)Pσ(nk , n ′k ′)i k − εn′ + δ(εn ¯k − ε¯n n′ (¯n¯k, q)Pσ(nk − q, n ′k ′ − q) n′σ(k′ − q)Πn′ k − εn′ n′ (¯n¯k, q)Pσ(nk , n ′k ′)i ¯k−q − ε¯n + δ(εn ¯k ) n (¯n¯k,−q)Pσ(nk − q, n ′k ′ − q) n′σ(k′ − q)Πn′ k′−q + Ωs + ε¯n k′−q + Ωs + ε¯n ¯k+q) n′ (¯n¯k,−q)Pσ(nk , n ′k ′)io, (B13) where Πn q = V nn′ V nn′ n′ (¯n¯k, q) = Pσ /ǫnn′ q V n¯n q ¯nσ(¯k) with (q) being the screened ee interaction. ¯nσ(¯k + q)f < f > V ¯nn′ q Appendix C: Scattering-induced frequency modification and decay of the polarization The polarization P for the i-th resonant pair in the presence of the scattering term ¯I σ nk,n′k′ given in Eq. (23) can be solved by treating Γi as perturbation and solve Eq. (13) order by order. By assuming Pσ(nk, n′k′) = ∞ Xj=0 one obtains P (j) σ (nk, n′k′; t), (C1) ∂tP (0) + ignn′ σ (nk, n′k′) = iωnn′ k−k′ pQs k−k′ ¯B† kk′ P (0) σ (nk, n′k′; t) shfnσ(k) − fn′σ(k′)i, for 0-th order and (C2) where N <(>) 2 with NLO = LO 1/[exp( ΩLO kB T ) − 1] representing the thermal LO phonon distribution. The ee scattering term can also be derived 2 − (+) 1 = NLO + 1 ∂tP (j+1) σ (nk, n′k′) = iωnn′ kk′ P (j+1) σ (nk, n′k′; t) Γi[P (j) σ (nk − q, n′k′ − q) − P (j) +Xq for (j + 1)-th order, with j ≥ 0. In the long time limit t → ∞, the solution of the above equations reads σ (nk, n′k′)], (C3) σ (nk, n′k′) = −gnn′ P (0) k−k′ pQs k−k′ ¯B† × [fnσ(k) − fn′σ(k′)]/(ωnn′ s kk′ + i0+), (C4) for 0-th order, and P (j+1) σ (nk, n′k′) = P (j) σ (nk, n′k′) ωnn′ kk′ + i0+ P (j) σ (nk − q, n′k′ − q) P (j) σ (nk, n′k′) − 1], (C5) [ × iΓiXq for (j + 1)-th order. 0 10 ) 1 - s p ( 1 - τ -1 10 -1 10 ) 1 - s p ( 1 - τ R=34 nm 17 nosc LO+SO LO -2 10 R=38.5 nm 50 100 150 200 250 300 T (K) FIG. 9: Temperature dependence of SPP damping rate in the presence of the ep scattering for R = 34 and 38.5 nm. Blue dashed curves represent the results with only the LO phonon scattering, while the brown solid curves represent the results with both the LO and SO phonon scatterings. The damping rates without any scattering are plotted with red double-dotted chain curves for comparison. Other parameters are all the same as Fig. 6. (C6) Appendix D: Effect of the SO phonons order, σ (nk−q,n′k′−q) σ (nk,n′k′) − 1]. For 0-th P (j) Now let's evaluate Pq[ P (j) Xq h P (0) σ (nk − q, n′k′ − q) =Xq h × ≈Xq k−q,k−q′ + i0+ − 1i k−q,k−q′ + i0+ − 1], ωnn′ ωnn′ kk′ ωnn′ [ − 1i P (0) σ (nk, n′k′) gnn′ k−k′ pQs k−k′ pQs k−k′ ¯B†[fnσ(k − q) − fn′σ(k′ − q)] gnn′ ωnn′ kk′ k−k′ ¯B†[fnσ(k) − fn′σ(k′)] where we have assumed that fnσ(k) varies slowly with respect to k and can be cancelled out. The 1-st order term can be evaluated as σ (nk − q, n′k′ − q) P (1) σ (nk, n′k′) − 1i σ (nk − q − q′, n′k′ − q − q′) [P (0) Xq h P (1) =Xq nXq′ − P (0) [P (0) σ (nk − q, n′k′ − q) σ (nk − q, n′k′ − q)]{Xq σ (nk, n′k′)]}−1 ωnn′ ωnn′ kk′ ωnn′ kk′ k−q,k−q′ + i0+ − 1], ωnn′ − P (0) ≈Xq [ k−q,k−q′ + i0+ − 1o where we have assumed that Pq[P (0) σ (nk − q, n′k′ − q) − P (0) σ (nk, n′k′)] varies slowly with respect to k(k′) and can be cancelled out. Following the similar procedure, for j-th order, one has P (j) σ (nk − q, n′k′ − q) P (j) σ (nk, n′k′) − 1] [ ωnn′ kk′ k−q,k−q′ + i0+ − 1], ωnn′ (C8) [ Xq ≈Xq with j ≥ 1. one obtains Eqs. (24-26). By combining Eqs. (C1), (C4), (C5), (C6) and (C8), (C7) D(ω) = The SO phonons in nanowires69 can also influence the damping of the SPP. However, for the typical InAs nanowires we considered here, the SO phonons are of marginal importance since their contribution is much smaller than that from the LO phonons. This will be shown in this Appendix. The induced potential due to the SO phonon is given by88 -- 90 with VSO = Xp (ΓSO p eiqzeipϕaSO p + h.c.), (D1) ΓSO p = s 2πe2R q D(ΩSO) Ip(qR)I ′ p(qR) ǫ2(ω) Ip(qr), (D2) ǫi(ω) = ǫ∞ i ∂ω ǫ2(ω) ǫ2(ω) ∂ ∂ω ǫ1(ω) − ǫ1(ω) ∂ Ω2 Li − ω2 Ti − ω2 , Ω2 , (D3) (D4) p where aSO is the annihilation operator for the SO phonons and ΩSO is the SO phonon energy. ΩLi(ΩTi) represents the LO(TO) phonon energy inside (i = 1) and/or outside (i = 2) the nanowire. The coordinates are chosen according to Fig. 1(a). As the SO phonon energy is very close to the LO phonon energy for typical InAs nanowires,69 we assume ΩSO = ΩLO in the calculation. By using the single elec- tron states in Eq. (5), the corresponding electron-SO- phonon interaction Hamiltonian can be written as HSO = Xpq Xnn′kσ ¯M nn′ pq [aSO pq + (aSO −p−q)†]c† nkσcn′k−qσ,(D5) where the interaction matrix element reads 18 1 − 1/ǫ0 1)R/q n )J m′+1(λ m′ n′ ) J m+1(λ m δ m, m′+p ¯M nn′ pq = pπe2ΩSO(1/ǫ∞ ×Z 1 n ¯ρ)J m′ (λ m′ d¯ρ¯ρJ m(λ m 0 n′ ¯ρ) Ip(qR¯ρ) qIp(qR)I ′ p(qR) . (D6) By using the analytic solution given in Secs. II D and II E, we calculate the temperature dependence of the SPP damping rates for R = 34 and 38.5 nm in the pres- ence of the ep scattering with and without the contribu- tion of the SO phonons, which is plotted in Fig. 9. One can see that the damping rates are dominated by the LO phonons and the SO phonons have very little effect. Appendix E: Visualization of the broadening and shifting We have seen that the SPP damping rate can be un- derstood as the broadening and shifting of the resonant pairs. In the analytic solution, the broadening and shift- ing can be seen clearly from both the resonant denomi- nator in the polarization [Eq. (24)] and the Lorentzian in the damping rate [Eq. (29)]. One may wonder whether the broadening and shifting can also be observed in a more intuitive way from the numerical results. In fact, inspired by Eq. (24), the broadening and shifting of the resonant pairs can be visualized from the normalized po- larization P (nk, n′k′)/(δf · ¯Bs), from which the structure of the resonant denominator can be identified. Before we show the numerical results of the normal- ized polarizations, we first explain what we expect from the polarizations. According to the analytic solution Eq. (15), the magnitude of the normalized polarization P (nk, n′k′)/(δf · ¯Bs) without scattering can exhibit a sharp Lorentzian peak around the resonance correspond- ing to the SPP central wavevector Qs as indicated by the resonant denominator. Several side peaks can also exist as the SPP wave packet is nonmonochromatic [e.g., the line-shape function pQs 6= δ(q − Qs)]. With scattering, these peaks can be shifted and broadened, indicating the shifting and broadening of the corresponding resonant pairs. Small side peaks may also be smeared out by the broadening. Note that with the scattering, the polariza- tions from the numerical results can have more complex behaviors. The peaks in the polarization can also be dis- torted by the scattering, which has been omitted in the analytic solution. q We first concentrate on the typical case with R = 34 nm and T = 150 K corresponding to the strong Lan- dau damping regime. The magnitudes of the normalized electron polarization P (nk, n′k′)/(δf· ¯Bs) from the com- putation are plotted in Fig. 10(a). In the figure, polar- izations with different scatterings are plotted by curves with different colors. The contour plots of the polariza- tions are also shown in the K-q plane, which are useful FIG. 10: (Color online) 3D plot and the corresponding con- tour plot of the normalized (a) and unnormalized (b) elec- tron polarization corresponding to the resonant pair (i) for R = 34 nm, T = 150 K. The corresponding population differ- ence is also plotted in the figure. The olive, green and blue curves represent the polarization with the ei, ee and ep scat- tering only, respectively. The red curves represent the polar- ization without scattering, while the brown curves represent the polarization in the presence of all the three scatterings. In the contour plots, the resonance corresponding to the SPP central wavevector Qs is shown with skyblue dot in the con- tour plot. The population difference is plotted with orange curves. Different curves are offset along q-axis for clarify. To provide a clear visualization, in (a) the normalized polariza- tions with the ep and all the scatterings are enlarged by a factor of 2. In (b), the unnormalized polarization with the ee/ep/all scattering is enlarged by a factor of 4/30/30. for identifying the shape and position of the polariza- tions. The population differences δf = fn′(k′)−fn(k) for the resonant pairs are also plotted with orange curves for comparison, with the corresponding contour map plot- ted in the K-q plane. The polarizations with different scatterings and the population difference have been off- set along the q-axis for clarification. The corresponding unnormalized electron polarizations P (nk, n′k′)/ ¯Bs are also plotted in the similar way in Fig. 10(b) for compar- ison. Note that all the polarizations are taken at time t = 2.86 ps. Polarizations taken at other time show sim- ilar behaviors. In Fig. 10(a), the polarization is plotted for the reso- nant pair (i) which is the only relevant resonant pair for the damping. The resonant pair is centralized around the resonance corresponding to the SPP central wavevector Qs, which is represented by the skyblue dots in Fig. 10(a). One finds that without scattering, the corresponding po- larization (red curve) exhibits a sharp main peak around the resonance. Several side peaks exist around the main 19 that the ei scattering introduces a small broadening. In the presence of the ee scattering, a shifting of the peaks can be seen clearly by comparing the contour plot of the corresponding polarization (green curve) to the one without scattering (red curve). Note that small side peaks can also be identified with the ee scattering. These features indicate that the ee scattering introduces a large shifting to the corresponding resonant pairs. The broadening due to the ee scattering is small since the small side peaks are not smeared out. It is also noted that the ee scattering can distort the polarization. The peak becomes fragmented and the profile becomes non- Lorentzian. However, as the influence to the SPP damp- ing comes from the summation of the polarizations as indicated by Eq. (12), the effect of these distortion on the SPP damping is marginal. In the presence of the ep scattering, the peak in the polarization (blue curve) is markedly broadened and all the side peaks are smeared out, indicating that the ef- fect of the ep scattering introduces a large broadening. Note that the maximum of the broadened peak is also shifted compared to the one without scattering. This shows that the ep scattering can also have contribution to the shifting of the resonant pairs. However, due to the large broadening, the effect of such shifting on the SPP damping is marginal. We also point out that due to the large broadening, the profile of the unnormalized polar- ization is mainly determined by the population difference rather than the resonance, which can be seen from the corresponding unnormalized polarization (blue curve) in Fig. 10(b). In the presence of all the three scatterings, one finds that the polarization (brown curve) has almost the same profile as the one with the ep scattering only. The broad- ening and shifting due to the ei and ee scatterings are negligible due to the large broadening introduced by the ep scattering. This indicates that the ep scattering plays the dominant role. For the weak Landau damping regime, as the scat- tering has different influence on the SPP damping rate in different temperature regimes. We plot the normal- ized polarizations in Fig. 11 for three typical cases: (a) R = 38.5 nm, T = 100 K, (b) R = 38.5 nm, T = 150 K and (c) R = 38.5 nm, T = 200 K, corresponding to the low, intermediate, and high temperature regimes, respec- tively. The polarizations are plotted for the resonant pair (ii) which dominates the SPP damping in these cases. Similar broadening and shifting can also be identified in the figure. Note that in these cases, the shifting due to the ee scattering is rather small and difficult to be iden- tified in the figures. FIG. 11: (Color online) 3D plot and the corresponding con- tour plot of the normalized electron polarization correspond- ing to the resonant pair (ii) for (a) R = 38.5 nm, T = 100 K, (b) R = 38.5 nm, T = 150 K and (c) R = 38.5 nm, T = 200 K. The corresponding population difference is also plotted in the figure. The resonance corresponding to the SPP central wavevector Qs is shown with green square in the contour plot. peak. These features agree with the previous discussion. Note that the main peak lies in the regime with large δf , indicating that the SPP absorption by the polarization is strong. Also note that due to the strong resonance, the profile of the unnormalized polarization is mainly de- cided by the resonance and shares a similar structure as the normalized one. In the presence of the ei scattering, the polarization is slightly modified. Some side peaks are smeared out. However, the broadening is rather small and the main peak is almost unchanged. This can be better seen by comparing the polarization with the ei scattering (olive curve) to the one without scattering (red curve). Ac- cording to the previous discussion, these features indicate ∗ Electronic address: yin80@ustc.edu.cn. † Electronic address: mwwu@ustc.edu.cn. 1 R. H. Ritchie, Phys. Rev. 106, 874 (1957). 2 C. J. Powell and J. B. Swan, Phys. Rev. 118, 640 (1960). 3 C. K. Chen, A. R. B. de Castron, and Y. R. Shen, Phys. Rev. Lett. 46, 145 (1981). 4 A. Wokaun, J. P. Gordon, and P. F. Liao, Phys. Rev. Lett. 48, 957 (1982). 20 5 B. Rothenhausler and W. Knoll, Nature 332, 615 (1988). 6 U. Kreibig and M. Vollmer, Optical properties of metal clusters (Springer-Verlag, Berlin, 1995). 7 A. V. Zayats, I. I. Smolyaninov, and A. A. Maradudin, Phys. Rep. 408, 131 (2005). 8 J. M. Pitarke, V. M. Silkin, E. V. Chulkov, and P. M. 39 A. Liebsch, Phys. Rev. B 36, 7378 (1987). 40 K. D. Tsuei, E. W. Plummer, and P. J. Feibelman, Phys. Rev. Lett. 63, 2256 (1989); K. D. Tsuei, E. W. Plummer, A. Liebsch, E. Pehlke, K. Kempa, and P. Bakshi, Surf. Sci. 247, 302 (1991). 41 P. T. Sprunger, G. M. Watson, and E. W. Plummer, Surf. Echenique, Rep. Prog. Phys. 70, 1 (2007). Sci. 269, 551 (1992). 9 F. J. Garcia-Vidal, L. Martin-Moreno, T. W. Ebbesen, and 42 E. Zaremba and B. N. J. Persson, Phys. Rev. B 35, 596 L. Kuipers, Rev. Mod. Phys. 82, 729 (2010). (1987). 10 H. Raether, Surface plasmons (Springer, Berlin and New 43 R. A. Molina, D. Weinmann, and R. A. Jalabert, Phys. York, 1988). Rev. B 65, 155427 (2002). 11 W. L. Barnes, A. Dereux, and T. W. Ebbesen, Nature 424, 44 G. Weick, R. A. Molina, D. Weinmann, and R. A. Jalabert, 824 (2003). Phys. Rev. B 72, 115410 (2005). 12 B. Hecht, H. Bielefeldt, L. Novotny, Y. Inouye, and D. W. 45 Y. Gao, Z. Yuan, and S. Gao, J. Chem. Phys. 134, 134702 Pohl, Phys. Rev. Lett. 77, 1889 (1996). 13 K. Kneipp, Y. Wang, H. Kneipp, L. T. Perelman, I. Itzkan, R. R. Dasari, and M. S. Feld, Phys. Rev. Lett. 78, 1667 (1997). 14 S. M. Nie and S. R. Emery, Science 275, 1102 (1997). 15 H. X. Xu, E. J. Bjerneld, M. Kll, and L. Borjesson, Phys. Rev. Lett. 83, 4357 (1999). 16 J. B. Pendry, L. Martin-Moreno, and F. J. Garcia-Vidal, Science 305, 847 (2004). 17 E. Ozbay, Science 311, 189 (2006). 18 M. Sandtke and L. Kuipers, Nature Photon. 1, 573 (2007). 19 S. A. Maier, Plasmonics: Fundamentals and Applications (Springer, New York, 2007). 20 S. Zhang, D. A. Genov, Y. Wang, M. Liu, and X. Zhang, Phys. Rev. Lett. 101, 047401 (2008). 21 J. Homola, S. S. Yee, and G. Gauglitz, Sensors Actuat. B 54, 3 (1999). 22 H. A. Atwater and A. Polman, Nature Mater. 9, 205 (2010). 23 D. E. Chang, A. S. Sørensen, P. R. Hemmer, and M. D. Lukin, Phys. Rev. Lett. 97, 053002 (2006). 24 A. V. Akimov, A. Mukherjee, C. L. Yu, D. E. Chang, A. S. Zibrov, P. R. Hemmer, H. Park, and M. D. Lukin, Nature 450, 402 (2007). 25 R. Kolesov, B. Grotz, G. Balasubramanian, R. J. Stohr, A. A. L. Nicolet, P. R. Hemmer, F. Jelezko, and J. Wrachtrup, Nature Phys. 5, 470 (2009). 26 A. V. Zayats, J. Elliott, I. I. Smolyaninov, and C. C. Davis, Appl. Phys. Lett. 86, 151114 (2005). 27 I. I. Smolyaninov, J. Elliott, A. V. Zayats, and C. C. Davis, Phys. Rev. Lett. 94, 057401 (2005). 28 J. Takahara, S. Yamagishi, H. Taki, A. Morimoto, and T. Kobayashi, Opt. Lett. 22, 475 (1997). 29 Q. Cao and J. Jahns, Opt. Express 13, 511 (2005). 30 A. Archambault, F. Marquier, J. J Greffet, and C. Arnold, Phys. Rev. B 82, 35411 (2010). 31 D. J. Bergman and M. I. Stockman, Phys. Rev. Lett. 90, 027402 (2003). 32 A. Boltasseva and H. A. Atwater, Science 331, 290 (2011). 33 J. A. Schuller, E. S. Barnard, W. Cai, Y. C. Jun, J. S. White, and M. L. Brongersma, Nature Mater. 9, 193 (2010). 34 A. V. Krasavin and N. I. Zheludev, Appl. Phys. Lett. 84, 1416 (2004). (2011). 46 V. M. Silkin, E. V. Chulkov, and P. M. Echenique, Phys. Rev. Lett. 93, 176801 (2004). 47 V. M. Silkin and E. V. Chulkov, Vacuum 81, 186 (2006). 48 A. J. Hoffman, L. Alekseyev, S. S. Howard, K. J. Franz, D. Wasserman, V. A. Podolskiy, E. E. Narimanov, D. L. Sivco, and C. Gmachl, Nature Mater. 6, 946 (2007). 49 T. Taubner, D. Korobkin, Y. Urzhumov, G. Shvets, and R. Hillenbrand, Science 313, 1595 (2006). 50 J. M. Luther, P. K. Jain, T. Ewers, and A. P. Alivisatos, Nature Mater. 10, 361 (2011). 51 D. V. Seletskiy, M. P. Hasselbeck, J. G. Cederberg, A. Katzenmeyer, M. E. Toimil-Molares, F. L´eonard, A. A. Talin, and M. Sheik-Bahae, Phys. Rev. B 84, 115421 (2011). 52 M. S. Vitiello, D. Coquillat, L. Viti, D. Ercolani, F. Teppe, A. Pitanti, F. Beltram, L. Sorba, W. Knap, and A. Tredicucci, Nano Lett. 12, 96 (2011). 53 J. G´omez Rivas, J. A. S´anchez-Gil, M. Kuttge, P. Haring Bolivar, and H. Kurz, Phys. Rev. B 74, 245324 (2006). 54 J. G´omez Rivas, M. Kuttge, H. Kurz, P. H. Bolivar, and J. A. S´anchez-Gil, Appl. Phys. Lett. 88, 082106 (2006). 55 T. H. Isaac, W. L. Barnes, and E. Hendry, Appl. Phys. Lett. 93, 241115 (2008). 56 G. V. Naik and A. Boltasseva, phys. stat. sol. (RRL) 4, 295 (2010). 57 M. Galli, D. Gerace, A. Politi, M. Liscidini, M. Patrini, L. C. Andreani, A. Canino, M. Miritello, R. L. Savio, A. Irrera, and F. Priolo, Appl. Phys. Lett. 89, 241114 (2006). 58 L. Tang, S. E. Kocabas, S. Latif, A. K. Okyay, D.-S. Ly- Gagnon, K. C. Saraswat, and D. A. B. Miller, Nature Pho- ton. 2, 226 (2008). 59 R. J. Walters, R. V. A van Loon, I. Brunets, J. Schmitz, and A. Polman, Nature Mater. 9, 21 (2010). 60 M. P. Hasselbeck, D. Seletskiy, L. R. Dawson, and M. Sheik-Bahae, phys. stat. sol. (c) 5, 253 (2008). 61 For semiconductors, the scattering between the SPP and the LO phonons can also lead to the SPP damping [G. C. Cho, T. Dekorsy, H. J. Bakker, R. Hvel, and H. Kurz, Phys. Rev. Lett. 77, 4062 (1996)], which plays the dominant role for the SPP with small wavevectors (Ref. 51). 62 D. H. Ehlers and D. L. Mills, Phys. Rev. B 34, 3939 (1986); ibid. 36, 1051 (1987). 63 G. R. Bell, C. F. McConville, and T. S. Jones, Phys. Rev. 35 T. Nikolajsen, K. Leosson, and S. I. Bozhevolnyi, Appl. B 54, 2654 (1996). Phys. Lett. 85, 5833 (2004). 64 G. R. Bell, T. D. Veal, J. A. Frost, and C. F. McConville, 36 K. F. MacDonald, Z. L. S´amson, M. I. Stockman and N. Phys. Rev. B 73, 153302 (2006). I. Zheludev, Nature Photon. 3, 55 (2009). 37 P.J. Feibelman, Prog. Surf. Sci. 12, 287 (1982). 38 W. Ekardt, Phys. Rev. B 31, 6360 (1985). 65 F. J. Garc´ıa de Abajo, Rev. Mod. Phys. 82, 209 (2010). 66 H. Haug and A.-P. Jauho, Quantum Kinetics in Transport and Optics of Semiconductors (Springer-Verlag, Berlin, 21 1996). 67 M. W. Wu, J. H. Jiang, and M. Q. Weng, Phys. Rep. 493, 61 (2010). 68 A. Konar and D. Jena, J. Appl. Phys. 102, 123705 (2007). 69 N. G. Hormann, I. Zardo, S. Hertenberger, S. Funk, S. Bolte, M. Doblinger, G. Koblmuller, and G. Abstreiter, Phys. Rev. B 84, 155301 (2011); M. Moller, M. M. de Lima jr., A. Cantarero, L. C. O. Dacal, and J. R. Madureira, Phys. Rev. B 84, 085318 (2011). 70 R. Loudon, The Quantum Theory of Light, 3rd ed. (Oxford University Press, Oxford, 2000). 71 H. Ditlbacher, A. Hohenau, D. Wagner, U. Kreibig, M. Rogers, F. Hofer, F. R. Aussenegg, and J. R. Krenn, Phys. Rev. Lett. 95, 257403 (2005). 72 H. T. Miyazaki and Y. Kurokawa, Phys. Rev. Lett. 96, 097401 (2006). 73 L. Cao, R. A. Nome, J. M. Montgomery, S. K. Gray, and N. F. Scherer, Nano Lett. 10, 3389 (2010). purity configuration and the impurity density. In addition to the surface impurities, we have also performed the cal- culation for bulk-like impurities (i.e., impurities evenly dis- tributed inside the nanowire. The corresponding line den- sity of the bulk impurities is set equal to the line density of the surface impurities ni). For both impurity configura- tions, we have checked that the influence of the ei scatter- ing remains very small for ni < 5ne. 80 A. Bergara, J. M. Pitarke, and R. H. Ritchie, Phys. Rev. B 60, 16176 (1999). 81 R. H. Ritchie and R. E. Wilems, Phys. Rev. 178, 372 (1969). 82 G. Barton, Rep. Prog. Phys. 42, 963 (1979). 83 I. Vill´o-P´erez and N. R. Arista, Surf. Sci. 603, 1 (2009). 84 E. Prodan, C. Radloff, N. Halas, and P. Nordlander, Sci- ence 302, 419 (2003); E. Prodan and P. Nordlander, J. Chem. Phys. 120, 5444 (2004). 85 A. F. Slachmuylders, B. Partoens, W. Magnus, and F. M. 74 S. A. Dayeh, D. P. R. Aplin, X. Zhou, P. K. L. Yu, E. T. Peeters, Phys. Rev. B 74, 235321 (2006). Yu, and D. Wang, Small 3, 326 (2007). 75 For sufficiently long SPP wave packet (L & 100R), the 86 C. Ell and H. Haug, phys. stat. sol. (b) 159, 117 (1990). 87 D. B. Tran Thoai and H. T. Cao, Solid State Commun. results are not sensitive to the wave packet length L. 111, 67 (1999). 76 K.P. Charl´e, W. Schulze, and B. Winter, Z. Phys. D: At., 88 S. N. Klimin, E. P. Pokatilov, and V. M. Fomin, phys. stat. Mol. Clusters 12, 471 (1989). sol. (b) 184, 373 (1994). 77 C. Bre´chignac, Ph. Cahuzac, J. Leygnier, and A. Sarfati, 89 C. R. Bennett, N. C. Constantinou, M. Babiker, and B. K. Phys. Rev. Lett. 70, 2036 (1993). Ridley, J. Phys.:Condens. Matter 7, 9819 (1995). 78 S. Mochizuki, M. Sasaki, and R. Ruppin, J. Phys.: Con- 90 A. L. Vartanian, phys. stat. sol. (b) 242, 1482 (2005). dens. Matter 9, 5801 (1997). 79 The influence of the ei scattering depends on both the im- 

1309.5559

1

1309

Exact wave functions for the edge state of a disk-shaped two dimensional topological insulator

We report the exact wave functions for the eigen state of a disk-shaped two dimensional topological insulator. The property of the edge state whose energy lies inside the bulk gap is studied. It is found that the edge state energy is affected by the radius of the disk. For a fixed angular momentum index, there is a critical disk radius below which there exists no edge state. The value of this critical radius increases as the angular momentum index increases. In the limit of large disk radius, the energy of the edge state approaches a limiting value determined by the system parameters and independent of the angular momentum index. The derivation from this limiting value is inversely proportional to the radius with a coefficient proportional to the angular momentum index. In the general case, the energy differences between two edge states with adjacent angular momentum indexes are not equal. The exact and analytical wave functions also facilitates the investigation of electronic state in other structures of the two dimensional topological insulator.

cond-mat.mes-hall

cond-mat

Exact wave functions for the edge state of a disk-shaped two dimensional topological insulator M. Pang and X. G. Wu SKLSM, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China We report the exact wave functions for the eigen state of a disk-shaped two dimensional topological insulator. The property of the edge state whose energy lies inside the bulk gap is studied. It is found that the edge state energy is affected by the radius of the disk. For a fixed angular momentum index, there is a critical disk radius below which there exists no edge state. The value of this critical radius increases as the angular momentum index increases. In the limit of large disk radius, the energy of the edge state approaches a limiting value determined by the system parameters and independent of the angular momentum index. The derivation from this limiting value is inversely proportional to the radius with a coefficient proportional to the angular momentum index. In the general case, the energy differences between two edge states with adjacent angular momentum indexes are not equal. The exact and analytical wave functions also facilitates the investigation of electronic state in other structures of the two dimensional topological insulator. PACS numbers: 72.25.Dc, 73.23.Ra Topological insulator has attracted considerable atten- tions in recent years [1 -- 3]. In two dimensions, the topo- logical insulator is described by an effective model [4]. Many theoretical investigations into the exotic proper- ties of two dimensional topological insulator (2DTI) are based on this model [5 -- 9]. Despite those great progresses achieved, exact wave function for the 2DTI is rarely reported so far. When a 2DTI is cut into a ribbon like structure, exact wave func- tions were obtained [5]. It is found that due to the finite width of ribbon, there is a gap in the energy spectrum which decreases exponentially as the width of the ribbon increases [5]. In the present paper, we report exact wave functions for the 2DTI with a circular geometry. This allows us to probe some exact properties of 2DTI. In particular, we focus on the edge state of a disk shaped 2DTI with the open boundary condition that both components of wave functions vanish at the disk edge. The electronic state in other geometric structure will also be briefly discussed. The 2DTI is described by the following well-known Hamiltonian [4] (cid:18) h(k) 0 0 h∗(−k) (cid:19) , (1) x + k2 x + k2 where k = (kx, ky). h(k) = ǫ(k) + Pα dα(k)σα with σα the Pauli matrix. ǫ(k) = C − D(k2 y), d1 = Akx, d2 = Aky, and d3 = M − B(k2 y). A, B, C, D, and M are parameters determined by the structure of the quan- tum well [4]. Operators kx and ky represent differential operators −i∂x and −i∂y respectively. The parameter C gives the zero point of energy and we can safely set it to zero for simplicity in this paper. The spin-up block h(k) and spin-down block h∗(−k) in the Hamiltonian are decoupled and they can be solved separately [4]. Since the the system under consideration has a circular geom- etry, we will adopt the polar coordinate system. It can be shown that the radial and angular part of the wave function can be separated. This simplifies the problem and one only needs to solve the radial wave function. For the spin-up block h(k), the wave function can be written as Ψm(r, θ) = (cid:18) eimθφ1(r) ei(m+1)θφ2(r) (cid:19) , (2) where m = 0,±1,±2,±3, ... is the angular momentum index. The differential equations for the radial part can be written as (cid:18) Om + X1 X3Pm+1 X4P−m Om+1 + X2 (cid:19)(cid:18) φ1(r) φ2(r) (cid:19) = 0, (3) with Om = d2/d2r + d/rdr − m2/r2, Pm = d/dr + m/r, X1 = (−E + V (r) + M )/(B + D), X2 = (E − V (r) + M )/(B−D), X3 = −iA/(B +D), and X4 = iA/(B−D). E is the eigen energy of the system and V (r) the exter- nally applied potential that depends only on the radial coordinate. In this paper, we consider V (r) taking different but constant values in the different regions of r. In this case, the exact wave function can be written as Φm(λr) = (cid:18) φ1(r) φ2(r) (cid:19) = (cid:18) C1Zm(λr) C2Zm+1(λr) (cid:19) , (4) with Zm(λr) the Bessel functions Jm(λr) or Ym(λr). λ, C1, and C2 are solution of a secular equation (cid:18) X1 − λ2 X3λ −X4λ X2 − λ2 (cid:19)(cid:18) C1(λ) C2(λ) (cid:19) = 0. The equation gives four roots of λ: ±λ1 and ±λ2, as λ1 = [(F + √F 2 − 4X1X2)/2]1/2, λ2 = [(F − √F 2 − 4X1X2)/2]1/2, (5) (6) with F = X1 + X2 − X3X4. Note that λ is a function of potential V and energy E. C1 and C2 are only deter- mined up to factor. The value of λ obtained in Eq.(6) can become imag- In that case, it is more convenient to write the inary. exact wave function as Φm(ξr) = (cid:18) φ1(r) φ2(r) (cid:19) = (cid:18) C3Zm(ξr) C4Zm+1(ξr) (cid:19) , (7) with Zm(ξr) (−1)mKm(ξr). secular equation or the Bessel ξ, C3, and C4 are solution of a functions Im(ξr) (cid:18) X1 + ξ2 X3ξ X4ξ X2 + ξ2 (cid:19)(cid:18) C3(ξ) C4(ξ) (cid:19) = 0. The equation gives four roots of ξ: ±ξ1 and ±ξ2, as ξ1 = [(−F + √F 2 − 4X1X2)/2]1/2, ξ2 = [(−F − √F 2 − 4X1X2)/2]1/2, (8) (9) with F = X1 + X2 − X3X4 the same as given before. The E dependence of ξ in Eq.(9) depends also on other model parameters. With the parameters given in [4], one can readily verify that when E − V < M, both ξ1 and ξ2 are real and non-zero. When E − V = M, ξ1 is non-zero and real, and ξ2 = 0. When E − V > M, both ξ1 and ξ2 are non-zero, ξ1 remains real, and ξ2 is purely imaginary. m(λr), or ΦI m(ξr) and ΦK It can be shown that Φm(λr) and Φm(−λr) (or Φm(ξr) and Φm(−ξr)) are not linearly independent solutions. Therefore we have linearly independent solutions ΦJ m(λr) and ΦY m(ξr). The superscripts J, Y , I, and K denotes the kind of the Bessel functions involved. The desired wave function can be constructed as a linear combination of the linearly independent solu- tions. In the case of λ = 0 or ξ = 0, a careful treatment of λ → 0 or ξ → 0 limit is required in order to obtain linearly independent solutions. For the spin-down block h∗(−k), the exact wave func- tion can be obtained in the same way. A careful exami- nation shows that the energy of spin-up state with angu- lar momentum index −m, denoted as E−m,↑, exactly equals to Em−1,↓ the energy of spin-down state with angular momentum index m − 1. One has E−m,↑ = Em−1,↓ and E−m,↓ = Em−1,↑. Next, we use the exact wave functions obtained above to construct the edge state for a disk shaped 2DTI. In this quantum disk system, one has V (r) = 0 for r < R with R the radius of the disk. The boundary condition is that both components of the wave functions must van- ish at r = R. When the energy falls inside the bulk gap E2 < M 2, Eq.(9) gives to two real ξ1 and ξ2 with model parameters given in Ref.[4]. The wave functions that involve the Bessel function Km(ξr) are divergent at r = 0 and must be discarded. Therefore, we construct the desired wave function from ΦI m(ξ2r) by requiring d1ΦI m(ξ2R) = 0, with d1 and d2 the coefficients to be determined. This leads to the fol- lowing equation for the spin-up case m(ξ1R) + d2ΦI m(ξ1r) and ΦI ξ1(ξ2 ξ2(ξ2 2 + X1) 1 + X1) = Im(ξ2R)Im+1(ξ1R) Im(ξ1R)Im+1(ξ2R) , (10) 2 from which the eigen state energy can be obtained. The nature of function Im(ξr) guarantees that the amplitude of the edge state wave function will be large near the disk edge and small in the disk center. In the large disk radius limit, i.e. R → ∞, Im(ξr) ap- proaches to eξr/√2πξr, a value independent on m. From Eq.(10) we obtain eigen state energy E0 = −DM/B for both spin-up and spin-down states. For large radius ξR >> 1, the eigen state energy is given by Em,s = E0 + δEm,s (s denotes the spin index) with δEm,s ≃ C′ s(m + 1 2 )R−1, (11) where C′ s is a factor only relies on the model parameters and spin. Eq.(11) suggests that for a large radius of the disk, the energy of the edge states approach linearly in R−1 to the limiting value E0, and for a fixed R one get equal energy spacing for eigen states with adjacent m index. In general, the eigen state energy denoted as Em,s(R) is a function of m and R. In Fig.1, Em,s(R) is shown as a function of the angular momentum index m for three values of R. The energy of spin-up states is depicted by 6 4 2 0 ) V e m ( E -2 -4 -6 0.5µm ↑ 1µm ↑ 2µm ↑ 0.5µm ↓ 1µm ↓ 2µm ↓ -30 -20 -10 0 m 10 20 30 FIG. 1: (Color online) The energy spectrum versus the angu- lar momentum index m for three values of the disk radius R: 0.5 µm (square dots), 1 µm (round dots), and 2 µm (triangu- lar dots). The energy of spin-up (spin-down) state is depicted by solid (open) symbols, respectively. Parameters A, B, D, M are taken from Ref.[4]: A = −342 meV nm, B = −169 meV nm2, D = 5.14 meV nm2, M = −6.86 meV. solid symbols and the energy of spin-down states is shown by open symbols. The energy exhibits an approximately linear dependence on m and the slope of the curves be- comes smaller as the disk radius R increases. It is found that for each m, there is a critical value Rc m 6= 0 such that for R < Rc m no edge states can exist inside the bulk gap as one can not find any energy E < M that Eq.(10) m becomes larger as m holds. increases. It is also found that Rc In Fig.2, the eigen state energy is shown as a function of R−1 for several values of m. It is clear that the R−1 5 6 6 4 2 0 ) V e m ( E -2 -4 -6 4 3 2 1 1 2 m=0 m=0 spin up spin down ) V e m ( E 0.8 0.4 0.0 -0.4 -0.8 0.0 0.5 R-1 (µm-1) 1.0 6 5 3 4 0.000 0.004 0.008 0.012 R-1 (nm-1) 0.016 0.020 FIG. 2: (Color online) The R−1 dependence of the edge state energy for several values of m. The black solid curves are for the spin-up states and the red dashed curves are for the spin-down states. The inset shows an enlarged portion of the figure near R−1 ≈ 0. dependence is not linear. For a fixed m, this non-linear dependence is different for spin-up and spin-down states. Eq.(11) shows that for large disk radius the energy spacing ∆Em,s = Em+1,s − Em,s depends linearly on R−1 and is independent of m. In Fig.3, ∆Em,s − ∆E0,s is shown as a function of R−1 for several values of m. The upper panel (a) is for the spin-up states and the lower panel (b) is for the spin-down states. The energy spacing becomes smaller as m increases. This clearly demonstrates that the m dependence of eigen state en- ergy shown in Fig.1 is non-linear intrinsically. ∆E 4 -∆E 0 ∆E 3 -∆E 0 (a) spin up spin down ∆E 2 -∆E 0 ∆E 1 -∆E 0 ∆E 1 -∆E 0 (b) ∆E 2 -∆E 0 ∆E 3 -∆E 0 ∆E 4 -∆E 0 0.00 -0.02 -0.04 -0.06 -0.08 ) V e m ( 0 E ∆ - n E ∆ -0.10 0.10 ) V e m ( 0 E ∆ - n E ∆ 0.08 0.06 0.04 0.02 0.00 0.000 0.002 0.004 R-1 (nm-1) 0.006 FIG. 3: (Color online) The R−1 dependence of the energy spacing ∆Em,s − ∆E0,s. The upper panel (a) is for the spin- up state and the lower panel (b) is for the spin-down state. In the remaining part of this paper, base upon our ex- act wave functions, we discuss some interesting aspects of 2DTI systems with circular geometry. In the case of the disk shaped 2DTI, with the open boundary condi- 3 m(λ1r) and ΦI tion, one may also seek eigen state with energy outside the bulk gap. In this case, the wave function is a linear combination of ΦJ m(ξ1r). Since Jm(λr) is an oscillatory function, but Im(ξr) grows exponentially, the boundary condition results in a larger weight for the ΦJ m(λ1r) term. Thus, the amplitude of wave function can not be mainly concentrated near the edge and decay exponentially toward the center of disk. m(ξ1r), ΦJ m(ξ1r) and ΦK Let us consider the anti-disk system: a hole in the 2DTI plane. The potential is given by V (r) = 0 when r > R. The open boundary condition is adopted. For energy inside the bulk gap, The wave function is a linear combination of ΦK m(ξ2r). The wave function will mainly concentrated near the edge, thus one has an edge state. When the energy is outside the bulk gap, the physically allowed wave function is a linear combina- tions of ΦK m(λ1r). The boundary condition can only provide two equations, but there are three coefficients to be determined. This means that one can find an eigen state for any energy outside the bulk gap, completely different from the disk case. It is also found that for each m, there is a critical value Rc m 6= 0 (different from the disk case) such that for R < Rc m no edge states can exist inside the bulk gap. It is found that Rc m becomes larger as m increases. This indicates that, an infinite 2DTI may have no edge state though it has a finite length edge. m(λ1r) and ΦY Next, we consider a ring-like geometry of the 2DTI, i.e. V (r) = 0 for R1 < r < R2, with the open boundary condition adopted for both edges. When the energy falls inside the bulk gap, wave functions are a linear combina- tion of ΦK m(ξ2r). The boundary conditions give four equations for the four co- efficients to be determined. One should have a discrete energy spectrum. This system is similar to the ribbon structure [5] but the two edges are different. m(ξ1r), and ΦI m(ξ1r), ΦK m(ξ2r), ΦI Let us finally consider the case that the potential V (r) is a step function, i.e., V (r) = V0 6= 0 for r < R, and V (r) = 0 otherwise. This system can be implemented experimentally by depositing metal gates on top of the 2DTI. As the potential V takes different values for r < R and r > R, the values of λ in Eq.(6) or ξ in Eq.(9) will be different in the different r regions. In the following, we introduce a superscript < for λ and ξ in the region r < R, and a superscript > in the region r > R. m(ξ< m(λ< m(ξ< 1 and λ< 1 r) and ΦI 2 imaginary); (2) ΦI 1 real, λ< 1 r) when E − V0 > M (λ< 2 r) when E < M (both λ> m(λ> m(ξ> 1 real, λ> In the region r < R, the allowed contributions to the wave functions are: (1) ΦI 2 r), when m(ξ< E−V0 < M (both λ< 1 r) and ΦJ 2 imag- inary). For r > R, the allowed contributions are: (1) 1 and λ> m(ξ> ΦK 2 1 r), ΦJ imaginary); (2) ΦK 1 r) when E > M (λ> 2 imaginary). The boundary con- dition is that the wave function and its first order deriva- tive should be both continuous at r = R. The boundary condition now leads to four equations. 1 r) and ΦY 1 r) and ΦK m(λ> m(ξ> When E > M or E < −M, the system should have a continuous energy spectrum, for any V0 6= 0. This is because that the number of coefficients needed to be de- termined is larger than the number of equations due to the boundary condition. When E < M, the quantiza- tion of energy level is expected. In summary, the exact wave functions for the eigen state of a disk-shaped two dimensional topological insu- lator is reported. The property of edge state is studied. For a fixed angular momentum index m, there is a criti- cal disk radius below which the edge state is not possible. The value of this critical radius increases as m increases. The R−1 dependence of the edge state energy is non- 4 linear. The m dependence of the edge state energy is non-linear as well. The exact wave functions also make it easy for us to investigate the electronic state in other structures of the two dimensional topological insulator. Acknowledgments This work was partly supported by NSF and MOST of China. [1] S. C. Zhang, Physics 1, 6 (2008); X. L. Qi and S. C. Zhang, [6] C. X. Liu, T. L. Hughes, X. L. Qi, K. Wang, and S. C. Physics Today 63, 33 (2010). Zhang, Phys. Rev. Lett. 100, 236601 (2008). [2] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 [7] M. J. Schmidt, E. G. Novik, M. Kindermann, and B. (2010). [3] J. E. Moore, Nature 464, 194 (2010). [4] B. A. Bernevig, T. L. Hughes, and S. C. Zhang, Science Trauzettel, Phys. Rev. B 79, 241306 (2009). [8] J. Li, R. L Chu, J. K. Jain, and S. Q. Shen, Phys. Rev. Lett. 102, 136806 (2009). 314, 1757 (2006). [9] K. Chang and W. K. Lou, Phys. Rev. Lett. 106, 206802 [5] B. Zhou, H. Z. Lu, R. L. Chu, S. Q. Shen, and Q. Niu, (2011). Phys. Rev. Lett. 101, 246807 (2008). 

1001.5330

1

1001

Orbital magnetization of the electron gas on a two-dimensional kagome lattice under a perpendicular magnetic field

The orbital magnetization of the electron gas on a two-dimensional kagome lattice under a perpendicular magnetic field is theoretically investigated. The interplay between the lattice geometry and magnetic field induce nontrivial $k$-space Chern invariant in the magnetic Brillouin zone, which turns to result in profound effects on the magnetization properties. We show that the Berry-phase term in the magnetization gives a paramagnetic contribution, while the conventional term brought about by the magnetic response of the magnetic Bloch bands produces a diamagnetic contribution. As a result, the superposition of these two components gives rise to a delicate oscillatory structure in the magnetization curve when varying the electron filling factor. The relationship between this oscillatory behavior and the Hofstadter energy spectrum is revealed by selectively discussing the magnetization and its two components at the commensurate fluxes of $f$=1/4, 1/3, and 1/6, respectively. In particular, we reveal as a typical example the fractal structure in the magnetic oscillations by tuning the commensurate flux around $f$=1/4. The finite-temperature effect on the magnetization is also discussed.

cond-mat.mes-hall

cond-mat

Orbital magnetization of the electron gas on a two-dimensional kagom´e lattice under a perpendicular magnetic field Zhigang Wang,1 Zi-Gang Yuan,2, 3 Zhen-Guo Fu,1, 2 Shu-Shen Li,2 and Ping Zhang1, 4, ∗ 1LCP, Institute of Applied Physics and Computational Mathematics, P.O. Box 8009, Beijing 100088, People's Republic of China 2State Key Laboratory for Superlattices and Microstructures, Institute of Semiconductors, Chinese Academy of Sciences, P. O. Box 912, Beijing 100083, People's Republic of China 3College of Science, Beijing University of Chemical Technology, Beijing 100029, People's Republic of China 4Center for Applied Physics and Technology, Peking University, Beijing 100871, People's Republic of China Abstract The orbital magnetization of the electron gas on a two-dimensional kagom´e lattice under a per- pendicular magnetic field is theoretically investigated. The interplay between the lattice geometry and magnetic field induce nontrivial k-space Chern invariant in the magnetic Brillouin zone, which turns to result in profound effects on the magnetization properties. We show that the Berry-phase term in the magnetization gives a paramagnetic contribution, while the conventional term brought about by the magnetic response of the magnetic Bloch bands produces a diamagnetic contribution. As a result, the superposition of these two components gives rise to a delicate oscillatory struc- ture in the magnetization curve when varying the electron filling factor. The relationship between this oscillatory behavior and the Hofstadter energy spectrum is revealed by selectively discussing the magnetization and its two components at the commensurate fluxes of f =1/4, 1/3, and 1/6, respectively. In particular, we reveal as a typical example the fractal structure in the magnetic oscillations by tuning the commensurate flux around f =1/4. The finite-temperature effect on the magnetization is also discussed. PACS numbers: 73.20.At, 71.10.Ca, 72.15.Gd ∗Author to whom correspondence should be addressed. Electronic address: zhang ping@iapcm.ac.cn 1 I. INTRODUCTION Over the past three decades, the orbital dynamics of two-dimensional (2D) electrons coupled to a uniform perpendicular magnetic field has been a subject of special interest in condensed matter physics. Great attention has been paid to investigate the well-known Hofstadter butterfly-like diagram, a remarkable intricate, detailed, and self-similar feature for the subband spectrum of single-particle eigenenergies, in different 2D structures such as square [1], triangular [2], honeycomb [3], and kagom´e lattices [4], lateral superlattices [5], superstructures with flat bands [6], quasiperiodic tilings [7], fractal networks [8], and rhombus tilings [9]. This magnetic-field-induced frustration in 2D structures, caused by the competition between the periodic potential imposed by the lattice structure and the length scale provided by the magnetic field, is the headstream of rich and novel physical phenomena in these 2D systems. Among them the orbital magnetization is a very interesting one and needs to be paid special attention due to its intrinsic relationship with the unique topological structure brought about by the interplay between the special lattice geometry and magnetic subbands. Unfortunately, to date, there are very scarce studies [10, 11] in literature that devote to enlightening this revealing relationship. Motivated by this observation, in this paper, we study the orbital magnetization proper- ties of the electron gas on a two-dimensional kagom´e lattice under a perpendicular magnetic field. Since our attention is solely on the orbital character in magnetization brought about by the interplay between the kagom´e lattice structure and the magnetic field, thus unlike most of previous work, the kagom´e lattice considered in this paper is spin nonmagnetic in itself, namely, no spin structure is exposed in the absence of the external magnetic field. The nonmagnetic kagom´e lattice structure has been either fabricated by modern patterning techniques [12, 13] or observed in reconstructed semiconductor surfaces [14]. In the former case, remarkably, the electron filling factor (namely, the Fermi energy) can be readily con- trolled by applying a gate voltage [15]. Our lattice model is free from the constraint imposed on the k·p approximation used in the extensively studied GaAs two-dimensional electron gas (2DEG), in which the k·p Hamiltonian is only valid around the Γ point in the Brillouin zone (BZ). In contrast, our lattice model allows for any electron filling, which result in var- ious Fermi-surface topologies in the magnetic BZ, which in turn, as will be shown below, produces profound effects on the orbital magnetization properties and the related transport 2 phenomenon. The orbital magnetization for Bloch electrons combines two terms [10, 16], one is the conventional orbital magnetic moment to characterize the magnetic response of the single- particle energy spectrum, and the other is a Berry-phase correction. These two terms play different roles in metallic and insulating regions [17, 18]. By varying the Fermi energy, we obtain the follows: (i) Both the magnitudes of these two terms increase (decrease) with increasing (decreasing) the Fermi energy in metallic regions; (ii) In insulating regions the conventional term keeps a constant unchanged value (quantum step), while the Berry-phase linearly varies with the Fermi energy with a slope proportional to the system's Hall conduc- tance; (iii) A general fractal structure may occur in the orbital magnetization curve (as a function of the Fermi energy) by tuning the commensurate flux. The rest of this paper is organized as follows. In Sec. II we give the Hofstadter spectrum of the kagom´e lattice by diagonalizing the corresponding tight-binding Hamiltonian in an external magnetic field. In Sec. III the orbital magnetization and its two components are systematically discussed by varying various system parameters. A summary is given in Sec. IV. II. KAGOM´E LATTICE IN A MAGNETIC FIELD The kagom´e lattice is a two-dimensional periodic array of corner-sharing triangles with three sites per unit cell, as illustrated in Fig. 1. Here a is the triangle edge length (we set it as the length unit in the following), and A, B, and C denote the three sites in a unit cell. Let us begin with the tight-binding Hamiltonian for the spinless electrons on the kagom´e lattice with zero magnetic field (B=0), which is given by c†i cj, H = tXhi,ji (1) where t is hopping amplitude between the nearest-neighbor link hi, ji and c†i (ci) is the cre- ation (annihilation) operator of an electron on lattice i. Hamiltonian (1) can be diagonalized in the momentum space as H = tXk Ψ†kH(k)Ψk, (2) 3 where Ψk=(ckA, ckB, ckC)T is the three-component electron field operator. Each component of Ψk is the Fourier transform of ci, i.e., cks = Xmn cmnseik·rmns, (3) where we have changed notation i → (mns) by using (mn) to label the kagom´e unit cells and s to label the three sites in a unit cell. H(k) is given by 0 p1 k p3 k k 0 p2 p1 k k p2 p3 k 0     H(k) = , (4) where pi k=2 cos (k · ai) and a1=(−1/2,−√3/2), a2=(1, 0), and a3=(−1/2,√3/2) represent the displacements in a unit cell from the A to B site, from the B to C site, and from the C to A site, respectively. K=±(2/3)a1, ±(2/3)a2, and ±(2/3)a3. In this notation, the first BZ is a hexagon with the corners of In the presence of a magnetic filed (B 6= 0), the hopping term of the tight-binding Hamiltonian (1) is modified by phase factors from the vector potential A, where γij is the phase factor (the well known Peierls phase) between sites j and i: t → teiγij γij = with Φ0=h/e being the flux quantum. 2π Φ0 Z i j A · dl (5) (6) One can take different gauge potential A mathematically to satisfy the physical confine- ment ∇ × A=B. However, the flux through one unit cell is only linearly dependent on the external magnetic field B, and has no relation with the choice of the gauge potential A. So, it is convenient to measure the magnetic field in units of the flux quantum per elementary triangular plaquette of the kagom´e lattice. After a simple algebraic calculation, one obtains the flux through one triangle is Φ=√3B/4. Thus we can define a parameter, called the filling ratio f , as the fraction of a flux quantum throng each triangle, i.e, f =Φ/Φ0=√3eB/4h [4]. The total flux through a unit cell of the kagom´e lattice is then given as 8f Φ0. When f is a rational number that can be written as f =p/q, where p and q are integers with no common factors, the total flux through q triangular plaquette of the kagom´e lattice 4 A B C FIG. 1: (Color online). Schematic picture of the 2D kagom´e lattice with bond length a. The dashed line represents the Wigner-Seitz unit cell, which contains three independent sites (A, B, and C). FIG. 2: (Color online). Hofstadter butterfly-like spectrum of the 2D Kagom´e lattice. The numbers recorded in gaps represent the Hall conductance of this system in units of e2/h when the Fermi energy lies in these gaps. is an integer multiple of Φ0. In this case, the Hamiltonian (1) should be diagonalized in a "magnetic" unit cell, which contains q plaquette in order to guarantee k being good quantum numbers. As a result, the magnetic Brillouin zone (MBZ) is q times smaller than the usual BZ. Furthermore, because of the magnetic translation symmetry, the MBZ has exactly a q- fold degeneracy. For convenience we take the gauge potential as A=B(− 1 2(x− 1√3 y), 0). Under this gauge, the Peierls phases, which reflect the information about the external magnetic field, are only related to n and have nothing with m. The Hamiltonian of the kagom´e lattice now can be diagonalized in the momentum space as the following 3q × 3q 2(−√3x+y), 1 5 matrix, where and H(k) = q−1 Xn=0hhBA n c†nBcnA + hBA∗n c†nBcnA + hCB +hAC n c†nCcnB + exp(−i4πf )hCB∗n+1 c†nCcn+1B n c†nAcnC + hAC∗n c†n+1AcnC + H.c.i , (7) hBA n = t exp(−ik · a1) exp(cid:20)−i8πf (cid:18)n − n = t exp(−ik · a2) exp(cid:20)i8πf (cid:18)n + 6(cid:19)(cid:21) , 12(cid:19)(cid:21) , 1 1 hCB hAC n = t exp(−ik · a3). With the above Hamiltonian (7), it is straightforward to obtain the Hofstadter butterfly-like energy spectrum of the kagom´e lattice, which is shown in Fig. 2. From Fig. 2, one can easily observe three symmetries of this Hofstadter butterfly-like spectrum of the kagom´e lattice [4]: (i) The spectrum at f is the same as that at f + j with j any integer; (ii) The spectrum is also unchanged on changing f to −f , because if there is an eigenstate with energy ǫ for field f , then the corresponding complex conjugate of this eigenstate is also an eigenstate with the same energy ǫ for field −f ; (iii) The spectrum is inverted when the field f changes to f + 1/2. Thus the highest-energy states for field near f =1/2 are equivalent to the lowest-energy states near zero field. III. THE PROPERTIES OF THE ORBITAL MAGNETIZATION To obtain the orbital magnetization of the kagom´e lattice in an external magnetic field and with finite temperature, let us first write down the single-particle free energy as follows [16]: F = − 1 β Xn ZMBZ d2kh1 + e ℏ B · Ωn(k)i ln[1 + eβ(µ−ǫnk)]. (8) Here, Ωn(k) is the Berry curvature of electronic Bloch states defined by Ωn(k)=ih∇kunk × Its ∇kunki with unki being the periodic part of Bloch wave for the nth band. 6 integral over the MBZ gives the topological Cn= − 1 the quantized Hall conductance in unit of e2/h (see numbers in Fig. 2). µ in Eq. (8) is the 2π RMBZ d2kΩn(k). The sum of Chern numbers over the integer occupied bands gives the Chern number invariant, namely, electron chemical potential, β=1/kBT , and ǫnk is the magnetic band energy. When the field varies by an infinitesimal quantity, i.e., from B to B+δB, then the magnetic band energy can be linearly expanded as ǫnk(B+δB) = ǫnk(B)−mnk(B) · δB, where mnk is the conventional crystal orbital magnetic moment defined by mnk=−i(e/2ℏ)h∇kunk × [H(k) − ǫnk]∇kunki in the MBZ. The magnetization is then given by the field derivative at fixed temperature and chemical potential, M = − (∂F/∂B)µ,T , with the result d2kmn(k)fn(k) (9) 1 M = Xn ZMBZ β Xn ZMBZ ≡ Mc + MΩ, + d2k e  Ωn ln(cid:2)1 + eβ(µ−ǫnk)(cid:3) where fn(k) is the equilibrium Fermi-Dirac distribution function for nth magnetic Bloch band. The first term in Eq. (9) is just a statistical sum of the orbital magnetic moments of the carriers originating from the self-rotation of the carrier wave packet [19, 20], thus we call this term the conventional part Mc of the orbital magnetization. Whereas the second term MΩ is the Berry-phase correction to the orbital magnetization. This term is of topological nature. Interestingly, it is this Berry-phase term that eventually enters the transport current [21]. At zero temperature the general expression (9) reduces to [10, 16, 22] M = Xn Z µ0 MBZ d2khmn(k) + e  Ωn(k)(µ0 − ǫnk)i , (10) where the upper limit means that the integral is over magnetic Bloch states with energies below the Fermi energy µ0. Now with the help of the knowledge on the magnetic energy spectrum obtained from diagonalizing the Hamiltonian (7), we can numerically obtain the magnetization M for different field f . First, let us consider the case with field f =1/4. In this case, the energy spectrum splits into three bands between which there are two band gaps with the same gap width ∆=1.74 (see Fig. 3(a)). The middle band collapse into a plane in the MBZ. The calculated zero-temperature magnetization M is plotted in Fig. 3(b). From this figure, one can find that with the Fermi energy µ0 increases from −3.26, the lower band begins to be 7 (a) 2 y g r e n E 0 -2 n o i t a z i t e n g a m l a t i b r O 1 (b) 0 -1 -4 (c) 3 0 M , c M -3 -2 Fermi energy 0 2 4 0.0 0.5 kx 1.0 -4 -2 0 Fermi energy 2 4 FIG. 3: (Color online). (a) Energy spectrum of the 2D kagom´e lattice with a field f =1/4. The shadow areas are the energy bands. The corresponding magnetization M and its two components are respectively drawn in (b) and (c) as functions of the Fermi energy µ0. The shaded areas in (b) and (c) are the insulating regions. The dashed and dotted lines in (c) represent Mc and MΩ, respectively. occupied and M begins to decrease from 0. When µ0 increases to −1.74, the lower band is fully occupied and the magnetization decreases to −0.7e/h. When continue increasing µ0, the system enters the first insulating region. The magnetization then linearly increases with the Fermi energy until the middle band is occupied (µ0=0−, at this time M=1.03e/h). When further increasing the Fermi energy, the system immediately comes into the second insulating region and the magnetization jumps down to −1.03e/h and then linearly increases up to 0.7e/h until the upper band begins to be occupied (µ0=1.74). Then the magnetization decreases to 0 with the Fermi energy going through the upper band. As shown in Refs. [17, 18], the totally different behavior of the magnetization in the metallic and insulating regions is due to the different roles Mc and MΩ play in these two regions. For further illustration, we show in Fig. 3(c) Mc (dashed line) and MΩ (dotted line) as functions of the Fermi energy, their sum gives M in Fig. 3(b). One can see that overall Mc and MΩ have opposite contributions to M, which implies that these two parts In each insulating regime the conventional term Mc carry opposite-circulating currents. keeps a constant, which is due to the fact that the upper limit of the k-integral of mn(k) is invariant as the chemical potential varies in the gap. In the metallic region, however, since the occupied states varies with the Fermi energy, thus Mc also varies with µ0, resulting in a decreasing slope shown in Fig. 3(c). The Berry phase term MΩ also displays different 8  (a) 2 0 y g r e n E -2 0.0 0.5 kx 1 (b) 0 n o i t a z i t e n g a m l a t i b r O -1 -4 5 (c) 0 M , c M -5 -4 1.0 -2 Fermi energy 0 2 4 -2 0 Fermi energy 2 4 FIG. 4: (Color online). Energy spectrum and orbital magnetization for f =1/3. (a) 2 y g r e n E 0 -2 0.0 n o i t a z i t e n g a m l a t i b r O 1 (b) 0 -1 -4 5 (c) 0 M , M c -2 Fermi energy 0 2 4 -5 -4 1.0 0.5 kx -2 0 Fermi energy 2 4 FIG. 5: (Color online). Energy spectrum and orbital magnetization for f =1/6. behavior between insulating and metallic regions. In the insulating region, MΩ linearly increases with µ0, as is expected from Eq. (10). The slope of the Berry-phase correction term in insulating regions is proportional to the system's Hall conductance. In the metallic region, however, this term sensitively depends on the topological property of the band in which the chemical potential is located. On the whole the comparison between Fig. 3(b) and Fig. 3(c) shows that the metallic behavior of M is dominated by its conventional term Mc, while in the insulating regime MΩ plays a main role in determining the behavior of M. Now let us consider more complex cases, for example, with a field f =1/3 and with a field f =1/6. In these two case, there are more energy bands and gaps appearing in the energy spectrum [see Figs. 4(a) and 5(a)]. Figures 4(b) and 4(c), and Figs. 5(b) and 5(c), for f =1/3 and f =1/6, respectively, plot the magnetization M and its two components as functions of µ0. Clearly from these figures, one can observe a similar variation of the magnetization by changing the electron's fillings (the Fermi energy µ0). Comparing the case of f =1/3 [Fig. 9  n o 0.8 i t a z i t e n g a m 0.6 l a t i b r O 0.4 0.06 0.09 Fermi energy f=12/49 f=1/4 -2 0 2 4 Fermi energy n o i t a z i t e n g a m l a t i b r O 2 1 0 -1 -4 -6 -9 -2.4 -2.2 8 M 0 -2 2 Fermi energy 4 4 0 -4 6 -2.4 -2.2 -2 0 2 Fermi energy 4 FIG. 6: (Color online). The magnetization M as a function of the Fermi energy µ0 in the fields f =1/4 and f′=12/49. The inset is the enlarged vision of the magnetization M at f′=12/49 when the Fermi energy varies between 0.05 and 0.1. (a) 16 12 (b) 9 0 -4 c M -8 -12 -16 -4 FIG. 7: (Color online). The two components of the magnetization M as functions of the Fermi energy µ0 at f =12/49. 4(a)] to the case of f =1/6 [Fig. 5(a)], one can find that the orbital magnetization in these two fields have a relation M(f = 1 6 , µ0) = M(f = 1 3 ,−µ0). (11) This conclusion is a combined exhibition of the above mentioned symmetries (ii) and (iii) in the Hofstadter energy spectrum. We have known that the topology of the energy spectrum of the 2D lattice system arises from the competition between the periodic potential-induced level broadening and the mag- netic field-induced level discretization. Once the external magnetic field changes, the topol- ogy of the energy spectrum also changes, which results in the simultaneous prominent vari- ation in the magnetization of the 2D system. The above discussions on the magnetization for different fields f =1/4, 1/3 and 1/6 are typical examples. Now a question is put forward: 10 If there is a very small perturbation δf to the field f , then how about the change in the magnetization? The answer is the occurrence of fractal structure in the magnetic de Haas- van Alphen (dHvA) oscillations, which is very important for the experimental measurement of the Fermi surface topology. Now we investigate this fractal structure in the magnetic oscillations for the present system. For convenience and simplicity, we choose the field f =1/4 and the perturbation δf =−1/196 (f′=f + δf =12/49). The calculated magnetization M at the fields f and f′ are plotted in Fig. 6. From Fig. 6 one can observe the following features: (i) In the insulating region, the magnitudes of the magnetization M have little difference at the fields f and f′. The perturbation δf is more small, more little in the changes of the magnetization M; (ii) In the metallic regions at the field f , however, there are dHvA oscillations appearing in the magnetization M at the field f + δf . The perturbation δf is more small, the magnetization oscillates more rapidly; (iii) The collapsed middle band originally at field f =1/4 now spreads at f′=12/49. And in this middle band there are also dHvA oscillations appearing in the orbital magnetization. To see this fractal feature more clearly, we enlarge the magnetization in the inset in Fig. 6; (iv) There are similar variations in the two components of M (see Fig. 7). In the insulating regions at f =1/4, there are little difference in the conventional and Berry-phase terms. However, both terms exhibit different behaviors in the metallic regions. While the Berry-phase term has little difference at different fields as it in the insulating regions, there are many quantum steps appearing in the conventional term at f′=12/49. The reason for these quantum steps appearing in the conventional magnetization is that the topology of the energy spectrum at the field f′=f + δf is different from that at the field f . When the external field f changes to f + δf , the eigenbands at the field f split into lots of subbands. The perturbation δf is more small, the number of the splitting subbands is more large. Because the upper limit of the k-integral of mn(k) is invariant as the Fermi level µ0 varies in the subgap, then the conventional term in the subgap keeps a constant, which is the reason for the quantum step appearing in the conventional term. On the other hand, the integral of the Berry-phase term intimately depends on the Fermi level and has no such relation with the energy gaps (or bands). That results in the little difference in the Berry- phase term at different fields. So, when one changes the electron's fillings through these splitting bands, the total magnetization then exhibits the fractal structure in the magnetic dHvA oscillations. 11 n o i t a z i t e n g a m l a t i b r O 0 0.005t 0.01t 0.02t 4 -2 0 2 Chemical potential 1 0 -1 -4 FIG. 8: (Color online). The magnetization M as a function of the electron chemical potential at the field f =12/49 with different temperatures. In the above discussions on the orbital magnetization of the 2D kagom´e lattice, we have concentrated on the zero-temperature limit and omitted finite-temperature effect. Now let us briefly consider the more realistic cases in which the finite-temperature effect is included. Figure 8 plots the orbital magnetization at the field f =12/49 with different temperatures kBT =0, 0.005, 0.01, and 0.02, respectively. From Fig. 8, one can find that the magnetic oscillations are suppressed by thermal broadening. IV. SUMMARY In summary, we have theoretically investigated the orbital magnetization of a 2D kagom´e lattice in a perpendicular magnetic field. Here, the orbital magnetization includes a con- ventional term and a Berry-phase term, which play different roles in metallic and insulating regions. As examples, we have carefully discussed the orbital magnetization and its two components at the fields f =1/4, 1/3, and 1/6, respectively. By varying the Fermi energy µ0, we have obtained the following results: (i) The conventional term and the Berry-phase term give the opposite contributions, with their magnitudes increasing (decreasing) with in- creasing (decreasing) the Fermi energy in metallic regions; (ii) The conventional term keeps unchanged in insulating regions; (iii) The slope of the Berry-phase term in insulating regions is proportional to the system's Hall conductance. When the flux is applied near a commen- surate one (for example, 1/4), the magnetic dHvA oscillations develop a fractal structure, i.e., the orbital magnetization rapidly oscillates when the Fermi energy varies through the split subbands. The finite-temperature effect has also been shown to suppress the oscillating 12 amplitude of the orbital magnetization. Acknowledgments This work was supported by NSFC under Grants No. 90921003, No. 10904005, No. 60776061 and No. 60776063, and by the National Basic Research Program of China (973 Program) under Grants No. 2009CB929103 and No. 2009CB929300. [1] D. R. Hofstadter, Phys. Rev. B 14, 2239 (1976); G. H. Wannier, Phys. Status Solidi B 88, 757 (1978); G. H. Wannier, G. M. Obermair, and R. Ray, Phys. Status Solidi B 93, 337 (1979). [2] F. H. Claro and G. H. Wannier, Phys. Rev. B 19, 6068 (1979). [3] R. Rammal, J. Phys. (Paris) 46, 1345 (1985). [4] Y. Xiao, V. Pelletier, P. M. Chaikin, and D. A. Huse, Phys. Rev. B 67, 104505 (2003). [5] M. A. Andrade Neto and P. A. Schulz, Phys. Rev. B 52, 14093 (1995). [6] H. Aoki, M. Ando, and H. Matsumura, Phys. Rev. B 54, R17296 (1996). [7] A. Behrooz, M. J. Burns, H. Deckman, D. Levine, B. Whitehead, and P. M. Chaikin, Phys. Rev. Lett. 57, 368 (1986); K. Springer and D. van Harlingen, Phys. Rev. B 36, 7273 (1987); H. Schwabe, G. Kasner, and H. Bottger, Phys. Rev. B 56, 8026 (1997). [8] B. Douc¸ot, W. Wang, B. Pannetier, P. Rammal, A. Vareille, and D. Henry, Phys. Rev. Lett. 57, 1235 (1986); J. M. Gordon, A. M. Goldman, J. Maps, D. Costello, R. Tiberio, and B. Whitehead, Phys. Rev. Lett. 56, 2280 (1986); Q. Niu and F. Nori, Phys. Rev. B 39, 2134 (1989). [9] J. Vidal, R. Mosseri, and B. Douc¸ot, Phys. Rev. Lett. 81, 5888 (1998). [10] O. Gat and J. E. Avron, Phys. Rev. Lett. 91, 186801 (2003). [11] O. Gat and J. E. Avron, New J. Phys. 5, 441 (2003). [12] P. Mohan, F. Nakajima, M. Akabori, J. Motohisa, and T. Fukui, Appl. Phys. Lett. 83, 689 (2003); P. Mohan, J. Motohisa, and T. Fukui, Appl. Phys. Lett. 84, 2664 (2004). [13] M. J. Higgins, Y. Xiao, S. Bhattacharya, P. M. Chaikin, S. Sethuraman, R. Bojko, and D. Spencer, Phys. Rev. B 61, R894 (2000); Y. Xiao, D. A. Huse, P. M. Chaikin, M. J. Higgins, S. Bhattacharya, and D. Spencer, ibid. 65, 214503 (2002). 13 [14] S. Y. Tong, G. Xu, W. Y. Hu, and M. W. Puga, J. Vac. Sci. Technol. B 3, 1076 (1985). [15] K. Shiraishi, H. Tamura, and H. Takayanagi, Appl. Phys. Lett. 78, 3702 (2001). [16] D. Xiao, J. Shi, and Q. Niu, Phys. Rev. Lett. 95, 137204 (2005). [17] Z. Wang and P. Zhang, Phys. Rev. B 76, 064406 (2007). [18] Z. Wang, P. Zhang, and J. Shi, Phys. Rev. B 76, 094406 (2007). [19] M.-C. Chang and Q. Niu, Phys. Rev. B 53, 7010 (1996). [20] G. Sundaram and Q. Niu, Phys. Rev. B 59, 14915 (1999). [21] D. Xiao, Y. Yao, Z. Fang, and Q. Niu, Phys. Rev. Lett. 97, 026603 (2006). [22] T. Thonhauser, D. Ceresoli, D. Vanderbilt, and R. Resta, Phys. Rev. Lett. 95, 137205 (2005). 14 

1906.00644

2

1906

Many-body localized quantum batteries

The collective and quantum behavior of many-body systems may be harnessed to achieve fast charging of energy storage devices, which have been recently dubbed quantum batteries. In this paper, we present an extensive numerical analysis of energy flow in a quantum battery described by a disordered quantum Ising chain Hamiltonian, whose equilibrium phase diagram presents many-body localized (MBL), Anderson localized (AL), and ergodic phases. We demonstrate that i) the low amount of entanglement of the MBL phase guarantees much better work extraction capabilities than the ergodic phase and ii) interactions suppress temporal energy fluctuations in comparison with those of the non-interacting AL phase. Finally, we show that the statistical distribution of values of the optimal charging time is a clear-cut diagnostic tool of the MBL phase.

cond-mat.mes-hall

cond-mat

Many-body localized quantum batteries Davide Rossini,1, ∗ Gian Marcello Andolina,2, 3 and Marco Polini3 1Dipartimento di Fisica dell'Universit`a di Pisa and INFN, Largo Pontecorvo 3, I-56127 Pisa, Italy 3Istituto Italiano di Tecnologia, Graphene Labs, Via Morego 30, I-16163 Genova, Italy 2NEST, Scuola Normale Superiore, I-56126 Pisa, Italy The collective and quantum behavior of many-body systems may be harnessed to achieve fast charging of energy storage devices, which have been recently dubbed "quantum batteries". In this Article, we present an extensive numerical analysis of energy flow in a quantum battery described by a disordered quantum Ising chain Hamiltonian, whose equilibrium phase diagram presents many- body localized (MBL), Anderson localized (AL), and ergodic phases. We demonstrate that i) the low amount of entanglement of the MBL phase guarantees much better work extraction capabilities, measured by the ergotropy, than the ergodic phase and ii) interactions suppress temporal energy fluctuations in comparison with those of the non-interacting AL phase. Finally, we show that the statistical distribution of values of the optimal charging time is a clear-cut diagnostic tool of the MBL phase. I. INTRODUCTION Recently, there has been a great deal of interest in studying charging times and work extraction in "quan- tum batteries" (QBs)1 -- 15. A QB is composed by N iden- tical quantum cells where energy is stored and from which work can be extracted (at least in principle). Prototypi- cal examples of QBs that have been studied include ar- rays of N qubits6 -- 10,13 and XXZ Heisenberg spin chains with N spins5. The quantum cells are either coupled to an energy source, e.g. a photonic cavity6, or they are charged via a non-equilibrium quantum quench16 after they have been prepared in a given state, physically rep- resenting the discharged battery. In the first instance, the coupling allows energy flow from the energy source into the battery. (In the case of Dicke QBs comprising N qubits in a photonic cavity6, the latter also mediates long-range interactions between the qubits.) In the sec- ond instance, the work needed to perform the quench charges the quantum cells during the non-equilibrium dy- namics. In a QB, one is interested in minimizing the charg- ing time, exploiting the collective behavior of the ensem- ble of quantum cells and, possibly, genuine quantum fea- tures of the charging dynamics. At the same time, one needs to maximize the fraction of energy stored in the QB that can be extracted in order to perform thermody- namic work. If the energy stored in the battery is indeed locked by correlations between the N quantum cells8, it cannot be extracted and, despite being maximal at the optimal charging time, is useless from the point of view of performing work. In this Article, we consider a QB model, which, at equi- librium, displays a many-body localized (MBL) phase. MBL phases17 -- 19 have been and still are at the center of intensive theoretical17 -- 19 and experimental20 -- 29 inves- tigations. Many-body localization30,31 is a phenomenon occurring in an interacting disordered quantum many- particle system, in the regime where all single-particle eigenstates are localized by disorder. In condensed mat- ter systems, it is known that electron-phonon collisions induce variable-range-hopping transport at low temper- atures. Surprisingly, only recently it has been demon- strated30,31 that electron-electron interactions are inca- pable of doing so and that, in the absence of phonons, the electron system is locked into an insulating state with zero dc electrical conductivity, up to a critical temper- ature Tc. For T > Tc, interactions produce a metallic state. Since for T < Tc electron-electron interactions alone are unable to establish thermal equilibrium, the study of many-body localization has greatly increased our knowledge of the approach to equilibrium in a closed quantum system, providing us with the only robust mech- anism known so far to avoid thermalization17 -- 19. An- other aspect of MBL states of interest to this work is the fact that these display a low amount of entangle- ment and obey17 -- 19 an area-law rather than a volume- law. Indeed, the entanglement entropy of a subsystem in an MBL eigenstate scales proportionally to the vol- ume of the boundary of the subsystem32. This is in stark contrast to ergodic states, whose entanglement scales like the volume of the subsystem. The reasons to investigate MBL states in the con- text of QBs are threefold. i) All quantum many-body battery models studied so far5,6,8,10,12,13 display ground- state quantum phase transitions33. The charging dynam- ics in these models is therefore qualitatively insensitive to these phase transitions since it involves highly excited states. On the other hand, many-body localization in- volves excited states with finite energy density and is thus expected to imprint qualitative features on the charging dynamics. ii) The low amount of entanglement carried by MBL states is expected8 to lead to a larger value of extractable work (ergotropy) from the QB in comparison with phases which obey a volume-law, like the ergodic one. iii) Finally, interactions in the MBL phase are ex- pected to suppress temporal energy fluctuations with re- spect to other phases obeying an area-law, like the Ander- son localized (AL) phase. In practice, this means much more stable batteries. Below, we present the outcome of extensive numerical 9 1 0 2 p e S 9 1 ] l l a h - s e m . t a m - d n o c [ 2 v 4 4 6 0 0 . 6 0 9 1 : v i X r a calculations on a specific QB model Hamiltonian, namely that of a disordered quantum Ising chain. These fully support the above expectations. Moreover, we show that the statistical distribution of optimal charging times of a QB provides a powerful diagnostic tool of MBL phases and, most importantly, a tool that does not rely on looking at the long-time dynamics of the many-particle system19. We emphasize that we have also simulated an alternative QB model, that of a disordered quantum Heisenberg spin chain. We have reached the same conclu- sions. This makes us reasonably confident on the general nature of our findings, in the context of models displaying MBL, AL, and ergodic phases. Our Article is organized as following. In Sect. II we provide a general definition of a many-body QB and discuss a suitable quench protocol for its charging pro- cess, following the strategy outlined in Ref. 5. We also present the figures of merit we use in our analysis as quantifiers of the "performance" of a many-body QB. The specific model Hamiltonian of our QB is then intro- duced in Sect. III, together with a brief presentation of its phase diagram. Sect. IV contains the main results we obtained, and constitutes the core of the Article, while in Sect. V we draw our main conclusions. Five appendices deal with technical issues concerning the scaling of the energy stored in the QB with its size (App. A), the influ- ence of temporal fluctuations (App. B) and that of the observation time window (App. C). We also discuss the cases of a non-disordered spin-chain QB (App. D) and of a QB modelled by a Heisenberg spin chain in a random magnetic field (App. E). II. MANY-BODY QUANTUM BATTERIES AND FIGURES OF MERIT A. Charging protocol We start by defining a quantum-quench-based protocol for the charging process of a QB. We assume that our battery is made of N identical quantum cells, which are governed by a free and local Hamiltonian34, N(cid:88) j=1 H0 = hj . (1) At time t = 0, the system is prepared in its own ground state 0(cid:105) (representing the discharged battery). By sud- denly switching on a suitable interaction Hamiltonian H1 for a finite amount of time τ , one aims to inject as much energy as possible into the quantum cells3 -- 5. The time interval τ is called the charging time of the protocol. The full Hamiltonian of the model can be written as H(t) = H0 + λ(t)H1 , (2) where λ(t) is a classical parameter that represents the external control exerted on the system, and which is 2 assumed to be given by a step function equal to 1 for t ∈ [0, τ ] and zero elsewhere. (This fast switch can be implemented experimentally, e.g., by controlling an ex- ternal magnetic field in a setup where the quantum cells are realized via superconducting qubits35.) Accordingly, denoting by ψ(t)(cid:105) the evolved state of the system at time t, its total energy Etot N (t) = (cid:104)ψ(t)H(t)ψ(t)(cid:105) is constant at all times, with the exception of the switching points, t = 0 and t = τ . The energy injected into the N quan- tum cells can be thus expressed in terms of the mean local energy at the end of the protocol: EN (τ ) = (cid:104)ψ(τ )H0ψ(τ )(cid:105) . The optimal charging time ¯τ is defined by36 EN (¯τ ) = max τ >0 EN (τ ) . B. Ergotropy (3) (4) An important quantity in the above protocol is the fraction of EN (τ ) that can be effectively extracted from M ≤ N quantum cells, without having access to the full system. Indeed, part of the energy injected into the battery is typically locked by correlations that are estab- lished between the N quantum cells during the charging dynamics, and is thus useless for practical purposes8. 0 =(cid:80)M A proper measure of the fraction of EN (τ ) that can be utilized to perform thermodynamic work is provided by the so-called ergotropy37 -- 39 of the local state ρM (τ ). Here, ρM (τ ) is the reduced density matrix -- evaluated at time τ -- of M ≤ N quantum cells, and governed by the Hamiltonian H(M ) We remind that, for a quantum system X character- ized by a local Hamiltonian H, the ergotropy E(ρ,H) is a functional measuring the maximum amount of energy that can be extracted from a density matrix ρ of X, us- ing an arbitrary unitary operator. A closed expression for this quantity can be obtained in terms of the difference j=1 hj. E(ρ,H) = E(ρ) − E(ρ) (5) between the mean energy E(ρ) = tr[Hρ] of the state ρ and that of the passive counterpart ρ of ρ37 -- 39, i.e. E(ρ) = tr[H ρ]. The latter is defined as the density matrix of X which is diagonal on the eigenbasis of H and whose eigenvalues correspond to a proper reordering of those of ρ, i.e., ρ = rn
n(cid:105)(cid:104)
n , (6) (cid:88) n where ρ =(cid:80) H =(cid:80) Therefore E(ρ) =(cid:80) n rn rn(cid:105)(cid:104)rn , n
n
n(cid:105)(cid:104)
n , r0 ≥ r1 ≥ r2 ≥ ··· ,
0 ≤
1 ≤
2 ≤ ··· . (7) (8) n rn
n. Notice that, if
0 = 0 and the state ρ is pure, then E(ρ) = 0 and the ergotropy coincides with the mean energy of ρ, i.e. E(ρ,H) = E(ρ). On the contrary, if the state ρ is mixed, the extractable work is in general smaller than the mean energy, i.e. E(ρ,H) < E(ρ). Since the global evolution is unitary, ρN (t) = ψ(t)(cid:105)(cid:104)ψ(t) remains pure at all times. In contrast, the local state ρM (τ ) will be in general mixed, because of its entanglement with the remaining N − M elements of the battery, introducing a non-trivial "gap" between its ergotropy EM (τ ) ≡ E end of the charging process, see Eq. (3). (cid:3) and the energy ρM (τ )(cid:3) stored in the subsystem at the (cid:2)ρM (τ ),H(M ) EM (τ ) = tr(cid:2) 0 H(M ) 0 III. MODEL We consider a spin-chain QB. In the absence of charg- ing operations, the system is described by the following non-interacting Hamiltonian: H0 = h σz j , (9) where h > 0 represents a transverse magnetic field (with units of energy), σα j (α = x, y, z) are standard spin-1/2 Pauli matrices, and N is the number of spins (i.e. our quantum cells) in the QB. At time t = 0 the system is initialized in the ground state N(cid:88) j=1 N(cid:79) j=1 0(cid:105) = ↓(cid:105)j . (10) In order to inject energy into the system, we perform a sudden quench by switching on at time t = 0 the Hamil- tonian H1 in Eq. (2). We take N(cid:88) H1 = − N(cid:88) Jjσx j σx j+1 + J2 σx j σx j+2 . (11) j=1 j=1 The two terms entering H1 describe nearest-neighbor and next-to-nearest-neighbor spin-spin interactions. The couplings Jj = J + δJj are the sum of a fixed part J and a randomly-varying contribution δJj, which depends on the site index j, and is taken from a uniform distri- bution: δJj ∈ [−δJ, δJ]. In what follows, we assume periodic boundary conditions, i.e. σα (cid:96) , and fix J =  = 1. (The optimal charging time is therefore measured in units of /J.) We also keep the transverse field strength fixed and equal to h = 0.6 (in units of J), so that the only Hamiltonian parameters that will be changed below are the strength J2 > 0 of next-to- nearest-neighbor interactions and the disorder strength δJ > 0. After mapping the spin-chain model into a fermionic model through a Jordan-Wigner transforma- tion, the next-to-nearest-neighbor interaction term can be seen to be responsible for the breakdown of integra- bility. N +(cid:96) ≡ σα 3 When varying the strength of J2 and δJ, the disor- dered quantum Ising chain Hamiltonian H = H0 + H1 in Eqs. (9)-(11) presents a rich phase diagram, which has been scrutinized in Ref. 40. In particular, for suf- ficiently strong disorder, i.e. for δJ > δJc, its excita- tion spectrum presents a MBL phase. Conversely, for δJ < δJc, the excitation spectrum presents a many-body mobility edge so that eigenstates with energy larger than a given threshold E0 behave as thermal, and are believed to satisfy the eigenstate thermalization hypothesis16,17,41, while low-energy states are localized. If next-to-nearest neighbor couplings are switched off (J2 = 0), the Hamil- tonian H reduces to a conventional spin-1/2 quantum Ising chain in a transverse field. In this case, any finite amount of disorder (δJ > 0) in the nearest-neighbor in- teraction term localizes all the eigenstates, because of the one-dimensional nature of the model42. Setting J2 = 0 yields therefore an AL phase. In order to evaluate the properties of the time evolved state ψ(τ )(cid:105) = e−i(H0+H1)τ0(cid:105), it is convenient to em- ploy a fourth-order Runge-Kutta integration of the time evolution operator43. Moreover, since {δJi} is a random variable, in the following we average the quantities of in- terest O over the distribution of {δJi}. We denote by (cid:104)(cid:104)O(cid:105)(cid:105) the averaged quantity, i.e., (cid:90) (cid:104)(cid:104)O(cid:105)(cid:105) ≡ P ({δJi})O({δJi}) d{δJi} , (12) where P ({δJi}) is the uniform probability to have a fixed value of δJi ∈ [−δJ, δJ]. IV. RESULTS We now analyze the charging process of our quantum Ising chain battery, focusing on the three different phases of the Hamiltonian H0 + H1. Data shown in this section have been obtained by fixing suitable values of the pa- rameters J2 and δJ, as suggested by the phase diagram worked out in Ref. 40. We therefore set J2 = 0 and δJ = 1 to yield a representative AL phase (red color in all the figures); J2 = 0.3 and δJ = 5 to yield a MBL phase (blue color); and J2 = 0.3 and δJ = 1 to yield an ergodic phase (green color). Of course, we have exten- sively tested (not shown) the dependence of all our results upon changes in the values of the parameters, finding that the main features highlighted below are robust and independent of the aforementioned particular choices. A. Optimal charging time First of all, the energy stored in the QB at the optimal time always scales linearly with the number of elements in the battery, thus it is an extensive quantity irrespective of the phase of the system (see App. A). Looking more in detail at the temporal behavior of EN (τ ), we have ob- served consistent temporal fluctuations which have the (a) (b) (c) (d) (e) FIG. 1. (Color online) Panel (a) The energy EN (τ ) as a func- tion of the time τ for a single realization of disorder and N = 14. The three curves correspond to different phases, according to the following color code that will be adopted hereafter: AL phase (red), MBL phase (blue), and ergodic phase (green). Panel (b) The optimal charging time (cid:104)(cid:104)¯τ(cid:105)(cid:105) is plotted as a function of the number of units N in the bat- tery. Data have been averaged over 104 disorder realizations. Panels (c)-(e) The probability P (¯τ ) to find a certain value of the optimal time in a given phase is plotted as a function of ¯τ . Histograms in the three panels refer to a battery with N = 14 units, in the above mentioned phases, and have been obtained by averaging over 105 realizations. The observation time window has been fixed to36 τmax = 50. tendency to be suppressed with increasing N , provided interactions J2 are switched on. An example is reported in Fig. 1(a), where we show EN (τ ) as a function of τ , for a single realization of disorder. We note that, while the behavior at short times does not depend on the micro- scopic details, at longer times the AL phase is character- ized by wild temporal fluctuations of the energy. These are clearly detrimental, as one would like to work with QBs that deliver energy and power in a stable manner. Interactions greatly suppress these fluctuations, yielding stable QBs. We further elaborate on temporal fluctua- 4 tions in App. B. The optimal charging time (cid:104)(cid:104)¯τ(cid:105)(cid:105), averaged over the distribution of {δJj}, is found to decrease with N , see Fig. 1(b). Quantitative differences emerge between the various phases, with a decrease that is much faster in the MBL phase than in the ergodic and AL phases. These results are strongly affected by the finite observation time window, especially in the AL phase where time fluctua- tions are most prominent and cannot be neglected. How- ever, the fact that the optimal charging time (cid:104)(cid:104)¯τ(cid:105)(cid:105) is short- est in the MBL phase is independent of τmax. More insight about the influence of the quantum many- body dynamics on the charging process can be inferred by analyzing the full statistical distribution of the opti- mal times, namely the probability P (¯τ ) to find a certain value of the optimal time in a given phase. The results of our simulations are reported in Figs. 1(c)-(e). In the AL phase [panel (c)], an initial peak in such distribution is followed by a completely flat region, which witnesses the presence of strong temporal fluctuations for any re- alization. P (¯τ ) drastically changes upon switching on interactions while maintaining the system in a localized state (i.e. in the MBL phase). A series of distinct peaks emerges, separated by forbidden regions [panel (d)]. Fi- nally, P (¯τ ) displays a yet different behavior in the ergodic phase [panel (e)], since in this case it exhibits a (nearly) monotonic drop with ¯τ . The growth of P (¯τ ) at long times is an artifact due to the choice of a finite τmax (see App. C). The statistics of the optimal times can be thus used to discriminate between the three different phases of the model, due to clear qualitative differences between the corresponding distributions P (¯τ ). B. Extractable energy In particular, we are interested in: We now focus on the amount of extractable energy EM (τ ). (i) the disorder-averaged fraction of useful energy in half of the battery, i.e. (cid:104)(cid:104)EN/2(τ )/EN/2(τ )(cid:105)(cid:105) as a function of τ ; (ii) the same fraction at fixed τ , viewed as a function of the subsystem size M ≤ N , i.e. (cid:104)(cid:104)EM (τ )/EM (τ )(cid:105)(cid:105). These quantities are of interest because performing operations on all qubits may be experimentally challenging, while it is reasonable to assume fully global control only on a restricted number M of battery unities. These two quantities are reported in Fig. 2, upper and lower panels, respectively. Results in the AL phase [panels (a,d)], are compatible with the following asymptotics8: N→∞(cid:104)(cid:104)EN/2(τ )/EN/2(τ )(cid:105)(cid:105) = 1 , lim M→∞(cid:104)(cid:104)EM (τ )/EM (τ )(cid:105)(cid:105) = 1 . lim (13) (14) Indeed, due to the integrability of the model at J2 = 0, only a small portion of the entire Hilbert space is vis- ited during the quantum dynamics. Consequently, the difference between energy and ergotropy is expected to 10−1110102103¯τ0505EN(τ)05106810121416N0102030hh¯τii01020304050¯τ10−510−410−310−210−1101020304050¯τ10−510−410−310−210−1101020304050¯τ10−510−410−310−210−11(a) (b) (c) 5 (d) (e) (f) FIG. 2. (Color online) Panels (a,b,c) The fraction of extractable energy (cid:104)(cid:104)EN/2(τ )/EN/2(τ )(cid:105)(cid:105) as a function of τ . Panels (d,e,f) The fraction of useful energy (cid:104)(cid:104)EM (τ )/EM (τ )(cid:105)(cid:105) as a function of M/N , at fixed time τ = 32. Panels (a,d) correspond to the AL phase, panels (b,e) to the MBL phase, and panels (c,f) to the ergodic phase. The various solid lines correspond to different values of the number N of quantum cells: N = 8, 10, 12, 14, 16, 18. The lines thicken and brighten with increasing N . Data are averaged over 103 disorder realizations for N up to 14, and over 102 realizations in the cases N = 16 and 18. become negligible in the thermodynamic limit8, in ac- cordance with the trends found numerically. In the MBL phase [panels (b,e)], the presence of an extensive number of exponentially localized constants of motion17,19 keeps the system's dynamics "frozen" as well, and the bipartite entanglement entropy fulfills an area-law scaling (up to logarithmic corrections)17,19. Therefore a scenario quali- tatively analogous to the AL phase occurs8. We checked that the same conclusions apply for simpler integrable non-disordered systems (see App. D). A completely different situation occurs in the er- Indeed, the behavior of godic phase [panels (c,f)]. the fraction of extractable energy with τ suggests that (cid:104)(cid:104)EN/2(τ )/EN/2(τ )(cid:105)(cid:105) saturates to a finite constant when the thermodynamic limit is taken. This is a signature of ergodicity. All the Hilbert space is visited during the dy- namical evolution, and the evolved state locally resembles a canonical ensemble. Hence, the arguments of Ref. 8 do not hold in this case. Similar conclusions apply to generic non-integrable and non-disordered systems (see App. D). We note that (cid:104)(cid:104)EM (τ )/EM (τ )(cid:105)(cid:105) tends to form a plateau in the region M/N < 1/2, which is reminiscent of the re- dundancy plateau emerging in the mutual information as a function of M/N witnessing quantum Darwinism44,45. V. DISCUSSION AND CONCLUSIONS The results presented in this work are not qualita- tively affected by the specific choice of the Hamiltonian in Eqs. (9)-(11). We have performed numerical simulations (see App. E) to verify that similar conclusions apply for a QB described by a XXZ Heisenberg spin chain in the presence of a random field along the z axis46 -- 49. Our findings are amenable to experimental verification. Interacting spin-chain Hamiltonians can be implemented, e.g., by using trapped ions19,23,50. In Ref. 50, for exam- ple, energy transport has been studied in a trapped-ion setup containing N = 10 spins, in the presence of static disorder and dephasing noise, paving the way for stud- ies of energy flow in MBL spin-chain-based QBs. Exact numerical simulations for relatively small values of N , as those carried out in this work, are therefore relevant for interpreting experiments in trapped-ion set-ups and other set-ups simulating a variety of interacting-qubit models, such as superconducting circuits26,28. In the future it will be interesting to study the role of interactions and disorder in the more general context of thermal nano-machines and quantum thermodynam- ics51 -- 58. During the completion of this work we became aware of a recent interesting analysis of disordered QBs59. The 01020304050τ0.40.60.81hhEN/2(τ)/EN/2(τ)ii01020304050τ0.40.60.8101020304050τ0.40.60.8101/21M/N00.51hhEM(τ)/EM(τ)ii01/21M/N00.5101/21M/N00.51authors of this work, however, did not study the role of many-body localization on the figures of merit. ACKNOWLEDGMENTS We wish to thank V. Giovannetti and E. Vicari for useful discussions. Appendix A: Scaling of the stored energy and the average power with N In order to ensure that the mean local energy (3) stored in the QB at the optimal time is an extensive quantity and is thus well defined in the thermodynamic limit, we have verified that it increases linearly with the number of quantum cells N : (cid:104)(cid:104)EN (¯τ )(cid:105)(cid:105) ∼ N . (A1) This is shown in panel (a) of Fig. 3, where the various curves refer to the three different phases introduced in Sect. IV. We observe that the injected energy is generally smaller in the ergodic phase, thus confirming the fact that localized phases yield the optimal scenario where a QB can be operated. Indeed, in these phases, better performances can be achieved, both in terms of charging energy and of work extraction. More in detail, we have found numerical evidence that increasing both the disorder strength δJ and the next-to- nearest neighbor interactions, generally leads to an am- plification of the energy stored in the QB, for fixed N . Indeed, while for the three representative cases reported in the figure the MBL phase appears as the one which optimizes the amount of stored energy, for a given value of N , this observation depends on the choice of δJ. (We remind the reader that, as discussed in Sect. IV, data for the AL phase correspond to δJ = 1, while those for the MBL phase correspond to δJ = 5.) Choosing the same value of δJ for both cases would result in a slightly larger amount of injected energy (≈ 15%) in the AL phase, de- spite the presence of detrimental wild temporal fluctua- tions [see Fig. 1(a) and App. C]. However, one can show that the average value of the injected energy, defined in Eq. (B2) below, is the same for the two phases at fixed δJ. It is also worth mentioning that even the total energy Etot N (t) of the time evolved state ψ(t)(cid:105) scales linearly with N . Indeed, as already commented in Sect. II A, such quantity remains constant at all times, except for the switching points t = 0 and t = τ . Therefore Etot N (τ ) = (cid:104)ψ(0)H(0+)ψ(0)(cid:105) = (cid:104)0H0 + H10(cid:105) . Since 0(cid:105) is the ground state of H0 =(cid:80)N that (cid:104)0H10(cid:105) = 0, and (cid:104)0H00(cid:105) = 0. Furthermore, we checked numerically (not shown) that the variance of the total Hamiltonian H(0+) = H0 + H1 (A2) j=1 hj, we have 6 (a) (b) FIG. 3. (Color online) Panel (a) The energy stored in the QB at the optimal time (cid:104)(cid:104)EN (¯τ )(cid:105)(cid:105), as a function of N , in the AL phase (red), MBL phase (blue), and ergodic phase (green). Panel (b) The average power in the QB at the optimal time (cid:104)(cid:104)PN (¯τ )(cid:105)(cid:105), as a function of N . Same color code as in panel (a). Data have been obtained by averaging over 104 realizations. scales linearly with N , namely: (cid:104)ψ(0)H2(0+)ψ(0)(cid:105) = (cid:104)0H2 10(cid:105) ∼ N , (A3) where we used the fact H00(cid:105) = 0. The fact that the mean and the variance scale extensively with N ensures that the model is well defined in the thermodynamic limit. Another relevant figure of merit for a QB is the average charging power, namely (cid:104)(cid:104)PN (¯τ )(cid:105)(cid:105) ≡ (cid:104)(cid:104)EN (¯τ )/¯τ(cid:105)(cid:105) . (A4) As the average time (cid:104)(cid:104)¯τ(cid:105)(cid:105), this quantity measures the speed of the charging process. The average charging power is shown in panel (b) of Fig. 3, which displays a linear behavior for large enough values of N , which is independent of the particular quantum phase. It is in- teresting to note that, even with respect to this figure of merit, the MBL phase proves to be quantitatively opti- mal. 6810121416N246810hhEN(¯τ)ii6810121416N05101520hhPN(¯τ)iiAppendix B: Temporal fluctuations Any energy storage device is expected to deliver energy and power in a stable fashion. We have investigated the stability of the charging process by analyzing the tempo- ral fluctuations of the mean local energy EN (τ ). These can be quantified by means of the time-averaged variance (cid:90) τmax (cid:16) δE2 N = 1 τmax 0 (cid:17)2 dτ EN (τ ) − EN , (B1) where τmax denotes a given observation time window and EN stands for the average value of the energy of the battery, in the time interval τ ∈ [0, τmax]: dτ EN (τ ) . (cid:90) τmax EN = (B2) 1 τmax 0 7 (a) (b) N(cid:105)(cid:105) and (cid:104)(cid:104)EN(cid:105)(cid:105). In the cases in which disorder is present (δJ (cid:54)= 0), we have also performed averages of the above quantities over a number of different disorder realizations, thus obtaining the averaged quantities (cid:104)(cid:104)δE2 The outcomes of our numerical simulations for the averaged temporal fluctuations (cid:104)(cid:104)δE2 N(cid:105)(cid:105) are reported in Fig. 4. We separately discuss the cases J2 = 0 and J2 (cid:54)= 0 [panels (a) and (b), respectively], since they display dif- ferent qualitative behaviors. Panel (a) shows the data corresponding to two situations within the AL phase (red data sets, with δJ = 1 or 5), compared with the clean (i.e., non-disordered) non-interacting case (purple data set). We observe that fluctuations always increase with the number N of quantum cells, thus mining the useful- ness of any QB operating in this regime. The trend is compatible with a large-N linear growth, even though we have not further investigated this issue. Panel (b) focuses on prototypical interacting situations for J2 = 0.3, comparing the data for the MBL phase (blue) with those for the ergodic (green) and the clean interacting case (cyan data set). In all such cases we observe a general decrease of fluctuations with N , in stark contrast with the non-interacting situations of panel (a). In terms of magnitudes, the MBL phase results in smaller fluctuations at fixed N , thus representing the optimal regime at which a disordered QB should be operated, from a stability point of view. Appendix C: Dependence of P (¯τ ) on the observation time window The analysis of the optimal charging times for the QB, reported in Fig. 1, has been performed for a fixed obser- vation time window, i.e. τmax = 50. This may have some influence on the obtained results, since it introduces an artificial temporal cutoff. We have indeed checked that, if temporal fluctuations of the local energy EN (τ ) are not suppressed in time nor in N , such cutoff becomes partic- ularly important. Specifically, the presence of next-to- nearest neighbor spin-spin interactions is seen to system- atically suppress time fluctuations with N , while this is (Color online) The time average of the variance FIG. 4. of the mean local energy as a function of time, δE2 N , aver- aged over 102 realizations and with an observation time win- dow τmax = 103, as a function of N . Panel (a) describes non-interacting situations (J2 = 0), including the AL phase (with δJ = 1 or 5, corresponding to red lines with circles or diamonds respectively) and the clean integrable model (δJ = 0, purple line). Panel (b) describes interacting situ- ations (J2 = 0.3), including the ergodic phase (δJ = 1, green line), the MBL phase (δJ = 5, blue line), and the clean non- integrable model (δJ = 0, cyan line). not the case for J2 = 0 (i.e., in the AL phase, where they are particularly pronounced). As a consequence, enlarg- ing the observation time window would cause a consis- tent increase of the measured optimal charging time in the AL phase. The corresponding histogram would nev- ertheless remain qualitatively unaffected, with a charac- teristic long and flat tail extending at arbitrarily large values of ¯τ [see Fig. 1(c)]. A less pronounced influence of τmax on ¯τ can be wit- nessed in the ergodic phase (J2 (cid:54)= 0), since time fluctua- tions are indeed suppressed with N . However, the shape of P (¯τ ), shown in Fig. 1(e), may be significantly modify in its tail. This is shown in Fig. 5, where we compare the statistical distribution of optimal times for two different values of τmax, i.e. τmax = 50 (light green) and τmax = 100 (dark green). It is clearly visible that, while the distri- 6810121416N10−113hhδE2Nii6810121416N10−210−11hhδE2Nii(a) (b) 8 (c) (d) FIG. 6. (Color online) Analysis of the extractable work for a QB modeled by a clean spin chain (δJ = 0). Panels (a,b) show the fraction of useful energy EN/2(τ )/EN/2(τ ) as a func- tion of τ . Panels (c,d) show the fraction of useful energy EM (τ )/EM (τ ) as a function of the fraction M/N at the given time τ = 32. Panels (a,c) [respectively, Panels (b,d)] corre- spond to the non-interacting (respectively, interacting) phase, which is shown in purple (respectively, cyan). Different solid lines correspond to different values of the number of cells in the QB: N = 8, 10, . . . , 18. An increase in thickness and brightness of the various lines corresponds to an increase of N . j σx coupling term (σx j+2) into a quartic fermion operator. We start by focusing on the non-interacting case (quan- tum Ising chain in a transverse field). Panel (a) of Fig. 6 displays the fraction of useful energy for half of the sys- tem EN/2(τ )/EN/2(τ ) as a function of τ , while panel (c) displays the same fraction as a function of the fraction M/N , i.e. EM (τ )/EM (τ ), at the given time τ = 32. We first recognize the presence of strong oscillations in time, reminiscent of the wild temporal fluctuations for J2 = 0 (see above). Moreover, the trend with N suggests that EN/2(τ )/EN/2(τ ) → 1 and EM (τ )/EM (τ ) → 1, analo- gously to what has been reported in the AL and MBL phases, cf., Eqs. (13)-(14) [see panels (a,d) and (b,e) of Fig. 2, respectively]. Indeed, as it occurs in a localized phase, due to the presence of multiple constants of mo- tion (i.e. the number operators corresponding to the fermionic quasiparticles which diagonalize the model), the quantum dynamics is constrained in a small subpor- tion of the full Hilbert space, and the difference between energy and ergotropy is expected to become negligible with increasing N . Panels (b,d) of Fig. 6 show the same quantities of pan- els (a,c), but for the interacting case (quantum Ising chain in a transverse field, with next-to-nearest neigh- bor couplings). The behavior of the useful energy versus FIG. 5. Probability P (¯τ ) in the ergodic phase as a function of ¯τ , for two fixed values of τmax = 50 (green histogram) and τmax = 100 (dark green histogram). All other parameters are the same as in Fig. 1 panel (e), the representative case for the ergodic phase. In order to calculate the distribution, we simulated 105 realizations of disorder. bution for small values of ¯τ is insensitive to the choice of τmax, the late-time growth for ¯τ ∼ τmax is a finite-time window effect, which depends on the particular choice of τmax. Finally, the charging time behavior in the MBL phase is less affected by τmax, as, for a typical disorder realiza- tion, time fluctuations are smaller and ¯τ (cid:28) τmax. Appendix D: Ergotropy for a non-disordered spin-chain QB We now discuss the extractable work in a QB where disorder is absent (δJ = 0), focusing on the two paradig- matic cases where interactions are either absent (J2 = 0, purple) or present (J2 = 0.3, cyan). For the sake of clarity, we keep the other parameters fixed to the same values as in Sect. IV, i.e. J = 1 and h = 0.6. The consid- erations reported below are not qualitatively affected by this choice. We also remind the reader that, in this case, statistical averages are not required, since the model is completely deterministic. The spin-chain Hamiltonian H = H0 +H1 [see Eqs. (9) and (11)], with δJ = 0 and J2 = 0, reduces to the integrable quantum Ising chain in a transverse field. Such model admits a simple and manageable solution in terms of free quasiparticles, after first mapping it into a quadratic fermion model through a Jordan-Wigner trans- formation, and then performing a Bogoliubov rotation33. On the other hand, for δJ = 0 and J2 (cid:54)= 0, it turns out to be non-integrable, since the Jordan-Wigner trans- formation maps the next-to-nearest neighbor spin-spin 0255075100¯τ10−510−410−310−210−1101020304050τ0.20.40.60.81EN/2(τ)/EN/2(τ)01020304050τ0.20.40.60.8101/21M/N00.51EM(τ)/EM(τ)01/21M/N00.51time and the fraction of the system size suggests that it saturates to a finite constant (different from one), when the thermodynamic limit is taken. Again, we can ascribe this fact to the presence of interactions (i.e. J2), which are known to break integrability and thus to induce er- godicity, similarly to what has been reported in the er- godic phase for the disordered QB model [panels (c,f) of Fig. 2]. We conclude that the asymptotic behavior of the ergotropy is strongly connected with the integrability of the underlying dynamical model. Appendix E: Disordered Heisenberg spin-chain model In order to verify the robustness of our findings and claims, we have also performed numerical simulations for a different spin-chain QB. Specifically, we have consid- ered an alternative QB model which undergoes the same charging protocol described in Sect. II A, and is charac- terized by a time-dependent Hamiltonian of the form as in Eq. (1). The free Hamiltonian H0 is constituted by single-particle spin-flip terms 9 FIG. 7. (Color online) The mean local energy EN (τ ) as a function of the time τ , for a disordered Heisenberg spin-chain QB operating in the various phases (from top to bottom, AL, MBL, ergodic). Here, and in the next two figures, the color code is the same used in Figs. 1 and 2, while the Hamiltonian parameters for the representative cases corresponding to the three above mentioned phases are reported in this section. Data are for a single realization of disorder and N = 12. (cid:88) (cid:0)σx (cid:88) (cid:0)σ+ j j H0 = J 2 = J (cid:1) j σx j σ− j+1 j σy j+1 + σy j+1 + h.c.(cid:1) , with the conservation of the total magnetization along the z axis, i.e., (cid:2) H0 + H1, Sz(cid:3) = 0 , where Sz = σz j , (E3) (cid:88) j (E1) 2 (σx j = 1 j ± σy where σ± j ) denotes raising/lowering opera- tors for spin-1/2 particles. The charging part H1 is the sum of a nearest-neighbor spin-spin coupling along the z axis (namely, the interaction), and a random field along the same direction: (cid:88) H1 = ∆ 2 (cid:88) σz j σz j+1 + hz j σz j , (E2) j j where ∆ denotes the strength of the z-axis anisotropy and hz j is a randomly-varying field strength, which is taken from a uniform distribution hz j ∈ [−W, W ]. The model defined above is usually referred to as the "disordered XXZ Heisenberg spin chain", and consti- tutes a prototypical example where a many-body lo- calization/delocalization transition can be studied46 -- 49. Extensive numerical simulations have shown that, for J = ∆ = 1, such transition occurs at W (cid:63) ≈ 3.549. We also recall that, by means of a Jordan-Wigner transfor- mation that maps spins into fermions, the model is uni- tarily equivalent to a one-dimensional chain of spinless fermions, interacting through a nearest-neighbor density- density term, and in the presence of a random onsite chemical potential. the parity operator P = (cid:81) The reason why we preferred to show results for the disordered quantum Ising chain model in Sect. IV is that the latter only preserves a global Z2 symmetry given by In contrast, the XXZ Heisenberg spin chain hosts a U (1) symmetry associated j σz j . and therefore its dynamics is constrained onto a smaller manifold of the Hilbert space. However, as we shall see below, the same qualitative features emerge. As a con- sequence, we can reasonably assure that the MBL phase represents the optimal compromise, in terms of stabil- ity properties and work extraction, where a many-body disordered QB can operate. Analogously to the quantum Ising chain battery, by changing the various Hamiltonian parameters of H0 and H1, one can probe the phase diagram of the model, which is characterized by the same three distinct phases ana- lyzed before: the AL, MBL, and ergodic phases. We ex- plicitly focus on one representative of each of such phases. However, we have checked that qualitatively analogous features emerge when the values of the parameters are changed in such a way to stay within the same phase. For the sake of simplicity, we express all the various pa- rameters in units of J = 1, which is thus taken as the reference energy scale. We set ∆ = 0 and W = 5 to yield a representative point in the AL phase (red color in all the figures of this section); ∆ = 1 and W = 5 for the MBL phase (blue color); ∆ = 1 and W = 1 for the ergodic phase (green color). The mean energy injected in the QB, for a single dis- order realization, is shown in Fig. 7. We observe that a QB operating in the AL phase is characterized by large temporal fluctuations of the energy, which are generally suppressed when interactions are switched on (not shown, for this model -- see, however, Fig. 4 for the quantum Ising chain). While the energy at the optimal time ¯τ can 10−1110102103¯τ0505EN(τ)0510(a) (b) (c) 10 FIG. 8. (Color online) The probability P (¯τ ) to find a certain value of the optimal charging time in a given phase, for a disordered Heisenberg spin-chain battery, is plotted as a function of ¯τ . Histograms in the three panels refer to a battery with N = 12 units, in the above mentioned phases (from left to right, AL, MBL, ergodic), and have been obtained by averaging over 105 realizations. The observation time is τmax = 50. (a) (b) (Color online) The fraction of extractable energy FIG. 9. (cid:104)(cid:104)EN/2(τ )/EN/2(τ )(cid:105)(cid:105) as a function of τ , for a disordered Heisenberg spin-chain battery. The two panels refer to MBL (a) and ergodic (b) phase. The various solid lines corre- spond to different values of the number N of quantum cells: N = 8, 10, 12, 14, 16, 18. The lines thicken and brighten with increasing N . Data are averaged over 103 disorder realiza- tions for N up to 14, and over 102 realizations for N = 16 and 18. be slightly larger in the AL phase, rather than in the MBL phase, its time average is the same up to a (cid:46) 2% discrepancy. In contrast, for the ergodic phase, a lower amount of disorder reflects into a smaller injected energy. A closer look at the statistics of the optimal charg- ing times ¯τ in the three quantum phases, displayed in Fig. 8, reveals the same qualitative features we observed in Sect. IV A. Namely, after a pronounced peak for short times, both the AL phase [panel (a)] and the ergodic phase [panel (c)] present an approximately flat distri- bution. This witnesses the presence of temporal fluc- tuations at long times. The MBL phase [panel (b)] is characterized by a rather distinct behavior: the emerg- ing sequence of peaks is reminiscent of those observed in the Ising chain model [cf., Fig 1, panel (d)]. The more spurious behavior observed here may be due to the re- duced number N of quantum cells that we considered, and to the constraint in the quantum dynamics imposed by the conservation of the magnetization along the z axis. Specifically, at N = 12 as is the case for this figure, the quantum dynamics starting from the ground state of H0 is constrained in a subspace of the system's Hilbert space size of the Hilbert space 212 = 4096). having dimension (cid:0)12 (cid:1) = 924 (to be compared with the 6 The absence of complete ergodicity in the full Hilbert space, even in the ergodic phase, also emerges in the anal- ysis of the fraction of extractable energy, over half of the system. We report the corresponding data for the MBL and the ergodic phase in Fig. 9. We clearly see, as ob- served in Sect. IV B, that by increasing N , better per- formances in terms of work extraction are achieved in a localized situation [panel (a)], rather than in the ergodic case [panel (b)]. However, we report a less distinct behav- ior for the two regimes, as compared with those observed in the Ising model [cf., Fig. 2, panels (b) and (c)]. davide.rossini@unipi.it ∗ 1 R. Alicki and M. Fannes, Phys. Rev. E 87, 042123 (2013). 2 K.V. Hovhannisyan, M. Perarnau-Llobet, M. Huber, and A. Ac´ın, Phys. Rev. Lett. 111, 240201 (2013). 3 F.C. Binder, S. Vinjanampathy, K. Modi, and J. Goold, New J. Phys. 17, 075015 (2015). 4 F. Campaioli, F.A. Pollock, F.C. Binder, L. C´eleri, J. Goold, S. Vinjanampathy, and K. Modi, Phys. Rev. Lett. 118, 150601 (2017). 5 T.P. Le, J. Levinsen, K. Modi, M. Parish, and F.A. Pollock, Phys. Rev. A 97, 022106 (2018). 6 D. Ferraro, M. Campisi, G.M. Andolina, V. Pellegrini, and M. Polini, Phys. Rev. Lett. 120, 117702 (2018). 7 G.M. Andolina, D. Farina, A. Mari, V. Pellegrini, V. Gio- vannetti, and M. Polini, Phys. Rev. B 98, 205423 (2018). 8 G.M. Andolina, M. Keck, A. Mari, M. Campisi, V. Giovan- 01020304050¯τ10−510−410−310−210−1101020304050¯τ10−510−410−310−210−1101020304050¯τ10−510−410−310−210−1101020304050τ0.60.81hhEN/2(τ)/EN/2(τ)ii01020304050τ0.60.81netti, and M. Polini, Phys. Rev. Lett. 122, 047702 (2019). 9 D. Farina, G.M. Andolina, A. Mari, M. Polini, and V. Giovannetti, Phys. Rev. B 99, 035421 (2019). 10 S. Juli´a-Farr`e, T. Salamon, A. Riera, M.N. Bera, and M. Lewenstein, arXiv:1811.04005. 11 Y.-Y. Zhang, T.-R. Yang, L. Fu, and X. Wang, Phys. Rev. E 99, 052106 (2019). 12 G.M. Andolina, M. Keck, A. Mari, V. Giovannetti, and M. Polini, Phys. Rev. B 99, 205437 (2019). 13 X. Zhang and M. Blaauboer, arXiv:1812.10139. 14 For a recent review on QBs see e.g. F. Campaioli, F.A. Pollock, and S. Vinjanampathy, in Thermodynamics in the Quantum Regime, F. Binder, L.A. Correa, C. Gogolin, J. Anders, and G. Adesso (eds.), p. 207-225, (Springer, 2018). Also available as arXiv:1805.05507. 15 K. Kaneko, E. Iyoda, and T. Sagawa, Phys. Rev. E 99, 032128 (2019). 16 For a recent review on the non-equilibrium dynamics of isolated quantum systems see e.g. A. Polkovnikov, K. Sen- gupta, A. Silva, and M. Vengalattore, Rev. Mod. Phys. 83, 863 (2011). 17 R. Nandkishore and D.A. Huse, Annu. Rev. Condens. Mat- ter Phys. 6, 15 (2015). 18 F. Alet and N. Laflorencie, C. R. Phys. 19, 498 (2018). 19 D.A. Abanin, E. Altman, I. Bloch, and M. Serbyn, Rev. Mod. Phys. 91, 021001 (2019). 20 M. Schreiber, S.S. Hodgman, P. Bordia, H.P. Lschen, M.H. Fischer, R. Vosk, E. Altman, U. Schneider, and I. Bloch, Science 349, 842 (2015). 21 J.-y. Choi, S. Hild, J. Zeiher, P. Schauss, A. Rubio-Abadal, T. Yefsah, V. Khemani, D.A. Huse, I. Bloch, and C. Gross, Science 352, 1547 (2016). 22 P. Bordia, H.P. Luschen, S.S. Hodgman, M. Schreiber, I. Bloch, and U. Schneider, Phys. Rev. Lett. 116, 140401 (2016). 23 J. Smith, A. Lee, P. Richerme, B. Neyenhuis, P.W. Hess, P. Hauke, M. Heyl, D.A. Huse, and C. Monroe, Nat. Phys. 12, 907 (2016). 24 P. Bordia, H. Luschen, S. Scherg, S. Gopalakrishnan, M. Knap, U. Schneider, and I. Bloch, Phys. Rev. X 7, 041047 (2017). 25 H.P. Luschen, P. Bordia, S.S. Hodgman, M. Schreiber, S. Sarkar, A.J. Daley, M.H. Fischer, E. Altman, I. Bloch, and U. Schneider, Phys. Rev. X 7, 011034 (2017). 26 P. Roushan, C. Neill, J. Tangpanitanon, V.M. Bastidas, A. Megrant, R. Barends, Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth, A. Fowler, B. Foxen, M. Giustina, E. Jeffrey, J. Kelly, E. Lucero, J. Mutus, M. Neeley, C. Quintana, D. Sank, A. Vainsencher, J. Wenner, T. White, H. Neven, D. G. Angelakis, and J. Martinis, Science 358, 1175 (2017). 27 K.X. Wei, C. Ramanathan, and P. Cappellaro, Phys. Rev. Lett. 120, 070501 (2018). 28 K. Xu, J.-J. Chen, Y. Zeng, Y.-R. Zhang, C. Song, W. Liu, Q. Guo, P. Zhang, D. Xu, H. Deng, K. Huang, H. Wang, X. Zhu, D. Zheng, and H. Fan, Phys. Rev. Lett. 120, 050507 (2018). 29 A. Lukin, M. Rispoli, R. Schittko, M.E. Tai, A.M. Kauf- man, S. Choi, V. Khemani, J. L´eonard, and M. Greiner, Science 364, 256 (2019). 30 D.M. Basko, I.L. Aleiner, and B.L. Altshuler, Ann. Phys. 321, 1126 (2006). 31 I.V. Gornyi, A.D. Mirlin, and D.G. Polyakov, Phys. Rev. Lett. 95, 206603 (2005). 32 J. Eisert, M. Cramer, and M.B. Plenio, Rev. Mod. Phys. 82, 277 (2010). 33 S. Sachdev, Quantum Phase Transitions (Cambridge Uni- 11 versity Press, Cambridge, 1999). 34 By "free" and "local" we mean that the Hamiltonian H0 is a sum of terms which only involve a single quantum cell of the battery. For the sake of convenience and without loss of generality, we also set to zero the ground-state energy
0 of H0, by adding to the Hamiltonian a trivial constant, i.e. +N h. 35 J.M. Fink, R. Bianchetti, M. Baur, M. Goppl, L. Steffen, S. Filipp, P.J. Leek, A. Blais, and A. Wallraff, Phys. Rev. Lett. 103, 083601 (2009). 36 For practical reasons, of relevance for both numerical calcu- lations and real experiments, such maximum can be taken by calculating (measuring) EN (τ ) over a finite observation time window τmax. 37 A.E. Allahverdyan, R. Balian, and T.M. Nieuwenhuizen, Europhys. Lett. 67, 565 (2004). 38 A. Lenard, J. Stat. Phys. 19, 575 (1978). 39 W. Pusz and S.L. Woronowicz, Commun. Math. Phys. 58, 273 (1978). 40 J.A. Kjall, J. H. Bardarson, and F. Pollmann, Phys. Rev. Lett. 113, 107204 (2014). 41 L. D'Alessio, Y. Kafri, A. Polkovnikov, and M. Rigol, Adv. Phys. 65, 239 (2016). 42 P.W. Anderson, Phys. Rev. 109, 1492 (1958). 43 We simulated systems with up to N = 18 spins, over a maximum observation time window τmax = 100. We checked that, under these conditions, an integration time −2 is sufficient to guarantee convergence of all step δt = 10 our results, with an error bar that is negligible on the scale of the plotted figures. 44 W.H. Zurek, Nat. Phys. 5, 181 (2009). 45 C.J. Riedel, W.H. Zurek, and M. Zwolak, New J. Phys. 14, 083010 (2012). 46 V. Oganesyan and D.A. Huse, Phys. Rev. B 75, 155111 (2007). 47 M. Znidaric, T. Prosen, and P. Prelovsek, Phys. Rev. B 77, 064426 (2008). 48 A. Pal and D.A. Huse, Phys. Rev. B 82, 174411 (2010). 49 D.J. Luitz, N. Laflorencie, and F. Alet, Phys. Rev. B 91, 081103(R) (2015). 50 C. Maier, T. Brydges, P. Jurcevic, N. Trautmann, C. Hempel, B.P. Lanyon, P. Hauke, R. Blatt, and C.F. Roos, Phys. Rev. Lett. 122, 050501 (2019). 51 M. Campisi, P. Hanggi, and P. Talkner, Rev. Mod. Phys. 83, 1653 (2011). 52 D. Gelbwaser-Klimovsky, W. Niedenzu, and G. Kurizki, Adv. At. Mol. Opt. Phys. 64, 329 (2015). 53 J. Goold, M. Huber, A. Riera, L. del Rio, and P. Skrzypczyk, J. Phys. A: Math. Theor. 49, 143001 (2016). 54 S. Vinjanampathy and J. Anders, Contemp. Phys. 57, 545 (2016). 55 P. Strasberg, G. Schaller, T. Brandes, and M. Esposito, Phys. Rev. X 7, 021003 (2017). 56 W. Niedenzu and G. Kurizki, New J. Phys. 20, 113038 (2018). 57 N.Y. Halpern, C.D. White, S. Gopalakrishnan, and G. Re- fael, Phys. Rev. B 99, 024203 (2019). 58 G. Watanabe, B. Prasanna Venkatesh, P. Talkner, M.-J. Hwang, and A. del Campo, arXiv:1904.07811. 59 S. Ghosh, T. Chanda, and A. Sen De, arXiv:1905.12377. 

1304.1011

2

1304

Mechanics of Adhered, Pressurized Graphene Blisters

We study the mechanics of pressurized graphene membranes using an experimental configuration that allows the determination of the elasticity of graphene and the adhesion energy between a substrate and a graphene (or other two-dimensional solid) membrane. The test consists of a monolayer graphene membrane adhered to a substrate by surface forces. The substrate is patterned with etched microcavities of a prescribed volume and when they are covered with the graphene monolayer it traps a fixed number (N) of gas molecules in the microchamber. By lowering the ambient pressure, and thus changing the pressure difference across the graphene membrane, the membrane can be made to bulge and delaminate in a stable manner from the substrate. Here we describe the analysis of the membrane/substrate as a thermodynamic system and explore the behavior of the system over representative experimentally-accessible geometry and loading parameters. We carry out companion experiments and compare them to the theoretical predictions and then use the theory and experiments together to determine the adhesion energy of graphene/SiO2 interfaces. We find an average adhesion energy of 0.24 J/m2 which is lower, but in line with our previously reported values. We assert that this test, which we call the constant N blister test, is a valuable approach to determine the adhesion energy between two-dimensional solid membranes and a substrate, which is an important, but not well-understood aspect of behavior. The test also provides valuable information that can serve as the basis for subsequent research to understand the mechanisms contributing to the observed adhesion energy. Finally, we show how in the limit of a large microcavity, the constant N test approaches the behavior observed in a constant pressure blister test and we provide an experimental observation that suggests this behavior.

cond-mat.mes-hall

cond-mat

Mechanics of Adhered, Pressurized Graphene Blisters Narasimha G. Boddeti1, Steven P. Koenig1, Rong Long1,2, Jianliang Xiao1, J. Scott Bunch1, and Martin L. Dunn3 1Department of Mechanical Engineering University of Colorado at Boulder Boulder, Colorado 80309 2Current address: Department of Mechanical Engineering University of Alberta Edmonton, Alberta T6G 2G8, Canada 3Singapore University of Technology and Design Singapore, 138682 Abstract We study the mechanics of pressurized graphene membranes using an experimental configuration that allows the determination of the elasticity of graphene and the adhesion energy between a substrate and a graphene (or other two-dimensional solid) membrane. The test consists of a monolayer graphene membrane adhered to a substrate by surface forces. The substrate is patterned with etched microcavities of a prescribed volume and when they are covered with the graphene monolayer it traps a fixed number (N) of gas molecules in the microchamber. By lowering the ambient pressure, and thus changing the pressure difference across the graphene membrane, the membrane can be made to bulge and delaminate in a stable manner from the substrate. This is in contrast to the more common scenario of a constant pressure membrane blister test where membrane delamination is unstable and so this is not an appealing test to determine adhesion energy. Here we describe the analysis of the membrane/substrate as a thermodynamic system and explore the behavior of the system over representative experimentally-accessible geometry and loading parameters. We carry out companion experiments and compare them to the theoretical predictions and then use the theory and experiments together to determine the adhesion energy of graphene/SiO2 interfaces. We find an average adhesion energy of 0.24 J/m2 which is lower, but in line with our previously reported values. We assert that this test – which we call the constant N blister test – is a valuable approach to determine the adhesion energy between two-dimensional solid membranes and a substrate, which is an important, but not well-understood aspect of behavior. The test also provides valuable information that can serve as the basis for subsequent research to understand the mechanisms contributing to the observed adhesion energy. Finally, we show how in the limit of a large microcavity, the constant N test approaches the behavior observed in a constant pressure blister test and we provide an experimental observation that suggests this behavior. 1. Introduction Graphene consists of a single, or a few, layers of carbon atoms bonded by strong covalent bonds within a layer, but weaker van der Waals bonds between layers. A monolayer of graphene represents the ultimate limit in thickness for two-dimensional solids. Graphene has impressive electrical, physical, and mechanical properties (Geim, 2009) and as a result has been pursued for many technological applications including electronics, barriers, and energy storage (Lin et al., 2010; Chen et al., 2011; El-kady et al., 2012; Zhu et al., 2012). Because graphene is so thin, it can also be extremely compliant when it has in-plane dimensions on the order of only a few microns and this makes structures fabricated from graphene susceptible to adhesion to a substrate or neighboring structures. Myriad structures have been created from graphene sheets; some in reality (Low et al., 2012; Scharfenberg et al., 2011; Levy et al., 2010; Pan et al., 2012; Meyer et al., 2007; Kim et al., 2011) and many more in computational simulations (Li and Zhang, 2010; Aitken and Huang, 2010; Lu and Dunn, 2010) that provide important future directions. Blisters are a seemingly simple class of structures that have been observed in various shapes and sizes as a result of graphene fabrication processes and intentionally fabricated to yield attractive technological characteristics, such as strain-engineered electronic properties (Georgiou et al., 2011). Graphene membranes deformed by indentation with an atomic force microscope (AFM) (Lee et al., 2008), intercalation of nanoparticles (Zong et al., 2010), and controlled pressurization by a gas (Bunch et al., 2008; Koenig et al., 2011) have been used to determine various mechanical, and more recently adhesive, properties of graphene. Specifically, in a previous rapid communication (Koenig et al., 2011), we developed a particularly attractive graphene blister test where we mechanically exfoliated graphene membranes (from 1 to 5 layers) on top of a circular cavity (~5 m diameter and ~300 nm depth) that was microfabricated on a silicon substrate with a thick layer of SiO2 on its surface. This resulted in a graphene membrane adhered to SiO2, presumably by van der Waals forces, and suspended over the cavity with gas trapped inside of it because graphene is impermeable to gas molecules (Bunch et al., 2008; Koenig et al., 2011). We charged this cavity/membrane device in a high-pressure chamber so that the pressure inside and outside the cavity equilibrated at a prescribed value, and then we removed it to ambient at which point the pressure outside the cavity was less than that inside of it and this caused the membrane to bulge. The membrane, or blister, bulged under the condition that a fixed number of molecules of gas was trapped in the chamber. If the charging pressure exceeded a critical value, the blister not only bulged, but it also delaminated from the SiO2 substrate in a stable manner. After delamination the graphene retained the form of a circular blister, but with an increased radius and height and a decreased pressure in the cavity due to the increased volume under the bulge. In our previous communication we used this cavity/blister system, which we termed a constant N blister test (N denotes number of molecules) to determine elastic properties of monolayer and multilayer graphene as well as the adhesion energy between graphene and SiO2. Here we describe the mechanics of this test in detail; although it seems fairly straightforward, it admits rich and interesting phenomena across experimentally-accessible system parameters. We demonstrate, through a series of examples, the phenomena of deformation, stability, and interfacial delamination and show how the analysis can be combined with measurements of blister shapes with an AFM to determine elastic and adhesive properties. We use a combination of our previously-reported data and new measurements to demonstrate the utility of these blister tests. In our study of graphene blisters adhered to a substrate we adopt a continuum viewpoint 2 and describe the interaction in terms of an effective adhesion energy that results from the surface forces between the graphene membrane and substrate. We do not consider the origin of these surface forces, e.g., van der Waals, capillary, etc. (DelRio et al., 2007, 2008), which is itself not well understood, and remains a fruitful area for future research. (cid:1848)(cid:2868) (radius (cid:1853)(cid:2868) ) containing N molecules of a gas, an isotropic elastic membrane (Poisson’s ratio, 2. Graphene Blisters and the Constant N Pressurized Membrane Test energy, (cid:1985)), and an external environment at a prescribed pressure (cid:1868)(cid:3032) . We realized structures We consider a blister test structure (Fig. 1) that consists of a circular cylindrical cavity of volume , Young’s modulus, E, and thickness, t) adhered to the surface of the substrate (adhesion photolithographically-defined cylindrical cavities of radii (cid:1853)(cid:2868) = 2.32 μm and 2.55 μm on Chip 1 consisting of monolayer graphene membranes adhered to a SiO2 surface through a combination of microfabrication and mechanical exfoliation of graphene. Specifically, we prepared the graphene blisters on two Si wafers, referred to as Chip 1 and Chip 2 hereafter. We and Chip 2, respectively, and the Si surface was thermally oxidized to realize a 285 nm thick layer of SiO2. We etched multiple cylindrical cavities to nominal depths of 293nm and 290nm with reactive ion etching for Chip 1 and Chip 2, respectively. We then deposited suspended graphene sheets over the microcavities via mechanical exfoliation with natural graphite. Our samples consisted of 5 monolayer membranes on Chip 1 and 4 monolayer membranes on Chip 2. We verified that the graphene was a monolayer using a combination of measurement techniques including Raman spectroscopy, optical contrast, AFM, and elastic constants; the procedures are After exfoliation we charge the system in a chamber so that the internal pressure (cid:1868)(cid:3036) and external similar to that used in our previous studies (Koenig et al., 2011). The monolayers appear to be pressure are equal at a prescribed value (cid:1868)(cid:2868) . Practically the charging occurs over a period of quite flat on the substrate, with insignificant pull-in into the cavities. about seven days as gas molecules (N2 in our study) diffuse through the SiO2 layer and become trapped within the microchamber over the time scale of the remainder of the test. Further details regarding the gas diffusion are given in Bunch et al. (2008) and Koenig et al. (2011). chamber has the effect of fixing (cid:1868)(cid:3032) at a new value (cid:1868)(cid:3032) (cid:3407) (cid:1868)(cid:2868) , which results in a pressure difference across the membrane that causes it to bulge and increases the volume by (cid:1848)(cid:3029) . Over the time scale At this state the membrane is flat, adhered to the substrate at outer perimeter, and spans the cylindrical cavity which holds N gas molecules (Fig. 1a). Removing the device from the constant pressure membrane inflation tests. As a result, the internal cavity pressure (cid:1868)(cid:3036) drops to a value (cid:1868)(cid:3036) (cid:3407) (cid:1868)(cid:2868) . If the charging pressure is below a critical value, (cid:1868)(cid:3030)(cid:3045) , the pressure difference of the subsequent measurements diffusion of the gas through the SiO2 is insignificant and so N (cid:1868) (cid:3404) (cid:1868)(cid:3036) (cid:3398) (cid:1868)(cid:3032) across the membrane causes it to bulge into a nearly-spherical cap, while can be considered fixed; we refer to this as a Constant N test as opposed to more common (cid:1868)(cid:3030)(cid:3045) , the membrane will delaminate from the outer perimeter of the cavity. In the final equilibrium configuration the cavity volume is (cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029) where (cid:1848)(cid:3029) is the volume of the blister and maintaining its adherence to the surrounding substrate. If the charging pressure is greater than depends on whether the membrane has delaminated or not. During our experiments, we use an AFM to measure the shape of the graphene membrane during each stage of the deformation described above. From full-field measurements of the membrane, 3 we extract the maximum deflection, (cid:2012) , and the blister radius (cid:1853). Initially the radius (cid:1853) is equal to the cavity radius (cid:1853)(cid:2868) , and then becomes larger than (cid:1853)(cid:2868) due to membrane delamination. 3. Analysis of the Blister Test We model the blister/cavity/substrate configuration as a thermodynamic system with the goal of developing relations among the system parameters (geometry, loading, elastic properties, and the membrane-substrate interface adhesion energy). Our approach is to determine free energy of the thermodynamic system by modelling the gas as ideal and adopting a nonlinear membrane model to describe the deformation of the membrane. We then calculate minimum energy configurations as a function of system parameters and study their stability. In the following we describe the details of this process. Of course our work is related to many other studies of graphene membranes specifically, and membranes more generally and we note specifically that of Yue et al. (2012) that analyzes similar blister configurations and studies the effect of the approximations made in the membrane mechanics, Wan and Mai (1995) who to the best of our knowledge first proposed the blister test with a trapped mass of gas, and Gent and Lewandowski (1987) who analyzed delamination in the constant pressure loading case. position; before delamination the radial boundary is located at (cid:1870) (cid:3404) (cid:1853)(cid:3042) and afterwards it is at (cid:1870) (cid:3404) (cid:1853) with (cid:1853) (cid:3408) (cid:1853)(cid:3042) . The mechanical behavior of thin structures can be described by Foppl-von Mechanics of Pressurized Blisters We model the bulged graphene blister as an axisymmetric thin structure clamped at a radial culminates in a relation between the maximum deflection (cid:2012) , pressure difference across the Karman (FvK) plate equations which include contributions from both bending and stretching. membrane (cid:1868) (cid:3404) (cid:1868)(cid:3036) (cid:3398) (cid:1868)(cid:3032) and the radius of the pressurized circular region (cid:1853): For the graphene blisters considered here, we assume that the bending rigidity is negligible and adopt the series solution of the simplified FvK equations obtained by Hencky (1915) that (cid:2012) (cid:3404) (cid:1829)(cid:2870) (cid:4678)(cid:1868)(cid:1853)(cid:2872)(cid:1831)(cid:1872) (cid:4679)(cid:2869)(cid:2871) The volume (cid:1848)(cid:3029) under the bulge is given by: (1)  (cid:1848)(cid:3029) (cid:3404) (cid:1829)(cid:2869)(cid:2024)(cid:1853)(cid:2870)(cid:2012) Here (cid:1829)(cid:2869) and (cid:1829)(cid:2870) are constants dependent on the Poisson’s ratio (it is well-known that (cid:1829)(cid:2869) and (cid:1829)(cid:2870) (2)  have errors in Hencky’s paper, see, e.g., Williams (1987) and Wan and Mai (1995) for corrected versions); we use C1 = 0.524 and C2 = 0.687, consistent with  = 0.16. Hencky’s solution is formally for the case of a uniformly-distributed load on the membrane which simplifies the analysis. Fichter (1997) treated the case of a uniform pressure load on the membrane which is membrane. Campbell (1956) extended Hencky’s solution to cases with an initial tension (cid:1840)(cid:2868) (cid:3405) 0 more complicated, but still analytically tractable. For the scenarios considered in this paper, the difference between the uniform load and uniform pressure are small and we neglect them. and showed that when the non-dimensional parameter (cid:1842) (cid:3404) (cid:3043)(cid:3028)(cid:3006)(cid:3047) (cid:4672)(cid:3006)(cid:3047)(cid:3015)(cid:3116)(cid:4673)(cid:3119)(cid:3118) (cid:3408) 100, the deflection given Furthermore, Hencky’s solution does not consider the effects of initial stress in the bulged by eq. (1) is within 5% of the solution with (cid:1840)(cid:2868) taken into account. Mechanically exfoliated 4 graphene blisters like the ones of our study often have an initial tension, (cid:1840)(cid:2868) between 0.03-0.15 N/m (e.g., Wang et al., 2012; Bunch et al., 2008; Barton et al., 2011). With typical values of (cid:1853) = 2 μm, (cid:1831)(cid:1872) = 340 N/m (Lee et al., 2008) and (cid:1840)(cid:2868)=0.07 N/m, the non-dimensional parameter (cid:1842) is experiment are done well above 500 kPa, hence we neglect the effect of (cid:1840)(cid:2868) and use Hencky’s incorporation of (cid:1840)(cid:2868) is straightforward in practice. about 100 when the pressure load is about 500 kPa. The majority of measurements in our solution to completely describe the mechanics of the pressurized blisters. Nevertheless, the inside and outside the cavity equal to (cid:1868)(cid:2868) (Fig. 1a). The pressure outside the cavity is then Thermodynamic Model of the Blister Configuration reduced to (cid:1868)(cid:3032) which causes the membrane to deform due to the pressure difference across it We model the behavior of the blister considering the three stages identified in Fig. 1. Initially (cid:1868) (cid:3404) (cid:1868)(cid:3036) (cid:3398) (cid:1868)(cid:3032) . The gas inside the cavity is assumed to isothermally expand to its final equilibrium the system is at equilibrium with the graphene membrane flat and stress free and the pressure pressure (cid:1868)(cid:3036) . Depending on the magnitude of (cid:1868)(cid:3032) , one of two configurations will arise: i) the solution and parameterize the deformed shape by the radius (cid:1853) and maximum deflection (cid:2012) ; in the former (cid:1853) (cid:3404) (cid:1853)(cid:2868) and in the latter (cid:1853) (cid:3408) (cid:1853)(cid:2868) . membrane will bulge, but not delaminate (Fig. 1b), or ii) the membrane will both bulge and delaminate (Fig. 1c). In both cases we describe the membrane mechanics using the Hencky Our strategy is to determine equilibrium configurations of the deformed membrane by seeking (cid:1832) (cid:3404) (cid:1832)(cid:3040)(cid:3032)(cid:3040) (cid:3397) (cid:1832)(cid:3034)(cid:3028)(cid:3046) (cid:3397) (cid:1832)(cid:3032)(cid:3051)(cid:3047) (cid:3397) (cid:1832)(cid:3028)(cid:3031)(cid:3035) minima in the system free energy F. To this end we recognize that the change in free energy of the system can be expressed as: In eq. (3), (cid:1832)(cid:3040)(cid:3032)(cid:3040) is the strain energy stored in the membrane as it deforms when subjected to a pressure difference across it (cid:1868), (cid:1832)(cid:3034)(cid:3028)(cid:3046) is the free energy change associated with expansion of the N (3)  gas molecules in the microchamber, (cid:1832)(cid:3032)(cid:3051)(cid:3047) is the free energy change of the external environment that is held at a constant pressure (cid:1868)(cid:3032) , and (cid:1832)(cid:3028)(cid:3031)(cid:3035) is the adhesion energy of the membrane-substrate For a fixed (cid:1853), we can compute (cid:1832)(cid:3040)(cid:3032)(cid:3040) assuming quasistatic expansion of the gas and using the interface. (cid:1832)(cid:3040)(cid:3032)(cid:3040) (cid:3404) ∬ (cid:1840)(cid:3036) (cid:1856)(cid:2035)(cid:3036) (cid:1856)(cid:1827)(cid:3040)(cid:3032)(cid:3040) (cid:3404) (cid:1868)(cid:1848)(cid:3029)4 relations from eqs. (1) and (2): where (cid:1840)(cid:3036) is the membrane force resultant, (cid:2035)(cid:3036) is the associated strain, and (cid:1856)(cid:1827)(cid:3040)(cid:3032)(cid:3040) is an (4)  the microchamber, from an initial pressure and volume (cid:4666)(cid:1868)(cid:2868) , (cid:1848)(cid:2868) (cid:4667) to final pressure and volume infinitesimal element of membrane cross sectional area. (cid:4666)(cid:1868)(cid:3036) , (cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029) (cid:4667) is: The free energy change due to isothermal expansion of the fixed number of gas molecules N in 5 (cid:1832)(cid:3034)(cid:3028)(cid:3046) (cid:3404) (cid:3398)(cid:1516) (cid:1842)(cid:1856)(cid:1848) (cid:3404) (cid:3398)(cid:1868)(cid:2868)(cid:1848)(cid:2868) ln (cid:3428)(cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029)(cid:1848)(cid:2868) (cid:3432) As the blister expands by (cid:1848)(cid:3029) , the volume of the surroundings decreases by an equal amount (5)  constant pressure (cid:1868)(cid:3032) , the free energy then changes by: (cid:1832)(cid:3032)(cid:3051)(cid:3047) (cid:3404) (cid:1516) (cid:1868)(cid:3032) (cid:1856)(cid:1848) (cid:3404) (cid:1868)(cid:3032)(cid:1848)(cid:3029) (assuming no volume change of the membrane). Assuming the surroundings are maintained at a For a constant value of adhesion energy per unit area , (cid:1832)(cid:3028)(cid:3031)(cid:3035) is then: (6)  (cid:1832)(cid:3028)(cid:3031)(cid:3035) (cid:3404) (cid:1516) (cid:1985)(cid:1856)(cid:1827) (cid:3404) (cid:1985)(cid:2024)(cid:4666)(cid:1853)(cid:2870) (cid:3398) (cid:1853)(cid:2868)(cid:2870) (cid:4667) Equations (4) – (7) show that the system energetics are described by three unknowns: (cid:1868)(cid:3036) , (cid:2012) and (cid:1853). The constitutive eq. (1) along with the ideal gas equation (cid:1868)(cid:2868)(cid:1848)(cid:2868) (cid:3404) (cid:1868)(cid:3036) (cid:4666)(cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029) (cid:4667) provides two (7)  single unknown (cid:1853): (cid:1832)(cid:4666)(cid:1853)(cid:4667) (cid:3404) (cid:1868)(cid:1848)(cid:3029)4 (cid:3398) (cid:1868)(cid:2868)(cid:1848)(cid:2868) ln (cid:3428)(cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029)(cid:1848)(cid:2868) relations between these three unknowns; we use these to express the free energy in terms of the (cid:3432) (cid:3397) (cid:1868)(cid:3032) (cid:1848)(cid:3029) (cid:3397) (cid:1985)(cid:2024)(cid:4666)(cid:1853)(cid:2870) (cid:3398) (cid:1853)(cid:2868)(cid:2870)(cid:4667) Recall that (cid:1848)(cid:3029) is a function of (cid:1853) as given by eqs. (1) and (2). We determine equilibrium (8)  configurations by computing extrema of (cid:1832)(cid:4666)(cid:1853)(cid:4667): (cid:1856)(cid:1832)(cid:4666)(cid:1853)(cid:4667)(cid:1856)(cid:1853) (cid:3404) 0 When there is no delamination ((cid:1853) (cid:3404) (cid:1853)(cid:2868) (cid:4667), the equilibrium solution is obtained simply from eqs. (9)  (cid:1856)(cid:1832)(cid:4666)(cid:1853)(cid:4667)(cid:1856)(cid:1853) (cid:3404) (cid:3398) 3(cid:1868)4 (cid:1856)(cid:1848)(cid:3029)(cid:1856)(cid:1853) (cid:3397) (cid:1848)(cid:3029)4 (cid:1856)(cid:1868)(cid:1856)(cid:1853) (cid:3397) 2(cid:2024)(cid:1985)(cid:1853) (cid:3404) 0 (1) and (2) along with the ideal gas equation. When there is delamination, the equilibrium configuration obtained by solving eq. (9) can be expressed as: Here (cid:1868) depends on (cid:1853) through the relation obtained from eq. (1) and ideal gas equation:   (10)  (cid:1853) (cid:3404) (cid:3436)(cid:1868)(cid:2868)(cid:1868)(cid:3036) (cid:3398) 1(cid:3440) (cid:2871)(cid:2869)(cid:2868) (cid:3436) (cid:1848)(cid:2868)(cid:2024)(cid:1829)(cid:2869)(cid:1829)(cid:2870)(cid:3440) (cid:2871)(cid:2869)(cid:2868) (cid:3436)(cid:1831)(cid:1872)(cid:1868) (cid:3440) (cid:2869)(cid:2869)(cid:2868) Using eqs. (1) and (2), we can write: (11)  (12) 6 (13)  (cid:1856)(cid:1848)(cid:3029)(cid:1856)(cid:1853) (cid:3404) (cid:2034)(cid:1848)(cid:3029)(cid:2034)(cid:1868) │(cid:3028) (cid:2034)(cid:1868)(cid:2034)(cid:1853) (cid:3397) (cid:2034)(cid:1848)(cid:3029)(cid:2034)(cid:1853) │(cid:3043) (cid:3404) 13 (cid:1848)(cid:3029)(cid:1868) (cid:2034)(cid:1868)(cid:2034)(cid:1853) │(cid:3028) (cid:3397) (cid:2034)(cid:1848)(cid:3029)(cid:2034)(cid:1853) │(cid:3043) (cid:1856)(cid:2272)(cid:4666)(cid:1853)(cid:4667)(cid:1856)(cid:1853) (cid:3404) (cid:3398) 3(cid:1868)4 (cid:2034)(cid:1848)(cid:3029)(cid:2034)(cid:1853) │(cid:3043) (cid:3397) 2(cid:2024)(cid:1985)(cid:1853) (cid:3404) 0 Substituting eq. (12) into eq. (10) results in the relation: (cid:1985) (cid:3404) 5(cid:1829)(cid:2869)4 (cid:3436) (cid:1868)(cid:2868)(cid:1848)(cid:2868) (cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029) (cid:4666)(cid:1853)(cid:4667) (cid:3398) (cid:1868)(cid:3032) (cid:3440) (cid:2012)(cid:4666)(cid:1853)(cid:4667) Rearranging and using ideal gas equation, we finally obtain: (14)  ((cid:1868)(cid:2868) , (cid:1868)(cid:3032) , (cid:1860), (cid:1853), (cid:2012) , Γ). We use eq. (14) with typical experiments, to determine (cid:1985) with prescribed values of (cid:1868)(cid:2868) and (cid:1868)(cid:3032) , ((cid:1853), (cid:2012) ) pairs measured with an atomic force microscope, (cid:1848)(cid:2868) (cid:3404) (cid:2024)(cid:1853)(cid:2868)(cid:2870)(cid:1860) determined by the device geometry and (cid:1848)(cid:3029) (cid:4666)(cid:1853)(cid:4667) given by eq. (2). terms of system parameters in (14) describes equilibrium configurations Equation In an experiment if we systematically increase (cid:1868)(cid:2868) , we find that at a critical value, the membrane will begin to delaminate. We determine (cid:1868)(cid:3030)(cid:3045) by substituting (cid:1853) (cid:3404) (cid:1853)(cid:2868) in eq. (14) and solving for (cid:1868)(cid:2868) : 4(cid:1985) 5(cid:1829)1(cid:2012)(cid:4666)(cid:1853)0(cid:4667)(cid:4679) (cid:3397) (cid:1868)(cid:1857)(cid:4679) (cid:1848)0 (cid:3397) (cid:1848)(cid:1854) (cid:4666)(cid:1853)0 (cid:4667) (cid:1868)(cid:1855)(cid:1870) (cid:3404) (cid:4678)(cid:4678) (cid:1848)0 (15)  In eq. (15), as (cid:1848)(cid:2868) → ∞, (cid:3023)(cid:3116)(cid:2878)(cid:3023)(cid:3277)(cid:4666)(cid:3028)(cid:3116)(cid:4667) → 1 and we can express eq. (15) as: (cid:3023)(cid:3116) (cid:1985) (cid:3404) 5(cid:1829)(cid:2869)4 (cid:4666)(cid:1868)(cid:3030)(cid:3045) (cid:3398) (cid:1868)(cid:3032) (cid:4667)(cid:2012) (cid:4666)(cid:1853)(cid:2868)(cid:4667) (16)  (cid:1848)(cid:2868) → ∞, the isothermal expansion approaches a constant pressure process; hence the constant This agrees with the constant–pressure result obtained by Williams (1997). In essence, as pressure blister configuration results as a limiting case of the constant N blister configuration as the cavity size becomes large. (cid:1856)(cid:2870)(cid:1832)(cid:1856)(cid:1853)(cid:2870) (cid:3404) 10(cid:1868)(cid:1848)(cid:3029)(cid:1853)(cid:2870) (cid:4678)2(cid:1868)(cid:2868)(cid:1868)(cid:3036) (cid:3398) 3(cid:1868)(cid:3036)(cid:2870) (cid:3397) (cid:1868)(cid:2868)(cid:1868)(cid:3032) 3(cid:1868)(cid:2868)(cid:1868) (cid:3397) (cid:1868)(cid:3036) (cid:4666)(cid:1868)(cid:2868) (cid:3398) (cid:1868)(cid:3036) (cid:4667) (cid:4679) Finally, we evaluate the stability of the system by computing:   (17)  7 If (cid:3031)(cid:3118)(cid:3007)(cid:3031)(cid:3028)(cid:3118) (cid:3408) 0 the delamination will be stable. Assuming (cid:1868)(cid:3032) ≪ (cid:1868)(cid:3036) , (cid:1868)(cid:2868) (which is the case in our experiments) then we require (cid:1868)(cid:3036) (cid:3407) 2(cid:1868)(cid:2868)/3 for stable delamination. This inequality is equivalent to requiring (cid:1848)(cid:2868) (cid:3407) 2(cid:1848)(cid:3029) which can be satisfied experimentally by tailoring the geometry of the microcavity. 4. Results and Discussion In this section we have three goals: i) to demonstrate the behaviour of the blister system, ii) to use the blister analysis in conjunction with experiments to determine the adhesion energy of graphene-SiO2 interfaces, and iii) to show how the model describes measurements of monolayer graphene blisters in the constant N experimental configurations. Previously (Koenig et al., 2011) we used this blister test to determine the elastic moduli (Et) of graphene monolayers and multilayers, but this required more measurements in the elastic regime before delamination than we made here. Since our emphasis here is on the adhesion energy, we did not make as many measurements in the elastic regime and instead used our previous measured modulus results as inputs to our calculations. System Behavior: Equilibrium Configurations and Stability As mentioned earlier, we obtain equilibrium configurations of the blister system by solving eq. (14) and its stability is described by eq. (17). In general these are implicit equations involving cavity radius (cid:1853)(cid:2868) and cavity depth (cid:1860): the system parameters, but explicit relations in general are elusive or not particularly revealing so here we describe three specific examples by which we intend to demonstrate the rich behavior of Case 1 (cid:1853)(cid:2868) = 2 μm and (cid:1860) = 0.25 μm the system for experimentally-accessible system parameters. For each case we prescribe the Case 2 (cid:1853)(cid:2868) = 3 μm and (cid:1860) = 0.25 μm Case 3 (cid:1853)(cid:2868) = 2 μm and (cid:1860) = 1.25 μm with existing measurements and theory (cid:1831)(cid:1872) = 340 N/m, (cid:2021) = 0.16, (Blakslee et al., 1970) and we take (cid:1985) = 0.2 J/m2. For each case we take the membrane to be a graphene monolayer with elastic properties in line The system in Case 1 has an initial volume (cid:1848)(cid:2868) (cid:3404) (cid:2024)(cid:1853)(cid:3042)(cid:2870)(cid:1860) (cid:3406)3.14 μm(cid:2871) . This geometry is similar to pressure for delamination (cid:1868)(cid:3030)(cid:3045) = 1.94 MPa. The free energy of eq. (8) is plotted as a function of the experimental devices used in our study. From eq. (14) we calculate the critical charging equilibrium configuration where (cid:1856)(cid:1832)/(cid:1856)(cid:1853) (cid:3404) 0 is satisfied respectively. The dashed part of each curve corresponds to (cid:1853) (cid:3407) (cid:1853)(cid:2868) which is physically not realizable. When (cid:1868)(cid:2868) (cid:3407) (cid:1868)(cid:3030)(cid:3045) (blue curve), the blister radius at three different input/charging pressures as shown in Fig. 2a. The green and magenta colored points on the curves signify the initial configuration of the system and the final be no delamination and (cid:1853) remains equal to (cid:1853)(cid:2868) . When (cid:1868)(cid:2868) (cid:3404) (cid:1868)(cid:3030)(cid:3045) (black curve), the system finds an equilibrium configuration exactly at (cid:1853) (cid:3404) (cid:1853)(cid:2868) , an inflection point. If (cid:1868)(cid:2868) is increased to a value beyond (cid:1868)(cid:3030)(cid:3045) there is no configuration with free energy less than the initial configuration, implying there will configurations –a local maximum with (cid:1853) (cid:3407) (cid:1853)(cid:2868) (not identified with a symbol and unrealizable) and a local minimum with (cid:1853) (cid:3408) (cid:1853)(cid:2868) which is evident from the red curve in the Fig. 2a. The this unique equilibrium configuration degenerates into two equilibrium presence of this minimum makes the stable delamination possible in the constant N blister test. 8 From the equilibrium configurations as a function of charging pressure ((cid:1868)(cid:2868) ), we obtain various charging pressure: maximum blister deflection ((cid:2012) ), blister radius ((cid:1853)), and cavity pressure ((cid:1868)(cid:3036) ). As representations of the system behavior. Figure 3a-c shows three quantities as a function of the 3a. At (cid:1868)(cid:2868) = (cid:1868)(cid:3030)(cid:3045) = 1.94 MPa, delamination begins and as (cid:1868)(cid:2868) continues to increase the blister the charging pressure is increased, the graphene blister deflection increases, and the membrane stiffens resulting in the nonlinear behavior given by eq. (1) and shown by the black curve in Fig. Figure 3c shows the evolution of the cavity pressure (cid:1868)(cid:3036) with increasing (cid:1868)(cid:2868) . Before delamination, (cid:1868)(cid:3036) increases nearly linearly with (cid:1868)(cid:2868) ; the gentle softening of the curve results because as the continues to delaminate and the deflection increases as given by eq. (14). This is shown by the red line in Figs. 3a and 3b, the latter showing the blister radius after the onset of delamination. pressure decreases consistent with the ideal gas law. After delamination, (cid:1868)(cid:3036) decreases rapidly with increasing (cid:1868)(cid:2868) because the volume increases at higher rate than before delamination, thereby decreasing the equilibrium pressure. Formally, as (cid:1868)(cid:2868) → ∞, (cid:1868)(cid:3036) → (cid:1868)(cid:3032) . blister volume increases with a constant number of gas molecules trapped in the cavity, the In the system of Case 2, the radius of the cavity is increased from (cid:1853)(cid:2868) = 2 μm to 3 μm. In this from 1.94 MPa to 1.57 MPa. From the (cid:1832)(cid:4666)(cid:1853)(cid:4667) plots in Fig. 2b, at the critical charging pressure the case, the membrane system is more compliant and, as a result, the critical pressure is lowered equilibrium now occurs at a minimum rather than at an inflection point. However, this subtle Finally in Case 3, we increase the cavity depth (cid:1860) = 0.25 to 1.25 μm while keeping the cavity difference from Case 1 does not qualitatively change the system behavior; it behaves similar to radius at (cid:1853)(cid:2868) = 2 μm. The critical charging pressure is again decreased from the original 1.94 that of Case 1 (Fig 3a-c) and so we do not show plots. MPa to 1.39 MPa. The plot of (cid:1832)(cid:4666)(cid:1853)(cid:4667) in Fig. 2c shows that now when (cid:1868)(cid:2868) (cid:3407) (cid:1868)(cid:3030)(cid:3045) (blue curve), the overcome to reach the minimum energy delaminated configuration. When (cid:1868)(cid:2868) (cid:3404) (cid:1868)(cid:3030)(cid:3045) (black curve has two possible extrema instead of none as in the previous two cases. When the system starts in the prescribed initial configuration, an energy barrier has to be the minimum energy delaminated configuration with (cid:1853) (cid:3408) (cid:1853)(cid:2868) . Therefore, when the charging curve), however, the barrier is removed and the initial configuration coincides with a local with a rapid advance in the membrane radius (cid:1853). This also results in a discontinuity in the maximum. This is an unstable equilibrium and with a small perturbation the system can move to pressure is increased beyond the critical pressure (1.39 MPa), delamination can occur suddenly equilibrium system parameters as illustrated in Fig. 3d-f. Such a discontinuity is in contrast to the previous two cases where delamination progresses in a stable manner as the charging pressure can be an inflection point, a local minimum, or a local maximum of (cid:1832)(cid:4666)(cid:1853)(cid:4667). What this pressure is increased. In summary, these case studies show that the equilibrium configuration at the critical charging an unstable equilibrium there can be a jump in the observable/measured quantities (cid:1853), (cid:2012) , and (cid:1868)(cid:3036) . suggests for experiments is that in the first two cases the blister radius and deflection will evolve as a steady, continuous change from the initial values as the membrane starts delaminating, and similarly the cavity pressure will decrease. In Case 3, however, because the initial condition is 9 Looking more closely at the behavior we find that as (cid:1860) is increased at a fixed (cid:1853)(cid:2868) , the initial volume (cid:1848)(cid:2868) can become much larger than the volume of the membrane blister (cid:1848)(cid:3029) . From the ideal gas law for isothermal conditions, (cid:1868)(cid:3036) (cid:3404) (cid:1868)(cid:2868)(cid:1848)(cid:2868)/(cid:4666)(cid:1848)(cid:2868) (cid:3397) (cid:1848)(cid:3029) (cid:4667), we see that the pressure (cid:1868)(cid:3036) approaches the charging pressure (cid:1868)(cid:2868) when (cid:1848)(cid:3029) ≪ (cid:1848)(cid:2868) . It is well-known that in a constant pressure (P) blister proceeds, the blister volume (cid:1848)(cid:3029) increases and eventually becomes comparable to (cid:1848)(cid:2868) . This leads test, delamination is unstable (Gent and Lewandowski, 1987), i.e., once the critical pressure is reached, the entire adhered membrane delaminates. Therefore, for large cavity depths, membrane delamination may initiate in an unstable manner. However, as delamination to a significant decrease in the cavity pressure and a stable equilibrium is then approached. To further illustrate the connection between the constant N and constant P blister tests, we plot the critical pressure versus the cavity depth in Fig. 4a, and see that the constant N blister test curve asymptotically approaches the constant P blister test value which is independent of the cavity depth. Also, the critical pressure as a function of the cavity radius and the adhesion energy is shown in Figs. 4b and 4c, respectively. As the adhesion energy is increased, the rapidly decreases initially with increasing (cid:1853)(cid:2868) but reverses this trend after reaching a minimum critical delamination pressure increases as expected in both constant P and constant N blister tests. While with increasing cavity radius and a fixed cavity depth, the delamination pressure decreases rapidly and continuously in the constant P case whereas in the constant N case it We can determine the adhesion energy (cid:1985) between the graphene membrane and the substrate value. using the measured deflection (cid:2012)and radius (cid:1853) of the equilibrated blister membrane after Combining the Model and Measurements to Determine Adhesion Energy delamination (the red symbols in Fig. 7) and the prescribed charging pressure (cid:1868)(cid:2868) , we can calculate the adhesion energy (cid:1985) from eq. (14). In Figure 5 we plot results obtained in this (SiO2 in our case) by combining the theory and the experimental measurements. Specifically, average adhesion energy of (cid:1985) = 0.44 J/m2. The results shown for Chip 2 are new measurements and show a lower value of (cid:1985) = 0.24 J/m2. The data for both chips are self-consistent, suggesting manner for two different sets of monolayer graphene blisters fabricated on two different chips. The results for Chip 1 are our previously-reported values (Koenig et al., 2011) and show an that the difference is not due to errors in measurements but that it reflects the actual difference in the operant surface forces on the two chips. This in turn could arise from differences in surface properties such as roughness and chemical reactivity, and thus change the apparent adhesion energy. Although the exact cause of the variation in adhesion energies remains to be elucidated with more experimental efforts, these results demonstrate the usefulness of the constant N blister test to determine adhesion energy. Blister System Behavior – Measurements and Theory Here we compare measurements of monolayer graphene membranes and theory. As mentioned, we used an AFM to measure the deformation of graphene blisters in the constant N configuration. In our measurements we estimate the resolution in blister height to be subnanometerand that in blister radius to be about 90 nm. Figure 6a shows a representative three dimensional profile of a bulged monolayer graphene blister (from Chip 2) and confirms the axisymmetric deformation of the membrane. In Fig. 6b we plot the cross-section of the 10 membrane profile for various values of the prescribed charging pressure (cid:1868)(cid:2868) . When (cid:1868)(cid:2868) is below 1.32 MPa, the graphene membrane remains attached to the edge of the cavity but as (cid:1868)(cid:2868) increases the graphene membrane delaminates from the substrate, resulting in a larger radius as shown in Fig. 6b. Also plotted in Fig. 6b are theoretical fits of membrane profiles according to Hencky’s solution, with the maximum deflection (eq. (1)) fit to the measurements. The Hencky solution, with the measured maximum deflection as a fitting parameter, is in excellent agreement with the maximum deflection (cid:2012) , blister radius (cid:1853) and the calculated equilibrium cavity pressure (cid:1868)(cid:3036) versus measurements, both in terms of the shape of the profile, but also in terms of the boundary the charging pressure (cid:1868)(cid:2868) along with theoretical predictions. The behavior is as described in Fig. conditions. This reinforces the appropriateness of using Hencky’s solution to describe the membrane mechanics in the model of the constant N blister test. In Fig. 7, we plot the measured measured value of (cid:1985) = J/m2 to illustrate the sensitivity of the measured parameters to the 3a-c but we also include plots for multiple values of the adhesion energy, centered around the adhesion energy. In Fig. 7 we show the measurements with symbols of two colors, red and black. The black symbols show results before the clear onset of delamination. The red symbols As we discussed earlier, the theory predicts that when the cavity depth (cid:1860) is large, the blister test indicate measurements after delamination has occurred and these are used to determine the adhesion energy in Fig. 5. In summary, the theory describes the measurements well. blister radius. We observed such behavior in tests with microcavities with a cavity radius (cid:1853)(cid:2868) = 2.2m and depth (cid:1860) = 5m, a geometry similar to the third example discussed above. We find that with increasing charging pressure (cid:1868)(cid:2868) , graphene membranes bulge as previously described, system may exhibit an unstable delamination with a jump in the system parameters, including the In this case, (cid:1868)(cid:2868)= 2.8 MPa was the pressure at which delamination was observed. We think that but that above a critical pressure, the membrane appears to undergo severe delamination with a resulting blister of irregular shape that is very large and covers multiple microcavities; see Fig. 8. this large blister is a consequence of the unstable delamination as predicted by theory and shown in Fig.3c. Conceivably, the membrane delaminated over a large region, neighboring blisters observation where delamination was observed at (cid:1868)(cid:2868)= 2.8 MPa, but not at a lower pressure of at coalesced, and the result is a large irregular shaped blister. Assuming the adhesion energy is (cid:1868)(cid:2868)= 2.2 MPa. We did not do tests at pressures between these two values. between 0.2-0.4 J/m2 and graphene is 8 layered, the predicted critical input pressure for delamination is between 1.90-3.15 MPa. This is in reasonable agreement with the experimental 5. Conclusions We studied the mechanics of a graphene membrane adhered to a substrate patterned with etched microcavities of a prescribed volume that trap a fixed number of gas molecules. By lowering the ambient pressure, and thus changing the pressure difference across the graphene membrane, the membrane can be made to bulge and delaminate in a stable manner from the substrate. We analyzed the membrane/substrate as a thermodynamic system and studied the behavior of this the constant N blister test over representative experimentally-accessible geometry and loading parameters. We found that depending on the system parameters, the membrane will deform in a nonlinear elastic manner until a critical charging pressure is reached. At that point, the membrane will delaminate from the substrate in a stable manner. We carried out companion experiments of the membrane deformation as the charging pressure was increased and used them 11 with the theory to determine the adhesion energy of graphene/SiO2 interfaces. We found an average adhesion energy that is lower, but in line with previously reported values by us and others. We also showed that the theoretical predictions described the experiments well, both before and after stable delamination. For deep cavities, the membrane can delaminate in an unstable manner and we demonstrated this experimentally. Although we did not study the nature of the surface forces that influence the adhesion energy, the constant N blister test is an attractive approach to enable the study of important effects on adhesion including substrate topography, membrane stiffness, and the surface force law. 12 6. References Aitken, Z. H., and Huang, R., 2010, “Effects of mismatch strain and substrate surface corrugation on morphology of supported monolayer graphene,” Journal of Applied Physics, 107(12), 123531. Barton, R. A., Ilic, B., van der Zande, A. M., Whitney, W. S., McEuen, P. L., Parpia, J. M., and Craighead, H. G., 2011, “High, Size-Dependent Quality Factor in an Array of Graphene Mechanical Resonators,” Nano Letters, 11(3), pp. 1232-1236. Blakslee, O. L., Proctor, D. G., Seldin, E. J., Spence, G. B., and Weng, T., 1970, “Elastic constants of compression-annealed pyrolytic graphite,” J. Appl. Phys., 41, pp. 3373–3382. Bunch, J. S., Verbridge, S. S., Alden, J.S., van der Zande, A. M., Parpia, J. M., Craighead, H. G., and McEuen, P. L., 2008, “Impermeable Atomic Membranes from Graphene Sheets," Nano Letters, 8, pp. 2458-2462. Bunch, J. S., van der Zande, A. M., Verbridge, S. S., Frank, I. W., Tanenbaum, D. M., Parpia, J. M., Craighead, H. G., and McEuen, P. L., 2007, "Electromechanical Resonators from Graphene Sheets," Science, 315, pp. 490-493. Campbell, J. D., 1956, “On the theory of initially tensioned circular membranes subjected to uniform pressure,” Q J Mechanics Appl Math, 9(1), pp. 84-93. Chen, S., Brown, L., Levendorf, M., Cai, W., Ju, S-Y., Edgeworth, J., Li, X., Magnuson, C. W., Velamakanni, A., Piner, R. D., Kang, J., Park, J., and Ruoff, R. S., 2011, “Oxidation Resistance of Graphene-Coated Cu and Cu/Ni Alloy,” ACS Nano, 5(2), pp. 1321-1327. DelRio, F., Dunn, M. L., and, de Boer, M. P., 2008, “Capillary Adhesion Model for Contacting Micromachined Surfaces,” Scripta Materialia, 59, Viewpoint Set No. 44, pp.916-920. DelRio, F., Dunn, M. L., Phinney, L. M., Bourdon, C. J., and, de Boer, M. P., 2007, “Rough Surface Adhesion in the Presence of Capillary Condensation,” Applied Physics Letters, 90, 163104. El-Kady, M. F., Strong, V., Dubin, S., and Kaner, R. B., 2012, “Laser Scribing of High- Performance and Flexible Graphene-Based Electrochemical Capacitors,” Science, 335 (6074), pp. 1326-1330. Fichter, W. B., 1997, “Some solutions for the large deflections of uniformly loaded circular membranes,” NASA Technical Paper 3658, Langley Research Center, Hampton, Virginia. Geim, A. K., 2009, “Graphene: Status and Prospects,” Science, 324(5934), pp. 1530-1534. Gent, A. N., and Lewandowski, L. H., 1987, “Blow-off pressures for adhering layers,” Journal of Applied Polymer Science, 33(5), pp. 1567-1577. Georgiou, T., Britnell, L., Blake, P., Gorbachev, R. V., Gholinia, A., Geim, A. K., Casiraghi, C., and Novoselov, K. S., 2011, “Graphene bubbles with controllable curvature,”, Appl. Phys. Lett., 99, 093103. Hencky, H., 1915, “Über den spannungszustand in kreisrunden platten mit verschwindender biegungssteiflgkeit,” Z. Fur Mathematik Und Physik, 63, pp. 311-317. Jiang, L. Y., Huang, Y., Jiang, H., Ravichandran, G., Gao, H., Hwang, K. C., and Liu, B., 2006, “A Cohesive Law for Carbon Nanotube/polymer Interfaces Based on the Van Der Waals Force,” J. Mech. Phys. Solids, 54, pp. 2436-2452. Kim, K., Lee, Z., Malone, B. D., Chan, K. T., Alemán, B., Regan, W., Gannett, W., Crommie, M. F., Cohen, M. L., and Zettl, A., 2011, “ Multiply folded graphene,” Phys. Rev. B, 83, 245433. Koenig, S. P., Boddeti, N. G., Dunn, M. L., Bunch, J. S., 2011, “Ultrastrong adhesion of graphene membranes,” Nature Nanotechnology, 6, pp. 543-546. 13 Lee, C. G., Wei, X. D., Kysar, J. W., and Hone, J., 2008, "Measurement of the Elastic Properties and Intrinsic Strength of Monolayer Graphene," Science, 321, pp. 385-388. Levy, N., Burke, S. A., Meaker, K. L., Panlasigui, M., Zettl, A., Guinea, F., Castro Neto, A. H., and Crommie, M. F., 2010, “Strain-Induced Pseudo–Magnetic Fields Greater Than 300 Tesla in Graphene Nanobubbles,” Science, 329(5991), pp. 544-547. Li, T., and Zhang, Z., 2010, “Substrate-regulated morphology of graphene,” J. Phys. D: Appl. Phys., 43, 075303. Lin, Y.-M., Dimitrakopoulos, C., Jenkins, K. A., Farmer, D. B., Chiu, H.-Y., Grill, A., and Avouris, Ph., 2010, “100-GHz Transistors from Wafer-Scale Epitaxial Graphene,” Science, 327(5966), pp. 662. Low, T., Perebeinos, V., Tersoff, J., and Avouris, P., 2012, “Deformation and Scattering in Graphene over Substrate Steps,” Phys. Rev. Lett., 108(9), 096601. Lu, Z. and Dunn, M. L., 2010, “van der Waals Adhesion of Graphene Membranes,” J. Appl. Phys., 96, 111902. Meyer, J. C., Geim, A. K., Katsnelson, M. I., Novoselov, K. S., Booth, T. J., and Roth, S., 2007, “The structure of suspended graphene sheets,” Nature, 446, pp. 60-63. Meyer, J. C., Kisielowski, C., Erni, R., Rossell, M. D., Crommie, M. F., and Zettl, A., 2008, “Direct Imaging of Lattice Atoms and Topological Defects in Graphene Membranes,” Nano Lett., 8(11), pp. 3582-3586. Pan, W., Xiao, J., Zhu, J., Yu, C., Zhang, G., Ni, Z., Watanabe, K., Taniguchi, T., Shi, Y. and Wang, X., 2012, “.Biaxial Compressive Strain Engineering in Graphene/Boron Nitride Heterostructures,” Scientific Reports, 2, 893. Scharfenberg, S., Rocklin, D. Z., Chialvo, C., Weaver, R. L., Goldbart, P. M., and Mason, N., 2011, “Probing the mechanical properties of graphene using a corrugated elastic substrate,” Appl. Phys. Lett., 98(9), 091908. Springman, R. M., and Bassani, J. L., 2008, “Snap Transitions in Adhesion,” J. Mech. Phys. Solids., 56, pp. 2358-2380. Tang, T., Jagota, A., and Hui, C. Y., 2005, “Adhesion between Single-walled Carbon Nanotubes,” J. Appl. Phys., 97, 074304. Wan, K.-T., and Mai, Y.-W., 1995, “Fracture mechanics of a new blister test with stable crack growth,” Acta Metallurgica et Materialia, 43(11), pp. 4109-4115. Wang, L., Travis, J. J., Cavanagh, A. S., Liu, X., Koenig, S. P., Huang, P. Y., George, S. M., and Bunch, J. S., 2012, “Ultrathin Oxide Films by Atomic Layer Deposition on Graphene,” Nano Letters, 12(7), pp. 3706-3710. Williams, J. G., 1997, “Energy Release Rates for the Peeling of Flexible Membranes and the Analysis of Blister Tests,” Int. J. Fracture, 87, pp. 265-288. Yue, K., Gao, W., Huang, R., and Liechti, K. M., 2012, “Analytical methods for mechanics of graphene bubbles,”, J. Appl. Phys., 112(8), 083512. Zhu, Y., Li, L., Zhang, C., Casillas, G., Sun, Z., Yan, Z., Ruan, G., Peng, Z., Raji, A. O., Kittrell, C., Hauge, R. H., and Tour, J. M., 2012, “A seamless three-dimensional carbon nanotube graphene hybrid material,” Nature Communications, 3, 1225. Zong, Z., Chen, C.-L., Dokmeci, M. R., and Wan, K.-T., 2010, “Direct measurement of graphene adhesion on silicon surface by intercalation of nanoparticles.” J. Appl. Phys., 107(2), 026104. 14  b  Figure 1 2(cid:1853)(cid:2868) (cid:1868)(cid:2868) (cid:1868)(cid:2868) SiO2 (cid:1868)(cid:2868) (cid:3407) (cid:1868)(cid:3030)(cid:3045) a  Graphene (cid:1860) (cid:1868)(cid:2868) (cid:3408) (cid:1868)(cid:3030)(cid:3045) (cid:1868)(cid:3028) (cid:1868)(cid:3028) (cid:1868)(cid:3036) (cid:1868)(cid:3036) 2(cid:1853) (cid:2012) Schematic cross sections of test structures illustrating: (a) the initial configuration of the system, charged to a pressure p0 in a pressure chamber – the blue color indicates gas and the red curve is the graphene membrane; possible final configurations when the external pressure is reduced with graphene membranes deformed due to the expanding gas molecules (b) with; and (c) without delamination from the substrate. The change of the blue color from a darker to a lighter shade indicates decreasing pressure. c  15     ▬ (cid:1868)(cid:2868) = 1.7 MPa ▬ (cid:1868)(cid:2868) = (cid:1868)(cid:3030)(cid:3045) = 1.94 MPa ▬ (cid:1868)(cid:2868) = 2.2 MPa ▬ (cid:1868)(cid:2868) = 1.4 MPa ▬ (cid:1868)(cid:2868) = (cid:1868)(cid:3030)(cid:3045) = 1.57 MPa   ▬ (cid:1868)(cid:2868) = 1.7 MPa ▬ (cid:1868)(cid:2868) = 1.25 MPa ▬ (cid:1868)(cid:2868) = (cid:1868)(cid:3030)(cid:3045) = 1.39 MPa ▬ (cid:1868)(cid:2868) = 1.45 MPa a  b c Variation of free energy with blister radius, at a fixed pressure (cid:1868)(cid:2868) with (a) (cid:1853)(cid:2868) = 2 (cid:2020)(cid:1865) and (cid:1860) = 0.25 (cid:2020)(cid:1865), (b) (cid:1853)(cid:2868) = 3 (cid:2020)(cid:1865) and (cid:1860) = 0.25 (cid:2020)(cid:1865) and (c) (cid:1853)(cid:2868) = 2 (cid:2020)(cid:1865) and (cid:1860) = 1.25 (cid:2020)(cid:1865)     Figure 2 16 a  d  ▬   Before Delam.  ▬   After Delam.        b e c f   (a, d) Maximum deflection, (cid:2012) , (b, e) blister radius, a, and (c, f) final equilibrium microchamber pressure, (cid:1868)(cid:3036) plotted as functions of the input pressure, (cid:1868)(cid:2868) with (cid:1985) = 0.2 J/m2. The cavity dimensions   are (a-c) (cid:1853)(cid:2868) = 2 (cid:2020)(cid:1865) and (cid:1860) = 0.25 (cid:2020)(cid:1865) and (d-f) (cid:1853)(cid:2868) = 2 (cid:2020)(cid:1865) and (cid:1860) = 1.25 (cid:2020)(cid:1865)   Figure 3 17       a  b c   Figure 4 When not being varied, (cid:1860) = 400 nm, (cid:1853)(cid:2868) = 2 (cid:2020)(cid:1865), and (cid:1985) = 0.2 J/m2. Critical pressure for the onset of delamination as a function of: (a) cavity depth, (b) cavity radius and (c) adhesion energy for the constant pressure (black curves) and constant N (red curves) blister tests.   18 ∎ Chip2  ● Chip1         0.44 J/m2  0.24 J/m2  Figure 5 Adhesion energies for monolayer graphene membranes on two different SiO2 substrates/chips. The average adhesion energy is 0.44 J/m2 for Chip 1 and 0.24 J/m2 for Chip 2. 19 a  b Figure 6 (a) Three dimensional rendering of AFM height scan of a graphene blister pressurized to 2.4 MPa (Chip 2). The maximum height is about 520 nm; (b) cross sections of the AFM height measurements (Chip 2) at different input pressures, – 0.48 MPa (red), 1.32 MPa (green), 1.83 MPa (cyan) and 2.40 MPa (magenta). The black curves are the deflection profiles from Hencky’s solution, with the maximum deflection fit to the measured value. 20 a  b c delamination for different values of adhesion energy: (cid:1985) = 0.2 J/m2 (dashed blue), (cid:1985) = 0.24 J/m2, and (cid:1985) (a) Maximum deflection, (b) blister radius, and (c) final internal pressure. The black symbols are from measurements and the solid curves are from the analysis with: no delamination (red), and = 0.28 J/m2 (long dashed blue). The red symbols are those that were used to determine the adhesion energies in Fig. 5. Figure 7 21 (a) AFM amplitude image (40(cid:3400)40 (cid:2020)(cid:1865)) of a graphene membrane that has undergone large-scale delamination at (cid:1868)(cid:2868) = 2.8 MPa with (cid:1853)(cid:2868) (cid:3406) 2.2 (cid:2020)(cid:1865) and (cid:1860) (cid:3406) 5 (cid:2020)(cid:1865). Assuming the adhesion energy is between 0.2-0.4 J/m2 and the graphene has 8 layers, the critical pressure is between 1.2-1.9 MPa. Figure 8 22 

1407.6345

4

1407

Topological Yu-Shiba-Rusinov chain from spin-orbit coupling

We investigate the possibility of realizing a topological state in the impurity band formed by a chain of classical spins embedded in a two-dimensional singlet superconductor with Rashba spin-orbit coupling. In contrast to similar proposals which require a helical spin texture of the impurity spins for a nontrivial topology, here we show that spin-flip correlations due to the spin-orbit coupling in the superconductor produces a topological state for ferromagnetic alignment of the impurity spins. From the Bogoliubov-de Gennes equations we derive an effective tight-binding model for the subgap states which resembles a spinless superconductor with long-range hopping and pairing terms. We evaluate the topological invariant, and show that a topologically non-trivial state is generically present in this model.

cond-mat.mes-hall

cond-mat

Topological Yu-Shiba-Rusinov chain from spin-orbit coupling P. M. R. Brydon,∗ S. Das Sarma, Hoi-Yin Hui, and Jay D. Sau Condensed Matter Theory Center and Joint Quantum Institute, University of Maryland, College Park, Maryland 20742-4111, USA (Dated: January 22, 2015) We investigate the possibility of realizing a topological state in the impurity band formed by a chain of classical spins embedded in a two-dimensional singlet superconductor with Rashba spin- orbit coupling. In contrast to similar proposals which require a helical spin texture of the impurity spins for a nontrivial topology, here we show that spin-flip correlations due to the spin-orbit coupling in the superconductor produces a topological state for ferromagnetic alignment of the impurity spins. From the Bogoliubov-de Gennes equations we derive an effective tight-binding model for the subgap states which resembles a spinless superconductor with long-range hopping and pairing terms. We evaluate the topological invariant, and show that a topologically non-trivial state is generically present in this model. PACS numbers: 74.20.-z,75.70.Tj,73.63.Nm,03.67.Lx I. INTRODUCTION turn out to be a strong detrimental factor.26 The search for a condensed-matter realization of the Majorana fermion continues, motivated both by the un- derlying fundamental physics and potential technological applications. Such states are predicted to occur in vor- tices of unconventional superconductors.1 They may also be realized as edge states of engineered spinless supercon- ductors with nontrivial topology, e.g. a Kitaev chain,2 in a superconducting heterostructure.3 Such a phase was predicted to occur in a semiconducting nanowire in prox- imity contact with a superconductor and in an applied magnetic field.4,5 Transport signatures consistent with this theoretical prediction were subsequently detected,6,7 although the definitive existence of the Majorana mode in such devices is still debated.8 Much attention has recently been directed at an alter- native proposal, where a topological band arises from the overlapping Yu-Shiba-Rusinov (YSR) states9 in a chain of magnetic impurities with helical spin order on the sur- face of a superconductor.10–21 The helical spin texture plays a critical role combining the effect of the spin-orbit coupling (SOC) and external field in the nanowire pro- posal. Topological states are similarly predicted in metal- lic systems with coexisting superconductivity and helical magnetic order.23–25 A significant advantage of the YSR chain proposal is that it is possible to unambiguously im- age the Majorana end modes using scanning tunneling microscopy (STM), in contrast to relying on difficult-to- interpret transport measurements of nanowire systems. Although critical to ensuring a topological state, the he- lical order also represents the main experimental diffi- culty since it is impossible to control externally. The helical order is stable when the magnetic ions are placed on a quasi-one-dimensional substrate,12,14,16 but for the physically-relevant case of a planar surface the chain is generically unstable towards a ferromagnetic or antifer- romagnetic configuration.26 A pair-breaking effect in the superconducting state might nevertheless restore the sta- bility of the helical order,20 but disorder effects may still The prospect of unambiguously verifying the existence of Majorana end modes in a YSR chain motivates the search for a way to realize a topological state in this sys- tem without relying upon an intrinsic helical ordering of the impurity spins. For example, it has been proposed to use external magnetic fields and a supercurrent flow to tune a nontopological antiferromagnetic chain into a topological regime.22 In view of the importance of SOC in the nanowire proposal, it is also interesting to include SOC in the description of the superconducting host of the impurity chain. Indeed, one typically expects the pres- ence of a Rashba SOC at the surface of the superconduc- tor due to the broken inversion symmetry. We note that SOC intrinsic to the superconductor has been considered in other proposals for realizing topological systems,27 and the possible relevance of SOC in the context of a topolog- ically nontrivial magnetic impurity chains has been men- tioned in Ref. 26. This scenario has recently been invoked to explain STM measurements of zero-bias peaks at the ends of a ferromagnetic chain on a superconductor,28 al- though the relevance of YSR physics to this situation is uncertain, as we discuss below. In spite of the exten- sive activity on the interplay between magnetic chains and superconductivity in generating emergent topologi- cal phases,10–22 the specific problem of combining both spin-orbit coupling and magnetic YSR chain physics to- gether in a model has been conspicuously lacking. We study this particular issue in our current work by general- izing and synthesizing the existing work in the literature, most specifically Refs. 10, 13, and 26. We mention here that very recently there has been a spurt in the activity29–32 on topological superconductiv- ity and emergent Majorana fermions, following Ref. 28, in ferromagnetic chains fabricated on the surface of bulk su- perconductors. In particular, a report of impressive STM experiments29 has just appeared claiming the generic ob- servation of Majorana fermions at the ends of Fe chains on superconducting Pb. These experiments are the main reason for this enhanced activity, but the question of the 5 1 0 2 b e F 0 2 ] l l a h - s e m . t a m - d n o c [ 4 v 5 4 3 6 . 7 0 4 1 : v i X r a topological nature of purely ferromagnetic chains prox- imity coupled to s-wave superconductors is of intrinsic theoretical interest independent of experimental develop- ments. Rather surprisingly, the theory of such systems has not yet been dealt with in any detail, in contrast to the extensive theoretical analyses of spiral YSR chains and spin-polarized semiconducting nanowires. Although there is superficial similarity between the model used in the current work and the experimental systems,28,29 it is far too early to tell whether there is any connection between the theory and predictions presented in the cur- rent work involving a weakly coupled YSR chain and the experimental system involving Fe chains where tunneling between the atoms may well be strong. The correspond- ing topological theory for strongly-coupled ferromagnetic chains on superconducting substrates has been consid- ered in Refs. 29–32, and is a generalization of previous proposals to realize Majorana fermions in half-metals (i.e. fully spin-polarized ferromagnets) deposited on supercon- ductors.27 Further discussion of such strongly tunnel cou- pled ferromagnetic nanowire systems and the experimen- tal results of Ref. 29 is beyond the scope of the current work. In this paper we show that the SOC indeed induces a topological state in a YSR chain formed from ferro- magnetically aligned impurity spins, and so demonstrate that the more delicate helical order is not essential to such proposals. To this end, we analytically construct a tight-binding model for the YSR states valid in the limit of "deep" impurities, when the impurity band lies close to the middle of the superconducting gap. Although the SOC does not affect the YSR states for an isolated im- purity, it dramatically alters the results for the chain. Specifically, spin-flip correlations in the bulk supercon- ductor, induced by the antisymmetric SOC, mix the two branches of the impurity band when the polarization of the impurity spins is transverse to the SOC along the chain. This can be interpreted as a triplet pairing ampli- tude in a Kitaev-like model, and is thus responsible for the topologically nontrivial state. A magnetic polariza- tion parallel to the SOC, on the other hand, produces no such mixing but instead results in an asymmetric disper- sion with trivial topology. We construct a phase diagram, demonstrating that a topological state is possible for in- finitesimal SOC strength. Our analysis closely follows that of Ref. 13, where a similar tight-binding model for the impurity band was obtained for a chain with spiral magnetic texture embedded in a three-dimensional su- perconductor. II. MODEL (cid:80) † k k Ψ A bulk two-dimensional singlet s-wave superconductor with Rashba SOC is described by the Hamiltonian H = HkΨk where Hk = τz ⊗ (ξk σ0 + lk · σσσ) + ∆τx ⊗ σ0 . (1) 2 Here τµ (σµ) are the Pauli matrices in Nambu (spin) † † −k,↓,−c −k,↑)T is the spinor space, and Ψk = (ck,↑, ck,↓, c of creation and annihilation operators. We have adopted the notation that . . . and . . . indicate 2×2 and 4×4 matri- ces, respectively. The noninteracting dispersion is given by ξk = 2k2/2m−µ where m is the effective mass and µ the chemical potential, the Rashba SOC is parametrized by lk = λ(kyex − kxey) = λk(sin θex − cos θey) where λ is the SOC strength, and ∆ is the superconducting gap. The SOC lifts the spin degeneracy in the normal state, resulting in the dispersions ξk,± = ξk ± lk, where the plus (minus) sign corresponds to the positive (negative) helicity band. As time-reversal symmetry remains in- tact, however, in the superconducting phase there is only pairing between states in the same helicity band. The bulk Green's function can then be written as Gk(ω) = 2{ G+ 1 k (ω) = (ωτ0 + ξ± τz + ∆τx) ⊗ (σ0 ± sin θσx ∓ cos θσy) G± (2) ×(cid:0)ω2 − ξ2± − ∆2(cid:1)−1 k (ω)}, where k (ω) + G− , is the Green's function in each helicity sector. Note that the SOC produces normal spin-flip and triplet pairing terms in the Green's function.33 For clarity we suppress the momentum index in the dispersion of the helical bands, i.e. ξk,± ≡ ξ±. III. SINGLE IMPURITY to the bulk Hamiltonian, where Ψ(r) = (cid:82) d2k We first consider a single (classical) magnetic impurity with spin S at the origin, interacting with the electron states with exchange strength −J. We include this in our model by adding Himp = −JS · [Ψ†(0)τ0 ⊗ σσσΨ(0)] (2π)2 Ψkeik·r. We aim to solve the Bogoliubov-de Gennes equation (H + Himp)ψ(r) = ωψ(r) for the impurity bound states, for energy ω < ∆. By straightforward manipula- i.e. tion,13 the spinor of the bound state at the impurity ψ(0) satisfies the equation (cid:26) (cid:90) (cid:27) 1 + d2k (2π)2 Gk(ω)JS · (τ0 ⊗ σσσ) ψ(0) = 0 . (3) (cid:90) D To evaluate this equation, we split the Green's function into positive and negative helicity components and then convert the integral over the momentum to an integral over the appropriate dispersion ξ± and the angle θ (cid:90) where Nν = (m/2π2)[1 ∓(cid:101)λ/(1 +(cid:101)λ2)1/2] is the density (cid:101)λ = λm/2kF is the ratio of SOC splitting to the Fermi of states of the ν = ± helicity band at the Fermi level, k (ω) ≈ N± G± (cid:90) 2π d2k (2π)2 dθ G± k (ω) , 2π −D dξ± 0 (4) energy and gives a dimensionless measure of the SOC strength, kF the Fermi wavevector in the absence of SOC, and D → ∞ is a cutoff. The symmetric cutoff in Eq. (4) is χ↓ =(cid:0) e−iη sin ζ/2 , − cos ζ/2(cid:1)T . 3 (8) IV. FERROMAGNETIC CHAIN The above analysis can be extended to a chain of ferromagnetically-aligned impurity spins, with the impu- rity Hamiltonian now written as Himp = −J S · [Ψ†(rj)τ0 ⊗ σσσΨ(rj)] , (9) (cid:88) j where rj is the position of the jth impurity. We have suppressed the site index of the spins since they all point in the same direction. Without loss of generality, we assume that the chain runs along the x-axis, and so rj = xjex. After similar manipulations as in the sin- gle impurity problem, the BdG equations for the subgap YSR states on the chain can be written 1 − α√ ∆2 − ω2 [ωτ0 + ∆τx] ⊗ (eS · σσσ) ψ(xi) J(xij)eS · (τ0 ⊗ σσσ)ψ(xj) = −(cid:88) j(cid:54)=i (10) (cid:26) (cid:27) (7) where xij = xi − xj and the matrix J(xij) is defined (cid:27) (cid:26) used for simplicity; Although it implies particle-hole sym- metry of the normal dispersion, relaxing this assumption does not qualitatively change our results. The result- ing integrals are presented in the appendix. Due to the isotropic δ-function structure of the potential, the inte- grals involving the spin-flip and triplet pairing terms in the Green's function vanish, and Eq. (3) therefore has ex- actly the same form as a magnetic impurity in an s-wave superconductor without SOC,9,13 specifically 1 − ψ(0) = 0 , (5) [ωτ0 + ∆τx] ⊗ (eS · σσσ) α√ ∆2 − ω2 2 (N+ +N−)JS, S = S, and eS = S/S. The where α = π solutions of this equation occur at ω = ±
0, where
0 = ∆(1−α2)/(1+α2). The form of the corresponding spinors ψ±(0) is dictated by the orientation of the impurity spin. Parametrizing S = S(cos η sin ζ, sin η sin ζ, cos ζ), these spinors can then be written13 up to unimportant nor- malization constant as ψ+(0) = (cid:19) (cid:19) (cid:18) χ↑ (cid:18) χ↓−χ↓ χ↑ =(cid:0) cos ζ/2 , eiη sin ζ/2(cid:1)T ψ−(0) = χ↑ , , , (6) where J(xij) = JS Gk(ω)eikxxij (cid:90) (cid:8)[I− d2k (2π)2 = JS 2 +[I− 1 (xij) + I + 1 (xij)]τz ⊗ σ0 + ω[I− 3 (xij) + I + 3 (xij)]τ0 ⊗ σ0 + ∆[I− 3 (xij) + I + 3 (xij)]τx ⊗ σ0 2 (xij) − I + 2 (xij)]τz ⊗ σy + ω[I− 4 (xij) − I + 4 (xij)]τ0 ⊗ σy + ∆[I− 4 (xij) − I + 4 (xij)]τx ⊗ σy (11) (cid:9) . We have expressed this in terms of the integrals I ν 1 (x) = I ν 2 (x) = I ν 3 (x) = 0 (cid:90) 2π (cid:90) 2π (cid:90) 2π (cid:90) 2π 0 0 dξ dξ dξ −D (cid:90) D (cid:90) D (cid:90) D (cid:90) D −D −D Nν 2π Nν 2π Nν 2π Nν 2π dθ dθ dθ (12a) ξeikν (ξ)x cos θ ω2 − ξ2 − ∆2 , ξeiθeikν (ξ)x cos θ ω2 − ξ2 − ∆2 , (12b) eikν (ξ)x cos θ ω2 − ξ2 − ∆2 , eiθeikν (ξ)x cos θ ω2 − ξ2 − ∆2 , (12c) I ν 4 (x) = dξ (12d) where kν(ξ) = kF,ν + ξ/vF,ν, while kF,ν = kF [(1 + (cid:101)λ2)1/2−ν(cid:101)λ] and vF,ν = (kF /m)(1+(cid:101)λ2)1/2 are the Fermi −D dθ 0 vector and velocity for the ν helicity band, respectively. These integrals are explicitly evaluated in the appendix for D → ∞, where we also provide asymptotic expan- sions valid for kF,νx (cid:29) 1. Note that I ν 3 (x) are even functions of x, whereas I ν 4 (x) are odd. In contrast to the single-impurity system considered above, the presence of SOC makes a significant differ- 1 (x) and I ν 2 (x) and I ν ence to the BdG equations for the multi-impurity prob- lem: while the first line of Eq. (11) is identical to the result found in Ref. 13, the second line is only present for nonzero SOC. This line contains explicitly magnetic terms ∝ σy, reflecting the orientation of the SOC vector lkey for k pointing along the magnetic chain. V. TIGHT-BINDING MODEL We do not attempt a general solution of Eq. (10), but instead consider the analytically-tractable limit of dilute "deep" impurities, as discussed in Ref. 13. Specifically, we assume that α ≈ 1, so that the energy
0 of the isolated YSR state lies close to the center of the gap, and that the spacing a between impurities is sufficiently large that the impurity band formed from the hybridized YSR states lies entirely within the superconducting gap. Linearizing the BdG equations Eq. (10) in the energy ω and the coupling between impurity sites, we obtain after ∆ [eS · (τ0 ⊗ σσσ) − ατx ⊗ σ0] ψ(xi) + ∆ (cid:88) j(cid:54)=i straightforward manipulation 4 eS · (τ0 ⊗ σσσ) lim ω→0 J(xij)eS · (τ0 ⊗ σσσ)ψ(xj) = ωψ(xi) (13) This equation is now projected into the YSR states [Eq. (6)] at each site, to obtain a BdG-type equation where φi = (ui,+, ui,−)T is the vector of the wavefunc- tions for the + and − YSR states at site i and C∗ ji Cij −Aij + Bij (cid:101)H(i, j) = for the impurity band(cid:101)H(i, j)φj = ωφi (cid:18) Aij + Bij (cid:2)I + (cid:2)I− 2 e−2iη(cid:17) 2 (xij)(cid:3) . (cid:16) (cid:2)I− cos2 ζ 2 (xij) − I + 1 Bij = 2 Cij = − i 2 JS∆2 sin η sin ζ lim ω→0 JS∆2 lim ω→0 Aij =
0δij + 2 + sin2 ζ JS∆ × lim ω→0 where 1 2 3 (xij) + I− 4 (xij) − I + (14) (15) (cid:19) 3 (xij)(cid:3) , 4 (xij)(cid:3) ,(17) (16) (18) Note that the integrals in these expressions are to be regarded as vanishing for i = j. The effective tight-binding Hamiltonian Eq. (15) is the central result of this paper. Due to the antisymme- try of the integrals I ν 2 (x) in the off-diagonal terms, it can be interpreted as describing superconducting spin- less fermions, recalling the Kitaev model,2 albeit with long-range hopping and pairing terms. The properties of this system depend crucially on the SOC in the bulk su- perconductor and the polarization of the impurity spins. Specifically, the pairing term Cij is only present for non- vanishing SOC, and when the polarization of the ferro- magnetic chain has a component perpendicular to the y-axis. Examining Eq. (11), we observe that the pair- ing term originates from the spin-flip correlations in the host superconductor induced by the SOC. A polarization component along the y-axis contributes an antisymmet- ric hopping Bij in the presence of SOC. This echoes the asymmetric dispersion of a spin-orbit coupled electron gas in the direction of an applied magnetic field, and its appearance here is due to the triplet pairing correlations in the bulk Green's function Eq. (2). A similar tight-binding model was derived in Ref. 13, but there the odd-parity pairing term arose from the spi- ral magnetic texture of the impurity chain. This mech- anism for generating a pairing term is still valid in the presence of the SOC considered here. Examining the in- terplay of spiral spin texture and SOC is an interesting topic which we leave to later work. VI. TOPOLOGICAL PROPERTIES To conclude we examine the topology of the impurity band. For an infinite chain with uniform spacing a of the impurities, we define the Fourier transform of the Hamiltonian Eq. (15) (cid:19) C(k) −A(k) + B(k) (19) (cid:18) A(k) + B(k) C∗(k) (cid:101)H(k) = where A(k) = (cid:80) j A0jeikja, etc. Using the asymptotic forms for the integrals, it is possible to obtain analytical expressions for these quantities in the limit kF,νa (cid:29) 1, which are presented in the appendix. The Hamilto- nian Eq. (19) is in Altland-Zirnbauer symmetry class D, and for a fully-gapped system it is therefore characterized by the Z2 topological invariant2 Q = sgn{A(0)A(π/a)} . (20) The system is topologically nontrivial for Q = −1; con- versely, Q = 1 indicates a trivial state. To demonstrate that our model supports a topolog- ically nontrivial state, in Fig. (1) we present a phase diagram as a function of the dimensionless SOC (cid:101)λ and the parameter kF a, which gives a measure of the Fermi surface volume or alternatively the spacing of the chain. We consider only a polarization in the x-z plane. In the topologically non-trivial regions, we plot the mini- mum gap magnitude, demonstrating the existence of a fully gapped state; the non-topological regions are left white. The most important aspect of this phase diagram is that a topological state is revealed to be possible even for infinitesimal SOC. Remarkably, the excitation spec- trum can display a substantial gap even for very small SOC strength (cid:101)λ (cid:28) 1. We emphasize that our analy- sis is only valid for
0 sufficiently close to zero, and so other methods are required to comprehensively survey the phase diagram. VII. SUMMARY In this paper we have studied the appearance of a topo- logical impurity band when a ferromagnetic chain of clas- sical spins are embedded in a two-dimensional singlet s- wave superconductor with Rashba SOC. To this end, we have derived an effective tight-binding model for the over- lapping YSR states of the impurities. When the spins are polarized perpendicular to the SOC along the chain, an 5 to confined ourselves Although we have the analytically-tractable limit of a dilute chain of deep impurities, we expect that our results are of more general validity since they rely only upon the low-energy form of the Green's function. We have also neglected complicating factors such as particle-hole asymmetry in the normal state dispersion, the suppression of the superconducting gap close to the impurity spins, and the three-dimensional nature of the superconducting host. These issues must certainly be accounted for when mod- elling a realistic system, but can only be addressed using large-scale computer simulations. Nevertheless, none of these effects should invalidate the mechanism giving rise to the topological state of our basic model which arises simply from the interplay between ferromagnetism, superconductivity, and SOC. ACKNOWLEDGMENTS The authors thank P. Kotetes and A. Yazdani for use- ful discussions. This work is supported by JQI-NSF-PFC and LPS-CMTC. Appendix A: Important integrals In this appendix we present analytic forms for the four integrals Eq. (12) encountered in our solution of the YSR chain. We perform by these integrals by extending the cutoff D → ∞. We distinguish two cases for the argu- ment: x = 0 for the isolated YSR impurity, and x (cid:54)= 0 for the YSR chain. 1. Isolated impurity: x = 0 In this case all the integrals except I ν 3 (0) are vanishing, which evaluates to 3 (0) = − πNν √ I ν ∆2 − ω2 . (A1) Impurity chain: x (cid:54)= 0 2. For x (cid:54)= 0, we first evaluate the integral over ξ using elementary contour integral methods, and then evaluate the angular integral. We hence find FIG. 1. (Color online). Topological phase diagram for the effective model as a function of kF a and (cid:101)λ. The topological regions are shaded according to the magnitude of the gap, while the nontopological regions are left blank. Red lines indicate the boundary between topological and nontopological phases. We have chosen
0 = 0 for the isolated impurity level and ξ0 = 5a for the superconducting coherence length at (cid:101)λ = 0, which ensures that the impurity band remains within the superconducting gap. The impurity spins point in the x-z plane. The large values of kF a allow us to utilize the asymptotic expressions for the entries in Eq. (19). odd-parity pairing term is induced in the effective Hamil- tonian, thus realizing a Kitaev-like model with generi- cally non-trivial topology. Our work and recent others22 explore alternative routes to a topological YSR chain which do not rely upon helical spin texture.10–21 This is a significant result, as the stability of the helical spin texture is debated.20,26 In contrast, the SOC mechanism examined here is intrinsic to the superconductor surface. This implies that topological phases are possible for a much wider variety of impurity spin configurations than hitherto realized, which grants the YSR chain proposal additional robustness and lends strong theoretical sup- port to experimental efforts to detect Majorana fermions in such a setting. As revealed by our calculated quantum phase diagram Fig. (1), however, the topological phase in the ferromagnetic YSR chain system is not generic. Some fine-tuning of the system is therefore required in order to observe topological Majorana fermions through the measurement, for example, of zero-bias-conductance peaks in tunneling spectroscopy experiments. 1 (x) = πNνIm(cid:8)J0((kF,ν + iξ−1 2 (x) = −iπNν sgn(x)Re(cid:8)iJ1((kF,ν + iξ−1 Re(cid:8)J0((kF,ν + iξ−1 3 (x) = − πNν √ I ν I ν I ν ∆2 − ω2 ν )x) + H−1((kF,ν + iξ−1 ν )x) + iH0((kF,ν + iξ−1 ν )x)(cid:9) , ν )x)(cid:9) , ν )x) + iH0((kF,ν + iξ−1 ν )x)(cid:9) , (A2) (A3) (A4) E
(cid:68)00.0250.0500.0750.1000.125Im(cid:8)iJ1((kF,ν + iξ−1 √ iπNν ∆2 − ω2 ν )x) + H−1((kF,ν + iξ−1 ν )x)(cid:9) , √ 6 (A5) ∆2 − ω2. Using where Jn(z) and Hn(z) are Bessel and Struve functions of order n, respectively, and ξν = vF,ν/ asymptotic forms34 valid for large values of the argument close to the positive real axis, we can approximate these as 4 (x) = − sgn(x) I ν (cid:115) 4 2 1 (x) ≈ πNν I ν 2Nν kF,νx , πkF,νx sin(cid:0)kF,νx − π (cid:1) e−x/ξν + (cid:115) πkF,νx sin(cid:0)kF,νx − 3π 2 (x) ≈ iπNν sgn(x) (cid:115) I ν πkF,νx cos(cid:0)kF,νx − π 3 (x) ≈ − πNν (cid:115) (cid:1) e−x/ξν + sgn(x) (cid:1) e−x/ξν , πkF,νx cos(cid:0)kF,νx − 3π (cid:1) e−x/ξν . 4 (x) ≈ − sgn(x) I ν iπNν ∆2 − ω2 ∆2 − ω2 √ √ I ν 2 2 2 4 4 4 2iNν (kF,νx)2 , (A6) (A7) (A8) (A9) The nonoscillating component is valid up to O((kF,νx)−3). 3. Fourier transforms The Fourier transform of the effective Hamiltonian Eq. (15) can be carried out analytically when we utilize the asymptotic expressions. Defining the Fourier transform as A(k) = A0jeikja , (A10) (cid:88) j (cid:17) ei(−kF,ν a+ka)−a/ξν (cid:16) (cid:17) − e3πi/4Li 1 (cid:16) 2 ei(−kF,ν a−ka)−a/ξν we obtain 2 (k) = Nν I ν (cid:114) π 2kF,νa −e−3πi/4Li 1 ei(kF,ν a−ka)−a/ξν 2 2 (cid:16) e−3πi/4Li 1 (cid:26) (cid:16) (cid:0)eika(cid:1) − Li2 (cid:8)Li2 (cid:26) (cid:114) π (cid:16) (cid:114) π (cid:16) (cid:26) 2kF,νa ei(kF,ν a−ka)−a/ξν 2kF,νa ei(kF,ν a−ka)−a/ξν 2 ei(kF,ν a+ka)−a/ξν (cid:17) − e3πi/4Li 1 (cid:0)e−ika(cid:1)(cid:9) , (cid:16) (cid:17) (cid:16) (cid:16) (cid:17) − e3πi/4Li 1 + eπi/4Li 1 2 2 2 + 3 (k) = − I ν 2iNν (kF,νa)2 Nν√ ∆2 − ω2 +e−πi/4Li 1 4 (k) = − iNν √ I ν 2 ∆2 − ω2 −e−3πi/4Li 1 e−πi/4Li 1 ei(kF,ν a+ka)−a/ξν + eπi/4Li 1 2 ei(−kF,ν a+ka)−a/ξν e−3πi/4Li 1 ei(kF,ν a+ka)−a/ξν ei(−kF,ν a−ka)−a/ξν (cid:16) + e3πi/4Li 1 2 ei(−kF,ν a−ka)−a/ξν , (cid:17) ei(−kF,ν a+ka)−a/ξν (cid:17) (A11) (A12) (A13) (cid:17) (cid:17) (cid:17)(cid:27) (cid:16) (cid:17)(cid:27) (cid:16) (cid:17)(cid:27) , where Lis(z) is the polylogarithm of order s. 2 2 ∗ pbrydon@umd.edu 1 N. Read and D. Green, Phys. Rev. B 61, 10267 (2000); S. Das Sarma, C. Nayak, and S. Tewari, Phys. Rev. B 73, 220502(R) (2006); C. Zhang, S. Tewari, R. M. Lutchyn, and S. Das Sarma, Phys. Rev. Lett. 101, 160401 (2008); M. Sato and S. Fujimoto, Phys. Rev. B 79, 094504 (2009); P. Hosur, P. Ghaemi, R. S. K. Mong, and A. Vishwanath, Phys. Rev. Lett. 107, 097001 (2011). 7 2 A. Y. Kitaev, Phys. Usp. 44, 131 (2001). 3 L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008); J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, ibid 104, 040502 (2010); J. D. Sau, S. Tewari, R. M. Lutchyn, T. Stanescu, and S. Das Sarma, Phys. Rev. B 82, 214509 (2010). 4 R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. 105, 077001 (2010). 5 Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105, 177002 (2010). 6 V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Science 336, 1003 (2012). 7 A. Das, Y. Ronen, Y. Most, Y. Oreg, M. Heiblum, and H. Shtrikman, Nature Phys. 8, 887 (2012). 8 D. Bagrets and A. Altland, Phys. Rev. Lett. 109, 227005 (2012); D. I. Pikulin, J. P. Dahlhaus, M. Wimmer, H. Schomerus, and C. W. J. Beenakker, New. J. Phys. 14, 125011 (2012). 9 L. Yu, Acta Phys. Sin. 21, 75 (1965); H. Shiba, Prog. Theor. Phys. 40, 435 (1968); A. I. Rusinov, Zh. Eksp. Teor. Fiz. Pisma. Red. 9, 146 (1968) [JETP Lett. 9, 85 (1969)]. 10 T.-P. Choy, J. M. Edge, A. R. Akhmerov, and C. W. J. Beenakker, Phys. Rev. B 84, 195442 (2011). 11 S. Nadj-Perge, I. K. Drozdov, B. A. Bernevig, and A. Yaz- dani, Phys. Rev. B 88, 020407(R) (2013). 12 B. Braunecker and P. Simon, Phys. Rev. Lett. 111, 147202 (2013). 13 F. Pientka, L. I. Glazman, and F. von Oppen, Phys. Rev. B 88, 155420 (2013). 17 K. Poyhonen, A. Weststrom, J. Rontynen, and T. Ojanen, Phys. Rev. B 89, 115109 (2014). 18 F. Pientka, L. I. Glazman, and F. von Oppen, Phys. Rev. B 89, 180505(R) (2014). 19 J. Rontynen and T. Ojanen, Phys. Rev. B 90, 180503(R) (2014). 20 I. Reis, D. J. J. Marchand, and M. Franz, Phys. Rev. B 90, 085124 (2014). 21 S. Nadj-Perge, I. Drozdov, S. Jeon, J. Seo, A. Bernevig, and A. Yazdani, Bull. Am. Phys. Soc. 59, Z46.10 (2014). 22 A. Heimes, P. Kotetes, and G. Schon, Phys. Rev. B 90, 060507(R) (2014). 23 M. Kjaergaard, K. Wolms, and K. Flensberg, Phys. Rev. B 85, 020503(R) (2012). 24 I. Martin and A. F. Morpurgo, Phys. Rev. B 85, 144505 (2012). 25 P. Kotetes, New J. Phys. 15, 105027 (2013). 26 Y. Kim, M. Cheng, B. Bauer, R. M. Lutchyn, and S. Das Sarma, Phys. Rev. B 90, 060401(R) (2014). 27 P. A. Lee, arXiv:0907.2681; M. Duckheim and P. W. Brouwer, Phys. Rev. B 83, 054513 (2011); S.-B. Chung, H.-J. Zhang, X.-L. Qi, and S.-C. Zhang, ibid 84, 060510(R) (2011). 28 A. Yazdani, Nobel Symposium 156 on New Forms of Mat- ter: Topological Insulators and Superconductors, Sweden, June 12-15 (2014) . 29 S. Nadj-Perge, I. K. Drozdov, J. Li, H Chen, S. Jeon, J. Seo, A. H. MacDonald, B. A. Bernevig, and A. Yazdani, Science 346, 602 (2014). 30 H.-Y. Hui, P. M. R. Brydon, J. D. Sau, S. Tewari, and S. 14 J. Klinovaja, P. Stano, A. Yazdani, and D. Loss, Phys. Das Sarma, arXiv:1407.7519. Rev. Lett. 111, 186805 (2013). 15 S. Nakosai, Y. Tanaka, and N. Nagaosa, Phys. Rev. B 88 180503 (2013). 31 J. Li, H. Chen, I. K. Drozdov, A. Yazdani, B. A. Bernevig, and A.H. MacDonald, Phys. Rev. B 90, 235433 (2014). 32 E. Dumitrescu, B. Roberts, S. Tewari, J. D. Sau, and S. 16 M. M. Vazifeh and M. Franz, Phys. Rev. Lett. 111, 206802 Das Sarma, arXiv:1410.5412. (2013). 33 L. P. Gorkov and E. I. Rashba, Phys. Rev. Lett. 87, 037004 (2001). 34 M. Abramowitz and I. A. Stegun, Handbook of mathemat- ical functions, (Dover, New York, 1964). 

1606.03232

2

1606

Topological origin of edge states in two-dimensional inversion-symmetric insulators and semimetals

Symmetries play an essential role in identifying and characterizing topological states of matter. Here, we classify topologically two-dimensional (2D) insulators and semimetals with vanishing spin-orbit coupling using time-reversal ($\mathcal{T}$) and inversion ($\mathcal{I}$) symmetry. This allows us to link the presence of edge states in $\mathcal{I}$ and $\mathcal{T}$ symmetric 2D insulators, which are topologically trivial according to the Altland-Zirnbauer table, to a $\mathbb{Z}_2$ topological invariant. This invariant is directly related to the quantization of the Zak phase. It also predicts the generic presence of edge states in Dirac semimetals, in the absence of chiral symmetry. We then apply our findings to bilayer black phosphorus and show the occurrence of a gate-induced topological phase transition, where the $\mathbb{Z}_2$ invariant changes.

cond-mat.mes-hall

cond-mat

Topological origin of edge states in two-dimensional inversion-symmetric insulators and semimetals Guido van Miert,1 Carmine Ortix,2, 1 and Cristiane Morais Smith1, 3 1Institute for Theoretical Physics, Centre for Extreme Matter and Emergent Phenomena, Utrecht University, Princetonplein 5, 3584 CC Utrecht, The Netherlands 2Institute for Theoretical Solid State Physics, IFW Dresden, PF 270116, 01171 Dresden, Germany 3Wilczek Quantum Center, Zhejiang University of Technology, Hangzhou 310023, China Symmetries play an essential role in identifying and characterizing topological states of mat- ter. Here, we classify topologically two-dimensional (2D) insulators and semimetals with vanishing spin-orbit coupling using time-reversal (T ) and inversion (I) symmetry. This allows us to link the presence of edge states in I and T symmetric 2D insulators, which are topologically trivial according to the Altland-Zirnbauer table, to a Z2 topological invariant. This invariant is directly related to the quantization of the Zak phase. It also predicts the generic presence of edge states in Dirac semimetals, in the absence of chiral symmetry. We then apply our findings to bilayer black phosphorus and show the occurrence of a gate-induced topological phase transition, where the Z2 invariant changes. PACS numbers: 03.65.Vf,73.20.-r, 73.22.-f I. INTRODUCTION The characterization of topological states of matter is a central topic in condensed-matter physics.1,2 A beau- tiful example is given by the Altland-Zirnbauer (AZ) table3, which classifies topological insulators and super- conductors depending on their dimensions and discrete symmetries.4–6 A 2D Chern insulator, for example, is characterized by a Z invariant and relies on the absence of time-reversal (T ) symmetry.7 The presence of this sym- metry is instead crucial for Z2 topological insulators that exhibit the quantum spin Hall effect.8 In these topolog- ically non-trivial insulators, there is a one-to-one cor- respondence between the topological invariant and the number of gapless modes localized at the edge, known as the bulk-boundary correspondence.9 In Chern insula- tors the edge states are chiral, which means that they all propagate in the same direction, whereas topological T invariant insulators exhibit helical edge states, with elec- trons of opposite spins counterpropagating at the sample boundaries. There are, however, insulators and semimetals, which are topologically trivial according to the AZ-table, al- though they generally do exhibit edge states. A natural question is then whether these "trivial " edge states are related to a topological invariant that exists in the pres- ence of a discrete or continuous symmetry. For 2D chiral systems, for instance, the existence of zero-energy edge states can be inferred from a 1D winding number.10,11 This explains the origin of trivial edge states in a min- imal tight-binding model for graphene. More recently, this symmetry has been used to predict edge states in single-layer black phosphorus (sometimes also called phosphorene).12 However, this explanation is quite un- satisfactory for insulators and semimetals because we do not expect the presence of a chiral symmetry: both in graphene and black phosphorus, the next-nearest neigh- bor hopping breaks the chiral symmetry. Here, we reveal the importance of inversion (I)- symmetry, which has been overlooked in previous stud- ies. We demonstrate that the existence of edge states in 2D crystalline insulators without spin-orbit coupling (SOC) in the presence of T and I symmetry is related to a one-dimensional (1D) Z2 invariant. We then apply this result to a generic toy model, and elucidate the relation between the edge states and this topological invariant. Moreover, we discuss the quantization of the edge charge in the absence of edge states. Finally, we apply our re- sults to single and bilayer black phosphorus. II. SYMMETRY OF THE BLOCH HAMILTONIAN Let us consider a 2D crystalline insulator, described by a Bloch Hamiltonian H((cid:126)k). We assume negligible FIG. 1: (Color online) (a) Oblique lattice with lattice vectors (cid:126)a1 and (cid:126)a2. The red dashed rhomboid shows the unit-cell. (b) The corresponding 2D Brillouin zone, with the reciprocal lattice vectors (cid:126)b1 and (cid:126)b2. The red dashed line corresponds to the contour C0, whereas the solid line corresponds to Ck , the blue circle denotes the contour Cs, and the enclosed area S is shown in yellow. (a)(b)SOC, such that we can model the system using spin- less fermions, for which T 2 = +1. In the presence of I and T symmetry, the Hamiltonian satisfies the following constraints: H((cid:126)k) = IH(−(cid:126)k) I, H((cid:126)k) = H(−(cid:126)k)∗, (1) where I is the inversion operator and ∗ denotes complex conjugation. Since we are interested in the existence of edge states, we put the system on a cylinder, as in Fig. 1(a), and limit ourselves to crystalline edges, char- acterized by one of the lattice vectors (cid:126)a1 = (a1, 0). We will assign a topological invariant to the projected Bloch Hamiltonian Hk(k⊥) = H(k, k⊥), where k(k⊥) de- notes the momentum along (perpendicular to) the edge. To discuss the consequences of the constraints in Eq. (1), we first briefly revisit the topology of band insulators in 1D. III. BAND TOPOLOGY IN 1D According to the AZ-table, insulators in 1D, belong- ing to classes A, AI, and AII, are trivial. Despite this fact, band topology does play an important role for these crystalline insulators. In particular, the surface charge at the end of such an insulator is directly related to the Zak phase13,14 (cid:90) π/a −π/a γ = dkA(k), (2) the lattice constant (cid:80) and A(k) = where a is i∈occ(cid:104)uk,ii∇kuk,i(cid:105) denotes the Berry connection, with uk,i(cid:105) the periodic part of the Bloch wave function with band index i, momentum k, and "occ" denotes the set of occupied bands. Generically, the Zak phase is not quan- tized. However, in the presence of 1D I-symmetry, one finds γ = −γ mod 2π (see Appendix A). Hence, either γ = 0 or γ = π, and as such it defines a Z2 invariant.15 Moreover, we do not need to compute the integral in Eq. (2); instead, we can write (cid:89) i∈occ T (cid:88) × (cid:88) × × eiγ = I (cid:88) (cid:88) × × × ξi(0)ξi(π/a) ∈ Z2, (3) IT (cid:88) × × (cid:88) × 1D Z2 Z2 0 Z2 0 TABLE I: Classification of 1D insulators in the presence or absence of I, T , and T I symmetry, based on the Zak phase. Here, we consider spinless fermions, meaning T 2 = +1. 2 where ξi(kinv) = (cid:104)Ψkinv,i IΨkinv,i(cid:105) is the parity of the full Bloch wave function Ψkinv,i(cid:105) at the I invariant momenta kinv. By considering T -symmetry, one can construct five In 1D, three of these allow for a Z2 different classes. classification, based on the Zak phase, see Table I and Appendix A. Next, we employ this invariant to charac- terize a 2D system. IV. Z2 CLASSIFICATION Let us now return to the set of 1D Hamiltonians Hk(k⊥) parameterized by k. Note that for k = 0, the symmetry constraints given in Eq. (1) are inher- ited. In particular, the 2D I symmetry yields H0(k⊥) = IH0(−k⊥) I. Therefore, the associated Zak phase γ(0) is quantized, and defines a Z2 invariant16,17, χ1 = eiγ(0) = ξi(0)ξi((cid:126)b2/2). (4) (cid:89) i∈occ invariant χ2 = (cid:81) Here, (cid:126)b2 is the reciprocal-lattice vector pointing in the direction perpendicular to the edge, see Fig. 1(b), and the subscript in χ1 reminds us of the fact that this invariant is associated with an edge parallel to (cid:126)a1. If the edge would be along (cid:126)a2, then we should simply consider the i∈occ ξi(0)ξi((cid:126)b1/2). The definition of the invariant in Eq. (4) is, strictly speaking, only valid for k = 0. However, we can express the difference in the Zak phases γ(k) − γ(0) as an integral of the Berry connection (cid:126)A along the contour Ck − C0, see Fig. 1(b). Using Stokes' theorem, this can be rewritten as a surface integral of the Berry curvature F (k, k⊥) = ∂kA⊥ − ∂k⊥ A, see Fig. 1(b). Since the two symmetry constraints in Eq. (1) ensure that F = 0 (see Appendix B), we find γ(k) = γ(0). It follows that even in the absence of chiral symmetry, we can still associate a Z2 invariant with each of the 1D Hamiltonians Hk. So far, we have implicitly assumed that Hk and H0 are adiabatically connected, such that F is actually well defined within the yellow area in Fig. 1(b). Therefore, our proof does not apply to systems with band-crossing points, like semimetals. In the latter case, we can modify our proof, by assuming that the band-crossing point is located inside the contour Cs, as indicated in Fig. 1(b). Then, we can apply Stokes' theorem to the new contour (cid:73) Ck − C0 − Cs, which yields γ(k) − γ(0) = d(cid:126)k · (cid:126)A(kx, ky) = jπ, Cs tization of the Zak phase γs =(cid:72) where j is an integer. This result follows from the quan- d(cid:126)k · (cid:126)A(kx, ky) associ- ated with the band crossing in multiples of π (see Ap- pendix A). Thus, for a semimetal, the Zak phase γ(k) is quantized, but changes by jπ as one encloses a jπ- Berry phase degeneracy. Henceforth, for both 2D insula- tors and semimetals, we can introduce a topological Z2 Cs (5) (a) ξ(0)ξ((cid:126)b2/2) = +1 (b) ξ(0)ξ((cid:126)b2/2) = +1 (c) ξ(0)ξ((cid:126)b2/2) = −1 (d) ξ(0)ξ((cid:126)b2/2) = −1 3 TABLE II: (Color online) Four different band structures for a two-band model, with both I and T symmetry. Bulk (edge) states are indicated by solid (dashed) black (red) lines. (a) Spectrum of an insulator with a trivial Z2 invariant; (b) Dirac semimetal with a trivial Z2 invariant; (c) Again an insulator, but now with a non-trivial Z2 invariant; (d) Dirac semimetal with a non-trivial Z2 invariant. The parameters used for these four dispersion relations are given in App. C. invariant χ1 protected by I and T symmetry, which re- flects the fact that the Zak phase γ(k) is quantized for all values of k. V. BULK-BOUNDARY CORRESPONDENCE Next, we discuss the bulk-boundary correspondence, which relates the Z2 invariant χ1 to the edge behavior. For this purpose, we consider a two-band toy model on an oblique lattice. Specifically, we study a system with edges as in Fig. 1(a), for which every site hosts an s and a p orbital. The corresponding bulk Hamiltonian can be written as H((cid:126)k) = hI ((cid:126)k)1 + hy((cid:126)k)σy + hz((cid:126)k)σz, where hI , hy, and hz are real-valued even functions due to the constraints in Eq. (1), and σi are the Pauli matrices. Moreover, in this basis the inversion operator is given by I = σz, such that Is(cid:105) = s(cid:105) and Ip(cid:105) = −p(cid:105). At half- filling, we have ξ((cid:126)kinv) = sgn[hz((cid:126)kinv)]. Hence, we can express the Z2 invariant as χ1 = sgn[hz(0)]sgn[hz((cid:126)b2/2)]. (6) If one only includes nearest-neighbor hopping, the bulk Hamiltonian is specified by hI + hz = es + 2ts,1 cos((cid:126)k · (cid:126)a1) + 2ts,2 cos((cid:126)k · (cid:126)a2), hI − hz = ep + 2tp,1 cos((cid:126)k · (cid:126)a1) + 2tp,2 cos((cid:126)k · (cid:126)a2), hy = 2tsp,1 sin((cid:126)k · (cid:126)a1) + 2tsp,2 sin((cid:126)k · (cid:126)a2), where es and ep denote the on-site energies, ts,i and tp,i are the nearest-neighbor-hopping parameters in the di- rection i, and tsp,i is the hybridization among s and p orbitals. In Table II, we display four spectra, which are realized for values of the above parameters chosen ad hoc, in a way to provide an example of the four qualitatively different scenarios (see Appendix C). We will shortly dis- cuss below how their distinct features can be understood in terms of the Z2 invariant. (i) Gapped bulk with a trivial Z2 invariant: For χ1 = +1, and in the absence of band-crossing points, the Zak phase γk = 0 for all k. In this case, the trivial value of the Zak phase for all values of the momentum is reflected in the absence of edge states. This behavior is confirmed by plotting the spectrum for an insulator with χ1 = +1 in Table II(a), which only shows bulk states. (ii) Gapless bulk with a trivial Z2 invariant: The combination of a trivial Z2 invariant, χ1 = +1, and two π-Berry phase band-crossing points yields a trivial Zak phase γ(k) = 0 for momenta adiabatically con- nected to 0, and a non-trivial Zak phase γ(k) = π for k outside this region. Hence, for momenta contained in the latter region, the bulk topology gives rise to edge states. This is indeed the case for the spectrum shown in Table II(b), which corresponds to a Dirac semimetal with χ1 = +1. (iii) Gapped bulk with a non-trivial Z2 invariant: For a non-trivial insulator, with χ1 = −1, the Zak phase γ(k) = π for all k. The non-trivial Zak phase man- ifests itself via the presence of edge states for all mo- menta. This is verified in the example shown in Ta- ble II(c) (where χ1 = −1), which features in-gap edge states for all momenta. (iv) Gapless bulk with a non-trivial Z2 invariant: Finally, we consider a gapless system with a non-trivial invariant, χ1 = −1. Then, the Zak phase γk = π for the values of k that are adiabatically connected to k = 0, whereas for k outside this region γ(k) = 0. Hence, we expect the presence of edge states for k that are adia- batically connected to k = 0. The band structure shown in Table II(d) confirms this expectation. Thus, for insulators the Z2 invariant determines the presence or absence of edge states, whereas for a Dirac semimetal it encodes whether the two Dirac cones are connected by edge states which go through zero or π. In particular, we might view Dirac semimetals as systems that interpolate between different Z2 insulators. More- over, our results show that edge states are a robust fea- ture of Dirac semi-metals. The presence of edge states is guaranteed, as long as the two Dirac cones are separated in the one-dimensional Brillouin zone. -3-2-1012ka-10-5051015E(ka)3-3-2-1012ka-10-5051015E(ka)3-3-2-1012kaE(ka)3-10-5051015-3-2-1012ka-10-5051015E(ka)34 tiples of 2e. In principle, this 2e ambiguity allows for dif- ferent right- and left-surface charges; however, when the open chain is I-symmetric, we find that σleft = σright. Hence, the net charge present on the open chain is then Qnet = σright+σleft = 2σright modulo 4e. The relation be- tween the net charge and in-gap states can be brought to light using the following adiabatic continuity argument, which we illustrate for a Su-Schrieffer-Heeger chain that we put in a ring geometry, as depicted in Fig. 2(a). Elec- trons on the chain can hop with hopping parameters t and t(cid:48), corresponding to the single and double bonds. For simplicity, we set t = −0.8eV and t(cid:48) = −1.2eV. In addition, we consider a weak link with hopping parame- ter s, that can be varied. The presence of the weak link defines a preferential unit cell, which contains a single bond in our example. The Zak phase is then γ = π. From the boundary-charge theorem, it follows that, for an open chain (s = 0), both ends of the chain will have a surface charge σ = e. Therefore, the net charge will be 2e modulo 4e. This implies, that if one adiabatically changes the weak-link hopping parameter from s = t(cid:48) to s = 0, an odd number of spin-degenerate states crosses the Fermi level: an in-gap state must have appeared, see Figs. 2(b) and (c). This does not guarantee that in-gap states will be present at the end of the adiabatic defor- mation (s = 0). In-gap states can dissolve into the bulk, whereas they are pinned at zero energy in systems with particle-hole symmetry. To demonstrate this statement, we consider the band structure shown in Fig. 3(a). The corresponding bulk Hamiltonian is still described by the two-band toy model, with χ1 = −1 (see Appendix D for the details). The in-gap edge states are absent for k = 0, while for suffi- ciently large values of k, they emerge from the bulk. Hence, the in-gap state that must have traversed the band gap leaves no clear trace in the spectrum at k = 0. FIG. 2: (Color online) (a) Sketch of a periodic Su-Schrieffer- Heeger chain, which is adiabatically deformed into a chain with open ends. The parameter s is the hopping parameter that connects the two blue unit-cells. (b) Net charge present on the chain as a function of the hopping parameter s, which is adiabatically turned off. (c) Corresponding spectrum. Here, we have considered a chain containing 100 unit cells. VI. EDGE CHARGE QUANTIZATION We now explain the deep connection between the Z2 invariant χ1 and the emergence of in-gap states. It is a direct consequence of the boundary-charge theorem,13 which states that for a one-dimensional crystalline spin- degenerate insulator the surface charge σ is well defined modulo 2e, and is given by + e Zjuj + e (7)  eγ π σ = ± N(cid:88) j=1  . (cid:88) Zj j∈surf Here, e is the electron charge, γ is the Zak phase, N is the total number of atoms within a unit cell, and Zj and uj denote, respectively, the atomic number and position of the jth atom within the unit cell. The second term vanishes for I-symmetric insulators. Moreover, the third term counts the total ionic charge of the atoms contained in the set "surf". These are the atoms that remain at the edges when tiling the finite system with unit cells. Note that this term precisely cancels the unit-cell ambi- guity stemming from the first two terms. The ± refers to the left and right surface charge. In the presence of I-symmetry, σ is quantized to 0 or e modulo integer mul- FIG. 3: (Color online) (a) Spectrum of a two-band model with χ1 = −1. (b) Net charge density ρ(i) as a function of the lattice site for k = 0 and chain length N = 120. (c) IPR of one of the in-gap states at k = π/a. (d) IPR of the highest-valence band states at k = 0. ✂s=t´-202EFt´-st´0.51.0spectrumnet charge02(a)(b)(c)tt´s=t´/2s=0-3-2-10123ka-3-2-1012E[ka]02468612k=01/N (10-3) 1/N (10-3) IPR[1/N] (10-3) IPR[1/N](a)(b)(c)(d)0.01.0020406080100iρ[i]k=0N=120k=π/a024680.50.250.5Despite the trivial spectrum at k = 0, we find that the non-trivial topology is captured by the net charge dis- tribution ρ(i), as shown in Fig. 3(b), thereby verifying the validity of the boundary-charge theorem, Eq. (7). In particular, it follows that for 2D insulators in the pres- ence of I- and T -symmetry, the edge charge associated with Hk is either 0 or e, as numerically confirmed in Fig. 3(b). To verify the absence of bound states at k = 0, we have analyzed the inverse participation ratio (IPR), which quantifies over how many sites a particular state is distributed.18 For a given state Ψ(cid:105) in a 1D system, the IPR is defined as IP R(Ψ(cid:105)) = (cid:104)i, αΨ(cid:105)4, (8) (cid:88) i,α where i, α(cid:105) denotes the state localized at site i, and α labels the orbital. For a proper bulk state, the IPR as a function of the chain length N should be proportional to 1/N , whereas for an edge state the IPR goes to a constant value. In Fig. 3(c), we plot the IPR for one of the in-gap states at k = π/a. We can clearly see that this is indeed an edge state. In contrast, in Fig. 3(d) we plot the IPR for the highest-valence band state at k = 0, which can be identified as a bulk state. VII. FEW-LAYER BLACK PHOSPHORUS Next, we apply our results to an actual 2D material: black phosphorus, which consists of stacked sheets cou- pled by Van der Waals forces. Recently, it has been possi- ble to isolate individual layers of black phosphorus (phos- phorene) by mechanical exfoliation.19–23 In Figs. 4(a) and (b), we display a sketch of such a layer. Like graphene, phosphorene has a honeycomb lattice; how- ever, the bonds in phosphorene result from sp3 hybridiza- (b) Top view. FIG. 4: (Color online) (a) Bird's eye view of single-layer black phosphorus. (c) Bulk band structure along high-symmetry lines, where the ± signs indicate the parities of the Bloch waves. (d), (e), and (f) Spectra of rib- bons single-layer black phosphorus with zigzag, bearded, and armchair edges, respectively. 5 tion, which leads to the puckered structure. Phosphorene is a semiconductor, with a band gap that goes from 0.3eV to 2eV, depending on the number of layers. This sizable gap makes it a very promising material for electronic ap- plications. Here, we demonstrate that this novel material fea- tures edge states. For this purpose, we use a sim- ple tight-binding model, where each atom hosts one pz- like orbital.24 The model can be used to describe both valence- and conduction-band edges in black phosphorus. For a single layer, we can write (cid:107) i,jc (cid:88) † i cj, H = (9) t i(cid:54)=j † i (ci) creates (annihilates) an electron at site i, where c (cid:107) and t i,j denotes the hopping parameter from site i to site j. In Fig. 4(c), we plot the resulting band structure along the high-symmetry lines connecting the X − Γ− Y points, evaluated for the hopping parameters obtained from first-principles calculations.24 This parametrization includes in total ten different parameters, although qual- itatively only the two nearest-neighbor-hopping parame- 1 = −1.486eV and t ters t 2 = 3.729eV, shown in Fig. 4(a), Inspection of Fig. 4(b) indeed confirms are required. that phosphorene exhibits I symmetry around the center of the unit-cell. In particular, if we label the inequiva- lent sites in the unit-cell as in Fig. 4(b), we find that I = σx ⊗ σx. The ± signs in the band structure denote the eigenvalues of I. Using these eigenvalues, we can eas- ily calculate the Z2 invariants χx and χy associated with edges along the x and y directions, respectively; we find χx = χy = +1. From this result, we can immediately infer that both bearded and armchair phosphorene [see Fig. 4(b)], do not feature edge states. However, zigzag- terminated phosphorene cannot be tiled with an integer number of unit-cells. Therefore, we cannot simply infer the presence of edge states from χy. Instead, Eq. (7) states that we need to account for the ionic charge e con- tained in the broken unit cell at the edge, see Fig. 4(b). The edge charge is given by σ(k) = eγ(k)/π + e = e. Hence, we expect the presence of edge states for zigzag- terminated phosphorene. These conclusions are con- firmed by plotting the three different spectra, see Figs. 4(d), (e), and (f). Here, only zigzag-terminated phospho- rene exhibits one pair of in-gap edge states, which can be attributed to the edge charges (shown in red). We can repeat this analysis for bilayer black phospho- rus, for which the structure is shown in Figs. 5(a) and (e). Note that the second layer is displaced by half a lattice vector in the zigzag direction, but the stacking respects the I symmetry. One can also study this system using a tight-binding model (cid:107) † i,jc t i cj + t⊥ i,jc † i cj, (10) (cid:88) i(cid:54)=j H = (cid:88) i(cid:54)=j where we have now included interlayer hopping t⊥ i,j. In Fig. 5(a), we have pictured the dominant interlayer hop- -6-4-2246xyzigzagbeardedarmchair630-3-6++−−±±±±630-3-6XYΓπ/ay-π/ay0Energy(a)(b)(c)(d)(e)(f)kkkπ/ay-π/ay0π/ax-π/ax0Energy31246 FIG. 5: (Color online) (a) Bird's eye view of bilayer black phosphorus. (b) , (c), and (d) Spectra of a ribbon of bilayer black phosphorus with zigzag, bearded and armchair edges, respectively. (e) Top view of bilayer black phosphorus. (f), (g), and (h) Spectra of a biased ribbon of bilayer black phosphorus (∆V = 2eV), with zigzag, bearded and armchair edges, respectively. (i) and (j) Bulk band structure along high-symmetry lines, for bilayer black phosphorus with ∆V = 0 and ∆V = 2eV, respectively. (k), (l), and (m) close-up of Figs. (f), (g), and (h) around the band-crossing point. ping t⊥ 1 = 0.524eV. The resulting bulk-band structure is shown in Fig. 5(i). The two bulk Z2 invariants are trivial, χx = χy = +1. This implies that both bearded and armchair bilayer ribbons do not exhibit edge states. For zigzag bilayer phosphorene, we still need to account for the ionic charge in the broken unit-cell at the edge. Inspection of Fig. 5(e) reveals that this contribution is equal to 2e. However, the surface charge is only well defined modulo 2e, and thus we expect that all three ter- minations are topologically trivial. This is confirmed by plotting the three spectra, see Figs. 5(b), (c), and (d). Both armchair and bearded bilayer phosphorene do not feature any in-gap states, whereas the zigzag terminated one exhibits two pairs. Although bearded and zigzag ter- minated bilayer phosphorene feature very different edge physics, they are topologically identical, owing to the Z2 nature of the invariant. It has been noticed in ab initio calculations that a potential bias ∆V applied between the two layers can drastically affect the band structure.25 In particular, one can induce a Lifshitz transition if ∆V exceeds a critical value. Then, the valence and conduction band invert, and a band-crossing point emerges along the line connecting the Γ−Y points, whereas a small gap is opened along the line connecting the Γ−X points, see Fig. 5(j). This band inversion is accompanied by a topological phase transi- tion, such that now χx = χy = −1. Hence, by varying this bias potential one transforms an insulator with a trivial Z2 invariant, as in Table II(a), into a semimetal with a non-trivial Z2 invariant, as in Table II(d). In Figs. 5(f) and (g), we plot the spectra for zigzag and bearded bilayer phosphorene. Here, one can indeed see 005xyzigzagbeardedarmchairarmchairzigzagbearded-6-3036XYΓ−±±±±±±±±−−−++++-3-2-1123-8-6-4-2246++−−±±±±XYΓ-6-3036(a)(b)(c)(d)(k)(l)(m)(e)(i)(j)π/axπ/axπ/ayπ/aykkkkkkEnergyEnergyEnergyEnergyπ/ayπ/ay-π/ax-π/ax-π/ay-π/ay-π/ay-π/ay000000(f)(g)(h)-505-55the in-gap edge states located between the two band- crossing points [see also the close-ups in Figs. 5(k) and (l)]. These spectra are qualitatively similar to the band structure shown in Table II(d). However, for the rib- bon with armchair termination, the spectrum does not exhibit any edge states, see Figs. 5(h) and (m). This be- havior is grounded on the fact that for this termination, the two band-crossing points coincide at k = 0. This ex- ample provides a good illustration of our claim that edge states are a robust feature of Dirac semimetals, and that their existence can be attributed to the Zak phase γ(k), which changes from π to zero as one traverses the band- crossing point. Although we have limited ourselves to single and bilayer black phosphorus, our conclusion can easily be generalized to other few-layer configurations. VIII. CONCLUSION AND DISCUSSION In conclusion, we show that the interplay between T - and I-symmetry gives rise to a topological Z2 invariant χ1, which is directly related to the quantization of the Zak phase γ(k) in both insulators and semimetals. In particular, we find that a non-trivial Zak phase generally leads to edge states. Hereby, we have generalized the re- sult by Ryu and Hatsugai [10] to systems lacking chiral symmetry. Moreover, we have extended the usual classifi- cation of 2D I-symmetric insulators given in Ref. [15,26]. These results are relevant for a broad range of 2D ma- terials, including graphene, phosphorene and their multi- layer configurations. Our results explain the robust topo- logical origin of edge states in Dirac semimetals, due to the π Berry phase of the Dirac cone. This work, therefore, complements earlier studies on edge states in graphene27 based on the Dirac equation. We note that silicene,35 germanene, stanene, and transition-metal dichalcogenides, which also exhibit similar properties to the previous materials, are excluded from our analysis be- cause of a significant SOC and/or the lack of I-symmetry. Experimentally, the presence of edge states may be most easily detected via scanning-tunneling microscope (STM) experiments, which probe the local density of states.28 For an insulator, the excess density of states at the surface will be quantized, whereas for semimet- 7 als it will be proportional to the distance between the two band-crossing points in the reduced 1D BZ.11 This is particularly relevant for few-layer phosphorene, where a gate voltage can induce an insulator to semimetal transition.25 In the semimetallic regime, the gate voltage controls the distance in momentum space between the two band-crossing points, and as such it provides new experimental possibilities to verify our predictions. Ex- perimentally, this insulator to semimetal transition has already been realized by depositing potassium atoms on black phosphorus.29,30 Therefore, we hope that our work will motivate future STM experiments in few-layer black phosphorus. In order to avoid contamination, one should cleave the black phosphorus and perform the STM ex- periments in an ultra-high vacuum environment.29 We would still like to comment on the role of disor- der. It has been shown that in 1D insulators, the surface charge is immune to disorder near the edges.31 More- over, the edge charges are stable against small amounts of disorder in the bulk, which preserve I-symmetry on average.32,33 Finally, we would like to point out that the relevance of our results is not restricted to 2D materials because the Zak phase has been recently used to explain the exis- tence of drumhead surface states in the three-dimensional materials Cu3N and Ca3P2.32,34 IX. ACKNOWLEDGEMENTS G.v.M. and C.M.S. acknowledge financial support from NWO and the Dutch FOM association with the pro- gram Designing Dirac carriers in semiconductor hon- eycomb lattices. C.O. acknowledges the financial sup- port of the Future and Emerging Technologies (FET) programme within the Seventh Framework Programme for Research of the European Commission under FET- Open grant number: 618083 (CNTQC), and Deutsche Forschungsgemeinschaft under Grant No. OR 404/1-1. This work is part of the D-ITP consortium, a program of the Netherlands Organization for Scientific Research (NWO) that is funded by the Dutch Ministry of Educa- tion, Culture and Science (OCW). Appendix A: Properties of Berry connection and curvature In the presence of T -symmetry, the Berry connection is even up to a total derivative. For simplicity, we consider the case where the T -operator is represented by complex conjugation K. Then, A(k) = i(cid:104)uk∇kuk(cid:105) = i(cid:104)Ku−ke−iφ(k)∇keiφ(k)Ku−k(cid:105) = i(cid:104)Ku−k∇kKu−k(cid:105) − ∇kφ(k) = −i(cid:104)u−k∇ku−k(cid:105) − ∇kφ(k) = i(cid:104)u−k∇−ku−k(cid:105) − ∇kφ(k) = A(−k) − ∇kφ(k). In the presence of I-symmetry, the Berry connection is odd up to a total derivative, A(k) = i(cid:104)uk∇kuk(cid:105) = i(cid:104)u−ke−iχ(k) I∇k Ieiχ(k)u−k(cid:105) = i(cid:104)u−k∇ku−k(cid:105) − ∇kχ(k) = −i(cid:104)u−k∇−ku−k(cid:105) − ∇kχ(k) = −A(−k) − ∇kχ(k). Hence, from this it follows that the polarization is quantized for an I-symmetric system, as(cid:82) A(k) = −(cid:82) A(−k) + (cid:82) ∇χ(k) = −(cid:82) A(k) + 2πj where the integral is over a symmetric domain. In the presence of both I and T -symmetry, the integral along any closed contour is quantized, as we find that 2A(k) = ∇χ(k). Finally, at the level of the Berry curvature,T -symmetry dictates F (k) = −F (k), as the curl of an even function is odd, and I dictates F (k) = F (−k) as the curl of an odd function is even. Hence, when both T and I-symmetry are present, we find F = 0. 8 The Zak phase is defined as γ = i(cid:82) dk(cid:104)uk∇kuk(cid:105), where uk is the periodic part of the full Bloch wave function Ψk. Appendix B: Relation between the Zak phase and eigenvalues of inversion In a tight-binding model, we find uk,j,α = e−ik(ja+rα)Ψk,j,α, where j labels the unit-cells, α is an orbital index, and rα the corresponding location with respect to the center of the jth unit-cell. Note that the inner product is restricted to one unit cell. Furthermore, the Zak phase should be calculated for the periodic gauge, which in terms of the full wave function, translates into Ψk = Ψk+2π/a. Using the relation between Ψk and uk, we can then rewrite the Zak phase as γ = i dk(cid:104)Ψ∗ k∇kΨk(cid:105) + dkΨk,α2rα symmetric around the center of the unit cell. Hence, we can write γ = i(cid:82) dk(cid:104)Ψ∗ The second term on the right-hand side vanishes in the presence of I-symmetry, since the charge distribution is k∇kΨk(cid:105). Then, for a system without degeneracies, I-symmetry guarantees that Ψk(cid:105) = e−iφ(k) IΨ−k(cid:105), with φ some arbitrary phase. Hence, we can rewrite (cid:90) 0 the Zak phase as α dk(cid:104)Ψk∇kΨk(cid:105) + i dk(cid:104)Ψk∇kΨk(cid:105) = i dk(cid:104)Ψk∇kΨk(cid:105) + i dk(cid:104)Ψ−k I†eiφ(k)∇ke−iφ(k) IΨ−k(cid:105) (cid:90) (cid:88) (cid:90) −π (cid:90) 0 (cid:90) 0 (cid:90) 0 −π −π π (cid:90) π (cid:90) 0 0 (cid:90) 0 −π dk(cid:104)Ψk∇kΨk(cid:105) + i dk(cid:104)Ψk∇kΨk(cid:105) + i dk(cid:104)Ψ−k∇kΨ−k(cid:105) + dk∇kφ(k) dk(cid:104)Ψk∇kΨk(cid:105) + −π dk∇kφ(k) = φ(0) − φ(−π). (cid:90) π (cid:90) π (cid:90) π 0 0 0 γ = i = i = i In the penultimate step, we used that ∇k = −∇−k. Although the phase φ(k) is arbitrary for generic k, this is not true for I-invariant momenta, as follows from the identity Hence, φ(kinv) = 2πj + [ξ(kinv) − 1]π/2, and thus Ψkinv(cid:105) = eiφ(kinv) IΨkinv(cid:105) = eiφ(kinv)ξ(kinv)Ψkinv(cid:105). γ = φ(0) − φ(−π) = 2π(j − j(cid:48)) + [ξ(0) − ξ(−π)]π/2. Appendix C: parameters for the two-band toy model Table III lists the hopping parameters that have been used to obtain the spectra shown in Table II. Appendix D For the band structure shown in Fig. 2(a), we have used a tight-binding model that includes long-range hopping. The Fourier transformed bulk Hamiltonian reads H((cid:126)k) = hI ((cid:126)k)1 + hy((cid:126)k)σy + hz((cid:126)k)σz, with hI ((cid:126)k) + hz((cid:126)k) = −0.2 − 0.46 cos((cid:126)k · (cid:126)a1) + 2.15 cos((cid:126)k · (cid:126)a2), hI ((cid:126)k) − hz((cid:126)k) = −0.52 − 0.29 cos((cid:126)k · (cid:126)a1) − 0.58 cos((cid:126)k · (cid:126)a2) + 0.6 cos(2(cid:126)k · (cid:126)a2) + 0.3 cos((cid:126)k · ((cid:126)a1 + 2(cid:126)a2)) + 0.3 cos((cid:126)k · (−(cid:126)a1 + 2(cid:126)a2)), hy((cid:126)k) = 1.81 sin((cid:126)k · (cid:126)a2). (a) (b) (c) (d) 9 es ep t1s t2s t1p t2p t1sp t2sp 3 -3 -5 -6 -3 4 1 7 3 -2 -2 -1 2 2 3 1 2 -1 -4 -3 -1 -3 -2 -2 -2 -2 4 3 3 4 3 1 TABLE III: Parameters used for the band structures from Table II. 1 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). Neto, and B. Ozyilmaz, Appl. Phys. Lett. 104, 103106 (2014). 2 X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 22 H. Liu, A. T. Neal, Z. Zhu, D. Tomanek, and P. D. Ye, (2011). ACS Nano 8, 4033 (2014). 3 A. Altland and M. R. Zirnbauer, Phys. Rev. B 55, 1142 23 F. Xia, H. Wang, and Y. Jia, Nat. Commun. 5, 4458 (1997). (2014). 4 A.P. Schnyder, S. Ryu, A. Furusaki, and A.W.W. Ludwig, 24 A. N. Rudenko, S. Yuan, and M. I. Katsnelson, Phys. Rev. Phys. Rev. B 78, 195125 (2008). 5 A. Kitaev, AIP Conf. Proc. 1134, 22 (2009). 6 S. Ryu, A.P. Schnyder, A. Furusaki, and A.W.W. Ludwig, New J. Phys. 12, 065010 (2010). B 92, 085419 (2015). 25 Q. Liu, X. Zhang, L. B. Abdalla, A. Fazzio, and A. Zunger, Nano Lett. 15, 1222 (2015). 26 R.J. Slager, A. Mesaros, V. Juricic, and J. Zaanen, Nature 7 D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. Phys. 9 98 (2013). den Nijs, Phys. Rev. Lett. 49, 405 (1982). 27 A. R. Akhmerov and C. W. J. Beenakker, Phys. Rev. B 8 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 77, 085423 (2008). (2005). 9 R. B. Laughlin, Phys. Rev. B 23, 5632 (1981). 10 S. Ryu and Y. Hatsugai, Phys. Rev. Lett. 89, 077002 (2002). 11 P. Delplace, D. Ullmo, and G. Montambaux, Phys. Rev. B 84, 195452 (2011). 12 M. Ezawa, New J. Phys. 16, 115004 (2014). 13 R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651(R) (1993). 14 J. Zak, Phys. Rev. Lett. 62, 2747 (1989). 15 T.L. Hughes, E. Prodan, B.A. Bernevig, Phys. Rev. B 83, 245132 (2011). 28 C. Pauly, B. Rasche, K. Koepernik, M. Liebmann, M. Pratzer, M. Richter, J. Kellner, M. Eschbach, B. Kauf- mann, L. Plucinski, C. M. Schneider, M. Ruck, J. van den Brink, and M. Morgenstern, Nat. Phys. 11, 338 (2015). 29 J. Kim, S. S. Baik, S. H. Ryu, Y. Sohn, S. Park, B.-G. Park, J. Denlinger, Y. Yi, H. J. Choi, and K. S. Kim, Science 349, 723 (2015). 30 S.S. Baik, K.S. Kim, Y. Yi, and H.J. Choi, Nano Lett. 15, 7788 (2015). 31 J.-H. Park, G. Yang, J. Klinovaja, P. Stano, D. Loss, arXiv:1604.05437 32 Y.-H. Chan, C.-K. Chiu, M. Y. Chou, and A. P. Schnyder, 16 A. Lau, C. Ortix and J. van den Brink, Phys. Rev. Lett. Phys. Rev. B 93, 205132 (2016). 115 216805 (2015). 33 M. Diez, D.I. Pikulin, I.C. Fulga, and J. Tworzyd(cid:32)lo, New. 17 T. Kariyado and Y. Hatsugai, Phys. Rev. B 88 245126 J. Phys. 17, 043014 (2015). (2013). 18 F. Wegner, Z. Phys. B 36, 209 (1980). 19 L. Li, Y. Yu, G. Jun Ye, Q. Ge, X. Ou, H. Wu, D. Feng, X. Hui Chen, and Y. Zhang, Nat Nanotechnol 9, 372 (2014). 20 M. Buscema, D.J. Groenendijk, S.I. Blanter, G.A. Steele, H.S.J. van der Zant, and A. Castellanos-Gomez, Nano Lett. 14, 3347 (2014). 21 S. P. Koenig, R. A. Doganov, H. Schmidt, A. H. Castro 34 Y. Kim, B.J. Wieder, C.L. Kane, and A.M. Rappe, Phys. Rev. Lett. 115, 036806 (2015). 35 Since the SOC in silicene is of the order of 20K, STM experiments must be performed at 4K to resolve the gap. At room temperature the material can be considered, for all effects, as gapless, and hence can be described by the theory presented here. 

1701.03622

2

1701

Robust Single-Shot Spin Measurement with 99.5% Fidelity in a Quantum Dot Array

We demonstrate a new method for projective single-shot measurement of two electron spin states (singlet versus triplet) in an array of gate-defined lateral quantum dots in GaAs. The measurement has very high fidelity and is robust with respect to electric and magnetic fluctuations in the environment. It exploits a long-lived metastable charge state, which increases both the contrast and the duration of the charge signal distinguishing the two measurement outcomes. This method allows us to evaluate the charge measurement error and the spin-to-charge conversion error separately. We specify conditions under which this method can be used, and project its general applicability to scalable quantum dot arrays in GaAs or silicon.

cond-mat.mes-hall

cond-mat

Robust single-shot spin measurement with 99.5% fidelity in a quantum dot array Takashi Nakajima,1, ∗ Matthieu R. Delbecq,1 Tomohiro Otsuka,1, 2 Peter Stano,1 Shinichi Amaha,1 Jun Yoneda,1 Akito Noiri,3 Kento Kawasaki,3 Kenta Takeda,1 Giles Allison,1 Arne Ludwig,4 Andreas D. Wieck,4 Daniel Loss,5, 1 and Seigo Tarucha3, 1, † 1Center for Emergent Matter Science, RIKEN, 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan 2JST, PRESTO, 4-1-8 Honcho, Kawaguchi, Saitama, 332-0012, Japan 3Department of Applied Physics, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan 4Lehrstuhl fur Angewandte Festkorperphysik, Ruhr-Universitat Bochum, D-44780 Bochum, Germany 5Department of Physics, University of Basel, Klingelbergstrasse 82, 4056 Basel, Switzerland (Dated: July 21, 2017) Abstract We demonstrate a new method for projective single-shot measurement of two electron spin states (singlet versus triplet) in an array of gate-defined lateral quantum dots in GaAs. The measurement has very high fidelity and is robust with respect to electric and magnetic fluctuations in the environment. It exploits a long-lived metastable charge state, which increases both the contrast and the duration of the charge signal distinguishing the two measurement outcomes. This method allows us to evaluate the charge measurement error and the spin-to-charge conversion error separately. We specify conditions under which this method can be used, and project its general applicability to scalable quantum dot arrays in GaAs or silicon. 7 1 0 2 l u J 0 2 ] l l a h - s e m . t a m - d n o c [ 2 v 2 2 6 3 0 . 1 0 7 1 : v i X r a 1 Improving measurement fidelities of qubits is an important step to progress with quantum technologies. Apart from being one of the basic constituents of quantum computation[1], or even means to perform it[2], precise measurements of qubits are indispensable for error correction protocols[3–5], or any feedback method in general [6]. Suppressing measurement errors also boosts sensitivity and time resolution of sensors[7, 8] and, by allowing the ma- nipulations to be performed with less averaging and thus faster, can directly enhance the qubit quality factor[9, 10]. For spin qubits in gate-defined quantum dots, which are among prime candidates to realize scalable qubits in solid state [11–13], the first single-shot measurements of a spin-half qubit exploited the spin-dependent energy and tunnel rate and reached fidelities around 80-90%[14, 15]. Later, the development of the rf-reflectometry technique[16] permitted to use the Pauli spin blockade [17, 18] for a single-shot measurement of a singlet-triplet qubit in double quantum dots with 90% fidelity[19]. This was further advanced by optimizing the charge sensor sensitivity[20] up to the recent value of 98% reported in Ref. [9]. Despite the impressive progress, quantum dot spin qubits have been falling short in this respect to other systems, most notably those based on nuclear spins of impurities accessed electrically [21] or optically [22]. To further increase fidelity is not easy, as the signal- to-noise ratio of the charge sensor is limited by the electrical noise in the measurement circuitry and the short lifetime of the spin-blockade state. The latter issue becomes even more serious in the presence of a micromagnet-induced field gradient, which is necessary for fast [23] and addressable [24] spin manipulations. More importantly, the lifetime of the state being detected is sensitive to both electric and magnetic disturbances, which can drastically degrade the measurement fidelity [25]. Here we implement a single-shot measurement distinguishing two-electron spin states (singlet S versus spin-unpolarized triplet T0) inside a quantum dot array with 99.5% fidelity. It relies on the Pauli spin blockade, but using a different spin to charge conversion, first identified in Ref. [26]. We find that the method leads to a substantial fidelity boost and is robust with respect to environmental fluctuations, both magnetic and electric. This demonstrates that electronic spin qubits can reach measurement fidelities comparable to the highest achieved in solid state, and above the threshold for fault-tolerant quantum computing[5, 27], without sacrificing their essential advantages of speed [28] and scalability [29, 30]. Furthermore, the high-fidelity measurement allows us to unravel the underlying 2 FIG. 1. (a) SEM micrograph of a device similar to the one measured. An array of quantum dots is fabricated with the proximal charge sensor and a top cobalt micromagnet layer (orange-shaded area). The dot Q1 (dashed white circle) is idle and singly occupied throughout the measurement, while the spin pair in Q2 and Q3 (filled white circles) is measured. The fourth dot is not used. (b) Charge stability diagram around the (111)-(102) transition, taken with the application of the I/O/M pulse cycles shown in the inset. The positions of the initialization (I), operation (O), and measurement (M, R) configurations are denoted by circles along with the detuning axis ε. The main panel shows VCDS = VM − Vref, the difference of the charge sensor signal before (Vref) and after (VM) pulsing to O to cancel out the slow drift and the smooth landscape of the background signal. (c) Energy spectrum (solid lines) as a function of ε for the states labeled according to their charge and spin. The dotted arrows show transitions upon gate pulses and the system can make relaxations (solid arrows) with the corresponding rates labeled during each step. (d) Energy configurations in each pulse step. mechanism of the spin-to-charge conversion error, which is generally present but has been obscured in spin-blockade measurement. The device is a gate-defined array of quantum dots fabricated on a GaAs/AlGaAs het- erostructure with a charge sensor and a cobalt micromagnet on the top, as shown in Fig. 1(a). It was placed in a dilution refrigerator with a base temperature of ∼ 20 mK and an in-plane magnetic field of Bext = 0.7 T was applied. The three left-most dots (Q1-Q3) are used, while 3 Cobalt MMQ1Q2Q3P1P2P3Charge sensor(111)(102)(112)!↑#↑$↑%MOIabc&'()200nm↑#S%↑#↓$↑%↑#↑$↓%MOI↑#↑$S%(112)(102)MOIt,-'.,/!0-1232)1##$1#RR(112)0-1#(102)M1:O:R:M2:dthe gate electrodes for the fourth dot are grounded. The dot Q1 is kept singly occupied and decoupled from the rest by a high tunneling barrier throughout the experiment. Figure 1(b) shows the relevant part of the charge stability diagram taken as a function of DC gate voltages on P1 and P3, with applying the voltage pulse cycles of the shape shown in the inset. The high-fidelity measurement of the spin states is realized in the bright triangular region shown in Fig. 1b by using the metastable (112) charge state, as discussed later in detail. The energy spectrum along the black line with an arrow (parameterized by ε, the detuning energy between Q1 and Q3) is given in Fig. 1(c). The standard single-shot measurement based on the spin blockade works as follows. A gate voltage pulse is applied to alternate between (102) and (111) charge configurations, denoted in Fig. 1(b) and (c) by R and O, respectively. Waiting in R (reset), the system relaxes into its ground state σ1S3(cid:105), meaning the first dot contains an electron with spin σ1, the second dot is empty and the third dot is occupied by the singlet. Pulsing from here to O (operation), the system starts singlet (S)-triplet (T0) precession. Here S and T0 are coherent superpositions of two system eigenstates, σ1 ↑2↓3(cid:105) and σ1 ↓2↑3(cid:105), split in energy according to the gradient of the micromagnet field ∆B23. The precession, represented in Fig.1(c) by an orange circular arrow, is ended by pulsing back to R. It converts the spin to charge information, because S goes adiabatically (in the nanosecond ramp time determined by the circuit bandwidth) over to (102), while T0 will remain in (111) until it decays to (102) by a nearest-neighbor hopping. Since the latter transition requires a change of the spin, it is slow enough (typically microseconds) such that the charge sensor can distinguish (111) from (102) and thus the single-shot spin measurement is accomplished. The measured histogram of the charge sensor signal integrated over time tM = 4 µs after pulsing into R is plotted in Fig 2(a). It is well fitted by assuming that it originates from two discrete values VT and VS, assigned to (111) and (102), smeared by the Gaussian noise of the charge sensor [19]. While integrating the signal longer averages out the noise, it also leads to an overall shift of the signal towards VS, because of the finite lifetime of (111), T1. The latter can be found from the time dependence of the mean value of the sensor signal. This is plotted in Fig. 2(c), and an exponential fit gives T1 ≈ 9 µs. Therefore, there is an optimal integration time tM and a threshold voltage Vth which maximizes the contrast by minimizing the overlap of the two Gaussian-like distributions forming the histogram. This overlap is the infidelity (one minus the fidelity) of the specific charge measurement, being 4 the measure of the reliability with which one can discriminate the system being initially in (111) versus (102). For the data plotted in Fig. 2(a) the fidelity is 83.8 ± 0.8%. Once the (102)/(111) charge state is identified, it is interpreted as the spin singlet/triplet measurement outcome. However, we stress that the fidelities of the charge and spin mea- surement are not identical, since the spin measurement fidelity is further diminished by the fidelity of the spin to charge conversion. Our measurement scheme explained below signif- icantly improves the charge measurement fidelity. Its robustness against the magnetic and electric fluctuations also allows us to separately analyze the spin measurement fidelity as discussed below. The high-fidelity single-shot measurement is performed at the readout point M inside the bright triangular region shown in Fig. 1(b), by taking advantage of the presence of an excited, additionally charged, state (112). As shown in Fig. 1(c), the energy of this state is below that of (111) at M. Upon applying the measurement pulse from O, the singlet goes over to (102) as before, but the charge sequence for a triplet changes. It first loads (with a fast rate τ−1 r ) an additional electron from the lead into Q3, going to (112), before relaxing (with a slow rate T −1 112) to the system ground state (102). This has two decisive advantages. First, the charge states to be distinguished differ by the total number of electrons in the system [(102) versus (112)] and not just by their position [(102) versus (111)], which gives a larger signal contrast. Second, the lifetime of the metastable state is longer, which diminishes the shift of the triplet signal due to the relaxation in a given integration time[31]. The resulting improvement is clearly visible from the histogram in Fig. 2(b), the analog of Fig. 2(a), with fidelity 99.68± 0.06%. The fidelity boost is possible because of the hierarchy of the relaxation times, T112 (cid:29) T1 (cid:29) τr, which is easily realized in quantum dot arrays. The lifetime of (112), T112, is large [0.7 ms here, see Fig. 2(d)], because the relaxation from (112) to (102) requires to remove an electron from a dot without a direct access to a lead, a next nearest-neighbor tunneling. On the other hand, the relaxation of the triplet in (111) to the singlet in (102) is caused by a nearest-neighbor tunneling accompanied by spin mixing, resulting in T1 ≈ 9 µs [see Fig. 2(c)]. Finally, since the loading of an extra electron from a lead to Q3 is blocked by neither spin nor charge, τr is the smallest. The value of τ−1 is well r (cid:29) 10 MHz, and we above the measurement bandwidth of the charge sensor such that τ−1 expect τ−1 to be equal to the Q3-lead tunnel rate, which can easily reach 100 MHz[32]. r r The infidelity of the charge measurement is of the order of the small ratios τr/T1 and 5 FIG. 2. (a-b) Histograms of the single-shot signals in configuration (a) R ((111) readout; VS = 158.5 mV, VT = 171.7 mV) and (b) M ((112) readout; VS = 157.9 mV, VT = 177.7 mV). Thin solid lines show Gaussian distributions for S and T0 that would have been observed if no relaxation occurred. Dashed vertical lines show the threshold voltages. (c-e) Waiting time dependence of the mean signal VM at (c) R (d) M, and (e) I. The values of VM differ from those in (a-b) due to the change of the charge sensor condition. The curves are fits to A + B exp(−t/T ) with the corresponding relaxation time T given in each panel. tM/T112. Since we estimate τr/T1 < 10−3 and tM/T112 ≈ 5 × 10−3, the latter dominates the infidelity in the present setup. Employing the theory developed in Ref. [19] allows- from a fit to the histogram-to both evaluate the fidelity and optimize it by choosing the proper integration time tM and the threshold voltage [denoted by vertical dashed lines in Fig. 2(a),(b)] used to assign the binary result. This is how we arrived at the value 99.68%, and dependence of the maximal fidelity on the metastable state lifetime is further illus- 6 ecdST0(112)(111)(102)S(102)T0abtrated in the Supplemental Material (SM)[33]. More importantly, the condition τr/T1 (cid:28) 1 makes the measurement fidelity insensitive to modest variations of T1 due to fluctuations of the Overhauser field and electrostatic potential[25]. This insensitivity to T1 makes our measurement robust throughout a long-term experimental run, which is a major advantage. Despite the long lifetime of (112), one can perform the spin initialization by inserting an additional pulse step positioned at I in Fig. 1(b). This takes advantage of the increased efficiency of the relaxation at the degeneracy of (112) with (111)[34], which is visible as the bright line (larger signal) along the edge of the triangular readout region in Fig. 1(b). The corresponding relaxation time is fitted to Tinit ≈ 3 µs from the data shown in Fig. 2e, being more than three times smaller than T1. To evaluate the fidelity of spin measurement, however, one has to consider additional errors arising in its conversion to charge by pulsing from O to M. The high-fidelity charge measurement developed here allows us to study this effect separately. The dominant source of errors is the deviation from the pulse being perfectly adiabatic with respect to (111) and (102) singlet-singlet anticrossing. Using the Landau-Zener formula, the probability to move through a state crossing non-adiabatically would give this error as (cid:18) (cid:19) pn ∼ exp −2πt2 c∆
∆t , (1) where 2tc is the energy splitting at the anticrossing, and ∆
is the change of the energy difference of the crossing states during the pulse time ∆t. Additional errors, such as photon- assisted charging, spin decay by co-tunneling, or spin relaxation by phonon-emission are, first, not specific to the measurement pulse, and, second, we find these negligible compared to pn based on estimates given in SM[33]. Instead of estimating pn from Eq. (1), we directly measure it. To this end, we set up a rate-equation model (see section II of SM[33]) for the previously described I→O→M cycle and derive PS(t) = a + v 2 e −(t/T ∗ 2 )2 cos(ωt + φ) + c e−Γt, (2) as the probability to measure signal 'S' after the S-T0 precession with an angular frequency ω for a duration t, with φ an additional phase shift and T ∗ 2 the ensemble dephasing time. The idea is that the same non-adiabaticity as the one causing the error in the spin to charge conversion, pn, results in errorneous initialization to the excited (102) state rather than the (111) singlet state at O [see Fig. 3(a)]. If this excited state lifetime 1/Γ is relatively long, as 7 FIG. 3. (a) Schematics of the transition through the singlet S-S3 (black lines) anticrossing. The blue horizontal line is the energy of the (111) triplet T0. Notation similar to Fig. 1 is used. (b) Signal oscillation observed in the I→O→M cycle. (c) The probability pn as a function of the pulse ramp time ∆t. A solid line is the fit according to Eq. (1) written in form exp(−t/t0), which gives t0 = 1.16 ns. The inset shows the random fluctuation of ω in the course of measurement. (d) The values of the phase shift φfit obtained from the fit with Eq. (2) versus the values φcalc calculated from pn, Γ, ω, and ∆t using the theoretical model (see SM[33]). is the case here, the imperfect initialization is directly visible as an exponentially decaying signal downshift by c ∝ pn, described by the last term in Eq. (2). Before discussing the other terms of Eq. (2), Fig. 3(b) shows an exemplary data set, together with the fit according to Eq. (2). The downshift of the oscillating signal with t is apparent and allows us to extract pn and Γ. The fit results in Γ = 13.8 ± 4.5 MHz which complies very well with a microscopic model of the quantum dot (see SM). As shown in Fig. 3(c), we find that pn is suppressed substantially by slowing down the pulse ramp 8 ab↑#S%431−73Γcd↑#↓$↑%↑#↑$↓%between O and M. The observed dependence on the pulse ramp time ∆t follows the scaling suggested by Eq. (1). By suppressing the non-adiabaticity error to the value fitted for ∆t = 8 ns to be pn ≈ 2 × 10−3, we arrive at the spin measurement fidelity of 99.5%, with the 0.2% error of the spin to charge transfer[35] and 0.3% error of the charge readout. This constitutes our main result. We now turn to the remaining parameters of Eq. (2). Figure 3(d) shows the fitted phase shift φ. We find that it is dominated by the phase acquired during the pulse ramp time of ∆t, rather than by a contribution from pn, and therefore does not allow us to independently estimate pn. Similarly, we find that the values of the offset a and the visibility v are much more susceptible to noise and therefore not reliable to estimate other parameters involved in the model, especially the initialization fidelities into various possible states during waiting at I (see SM). We believe this is because of the Overhauser field fluctuations. We take them partially into account in Eq. (2) by introducing the dephasing time T ∗ 2 , appropriate for weak Gaussian noise in ω. However, short acquisition times which we employ to prolong T ∗ 2 [10], at the same time lead to these fluctuations varying non-uniformly over different, or even during, measurements. These fluctuations are not weak, as we estimate that the magnetic field gradient can be sometimes as small as the exchange coupling due to the fluctuations. This leads to changes of the precession axis direction and additional initialization errors[36], which our model resulting in Eq. (2) does not take into account. We would like to make several comments now. First, metastable states such as the one used here are a typical feature found in quantum dot arrays. Second, the presented method is applicable to larger arrays without extensive tuning of the tunnel rates. Third, we stress that the measurement fidelity is stable with respect to the variation of the Overhauser field, which here leads to variations of the precession frequency. Despite the variation of ω/2π in a wide range of 35-95 MHz in the course of the measurement as shown in the inset of Fig. 3(c), we did not find any apparent effects on the histogram in Fig. 2(b). Third, the very long lifetime of the metastable state would enable sequential readout of many spins using a switch matrix and a single transmission line[37], which will be an important technical simplification of the circuitry for large-scale quantum computing. With spin measurement fidelities achieved here, we estimate that 19 qubits can be read out with the fidelity above 90%[33]. Finally, we suppose that it will be possible to increase the measurement fidelity much further by tuning the dot parameters, especially the dot-dot and dot-lead tunnel rates, 9 that are not extensively optimized in this work. Before concluding, let us discuss the results presented here from a broader view. Even though we believe that the achieved high fidelity characterizes the measurement of the spin (and not just a charge), it cannot be strictly proven unless the fidelities in other parts of the experiment-spin initialization and manipulation-are higher than, or at least comparable to, the measurement fidelity[38]. The whole cycle as we do here is aimed at observing the S-T0 oscillations. The qubit initialization, coherent rotation and measurement, taken all together can be regarded as a quantum algorithm, perhaps the most simple one. The overall precision of this specific algorithm is revealed by the visibility of the oscillations, to which imperfections of all parts contribute. Interestingly, we observe a non-monotonic change of the visibility v upon suppressing the measurement errors (see SM), suggesting that fidelities of these other parts are influenced upon changing the pulse time [39]. Nevertheless, a precise measurement is the first requirement for being able to characterize and confirm the suppression of these imperfections, for which many methods have been suggested. In conclusion, we reached 99.5% fidelity of the single-shot spin measurement in a quantum dot array using a metastable state for the charge readout. It has two advantages, a stronger and a longer lived charge signal corresponding to the two possible measurement results. Requirements for using this method are simple, and we therefore find it generally suited for scalable structures of gate-defined quantum dots in GaAs as well as Si. The high-fidelity measurement will bring the spin qubit platform closer to the error-correction threshold and serve as a useful tool for distant quantum communications in which projection measurement onto a 'Bell basis' is essential. We thank the Microwave Research Group in Caltech for technical support. This work is fi- nancially supported by CREST, JST (JPMJCR15N2, JPMJCR1675), the ImPACT Program of Council for Science, Technology and Innovation (Cabinet Office, Government of Japan). TN, TO and JY acknowledge financial support from RIKEN Incentive Research Projects. PS acknowledges financial support from JSPS KAKENHI Grant Number 16K05411. TO ac- knowledges financial support from PRESTO, JST (JPMJPR16N3), JSPS KAKENHI Grant Numbers 25800173 and 16H00817, Strategic Information and Communications R&D Promo- tion Programme, Yazaki Memorial Foundation for Science and Technology Research Grant, Japan Prize Foundation Research Grant, Advanced Technology Institute Research Grant, the Murata Science Foundation Research Grant, Izumi Science and Technology Foundation 10 Research Grant, TEPCO Memorial Foundation Research Grant, The Thermal & Electric Energy Technology Foundation Research Grant. AN acknowledges support from Advanced Leading Graduate Course for Photon Science (ALPS). ST acknowledges financial support by JSPS KAKENHI Grant Numbers 26220710 and JP16H02204. AL and ADW acknowledge gratefully support of Mercur Pr-2013-0001, DFG-TRR160, BMBF Q.com-H 16KIS0109, and the DFH/UFA CDFA-05-06. ∗ Corresponding authors: nakajima.physics@icloud.com † tarucha@ap.t.u-tokyo.ac.jp [1] D. P. DiVincenzo, Fortsch. Phys. 48, 771 (2000). [2] R. Raussendorf, and H. J. Briegel, Phys. Rev. Lett. 86, 5188 (2001). [3] P. W. Shor, Phys. Rev. A 52, R2493 (1995). [4] A. Yu. Kitaev, Ann. Phys. (N.Y.) 303, 2 (2003). [5] A. G. Fowler, A. M. Stephens and P. Groszkowski, Phys. Rev. A 80 052312 (2009) [6] R. B. Griffiths and Ch.-S. Niu, Phys. Rev. Lett. 76, 3228 (1996). [7] M. A. Armen, J. K. Au, J. K. Stockton, A. C. Doherty, and H. Mabuchi, Phys. Rev. Lett. 89, 133602 (2002) [8] J. M. Taylor, P. Cappellaro, L. Childress, L. Jiang, D. Budker, P. R. Hemmer, A. Yacoby, R. Walsworth, and M. D. Lukin, Nature Phys. 4, 810 (2008). [9] M. D. Shulman, S. P. Harvey, J. M. Nichol, S. D. Bartlett, A. C. Doherty, V. Umansky, and A. Yacoby, Nat. Commun. 5, 5156 (2014). [10] M. R. Delbecq, T. Nakajima, P. Stano, T. Otsuka, S. Amaha, J. Yoneda, K. Takeda, G. Allison, A. Ludwig, A. D. Wieck, S. Tarucha, Phys. Rev. Lett. 116, 046802 (2016). [11] D. Loss and D. DiVincenzo, Phys. Rev. A 57, 120 (1998). [12] C. Kloeffel and D. Loss, Annu. Rev. Condens. Matter Phys. 4, 51 (2013). [13] J. M. Taylor, H.-A. Engel, W. Dur, A. Yacoby, C. M. Marcus, P. Zoller, and M. D. Lukin, Nat. Phys. 1, 177 (2005). [14] J. M. Elzerman, R. Hanson, L. H. Willems van Beveren, B. Witkamp, L. M. K. Vandersypen, and L. P. Kouwenhoven, Nature 430, 431 (2004). [15] R. Hanson, L. H. Willems van Beveren, I. T. Vink, J. M. Elzerman, W. J. M. Naber, F. H. L. 11 Koppens, L. P. Kouwenhoven, and L. M. K. Vandersypen, Phys. Rev. Lett. 94, 196802 (2005) [16] D. J. Reilly, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Appl. Phys. Lett. 91, 162101 (2007) [17] K. Ono, D. G. Austing, Y. Tokura, and S. Tarucha, Science 297, 1313 (2002). [18] J. R. Petta, A. C. Johnson, J. M. Taylor, E. A. Laird, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Science 309, 2180 (2005). [19] C. Barthel, D. J. Reilly, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 103, 160503 (2009) [20] C. Barthel, M. Kjaergaard, J. Medford, M. Stopa, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. B 81, 161308(R) (2010). [21] J. J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. L. Morton, F. A. Zwanenburg, D. N. Jamieson, A. S. Dzurak, and A. Morello, Nature 496, 334 (2013). [22] G. Waldherr, Y. Wang, S. Zaiser, M. Jamali, T. Schulte-Herbruggen, H. Abe, T. Ohshima, J. Isoya, J. F. Du, P. Neumann, and J. Wrachtrup, Nature 506, 204 (2014). [23] J. Yoneda, T. Otsuka, T. Takakura, M. Pioro-Ladri´ere, R. Brunner, H. Lu, T. Nakajima, T. Obata, A. Noiri, C. J. Palmstrøm, A. C. Gossard, and S. Tarucha, App. Phys. Exp. 8, 084401 (2015). [24] M. Pioro-Ladri`ere, T. Obata, Y. Tokura, Y. S. Shin, T. Kubo, K. Yoshida, T. Taniyama, and S. Tarucha, Nature Physics 4 776 (2008). [25] C. Barthel, J. Medford, H. Bluhm, A. Yacoby, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. B 85, 035306 (2012). [26] S. A. Studenikin, J. Thorgrimson, G. C. Aers, A. Kam, P. Zawadzki, Z. R. Wasilewski, A. Bogan, and A. S. Sachrajda, Appl. Phys. Lett. 101, 233101 (2012); J. D. Mason, S. A. Studenikin, A. Kam, Z. R. Wasilewski, A. S. Sachrajda, and J. B. Kycia, Phys. Rev. B 92, 125434 (2015). [27] J. M. Martinis, Npj Quantum Inf. 1, 15005 (2015). [28] J. Yoneda, T. Otsuka, T. Nakajima, T. Takakura, T. Obata, M. Pioro-Ladri´ere, H. Lu, C. J. Palmstrøm, A. C. Gossard, and S. Tarucha, Phys. Rev. Lett. 113, 267601 (2014). [29] A. Noiri, J. Yoneda, T. Nakajima, T. Otsuka, M. R. Delbecq, K. Takeda, S. Amaha, G. Allison, A. Ludwig, A. D. Wieck, and S. Tarucha, Appl. Phys. Lett. 108, 153101 (2016). [30] T. Otsuka, T. Nakajima, M. R. Delbecq, S. Amaha, J. Yoneda, K. Takeda, G. Allison, T. Ito, 12 R. Sugawara, A. Noiri, A. Ludwig, A. D. Wieck, and S. Tarucha, Sci. Rep. 6, 31820 (2016). [31] Alternatively, one could integrate signal longer to suppress the noise, though this kind of optimization was not done here. [32] M. G. House, T. Kobayashi, B. Weber, S. J. Hile, T. F. Watson, J. van der Heijden, S. Rogge, and M. Y. Simmons, Nat. Commun. 6, 8848 (2015). [33] See Supplemental Material [url], which includes Refs. [40–43]. [34] A. C. Johnson, J. R. Petta, J. M. Taylor, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Nature 435, 925 (2005). [35] The spin to charge conversion error is most probably even smaller than this value, see the discussion around Eq. (19) in SM[33]. [36] M. D. Shulman, O. E. Dial, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Science 336, 202 (2012). [37] J. M. Hornibrook, J. I. Colless, I. D. Conway Lamb, S. J. Pauka, H. Lu, A. C. Gossard, J. D. Watson, G. C. Gardner, S. Fallahi, M. J. Manfra, and D. J. Reilly, Phys. Rev. Appl. 3, 024010 (2015). [38] Their characterization in turn requires a precise measurement, so that non-trivial bootstrap methods will be necessary in confirming high fidelities of a quantum algorithm. [39] This effect is currently under investigations. [40] F. Baruffa, P. Stano, and J. Fabian, Phys. Rev. Lett. 104, 126401 (2010) . [41] P. Stano, and J. Fabian, Phys. Rev. B 74, 045320 (2006). [42] M. Raith, P. Stano, F. Baruffa, and J. Fabian, Phys. Rev. Lett. 108, 246602 (2012). [43] F. Baruffa, P. Stano, and J. Fabian, Phys. Rev. B 82, 045311 (2010). 13 

1709.07029

1

1709

Spin and charge caloritronics in bilayer graphene flakes with magnetic contacts

We investigate the coupling of spin and thermal currents as a means to rise the thermoelectric efficiency of nanoscale graphene devices. We consider nanostructures composed of overlapping graphene nanoribbons with ferromagnetic contacts in different magnetic configurations. Our results show that the charge Seebeck effect is greatly enhanced when the magnetic leads are in an antiparallel configuration, due to the enlargement of the transport gap. However, for the optimization of the charge figure of merit ZT it is better to choose a parallel alignment of the magnetization in the leads, because the electron-hole symmetry is broken in this magnetic configuration. We also obtain the spin-dependent Seebeck coefficient and spin figure of merit. In fact, the spin ZT can double its value with respect to the charge ZT for a wide temperature range, above 300 K. These findings suggest the potential value of graphene nanosystems as energy harvesting devices employing spin currents.

cond-mat.mes-hall

cond-mat

Spin and charge caloritronics in bilayer graphene flakes with magnetic contacts Instituto de Ciencia de Materiales de Madrid, Consejo Superior de Investigaciones Cient´ıficas, C/ Sor Juana In´es de la Cruz 3, 28049 Madrid, Spain Leonor Chico P.A. Orellana, L. Rosales, and M. Pacheco∗ Departamento de F´ısica, Universidad T´ecnica Federico Santa Mar´ıa, Casilla 110-V, Valpara´ıso, Chile (Dated: May 7, 2019) We investigate the coupling of spin and thermal currents as a means to rise the thermoelec- tric efficiency of nanoscale graphene devices. We consider nanostructures composed of overlapping graphene nanoribbons with ferromagnetic contacts in different magnetic configurations. Our results show that the charge Seebeck effect is greatly enhanced when the magnetic leads are in an antipar- allel configuration, due to the enlargement of the transport gap. However, for the optimization of the charge figure of merit ZT it is better to choose a parallel alignment of the magnetization in the leads, because the electron-hole symmetry is broken in this magnetic configuration. We also obtain the spin-dependent Seebeck coefficient and spin figure of merit. In fact, the spin ZT can double its value with respect to the charge ZT for a wide temperature range, above 300 K. These findings suggest the potential value of graphene nanosystems as energy harvesting devices employing spin currents. I. INTRODUCTION Thermoelectric materials allow for the conversion of heat into electricity and vice versa. Besides the funda- mental interest in thermoelectric properties, that can be analyzed from the general viewpoint of transport phe- nomena, their potential applications for energy recovery or harvesting in processes where a large amount of heat is dissipated are one of the main motivations for their study [1]. The efficiency of thermoelectric devices is related to their figure of merit, defined as ZT = S2GT /κ where S is the Seebeck coefficient or thermopower, G is the electronic conductance, κ the thermal conductance and T is the temperature. In principle, the Seebeck coef- ficient is large for insulators, whereas a high electronic conductance is obtained in conductors, so usually these two magnitudes are related in a conflicting way. Addi- tionally, a low thermal conductance κ would also enhance ZT , which includes an electronic and a phononic contri- bution. A possible way to tune these properties in an independent fashion is to lower the dimension and go to the nanoscale [2 -- 4]. In the last years the thermoelectric properties of nanos- tructured materials have been investigated with the aim to enhance thermoelectric efficiencies. Quantum size ef- fects permit the variation of the electronic properties [5 -- 9] and the reduction of the phonon thermal conduc- tance [10]. Nanopatterning, either with antidots, de- fects or edge modification, allows for further suppression of phonons [11, 12]. So despite the fact that pristine graphene is not a good material for thermoelectrics, be- ing gapless and very good thermal conductor, graphene ∗ monica.pacheco@usm.cl nanostructures have been proposed as potentially inter- esting for thermoelectric applications [13], with excellent values of the thermopower coefficient and figure of merit [14 -- 21]. One way to increase the thermopower is to enhance the electron-hole asymmetry at the Fermi energy [21, 22]. Strong asymmetries can be found in ferromagnetic ma- terials, so the coupling of the spin and thermal currents have been proposed as a means to further improve the thermoelectric properties of nanoscale devices [22 -- 25]. In such systems, charge, spin and heat transport are related, and may provide opportunities to optimize their perfor- mance. This active field has been called spin caloritron- ics [26]; although its bases were settled in the pioneer- ing work by Johnson and Silsbee [27], its development in low-dimensional systems has boosted the interest in these phenomena. Spin-dependent thermoelectric prop- erties of magnetic graphene have been previously studied [28], but without taking advantage of the reduced dimen- sionality and neglecting the phonon contribution to the thermal conductance. In this work, we explore the spin-dependent ther- moelectric properties of quasi-one-dimensional graphene structures composed of overlapping nanoribbons. Such systems were studied before due to their remarkable transport properties; if the ribbons are metallic, a se- ries of resonances and antiresonances appear related to the length of the bilayer region [29, 30]. If the ribbons are semiconducting, they present important advantages due to the reduction of the thermal conductance in roughly an order of magnitude, yielding an enhanced figure of merit [31]. This is due to the fact that only van der Waals forces couple the bilayer central part, so the strong graphene σ-bond lattice is interrupted in the overlapping region, hindering phonon transport. Here we explore the ad- vantages of considering spin-dependent transport in the thermoelectric properties; specifically, the role of mag- 7 1 0 2 p e S 0 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 2 0 7 0 . 9 0 7 1 : v i X r a netic contacts in the spin-thermoelectric response of these structures. Our main results are the following: (1) The charge Seebeck coefficient and figure of merit are strongly dependent of the magnetic configuration of the leads. We find that the charge Seebeck coefficient is greatly enhanced when the magnetic leads are in an antiparallel configuration, due to the enlargement of the transport gap. (2) Conversely, the charge figure of merit improves if the magnetizations of the leads are parallel, due to the electron-hole asymmetry. (3) Large values for the spin-dependent Seebeck coeffi- cient can be achieved for a wide range of energies, up to twice the charge Seebeck coefficient. Due to the changes on the spin-dependent Seebeck coefficient, the spin fig- ure of merit may have more than one maximum with an opposite temperature dependence. In Section II of the paper we describe the theory and model we employed to calculate the charge and spin-dependent thermoelectric coefficients of the bilayer graphene structure. In Section III we present results for the conductances, Seebeck coefficients and figure of mer- its as functions of the chemical potential. Finally, Section IV summarizes our conclusions. II. SYSTEM AND MODEL A. Geometry and configuration of the system As commented in the Introduction, we consider two overlapping armchair nanoribbons with AA stacking. The geometry is shown in Fig. 1. This system can be al- ternatively viewed as a finite-size bilayer flake connected to two monolayer ribbons acting as electrodes, which can be ferromagnetic due to proximity effects [32, 33]. We take into account two configurations for the leads, with the magnetic moments of the left and right lead being parallel (P) or antiparallel (AP), as depicted in the figure. There is a temperature difference between the monolayer electrodes, giving rise to thermoelectric effects. B. Model We employ a π-orbital tight-binding model, which gives an excellent description of the electronic proper- ties of graphene systems around the Fermi level. The in-plane nearest-neighbor interaction is given by a single hopping parameter γ, which we take as −3 eV. The inter- layer coupling in the flakes is considered by one hopping parameter between atoms directly placed on top of each other, being γ(cid:48) = 0.1γ. The ferromagnetic electrodes are modeled by the inclusion of a spin-dependent onsite term ∆ = ±0.1 eV, which produces a 0.2 eV splitting between majority and minority bands in the leads. 2 FIG. 1. Depiction of the graphene flake of length L and width W made by two overlapping nanoribbons. The left and right leads, i.e., the monolayer ribbons, are deposited on top of a ferromagnetic material. The magnetic moment of the right lead may be parallel or antiparallel to that of the left lead. A difference of temperature between the electrodes is applied. As it is customary, armchair (AC) ribbons are labeled with the number of dimers n across its width, i.e., ACn. Thus, an AC12 ribbon has a width of 13.53 A. The length of the flake is given in terms of the number m of translational unit cells. The length of one unit cell for this type of ribbon is equal to 3ac, being ac = 1.42 A the carbon-carbon bond length. In this way we can identify the systems as ACnLm, i.e., giving the width of the ribbon and the length of the flake. The electronic transmission τ as a function of the en- ergy carrier E through the system can be numerically ob- tained by standard Green function techniques, following a real-space renormalization scheme [34]. From the trans- mission all the thermoelectric magnitudes of interest can be computed. We neglect spin-scattering effects, mean- ing that each spin channel contributes independently to the conductance and spin-flip processes are neglected. As the right (R) and left (L) leads can be in two differ- ent ferromagnetic states, we have two different situations with respect to the spin-dependent transport through the flake, depicted in Fig. 2. In the parallel configuration (P), the band structure of the left and right lead are exactly equal, whereas in the AP case they are not. FIG. 2. Schematic band structures of the left lead, central part (flake) and right lead in the P (left panel) and AP (right panel) magnetic configurations. ()WLeft electrodeRight electrodeLPAPLLRRPAPC. Spin-dependent thermoelectric magnitudes With these definitions, we can obtain the correspond- 3 ing charge and spin figures of merit, Zc,sT = c,sT Gc,sS2 κ (7) where κ = κe↑ + κe↓ + κph, with κeσ and κph are the electronic and phonon contribution to the charge thermal conductance, respectively. In what follows we show results for charge-dependent magnitudes Gc, Sc and ZcT and spin-dependent magni- tudes Gs, Ss. Note that in the AP configuration Gs = 0, Ss = 0 and ZsT = 0 due to the symmetry of the spin- split bands in the leads, as can be inferred from Fig. 2. III. RESULTS AND DISCUSSION As are the to compute ingredients the main spin- spin-resolved thermoelectric magnitudes transmissions, we first show these values for the systems studied in this work. We choose the ribbons to be semiconducting, because they yield better values of the thermoelectric properties than the metallic counterparts [31], for which a fine tuning of their geometry is needed to increase their thermoelectric response [9]. We have verified that certainly is more interesting to focus in semiconducting flakes. The spin-dependent electric current can be obtained in terms of the transmission function within the Landauer- Buttiker formalism τσ(E)(fL(E) − fR(E))dE , (1) (cid:90) ∞ −∞ Iσ = e h where fR,L are the Fermi distributions of the right and left electrodes and σ =↑ (↓) denotes the spin projection direction. Gσ = ∆Iσ/∆V , in the limit of ∆V → 0. The spin-resolved electronic conductance is given by Analogously, the spin-resolved thermal current is given by IQ,σ = (cid:90) ∞ −∞ (E−µ+∆V )τσ(E)(fL(E)−fR(E))dE , (2) The Seebeck coefficient or thermopower S is defined as the voltage drop induced by a temperature gradient at vanishing current, S = −∆V /∆TIσ=0, in the limit of ∆T → 0. The electronic thermal conductance defined as κe,σ = ∆IQ,σ/∆V for ∆V → 0 is calculated in the linear response regime i.e., ∆T << T and e∆V << µ, with µ being the equilibrium chemical potential, and T the temperature. The above-defined thermoelectric coefficients can be written in compact form in terms of the following thermal integrals: (cid:90) ∞ −∞ Ln,σ(µ, T ) = xn ex (ex + 1)2 τσ(x, µ, T ) dx. Thus the electronic conductance is just Gσ = L0,σ, e2 h (3) (4) where e is the elementary charge and h the Planck's con- stant; the Seebeck coefficient is given by Sσ = − kB e L1,σL0,σ , (5) with kB being the Boltzmann's constant; and finally, the electronic thermal conductance is (cid:34) (cid:35) κe,σ = k2 BT h L2,σ − L2 1,σL0,σ . (6) FIG. 3. Spin-resolved electronic transmissions for AC12L12 and AC21L12 graphene flakes in the non-magnetic (black), P (blue for majority spin, green for minority spin) and AP (red) configurations. The charge and spin conductance are defined as Gc = G↑ + G↓ and Gs = G↑ − G↓, and the charge Seebeck and spin-dependent Seebeck coefficients as Sc = (S↑ + S↓)/2 and Ss = S↑ − S↓, respectively [23]. Fig. 3 shows the spin-resolved transmissions for the AC21L12 and AC12L12 flakes in the AP and P configu- rations; for the sake of comparison, we include also the transmission in the case of non-magnetic (NM) leads. As 01τστσ NMτσ AP01τστ↑ Pτ↓ P01τστσ NMτσ AP-1-0.500.51Energy (eV)01τστ↑ Pτ↓ PAC21AC21AC12AC12mentioned before, the transmission for the AP configu- ration is the same for both spin directions (see Fig. 2); obviously, this also happens trivially for the NM case. The asymmetry in the transmission for the P configu- ration is noteworthy; this can be a way to increase the thermoelectric response. Nonetheless, there is a clear re- lation between τ↑ and τ↓ in the P configuration due to the electron-hole symmetry present in graphene systems, namely, τ P↑ (E) = τ P↓ (−E). Notice that the conductance gap increases in the AP configuration; however, for the P case, the gap for each spin channel is unchanged in com- parison to the NM case, although the total transmission has a smaller conductance gap, as it can be clearly seen in Fig. 3. 4 flake length for all the magnetic configurations. This can be understood because the Seebeck coefficient is mainly dependent on the transmission gap, for which the flake characteristics are not relevant. Indeed, our Sc for the NM configuration is similar to that reported in Ref. [3] for the corresponding perfect ribbons. In order to inspect the role of the nanoribbon width, we show the temperature dependence for the AC21L12 flake in the AP magnetic configuration in the panel of Fig. 4 (b). This is a wider ribbon, so its gap is smaller; for this reason the Seebeck coefficients are reduced with respect to the AC12 flakes. We observe a decrease of the thermopower with increasing temperature for both magnetic configurations; Fig.4 (c) shows a wide range of temperatures for the AC12L12 system in the P magnetic configuration. This is consistent with previous calcula- tions in pristine semiconducting nanoribbons [3]. For the sake of obtaining a reliable quantity for the fig- ure of merit, it is necessary to include the phonon thermal conductance. We draw these values from the literature. In Ref. [31] the same geometry (AC12) for different flake lengths is investigated, albeit without including the spin dependence. In their Fig. 3 (a) they present the phonon conductance as a function of temperature. For temper- atures above 300K, the phonon conductance is almost constant, so we take their values at 600 K for all the tem- peratures studied, namely, κph = 0.28 nW/K for L=12, κph = 0.22 nW/K for L=8, and κph = 0.15 nW/K for L=4. In all cases these are reasonable upper bounds for the phonon thermal conductance. For the wider flake, AC21L12, we extrapolate the value from the calculated phonon thermal conductance presented in Ref. [35]; we take κph = 0.38 nW/K. FIG. 4. a) Charge Seebeck coefficients of the AC12L4 system at T = 300 K, for the P, AP and NM configurations. b) Temperature dependence of the charge Seebeck coefficient for the AC21L12 case in the AP configuration and c) temperature dependence of the charge Seebeck coefficient for the AC12L12 case in the P configuration. In Fig. 4 (a) we present the charge Seebeck coefficients (in units of S0 = kB/e) for the three magnetic configura- tions, NM, P and AP, for the AC12L4 case, at T = 300 K. The optimal charge thermopower is obtained for the AP configuration, being better than the NM case. This is due to the increase of the gap for the AP alignment. The smallest Sc is obtained for the P configuration; this is due to the reduction of the conductance gap. As Sc is the sum of the spin-up and down contributions, the P charge thermopower has more structure due to the ad- dition of the two displaced S↑ and S↓ spin parts, shifted in 0.1 eV with respect to the NM case. The reduction of Sc has also been found in quantum dots with ferro- magnetic leads with parallel magnetization [7]. We have checked that Sc is almost unchanged with respect to the FIG. 5. Charge ZT for AC12L12, AC21L12 and AC12L8 flakes in the P and AP magnetic configurations for different temperatures. The NM case is included in black dotted line for 300K. In Fig. 5 we present our results for the charge figure of merit ZcT corresponding to the flakes AC12L12, AC12L8 and AC21L12 in the P and AP magnetic configurations and three different temperatures, 300 K, 400 K and 500 K. For comparison, we have also plotted the results for nonmagnetic (NM) leads in black dotted lines for 300 K. -0.4-0.200.20.4-20-1001020Sc/S0APPNM-0.4-0.200.20.4-20-1001020Sc/S0300K400K500K-0.4-0.200.20.4Energy (eV)-20-1001020Sc/S0300K400K500K600K700K800Ka)b)c)AC12L12 (P)AC12L4 (300K)AC21L12 (AP)00.20.40.6Energy (eV)012345Zc T NM300K400K500K00.20.40.6Energy (eV)012345NM300K400K500K00.20.40.6Energy (eV)012345NM300K400K500KAC12L12AC21L12AC12L8PAPPAPPAP5 FIG. 6. Spin-dependent Seebeck coefficient for AC12L12 and AC21L12 for different temperatures. FIG. 7. Spin-dependent figure of merit for AC12L12 and AC12L8 for several temperatures. We observe that for all the systems the maximum value of ZcT (ZcTmax) is greatly enhanced if the ferromagnetic leads are in the parallel magnetic configuration, while in the antiparallel case ZcTmax takes values slightly greater than those for NM leads. This behavior is a consequence of the decrease in the transmission gap in the P config- uration, as it can be observed in Fig. 3. The increase in the electronic conductance permits to reach maximal intensities for the figure of merit, despite the fact that this is not the optimal magnetic arrangement to obtain high Seebeck coefficients. For the range of temperatures studied, ZcTmax in the AP configuration increases with temperature. This is an interesting behavior which we also observe for the NM case. It is a consequence of the increase in the electronic conductance with T and the independence of the phonon thermal conductance for T ≥ 300 K in these structures (see Ref. [31]). In the P configuration, ZcT has a dual behavior with T as a func- tion of the chemical potential. Depending on the energy range, the figure of merit decreases with increasing tem- perature, as expected in a normal semiconductor device, whereas in other energy intervals the behavior is the op- posite. This is most clearly seen in the AC12 cases (the left and right panels of Fig. 5). Such changes in the tem- perature dependence are very sensitive to the gap value and spin splitting energy ∆. In Fig. 6 we present the spin-dependent Seebeck coef- ficients for the AC12L4 (upper panel) and the ACL21L12 (bottom panel) systems for three different temperatures. As the spin-dependent thermopower is the subtraction of the Seebeck coefficients of the two spin channels, for a nonzero ∆ there will always be a finite contribution due to the breaking of electron-hole symmetry. For larger spin splitting ∆ > 0.3 eV, the Ss will have a distinct shape, being possible to discern the two contributions of the two spin channels separately. However, for ∆ = 0.1 eV, chosen in this work, we have a plateau around 0 eV due to the overlap of the Seebeck coefficients of the two spin channels. The maximum Ss is considerably high, slightly larger than that obtained for the charge Seebeck coefficient. If electron-symmetry holds, the charge Seebeck coef- ficient is an odd function of the energy, as it is evident from Fig. 4 and readily inferred from Eqs. 3 and 5. Since in our system in the P configuration the spin splittings at the leads have the same value, the spin-dependent See- beck coefficient is also symmetric, although now it is an even function of the energy, having at least two zeros close to ±∆. In Fig. 7 we present our results for the spin-dependent figure of merit ZsT for the flakes AC12L12, AC12L8 in the temperature interval from 300 K to 800 K. There are two peaks related to the two intervals of positive and negative values of Ss, which are shifted with respect to the non-magnetic case. Although the maximum abso- lute value of the spin-dependent Seebeck coefficient is attained in the central plateau, the role of the electronic conductance is crucial to set the maximum value of ZsT , ZsTmax. Notice that ZsTmax takes place near the band edges, boosting up to twice the value of the charge figure of merit (see Fig. 5). The two peaks have quite dissimilar value, due to the electronic conductance, and also exhibit an opposite temperature dependence, which changes at the minimum imposed by the zero of the spin-dependent Seebeck coefficient. These findings indicate that it is advantageous to ex- ploit spin currents with the aim of maximizing the figure of merit, especially for nanostructured materials. Notice that similar effects to those achieved in the AP configu- ration can be obtained by applying a gate voltage at the leads: if the chemical potential at the electrodes if varied, one can mimic the role of the ferromagnetic leads ana- lyzed in this work, therefore tuning by electrical means -0.4-0.200.20.4-100102030Ss /S0-0.4-0.200.20.4Energy (eV)-100102030Ss /S0300K400K500KAC12L4AC21L1200.20.40.6Energy (eV)0246810ZsT00.20.40.6Energy (eV)0246810300K400K500K600K700K800KL=8L=12the thermoelectric behavior of the system. IV. CONCLUSIONS We have shown that the coupling of spin with charge and thermal currents can be harnessed to rise the thermo- electric efficiency of nanoscale graphene devices. The use of magnetic leads in nanostructures composed of overlap- ping graphene nanoribbons allows for the optimization of the charge thermoelectric properties, as well as the spin- dependent magnitudes. We have demonstrated that the charge Seebeck effect is greatly enhanced when the magnetic leads are in an antiparallel configuration, due to the enlargement of the transport gap. However, for the optimization of the charge figure of merit, it is better to choose a paral- 6 lel alignment of the magnetization in the leads, due to the electron-hole asymmetry present in this configura- tion. We also obtained the spin-dependent Seebeck coef- ficient and figure of merit. In fact, the latter shows much better values than their charge counterparts, up to 10 at room temperature. These findings suggest the poten- tial value of graphene nanosystems as energy harvesting devices employing spin currents and magnetic electrodes. ACKNOWLEDGMENTS We acknowledge helpful discussions with P. Vargas. This work has been partially supported by the Span- ish MINECO under grant FIS2015-64654-P, Chilean FONDECYT grants 1140571 (P.O.), 1140388 (L. R.), 1151316 (M.P.). L. C. gratefully thanks the hospitality of the Universidad T´ecnica Federico Santa Mar´ıa (Chile). [1] T. M. Tritt, Annu. Rev. Mater. Res. 41, 433 (2011). [2] R. Kim, S. Datta, and M. S. Lundstrom, Journal of Applied Physics 105, 034506 (2009). [3] Y. Ouyang and J. Guo, Applied Physics Letters 94, 263107 (2009). [4] Y. Dubi and M. Di Ventra, Rev. Mod. Phys. 83, 131 (2011). [5] C. M. Finch, V. M. Garc´ıa-Su´arez, and C. J. Lambert, Phys. Rev. B 79, 033405 (2009). [6] G. G´omez-Silva, O. ´Avalos-Ovando, M. L. Ladr´on de Guevara, and P. A. Orellana, J. Appl. Phys. 111, 053704 (2012). [7] P. Trocha and J. Barna´s, Phys. Rev. B 85, 085408 (2012). [8] V. M. Garc´ıa-Su´arez, R. Ferrad´as, and J. Ferrer, Phys. Rev. B 88, 235417 (2013). [9] N. Cort´es, L. Rosales, L. Chico, M. Pacheco, and P. A. Orellana, Journal of Physics: Condensed Matter 29, 015004 (2017). [10] W.-X. Zhou and K.-Q. Chen, Scientific Reports 4, 7150 (2014). [19] N. Cort´es, L. Chico, M. Pacheco, L. Rosales, and P. A. Orellana, EPL (Europhysics Letters) 108, 46008 (2014). [20] M. Saiz-Bret´ın, A. V. Malyshev, P. A. Orellana, and F. Dom´ınguez-Adame, Phys. Rev. B 91, 085431 (2015). [21] M. S. Hossain, D. H. Huynh, P. D. Nguyen, L. Jiang, T. C. Nguyen, F. Al-Dirini, F. M. Hossain, and E. Skafi- das, Journal of Applied Physics 119, 125106 (2016). [22] 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). [23] A. Slachter, F. L. Bakker, J.-P. Adam, and B. J. van Wees, Nat. Phys. 6, 879 (2010). [24] A. Slachter, F. L. Bakker, and B. J. van Wees, Phys. Rev. B 84, 174408 (2011). [25] G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nat. Mater. 11, 391 (2012). [26] G. E. Bauer, A. H. MacDonald, and S. Maekawa, Solid State Communications 150, 459 (2010). [27] M. Johnson and R. H. Silsbee, Phys. Rev. B 35, 4959 [11] H. Karamitaheri, M. Pourfath, R. Faez, and H. Kosina, (1987). J. Appl. Phys. 110, 054506 (2011). [28] B. Z. Rameshti and A. G. Moghaddam, Phys. Rev. B 91, [12] P.-H. Chang and B. K. Nikoli´c, Phys. Rev. B 86, 041406 155407 (2015). (2012). [13] P. Dollfus, V. H. Nguyen, and J. Saint-Martin, Journal of Physics: Condensed Matter 27, 133204 (2015). [14] F. Mazzamuto, V. Hung Nguyen, Y. Apertet, C. Caer, C. Chassat, J. Saint-Martin, and P. Dollfus, Phys. Rev. B 83, 235426 (2011). [15] J. Haskins, A. Kınacı, C. Sevik, H. Sevin¸cli, G. Cuniberti, and T. C¸ agın, ACS Nano 5, 3779 (2011). [29] J. W. Gonz´alez, H. Santos, M. Pacheco, L. Chico, and L. Brey, Phys. Rev. B 81, 195406 (2010). [30] J. W. Gonz´alez, H. Santos, E. Prada, L. Brey, and L. Chico, Phys. Rev. B 83, 205402 (2011). [31] V. H. Nguyen, M. C. Nguyen, H.-V. Nguyen, J. Saint- Martin, and P. Dollfus, Applied Physics Letters 105, 133105 (2014). [32] Y.-L. Lee, S. Kim, C. Park, J. Ihm, and Y.-W. Son, ACS [16] H. Sevin¸cli, C. Sevik, T. C¸ agın, and G. Cuniberti, Sci- Nano 4, 1345 (2010). entific Reports 3, 1228 (2013). [17] L. Rosales, C. D. Nunez, M. Pacheco, A. Latg´e, and P. A. Orellana, J. Appl. Phys. 114, 153711 (2013). [18] K.-M. Li, Z.-X. Xie, K.-L. Su, W.-H. Luo, and Y. Zhang, Physics Letters A 378, 1383 (2014). [33] Z. Wang, C. Tang, R. Sachs, Y. Barlas, and J. Shi, Phys. Rev. Lett. 114, 016603 (2015). [34] L. Chico, L. X. Benedict, S. G. Louie, and M. L. Cohen, Phys. Rev. B 54, 2600 (1996). [35] Z. Huang, T. S. Fisher, and J. Y. Murthy, Journal of Applied Physics 108, 094319 (2010). 

1010.1118

1

1010

Tuning Fano resonances by magnetic forces for electron transport through a quantum wire side-coupled to a quantum ring

We consider electron transport in a quantum wire with a side-coupled quantum ring in a two-dimensional model that accounts for a finite width of the channels. We use the finite difference technique to solve the scattering problem as well as to determine the ring-localized states of the energy continuum. The backscattering probability exhibits Fano peaks for magnetic fields for which a ring-localized states appear at the Fermi level. We find that the width of the Fano resonances changes at high magnetic field. The width is increased (decreased) for resonant states with current circulation that produce the magnetic dipole moment that is parallel (antiparallel) to the external magnetic field. We indicate that the opposite behavior of Fano resonances due to localized states with clockwise and counterclockwise currents results from the magnetic forces which change the strength of their coupling to the channel and modify the lifetime of localized states.

cond-mat.mes-hall

cond-mat

Tuning Fano resonances by magnetic forces for electron transport through a quantum wire side-coupled to a quantum ring B. Szafran and M.R. Poniedzia lek Faculty of Physics and Applied Computer Science, AGH University of Science and Technology, al. Mickiewicza 30, 30-059 Krak´ow, Poland (Dated: November 25, 2018) We consider electron transport in a quantum wire with a side-coupled quantum ring in a two- dimensional model that accounts for a finite width of the channels. We use the finite difference technique to solve the scattering problem as well as to determine the ring-localized states of the energy continuum. The backscattering probability exhibits Fano peaks for magnetic fields for which a ring-localized states appear at the Fermi level. We find that the width of the Fano resonances changes at high magnetic field. The width is increased (decreased) for resonant states with current circulation that produce the magnetic dipole moment that is parallel (antiparallel) to the external magnetic field. We indicate that the opposite behavior of Fano resonances due to localized states with clockwise and counterclockwise currents results from the magnetic forces which change the strength of their coupling to the channel and modify the lifetime of localized states. PACS numbers: 73.63.-b, 73.63.Nm, 73.63.Kv I. INTRODUCTION Coherent transport properties of mesoscopic and nanoscale conductors is determined by interference con- ditions for the electron wave function at the Fermi level. Quantum dots and rings that are attached by a single contact to a conducting channel modify its magneto- transport properties although they lie outside the classi- cal current path.1 Single subband conductance of the con- tact is proportional to the electron transfer probability2 and the latter is particularly sensitive to existence of localized states in the side-coupled structures. The lo- calized states that belong to the energy continuum in- terfere with delocalized states of the channel leading to an appearance of Fano resonances3 in magnetoconduc- tance when the Fermi level of the channel is degenerate with the localized energy level. The Fano resonances for quantum dots and rings connected to a semiconducting channel by one or two contacts are extensively studied in the context of phase coherence probes,4 Aharonov-Bohm interferometry,5 Kondo effect,6 -- 9 construction of spin filters,10,11 conductance of single-electron transistors,12 arrays of quantum dots13 -- 15 and artificial defects.16,17 The of singly transport properties connected nanostructures are usually studied using theoretical models9 -- 11,15 -- 20 that neglect the finite width of the channels. These models account for the phase shifts due to the Aharonov-Bohm effect but naturally over- look the deflection of electron trajectories by classical magnetic forces, which occurs when the Larmor radius is comparable to the width of the channel. The purpose of the present paper is to establish the role of magnetic deflection for the ballistic transport properties of a quantum ring side-coupled to a channel. The magnetic forces were previously considered21,22 for quantum rings that are embedded within the channel.23 -- 27 In these structures the Lorentz force leads to a preferential injection of the electron from the input channel to one of the arms of the ring.21 The preferential injection implies reduction of the Aharonov-Bohm inter- ference amplitude at the exit to the output channel.21 A theoretical study of a ring connected to three channels,22 indicated that the reduction of the Aharonov-Bohm con- ductance oscillations due to magnetic forces is accompa- nied by a distinct imbalance of the electron transfer to the two output terminals. This prediction22 was confirmed in a subsequent experiment.28 In a recent proposal29,30 of electronic interaction-free measurement, the idea for the solid state device employs magnetic forces to de- flect the electron trajectory in a similar manner as beam splitters deflect the photons in the optical experimental setup,31 which introduces an additional interest in mag- netic forces. In this paper we solve the single subband scattering problem for a quantum ring singly connected to a quan- tum wire in presence of perpendicular magnetic field. The resonances that are found in the electron backscat- tering probability are confronted with the results of the stabilization method32 which allows for detection of lo- calized states in the energy continuum. We find that injection of the electron from the channel to the ring is enhanced at the resonances. The charge distribution within the structure is not an even function of the mag- netic field. This direct effect of magnetic forces does not influence the linear conductance which is necessarily an even function of the magnetic field due to the Onsager symmetry.33 We demonstrate that for a finite width of the channel the Fano resonances of backscattering proba- bility change in a characteristic manner at high magnetic field. Typically these resonances appear in pairs. We find that at high magnetic field one of the resonances widens, and the other becomes extremely thin. We argue that modification of the width of resonances is uniquely due to magnetic forces. Thermal stability of Fano resonances of backscattering probability is also discussed. The Fano resonances in weak magnetic fields were con- sidered in particular in Refs. 34 and 35 for quantum dots embedded within a channel. Ref. 35 describes the quantum-dot-localized states of the energy contin- uum between the first and second subband propagation thresholds that are of the odd parity symmetry with re- spect to the axis of the system. These states are bound, i.e. have an infinite lifetime, since the leakage to the channel is blocked by the opposite - even parity - symme- try of the lowest subband of the channel. The magnetic field breaks the symmetry with respect to the axis of the system and allows for the electron leakage. The bound states turn into metastable ones with a finite lifetime and the backscattering probability exhibits Fano peaks that shift on the energy scale with the magnetic field accord- ing to the sign of the dipole moment produced by the current circulation in the quantum-dot-localized states. The study35 is limited to weak magnetic fields, and does not cover the modification of the lifetime due to magnetic forces which are the subject of the present paper. This paper is organized as follows. In Section II we briefly describe the approach applied for the scattering problem. The stabilization method for localized states detection is sketched in Section III. The results of these two approaches are confronted in Section IV. Summary and conclusions are given in Section V. II. TREATMENT OF SCATTERING PROBLEM The studied system is presented in Fig. 1. We as- sume that the channels are made of GaAs embedded in Al0.45Ga0.55As matrix. The width of the channels is taken equal to 64 nm, unless explicitly stated otherwise. The inner (outer) radius of the ring is 60 nm (124 nm). The electron confinement in the growth direction is usu- ally much stronger than the planar one, which justifies application of a two-dimensional model that is employed below. We consider the Hamiltonian H = (p + eA(r))2 /2m∗ + V (x, y) (1) where −e is the electron charge (e > 0) and m∗ = 0.067m0 is the GaAs electron band effective mass. We apply the Lorentz gauge A = (Ax, Ay, 0) = (0, Bx, 0) and assume that the confinement potential V (x, y) is zero in- side (the blue area in Fig. 1) and V0 = 200 meV36 outside the channels (the white area in Fig. 1). We solve the eigenequation HΨ = EΨ (2) with a finite difference approach using a square grid and lattice spacings ∆x = ∆y = 2 nm to determine the wave function on a mesh Ψµ,ν = Ψ(xµ, yν) . In the calculation we use the gauge-invariant discretization of the kinetic energy operator39 1 2m∗ (p + eA(r))2 Ψµ,ν = ¯h2 2m∗∆x2 × 2 64nm 400 350 300 250 200 150 100 50 ] m n [ y 1 2 4 n m 6 0 n m 0 100 200 300 400 x [nm] (b) ] m n [ y 200 150 100 50 0 -50 -100 -150 -200 L 0 100 200 300 400 x [nm] FIG. 1: (a) The system studied in the scattering problem. The electron is assumed to come from the channel below the contact to the ring and the vertical channel has an infinite length. (b) The model system used for determination of the ring-localized states. L is the finite length of the channel, which varies in the calculation. The confinement potential is assumed 0 within the blue area and 200 meV in the outside. (cid:0)4Ψµ,ν − CyΨµ,ν−1 − C ∗ −Ψµ−1,ν − Ψµ+1,ν) , y Ψµ,ν+1 (3) with Cy = exp(cid:2)−i e ¯h ∆xBx(cid:3). For description of electron scattering we assume that the input channel is the one below the contact to the ring [Fig. 1(a)]. For the chosen gauge the Hamiltonian eigen- functions in the channel can be written in a separable form Ψ(x, y) = exp(iky)ψk(x), (4) where k is the wave vector and ψk(x) is the transverse eigenfunction of the electron in the vertical channel, which is determined by solution of a one-dimensional equation obtained by plugging Eq. into the Schroedinger equation (2). For the energy that corre- sponds to the lowest subband the wave function in the incoming lead far from the ring is a superposition of the incident and reflected waves (4) Ψ = ψk(x) exp(iky) + c−kψ−k(x) exp(−iky), (5) and far in the output lead one has only the transferred wave function Ψ = dkψk(x) exp(iky). (6) The values of c−k and dk are extracted from the finite dif- ference wave functions at the ends of the computational box. The boundary condition of the output lead (6) is intro- duced to the discretized eigenequation (2) in a Neumann form Ψµ,ν+1 = Ψµ,ν exp(ik∆y) (7) that results of Eq. (6). The boundary condition for the input lead (5) is taken in the Dirichlet form with an initial assumption that c−k = 0. After solution of the algebraic form of the eigenequation we extract c−k and dk by an- alyzing the finite difference wave function at the ends of the channels. The eigenequation is solved for the new value of c−k introduced in the boundary condition. The procedure converges after just a few iterations.37 After the convergence is reached we evaluate the backscattering probability by calculating the current fluxes in the incoming channel R = R dxj−k(x) R dxjk(x) , (8) where jk is the probability density current associated to the incoming part of the wave function of Eq. (5), jk(x) = ¯h m∗ ψk(x)2(¯hk + eBx) and j−k corresponds to the backscattered one (9) j−k(x) = ¯h m∗ c−k2ψ−k(x)2(−¯hk + eBx). (10) is related to h [1 − R(B)] . backscattering The probability magnetoconductance2 as G(B) = 2e2 The magnetic forces modify the width of the Fano res- onances of the backscattering probability not only as a function of the magnetic field but also as a function of the energy. In finite temperature a transport window is opened near the Fermi level EF and the narrow Fano res- onances are likely to be thermally unstable. The stability of Fano resonances against thermal excitations is below estimated by the linear response formula R(EF ) = Z R(E)(cid:18)− ∂f ∂E(cid:19) dE, (11) where f is the Fermi function f = (cid:0)e(E−EF )/kB T + 1(cid:1)−1 . T stands for the temperature and kB for the Boltzmann constant. III. DETECTION OF LOCALIZED STATES In order to determine the resonant states localized in the ring we use the stabilization method.32 For this pur- pose we assume that the vertical channel has a finite length [see Fig. 1(b)]. Then we solve the eigenequation 3 FIG. 2: (a) Energy spectrum for the channel of a finite length (L) for B = 0. (b) The peaks indicate the energies of the localized states extracted of the energy spectrum according to Eq. (12). (c-e) Charge densities of the five lowest energy eigenstates for L = 352 nm. Plot (c) corresponds to the ground-state, (d) to the first excited state etc. (4) with boundary conditions that require the wave func- tions to vanish at all the edges of the computational box. We calculate the energy spectrum as a function of the length of the channel L. The energy spectrum that is discrete for finite L is displayed in Fig. 2(a) for B = 0. We see that some energy levels decrease as L grows. They correspond to electron states in the channel. The energy of states localized in the ring are independent of L. The two lowest energy levels of Fig. 2(a) correspond to states localized at one of T-junctions38 present in the system -- one between the main vertical channel and the short horizontal one [Fig. 2(c)], and the other between the latter and the quantum ring [Fig. 2(d)]. These two states are not only localized but energetically bound - their energies (0.925 meV and 1.006 meV) lie below the continuum threshold. The threshold is determined by the bottom of the lowest subband (k = 0) for the infinite vertical channel. For 64 nm - wide channel the energy continuum starts above 1.147 meV (B = 0). The energy levels of Fig. 2(a) above the continuum threshold that are independent of L correspond to the ring-localized states, see Figs. 2(e) and 2(f). The fourth excited state [Fig. 2(g)] found for L = 352 nm corresponds to the electron within the channel, and its energy distinctly depends on L [see Fig. 2(a) for L = 352 nm]. The spectral positions of the localized states are de- termined by counting the states of the energy close to E, N (E) = Z dLXl δ(E − El(E); dE), (12) where l numbers the Hamiltonian eigenvalues for the fi- nite size of the system, and δ(E − El(L); dE) is equal to 1 for E − El(L) < dE and 0 otherwise. In Fig. 2(b) we plotted N (E) calculated for the energy spectrum of Fig. 2(a) with the energy window dE = 5µeV. FIG. 3: (a) Positions of localized states in the energy contin- uum as calculated by formula (12). The darker the shade of red the larger value of N (E). The thin vertical lines indicate energies of 2.4, 2.65 and 3.4 meV that are considered in detail below. The green dashed line indicates the continuum thresh- old (ground-state energy of the electron within the channel for k = 0). (b) Energy spectrum of the closed circular quantum ring that is not connected to the channel. Figure 3 shows how the maxima of N (E) shift with B [Fig. 3(a)] as compared to the energy spectrum of a closed40 quantum ring [Fig. 3(b)] that is not connected to the channel. The two lowest-energy lines of Fig. 3(a) correspond to states localized at the junctions. The reso- nances that are higher in the energy correspond to states localized in the ring. The positions of resonances oscillate with magnetic field with a period of 0.165 T that results from the Aharonov-Bohm effect for a ring of an effec- tive radius of 92 nm. The resonances enter into avoided crossings that result from angular momentum mixing of closed-ring states. The mixing is due to the presence of the contact that breaks the rotational symmetry and is the strongest at odd multiples of half of the flux quan- tum threading the ring. Above 4.6 meV the states of the second subband - with wave functions that change sign across the channel appear in Fig. 3(a). For the closed ring these states appear near 4.9 meV [Fig. 3(b)]. IV. RESONANCES IN THE QUANTUM RESISTANCE 4 0.7 R 0.03 0.02 0.01 0.00 100 0.9 80 (a) 0.8 B [T] N 40 0 50 N 0 1 r (b) 0 1.0 -0.2 0.0 0.2 0.4 0.6 0.8 B [T] 1 R 0 1 0 -1 χ FIG. 4: (a) Black line shows the electron backscattering prob- ability for energy E = 2.65 meV and the red one -- the value of the resonance detection counter N (E) [Eq.(12)]. The inset shows a zoom of high B part of the figure. (b) The value of χ plotted in green indicates the direction of the current circulation inside the ring: χ = 1 (-1) corresponds to coun- terclockwise (clockwise) direction. The blue line shows the fraction of the probability density r that is contained within the ring, i.e. for x > 32.5 nm of the computational box of Fig. 1(a). FIG. 5: The contours show the charge density for two peaks of backscattering probability of Fig. 4 obtained for E = 2.65 meV at B = 0.5 T (a) and at B = 0.751 T (b). The arrows show the probability current distribution. Now let us look at the electron backscattering prob- ability that is plotted with the black line in Fig. 4(a) for E = 2.65 meV. The red lines in Fig. 4(a) show the resonance detection counter N (E) of Eq. (12). R ex- hibits peaks as a function of the magnetic field. These peeks perfectly coincide on the magnetic field scale with the positions of the ring-localized states as determined by the stabilization method. The backscattering occurs only when the channel state carrying the current is degenerate with a localized state within the ring. Outside the de- generacy the structure is nearly transparent for the elec- tron transfer. The peaks of R have distinctly asymmetric 1 R 0.5 0 0.0 1 R 0.5 0 0.0 +1 E=2.4 meV χ -1 0.2 0.4 B [T] 0.6 E=3.4 meV +1 χ -1 80 (a) 40 N 0 80 (b) 40 N 0.2 0.4 B [T] 0 0.6 FIG. 6: Backscattering probability (black line, left axis), the resonance counter [Eq.(12), red line, right axis] and the di- rection of the current circulation within the ring in the scat- tering eigenstates [green line, χ = +1 (-1) corresponds to counterclockwise (clockwise) orientation]. Plots (a) and (b) were prepared for E = 2.4 meV and 3.4 meV, respectively. 0.8 0K =2.65 meV E F R 0.4 0.0 0.0 100mK K m 0 0 3 0.2 0.4 0.6 B [T] FIG. 7: The backscattering probability averaged according to the linear response formula (11) for the Fermi energy assumed at 2.65 meV for 0K (black line), 100 mK (blue line) and 300 mK (red line). shape which is a characteristic signature of a resonance involving a localized state of the energy continuum.3 In Fig. 4(b) we plotted with the blue line the fraction r of the probability density stored by the computational box of Fig. 1(a) that is contained within the ring (outside the vertical channel in fact). We notice that the electron enters the ring for magnetic fields for which a localized state is present at the considered energy. Note that R is an even function of B but r is not [r(B) 6= r(−B)]. For B > 0 the Lorentz force deflects the electron trajectories to the left of its momentum vector, hence the penetration of the wave function to the ring is hampered for positive and enhanced for negative magnetic field. 5 FIG. 8: Same as Fig. 4(a) but for the channel width reduced to 32 nm. The inner and outer radii of the side-coupled ring 76 and 108 nm, respectively. The geometry is displayed in the inset. The thin vertical lines indicate energies of 5.05, 5.16 and 5.26 meV that are considered in Fig. 9. The height of R maxima is reduced for high magnetic field. For B > 0 the Lorentz force pushes the electron to the left edge of the vertical channel. The electron with wave function shifted to the left edge of the channel seems not to notice the presence of the ring and passes through the contact with a nearly 100% probability. At high B the presence of the ring still produces the Fano peaks but on a tiny scale [see the inset to Fig. 4(a)]. In this sense the Lorentz force for B > 0 assists in the electron transport across the contact. Note, that due to the microreversibility relation33 we have R(B) = R(−B) although for negative B the electron penetration to the ring is enhanced. The results for N (E) as presented in Fig. 4(a) are cross section of Fig. 3(a) taken along the constant energy line. We can see that the line for E = 2.65 meV crosses the resonances of Fig. 3(a) that grow or decrease in energy as the magnetic field grows. The magnetic dipole moment generated by currents is µ = − dE dB . For a strictly one dimensional closed circular quantum ring µ = − 1 2 er × j, where j stands for the probability density current. Thus, the resonances that grow (decrease) in the energy corre- spond to localized states in which the current circulates counterclockwise (clockwise) around the ring and pro- duce magnetic dipole moment that is antiparallel (paral- lel) to the external magnetic field. The direction of the current flow within the ring as found in the solution of the scattering problem is denoted by χ = ±1 [green line in Fig. 4(b)], where the plus sign stands for the counter- clockwise circulation. Note, that χ changes sign between each minimum of the transfer probability. For B > 0 the resonances that increase (decrease) in width corresponds to clockwise (counterclockwise) current circulation. At high positive magnetic field the current circulation within the ring in the scattering eigenstates is counterclockwise with the exception of B ranges that surround the sharp Fano resonances [Fig. 4(b)]. Figure 5 shows the charge density and current distri- 1.0 0.8 0.6 R 0.4 0.2 0.0 0.0 0.3 R 0.2 E=5.05 meV (a) 1 0 -1 χ 0.2 0.4 0.6 0.8 1.0 B [T] E=5.16 meV (b) 1 0 -1 χ 0.0 0.2 0.4 0.6 0.8 1.0 B [T] 1.0 0.8 0.6 R 0.4 0.2 0.0 0.0 E=5.26 meV (c) 1 0 -1 χ 0.2 0.4 0.6 0.8 1.0 B [T] FIG. 9: Backscattering probability (black line, left axis), and the direction of the current circulation within the ring in the scattering eigenstates [green line, χ = +1 (-1) corresponds to counterclockwise (clockwise) orientation]. Plots (a,b,c) were prepared for E = 5.05, 5.16 and 5.26 meV, for the structure parameters of Fig. 8. bution for two backscattering peaks of Fig. 4(a), a wide one [Fig. 5(a)] and a narrow one [Fig. 5(b)]. In both these cases the charge density is pushed to the left with respect to the electron "velocity", i.e. to the external edge of the ring for the wide peak [Fig. 5(a)] and to the internal edge of the ring for the narrow one [Fig. 5(b)]. The clockwise orientation of the current circula- tion is at high magnetic field translated by the Lorentz force into stronger coupling of the localized states to the conducting channel [Fig. 5(a)]. The electron of the ring- localized state is "ejected" into the horizontal link to the main channel by the Lorentz force. The charge density within the horizontal channel acquires similar values to the ones within the ring. The stronger coupling for these 6 states leads to an increased width of the resonance, or in other words to a decreased lifetime of the ring localized state. Opposite effects are observed for resonances with localized states of counterclockwise current circulation, for which the Lorentz force tends to keep the electron within the ring, increases the localized state lifetime and thus reduces the width of the resonance in the transfer probability. Let us count the resonances for positive B starting from zero field, for the energies marked by thin blue lines in Fig. 3(a). For E = 2.65 meV -- the case that was dis- cussed above -- the Fano resonances that become narrow at high B correspond to even resonance numbers. For E = 2.4 meV the first resonance is obtained for a lo- calized state with a counterclockwise current circulation that grows in the energy for increasing B [Fig. 3(a)]. Accordingly, Fig. 6(a) shows that the resonances that become narrow at high B correspond to odd resonance numbers [see Fig. 3(a)]. For E = 3.4 meV [Fig. 9(b)] the odd (even) peaks increase (decrease) in width as for E = 2.65 meV in consistence with the order of resonances crossed at this energy in function of B of Fig. 3(a). Fig. 7 shows the backscattering probability averaged over the transport window opened near the Fermi energy assumed equal 2.65 meV for T = 100 mK and 300 mK according to Eq.(11) . The resonances that corresponds to ring localized states with counterclockwise current cir- culation are less thermally stable than the ones with the clockwise flowing current. For completeness we present results for the structure that is closer to the one-dimensional limit. For this pur- pose we reduce the channel width from 64 to 32 nm. The shape of the ring with thin channels is displayed in the inset to Fig. 8. We keep the axis of the vertical channel as well as the center of the ring in unchanged positions, and set the outer and inner radii of the ring equal to 108 and 76 nm, respectively to maintain the same Aharonov- Bohm period as in the structure considered above. Fig. 8 shows the positions of localized states as deter- mined by the stabilization method. The stronger trans- verse confinement increases significantly the energy of the low part of the spectrum and enlarges the avoided crossing due to the presence of the contact. The vertical lines in Fig. 8 indicate the energies for which we display also the 0K backscattering probability: 5.05 meV [Fig. 9(a)], 5.16 meV [Fig. 9(b)] and 5.26 meV [Fig. 9(c)]. The energy of 5.05 meV lies within the wide gap opened by the avoided crossing of localized energy levels. The case presented in Fig. 9(b) corresponds to the energy for which no resonances are found in function of the mag- netic field. The oscillations of R have a small amplitude and are due to the Aharonov-Bohm phase shifts within the ring. For the two other energies considered here, the Fano resonances are distinctly present in R(B) depen- dence [Fig. 9(a) and 9(c)]. The maxima of R are close to 1 and decrease in function of B more slowly than for the wide structure considered above [see Figs. 4 and 6], since the shift of the electron density within the channel due 7 to the Lorentz force is hampered by the much stronger confinement. The Fano resonances that become thin in both cases considered in Figs. 9(a) and 9(c) correspond to counterclockwise circulation of the current within the ring. Concluding, the modification of the width and height of resonances are of the same origin and form as for the wide channels, only the effects of magnetic forces are of a reduced strength. We also note that for the thin channels the backscattering probability is of the order of 0.25% outside the resonances as compared to nearly zero for the wider channels discussed above. V. SUMMARY AND CONCLUSION We have studied the electron transport in a channel side-coupled to a quantum ring taking into account the finite width of the structure. The evaluated backscatter- ing probability exhibits Fano peaks for magnetic fields for which the ring-localized states of the energy continuum are degenerate with the Fermi level. We have demon- strated that the ring-localized states that correspond to current circulation producing a magnetic dipole moment that is antiparallel to the external magnetic field produce very thin Fano resonances. The localized states with the dipole moment that is aligned with the magnetic field vector result in resonances that become wider at high magnetic field. The latter are also more thermally stable. The opposite behavior of these two types of resonances is due to shifts of the radial wave function of ring-localized states induced by magnetic forces which increase or de- crease the coupling to the channel depending on the di- rection of the current circulation. The described effect can be used to tune the Fano resonances for potential applications10,11,13 or for magnetic forces detection. Acknowledgements This work was performed within a research project N N202 103938 supported by Min- istry of Science an Higher Education (MNiSW) for 2010- 2013. Calculations were performed in ACK -- CYFRO- NET -- AGH on the RackServer Zeus. 1 R.A. Webb and S. Washburn, Phys. Today 41, 46 (1988). 2 M. Buttiker, Phys. Rev. B 38, 9375 (1988). 3 U. Fano, Phys. Rev. 124, 1866 (1961). 4 A.A. Clerk, X. Waital, and P.W. Brouwer, Phys. Rev. Lett. 86, 4636 (2001). 5 K. Kobayashi, H. Aikawa, S. Katsumoto, and Y. Ire, Phys. Rev. Lett. 88, 256806 (2002). 6 M.E. Torio, K. Hallberg, A.H. Ceccatto, and C.R. Proetto, Phys. Rev. B 65, 085302 (2002). 7 T.F. Fang, W. Zuo, and J.Y. Chen, Phys. Rev. B 77, 125136 (2008). 8 B.P. Bu lka and P. Stefa´nski, Phys. Rev. Lett. 86, 5128 (2001). 9 K. Kang, S.Y. Cho, J.J. Kim, and S.C. Shin, Phys. Rev. B 63, 113304 (2001). 10 M.E. Torio, K. Hallberg, S. Flach, A.E. Miroshnichenko, and M. Titov, Eur. Phys. J. B 37, 399 (2004). 11 M. Lee and C. Bruder, Phys. Rev. B 73, 085315 (2006). 12 J. Gores, D. Goldhaber-Gordon, S. Heemeyer, M.A. Kast- ner, H. Shtrikman, D. Mahalu, and U. Meirav, Phys. Rev. B 62, 2188 (2000). 13 C. Morfonios, D. Buchholz, and P. Schmelcher, Phys. Rev. B 80, 035301 (2009). 14 Z.Y. Zeng, F. Claro, and A. P´erez, Phys. Rev. B 65, 085308 (2002). 15 R. Zitko, Phys. Rev. B 81, 115316 (2010). 16 A.E. Miroshnichenko and Y.S. Kivshar, Phys. Rev. E 72, 056611 (2005). (2005). 23 A. Fuhrer, S. Luscher, T. Ihn, T. Heinzel, K. Ensslin, W. Wegscheider, and M. Bichler, Nature (London) 413, 822 (2001). 24 U. F. Keyser, C. Fuhner, S. Borck, R. J. Haug, M. Bichler, G. Abstreiter, and W. Wegscheider, Phys. Rev. Lett. 90, 196601 (2003). 25 W. G. van der Wiel, Yu. V. Nazarov, S. De Franceschi, T. Fujisawa, J. M. Elzerman, E. W. G. M. Huizeling, S. Tarucha, and L. P. Kouwenhoven, Phys. Rev. B, 67 033307 (2003). 26 A. Muhle, W. Wegscheider, and R. J. Haug, Appl. Phys. Lett. 91, 133116 (2007). 27 F. Martins, B. Hackens, M.G. Pala, T. Ouisse, H. Sellier, X. Wallart, S. Bollaert, A. Cappy, J. Chevrier, V. Bayot, and S. Huant, Phys. Rev. Lett. 99, 136807 (2007). 28 E. Strambini, V. Piazza, G. Biasiol, L. Sorba, and F. Bel- tram, Phys. Rev. B 79, 195443 (2009). 29 E. Strambini, L. Chirolli, V. Giovannetti, F. Taddei, R. Fazio, V. Piazza, and F. Beltram, Phys. Rev. Lett. 104, 170403 (2010). 30 L. Chirolli, E. Strambini, V. Giovannetti, F. Taddei, V. Piazza, R. Fazio, F. Beltram, and G. Burkard, arXiv:1004.1895v1. 31 P. Kwiat, H. Weinfurter, T. Herzog, A. Zeilinger, and M.A. Kasevich, Phys. Rev. Lett. 74, 4763 (1995). 32 V.A. Mandelshtam, T.R. Ravuri, and H.S. Taylor, Phys. Rev. Lett. 70, 1932 (1993). 17 A. Chakrabarti, Phys. Rev. B 74, 205315 (2006). 18 K. Kobayashi, H. Aikawa, A. Sano, S. Katsumoto, and Y. 33 M. Buttiker, Phys. Rev. Lett. 57, 1761 (1986). 34 R. Akis, P. Vasilopoulos, and P. Debray, Phys. Rev. B 56, Iye, Phys. Rev. B 70, 035319 (2004). 19 A.A. Aligia, and L.A. Salguero, Phys. Rev. B 70, 075307 (2004). 20 R. Zitko and J. Bonca, Phys. Rev. B 73, 035332 (2006). 21 B. Szafran and F.M. Peeters, Phys. Rev. B 72, 165301 (2005). 22 B. Szafran and F.M. Peeters, Europhys. Lett. 70, 810 9594 (1997). 35 J.U. Nockel, Phys. Rev. B 46, 15348 (1992). 36 The barrier height V = 200 meV is large as compared to the energy range of the lowest subband transport which for the system of Fig. 1 corresponds to the energy E lower than 4.5 meV. The wave functions vanish exponen- tially in the barriers with a penetration length equal to ¯h/p2m∗(V − E) ≃ ¯h/p2m∗(V ), which equals roughly 1.7 nm. This is small as compared to the width of the channels (64 nm). The effect of the finite barrier is an in- creased effective width of the channels. The bridge in the structure of Fig. 1 is necessary to allow the channel elec- trons to reach the ring. 37 The finite difference approach naturally accounts for the evanescent modes [P.F. Bagwell, Phys.Rev. B 41, 10 354 (1990)] which appear near the scattering region. The evanescent modes correspond to higher subbands and imaginary wave vectors (negative kinetic energy). They vanish exponentially away from the scattering point with a characteristic length equal to λ = ¯h/p2m(En − E), where En is the threshold energy for the n − th subband. For the parameters considered in this paper the length is maxi- mal in Fig. 6 (b) and equal to λ = 23 nm which results from the second subband threshold En = 4.5 meV and the considered energy E = 3.4 meV. This length is small as 8 compared to the length of the computational box which covers 200 nm below and 200 nm above the contact to the ring. The evanescent modes do have influence on the c−k and dk values (reflected and transmitted amplitudes), but do not perturb their extraction from the finite difference wave function which is performed at the ends of the box of a proper size, where it takes the asymptotic form given by Eqs. (5) and (6). 38 F.M. Peeters, in Science and Engineering of One- and Zero-Dimensional Semiconductors, edited by S.P. Beau- mont and C.M. Sotomajor Torres (Plenum, New York, 1990), p. 107. 39 M. Governale and C. Ungarelli, Phys. Rev. B 58, 7816 (1998). 40 L. Wendler, V.M. Fomin, and A.A. Krokhin, Phys. Rev. B 50, 4642 (1994); L. Wendler, V.M. Fomin, A.V. Chaplik, and A.O. Govorov, Phys. Rev. B 54, 4794 (1996). 

1907.02815

1

1907

Efficient heating of single-molecule junctions for thermoelectric studies at cryogenic temperatures

The energy dependent thermoelectric response of a single molecule contains valuable information about its transmission function and its excited states. However, measuring it requires devices that can efficiently heat up one side of the molecule while being able to tune its electrochemical potential over a wide energy range. Furthermore, to increase junction stability devices need to operate at cryogenic temperatures. In this work we report on a new device architecture to study the thermoelectric properties and the conductance of single molecules simultaneously over a wide energy range. We employ a sample heater in direct contact with the metallic electrodes contacting the single molecule which allows us to apply temperature biases up to $\Delta T = 60$K with minimal heating of the molecular junction. This makes these devices compatible with base temperatures $T_\mathrm{bath} <2$K and enables studies in the linear ($\Delta T \ll T_\mathrm{molecule}$) and non-linear ($\Delta T \gg T_\mathrm{molecule}$) thermoelectric transport regimes.

cond-mat.mes-hall

cond-mat

Efficient heating of single-molecule junctions for thermoelectric studies at cryogenic temperatures Pascal Gehring,1, a) Martijn van der Star,1 Charalambos Evangeli,2, 3 Jennifer J. Le Roy,2 Lapo Bogani,2 Oleg V. Kolosov,3 and Herre S. J. van der Zant1 1)Kavli Institute of Nanoscience, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands 2)Department of Materials, University of Oxford, Parks Road, OX1 3PH, Oxford, United Kingdom 3)Department of Physics, Lancaster University, Bailrigg, LA1 4YB, Lancaster, United Kingdom (Dated: 8 July 2019) The energy dependent thermoelectric response of a single molecule contains valuable information about its transmission function and its excited states. However, measur- ing it requires devices that can efficiently heat up one side of the molecule while being able to tune its electrochemical potential over a wide energy range. Furthermore, to increase junction stability devices need to operate at cryogenic temperatures. In this work we report on a new device architecture to study the thermoelectric properties and the conductance of single molecules simultaneously over a wide energy range. We employ a sample heater in direct contact with the metallic electrodes contacting the single molecule which allows us to apply temperature biases up to ∆T = 60 K with minimal heating of the molecular junction. This makes these devices compatible with base temperatures Tbath < 2 K and enables studies in the linear (∆T ≪ Tmolecule) and non-linear (∆T ≫ Tmolecule) thermoelectric transport regimes. 9 1 0 2 l u J 5 ] l l a h - s e m . t a m - d n o c [ 1 v 5 1 8 2 0 . 7 0 9 1 : v i X r a a)Electronic mail: p.gehring@tudelft.nl 1 Theory predicts that electrical and thermoelectric properties of single molecules can be tailored by chemical design. For example, adding pendant groups to a conjugated molecule backbone can introduce sharp features in its energy dependent transmission probability, because of quantum interference effects,1 and such sharp features should generate excep- tionally high thermoelectric efficiencies2. Furthermore, single molecules can host a rich variety of physical effects:3 strong electron-phonon interactions4, strong correlations and Kondo effects5, or exotic blockade phenomena.6 All these are predicted to strongly influence the thermoelectric properties7 -- 10, but these predictions remain untested, because of a lack of appropriate experimental platforms. In order to perform detailed thermoelectric characterisations of single molecules, the de- vice architecture needs to fulfill the following conditions: the device needs to be compatible with methods to contact single molecules; a gate electrode is necessary for a full characterisa- tion of the thermoelectric properties of the single-molecule junction; because of the thermal instabilities in molecular junctions, the devices need to be compatible with cryogenic tem- peratures; and for the same reason the temperature difference between the hot and the cold side ∆T = Thot − Tcold in the molecular junction must not heat excessively the molecule itself. So far, only a few device architectures exist that fulfill some of the aforementioned conditions, based on graphene11 or Au electrodes12. These devices suffer, however, from low heating efficiencies (50− 150 mK mW−1) and, in devices with a side heater, the temperature profile along the channel is approximately linear13 so that high heater powers are necessary to apply ∆T across short junctions. For the case of graphene junctions, a side heater pro- duces strong heating of the cold side of the junction, characterized by (Tcold−Tbath)/∆T ≈ 5, where Tbath is the temperature of the cryostat. This makes these devices not compatible with measurements at low cryogenic temperatures. Here, we develop a novel device architecture for simultaneously studying the electric and thermoelectric properties of single molecules as a function of the gate voltage Vg. Fabrication is based on electromigration and self-breaking of Au, leading to several key advantages: Au enables access to different tunnel coupling strengths (e.g. by using thiol bonds, and spacer linkers)14; self-breaking15 of the Au bridges can prevent the formation of spurious quantum dots, which is sometimes a problem for carbon-based leads16; the close proximity of the sample heater to the leads enables efficient heat transfer, while reducing heating of the single molecule at temperature Tmolecule = (Thot + Tcold)/2 (thus ensuring device stability) 2 and enabling experiments at Tbath < 2 K. The improved heating efficiency also provides access to a wide ∆T range (mK to few tens of K), opening the way to the study of the thermoelectric properties of single-molecule junctions in the linear (∆T ≪ Tmolecule) and non-linear (∆T ≫ Tmolecule) thermal bias regimes. Moreover, this novel method allows the simultaneous measurement of the gate-dependent conductance G(Vg) and thermoelectric current Ith(Vg). This eliminates the problem that small drifts of the signals (because of hysteresis effects of the gates or activation of charge traps in the gate oxide) can hinder a direct comparison of data sets when the two quantities are measured subsequently, as in previous devices. (a) (c) (b) (d) (e) I heater V g V sd I FIG. 1: (a) - (d) Overview of the fabrication process. (a) Fabrication of the local back gate electrode (purple) and sample heaters (blue). (b) Deposition of a thin Al2O3 insulating layer on top of the whole device. (c) Deposition of a Au bridge (yellow) which is (d) contacted by two four-terminal thermometers (orange). (e) False-color scanning electron micrograph of the single-molecule transistor architecture, consisting of a thin Au bridge (yellow) on top of a gate electrode (purple) connected by two four-terminal thermometers (orange) which are on top of the sample heaters (blue). The schematic circuit diagram indicates the terminals used for G and Ith measurements: a source-drain voltage Vsd can be applied to the drain while a current to ground I is measured at the source. Vg is applied via the back gate with respect to ground. Heater currents Iheater are applied to the sample heater. Scale bar: 2 µm. 3 The junctions are fabricated following the scheme depicted in Figures 1(a) - (d). A Pd sample heater (3 nm Ti/27 nm Pd) and Pd gate electrode (1 nm Ti/6 nm Pd) were patterned on a Si wafer with 817 nm SiO2 using standard electron beam lithography and electron beam evaporation (Figure 1a). A thin gate electrode is used to reduce thermal transport between drain and source lead. Pd is used because it is known to form uniform thin layers with low surface roughness.17 In a second step a 10 nm Al2O3 insulating layer is globally applied by atomic layer deposition (Figure 1b). This layer serves as a gate dielectric and as an insulation layer to electrically insulate the sample heater from the drain and source leads.18 Thereafter, a 12 nm thick bow-tie shaped Au bridge (narrowest part < 60 nm) is evaporated (Figure 1c) and electrically contacted by two four-terminal thermometers (5 nm Ti/65 nm Au, Figure 1d). The effective temperature drop on a molecule trapped between the two Au contacts depends on the thermal resistances of the Au bridge. Therefore a short channel length should be used to reduce its thermal resistance which ensures thermalisation with the heated Au contact. On the other hand, very short channels promote direct heating of the 'cold' contact by the sample heater. In this study we chose a short channel length of 1 µm. Figure 1e shows a false-color scanning electron microscopy image of a final device. To use these devices for studying the thermoelectric properties of single molecules we open a nm sized gap in the Au bridge by electromigration19 followed by self-breaking15 to avoid the formation of Au clusters inside the junction. In the following we describe the methods for estimating ∆T created by the sample heater after electromigration. We employed two calibration techniques: Scanning thermal micro- scope (SThM) mapping in high vacuum and resistance thermometer method using the drain and source contacts as thermometers. For the former, we used a home-built high vacuum SThM20 with commercially available (Anasys Instruments, AN-300) doped silicon probes which are geometrically similar to standard micromachined AFM probes. The probe tem- perature Tprobe can be controlled with an integrated heater at the end of the cantilever, which also acts as a temperature sensor when the tip is in contact with the sample. The electrical response of the probe heater as a function of excess mean probe temperature (∆Tprobe = Tprobe − Tbath) was calibrated on a heated stage inside the high vacuum chamber, following a procedure described elsewhere21. Two different quantitative SThM methods were employed to estimate ∆T : the null-point method22 and a non-equilibrium thermometry method23,24. In the null-point method the 4 probe is brought into contact with the sample for different Tprobe while the SThM response is recorded. A jump in the SThM response signal is typically observed at the tip-sample mechanical contact when the probe apex and sample are at different temperatures (examples in Figure S1, Supporting Information). The jump is positive/negative when the temperature of the probe apex, Tapex, is larger/smaller than that of the sample, Tsample, and zero when they are the same. Tapex in contact with the sample has been found21 to be 88% of Tprobe. Using this procedure, we measured the Texcess = Tsample − Tbath of the drain (hot) lead for 4 different powers applied to the sample heater which is plotted in Figure 2 (a). Linear regression yields a conversion factor of 9.8 ± 1.2 K mW−1, with an error originating mainly from the temperature calibration of the probe and the estimation of the jump of the SThM signal (see Supporting Information), especially for low Tprobe where the SThM signal noise is comparable to the signal jump. The second SThM method relies on non-equilibrium thermometry where an AC bias voltage is applied to the sample heater and the resulting variations of Texcess are detected by the SThM tip. The Texcess map is extracted through the relation Texcess = ∆Tprobe ∆VAC−∆VDC where ∆VAC is the AC SThM response detected at the second harmonic and ∆VDC the DC ∆VAC , SThM signal due to heat flux from the sample to the tip. Modulation of the sample heater with high frequencies can lead to damping of the SThM signal since thermal equilibrium can only be reached within a time scale τth given by the total thermal capacitance and all thermal resistances of our device. For the temperature mapping, a modulation frequency of 7 Hz is used, due to limitations in the lowest possible scanning speed, which is slightly bigger than 1/τth and which results in a reduction of SThM signal by about 10%. We account for this damping by rescaling of the Texcess maps using Figure 4 (c). The resulting map for a device with Pheat = 0.38 mW applied to the sample heater is shown in Figure 2 (b). From this Texcess map and a line cut through this map in Figure 2 (c) we observe that for a heating of the hot (left) contact by about 3 K the cold (right) contact only heats up by about 0.14 K, which yields a very low (Tcold − Tbath)/∆T ≈ 0.05. This low heating of the cold side allows us to estimate ∆T from the excess temperature of the drain lead in Figure 2 (a) using ∆T ≈ Texcess. It is worth to mention that the temperature of the gold bridge differs noticeably from that of the drain and source contacts. This has been observed in previous studies12 and would result in an overestimation of ∆T across the molecule in the centre of the junction. However, SThM only accesses the phonon (lattice) temperature Tph, 5 40 30 20 10 0 0 (a) ) K ( s s e c x e T (d) T excess (K) 0 5 (b) ) K ( 1 P 2 3 4 heater (mW) s s e c x e T (e) 4 (c) 0 0 1 3 4 2 x (μm) 5 ) K ( T T drain T source 130 110 90 70 50 0 4 6 2 heater (mW) P ) K ( T Δ 80 60 40 20 0 0 4 6 2 heater (mW) P FIG. 2: (a) Results of the SThM null-point method. Excess temperature Texcess of the drain (hot) lead as a function of heater power. The error of the linear fit is indicated by the red shaded area. (b) Temperature map of the device recorded using non-equilibrium thermometry method at Pheater = 0.38 mW. The dotted lines indicate the position of the drain and source leads, and the gold bridge, respectively. A line cut along the device (indicated by arrows) is shown in (c). (d),(e) Results of the calibration using the resistance thermometer method. (d) Temperature of the drain and source lead as a function of heater power. (e) Temperature drop ∆T = Tdrain − Tsource across the junction as a function of heater power. The red shaded area indicates the error of the linear fit. and the electron temperature Te (which drives thermoelectric effects) can be much higher when using efficient sample heater in direct contact with leads18. Since we cannot access the real drop in Te on the single-molecule junction, in the remainder of this paper we use the ∆T between the drain and source lead for calculations, which leads to an underestimation of the thermoelectric coefficients and efficiencies. The second technique used to estimate ∆T is the resistance thermometer method.25 -- 27 To this end, we use the four contacts connecting the drain and source lead to first measure their 4-terminal resistance as a function of Tbath in a cryostat. Thereafter the sample temperature is held constant (here Tbath = 50 K) and the 4-terminal resistance of the drain and source leads are measured as a function of dissipated heater power. Combining of both measurement 6 results allows estimating Tdrain and Tsource as a function of heater power Pheater (see Figure 2 (d)). It can be seen that the (hot) drain lead in direct contact with the sample heater heats up by tens of Kelvin when increasing the heater power while the (cold) source lead stays almost at Tbath. Using this data we estimate ∆T as a function of Pheater (Figure 2 (e)). We find that ∆T increases linearly with Pheater, which allows to accurately apply small ∆T biases. Extracting the slope of 10.7 ± 0.8 K/mW, we find a heating efficiency of ∆T /(PheatL) = 10.7 ± 0.8 K mW−1 µm−1 at 50 K. This value is close to the value found using the SThM methods above. The efficiency found in our devices is orders of magnitude higher than that found in devices with side heaters11 and it is comparable to similar devices designed to study ther- moelectric properties of nanowires which use sample heater patterned on top of the leads18. Such a high heating efficiency allows to drive systems into the non-linear regime where ∆T becomes comparable to, or even exceeds Tbath. This is demonstrated in Figure 2 (e) (which was recorded at Tbath = 50 K) for Pheater > 5 mW, where ∆T > 50 K. Moreover, from the data in Figure 2 (d) we find a low (Tcold − Tbath)/∆T < 0.026, which indicates minimal heating of the cold reservoir and the molecule. This value, which is significantly lower than previously-reported values11,18), ensures stability of the molecular junction and enables experiments at Tbath < 2 K. In the following we test the device architecture to measure the thermocurrent of a single [Gd(tpy-SH)2(NCS)3] molecule. by immersing the sample in a 0.5 mM molecule solution in dichlormethan after electromigration and self breaking. We observe molecular junction formation indicated by occurrence of gate dependent transport features at Tbath = 1.8 K in 7 out of 47 junctions. This junction formation yield of ≈ 15% is similar to values that we typically observe for electromigrated Au electrodes.28 In this paper we focus on demonstrat- ing the suitability of our junctions for thermoelectric characterisation of single molecules and present the data for one selected device. Figure 3 (a) shows the differential conductance dI/dVsd of a molecular junction as a function of bias voltage Vsd and Vg. Two regions with low dI/dVsd (yellow) are separated by two crossing lines of high dI/dVsd. These lines are attributed to the borders of so- called Coulomb diamonds. The current inside the two adjacent diamonds is suppressed due to Coulomb blockade, whereas sequential electron tunneling occurs inside the hour-glass 7 (a) (b) I sd I 0 I th,+ I sd I 0 I th,+ I / V d d sd (nS) 8 6 4 2 0 4 2 0 -2 -4 -1.05 -1 -0.95 -0.9 V g (V) (d) ) A p ( ) 0 I - + , h t I ( 0.4 0.2 0 -0.2 -0.4 ) V m ( d s V (c) ) S n ( d s V / ) 0 I - d s I ( 3 2 1 0 ) V ( g V ) V m ( d s V 0 ) A m ( r e t a e h I 0 0 2 4 6 8 10 t (s) (e) ) h / B2 k ( G / 2 L 0.6 0.4 0.2 0 -1 -0.95 -0.9 -1 -0.95 -0.9 -1 -0.95 -0.9 V g (V) V g (V) V g (V) FIG. 3: (a) Differential conductance dI/dVsd as a function of applied gate Vg and bias Vsd voltages. (b) Measurement scheme used to measure Ith(Vg) and G(Vg) as a function of gate voltage. The shaded regions indicate the time windows in which current measurements are performed. (c) Conductance, (d) Ith and (e) power factor as a function of Vg. shaped region.3 Ith and G were then measured simultaneously in the device configuration shown in Figure 1 (e) following the measurement protocol depicted in Figure 3 (b). Vg is first ramped to the desired value and a small Vsd = 0.5 mV is applied. After a short settling time Isd is measured, Vsd is set to zero and a offset current I0 may be measured, which can originate from gate leakage currents or offsets in the current pre-amplifier. Subsequently, a heater current Iheater = 0.1 mA (P = 2.6µW) is applied to the sample heater, followed by a settling time and a measurement of the raw thermocurrent, Ith,+. These measurement steps 8 are repeated for each gate voltage value. Using the three measured current values the conductance G = (Isd − I0)/Vsd and the thermocurrent Ith = Ith,+ − I0 are calculated. The power factor S2G = (Vth/∆T )2G = (Ith/∆T )2/G, which is a measure for the amount of energy that can be generated from a certain ∆T , is thus determined directly. Figure 3 (c) and (d) show the results of this measurement on the molecular junction. The conductance Isd/Vsd peaks at around Vg = −0.96 V. This indicates the energetic position of the charge degeneracy point where the transition from the N to the N + 1 charge state of the molecule occurs (corresponds to closing point of the Coulomb diamonds in Figure 3 (a)). Furthermore, we extract the gate coupling factor α = Cg/(Cs + Cd + Cg) = 33 meV/V, from the slopes of the Coulomb diamond following Ref. 29. This gate coupling factor, which is a factor 4-5 higher than the typical values found for devices using SiO2 back gates30,31, enables efficient tuning of the single-molecule junction and allows thermoelectric studies over a wide energy range, of about ±400 meV, as estimated using the typical break down voltages of 12-14 V found in our devices. Figure 3 (d) shows Ith = Ith,+ − I0 as a function of Vg, displaying a resulting curve that is S-shaped and changes sign at the charge degeneracy point. This sign change indicates that the transition from electron- to hole-like thermocurrents occurs when crossing the charge degeneracy point, in agreement with theoretical predictions and previous experiments.3,32 By tuning the system far away from resonance, Ith vanishes. Combining the data in Figure 3 (c) and (d) and using ∆T ≈ 30 mK obtained from our calibration allows calculating the gate-dependent power factor S2G = L2/G, where S = −Vth/∆T is the Seebeck coefficient, Vth is the thermovoltage and L = −Ith/∆T is the thermal response coefficient. The result of this calculation is shown in Figure 3 (e). The power factor can be tuned from zero to about 0.4k2 B/h, which is close to the theoretical limit of (1/2.2)k2 B/h predicted for a single quantum level.11 In the remainder of this paper we test if the device platform developed in this study is suitable for AC thermoelectric measurements.26 For this purpose an AC current at frequency f is applied to the sample heater and Ith is measured at the second harmonic, 2f . As can be shown33 the maximum signal in the second harmonic is at a phase of 90◦ with respect to the excitation. Furthermore, the raw data needs to be multiplied by a factor of 2√2 to convert it from rms to peak-to-peak and to correct the shift in reference when locking to the second harmonic.11 Figure 4 (a) shows the AC thermocurrent as a function of gate voltage 9 measured with f = 3 Hz for the same device discussed above. Line shape and amplitude of (a) ) A p ( h t I 0.4 0.2 0 -0.2 -0.4 (b) ) A p ( h t I 0.2 0.1 0 (c) 1 ) . u . a ( M h T S V 0.5 0 -1 -0.95 V g (V) -0.9 100 101 102 f (Hz) FIG. 4: (a) AC thermocurrent (f = 3 Hz, Iheater = 0.1 mA, Pheater = 26 µW) as a function of gate voltage. (b) Thermocurrent at Vg = −0.965 V as a function of modulation frequency of the sample heater. (c) SThM signal on the drain (hot) contact as a function of modulation frequency of the sample heater. the AC measurement match the results of the DC measurement in Figure 3 (d) well. This changes if higher frequencies are used: in Figure 4 (b) the AC thermocurrent measured at fixed gate voltage (Vg = −0.965 V) as a function of modulation frequency of the sample heater is shown. Above a frequency of about 3 Hz the signal amplitude drops from its DC value to zero when reaching frequencies of about 30 Hz. This can be explained by the thermal equilibrium time of the system as discussed above. To illustrate this the SThM signal measured on the hot contact as a function of sample heater excitation frequency is shown in Figure 4 (c). A similar trend as for the thermocurrent signal can be observed where a deviation from the DC signal strength occurs at f > 3 Hz. In summary, we developed a new device architecture and the first robust measurement 10 protocol that allows measuring the thermoelectric properties of single molecules at cryogenic temperatures, over a wide energy range. The close proximity of the sample heater to the electrical contacts yields a high heating efficiency and low global heating of the molecular junction itself. This ensures device stability and allows to accurately study thermoelectric effects over wide ∆T ranges. Furthermore, we demonstrate that the gate dependent ther- mocurrent and conductance can be measured in parallel and that the devices are suitable for AC measurements, if the excitation frequency is chosen to be smaller than the thermal response time of the system. The devices presented in this study could thus be readily used to study the thermoelectric properties of single molecules in the non-linear regime34 or to investigate the thermoelectric response of single-molecule magnets9 or high-spin molecules in the Kondo regime10. What is more, the Gd-based molecules used in this study are promis- ing candidates for observing single-molecule magneto-cooling effects35 which are now within experimental reach. ACKNOWLEDGMENTS This work was supported by the EC H2020 FET Open project 767187 QuIET and ERC (StG-OptoQMol-338258 and CoG-MMGNRs-773048). P.G. and J.R. acknowledge Marie Skodowska-Curie Individual Fellowships (Grant TherSpinMol-748642 and SpinReMag- 707252) from the European Unions Horizon 2020 research and innovation programme. REFERENCES 1H. Valkenier, C. M. Gudon, T. Markussen, K. S. Thygesen, S. J. van der Molen, and J. C. Hummelen, "Cross-conjugation and quantum interference: a general correlation?" Phys. Chem. Chem. Phys. 16, 653 -- 662 (2014). 2C. M. Finch, V. M. Garc´ıa-Su´arez, and C. J. Lambert, "Giant thermopower and figure of merit in single-molecule devices," Phys. Rev. B 79, 033405 (2009). 3P. Gehring, J. M. Thijssen, and H. S. J. van der Zant, "Single-molecule quantum-transport phenomena in break junctions," Nature Reviews Physics 1, 381 -- 396 (2019). 4J. Koch, F. von Oppen, and A. V. Andreev, "Theory of the franck-condon blockade regime," Phys. Rev. B 74, 205438 (2006). 11 5W. Liang, M. P. Shores, M. Bockrath, J. R. Long, and H. Park, "Kondo resonance in a single-molecule transistor," Nature 417, 725 -- 729 (2002). 6J. de Bruijckere, P. Gehring, M. Palacios-Corella, M. Clemente-Le´on, E. Coronado, J. Paaske, P. Hedegard, and H. S. J. van der Zant, "Ground-state spin blockade in a single-molecule junction," Phys. Rev. Lett. 122, 197701 (2019). 7J. Koch, F. von Oppen, Y. Oreg, and E. Sela, "Thermopower of single-molecule devices," Phys. Rev. B 70, 195107 (2004). 8J. K. Sowa, J. A. Mol, and E. M. Gauger, "Marcus theory of thermoelectric- ity in molecular junctions," The Journal of Physical Chemistry C 123, 4103 -- 4108 (2019), https://doi.org/10.1021/acs.jpcc.8b12163. 9R.-Q. Wang, L. Sheng, R. Shen, B. Wang, and D. Y. Xing, "Thermoelectric effect in single-molecule-magnet junctions," Phys. Rev. Lett. 105, 057202 (2010). 10T. A. Costi and V. Zlati´c, "Thermoelectric transport through strongly correlated quantum dots," Phys. Rev. B 81, 235127 (2010). 11P. Gehring, A. Harzheim, J. Spice, Y. Sheng, G. Rogers, C. Evangeli, A. Mishra, B. J. Robinson, K. Porfyrakis, J. H. Warner, O. V. Kolosov, G. A. D. Briggs, and J. A. Mol, "Field-effect control of graphenefullerene thermoelectric nanodevices," Nano Letters 17, 7055 -- 7061 (2017). 12Y. Kim, W. Jeong, K. Kim, W. Lee, and P. Reddy, "Electrostatic control of thermoelec- tricity in molecular junctions," Nat. Nanotechnol. 9, 881 -- 885 (2014). 13J. Moon, J.-H. Kim, Z. C. Chen, J. Xiang, and R. Chen, "Gate-modulated thermoelectric power factor of hole gas in gesi coreshell nanowires," Nano Letters 13, 1196 -- 1202 (2013). 14T. A. Su, M. Neupane, M. L. Steigerwald, L. Venkataraman, and C. Nuckolls, "Chemical principles of single-molecule electronics," Nature Reviews Materials 1, 16002 (2016). 15K. ONeill, E. A. Osorio, and H. S. J. van der Zant, "Self-breaking in planar few-atom au constrictions for nanometer-spaced electrodes," Applied Physics Letters 90, 133109 (2007). 16P. 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, "Quantum interference in graphene nanoconstrictions," Nano Letters 16, 4210 -- 4216 (2016), pMID: 27295198, https://doi.org/10.1021/acs.nanolett.6b01104. 17S. Nazarpour, A. Cirera, and M. Varela, "Material properties of aupd thin alloy films," Thin Solid Films 518, 5715 -- 5719 (2010). 12 18J. G. Gluschke, S. Fahlvik Svensson, C. Thelander, and H. Linke, Nanotechnology 25, 385704 (2014). 19H. Park, A. K. L. Lim, A. P. Alivisatos, J. Park, and P. L. McEuen, "Fabrication of metallic electrodes with nanometer separation by electromigration," Applied Physics Letters 75, 301 -- 303 (1999). 20M. E. Pumarol, M. C. Rosamond, P. Tovee, M. C. Petty, D. A. Zeze, V. Falko, and O. V. Kolosov, "Direct nanoscale imaging of ballistic and diffusive thermal transport in graphene nanostructures," Nano Letters 12, 2906 -- 2911 (2012). 21P. Tovee, M. Pumarol, D. Zeze, K. Kjoller, and O. Kolosov, "Nanoscale spatial resolu- tion probes for scanning thermal microscopy of solid state materials," Journal of Applied Physics 112, 114317 (2012). 22G. Hwang, J. Chung, and O. Kwon, "Enabling low-noise null-point scanning thermal microscopy by the optimization of scanning thermal microscope probe through a rigorous theory of quantitative measurement," Review of Scientific Instruments 85, 114901 (2014). 23F. Menges, P. Mensch, H. Schmid, H. Riel, A. Stemmer, and B. Gotsmann, "Tem- perature mapping of operating nanoscale devices by scanning probe thermometry," Nature Communications 7, 10874 (2016). 24A. Harzheim, J. Spiece, C. Evangeli, E. McCann, V. Falko, Y. Sheng, J. H. Warner, G. A. D. Briggs, J. A. Mol, P. Gehring, and O. V. Kolosov, "Geometrically enhanced thermoelectric effects in graphene nanoconstrictions," Nano Letters 18, 7719 -- 7725 (2018). 25P. Kim, L. Shi, A. Majumdar, and P. L. McEuen, "Thermal transport measurements of individual multiwalled nanotubes," Phys. Rev. Lett. 87, 215502 (2001). 26J. P. Small, L. Shi, and P. Kim, "Mesoscopic thermal and thermoelectric measurements of individual carbon nanotubes," Solid State Communications 127, 181 -- 186 (2003). 27J. P. Small, K. M. Perez, and P. Kim, "Modulation of thermoelectric power of individual carbon nanotubes," Phys. Rev. Lett. 91, 256801 (2003). 28E. Burzur´ı, R. Gaudenzi, and H. S. J. van der Zant, "Observing mag- netic anisotropy in electronic transport through individual single-molecule magnets," Journal of Physics: Condensed Matter 27, 113202 (2015). 29R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen, "Spins in few-electron quantum dots," Rev. Mod. Phys. 79, 1217 -- 1265 (2007). 30E. A. Osorio, T. Bjørnholm, J.-M. Lehn, M. Ruben, and H. S. J. 13 van der Zant, "Single-molecule transport in three-terminal devices," Journal of Physics: Condensed Matter 20, 374121 (2008). 31P. Gehring, J. K. Sowa, J. Cremers, Q. Wu, H. Sadeghi, Y. Sheng, J. H. Warner, C. J. Lambert, G. A. D. Briggs, and J. A. Mol, "Distinguishing lead and molecule states in graphene-based single-electron transistors," ACS Nano 11, 5325 -- 5331 (2017), pMID: 28423272, https://doi.org/10.1021/acsnano.7b00570. 32L. Rinc´on-Garc´ıa, C. Evangeli, G. Rubio-Bollinger, and N. Agraıt, "Thermopower mea- surements in molecular junctions," Chem. Soc. Rev. 45, 4285 -- 4306 (2016). 33Y. M. Zuev, "Nanoscale thermoelectric energy conversion," PhD thesis, Columbia Univer- sity (2011). 34D. S´anchez and R. L´opez, "Scattering theory of nonlinear thermoelectric transport," Phys. Rev. Lett. 110, 026804 (2013). 35G. Karotsis, M. Evangelisti, S. Dalgarno, and E. Brechin, "A calix[4]arene 3d/4f magnetic cooler," Angewandte Chemie International Edition 48, 9928 -- 9931. 14 

1003.5430

1

1003

Superconducting transition in Nb nanowires fabricated using focused ion beam

Making use of focused Ga-ion beam (FIB) fabrication technology, the evolution with device dimension of the low-temperature electrical properties of Nb nanowires has been examined in a regime where crossover from Josephson-like to insulating behaviour is evident. Resistance-temperature data for devices with a physical width of order 100 nm demonstrate suppression of superconductivity, leading to dissipative behaviour that is shown to be consistent with the activation of phase-slip below Tc. This study suggests that by exploiting the Ga-impurity poisoning introduced by the FIB into the periphery of the nanowire, a central superconducting phase-slip nanowire with sub-10 nm dimensions may be engineered within the core of the nanowire.

cond-mat.mes-hall

cond-mat

Superconducting transition in Nb nanowires fabricated using focused ion beam G C Tettamanzi1,2*, C I Pakes3, A Potenza4, S. Rubanov5, C H Marrows4 and S Prawer¹ 1 School of Physics, University of Melbourne, Victoria 3010, Australia 2 Kavli Institute of Nanoscience, TU Delft, Lorentzweg 1, 2628 CJ DELFT, The Netherlands 3 Department of Physics, La Trobe University, Victoria 3086, Australia 4 School of Physics and Astronomy, University of Leeds, Leeds LS2 9JT, United Kingdom 5 Bio21 Institute, University of Melbourne, Australia *E-mail: G.C.Tettamanzi@tudelft.nl Abstract. Making use of focused Ga-ion beam (FIB) fabrication technology, the evolution with device dimension of the low-temperature electrical properties of Nb nanowires has been examined in a regime where crossover from Josephson-like to insulating behaviour is evident. Resistance-temperature data for devices with a physical width of order 100 nm demonstrate suppression of superconductivity, leading to dissipative behaviour that is shown to be consistent with the activation of phase-slip below Tc. This study suggests that by exploiting the Ga-impurity poisoning introduced by the FIB into the periphery of the nanowire, a central superconducting phase-slip nanowire with sub-10 nm dimensions may be engineered within the core of the nanowire. 1. Introduction The use of focused ion beam technology to fabricate nanoscale superconducting components has developed recent interest, particularly as an approach to engineer miniature, low-noise SQUID technology based upon nanowire junctions [1-3]. However, the manner in which superconductivity is suppressed as the nanowire width is reduced has not been explored for low-Tc FIB-engineered devices. This is of importance in assessing the extent to which FIB- patterned components can be further miniaturized for superconducting nano-electronics and motivated particularly by the possibility of introducing new device functionality that exploits phase-slip processes driven by macroscopic quantum tunneling in the one-dimensional limit [4].  For superconducting components with dimensions on the order of the Ginzburg-Landau (GL) coherence length, local fluctuations may lead to dissipation at temperatures below the superconducting transition temperature, Tc [5]. A residual resistance close to Tc has been extensively reported [4], and is understood within a theoretical framework of phase-slip arising from thermal activation over the free energy barrier [6]. Demonstration of quantum phase-slip (QPS), for which a significant residual resistance will extend far below Tc, is non-trivial. The main challenge comes from the need to fabricate a wire of width just a few nanometers, which is beyond the capability of conventional nanofabrication techniques utilizing electron beam lithography or atomic force microscopy. QPS has been established in several systems which have exploited novel nanoengineering techniques, most notably for MoGe nanowires evaporated onto carbon nanotube templates [7], In-based wires formed on ion-beam patterned glass substrates [8], Al nanowires formed on MBE-defined structures [9], and in Al nanowires formed using conventional EBL techniques and reduced by ion beam sputtering of 1 keV Ar+ ions [10]. Recently, new device architectures utilizing coherent QPS for quantum computation and metrology have been explored theoretically [11]. For these purposes the nanowire must be embedded as a functional component within an extended device architecture, incorporating on-chip resistive components. The possibility of engineering nanowires, on the scale of the superconducting coherence length, using conventional fabrication techniques is therefore of importance for the experimental implementation and application of phase-slip phenomena. This paper reports a study of the low-temperature electronic properties of Nb nanowires fabricated using a focused- ion beam (FIB) to define structures by milling of a thin Nb film with high precision, in a dimension regime where crossover between Josephson and phase-slip behavior is anticipated. As we shall demonstrate, FIB-based fabrication techniques provide access to devices that exhibit phase-slip behavior, aided by Ga-ion poisoning in the periphery of the structure, which reduces the effective electronic dimensions of the superconducting wire significantly below than the physical lithographic limits. The electronic properties observed for a series of nanowires is compared with theoretical models describing thermal and quantum phase-slip processes. 2. Experimental details This paper reports a study of the transport properties of a series of Nb nanowires with width, W , in the range 70 nm to 200 nm, and length, L, in the range 100 nm to 104 nm. The devices are fabricated by ion-beam milling Nb thin films, € of 100 nm thickness and of Tc~8.2K, prepared by dc magnetron sputtering on an underlying SiO2/Si substrate. FIB milling is carried out using a crossed-beam FIB/Scanning Electron Microscope (SEM) operating with 30 keV Ga- ions, and is utilized to define the nanowire and bonding pads in the Nb film within the same fabrication step. The SEM, which facilitates imaging of the nanowires (Figure 1a), has been used to estimate their physical dimensions, width, W, and length, L. A more detailed examination of some of the devices has been performed using High Resolution Transmission Electron Microscopy (HR-TEM) to obtain images of the engineered wires in cross-section. HR-TEM data, shown in Figures 1b and 1c, indicate that the width of the upper part of the wire is considerably smaller than the overall wire width, as determined by SEM imaging. Furthermore, the periphery of the wires is contaminated with Ga, introduced by the ion-beam processing; the TEM images show that in the centre of the upper region of the wire a Nb core free of Ga contamination is present (indicated by the dark region in Figure 1c). The size of this undamaged Ga region is plotted in Figure 1d as a function of the overall wire width determined from SEM imaging. DC current-voltage (I-V) measurements were performed between room temperature and 4.2 K, using a Star Cryoelectronics system with nVHz-1/2 sensitivity. The resistance of the devices was determined from the I-V data at zero-bias-current, so is therefore not influenced by resistive behavior close to the switching current [12], for wires which undergo a transition into a resistive state. For several long wires, with length exceeding 200 nm, this transition was observed to be hysteretic, with the number of steps increasing with increased wire length and reduced width; this behavior will be discussed elsewhere [13]. 3. Experimental results The measured variation in R T( ) of a series of nanowires with length 100 nm, and varying width, is illustrated in Figure 2. At low temperatures a gradual evolution in the electronic properties of the nanowires is observed with € decreasing wire width. Several devices with width exceeding 200 nm, indicated by the open symbols in Figure 2, undergo a superconducting transition at a temperature in the range 7 – 7.3 K, corresponding to the onset of superconductivity in the nanowire. At temperatures below the superconducting transition these devices exhibit Josephson-like behaviour with a critical current that scales appropriately with the device width. A transition at higher temperature, 7.7 K, corresponds to the superconducting transition in the Nb contact pads, which are measured in series with the nanowire [14]. In several narrower devices, indicated by the closed symbols in Figure 2, the superconducting transition is suppressed and a residual resistance is observed at low temperature. In one such device, with width 118 nm (labelled A in Figure 3), the observed R T( ) data is suggestive of a phase-slip mechanism extending to low temperature; the behaviour of this device is compared to thermal and quantum phase-slip theory in Section 5. A € further five narrow devices, with width W ≤ 100 nm , exhibit complete suppression of superconductivity. The two narrowest devices, of widths 70 nm and 80 nm, exhibit a negative € behaviour at low temperature. The side effect of Ga milling is the formation of amorphous damage layers on the walls € of the milled trenches with significant Ga concentration (poisoning). Ga-ion poisoning may significantly influence the dR dT extending to 4 K, suggesting insulating electronic properties of FIB-defined devices, and it is known that the superconducting properties of Nb are suppressed by the implantation of impurity atoms into interstitial sites [15]. The lateral implantation of Ga ions into the nanowires will be considerable and clearly visible on HR-TEM images in silicon and Nb wires with typical amorphous contrast while the core of the Nb wire remains pristine (Fig 1c). Based on SRIM calculations [16] simulating the implantation of 30 keV Ga-ions into an Nb substrate we estimate a lateral ion straggle of about 35 nm, consistent with measurements made by other groups [1]. We would therefore anticipate damage to the Nb crystal structure over length scales significantly exceeding this range from each surface of the wire, as indicated by the HR-TEM analysis, with the existence of an un-contaminated Nb region at the core of the wire. The observed gradual evolution in the electronic properties with nanowire width of these short devices gives confidence that the observed behavior is not a result of granularity in the films [17]. A similar evolution has been observed in devices with a range of length from 100 nm to 104 nm. The temperature dependence of the resistance for all devices has been compared to theoretical models describing thermal and quantum phase-slip processes in one-dimension. For several devices, of length ≥ 500 nm, R T( ) characteristics have been found to be consistent with thermal phase slip behaviour. These are summarised in Figure 3 along with the corresponding data for device A. € 4. Theoretical background Theoretical models have been developed by several authors to explain the temperature dependence arising from thermal and quantum phase slip effects arising in the one-dimensional limit. This section briefly reviews key results, and presents a formulism for modeling the R(T) data in the present experiment; this is implemented in Section 5. Following the approach of Lau [8], the resistance below T c of each nanowire is fitted to an expression of the form ( −1 + ( RQPS + RLAMH )−1 R = RN € )−1 , (1) where RLAMH and RQPS are contributions to the resistance arising from thermal and quantum phase-slip respectively, € and RN is the resistance caused by electronic quasi-particles, which represent a parallel conduction channel and can € € be taken equal to the normal state resistance [4]. Langer, Ambegaokar, McCumber and Halperin (LAMH) [6], derived a model that accounts for phase slip arising due to thermal activation of the order parameter close to T c . They proposed the following formula for R T( ) below T c [6], RLAMH T( ( ) = h 4e 2 ) Ω kBT ( energy barrier, )e−ΔF K B T , where ) is the attempt frequency for activation over the Ω = 3ΔF 4πkBT L ξτGL ( € € € ΔF is energy barrier, and is the GL relaxation time, L is the length of the wire, € ) π 8kB T c − T ( τGL = )−1 2 is ξ = ξ0 1 − T / T c ( € ( ΔF = 0.83 LRq ξ0Rn the GL coherence length. )kBT c (1 − T / T c ) 3 2 , where ( Rq = h 4e 2 ) and The barrier energy € Rn is the normal state resistance [7], so that takes the form RLAMH T( € [ ) = A 1 − T TC € ( ) 3 2 T c T ] 3 2 ( exp −B 1 − T T c ( ) 3 2 T c T ) (2) where A = Rq 8 3 5 2π 2 L ξ0 0.83 L  €  ξ0  Rq Rn 2 1    and B = 0.83 LRq ξ0Rn ( ) . The form of RQPS T( ) has been considered theoretically by several authors, who identify a similar form for an € € exponential dependence on temperature, but with significant variation in the value of the pre-exponential factors [18- € 20]. Following Chang [20], we write a contribution to the nanowire resistance of the form RQPS T( ) = Φ0 I 0 ( )Ce− S  , € € € € € 3 S = H c where ) , ( C = S  L π 2τGLξ dependence as described by the analysis of Lau [7], and Giordano [8]. Simplifying this expression gives € € € € ) and Φ0 = h 2e( 2στGLξ 81 , I 0 = kBT /Φ0 , giving essentially the same temperature RQPS T(   ) = α1 A 1 −      T Tc 7    4 Tc T   exp −α2B 1 −      T Tc   ( 3),  where α1 ≈ 1 and α2 ≈ 0.03 . € € € 5. Discussion The solid curves in Figure 3 represent fits to theory. Theoretical fitting of the R T( ) data of device A was performed with measured values for W and Rn , and utilizing the other variables in Table I as fitting parameters. We note that in € this case the fitted curves yield unrealistic values for the variable parameter € for example in the first column of Table I, which significantly exceeds the expected coherence length for Nb. For the € α2 also differ significantly from the expected values. An excessive ξ0 , of the order of 105 nm as illustrated case of Device A, fitted values given for α1 and value of the coherence length has been deduced from related studies of thermal phase slip in YBaCu3O7-d nanowires € € by Mikheenko [21], suggesting that the apparent large coherence length can be accounted for by noting that the nanowire consists of an effective superconducting filament that is considerably smaller than the physical dimensions of the FIB-patterned device (as it is also shown in Fig. 1c). Following this approach, the theoretical curves in Figure 3 have been produced using the nanowire width, W, as a fitting parameter and setting the coherence length ξ0 to a fixed value. NbGa regions in the periphery of the nanowire, which are expected to have a lower superconducting transition € ξ0 in both the NbGa temperature than Nb, may be proximitized by the Nb nanowire, giving rise to differences of region and the Nb region. However, fluctuations would drive the Nb core into a normal state, so that the NbGa € ξ0 is therefore estimated for the Nb core, with a value of periphery would no longer be superconducting by proximity. order 10 nm [12]. The results of these fits are summarized in Table I. € Curve AI represents a fit of the form given by (1), incorporating contributions to the nanowire resistance from both thermal (2) and quantum phase-slip (3) processes, in addition to the quasiparticle resistance channel. Curve AII, and curves B-D, consider thermal activation of phase-slip and ignore any contribution to the resistance from quantum phase slip. Curves B-D show the corresponding experimental data for several nanowires with length exceeding 100 nm to be consistent with a residual resistance described by the LAMH model. Evidence of thermal phase-slip has been reported in several superconducting nanowire systems [4] and is well understood; the remainder of this letter therefore focuses on a theoretical description of the device A. We note that attempts to fit the R T( ) data for device A at low temperature were not possible if the term RQPS T( for our data close to ) was omitted. The LAMH expression (curve AII) is found to account € T c , with the fitting parameters given in Table I, but is not consistent with the measured resistance € below about 5.4 K. The theoretical model incorporating both thermal and quantum phase-slip processes describes the € data over the full temperature range very well, using a value for T c which is close to that of the Nb film and a nanowire width of 1.8 nm. The evolution of the Nb core width, determined by HR-TEM, with the overall physical € width of the wire, determined by SEM (Figure 1d), indicates that wires of width about 100 nm would possess a Nb core size of just a few nanometres. This is consistent with the value for the Nb wire width obtained from theoretical fitting of the electronic properties. The R(T) data suggests that thermal activation of phase-slip is dominant close to Tc, as anticipated. However, below a temperature of about 5.4 K, the form of R(T) can no longer be described by a thermal activation model and the experimental data is suggestive of a quantum-activated process at low temperature. 6. Summary The low temperature electronic properties of Nb nanowires, fabricated by FIB milling, have been explored in a dimension regime where crossover between Josephson-like to dissipative, phase-slip behavior is evident as the nanowire width is reduced. Wide nanowires (W>200 nm) display Josephson characteristics with a well-defined critical current. Devices with width in the range 100 - 120 nm exhibit behavior consistent with thermal or quantum phase-slip in a central Nb filament of nanoscale dimensions. Nanowires with a physical width less than 100 nm exhibit insulating behavior extending to high temperature, indicating that Ga-ion poisoning leads to complete suppression of superconductivity in this case. The current study demonstrates the unique possibility that FIB processing affords for the fabrication of phase-slip components with nanoscale dimensions embedded in a wire with physical dimensions of order 100 nm. This work was supported by the Australian Research Council. G.C. Tettamanzi acknowledge the Dutch Foundation for Fundamental Research on Matter (FOM) for financial support. C.H.M. acknowledges funding from the EPSRC and the European Commission via project NMP2-CT-2003-505587 “SFINx”. [1] A.P.G. Troeman, H. Derking, B. Borger, J. Pleikies, D. Veldhius and H. Hilgenkamp, Nano Lett. 7, 2152 (2007). [2] G.C. Tettamanzi, C.I. Pakes, S.K.H. Lam and S. Prawer, Superconductor Science and Technology, 22, 064006 (2009). [3] Hao L, Macfarlane J C, Gallop J C, Cox D, Beyer J, Drung D, Schurig T, Appl. Phys. Lett. 92, 192507 (2008). [4] M. Tinkham, Introduction to Superconductivity (McGraw-Hill, New York, 1996). [5] K.Yu. Arutyunov, D.S. Golubev and A.D. Zaikin, Physics Reports 464, 1 (2008). [6] J.S. Langer and V. Ambegaokar, Phys. Rev. 164, 498 (1967); D.E. McCumber and B.I. Halperin, Phys. Rev. B, 1, 1054 (1970). [7] C.N. Lau, N. Markovic, M. Bockrath, A. Bezryadin, M. Tinkham, Phys. Rev. Lett. 87, 217003 (2001); A. Bezryadin, C.N. Lau, M. Tinkham, Nature 404, 971 (2000). [8] N. Giordano, Phys. Rev. Lett. 61, 2137 (1988); N. Giordano and E.R. Schuler, Phys. Rev. Lett. 63, 2417 (1989); N. Giordano, Phys. Rev. B 41, 6350 (1990); N. Giordano, Phys. Rev. B 43, 160 (1991). [9] F. Altomare, A.M. Chang, M.R. Melloch, Y. Hong, C.W. Tu, Phys. Rev. Lett. 97, 017001 (2006). [10] M. Zgirski and K. Y. Arutyunov. Phys. Rev. B, 75, 172509 (2007), M. Zgirski, K.P. Riikonen, V. Touboltsev, and K. Y. Arutyunov, Nano Lett. 5, 1029 (2005). [11] J.E. Mooij and C.J.P.M. Harmans, New J. Phys. 7, 219 (2005); J.E. Mooij and Yu.V. Nazarov, Nature Physics 2, 169 (2006). [12] A. Rogachev, A.T. Bollinger and A. Bezryadin, Phys. Rev. Lett. 94, 017004 (2005). [13] G. C. Tettamanzi et al, in prep. [14] Bezryadin, J. Phys.: Condens. Matter 20, 043202 (2008), A. Potenza et al, Phys. Rev. B 76, 014534 (2007). [15] C. Camerlingo, P. Scardi, C. Tosello and R. Vaglio, Phys. Rev. B 31, 3121 (1985). [16] J.F. Ziegler, J.P. Biersack, and U. Littmark, The Stopping and Range of Ions in Matter (SRIM), Pergamon Press, New York (1985). [17] S. Bose, R. Banerjee, A. Genc, P. Raychaudhuri, H. L. Fraser and P. Ayyub, J. Physics: Condens Matter 18, 4553 (2006). [18] S. Saito and Y. Murayama, Phys. Lett. A 135, 55 (1989); S. Saito and Y. Murayama, Phys. Lett. A 139, 85 (1989). [19] D.S. Golubev and A.D. Zaikin, Phys. Rev. B 64, 14504 (2001). [20] Y. Chang, Phys. Rev. B 54, 9436 (1996). [21] P. Mikheenko , X. Deng, S. Gildert, M.S. Colclough, R.A. Smith, C.M. Muirhead, P.D. Prewett and J. Teng, Phys. Rev. B 72, 174506 (2005).  ξ0 (nm) W (nm) RN (Ω) Tc (K) α1 α2 AI (LAMH + QPS) 2.5 × 105 ± 1.0 × 105 118 15 ± 1 7 ± 1 189 ± 10 55 ± 1 AI (LAMH + QPS) 10 1.8 ± 0.1 13 ± 1 8 ± 1 0.8 ± 0.1 0.34 ± 0.01 AII (LAMH) B (LAMH) C (LAMH) D (LAMH) 10 10 10 10 0.7 ± 0.1 2.9 ± 0.1 14 ± 1 - - 1.0 ± 0.1 2.5 ± 0.1 11.1 ± 0.2 - - 0.9 ± 0.1 2.0 ± 0.1 12 ± 1 - - 1.47 ± 0.01 9.9± 0.1 9 ± 0.1 - - Table I. Theoretical fitting results illustrated in Figure 3. Parameters indicated in italic text were regarded as being fixed in the corresponding fit.  Figure 1: a) Scanning electron microscopy image of a long Nb wire. The vertical trenches parallel to the wire, which are in the ion-beam scan direction, arise presumably due to variation in the ion current during processing. In the inset, a schematic of the general structure of the devices is given. b) HR-TEM cross-section of a series of wires fabricated in the similar fashion to the measured wires. The width for these wires when measured by SEM is in the range 100 nm to 200 nm (left to right). c) Enlarged HR-TEM image for a wire of width 130 nm, determined by SEM. A Nb core (in black) of diameter approximately 20 nm, surrounded by Ga-implanted regions is clearly visible. d) Comparison between the wire and Nb core dimensions, determined from SEM and from HR-TEM images. From this graph, a Nb core of a few nanometres in diameter can be estimated for wires that shows a SEM width of ~ 100 nm.  Figure 2. R T( ) data for several devices with physical length 100 nm. The physical width, determined from SEM images, of each device is indicated. Suppression of superconductivity is evident in the devices represented by closed symbols. €  Figure 3. Theoretical fits of R T( ) for a series of devices exhibiting phase-slip behaviour. AI represents the fit done using both quantum and thermal phase slip theories. AII, B, C and D represent the fits done using only thermal phase slip theory. The € symbols represent the experimental data (Device label, Length (nm), Width (nm); A (open circles): 100, 118; B (open pentagons): 5000, 170; C (open hexagons): 10000, 200; D (open triangles): 500, 150). These data are showing trend similar to the one observed by other groups [14] 

1101.4117

1

1101

Single-layer and bilayer graphene superlattices: collimation, additional Dirac points and Dirac lines

We review the energy spectrum and transport properties of several types of one- dimensional superlattices (SLs) on single-layer and bilayer graphene. In single-layer graphene, for certain SL parameters an electron beam incident on a SL is highly collimated. On the other hand there are extra Dirac points generated for other SL parameters. Using rectangular barriers allows us to find analytic expressions for the location of new Dirac points in the spectrum and for the renormalization of the electron velocities. The influence of these extra Dirac points on the conductivity is investigated. In the limit of {\delta}-function barriers, the transmission T through, conductance G of a finite number of barriers as well as the energy spectra of SLs are periodic functions of the dimensionless strength P of the barriers, P{\delta}(x) ~ V (x). For a Kronig-Penney SL with alternating sign of the height of the barriers the Dirac point becomes a Dirac line for P = {\pi}/2 + n{\pi} with n an integer. In bilayer graphene, with an appropriate bias applied to the barriers and wells, we show that several new types of SLs are produced and two of them are similar to type I and type II semiconductor SLs. Similar as in single-layer graphene extra "Dirac" points are found. Non-ballistic transport is also considered.

cond-mat.mes-hall

cond-mat

Single-layer and bilayer graphene superlattices: collimation, additional Dirac points and Dirac lines By Michael Barbier, Panagiotis Vasilopoulos, and Franc¸ois M. Peeters 1Department of Physics, University of Antwerp, Groenenborgerlaan 171, B-2020 Antwerpen, Belgium 2Department of Physics, Concordia University, 7141 Sherbrooke Ouest, Montr´eal, Quebec, Canada H4B 1R6 We review the energy spectrum and transport properties of several types of one- dimensional superlattices (SLs) on single-layer and bilayer graphene. In single-layer graphene, for certain SL parameters an electron beam incident on a SL is highly collimated. On the other hand there are extra Dirac points generated for other SL parameters. Using rectangular barriers allows us to find analytic expressions for the location of new Dirac points in the spectrum and for the renormalization of the electron velocities. The influence of these extra Dirac points on the conductivity is investigated. In the limit of δ-function barriers, the transmission T through, conductance G of a finite number of barriers as well as the energy spectra of SLs are periodic functions of the dimensionless strength P of the barriers, P δ(x) = V (x)/vF , with vF the Fermi velocity. For a Kronig-Penney SL with alternating sign of the height of the barriers the Dirac point becomes a Dirac line for P = π/2 + nπ with n an integer. In bilayer graphene, with an appropriate bias applied to the barriers and wells, we show that several new types of SLs are produced and two of them are similar to type I and type II semiconductor SLs. Similar as in single-layer graphene extra "Dirac" points are found. Non-ballistic transport is also considered. Keywords: graphene; electron transport; two-dimensional crystals 1. Introduction Since the experimental realisation of graphene (Novoselov et al., 2004) in 2004, this one-atom thick layer of carbon atoms has attracted the attention of the scien- tific world. This attraction pole was created by the prediction that the carriers in graphene behave as massless relativistic fermions moving in two dimensions. The latter particles, which are described by the Dirac-Weyl Hamiltonian, possess inter- esting properties such as a gapless and linear-in-wave vector electronic spectrum, a perfect transmission, at normal incidence, through any potential barrier, i.e., the Klein paradox (Klein, 1929; Katsnelson et al., 2006; Pereira Jr et al., 2010; Roslyak et al., 2010), which was recently addressed experimentally (Young & Kim, 2009; Huard et al., 2007), the zitterbewegung (Schliemann et al., 2005; Winkler et al., 2007; Zawadzki, 2005), etc., see Ref. (Castro Neto et al., 2009) and (Abergel et al., 2010) for recent reviews. On the other hand, in bilayer graphene the carriers exhibit Article submitted to Royal Society TEX Paper 1 1 0 2 n a J 1 2 ] l l a h - s e m . t a m - d n o c [ 1 v 7 1 1 4 . 1 0 1 1 : v i X r a 2 M. Barbier, P. Vasilopoulos, and F. M. Peeters a very different but extraordinary electronic behaviour, such as being chiral (Mc- Cann, 2006; Katsnelson et al., 2006) but with a different pseudospin (=1) than in single-layer graphene (=1/2). Although their spectrum is parabolic in wave vector and also gapless, it is possible to create an energy gap by applying a perpendicular electric field on a bilayer graphene sample (Castro et al., 2007). This allows one to electrostatically create quantum dots in bilayer graphene (Pereira Jr et al., 2007b) and enrich its technological capabilities. In previous work we studied the band structure and other properties of single- layer and bilayer graphene (Barbier et al., 2008, 2009b) in the presence of one- dimensional (1D) periodic potential, i.e., a superlattice (SL). SLs are known to be useful in altering the band structure of materials and thereby broadening their technological applicability. The already peculiar, cone-shaped band structure of single-layer graphene can be drastically changed in a SL. An interesting feature is that for certain SL parameters the carriers are restricted to move along one direction, i.e. they are collimated (Park et al., 2009a). Furthermore, it was found that for other parameters of the SL instead of the single-valley ( the K or K(cid:48)-point) Dirac cone, "extra Dirac points" appeared at the Fermi level in addition to the original one (Ho et al., 2009). The latter extra Dirac points are interesting because of their accompagning zero modes (Brey & Fertig, 2009) and their influence on many physical properties such as the density of states (Ho et al., 2009), the conductivity (Barbier et al., 2010; Wang & Zhu, 2010), and the Landau levels upon applying a magnetic field (Park et al., 2009b; Sun et al., 2010). One can also obtain "extra Dirac points" in bilayer graphene SLs. The possibility of locally altering the gap (Castro et al., 2007) of bilayer graphene by applying a bias is another way of tuning the band structure. In this review we classify these SLs in four types. Another interesting result of applying a bias locally is that sign flips of the bias introduce bound states along the interfaces (Martin et al., 2008; Martinez et al., 2009). These bound states break the time reversal symmetry and are distinct for the two K and K(cid:48) valleys; this opens up perspectives for valley-filter devices (San-Jose et al., 2009). In this review we will use the following methods to describe our findings. For both single-layer and bilayer graphene we will use the nearest neighbour, tight- binding Hamiltonian in the continuum approximation, and restrict ourselves to the electronic structure in the neighbourhood of the K point. We then apply the transfer-matrix method to study the spectrum of and transmission through various potential barrier structures, which we approximate by piecewise constant potentials. We consider structures with a finite number of barriers and SLs. We will study ballistic transport in systems with a finite number of barriers using the two-probe Landauer conductance while in a SL (infinite number of barriers) we will evaluate the spectrum and the diffusive conductivity, i.e., we will study non- ballistic transport. The work is organized as follows. In Sec. 2 we investigate various aspects of ballistic transport through a finite number of barriers on single-layer graphene as well as the spectrum of SLs, with emphasis on collimation and extra Dirac points and their influence on non-ballistic transport. In Sec. 3 we carry on the same stud- ies, whenever possible, on bilayer graphene. In addition, we consider various types Article submitted to Royal Society Review: Single-layer and bilayer graphene SLs 3 of band alignments in the presence of a bias that can lead to different types of heterostructures and SLs. We make a summary and concluding remarks in Sec. 4. 2. Single-layer graphene We describe the electronic structure of an infinitely large, flat graphene flake by the nearest-neighbour tight-binding model and consider wave vectors close to the K point. The relevant Hamiltonian in the continuum approximation is H = vF σ · F σz, with p the momentum operator, V the potential, 1 the 2× 2 unit p + V 1 + mv2 matrix, σ = (σxσy), σz the Pauli-matrices and vF ≈ 106m/s the Fermi velocity. Explicitly H is given by H = V + mv2 F −ivF (∂x + i∂y) −ivF (∂x − i∂y) V − mv2 F . (2.1) (cid:18) (cid:19) The mass term is in principle zero in the nearest-neighbour, tight-binding model but due to interaction with a substrate (Giovannetti et al., 2007) an effective mass term can be induced and results in the opening of an energy gap. Recently there have been proposals to induce an energy gap in single-layer graphene, and it is appropriate that we consider this mass term where relevant. In the presence of a 1D rectangular potential V (x), such as the one shown in Fig. 1, the equation (H − E)ψ = 0 admits (right- and left-travelling) plane wave solutions of the form ψl,r(x)eikyy with e−iλx, (2.2) ψr(x) = eiλx, ψl(x) = λ + iky here λ = [(ε − u(x))2 − k2 y − µ2]1/2 is the x component of the wave vector, ε = EL/vF , u(x) = V (x)L/vF , and µ = mvF L/. The dimensionless parameters ε, u(x) and µ scale with the characteristic length L of the potential barrier structure. For the single or double barrier system this L will be equal to the barrier width while for a SL it will be its period. Neglecting the mass term one rewrites Eq. (2.2) in the simpler form (cid:18) ε + µ (cid:19) (cid:19) (cid:18) ε + µ −λ + iky (cid:19) (cid:18) 1 −se−iφ ψr(x) = eiλx, ψl(x) = seiφ y]1/2, tan φ = ky/λ, and s = sgn(ε − u(x)). e−iλx, with λ = [(ε − u(x))2 − k2 (2.3) (cid:18) 1 (cid:19) (a) A single or double barrier The model barriers and wells we consider are shown in Fig. 1(a). It is interesting to look at the tunneling through such barriers, which was previously studied by (Katsnelson et al., 2006) for a single barrier. This was later extended to massive electrons with spatially varying mass (Gomes & Peres, 2008). Transmission. To find the transmission T through a square-barrier structure one first observes that the wave function in the jth region ψj(x) of the constant potential Vj is given by a superposition of the eigenstates given by Eq. (2.2), ψj(x) = Ajψr j + Bjψlj. (2.4) Article submitted to Royal Society 4 M. Barbier, P. Vasilopoulos, and F. M. Peeters Figure 1. (a) A 1D potential barrier of height Vb and width Wb. (b) A single unit of a potential well next to a potential barrier. The wave function should be continuous at the interfaces. This boundary condition gives the transfer matrix Nj relating the coefficients Aj and Bj of region j with those of the region j + 1 in the manner . (2.5) (cid:18)Aj (cid:19) (cid:18)A0 Bj (cid:19) B0 = Nj+1 n(cid:89) j=1 (cid:19) (cid:18)Aj+1 (cid:18)An (cid:19) Bj+1 Bn By employing the transfer matrix at each potential step we obtain, after n steps, the relation = Nj . (2.6) In the region to the left of the barrier we assume A0 = 1 and denote by B0 = r the reflection amplitude. Likewise, to the right of the nth barrier we have Bn = 0 and denote by An = t the transmission amplitude. The transmission probability T can be expressed as the ratio of the transmitted current density jx over the incident one, where jx = vF ψ†σxψ. This results in T = (λ(cid:48)/λ)t2, with λ(cid:48)/λ the ratio between the wave vector λ(cid:48) to the right and λ to the left of the barrier. If the potential to the right and left of the barrier is the same we have λ(cid:48) = λ. For a single barrier the transmission amplitude is given by T = t2 = N11−1, with Nij the elements of the transfer matrix N . Explicitly, t can be written as 1/t = cos(λbWb) − iQ sin(λbWb), Q = (ε0εb − k2 y − µ0µb)/λ0λb; (2.7) the indices 0 and b refer, respectively, to the region outside and inside the barrier and εb = ε − u. A contour plot of the transmission is shown in Fig. 2(a). We clearly see: 1) T = 1 for φ = 0 which is the well-known Klein tunneling, and 2) strong resonances, in particular for E < 0, when λbWb = nπ, which describe hole scattering above a potential well. In the limit of a very thin and high barrier, one can model it by a δ-function barrier V (x)/vF = P δ(x). Using Eq. (2.7) for t gives (Barbier et al., 2009a) T = 1/[1 + sin2 P tan2 φ], (2.8) with tan φ = ky/λ0 the angle of incidence. Notice that this transmission is inde- pendent of the energy and is a periodic function of P . The latter is very different from the non-relativistic case where T is a decreasing function of P. A contour plot of the transmission is shown in Fig. 2(b) and T = 1 for φ ≈ 0 which is nothing else than Klein tunneling. Notice also the symmetry T (π − P ) = T (P ). Article submitted to Royal Society (a)(b)V(x)xWWbwVVbwWbVbReview: Single-layer and bilayer graphene SLs 5 Figure 2. (a) Contour plot of the transmission through a single barrier with µ = 0, Wb = L, and ub = 10. (b) As in (a) for a single δ-function barrier with µ = 0 and u(x) = P δ(x); the transmission is independent of the energy. (c) As in (a) for two barriers with µ = 0, ub = 10, uw = 0, Wb = 0.5L, and Ww = L. (d) Spectrum of the bound states vs ky for a single (L = 1, solid red line), two parallel (dashed blue curves), and two anti-parallel (green dash-dotted curves) δ-function barriers (L is the inter-barrier distance). For two barriers the system becomes a resonant structure, for which it was found that the resonances in the transmission depend mostly on the width Ww of the well between the barriers (Pereira Jr et al., 2007a). A plot of the transmission is shown in Fig. 2(c). In the limit of two parallel δ-function barriers of equal strength P we obtain the transmission T =(cid:2)1 + tan2 φ(cos λ0 sin 2P − 2s sin λ0 sin2 P/ cos φ)2(cid:3)−1 T =(cid:2) cos2 λ0 + sin2 λ0(1 − sin2 φ cos 2P )2/ cos4 φ(cid:3)−1 . The case of two anti-parallel δ-function barriers of equal strength is also interesting. The relevant transmission is . (2.9) (2.10) Conductance. The two-terminal conductance is given by (cid:90) π/2 −π/2 G(EF ) = G0 T (EF , φ) cos φ dφ, (2.11) with G0 = 2EF Lye2/(vF h2) for single-layer graphene, and Ly the width of the system. For a single and double barrier, the transmission through which is plotted Article submitted to Royal Society 6 M. Barbier, P. Vasilopoulos, and F. M. Peeters in Fig. 2(a) and 2(c), the conductance G is shown in Fig. 3(b) and exhibits multiple resonances despite the integration over the angle φ. Taking the limit of a δ-function barrier leads to G periodic in P and given by G/G0 = 2(cid:2)1 − artanh(cos P ) sin P tan P(cid:3)/ cos2 P. (2.12) For one period G is shown in Fig. 3(a). Figure 3. (a) Conductance G vs strength P of a δ-function barrier in single-layer graphene; the conductance is independent of the energy. (b) Conductance G vs energy for the single (solid blue curve) and double (dashed green curve) square barrier of Fig. 2(a) and 2(c). y + µ2 Bound states. For k2 0 > ε2 the wave function outside the barrier (well) becomes an exponentially decaying function of x, ψ(x) ∝ exp{±kxx} with kx = 0−ε2]1/2. Localized states form near the barrier boundaries (Pereira Jr et al., y +µ2 [k2 2006); however, they are propagating freely along the y-direction. The spectrum of these bound states can be found by setting the determinant of the transfer matrix equal to zero. For a single potential barrier (well) it is given by the solution of the transcendental equation λ0λb cos(λbWb) + (k2 y + µ0µb − ε(ε − u)) sin(λbWb) = 0. (2.13) In Fig. 4(b) these bound states are shown, as a function of ky, by the dashed blue (red) curves. An interesting structure to study is that of a potential barrier next to a well but with average potential equal to zero, considered by (Arovas et al., 2010). This is the unit cell (shown in Fig. 1(b)) of the SL we will use in Sec. 3 where extra Dirac points will be found. In Fig. 4(a) the Dirac cone outside the barrier is shown as a grey area, inside this region there are no bound states. Superimposed are grey lines corresponding to the edges of the Dirac cones inside the well and barrier that divide the (E, ky) plane into four regions. Region I corresponds to propagating states inside both the barrier and well while region II (III) corresponds to propagating states only inside the well (barrier). In region IV no propagating modes are possible, neither in the barrier nor in the well. For high thin barriers, region I will become a thin area adjacent to the upper cone, converging to the dark green line in the limit of a δ-function barrier. Figure 4(b) shows that the bound states of this structure are composed of the ones of a single barrier and those of a single well. Anticrossings take place where the bands otherwise would cross. The resulting spectrum is clearly a starter of the spectrum of a SL shown in Fig. 4(d). Article submitted to Royal Society (a)Review: Single-layer and bilayer graphene SLs 7 Figure 4. (a) Four different regions for a single unit of Fig. 1(b) with ub = 24, uw = 16, Wb = 0.4 and Ww = 0.6. The green line corresponds to region I in the limit of a δ-function barrier. (b) Bound states for a single barrier (dashed blue curves) and well (dashed red curves) and the combined barrier-well unit (black curves). (c) Contour plot of the trans- mission through a unit with µ = 2, ub = −uw = 20 and Wb = Ww = 0.5; the red curves show the bound states. (d) Spectrum of a SL whose unit cell is shown in Fig. 1(b), for kx = 0 (blue curves) and kxL = π/2 (red curves). In the limit of δ-function barriers and wells the expressions for the dispersion relation are strongly simplified by setting µ = 0 in all regions. For a single δ-function barrier the bound state is given by ε = sgn(sin P )ky cos P, (2.14) which is a straight line with a reduced group velocity vy; the result is shown in Fig. 2(d) by the red curve. Comparing with the single-barrier case we notice that due to the periodicity in P , the δ-function barrier can act as a barrier or as a well depending on the value of P . For two δ-function barriers there are two important cases: the parallel and the anti-parallel case. For parallel barriers one finds an implicit equation for the energy ky sin P, where λ(cid:48) = λ0, while for anti-parallel barriers one obtains λ(cid:48) cos P + ε sin P = e−λ(cid:48) y sin2 P = λ(cid:48)2/(1 − e−2λ(cid:48) k2 ). (2.15) (2.16) Article submitted to Royal Society 01020-10010EL / hvFkyLIIIIIIIV(a)01020-10010EL / hvFkyL(b)(d)8 M. Barbier, P. Vasilopoulos, and F. M. Peeters For two (anti-)parallel δ-function barriers we have, for each fixed ky and P , two energy values ±ε, and therefore two bound states. In both cases, for P = nπ the spectrum is simplified to the one in the absence of any potential ε = ±ky. In Fig. 2(d) the bound states for double (anti-)parallel δ-function barriers are shown, as a function of kyL, by the dashed blue (dash-dotted green) curves. For anti- parallel barriers we see that there is a symmetry around E = 0, which is absent when the barriers are parallel. (b) Superlattice Now we will consider the system of a superlattice with a corresponding 1D periodic potential, with square barriers, given by V (x) = V0 [Θ(x − jL) − Θ(x − jL − Wb)]. (2.17) ∞(cid:88) j=−∞ with Θ(x) the step function. The corresponding wave function is a Bloch function and satisfies the periodicity condition ψ(L) = ψ(0) exp(ikx), with kx now the Bloch phase. Using this relation together with the transfer matrix for a single unit ψ(L) = Mψ(0) leads to the condition det[M − exp(ikx)] = 0. This gives the transcendental equation cos kx = cos λwWw cos λbWb − Q sin λwWw sin λbWb, (2.18) (2.19) from which we obtain the energy spectrum of the system. In Eq. (2.19) we used the following notation: εw = ε + uWb, y − µ2 w]1/2, w − k2 λb = [ε2 λw = [ε2 εb = ε − uWw, y − µ2 b − k2 u = V0L/vF , Wb,w → Wb,w/L, y − µbµw)/λwλb. Numerical results for the dispersion relation E(ky) are shown in Fig. 4(d). We see the appearance of bands (green areas) which for large ky values collapse into the bound states (where the red and blue curves meet) while the charge carriers move freely along the y direction. b]1/2, Q = (εwεb − k2 (c) Collimation and extra Dirac points As shown by various studies, carriers in graphene SLs exhibit several interesting pecularities that result from the particular electronic SL band structure. In a 1D SL it was found that the spectrum can be altered anisotropically (Park et al., 2008a; Bliokh et al., 2009). Moreover, this anisotropy can be made very large such that for a broad region in k space the spectrum is dispersionless in one direction, and thus electrons are collimated along the other direction (Park et al., 2009a). Even more intriguing was the ability to split off "extra Dirac points" (Ho et al., 2009) with accompanying zero modes (Brey & Fertig, 2009) which move away from the K point along the ky direction with increasing potential strength. Here we will describe these phenomena for a SL of square potential barriers. Article submitted to Royal Society Review: Single-layer and bilayer graphene SLs 9 We start by describing the collimation as done by (Park et al., 2009a); sub- sequently we will find the conditions on the parameters of the SL for which a collimation appears. It turns out that they are the same as those needed to create a pair of extra Dirac points. Figure 5. The lowest conduction band of the spectrum of graphene near the K point in the absence of SL potential (a), (b) and in its presence (c), (d) with u = 4π. (a) and (c) are contour plots of the conduction band with a contour step of 0.5 vF /L. (b) and (d) show slices along constant kyL = 0, 0.2, 0.4π. Following (Park et al., 2009a) the condition for collimation to occur is(cid:82) 0, where the function α(x) = 2(cid:82) x ε ≈ ±(cid:2)k2 0 u(x(cid:48))dx(cid:48) embodies the influence of the potential, s = sign(ε) and s = sign(kx). For a symmetric rectangular lattice this corresponds to u/4 = nπ. The spectrum for the lowest energy bands is then given by (Park et al., 2008b) (cid:3)1/2 y BZ eissα(x) = x + fl2k2 with fl being the coefficients of the Fourier expansion eiα(x) =(cid:80)∞ (2.20) l=−∞ flei2πlx/L. The coefficients fl depend on the potential profile V (x), with fl < 1. For a sym- metric SL of square barriers we have fl = u sin(lπ/2 − u/2)/(l2u2 − u2/4). The inequality fl < 1 implies a group velocity in the y direction vy < vF which can be seen from Eq. (2.20). + πl/L In Fig. 5(b),(d) we show the dispersion relation E vs kx for u = 4π at constant ky. As can be seen, when a SL is present in most of the Brillouin zone the spectrum, partially shown in (c), is nearly independent of ky. That is, we have collimation Article submitted to Royal Society 10 M. Barbier, P. Vasilopoulos, and F. M. Peeters of an electron beam along the SL axis. The condition u = V0L/vF = 4nπ shows that altering the period of the SL or the potential height of the barriers is sufficient to produce collimation. This makes a SL a versatile tool for tuning the spectrum. Comparing with Figs. 5(a), (b) we see that the cone-shaped spectrum for u = 0, is transformed into a wedge-shaped spectrum (Park et al., 2009a). We will compare this result now with an other approximate result for the spec- trum, where we suppose ε small instead of ky small. We start with the transcenden- tal equation (2.19). As we are interested in an analytical approximate expression for the spectrum, we choose to expand the dispersion relation around ε = 0 up to second order in ε. The resulting spectrum is (cid:34) 4a22(cid:2)k2 y sin2(a/2) + a2 sin2(kx/2)(cid:3) ya sin a + a2u4/16 − 2k2 k4 yu2 sin2(a/2) (cid:35)1/2 ε± = ± , (2.21) with a = [u2/4 − k2 et al., 2009a), we expand Eq. (2.19) for small k and ε; this leads to y]1/2. In order to compare this spectrum with that by (Park ε ≈ ±(cid:2)k2 y sin2(u/4)/(u/4)2(cid:3)1/2 x + k2 . (2.22) This spectrum has the form of an anisotropic cone and corresponds to that of Eq. (2.20) for l = 0 (higher l correspond to higher energy bands). In Fig. 6(a), (b) we see that the cone-shaped spectrum in (a), for u = 0, is transformed into a anisotropic spectrum in (b), for u = 4.5π, having peculiar extra Dirac points. These extra Dirac points cannot be described by a spectrum having an anisotropic cone-shape, therefore we compare the two approximate spectra. In Fig. 6(c), (d) we show how Eq. (2.21)) and Eq. (2.22) differ from the "exact" numerically obtained spectrum. From this figure one can see that Eq. (2.21) describes the lowest bands rather well for ε < 1, while Eq. (2.22) is sufficient to describe the spectrum near the Dirac point. The former equation will be usefull when describing the spectrum near the extra Dirac points and we will use it to obtain the velocity. We now move on to another important feature of the spectrum, the extra Dirac points first obtained by (Ho et al., 2009) using tight-binding calculations. These extra Dirac points are found as the zero-energy solutions of the dispersion relation in Eq. (2.19) for zero energy (Barbier et al., 2010). In order to find the location of the Dirac points we assume kx = 0, ε = 0, µb = µw = 0, and consider the special case of Wb = Ww = 1/2 in Eq. (2.19). The resulting equation 1 = cos2 λ/2 +(cid:2)(u2/4 + k2 (cid:114) y = u2/4 + k2 has solutions for u2/4− k2 of ky = 0 (at the Dirac points) as y)(cid:3) sin2 λ/2, y)/(u2/4 − k2 (2.23) y or sin2 λ/2 = 0. This determines the values ky,j± = ± − 4j2π2; u2 4 (2.24) the extra Dirac points are for j (cid:54)= 0. For a SL spectrum symmetric around zero energy, the extra Dirac points are at ε = 0. We expect from the considerations of Sec. 2(b) (and Fig. 4(b)) that for unequal barrier and well widths this will no longer Article submitted to Royal Society Review: Single-layer and bilayer graphene SLs 11 be true. Indeed, in such a case the extra Dirac points shift in energy, as seen in Fig. 4(d), and their position in the spectrum is given, for kx = 0, by (Barbier et al., 2010) u 2 εj,m = ky j,m = ±(cid:104) π2 2u (1 − 2Wb) + (εj,m + uWb)2 − (jπ/Ww)2(cid:105)1/2 − (j + 2m)2 W 2 b , (2.25) (cid:18) j2 W 2 w (cid:19) , where j and m are integers, and m (cid:54)= 0 corresponds to higher and lower crossing points. Also, perturbing the potential with an asymmetric term, as done by (Park et al., 2009b), leads to qualitatively similar results. Figure 6. The spectrum of graphene near the K point in the absence of a SL (a) and in its presence (b) with u = 4.5π. (c) and (d) The SL spectrum with u = 10π, the lowest conduction bands are coloured in cyan, red, and green for, respectively, the exact, and the approximations given by (c) Eq. (2.21) and (d) Eq. (2.22), respectively. The approximate spectra are delimited by the dashed curves. An investigation of the group velocity near the (extra) Dirac points is appropri- ate for understanding the transport of carriers in the energy bands close to zero en- ergy. Near the extra Dirac points the group velocity tends to renormalise differently as compared to the original Dirac point. Near them v is oriented along the y direc- tion, while near the latter one v is oriented along the x direction (Ho et al., 2009). The group velocity near the extra Dirac points can be calculated from Eq. (2.21). At the jth extra Dirac point the magnitude of the velocity v/vF = (∂ε/∂kx, ∂ε/∂ky) is given by vx/vF = 16π2j2 cos(kx/2)/u2 vy/vF = (u2/4 − 4j2π2)/u2, (2.26) Article submitted to Royal Society -6-4-202460123EL / hvFkyL / π(c)-6-4-202460123EL / hvFkyL / π(d)12 M. Barbier, P. Vasilopoulos, and F. M. Peeters while at the main Dirac point it is given by vx/vF = 1 and vy/vF = 4 sin(u/4)/u. The dependence of the velocity components on the strength of the potential barriers is shown in Fig. 7. From this figure we observe that new extra Dirac points emerge upon increasing u = V0L/vF (consistent with Eq. (2.24)) and vx decreases while vy increases. The Dirac point itself, however, shows a different behaviour upon increasing u, namely vx = vF constant, and vy is here a globally decaying function showing vy = 0 for periodic values of u, u = 4nπ, with n a nonzero positive integer. Figure 7. The group velocity components vy and vx at the Dirac point j = 0 (shown, respectively, by the solid blue and the dot-dot-dashed red curve), and at the extra Dirac points j = 1, 2, 3 (shown, respectively, by the dot-dashed blue and the dashed red curves) as a function of the barrier parameter u = V0L/vF . Conductivity. We now turn to the transport properties of a SL and look at the influence of these extra Dirac points on the conductivity. The diffusive dc con- ductivity σµν for the SL system can be readily calculated from the spectrum if we assume a nearly constant relaxation time τ (EF ) ≡ τF . It is given by (Charbonneau et al., 1982) (cid:88) n,k σµν(EF ) = e2βτF A vnµvnνfnk(1 − fnk), (2.27) with A the area of the system, n the energy band index, µ, ν = x, y, and fnk = 1/[exp(β(EF − Enk)) + 1] the equilibrium Fermi-Dirac distribution function; β = 1/kBT and the temperature enters the results through the dimensionless value for β which is β = vF /kBT L = 20. For comparison we first look at the conductivity tensor at zero temperature and in the absence of a SL. For single-layer graphene the conductivity is given by σµµ(εF )/σ0 = εF /4π (2.28) with σ0 = e2/, In Figs. 8(a), (b) the conductivities σxx and σyy are shown for a SL as functions of the energy. Notice that for small energies the slope of the conductivity σyy is tunable to a large extent by altering the parameter u of the SL. The dashed blue curves correspond to u = 4π and the rather flat dispersion in the y direction for Article submitted to Royal Society Review: Single-layer and bilayer graphene SLs 13 the lowest conduction band (see Fig. 5(c,d)) translates to a small σyy (for energies EL/vF < 1) compared to the conductivity in the absence of a SL. The solid red curves on the other hand correspond to u = 6π and due to the extra Dirac points, which have a rather flat dispersion in the x direction (Ho et al., 2009), the conductivity σyy is large. Figure 8. (Color online) Conductivities, σxx in (a) and σyy in (b), vs Fermi energy for a SL on single-layer graphene with u = 4π and 6π for, respectively, the blue dashed and red solid curves. In both cases Wb = Ww = 0.5. The dash-dotted black curves show the conductivities in the absence of the SL potential, σxx = σyy = εF σ0/4π. (d ) Dirac lines Figure 9. (a) Schematics of Kronig-Penney SL on single-layer graphene. (b) Extended Kronig-Penney SL. In an effort to simplify the expressions for the dispersion relation we replace, as we did for the few-barrier structures, the SL barriers by δ-function barriers. The square SL potential is then approximated by V (x) = P δ(x − jL). (2.29) ∞(cid:88) j=−∞ This potential leads to the dispersion relation cos kx = cos λ cos P + (ε/λ) sin λ sin P, (2.30) which is periodic in P . This is in sharp contrast with that for standard electrons which is not periodic in P and which in our notation reads cos kx = cos λ(cid:48) + (µP/λ(cid:48)) sin λ(cid:48), (2.31) Article submitted to Royal Society P-P(a)(b)PV(x)x14 M. Barbier, P. Vasilopoulos, and F. M. Peeters where µ = mvF L/ and λ(cid:48) = [2µε − k2 y]1/2. As can be seen from Fig. 10(a), the energy band near the Dirac point has the interesting property that it becomes nearly flat in kx, forming a plane, for large ky. The angle which the asymptotic plane makes with the zero-energy plane depends on P and the group velocity vy corresponding to this asymptotic plane varies from −vF to vF in each period nπ < P < (n + 1)π. Notice that no extra Dirac points are found and the reason is the same as that for the asymmetric SL potential, i.e., the extra Dirac points shift away from zero energy. Alternatively, we can try to shed some light by comparing with Sec. 2(b), where it is explained that the bound states for a single unit of the SL potential are similar to those of the combined single barrier and well. In the region where the bound states cross (denoted by I in Fig. 4(a)) anti-crossings occur and corresponding crossings in the SL spectrum (extra Dirac points) are expected. In the limit of a δ-function barrier this region is reduced to a line (the dark green line in Fig. 4(a)). This prevents anti-crossings from occurring and in this way no extra Dirac points are expected. Figure 10. (a) Spectrum for a Kronig-Penney SL with P = 0.4π. The blue and red curves show, respectively, the kx = 0 and kx = π/L results which delimit the energy bands (green coloured regions). (b) Spectrum for an extended Kronig-Penney SL with P = π/2. Notice that the Dirac point has become a Dirac line. Extended Kronig-Penney model. To re-establish the symmetry between electrons and holes, as in the case of square barriers with Wb = Ww, we can use alternating- in-sign δ-function barriers. The unit cell of the periodic potential contains one such barrier up, at x = 0, followed by a barrier down, at x = L/2, see Fig. 9(b). The potential is given by V (x) = P [δ(x − jL) − δ(x − jL − L/2)], (2.32) ∞(cid:88) j=−∞ and is the asymptotic limit of the potential shown in Fig. 1(b). The resulting transfer matrix leads to the dispersion relation cos kx = cos λ − (2k2 y/λ2) sin2(λ/2) sin2 P. (2.33) This dispersion relation is periodic in P . As shown in Fig. 10(b) no extra Dirac points occur, but for the particular case of P = (n + 1/2)π, n an integer, the Article submitted to Royal Society (a)-101-101-101EL / πhvFkxL / πkyL / π(b)Review: Single-layer and bilayer graphene SLs 15 spectrum shows an interesting feature: for all ky we see that Eq. (2.33) has a solution with ε = kx = 0, which means the Dirac point at kx = ky = 0 turned into a Dirac line along the ky axis. If we take ky not too large (of the order of kx), this spectrum has a wedge structure as was also found for rectangular SLs. For ky → ∞, though, the spectrum becomes a horizontal plane situated at ε = 0. We can generalize this model by taking the distance W between the two barriers of the unit cell not equal to L/2. This was done by (Ramizani Masir et al., 2010, unpublished work). They found an approximate analytic expression for the dispersion given by ε ≈ [k2 x + F k2 y]1/2, F = W 2 + (L − W )2 + 2W (L − W ) cos(2P ). (2.34) This dispersion has the shape of an anisotropic cone with a renormalized velocity in the y direction. Comparing with Eqs. (2.20) and (2.22), we observe that the condition for collimation and the velocity renormalization in the y direction is quite different for square barriers. For instance, in the extended KP model, with W = L/2, we find vy/vF = cos P while for square barriers the result is vy/vF = sin(u/4)/(u/4). The latter means that if we consider P ≡ u/4, the velocity in the y direction is maximum vy = vF for P = (1/2 + n)π in the extended KP model while for square barriers vy = 0 at these points. 3. Bilayer graphene We now turn to bilayer graphene and use again the nearest-neighbour, tight-binding Hamiltonian in the continuum approximation with k close to the K point. If we include a potential difference between the two layers, the Hamiltonian is given by  U1 vF π vF π† U1 t⊥ 0 0 0 H =  . t⊥ 0 U2 vF π 0 0 vF π† U2  e±iλx+ikyy,  1 f± h± g±h± Ψ±(x) = Article submitted to Royal Society Here U1 and U2 are the potentials on layers 1 and 2, respectively, 2∆ = U1 − U2 is the potential difference, and t⊥ describes the coupling between the layers. The energy spectrum for free electrons is given by (McCann, 2006; Barbier et al., 2009b) ε = u0 ±(cid:104) ε = u0 ±(cid:104) ∆2 + k2 + ∆2 + k2 + + (4∆2k2 + k2t2⊥ + − (4∆2k2 + k2t2⊥ + t2⊥ 2 t2⊥ 2 )1/2(cid:105)1/2 )1/2(cid:105)1/2 t2⊥ 4 t2⊥ 4 , , (3.2) with u1 = u0 + ∆ and u2 = u0 − ∆. Contrary to Sec. 3 we use units in inverse distance, namely, ε = E/vF , uj = Uj/vF , and k = [λ2 + k2 y]1/2. This spectrum exhibits an energy gap that for 2∆ (cid:28) t⊥ equals the difference 2∆ between the conduction and valence band at the K point (McCann, 2006). Solutions for this Hamiltonian are four-vectors ψ and for 1D potentials we can write ψ(x, y) = ψ(x) exp(ikyy). If the potentials U1 and U2 do not vary in space, these solutions are of the form (3.1) (3.3) 16 M. Barbier, P. Vasilopoulos, and F. M. Peeters with f± = [−iky ± λ]/[ε(cid:48) − δ], h± = [(ε(cid:48) − δ)2 − k2 [iky ± λ]/[ε(cid:48) + δ]; the wave vector λ is given by y − λ2]/[t⊥(ε(cid:48) − δ)], and g± = λ± = ε(cid:48)2 + δ2 − ky 4ε(cid:48)2δ2 + t2⊥(ε(cid:48)2 − δ2) . (3.4) (cid:20) 2 ±(cid:113) (cid:21)1/2 We will write λ+ = α and λ− = β. (a) Tuning of the band offsets It was shown before that using a 1D biasing, indicated in Figs. 11(a,b,c) by 2∆, one can create three types of heterostructures in graphene (Dragoman et al., 2010). A fourth type, where the energy gap is spatially kept constant but the bias periodically changes sign along the interfaces, can be introduced (see Fig. 11(d)). We characterize these heterostructures as follows: 1) Type I: The gate bias applied in the barrier regions is larger than in the well regions. 2) Type II: The gaps, not necessarily equal, are shifted in energy but they have an overlap as shown. 3) Type III: The gaps, not necessarily equal, are shifted in energy and have no overlap. 4) Type IV: The bias changes sign between successive barriers and wells but its magnitude remains constant. Type IV structures have been shown to localize the wave function at the inter- faces (Martin et al., 2008; Martinez et al., 2009). To understand the influence of such interfaces in this section we will separately investigate structures with such a single interface embedded by an anti-symmetric potential. Figure 11. Four different types of band alignments in bilayer graphene. Ec,b, Ec,w, Ev,c, and Ev,b denote the energies of the conduction (c) and valence (v) bands in the barrier (b) and well (w) regions. The corresponding gap is, respectively, 2∆b and 2∆w. Article submitted to Royal Society Ec,bEc,wEv,wEv,b2D2DwbType IEc,bEc,wEv,wEv,b2D2DwbType IIEc,bEc,wEv,wEv,b2D2DwbType III2D2DwbType IVEcEvReview: Single-layer and bilayer graphene SLs 17 To describe the transmission and bound states of some simple structures we notice that in the energy region of interest, i.e., for E < t⊥, the eigenstates which are propagating are the ones with λ = α. Accordingly, from now on we will assume that β is complex. In this way we can simply use the transfer-matrix approach of Sec. 2 in the transmission calculations. This leads to the relation  = N  t 0 ed 0  .  1 r 0 eg (3.5) Again the transmission is given by T = t2. For a single barrier the transmission in bilayer graphene is given by a com- plicated expression. Therefore, we will first look at a few limiting cases. First we assume a zero bias ∆ = 0 that corresponds to a particular case of type III het- erostructures. In this case we slightly change the definition of the wave vectors: for ∆ = 0 we assume α(β) = [ε2 + (−)εt⊥ − k2 y]1/2. If we restrict the motion along the x axis, by taking ky = 0, and assume a bias ∆ = 0, then the transmission T = t2 is given via 1/t = eiα0D[cos(αbD) − iQ sin(αbD)], (cid:18) αbε0 α0εb Q = 1 2 + α0εb αbε0 (cid:19) . (3.6) This expression depends only on the propagating wave vector α (β for E < 0) as propagating and localized states are decoupled in this approximation. This also means that one does not find any resonances in the transmission for energies in the barrier region, i.e., for 0 < ε < u. Due to the coupling for nonzero ky with the localized states, resonances in the transmission will occur (see Fig. 12). We can easily generalize this expression to account for the double barrier case under the same assumptions. With an inter-barrier distance Ww one obtains the transmission (Barbier et al., 2009b) Td = td2 from ei2α0(Ww+2Wb)t2ei2φt 1 − r2ei2φr ei2α0Ww , td = (3.7) with r = reiφr , and t = teiφt, corresponding to the single barrier transmission and reflection amplitudes. In this case we do have resonances due to the well states; they occur for ei2φr ei2α0Ww = 1. As φr is independent of Ww, one obtains more resonances by increasing Ww. For a single δ-function barrier with potential V (x)/vF = P δ(x) under zero bias, we find the transmission amplitude 1/t = cos P + iµ sin P + (α − β)2k2 y sin P 4αβε2 cot P + iν , (3.8) where µ = (ε + 1/2)/α and ν = (ε − 1/2)/β. Notice that this formula is periodic in the strength of the barrier P as in the single-layer case. For the general case we obtained numerical results for the transmission through various types of single and double barrier structures; they are shown in Fig. 13. Article submitted to Royal Society 18 M. Barbier, P. Vasilopoulos, and F. M. Peeters Figure 12. (a) Contour plot of the transmission for the potential of Fig. 1(b) in bilayer graphene with Wb = Ww = 40 nm, Vb = −Vw = 100 meV and zero bias. Bound states are shown by the red curves. (b) Spectrum for a SL whose unit is the potential structure of Fig. 1(b). Blue and red curves show, respectively, the kx = 0 and kx = π/L results which delimit the energy bands (green coloured regions). The different types of structures clearly lead to different behaviour of the tunnelling resonances. An interesting structure to study is the fourth type of SLs shown in Fig. 11(d). To investigate the influence of the localized states (Martin et al., 2008; Martinez et al., 2009) on the transport properties we embed the anti-symmetric potential profile in a structure with unbiased layers. Conductance At zero temperature G can be calculated from the transmission F + t⊥EF )1/2/vF for bilayer graphene using Eq. (13) with G0 = (4e2Ly/2πh) (E2 and Ly the width of the sample. The angle of incidence φ is given by tan φ = ky/α with α the wave vector outside the barrier. Figure 14 shows G for the four SL types. Notice the clear differences in 1) the onset of the conductance and 2) the number and amplitude of the oscillations. Bound states. To describe bound states we assume that there are no propagat- ing states, i.e., α and β are imaginary or complex (the latter case can be solved separately), and only the eigenstates with exponentially decaying behaviour are nonzero leading to the relation fd  = N 0 ed 0  .  0 fg 0 eg (3.9) From this relation we can find the dispersion relation for the bound states. To study the localized states for the anti-symmetric potential profile (Martin et al., 2008; Martinez et al., 2009) we will use a sharp kink profile (step function). The spectrum found by the method above is shown in Fig. 15(a). We see that there are two bound states, both with negative group velocity vy ∝ ∂ε/∂ky, as found previously by (Martin et al., 2008). No bound state near zero energy was found for ky → ∞ in contradiction with (Martinez et al., 2009). For zero energy we find the Article submitted to Royal Society (b)Review: Single-layer and bilayer graphene SLs 19 Figure 13. (Color online) Contour plot of the transmission through a single barrier in (a) and (b), for width Wb = 50 nm, and through double barriers in (c), (d), (e), and (f) of equal widths Wb = 20 nm that are separated by Ww = 20 nm. Other parameters are as follows: (a) ∆b = 100 meV, Vb = 0 meV. (b) ∆b = 20 meV, Vb = 50 meV. (c) Type I : Vb = Vw = 0 meV, ∆w = 20 meV, and ∆b = 100 meV. (d) Type II : V b = −V w = 20 meV, ∆w = ∆ = 50 meV, (e) Type III : Vb = −Vw = 50 meV, ∆w = ∆b = 20 meV. (f) Type IV : Vb = Vw = 0 meV, ∆b = −∆w = 100 meV. solution ky = ± 1 2 ≈ ±(cid:112) [∆2 + (∆4 + 2∆2t2⊥)1/2]1/2 ∆t⊥/23/4, ∆ (cid:28) t⊥; (3.10) the approximation on the second line leads to the expression found by (Martin et al., 2008). (b) Superlattices The heterostructures above (see Fig. 11), can be used to create four different types of SLs (Dragoman et al., 2010). We will especially focus on type IV and type III SLs in certain limiting cases. Article submitted to Royal Society 20 M. Barbier, P. Vasilopoulos, and F. M. Peeters Figure 14. (Color online) Two-terminal conductance of four equally spaced barriers vs energy for Wb = Ww = 10 nm and different SL types I-IV. The solid red curve (type I) is for ∆b = 50 meV, ∆w = 20 meV, and Vw = Vb = 0. The blue dashed curve (type II) is for ∆b = ∆w = 50 meV and Vb = −Vw = 20 meV. The green dotted curve (type III) is for ∆b = ∆w = 20 meV and Vb = −Vw = 50 meV. The black dash-dotted curve (type IV) is for ∆b = −∆w = 50 meV and Vw = Vb = 0. Figure 15. (a) Bound states of the anti-symmetric potential profile (type IV) with bias ∆w = −∆b = 200 meV. (b) Contour plot of the transmission through a 20 nm wide barrier consisting of two regions with opposite biases ∆ = ±100 meV. For a type I SL we see in Fig. 16(a) that the conduction and valence band of the bilayer structure are qualitatively similar to those in the presence of a uniform bias. Type II structures maintain this gap, see Fig. 16(b), as there is a range in energy for which there is a gap in the SL potential in the barrier and well regions. In type III structures we have two interesting features, which can close the gap. First we see from Fig. 12(b) that for zero bias, similar to single-layer graphene, extra Dirac points appear for kx = 0, likewise for Fig. 4(d). In the case Wb = Ww = L/2 = W , kx = 0 and E = 0 the values for the ky where extra Dirac points occur are given by the following transcendental equation [cos(αW ) cos(βW ) − 1] + α2 + β2 − 4ky2 2αβ sin(αW ) sin(βW ) = 0. (3.11) Comparing the figures 12(b) and 4(d) we remark that, different from the single- layer case, for bilayer graphene the bands in the barrier region are not only flat in the x direction for large ky values but also for small ky. The latter corresponds to the zero transmission value inside the barrier region for tunneling through a Article submitted to Royal Society (a)-1.0-0.50.00.51.0-0.50.00.5kyhvF/ ttE / tt00.20.40.60.81(b)Review: Single-layer and bilayer graphene SLs 21 Figure 16. (Color online) Lowest conduction and highest valence band of the spectrum for a square SL with period L = 20 nm and Wb = Ww = 10 nm. (a) Type I : ∆b = 100 meV and ∆w = 0. (b) Type II : As in (a) for ∆b = ∆w = 50 meV, and Vb = −Vw = 25 meV. (c) Type III : Vb = −Vw = 25 meV, and ∆b = ∆w = 0. (d) Type III : Vb = −Vw = 50 meV and ∆b = ∆w = 0. (e) Type IV : Plot of the spectrum for a square SL with average potential Vb = Vw = 0 and ∆b = −∆w = 100 meV. The contours are for the conduction band and show that the dispersion is almost flat in the x direction. single unbiased barrier in bilayer graphene. Secondly, if there are no extra Dirac points (small parameter uL) for certain SL parameters, the gap closes at two points at the Fermi-level for ky = 0. The latter we will investigate a bit more in the extended Kronig-Penney model. Periodically changing the sign of the bias (type IV) introduces a splitting of the charge neutrality point along the ky axis; this agrees with what was found by (Martin et al., 2008). We illustrate that in Fig. 13(e) for a SL with ∆b = −∆w = 100 meV. We also see that the two valleys in the spectrum are rather flat in the x direction. Upon increasing the parameter ∆L, the two touching points shift to larger ±ky and the valleys become flatter in the x direction. For all four types of SLs the spectrum is anisotropic and results in very different velocities along the x and y directions. Extended Kronig-Penney model. To understand which SL parameters lead to the creation of a gap we look at the Kronig-Penney limit of type III SLs for zero bias (Barbier et al. 2010, unpublished work). Also we choose the extended Kronig- Penney model to ensure spectra symmetric with respect to the zero-energy value, such that the zero-energy solutions can be traced down more easily. If the latter zero modes exist, there is no gap. To simplify the calculations we restrict the spectrum to that for ky = 0. This assumption is certainly not valid if the parameter uL is large because in that case we expect extra Dirac points (not in the KP limit) to appear that will close the gap. The spectrum for ky = 0 is determined by the Article submitted to Royal Society 22 M. Barbier, P. Vasilopoulos, and F. M. Peeters transcendental equations with Dλ =(cid:2)(λ2 + ε2) cos λL − λ2 + ε2(cid:3) /4λ2ε2, and λ = α, β. To see whether there cos kxL = cos αL cos2 P + Dα sin2 P, cos kxL = cos βL cos2 P + Dβ sin2 P, (3.12b) (3.12a) is a gap in the spectrum we look for a solution with ε = 0 in the dispersion relations. This gives two values for kx where zero energy solutions occur kx,0 = ± arccos[1 − (L2/8) sin2 P ]/L, (3.13) and the crossing points are at (ε, kx, ky) = (0, ±kx,0, 0). If the kx,0 value is not real, then there is no solution at zero energy and a gap arises in the spectrum. From Eq. (3.12a) we see that for sin2 P > 16/L2 a band gap arises. Conductivity. In bilayer graphene the diffusive dc conductivity, given by Eq. (2.27), takes the form σµµ(εF )/σ0 = (k3 F /4πε2 F ) 1 ± δ/2(k2 (cid:104) F δ + 1/4)1/2(cid:105)2 , (3.14) with kF = [ε2 F + ∆2 ∓ (ε2 F δ − ∆2)1/2]1/2, δ = 1 + 4∆2, and σ0 = e2τF t⊥/2. Figure 17. (Color online) Conductivities, σxx in (a) and σyy in (b), vs Fermi energy for the four types of SLs with L = 20 nm and Wb = Ww = 10 nm, at temperature T = 45K; σ0 = e2τF t⊥/2. Type I : ∆b = 50 meV, ∆w = 25 meV and Vb = Vw = 0. Type II : ∆b = ∆w = 25 meV and Vb = −Vw = 50 meV. Type III : ∆b = ∆w = 50 meV and Vb = −Vw = 25 meV. Type IV : ∆b = −∆w = 100 meV and Vb = Vw = 0. In Figs. 17(a), (b) the conductivities σxx in (a) and σyy in (b) for bilayer graphene are shown for the various types of SLs defined in Sec. 3(b). Notice that for type IV SL the conductivities σxx and σyy differ substantially due to the anisotropy in the spectrum. 4. Conclusions We reviewed the electronic band structure of single-layer and bilayer graphene in the presence of 1D periodic potentials. In addition, we investigated the conditions that lead to carrier collimation in single-layer graphene and determined when extra Dirac points appear in the spectrum and what their influence is on the conductivity. Furthermore, we investigated the tunnelling through, and bound states created by, Article submitted to Royal Society Review: Single-layer and bilayer graphene SLs 23 simple barrier structures. In single-layer graphene we found that the SL spectrum can be linked to the bound states of a combined barrier and a well. In bilayer graphene we considered transport through different types of het- erostructures, where we distinguished between four types of band alignments. We also connected the bound states in an anti-symmetric potential (type IV) with the transmission through such a potential barrier. Furthermore, we investigated the same four types of band alignments in SLs. The differences between the four types of SLs are reflected not only in the spectrum but also in the conductivities parallel and perpendicular to the SL direction. For type III SLs, which have a zero bias, we found a feature in the spectrum similar to the extra Dirac points found for single-layer graphene. Also, for not too large strengths of the SL barriers we found that the valence and condunction bands touch at points in k space with ky = 0 and nonzero ky. Type IV SLs tend to split the K (K') valley into two valleys. In the Kronig-Penney limit, where we take the barriers to be δ functions V (x)/vF = P δ(x), we saw that the SL spectra, the transmission, conductance, etc., are periodic in the strength of the barriers. As s well known, this is not the case for standard electrons. An important qualitatively new feature is encountered in the extended Kronig-Penney limit for P = (n + 1/2)π, see Sec. 2(d): the Dirac point becomes a Dirac line. We expect that these relatively recent findings, that we reviewed in this work, will be tested experimentally in the near future. This work was supported by IMEC, the Flemish Science Foundation (FWO-Vl), the Bel- gian Science Policy (IAP), and the Canadian NSERC Grant No. OGP0121756. References Abergel, D. S. L., Apalkov, V., Berashevich, J., Ziegler, K. & Chakraborty, T. 2010 Properties of graphene: A theoretical perspective. Abergel, D. S. L., Apalkov, V., Berashevich, J., Ziegler, K. and Chakraborty, Tapash 2010 Properties of graphene: a theoretical perspective, Advances in Physics, 59: 4, 261 482 Arovas, D. P., Brey, L., Fertig, H. A., Kim, E. & Ziegler, K. 2010 Dirac Spectrum in Piecewise Constant One-Dimensional Potentials. ArXiv e-prints. Barbier, M., Peeters, F. M., Vasilopoulos, P. & Pereira Jr, J. M. 2008 Dirac and klein-gordon particles in one-dimensional periodic potentials. Phys. Rev. B, 77(11), 115 446. (doi:10.1103/PhysRevB.77.115446) Barbier, M., Vasilopoulos, P. & Peeters, F. M. 2009a Dirac electrons in a kronig- penney potential: Dispersion relation and transmission periodic in the strength of the barriers. Phys. Rev. B, 80(20), 205 415. (doi:10.1103/PhysRevB.80.205415) Barbier, M., Vasilopoulos, P. & Peeters, F. M. 2010 Extra dirac points in the energy spectrum for superlattices on single-layer graphene. Phys. Rev. B, 81(7), 075 438. (doi:10.1103/PhysRevB.81.075438) Barbier, M., Vasilopoulos, P., Peeters, F. M. & Pereira Jr, J. M. 2009b Bilayer graphene with single and multiple electrostatic barriers: Band structure and transmission. Phys. Rev. B, 79(15), 155 402. (doi:10.1103/PhysRevB.79.155402) Article submitted to Royal Society 24 M. Barbier, P. Vasilopoulos, and F. M. Peeters Bliokh, Y. P., Freilikher, V., Savel'ev, S. & Nori, F. 2009 Transport and localization in periodic and disordered graphene superlattices. Phys. Rev. B, 79(7), 075 123. (doi:10.1103/PhysRevB.79.075123) Brey, L. & Fertig, H. A. 2009 Emerging zero modes for graphene in a periodic po- tential. Phys. Rev. Lett., 103(4), 046 809. (doi:10.1103/PhysRevLett.103.046809) Castro, E. V., Novoselov, K. S., Morozov, S. V., Peres, N. M. R., dos Santos, J. M. B. L., Nilsson, J., Guinea, F., Geim, A. K. & Neto, A. H. C. 2007 Biased bilayer graphene: Semiconductor with a gap tunable by the electric field effect. Phys. Rev. Lett., 99(21), 216 802. (doi:10.1103/PhysRevLett.99.216802) Castro Neto, A. H., Guinea, F., Peres, N. M. R., Novoselov, K. S. & Geim, A. K. 2009 The electronic properties of graphene. Rev. Mod. Phys., 81(1), 109 -- 162. (doi:10.1103/RevModPhys.81.109) Charbonneau, M., van Vliet, K. M. & Vasilopoulos, P. 1982 Linear response theory revisited. III. one-body response formulas and generalized Boltzmann equations. J. Math. Phys., 23(2), 318 -- 336. Dragoman, D., Dragoman, M. & Plana, R. 2010 Tunable electrical superlattices in periodically gated bilayer graphene. J. Appl. Phys., 107(4), 044 312. (doi: 10.1063/1.3309408) Giovannetti, G., Khomyakov, P. A., Brocks, G., Kelly, P. J. & van den Brink, J. 2007 Substrate-induced band gap in graphene on hexagonal boron nitride: Ab initio density functional calculations. Phys. Rev. B, 76(7), 073 103. (doi: 10.1103/PhysRevB.76.073103) Gomes, J. V. & Peres, N. M. R. 2008 Tunneling of dirac electrons through spatial regions of finite mass. J. Phys.: Condensed Matter, 20(32), 325 221. Ho, J. H., Chiu, Y. H., Tsai, S. J. & Lin, M. F. 2009 Semimetallic graphene in (doi:10.1103/ a modulated electric potential. Phys. Rev. B, 79(11), 115 427. PhysRevB.79.115427) Huard, B., Sulpizio, J. A., Stander, N., Todd, K., Yang, B. & Gordon, D. G. 2007 Transport measurements across a tunable potential barrier in graphene. Phys. Rev. Lett., 98(23), 236 803. (doi:10.1103/PhysRevLett.98.236803) Katsnelson, M. I., Novoselov, K. S. & Geim, A. K. 2006 Chiral tunnelling and the klein paradox ingraphene. Nat. Phys., 2(9), 620 -- 625. (doi:10.1038/nphys384) Klein, O. 1929 Die reflexion von elektronen an einem potentialsprung nach der relativistischen dynamik von dirac. Zeitschrift fur Physik A Hadrons and Nuclei, 53(3), 157 -- 165. (doi:10.1007/BF01339716) Martin, I., Blanter, Y. M. & Morpurgo, A. F. 2008 Topological confinement in bilayer graphene. Phys. Rev. Lett., 100(3), 036 804. (doi:10.1103/PhysRevLett. 100.036804) Article submitted to Royal Society Review: Single-layer and bilayer graphene SLs 25 Martinez, J. C., Jalil, M. B. A. & Tan, S. G. 2009 Robust localized modes in bilayer graphene induced by an antisymmetric kink potential. Appl. Phys. Lett., 95(21), 213106. (doi:10.1063/1.3263150) McCann, E. 2006 Asymmetry gap in the electronic band structure of bilayer graphene. Phys. Rev. B, 74(16), 161 403. (doi:10.1103/PhysRevB.74.161403) Novoselov, K. S., Geim, A. K., Morozov, S. V., Jiang, D., Zhang, Y., Dubonos, S. V., Grigorieva, I. V. & Firsov, A. A. 2004 Electric field effect in atomically thin carbon films. Science, 306(5696), 666 -- 669. (doi:10.1126/science.1102896) Park, C.-H., Son, Y.-W., Yang, L., Cohen, M. L. & Louie, S. G. 2009a Electron beam supercollimation in graphene superlattices. Nano Lett., 8(9), 2920 -- 2924. (doi:10.1021/nl801752r) Park, C.-H., Son, Y.-W., Yang, L., Cohen, M. L. & Louie, S. G. 2009b Landau levels and quantum hall effect in graphene superlattices. Phys. Rev. Lett., 103(4), 046 808. (doi:10.1103/PhysRevLett.103.046808) Park, C.-H., Yang, L., Son, Y.-W., Cohen, M. L. & Louie, S. G. 2008a Anisotropic behaviours of massless dirac fermions in graphene under periodic potentials. Nat. Phys., 4(3), 213 -- 217. Park, C.-H., Yang, L., Son, Y.-W., Cohen, M. L. & Louie, S. G. 2008b New gener- ation of massless dirac fermions in graphene under external periodic potentials. Phys. Rev. Lett., 101(12), 126 804. (doi:10.1103/PhysRevLett.101.126804) Pereira Jr, J. M., Mlinar, V., Peeters, F. M. & Vasilopoulos, P. 2006 Confined states and direction-dependent transmission in graphene quantum wells. Phys. Rev. B, 74(4), 045 424. (doi:10.1103/PhysRevB.74.045424) Pereira Jr, J. M., Peeters, F. M., Chaves, A. & Farias, G. A. 2010 Klein tunneling in single and multiple barriers in graphene. Semiconductor Science and Technology, 25(3), 033 002. (doi:10.1088/0268-1242/25/3/033002) Pereira Jr, J. M., Vasilopoulos, P. & Peeters, F. M. 2007a Graphene-based resonant- tunneling structures. Appl. Phys. Lett., 90(13), 132122. (doi:10.1063/1.2717092) Pereira Jr, J. M., Vasilopoulos, P. & Peeters, F. M. 2007b Tunable quantum dots in bilayer graphene. Nano Lett., 7(4), 946 -- 949. (doi:10.1021/nl062967s) Roslyak, O., Iurov, A., Gumbs, G. & Huang, D. 2010 Unimpeded tunneling in graphene nanoribbons. J. Phys.: Condensed Matter, 22(16), 165 301. San-Jose, P., Prada, E., McCann, E. & Schomerus, H. 2009 Pseudospin valve in bilayer graphene: Towards graphene-based pseudospintronics. Phys. Rev. Lett., 102(24), 247 204. (doi:10.1103/PhysRevLett.102.247204) Schliemann, J., Loss, D. & Westervelt, R. M. 2005 Zitterbewegung of electronic wave packets in iii-v zinc-blende semiconductor quantum wells. Phys. Rev. Lett., 94(20), 206 801. (doi:10.1103/PhysRevLett.94.206801) Article submitted to Royal Society 26 M. Barbier, P. Vasilopoulos, and F. M. Peeters Sun, J., Fertig, H. A. & Brey, L. 2010 Effective Magnetic Fields in Graphene Su- perlattices. ArXiv e-prints. Wang, L.-G. & Zhu, S.-Y. 2010 Electronic band gaps and transport properties in graphene superlattices with one-dimensional periodic potentials of square barri- ers. Phys. Rev. B, 81(20), 205 444. (doi:10.1103/PhysRevB.81.205444) Winkler, R., Zulicke, U. & Bolte, J. 2007 Oscillatory multiband dynamics of free particles: The ubiquity of zitterbewegung effects. Phys. Rev. B, 75(20), 205 314. (doi:10.1103/PhysRevB.75.205314) Young, A. F. & Kim, P. 2009 Quantum interference and klein tunnelling in graphene heterojunctions. Nat. Phys., 5(3), 222 -- 226. (doi:10.1103/PhysRevLett. 98.236803) Zawadzki, W. 2005 Zitterbewegung and its effects on electrons in semiconductors. Phys. Rev. B, 72(8), 085 217. (doi:10.1103/PhysRevB.72.085217) Article submitted to Royal Society 

1808.00342

3

1808

Giant tunable nonreciprocity of light in Weyl semimetals

The propagation of light in Weyl semimetal films is analyzed. The magnetic family of these materials is known by anomalous Hall effect, which, being enhanced by the large Berry curvature, allows one to create strong gyrotropic and nonreciprocity effects without external magnetic field. The existence of nonreciprocal waveguide electromagnetic modes in ferromagnetic Weyl semimetal films in the Voigt configuration is predicted. Thanks to the strong dielectric response caused by the gapless Weyl spectrum and the large Berry curvature, ferromagnetic Weyl semimetals combine the best waveguide properties of magnetic dielectrics or semiconductors with strong anomalous Hall effect in ferromagnets. The magnitude of the nonreciprocity depends both on the internal Weyl semimetal properties, the separation of Weyl nodes, and the external factor, the optical contrast between the media surrounding the film. By tuning the Fermi level in Weyl semimetals, one can vary the operation frequencies of the waveguide modes in THz and mid-IR ranges. Our findings pave the way to the design of compact, tunable, and effective nonreciprocal optical elements.

cond-mat.mes-hall

cond-mat

Giant tunable nonreciprocity of light in Weyl semimetals O. V. Kotov1, ∗ and Yu. E. Lozovik2, 1, 3, † 1N. L. Dukhov Research Institute of Automatics (VNIIA), 127055 Moscow, Russia 2Institute for Spectroscopy, Russian Academy of Sciences, 142190 Troitsk, Moscow, Russia 3National Research University Higher School of Economics, 101000 Moscow, Russia The propagation of light in Weyl semimetal films is analyzed. The magnetic family of these materials is known by anomalous Hall effect, which, being enhanced by the large Berry curvature, allows one to create strong gyrotropic and nonreciprocity effects without external magnetic field. The existence of nonreciprocal waveguide electromagnetic modes in ferromagnetic Weyl semimetal films in the Voigt configuration is predicted. Thanks to the strong dielectric response caused by the gapless Weyl spectrum and the large Berry curvature, ferromagnetic Weyl semimetals combine the best waveguide properties of magnetic dielectrics or semiconductors with strong anomalous Hall effect in ferromagnets. The magnitude of the nonreciprocity depends both on the internal Weyl semimetal properties, the separation of Weyl nodes, and the external factor, the optical contrast between the media surrounding the film. By tuning the Fermi level in Weyl semimetals, one can vary the operation frequencies of the waveguide modes in THz and mid-IR ranges. Our findings pave the way to the design of compact, tunable, and effective nonreciprocal optical elements. I. INTRODUCTION Weyl semimetals (WSs), being topologically nontriv- ial phase of matter, have recently attracted significant attention due to their massless bulk fermions and pro- tected Fermi arc surface states with the corresponding topological transport phenomena [1 -- 5]. WS band struc- ture contains an even number [6] of nondegenerate band- touching points (Weyl nodes), which are topologically stable and can be regarded as magnetic monopoles and antimonopoles in the momentum space with positive or negative chiral charges and corresponding nonzero Chern numbers acting as the source and drain for the Berry cur- vature field [7, 8]. The topological protection of mass- less fermions in WSs against weak perturbations follows from the separation of the individual Weyl nodes with opposite topological charges in momentum space, as the chiral Weyl nodes can only be destroyed by chirality mix- ing, which requires two opposite chirality Weyl nodes to meet. Such a separation demands breaking of either time-reversal (T ) or inversion (P) symmetry, or both [2]. In WSs with lack of P symmetry, the Weyl nodes sep- aration is roughly proportional to the strength of the spin-orbit coupling (SOC), which indicates the crucial role played by SOC in the formation of WSs [9]. By contrast, in T - and P-invariant bulk Dirac semimetals (BDSs) where, according to Kramers theorem, all bands are doubly degenerate, the massless bulk fermions require additional crystal symmetries to be stable [10]. The realization of a BDS phase in Na3Bi, Cd3As2, and ZrTe5 compounds was predicted [11, 12] and confirmed experimentally [13 -- 17]. WS phase natural realizations contain the family of T -broken magnetic materials in- cluding pyrochlore iridates Y2IrO7, Eu2IrO7 [18, 19], fer- ∗ oleg.v.kotov@yandex.ru † lozovik@isan.troitsk.ru romagnetic spinels HgCr2Se4 [20], and Heusler ferromag- nets Co3S2Sn2, Co3S2Se2 [21 -- 23]. This family also in- cludes spin gapless compensated ferrimagnets Ti2MnAl, where, in contrast to ferromagnetic WSs, the spin de- generacy is broken even without SOC, and T -broken WS phase exist despite a zero net magnetic moment [24]. The WS family of P-broken nonmagnetic materials includes noncentrosymmetric compounds TaAs, TaP, NbAs, and NbP [25 -- 33] (the detailed WS classification can be found in Refs. [34 -- 36]). Moreover, in some compounds, e.g., WTe2 [37, 38] and MoTe2 [39 -- 41], the tilt of the Weyl cones exceeds the Fermi velocity giving rise to a type -- II WS with open Fermi surface and a different type of Weyl fermions at the boundary between electron and hole pockets [37, 39, 42]. The nontrivial bulk band topology of WSs manifests in a number of exotic physical effects such as the protected against weak perturbations Fermi arc surface states [43 -- 47] that connect the projections of the Weyl nodes in the surface Brillouin zone, the chiral anomaly [6, 48 -- 51] (nonconservation of the chiral charge transferred be- tween Weyl nodes of opposite chirality), and related neg- ative longitudinal magnetoresistance [5, 52, 53] quadratic in magnetic field, which appears if parallel electric and magnetic fields are applied. Also, WSs possess two basic phenomena related to the chiral anomaly: the chiral mag- netic effect (CME) [5, 17, 54 -- 57] and the anomalous Hall effect (AHE) [5, 58 -- 60], which are closely related to the topological magnetoelectric effect in T -invariant topolog- ical insulators [8]. The CME, manifested in P-broken WSs as the electrical currents induced along the mag- netic field, hypothetically could be caused by only a mag- netic external field and not be associated with the chiral anomaly [61]. However, in an equilibrium state, when all contributions from filled electronic states are taken into account, the static magnetic-field-driven current must vanish [62]. Thus, the nonvanishing CME implies the nonzero chiral chemical potential (the difference between local chemical potentials in Weyl nodes), which can be re- 9 1 0 2 n a J 8 1 ] l l a h - s e m . t a m - d n o c [ 3 v 2 4 3 0 0 . 8 0 8 1 : v i X r a alized only in the nonequilibrium state dynamically gen- erated by DC parallel electric and magnetic fields and associated with the chiral anomaly [36, 63]. While the dynamic CME with the violation of the chiral current conservation is the consequence of the chiral anomaly, the AHE in any T -broken system, being the Hall effect in the absence of a magnetic field, strictly speaking, may be not a part of the chiral anomaly in WSs. We underline that due to the nonzero Chern numbers of the Weyl nodes, magnetic WSs are distinguished from ordinary ferromag- nets by a lack of spin-dependent charge carrier scattering (extrinsic factor) and Fermi-surface contributions to the AHE. Instead, the AHE in ideal WSs (with two Weyl nodes in the vicinity of the Fermi level) is purely intrinsic and determined only by the distance between the Weyl nodes in momentum space [58]. However, this is true only for the type-I WSs which have a point-like Fermi surface, while the AHE in the type-II WSs with tilted conical spectrum around the Weyl node is not univer- sal and can change sign as a function of the parameters quantifying the tilt [60]. This universality can also be violated in the nodal-line WSs, such as Co3S2Sn2, where the gapped nodal lines contribution to the AHE may be higher than the impact of the Weyl nodes themselves [22]. Nevertheless, an ideal WS, possessing purely topological AHE without nodal lines or magnetic moment contri- butions, can be found among spin gapless compensated ferrimagnets (e.g., Ti2MnAl [24]). Notice that the cubic lattice symmetry of the typical magnetic WS crystals, such as pyrochlore iridates [18, 19], enforces vanishing of the AHE due to the absence of a preferred axis. Never- theless, the AHE can be recovered by applying a uniaxial strain that lowers the symmetry [59]. The effects caused by WSs' nontrivial topology man- ifest in the optical [55, 62, 64 -- 71] and electron density responses [72 -- 85]. In particular, the AHE and CME give rise to gyrotropic terms in dielectric function [86], which lead to the Faraday and Kerr magneto-optical ef- fects in T -broken WS [65, 66] and to the natural optical activity in P-broken WSs [55, 67, 87]. Moreover, the AHE, CME, and corresponding photocurrents in WSs can be generated by illuminating with circularly polar- ized light [68, 69, 88]. Nontrivial topology of T -broken WSs also results in the chiral Fermi arc plasmons with hyperbolic isofrequency contours [82, 83], in the chiral electromagnetic (EM) waves propagating at the vicinity of the magnetic domain wall in WSs [89], in the trans- verse EM waves in a static magnetic field (helicons) [90], and in the unusual EM modes with a linear dispersion in a neutral WS [79, 80]. Besides, the AHE also makes the surface plasmon polaritons (SPPs) in WS chiral with- out applying an external magnetic filed (compare with Ref. [91]). Particularly, in Ref. [81], the behavior of SPP on the surface of WSs is calculated at different orien- tations of the AHE vector (b) and the direction of SPP propagation (q). The existence of the nonreciprocal SPP in WS, whose dispersion depends on the sign of the wave vector, is predicted in the Voigt configuration, when both 2 b and q are in the plane of WS film, but perpendicular to each other. The nonreciprocal unidirectional EM waves are widely known in magneto-optics, and dielectric waveguides (WGs) with ferrite cores or substrates, as well as films of magnetic dielectrics (MDs) (see Refs. [92 -- 94]) are usu- ally used for their transmission. Nonreciprocal optical elements are used in optical radiation control systems to create unidirectional optical circuits [95], for the di- rected excitations in a ring laser [96], in a laser gyro- scope to eliminate the capture of the frequencies of coun- terpropagating modes [97], as well as in fiber optic gy- roscopes for the initial phase shift between the counter waves [98]. The theoretical description of nonreciprocal SPP was given in Refs. [99],[100], and the generalization for nonreciprocal WG modes in a film in the Voigt con- figuration was made in Ref. [101]. Notice that in the Faraday configuration (magnetic field is along the propa- gation direction, parallel to the film) also the WG modes may exist but they will be reciprocal [102, 103]. For the design of compact optical elements with strong nonre- ciprocity effects, which do not need external sources of magnetic field, it is better to use materials with strong AHE and good WG properties. On the one hand, the MDs [104] or delute magnetic semiconductors [105, 106] may be good WGs but they possess weak AHE. On the other hand, the ferromagnets may have strong AHE but they allow only anomalous light penetration, while or- dinary dielectric response is suppressed due to the large electronic band gap. A compromise solution to this prob- lem could become the Dirac (Weyl) ferromagnets, where, as it was shown in our previous work [107], the weakly damped WG modes may arise due to the gapless spec- trum. In this paper, we propose to consider ferromagnetic WSs as the best candidates for the material which com- bines the WG properties of MDs or magnetic semicon- ductors with strong AHE in ferromagnets. Thanks to the strong dielectric response caused by the gapless Weyl spectrum and the large Berry curvature coming from the entangled Bloch electronic bands with SOC, ferro- magnetic WSs may demonstrate good WG properties to- gether with strong AHE, even stronger than in ordinary ferromagnets. We study the propagation of light in ferro- magnetic WS films in the Voigt configuration without an external magnetic field. The role of a magnetic field plays the AHE in WS. We predict not only the nonreciprocity of the SPP on both sides of a WS film but also the ex- istence of the nonreciprocal WG EM modes inside the film. The dispersions of the WG modes were obtained within the two-band model, accounting for the gapless nature of the Weyl spectrum. We also underline the key role played by the optical contrast between the media surrounding the film in the nonreciprocity magnitude of the predicted WG modes. Besides, the possibilities of varying of the nonreciprocity magnitude and operation frequencies of these modes by tuning the Fermi level in WS are discussed. Finally, we compare the AHE parame- ters in some real WS compounds. Our calculations show that WSs may become a good platform for the compact and tunable optical elements with strong nonreciprocity. II. WEYL SEMIMETALS OPTICAL RESPONSE 3 Generally, the nonreciprocity effects in the Voigt con- figuration, as well as the magneto-optical effects, arise in a T -broken media, and the violation of P symmetry leads to the natural optical activity effects, like in chiral media. In the case of BDS, the breaking of T or P sym- metry splits each doubly degenerated Dirac point into a pair of Weyl nodes of opposite chirality, which are sepa- rated in the momentum space by vector 2b or in energy space by 2b0 (the chiral chemical potential). In the first case, there can be the AHE with the currents across the electric field, and in the second case the CME may occur with the currents induced along the magnetic field. The manifestation of WS topological nature in the optical re- sponse can be described by the additional axion term in the EM action [61, 63, 108]: Sθ = −e2(cid:14)(cid:0)4π2c(cid:1)(cid:90) dt d3r θ(r, t) E·B, (1) where θ(r, t) = 2(b·r − b0t) is the axion angle. Varying this axion action with respect to the EM vector potential A we obtain the corresponding currents jθ = δSθ/δA = −e2(cid:14)(cid:0)4π2(cid:1)(cid:104)∇θ(r, t) × E + θ(r, t)/c B (cid:105) , (2) where the first term corresponds to AHE and the sec- ond one to the CME currents. These currents result in additional terms of the displacement vector [81]: D = εE + ie2 πω 2b × E − ie2 πωc 2b0B. (3) Thus, to account for WS topological properties in the op- tical response, one may use the standard form of Maxwell equations with D = (cid:98)ε2E taking WS dielectric tensor in the form ε2 (cid:98)ε2 = 0 ε2 0 0 −iε2b 0 iε2b ε2  , (4) where ε2 is a BDS dielectric function and ε2b is a nondi- agonal component caused by the AHE and CME. Since the nonreciprocal properties are always associated with the Hall response, we will consider the ferromagnetic WS in an equilibrium state without any external fields, with Weyl nodes separated only in momentum space (i.e., b0 = 0). For this case, as follows from Eq. (3), the non- diagonal component of the tensor Eq. (4) can be written as iε2b = i2be2(cid:14)(πω) = i
∞Ωb (cid:14)Ω, (5) FIG. 1. (a) The dispersion of BDS dielectric functions in the one-band (Drude) εD Eq. (6) and two-band ε2 Eq. (7) models. (b) The dispersion of the components of WS dielectric tensor [diagonal component ε2, nondiagonal component caused by the AHE ε2b Eq. (5)] and the Voigt dielectric function εV. The parameters of WS are set as EF = 150meV, vF = 106m/s, g = 24, εc = 3, εb = 6.2, ε∞ = 13, Ωb = Ωp ≈ 0.93 (i.e., 2b ≈ 0.4A−1 and ε2b(Ω = Ωp) = 12). (cid:14)(kFπε∞), Ω = ω/EF, EF is the Fermi where Ωb = 2brs level, kF = EF/vF is the Fermi momentum, vF is the Fermi velocity, and ε∞ is the effective dielectric constant taking into account all interband electronic transitions. In Ref. [81], the standard one-band Drude model was used for ε2, accounting only for the intraband electronic transitions: p = 2rsg(cid:14)(3πε∞) denotes the bulk plasma fre- e2(cid:14)vF is the effective fine structure constant, and g is the where Ω2 quency constant normalized to the Fermi level, rs = (6) p degeneracy factor (the number of nondegenerated Weyl nodes). To describe the dielectric response in BDS more accurately, one should use the two-band model, taking into account the interband electronic transitions in the Dirac cone. As we have shown in Ref. [107], according to this model, a BDS dielectric function in the local re- sponse approximation at zero temperature has the form εD = ε∞(cid:0)1 − Ω2 (cid:14)Ω2(cid:1) , (cid:20) (cid:18) 4Λ2 ln Ω2 − 4 (cid:19) (cid:21) + iπθ(Ω − 2) , ε2 = εb − 2rsg 3π 1 Ω2 + rsg 6π (7) where Λ = Ec/EF (Ec is the cutoff energy beyond which the Dirac spectrum is no longer linear), εb is the effective background dielectric constant accounting the contribu- tions from all bands below the Dirac cone. In Ref. [107], we obtained εb = 6.2 for g = 24 and
∞ = 13 (Eu2IrO7 [19]). The difference between Eqs. (6) and (7) [see Fig. 1(a)] is manifested at frequencies above the Fermi level when the dielectric behavior (ε > 1) occurs and WG modes can exist. Notice that all the Weyl nodes have an equal contri- bution to the diagonal component of the dielectric ten- sor, which after summation gives the g-factor in Eq. (6). In contrast, to calculate the nondiagonal component, strictly speaking, one should integrate the Berry curva- ture over all occupied states in the first Brillouin zone, and Weyl nodes with different Chern numbers may even compensate each other, resulting in the vanishing of the AHE [59]. Moreover, in the nodal-line WSs with strong AHE, such as Co3S2Sn2, the gapped nodal lines contri- bution to the integrated Berry curvature is higher than the impact of the Weyl nodes themselves, and the anoma- lous Hall conductivity is more determined by the shape of the nodal lines than by the Weyl nodes separation [22]. Thus, strictly speaking, Eq. (5) describes the con- tribution of only a one Weyl pair. To account for all the pairs and other possible sources of the Berry curva- ture the nondiagonal component given by Eq. (5) should be multiplied by a coefficient depending on the Brillouin zone topology of a particular compound, which in general may be not directly expressed through the g-factor. For example, for Co3S2Sn2 the numerical calculations and experimental measurements of the anomalous Hall con- ductivity gives the value about 1130 S/cm [22], while the expression σAH = 2be2(cid:14)(cid:0)4π2(cid:1) following from Eq. (5) gives at 2b = 0.47A−1 about 290 S/cm, which is approx- imately four times lower. Nevertheless, Eq. (5) can be used as a good estimation of the minimum value of the nonvanishing AHE. III. NONRECIPROCAL WAVES Let us consider the propagation of EM waves in the WS film with Weyl pairs where, in each pair, nodes are separated in momentum space by the wave vector 2b. In the Voigt configuration, where nonreciprocal solutions can be found, the waves propagate in the plane of the film but perpendicular to the magnetic field (see Fig. 2). In our case, the AHE plays the role of "internal magnetic field"with the direction determined by the vector b. The WS film is asymmetrically bounded by two semi-infinite dielectric media with dielectric functions ε1 and ε3. As it will be shown below, for the nonreciprocal WG modes, it is important that ε1 (cid:54)= ε3. The wave equation ∇ × (∇ × E)− k2 ω/c for the considered system in the Voigt configuration has the form V − k2 0ε2 0 0(cid:98)ε2E = 0 with the vacuum wave vector k0 =  = 0, Ex (cid:14)ε2 are the Voigt wave (8) where qy is the wave vector of EM waves, kV = q2 + k2 (cid:112)k2 0 V − k2 k2 0ε2 qkV + k2 0iε2b 0εV − q2 and εV = ε2 − ε2 2b 0  0 qkV − k2 0iε2b q2 − k2 0ε2 Ey Ez vector and dielectric function, respectively, which deter- mine a light behavior inside the film in the considered configuration. This function has the resonance at the plasma frequency (ε2 = 0), which leads to the splitting of the WG region (εV > 1) into two parts: the lower one is below the plasma frequency and the upper one is above it [Fig. 1(b)]. The lower region, where the anoma- lous light penetration into a metal at frequencies below the plasma one occurs, is typical for any magnetoplasma system in the Voigt configuration and is connected with 4 FIG. 2. The schematics of the nonreciprocal EM wave prop- agation in the WS film. In the Voigt configuration, the wave vector qy lies in the plane of the film but perpendicular to the AHE vector 2bx separating the Weyl nodes in momen- space dielectric constant, and ε3 (cid:54)= ε1 is the dielectric func- tion of a thick substrate. tum space. (cid:98)ε2 is the dielectric tensor of WS, ε1 is the free the modification of the plasma frequency by the cyclotron resonance (see, e.g., Ref. [109]). Interestingly, this phe- nomenon is accompanied by the effect of negative refrac- tion, which can be observed not only in metamaterials but also in any gyrotropic (magnetic or chiral) system [110 -- 113]. This effect can also be observed in WSs, which has been recently predicted in Ref. [114, 115]. The up- per WG region is the manifestation of the dielectric re- sponse in any MD, magnetic semiconductor, or magnetic semimetal and may exist not only in the Voigt configura- tion [102, 103]. So, semiconductors or semimetals in an external magnetic field [102, 103, 109] or with intrinsic magnetic moment, such as dilute magnetic semiconduc- tors [105], in the Voigt configuration may possess both lower and upper WG regions. However, for the design of the WGs with strong nonreciprocity effects, which do not need external sources of magnetic field, it is better to use ferromagnets, where the AHE is much stronger than in MDs [104] or magnetic semiconductors [106]. Thus, ferromagnetic WSs are the most suitable materials for these purposes. On the one hand, unlike ordinary ferro- magnets, due to the gapless Weyl spectrum they possess a strong dielectric response and corresponding upper WG region, but on the other hand, unlike ordinary magnetic semiconductors, WSs due to the large Berry curvature have very strong AHE, even stronger than in ordinary fer- romagnets [22]. Notice that in the Voigt configuration, the TE-polorized (s) EM waves will not feel the AHE, like in the case of external magnetic field, the carriers drifting parallel to the applied field do not experience a magnetic force. So, all the above comments are related to TM-polorized (p) WG modes. Thus, we consider only the TM waves with field com- ponents Hx, Ey, Ez, and magnetic field in the form Hx(r, t) = Hx(z)ei(qy−ωt), where Hx(z) in the media Fig. 2) is expressed as H1x(z) = H1e−k1z, H2x(z) = with ε1 (z > d), (cid:98)ε2 (0 < z < d), and ε3 (z < 0) (see (cid:2)k1k3ε2 2 − k2 2ε1ε3 (cid:3) tan (k2d) + k2ε2 (k1ε3 + k3ε1) = 0, (cid:3) tan (k2d) + k2 (k1 + k3) = 0, 0ε2 − q2 and k1,3 =(cid:112)q2 − k2 2 (10) (11) and for the TE waves, (cid:2)k1k3 − k2 where k2 =(cid:112)k2 H2eikVz +(cid:102)H2e−ikVz, and H3x(z) = H3ek3z, respectively. These fields correspond to the WG modes propagating inside the film with the Voigt wave vector kV and ex- ponentially decaying out of it. Employing the boundary conditions at the two interfaces z = d and z = 0, we obtain the dispersion relation for the TM waves in the Voigt configuration (compare with Ref. [101]): 2ε1ε3 ± qε2b (k1ε3 − k3ε1)(cid:3) tan (kVd) 2 − ε2 (cid:0)ε2 (cid:2)k1k3 (cid:1) + k2 where k1,2,3 =(cid:112)q2 − k2 2b + kVε2 (k1ε3 + k3ε1) = 0, (9) 0ε1,2,3 and ± sign before q corre- sponds to the forward and backward propagation direc- tions. Thus, the TM waves in the considered configura- tion will be nonreciprocal, which means that at certain frequencies they may propagate only forward (p>) but at another frequencies only backward (p<). For BDS films without Weyl features, the dispersion relations have a standard form [107]: for the TM waves, 0ε1,3. In WS film, the TE waves obey Eq. (11) but the TM waves de- fined by Eq. (9), which in the absence of the AHE turns into Eq. (10) in the limit ε2b → 0 and by successive sub- stitutions: k2 → ik2, then kV → k2. The dispersion of SPP in WS or BDS films can be obtained from Eqs. (9) or (10) by substitution kV → ikV and k2 → ik2, respec- tively. ε3 > ω > qc(cid:14)√ Using Eqs. (4), (5), (7), and (9) -- (11) on Fig. 3 at the same model parameters as for Fig. 1, we compared the dispersions of light and SPP in BDS (ε2) film and WS ((cid:98)ε2) film in the Voigt configuration with thicknesses gion at qc(cid:14)√ ω > qc(cid:14)√ SPP branches at qc(cid:14)√ d = 0.5 µm on the semi-infinite dielectric substrate (ε3 = 4). For the case of BDS [Fig. 3(a)] we reproduced our previous result from Ref. [107], obtaining the WG re- ε2, leaky waves region at ε3, and high (in-phase) and low (out-of-phase) ε2 > ω. Both WG modes and SPP in this case are reciprocal. In the case of WS [Fig. 3(b)] for TM waves, we get the splitting of the WG region into two parts, one of which lies below the plasma frequency. Both of these parts may contain the nonreciprocal TM WG modes. We also obtain the two pairs of the nonrecip- rocal SPP branches, which agrees with the results from Ref. [81]. Moreover, in the upper pair the nonreciprocity effect is much larger. Remarkably, from Eq. (9) it follows that the nonreciprocity effect of TM waves in the WS film is determined not only by the component ε2b, but also by the term (k1ε3 − k3ε1). Therefore, the nonreciprocity effect grows with the optical contrast ε1 − ε3 between the media above and below the WS film. This can be 5 FIG. 3. The dispersions of light and SPP in BDS (ε2) film (a) and WS ((cid:98)ε2) film in the Voigt configuration (b) with thick- nesses d = 0.5 µm on the semi-infinite dielectric substrate with ε3 = 4, while the medium above the films with ε1 = 1. SPP> and SPP< are the forward and backward nonreciprocal SPP, respectively. SPE denotes the interband Landau damp- ing region. The parameters of BDS and WS are the same as for Fig. 1. understood from the fact that nonreciprocal SPP excited on both sides of the film will compensate each other if the media from both sides are the same. In the measure of the optical contrast between these media, the nonre- ciprocity effect will appear in the collective SPP or WG TM modes propagating along the film. To demonstrate these phenomena, we considered the case of ordinary con- trast, when the WS film lies on a dielectric substrate, and the case of high contrast, when the substrate is metallic. In the case of ordinary contrast, we compared the dis- persions of WG modes and SPP in BDS and WS films with thickness d = 0.5 µm, as well as in the film of MD with thickness d = 80 nm and the same direction of magnetization as in WS. All the films are placed on a semi-infinite dielectric substrate (ε3 = 4). The optical response of the MD we described by the same dielec- tric tensor Eq. (4) as for the WS but with frequency independent components (ε∞ = 13, ε2b = 4). Compar- ing BDS and WS films [see Figs. 4(a) and 4(b)], we get that the WG TM mode in WS becomes nonreciprocal and splits into two branches corresponding to the oppo- site directions of propagation, while the WG TE mode remains unchanged. There are also nonreciprocal TM waves in the WG region below the plasma frequency and two pairs of the nonreciprocal SPP in the evanescent re- gion. In the MD film, certainly, there is no SPP and only one nonreciprocal WG region with a linear dispersion law [Fig. 4(c)]. ε3 = 3.7−Ω2 [116]), at the considered frequencies its dielectric con- stant is very large and negative (ε3 ∼ −103) which leads to a high optical contrast, and hence to a strong non- reciprocity effect. In the BDS film on the metallic sub- strate, the WG TM and TE modes swap places by fre- quency, and also only the high (in-phase) SPP branch exists [Fig. 4(d)]. In the WS film, on the metal there is really a large difference between the dispersion of the (cid:14)EF and ωp = 9.2eV (cid:14)Ω2, where Ωp = ωp In the case of metallic substrate (we take silver with p 6 The dispersions of light and SPP in BDS (ε2) film with d = 0.5 µm (a), in WS ((cid:98)ε2) film with d = 0.5 µm and (cid:14)EF and ωp = 9.2eV ). The WG TE (s) modes do not feel the AHE and WG nonreciprocal (SPP, WG TM modes), respectively. The MD dielectric tensor ((cid:98)ε2) is the same as for the WS but with frequency FIG. 4. in MD film with d = 80 nm in the Voigt configuration. (a)-(c) the case of the ordinary contrast when all the films on the semi-infinite dielectric substrate (ε3 = 4), (d)-(f) the case of the high contrast when all the films on the semi-infinite silver substrate (ε3 = 3.7 − Ω2 TM (p) modes in (b) and (c), (e) and (f) become nonreciprocal: (SPP>, p>) and (SPP<, p<) are the forward and backward independent components (ε∞ = 13, ε2b = 4). The medium above the films for all cases with ε1 = 1. The parameters of BDS and WS are the same as for Fig. 1. (cid:14)Ω2, where Ωp = ωp p nonreciprocal waves propagating in the opposite direc- tions [Fig. 4(e)]. In particular, the TM mode in one direction remains WG (p>) and in the opposite direc- tion it becomes evanescent (SPP<). Also, in this case, the in-phase SPP splits to the pair of the nonreciprocal SPP with very high difference between SPP> and SPP<. The similar behavior demonstrates the MD on the metal, where the WG TM and TE modes also swap places by frequency, the strong nonreciprocity effect of the WG TM modes takes place, and the nonreciprocal SPP arise due to the metallic substrate [Fig. 4(f)]. IV. DISCUSSION For both ordinary and high optical contrasts between the media above and below the films, in the WS film the nonreciprocal WG modes and SPP can exist, similar to the waves which can be observed in WG with ferrite rods or MD films. However, in contrast to a MD film, in the WS film the WG mode frequency depends nonlinearly on the wave vector, and also there are two WG regions, one of which lies below the plasma frequency where the negative refraction in WS can be observed [114]. But the main advantage of WS over MD films is the magnitude of the nonreciprocity effect. In the WS, it depends not only on the surrounding media optical contrast, but also on the separation of the Weyl nodes in momentum space 2b = ΩbkFπε∞/rs. For all the figures, we took model parameters EF = 150meV, g = 24, Ωb = Ωp ≈ 0.93, i.e., 2b ≈ 0.4A−1 and ε2b(Ω = Ωp) = 12. Such characteristics can be observed in the real compounds listed in Table I. In WS, the separation of the Weyl nodes in momentum space can be so large 2b ≈ 0.5A−1 (Co3S2Se2) that the AHE dielectric tensor component ε2b ≈ 2.3/ω[eV ] even in the optical range (ω ∼ 2eV ) may be of the order of ε2b ∼ 1. While for the typical MD film (bismuth iron garnet), in the optical range ε2b = 0.003 [104] is by three orders of magnitude less than in some WS films. How- ever, while bismuth iron garnet retains the magnetization at room temperature, all the WSs listed in Table I can be used only at T < TC ∼ 150K. Nevertheless, WS such as compensated ferrimagnet Ti2MnAl [24], with a high Curie temperature TC > 650K and similar parameters as 7 (a) The dispersion of the Voigt dielectric function εV in WS at different Fermi levels EF = 150meV (solid line) and FIG. 5. E(cid:48) F = 100meV (dashed line); the WG regions where εV > 1 are shaded by color. (b) The slice of the Voigt dielectric function vs. frequency and Fermi level at εV (ω, EF) = 1; weakly damped WG modes regions are shaded by blue and the damping region is gray. (c) The dispersion of the nonreciprocal WG TM modes [forward p> (red) and backward p< (blue)] in WS film with d = 0.7 µm in the Voigt configuration at different Fermi levels EF = 150meV (solid lines) and E(cid:48) F = 100meV (dashed lines); dotted line denotes the dispersion of light in WS at changed Fermi level E(cid:48) F. Other parameters of WS are the same as for Fig. 1. listed in Table I for Eu2IrO7, may be the best candidate for room-temperature applications. ε2b <(cid:112)ε2 ε2b > (cid:112)ε2 2b (cid:14)ε2 > 1, i.e., at ε2 > 1 for the upper region By tuning the Fermi level in WS, one can shift in fre- quency the nonreciprocal WG regions [see Fig. 5(a)], but a number of limitations should be considered. First, the existence of the WG region assumes that εV = ε2 − ε2 2 − ε2 ≈ ε2 and at ε2 < 0 for the lower region 2 − ε2 ≈ ε2. Thus, in the lower region, the AHE response dominates ε2b > ε2 and in the upper one the diagonal component should be larger ε2b < ε2. This leads to the different influence of the Fermi level tuning on the upper and lower WG regions. By making the slice of the Voigt dielectric function at εV (ω, EF) = 1, we find that the upper WG region is much more sensitive to the Fermi level changes than the lower one [see Fig. 5(b)]. In particular, the upper weakly damped WG modes re- gion vanishes with the decrease of the Fermi level, while the lower one only slightly narrows. The damping region is located at ω > 2EF, where the interband Landau damping takes place [see imaginary part of Eq. (7)]. Nev- ertheless, without lowering the Fermi level too low one can change the dispersion of the nonreciprocal TM WG modes in the upper region as well [see Fig. 5(c)]. Besides, there is also an upper limit on the Fermi level: when it is high enough that the Fermi surfaces, enclosing the two Weyl nodes with opposite topological charges, merge, the magnitude of the AHE and corresponding WG modes nonreciprocity may dramatically change [58]. Thus, tun- ing the Fermi level in WS, one can vary the operation frequencies (which are near ω ∼ EF) of the predicted nonreciprocal waves in THz and mid-IR ranges. In par- ticular, in WSs with a low Fermi level EF ∼ 10meV such as Eu2IrO7 [19], only the lower weakly damped nonrecip- rocal WG modes region can exist [see Fig. 5(b)]; its fre- quencies lie in THz range, where ε2b ∼ 170 (see Table I) and the nonreciprocity can be very strong. However, as we discussed in Sec. III, in contrast to the upper WG re- gion, which is the manifestation of the dielectric response in WSs, the lower one lying below the plasma frequency may exist in any magnetoplasma system in the Voigt con- figuration. Nevertheless, in WSs with Fermi level around EF ∼ 100meV such as Co3S2Se2 [23], both of the WG regions can exist and will belong to mid-IR frequency range, where the nonreciprocity is moderate (ε2b ∼ 10), as at the model parameters used for Figs. 1-5. Notice that the nonlocal response in any materials may destroy the nonreciprocal effects [117], however, all our results for WSs were obtained at q (cid:28) kF, where the local response approximation [see Eq. (7)] works well. Compounds Y2IrO7 [18] Eu2IrO7 [19] Co3S2Sn2 [22] Co3S2Se2 [23] EF(meV) g 2b(A−1) ε2b ω0(THz) 10 60 110 24 12 12 0.37 170 2.4 0.47 0.5 36 21 14.5 26.6 TABLE I. The list of magnetic WSs with different Fermi lev- els EF, numbers of nondegenerated Weyl nodes g, separations of the Weyl nodes in momentum space 2b, corresponding di- electric tensor AHE components ε2b taken at ω0 = EF, and operation frequencies ω0. V. CONCLUSION In summary, we predict the existence of nonreciprocal WG modes in ferromagnetic WS films in the Voigt config- uration without an external magnetic field. The role of a magnetic field plays the AHE in WS, which, being purely intrinsic and universal in ideal WSs, depends only on the separation of the Weyl nodes in momentum space. The nonreciprocity value also depends on the optical contrast between the media surrounding a WS film, particularly, a metallic substrate leads to a significant increase of the nonreciprocity due to the high optical contrast with the medium above the film. We show that the nonrecipro- cal WG modes may exist in the two frequency regions: the lower one is below the WS plasma frequency and the upper one is above it. The lower WG region, where the negative refraction can be observed, is typical for any magnetoplasma system in the Voigt configuration, while the upper one is the manifestation of the dielec- tric response in WSs. We provide the AHE parameters of the real WS materials where a strong nonreciprocity can be observed even without a help of the surrounding media optical contrast. Such high values of the AHE in ferromagnetic WSs may be useful not only for the non- reciprocity but also for the gyrotropic effects. Moreover, tuning the Fermi level in WSs, one can vary the operation frequencies of the WG modes in THz and mid-IR ranges. We find that the upper WG region is much more sensi- tive to the Fermi level changes than the lower one. In particular, the upper weakly damped WG modes region vanishes with the decrease of the Fermi level, while the lower one only slightly narrows. So, to work with both WG regions, one should use WSs with rather high Fermi 8 levels. Thanks to the strong dielectric response caused by the gapless Weyl spectrum and the large Berry curvature coming from the entangled Bloch electronic bands with SOC, ferromagnetic WSs combine the best WG proper- ties of MDs or magnetic semiconductors with strong AHE in ferromagnets. Thus, ferromagnetic WSs allow one to realize giant tunable gyrotropic and nonreciprocity ef- fects for a propagating light, which paves the way to the design of compact, tunable, and effective nonreciprocal optical elements. ACKNOWLEDGMENTS The authors are grateful to A. A. Sokolik for useful dis- cussions. The work was supported by the Russian Foun- dation for Basic Research (17-02-01322, 17-02-01134, 18- 52-00002). Yu. E. L. thanks the Basic Research Pro- gram of the National Research University Higher School of Economics. [1] G. E. Volovik, JETP Lett. 46, 98 (1987). [2] S. Murakami, New. J. Phys. 9, 356 (2007). [3] G. B. Hal´asz and L. Balents, Phys. Rev. B 85, 035103 (2012). [4] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011). [5] P. Hosur and X. Qi, Compt. Rend. Phys. 14, 857 (2013). [6] H. Nielsen and M. Ninomiya, Phys.Lett. B 130, 389 (1983). [7] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [8] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). [9] Z. K. Liu, L. X. Yang, Y. Sun, T. Zhang, H. Peng, H. F. Yang, C. Chen, Y. Zhang, Y. F. Guo, D. Prabhakaran, M. Schmidt, Z. Hussain, S.-K. Mo, C. Felser, B. Yan, and Y. L. Chen, Nat. Mater. 15, 27 (2016). [10] B.-J. Yang and N. Nagaosa, Nat. Commun. 5, 4898 (2014). [11] Z. Wang, Y. Sun, X.-Q. Chen, C. Franchini, G. Xu, H. Weng, X. Dai, and Z. Fang, Phys. Rev. B 85, 195320 (2012). [16] 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). [17] Q. Li, D. E. Kharzeev, C. Zhang, Y. Huang, I. Pletikosic, A. V. Fedorov, R. D. Zhong, J. A. Schnee- loch, G. D. Gu, and T. Valla, Nat. Phys. 12, 550 (2016). and S. Y. [18] X. Wan, A. M. Turner, A. Vishwanath, Savrasov, Phys. Rev. B 83, 205101 (2011). [19] A. B. Sushkov, J. B. Hofmann, G. S. Jenkins, J. Ishikawa, S. Nakatsuji, S. Das Sarma, and H. D. Drew, Phys. Rev. B 92, 241108 (2015). [20] G. Xu, H. Weng, Z. Wang, X. Dai, and Z. Fang, Phys. Rev. Lett. 107, 186806 (2011). [21] Q. Wang, Y. Xu, R. Lou, Z. Liu, M. Li, Y. Huang, D. Shen, H. Weng, S. Wang, and H. Lei, Nat. Commun. 9, 3681 (2018). [22] E. Liu, Y. Sun, N. Kumar, L. Muechler, A. Sun, L. Jiao, S.-Y. Yang, D. Liu, A. Liang, Q. Xu, J. Kroder, V. Suss, H. Borrmann, C. Shekhar, Z. Wang, C. Xi, W. Wang, W. Schnelle, S. Wirth, Y. Chen, S. T. B. Goennenwein, and C. Felser, Nat. Physics 14, 1125 (2018). [12] Z. Wang, H. Weng, Q. Wu, X. Dai, and Z. Fang, Phys. [23] Q. Xu, E. Liu, W. Shi, L. Muechler, J. Gayles, C. Felser, Rev. B 88, 125427 (2013). [13] 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). [14] S. Borisenko, Q. Gibson, D. Evtushinsky, V. Zabolot- nyy, B. Buchner, and R. J. Cava, Phys. Rev. Lett. 113, 027603 (2014). [15] 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). and Y. Sun, Phys. Rev. B 97, 235416 (2018). [24] W. Shi, L. Muechler, K. Manna, Y. Zhang, K. Koepernik, R. Car, J. van den Brink, C. Felser, and Y. Sun, Phys. Rev. B 97, 060406 (2018). [25] S.-M. Huang, S.-Y. Xu, I. Belopolski, C.-C. Lee, G. Chang, B. Wang, N. Alidoust, G. Bian, M. Neu- pane, C. Zhang, S. Jia, A. Bansil, H. Lin, and M. Z. Hasan, Nat. Commun. 6, 7373 (2015). [26] H. Weng, C. Fang, Z. Fang, B. A. Bernevig, and X. Dai, Phys. Rev. X 5, 011029 (2015). [27] C. Shekhar, A. K. Nayak, Y. Sun, M. Schmidt, M. Nick- las, I. Leermakers, U. Zeitler, Y. Skourski, J. Wosnitza, Z. Liu, Y. Chen, W. Schnelle, H. Borrmann, Y. Grin, C. Felser, and B. Yan, Nat. Phys. 11, 645 (2015). [28] 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). [29] J. Behrends, A. G. Grushin, T. Ojanen, and J. H. Bar- darson, Phys. Rev. B 93, 075114 (2016). [30] 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). [31] S.-Y. Xu, N. Alidoust, I. Belopolski, Z. Yuan, G. Bian, T.-R. Chang, H. Zheng, V. N. Strocov, D. S. Sanchez, G. Chang, C. Zhang, D. Mou, Y. Wu, L. Huang, C.- C. Lee, S.-M. Huang, B. Wang, A. Bansil, H.-T. Jeng, T. Neupert, A. Kaminski, H. Lin, S. Jia, and M. Za- hid Hasan, Nat. Phys. 11, 748 (2015). [32] S.-Y. Xu, I. Belopolski, D. S. Sanchez, C. Zhang, G. Chang, C. Guo, G. Bian, Z. Yuan, H. Lu, T.-R. Chang, P. P. Shibayev, M. L. Prokopovych, N. Alidoust, H. Zheng, C.-C. Lee, S.-M. Huang, R. Sankar, F. Chou, C.-H. Hsu, H.-T. Jeng, A. Bansil, T. Neupert, V. N. Strocov, H. Lin, S. Jia, and M. Z. Hasan, Sci. Adv. 1, e1501092 (2015). [33] B. Xu, Y. M. Dai, L. X. Zhao, K. Wang, R. Yang, W. Zhang, J. Y. Liu, H. Xiao, G. F. Chen, A. J. Taylor, D. A. Yarotski, R. P. Prasankumar, and X. G. Qiu, Phys. Rev. B 93, 121110 (2016). 9 [44] Y. Sun, S.-C. Wu, and B. Yan, Phys. Rev. B 92, 115428 (2015). [45] T. Ojanen, Phys. Rev. B 87, 245112 (2013). [46] P. Hosur, Phys. Rev. B 86, 195102 (2012). [47] A. C. Potter, I. Kimchi, and A. Vishwanath, Nat. Com- mun. 5, (2014). [48] S. L. Adler, Phys. Rev. 177, 2426 (1969). [49] J. S. Bell and R. Jackiw, Il Nuovo Cimento A 60, 47 (1969). [50] V. Aji, Phys. Rev. B 85, 241101 (2012). [51] S. A. Parameswaran, T. Grover, D. A. Abanin, D. A. and A. Vishwanath, Phys. Rev. X 4, 031035 Pesin, (2014). [52] X. Huang, L. Zhao, Y. Long, P. Wang, D. Chen, Z. Yang, H. Liang, M. Xue, H. Weng, Z. Fang, X. Dai, and G. Chen, Phys. Rev. X 5, 031023 (2015). [53] C.-L. Zhang, S.-Y. Xu, I. Belopolski, Z. Yuan, Z. Lin, B. Tong, G. Bian, N. Alidoust, C.-C. Lee, S.-M. Huang, T.-R. Chang, G. Chang, C.-H. Hsu, H.-T. Jeng, M. Ne- upane, D. S. Sanchez, H. Zheng, J. Wang, H. Lin, C. Zhang, H.-Z. Lu, S.-Q. Shen, T. Neupert, M. Za- hid Hasan, and S. Jia, Nat. Commun. 7 (2016). [54] A. A. Burkov, J. Phys.: Condens. Matter 27, 113201 (2015). [55] J. Ma and D. A. Pesin, Phys. Rev. B 92, 235205 (2015). [56] P. Baireuther, J. A. Hutasoit, J. Tworzydo, and C. W. J. Beenakker, New. J. Phys. 18, 045009 (2016). [57] J. Zhou and H.-R. Chang, Phys. Rev. B 97, 075202 (2018). [58] A. A. Burkov, Phys. Rev. Lett. 113, 187202 (2014). [59] K.-Y. Yang, Y.-M. Lu, and Y. Ran, Phys. Rev. B 84, 075129 (2011). [34] H. Weng, X. Dai, and Z. Fang, J. Phys.: Condens. [60] A. A. Zyuzin and R. P. Tiwari, JETP Lett. 103, 717 Matter 28, 303001 (2016). (2016). [35] M. Z. Hasan, S.-Y. Xu, I. Belopolski, and S.-M. Huang, [61] A. A. Zyuzin and A. A. Burkov, Phys. Rev. B 86, 115133 Annu. Rev. Condens. Matter Phys. 8, 289 (2017). (2012). [36] N. P. Armitage, E. J. Mele, and A. Vishwanath, Rev. [62] M. M. Vazifeh and M. Franz, Phys. Rev. Lett. 111, Mod. Phys. 90, 015001 (2018). 027201 (2013). [37] A. A. Soluyanov, D. Gresch, Z. Wang, Q. Wu, M. Troyer, X. Dai, and B. A. Bernevig, Nature 527, 495 (2015). [38] C. Wang, Y. Zhang, J. Huang, S. Nie, G. Liu, A. Liang, Y. Zhang, B. Shen, J. Liu, C. Hu, Y. Ding, D. Liu, Y. Hu, S. He, L. Zhao, L. Yu, J. Hu, J. Wei, Z. Mao, Y. Shi, X. Jia, F. Zhang, S. Zhang, F. Yang, Z. Wang, Q. Peng, H. Weng, X. Dai, Z. Fang, Z. Xu, C. Chen, and X. J. Zhou, Phys. Rev. B 94, 241119 (2016). [63] A. Sekine, D. Culcer, and A. H. MacDonald, Phys. Rev. B 96, 235134 (2017). [64] P. E. C. Ashby and J. P. Carbotte, Phys. Rev. B 89, 245121 (2014). [65] P. Hosur and X.-L. Qi, Phys. Rev. B 91, 081106 (2015). [66] M. Kargarian, M. Randeria, and N. Trivedi, Sci. Rep. 5, 12683 (2015). [67] P. Goswami, G. Sharma, and S. Tewari, Phys. Rev. B 92, 161110 (2015). [39] Y. Sun, S.-C. Wu, M. N. Ali, C. Felser, and B. Yan, [68] K. Taguchi, T. Imaeda, M. Sato, and Y. Tanaka, Phys. Phys. Rev. B 92, 161107 (2015). Rev. B 93, 201202 (2016). [40] A. Tamai, Q. S. Wu, I. Cucchi, F. Y. Bruno, S. Ricc`o, T. K. Kim, M. Hoesch, C. Barreteau, E. Giannini, C. Besnard, A. A. Soluyanov, and F. Baumberger, Phys. Rev. X 6, 031021 (2016). [41] L. Huang, T. M. McCormick, M. Ochi, Z. Zhao, M.- T. Suzuki, R. Arita, Y. Wu, D. Mou, H. Cao, J. Yan, N. Trivedi, and A. Kaminski, Nat. Mater. 15, 1155 (2016). [69] C.-K. Chan, N. H. Lindner, G. Refael, and P. A. Lee, Phys. Rev. B 95, 041104 (2017). [70] E. Barnes, J. J. Heremans, and D. Minic, Phys. Rev. Lett. 117, 217204 (2016). [71] K. Halterman, M. Alidoust, and A. Zyuzin, Phys. Rev. B 98, 085109 (2018). [72] C.-X. Liu, P. Ye, and X.-L. Qi, Phys. Rev. B 87, 235306 (2013). [42] Y. Xu, F. Zhang, and C. Zhang, Phys. Rev. Lett. 115, [73] I. Panfilov, A. A. Burkov, and D. A. Pesin, Phys. Rev. 265304 (2015). [43] C.-C. Lee, S.-Y. Xu, S.-M. Huang, D. S. Sanchez, I. Be- lopolski, G. Chang, G. Bian, N. Alidoust, H. Zheng, M. Neupane, B. Wang, A. Bansil, M. Z. Hasan, and H. Lin, Phys. Rev. B 92, 235104 (2015). B 89, 245103 (2014). [74] J. Zhou, H.-R. Chang, and D. Xiao, Phys. Rev. B 91, 035114 (2015). [75] M. Lv and S. Zhang, Int. J. Mod. Phys. B 27, 1350177 (2013). 10 [76] S. Das Sarma, E. H. Hwang, and H. Min, Phys. Rev. 1997). B 91, 035201 (2015). [77] D. E. Kharzeev, R. D. Pisarski, and H.-U. Yee, Phys. [98] K. Petermann, Opt. Lett. 7, 623 (1982). [99] K. W. Chiu and J. J. Quinn, Il Nuovo Cimento B 10, 1 Rev. Lett. 115, 236402 (2015). (1972). [78] J. Hofmann and S. Das Sarma, Phys. Rev. B 91, 241108 [100] R. F. Wallis, J. J. Brion, E. Burstein, and A. Hartstein, (2015). Phys. Rev. B 9, 3424 (1974). [79] B. Rosenstein, H. C. Kao, and M. Lewkowicz, Phys. [101] M. S. Kushwaha and P. Halevi, Phys. Rev. B 36, 5960 Rev. B 95, 085148 (2017). (1987). [80] Y. Ferreiros and A. Cortijo, Phys. Rev. B 93, 195154 [102] S. Miyahara and S. Kobayashi, Jpn. J. Appl. Phys. 27, (2016). 2340 (1988). [81] J. Hofmann and S. Das Sarma, Phys. Rev. B 93, 241402 [103] S. T. Ivanov and N. I. Nikolaev, J. Phys. D: Appl. Phys. (2016). 32, 430 (1999). [82] J. C. W. Song and M. S. Rudner, Phys. Rev. B 96, 205443 (2017). [83] G. M. Andolina, F. M. D. Pellegrino, F. H. L. Koppens, and M. Polini, Phys. Rev. B 97, 125431 (2018). [84] E. V. Gorbar, V. A. Miransky, I. A. Shovkovy, and P. O. Sukhachov, Phys. Rev. B 95, 115202 (2017). [104] V. I. Belotelov, I. A. Akimov, M. Pohl, V. A. Ko- tov, S. Kasture, A. S. Vengurlekar, A. V. Gopal, D. R. Yakovlev, A. K. Zvezdin, and M. Bayer, Nat. Nanotech. 6, 370 (2011). [105] T. Dietl and H. Ohno, Rev. Mod. Phys. 86, 187 (2014). [106] T. Fukumura, H. Toyosaki, and Y. Yamada, Semicond. [85] Z. Long, Y. Wang, M. Erukhimova, M. Tokman, and Sci. Technol. 20, S103 (2005). A. Belyanin, Phys. Rev. Lett. 120, 037403 (2018). [107] O. V. Kotov and Y. E. Lozovik, Phys. Rev. B 93, 235417 [86] J. Shibata, A. Takeuchi, H. Kohno, and G. Tatara, J. (2016). Appl. Phys. 123, 063902 (2018). [87] S. Zhong, J. E. Moore, and I. Souza, Phys. Rev. Lett. 116, 077201 (2016). [88] C.-K. Chan, P. A. Lee, K. S. Burch, J. H. Han, and Y. Ran, Phys. Rev. Lett. 116, 026805 (2016). [89] A. A. Zyuzin and V. A. Zyuzin, Phys. Rev. B 92, 115310 (2015). [90] F. M. D. Pellegrino, M. I. Katsnelson, and M. Polini, [108] F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). [109] H. Huang, Y. Fan, B.-I. Wu, and J. A. Kong, Prog. Electromagn. Res. 85, 367 (2008). [110] S. Tretyakov, I. Nefedov, A. Sihvola, S. Maslovski, and C. Simovski, J. Electromagn. Waves Appl. 17, 695 (2003). [111] J. B. Pendry, Science 306, 1353 (2004). [112] T. G. Mackay, Microw. Opt. Technol. Lett. 45, 120 Phys. Rev. B 92, 201407 (2015). (2005). [91] J. C. W. Song and M. S. Rudner, Proc. Natl. Acad. Sci. [113] V. M. Agranovich, Y. N. Gartstein, and A. A. Zakhi- 113, 4658 (2016). [92] P. K. Tien, Rev. Mod. Phys. 49, 361 (1977). [93] A. M. Prokhorov, G. A. Smolenskii, and A. N. Ageev, Phys. Usp. 27, 339 (1984). [94] R. E. Camley, Surf. Sci. Rep. 7, 103 (1987). [95] H. Dotsch, N. Bahlmann, O. Zhuromskyy, M. Hammer, L. Wilkens, R. Gerhardt, P. Hertel, and A. F. Popkov, J. Opt. Soc. Am. B 22, 240 (2005). [96] N. V. Kravtsov and N. N. Kravtsov, Quantum Electron. 29, 378 (1999). [97] A. K. Zvezdin and V. A. Kotov, Modern Magnetooptics and Magnetooptical Materials (CRC Press, New York, dov, Phys. Rev. B 73, 045114 (2006). [114] M. S. Ukhtary, A. R. T. Nugraha, and R. Saito, J. Phys. Soc. Jpn. 86, 104703 (2017). [115] T. Hayata, arXiv: 1801.10272 (2018). [116] P. R. West, S. Ishii, G. V. Naik, N. K. Emani, V. M. Shalaev, and A. Boltasseva, Laser & Photonics Rev. 4, 795 (2010). [117] S. Buddhiraju, Y. Shi, A. Song, C. Wojcik, M. Minkov, and S. Fan, arXiv: I. A. D. Williamson, A. Dutt, 1809.05100 (2018). 

1808.01446

1

1808

Fully guided electrically-controlled exciton polaritons

We demonstrate two types of waveguide structures which optically confine exciton- polaritons in two dimensions and act as polaritonic channels. We show a strong optical confinement in an etched rectangular waveguide, that significantly increases the propa- gation distance of the polaritons and allow to direct them in curved trajectories. Also, we show low-loss optical guiding over a record-high of hundreds of microns which is com- bined seamlessly with electrical control of the polaritons, in a strip waveguide formed by electrically conductive and optically transparent strips deposited on top of a planar waveguide. Both structures are scalable and easy to fabricate and offer new possibilities for designing complex polaritonic devices.

cond-mat.mes-hall

cond-mat

Fully guided electrically-controlled exciton polaritons Dror Liran,†,§ Itamar Rosenberg,†,§ Kenneth West,‡ Loren Pfeiffer,‡ and Ronen Rapaport∗,† †Racah Institute of Physics, The Hebrew University of Jerusalem, Jerusalem 91904, Israel ‡Department of Electrical Engineering, Princeton University, Princeton, New Jersey ¶The Applied Physics Department, The Hebrew University of Jerusalem, Jerusalem 91904, 08544,USA Israel §These two authors contributed equally E-mail: ronenr@phys.huji.ac.il Abstract We demonstrate two types of waveguide structures which optically confine exciton- polaritons in two dimensions and act as polaritonic channels. We show a strong optical confinement in an etched rectangular waveguide, that significantly increases the propa- gation distance of the polaritons and allow to direct them in curved trajectories. Also, we show low-loss optical guiding over a record-high of hundreds of microns which is com- bined seamlessly with electrical control of the polaritons, in a strip waveguide formed by electrically conductive and optically transparent strips deposited on top of a planar waveguide. Both structures are scalable and easy to fabricate and offer new possibilities for designing complex polaritonic devices. 1 Introduction Exciton-polaritons are quantum superpositions of light and matter resulting from the strong coupling between confined optical modes and confined electronic excitations (excitons). They have very small effective masses ∼ 10−5m0 and very large propagation velocities, on the one hand, while maintaining the ability to interact with each other, as well with external elec- tric or magnetic fields on the other hand. This conjunction of properties induces effectively strong nonlinearities to the medium leading to "interacting photons". As such, polariton quasi-particles hold promise for realizations of light-based quantum devices with new func- tionalities. Typically, semiconductor microcavities (MC) are used as means for optical confinement of the photonic part of the polaritons. In these structures, several quantum wells (QWs) are positioned at the anti-nodes of the optical mode, which is confined between two distributed Bragg reflectors (DBRs). Recently, several proofs-of-concept of non-linear photonic devices based on MC-polaritons have been demostrated.1 -- 11 MC-based polariton devices have how- ever some drawbacks when considering the feasibility of large-scale polaritonic-based circuits. Firstly, DBRs are monolithically grown thick multi-layers, making the fabrication process of even the simplest lateral structures complex and prone to surface defects and imperfections. Secondly, even very thick high-quality DBRs have a finite reflectivity, causing a constant photon leak as polaritons propagate inside the sample. This limits polariton propagation lengths even with high quality mirrors12 to distances typically below 200µm.13,14 Together with the inherent tendency of MC-polaritons to relax to their lowest energy having a zero in-plane momentum (and thus low in-plane propagation velocities), these short propagation distances limits the ability to create extended optical circuits based on MC-polaritons. Several years ago, it was demonstrated that strong coupling between excitons in QWs and propagating optical modes of a planar slab waveguide (WG) leads to the formation of propagating ('flying') WG-polaritons.15 Since the optical modes in a WG are confined by total internal reflection, the optical losses are significantly reduced, and the WG-polaritons 2 display very high propagation velocities and long propagation distances,15,16 and recently WG-polaritons operating up to room temperature have been reported.17 Significant non- linearity of waveguide polaritons has also been reported, leading to the formation of dark solitons in a WG-polariton fluid.18,19 Since the WG-polaritons can be made in thin structures (since no thick cladding DBRs are required), it enables easy local access to both the optical modes, the photonic part (e.g. by patterning the WG surface), as well as to the matter, the excitonic part. In this regard, we showed that surface electrodes can be used to easily apply voltages across the QW planes, resulting in an electrical polarization of the excitons, that acquire an electrical dipole moment. This allows an electrical control of the polari- ton signal16 and results in a significant electrically controlled enhancement of the mutual interactions between WG-polaritons.20,21 Therefore, flying WG-polaritons can be excellent candidates for polariton-based circuitry, due to the simple access to local control of both the light and the matter parts of the polaritons, where a complex spatial control can be potentially designed and achieved by methods as simple as single-layer shallow etching or simple deposition processes. Until now, all experiments in WG-polaritons were done in a slab-WG geometry, with no lateral optical confinement, where polaritons are confined only in one dimension. This simple geometry limits the ability to construct complex lightwave structures, as well as the ability to maintain control of the direction and density of the WG- polaritons, since polaritons in a slab geometry display a significant lateral spreading as they propagate. In this paper we demonstrate two methods to confine polaritons laterally in channel-WG ge- ometries. In the first part of this paper we show that polaritons can be confined in an etched rectangular WG (RWG)22 where polaritons propagate to distances much larger than those possible in a slab WG, and where they can also propagate along a bent channel. We then show that the dielectric mismatch between a slab-WG and a thin dielectric strip is enough to create a low-loss strip-loaded polariton WG (SWG) leading to a very large increase of the propagation length. We demonstrate that a thin strip of ITO can be used to both optically 3 Figure 1: (a) Slab-waveguide experiments: a non-resonant excitation beam is focused on a planar waveguide at a distance of ∆x away from the grating coupler to generate a cloud of polaritons that propagate isotropically. The fraction of these polaritons that propagate toward the grating coupler, illustrated by the θ red cone, are coupled out and collected by the collection objective. (b) Illustration of an RWG and a bent-RWG. The dashed dotted lines depict the line of the non-resonant excitation with varying ∆x. (c) An SEM image of a section of the bent-RWG, with the metallic grating at the left part of the channel. (d) PL measured dispersion of the planar WG polaritons emitted from the grating coupler. The bare exciton and photons modes are marked by dashed lines and the fitted LP branch is marked by a solid line. guide polaritons and electrically control over the excitonic part of the optically confined polaritons by using it as an electrode which can electrically polarize the SWG-polaritons. Results and discussion A rectangular polariton waveguide In this section we compare between the propagation lengths of polaritons in three different WG geometries, a slab WG, a 10 µm-wide straight RWG and a 10 µm-wide bent-RWG with a radius of curvature of 100 µm, all illustrated in Fig.1(a,b). The sample, and the planar WG used in this work is identical to the one used by Rosenberg et. al.16 and incor- porates twelve 20nm wide GaAs/Al0.4Ga0.6As QWs in its core. Both RWGs were fabricated 4 by etching two parallel ∼ 600nm-deep trenches resulting in a rectangular WG channel. A comparison between the cross-sections of the slab-WG and the RWG is shown in Fig. 2(a,b). On each RWG a 40 µm long and 10 µm wide metallic grating-coupler were deposited (pe- riod=240 nm, duty cycle=0.5) to allow out-coupling of the propagating polaritonic signal.16 The method for measuring the propagation lengths of the polaritons is the following: a non-resonant excitation beam is focused at a specific distance ∆x from the grating-coupler, generating a reservoir of uncoupled excitons at the excitation spot. A large fraction of these excitons relaxes and accumulates at the bottleneck of the lower polariton (LP) dispersion and propagates inside the WG with a wave vector β. The fraction of these polaritons which propagates toward the grating-coupler is coupled out to be collected and angularly resolved by the acquisition setup. In Fig .1(d) We show the spectrally and angularly resolved pho- toluminescence (PL) emitted from a slab WG through the grating coupler. The intensity of the PL is determined by the polariton density reaching the grating from excitation spot after traveling a distance of ∆x. In Fig.2(c) we show examples of the PL dispersions measured when exciting at three locations along each of the WGs. Here the different rows correspond to a slab-WG, a straight-RWG and a bent-RWG respectively, and the excitation distance ∆x increases from left to right. Several things can be learned from Fig.2(c). In all these measurements, the dispersion of the LP can be clearly seen and the total intensity decreases when increasing ∆x. The variation in the emitted intensity as a function of ∆x, fitted to exponential decay for each of the RWG geometries, I(∆x) = I0 exp(−∆x/L) is shown in It is apparent that the propagation length in the straight-RWG is much larger Fig.2(g). than that in a slab-WG. This can be explained when considering geometrical losses due to spatial spreading in a slab-WG due to the lack of lateral confinement: non-resonantly ex- cited polaritons will propagate in all directions away from their excitation spot. We only collect PL from polaritons that passes under the grating-coupler, which propagate within an angle of θ = 2 tan−1( w/2 ∆x ), where w is the width of the grating-coupler as is illustrated in Fig.1(a). Obviously, as we increase ∆x, θ reduces and the measured intensity decreases. 5 Figure 2: A comparison between the cross-sections of a slab-WG (a) and a RWG (b). (c) PL dispersion measurements: The different rows correspond to a slab-WG, a straight-RWG and a bent-RWG correspondingly. ∆x increases from left to right in each row. For the bent-RWG insets are added to illustrate the excitation position (red dot). (d-f) Calculated profiles of the fundamental WG modes in the slab-WG, the RWG and the bent-RWG respectively. The intensity in each panel is normalized to the intensity from the shortest ∆x. (g) The variation in the emitted intensity as a function of ∆x. The straight and bent RWG channels are fitted to exponential decay, a curve of a × tan−1( w 2x) is plotted with w = 10µm as a guideline for the slab-WG measurements. Good agreement between the above geometrical dependence and the experiment is seen in red in Fig.2(g). Due to the lateral confinement in the case of RWG, the polaritons are guided towards the grating-coupler and are subjected only to physical losses which affect the polari- ton population. This difference can be understood when comparing the calculated optical modes cross-sections in each geometry. The fundamental optical modes for each of the WG geometries, calculated using a commercial finite difference eigenmode solver (Lumerical), are presented In Fig.2(d-f). It can be seen that while the mode of the slab-WG is confined only along the growth direction, the modes of the RWGs are confined in two dimensions. Particularly, the calculations show that a mode-confinement is expected also in a bent-RWG geometry which suggests that polaritons can be guided along curved trajectories, a feature which is demonstrated experimentally in the bottom row of Fig.2(c). Due to the lateral confinement of the RWG and the Bent-RWG we see an increased propagation length L as 6 shown in Fig.2(g). We measured L = 179 ± 60µm in the straight-RWG and L = 83 ± 30µm in the bent-RWG. Most of the loss of the RWG originates from scattering of polaritons from the rough surface of the side walls of the RWG. These scatterings are especially dominant in the bent-RWG where the mode is squeezed to the side and has a higher overlap with the rough wall of the WG (see Fig.2(f)) which explains the relatively small value of L of the bent-RWG. The side walls of both RWG channels were created using a wet-etch process. We note that optimization of the process and using dry-etch techniques will significantly reduce the roughness and should result in significantly larger propagation lengths.. An electrically-active striploaded polariton waveguide Next we turn to present an electrically active striploaded-WG (SWG), where lateral optical confinement is achieved together with an electrical control over the QW-excitons composing the polaritons. The SWG was fabricated by laying a 50 nm thick, 20 µm wide strip of ITO on top of the planar WG, as is shown in In Fig.3(a,b). The ITO strip introduces a higher effective refractive index than its surrounding, leading to a laterally confined WG mode, as is shown in the numerical calculation presented in Fig.3(c). The measured propagation decay of the polariton intensity along the SWG is presented in Fig.3(d) where it is compared to a calculation of the expected geometrical losses due to spatial spreading in slab-WG with a 20 µm wide grating-coupler. The fact that the propagation decay in the SWG is much slower than the calculated spatial spreading in slab-WG, confirms that the polaritons are indeed laterally confined by the SWG. Furthermore, we find a propagation length L = 610± 50µm, longer than that of the RWG. This is attributed to the lack of etching, which significantly reduces roughness-induced optical losses. As far as we know, such long polariton propagation lengths, approaching a millimeter, have not been achieved in any other geometry. Finally, we want to emphasize the added functionality that is easily introduced in the ITO-based SWG geometry. Here the conductive ITO strip also forms a transparent top electrode through which voltage can be applied across the sample. In Fig.3(e) we plot three 7 Figure 3: (a) Illustration of an electrically biased SWG. (b,c) The cross section, and the calculated profile of the fundamental mode of the SWG. (d) The variation of the measured emitted intensity as a function of ∆x fitted to a decaying exponent (blue). The red line shows the calculated expected intensity in a slab-WG with only geometrical losses due to unconfined lateral spreading. (e) PL dispersions measured while inducing different values of applied electric field F . The Stark red-shift of the polariton signal is a signature of the formation of dipolaritons.16,20 8 PL measurements measured when exciting at a distance of 187µm from the grating-coupler while applying different values of voltage across the sample with respect to the n+ doped substrate. The effect of the Stark red-shift, induced by the electric field, on the dispersion can be clearly seen. In addition, the ability to measure the WG-polariton signal when exciting this far away from the grating-coupler, indicates that the effect of the electric field on the confinement is negligible. This dual use of the ITO strip allows therefore to form fully guided modes of electrically polarized WG-polaritons, with electrical control over the polariton energy, dispersion, and interactions as has been recently demonstrated.16,20 Summary and conclusions We demonstrated two types of WG geometries which can be used to laterally confine and guide polaritons. This lateral confinement maintains well-defined polariton modes over long propagation distances, prevents spreading and density reduction, and provides the ability to guide polaritons in curved trajectories. These guiding capabilities are a necessary step to- wards the demonstration of large-scale on-chip polaritonic optical circuits. The demonstrated ability, to control simultaneously both the photonic part of the flying polaritons through op- tical confinement and the excitonic part using electrical fields, is a milestone towards a full control of polaritons. Since electrically polarized polaritons exhibit stronger nonlinearities than unpolarized polaritons,20,21 combined with the ability of strong optical confinement, the enhanced polariton interactions may allow observation of polariton-polariton interactions on the quantum level, which can open up routes towards polaritonic-based quantum gates. Acknowledgement The authors thank to financial support from the U.S. Department of Energy: Office of Basic Energy Sciences - Division of Materials Sciences and Engineering, from the United State - Israel Binational Science Foundation (BSF grant 2016112), and from the Israeli Science 9 Foundation (grant No. 1319/12). The work at Princeton University was funded by the Gordon and Betty Moore Foundation through the EPiQS initiative Grant GBMF4420, and by the National Science Foundation MRSEC Grant DMR-1420541. References (1) Amo, A.; Liew, T. C. H.; Adrados, C.; Houdré, R.; Giacobino, E.; Kavokin, A. V.; Bramati, A. ExcitonâĂŞpolariton spin switches. Nat Photon 2010, 4, 361 -- 366. (2) Ferrier, L.; Wertz, E.; Johne, R.; Solnyshkov, D. D.; Senellart, P.; Sagnes, I.; Lemaître, A.; Malpuech, G.; Bloch, J. Interactions in Confined Polariton Condensates. Physical Review Letters 2011, 106, 126401. (3) Cristofolini, P.; Dreismann, A.; Christmann, G.; Franchetti, G.; Berloff, N. G.; Tsot- sis, P.; Hatzopoulos, Z.; Savvidis, P. G.; Baumberg, J. J. Optical Superfluid Phase Transitions and Trapping of Polariton Condensates. Physical Review Letters 2013, 110, 186403. (4) Gao, T.; Eldridge, P. S.; Liew, T. C. H.; Tsintzos, S. I.; Stavrinidis, G.; Deligeorgis, G.; Hatzopoulos, Z.; Savvidis, P. G. Polariton condensate transistor switch. Physical Review B 2012, 85, 235102. (5) Ballarini, D.; De Giorgi, M.; Cancellieri, E.; Houdré, R.; Giacobino, E.; Cingolani, R.; Bramati, A.; Gigli, G.; Sanvitto, D. All-optical polariton transistor. Nature Communi- cations 2013, 4, 1778. (6) Sturm, C.; Tanese, D.; Nguyen, H. S.; Flayac, H.; Galopin, E.; Lemaître, A.; Sagnes, I.; Solnyshkov, D.; Amo, A.; Malpuech, G.; Bloch, J. All-optical phase modulation in a cavity-polariton MachâĂŞZehnder interferometer. Nature Communications 2014, 5 . (7) Nguyen, H. S.; Vishnevsky, D.; Sturm, C.; Tanese, D.; Solnyshkov, D.; Galopin, E.; 10 Lemaître, A.; Sagnes, I.; Amo, A.; Malpuech, G.; Bloch, J. Realization of a Double- Barrier Resonant Tunneling Diode for Cavity Polaritons. Physical Review Letters 2013, 110, 236601. (8) Delteil, A.; Fink, T.; Schade, A.; Höfling, S.; Schneider, C.; Imamoğlu, A. Quantum correlations of confined exciton-polaritons. arXiv:1805.04020 [cond-mat] 2018, (9) Sanvitto, D.; Kéna-Cohen, S. The road towards polaritonic devices. Nature Materials 2016, 15, 1061 -- 1073. (10) Liew, T. C. H.; Kavokin, A. V.; Shelykh, I. A. Optical Circuits Based on Polariton Neurons in Semiconductor Microcavities. Physical Review Letters 2008, 101, 016402. (11) Menon, V. M.; Deych, L. I.; Lisyansky, A. A. Nonlinear optics: Towards polaritonic logic circuits. Nat Photon 2010, 4, 345 -- 346. (12) Nelsen, B.; Liu, G.; Steger, M.; Snoke, D. W.; Balili, R.; West, K.; Pfeiffer, L. Dissi- pationless Flow and Sharp Threshold of a Polariton Condensate with Long Lifetime. Physical Review X 2013, 3, 041015. (13) Wertz, E.; Ferrier, L.; Solnyshkov, D. D.; Johne, R.; Sanvitto, D.; Lemaître, A.; Sagnes, I.; Grousson, R.; Kavokin, A. V.; Senellart, P.; Malpuech, G.; Bloch, J. Spon- taneous formation and optical manipulation of extended polariton condensates. Nature Physics 2010, 6, 860 -- 864. (14) Liew, T. C. H.; Kavokin, A. V.; Ostatnický, T.; Kaliteevski, M.; Shelykh, I. A.; Abram, R. A. Exciton-polariton integrated circuits. Physical Review B 2010, 82, 033302. (15) Walker, P. M.; Tinkler, L.; Durska, M.; Whittaker, D. M.; Luxmoore, I. J.; Royall, B.; Krizhanovskii, D. N.; Skolnick, M. S.; Farrer, I.; Ritchie, D. A. Exciton polaritons in semiconductor waveguides. Applied Physics Letters 2013, 102, 012109. 11 (16) Rosenberg, I.; Mazuz-Harpaz, Y.; Rapaport, R.; West, K.; Pfeiffer, L. Electrically con- trolled mutual interactions of flying waveguide dipolaritons. Physical Review B 2016, 93, 195151. (17) Ciers, J.; Roch, J.; Carlin, J.-F.; Jacopin, G.; ButtÃľ, R.; Grandjean, N. Propagating Polaritons in III-Nitride Slab Waveguides. Physical Review Applied 2017, 7, 034019. (18) Walker, P.; Tinkler, L.; Royall, B.; Skryabin, D.; Farrer, I.; Ritchie, D.; Skolnick, M.; Krizhanovskii, D. Dark Solitons in High Velocity Waveguide Polariton Fluids. Physical Review Letters 2017, 119, 097403. (19) Walker, P. M.; Tinkler, L.; Skryabin, D. V.; Yulin, A.; Royall, B.; Farrer, I.; Ritchie, D. A.; Skolnick, M. S.; Krizhanovskii, D. N. Ultra-low-power hybrid light- matter solitons. Nature Communications 2015, 6, 8317. (20) Rosenberg, I.; Liran, D.; Mazuz-Harpaz, Y.; West, K.; Pfeiffer, L.; Rapaport, R. Strongly interacting dipolar-polaritons. arXiv:1802.01123 [cond-mat] 2018, arXiv: 1802.01123. (21) Togan, E.; Lim, H.-T.; Faelt, S.; Wegscheider, W.; Imamoglu, A. Strong interactions between dipolar polaritons. arXiv:1804.04975 [cond-mat] 2018, arXiv:1804.04975. (22) Hunsperger, R. Integrated Optics: Theory and Technology, 6th ed.; Springer-Verlag: New York, 2009. 12 

1201.0920

1

1201

Pseudospin dynamics in multimode polaritonic Josephson junctions

We analyzed multimode Josephson junctions with exciton-polaritons (polaritonic Josephson junctions) when several coupling mechanisms of fundamental and excited states are present. The applied method is based on Keldysh-Green function formalism and takes into account polariton pseudospin. We found that mean value of circular polarization degree in intrinsic Josephson oscillations and microscopic quantum self-trapping follow an oscillator behavior whose renormalizes due to intermode interactions. The effect of an additional transfer of particles over junction barrier occurring in multimode approximation in combination with common Josephson tunneling is discussed in regime of dynamical separation of two polarizations.

cond-mat.mes-hall

cond-mat

Pseudospin dynamics in multimode polaritonic Josephson junctions 1G. Pavlovic,2G. Malpuech,3,4I.A. Shelykh 1International Institute of Physics, Av. Odilon Gomes de Lima, 1722, CEP 59078-400 Capim Macio Natal, RN, Brazil 2LASMEA, UMR CNRS-Universit´e Blaise Pascal 6602, 24 Avenue des Landais, 63177 Aubi`ere Cedex France 3Physics Department, University of Iceland, Dunhaga-3, IS-107, Reykjavik, Iceland 3 A.F. Ioffe Physico-Technical Institute of RAS, 194021 St. Petesburg, Russia E-mail: shelykh@raunvis.hi.is Abstract. We analyzed multimode Josephson junctions with exciton-polaritons (polaritonic Josephson junctions) when several coupling mechanisms of fundamental and excited states are present. The applied method is based on Keldysh-Green function formalism and takes into account polariton pseudospin. We found that mean value of circular polarization degree in intrinsic Josephson oscillations and microscopic quantum self-trapping follow an oscillator behavior whose renormalizes due to inter- mode interactions. The effect of an additional transfer of particles over junction barrier occurring in multimode approximation in combination with common Josephson tunneling is discussed in regime of dynamical separation of two polarizations. PACS numbers: 71.36.+c,71.35.Lk,03.75.Mn 1. Introduction After theoretical prediction [1] and experimental detection [2] of Josephson tunneling between two superconductors separated by a thick insulator under application of an external voltage, analogous quantum oscillations were found also in a superfluid system: two vessels of superfluid helium connected by a nanoscale aperture [3]. Similar type of dynamics was observed for two weakly coupled atomic Bose-Einstein condensates (BECs) created in a double-trap potential. Differently from superconductors and superfluid helium in Josephson effect for atomic BECs inter-particle interactions play essential role [4]. An an-harmonic behavior occurs in these systems additionally to common Josephson oscillations. Under certain initial conditions BECs localize in one of the traps formed by external potential due to suppression of the tunneling current [5]. The phenomenon is known as macroscopic quantum self-trapping (MQST). It is a consequence of domination of nonlinearities, e.g. interactions, over Josephson coupling in the system. In some other range of the initial conditions delocalized phase is formed with unsuppressed oscillations between the traps. In general, such a system - interacting BECs in a Josephson junction (JJ) is 2 described by well-known Hamiltonian H = H0(z(0), θ(0)) = Λ z(t)2 2 − cos θ(t)q1 − z(t)2 (1) written in terms of population imbalance z(t) = (N1(t) − N1(t))/(N1(t) + N1(t)), where N1 and N2 are populations in trap one and trap two, and phase difference θ(t) between BECs. Two-mode approximation including only lowest energy states of symmetric double-trap is used in Hamiltonian (1) [5] and contains single parameter Λ = U0NT /2J. In the last expression J figures as Josephson coupling constant which is equal to the difference in energies of symmetric and anti-symmetric states of the double- trap potential is less then zero, as the symmetric state is usually lower in energy then the anti-symmetric one. NT is total population of particles in both traps which interact with energy U0. Inspecting the phase-space diagram (z,θ) of Hamiltonian (1) one can observe two distinct regions separated by separatrix line H = 1 ( full/red contour in Fig. 1). Fixing the value of H in (1) by choosing initial conditions z(0) and θ(0), delocalized regime will establish if H < 1. It is characterized by the population imbalance being zero in average (< z(t) >= 0) for the system's cyclic motion on closed orbits (see Fig. 1). Approaching the separatrix from the inner side the orbits starts to deviate from regular circles as nonlinearities start to be important. Outside the separatrix where H > 1 the population imbalance evolves along open lines with small oscillatins around constant mean value < z(t) >= const indicating the transition to MQST effect. Besides cold atoms there is an another kind of particles which can undergo a transition into collective state with properties similar to Bose- Einstein condensate (BEC). These are exciton-polaritons whose condensation has been repported several years ago in in CdTe microcavities [6, 7] at about 20 K and later on in GaN-based microcavities even at room-temperatures [8, 9]. The possibility to obtain BEC at such conditions is extremely interesting from fundamantal point of view. Besides, it opens a way for design of novel optoelectronic components, such as polariton lasers [11] and various devices based on polariton superfluidity [10] Peculiar properties of exciton-polaritons are consequences of Polaritons are two-dimensional quasi-particles appearing due to strong coupling of excitons in semiconductor quantum-well(s) and photons confined within a microcavity structure. their hybrid, half-light half-mater nature coming from photonic and excitonic components respectively. In particular, ultra-small effective mass of polaritons, which is typically four or five orders of magnitude less then the free electron mass, makes quantum collective phenomena very pronounced in polaritonic systems and leads extremely high critical temperatures of the polariton BEC transition. For the purposes of our further discussion, another property of polaritons is very of a polariton state on z = ±1 z = ±1) and photons with right or left important. That is projection of total angular momentum J pol the structure's growth axis (chosen as z direction) [12]. It can take two values: J pol corresponding to mixing of bright excitons (J exc z 3 Figure 1. Hamiltonian (1) in the phase space (z,θ). Contour lines for H=0,1,2 and 3 are labeled. Separatrix is shown with full/red line circular polarizations. Dark excitons with J exc z = ±2 are uncoupled from cavity mode due to optical selection rules and thus do not contribute to the formation of a polariton doublet. From the point of view os spin polariton is thus analogical to electron, as both of them are two-level systems. Consequently one can apply pseudospin formalism for description of spin dynamics of the polaritons. It is convenient to represent a pseudo spin state of polariton system by a point on Poincar´e sphere (also known as Bloch sphere). Importantly, polariton pseudospin unambiguously defines the polarization state of photoemission from the cavity. The states lying at the poles of Poincar´e sphere will correspond to circular polarization of the polaritons, the states on equator- to linear polarizations and all other states to elliptical polarizations. Being bosons, exciton-polaritons should manifest phenomena related to Josephson effect. Polaritonic Josephson junctions (PJJ) were considered theoretically in Refs. [17, 18, 19] and experimentally in Ref.[22]. In these articles polariton pseudospin was neglected, and and Josephson coupling occurred between the polaritons located in spatially separated traps. The introduction of polarization degree of freedom makes Josephson dynamics far more rich. Indeed, in addition to the coupling between states localized in different traps, two polarization states in single asymmetric trap can be coupled as well by mechanism analogical to TE- TM splitting [21]. Josephson-type 4 oscillations can occur between different polarizations, the phenomenon which was called intristic Josephson effect [20]. The interplay between intrinsic and extrinsic Josephson dynamics can lead to dynamic separation of different polarizations in the real space and other intriguing phenomena [20]. The role of the polarization coupling in PJJ was studied for several configurations [23, 24, 25]. In most of them so called two-mode approximation was applied, in which only two lowest levels, one in each of two traps, were considered. An exception is the reference [25] where several modes were included in analysis in order damping of Josephson oscillations due to the interactions of the polaritons with acoustic phonons. However, even in this case the separation between two fundamental modes and excited ones was considered to be much greater then characteristic values of blueshifts provided by polariton- polariton interactions, and transitions to the excited modes were only possible due to absorbtion of phonons. In this paper we analyze multimode PJJ with confining potential created in such a way that interactions between the lowest and excited levels are not negligible. This kind of coupling have been already considered for Josephson Junctions based on atomic BECs [13]. Our goal is to extend the analysis of Ref.[13] for the case of cavity polaritons accounting the polarization degree of freedom, and clarify the effects of the coupling between fundamental and excited modes on nonlinear polarization dynamics of PJJs. 2. Model In Ref.[13] Keldysh-Green functions technique was employed to study dynamics for atomic Josephson Junction with multimode structure. Here we shall follow similar approach for description of PJJs accounting for the spin degree of freedom. Because of spinor nature of polariton condensate in addition to normal Josephson coupling we have to consider also and the coupling of two polarization components, normally irrelevant for atomic JJs. We start by expanding exciton-polariton field operators Ψσ±(r, t) over complete set of eigenstates {f1, f2, ...} of a double-well potential Ψσ±(r, t) = f1(r)a1σ±(t) + f2(r)a2σ±(t) + fn(r)bnσ±(t) (2) ∞ Xn=3 where a1σ± and a2σ± denote annihilation operators for two lowest modes of PJJ localized in the traps one and two, and bnσ± stands for the anihilation operator on excited level n taking values. σ± denotes circular polarization of the state. The system we study is described by the Hamiltonian which can be represented as a sum of three terms: H = Hσ+ + Hσ− + Hσ+σ−. (3) The terms conserving z- projection of pseudospin σ+ and σ− denoted by Hσ+ and Hσ− are given by following expressions Hσ± = Z d2r Ψ† σ±(r, t)(− ¯h2 2m ∆ + Vext(r, t)) Ψσ±(r, t) (4) ′ , t)V (r − r ′ ) Ψσ±(r ′ , t) Ψσ±(r, t). 5 ′ Ψ+ σ±(r, t) Ψ† σ±(r + 1 2 Z d2rZ d2r The effective mass of polaroton is denoted by m and Vext(r, t) stands for external double- well potential. V (r, r′) = gδ(r − r′) is contact interaction described by delta function and interaction constant can be estimated as g ≈ EBa2 B with EB and aB being being exciton binding energy and Bohr radius of the exciton respectively [14]. The last term in the Eq. (3) accounts for the Josephson- type coupling of the particles having opposite pseudospin projections. It can be viewed as a consequence of one presence of effective in-plane magnetic field Ω(r) arising from the asymmetry of the structure [21] and acting on polaritons spins and can be represented as Hσ+σ− = Z d2r Ψ† σ±(r, t)Ω(r) Ψσ∓(r, t). (5) After subtitution of the expensions (2) into the Hamiltonians (4) and (5) the starting, i.e. full Hamiltonian (3) can be recast as H = H 0 + H exc + H int. (6) where the first term describes the dynamics of the four fundamental modes (accounting to polarization degreee of freedom), the second term describes the dynamics of the delocalized excited modes and the last term corresponds to the coupling between fundamantal and excited modes. These terms read: H 0 = E0Xi;σ H exc = Xn,m;σ a† iσaiσ + J0Xσ (En + Unm < b† (a† 1σa2σ + a† 2σa1σ) + bmσ >)b† nσ mσ U0 2 Xi;σ b† bnσ + ΩnXn;σ iσa† a† iσaiσaiσ + Ω0Xi;σ bn−σ nσ H int = Kn Xi,n;σ mσ b† mσ > bnσbnσ+ < bmσbmσ > b† nσ b† nσ(cid:17); iσaiσ b† nσ bnσ] + Unm [ 1 + 2 Xn,m;σ(cid:16)< b† 2 (cid:16)a† iσa† +JnXn;σ [2(a† iσ bnσbnσ + h.c.(cid:17) + 2a† 2σa1σ)b† 1σa2σ + a† nσ bnσ + (a† 1σa† 2σ bnσbnσ + h.c.)]. a† iσai−σ; (7) (8) (9) In the above expressions En are energies of the modes, J0 is a Josephson coupling strength between two fundamental modes localized in right and left traps of the double- well potential, Ωn are coupling strengthes between states of different polarizations at level n corresponding to intrinsic Josephson effect, U0 describes polariton- polariton interaction in the fundamental modes, Umn- interactions in excited modes, Kn- interactions between fundamental and excited modes and Jn describes the renormalization of the Josephson tunneling due to the interaction between fundamental and excited modes. The parameters entering into Hamiltonian 6 are related to those entering into the initial Hamiltonian 4 as E0(n) = Z d3rf ∗ 1,2(n)(r) − ¯h2 2m ∆ + Vext(r)! f1,2(n)(r), (10) 1,2(r) − ¯h2 2m ∆ + Vext(r)! f2,1(r), 1,2(r)f ∗ f ∗ 1,2(r ′ )V (r, r ′ ′ )f1,2(r )f1,2)(r) = 4 , ′ J0 = Z d3rf ∗ U0 = Z d3rd3r = gZ d3r(cid:12)(cid:12)(cid:12) f1(2)(r)(cid:12)(cid:12)(cid:12) Unm = gZ d3rf ∗2 Ω0(n) = gZ d3rf ∗ Kn = gZ d3rf ∗2 Jn = gZ d3rf ∗ m (r)f 2 n(r), 6 (11) (12) (13) (14) (15) (16) 1σ,2σ(nσ)(r)Ω(r)f1−σ,2−σ(n−σ)(r), 1,2(r)f 2 n(r) = gZ d3rf1,2(r)fn(r)2, 1 (r)f ∗ 1 (r)f2(r) fn(r)2 = gZ d3rf ∗ 2 (r)f 2 n(r). In the part corresponding to the excited states Hexc the interactions are treated using the mean field approximation [26] b† bnσbmσ = 2 < b† bmσ > b† bnσ+ < b† b† mσ nσ mσ nσ mσ b† mσ > bnσbnσ+ < bmσbmσ > b† nσ b† nσ. (17) In order to obtain a closed system of dynamical equations describing interacting PJJ we write Keldysh propagators of the system in the following representation Gαβ(t, t ′ ) = Gαβσσ(t, t′) Gαβ−σσ(t, t′) Gαβσ−σ(t, t′) Gαβ−σ−σ(t, t′) ! , (18) where the general indices α and β become i or j for the ground states and take values 1 or 2 as there are two lowest lying modes. The excited states are counted by associating m or n to both α and β. As previously indices ±σ denotes pseudospin degrees of freedom. The elements of the above matrix (18) are themselves 2 × 2 block-matrices of the form i Gαβσσ(t, t ′ ) = Gαβσσ(t, t′) Fαβσσ(t, t′) ¯Gαβσσ(t, t′) ! . ¯Fαβσσ(t, t′) with the elements being, for example for the reservoir modes Gαβσσ(t, t Fαβσσ(t, t ′ ′ ) =< T blσ(t)b† mσ(t ) =< T blσ(t)bmσ(t ′ ′ ) > ) > . (19) (20) (21) and similarly for the fundamental modes localized on the traps 1 and 2. Time-ordering T in the previous formulae is performed on the Keldysh contour [27] and appears because of non-adiabatic switching of Josephson coupling in the initial time instant. As it is not possible to guarantee the behavior of the system with such a kind of the irreversibility for asymptotically large times [28] the time contour in Keldysh formulation is adapted so that system evolves forwardly on time axis to some time and then from this point it makes backward evolution to the initial state. Any Keldysh time-ordered product of two operators, for example, operators bn and bm, has form 7 ′ ′ ′ )] >= θ(t, t ) < bm(t)bn(t < T [bm(t)bn(t )bm(t) > .(22) where θ(t, t′) is "step" function defined on the Keldysh contour for two arbitrary times t and t′ in way that is one always when the first argument is later then the second one, and zero otherwise. In this sense the first Green function in the formula (22) is called "greater", denoted usually with F > and the second one is "lesser" Green function F <. They act only on the forward or the backward Keldysh contour branch, respectively. , t) < bn(t ) > +θ(t ′ ′ Using Wigner transformation, the matrix elements Gαβσσ(t, t′) can be written in terms of center of "mass" and "relative" time coordinates T = (t + t′)/2 and τ = t − t′. We are interested here in external and internal Josephson dynamics which are much slower processes then the others occuring in the system so that we can work in the limit in which τ = 0. The new matrix elements Gαβσσ(T, τ ) will then only depend on the macroscopical time T [13]. According to the expression (22) we will then deal only with "lesser" functions F <(T ) and G<(T ). Equations of motion techniques combined with use of the mean- field approximation allows us to obtain the closed system of equations for Green functions defined by Eq.19: = Ξnσ ¯F < nσnσ − ¯ΞnσF < nσnσ − Ωn(G< nσn−σ − G< n−σnσ); (23) + Υnσ − Υn−σ)G< nσn−σ = 2Ξnσ ¯F < nσn−σ − 2¯ΞnσF < nσn−σ − Ωn(G< nσjσ − G< n−σn−σ); (24) − 2Enσ − 2Υnσ)F < nσnσ = 2¯ΞnσG< nσnσ + 2Ξnσ ¯G< nσnσ − ΩnF < nσn−σ; i (i (i dG< nσnσ dT d dT d dT d dT (25) (26) (i + 2Enσ + Υnσ + Υn−σ)F < nσn−σ = −2(Ξnσ − Ξn−σ)G< nσn−σ − Ωn(F < nσnσ + F < n−σn−σ). where n = 3, 4, ... stands for the n th excited level. Dynamics of the fundamental states i = 1, 2 is given by following expressions: dG< dT = (E0 + U0G< nσnσ)G< nσnσ)G< iσjσ + (27) G< G< iσiσ i iσiσ + 2KnXnσ iσiσ + Jn ¯F < +i(Kn ¯F < iσiσ + (J + 2JnXnσ ¯F < nσnσ − Ω0G< iσi−σ. with i, j = 1, i 6= j. Self-energies Υ and Ξ read jσjσ)Xnσ −iΥnσ = U ∗Xm 2 Xm −iΞnσ = U ∗ G< mσmσ + 2Kn(G< 1σ + G2σ <) + 2Jn(G< 1σ2σ + G< 2σ1σ), F < mσmσ + Kn 2 (F < 1σ1σ + F < 2σ2σ) + JnF < 1σ2σ. (28) (29) The normal and anomalous self-energies represent the energy renormalizations entering to the diagonal and off-diagonal propagators in the matrix (19) due to particle- particle 8 interactions present in PJJ. For simplicity we take them diagonal in excited level index n neglecting collisions between particles situated at different reservoir levels, Unm = U ∗δnm. The system of (23)-(29) is the closed system of nonlinear first order ordinary differential equations which can be solved numerically. The corresponding analysis is presented in the next section. 3. Results and Discussions We consider a PJJ inside GaAs- based quantum microcavity with parameters similar to those used in Ref.[20]. We numerically studied the system of equations (23)-(29) in order to analyze the effects of the interactions between fundamental and excited states given by parameters Kn and Jn on various types of Josephson deynamics. In the reference [13] it was found that for atomic condensates such kinds of interactions in general lead to chaotization of the Josephson oscillations after some initial period of regular dynamics. The effect is due to the intensive exchange of the particles between fundamental states and multi- mode reservoir of the excited states. Here we consider time intervals smaller then those necessary for the transition to chaotic regime. The reason is that polaritons have finite lifetimes, and in the regime of pulsed excitation they will simply disappear before the system will demonstrate the characteristics of chaos [20]. The quantity ρn(T ) = G< G< nσ+iσ+(T ) − G< nσ+iσ+(T ) − G< n−σ−i−σ−(T ) n−σ−i−σ−(T ) (30) describes circular polarization degree at the state n. Its dynamics is shown at Figure 2. For a while, we neglected the extrinsic Josephson coupling puting J0 = 0, Jn = 0. Panels a) and b) show a profile of the oscillations of circular polarization degree in the fundamental states corresponding to intrinsic Josephson effect and corresponding Fourier spectrum. The dashed line correspond to the case when coupling to excited states is switched off, while solid line accounts for this coupling. One sees, that introduction of the term Kn slightly renormalizes the frequency of the oscillations. However, the effect remains quite weak, as population of the excited levels is sufficiently small, less then ten percents of total number of particles. On the contrary, the effect of coupling becomes more pronounced if one monitors the intrinsic Josephson dynamics on the excited states, shown at panels c) and d). The interaction with fundamental modes changes the oscillation pattern on excited modes quite radically. The effect is clearly seen at panel d) showing Fourier power spectrum of the oscillations. Account for the terms Kn leads to the appearance of higher harmonics in the spectrum. Besides, in place of the intrinsic Josephson oscillations with zero time average, novel regime establishes in which < ρ(T ) >6= 0. In the absence of coupling Kn the frequencies of the intrinsic oscillations in the ground state and the reservoir are different (dashed lines in panels a) and c)) although polarization couplings on these levels are equal Ω0 = Ωn. It is result of the renormalization of the oscillation frequency for the fundamental mode ) t ( 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 0.10 0.05 ) t ( 0.00 -0.05 -0.10 0 9 a) b) 1.0 0.8 0.6 0.4 0.2 0.0 0 20 40 60 80 100 0.00 0.02 0.04 0.06 Energy(meV) 0.08 0.10 c) 1.0 d) 0.8 0.6 0.4 0.2 0.0 20 40 60 80 100 0.00 0.02 t(ps) 0.04 0.06 Energy(meV) 0.08 0.10 Figure 2. Panels a) and b) Profile of the intrinsic Josephson oscillations of circular polarization degree at fundamental states and its Fourier power spectrum. Panels c) and d) Profile of the intrinsic Josephson oscillations of circular polarization degree at first excited state and its Fourier power spectrum. Dashed lines correspond to the absence of the coupling between fundamantal modes and excited states, Kn = 0. Solid lines correspond to the case Kn = 0.3U0. due to nonlinearities, which is negligible in the low populated reservoir. Account of the reservoir- fundamental mode coupling makes the frequencies of intrinsic Josephson oscillations comparable for all modes of the system. Fig. 3 illustrates behavior of the polarization degree when ground state MQST phase in the intrinsic Josephson effect interacts with the reservoir. As compare the situation analyzed in Fig 2 particle-particle interactions characterized by a parameter U0 dominate over Josephson coupling Ω in the fundamental mode . The profiles of the oscillations of circular polarization degree for the fundamental modes are shown in the top of the panel a). The solid line corresponds to the case Kn 6= 0 and dashed one for Kn = 0. The dashed line is characterized by a single peak in the Fourier spectrum, usual for MQST regime in two- mode PJJ. On the contrary, the introduction of the coupling with excited level leads to the appearance of the two additional peaks in Fourier spectrum as it is shown at the panel b). The similar trends can be seen for oscillations in the excited level, illustrated by lower curves at panel a) and blue line at panel b). 0.6 a) 10 ) t ( 0.5 0.0 -0.1 1.0 0.8 0.6 0.4 0.2 0.0 0 10 20 30 t(ps) 40 50 b) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Energy(meV) Figure 3. Top of the panel a) shows the polarization degree of the intrinsic Josephson oscillations in the fundamental mode. Bottom of the panel a) shows the same quantity in the reservoir. Dashed lines: Kn = 0. Thick/blue and solid/red lines stands for K3 = 0.3U0. Panel b) shows corresponding Fourier power spectrum when the interactions are present (red for the fundamental mode, blue for the reservoir). Finally we consider the scenario in which two fundamental modes localized in traps one and two, initially elliptically polarized, are coupled with each other by means of extrinsic Josephson tunneling J. It was already pointed out that spatial separation of polarization occurs in this case [20] for sufficiently high polariton concentrations and circular polarization degrees. In this regime the dominant polarization component is trapped in one of the wells (dotted line in lower panel of Fig. 4), while another one undergoes Josephson oscillations (dotted line in upper panel of Fig. 4). In the presence of the higher modes there is an extra pseudospin-conserving exchange of the particles through the excited levels between the traps described by reservoir- assisted coupling term Jn in Eq.9). The influence of this term is illustrated in Fig.4 by solid red and blue lines corresponding to Jn = 0.1U0 and Jn = 0.3U0 respectively. For freely oscillating component the presence of the reservoir-assisted term leads to self-trapping. This transition is provided by competition of two effects: common Josephson tunneling J and reservoir-assisted coupling given by Jn. As their signs are opposite (J < 0 and Jn > 0), increase of Jn decrease in fact absolute value of the total Josephson coupling. 11 Figure 4. Upper panel: Josephson oscillations of the population imbalances between two traps for minor polarization component. Lower panel: Josephson oscillations of the population imbalances between two traps for major polarization component. J0 = 50µeV . Dashed line: no reservoir assisted coupling, Jn = 0. Red line: J3 = 0.1U0, blue line J3 = 0.2U0. This results in overwhelming of free oscillations by nonlinearities and establishing of MQST in the minor polarization component. The smaller is the total Josephson coupling the trapping becomes stronger (compare blue and red line of the upper panel). The dominant component rests relatively robust to the influence of the reservoir-assisted coupling. Note, that in hypothetical case of positive J the result will be opposite to those considered here: the reservoir- assisting terms will increase the absolute value of the effective tunneling constant, thus contributing to the distruction of MQST regime for major polarization component. 4. Conclusions In conclusion, we analyzed polarization dynamics in multimode Polartonic Josephson Junctions. We have found that the coupling between fundamental and excited modes changes the patterns of intrinsic and extrinsic Josephson oscillations, leading to the 12 appearance of higher harmonics and transitions between Josephson and MQST regimes. 5. Acknowledgment We thank Dr.D.D. Solnyshkov and Prof. N. Gippius for stimulating discussions. The work was supported by FP7 IRSES project "SPINMET" and Rannis "Center of excellence in polaritonics". I.A.S. acknowledges the support from COST "POLATOM" project and thanks Mediterranean Institute of Fundamental Physics for hospitality. References [1] B.D. Josephson, Phys. Lett. 1, 251 (1962). [2] A.G. Likharev, Rev. Mod. Phys. 51, 101 (1979). [3] S.V. Pereverzev, A. Loshak, S. Backhaus, J.C. Davis and R.E. Packard, Nature 388, 449 (1997). [4] M. Albiez, R. Gati, J. Folling, S. Hunsmann, M. Cristiani and M. K. Oberthaler, Phys. Rev. Lett. 95, 010402 (2005). [5] A. Smerzi, S. Fantoni, S. Giovanazzi and S. R. Shenoy, Phys. Rev. Lett. 79, 4950 (1997); G.J. Milburn, J. Corney, E.M. Wright and D. F. Walls, Phys. Rev. A 55, 4318 (1997); S. Raghavan, A. Smerzi, S. Fantoni and S. R. Shenoy, Phys. Rev A 59, 620 (1999). [6] J. Kasprzak, M. Richard, S. Kundermann, A. Baas, P. Jeambrun, J. M. J. Keeling, F. M. Marchetti, M. H. Szymaska, R. Andr´e, J.L. Staehli, V. Savona, P.B. Littlewood, B. Deveaud and L.S. Dang, Nature 443, 409 (2006); J. Kasprzak, R. Andr´e, L.S. Dang, I.A. Shelykh, A. V. Kavokin, Y.G. Rubo, K. V. Kavokin and G. Malpuech, Phys. Rev. B 75, 045326 (2007). [7] R. Balili, V. Hartwell, D. Snoke, L. Pfeiffer and K. West, Science 316, 1007 (2007) [8] J. Levrat, R. Butt´e, E. Feltin, J.-F. Carlin, N. Grandjean , D. Solnyshkov and G. Malpuech Phys. Rev. B 81, 125305 (2010). [9] J. J. Baumberg, A. V. Kavokin, S. Christopoulos, A. J. D. Grundy, R. Butt´e, G. Christmann, D. D. Solnyshkov, G. Malpuech, G. Baldassarri Hoger von Hogersthal, E. Feltin, J.-F. Carlin and N. Grandjean, Phys. Rev. Lett. 101, 136409 (2008). [10] T.C.H. Liew, I.A. Shelykh, G. Malpuech, Physica E 43, 1543 (2011) [11] A. Imamoglu and J. R. Ram, Phys. Lett. A, 214, 193, (1996). [12] I.A. Shelykh, A.V. Kavokin, Y.G. Rubo, T.C.H. Liew and G. Malpuech, Semicond. Sci. Technol. 25, 013001 (2010) [13] M. Truijjilo-Martinez, A. Posazhennikova and J. Kroha, Phys. Rev. Lett., 103, 105302, (2009). [14] C. Ciuti, V. Savona, C. Piermarocchi and A. Quattropani , Phys. Rev. B 58, 7926 (1998) [15] J.J. Hopfield, Phys. Rev. 112, 1555 (1958). [16] C. Weisbuch, M. Nishioka, A. Ishikawa and Y. Arakawa , Phys. Rev. Lett. 69, 3314 (1992). [17] M.Wouters and I. Carusotto, Phys. Rev. Lett., 99, 140402, (2007). [18] M. Wouters, Phys. Rev. B,77, 121302, (2008). [19] D. Sarchi, I. Carusotto, M. Wouters and V. Savona, Phys. Rev. B,77, 125324, (2008). [20] I.A. Shelykh, D. D. Solnyshkov, G. Pavlovic and G. Malpuech , Phys. Rev. B,78, 041302(R), (2008). [21] I.L. Aleiner and E.L. Ivchenko, JETP Letters 55, 692 (1992). [22] K.G. Lagoudakis, B. Pietka, M. Wouters, R. Andr´e and B. Deveaud-Pl´edran, Phys. Rev. Lett., 105, 120403, (2010). [23] D.D. Solnyshkov, R. Johne, I. A. Shelykh and G. Malpuech, Phys. Rev. B, 80, 235303, (2009). [24] D. Read, Y.G. Rubo and A. V. Kavokin, Phys. Rev. B, 81, 235315, (2010). [25] E.B. Magnusson, H. Flayac, G. Malpuech and I.A. Shelykh, Phys. Rev. B, 82, 195312, (2010). [26] A. Griffin, Phys. Rev. B, 53, 14, (1996). [27] J. Rammer and H. Smith, Rev. Mod. Phys., 58, 323, (1986). [28] H. Haug and A. P. Jauho, Quantum kinetics in transport and optics of semi-conductors, Springer Solid-State Sciences 59 (1996). 13 

1107.4560

3

1107

Stabilization of single-electron pumps by high magnetic fields

We study the effect of perpendicular magnetic fields on a single-electron system with a strongly time-dependent electrostatic potential. Continuous improvements to the current quantization in these electron pumps are revealed by high-resolution measurements. Simulations show that the sensitivity of tunnel rates to the barrier potential is enhanced, stabilizing particular charge states. Nonadiabatic excitations are also suppressed due to a reduced sensitivity of the Fock-Darwin states to electrostatic potential. The combination of these effects leads to significantly more accurate current quantization.

cond-mat.mes-hall

cond-mat

Stabilization of single-electron pumps by high magnetic fields J. D. Fletcher,1 M. Kataoka,1 S. P. Giblin,1 Sunghun Park,2 H.-S. Sim,2 P. See,1 T. J. B. M. Janssen,1 J. P. Griffiths,3 G. A. C. Jones,3 H. E. Beere,3 and D. A. Ritchie3 1 National Physical Laboratory, Hampton Road, Teddington, Middlesex TW11 0LW, United Kingdom 2 Department of Physics, Korea Advanced Institute of Science and Technology, Daejeon 305-701, Korea and 3 Cavendish Laboratory, University of Cambridge, J. J. Thomson Avenue, Cambridge CB3 0HE, United Kingdom (Dated: July 15, 2021) We demonstrate theoretically and experimentally how magnetic fields influence the single-electron tunnel- ing dynamics in electron pumps, giving a massively enhanced quantization accuracy and providing a route to a quantum current standard based on the elementary charge. The field dependence is explained by two ef- fects: Field-induced changes in the sensitivity of tunneling rates to the barrier potential and the suppression of nonadiabatic excitations due to a reduced sensitivity of the Fock-Darwin states to electrostatic potential. 2 1 0 2 g u A 1 3 ] l l a h - s e m . t a m - d n o c [ 3 v 0 6 5 4 . 7 0 1 1 : v i X r a PACS numbers: 73.23.Hk, 73.63.Kv I. INTRODUCTION Single-electron devices proposed for quantum information technologies1 -- 3 and quantum electrical metrology4,5 can be used to capture, manipulate, and release electrons through a series of gate pulses. To design such devices it is important to understand the electronic response to a rapid time-varying electrostatic potential, often in the presence of externally ap- plied magnetic fields. The effects of magnetic confinement on electronic states6,7 and on electron-electron interactions8 have been studied extensively. However, the effect of mag- netic fields on the electron dynamics in time-varying poten- tials is less established. Semiconductor single-electron pumps in magnetic fields are an example of a system which requires a consideration of these effects. It was found experimentally that the accuracy of the quantization current produced by these devices was strongly enhanced in magnetic field.9,10 More re- cently it has been shown how important this effect is for pro- viding a levels of quantization accuracy (at the part per million level and below) that make these devices useful in metrolog- ical applications11. The origin of this magnetic field depen- dence has not been explained. We explain here how magnetic fields influence the single- electron tunneling dynamics in electron pumps and show how large magnetic fields reduce back-tunneling errors by more than five orders of magnitude. We show that there are two distinguishable components to the field dependence of the pump accuracy. Firstly, we show through numerical calcu- lations how magnetic fields change the back-tunneling rates in the pump; magnetic fields enhance the sensitivity of tun- neling rates to the confinement barriers, which stabilizes the number of pumped electrons. Secondly, we report that the spillage of electrons through non-adiabatic processes12, which only appears when pumping at high frequencies, has a distinc- tive non-monotonic field dependence. Intriguingly, there is a also recovery of quantization accuracy at high field due to the suppression of excitations. Both effects are important in de- termining the ultimate current quantization accuracy in these pumps at high field, which is a crucial factor for their use in quantum metrology.11,13 FIG. 1. (a) Scanning electron microscope image of a typical device. (b) Schematic electrical connections. Electrons are pumped from left to right. (c) Potential profile during the pumping cycle (offset vertically): i. loading, ii. back-tunneling, iii. trapping, iv. ejection. (d) Pump current at f = 0.4 GHz as a function of VG2. Curves are offset vertically by a fixed amount as magnetic field is stepped in intervals of 2 T from 0 T to 14 T. Data for B <14 T have been shifted horizontally to align the I = 1ef to 2ef transition. (e) High resolution scans at f = 0.4 GHz at B = 10, 11,..14 T. Scans are offset by 10 fA. Dashed line is ef for each field. (f-h) dI/dVG2 on a color scale as a function of VG2 and B for f = 0.1, 0.4 and 1 GHz. Structure arising from the first excited state is labeled L1. II. ELECTRON PUMPS IN A MAGNETIC FIELD Our pumps use a dynamically formed quantum dot defined in a 2DEG AlGaAs/GaAs heterostructure14 by two surface gates [Fig. 1(a)]. The gates cross an etch-defined wire ter- minated with ohmic electrical contacts [see Fig. 1(b)]. The potential on the entrance gate (left) is modulated sinusoidally by VRF around a constant value VG1 while the exit gate (right) is held at constant voltage VG2.15 Pump operation is illustrated in Fig. 1(c)i-iv: (i) Electrons from the source reservoir (left) are loaded into a quantum dot formed in the space between the gates. (ii) While the dot is progressively isolated by the rising entrance barrier, some initially-trapped electrons tun- nel back to the source before tunneling is eventually cut off. (iii) The back-tunneling rate Γ depends strongly on the num- ber of trapped electrons n due to the charging energy, leading to the same number of electrons being trapped in every cy- cle. (iv) Electrons that remain trapped are forced over the exit barrier into the drain lead, producing a quantized current in an external circuit. The number of electrons pumped can be changed by adjusting the values of VG1 and VG2 giving current plateaux at integer multiples of ef, where f is the operating frequency and e is the elementary charge. Figure 1(d) shows how the pump current I varies with VG2 concentrating on the plateau at I = nef with n = 1 (one electron per cycle). Data are shown for a pump fre- quency f = 0.4 GHz in perpendicular magnetic field B up to 14 T. Measurements were performed in a 3He cryostat with a base temperature of ∼ 300 mK. At higher fields the cur- rent plateaux become markedly flatter and somewhat longer, increasing the accuracy of current quantization similar to pre- vious studies.9,10 To allow for detailed comparison with our model we have studied these effects in several samples in de- tail. High-resolution measurements, using the same experi- mental configuration as Ref. 11 are shown in Fig. 1(e). Mea- surement for B = 10 − 14 T indicate that the pumped current gets continuously closer to the expected value of ef at higher fields. This is the first time that variations with magnetic field at this level have been reported and underlines the significance of the magnetic field effect. Figure 1(f-h) shows dI/dVG2 as a function of VG2 and B for f = 0.1, 0.4 and 1 GHz. All three data sets show the movement of plateaux boundaries in magnetic field. Some of this behavior has been linked to magnetic confinement16. At high magnetic fields, where we expect tunneling rates to be suppressed, it appears that a shift in VG2 is required to recover the same current. This is consistent with the number of elec- trons loaded per cycle being determined by 'back-tunneling', which is sensitive to both magnetic field and the electrostatic potential. This explanation does not give any clues as to why the shape of the plateaux changes, which is also visible in this data. In fig. 1(f) where f = 0.1 GHz, sharpening of the plateaux boundaries and lengthening of the plateaux are visible, massively enhancing the pump accuracy from a few percent accuracy at zero field to the part per million level.11 We find that at higher frequency, for instance at f = 0.4 GHz in 1(g), similar behavior is seen except in a certain field range (near 4 T) where step-like features appear in the VG2 scans and broaden the plateau edge (labeled L1). This second ef- fect is identified as the non-adiabatic population of excited dot states12. Further increasing the frequency to 1 GHz, these step features destroy plateaux flatness over a wide field range as seen in Fig. 1(h). From detailed studies of several samples we have seen that there are two field dependent contributions to the pump accu- racy. At sufficiently low frequencies, where there is no evi- dence for excitation effects, the pump current accuracy is de- 2 FIG. 2. Model calculations: (a) Contour plot of the model 2D poten- tial well with a line cut along the pumping direction. (b) Tunneling rate Γ from the potential well as a function of V1, the entrance barrier for different values of B for a constant V2 = −50 mV. Solid lines are fits to an exponential function.(c) Schematic diagram indicating the relative tunneling rate for n = 1 and n = 2 (solid and dashed lines respectively) for states in zero and large field. Γ2 = eδ(B)Γ1 and Γ2(eV1) = Γ1(eV1 + ∆). See text for definition of δ. (d) Calculated pump current as a function of control voltage in the back-tunneling model (e) Fit of experimental data to determine δ(B) at 100 MHz (sample A). (f) Comparison of field induced changes in δ found ex- perimentally in sample A (as in Fig. 1) alongside model predictions (symbols). Shaded region indicates where non-adiabatic effects in- fluence plateau flatness at f = 0.4 GHz (g) Similar data for sample B. termined by the back-tunneling of excess electrons before the dot is isolated from the leads during the time when the en- trance barrier is rising rapidly [Fig. 1(b) ii]. This process can be described with time-dependent tunneling rates Γn(t) for the nth electron out of the dot, determined by the confining geometry.5,17,18 The disparity in tunneling rates for different numbers of electrons Γn (cid:28) Γn+1, combined with the increas- ing opacity of the tunneling barrier gives a mean number of electrons captured n (cid:39) an integer. To understand the effect of magnetic field on the accuracy of this process a calculation of Γn including the effects of magnetic confinement is required. Below we show the results of numerical calculation for a model of the pump. We use these to show how the back- tunneling quantization process is improved in a magnetic field. We then separately consider the field dependence of non-adiabatic excitation effect, which has a different origin. III. NUMERICAL CALCULATIONS OF TUNNELING RATE Previous work has illustrated schematically the significance of the time dependent tunneling rates for the single param- eter pumping using a one-dimensional model.17,18 However, to include a magnetic field a two dimensional calculation is required. We have calculated the tunnel-coupling for the two- dimensional Hamiltionian H = (i¯h∇−eA)2/2m∗−eV (x, y) where m∗ is the effective mass, A the magnetic vector poten- tial and V (x, y) the model potential (cid:20)−4(x − xb)2 (cid:21) (cid:88) b=1,2 − eV (x, y) = m∗ω2 yy2 − e 1 2 Vb exp d2 (1) consisting of two Gaussian barriers of width d = 60 nm, po- sitioned x2 − x1 =120 nm apart with amplitudes V1, V2 19 and with parabolic lateral confinement ¯hωy (cid:39) 5 meV. This poten- tial, shown in Fig. 2(a), was chosen to approximate the ex- perimental geometry and gives an orbital energy level spacing similar to that found in our pumps.12 The broadening of the electron energy is calculated by the lattice Green's function method.20,21 The two-dimensional continuous system is mod- eled by a discrete square lattice with a tight-binding Hamil- tonian21. The on-site energies of the Hamiltonian contain the position-dependent potential, while off-diagonal elements, describing the hopping between neighboring sites, include the Peierls phase factor from the magnetic field22. The reservoir regions away from the region Fig. 2(a) are treated as semi- infinite leads.21 Figure 2(b) shows the calculated back-tunneling rate Γ as a function of V1 for exit barrier height V2 = -50 mV at B = 0, 5, and 10 T. This shows the expected exponential variation and a reduction in Γ at higher fields -- the dot is increasingly decoupled from the leads by magnetic confinement. How- ever, fitting the data to the expression Γ = Γ0 exp(V1/
) we can see that the sensitivity of Γ to the barrier height is also strongly enhanced in higher magnetic fields, with
changing by a factor (cid:39) 6. We show in the next section that this effect drives a very large enhancement in quantization accuracy, but first discuss the origin of this effect. At low field the electronic wavefunction is determined solely by the electrostatic confin- ing potential, with the penetration of the wavefunction into the confining barrier determining the sensitivity of the tunnel- ing rate to barrier height. At high field, magnetic confinement reduces the size of the electronic wavefunction, causing the tunneling rate to drop. In our experiment this is compensated by forcing the electron closer to the barrier, but the probability density is then so concentrated that small variations in barrier height change the tunnel-coupling very rapidly. IV. EFFECT ON PUMPING ACCURACY The accuracy of the back-tunneling process is determined by the disparity in back-tunneling rates for different electron numbers. We use the above calculation for a single electron 3 occupying the dot and use some simple approximations to de- duce the effect of field on these relative tunneling rates. To estimate the tunneling rate for a state with two electrons we as- sume that energy of this system E2 is increased by an amount ∆ over the single electron case E1, effectively lowering the barrier by ∆/e. In this case Γ2 then has the same exponential dependence on V1 as Γ1 but is shifted to higher tunneling rates by a factor exp(δ) where δ = ∆/
e, as shown in Fig. 2(c), where
is related to the slope of the exponential behavior in Fig. 2(b). The ratio Γ2/Γ1, which determines quantization accuracy, can be enhanced either by increasing ∆ (increas- ing the charging energy) or by increasing the sensitivity of the tunneling rate to V1 (decreasing
). While the field depen- dence of the charging energy is typically observed to be very weak,7,23 -- 25 the large changes in
(B) seen in Fig. 2(b) will give large enhancements in quantization accuracy. We show in Fig. 2(d) the calculated pump current I(V2) using a model based on the back-tunneling process5,18 but in- cluding the field dependent effects found above. The func- tional form of I(V2) is given by (cid:20) (cid:88) n=1,2 (cid:18) − α(V2 − V0)
I = ef exp − exp + (1 − n)δ (cid:19)(cid:21) (2) where V0 sets the position of the first plateau and δ sets the plateaux flatness (larger values correspond to more accurate quantization). Equation 2 arises from the sensitivity of Γ1,2 to exit barrier height, which can be used to select the num- ber of electrons trapped by increasing the dot energy and in- creasing tunneling rates.18 The parameter α is defined as the proportionality constant between exit barrier height and dot potential. Figure 2(d) shows that, using the values of
derived from our numerical calculations, Eq. 2 predicts a pronounced en- hancement of plateaux flatness like that seen experimentally. One small difference is that while it reproduces the experi- mentally observed sharpening of the plateau risers, plateaux length is fixed. The change in plateaux length suggests a slight difference in the way that magnetic field enhances the sensi- tivity of Γn to V2 compared to its effect on Γ(V1). This would be equivalent to making α field dependent, which would allow the plateaux length to change with field as seen experimentally [see Fig. 1(d)]. We fit our experimental data to Eq. 2 and extract an ef- fective value of δ as a function of field to compare with our calculations.26 This is shown in Fig. 2(f) and Fig. 2(g) for two samples (Sample A is the same as in Fig. 1). At a frequency of 0.1 GHz, Eq. 2 fits the data well and there is no sign of any excitation effects. In both samples there is a strong monotonic increase in δ. This field dependent enhancement is similar in size to that estimated in our model. In the next section we illustrate where the scale and field de- pendence of this effect arises by a simple analytical estimate using the WKB model, which serves to corroborate these nu- merical calculations. FIG. 3. Field dependence of δ(B) assuming that it is determined by the effective barrier thickness. V. EFFECT OF MAGNETIC FIELD ON TUNNELING RATES IN WKB MODEL To complement the detailed numerical calculation we can illustrate the qualitative origin of the relevant field scales and how these relate to experimental dimensions. We take a 1D WKB approximation27 applied to the transmission coefficient for a particle tunnelling through a barrier with shape V (x) and length d. The transmission coefficient is given by (cid:16)−2(cid:82) x2 (cid:16)−2(cid:82) x2 x1 x1 dx (cid:113) 2m∗ (cid:17) (cid:113) 2m∗ ¯h2 (V (x) − E) ¯h2 (V (x) − E) dx (cid:104) T = exp 1 + 1 4 exp (cid:17)(cid:105)2 . (3) In the case of a constant barrier potential V0 of width d and considering small transmission probabilities this gives a tun- neling rate −2d(cid:112)2m∗(V0 − E) Γ ∝ exp (4) where an effective barrier height is defined by the quantity Vb = (V0 − E). This effective barrier can be reduced by low- ering the potential barrier by an amount ∆V0, or increasing the energy of the electron by a small amount ∆E, either of which will increase the tunneling rate. We then write −2d(cid:112)2m∗(V0 + ∆V0 − E − ∆E) Γ ∝ exp ¯h ¯h 4 rates Γ2/Γ1 = exp(δ) from two different energy levels, corre- sponding to different numbers of electrons in the dot. Within the above approximations we find that −d 2m∗ √ δ = ln Γ2 − ln Γ1 = ¯h(cid:112)(V0 − E) (7) ∆. where the charging energy ∆ = E2 − E1 gives rise to a large difference in tunnel rates whose ratio depends on barrier ge- ometry. To consider how the magnetic field changes the tun- neling rate we introduce an effective barrier thickness de(B) which increases with magnetic field28. This describes the fact that, under magnetic field, the spatial extent of the wavefunc- tion is reduced, decreasing the penetration into the barrier. This will lead to an increase in the ratio of tunneling rates. In our model of pump operation this will change the plateaux quality parameter δ according to δ(B) δ(0) ∼ de(B) de(0) (8) 24 (cid:16) m∗(cid:112)ω2 c /4/¯h2(cid:17)−1/2 An estimate of the field dependence of this enhancement can be found by taking the expected changes in magnetic length (cid:96)B(B) from the solution to the 2D harmonic poten- tial well in a perpendicular magnetic field24 with de = d0 − (cid:96)B(B) + (cid:96)B(0) and (cid:96)B(B) = where ωc = eB/m∗ and ¯hω0 is the electrostatic confinement strength. There is a gradual crossover from an electrostatic to magnetic dominated regime, for example for ¯hω0 = 2 meV, (cid:96)B shrinks from ∼ 25 nm at B = 0 T to ∼10 nm at B = 14 T. The result is an increase in δ(B) at high field by an amount given by the relative change in (cid:96)B(B) compared to the zero field length d0. The onset field of the effect is determined by the ratio ¯hω0/¯hωc. 0 + ω2 Fig. 3 shows the enhancement this for a range of values of ¯hω0 similar to that expected12 and an effective value of d0 = 4 nm. Choosing a larger or smaller value of d0 simply magnifies or weakens the enhancement effect, without chang- ing the result qualitatively. This simple model reproduces the qualitative effects that are more accurately probed by the above numerical calculations. , (5) VI. NON-ADIABATIC EFFECTS expand in ∆V0 and retaining only the term that depends on ∆V0 gives us Γ ∝ exp √ −d ¯h(cid:112)(V0 − E) 2m∗ ∆V0 (6) showing that the sensitivity of tunneling rates to effective bar- rier height is determined by the thickness of the barrier. Con- sidering for changes in energy ∆E gives an equivalent expres- sion. We can see qualitatively that any effect which modifies this barrier thickness, in the present case a strong magnetic confinement effect, can modify the sensitivity of the tunnel- ing rate to confinement parameters. In our model of pump operation we are normally interested in the ratio of tunneling Fig. 2(f) shows that at 0.4 GHz (sample A) there is also a strong increase in δ with field, although there is a difference in the maximum value of δ reached. This effect may be due to the slightly different confinement potential shape at these higher frequencies as the value of VG2 at which the plateaux appear are different. The deterioration of pump accuracy at these higher frequencies can be overcome by the use of a spe- cially tailored wave-form,11 which gives an effective five-fold increase in frequency for the same pump accuracy. A very distinctive frequency dependent effect is visible in the form of a dip in plateaux flatness around 4 T. This departure from a monotonic field dependence is associated with the onset of non-adiabatic effects. In this regime δ is strongly suppressed 5 duce nonadiabatic excitations of the dot. Excitation features emerge above a few Tesla12 but as the field is increased above 12 T these features (peaks in the derivative) become signifi- cantly weaker. Figure 4(b) and (c) show this effect in more detail. This leads to the recovery of δ seen in Fig. 2(g). Ac- cording to Ref.18, the recovery of a δ (cid:39) 18 is sufficient to give a quantization accuracy of (cid:39) 3 parts in 107. This represents an enhancement of 105 over the zero field case where δ ∼ 5 gives only ∼5% accuracy. The observation of a limited field range where nonadiabatic effects are visible can be explained by a competition between magnetic and electrostatic effects on the electronic wavefunc- tion. Nonadiabatic transition rates depend both on the strength and rapidity of the perturbation of the wavefunction.12 At high field ¯hωc (1.7 meV/T) can exceed ¯hω0 ∼ a few meV in our system. The relative contribution of the electrostatic com- ponent is diminished and the magnetic field determines the size of the wavefunction. As a result, changes in the con- finement potential during pumping have a weaker effect on the dot wavefunction and nonadiabatic transition rates are re- duced. Figures 4(d) and 4(e) show, for example, that the prob- ability density ψ2 and eigenenergy ε0 of the lowest energy orbital Fock-Darwin state29 become increasingly insensitive to changes in ω0 at higher field, consistent with the disappear- ance of the excitation features. At lower fields excitation features become weaker due to the significant increase in the excitation gap between ground and excited states. This field dependent gap can be seen di- rectly in the spectrum of excitation features in Fig. 4(c)12. The excitation gap between the ground and first excited orbital en- ergy levels ε1 − ε0 is reduced strongly with field, which may be sufficient to suppress excitations at low field. However, the magnitude of this field dependence is sensitive to ω0 [see Fig. 4(f)], which is time dependent in our case. Excitation pro- cesses happening earlier in the pumping cycle, when the dot confinement is smaller, would be more sensitive to magnetic fields. In summary, we have shown that electron dynamics in single-electron tunable-barrier pumps are sensitive to mag- netic fields via two mechanisms. Firstly, the increased sen- sitivity of tunneling rates to the confining barriers enhances the separation of back-tunneling times that is relevant for en- suring stability in the number of electrons trapped. Secondly the magnetic field plays a role in suppressing excitations of the quantum dot, which would other wise lead to unwanted spillage of trapped electrons. These effects are both important in allowing pumps to operate at high speed with error rates smaller than one part per million in high fields.11 This research is supported by the UK Department for Busi- ness, Innovation and Skills, the European Metrology Research Programme, grant no. 217257, the UK EPSRC and Korea NRF (2011-0022955). Phys. Rev. Lett. 91, 226804 (2003). 3 D. Loss and D. DiVincenzo, Phys. Rev. A 57, 120 (1998). 4 M. D. Blumenthal, B. Kaestner, L. Li, S. Giblin, T. J. B. M. FIG. 4. (a) Current plateaux at 1 GHz for fields of 6, 10 and 14 T along with dI/dVG2 (offset vertically, VG1 = −0.4 V ). Voltage changes ∆VG2 are measured from the rising edge of the first plateau. All data in this figure is from sample B. Arrows indicate excitation features. (b) Maps of dI/dVG2 (color scale) as a function of both VG2 and VG1 for 6 T, 10 T and 14 T at 1 GHz. (c) Color map of dI/dVG2(∆VG2) as a function of field up to 14 T. Excitation gap ∆ε = ε1−ε0 is indicated. (d) Fock-Darwin state probability density for different combinations of B and ¯hω0. Open curves are for ¯hω0 = 8 meV and filled curves are for ¯hω0 = 2 meV. (e) Eigenenergy of the ground state solution of the Fock-Darwin confinement potential7 for different combinations of B and electrostatic confinement energy ¯hω0. (f) Field dependence of the first excitation gap between (n, l) = (0, 0) to (n, l) = (0, 1) states, where n, l are the radial quantum number and orbital angular momentum respectively, for ¯hω0 = 2, 4, 8 meV. by excitation features, which appear as 'shoulder' features in the VG2 scans.12 Data at 1 GHz shows further suppression of δ over a wide field range, but with a recovery at high field. In these devices rapid changes in the electrostatic confine- ment potential can populate excited states of the dot by nona- diabatic processes12, as observed in Fig. 2 and in Fig. 4. Tun- neling rates out of excited states are larger so electrons "spill" out of the pump and the current plateaux are eroded into a number of unquantized steps. Figure 4(a) shows I(VG2) and dI/dVG2 for f = 1 GHz (Sample B), high enough to in- 1 J. Elzerman, R. Hanson, L. van Beveren, B. Witkamp, L. Vander- sypen, and L. Kouwenhoven, Nature 430, 431 (2004). 2 T. Hayashi, T. Fujisawa, H. Cheong, Y. Jeong, and Y. Hirayama, Janssen, M. Pepper, D. Anderson, G. Jones, and D. A. Ritchie, Nat. Phys. 3, 343 (2007). 5 A. Fujiwara, K. Nishiguchi, and Y. Ono, Appl. Phys. Lett. 92, 042102 (2008). 6 C. Darwin, in Mathematical Proceedings of the Cambridge Philo- sophical Society, Vol. 27 (Cambridge Univ Press, 1931) p. 86. 7 S. Tarucha, D. Austing, T. Honda, R. vanderHage, and L. Kouwenhoven, Phys. Rev. Lett. 77, 3613 (1996). 8 M. Ciorga, A. Wensauer, M. Pioro-Ladriere, M. Korkusinski, J. Kyriakidis, A. Sachrajda, and P. Hawrylak, Phys. Rev. Lett. 88, 256804 (2002). 9 S. J. Wright, M. D. Blumenthal, G. Gumbs, A. L. Thorn, M. Pep- per, T. J. B. M. Janssen, S. N. Holmes, D. Anderson, G. A. C. Jones, C. A. Nicoll, and D. A. Ritchie, Phys. Rev. B 78, 233311 (2008). 10 B. Kaestner, C. Leicht, V. Kashcheyevs, K. Pierz, U. Siegner, and H. W. Schumacher, Appl. Phys. Lett. 94, 012106 (2009). 11 S. Giblin, M. Kataoka, J. Fletcher, P. See, T. Janssen, J. Griffiths, G. Jones, I. Farrer, and D. Ritchie, Nat Commun 3, 930 (2012). 12 M. Kataoka, J. D. Fletcher, P. See, S. P. Giblin, T. J. B. M. Janssen, J. P. Griffiths, G. A. C. Jones, I. Farrer, and D. A. Ritchie, Phys. Rev. Lett. 106, 126801 (2011). Technology 14, 1237 (2003). 13 N. M. Zimmerman and M. W. Keller, Measurement Science and 14 Carrier density 2.7 × 1015 m2 at 1.5 K, mobility is 70 m2/Vs. 15 Peak to peak VRF (cid:39) VG1, VG2 (cid:39) 0.4 V. 16 S. J. Wright, A. L. Thorn, M. D. Blumenthal, S. P. Giblin, M. Pep- per, T. J. B. M. Janssen, M. Kataoka, J. D. Fletcher, G. A. C. Jones, C. A. Nicoll, G. Gumbs, and D. A. Ritchie, J. Appl. Phys. 109, 102422 (2011). 6 17 B. Kaestner, V. Kashcheyevs, S. Amakawa, M. D. Blumenthal, L. Li, T. J. B. M. Janssen, G. Hein, K. Pierz, T. Weimann, U. Sieg- ner, and H. W. Schumacher, Phys. Rev. B 77, 153301 (2008). 18 V. Kashcheyevs and B. Kaestner, Phys. Rev. Lett. 104, 186805 (2010). 19 We distinguish the numerical values V1, V2 of potential within the 2DEG from the experimental voltages VG1,G2 applied to surface gates. 20 H.-S. Sim, G. Ihm, N. Kim, and K. Chang, Phys. Rev. Lett. 87, 146601 (2001). Univ Press, 1997). 21 S. Datta, Electronic transport in mesoscopic systems (Cambridge 22 R. Feynman, R. Leighton, M. Sands, et al., The Feynman lectures on physics III (Addison-Wesley Reading, MA, 1964). 23 M. Ciorga, A. S. Sachrajda, P. Hawrylak, C. Gould, P. Zawadzki, S. Jullian, Y. Feng, and Z. Wasilewski, Phys. Rev. B 61, R16315 (2000). 24 L. Kouwenhoven, D. Austing, Progress in Physics 64, 701 (2001). and S. Tarucha, Reports on 25 R. C. Ashoori, H. L. Stormer, J. S. Weiner, L. N. Pfeiffer, K. W. Baldwin, and K. W. West, Phys. Rev. Lett. 71, 613 (1993). 26 The experimental scaling of VG2 which is (α/
) is also treated as a free parameter. In the calculations the zero field value of δ is set at the experimental value. 27 R. E. Langer, Phys. Rev. 51, 669 (1937). 28 A. Huttel, S. Ludwig, H. Lorenz, K. Eberl, and J. Kotthaus, Phys- ical Review B 72, 081310 (2005). 29 L. Kouwenhoven, T. Oosterkamp, M. Danoesastro, M. Eto, D. Austing, T. Honda, and S. Tarucha, Science 278, 1788 (1997) 

1806.10769

1

1806

Transient thermal characterization of suspended monolayer MoS$_2$

We measure the thermal time constants of suspended single layer molybdenum disulfide drums by their thermomechanical response to a high-frequency modulated laser. From this measurement the thermal diffusivity of single layer MoS$_2$ is found to be 1.14 $\times$ 10$^{-5}$ m$^2$/s on average. Using a model for the thermal time constants and a model assuming continuum heat transport, we extract thermal conductivities at room temperature between 10 to 40 W/(m$\cdot$K). Significant device-to-device variation in the thermal diffusivity is observed. Based on statistical analysis we conclude that these variations in thermal diffusivity are caused by microscopic defects that have a large impact on phonon scattering, but do not affect the resonance frequency and damping of the membrane's lowest eigenmode. By combining the experimental thermal diffusivity with literature values of the thermal conductivity, a method is presented to determine the specific heat of suspended 2D materials, which is estimated to be 255 $\pm$ 104 J/(kg$\cdot$K) for single layer MoS$_2$.

cond-mat.mes-hall

cond-mat

Transient thermal characterization of suspended monolayer MoS2 Robin J. Dolleman,1, ∗ David Lloyd,2 J. Scott Bunch,2, 3 Herre S. J. van der Zant,1 and Peter G. Steeneken1, 4 1Kavli Institute of Nanoscience, Delft University of Technology, Lorentzweg 1, 2628 CJ, Delft, The Netherlands 2Department of Mechanical Engineering, Boston University, Boston, Massachusetts 02215 United States 3Boston University, Division of Materials Science and Engineering, Brookline, Massachusetts 02446 United States 4Department of Precision and Microsystems Engineering, Delft University of Technology, Mekelweg 2, 2628 CD, Delft, The Netherlands 8 1 0 2 n u J 8 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 6 7 0 1 . 6 0 8 1 : v i X r a We measure the thermal time constants of suspended single layer molybdenum disulfide drums by their thermomechanical response to a high-frequency modulated laser. From this measurement the thermal diffusivity of single layer MoS2 is found to be 1.14 × 10−5 m2/s on average. Using a model for the thermal time constants and a model assuming continuum heat transport, we extract thermal conductivities at room temperature between 10 to 40 W/(m·K). Significant device-to-device variation in the thermal diffusivity is observed. Based on statistical analysis we conclude that these variations in thermal diffusivity are caused by microscopic defects that have a large impact on phonon scattering, but do not affect the resonance frequency and damping of the membrane's lowest eigenmode. By combining the experimental thermal diffusivity with literature values of the thermal conductivity, a method is presented to determine the specific heat of suspended 2D materials, which is estimated to be 255 ± 104 J/(kg·K) for single layer MoS2. I. INTRODUCTION The distinct electronic [1–3] and mechanical [4, 5] prop- erties of atomically thin molybdenum disulfide opens up possibilities for novel nanoscale electronic [6] and opto- electronic [7–9] devices. The large and tunable Seebeck coefficient of single-layer MoS2 makes this material inter- esting for on-chip thermopower generation and thermal waste energy harvesting [10]. Since the power efficiency of these devices depends on the thermal conductivity, it is of interest to study the transport of heat in single- layer MoS2. Several theoretical works have found values of the thermal conductivity k of single layer MoS2 rang- ing between k = 1.35 up to 83 W/(m·K) [11–15]. By exploiting the temperature-dependent phonon frequency shifts in Raman spectroscopy [16], several experimental works have measured the thermal conductivity of single- layer MoS2 . Experimental values of k = 34.5 and 84 W/(m·K) of exfoliated single layer MoS2 have been re- ported [17, 18], while single-layer MoS2 grown by chem- ical vapor deposition was found to show a significantly lower thermal conductivity of 13.3 W/(m·K) [19]. Here, we thermally characterize suspended single- layer MoS2 drum resonators by measuring their ther- mal time constants. This was achieved by measuring the frequency-dependent vibration amplitude in response to rapidly varying heat flux delivered by a modulated diode laser, similar to previously reported characteriza- tion on single-layer graphene [20]. Since these are fre- quency based measurements, the result is to first order independent of the absorbed laser power, which greatly facilitates calibration compared to Raman spectroscopy based methods. Furthermore, the method allows one to study relations between the mechanical and thermal ∗ R.J.Dolleman@tudelft.nl properties of the material. From measurements of the thermal time constant τ, we find the thermal diffusiv- ity of MoS2 to be on average 1.05 × 10−5 m2/s for 5 µm diameter drums and 1.29 × 10−5 m2/s for 8 µm drums. Assuming a specific heat value of 373 J/(kg·K), this cor- responds to k = 19.8 W/(m·K) and k = 24.7 W/(m·K). The remainder of this article is structured as follows. Section II describes the experimental setup, fabrication, actuation and read-out of the motion of the single-layered MoS2 drums. The following section III describes the ther- mal model of the system and how τ is extracted from the experiments. Section IV shows the experimental re- sults of τ and extracts the value of the thermal diffu- sivity. This section also examines the relation between mechanics and thermal transport. Section V contains an extensive discussion elaborating on the possible causes of the large device-to-device variation, the specific heat of MoS2 and compares the present results to single-layer graphene. The conclusions of this work are then outlined in section VI. II. EXPERIMENTAL SETUP We use a substrate with many circular cavities to per- form the experiment. The fabrication starts with a sili- con chip with 285 nm of silicon dioxide. Circular cavities of approximately 300 nm deep and with a diameter of 8 and 5 micron are etched in the oxide layer. Many single layer MoS2 flakes grown by chemical vapor deposition are transferred over the substrate by a dry transfer method to create suspended drum resonators as drawn in Fig. 1(a). An optical image of several devices is shown in Fig. 1(d). The Raman and photoluminescence (PL) spectra of both the suspended MoS2 flakes are shown in in Figs. 1(e) and (f), data was taken on suspended drums to prevent the effects of substrate doping [9, 21, 22]. These measure- ments ensure that the MoS2 flakes are single-layer, since 2 FIG. 1. (a) Schematic of the samples and the laser interferometer setup to actuate and detect the motion of the single-layer MoS2 drum resonators. (b) Overview of the physical processes involved in measuring the transient properties of heat transport in the drum. (c) Typical experimental result of the real and imaginary part of the amplitude and the fits to find the thermal time constant. At lower frequencies a feature due to electrical cross talk becomes visible due to the low optical gain during the experiment. At higher frequencies the fundamental resonance is clearly visible. (d) Optical image of the device showing a single-layer MoS2 sheet on top of the substrate and several suspended drums. (e) Raman spectrum of the suspended and supported MoS2. (f) Photoluminescence spectrum of the suspended and supported MoS2. The A0 peak position is found at 1.89 eV. no indirect transition is observed in the PL spectrum (Fig. 1(f)) [22]. In the Raman spectra (Fig. 1(e)) the 2g peak is found at 384.9 cm−1 and the A1g peak at E1 404.5 cm−1, also in accordance with single-layer MoS2 [23]. Furthermore, the positions of both the E1 2g Raman peak and PL A0 (1.89 eV) suggests that no large strains (> 1%) are induced by the transfer [24]. More details on the CVD growth and transfer can be found in ref. [25]. The samples are kept in an atmosphere with a maximum pressure of 1 × 10−6 mbar for two weeks before and dur- ing the experiment to ensure all gas has escaped from the cavity. Figure 1(a) also shows a schematic drawing of the in- terferometer setup used to actuate and read-out the mo- tion of the membrane. The red laser intensity on the photodiode is used to read-out the motion using Fabry- Perot interferometry between the moving membrane and the fixed back-mirror, which is the silicon [4, 26, 27]. The blue laser heats up the membrane, which causes the membrane to move due to thermal expansion [20, 28]. The blue laser is power modulated using the output of a vector network analyzer (VNA). The input of the VNA is connected to the photodiode that detects the reflected red laser intensity. A dichroic mirror is used to prevent the blue laser light from reaching the photodiode. The VNA measures both the amplitude and the phase of the transmitted signal. All parasitic phase shifts in the elec- trical and optical components are measured by directly pointing the blue laser at the photodetector and are elim- inated by using the measured transmission function to deconvolve the experimental results [20]. III. THERMAL TIME CONSTANT Due to the diffusion of heat through the membrane, there will be a time delay between the optical power delivered to the membrane and the membrane's motion (Fig. 1(b)). The diffusion of heat can be described by the heat equation: ρcp dT dt − k∇2T = P, (1) where T (x, t) is the temperature and P (x, t) the heat flux applied to the membrane. ρ is the density of the material, cp the specific heat, k the thermal conductivity, x is the position vector and t is time. By separation of variables, and by using a lumped element model with incident laser heat flux P = Paceiωt, Eq. 1 can be simplified, which results in a single relaxation time approximation for the time-dependent temperature: C d∆T dt + 1 R ∆T = Paceiωt, (2) where C is the heat capacity and R the thermal re- sistance. Below the resonance frequency, the motion z = zωeiωt is proportional to the temperature change, z = A∆T , such that it follows from Eq. 2 that [20, 29]: zω = APacR iωτ + 1 = APacR 1 − iωt 1 + ω2τ 2 (3) where A is a proportionality constant that will be ob- tained by fitting and τ = RC the thermal time constant of the suspended drum. 405 nmdiode laser633 nmHe-Ne laserPhoto-detectorVNAFrequency (Hz)0Amplitde (a.u.)10k100k1M10M100MImaginary partFitReal partFitElectrical cross talkCut-off frequencyωc = 1/τ = 227 ns Mechanical resonancepeak (a)ObjectiveDichroic mirrorQuarter waveplatePolarizing beamsplitterVacuum chamberSilicon substrateSilicon dioxide (300 nm)Single layer MoS2Suspended drumOptical powerHeat fluxHeat diffusionMechanicalresponse(b)(c)MoS2SiO2Suspended drums20 µm(d)300400500600Wavenumber (cm-1)A1g1.41.61.82.02.2eVIntensity (a.u.)Intensity (a.u.)(e)(f)E12gSiThe thermal time constant τ can be determined from the measured thermomechanical frequency response of the drum over several decades using the setup in Fig. 1(a). Figure 1(c) shows the real and imaginary part of the experimentally obtained frequency response from a MoS2 drum with a diameter of 8 µm. It follows from eq. 3 that the imaginary part of the response function has a maximum amplitude at ωτ = 1. This maximum is indeed observed at a cut-off frequency of ωc = 2π × 800 kHz in Fig. 1d, which is far below the membrane's low- est resonance frequency such that the relation zω = A∆T is valid. By fitting the imaginary part using eq. 3 the thermal time constant of the membrane is determined to be τ = 1/ωc = 227 ns. The resonance peaks were ana- lyzed by fitting a harmonic oscillator model to the data, from which the resonance frequency and quality factor is found. Although both the real and imaginary part of the response function fit well to equation (3), deviations around 300 kHz are observed which are attributed to elec- trical cross-talk, most likely due to capacitive coupling to the optical table containing the experimental setup. Be- cause the laser powers are low in these experiments to prevent damage to the drums, the total optical signal on the photodiode is very low, making the system very sus- ceptible to parasitic cross-talk. The low frequency data was excluded for the fit in order to prevent cross-talk from affecting the value of τ. IV. RESULTS Frequency response fits as shown in Fig. 1(c) are ob- tained on a total of 32 single layer MoS2 drums with a 5 micron diameter and 18 drums with a 8 µm diame- ter. Figure 2(a) shows the experimentally obtained val- ues from all the drums as function of drum size and Fig. 2(b) shows a density plot for both diameters. Significant spread in the value of τ is found, even for drums of the same diameter. To exclude large effects of outliers, we only analyzed 80% of the samples with value τ closest to the mean and find ¯τ = 126 ns for the 5 micron diameter drums and ¯τ = 253 ns for the 8 micron drums. Aubin derived an expression for the thermal time con- stant for a uniformly heated circular drum [30, 31]: τ = a2ρcp µ2k , (4) where a is the drum radius, ρ the density, cp the specific heat and k the thermal conductivity the material. For a uniformly heated drum, µ = 2.4048 is the first root of the Bessel function J0(x). However, in the experiments the membrane is heated by a focused laser spot in the center of the drum. We therefore use a numerical COM- SOL model that adapts the value of µ by taking a point heat source in the center of the membrane. The measure- ment of the temperature is taken as the average temper- ature over the surface over the drum, since we expect the mechanical response to depend on the temperature field 3 in the entire drum. From the simulations it was found that µ2 = 5.0 is an accurate representation of the exper- iments. This should predict the value of k with an error less than 10% as long as 15 < k < 100 W/(m·K) and assuming that cp = 373.5 J/(kg·K) (See Supplemental Information). Using Eq. 8 we can estimate the thermal diffusivity of MoS2 α = k/ρcp: α = a2 5τ . (5) This expression was used to estimate the thermal diffu- sivity for each drum as shown in Fig. 2(c). We find the diffusivity is slightly diameter-dependent with an average diffusivity ¯α = 1.05 × 10−5 m2/s for the 5 micron drums and ¯α = 1.29 × 10−5 m2/s for the 8 micron drums. Based on known values of cp and ρ of molybdenum disulfide at room temperature (cp = 373.5 J/(kg·K) and ρ = 5060 kg/m3) we can estimate k = a2ρcp/(5τ ) from experimental values of τ. Fig. 2(c) shows the cumula- tive density function calculated for each drum. We find a mean of k, ¯k = 19.8 W/(m· K) with a standard devi- ation of 9.3 W/(m· K) for the 5 micron drums and for the 8 µm drums we find ¯k = 24.7 W/(m·K) with stan- dard deviation σk = 8.4 W/(m· K). We thus observe a considerable spread between devices. Moreover, most of the values of k found here are smaller compared to previ- ous observations in literature that used exfoliated MoS2 devices [17, 18], but are larger than CVD MoS2 values [19]. A. Comparison to the resonant properties The transient mechanical characterization allows one to study whether the mechanical properties of the sus- pended drums are correlated to the thermal properties. This might be expected, since the acoustic phonon ve- locities can be tension dependent, which would result in a correlation between the resonance frequency and the thermal diffusivity. Also mechanical damping in graphene due to defects could cause increased phonon scattering, which would lead to a lower thermal conduc- tivity for drums with a low mechanical Q. To study this, the resonance peaks were fitted by a harmonic oscillator model to extract the resonance fre- quency and the quality factor. The distribution of all the resonance frequencies is shown in Fig. 3(a) and the quality factors are shown in Fig. 3(c). We first investi- gate whether the thermal diffusivity is affected by strain in the resonator. The fundamental resonance frequency f of a circular drum resonator is given by: (cid:114) n0 2.4048 f = , (6) 2πa ρh where h is the thickness of the drum and n0 the tension in the membrane. From this, we deduce that f 2a2 ∝ n0 if ρh is the same for each drum. Figure 3(b) shows a scatter 4 FIG. 2. (a) Thermal time constants as function of diameter. Predictions using eq. 8 are plotted with several values of k obtained from literature: k = 23.3 W/(m· K) corresponding to α = 1.23 × 10−5 m2/s, [12] and 84 W/(m·K) to α = 4.44 × 10−5 m2/s. [18] (b) Density plot of the thermal time constant for both diameter, drums with extremely large values of τ and low resonance frequency were excluded. (c) CDF of the thermal conductivity k estimated from the values of τ using cp = 373.5 J/(kg·K) and ρ = 5060 kg/m3. mental data. The quality factor is nearly independent of diameter as shown in Fig. 3(c), we find ¯Q = 26.0 with standard deviation 10.4 for the 5 µm drums and ¯Q = 24.3 with standard deviation σQ = 10.3 for the 8 µm drums. B. Phonon relaxation time and mean free path The thermal conductivity can be expressed as k ≈ ρcpvλ, [33] where v and λ are appropriately averaged phonon group velocity and mean free path, respectively. Substituting this expression in eq. 8 gives: τ = a2 5vλ = a2 5v2τph , (7) where τph is the phonon relaxation time. We take the averaged velocity as v ≈ 300 m/s based on calculations from several theoretical works [12, 34, 35] and use Eq. 7 to estimate τph and λ. For the 5 micron drums we find an average phonon relaxation time and mean free path of 116 ps and 34.9 nm, respectively. For the 8 micron drums we find 143 ps and 43.2 nm. For both cases we again find device-to-device variations due to the spread in the measured values of τ. FIG. 3. Investigation of correlations between the mechanical and thermal properties. (a) CDF of the resonance frequency for both diameters. (b) Scatter plot with the thermal diffusiv- ity on the horizontal axis and frequency times radius squared (which is proportional to tension) on the vertical axis. (c) CDF of the quality factor for both diameters. (d) Scatter plot with the thermal time constant on the horizontal and quality factor of resonance on the vertical axis. plot of f 2a2 versus the thermal diffusivity for each drum, the strain was estimated assuming the membrane has the ideal mass and the 2D Young's modulus was taken as 160 N/m [25, 32]. No meaningful correlation between tension and the thermal diffusivity could be uncovered. We further investigate whether the mechanical dissipa- tion is related to the heat transport properties of these drums by examining the correlations to the quality fac- tor. Figure 3(d) shows a scatter plot of the quality factor of resonance versus the thermal time constant. No signif- icant correlation between the thermal time constant and the quality factor of resonance is found from the experi- V. DISCUSSION A. Comparison to single-layer graphene In Fig. 4 we compare the experimentally obtained val- ues of τ with experimentally obtained values of single layer graphene (data from previous work in ref. [20]) for drums with a 5 µm diameter. From the CDF in Fig. 4(a) it can be seen that both materials have a ther- mal time constant with the same order of magnitude. This is striking because even in the worst case scenario (CVD graphene with a lot of defects, k ≈ 600 W/(m·K)) 02004006008001000τ (ns)MoS2 DataMean (trimmed)α = 4.44 x 10-5 m2/sα = 1.23 x 10-5 m2/s068Diameter (µm)(a)Thermal Diffusivity α (x10-5 m2/s)00.20.40.60.81Cumulative probability8 micronmean α = 1.29 x 10-5 m2/sstd α = 0.45 x 10-5 m2/smean k = 24.7 W/mKstd k = 8.4 W/mK 0125 micronmean α = 1.05 x 10-5 m2/sstd α = 0.49 x 10-5 m2/s mean k = 19.8 W/mKstd k = 9.3 W/mK (c)Thermal conductivity k (W/mK)420.51.510203040001002003004005000246Density (x10-3)Thermal time constant τ (ns)(b)5 micron8 micron02004006008001000τ (ns)01020304050Quality factor5 µm8 µm(d)1020304050Quality factor00.20.40.60.81Cumulative probability5 micronmean = 26.0std = 10.4 8 micronmean = 24.3std = 10.3 (c)102030405060Frequency (MHz)00.20.40.60.81Cumulative probability00.51.522.5Thermal diffusivity (x10-5 m2/s) 024681012ω2xa2 (x105 m2/s2)1.08 micronmean = 39 MHzstd = 15 MHz5 micronmean = 53 MHzstd = 16 MHz 5 µm8 µm(b)(a)0.30.20.10Estimated strain (%)0.45 vices estimated from the resonance frequency is no more than 0.4%, which should result in a spread in the thermal conductivity of approximately 3% [36]. The measured device-to-device spread is significantly larger and strain- dependence is thus not the cause of the observed varia- tions. It should be considered however that the value of f 2a2 could actually show spread between devices due to variations in the mass due to polymer contamination. C. Device-to-device spread The observed device-to-device variations in τ might be attributed to variations in microscopic (point de- fects) and macroscopic imperfections between devices, that could alter the phonon relaxation times between de- vices explaining our result in Fig. 2 (c). From calcula- tions from the literature [34] using the Boltzmann trans- port equation for phonons, we would expect a mean free path of 316.5 nm for naturally occurring MoS2. The sig- nificantly shorter mean free paths (∼20 to 60 nm) found here might be related to our use of CVD MoS2 rather than pristine exfoliated samples. Additional defects can increase the phonon scattering rate, lowering the phonon relaxation time and the mean free path. This could also explain why our experimentally obtained value of k is smaller than other experimental observations, since previous works employed exfoliated membranes [17, 18]. Most of the drums show a higher value of k than previ- ous observations on CVD-grown MoS2 [19], which could be related to differences in quality of the sample. The value of the mean free path shows that λ << a, this sup- ports our notion that heat transport can be described by continuum models in these device. D. Specific heat Given the arguments above, the significant spread in τ is most likely related to the scattering mechanisms. However we cannot fully exclude the possibility that the heat capacity of the drums is responsible for the spread in τ. Little is known about potential mechanisms that can affect the specific heat of single-layered two-dimensional materials due to the lack of experimental data. However, the specific heat is most likely not very different from the bulk material since the number of vibrational degrees of freedom is the same. Also weak temperature dependence of the value of cp is expected since the experiments are performed above the Debye temperature, therefore most degrees of freedom in the lattice are thermalized. What we can conclude is that some of the literature values of k are impossible to have occurred in our mea- surements, since they would violate the Petit-Dulong limit (cp = 468.8 J/(kg·K)). The fastest 5 micron di- ameter drum has τ = 61 ns, which means that there is a limit on the thermal conductivity: k ≤ 48 W/(m·K). For the fastest 8 micron diameter drum, τ = 138 ns and it FIG. 4. (a) Cumulative probabilities from the experimental values of the thermal time constant in this work and for the case of single layer graphene for drums with a 5 micron diam- eter. (b) Empirical distribution functions found by fitting a Kernel distribution with a 30 ns bandwidth to the data. graphene should have a thermal diffusivity at least ten times higher than MoS2. In this previous work on single- layer graphene we attributed the anomalous diameter- dependence of τ to boundary effects that were limiting the heat transport. Since we only measured two diame- ters in this work, we cannot use diameter dependence to draw conclusions. Nevertheless, the values of τ on MoS2 are in good agreement with the theory of diffusive heat transport. This can be seen by comparing the measured values of τ to the theoretical predictions from literature as shown in Fig. 2(a). Any effects of a thermal boundary resistance based on the measurements on MoS2 are too small to be discerned. Molybdenum disulfide has a much lower thermal conductivity than graphene, which means that the intrinsic thermal resistance is more important than thermal resistance at the boundary of the drum, if such a resistance is present at all in the case of MoS2. B. Relation between mechanical and thermal properties We could not uncover any meaningful correlation be- tween strain and the thermal diffusivity from the exper- imental data. The spread in the strain between the de- 0100200300400 τ (ns)00.0020.0040.0060.0080.01Probability00.20.40.60.81Cumulative probabilityMoS2Graphene on goldGraphene on SiO2(a)(b)is impossible that the thermal conductivity of this drum exceeded 55 W/(m·K). Therefore, the highest reported value of k = 84 W/(m·K), [18] cannot have occurred in the drums used in this study. Also, the reported value of k = 34.5 W/(m·K), [17] would implicate that the Petit- Dulong limit is violated in most of the devices. The most representative study, since it uses both CVD MoS2 and conducted the experiment in vacuum, is k = 13.3 ± 1.4 W/(m·K) [19]. Using this value, we can use the experimentally obtained values of τ to estimate the specific heat of MoS2. For the 5 micron drums, we find cp = 278 ± 118 J/(kg·K) and for the 8 micron drums we find cp = 215 ± 73 J/(kg·K). The errors represent the standard deviation due to the large device-to-device spread, nevertheless this analysis suggests that most of the devices have a specific heat that is significantly lower than the bulk value. Future work can combine the tran- sient characterization with existing methods, such as Ra- man spectroscopy or electrical heaters, to extract the thermal resistance R. In that case the heat capacity C can be derived and provide more accurate measurements on the specific heat of 2D materials. The transient char- acterization thus provides a means to perform calorime- try on suspended 2D materials. VI. CONCLUSION We measured the thermal time constants of suspended monolayer molybdenum disulfide drums. In contrast to previous measurements on single layer graphene, we find that the values of τ are in agreement with classi- cal Fourier theory of heat transport. From the values of τ we can estimate the thermal conductivity to be be- 6 tween 10 and 40 W/(m·K), which is lower than previous measurements on exfoliated MoS2 but in agreement with measurements on CVD-grown MoS2. Significant device- to-device variation in thermal time constants is observed. This variation is not correlated to the resonance fre- quency or Q-factor of the membranes, which shows that mechanisms that determine the macroscopic damping are probably not responsible for the observed spread. We therefore conclude that the variations in thermal diffu- sivity are caused by microscopic defects that have a large impact on phonon scattering, but do not affect the reso- nance frequency and damping of the membrane's lowest eigenmode. The method can be used to estimate the specific heat of single layer MoS2, with our results sug- gesting its value might be lower than the bulk value. Fu- ture work can combine this technique with existing ther- mal conductivity measurements to perform calorimetry on suspended 2D materials, enabling one to determine whether the specific heat of 2D materials is equal to its bulk value. ACKNOWLEDGMENTS This work is part of the research programme Inte- grated Graphene Pressure Sensors (IGPS) with project number 13307 which is financed by the Netherlands Or- ganisation for Scientific Research (NWO). The research leading to these results also received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 785219 Graphene Flagship. J.S.B. and D.L. were funded by the National Science Foundation (NSF) grant no. 1706322 (CBET: Bioengineering of Channelrhodopsins for Neurophotonic and Nanophotonic Applications) and Boston University. [1] Kin Fai Mak, Changgu Lee, James Hone, Jie Shan and Tony F. Heinz, "Atomically thin MoS2: A new direct-gap semiconductor," Phys. Rev. Lett. 105, 136805 (2010). [2] Andrea Splendiani, Liang Sun, Yuanbo Zhang, Tianshu Li, Jonghwan Kim, Chi-Yung Chim, Giulia Galli and Feng Wang, "Emerging photoluminescence in monolayer MoS2," Nano Letters 10, 1271–1275 (2010). [3] Goki Eda, Hisato Yamaguchi, Damien Voiry, Takeshi Fu- jita, Mingwei Chen and Manish Chhowalla, "Photolumi- nescence from chemically exfoliated MoS2," Nano Letters 11, 5111–5116 (2011). [4] Andres Castellanos-Gomez, Ronald van Leeuwen, Michele Buscema, Herre SJ van der Zant, Gary A Steele and Warner J Venstra, "Single-layer MoS2 mechanical resonators," Advanced Materials 25, 6719–6723 (2013). [5] Simone Bertolazzi, Jacopo Brivio and Andras Kis, "Stretching and breaking of ultrathin MoS2," ACS Nano 5, 9703–9709 (2011). [6] B Radisavljevic, A Radenovic, J Brivio, V Giacometti and A Kis, "Single-layer MoS2 transistors," Nature Nan- otechnology 6, 147 (2011). [7] Zongyou Yin, Hai Li, Hong Li, Lin Jiang, Yumeng Shi, Yinghui Sun, Gang Lu, Qing Zhang, Xiaodong Chen and Hua Zhang, "Single-layer MoS2 phototransistors," ACS Nano 6, 74–80 (2011). [8] Michele Buscema, Joshua O Island, Dirk J Groenendijk, Sofya I Blanter, Gary A Steele, Herre SJ van der Zant and Andres Castellanos-Gomez, "Photocurrent genera- tion with two-dimensional van der waals semiconduc- tors," Chemical Society Reviews 44, 3691–3718 (2015). [9] Michele Buscema, Gary A Steele, Herre SJ van der Zant and Andres Castellanos-Gomez, "The effect of the sub- strate on the raman and photoluminescence emission of single-layer MoS2," Nano Research 7, 561–571 (2014). [10] Michele Buscema, Maria Barkelid, Val Zwiller, Herre SJ van der Zant, Gary A Steele and Andres Castellanos- Gomez, "Large and tunable photothermoelectric effect in single-layer MoS2," Nano Letters 13, 358–363 (2013). [11] Xiangjun Liu, Gang Zhang, Qing-Xiang Pei and Yong- Wei Zhang, "Phonon thermal conductivity of monolayer MoS2 sheet and nanoribbons," Applied Physics Letters 103, 133113 (2013). 7 [24] David Lloyd, Xinghui Liu, Jason W Christopher, Lauren Cantley, Anubhav Wadehra, Brian L Kim, Bennett B Goldberg, Anna K Swan and J Scott Bunch, "Band gap engineering with ultralarge biaxial strains in suspended monolayer MoS2," Nano Letters 16, 5836–5841 (2016). [25] David Lloyd, Xinghui Liu, Narasimha Boddeti, Lauren Cantley, Rong Long, Martin L Dunn and J Scott Bunch, "Adhesion, stiffness, and instability in atomically thin MoS2 bubbles," Nano Letters 17, 5329–5334 (2017). [26] Robin J Dolleman, Dejan Davidovikj, Herre SJ van der Zant and Peter G Steeneken, "Amplitude calibration of 2D mechanical resonators by nonlinear optical transduc- tion," Applied Physics Letters 111, 253104 (2017). [27] J Scott Bunch, Arend M Van Der Zande, Scott S Ver- bridge, Ian W Frank, David M Tanenbaum, Jeevak M Parpia, Harold G Craighead and Paul L McEuen, "Elec- tromechanical resonators from graphene sheets," Science 315, 490–493 (2007). [28] Robin J Dolleman, Samer Houri, Abhilash Chan- drashekar, Farbod Alijani, Herre SJ van der Zant and Peter G Steeneken, "Opto-thermally excited multimode parametric resonance in graphene membranes," Scientific Reports 8, 9366 (2018). [29] Constanze Metzger, Ivan Favero, Alexander Ortlieb and Khaled Karrai, "Optical self cooling of a deformable Fabry-Perot cavity in the classical limit," Physical Re- view B 78, 035309 (2008). [30] Keith Lewis Aubin, nano/microelectromechanical and nonlinear dynamics studies. University Ithaca, NY, 2004). Radio resonators: frequency Thermal (PhD Thesis, Cornell [31] Joseph Scott Bunch, Mechanical and electrical proper- ties of graphene sheets (PhD Thesis, Cornell University Ithaca, NY, 2008). [32] Kai Liu, Qimin Yan, Michelle Chen, Wen Fan, Yinghui Sun, Joonki Suh, Deyi Fu, Sangwook Lee, Jian Zhou, Se- faattin Tongay, Jie Ji, Jeffrey B Neaton and Junqiao Wu, "Elastic properties of chemical-vapor-deposited mono- layer MoS2, WS2, and their bilayer heterostructures," Nano Letters 14, 5097–5103 (2014). [33] Eric Pop, Vikas Varshney and Ajit K Roy, "Thermal properties of graphene: Fundamentals and applications," MRS Bulletin 37, 1273–1281 (2012). [34] Bo Peng, Hao Zhang, Hezhu Shao, Yuanfeng Xu, Xi- angchao Zhang and Heyuan Zhu, "Towards intrinsic phonon transport in single-layer MoS2," Annalen der Physik 528, 504–511 (2016). [35] Yingye Gan and Huijuan Zhao, "Chirality and vacancy effect on phonon dispersion of MoS2 with strain," Physics Letters A 380, 745 – 752 (2016). [36] Liyan Zhu, Tingting Zhang, Ziming Sun, Jianhua Li, Guibin Chen and Shengyuan A Yang, "Thermal conduc- tivity of biaxial-strained MoS2: sensitive strain depen- dence and size-dependent reduction rate," Nanotechnol- ogy 26, 465707 (2015). [12] Yongqing Cai, Jinghua Lan, Gang Zhang and Yong-Wei Zhang, "Lattice vibrational modes and phonon thermal conductivity of monolayer MoS2," Physical Review B 89, 035438 (2014). [13] Xiaolin Wei, Yongchun Wang, Yulu Shen, Guofeng Xie, Huaping Xiao, Jianxin Zhong and Gang Zhang, "Phonon thermal conductivity of monolayer MoS2: A comparison with single layer graphene," Applied Physics Letters 105, 103902 (2014). [14] Wu Li, Jesús Carrete and Natalio Mingo, "Thermal con- ductivity and phonon linewidths of monolayer MoS2 from first principles," Applied Physics Letters 103, 253103 (2013). [15] Jin-Wu Jiang, Harold S Park and Timon Rabczuk, "Molecular dynamics simulations of single-layer molyb- denum disulphide (MoS2): Stillinger-weber parametriza- tion, mechanical properties, and thermal conductivity," Journal of Applied Physics 114, 064307 (2013). [16] Nicholas A Lanzillo, A Glen Birdwell, Matin Amani, Frank J Crowne, Pankaj B Shah, Sina Najmaei, Zheng Liu, Pulickel M Ajayan, Jun Lou, Madan Dubey, Saroj K Najak and Terrance P O'Regan, "Temperature- dependent phonon shifts in monolayer MoS2," Applied Physics Letters 103, 093102 (2013). [17] Rusen Yan, Jeffrey R Simpson, Simone Bertolazzi, Ja- copo Brivio, Michael Watson, Xufei Wu, Andras Kis, Tengfei Luo, Angela R Hight Walker and Huili Grace Xing, "Thermal conductivity of monolayer molybdenum disulfide obtained from temperature-dependent Raman spectroscopy," ACS Nano 8, 986–993 (2014). [18] Xian Zhang, Dezheng Sun, Yilei Li, Gwan-Hyoung Lee, Xu Cui, Daniel Chenet, Yumeng You, Tony F Heinz and James C Hone, "Measurement of lateral and inter- facial thermal conductivity of single-and bilayer MoS2 and MoSe2 using refined optothermal Raman technique," ACS Applied Materials & Interfaces 7, 25923–25929 (2015). [19] Jung Jun Bae, Hye Yun Jeong, Gang Hee Han, Jaesu Kim, Hyun Kim, Min Su Kim, Byoung Hee Moon, Seong Chu Lim and Young Hee Lee, "Thickness- dependent in-plane thermal conductivity of suspended MoS2 grown by chemical vapor deposition," Nanoscale 9, 2541–2547 (2017). [20] Robin J Dolleman, Samer Houri, Dejan Davidovikj, San- tiago J Cartamil-Bueno, Yaroslav M Blanter, Herre SJ van der Zant and Peter G Steeneken, "Optomechanics for thermal characterization of suspended graphene," Physi- cal Review B 96, 165421 (2017). [21] D Sercombe, S Schwarz, O Del Pozo-Zamudio, F Liu, BJ Robinson, EA Chekhovich, II Tartakovskii, O Kolosov and AI Tartakovskii, "Optical investigation of the natural electron doping in thin MoS2 films deposited on dielectric substrates," Scientific Reports 3, 3489 (2013). [22] Nils Scheuschner, Oliver Ochedowski, Anne-Marie Kaulitz, Roland Gillen, Marika Schleberger and Janina Maultzsch, "Photoluminescence of freestanding single- and few-layer MoS2," Physical Review B 89, 125406 (2014). [23] Hong Li, Qing Zhang, Chin Chong Ray Yap, Beng Kang Tay, Teo Hang Tong Edwin, Aurelien Olivier and Do- minique Baillargeat, "From bulk to monolayer MoS2: evolution of Raman scattering," Advanced Functional Materials 22, 1385–1390 (2012). SUPPLEMENTAL INFORMATION COMSOL model 8 Here we show the COMSOL model used to derive the expression for τ used in the main section of the paper. An analytic expression was derived for the thermal time constant τ in the case of a uniformly heated circular disk [30]: τ = a2ρcp µ2k , (8) where a is the drum radius, ρ the density, cp the specific heat, k the thermal conductivity the material and µ = 2.4048 is the first root of the Bessel function J0(x). Since the experiment uses a laser spot with a size that is much smaller than the drum diameter to heat the drum, equation 8 needs to be modified in order to accurately describe the time constant of the system. Our approach is to choose a fixed value of the specific heat cp = 373.5 J/(kg·K) and vary both a and k to find a new value of µ that will enable us to accurately determine the value of k or the thermal diffusivity α from the experiment. Figure 5 shows the setup of the COMSOL simulation in order to find the thermal time constant τ. A simple circular domain was defined and the heat transport is simulated using the "heat transport in thin shells" module. A point source in the center was used to simulate the heat flux and the boundaries of the domain were kept at a fixed temperature. In order to find the time constant τ, a time-dependent simulation was performed that simulates the response to a step function in the heat source. The resulting time dependent temperature increase was calculated by taking the average over the entire domain. This results in the time-dependent traces shown in Fig. 6. For each trace, the time constant is found by fitting: T (t) = T0 + Tend exp (−t/τ ). (9) The diameter-dependence is simulated using a range of values of k, we selected a suitable range by selecting values found in literature. It is found that for the range between 10 < k < 100 we find values of µ2 ≈ 5.0. This model is shown as solid lines in Fig. 7. We find that µ2 = 5.0 yields good agreement at the higher values of k. Low values of k results in larger deviations at larger diameters, which can also be attributed to the a2 dependence of τ. From this model, we find that the value of τ with µ2 = 5.0 should produce the correct value of k or α within 10% error as long as 15 < k < 100 W/(m·K) and 2a ≤ 8 µm. FIG. 5. (a) Schematic drawing of the simulation. (b) Temperature profile at the end of the simulation for a 5 micron diameter drum. Point heat sourceThin conductive layerFixed temperatureTmaxTmin(a)(b)9 FIG. 6. Simulated temperature as function of time, from which τ can be derived for different diameters. The temperature is calculated using the average value over the drum surface. FIG. 7. Diameter dependence of τ for different diameters and different values of k. 1101001000Time (ns)1 ∆T/∆Tend00.51 µm, τ = 3.68 ns 2 µm, τ = 14.8 ns 3 µm, τ = 33.5 ns 4 µm, τ = 59.8 ns 5 µm, τ = 94.0 ns 6 µm, τ = 136 ns 7 µm, τ = 188 ns 8 µm, τ = 250 ns Diameter (mm)0100200300400500600700800Thermal time constant (ns)0123456789k = 10 W/(m K)k = 25 W/(m K)k = 50 W/(m K)k = 100 W/(m K)

1602.03432

2

1602

On Cherenkov Friction and Radiation of a Neutral Polarizable Particle Moving Parallel to a Transparent Dielectric Plate

We have obtained general expressions for the intensity of radiation, decelerating force, the rate of heating and acceleration of a small polarizable particle under the conditions of Cherenkov friction: at relativistic motion parallel to the surface of thick transparent dielectric plate. Comparison with the results of other authors is given.

cond-mat.mes-hall

cond-mat

Cherenkov Friction and Radiation of a Neutral Polarizable Particle Moving near a Transparent Dielectric Plate G. V. Dedkov and A. A. Kyasov Kabardino-Balkarian State University, Nalchik, Russia E-mail: gv_dedkov@mail.ru Received: January 10, 2017 Abstract. We have obtained general expressions for the intensity of radiation, decelerating force, the rate of heating and acceleration of a small polarizable particle under the conditions of Cherenkov friction: at relativistic motion parallel to the surface of thick transparent dielectric plate. Comparison with the results of other authors is given. Casimir friction or quantum friction is the effect of quantum electrodynamics caused by quantum fluctuations of the electromagnetic field at relative movement in a vacuum of parallel smooth surfaces of two bodies or when a small particle moves parallel to the surface of an extended body (plate) [1, 2]. The processes associated with quantum and "thermal" (at a finite temperature of bodies) friction have aroused great interest of researchers during recent decades (see [3-5] and references). An important consequence of quantum friction is the possibility of electromagnetic radiation , where V is the particle velocity (hereinafter we will under the Cherenkov condition consider just this case), c is the speed of light in vacuum, and n the refractive index of an immovable body (transparent dielectric plate) [6-8]. According to the interpretation of [7], the physical cause of radiation is an anomalous Doppler effect, at which the frequency of electromagnetic quanta absorbed by a particle is negative in the rest frame of the particle. As a result, the particle is excited into a quantum state with a higher energy and emits a photon within the Cherenkov cone. Unlike works [7, 8], in which the intensity of radiation was calculated using another method, we will use the general results of our theory of fluctuation-electromagnetic interaction of a relativistic particle moving near a surface [9, 10]. This makes it possible better understanding of the electrodynamic essence of the problem and to find the important relations between electrodynamical, mechanical and ncV />  2 termal quantities. Our final expression for the intensity of radiation coincides with [7], but it differs somewhat from [8]. In contrast to [7], there are some differences in the relationship between radiation force, radiation power and the rate of change of the rest mass in the reference frame @ of an immovable plate. We use the geometric configuration shown in the figure. The nature of the basic electrodynamic relationships with these quantities is determined by equations ∫ , (1) VBmEd & BmEd ( +⋅∇⋅ VFQ x+ & +⋅ Ej ⋅ rd 3 = + = ) & ⋅ ⋅ Q & =′ ∫ Ω′ Ej ′ ′ rd 3 =′ 2 γ BmEd & +⋅ ⋅ & = 2 γ Q & (2) − dW dt = ∫ σ S ⋅ d σr + ∫ Ω Ej ⋅ rd 3 (3) where γ is the Lorentz-factor, W = ( Eπ )8/1( ∫ Ω 2 + 2 B ) rd 3 is the energy of the field inside the volume Ω that is restricted by the wave surface surrounding the particle (see figure), is the Pointing vector, d and m are the fluctuation electric and magnetic S )4/( πc HE × = dipole moments of the particle, E, B, H and j are the fluctuation-electromagnetic fields and the electric current density; angular brackets and points above dipole moments denote complete quantum-statistical averaging and time differentiation. All unprimed variables refer to the frame is the heating rate of a particle in its frame of of reference Σ . In formula (2) tdQdQ & =′ / ′ ′ Q =& reference Σ′ . It should be noted that the quantity has an independent meaning and coincides with the heating rate of the particle or surface only in the nonrelativistic limit. In the quasistationary case I (4) from (3) it follows that =dt Ej ⋅ d σr rd 3 dW −= dQ dt S = 0 / / ⋅ ∫ σ ∫ Ω Moreover, from (1) and (4) it follows I dQ −= dt VF x+ ) ( / (5)  3 Formulas (1)-(5) were previously used in our calculations of the radiation of moving and rotating particles in a vacuum [11-13]. In the used configuration when the plate is transparent, xF and Q& all the above arguments also remain valid, and one can use the general results for presented in [9, 10] kdk y x + γωα ( ′′ i F x −= coth ⋅ ⎡ ⎢ ⎣ ∞ ∫ γ h 2 2 π ⎛ ⎜⎜ ⎝ 0 ω h Tk B 2 2 +∞ ∫ d ω ∞− dk x +∞ ∫ ∞− ⎛ + ωγ h ⎜⎜ Tk 2 ⎝ B 1 ⎞ −⎟⎟ ⎠ coth ∑ , mei = ⎤ ⎞ ⎟⎟ +⎥ ⎠ ⎦ (...) dQ dt ⎡ ⎢ ⎣ ⋅ = ∞ ∫ γ h 2 2 π ⎛ ⎜⎜ ⎝ 2 0 ω h Tk B 2 coth +∞ ∫ d ω ∞− dk x +∞ ∫ dk + ∑ + γωαω y ( ′′ i ∞− ⎛ + ωγ h ⎜⎜ Tk 2 ⎝ B 1 , mei = ⎤ ⎞ ⎟⎟ +⎥ ⎠ ⎦ ⎞ −⎟⎟ ⎠ coth (...) ⎛ Im) ⎜⎜ ⎝ exp( zq 0 ) 2 − q 0 R i , ( ω k ) ⎛ Im) ⎜⎜ ⎝ exp( − q 2 zq 0 ) 0 R i , ( ω k ) ⎞ ⎟⎟ ⎠ ⎞ ⎟⎟ ⎠ (6) (7) where =+ ωω , Vkx+ 1T and 2T are the local temperatures of particle (in Σ′ ) and the plate (in are the imaginary components of the electric and magnetic polarizability, the terms Σ ), ) (, ωα me′′ (…) describe the interaction with vacuum modes of the electromagnetic field in the absence of plate, and make no contribution to further results. The auxiliary quantities in (6) and (7) are given by the expressions R ωk ( , − e ( / 2 ωβω ∆+ / 2 ωβ /) 2 2 1)( − ( + ω + k x ck 2 ( + ω /) 2 ] + + c e 2 ) ) 2 2 2 2 2 ) ∆= [ k 2) y ck 2 ]2 c m (8) [ k (2) ( ω 1( 2 − [ ( (2) ω 1( 2 − R k ωk ( , m ( 2 ωβω ∆+ ∆= 2 k ) [ 2) m e y 2 2 2 k − x ck / 22 2 ωβ ) /) 2 1)( − ( + ω + ) + ( + ω /) 2 c 2 ] + / c ck 22 ]2 (9) ∆ e ) ( ω = q 0 = 2 ( k q q − ) ( ωε 0 ) ( ωε 0 c / 2 ω , q − q + ) 2/12 ∆ m ) ( ω = 2 , k = 2 k x q q + 0 ( ( 2 ) ωµ ) ωµ 0 k y − + q q , q = ( k 2 − 2 ( ) ωµωεω () c ( ) / 2 2/1 ) , (10) In the case of transparent dielectric ) ( ωε = 2 n Im, take the form ( ωµωε ,0 = ( ) 1) = , and coefficients ) (, ωme∆  4 ∆ e ) ( ω = 2 2 n n 2 2 k k − − 2 ω 2 ω / / 2 2 c c − + 2 2 k k − − n n 2 2 ω 2 2 ω / / 2 2 c c , ∆ m ) ( ω = 2 2 k k − − 2 ω 2 ω / / 2 2 c c − + 2 2 k k − − n n 2 2 ω 2 2 ω / / 2 2 c c (11) Making use the limiting transitions relations coth( ω 2/ Tk B 2 ) h → sign ), ( ω T 1 coth( → T ,0 2 + ω h → 0 in (6), (7) and taking into account the 2/ Tk B 1 ) → sign ( + ω ) one obtains F x = 2 γ h 2 π ∞ ∫ 0 d βθω − n ( )1 n / ω c kdk x V ∫ / ω n 2 2 ω x 2 x y ⋅ 2 k c − dk / ∫ 0 − γωα ( ′′ i ∑ mei = , exp( ⎛ Im) ⎜⎜ ⎝ zq 00 ) R i ( , ω − k ) x 2 − q 0 (12) ⎞ ⎟⎟ ⎠ Q & = 2 γ h 2 π ∞ ∫ 0 d βθω − n ( )1 n / ω c dk V ∫ n 2 2 ω x 2 k c − dk / ∫ 0 / ω 2 x − − γωαω ( ′′ i y ∑ mei = , exp( ⎛ Im) ⎜⎜ ⎝ zq 00 ) R i ( , ω − k ) x 2 − q 0 (13) ⎞ ⎟⎟ ⎠ where R i −ω coincides with ( ) k , x kωmeR , (, ) at x kk−=k , ( y ) and =− ωω Vk x− . Using (5), (12) and (13) we find I −= 2 γ h 2 π ∞ ∫ 0 d βθω − n ( )1 n / ω c dk V ∫ n 2 2 ω x 2 k c − dk / ∫ 0 2 x y / ω − γωαω ( ′′ i ∑ mei = , exp( ⎛ Im) ⎜⎜ ⎝ zq 00 ) R i ( , ω − k ) x 2 − q 0 (14) ⎞ ⎟⎟ ⎠ −)(xθ where is the Heaviside step-function. Formulas (12) and (14) coincide with the analogous formulas in [7] with the difference that (12) and (14) also include the contribution of magnetic polarizability of the particle. The limits of integration over the wave vectors in (12)- ) (ωαm′′ 0 ( ( ) 0 q = < /) =′ Vk x exp( −ωγ 0>I and and zq− 2 00 ) , ωα me′′ ωγω − properties of functions , from (12)-(14) it follows that (14) correspond to the condition of the anomalous Doppler effect, since in the rest frame of . Due to analytical particle Σ′ the photon frequency is negative: 0<xF 0>Q& . This means that the conversion of the kinetic energy of particle into radiation proves to be the dominant mechanism in the process of Cherenkov friction (due to (5)). In this case, the particle itself heats up (see (2)). It is worth noting that formulas (12)-(14) are also valid in the case of refractive index dispersion: , if we assume that integration by frequencies is performed in the domain of transparency of the dielectric plate, Let us focus briefly on the difference from the results in [7] as regards formula (13) and the equation of particle dynamics. With allowance for (2), the change in the particle rest mass caused by radiation can be written in the form ) ′′ ωε (ωn ) n = . = , 0 ( = Q & c 2 (15) Q ′ & γ= 2 c 2 dm td ′ Taking into account dm dt We compare this result with equation (16) in [7], which in our designations has the form = γ , from (15) it follows (16) Q &γ= 2c dt td 5 ′ FV ⋅− x += I dm td γ ′ (17) From (17) it then follows dm dt In contrast to (18), formula (13) in [7] for ( FV −= x + )I ⋅ hand side. In this case, according to [7], dm / dt = (18) dm / dt contains an additional γ-factor in the right , where µu and µF are the components µFu µ of four-vectors of velocity and force, and from (18), (17) and (5) it follows (cf. with (16)) dm dt (19) Q &= 2c The observed difference becomes significant at Cherenkov radiation is maximal and may affect the solution of the dynamics equation, 1>>γ , i. e. precisely when the intensity of d dt ⎛ ⎜ ⎜ ⎝ mc β 1 β − 2 ⎞ ⎟ ⎟ ⎠ = xF (20) which with allowance for (16) reduces to the form [12] 3 γ dmc β dt = F x − βγ Q &2 c (21) The right side of (21) can be simplified with allowance for the relation between the forces xF xF ′ defined in the frames of reference Σ and Σ′ . The corresponding relation is obtained and upon the time differentiation of the Lorentz transformation for the particle momentum F =′ x F x γ− dmV dt (22)  From (16), (21) and (22) it follows that 6 F =′ x F x Q &2βγ− c (23) mc γ3 d β dt xF ′= (24) and the explicit expression for the force xF ′ is obtained when substituting (12) and (13) into (23) F =′ x 2 2 γ h 2 π ∞ ∫ 0 d βθω − n ( )1 n / ω c dk V ∫ / ω n 2 2 ω ( k x x − / βω c ) 0 2 2 x / k c − ∑∫ dk y mei = , − γωα ( ′′ i exp( ⎛ Im) ⎜⎜ ⎝ zq 00 ) R i ( , ω − k ) x 2 − q 0 (25) ⎞ ⎟⎟ ⎠ k x βω− c . In contrast to this, if we substitute Eq. (19) (instead (16)) into dynamics equation (20), then xF ′ obtained with allowance for (12), (13), (19) and (20) will differ from (25) the formula for within the integrand expression in (25) is replaced by since the factor c )/ ( 2 ) − 1 +− ( / βωβγ k x In conclusion, it is worth noting that formulas (12) and (14) for the Cherenkov friction force and radiation power of a particle fully agree with results [7] obtained in an independent way, but the four-vector of force introduced in [7] leads to a discrepancy with the equation of dynamics, which follows from our analysis. References [1] J. B. Pendry, J. Phys.: Condens. Matter, 9, 10301 (1997); New J. Phys. 12, 033028 (2010). [2] K. A. Milton, J. S. Høye, and I. Brevik, Symmetry 8 (5), 29 (2016). [3] M. Kardar and R. Golestanian, Rev. Mod. Phys. 71, 1233 (1999). [4] P. V. W. Davies, J. Opt. B.: Quant. Semiclass. Opt. 7(3), S40 (2005). [5] F. Intravaia, C. Henkel, and M. Antezza, Lect. Notes Phys. 834, 345 (2011). [6] M. F. Maghrebi, R. Golestanian, and M. Kardar, Phys. Rev. A88, 042515 (2013); Phys. Rev. A90, 012515 (2014). [7] G. Pieplow and C. Henkel, J. Phys.: Condens. Matter 27, 214001 (2015).  [8] A. I. Volokitin, JETP Lett. 104 , 504 (2016). 7 [9] G. V. Dedkov, A. A. Kyasov, J. Phys.: Condens. Matter 2008. V. 20. P. 354006. [10] G. V. Dedkov, A. A. Kyasov, Surf. Sci. 604, 562 (2010). [11] G. V. Dedkov, A. A. Kyasov, Phys. Scripta 8, 105501 (2014). [12] G. V. Dedkov, A. A. Kyasov, Int. J. Mod. Phys. B29, 1550237 (2015). [13] G. V. Dedkov, A. A. Kyasov, Tech. Phys. Lett. 42, 8 (2016). The geometric configuration and reference frames Σ and Σ′ corresponding to the dielectric plate and particle. The dashed line shows the boundary of the wave zone of radiating particle. 

1807.01277

1

1807

Inversion-symmetry protected chiral hinge states in stacks of doped quantum Hall layers

We prove the existence of higher-order topological insulators with protected chiral hinge modes in quasi-two-dimensional systems made out of coupled layers stacked in an inversion-symmetric manner. In particular, we show that an external magnetic field drives a stack of alternating p- and n-doped buckled honeycomb layers into a higher-order topological phase, characterized by a non-trivial three-dimensional ${\mathbb Z}_2$ invariant. We identify silicene multilayers as a potential material platform for the experimental detection of this novel topological insulating phase.

cond-mat.mes-hall

cond-mat

Inversion-symmetry protected chiral hinge states in stacks of doped quantum Hall layers Sander H. Kooi,1 Guido van Miert,1 and Carmine Ortix1, 2 1Institute for Theoretical Physics, Center for Extreme Matter and Emergent Phenomena, Utrecht University, Princetonplein 5, 3584 CC Utrecht, Netherlands 2Dipartimento di Fisica "E. R. Caianiello", Universit´a di Salerno, IT-84084 Fisciano, Italy (Dated: July 4, 2018) We prove the existence of higher-order topological insulators with protected chiral hinge modes in quasi-two-dimensional systems made out of coupled layers stacked in an inversion-symmetric manner. In particular, we show that an external magnetic field drives a stack of alternating p- and n-doped buckled honeycomb layers into a higher-order topological phase, characterized by a non-trivial three-dimensional Z2 invariant. We identify silicene multilayers as a potential material platform for the experimental detection of this novel topological insulating phase. Introduction – A free-fermion symmetry protected topological (SPT) insulator is a quantum state of matter that cannot be adiabatically deformed to a trivial atomic insulator without either closing the insulating bulk band gap or breaking the protecting symmetry [1, 2]. Its topo- logical nature is reflected in the general appearance of gapless boundary modes in one dimension lower [3, 4]. However, when the protecting symmetry is a crystalline symmetry the gapless boundary modes appear only on surfaces which are left invariant under the protecting symmetry operation [5, 6]. Most importantly, these gap- less boundary modes are "anomalous": on a single sur- face the number of fermion flavors explicitly violates the fermion doubling theorem [7] or stronger version of it [8]. Very recently, it has been shown that point-group sym- metries can stabilize insulating states of matter in bulk crystals with conventional gapped surfaces, but with gap- less modes at the hinges connecting two surfaces related by the protecting crystalline symmetry [8–24]. For sys- tems of spinless electrons (negligible spin-orbit coupling) the hinge modes are chiral. Hence, they represent anoma- lous one-dimensional (1D) modes – they cannot be en- countered in any conventional 1D atomic chain – but now embedded in a three-dimensional crystal. These novel topological crystalline insulators, which have been dubbed higher-order topological insulators, have started to be classified in systems possessing different crystalline symmetries, including rotational and rotoinversion sym- metries [9, 10, 20, 22, 25]. In inversion-symmetric crystals, higher-order topolog- ical insulators can also exist [18]. However, in this case, inversion symmetry-related surfaces are connected via two hinges, one of which will host a chiral gapless mode. This, in turn, gives rise to an additional modulo two am- biguity in the microscopic hinge location of the chiral modes, reminiscent of the ambiguity in the Fermi arcs connectivity of Weyl semimetals [26]. The aim of this Letter is to show that such an inversion-symmetry protected higher-order topological insulator can be in principle obtained in stacks of doped buckled honeycomb layers (e.g. silicene [27]) subject to an external magnetic field. Two factors conspire to render this possible. First, the quantum Hall states in p- and n-doped honeycomb layers generally have oppo- site sign of the Hall conductance and hence are char- acterized by opposite Chern numbers C [28]. Second, in a simple AA stacking configuration, buckled honey- comb layers intrinsically break both the reflection sym- metry in the stacking direction and the twofold rota- tion about the stacking direction, but still preserve the three-dimensional bulk inversion symmetry. We first give an intuitive argument for the existence of an inversion- symmetric higher-order topological insulator in stacks of Chern insulators with alternating C = ±1 integer invari- ants, and show that this insulating phase can be derived from a parent mirror Chern insulator [29] by adding crys- talline symmetry-breaking terms. Then, we introduce a corresponding minimal tight-binding model consisting of quantum anomalous Hall layer stacks [30], and ver- ify its topological nature by computing the correspond- ing bulk Z2 invariant [25]. Finally, we perform a full Hofstadter [31] calculation in three-dimensions [32] for buckled honeycomb layers to show the existence of topo- logically protected chiral hinge states. Effective surface theory – Let us start out with an ef- fective low-energy approach for stacks of Chern insulating layers of alternating C = ±1 integer topological invariant [c.f. Fig. 1(a)]. At any edge perpendicular to the stack- ing direction, each Chern insulating layer is characterized by an anomalous chiral edge mode whose dispersion can be considered to be linear. For completely uncoupled layers, the effective surface Hamiltonian in the primi- tive two-layer surface unit cell then reads H0 = kxσz, where the Pauli matrix acts in the layer space and we explicitly considered a (010) surface. We next introduce an interlayer coupling between consecutive layers with a coupling strength, which, for simplicity, we assume to be real. The effective surface Hamiltonian is then modified accordingly to HS = H0 +t [1 + cos (kz)] σx−t sin (kz)σy. It preserves mirror symmetry in the stacking direction 8 1 0 2 l u J 3 ] l l a h - s e m . t a m - d n o c [ 1 v 7 7 2 1 0 . 7 0 8 1 : v i X r a 2 stood by considering the effective surface Hamiltonian as a collection of one-dimensional Rice-Mele models [34, 35], parametrized by kx. For a chain with an odd number of sites, the latter displays an in-gap boundary state at an energy corresponding precisely to the staggered chemical potential ≡ kx. If we assume the dimerization pattern to be equivalent at all the four surfaces perpendicular to the layer planes, the hinge modes at the boundary layer will be connected to create a circulating planar current, which is in agreement with the fact that a dimerized stack of an odd number of layers defines a (thicker) two-dimensional Chern insulating state [36]. Let us now instead assume that the dimerization pat- terns at the two opposite surfaces (010) and (0¯10) are de- signed to be opposite to each other as shown in Fig. 1(a). Although still breaking mirror symmetry along the stack- ing direction, this configuration preserves bulk inversion symmetry with the inversion center lying at the center of one layer. The presence of inversion symmetry also implies that the chiral hinge modes related to the (010) and (0¯10) surfaces will be localized on opposite bound- ary layers. The same clearly holds true at the (100) and (¯100) surfaces. Moreover, the Jackiw-Rebbi mech- anism [37] guarantees the existence of an additional chi- ral hinge mode at two inversion- symmetry related hinges between the x and y planes, and thus the configuration of in-gap hinge states schematically shown in Fig. 1(c) is re- alized. The latter represents nothing but the hallmark of a second-order topological insulator in three-dimensions protected by inversion symmetry. topological Quantum anomalous Hall stacks – Having presented our coupled-layer low-energy approach, we next intro- duce a microscopic model that features an inversion- symmetric higher-order insulating state. Specifically, we consider stacks of honeycomb layers, each of which possesses chiral orbital currents leading to a quantum anomalous Hall (QAH) insulating state [30]. In order to have alternating Chern numbers on the honey- comb layers, we further assume the direction of the or- bital currents to be opposite in two consecutive layers. For uncoupled layers, the corresponding tight-binding Hamiltonian reads H(cid:107) = −t † iαcjα − it2 c † (−1)ανijc iαcjα, † where c iα (ciα) creates (annihilates) an electron on site i in layer α, t is the intralayer nearest-neighbor hopping amplitude and t2 is the next-nearest neighbor hopping amplitude. As usual, the factor νij = 1 if the next- nearest neighbor hopping path rotates counterclockwise, and −1 if it rotates clockwise. We next introduce an in- terlayer coupling that explicitly breaks the mirror sym- metry in the stacking direction but still preserves bulk inversion symmetry with the inversion center in one layer (cid:88) (cid:104)i,j(cid:105),α (cid:88) (cid:104)(cid:104)i,j(cid:105)(cid:105),α Figure 1. a) Sketch of stacked Chern insulators with alternat- ing Chern numbers. We also indicate the effective inversion- symmetric coupling between the chiral edge states. b) Corre- sponding surface energy spectrum for mirror symmetric cou- plings (t1 = t2) and inversion-symmetric couplings (t1 (cid:54)= t2). c) Schematic figure of the inversion symmetric hinge states in a cube geometry. M(kz) = diag(cid:0)1, e−ikz(cid:1). The mirror symmetry con- around one layer [29] with the reflection operator that acquires an explicit momentum dependence and reads straint M(kz)HSM(kz)−1 ≡ HS (kz → −kz) implies a decoupling of the chiral modes on the kz = π line. Hence, the three-dimensional system is characterized by gapless surfaces with a mirror-symmetry protected single Dirac cone with the Dirac point at {kx, kz} = {0, π}. The presence of this single surface Dirac cone can also be seen as a consequence of the fact that the bulk three- dimensional Hamiltonian is characterized by a non-zero mirror Chern number [33] CM = 1 at kz = π. When considering systems with a finite number of layers still preserving reflection symmetry in the stacking direction – this constraint is fulfilled for stacks with an odd num- ber of layers – the surface spectrum in the remaining translational invariant kx direction will therefore display a single uncoupled chiral mode as schematically shown in Fig. 1(b). We emphasize that this chiral anomaly is regularized by the presence of a chiral mode partner with opposite chirality at the opposite (0¯10) surface. We next "trivialize" our system by introducing a per- turbation that explicitly breaks the mirror symmetry along the stacking direction. To this end, we consider a dimerization pattern in the interlayer coupling which fur- ther modifies the effective surface Hamiltonian as HS = H0 + [t1 + t2 cos (kz)] σx − t2 sin (kz)σy [c.f. Fig. 1(a)]. This, in turn, implies that the single surface Dirac cone acquires a mass ∝ t1 − t2 and the surface becomes a conventional gapped one. When considering, as before, a finite system with an odd number of layers there will be a single chiral mode traversing the full surface gap [c.f. Fig. 1(c)] localized at one of the two boundary layers depending upon the specific dimerization pattern. The existence of this chiral hinge mode can be under- 3 Figure 2. a) Edge spectra of the QAH model with peri- odic boundary conditions in the x- and z-direction for stacks of zigzag-terminated honeycomb flakes. The tight-binding Hamiltonian parameters have been chosen as t2/t = 0.2 and tz/t = 0.3. b) The hexagonal 3D Brillouin zone with the inversion symmetric points labeled. c) Table specifying the number of bands with inversion eigenvalue +1 and -1 at the inversion symmetric points in the Brillouin zone. Figure 3. Energy versus magnetic flux φ (measured in units of the magnetic flux quantum φ0) for the Hofstadter model on a honeycomb lattice. The gaps are coloured according to their Chern number. The two largest gaps have C = ±1. amplitudes, will gap the surface Dirac cone leading to the observation of chiral hinge modes even for stacks of armchair terminated ribbons. (cid:88) at the center of the bond between the two A and B hon- eycomb sublattices. In its simplest form the interlayer Hamiltonian is then H⊥ = − tz 2 † ×c iαciα+1 + h.c. (cid:2)1 − eiπα(cid:3) c (cid:2)1 + eiπα(cid:3) † iαciα+1 − tz 2 (cid:88) i∈B,α i∈A,α (1) As discussed below, this interlayer Hamiltonian is nat- urally realized assuming buckled honeycomb layers. Fig. 2(a) shows the edge energy spectrum as obtained by diagonalizing the full Hamiltonian H = H(cid:107) + H⊥ for stacks of zigzag terminated ribbons with an odd num- It agrees perfectly with the foregoing ber of layers. low-energy description. Inside the gapped bulk energy bands, we clearly observe conventional surface states, corresponding to a massive surface Dirac cone, in the gap of which two chiral hinge modes localized on opposite layers appear. Precisely the same features occur consid- ering periodic boundary conditions in the stacking direc- tion and zigzag terminations in the other two directions. We point out that we excluded from our analysis ribbons with armchair terminations since the latter would yield an unprotected single massless surface Dirac cone. This is due to the fundamental difference between the chiral edge states of a QAH insulator for zigzag and armchair termi- nated ribbons [38]. For the latter, the chiral edge states have an equal amplitude on both the two honeycomb sublattices. The interlayer coupling Hamiltonian intro- duced above thus yields an effective mirror-symmetric coupling between the QAH chiral edge states. However, additional symmetry- allowed terms in the bulk Hamilto- nian, e.g. intralayer real next- nearest-neighbor hopping To prove the topological origin of these chiral hinge states, we have calculated the bulk Z2 topological invari- ant for a second-order topological insulator with inver- sion symmetry [25]. It can be derived using the bulk formulation of the quantized corner charges for the ef- fective two-dimensional inversion-symmetric Hamiltoni- ans at kz = 0, π, and thereafter considering the corre- sponding corner charge flow. When expressing the cor- ner charges in terms of the multiplicities of the inversion symmetry eigenvalues ±1 of the occupied bands at the inversion-symmetric momenta of the 3D Brillouin zone [c.f. Fig. 2(b)], we then find the following expression for the bulk Z2 invariant: (cid:20) ν = 1 2 Γ−1 + −Γ1 − 1 2 A−1 − (L1)−1 − 1 2 1 2 (M1)−1 + +A1 + (L2)−1 − 1 2 (L3)−1 mod 2 (2) 1 2 (M2)−1 + 1 2 (M3)−1 (cid:21) A non-trivial value ν = 1 mod 2 of this invariant guar- antees the presence of chiral hinge modes provided the bulk and the surfaces are gapped. For our model at half-filling, the inversion symmetry labels [c.f. Fig. 2(c)] directly imply an higher-order topology. Therefore the chiral hinge states shown in Fig. 2(a) are the direct man- ifestation of a bulk-hinge correspondence. Stacks of quantum Hall layers – With these results in hand, we next introduce our main result: the possibil- ity to engineer an inversion symmetry protected second- order topological insulator in stacks of doped quantum Hall layers with a buckled honeycomb geometry. To show 4 Figure 4. a) Spectrum along the zigzag direction, all spectra are for tz/t = 0.5, φ/φ0 = 0.2, φ2/φ0 = 0.04, φ3/φ0 = 0.04, and V0 = 0.6t, where φ/φ0, φ/φ1 and φ/φ2 are the flux per plaquette in the z, x and y direction respectively. b) Spectrum along the armchair direction. c) Spectrum along the stacking direction. d) Schematic side-view of an AA stacked buckled honeycomb lattice. In the first layer the A sublattice is connected to the next layer, while in the second layer the B sublattice is connected to the next layer. e) Table showing the multiplicity of inversion eigenvalues at half-filling at inversion-symmetric points in the Brillouin zone for φ/φ0 = 1/3, V0 = 0.8t. this, we first recall that the quantum Hall effect on the honeycomb lattice exhibits a zeroth Landau level above (below) which the total Chern number C = +1(−1) for relatively weak magnetic fluxes per plaquette [see Fig. 3]. This also implies that the sign of the Hall conductance is opposite in p- and n-doped layers. We then take ad- vantage of this property to realize a quantum Hall ana- logue of the quantum anomalous Hall model introduced above. For the intralayer part of the Hamiltonian we thus consider layers with an alternating bias ±V0. The corresponding Hamiltonian is then H(cid:107) = −t † c i,αcj,α + V0 † (−1)α c i,αci,α. (cid:88) (cid:104)i,j(cid:105),α (cid:88) i,α The effect of the perpendicular magnetic field is taken t ei(cid:82) ds A, where ds is the line integral between the bonds into account via the usual Peierls substitution t → and we took the vector potential in the Landau gauge A = −(eB/)(y, 0, 0) with e the electron charge and B the magnetic field strength. The interlayer Hamiltonian is taken to be the same as in Eq. (1) to account for the buckling of the layers [c.f. Fig. 4d)]. We have first verified the non-trivial topology of this microscopic tight-binding model by computing the mul- tiplicities of the inversion symmetry eigenvalues at the inversion-symmetric momenta of the BZ [see Fig. 4(e)]. When computing Eq. (2) we then find a non- trivial value of the Z2 topological invariant. However, this model Hamiltonian does not possess gapped surfaces. This is a consequence of the fact that the honeycomb sublat- tice symmetry is not broken by the external magnetic field. Therefore, an effective coupling between the quan- tum Hall edge states is absent. Clearly, sublattice sym- metry breaking terms, the simplest of which are real in- tralayer next-nearest neighbor hoppings [see the Supple- mental Material], effectively open a scattering channel between the intralayer edge modes and thus will yield conventional gapped surface Dirac cones. An alternative approach relies on tilting the direction of the external magnetic field away from the stacking direction towards the [111] direction. As evident from Fig. 4(a-c) such a magnetic field direction tilting allows to open substantial surface gaps. Moreover, the chiral hinge states at positive and negative energies are sepa- rated by a zeroth surface Landau level, and therefore live on different hinges [see the Supplemental Material]. As discussed above, this change in the location of the chiral hinge modes by tuning the Fermi level is perfectly com- patible with the higher-order topology of our system. Conclusion – To sum up, we have proved the exis- tence of a second-order topological insulator protected by inversion symmetry using a coupled layer approach in which the layers have alternating Chern numbers. It can be derived from a parent topological mirror Chern insulator by crystalline symmetry breaking terms that retain the bulk inversion symmetry of the bulk crystal. The presence of the topologically protected chiral hinge modes can be inferred from a three-dimensional Z2 in- variant. We have shown that a non-trivial value of this invariant can be encountered in stacks of doped quantum Hall layers with a buckled honeycomb lattice structure. As a result, we believe that silicene multilayers provide an excellent platform to engineer chiral inversion-symmetric higher-order topological insulators. Acknowledgements – C.O. acknowledges support from a VIDI grant (Project 680-47-543) financed by the Netherlands Organization for Scientific Research (NWO). This work is part of the research programme of the Foundation for Fundamental Research on Matter (FOM), which is part of the Netherlands Organization for Scientific Research (NWO). S.K. acknowledges sup- port from a NWO-Graduate Program grant. [1] B. Bradlyn, L. Elcoro, J. Cano, M. G. Vergniory, Z. Wang, C. Felser, M. I. Aroyo, and B. A. Bernevig, Nature 547, 298 (2017). [2] H. C. Po, A. Vishwanath, and H. Watanabe, Nature Communications 8, 50 (2017). 5 [18] E. Khalaf, Physical Review B 97, 205136 (2018). [19] M. Ezawa, Phys. Rev. Lett. 120, 026801 (2018). [20] J. Langbehn, Y. Peng, L. Trifunovic, F. von Oppen, and P. W. Brouwer, Phys. Rev. Lett. 119, 246401 (2017). [21] M. Sitte, A. Rosch, E. Altman, and L. Fritz, Phys. Rev. Lett. 108, 126807 (2012). [22] Z. Song, Z. Fang, and C. Fang, Phys. Rev. Lett. 119, 246402 (2017). [23] M. Ezawa, Phys. Rev. B 97, 155305 (2018). [24] M. Ezawa, Phys. Rev. B 97, 241402 (2018). [25] G. van Miert and C. Ortix, ArXiv e-prints arXiv:1806.04023 [cond-mat.mes-hall]. (2018), [26] A. Lau, K. Koepernik, J. van den Brink, and C. Ortix, Phys. Rev. Lett. 119, 076801 (2017). [27] C. Kamal, A. Chakrabarti, A. Banerjee, and S. Deb, J. Phys.: Condens. Matter 25, 085508 (2013). [28] A. C. Neto, F. Guinea, N. M. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [29] I. Fulga, N. Avraham, H. Beidenkopf, and A. Stern, Phys. Rev. B 94, 125405 (2016). [30] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [31] D. R. Hofstadter, Phys. Rev. B 14, 2239 (1976). [32] B. A. Bernevig, T. L. Hughes, S. Raghu, Arovas, Phys. Rev. Lett. 99, 146804 (2007). and D. P. [33] J. C. Teo, L. Fu, and C. Kane, Phys. Rev. B 78, 045426 (2008). [34] M. Rice and E. Mele, Phys. Rev. Lett. 49, 1455 (1982). [35] D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, [3] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 1959 (2010). [36] L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). [37] R. Jackiw and C. Rebbi, Phys. Rev. D 13, 3398 (1976). [38] L. Cano-Cort´es, C. Ortix, and J. van den Brink, Phys. Rev. Lett. 111, 146801 (2013). (2010). [4] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [5] L. Fu, Phys. Rev. Lett. 106, 106802 (2011). [6] T. H. Hsieh, H. Lin, J. Liu, W. Duan, A. Bansil, and L. Fu, Nature Communications 3, 982 (2012). [7] H. Nielsen and M. Ninomiya, Physics Letters B 105, 219 (1981). [8] C. Fang and L. Fu, ArXiv e-prints (2017), arXiv:1709.01929 [cond-mat.mes-hall]. [9] W. A. Benalcazar, B. A. Bernevig, and T. L. Hughes, Science 357, 61 (2017). [10] W. A. Benalcazar, B. A. Bernevig, and T. L. Hughes, Physical Review B 96, 245115 (2017). [11] F. Schindler, A. M. Cook, M. G. Vergniory, Z. Wang, S. S. Parkin, B. A. Bernevig, and T. Neupert, Science Advances 4, eaat0346 (2018). [12] F. Schindler, Z. Wang, M. G. Vergniory, A. M. Cook, A. Murani, S. Sengupta, A. Y. Kasumov, R. Deblock, S. Jeon, I. Drozdov, H. Bouchiat, S. Gu´eron, A. Yazdani, B. A. Bernevig, and T. Neupert, ArXiv e-prints (2018), arXiv:1802.02585 [cond-mat.mtrl-sci]. [13] M. Geier, L. Trifunovic, M. Hoskam, and P. W. Brouwer, Physical Review B 97, 205135 (2018). [14] Y. Xu, R. Xue, and S. Wan, ArXiv e-prints (2017), arXiv:1711.09202 [cond-mat.str-el]. [15] C. W. Peterson, W. A. Benalcazar, T. L. Hughes, and G. Bahl, Nature 555, 346 (2018). [16] M. Serra-Garcia, V. Peri, R. Susstrunk, O. R. Bilal, T. Larsen, L. G. Villanueva, and S. D. Huber, Nature 555, 342 (2018). [17] S. Imhof, C. Berger, F. Bayer, J. Brehm, L. Molenkamp, T. Kiessling, F. Schindler, C. H. Lee, M. Greiter, T. Neupert, and R. Thomale, ArXiv e-prints (2017), arXiv:1708.03647 [cond-mat.mes-hall]. SUPPLEMENTAL MATERIAL 6 In this supplemental material we show that a real nearest-neighbor hopping in one of the buckled honeycomb layers is enough to open a surface gap in the model of doped quantum Hall layers presented in the main text. In addition, we briefly discuss the localization of the hinge states present. Surface gap opening by nearest-neighbor hopping Let us consider one layer of silicene in a zigzag ribbon geometry with a perpendicular magnetic field threading a flux φ, at a bias V0 and with real next-nearest neighbor hopping t(cid:48). The Hamiltonian is −t(cid:48)(cid:0)eiπφj + eikx e−iπφj(cid:1) V0 − 2t(cid:48) cos (kx) (cid:19) ψkx,j (3) H(kx) = (cid:88) −(cid:88) j j † kx,j ψ † kx,j+1 ψ (cid:18) V0 − 2t(cid:48) cos (kx) −t(cid:48)(cid:0)e−iπφj + e−ikx eiπφj(cid:1) (cid:18)t(cid:48) (cid:19) ψkx,j + h.c, t 0 t(cid:48) where ψkx,j = (akx,j, bkx,j), with akx,j and bkx,j the annihilation operators on the A and B site of unit cell j respectively. If we consider a stack of such layers with alternating ±V0, we will have a gapped surface if the edge state of a layer hybridizes differently with the edge state of the layer above and below it. This corresponds to t1 (cid:54)= t2 in the effective surface model presented in the main text. To check whether this is the case we take the Hamiltonian Eq. (3), and fix V0 and φ such that the Fermi energy lies in the gap with Chern number +1 (see Fig. 3 in the main text). We solve for the right-moving edge state at E = 0. We then also take the Hamiltonian at −V0 and solve for the left-moving edge state at E = 0 at the same momentum kx. Let us denote these states by ψ(cid:105) and χ(cid:105). There are two different hoppings between two layers: one connecting all the A sublattice sites, and one connecting all the B sublattice sites. Let us denote these by H⊥A and H⊥B. If the mixing due to these is different, (cid:104)χ H⊥A ψ(cid:105) (cid:54)= (cid:104)χ H⊥B ψ(cid:105) , (4) there is an effective dimerization and a gapped surface. If t(cid:48) = 0, ψ(cid:105) and χ(cid:105) are related to each other by chiral symmetry. In addition we have sublattice symmetry, which means that the edge states have equal weight on the A and B sublattices. From this it follows that Eq. (4) is not satisfied. Taking t(cid:48) (cid:54)= 0 breaks sublattice symmetry and leads to edge states that do not have equal weight on the A and B sublattices. We now take even layers to have t(cid:48)/t = 0.1, V0 = −0.8t, φ/φ0 = 1/5 and the odd layers to have t(cid:48) = 0, V0 = 0.6t, φ/φ0 = 1/5 (since t(cid:48) (cid:54)= 0 breaks particle-hole symmetry we slightly adjust V0 to remain in the gap). Solving for ψ(cid:105) and χ(cid:105) then gives t1 t2 = (cid:104)χ H⊥A ψ(cid:105) (cid:104)χ H⊥B ψ(cid:105) ≈ 0.985. This means that the surface is gapped. We note that the mixing is small, but it nevertheless shows that the gapless surfaces are not protected and can be gapped out by symmetry-allowed perturbations. Localization of hinge states 7 Figure 5. Figure showing the localization of the hinge states along the armchair direction. Going from the gap above the zeroth LL of the surface to the one below it, the chiral hinge states localize on different hinges. This realizes both inversion symmetric configurations, as alluded to in the main text. 

1706.02957

1

1706

Effect of oxygen plasma on nanomechanical silicon nitride resonators

Precise control of tensile stress and intrinsic damping is crucial for the optimal design of nanomechanical systems for sensor applications and quantum optomechanics in particular. In this letter we study the in uence of oxygen plasma on the tensile stress and intrinsic damping of nanomechanical silicon nitride resonators. Oxygen plasma treatments are common steps in micro and nanofabrication. We show that oxygen plasma of only a few minutes oxidizes the silicon nitride surface, creating several nanometer thick silicon dioxide layers with a compressive stress of 1.30(16)GPa. Such oxide layers can cause a reduction of the e ective tensile stress of a 50 nm thick stoichiometric silicon nitride membrane by almost 50%. Additionally, intrinsic damping linearly increases with the silicon dioxide lm thickness. An oxide layer of 1.5nm grown in just 10s in a 50W oxygen plasma almost doubled the intrinsic damping. The oxide surface layer can be e ciently removed in bu ered HF.

cond-mat.mes-hall

cond-mat

Effect of oxygen plasma on nanomechanical silicon nitride resonators Niklas Luhmann,1, 2 Artur Jachimowicz,1 Johannes Schalko,1 Pedram Sadeghi,1 Markus Sauer,3 Annette Foelske-Schmitz,3 and Silvan Schmid1, a) 1)Institute of Sensor and Actuator Systems, TU Wien, Vienna, Austria 2)Department of Physics, University of Konstanz, 78457 Konstanz, Germany 3)Analytical Instrumentation Center, TU Wien, Vienna, Austria (Dated: 2 April 2018) Precise control of tensile stress and intrinsic damping is crucial for the optimal design of nanomechanical systems for sensor applications and quantum optomechanics in particular. In this letter we study the influence of oxygen plasma on the tensile stress and intrinsic damping of nanomechanical silicon nitride resonators. Oxygen plasma treatments are common steps in micro and nanofabrication. We show that oxygen plasma of only a few minutes oxidizes the silicon nitride surface, creating several nanometer thick silicon dioxide layers with a compressive stress of 1.30(16) GPa. Such oxide layers can cause a reduction of the effective tensile stress of a 50 nm thick stoichiometric silicon nitride membrane by almost 50%. Additionally, intrinsic damping linearly increases with the silicon dioxide film thickness. An oxide layer of 1.5 nm grown in just 10 s in a 50 W oxygen plasma almost doubled the intrinsic damping. The oxide surface layer can be efficiently removed in buffered HF. I. INTRODUCTION Silicon nitride has become a much valued material for the fabrication of nanomechanical resonators due to its excellent mechanical and optical properties. A partic- ularly interesting feature of silicon nitride thin films is the large intrinsic tensile stress. This stress not only de- fines the resonance frequency (f ) of resonators such as strings or membranes but further dilutes intrinsic damp- ing mechanisms,1–5 which results in exceptionally high quality factors (Q).3,6–9 This has made nanomechanical silicon nitride resonators a favorite choice e.g. for cavity optomechanics experiments.10–15 In particular for applications in quantum optomechan- ics there are strong efforts underway in order to overcome the theoretically required limit of Q × f > 1 × 1013 Hz, which would enable quantum experiments at room tem- perature. There are basically two approaches on how to improve Q, which are by i) optimizing damping dilution either by "soft-clamping" of silicon nitride resonators in- side phononic crystal structures,16,17 or by increasing the tensile stress,18 and ii) by reducing intrinsic losses. More generally, the precise control of mechanical pa- rameters, such as tensile stress and intrinsic damping, are of fundamental significance for the optimal design of nanomechanical sensors. In particular, the responsiv- ity of spectrochemical sensors based on the photothermal heating of a silicon nitride resonator directly depends on the magnitude of tensile stress.19–23 In this letter we study the effect of oxygen plasma on both effective tensile stress and intrinsic loss of nanome- chanical silicon nitride membrane resonators. The incin- eration of polymeric photoresist residues with an oxygen plasma is common practice in nano and microfabrica- tion. Although oxygen plasma has long been known to a)Electronic mail: silvan.schmid@tuwien.ac.at not only effectively oxidize silicon24–26 but also silicon nitride,27–29 its effect on nanomechanical resonators has so far not been recognized. II. METHODS The experiments were done with rectangular silicon- rich (low-stress) and stoichiometric (high-stress) silicon nitride membranes. The membranes were fabricated from Si wafers coated with 50 nm silicon nitride by low-pressure chemical vapor deposition (LPCVD), pur- chased from Hahn-Schickard-Gesellschaft fur angewandte Forschung e.V. with a nominal tensile stress of σ0 ≈ 50 MPa and σ0 ≈ 1 GPa, respectively. The membranes were patterned by photolithography and dry etching of the backside silicon nitride layer and subsequently re- leased by anisotropic KOH (40 wt%) wet etching all through the silicon wafer. The oxygen plasma exposure was performed with a parallel plate STS320PC RIE plasma system from STS Systems with 49.5 sccm O2 flow and a chamber pressure of 20 Pa. The vibrational analysis of the membranes was con- ducted under high vacuum with a laser-Doppler vibrom- eter (MSA-500 from Polytec GmbH). The membranes were actuated thermoelastically by focusing an ampli- tude modulated diode laser (λ =635 nm, with a maximal power on the sample of 70 µW) onto the membrane rim. The quality factors Q were extracted from ring-down measurements performed with a lock-in amplifier (HF2LI from Zurich Instrument). The ring-down was prepared by first driving the specific resonance mode with a phase- locked loop before stopping the actuation. The EDX analysis was performed with a 20 µm area scan directly on the membrane using an X-MaxN detector provided by Oxford Instruments attached to a Hitachi SU8030 scanning electron microscope. The XPS measurements were performed with an 7 1 0 2 n u J 9 ] l l a h - s e m . t a m - d n o c [ 1 v 7 5 9 2 0 . 6 0 7 1 : v i X r a 2 FIG. 2. a) Microscope images of a low-stress silicon ni- tride membrane before and after 30 s exposure to 150 W oxy- gen plasma. b) EDX analysis of the atomic composition and tensile stress of a low-stress silicon nitride membrane (L = 500 µm) measured initially, after 30 s of 150 W oxygen plasma, and after 2 min in BHF. Each data point represents the average of 5 membranes. c) Normalized XPS detail spec- tra of low-stress silicon nitride samples, measured initially and after 30 s of Ar-ion sputtering. The sample treated by oxygen plasma was exposed for 21 s at 50 W. reversed from tensile to compressive, as can be seen from microscope images shown in Figure 2a. Figure 2b presents the atomic composition and tensile stress of low-stress silicon nitride membranes measured i) initially, ii) after a 30 s exposure to 150 W oxygen plasma, and iii) after a 2 min bath in buffered hydrofluoric acid (BHF). As mentioned before, 150 W oxygen plasma re- sulted in compressive stress, clearly visible by the ripples in the membrane in Figure 2a, and an increased oxygen content. A subsequent dip in buffered HF (BHF) recov- ered the tensile stress. The same recovery was also found for high-stress silicon nitride membranes, whose stress reached 96% of the initial value after a 1 min dip in BHF (data not shown). This is clear evidence that the stress reduction is caused by a surface layer with compressive stress and that is removable in BHF. It is known that oxy- gen plasma not only creates a compressive silicon dioxide layer in silicon substrate,26 but it also efficiently oxidises LPCVD silicon nitride thin films.27 The rise in atomic FIG. 1. Tensile stress of a) high-stress (L =500 µm) and b) low-stress silicon nitride membranes (L =500 µm) with re- spect to the oxygen plasma exposure time. Each stress value is the average of 5 membranes extracted from the fundamental mode (1,1). The red lines are exponential fits. SPECS XPS-spectrometer, equipped with a monochro- matic aluminium K-alpha X-ray source (µFocus 350) and a hemispheric WAL-150 analyser. Additional sample preparation was carried out using 3kV Ar+-ions from a SPECS IQ 12/38 ion sputter gun. The surface compo- sition analysis was supported by simulation of electron spectra for surface analysis (SESSA) software.30 The thin film thickness was measured with a Filmetrics F20-UVX thin film analyzer. III. RESULTS & DISCUSSION Figure 1 shows the tensile stress of high-stress and low- stress silicon nitride membranes for an increasing time in oxygen plasma. The tensile stress σ was extracted from the eigenfrequency fn,m model for membranes5 √ (cid:114) σ n2 + m2 2L ρ fn,m = , (1) with length L, mode numbers n and m, and assuming a mass density of ρ =3000 kg/m3. Comparisons with an ex- tended plate model have shown that the ideal membrane model (1) holds true for all analysed membranes under tensile stress. For both types of membranes, tensile stress drops exponentially with plasma exposure time. The intrinsic tensile stress in the high-stress silicon nitride membranes dropped by almost 150 MPa when shortly ex- posed to an oxygen plasma of 150 W (see Figure 1a). For low-stress silicon nitride membranes the initial 40 MPa tensile stress reduced to almost zero for an exposure to only 50 W (see Figure 1b). When treating the low-stress silicon nitride with 150 W oxygen plasma the stress even oxygen content after plasma seen in Figure 2b can be attributed to the creation of such an oxidation layer. From previous work it appears that the oxygen from the plasma substitutes the nitrogen in the silicon nitride film. In detail, these oxygen plasma grown silicon dioxide films feature a vertical gradient in composition ranging from i) a silicon dioxide layer at the surface, due to a loss of all nitrogen atoms into the plasma, to ii) pure silicon nitride at the interface.28,29,31 Evidence of the growth of such a silicon dioxide layer was obtained by XPS anal- ysis, which revealed that the surface of the plasma in- duced oxidation layer indeed is fully depleted of nitrogen. Figure 2c presents the chemical state analysis of the Si 2p region a silicon nitride sample before (reference sam- ple) and after oxygen plasma exposure. Deconvolution of these Si 2p peaks shows that both samples exhibit an SiO2 (103.5 eV) component in addition to Si3N4 (101.7 eV). However, the plasma treated silicon nitride shows a significant SiO2-related component, representing a sub- stantial SiO2 film, with a thickness of several nm. In contrast to the faint SiO2 signature of the reference sam- ple, which can be assigned to a few monolayers of native SiO2 at the silicon nitride surface. After sputtering of the sample surface with Ar ions, the SiO2 related peaks disappeared completely for both samples. In the plasma oxidation process, the volume of Si3N4 molecules of 77.5 A3 (calculated assuming a mass den- sity of 3000 kg/m3) grows to the volume of three SiO2 molecules of 45.5 A3 each (assuming a mass density of 2200 kg/m3). In the case of planar growth the con- sumed Si3N4 is replaced by a SiO2 film that is theoreti- cally 1.77× thicker, which has been verified with electron microscope images of oxidised silicon nitride surfaces.27 The growth of SiO2 leads to a one-dimensional unrelaxed strain of ε = 0.21, which would result in an enormous compressive stress of σSiO = εE/(1 − ν) ≈ 18 GPa. This is of the same order of magnitude of the initial stress building up during silicon oxidation.32 However, this enormous compressive stress is expected to relax to a magnitude close to the compressive strength of amor- phous silicon dioxide of approximately 1.1 GPa. In order to estimate the effective compressive stress of the silicon dioxide layer, high-stress silicon nitride mem- branes were oxidised with various plasma powers. Af- terward the grown SiO2 was removed in BHF. The SiO2 layer thickness was then calculated from the reduction of the silicon nitride film thickness (see inset of Figure 3a), taking into account the volume increase during oxidation and the BHF etch-rate of the silicon nitride. Figure 3a shows the tensile stress σ, obtained from measured mem- brane resonance frequencies, versus the measured oxide film thickness. The effective stress σ in the silicon ni- tride membrane can be modelled as the arithmetic mean of the tensile stress σSiN of the silicon nitride (of thick- ness hSiN ) and the compressive stress σSiO of the silicon dioxide layer (of thickness hSiO) σ = σSiOhSiO + σSiN hSiN hSiO + hSiN . (2) 3 FIG. 3. a) Stress reduction of high-stress silicon nitride mem- branes (L =500 µm) versus plasma power. The samples were kept in the oxygen plasma for 5 min. The silicon dioxide film thickness was estimated from (2). The inset shows the mea- sured silicon dioxide film thickness as a function of oxygen plasma power for an exposure time of 5 min each. b) Effective SiO2 film thickness versus plasma exposure time, calculated from (2) based on the measured tensile stress σ of high-stress and low-stress silicon nitride, as presented in Figure 1. Fitting the model (2) to the data in Figure 2a allows the extraction of a compressive stress of the SiO2 film of 1.30(16) GPa. The measured stress is of the expected magnitude of the compressive strength of amorphous sil- icon dioxide. With the gained value for the compressive stress σSiO, it is now possible to estimate the oxide film thickness based on the measured effective stress σ. Figure 3b plots the estimated SiO2 thickness of the silicon dioxide layer as a function of oxygen plasma time for the two sam- ples presented in Figure 1. The same samples have been analysed with XPS and quantitative results have sub- sequently been compared with simulations varying the SiO2 layer thickness, and the obtained thickness esti- mates match well with the estimated values from the ef- fective tensile stress by means of (2). It has been shown that oxygen plasma induced oxide growth at room tem- perature shows logarithmic behavior.33 The same behav- ior seems to hold true for the oxygen plasma grown from silicon dioxide thin films, as can be seen by the logarith- mic fits shown as red lines. 4 In order to study the effect of oxygen plasma on intrin- sic losses, the quality factor of low-stress silicon nitride membranes was measured, as presented in Figure 4a. The low tensile stress in these membranes produce a sufficient decoupling from the chip frame thereby minimizing ra- diation losses.34 Hence, the measured quality factors are exclusively limited by intrinsic damping, which is further confirmed by the fact that the measured values for each treatment step follow the prediction (red curves) from the damping dilution model (3) for intrinsic loss (1/Qintr) in membranes4,5 (cid:34) (cid:114) Q = h L E 3σ + π2(n2 + m2) 12 E σ h2 L2 (cid:35)−1 Qintr. (3) The intrinsic loss (1/Qintr) was then extracted by cor- recting the measured quality factor values Q for the stress induced damping dilution effect (3) assuming a Young's modulus of E = 200 GPa for silicon-rich silicon nitride.35 The respective intrinsic quality factors Qintr for the dif- ferent plasma exposure times are plotted in Figure 4b, to- gether with the correlation to the estimated silicon diox- ide film thickness. Apparently, intrinsic losses increase with oxygen plasma exposure. The comparison of Qintr with the estimated oxide thickness shows a linear corre- lation. This suggests that the increased loss can be at- tributed to the growth of the silicon dioxide surface layer, which has larger intrinsic damping than silicon nitride.36 The magnitude of the initial Qintr value matches the gen- eral value found for surface loss in silicon nitride.34 IV. CONCLUSION Silicon nitride structures with a thickness of only 50 nm are highly sensitive to oxygen plasma. Even short expo- sures cause a significant decreases in tensile stress and an increase in intrinsic loss. Both effects can be at- tributed to the plasma-induced oxidation of the silicon nitride surface. We found that the created silicon diox- ide film on the silicon nitride surface has a compressive stress of 1.30(16) GPa, which is probably limited by the layers' own compressive strength. This relatively large stress counteracts the intrinsic tensile stress in LPCVD silicon nitride, leading to a sig- nificant drop of the intrinsic stress and hence in the res- onance frequency. Oxygen plasma exposure of nanome- chanical silicon nitride resonators can reduce the tensile stress by several hundreds of MPa, which for low-stress silicon nitride structures can even cause a total rever- sal of the effective stress from tensile to compressive. Hence it is an interesting tool which allows a precise post- fabrication control of tensile stress. Additionally, the plasma grown silicon dioxide layer significantly increases energy loss in nanomechanical sil- icon nitride resonators. Hence, for applications where FIG. 4. Energy loss study of 5 low-stress silicon nitride mem- branes (L =200 µm) for the (1,1), (2,2), and (3,3) mode after different exposure times to 50 W oxygen plasma. The red lines represent fits with the damping dilution model (3) with Qintr as fitting parameter. b) Extracted average Qintr values for the different plasma exposure times. The intrinsic qual- ity factor values are further plotted versus the silicon dioxide thickness estimated from the measured effective stress. The red line is a linear fit. a maximum tensile stress as well as minimum intrinsic damping is desired, the silicon dioxide layer and its de- teriorating effects can fully be removed by a quick BHF dip. Since an oxygen plasma also oxidizes silicon surfaces, similar effects as observed for silicon nitride resonators will likely occur in nanomechanical structures made from silicon. ACKNOWLEDGMENTS We thank Sophia Ewert and Patrick Meyer for the cleanroom support. This project has received funding from the European Research Council (ERC) under the European Unions Horizon 2020 research and innovation program (grant agreement - 716087 - PLASMECS). 1G. I. Gonzfilez and P. R. Saulson, Journal of the Acoustical So- ciety of America 96, 207 (1994). 2S. Schmid and C. Hierold, Journal of Applied Physics 104, 093516 (2008). 3S. Schmid, K. D. Jensen, K. H. Nielsen, and A. Boisen, Physical Review B 84, 165307 (2011). 4P.-L. Yu, T. Purdy, and C. A. Regal, Physical Review Letters 108, 083603 (2012). 5S. Schmid, L. G. Villanueva, and M. L. Roukes, Fundamentals of Nanomechanical Resonators, 1st ed. (Springer International Publishing, 2016). 6S. S. Verbridge, J. M. Parpia, R. B. Reichenbach, L. M. Bellan, and H. G. Craighead, Journal of Applied Physics 99, 124304 (2006). 7B. M. Zwickl, W. E. Shanks, A. M. Jayich, C. Yang, B. Jayich, J. D. Thompson, and J. G. E. Harris, Applied Physics Letters 92, 103125 (2008). 8S. S. Verbridge, H. G. Craighead, and J. M. Parpia, Applied Physics Letters 92, 013112 (2008). 9S. Chakram, Y. S. Patil, L. Chang, and M. Vengalattore, Phys- ical Review Letters 112, 127201 (2014). 10T. P. Purdy, R. W. Peterson, and C. A. Regal, Science 339, 801 (2013). 11D. J. Wilson, Cavity Optomechanics with High-Stress Silicon Nitride Films, Ph.D. thesis, California Institute of Technology (2012). 12E. Gavartin, P. Verlot, and T. J. Kippenberg, Nature Nanotech- nology 7, 509 (2012). 13T. Bagci, A. Simonsen, S. Schmid, L. G. Villanueva, E. Zeuthen, J. Appel, J. M. Taylor, A. Sørensen, K. Usami, A. Schliesser, and E. S. Polzik, Nature 507, 81 (2014). 14J. D. Thompson, B. M. Zwickl, A. M. Jayich, F. Marquardt, S. M. Girvin, and J. G. E. Harris, Nature 452, 72 (2008). 15G. a. Brawley, M. R. Vanner, P. E. Larsen, S. Schmid, A. Boisen, and W. P. Bowen, Nature Communications 7 (2016), 10.1038/ncomms10988, arXiv:1404.5746. 16Y. Tsaturyan, A. Barg, E. S. Polzik, and A. Schliesser, , 1 (2016), arXiv:1608.00937. 17A. H. Ghadimi, D. J. Wilson, and T. J. Kippenberg, Nano Letters ASAP (2017), 10.1021/acs.nanolett.7b00573, arXiv:1603.01605. 18R. A. Norte, J. P. Moura, and S. Groblacher, Physical Review Letters 116, 1 (2016), arXiv:1511.06235. 5 19M. Kurek, M. Carnoy, P. E. Larsen, L. H. Nielsen, O. Hansen, T. Rades, S. Schmid, and A. Boisen, Angewandte Chemie Inter- national Edition , 3901 (2017). 20A. J. Andersen, S. Yamada, P. Kumar E.K., T. L. Andresen, A. Boisen, and S. Schmid, Sensors and Actuators B: Chemical 233, 667 (2016). 21S. Yamada, S. Schmid, T. Larsen, O. Hansen, and A. Boisen, Analytical Chemistry 85, 10531 (2013). 22T. S. Biswas, N. Miriyala, C. Doolin, X. Liu, T. Thundat, and J. P. Davis, Analytical chemistry 86, 11368 (2014). 23T. Larsen, S. Schmid, L. G. Villanueva, and A. Boisen, ACS Nano 7, 6188 (2013). 24J. R. Ligenza, Journal of Applied Physics 36, 2703 (1965). 25J. Kraitchman, Journal of Applied Physics 38, 4323 (1967). 26D. Pulfrey and J. Reche, "Preparation and properties of plasma- anodized silicon dioxide films," (1974). 27S. Taylor, W. Eccleston, J. Ringnalda, D. M. Maher, D. J. Eagle- sham, C. J. Humphreys, and D. J. Godfrey, Journal de Physique 49, 393 (1988). 28G. P. Kennedy, S. Taylor, W. Eccleston, W. M. Arnoldbik, and F. H. M. Habraken, Microelectronic Engineering 28, 141 (1995). and S. Taylor, Journal of Applied 29G. P. Kennedy, O. Buiu, Physics 85, 3319 (1999). 30W. Smekal, W. S. M. Werner, and C. J. Powell, Surface and Interface Analysis 37, 1059 (2005). 31O. Buiu, G. Kennedy, M. Gartner, and S. Taylor, IEEE Trans- actions on Plasma Science 26, 1700 (1998). 32P. Sutardja and W. Oldham, IEEE Transactions on Electron De- vices 36, 2415 (1989). 33K. Kim, M. H. An, Y. G. Shin, M. S. Suh, C. J. Youn, Y. H. Lee, K. B. Lee, and H. J. Lee, Journal of Vacuum Science & Technology B 14, 2667 (1996). 34L. G. Villanueva and S. Schmid, Physical Review Letters 113, 1 (2014), arXiv:1405.6115. 35H. Guo and A. Lal, Journal of Microelectromechanical Systems 12, 53 (2003). 36D. R. Southworth, R. A. Barton, S. S. Verbridge, B. Ilic, A. D. Fefferman, H. G. Craighead, and J. M. Parpia, Physical review letters 102, 225503 (2009). 

1310.1234

1

1310

Magnetic states of single impurity in disordered environment

The charged and magnetic states of isolated impurities dissolved in amorphous metallic alloy are investigated. The Hamiltonian of the system under study is the generalization of Anderson impurity model. Namely, the processes of elastic and non-elastic scattering of conductive electrons on the ions of a metal and on a charged impurity are included. The configuration averaged one-particle Green's functions are obtained within Hartree-Fock approximation. A system of self-consistent equations is given for calculation of an electronic spectrum, the charged and the spin-polarized impurity states. Qualitative analysis of the effect of the metallic host structural disorder on the observed values is performed. Additional shift and broadening of virtual impurity level is caused by a structural disorder of impurity environment.

cond-mat.mes-hall

cond-mat

Condensed Matter Physics, 2013, Vol. 16, No 3, 33705: 1 -- 11 DOI: 10.5488/CMP.16.33705 http://www.icmp.lviv.ua/journal Magnetic states of single impurity in disordered environment G.V. Ponedilok, M.I. Klapchuk National University "Lviv Polytechnic", Institute of Applied Mathematics and Fundamental Sciences, 12 S. Bandera St., 79013 Lviv, Ukraine Received April 12, 2013, in final form June 18, 2013 The charged and magnetic states of isolated impurities dissolved in amorphous metallic alloy are investigated. The Hamiltonian of the system under study is the generalization of Anderson impurity model. Namely, the processes of elastic and non-elastic scattering of conductive electrons on the ions of a metal and on a charged impurity are included. The configuration averaged one-particle Green's functions are obtained within Hartree- Fock approximation. A system of self-consistent equations is given for calculation of an electronic spectrum, the charged and the spin-polarized impurity states. Qualitative analysis of the effect of the metallic host structural disorder on the observed values is performed. Additional shift and broadening of virtual impurity level is caused by a structural disorder of impurity environment. Key words: isolatedimpuritystates,structuraldisorder PACS: 71.23.-k,71.23.An,72.15.Rn 1. Introduction The purpose of this work is to explore the effects of the structural disorder on the states of electroneg- ative impurities dissolved in liquid alkali metal. The ions belonging to metal matrix form a complicated random field for an impurity atom. The model proposed in this study is applicable to a structurally dis- ordered system in which the tight-binding representation of the electronic wave function is appropriate and the effect of a short-range order is eminent. The liquid alkali metals as well as amorphous solids are the systems we would like to investigate. In this work we describe the states of isolated impurities of such elements as H, O, Cl, F, N using gener- alized microscopic theory based on the single-impurity Anderson model (SIAM) [1]. While discussing the macroscopic features of the single-electron properties of the system, the procedure of effective Green's function ensemble-averaging over all possible configurations of atoms R1, . . . , RN is necessary. The pro- cedure of configurational averaging is an enormously difficult problem in the multiple-scattering theory. Only the two-particle correlation functions are known from experimental data. In practice, of course, our knowledge of these density correlation functions is incomplete and various approximate theories for short-range order involve only the one- and two-site distribution functions [2 -- 6]. The short-range order is always present in liquid metals; its simplest manifestation is in the characteristic oscillation of the x-ray structure factor or in the oscillation of radial distribution function. The main goal of this paper is to show the effect of disordered local impurity environment on its charge and magnetic states. The experimentally observed magnetic moment decrease for some sorts of ferromagnetic solids or amorphous alloys is discussed in detail in [7]. Microscopic model to describe the electronegative impurity in a disordered system is proposed in section 1. The Hamiltonian of the system is a generalization of Anderson impurity model. It also includes the processes of elastic and non-elastic scattering of conductive electrons on the ions of a metal and on a charged impurity. Qualitative and quantitative estimates of the parameters of the Hamiltonian have been carried out in [8]. The formation of an effective charge and spin-polarized gaseous impurity states in a © G.V. Ponedilok, M.I. Klapchuk, 2013 33705-1 G.V. Ponedilok, M.I. Klapchuk liquid metal can be described as the process of hybridization of local level with quasi-free electron states under the effect of a polarizing impurity potential [9]. The two-time retarded Green's functions [10] are obtained within Hartree-Fock (HFA) approximation in section 3. The configuration averaged system of Green's functions is obtained in section 4. A system of self-consistent equations is given for the calculation of the electronic spectrum, as well as the charged and the spin-polarized impurity states. The qualitative analysis of the effect of the metallic host structural disorder on the observed values is performed. An additional shift and broadening of a virtual impurity level is caused by the structural disorder of an impurity environment. 2. Microscopic model of the system Let us consider a single impurity dissolved in liquid alkaline metal. The liquid metal phase will be described within the framework of electron-ion model which for such metals gives satisfactory compu- tational results for electronic and structural properties. Let R1, . . . , RN be the coordinates of atoms of metallic alloy which take arbitrary values in the volume V . The impurity has a coordinate R0. We have chosen the following model Hamiltonian in coordinate representation: The energy operator of electron-ion interaction is written as follows: H = Hcl + Hel-i + Hel-el . Hel−i = − ħ2 2m X1ÉiÉN ∆i + X1ÉiÉN X1É jÉN V ( ri − R j )+ X1ÉiÉNe V0( ri − R0 ). (2.1) (2.2) In this equation r1, . . . , rN are the electron coordinates of a metallic subsystem, the amount of which coincides with the number of metal atoms due to single valence of alkaline elements. It is assumed that the electrons of valence impurity orbital remain localized on the impurity. The potentials V (ri −R j) and V0(ri −R0) describe electron scattering on ions of metal and impurity, respectively. The first term in equation (2.2) is the operator of a kinetic energy of free electron subsystem. The last term in (2.1) describes the energy of pair electron-electron interaction Hel-el= 1 2 X1Éi , jÉN Φ( ri − r j )= 1 2 X1Éi , jÉN e2 ri − r j . (2.3) The non-operator part, Hcl, describes the energy of classical ion-ion interaction. In order to represent the secondary quantization, we use plane waves as a basis in order to decompose the field electronic operators and s-shell localized on the impurity ϕk(r) = 1 pV exp (ik· r) ψ0(r) =s 1 πr 3 p expµ−r− R0 rp ¶. The wave vector k in (2.4) takes specified values in the impulse quasi-continuous space Λ: Λ =n k : k = X1ÉαÉ3 2πV 1/3nαeα, nα ∈ Z , (eα, eβ) = δαβo. (2.4) (2.5) (2.6) Let us mention that ψ0(r) is not orthogonal to the plane waves (2.4). Apart from this, its inclusion into the basis set causes overfilling of the latter. However, the inaccuracy introduced by such an approximate procedure will not affect the qualitative picture. In the representation of the secondary quantization 33705-2 Magnetic states of single impurity in disordered environment operator (2.1) with allowance for only a certain class of Coulomb electron-electron interactions we then have the following expression, H = Hcl +Xk∈Λ Xσ=±1 Ek a+kσ akσ + Xσ=±1 E0 d+0σ d0σ +Xk∈ΛXq∈Λ Xσ=±1¡Vq a+kσ ak−q,σ + V0,q a+kσ ak−q,σ¢ U0 n0σ n0,−σ +Xk∈Λ Xσ=±1³Wk,0 a+kσ d0σ + W ∗k,0 d+0σ akσ´+Xk∈ΛXq∈Λ Xσ,σ′=±1 + Xσ=±1 +Xk∈Λ Xσ,σ′³Uk,0 nσ′ a+kσ d0,σ +U∗k,0 d+0,σ akσ nσ′´ . Pq,0 a+kσ ak−q,σ nσ′ (2.7) Here, akσ(a+kσ) and d0,σ(d+0,σ) are the annihilation (creation) Fermi-type operators for electrons in the states {k, σ} and {R0, σ}, where σ=±1 is quantum spin number, which takes two values due to two possible orientations of electronic spin relatively to the quantization axis. Ek = ħ2k2/2m is the energy spectrum of the electrons in states ϕk(r), and E0 is the energy of the localized electronic state ψ0(r). nσ = d+σ dσ is The matrix elements Vq and V0,q characterize the processes of elastic scattering of electrons on the the spin-dependent occupation number operator for the localized state. ions of the metal and on the impurity. Their explicit analytical forms are as follows: Vq = 1 N X1É jÉN e−iq·R j v(q), V0,q = e−iq·R0 v0(q). The formfactors of the scattering potentials v(q) =ZV V (r) e−iq·r dr, v0(q) = 1 V ZV V0(r) e−iq·r dr (2.8) (2.9) depend only on the absolute value of the momentum transfer q due to the locality of the potentials V (r) and V0(r). The processes of inelastic scattering of electrons caused by their transition from the state localized on the impurity into the conduction band and vice versa are characterized by the matrix element, Here, Wk,0 = 1 pV ZV VLF(r) = X1É jÉN e−ik·r·−ħ2∆r 2m + VLF(r)¸ ψ0(r) dr. V (r− R j)+ V0(r− R0) is the potential of a local field of metal ions and the impurity, which acts on the electron at a point r ∈ V . The termPσ U0 nσ n−σ in the Hamiltonian (2.7) arises from the operator of Coulomb electron interac- tion and describes the Hubbard repulsion of electrons localized on the impurity with the intensity U0. U0 =Z dr1Z dr2ψ0(r1)2 e2 r1 − r2ψ0(r2)2 = 5 8 e2 rp , (2.12) this value is approximately about 1÷ 5 eV for the atom of oxygen. The process of elastic and inelastic scattering of electrons on the charged impurity is described by the (2.10) (2.11) 33705-3 matrix elements, Here, the value Pq,0 =ZV Uk,0 = 1 pV ZV e−ik·reΦ(r)ψ0(r) dr. Φ(r− r′)ψ0(r′)2 dr′, (2.13) (2.14) e−iq·reΦ(r) dr, eΦ(r) =ZV gives the potential energy of the electron in a field which is generated by the electron localized on the ψ0(r) orbital. G.V. Ponedilok, M.I. Klapchuk The matrix elements (2.10) -- (2.13) can be written down in the other form by separating explicitly the structural multipliers Wk,0 = e−ik·R0 wk , Uk,0 = e−ik·R0 uk , Pk,0 = e−ik·R0 pk . (2.15) The coefficients wk , uk , pk do not depend here on the nodal index and are considered in the coordinate system related to the impurity. Their analytical form is given in [13]. Actually, only the electrostatic effects including two electrons are taken into account in the Hamilto- nian (2.7), while the processes of exchange are not considered. 3. Green's function method. Hartree-Fock approximation The method of equation of motion of Green's functions is one of the most important tools to solve the model Hamiltonian problems in condensed-matter physics [11]. Let us calculate the matrix of retarded time-dependent temperature Green's functions G(ω) =ÃG σ k,k′ (ω) M σ Lσ M σ 0,k′ (ω) k,0(ω) 0,0(ω)! ≡µ〈〈akσa+k′σ〉〉ω 〈〈d0σa+k′σ〉〉ω 〈〈d0σd+0σ〉〉ω¶ . 〈〈akσd+0σ〉〉ω (3.1) The equation of motion for each component of (3.1) is given in our earlier work [8, 9, 12, 13]. We make use of the decoupling scheme that corresponds to the Hartree-Fock approximation type for higher order Green functions. The limits of the HFA applicability for the description of real systems are considered in [11, 14 -- 18]. A set of connected Green's functions is obtained Ω∗p (ω− Ek )G σ (ω− E0,σ)M σ (ω− Ek )M σ (ω− E0,σ)Lσ k,q(ω) = δk,q +Xp 0,q(ω) =Xp k,0(ω) =Xq 0,0(ω) = 1+Xk p,q(ω)+ Ωσ k M σ 0,q(ω), Λk−pG σ σG σ p,q(ω), k Lσ 0,0(ω), Λk−qM σ Ω∗k q,0(ω)+ Ωσ σM σ k,0(ω). In the equations (3.2) -- (3.5), we use the notation The Fourier-component of an effective impurity potential, Λq = 2Xα=1 V (α) q = e−iq·R j v(q)+ev0(q). 1 N X1É jÉN e−iq·rhV0(r )+〈bn0〉ZV 1 V ZV ev0(q) = v0(q)+ pq〈bn0〉 = (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) dr′ψ0(r′)2 Φ(r− r′)i dr includes the Hartree-Fock potential, caused by the impurity atom 〈bn0〉 =Pσ〈bn0σ〉. q can be represented in the form, Ωσ In close similarity, the matrix elements Ωσ q = [uq〈bn−σ〉+ wq ], or (3.8) 1 Ωσ q = pV ZV e−iq·rh− ħ2∇∇∇2 LF(r) = VLF(r)+〈bn0,−σ〉ZV eV σ LF(r)iψ0(r) dr, 2m +eV σ dr′ψ0(r′)2Φ(r− r′). where 33705-4 Magnetic states of single impurity in disordered environment From the equations (3.3), (3.5) one can find locator Green's function Lσ 0,0(ω) = ω− E0,σ −Pk,q 1 Ωσ k Λk,q(ω) Ω∗q σ , (3.9) here, we introduced the effective potential Λk,q, which has the form of a series in terms of Λk−q Λk,q(ω)= δk,q ω−Ek−Λ0 + Λk−q (ω−Ek−Λ0)(ω−Eq−Λ0) +Xp Λk−pΛp−q (ω−Ek−Λ0)(ω−Ep−Λ0)(ω−Eq−Λ0) +··· . (3.10) Note that Λ0 = v(0)+ v0(0)− 2π〈 n0〉e2r 2 The non-diagonal Green function M σ locator Lσ 0,0(ω): p /V. 0,k(ω), M σ k,0(ω) and the propagator G σ k,k′ (ω) are expressed by the σΛq,k(ω), M σ Lσ 0,0(ω)Ω∗q 0,k(ω)=Xq M σ k,0(ω)=[M σ k,k′ (ω)=Λk,k′ (ω)+Xq,p 0,k(ω)]∗, G σ Λk,q(ω)Ωσ q Lσ 0,0(ω)Ω∗p σΛp,k′ (ω). The renormalized impurity level is E0,σ = E0 +U0〈bn0,−σ〉+Xk hUk,0〈a+k,−σd0,−σ〉+U∗k,0〈d+0,−σak,−σ〉i+Xk,qXσ′ 4. Configuration averaged Green's function Pq,0 〈a+k,σ′ ak−q,σ′〉. (3.11) One can start the averaging over all atomic configurations from equation (3.9). Here, the self-energy part describes the quasi-particles correlation. The problem is specified in terms of (3.10) describing the degree of correlation. As the first approximation, we take into account Λk−q only, while the higher order correlation functions are neglected. As common, we use a notation for the Fourier-transform of the atomic density fluctuations, The configuration averaged Green's function of localized electrons is given as 1 k , 0, ρk = e−ik···R j , Λk−q = ρk−qv(k− q)+ev0(k− q). E−Eσ−Σ0(k) q )∗Λk−q 1+ Xk,q pN X1É jÉN (E−Ek−Λ0)(E−Eq−Λ0)[E−Eσ−Σ0(k)]+··· Σ0(k) =Xk Ωσ k 2 k (Ωσ Ωσ (k,q) 1 , (4.1) Lσ 0,0(E ) = where E − Ek − Λ0 is the self-energy term in the quasi-crystalline approximation. We would like to discuss the case of inhomogeneous environment, where the impurity atom gives origin to the spherically symmetrical potential. Therefore, ρk−q = nk−q, in contrast to homogeneous case. Here, n(r) = (1/V )Pk nkeik·r. Taking into account the binary distribution function n(r ), the expression for the averaged locator Green function is obtained: Lσ 0,0(ω) = ω− E0,σ − Σ0(k)− Xk, q (k,q) Ωσ k Ωσ q£nk−qv(k− q)+ev0(k− q)¤ (E − Ek − Λ0)(E − Eq − Λ0)  −1 . (4.2) 33705-5 In order to get the density of states per atom for localized electrons, ρσ 0 (E ) with spin σ, we need to G.V. Ponedilok, M.I. Klapchuk calculate the sum over k in (4.2). Ωk2 lim ε→0Xk E − Ek − Λ0 + iε = PXk Ωk2 E − Ek − Λ0 − iπXk Ωk2δ(E − Ek − Λ0), For the sake of convenience let us introduce the notation, ∆σ(E ) = πXk Ωk2δ(E−Ek−Λ0); Λσ(E ) = PXk (E − Ek − Λ0) = Ωk2 1 π PZ d E′ ∆(E′) E − E′ . (4.3) (4.4) (4.5) The scattering of the s-electrons and localized electrons cause the impurity level to shift and become broader. Namely, Λ(E ) is the effective shift whereas ∆(E ) is the effective broadening of impurity level. For the sake of simplicity, the matrix elements (Ωσ)2 are estimated at the Fermi level [15]. Then, ∆σ(E ), Λσ(E ) are slowly varying functions of E over the band, and they can be treated as parameters, Here, ρ0(E ) = is density of states for the free electron gas and ∆σ(E ) = π(Ωσ)2ρ0(E ), Λσ(E ) = (Ωσ)2ρ0(E )g (E ). m3/2 pE 2p2ħ3π2 pEF/E + 1 pEF/E − 1¯¯¯− 2pEF/E . g (E ) = ln¯¯¯ In order to calculate the averaged density of localized states, we use the relation Ωσ k 2 Ωσ k lim Ω∗q ε→0Xk (E − Ek − Λ0 + iε)2 = − σ£nk−qv(k− q)+ev0(k− q)¤ Ωσ k 2Λσ(k, E ) E − Ek − Λ0 − iπXk Ωσ k 2∆σ(k, E ) Ωσ E − Ek − Λ0 − πXk Ωσ (E − Ek − Λ0)(E − Eq − Λ0) = PXk − iPXk = In the similar manner, Xk,q where we denote dΛσ(E ) dE + i d∆σ(E ) dE . k 2Λσ(k, E )δ(E − Ek − Λ0) k 2∆σ(k, E )δ(E − Ek − Λ0), σ Ω∗q f ( k− q ), E − Eq − Λ0 σ f ( k− q )δ(E − Eq − Λ0), Ω∗q Λσ(k, E ) = PXq ∆σ(k, E ) = πXq f ( k− q ) =£nk−qv( k− q )+ev0( k− q )¤. E − E0,σ −eΛσ(E )+ ie∆σ(E ) Lσ 0,0 = 1 , The configuration averaged Green function has the form, 33705-6 (4.6) (4.7) (4.8) (4.9) (4.10) (4.11) (4.12) (4.13) (4.14) Magnetic states of single impurity in disordered environment here, eΛσ(E ) = Λσ(E )+ e∆σ(E ) = ∆σ(E )+ dΛσ(E ) dE d∆σ(E ) dE f (0)+ PXk f (0)+ PXk Ωσ k 2Λσ(k, E ) E − Ek − Λ0 − πXk Ωσ Ωσ k 2∆σ(k, E ) E − Ek − Λ0 + πXk Ωσ k 2∆σ(k, E )δ(E − Ek − Λ0), k 2Λσ(k, E )δ(E − Ek − Λ0) (4.15) (4.16) are the effective shift and broadening of localized impurity level now contain the structural disorder contribution, besides the contribution from interactions. The occupation number of electrons for absolute zero temperature is 〈n0σ〉 = 〈d+0σd0σ〉 = EFZ−∞ ρσ 0 (E )dE , where 0,0(E + iε), is configurational density of localized states with the spin σ. ρσ 0 (E ) = − ImLσ ε → 0 1 π After simple transformation of the system of equations (4.15) -- (4.16) we obtain: e∆σ(E ) ρσ 0 (E ) = 1 π . [E − E0,σ −eΛσ(E )]2 + [e∆σ(E )]2 e∆σ(E ) = π[(Ωσ)2 + 2eF ρ0(E )+ g (E )]ρ0(E )+ π(Ωσ)2 f (0) eΛσ(E ) = (Ωσ)2g (E )ρ0(E )+ (Ωσ)2 f (0) dg (E )ρ0(E ) dE dρ0(E ) , dE +eF ρ0(E )g (E )2−π2eF ρ0(E )2. Here, the notation is introduced for the average value of matrix elements Ωσ k 〈F〉 =Pk,q Ωσ k Ωσ q f ( k− q )δ(E−Eq−Λ0)δ(E−Ek−Λ0) 1/V 2Pk,q δ(E−Eq−Λ0)δ(E−Ek−Λ0) Ωσ q f ( k− q ) at the Fermi level. (4.17) (4.18) (4.19) 5. Results and discussions We need to calculate the average values of matrix elements Ωσ k by using the formfactors of scatter- ing potentials (2.9). Ashcroft, Heine-Abarenkov, Cohen, Animalu model potentials are widely applicable in liquid metal physics. The parameters of these potentials are investigated and approved sufficiently completely, see e.g. [19 -- 22]. We have used the Ashcroft's potential (including screening by the conduction electrons) for the liquid sodium [19]. The Fourier-transform of Ashcroft's potential is v(q) = − 4πZ e2 Ωq 2 cos(qrc), (5.1) where rc is the core radius. The parameters for liquid sodium are r Na c =0.0878 nm, Ω = 270 a.u. -- atomic volume of liquid Na at 100 ◦C. They are taken from the experimental data of resistivity mea- surements [19]. The screened function by the conduction electrons in Heldart-Vosko approximation is as follows: ε(q) = 1+ 1 2 + λ(y) = f (q) = 3 4πZ Ωq 2µ 2 2kF¶£1− f (q)¤ , EF¶−1 λµ q ln¯¯¯ 1− y¯¯¯, 1− y 2 4y 1+ y 1/2q 2 , q 2 + 2kF/(1+ 0.01574(Ω/Z )1/3) (5.2) 33705-7 where kF = (3π2Z /Ω)1/3 = 0.4786 a.u.−1. Now, let us consider the interaction between the electron and the negative ion (3.7). Different forms of polarization potential were discussed in [23 -- 26]. We have proposed a new model potential for electron-negative ion interaction in [12]: G.V. Ponedilok, M.I. Klapchuk V0(r ) = A e−r /rp r 2 − α (r 2 + r 2 p )2 . p E0 + 3αI /r 2 where A = 3/8r 2 p − 3/8. The semi-empirical parameters α and rp do not arise naturally from the formalism. Thus, the only criterium available to establish the accuracy of the method is in the agree- ment with the experimental results. Hence, we use the values of rp [25] and α [26] taken from the exper- imental data for electron photodetachment from negative ions. The formfactor of the effective impurity potential in Hartree-Fock approximation is ev0(q) = The correlation function is as follows: 8πA q arctan(qrp)− 4απ2 rp 8π q 2h1− q 2r 2 p (4+ q 2r 2 p )³1+ 4 p´i. 4+ q 2r 2 (5.3) e−qrp +〈bn〉 3η nq = 1+ (qrc)3£qr ∗ cos(qr ∗)− sin(qr ∗)¤ , By using the expressions for the model potentials of liquid metal and impurity and for the correlation where r ∗ = rc + rp. function, one can calculate the average value of Ωσ, 〈F〉. The potentials w, u were discussed in the work [13], specifically at Fermi level (w)2 ≈ E 2/(E γ+ 1)4, p /ħ2, and z = u/w is the parameter of intensity of scattering process on the charged impurity. The γ = 2mr 2 function K (k, q) on the angles in a spherical coordinate system, was introduced to simplify the calculation of 〈F〉: K (k, q) = 2π πZ0 sin θ fµqk2 + q 2 − 2kq cos θ¶ dθ. tural contribution. Then, usingPk=[V /(2π)3]R∞0 k2dkR2π 0 dϕ, we obtain the averaged value 〈F〉 that characterizes the struc- The parameter δ = 〈F〉ρ0g /(Ω)2 measures the value of disorder, and we assume 0 < δ ≪ 1. The pa- rameter h = Λσ/∆σ has the meaning of a local level shift. Figure 1. (Color online) The dependence of the impu- rity magnetic moment on the structural parameter δ (y = 10, x = 0.2, z = 0.1, h = π/8). Figure 2. (Color online) The dependence of the im- purity magnetic moment on the degree of Coulomb repulsion y = U0/∆ at constant Zeff = 1. 33705-8 Magnetic states of single impurity in disordered environment Figure 3. (Color online) The phase diagram exhibits the regions with magnetic and nonmagnetic states. The dimensionless value x = (EF − E0)/U0 means that the local impurity level lies on the Fermi level for x = 0. For x = 1, the Fermi level lies on E0+U0. For magnetic solutions x = 1/2, which means, of course, that the Fermi level is exactly halfway between the case where only one electron is in a localized state and the one in which both are in the same state with opposite spins. The parameter y = U0/∆ measures the ratio of Coulomb integral respective to the width of virtual state. The spin-polarized magnetic impurity state m = 〈n+ − n−〉, m/m0 is shown in figure 1. Here, m0 = 0.849µB for δ = 0 (the quasi-crystalline case). The decrease of impurity local magnetic moment with the growing δ is shown in figure 1. The additional local level shift due to the interaction of condition electrons with the impurity leads to a decrease of the local magnetic moment. The behavior of magnetic moment at constant value of the effective impurity charge is presented in figure 2 when the parameter y = U0/∆ increases. When y is large but finite, magnetic solutions are still possible but as y is reduced they eventually disappear. The diagram describing the region of existence of magnetic and nonmagnetic states is presented in figure 3. The interplay of hybridization and local environment disorder produces a rich structure zero- temperature phase diagram. The region of impurity magnetic states in a disordered metal decreases in contrast to the quasi-crystalline case (δ = 0) [28]. This famous experimental fact for ferromagnetic alloys is discussed in various monographs, see e.g. [7]. The solvation free energy, ∆E , that determines the excess free energy associated with the insertion of an impurity atom into liquid metal, was calculated for this model in our earlier work [27] that corre- sponds to the quasi-crystalline case. The dependence of ∆E , caused by the impurity solvation in liquid metal, on Fermi level x, is shown in figure (4). The dotted lines correspond to the cases when the structural disorder is taken into account. The solid lines correspond to the quasi-crystalline approximation [27]. 6. Conclusions A generalized model proposed in this article permits to calculate the microscopic characteristics of impurity states in liquid metal and to analyze the effect of the structural disorder on the macroscopic properties. Using the equation of motion method for the two-time retarded Green function and using HFA, the system of self-consistent equations for average thermodynamic occupation numbers of localized impu- rity level is obtained. The region of impurity magnetic states in a disordered metal decreases in contrast to the quasi-crystalline case. The contribution to the broadening of virtual impurity level at T = 0 comes from the scattering processes on the charged impurity and from the structural disorder of the impurity 33705-9 G.V. Ponedilok, M.I. Klapchuk Figure 4. (Color online) The solvation free energy of impurity atom in liquid metal. environment as well. This interplay may be relevant to experimental realizations of the system "liquid metal+electronegative impurity" in order to study its magnetic properties. The next possible step of exploration of the proposed model can be the study of Kondo regime taking into account the processes of exchange. In the discussed Hamiltonian (2.7), these processes are described by the following terms, a+k1σ a+k2σ′d0,σ′ d0,σ, d+0,σ′ d+0,σ′ ak2σ′ ak1σ, a+k1σ d+0,σ′ ak2σ′ d0,σ, that correspond to the spin flip processes. They were not accounted for because their matrix elements are of an order of mag- nitude less than the Coulomb matrix elements. However, using these terms and the decoupling scheme beyond the HFA one can analyse the Kondo effect, which is important at low temperatures. This exchange interaction is more likely to increase the polarization of the band electrons rather than to enhance the formation of a magnetic moment. References 1. Anderson P., Phys. Rev., 1961, 124, 41; doi:10.1103/PhysRev.124.41. 2. Edwards S.F., Proc. Roy. Soc., 1962, A267, 518; doi:10.1098/rspa.1962.0116. 3. Roth L.M., Phys. Rev. B, 1972, 7, 4321; doi:10.1103/PhysRevB.7.4321. 4. Ishida Y.,Yonezawa F., Prog. Theor. Phys., 1973, 49, 731; doi:10.1143/PTP.49.731. 5. Kaneyoshi T., J. Phys. C: Solid State Phys., 1972, 5, 3504; doi:10.1088/0022-3719/5/24/013. 6. Schwartz L., Phys. Rev. B, 1973, 7, 4425; doi:10.1103/PhysRevB.7.4425. 7. Xandrich K., Kobe S., Amorphe Ferro- und Ferrimagnetika, Akademie-Verlag, Berlin, 1980. 8. Rudavskii Yu., Ponedilok G., Klapchuk M., Preprint of the Institute for Condensed Matter Physics, ICMP-02-13U, Lviv, 2002 (in Ukrainian). 9. Rudavskii Yu., Ponedilok G., Klapchuk M., Preprint of the Institute for Condensed Matter Physics, ICMP-02-24U, Lviv, 2003 (in Ukrainian). 10. Zubarev D.N., Sov. Phys. Usp., 1960, 3, 320; doi:10.1070/PU1960v003n03ABEH003275 [Usp. Fiz. Nauk, 1960, 71, 71 (in Russian)]. 11. Luo H.G., Ying Ju., Wang S., Phys. Rev. B, 1999, 59, 9710; doi:10.1103/PhysRevB.59.9710. 12. Rudavskii Yu., Ponedilok G., Klapchuk M., Condens. Matter Phys., 2003, 36, 611; doi:10.5488/CMP.6.4.611. 13. Rudavskii Yu., Ponedilok G., Klapchuk M., J. Phys. Stud., 2005, 8, 352. 14. Hubbard J., Proc. R. Soc. Lond. A, 1963, 276, 238; doi:10.1098/rspa.1963.0204. 15. Hewson A.C., Phys. Rev., 1966, 144, 420; doi:10.1103/PhysRev.144.420. 16. Haldane F.D.M., Anderson P.W., Phys. Rev. B, 1976, 13, 2553; doi:10.1103/PhysRevB.13.2553. 17. Haldane F.D.M., Phys. Rev. B, 1977, 15, 281; doi:10.1103/PhysRevB.15.281. 18. Kishore R., Joshi S.K., Phys. Rev. B, 1970, 2, 1411; doi:10.1103/PhysRevB.2.1411. 33705-10 Magnetic states of single impurity in disordered environment 19. Ostrovskii O.I., Grygorian V.A., Vishkariov A.F., Properties of Metallic Alloys, Moscow, Metallurgia, 1988 (in Rus- sian). 20. Harrison U., Pseudopotentials in Theory of Metals, Moscow, Mir, 1968 (in Russian). 21. Yuhnovskii I.R., Gurskii Z.A., Quantum Statistic Theory of Disordered System, Kiev, Naukova Dumka, 1991 (in Ukrainian). 22. Aschkroft N., Lekner J., Phys. Rev., 1966, 145, 83; doi:10.1103/PhysRev.145.83. 23. Smirnow B.M., Negative Ions, Moscow, Atomizdat, 1978 (in Russian). 24. Golovinskii P.A., Zon B.A., Izv. An. SSSR Fiz.+, 1981, 45, 2305 (in Russian). 25. Cooper J.W., Martin J.B., Phys. Rev., 1962, 126, 1482; doi:10.1103/PhysRev.126.1482. 26. Robinson E.J., Geltman S., Phys. Rev., 1967, 153, 4; doi:10.1103/PhysRev.153.4. 27. Rudavskii Yu.K., Ponedilok G.V., Klapchuk M.I., Ukr. J. Phys., 2005, 50, 1308 (in Ukrainian). 28. Rudavskii Yu., Ponedilok G., Klapchuk M., Preprint of the Institute for Condensed Matter Physics, ICMP-06-19U, Lviv, 2006 (in Ukrainian). Магнiтнi стани iзольованої домiшки у невпорядкованому середовищi Г.В. Понедiлок, М.I. Клапчук Нацiональний унiверситет "Львiвська полiтехнiка", Iнститут прикладної математики та фундаментальних наук, вул. С. Бандери, 12, 79013 Львiв, Україна Дослiджується зарядовий та магнiтний стани домiшки, розчиненої в аморфному металiчному сплавi. Га- мiльтонiан системи є узагальненням моделi Андерсона, де додатково враховано процеси пружнього i не- пружнього розсiяння електронiв провiдностi на iонах металу та на зарядженiй домiшцi. Пропонується ме- тод розрахунку конфiгурацiйно усереднених одноелектронних функцiй Грiна в наближеннi Хартрi-Фока. Отримана система самоузгоджених рiвнянь для розрахунку зарядового та спiн-поляризованого стану до- мiшки. Подано якiсний аналiз впливу структурної невпорядкованостi металевої матрицi на спостережу- ванi величини. Показано, що структурний безлад середовища приводить до додаткового розширення та зсуву вiртуального енергетичного рiвня домiшки, зменшуючи магнiтний момент домiшки. Ключовi слова: домiшковiстани,структурнийбезлад 33705-11 

1204.0500

1

1204

Enabling single-mode behavior over large areas with photonic Dirac cones

Many of graphene's unique electronic properties emerge from its Dirac-like electronic energy spectrum. Similarly, it is expected that a nanophotonic system featuring Dirac dispersion will open a path to a number of important research avenues. To date, however, all proposed realizations of a photonic analog of graphene lack fully omnidirectional out-of-plane light confinement, which has prevented creating truly realistic implementations of this class of systems. Here we report on a novel route to achieve all-dielectric three-dimensional photonic materials featuring Dirac-like dispersion in a quasi-two-dimensional system. We further discuss how this finding could enable a dramatic enhancement of the spontaneous emission coupling efficiency (the \beta-factor) over large areas, defying the common wisdom that the \beta-factor degrades rapidly as the size of the system increases. These results might enable general new classes of large-area ultralow-threshold lasers, single-photon sources, quantum information processing devices and energy harvesting systems.

cond-mat.mes-hall

cond-mat

Enabling single-mode behavior over large areas with photonic Dirac cones J. Bravo-Abad,1 J. D. Joannopoulos,2 and M. Soljaci´c2 1Departamento de Fisica Teorica de la Materia Condensada, Universidad Autonoma de Madrid, 28049 Madrid, Spain 2Department of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139 USA Abstract Many of graphene's unique electronic properties emerge from its Dirac-like electronic energy spectrum [1 -- 7]. Similarly, it is expected that a nanophotonic system featuring Dirac dispersion will open a path to a number of important research avenues [9 -- 14]. To date, however, all proposed realizations of a photonic analog of graphene lack fully omnidirectional out-of-plane light confine- ment, which has prevented creating truly realistic implementations of this class of systems. Here we report on a novel route to achieve all-dielectric three-dimensional photonic materials featuring Dirac-like dispersion in a quasi-two-dimensional system. We further discuss how this finding could enable a dramatic enhancement of the spontaneous emission coupling efficiency (the β-factor) over large areas, defying the common wisdom that the β-factor degrades rapidly as the size of the system increases. These results might enable general new classes of large-area ultralow-threshold lasers, single-photon sources, quantum information processing devices and energy harvesting systems. 2 1 0 2 r p A 2 ] l l a h - s e m . t a m - d n o c [ 1 v 0 0 5 0 . 4 0 2 1 : v i X r a 1 Since its isolation from bulk graphite in 2004 [1], graphene -- a one-atom-thick sheet of carbon -- has attracted an ever increasing amount of interest [2 -- 7]; nowadays, the study of the electronic properties of this two-dimensional (2D) material has become one of the most active areas of condensed matter physics. This general endeavor has also stimulated new directions in related research fields, especially those originally inspired by the physics of electronic transport in crystalline solids. Of particular relevance in this context are photonic materials whose dielectric constant is periodically structured at the subwavelength scale, the so-called photonic crystals (PhCs) [8]. By exploiting the analogy between the propagation of electrons in graphene and the propagation of photons in suitably designed 2D PhCs, phenomena such as directional optical waveguiding [9], pseudodiffusive transport of light [10, 11], Klein tunneling [12], and the observation of the Zitterbewegung of photons [13] have been recently proposed. The existence of Dirac points at the center of the Brillouin zone induced by accidental degeneracy in square lattice 2D PhCs, has also been discussed recently [14]. However, all these systems share the common fundamental drawback that they lack fully omnidirectional out-of-plane light confinement, which has so far prevented the creation of a truly realistic implementation of a photonic counterpart of graphene. In this Letter, we propose a feasible approach to achieve simultaneously quasi-two- dimensional light propagation and Dirac cone dispersion in an all-dielectric 3D photonic material particularly suitable for optical device integration. We show how the unique light confining properties of a proper choice of 3D layered PhCs enables creating extended planar defect modes whose dispersion relation exhibits isolated Dirac points inside a complete 3D photonic band-gap. In the limit in which the emitter frequency virtually coincides with the Dirac point frequency (i.e., the frequency of the Dirac cone vertex) the number of photonic states available to the emitter approaches one, even if the system features a macroscopic area. Thereby, the photonic materials presented in this Letter enable for the first time the implementation of structures much larger than the wavelength, which nevertheless have β factors close to unity. Due to the crucial role played by the β-factor in various areas of physics (from optoelectronics to quantum computation or energy harvesting), we believe that these results hold a great promise for the development of novel types of nanodevices. A schematics of the considered photonic material is rendered in Fig. 1a. We start by considering the electromagnetic properties of the PhC before the defect plane is introduced; it is a face-centered cubic (fcc) 3D PhC of air (or low-index) cylinders embedded in a 2 FIG. 1. An isolated photonic Dirac point in a 3D photonic crystal. Panel (a) displays a sketch of the considered structure. The basic structure consists of face-centered cubic (fcc) lattice of overlapping air cylinders embedded in a high dielectric background (yellow regions). This structure can also be seen as an alternating stack of rod and hole layers (see labels in the figure). A planar defect is introduced in the structure by removing a hole layer and replacing it by a triangular array of finite-height dielectric cylinders (green cylinders). The geometrical parameters used to define the structure are also displayed. The inset renders the electric-field intensity corresponding to a guided mode in the defect plane at the Dirac frequency. Panel (b) displays the dispersion diagram corresponding to the structure shown in Panel (a), projected over the First Brillouin zone of the in-plane triangular lattice characterizing the hole and rod layers of that structure. The inset of Panel (b) shows an enlarged view of the dispersion diagram near the Dirac point. dielectric background and oriented along the (1 1 1) direction [15, 16]. This peculiar class of layered PhCs can actually be viewed as an alternating stack of two different types of layers. One of the layers has the form of a triangular lattice of finite-height dielectric rods in air (labeled as rod layer in Fig. 1a), whereas the other layer can be described as a triangular lattice of air holes milled in a dielectric slab (labeled as hole layer in Fig. 1a). This PhC features two important characteristics. First, each of the two types of layers displays a highly symmetric cross section that mimics a canonical 2D photonic-crystal structure: one is a periodic array of air holes in a dielectric dielectric slab, and the other is a periodic array of hexagonal-like rods in air. Second, although neither of the layers displays a complete 3 (omnidirectional) photonic band gap by itself, when the layers are periodically stacked as shown in Fig. 1a, a large complete photonic band-gap can be obtained, using practical values of the refractive index contrast [15, 16]. Next, consider creating an extended planar electromagnetic defect mode in the 3D PhC described above. In order to that, we introduce into the structure a single defect layer that perturbs the original periodic sequence rod-layer/hole-layer/rod-layer along the (1 1 1) direction. In particular, we remove a hole-layer of the structure and replace it by a triangular lattice of finite-height dielectric rods with circular cross section (green cylinders in Fig. 1a), whose radius, height and dielectric constant are given by rd, hd and
d, respectively. It is known [16, 17] that the introduction of line defects into layered 3D PhCs of the type described above enables the implementation of localized electromagnetic states whose dispersion relation, field profiles, and polarization are in a close correspondence with those associated with the corresponding 2D PhC geometries. Similarly, one would expect that the spectral properties of the planar extended states localized in the defect layer rendered in Fig. 1a should, to some extent, inherit properties associated with the Bloch states of the bona-fide 2D counterpart of this defect layer (i.e., those that are present in a triangular array of infinitely long high-index rods). On the other hand, it has been demonstrated that the intrinsic symmetry properties of 2D PhCs based on a triangular lattice can induce the pres- ence of Dirac points near high-symmetry points of the band structure [9 -- 11]. Thus, this line of reasoning suggests the feasibility of creating a single 3D physical system simultaneously featuring quasi-two-dimensional light propagation and Dirac cone dispersion. In addition to these two features, it is also crucial that the mentioned Dirac cone dispersion is isolated within a given frequency bandwidth, or, equivalently, that the Dirac cone is fully separated from the rest of the bands present in the band structure. Only by combining all three of these features in the same system, it is possible to fully exploit the analogy between electronic and photonic graphene. We emphasize that, although, in principle, it is not challenging to obtain each of these three characteristics separately in a photonic structure, achieving all three of them concurrently in the same 3D physical system is not straightforward. To explore to which extent these ideas can be implemented in a realistic system, we have carried out a systematic numerical analysis of the evolution of the corresponding band structures as a function of the geometrical parameters and the dielectric constants, defining both the underlying 3D PhC, as well the defect layer. The calculations were performed 4 by means of the plane-wave expansion method to Maxwell's equations [18] using a supercell large enough in the (1 1 1) direction so the properties of an isolated defect plane in an infinite 3D PhC are accurately reproduced. Our calculations show that, for the optimal structure, the air holes within the hole layer and the equivalent-cylinders [17] in the rod layer are rh = 0.41a and rc = 0.18a respectively (a is the lattice constant of the in-plane triangular lattice defined within each layer; see Fig. 1a). The thicknesses of the hole and rod layers are th = 0.32a and tc = 0.50a, respectively, whereas the refractive index of the high-dielectric material is assumed to be n = 2.5. The low-refractive index of the structure is taken to be air. On the other hand, the defect layer of this optimal system features nd = 3.1, rd = 0.32a, and hd = tc. In addition, from this numerical study we have also obtained that the optimal configuration is that in which the cylindrical rods of the defect layer are aligned with the rods forming the two rod-layers located right above and below the defect plane (see Fig. 1a). Figure 1b summarizes the dispersion diagram obtained for this structure. Shaded violet areas in this figure show the projected band structure for the perfectly periodic 3D PhC (i.e., without the defect layer). In this case, the dispersion diagram was obtained by plotting the frequencies ω of the extended bulk states of the system as a function of the in-plane wavevector k(cid:107) in the irreducible Brillouin zone of the underlying in-plane 2D triangular lattice. The considered system exhibits a large 3D complete photonic band gap (shaded yellow area), centered at frequency ω = 0.497(2πc/a) (c is the light velocity in vacuum) and featuring a gap-midgap ratio of approximately 8%. The dispersion relations of the guided modes of the defect plane are also rendered in Fig. 1b (red lines). As can be seen, these defect bands indeed display a Dirac point at ωD = 0.506(2πc/a) (see Fig. 1b and the inset of Fig. 1a), fully lying within the omnidirectional photonic-band gap of the periodic system (yellow area in Fig. 1b). Importantly, this Dirac point is completely isolated from all of the rest of the frequencies of the band structure of the system within a bandwith ∆ω = 0.026(2πc/a). Finally, the strong out-of-plane photonic-band gap confinement of the electromagnetic fields at ω = ωD is clearly observed in the inset of Fig. 1a, which displays the corresponding cross-section along the xy plane of the electric-field intensity distribution. We believe that these results represent the first time that a complete photonic analog of graphene is implemented in a realistic 3D physical system. We emphasize that the obtained results are scalable to many different frequency regimes, and therefore, could be used to enhance performance of different classes of active optical devices. 5 We now turn to demonstrate how the proposed class of photonic systems can enable an unprecedented control of light-matter interaction over large areas. In order to do that, we apply the commonly employed procedure to probe the properties of light-matter interac- tion in a complex electromagnetic environment: we study the radiation process of a point quantum emitter embedded in the system under analysis [3, 11]. The emitter is modeled as a two-level system, characterized by a transition frequency ωs, an emission bandwidth (i.e., a transition bandwidth) of ∆ωs, and a dipolar transition moment d = d d. In partic- ular, we analyze the spontaneous emission coupling efficiency, the so-called β-factor. In a given multimode photonic system, the β-factor quantifies the portion of all spontaneously emitted photons that couple into a certain targeted mode [22]. This physical magnitude is of a paramount importance in modern optoelectronics and quantum information processing since the performance of an active nanophotonic device, or a single photon emitter, can often be greatly enhanced by increasing the value of β [22 -- 24] (recent examples include ultralow-threshold lasers [25 -- 27] and single photon sources [28, 29] based on PhC cavities). The dependence of the β-factor on the particular electromagnetic environment in which the considered emitter is embedded can be elucidated by examining its link with the corre- sponding photonic local density of states (LDOS), ρ(r, d, ω). For a non-dissipative system, the LDOS can be written as [3] ρ(r, d, ω) = (cid:88) ν
(r) Eν(r) · d2 δ(ω − ων) (1) where the index ν labels the different source-free normal solutions to Maxwell's equations obtained for the considered photonic structure; Eν(r) and ων are their corresponding E-field profile and frequency, respectively.
(r) stands for the dielectric constant distribution. On the other hand, from Fermi's golden rule [3, 11, 19], one finds that, in 3D, the spontaneous emission rate of the considered quantum emitter Γ is proportional to the LDOS accessible to the emitter Γ = (πd2ωs/3¯h
0) ρ(rs, d, ωs). Thus, assuming the targeted mode to be a normal mode of the system, Et(r), of frequency ωt, one can write the β factor as (see Suppl. Information), (cid:82) dω g(ω) ω ρ(rs, d, ω) ωt g(ωt)
(r) Et(r) · d2 β = (2) where g(ω) is the lineshape of the transition, centered at ωs and characterized by full-width- half-maximum (FWHM) of ∆ωs. Note that the factor g(ωt) in the numerator accounts for 6 FIG. 2. Dispersion relations of three representative photonic materials and typical spectrum of a quantum emitter. Panel (a) and (b) sketch, respectively, the dispersion relations of a homogeneous material and a photonic-crystal exhibiting a photonic band gap near the peak of the emission ωs. Panel (c): same as (b) but for the case in which the structure exhibits a Dirac point near ωs. Panel (d) renders the lineshape of the emission spectrum g(ω) in the form of a Lorentzian centered at ωs and featuring a full-width at half-maximum (FWHM) of ∆ωs. ωg and ωD denote the central frequency of the band-gap and the Dirac point frequency, respectively. the fact that β decreases as the emission frequency is detuned from ωt. Hereafter ωt = ωs is assumed. Equation (2) clearly shows that the β factor can be enhanced by introducing a phys- ical mechanism that minimizes the density of photonic states lying within the transition linewidth. In fact, the large values for β reached in subwavelength volume photonic res- onators [25, 26, 28, 29] can be viewed as a particular instance of this physical picture. Such nanoresonators are designed to have a volume small enough so that only one resonant mode lies within the transition linewidth. This makes them to effectively act as single-mode struc- tures, which as deduced from Eq. (1) leads to values β ≈ 1 (provided that the coupling with the radiation modes existing outside the resonator is negligible). Although very relevant in a number of contexts, this cavity-based approach does not admit a straightforward extension for large-area control of β. In the rest of this Letter, we show how the general class of isolated Dirac points discussed above do enable such large-area control. To gain physical insight into the effect of the Dirac cone dispersion on the β factor, we 7 evaluate the magnitude of β in the following three cases (all of them 3D): first, the case of a homogeneous material; second, the case in which the dielectric material is periodically struc- tured so that ωs lies in the vicinity of the lower edge of a 3D photonic-band gap; and, third, the case of the structure displayed in Fig. 1, i.e., a system exhibiting simultaneously quasi- two dimensional light propagation and an isolated Dirac point near ωs. Figures 2a, 2b, and 2c illustrate each of these dispersion relations (for simplicity in the visualization, in Figs. 2a and 2b, only the 2D counterparts of the corresponding cases are rendered). Impor- tantly, in all three cases we also assume that the EM field in the system is confined in a finite volume V , such that V >> λ3 (the dependence on volume is addressed below). The homogeneous case is characterized by the following dispersion relation: ω(k) = ck/n (where n is the refractive index and k = k, with k = (k(cid:107), kz)). Then, making β factor: β = (1/V ) ωsg(ωs)/Fh(ω), where Fh(ω) ≡ (cid:82) dω g(ω) ω ρh(ω), with ρh(ω) = use of the isotropy of the medium, one can derive the following simple expression for the (1/2π)(n/c)3ω2. Similarly, for the case of the photonic band-gap, taking ω(k) = ωg − Agk2 (ωg being the center of the gap and, see Fig. 1b, and Ag is a constant which has to be determined from calculations of the band structure; physically, Ag defines the curvature of the dispersion relation close to ωg), one finds that the magnitude of β can be calculated using the same expression given above for the homogeneous case but replacing Fh(ω) by Fg(ω) ≡(cid:82) dω g(ω) ω ρg(ω), with ρg(ω) = (4π2A3/2 g )−1(ωg − ω)1/2. For the Dirac case, the calculation of β is more involved that in the previous two cases. First, we take into account that an excited dipole embedded in the defect layer displayed in Fig. 1a only decays via the guided modes confined within the layer; the decay into any other modes surrounding the layer (e.g., bulk Bloch modes) is prevented by an omnidirectional photonic band gap. Then, for small enough values of hd (so only the fundamental mode guided mode in the z-direction is excited), the quasi-2D light propagation inside the defect layer can be described by the dispersion relation corresponding to in-plane Bloch states (i.e., states with kz = 0): ω(k(cid:107)) = ωD ± ADk(cid:107) − k(cid:107),0 (see details in Suppl. Information). In this expression, AD is a constant that can be obtained from band-structure calculations and k(cid:107),0 defines the coordinates in k-space of the Dirac cone vertex, whereas the plus and minus signs correspond to ω > ωD and ω < ωD, respectively. (Note that, physically, AD corresponds to the slope of the Dirac cone). Thus, one finds that the total density of states accessible D)ω − ωD. This, in turn, yields the following expression to the emitter as ρD(ω) = 1/(2πA2 8 for the SE coupling efficiency β = (1/A)ωsg(ωs)/FD(ω), with FD(ω) = (cid:82) dωg(ω)ω ρD(ω). Here A is the transversal area of the defect layer (i.e., the total volume of the defect layer is V = A × hd). We now quantify the values of β for each of the above cases using realistic parameter values. To allow for a direct comparison among the three considered class of systems, we introduce a renormalized spontaneous emission coupling efficiency η. In the homogeneous and band-edge cases we define this magnitude as ηh,g = βh,g × V /a3, whereas for the Dirac case we define ηD = βD × S/a2 . This normalization allows us, on one hand, to remove from the discussion the obvious dependence of β on the size of the system, and thus, focus exclusively on the photonic properties. On the other hand, it also removes the geometrical factor V /A that enhances the β factor in the Dirac case which respect to the other two cases. This factor stems from the electromagnetic confinement in the z−direction of the guided modes in the Dirac structure, and therefore cannot be ascribed to the Dirac spectrum. Furthermore, in our calculations we assume that the transition lineshape is described by a Lorenztian centered at ωs = 2.1 × 1015 s−1 (i.e., an emission wavelength of 900 nm) and featuring a relative FWHM ∆ωs/ωs = 10−4; values more than one magnitude smaller for In the ∆ωs/ωs can be reached using, for instance, quantum dots at low temperatures. homogeneous case a refractive index n = nd is chosen, whereas for the band-edge and Dirac cases, the values of the dispersion relation parameters Ag and AD are taken from band structure calculations: Ag = 1.2× ca/(2π) and AD = 0.3× 1/c (a is lattice constant defined in Fig. 1a, which for operation at the considered emission wavelength, takes a value of 450 nm). Thus, using these parameters, from the numerical evaluation of the expression for β given above, we obtain ηh = 68.2, ηg = 4.0 × 105 and ηD = 3.6 × 106. As readily deduced from these values, the Dirac dispersion introduces an enhancement factor of about four orders of magnitude with respect to the homogeneous case and about one order of magnitude with respect to the band-edge case. This is an important result, since, as we show below, enables reaching values of β ≈ 1 over macroscopic areas. Physically, the origin of this dramatic increase of η (and consequently of β) can be understood in terms of the rapid decrease of the number of photonic states available to the emitter as its emission frequency approaches the frequency of the Dirac point. In particular, in contrast to the homogeneous and band- edge cases, in the Dirac case when ωs → ωD (i.e., as emission frequency approaches the 9 FIG. 3. Finite size effects on the β-factor. Dependence of the β-factor on the normalized emission linewidth ∆ωs/ωs, as computed for several sizes of the transverse area A of the system, with a being the periodicity of the in-plane triangular photonic crystal (see definition in Fig. 1a). Dirac vertex frequency), the number of modes accessible to the emitter approaches unity, making the whole structure to effectively behave as a single-mode system, even if it features a large area. Note that the LDOS is strictly zero at ωg and ωD in the band-edge and Dirac cases, respectively. Therefore, in the calculations for each case, we have assumed ωs to be slightly detuned (by a quantity much smaller than ωs and ∆ωs) from ωg and ωD. We now focus on analyzing the dependence on size of the enhancement of β enabled by isolated photonic Dirac cones. The physical origin of this dependence stems from the fact that for values of the area A such that A >> λ2 cannot be safely assumed, it is necessary to account for the discreteness of the eigenmodes in the system: the only states that will be allowed to exist are those characterized by a wavevector k = (kx, ky) whose value coincides with one of the nodes of the rectangular grid defined by the discrete set of values {2πnx/L, 2πny/L} (with nx and ny being arbitrary integers, and where the system is assumed to be square shaped with side length L, i.e., A = L2) [6]. Therefore, the influence of these finite-size effects on β can be computed by using the discrete version of Eq. (2) (see 10 Suppl. Information). Figure 3 shows the computed results for β as a function of the normalized emission bandwidth ∆ωs/ωs for several values of the lateral size of the system, ranging from L/a = 10 to L/a = 103. As observed, in the considered cases, β tends to 1 when ∆ωs/ωs approaches the lower limit of the interval displayed in the figure (∆ωs/ωs = 10−5). This is due to the <∼ 10−5 is smaller than the fact that for all the considered system sizes, a linewidth ∆ωs/ωs frequency interval between the adjacent modes, and therefore, the structure is actually acting as a single mode system (much in the same way as occurs in large-β photonic nanocavities). As ∆ωs/ωs is increased, a growing number of modes enter into the interval where g(ω) is not negligible and, therefore, the value of β starts decreasing. Since the frequency interval between adjacent modes is smaller for larger values of A, the decrease of the β factor with ∆ωs/ωs starts sooner for larger values of A. In conclusion, we have demonstrated for the first time that a 3D photonic material can exhibit isolated Dirac cones in its dispersion relation. The proposed approach is readily accessible experimentally, both for fabrication and measurement. In addition, we have shown how the proposed class of systems enables achieving unprecedentedly large values of β-factor over large areas. We believe that these results could open new avenues in a variety of different fields. Thus, the large β factors demonstrated in this work may lead to the realization of important applications such as low-threshold large-area lasers, enhanced single-photon sources and novel efficient platforms for solar energy harvesting. We acknowledge helpful discussions with Dr. L. Lu and S. L. Chua. JBA was supported in part by the MRSEC Program of the National Science Foundation under award number DMR-0819762 and in part by the Ramon-y-Cajal program of the Spanish MICINN under grant no. RyC-2009-05489. MS was supported in part by the MIT S3TEC Energy Research Frontier Center of the Department of Energy under Grant No. de-sc0001299. This work was also supported in part by the Army Research Office through the Institute for Soldier Nanotechnologies under Contract No. W911NF-07-D0004. [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). 11 [2] Y. Zhang, Y. W. Tan, H. L. Stormer, and P. Kim, Nature 438, 201 (2005). [3] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, Nature 438, 197 (2005). [4] C. Berger, Z. Song, X. Li, X. Wu, N. Brown, C. Naud, D. Mayou, T. Li, J. Hass, A. N. Marchenkov, E. H. Conrad, P. N. Frist, W. A. de Heer, Science 312 , 1191 (2006). [5] A. K. Geim and K. S. Novoselov, Nat. Mat. 6, 183 (2007). [6] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [7] A. K. Geim, Science 324, 1530 (2009). [8] J. D. Joannopoulos, S. G. Johnson, J. N. Winn, and R.D. Meade, Photonic Crystals: Molding the Flow of Light (Princeton University Press, Princeton, NJ, 2008), 2nd ed. [9] F. D. M. Haldane and S. Raghu, Phys. Rev. Lett. 100, 013904 (2008). [10] R. A. Sepkhanov, Ya. B. Bazaliy, and C. W. J. Beenakker, Phys. Rev. A 75, 063813 (2007). [11] S. R. Zandbergen and M. J. A. de Dood, Phys. Rev. Lett. 104, 043903 (2010). [12] O. Bahat-Treidel, O. Peleg, M. Grobman, N. Shapira, M. Segev, and T. Pereg-Barnea, Phys. Rev. Lett. 104, 063901 (2010) [13] X. Zhang, Phys. Rev. Lett. 100, 113903 (2008) [14] X. Huang, Y. Lai, Z. H. Hang, H. Zheng, and C. T. Chan, Nat. Mat. 10, 582 (2011). [15] S.G. Johnson and J.D. Joannopoulos, Appl. Phys. Lett. 77, 3490 (2000). [16] Mi. Qi, E. Lidorikis, P. T. Rakich, S. G. Johnson, J. D. Joannopoulos, E. P. Ippen, and H. I. Smith, Nature 429, 538 (2004). [17] M. L. Povinelli, S. G. Johnson, S. Fan, and J. D. Joannopoulos, Phys. Rev. B 64, 075313 (2001). [18] S. G. Johnson and J. D. Joannopoulos, Opt. Express 8, 173 (2001). [19] E. Purcell, Phys. Rev. 69, 37 (1946). [20] L. Novotny and B. Hecht, Principles of Nano-Optics (Cambridge University Press, Cambridge, England, 2006). [21] Y. Xu, R. K. Lee, and A. Yariv, Phys. Rev. A 61, 033807 (2000). [22] L. A. Coldren and S. W. Corzine, Diode lasers and photonic integrated circuits (Wiley- Interscience, New York, 1995) [23] H. Yokoyama, Science 256, 66 (1992). 12 [24] T. Baba, T. Hamano, and F. Koyama, IEEE J. Quantum Electron. 27, 1347 (1991). [25] O. Painter, R. K. Lee, A. Scherer, A. Yariv, J. D. O'Brien, P. D. Dapkus, and I. Kim, Science 284, 18191821 (1999). [26] H. Altug, D. Englund, and J. Vuckovic, Nat. Phys. 2, 484 (2006). [27] B. Ellis, M. A. Mayer, G. Shambat, T. Sarmiento, J. Harris, E. E. Haller, and J. Vuckovic, Nat. Phot. 5, 297 (2011). [28] C. Santori, D. Fattal, J. Vukovic, G. S. Solomon, and Y. Yamamoto, Nature 419, 594 (2002). [29] W. H. Chang, W. Y. Chen, H. S. Chang, T. P. Hsieh, J. I. Chyi, and T. M. Hsu, Phys. Rev. Lett. 96, 117401 (2006). [30] C. Kittel, Introduction to Solid State Physics (Wiley, New York, 1976). 13 Supporting Information for: Enabling single-mode behavior over large areas with photonic Dirac cones. In the following, we present a detailed description of the theoretical treatment of the spontaneous emission coupling efficiency (the β factor) used in the main text. We start by deriving a general expression for the β factor for the class of structures considered in our work. Within the Wigner-Weisskopf approximation [1], the rate of spontaneous emission Γ of a two-level quantum emitter embedded in a complex electromagnetic (EM) environment can be described by Fermi's golden rule [2, 3] Γ = πd2ωs 3¯h
0 ρ(rs, d, ωs) (1) where d = d d is the dipolar moment associated to the radiative transition of the emitter. Here ωs is the frequency of that transition, whereas
0 and ¯h are the vacuum permittivity and the reduced Planck's constant, respectively. The scalar function ρ(r, d, ω) represents the photonic local density of states (LDOS) accessible to the emitter at position r = rs. On the other hand, in the case of a non-dissipative system, in which the electric field can be expanded in terms of a complete basis of transverse orthonormal modes {Em(r)} of frequencies {ωm}, the LDOS can be expressed as [3 -- 5] ρ(r, d, ω) = δ(ω − ωm)
(r) d Em(r)2 (2) where
(r) is the position dependent dielectric constant characterizing the system. Each of m (cid:88) the modes in Eq. (2) satisfies the following orthonormality condition (cid:90) dr
(r) Em(r) E∗ n(r) = δmn with δnm standing for the Kronecker's delta. The transversality condition reads ∇ · [
(r) Em(r)] = 0 (3) (4) Note also that each mode profile Em(r) can be obtained by solving the following wave equation ∇ × [∇ × Em(r)] = µ0
(r)ω2 mEm(r) (5) where µ0 is the vacuum permeability. 14 Now, by definition, the β-factor can be calculated as β = Γt/Γall, where Γt is the spon- taneous emission rate into a given targeted mode (often a laser mode) and Γall is the total spontaneous emission rate into all the modes of the system (including the targeted one)[7]. Thus, assuming that the lineshape of the emission transition is defined by function g(ω), by inserting Eq. (2) into Eq. (1) and integrating the resulting expression over ω, we obtain the following expression for Γall, corresponding to an emitter located at r = rs (cid:90) Γall = πd2
(rs) 3¯h
0 dω g(ω) ω f (ω, rs) where function f (ω, r) is defined as f (ω, r) = (cid:88) m δ(ω − ωm) d · Em(r)2 (6) (7) Equations (6) and (7) summarize well the physical origin of the total spontaneous emission decay in the considered system: on the one hand, the different terms in the summand of Eq. (7) account for the different modes in which a single frequency component ω of considered emission transition can decay to. On the other hand, the integral in ω appearing in Eq. (6) accounts for the continuous sum of these possible radiative decay paths for all frequency components of the emission transition. Note that, as expected, the lineshape of the emission, g(ω), acts as a frequency dependent weight in this sum. The additional factor ω multiplying g(ω) in the integral of Eq. (6) comes just from the proportionality factor that links the spontaneous emission rate and the LDOS (see Eq. (1)). This physical picture of the decay process also allows us to obtain an expression for Γt simply by singling out the contribution to Γall that stems from the considered targeted mode. In particular, if we define Et(r) and ωt to be the targeted electric-field profile and its corresponding frequency, respectively, the magnitude of Γt can be obtained by substituting g(ω) by gt(ω) ≡ δ(ω − ωt) in Eq. (7) and by replacing f (ω, rs) by ft(ω, r) ≡ δ(ω − ωt)d · Et(r)2. This yields Γt = ωt g(ωt) d · Et(rs)2 πd2
(rs) 3¯h
0 (8) Note that dividing Eq. (6) by Eq. (8), and using the definition of the LDOS given in Eq. (2), we recover Eq. (1) of the main text. We now focus on the application of the above formalism to calculate the β factor in the case in which the emitter is embedded in a three-dimensional photonic crystal (PhC). We assume that the considered PhC is characterized by a finite volume V = Lx×Ly×Lz (where 15 Lx, Ly, Lz are the dimensions of the PhC along the x, y and z axis, respectively). Here we emphasize that in our theoretical calculations, this three-dimensional analysis is applied to the homogeneous and band-edge cases discussed in the main text. (The homogeneous case can be trivially considered as a periodic system with an arbitray periodicity). For the Dirac case, however, due to the out-of-plane subwavelength confinement of the EM fields introduced by the full photonic band gap, the analysis is performed in terms of the in-plane transverse area of the system, A = Lx × Ly (see discussion in the main text). To analyze the finite-size effects on the β-factor, without loss of generality, we assume the volume V (or transversal area A for the Dirac case) to be surrounded by Born-von-Karman boundary conditions (i.e., periodic boundary conditions [6]; our theoretical analysis admits a straightforward generalization to other types of boundary conditions). In this case, the index m used above to label the modes can be identified with {n, k, σ}, where n is the band-index, k is the wavevector of each Bloch mode (k lies inside the First Brillouin Zone (FBZ)) and σ labels the polarization (σ=1 and σ=2, for s- and p-polarization, respectively). In addition, since the system is finite, k can only take discrete values: k = 2π × (nx/Lx, ny/Ly, nz/Lz) for the homogeneous and band-edge cases, and k = 2π× (nx/Lx, ny/Ly, 0) for the Dirac case (in all three cases, nx, ny, and nz are arbitrary integers). Thus, once the normal modes of the system En,k,σ are computed (for which we have used the plane-wave expansion method to Maxwells equations[8]), from Eqs. (6) and (7) the β factor can be calculated from (cid:82) dω g(ω) ω β = (cid:110)(cid:80) n,k,σ δ(ω − ωn,k,σ) En,k,σ(r) · d2(cid:111) ωt g(ωt) Et(r) · d2 (9) In the limit in which the volume V of the system is such that V >> λ3 (or equivalently, for the Dirac case, when the area A is such that A >> λ2), with λ being the central emis- a continuous distribution of k vectors over the FBZ. Specifically, we can replace (cid:80) V /(2π)3(cid:82) sion wavelength, semi-analytical expressions for the β-factor can be obtained by assuming k → F BZ dk in the Dirac case (note that in all three cases the integral over k is performed over the whole F BZ dk in the homogeneous and band-edge cases, and (cid:80) k → A/(2π)2(cid:82) FBZ) . Then, if we expand the argument of the Dirac delta appearing in the denominator of the right-hand side of Eq. (9) using ω − ωn,σ(k0) = ∇k0ω[k0 − k0(ωn,σ)] + O(k0 − k0(ωn,σ)2) (10) and neglect the contribution of second order terms in k0− k0(ωn,σ) [6], after some straight- 16 forward algebra, one finds that Eq. (9) can be rewritten as (cid:82) dω g(ω) ω ρ(ω) ωs g(ωs) β = 1 V (11) where the function ρ(ω) determines the total density of photonic states per unit volume in the structure. Note that in the Dirac case, V must be replaced by the transversal area A. In analogy with standard analyses in solid-state physics [6], in the homogeneous and band-edge cases, ρ(ω) can be expressed as (cid:90) 1 (2π)3 A(ωs) 1 vg dkt ρ(ω) = (12) where A(ωs) denotes the equifrequency surface ω = ωs, and vg is the magnitude of the group velocity vg = dω/dk. At each point of the equifrequency surface, kt stands for the component of the 3D vector k that lies along the tangential direction to A(ωs) at each point of the k-space. In the Dirac case a similar expression holds for ρ(ω), but now the domain of integration in Eq. (12) is a equifrequency curve instead of equifrequency surface. The resulting expressions ρ(ω), obtained by performing the integral defined by Eq. (12) for the different dispersion relations considered in this work, are discussed in detail in the main text. Importantly, in deriving Eq. (11) we have assumed that for the range of parameters con- sidered in this work, the emission bandwidth is narrow enough so that Enkσ(r)2 ≈ Et(r)2 for all modes whose equifrequencies ωnkσ lie inside the interval where g(ω) is not negligible. To verify numerically the accuracy of this approximation in the Dirac case (a similar analysis holds for the band-edge case), we have probed directly the LDOS of the 2D counterpart of the defect layer structure shown in Fig. 1a of the main text. Specifically, in order to do that, we have computed the power radiated by a dipole placed in the low-refractive index regions of the structure (i.e., in the interstitial regions among cylinders), Pout(r, d, ω). To compute Pout(r, d, ω) we have employed a generalization of the conventional coupled-mode theory [9], in which each Bloch mode is considered as an independent input/output channel (see details in Ref. [10]). The computed results are summarized in Fig. 1 of this Supple- mentary Information, in which band structure calculations (Fig. 1a) are displayed together with the corresponding dependence of Pout(r, d, ω) with frequency (Fig. 1b). As observed in these results, the dependence of Pout(r, d, ω) with ω near the Dirac frequency ωD (and hence the LDOS [11]), obtained by assuming Enkσ(r)2 ≈ Es(r)2, is in good agreement 17 FIG. 1. (a) Numerical calculation of two-dimensional photonic bands displaying an isolated Dirac point. The analyzed system is formed by a triangular lattice of dielectric cylinders of refractive index nd = 3.1 and radius rd = 0.32a (a being the lattice constant) embedded in air. (b) Power emitted by a dipole located at the center of the unit cell of the considered photonic crystal and with its dipolar moment pointing along the x-axis (see sketch in top inset of this panel). In both panels, ωD marks the frequency of the Dirac point. Bottom inset in Fig. 1b shows the comparison between the predictions of semi-analytical and full numerical calculations (cyan and blue lines, respectively) for the emitted power. with full numerical calculations within a moderately large bandwidth of frequencies (see comparison between cyan line and blue line in the bottom inset of Fig. 1b). Finally, we note that although in the particular case considered in Fig. 1b the dipole has been placed at the center of the unit cell of the triangular lattice, with d pointing along the x-direction (see top inset of Fig. 1b), we have checked that similar good agreement between numerical and and semi-analytical results is obtained for other positions of the dipole in the unit cell, as well as for other orientations of d. 18 [1] M. O. Scully and M. S. Zuibary, Quantum Optics (Cambridge University Press, Cambridge, 1997). [2] R. Loudon, The quantum theory of light (Oxford Univ. Press, New York, 2000). [3] L. Novotny and B. Hecht, Principles of Nano-Optics (Cambridge University Press, Cambridge, England, 2006). [4] K. Busch and S. John, Phys. Rev. E 58 3896 (1998). [5] R. Sprik, B. A. van Tiggelen, and A. Lagendijk, Europhys. Lett. 35, 265 (1996). [6] C. Kittel, Introduction to Solid State Physics (Wiley, New York, 1976). [7] J. Vuckovic, O. Painter, Y. Xu, and A. Yariv, IEEE J. of Quantum Electr. 35, 1168 (1999). [8] S. G. Johnson and J. D. Joannopoulos, Opt. Express 8, 173 (2001). [9] H. A. Haus, Waves and Fields in Optoelectronics (Prentice-Hall, Englewood Cliffs, NJ, 1984). [10] R. Hamam, M. Ibanescu, E.J. Reed, P. Bermel, S.G. Johnson, E. Ippen, J. D. Joannopoulos, and M. Soljacic, Opt. Express 16, 12523 (2008). [11] Y. Xu, R. K. Lee, and A. Yariv, Phys. Rev. A 61, 033807 (2000). 19 

1504.06370

1

1504

Strain and Electric Field Control of Hyperfine Interactions for Donor Spin Qubits in Silicon

Control of hyperfine interactions is a fundamental requirement for quantum computing architecture schemes based on shallow donors in silicon. However, at present, there is lacking an atomistic approach including critical effects of central-cell corrections and non-static screening of the donor potential capable of describing the hyperfine interaction in the presence of both strain and electric fields in realistically sized devices. We establish and apply a theoretical framework, based on atomistic tight-binding theory, to quantitatively determine the strain and electric field dependent hyperfine couplings of donors. Our method is scalable to millions of atoms, and yet captures the strain effects with an accuracy level of DFT method. Excellent agreement with the available experimental data sets allow reliable investigation of the design space of multi-qubit architectures, based on both strain-only as well as hybrid (strain+field) control of qubits. The benefits of strain are uncovered by demonstrating that a hybrid control of qubits based on (001) compressive strain and in-plane (100 or 010) fields results in higher gate fidelities and/or faster gate operations, for all of the four donor species considered (P, As, Sb, and Bi). The comparison between different donor species in strained environments further highlights the trends of hyperfine shifts, providing predictions where no experimental data exists. Whilst faster gate operations are realisable with in-plane fields for P, As, and Sb donors, only for the Bi donor, our calculations predict faster gate response in the presence of both in-plane and out-of-plane fields, truly benefiting from the proposed planar field control mechanism of the hyperfine interactions.

cond-mat.mes-hall

cond-mat

Strain and Electric Field Control of Hyperfine Interactions for Donor Spin Qubits in Silicon M. Usman,1, ∗ C.D. Hill,1 R. Rahman,2 G. Klimeck,2 M.Y. Simmons,3 S. Rogge,3 and L.C.L. Hollenberg1 1Center for Quantum Computation and Communication Technology, School of Physics, The University of Melbourne, Parkville, 3010, VIC, Australia. 2Electrical and Computer Engineering Department, Purdue University, West Lafayette, IN, USA. 3Center for Quantum Computation and Communication Technology, School of Physics, The University of New South Wales, Sydney, 2052, NSW, Australia. Abstract Control of hyperfine interactions is a fundamental requirement for quantum computing architecture schemes based on shallow donors in silicon. However, at present, there is lacking an atomistic approach including critical effects of central-cell corrections and non-static screening of the donor potential capable of describing the hyperfine interaction in the presence of both strain and electric fields in realistically sized devices. We establish and apply a theoretical framework, based on atomistic tight-binding theory, to quantitatively determine the strain and electric field dependent hyperfine couplings of donors. Our method is scalable to millions of atoms, and yet captures the strain effects with an accuracy level of DFT method. Excellent agreement with the available experimental data sets allow reliable investigation of the design space of multi-qubit architectures, based on both strain-only as well as hybrid (strain+field) control of qubits. The benefits of strain are uncovered by demonstrating that a hybrid 5 1 0 2 r p A 4 2 ] l l a h - s e m . t a m - d n o c [ 1 v 0 7 3 6 0 . 4 0 5 1 : v i X r a control of qubits based on (001) compressive strain and in-plane (100 or 010) fields results in higher gate fidelities and/or faster gate operations, for all of the four donor species considered (P, As, Sb, and Bi). The comparison between different donor species in strained environments further highlights the trends of hyperfine shifts, providing predictions where no experimental data exists. Whilst faster gate operations are realisable with in-plane fields for P, As, and Sb donors, only for the Bi donor, our calculations predict faster gate response in the presence of both in-plane and out-of-plane fields, truly benefiting from the proposed planar field control mechanism of the hyperfine interactions. 1 1. INTRODUCTION During the last few years, there has been significant progress [1 -- 5] towards the realization of quantum computing architectures based on shallow donors in silicon (Si) [6, 7]. Several techniques have been explored to implement precise control of the nuclear or electronic spins through wave function engineering of donors using either electric fields [8] or strain fields [9, 10]. At the core of such approaches, controlled manipulation of the donor hyperfine coupling is a critical component. Previous theoretical studies have been primarily focused on the electric-field-dependent Stark shift of the hyperfine interaction [6] for donors in Si, which is now well understood from both theory [11 -- 15] and experiments [14, 16, 17]. In comparison, the strain- dependence of the hyperfine coupling is relatively less studied, despite offering a promising alternative to manipulate the hyperfine coupling of donors. The presence of strain, in contrast to the use of an electric field, eliminates the possibility of ionization, as the control of the donor wave function is mechanical rather than electrostatic. Additionally strain can drastically reduce valley oscillations of exchange coupling [18, 19], which would play an important role in field control of qubits in strained environments. Recent progress towards atomically precise fabrication of donors in strained Si provides a testbed to demonstrate the advantages of strain in the realization of donor-based qubit devices [20]. Whilst the previous studies have exclusively considered strain or electric field effects on the quantum control of the donors, it is clear that through valley physics there is a subtle interplay between these two effects. This work establishes a multi-scale theoretical approach to provide an understanding of the impact of strain and electric fields simultaneously present in the qubit devices, and predicts that such a hybrid quantum control scheme can open new avenues for architectures with faster single spin gates and spin-dependent tunneling read-out strategies. Existing theoretical studies of the impact of strain on the hyperfine interaction of donors have been based on either valley-repopulation model (VRM) derived from effective-mass theory (EMT) [18, 21 -- 23] or density functional theory (DFT) [9]. While the VRM model was useful in providing a first-order description of the hyperfine shifts for small strain fields, it failed to explain the experimentally measured hyperfine reduction at large strain fields [9]. The DFT calculations for strained Si:P exhibited good agreement with the experimental measurements for an extended range of strain fields [9], highlighting the importance of atomistic approaches. However, this method is limited to only few-atom systems, and is consequently unable to reproduce the donor binding energy spectra and provide a detailed picture of the wave functions [24]. Therefore the requirement for a theoretical framework with an atomistic accuracy accompanied by scalability to large-scale realistic systems remains a critical challenging problem. Our work fills this theory gap by establishing a multi-scale atomistic tight-binding framework, which in 2 contrast to other approaches [18, 25 -- 27] explicitly includes central-cell corrections and non-static dielectric screening of the donor potential. The multi-million-atom simulations for strain-dependent hyperfine inter- action are benchmarked to a high level against both ab-initio approaches and experiment. The electric field dependence of the hyperfine is accurately captured by demonstrating excellent agreement with the experimentally measured Stark shift data [14] for all of the four donor species considered (P, As, Sb, and Bi). A clear understanding of the influence of strain on the physical properties of the donor is presented in terms of underlying valley physics. The performance prospects of unstrained and strained Si substrates are explored, uncovering the benefits of strain for qubit devices, in particular by showing that a hybrid control of qubits based on (001) compressive strain and in-plane fields (100 or 010) results in higher gate fidelities and/or faster gate operations for all of the four donor species. Due to recent research interests for As, Sb, and Bi donors [16, 17, 28], we also present a comparison among different donor species in strained Si environments, further highlighting the trends of strain and electric field induced shifts in the hyperfine couplings, and providing predictions at large strain fields where no previous experimental or theoretical data exists. A novel scheme of two-dimensional hyperfine control in strained environments is explored based on electric fields from top and side gates. Whilst faster gate operations are realisable with the in-plane fields for P, As, and Sb donors, only for the Bi donor, faster gate response is predicted in the presence of both in-plane (100 or 010) and out-of-plane (001) fields, truly benefiting from the proposed planar field control mechanism of hyperfine control. 2. THEORETICAL FRAMEWORK In our TB approach, the Si bulk band structure is reproduced using a twenty-orbital (sp3d5s∗) basis [29]. The donor atom is placed at the center of a large Si box (40×40×40 nm3) consisting of roughly 3.1 million atoms, and is represented by a Coulomb potential, U(r), which is screened by non-static dielectric function for Si and is given by: (cid:0)1 + A
e−αr + (1 − A)
e−βr − e−γr(cid:1) U (r) = −e2
r (1) where e is the electronic charge, and the previously published values of
, A, α, β, and γ are taken from the literature [13, 30]. Recently we have demonstrated the importance of the central-cell corrections and the non-static dielectric screening of the donor potential to accurately reproduce the experimental Stark shift for the Si:As donors [13]. We now show that the non-static dielectric screening is also crucial to accurately reproduce the experimentally measured strained hyperfine interaction at large strain fields. Therefore this 3 work extends the TB model using the non-static screening function to P, Sb, and Bi donors. The donor potential is truncated to U0 at the donor site, r=0. The values of U0 are adjusted to reproduce the experimentally measured binding energy spectra of the donors [31]. By using the U0 values of 3.5 eV, 2.2 eV, 3.8365 eV, and 4.6668 eV for P, As, Sb, and Bi donors respectively, we calculate the binding energies of the ground states A1 within 1 µeV and the binding energies of the excited states (T2 and E) within 1 meV of the experimental values for all of the four donor species. It should be emphasized that whereas multi-valley EMT theories have been successful in fitting the ground state binding energies [14], the fitting of all of the three 1s states simultaneously with such a level of accuracy has been inaccessible. The accurate fitting of the excited state energies, as achieved in this work, is critically important for strain-dependent hyperfine studies, where the excited state E mixes with the ground state A1 as a function of the strain. 3. RESULTS AND DISCUSSION For the study of the strain-dependent hyperfine interaction of donors, we first benchmark our model against the recent experimental data set for the P donors [9]. A commonly adopted procedure to induce strain is by using the lattice mismatch technique, where two materials with different lattice constants are grown on top of each other. Such a technique is depicted in Figure 1(a), where P doped Si is shown on top of a Si1−xGex virtual substrate. The amount of strain in the Si region can be tuned by varying the Ge fraction (x) in the substrate. The lattice constant of Si1−xGex is larger than Si, and therefore induces a tensile strain in P-doped Si region along the in-plane directions (a > aSi). Consequently, the out-of-plane lattice constant (a⊥) shrinks in accordance with the Poisson's ratio, leading to a compressive strain along the growth direction (a⊥ < aSi). For a (001)-oriented Si1−xGex/Si system, the growth axis is the z-axis and the growth plane is the xy-plane, implying that the z-valleys (xy valleys) will primarily experience the effect of a compressive (tensile) strain. 3.1 Characterising strain effects through valley physics: Figure 1(b) schematically illustrates different effects on the splittings of 1s donor states. The valley con- figuration of the lowest energy ground state is also included. A simple effective mass theory without multi-valley effects, such as presented by Kohn-Luttinger [32], would lead to a six-fold degenerate 1s state as shown in Figure 1(b1). In reality, the effect of valley-orbit interactions results in a splitting of the 1s energies into three sets (Figure 1(b2)). The lowest ground state A1 is a singlet state, which is made up of {1, 1, 1, 1, 1, 1}. The first triply-degenerate excited state (T2) has all six valleys with a configuration of 1√ {0, 0, the following valley configurations: T2x= 1√ 0, 0, 1, -1}. The second doubly-degenerate excited states (E) are composed of the following valley config- {0, 0, 1, -1, 0, 0}, and T2z= 1√ 6 2 {1, -1, 0, 0, 0, 0}, T2y= 1√ 2 2 4 FIG. 1. Benchmarking tight-binding theory against experiment and DFT method for P donor: (a) An artist's view of a single P donor in the strained Si. (b) Schematic diagram to indicate the impact of (b2) valley-orbital interaction and (b3) strain on the splitting of 1s energy levels and the valley configuration of the lowest ground state A1. The degeneracy of each energy level is also labeled. (c) The energies of the lowest few states of the P donor in the strained Si as a function of the valley strain (χ), with (c1) for a large variation of strain and (c2) the low strain region (−4 < χ < 0) is zoomed-in to highlight the valley repopulation regime. (c3) The energy difference of the lowest two states as a function of χ. (d) The computed fractional change in the hyperfine interaction A(χ)/A(0) is plotted as a function of χ, and is compared with the previously published experimental measurements, and calculations based on valley-repopulation model and DFT. urations: Exy= 1 12 {-1, -1, -1, -1, 2, 2}. The influence of strain on the donor energies can be understood in terms of their valley configurations: the valleys in the direction of compres- 2{1, 1, -1, -1, 0, 0} and Exyz= 1√ sive strain experience a reduction in energy (higher population) and that valleys in the direction of tensile strain exhibit an increased energy (lower population) [21]. Since the excited states consist of assymetric valley contributions, they experience different effects of strain and therefore do not remain degenerate in the presence of strain (Figure 1(b3)). In our case, the tensile strain along the x and y directions will push 5 the states with xy-valley configurations (T2x, T2y, Exy) up and the compressive strain along the z-axis will shift the states with z-valley configurations (A1z and T2z) downward on the energy scale. The valley repopluation effect for A1z has also been shown by illustrated by showing z-valleys (indicated by the red color) larger in size when compared with the xy valleys (indicated by the green color). in the schemetic diagram of Figure 1(b3). The impact of strain on the donor states can be characterized either directly in terms of the Ge fraction x in the substrate, or can be described in terms of a dimensionless parameter so called the valley strain, χ, which was derived by Wilson and Feher [21], and is given by: (cid:18) aSi − aGe (cid:19)(cid:18) aSi (cid:19) x (2) χ = Ξu 3∆c 1 + 2C12 C11 Here the value of the uniaxial strain parameter Ξu is 8.6 eV, C11 and C12 are the elastic constants of Si and the value of their ratio C12/C11 is 2.6, 6∆c=12.96 eV is the energy splitting of the singlet (A1) and doublet (E) states for the unstrained bulk P donor, aSi=0.5431 nm and aGe=0.5658 nm are the bulk Si and Ge lattice constants respectively, and x is the concentration of Ge in the virtual Si1−xGex substrate. For this study, we vary x between 0 and 0.5, which corresponds to a variation of χ from 0 to ≈ −49. For each value of x, a strained TB Hamiltonian [33] is solved to compute the strained donor energies and states. Figure 1(c1) plots the P donor energies calculated from TB simulations for a large variation of the valley strain χ. The valley repopulation effect primarily occurs for small magnitudes of χ, so Figure 1(c2) presents zoomed-in version of the plot for −4 < χ < 0 to highlight this effect. The effect of applied strain is on donor energies is accurately captured by the TB theory, indicating a partial lift of the degeneracy of the T2 states, splitting them into a single T2z state whose energy decreases due to compressive strain along the z-axis, and a pair of degenerate T2x and T2y states with their energies increasing due to the effect of tensile strain. The T2x and T2y states remain degenerate as the same magnitude of strain is applied along the x and y-axis (ax=ay=a). The strain completely lifts the degeneracy of the E states. The energy of the Exy state increases due to the effect of tensile strain. The remaining two states A1 and Exyz with contributions from all of the six valleys experience the effect of both compressive strain along the z-axis and the tensile strain along the x and y-axis, and therefore exhibit a nonlinear dependence on χ. The strain mixes the Exyz excited state into the ground A1 state. For −4 < χ < 0, the A1 state experiences the competing effects of the tensile and compressive strains. Initially the increase in a is larger than the decrease in a⊥, so the energy of the A1 state slightly increases. However, at the same time strain depopulates x and y valleys and increases z-valley contribution. This reverses the change in A1 due to the effect of decrease in a⊥ being increasingly dominant on the increasingly z-valley like A1 state. For χ < −5, the lowest energy state is dominantly a z-valley state, with mixing from 6 FIG. 2. Theoretically predicted trends among P, Sb, As, and Bi donors : The plots of (a) the A1 energies and (b) the fractional change in the hyperfine couplings A(x)/A(0) are shown as a function of the Ge content x. the Exyz state to form a new ground state A1z. The Exyz state is primarily composed of x and y valleys, and therefore its energy increases with strain. It is noted that the T2z state does not mix with the A1z state as it is composed of two z-valleys with opposite signs. A recent study [14] has defined the ionization field as proportional to the energy splitting (δE) of the ground state A1 and the higher excited state 2p0, which is roughly 34.1 meV for P donors in bulk Si [31]. As evident from Figure 1(c1), the strain reduces δE, which becomes roughly 24 meV for χ ≈ −20, the predicted strain field to suppress the valley oscillations of the J-coupling between P donor pairs [18, 19]. This implies a reduction in the ionization fields for the strained P donors, which would be useful for recently proposed spin-dependent tunneling read-out schemes [34]. The energy difference between the lowest two donor states is relevant in estimating time scales which determine the adiabatic condition in time-dependent processes driven by the gate potential variation. Figure 1(c3) plots this energy difference (∆E12 = T2z−A1z) as a function of χ, indicating a reduction in its value due to the mixing of ground and excited states. At χ ≈ −20, we calculate ∆E12 ≈ 2.5 meV which is smaller than the EMT value of −3.3 meV [18]. 3.2 Hyperfine control by strain: The hyperfine interaction A(0) is directly related to the charge density at the donor site, ψ(0)2. Only the A1 state has a non-zero charge density at the donor site, thus only this state contributes in the determination of A(0). The excited states T2 and E do not contribute to A(0). The applied strain reduces the hyperfine coupling due to the following reasons: (1) Valley Repopulation Effect: strain removes contributions from the x and y valleys, and increases z-valley contribution of the A1 state due to mixing of the Exyz state. The Exyz state does not contribute in A(0), so A(χ) becomes less than A(0). (2) Crystal Deformation: strain deforms the crystal and changes the bond-lengths from their bulk unstrained values. This modifies the radial distribution of the donor states, which are scattered over several Si lattice sites around the donor 7 atom, leading to a reduction in hyperfine. The VRM model based on the EMT theory only considers the first effect and the reduction in A(χ) is represented by an analytical expression derived in Ref. [21]: (cid:34) (cid:16) (cid:17)(cid:18) (cid:19)− 1 2 (cid:35) (3) A (χ) A (0) = 1 2 1 + 1 + χ 6 1 + χ 3 + χ2 4 Based on this model, the fractional change in the hyperfine A(χ)/A(0) as a function of χ is plotted in Figure 1(d) using a dashed green line, along with the experimentally measured values (red dots) from Ref. [9]. Although the VRM method successfully describes A(χ)/A(0) for small values of strain, it fails to capture the strain effects for the larger values of the applied strain (χ < −10). In fact the VRM model limits the value of A(χ)/A(0) to 1/3 for χ < −20, based on the fact that all of the six valleys have equal contributions to the A1 state, whereas only two z-valleys contribute to the A1z state and hence A(χ)/A(0)=2/6. Adding radial redistribution effects in the VRM model only reduces A(χ)/A(0) to 0.3 for χ=−89 [22], still considerably different from the experimental data shown in Figure 1(d), indicating A(χ)/A(0) already reduced to ≈ 0.22±0.09 at χ ≈ −29.5. Therefore in order to fully understand the strain dependence of the hyperfine, a more complete theoretical approach is required which takes into account both the valley-repopulation effect, as well as the volume deformation effect at atomistic scale. Recently reported DFT simulations [9] confirmed this notion by exhibiting a good match with the experimental data for both small and large values of strain (diamonds in Figure 1(d)). Our TB calculations of A(χ)/A(0) as a function of χ are shown in the Figure 1(d) using the square symbols, which demonstrate an excellent agreement with the experimental data as well as with the DFT calculations. For example, at χ ≈ −29.5, we calculate A(χ)/A(0) as ≈0.284, compared to the experimental value of ≈0.22±0.09 and the DFT value of ≈0.27. It is noted that the previously applied static dielectric screening of the donor potential in the TB approach [12, 25, 27, 35] results in a significantly higher value of ≈0.364 for A(χ)/A(0), which emphasizes on the requirement of the non-static (k-dependent) screening of the donor potential for the study of the strain effects. The successful benchmarking of the TB method is in particular useful, because this approach has many advantages over the continuum EMT model and the computationally restricted DFT method. The TB theory not only accurately captures the atomistic physics, it is also scalable to simulation domains with millions of atoms, thereby enabling investigation of multi-qubit architectures. 3.3 Tight-binding predictions for As, Sb, and Bi donors: After benchmarking the TB theory against the experimental data set for the Si:P system, we apply it to predict the influence of strain for other three donors (As, Sb, Bi), which have drawn significant recent research interests [16, 17, 28]. Figure 2(a) plots the energies of the lowest donor state as a function of 8 the Ge concentration x in the Si1−xGex substrate, which is a more relevant parameter for experimentalists compared to χ mainly used in analytical theories. Overall the changes in the A1z energies follow similar trends for all four donors species. Figure 2(b) plots the strain-dependent hyperfine A(x)/A(0) as a function of the Ge fraction x. Again the overall trends are same for all the donors: in the valley-repopulation regime, the strain-dependent hyperfine decreases sharply but for larger strain values, it becomes much less dependent on the applied strain. Interestingly, for a given value of x, the order of ∆A(x)=A(x)/A(0) follows the same trend as the Stark shift parameter η2: ∆ASb(x) < ∆AP(x) < ∆AAs(x) < ∆ABi(x) which is same order as ηSb 2 . One would naively expect this order to depend on the absolute values of the hyperfine interactions (AP(0) < ASb(0) < AAs(0) < ABi(0)), which is not true and in fact this 2 < ηP 2 < ηAs 2 < ηBi sequence is directly related to the order of the binding energies of the donors (ASb 1 < AP 1 < AAs 1 < ABi 1 ). 3.4 Benchmarking Stark shifts of hyperfine couplings against experiments: We have hitherto discussed the strain effects on the hyperfine interactions of donors, which will be useful for the psoposed all-mechanical control of qubits [10]. However, an alternative quantum computing architecture scheme could be based on a hybrid control of qubits, where the donors are present in strained Si and the control is applied by electric fields. Such hybrid control mechanism has certain benefits over traditional unstrained Si based systems, as the applied strain is expected to reduce valley oscillations of exchange coupling [18, 19], as well as ionization fields. To study the effects of strain and electic fields simultaneously present in the qubit devices, we first calculate and benchmark the Stark shift of the hyperfine interactions for all the four donor species under study. The Stark shift is calculated by adding a potential corresponding to an electric field of magnitude varying from 0 and 0.5 MV/m in the diagonal elements of the TB Hamiltonian, and computing the hyperfine interaction A( (cid:126)E) from the field-dependent ground state A1. The change in the hyperfine relative to the absolute value of the hyperfine is then fitted to a quadratic field dependence given by Eq. 4, and the Stark shift parameter (η2) is computed by fitting to our simulation data [12, 13]: (cid:16) (cid:126)E (cid:17) − A (0) A A (0) = η2 (cid:126)E2 (4) The computed values of η2 are plotted in Figure 3 as a function of the A1 binding energies, which are in good agreement with the available experimental measurements. 3.5 Hybrid (Strain+Field) control of qubits: For a bulk donor in unstrained Si, the x, y, and z axes are equivalent with respect to the application of an electric field. However for the donors in strained Si, where a (cid:54)= a⊥, the effect of an electric field along the in-plane direction is expected to be different from its effect along the growth direction. Based on this notion, we suggest a new method of two-dimensional control of the donor hyperfine interaction by applying 9 FIG. 3. Comparison of the calculated Stark shifts with experiments: The calculated values of the Stark shift parameter η2 are plotted as a function of the corresponding ground state binding energies (A1) of shallow donors in bulk Si, which are in good agreement with the available experimental values. electric fields along both the growth direction (z-axis) and one of the in-plane directions (x or y-axis). Such a scheme is schematically illustrated in Figure 4(a), where the P-doped strained Si is displayed on top of Si1−xGex substrate. The direction of electric field can be controlled by top and side gates, which apply fields (cid:126)E⊥ and (cid:126)E along the growth (001) and in-plane (010) directions, respectively. In atomically precise structures, in-plane side gates are frequently used to create an electric field ( (cid:126)E) across the device [2, 36] . The impact of (cid:126)E and (cid:126)E⊥ fields is investigated by varying their magnitudes from 0 to 0.5 MV/m, and calculating the Stark shift η2(x) for each value of the strain, characterized in terms of x. Figure 4(b) and (c) plots the strain-dependent Stark shift values (η2(x)) as a function of x for all of the four donor species, when the applied field is (b) (cid:126)E and (c) (cid:126)E⊥. In both cases, the magnitudes of η2(x) increase overall as a function of the strain. The increase in η2(x) is larger for (cid:126)E compared to (cid:126)E⊥ for the same x. This is because of the fact that the strain compresses the spatial distribution of the donor wave function along the z-direction and therefore the effective Bohr radius along the z-axis is smaller than its values in the in-plane directions [37]. Consequently for the same magnitude of the electric field, the net Stark effect is stronger for the (cid:126)E fields than for the (cid:126)E⊥ fields. 3.6 Strain leads to faster gate operations: In a quantum computer placed in a static field, donors of the same species will lie at or close to resonance. The ability to Stark tune an individual, targeted, spin away from this resonance allows for the addressability of an individual qubit. However, due to a finite linewidth of an excited transition, Stark tuning by a larger frequency leads to higher fidelity and/or faster gates are achievable at the same fidelity. The timescale of an individual spin rotation is limited by the change in frequency provided by the Stark shift [6, 14]: ∆f ( (cid:126)E, 10 FIG. 4. Planar control of hyperfine interactions in strained environments: (a) The schematic diagram illustrates a two dimensional control of the hyperfine interaction by applying (cid:126)E⊥ and (cid:126)E fields along the z and y axes respectively. The plot of the strained hyperfine Stark shift η2(x) for (b) the in-plane field E and (c) the out-of-plane field E⊥ as a function of the applied strain (x). x) = η2(x) (cid:126)E2A(x)mI, where mI is the nuclear spin quantum number. We are interested in comparing the performance prospects of devices based on the strained Si with that of the unstrained Si, therefore we only compare the ratio of the two cases: ∆f ( (cid:126)E, x)/∆f ( (cid:126)E, 0) = η2(x)A(x)/η2A(0). It is interesting to note that for the strained donors, the A(x) value decreases (see Figure 2(b)) as a function of the strain, but η2(x) increases (see Figures 4 (b) and (c)). A multiplication of these two quantities leads to ∆f ( (cid:126)E, x)/∆f ( (cid:126)E, 0) > 1 for (cid:126)E fields, with its largest value being approximately 3 for the Bi donor. Therefore the (cid:126)E field allows for faster control of the qubits in the strained Si substrate in comparison to the unstrained Si. On the other hand, ∆f ( (cid:126)E, x)/∆f ( (cid:126)E, 0) < 1 for the (cid:126)E⊥ fields for the P, As, and Sb donors, thereby implying slower operation with the same fidelity for these donor species. Only for the Bi donor, we calculate that both the (cid:126)E and (cid:126)E⊥ fields exhibit ∆f ( (cid:126)E, x)/∆f ( (cid:126)E, 0) > 1, which is promising given that the decoherence times for Si:Bi have been reported as comparable to that of the Si:P [38]. 4. CONCLUSIONS A multi-scale atomistic framework, established by explicitly describing the cental-cell corrections and the non-static dielectric screening of the donor potential, was used to quantitatively study the individual as well as simultaneous effects of strain and electric fields, as present in various quantum computing architecture schemes currently in development. Our calculations were based on millions of atoms in the simulation domain, and yet described the experimentally measured strain induced hyperfine shifts with an 11 accuracy level of the DFT method. We showed that a hybrid control scheme, where the donors are placed in a strained environment and the control of qubits is by electric fields, offers several advantages such as lowering of the ionization energies/fields and an increased magnitude of the Stark shift from the in-plane electric fields leading to the higher fidelity of single spin gates for all of the four donor species considered. The work demonstrates that the application of both strain and electric fields, and understanding their subtle interplay, has important implications for quantum control in the implementation of Si-dopant based quantum computing architectures. Acknowledgements: This work is funded by the ARC Center of Excellence for Quantum Computation and Communication Technology (CE1100001027), and in part by the U.S. Army Research Office (W911NF- 08-1-0527). MYS acknowledges an ARC Laureate Fellowship. Computational resources are acknowledged from NCN/Nanohub. ∗ musman@unimelb.edu.au [1] M. Fuechsle, J. A. Miwa, S. Mahapatra, H. Ryu, S. Lee, O. Warschkow, L. C. L. Hollenberg, G. Klimeck, and M. Y. Simmons, Nature Nanotechnology 7, 242 (2012). [2] B. Weber, Y. H. M. Tan, S. Mahapatra, T. F. Watson, H. Ryu, R. Rahman, L. C. L. Hollenberg, G. Klimeck, and M. Y. Simmons, Nature Nanotechnology 9, 430 (2014). [3] J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. L. Morton, F. A. Zwanenburg, D. N. Jamieson, A. S. Dzurak, and A. Morello, Nature 496, 334 (2013). [4] K. Saeedi, S. Simmons, J. Z. Salvail, P. Dluhy, H. Riemann, N. V. Abrosimov, P. Becker, H.-J. Pohl, J. J. L. Morton, and M. L. W. Thewalt, Science 342, 830 (2013). [5] J. T. Muhonen, J. P. Dehollain, A. Laucht, F. E. Hudson, R. Kalra, T. Sekiguchi, K. M. Itoh, D. N. Jamieson, J. C. McCallum, A. S. Dzurak, and A. Morello, Nature Nanotechnology 9, 986 (2014). [6] B. E. Kane, Nature 393, 133 (1998). [7] L. Hollenberg, A. D. Greentree, A. G. Fowler, and C. J. Wellard, Phys. Rev. B 74, 045311 (2006). [8] G. P. Lansbergen, R. Rahman, C. J. Wellard, I. Woo, J. Caro, N. Collaert, S. Biesemans, G. Klimeck, L. C. L. Hollenberg, and S. Rogge, Nature Physics 4, 656 (2008). [9] H. Huebl, A. R. Stegner, M. Stutzmann, M. S. Brandt, G. Vogg, F. Bensch, E. Rauls, and U. Gerstmann, Phys. Rev. Lett. 97, 166402 (2006). [10] L. Dreher, T. A. Hilker, A. Brandlmaier, S. T. B. Goennenwein, H. Huebl, M. Stutzmann, and M. S. Brandt, Phys. Rev. Lett. 106, 037601 (2011). 12 [11] A. S. Martins, R. B. Capaz, and B. Koiller, Phys. Rev. B 69, 085320 (2004). [12] R. Rahman, C. J. Wellard, F. R. Bradbury, M. Prada, J. H. Cole, G. Klimeck, and L. C. L. Hollenberg, Phys. Rev. Lett. 99, 036403 (2007). [13] M. Usman, R. Rahman, J. Salfi, J. Bocquel, B. Voisin, S. Rogge, G. Klimeck, and L. C. L. Hollenberg, J. Phys.: Cond. Matt. 27, 154207 (2015). [14] G. Pica, G. Wolfowicz, M. Urdampilleta, M. L. W. Thewalt, H. Riemann, N. V. Abrosimov, P. Becker, H.-J. Pohl, J. J. L. Morton, R. N. Bhatt, S. A. Lyon, and B. W. Lovett, Phys. Rev. B 90, 195204 (2014). [15] F. Zwanenburg, A. S. Dzurak, A. Morello, M. Y. Simmons, L. C. L. Hollenberg, G. Klimeck, S. Rogge, S. N. Coppersmith, and M. A. Eriksson, Rev. Mod. Phys. 85, 961 (2013). [16] F. R. Bradbury, A. M. Tyryshkin, G. Sabouret, J. Bokor, T. Schenkel, and S. A. Lyon, Phys. Rev. Lett. 97, 176404 (2006). [17] C. C. Lo, S. Simmons, R. L. Nardo, C. D. Weis, A. M. Tyryshkin, J. Meijer, D. Rogalla, S. A. Lyon, J. Bokor, T. Schenkel, and J. J. L. Morton1, Appl. Phys. Lett. 104, 193502 (2014). [18] B. Koiller, X. Hu, and S. D. Sarma, Phys. Rev. B 66, 115201 (2002). [19] C. J. Wellard and L. C. L. Hollenberg, Phys. Rev. B 72, 085202 (2005). [20] W. Lee, S. R. McKibbin, D. L. Thompson, K. Xue, G. Scappucci, N. Bishop, G. K. Celler, M. S. Carroll, and M. Y. Simmons, Nanotechnology 25, 145302 (2014). [21] D. K. Wilson and G. Feher, Phys. Rev. 124, 1068 (1961). [22] H. Fritzsche, Phys. Rev. 125, 1560 (1962). [23] E. B. Hale and T. G. Castner, Phys. Rev. B 1, 4763 (1970). [24] H. Overhof and U. Gerstmann, Phys. Rev. Lett. 92, 087602 (2004). [25] G. Klimeck, S. Ahmed, H. Bae, N. Kharche, S. Clark, B. Haley, S. Lee, M. Naumov, H. Ryu, F. Saied, M. .Prada, M. Korkusinski, T. B. Boykin, and R. Rahman, IEEE Trans. Elect. Dev. 54, 2079 (2007). [26] G. Klimeck, S. Ahmed, N. Kharche, M. Korkusinski, M. Usman, M. Parada, and T. Boykin, IEEE Trans. Elect. Dev. 54, 2090 (2007). [27] A. Laucht, J. T. Muhonen, F. A. Mohiyaddin, R. Kalra, J. P. Dehollain, S. Freer, F. E. Hudson, M. Veldhorst, R. Rahman, G. Klimeck, K. M. Itoh, D. N. Jamieson, J. C. McCallum, A. S. Dzurak, and A. Morello, arXiv:1503.05985v1 (2015). [28] G. Wolfowicz, A. M. Tyryshkin, R. E. George, H. Riemann, N. V. Abrosimov, P. Becker, H.-J. Pohl, M. L. W. Thewalt, S. A. Lyon, and J. J. L. Morton, Nature Nanotechnolgy 8, 561 (2013). [29] T. B. Boykin, G. Klimeck, and F. Oyafuso, Phys. Rev. B 69, 115201 (2004). [30] We have used the parameters given by Nara and Morita for P and As donors, and the parameters given 13 by Pantelides and Sah for Sb and Bi donors, based on the best match with the experimental values for the binding energy spectra and η2. [31] A. K. Ramdas and S. Rodriguez, Rep. Prog. Phys. 44, 1297 (1981). [32] W. Kohn and J. M. Luttinger, Phys. Rev. 98, 915 (1955). [33] T. B. Boykin, G. Klimeck, R. C. Bowen, and F. Oyafuso, Phys. Rev. B 66, 125207 (2002). [34] D. McCamey, J. V. Tol, G. W. Morley, and C. Boehme, Science 330, 1652 (2010). [35] S. Ahmed et al., Springer Encyclopedia of Complexity and Systems Science (Berlin: Springer) , p5745 (2009). [36] A. Fuhrer, M. Fuchsle, T. C. G. Reusch, B. Weber, and M. Y. Simmons, Nanoletters 9, 707 (2009). [37] L. Kettle, H.-S. Goan, and S. C. Smith, Phys. Rev. B 73, 115205 (2006). [38] R. George, W. Witzel, H. Riemann, N. V. Abrosimov, N. Notzel, M. L. W. Thewalt, and J. J. L. Morton, Phys. Rev. Lett. 105, 067601 (2010). 14 

1703.10024

2

1703

Squeezed thermal reservoirs as a resource for a nano-mechanical engine beyond the Carnot limit

The efficient conversion of thermal energy to mechanical work by a heat engine is an ongoing technological challenge. Since the pioneering work of Carnot, it is known that the efficiency of heat engines is bounded by a fundamental upper limit, the Carnot limit. Theoretical studies suggest that heat engines may be operated beyond the Carnot limit by exploiting stationary, non-equilibrium reservoirs that are characterized by a temperature as well as further parameters. In a proof-of-principle experiment, we demonstrate that the efficiency of a nano-beam heat engine coupled to squeezed thermal noise is not bounded by the standard Carnot limit. Remarkably, we also show that it is possible to design a cyclic process that allows for extraction of mechanical work from a single squeezed thermal reservoir. Our results demonstrate a qualitatively new regime of non-equilibrium thermodynamics at small scales and provide a new perspective on the design of efficient, highly miniaturized engines.

cond-mat.mes-hall

cond-mat

Squeezed thermal reservoirs as a resource for a nano-mechanical engine beyond the Carnot limit Jan Klaers1∗ and Stefan Faelt1, Atac Imamoglu1, and Emre Togan1 1Institute for Quantum Electronics, ETH Zürich, CH-8093 Zürich, Switzerland The efficient conversion of thermal energy to mechanical work by a heat engine is an ongoing technological challenge. Since the pioneering work of Carnot, it is known that the efficiency of heat engines is bounded by a fundamental upper limit - the Carnot limit. Theoretical studies suggest that heat engines may be operated beyond the Carnot limit by exploiting stationary, non- equilibrium reservoirs that are characterized by a temperature as well as further parameters. In a proof-of-principle experiment, we demonstrate that the efficiency of a nano-beam heat engine coupled to squeezed thermal noise is not bounded by the standard Carnot limit. Remarkably, we also show that it is possible to design a cyclic process that allows for extraction of mechanical work from a single squeezed thermal reservoir. Our results demonstrate a qualitatively new regime of non-equilibrium thermodynamics at small scales and provide a new perspective on the design of efficient, highly miniaturized engines. I. INTRODUCTION Advances in micro- and nano-technology allow for test- ing concepts derived from classical thermodynamics in regimes where the underlying assumptions, such as the thermodynamic limit and thermal equilibrium, no longer hold [1 -- 6]. Extremely miniaturized forms of heat engines, where the working medium is represented by a single particle, have revealed a fluctuation-dominated regime in the conversion of heat to work far away from the ther- modynamic limit [7 -- 11]. By employing non-equilibrium reservoirs it is furthermore expected that the efficiency of work generation surpasses the standard Carnot limit [12], as has been theoretically suggested for quantum coherent [13], quantum correlated [14, 15] and squeezed thermal reservoirs [16 -- 20]. In optics, the electric field of a monochromatic wave can be decomposed into two quadrature components that vary as cos ωt and sin ωt respectively. For co- herent states such as laser light, the uncertainties in the two quadratures are equal and follow the lower bound of Heisenberg's uncertainty relation. By contrast, squeezed states of light have reduced fluctuations in one (squeezed) quadrature at the cost of enhanced fluctua- tions in the other (anti-squeezed) quadrature allowing for optical measurements with reduced quantum noise [21, 22]. Squeezed states are, however, neither restricted to electromagnetic waves nor to minimum uncertainty states. For example, a mechanical oscillator may be pre- pared in a squeezed thermal state [23 -- 25] by a periodic modulation of the spring constant [26]. This results in a state with reduced thermal fluctuations in one quadra- ture (e.g. momentum) and enhanced thermal fluctua- tions in the other quadrature (e.g. position) compared to the expected level of fluctuations at the given temper- ature. In the context of heat engines, a theoretical work by Rossnagel et al. [17] suggests that squeezed thermal ∗ jklaers@phys.ethz.ch states may be used as an additional resource to overcome the standard Carnot limit. Due to the non-equilibrium nature of squeezed thermal reservoirs, this result does not violate the second law of thermodynamics. In our work, we present a physical realization of such an engine with a working medium consisting of a vibrating nano-beam that is driven by squeezed electronic noise to perform work beyond the standard Carnot limit. Furthermore, we demonstrate that by a phase-selective coupling to the squeezed or anti-squeezed quadrature, work can be ex- tracted even from a single squeezed reservoir, which is not possible with a standard thermal reservoir [13, 27]. II. NANO-BEAM HEAT ENGINE Figure 1a shows a sketch of the experiments: The working medium of our heat engine consists of a sin- gle harmonic oscillator given by the fundamental flexural mode of a doubly-clamped (GaAs) nano-beam structure with eigenfrequency ν = ω/2π = 1.97 MHz, quality fac- tor of order Q (cid:39) 103 at room temperature and under vac- uum conditions (p (cid:39) 10−4 mbar). The beam has a length of 18.8 µm, width of 2 µm, thickness of 270 nm, and is fab- ricated using conventional nano-structuring techniques such as electron beam lithography and selective etching. In the growth direction the beam contains two doped layers, which allow us to apply electric fields across the beam material. These electric fields lead to forces being applied to the beam structure due to the piezo-electricity of the employed gallium arsenide material. When driven by a noisy waveform, this creates a random force that allows us to mimic an engineered thermal environment for the beam structure. The waveform is synthesized from two independent white noise signals ξ1,2(t) that are mixed with sine and cosine component of a phase reference at frequency ν leading to a stochastic force f (t) = a0 itive squeezing parameter r corresponds to an amplified cosine and attenuated sine component at frequency ν in the thermal bath, while the overall strength of the noise can be controlled by the amplitude a0. (cid:2)e+rξ1(t) cos(ωt) + e−rξ2(t) sin(ωt)(cid:3). A pos- 7 1 0 2 r p A 5 2 ] l l a h - s e m . t a m - d n o c [ 2 v 4 2 0 0 1 . 3 0 7 1 : v i X r a 2 1b. In the absence of additional noise (a0 = 0), the nano-beam motion is fully determined by the residual thermal noise at room temperature, see panel (i) in Fig. 1b. By increasing the noise amplitude a0, the state of the nano-beam can be prepared in a thermal state at higher temperature T = 1000 K, see panel (ii). Squeezed noise leads to the expected elliptical phase-space probability distribution, see panels (iii) and (iv). We emphasize that all results are presented in a position-momentum-frame that rotates with frequency ν with respect to the labo- ratory frame. The observed probability densities closely follow the theoretically expected Gaussian distribution [23] ρ(x0, p0) ∝ exp − ωx2 0 2kBT1 − ωp2 0 2kBT2 , (1) (cid:18) (cid:19) where x0, p0 are dimensionless position and momen- tum variables and T1, T2 are two temperature param- eters describing the level of fluctuations in the anti- squeezed and squeezed quadratures (proportional to the variance of the gaussian distribution). These parame- ters are related to the effective temperature and squeez- ing parameter of the system by T1,2 = T exp(±2r) [23, 25]. The corresponding energy histograms reveal exponential distributions for cases (i) (circles) and (ii) (boxes) as is expected from purely thermal states (Fig. 1c). The non-exponential decays for parameter sets (iii) (upright triangles) and (iv) (upside-down triangles) demonstrate the non-equilibrium nature of the state of the nano-beam. The solid lines in Fig. 1c show the theoretically expected energy distribution ρ(E) ∝ I0 (E sinh(2r) /kBT ) exp (−E cosh(2r) /kBT ) with I0 as the modified Bessel function of order zero [28]. Fi- nally, the mean energy U of the nano-beam motion as a function of the squeezing parameter r (at fixed tem- perature) closely follows the caloric equation of state U = kBT (1 + 2 sinh2 r), as demonstrated in Fig. 1d. In our experiment, the eigenfrequency of the funda- mental flexural mode can be tuned over a few 100 kHz by applying a DC electrical potential to the nano-beam [28] allowing for a cyclic process with work output. The extraction of work in an engine is normally realized by an increase in volume of a gaseous working medium driving a piston. For a harmonically trapped working medium, such as the one investigated here, an expansion of the working medium may be realized by a decrease in trap- ping frequency ω. The latter suggests to define the vol- ume of the system as the inverse trapping frequency V = ω−1, see Ref. [29]. With this, a relation simi- lar to the state functional formulation of the first law of thermodynamics holds [28] dU = T1dS1 + T2dS2 − pdV , ´ +∞ with S1 = −kB −∞ ρ(x0) ln(ρ(x0)) dx0 as the entropy of the anti-squeezed quadrature and the corresponding defi- nition for S2 being obtained by replacing x0 with p0. The (2) √ Figure 1. Nano-beam heat engine. (a) Doubly-clamped nano- beam piezo-electrically coupled to squeezed electronic noise. (b) Phase space density of the nano-beam motion (in ro- tating frame), (i) when no additional noise is applied, (ii)- (iv) when (squeezed) noise is applied. The fluctuations in position and momentum are characterized by two tempera- ture parameters T1,2 proportional to the Gaussian variance of the line integrated probability densities. The tempera- T1T2 (cid:39) 1000 K for ture is T = 293 K for (i), and T = (ii-iv). The squeezing parameter follows r = 0 for (i-ii), and r = ln(T1/T2)/4 = 0.25, 0.5 for (iii-iv). (c) Measured energy histogram (symbols) of the nano-beam motion in a (squeezed) thermal state. The solid lines show the theoretically expected distributions. Experimental parameters as in Fig. 1b. (d) Av- erage energy of the nano-beam motion (circles) as a function of the squeezing parameter r. The energies are normalized to the energy at vanishing squeezing Ur=0 = kBT , with the tem- perature of the system being kept fixed. The solid line corre- sponds to the theoretical expectation U/Ur=0 = 1 + 2 sinh2 r. Statistical errors (s.e.m.) are smaller than the symbol size for all data points. To demonstrate the generation of squeezed thermal states, we experimentally determine the motional state of the nano-beam by recording the instantaneous posi- tion (out-of-plane displacement) of the beam via Mach- Zehnder interferometry [28] and calculating the corre- sponding phase-space probability distribution, see Fig. 0 1 2 3 4 0 0.5 1Energy E/Er=0Squeezing r 10000 100000 1x106 1x107 1x108 0 0.5 1 1.5CountsEnergy (eV)105104(c)(d)+-(a)15x / 0.478,436 px166,855 px(b)iviiiiiiiviiiiii3 Figure 2. Otto cycle between a cold thermal and a hot squeezed thermal reservoir. (a) Frequency, squeezing and temperature of the nano-beam motion throughout one cycle of the engine (Tcycle amounts to several seconds). The shown protocol implements an Otto cycle at maximum power between a cold thermal reservoir at Tc = 9, 500 K and a hot squeezed reservoir at Th = 10, 000 K with squeezing parameter r = 0.4. The four consecutive strokes include an isentropic compression (A), isochoric heat addition (B), isentropic expansion (C) and isochoric heat rejection (D). (b) Pressure-volume diagram of the Otto cycle. The shaded area corresponds to the total work output W performed by the engine. The solid line represents the theoretically expected behavior. (c) Temperature-entropy diagram of the Otto cycle. T1 (T2) and S1 (S2) denote temperature and entropy of the anti- squeezed (squeezed) quadrature. The solid line shows the theoretical expectations. (d) Efficiency of the nano-beam heat engine as a function of the squeezing parameter of the hot reservoir (circles). The solid blue line shows the theoretical expectation η = 1 −(cid:112)Tc/Th / cosh(r). The measured engine efficiency surpasses the standard Carnot (C) and Curzon-Ahlborn (C.A.) efficiency for finite squeezing parameters (dashed horizontal lines), but obeys a generalized Carnot limit (gC) [17, 27]. All error bars indicate statistical errors (s.e.m.) except for Fig. 2d (see Ref. [28] for details). pressure is defined by p = −(∂U/∂V )S1,S2, which evalu- ates to p = U/V for the given system. In this way, the term −p dV represents the work associated to a change in volume or trapping frequency, whereas the terms Ti dSi with i = 1, 2 describe heat exchange with the environ- ment. III. OTTO CYCLE WITH SQUEEZED THERMAL RESERVOIRS In a first line of experiments, we construct an Otto cycle between a hot squeezed thermal reservoir at Th = 10, 000 K with squeezing factor r = 0.4 and a cold purely thermal bath at Tc = 9, 500 K under maximum power condition [17]. The four strokes include an adiabatic compression, isochoric heat addition, adiabatic expan- sion and isochoric heat rejection. Similar to colloidal heat engines [9], the working medium in our system can- not be fully decoupled from its environment rendering the implementation of adiabatic steps not obvious. Ex- perimentally, it is however feasible to replace the adia- batic steps by isentropic steps. The isentropic steps are implemented by simultaneously varying frequency and temperature such that the ratios T1,2/ω remain constant, which conserves the entropies S1,2 [28]. Figure 2a shows frequency, squeezing and temperature of the nano-beam state throughout one cycle of the engine under quasi- static operation (the cycle time Tcycle amounts to several seconds). The pressure-volume (p-V) and temperature- entropy (T-S) diagrams are shown in Fig. 2b,c. The shaded area in the p-V diagram corresponds to a net work output of W (cid:39) 26 meV performed per cycle. We point out that presently, this work is neither used to per- form a certain task, nor stored in some form of poten- tial energy. The T-S diagram is composed of two closed curves corresponding to temperature and entropy varia- tion of the squeezed (circles) and anti-squeezed quadra- ture (boxes) within one cycle. Note that the curve for the anti-squeezed quadrature is run through clockwise, while the curve for the squeezed quadrature is run through counterclockwise. The net amount of heat consumed from the environment Qh during the hot isochore (B) cor- responds to the difference of the two shaded areas, which is Qh = Q1 − Q2 (cid:39) 244 meV. The efficiency of the cycle finally evaluates to η = W/Qh = (10.6 ± 0.5)%, which is roughly twice the efficiency of a Carnot cycle operating between Th and Tc and four times the Curzon-Ahlborn ef- ficiency [30]. We emphasize that the costs of providing a squeezed thermal bath are not accounted for in our analy- sis, which is fully analogous to neglecting the costs of pro- viding a hot or cold bath in standard thermodynamics. Figure 2d shows the efficiency for varying squeezing pa- rameters of the hot bath. The solid blue line corresponds demonstrate that squeezing, given as a free resource, can be used to increase the work output and efficiency of an to the theoretical expectation η = 1 −(cid:112)Tc/Th / cosh(r), ηC.A. = 1 −(cid:112)Tc/Th respectively. These results clearly see Ref. [17, 28], whereas the two horizontal dashed lines show the (standard) Carnot (C) and Curzon-Ahlborn (C.A.) efficiencies given by ηCarnot = 1 − Tc/Th and 1.25 1.5 1.75 2 2.25 2.5 80 85 90Pressure (pW)Volume (ns)ACDB 1.7 1.8 1.9 2 0 0.25 0.5 0.75 1Frequency (MHz)Time (Tc) 0 0.2 0.4 0 0.25 0.5 0.75 1Squeezing rTime (Tc) 9000 10000 11000 0 0.25 0.5 0.75 1Temperature (K)Time (Tcycle)9k10k11kABCD116,17 px76,17 px76,17 px116,17 px116,17 px(a)(b)(c)BBDD 05k10k15k20k 10.5 11Temperature T1,2 (K)Entropy S1,2 (kB)(d) 0 2 4 6 8 10 12 0 0.1 0.2 0.3 0.4Efficiency (%)Squeezing r4 Figure 3. Work extraction from a single squeezed thermal reservoir. (a) By means of a radio-frequency switch (rf switch in Fig. 1a) the coupling between the squeezed electronic noise and the nano-beam is periodically switched on and off with the phase difference between the phase of the squeezed noise and the switching being denoted by ∆ (top). Mean energy of the working medium coupled to a squeezed bath as a function of the relative phase ∆ for three squeezing parameters r = 0, 0.25, 0.5 (bottom). The data points are interpolated by harmonic functions with varying amplitude and offset (solid lines). (b) Protocol to extract work from a single squeezed reservoir with T = 10, 000 K and r = 0.3 by periodically varying mechanical eigenfrequency ν and phase ∆ (circles in upper two graphs). The four strokes include an isophasal compression, during which the phase relation between squeezed noise and periodic coupling is kept fixed, an isochoric heat addition, isophasal expansion, and isochoric heat rejection. Temperature and squeezing of the nano-beam motional state as consequence of the applied protocol (circles in lower two graphs). The case of an unmodulated coupling is shown with box symbols. Lines between data points are guides to the eye. (c) Pressure-volume diagram revealing a finite amount of work (W (cid:39) 37 meV) extracted from the single squeezed reservoir per cycle (circles). No work output is observed for unmodulated coupling (boxes). Lines between data points are guides to the eye. Statistical errors (s.e.m.) are smaller than the symbol size for all data points. engine beyond the (standard) Carnot limit. The engine operation, however, does obey a generalized Carnot limit (solid line 'gC') [17, 27]. IV. WORK EXTRACTION FROM A SINGLE RESERVOIR In a second line of experiments, we eliminate the cold thermal heat bath in the operation of the engine and in- troduce a phase-selective thermal coupling that will allow to periodically extract work from a single squeezed heat bath. In these measurements, the coupling of the me- chanical oscillator to the squeezed electronic noise is pe- riodically switched on and off by means of a rectangular control signal with frequency 2ν and duty cycle 50%, see top panel in Fig. 3a, that is applied to a radio-frequency switch. The mean energy of the mechanical oscillator as a function of the relative phase ∆ between the phase of the switching function and the phase of the squeezed bath is shown in the bottom panel of Fig. 3a for three differ- ent squeezing parameters. For non-vanishing squeezing parameters (boxes and triangles), the measured energy reaches a maximum when the oscillator is coupled to the anti-squeezed quadrature of the bath at ∆max (cid:39) π/2, whereas an energy minimum is obtained for coupling to the squeezed quadrature at ∆min (cid:39) 3π/2. This phase- selective coupling allows us to extract work from a single squeezed reservoir by a periodic variation of mechanical frequency and phase difference ∆ as shown in the up- per two panels of Fig. 3b (circles). The four strokes in- clude an "isophasal" compression, during which the phase relation between squeezed noise and periodic coupling is kept fixed, an isochoric heat addition, isophasal ex- pansion, and isochoric heat rejection. Temperature and squeezing of the nano-beam state as consequence of the applied protocol are shown in the bottom two panels of Fig. 3b (circles). Finally, the p-V diagram in Fig. 3c clearly reveals a finite amount of work (W (cid:39) 37 meV) that is extracted from the single squeezed reservoir per cycle (circles). Compared to the case of an unmodulated coupling (switch in 'on' position), resulting in a squeezed thermal state with T = 10, 000 K and r = 0.3 through- out the cycle (boxes in Fig. 3b, c), the working medium operates at a reduced temperature. The latter originates from the incomplete decoupling of the oscillator from its environment during the 'off'-phases of the phase-selective coupling. Here, an improved isolation would further in- crease the extracted work per cycle. V. CONCLUSIONS Single-particle heat machines provide an excellent plat- form to test theoretical advances in and our understand- ing of thermodynamics at the micro- and nano-scale. In 1.8 1.85 1.9 1.95 0 0.25 0.5 0.75 1Frequency (MHz)Time (Tc)π/23π/2 0 0.25 0.5 0.75 1Rel. phase ∆Time (Tc)1k2.5k5k10k 0 0.2 0.4 0.6 0.8 1T (K)Time (Tcycle) 0 0.4 0.8 0 0.25 0.5 0.75 1rTime (Tc)99,984 px152,267 px149,598 px99,984 px152,264 pxABCD(b) 0 0.5 1 1.5 2 80 82 84 86 88 90Pressure (pW)Volume (ns)ABCD(c) 0 0.2 0.4 0.6 0π/2π3π/22πEnergy (eV)Relative phase ∆(a)this work, a minimalist heat engine has been realized that takes advantage of squeezed heat to outperform conven- tional heat engines. The non-equilibrium nature of these reservoirs permit work extraction from a single reser- voir and engine efficiencies unbounded by the standard Carnot limit. A major open question remains whether highly miniaturized heat engines, such as the one re- ported here, will be able to use the extracted work to ac- complish mesoscopic tasks as transporting particles [31] or manipulating biological matter [32]. Squeezed ther- mal noise may naturally arise in non-equilibrium envi- ronments with temporally or spatially modulated tem- perature profiles and could be used to fuel such devices. 5 [1] J. Liphardt, S. Dumont, S. B. Smith, I. Tinoco, and C. Bustamante, Equilibrium information from nonequi- librium measurements in an experimental test of Jarzyn- ski's equality, Science 296, 1832 (2002). [2] S. Toyabe, T. Sagawa, M. Ueda, E. Muneyuki, and M. Sano, Experimental demonstration of information- to-energy conversion and validation of the generalized Jarzynski equality, Nature Phys. 6, 988 (2010). [3] A. Bérut, A. Arakelyan, A. Petrosyan, S. Ciliberto, R. Dillenschneider, and E. Lutz, Experimental verification of Landauer's principle linking information and thermo- dynamics, Nature 483, 187 (2012). [4] S. Trotzky, Y. A. Chen, A. Flesch, I. P. McCulloch, U. Schollwöck, J. Eisert, and I. Bloch, Probing the relaxation towards equilibrium in an isolated strongly correlated one- dimensional Bose gas, Nature Phys. 8, 325 (2012). [5] É. Roldán, I. A. Martínez, J. M. R. Parrondo, and D. Petrov, Universal features in the energetics of symmetry breaking, Nature Phys. 10, 457 (2014). [6] For a recent review see: S. Vinjanampathy, and J. An- ders, Quantum thermodynamics, Contemp. Phys. 57, 1 (2016), arXiv preprint arXiv:1508.06099. [7] V. Blickle, and C. Bechinger, Realization of a micrometre-sized stochastic heat engine, Nature Phys. 8, 143 (2012). [8] J. V. Koski, V. F. Maisi, J. P. Pekola, and D. V. Averin, Experimental realization of a Szilard engine with a single electron, P. Natl. Acad. Sci. USA 111, 13786 (2014). [9] I.A. Martínez, É. Roldán, L. Dinis, D. Petrov, J. M. Par- rondo, and R. A. Rica, Brownian Carnot engine, Nature Phys. 12, 67 (2016). [10] J. Rossnagel, S. T. Dawkins, K. N. Tolazzi, O. Abah, E. Lutz, F. Schmidt-Kaler, and K. A. Singer, A single-atom heat engine, Science 352, 325 (2016). [11] S. Krishnamurthy, S. Ghosh, D. Chatterji, R. Ganapathy, and A. K. Sood, A micrometre-sized heat engine operat- ing between bacterial reservoirs, Nature Phys. 12, 1134 (2016). [12] S. Carnot, Réflexions sur la puissance motrice du feu et sur les machines propres a développer cette puissance, Bachelier, Paris (1824). [13] M. O. Scully, M. S. Zubairy, G. S. Agarwal, and H. Walther, Extracting work from a single heat bath via van- ishing quantum coherence, Science 299, 862 (2003). [14] R. Dillenschneider and E. Lutz, Energetics of quantum correlations, EPL 88, 50003 (2009). [15] M. Perarnau-Llobet, K. V. Hovhannisyan, M. Huber, P. Skrzypczyk, N. Brunner, and A. Acín, Extractable work from correlations, Phys. Rev. X 5, 041011 (2015). [16] X. L. Huang, T. Wang, and X. X. Yi, Effects of reservoir squeezing on quantum systems and work extraction, Phys. Rev. E 86, 051105 (2012). [17] J. Rossnagel, O. Abah, F. Schmidt-Kaler, K. Singer, and E. Lutz, Nanoscale heat engine beyond the Carnot limit, PRL 112, 030602 (2014). [18] L. A. Correa, J. P. Palao, D. Alonso, and G. Adesso, Quantum-enhanced absorption refrigerators, Sci. Rep. 4, 3949 (2014). [19] W. Niedenzu, D. Gelbwaser-Klimovsky, A. G. Kofman, and G. Kurizki, On the operation of machines powered by quantum non-thermal baths, New J. Phys. 18, 083012 (2016). [20] W. Niedenzu, V. Mukherjee, A. Ghosh, A. G. Kofman, and G. Kurizki, Universal thermodynamic limit of quan- tum engine efficiency, arXiv:1703.02911 (2017). [21] LIGO Scientific Collaboration, A gravitational wave ob- servatory operating beyond the quantum shot-noise limit, Nature Phys. 7, 962 (2011). [22] F. Wolfgramm, A. Cere, F. A. Beduini, A. Predojević, M. Koschorreck, and M. W. Mitchell, Squeezed-light optical magnetometry, PRL 105, 053601 (2010). [23] H. Fearn and M. J. Collett, Representations of squeezed states with thermal noise, J. Mod. Optic. 35, 553 (1988). [24] M. S. Kim, F. A. M. de Oliveira, and P. L. Knight, Prop- erties of squeezed number states and squeezed thermal states, Phys. Rev. A 40, 2494 (1989). [25] R. R. Tucci, Entropy of a harmonic oscillator in contact with a squeezed reservoir, Int. J. Mod. Phys. B 5, 545 (1991). [26] D. Rugar and P. Grütter, Mechanical parametric ampli- fication and thermomechanical noise squeezing, PRL 67, 699 (1991). [27] O. Abah and E. Lutz, Efficiency of heat engines coupled to non-equilibrium reservoirs, EPL 106, 20001 (2015). [28] See supplemental material for further experimental and theoretical details. [29] V. Romero-Rochín, Equation of state of an interacting Bose gas confined by a harmonic trap: The role of the "harmonic" pressure, PRL 94, 130601 (2005). [30] F. L. Curzon and B. Ahlborn, Efficiency of a Carnot engine at maximum power output, Am. J. Phys. 43, 22 (1975). [31] P. Hänggi and F. Marchesoni, Artificial Brownian mo- tors: Controlling transport on the nanoscale, Rev. Mod. Phys. 81, 387 (2009). [32] S. M. Douglas, I. Bachelet, and G. M. Church, A logic- gated nanorobot for targeted transport of molecular pay- loads, Science 335, 831 (2012). 

1703.08399

1

1703

Coupling graphene nanomechanical motion to a single-electron transistor

Graphene-based electromechanical resonators have attracted much interest recently because of the outstanding mechanical and electrical properties of graphene and their various applications. However, the coupling between mechanical motion and charge transport has not been explored in graphene. Here, we studied the mechanical properties of a suspended 50-nm-wide graphene nanoribbon, which also acts as a single-electron transistor (SET) at low temperature. Using the SET as a sensitive detector, we found that the resonance frequency could be tuned from 82 MHz to 100 MHz and the quality factor exceeded 30000. The strong charge-mechanical coupling was demonstrated by observing the SET induced ~140 kHz resonance frequency shifts and mechanical damping. We also found that the SET can enhance the nonlinearity of the resonator. Our SET-coupled graphene mechanical resonator could approach an ultra-sensitive mass resolution of ~0.55*10^(-21) g and a force sensitivity of ~1.9*10^(-19) N/(Hz)^(1/2), and can be further improved. These properties indicate that our device is a good platform both for fundamental physical studies and potential applications.

cond-mat.mes-hall

cond-mat

Coupling graphene nanomechanical motion to a single- electron transistor Gang Luo, Zhuo-Zhi Zhang, Guang-Wei Deng*, Hai-Ou Li, Gang Cao, Ming Xiao, Guang-Can Guo, and Guo-Ping Guo* Key Laboratory of Quantum Information, University of Science and Technology of China, Chinese Academy of Sciences, Hefei 230026, China. Synergetic Innovation Center of Quantum Information and Quantum Physics, University of Science and Technology of China, Hefei, Anhui 230026, China. *Correspondence to: gwdeng@ustc.edu.cn or gpguo@ustc.edu.cn 1 / 15 Abstract: Graphene-based electromechanical resonators have attracted much interest recently because of the outstanding mechanical and electrical properties of graphene and their various applications. However, the coupling between mechanical motion and charge transport has not been explored in graphene. Here, we studied the mechanical properties of a suspended 50-nm-wide graphene nanoribbon, which also acts as a single-electron transistor (SET) at low temperature. Using the SET as a sensitive detector, we found that the resonance frequency could be tuned from 82 MHz to 100 MHz and the quality factor exceeded 30000. The strong charge-mechanical coupling was demonstrated by observing the SET induced ~140 kHz resonance frequency shifts and mechanical damping. We also found that the SET can enhance the nonlinearity of the resonator. Our SET-coupled graphene mechanical resonator could approach an ultra-sensitive mass resolution of ~0.55 × 10−21 g and a force sensitivity of ~1.9 × 10−19 N/(Hz)1/2, and can be further improved. These properties indicate that our device is a good platform both for fundamental physical studies and potential applications. Keywords: Graphene, nanomechanical resonator, single electron transistor, strong TOC coupling, Duffing nonlinearity 2 / 15 Introduction: Because of its large Young's modulus (~1 TPa) and low mass, graphene is extremely promising for use in nanoelectromechanical systems (NEMS).1-4 A graphene mechanical resonator is considered an ideal force5 and mass sensor6 because of its large resonance frequency and high quality factors, and it is also expected to exhibit quantum behavior at low temperature.7-9 Conversely, because of its high electron mobility,10 graphene has also been studied for etched11 or gate-defined12 quantum dots, which have been proposed as quantum bits.13, 14 Suspended graphene has shown ultrahigh electron mobility,15 offers a platform to observe the fractional quantum Hall effect,16, 17 and it also performed well as quantum dots.18-20 Thus, these properties open the way to study the interplay between mechanical and electronic degrees of freedom in graphene. Previous experiments have demonstrated ripple texturing of suspended graphene,21 phonon softening of strained graphene,22 a strain-induced zero-field quantum Hall effect,23 and 300 Tesla pseudo-magnetic fields.24 The relationships between strain and electrical resistance of suspended graphene have also been studied by AFM.25, 26 Graphene has potential applications at room temperature: quantum Hall effect27 and quantum dot behavior28 at 300 K have been reported. Moreover, graphene nanoribbon is easy to scale up.3, 29, 30 However, an experiment studying the coupling between a mechanical resonator and single-electron transistor (or quantum dot) in a graphene system is still lacking. Many theoretical investigations have studied the interplay between mechanical motion and single-electron tunneling, where the mechanical motion is reported to affect the electron transport current31 and current noise;32 in turn, the back-action from the electron tunneling should cause frequency shifts33 and damping33, 34 in the mechanical modes. There are proposals to realize single-electron shuttles by mechanical motion35, 36 and ground state cooling of the mechanical motion by quantum dots;37 thus, further study could be performed in the quantum regime.38, 39 Several pioneering experiments have demonstrated superconducting single-electron transistors as position detectors,40- 42 and strong coupling between single-electron tunneling and mechanical motions have been reported in carbon nanotubes.43-46 In the case of graphene resonators, however, it 3 / 15 is unknown whether the charge transport can be coupled to the mechanical motion. Figure 1. (a) Scanning electron microscopy figure of the nanoribbon-type graphene mechanical resonator. The doubly-clamped resonator was fabricated using mechanically exfoliated graphene. Two Ti/Au metal contacts were used as the source (S) and drain (D) of the quantum dot. A bottom Ti/Au gate beneath the graphene ribbon was used to apply DC and AC voltages, which can tune the chemical potential of the quantum dot, actuate the resonator, and change the mechanical tension. The blue curve shows the schematic potential, and the red dot indicates the quantum dot. (b) Schematic diagram of the sample and the measuring circuit. The substrate was undoped Si with 300 nm of SiO2. A bias-T, shown in the dashed box, was used to combine DC and AC voltages applied to the bottom gate. In this letter, we report the first experimental realization of coupling between a single-electron transistor and mechanical motion in a graphene nanoribbon. We find 4 / 15 that the mechanical motion can be greatly affected by single-electron tunneling into and out of the graphene ribbon. The electron transport induced frequency shift is measured to be as large as 140 kHz, which is 56 times larger than the smallest resonance linewidth of our sample. We further observed a large reduction in the quality factor in the quantum dot region, which could be useful to study the nonlinear damping and cooling of graphene resonators.47, 48 Moreover, the quantum dot affected the nonlinear dynamics of the resonator. These results show a strong coupling between electron transport and mechanical motion in graphene, which could be very useful in future high-frequency resonator measurements and for cooling the mechanical oscillations to the quantum regime. Graphene resonator has been demonstrated to interact with magnetic field recently,49 and we expect more new physics when a quantum dot participates in this system in the future. Results: Our sample structure is shown in Figure 1, where a 50-nm-wide (𝑊), 2-μm-length ( 𝐿 ) graphene nanoribbon was suspended across two metal electrodes. A 5-layer (confirmed with AFM) graphene ribbon was transferred using an all-dry viscoelastic stamping technique,50 without any further chemical etching process, which may harm the edge of the ribbon and leave chemical residues. A 40-nm Ti/Au (5 nm Ti and 35 nm Au) bottom electrode was deposited in a 150-nm deep (confirmed with AFM) silicon- oxide groove by e-beam lithography and a lift-off procedure, together with the two contact electrodes. Thus, the graphene mechanical resonator is expected to have a 150- nm distance from its bottom gate, ideally. We chose undoped Si with 300 nm of SiO2 as the substrate because doped Si may absorb microwave power. However, this made it difficult to tune the chemical potential of the sample in a large range. The only way to control the potential in our sample is through the bottom gate. The sample was mounted in a He-3 refrigerator with a base temperature of approximately 270 mK and at pressures below 10−6 torr. We actuated the suspended resonator by applying an AC field (𝑉𝑔 𝐴𝐶 ) to the bottom gate, together with the DC voltages (𝑉𝑔 𝐷𝐶), using a bias-T (Figure 1b). We applied a DC bias (𝑉𝑠𝑑) to the source 5 / 15 contact and simply measured the DC current (𝐼𝑠𝑑) from the drain contact (Figure 1b). The ribbon was first annealed by the in-situ current annealing method (see the supplementary information). Previously, graphene nanomechanical resonators have been commonly analyzed by optical1, 3 and mixing2 techniques or by cavity-based capacitive readout methods.8, 51 These methods require careful treatment of the parasitic capacitance and become difficult when the resonance frequency is as large as 1 GHz.52 Quantum dots have been shown to detect mechanical motion with resonance frequencies up to 39 GHz in carbon nanotube devices,53 and we expect a similar result in graphene systems for future high-frequency samples. Our 50-nm-wide graphene nanomechanical resonator also acts as a quantum dot, and we show its transport properties in Figure 2b, where the Coulomb blockade is clearly observed as a function of the gate voltage. The charging energy is estimated to be ~1 meV (see the supplementary information), and this kind of nanoribbon-type graphene quantum dot has been previously studied by several groups.18, 19, 54 Figure 2c shows a typical DC current response as a function of the driving frequency Figure 2. (a) Current response as a function of the driving frequency and the gate 6 / 15 voltage. To obtain better resolution, the data were acquired using the differential of the frequency axis. Two resonance modes can be observed by the transport measurement. The first mode is "W" type, and the second one could not be resolved at small gate voltage. (b) Transport property of the sample as a function of the gate voltage, which shows clear Coulomb peak structures. The bias voltage 𝑉𝑠𝑑 is fixed at 20 μV for this curve. (c) Fitting the current as a function of the driving frequency. The data were obtained at -70 dBm, as the average of 10 measurements. (d) Zoom of the dashed black box in panel a. The related resonance frequency goes up and down, corresponding to the Coulomb peak and blockade regime. This diagram means that single-electron tunneling modulates the resonance frequency of the mechanical resonator, indicating that the two systems are coupled. near its resonance center. Here the DC and AC signal applied to the bottom gate lead to a mean electrostatic force on the resonator,45 𝐹 = 1 2 ∂𝐶𝑔 ∂𝑧 (𝑉𝑔 DC − 𝑉𝑑𝑜𝑡 + 𝑉𝑔 AC)2, (1) where 𝐶𝑔 is the capacitance between the bottom gate and the resonator, and 𝑉𝑑𝑜𝑡 is the potential of the graphene ribbon and depends on the occupation of the charge number;43, 44, 55 through the capacitance, the DC voltage 𝑉𝑔 DC can affect the tension of the resonator, which may change the resonance frequency of the resonator (see Figure 2a). In addition, the gate voltage also controls the electrochemical potential of the quantum dot, as we have discussed previously. The change in 𝑧 is with respect to the displacement vibration 𝛿𝑧(ω, 𝑡) = 𝐴(ω)cos(ω𝑡 + 𝜙). Here, 𝜔 = 2𝜋𝑓 is the circle frequency of the driving AC signal and 𝜙 comes from the phase difference between the driving field (𝑉𝑔 AC = 𝑉𝑔 RFcos (2𝜋𝑓𝑡)) and the vibration. When the AC driving frequency 𝑓 approaches the resonance frequency 𝑓0 = 𝜔0/2𝜋 , the mechanical vibration is effectively actuated and modulates the electrochemical potential of the quantum dot, resulting in a change in the conductance; thus, a current peak/dip (Figure 2c) can be observed. The current changes with time as 𝐼SD(𝑡) = 7 / 15 ∑ 1 𝑛! 𝑛 DC(𝑉𝑔) 𝑑𝑛𝐼SD 𝑛 𝑑𝑉𝑔 [ DC 𝑉𝑔 𝐶𝑔 ∂𝐶𝑔 ∂𝑧 𝑛 𝛿𝑧(ω, 𝑡)] , resulting in a measured current change of45 ∆𝐼sd(𝑡) ≈ 1 4 DC(𝑉𝑔) 𝑑2𝐼SD 2 𝑑𝑉𝑔 [ DC 𝑉𝑔 𝐶𝑔 ∂𝐶𝑔 ∂𝑧 𝐴(ω)]2, (2) where 𝐴(𝜔) = 𝜕𝐶𝑔 𝜕𝑧 RF DC𝛿𝑉𝑔 𝑉𝑔 𝑚eff × 1 √(2𝜔0(𝜔0−𝜔)) 2 4 𝜔0 𝑄2 + (3) is the amplitude of the mechanical resonator without considering of the nonlinearity at very low driving powers. In Figure 2c, we show a typical fitting of the current change as a function of the driving frequency and obtain a line width 𝛾 = 2.5 kHz, which corresponds to a quality factor 𝑄 = 𝜔0/𝛾 = 33447 and yields an energy relaxation time of 400 μs. The frequency-Q product is usually considered as the figure of merit when comparing different mechanical resonator systems. We reached a frequency-Q product of 𝑓𝑄~3 × 1012 Hz, which is comparable with that of carbon nanotube devices.56 We estimated the effective mass of the resonator to be 𝑚eff~1.85 × 10−19 kg for the 𝑁 = 5-layer graphene nanoribbon, where 𝑚eff = 0.5𝑁𝜌𝐿𝑊 and 𝜌 = 7.4 × 10−19 kg/μm2. In the situation with low driving power, the force can be simply treated as 𝐹 = 𝑘𝛿𝑧, where 𝑘 is the spring constant. Simply, the resonance frequency can be obtained as 𝑓0 = 1 2𝜋 √ 𝑘 𝑚eff , from which we extract 𝑘~0.05 N/m. We further study the interaction between the quantum dot and the mechanical resonator in Figure 3. Compared with Figure 2c, Figure 3 was measured with larger driving power (-50 dBm) and better resolution, but a smaller quality factor. We find that 𝜔0, 𝛾, and 𝑄 (Figure 3b,c,d) all oscillate as a function of the gate voltage, together with the Coulomb oscillations shown in Figure 3a. The resonance frequency shift is caused by the back action force induced by the fluctuation of electron transport on the resonator. We found that the largest frequency shift in our sample is approximately ∆𝜔0 = 140 kHz × 2𝜋, which is 56 times larger than the smallest resonance linewidth we have measured, indicating a strong coupling of the graphene mechanical motion and the single-electron tunneling. Even compared with the same measurement shown in Figure 3c, this frequency shift is 6 times larger than the largest linewidth around the Coulomb peak. From the blockade to the Coulomb peak, the electron-phonon coupling 8 / 15 Figure 3. (a) Current as a function of gate voltage, showing three Coulomb peaks. (b- d) The corresponding frequency response (b), dissipation rate (c) and quality factor (d). induced the mechanical damping, increasing the linewidth from 13 kHz to 26 kHz and accordingly reducing the quality factor from 6000 to 3000. This result means that charge transport plays an important role in the mechanical dissipation mechanism at low temperature. The large frequency shifts and damping rates caused by the Coulomb peaks indicate a strong coupling between graphene mechanical motion and single- electron tunneling.43-45 This kind of strong coupling could be very useful to cool graphene mechanical resonators to the ground state and for further applications. With large driving power, the resonator enters the nonlinear regime, which is a result 9 / 15 Figure 4. Nonlinear dynamics. (a) Transport property as a function of the driving frequency and the gate voltage at large driving power (-48 dBm), where the nonlinear response is clearly observed and enhanced by the quantum dot. (b) Current as a function of the driving power and frequency in the Coulomb blockade region (white arrow in panel a), where the background current is zero. Linewidth broadening can be clearly observed when the power is larger than -50 dBm. (c) Transport current as a function of the driving frequency, for various powers. The units are dBm. (d) Nonlinear fit of the measured data; the fitted Duffing coefficient in this figure is 𝛼 = 2 × 1016 kg/m2s2 10 / 15 and obtained from point A shown in panel a (𝛼 = 5 × 1016 kg/m2s2 for point B). of the resonator being two dimensional. Without the effect from the quantum dot, we consider the Duffing nonlinearity term 𝛼(𝛿𝑧)3 and high order nonlinear damping term 𝜂(𝛿𝑧)2 𝑑𝑧 𝑑𝑡 , the Newton equation for a nonlinear resonator can then be described as:47,56 𝐹 = 𝑚𝑒𝑓𝑓 𝑑2𝛿𝑧 𝑑𝑡2 + 𝑘𝛿𝑧 + 𝛾 𝑑𝛿𝑧 𝑑𝑡 + 𝛼(𝛿𝑧)3 + 𝜂(𝛿𝑧)2 𝑑𝛿𝑧 𝑑𝑡 . (4) Regarding the nonlinearity induced by the SET, we describe the SET force by:55 𝐹𝑆𝐸𝑇 = −𝑘𝑆𝐸𝑇𝛿𝑧 − 𝛽𝑆𝐸𝑇(𝛿𝑧)2 − 𝛼𝑆𝐸𝑇(𝛿𝑧)3, (5) where the SET induced nonlinearity 𝛼𝑆𝐸𝑇 can be simply renormalized to the total Duffing term.55 And 𝛿𝑧 is defined as the small vibrations around a static equilibrium 𝑧0 (a fixed reference point), which depends on the DC voltage 𝑉𝑔 DC. Figure 4 shows the nonlinear dynamics of the sample, where the nonlinear induced linewidth broadening becomes obvious when the driving power is larger than -50 dBm (Figure 4b). This gives a 20 dB dynamic range (DR), which is very useful for mass sensing. The mass resolution was previously given as δ𝑚 = 𝑚𝑒𝑓𝑓 𝑄 ∙ 10−𝐷𝑅/20,57 from which we obtained a mass resolution of ~0.55 × 10−21 g (or 0.55 zeptograms). The measured transport current was found to be saturated as a function of driving frequency at very high driving power (Figure 4c). Neglecting the high order damping term, we fit the Duffing coefficient to be 𝛼~2 × 1016 kg/m2s2 in a Coulomb blockade region (Figure 4d), which is comparable with a previous report.56 We also found that quantum dots could enhance the Duffing nonlinearity (Figure 4a), where the Duffing coefficient in the quantum dot region increased to 𝛼~5 × 1016 kg/m2s2. This single-electron tunneling enhancing Duffing nonlinearity has been reported in a carbon nanotube system;55 however, it has not been observed in graphene. Exploring the onset of nonlinearity in a two-dimensional material is quite interesting and important, and may bring new understanding of high-order nonlinear physics. The QD-coupled mechanical resonator has a small DR; however, the small mass and high quality factor contribute to a high sensitivity (at zg level) for the mass resolution. 11 / 15 A wider device will have higher DR; however, it will have a larger mass (see supplementary information). Because the device has a large frequency-Q product, we expect a large force sensitivity 𝐹𝑚𝑖𝑛 = (4𝑘𝑏𝑇𝑘/𝜔𝑄)1/2 ,58 where 𝑘𝑏 is the Boltzmann constant and 𝑇 = 270 mK is the temperature. We calculate the force sensitivity to be 𝐹𝑚𝑖𝑛~1.9 × 10−19 N/(Hz)1/2 (or 19 aN/(Hz)1/2) for the current setup. In future work, by cooling down the sample to lower temperature and enhancing the quality, this sensitivity could be further optimized. In conclusion, we have realized the first coupling between the mechanical mode and single-electron tunneling in a graphene device. We find that the 50-nm-wide graphene resonator shows a high quality factor of > 3 × 104 and resonance frequency as high as 100 MHz. Therefore, the resonator could act as a good mass and force sensor. The resonator also acts as a quantum dot at low temperature, and single-electron tunneling causes a > 100-kHz frequency shift of the resonator. Our device also offers a platform for the study of nonlinear physics with respect to quantum dots. Further investigations may use this device for detecting ultra-high-frequency resonator and cooling the mechanical modes to the quantum regime. Acknowledgment This work was supported by the National Key R & D Program (Grant No.2016YFA0301700), the Strategic Priority Research Program of the CAS (Grant No. XDB01030000), the National Natural Science Foundation (Grant Nos. 11304301, 11575172, 61306150, and 91421303), and the Fundamental Research Fund for the Central Universities. Competing financial interest: The authors declare that they have no competing financial interests. Supporting Information Available: Further information concerning the placement method for graphene, sample preparation, measurement setup, Coulomb diamond, Duffing coefficient fitting, and comparison with another sample is available. This material is available free of charge via the Internet at http://pubs.acs.org. 12 / 15  Reference: 1. Bunch, J. S.; van der Zande, A. M.; Verbridge, S. S.; Frank, I. W.; Tanenbaum, D. M.; Parpia, J. M.; Craighead, H. G.; McEuen, P. L. Science 2007, 315, (5811), 490-493. 2. Chen, C. Y.; Rosenblatt, S.; Bolotin, K. I.; Kalb, W.; Kim, P.; Kymissis, I.; Stormer, H. L.; Heinz, T. F.; Hone, J. Nat. Nanotechnol. 2009, 4, (12), 861-867. 3. van der Zande, A. M.; Barton, R. A.; Alden, J. S.; Ruiz-Vargas, C. S.; Whitney, W. S.; Pham, P. H.; Park, J.; Parpia, J. M.; Craighead, H. G.; McEuen, P. L. Nano Lett. 2010, 10, (12), 4869-73. 4. Barton, R. A.; Ilic, B.; van der Zande, A. M.; Whitney, W. S.; McEuen, P. L.; Parpia, J. M.; Craighead, H. G. Nano Lett. 2011, 11, (3), 1232-6. 5. Weber, P.; Guttinger, J.; Noury, A.; Vergara-Cruz, J.; Bachtold, A. Nat. Commun. 2016, 7, 12496. 6. Chaste, J.; Eichler, A.; Moser, J.; Ceballos, G.; Rurali, R.; Bachtold, A. Nat. Nanotechnol. 2012, 7, (5), 300-303. 7. 8. Poot, M.; van der Zant, H. S. J. Phys. Rep. 2012, 511, (5), 273-335. Singh, V.; Bosman, S. J.; Schneider, B. H.; Blanter, Y. M.; Castellanos-Gomez, A.; Steele, G. A. Nat. Nanotechnol. 2014, 9, (10), 820-4. 9. Teufel, J. D.; Donner, T.; Li, D.; Harlow, J. W.; Allman, M. S.; Cicak, K.; Sirois, A. J.; Whittaker, J. D.; Lehnert, K. W.; Simmonds, R. W. Nature 2011, 475, (7356), 359-63. 10. Rozhkov, A. V.; Sboychakov, A. O.; Rakhmanov, A. L.; Nori, F. Phys. Rep. 2016, 648, 1-104. 11. Wang, L.-J.; Cao, G.; Tu, T.; Li, H.-O.; Zhou, C.; Hao, X.-J.; Su, Z.; Guo, G.-C.; Jiang, H.-W.; Guo, G.-P. Appl. Phys. Lett. 2010, 97, (26), 262113. 12. Liu, X. L.; Hug, D.; Vandersypen, L. M. K. Nano Lett. 2010, 10, (5), 1623-1627. 13. Trauzettel, B.; Bulaev, D. V.; Loss, D.; Burkard, G. Nature Phys. 2007, 3, (3), 192-196. 14. Deng, G.-W.; Wei, D.; Johansson, J. R.; Zhang, M.-L.; Li, S.-X.; Li, H.-O.; Cao, G.; Xiao, M.; Tu, T.; Guo, G.-C.; Jiang, H.-W.; Nori, F.; Guo, G.-P. Phys. Rev. Lett. 2015, 115, (12), 126804. 15. Bolotin, K. I.; Sikes, K. J.; Jiang, Z.; Klima, M.; Fudenberg, G.; Hone, J.; Kim, P.; Stormer, H. L. Solid State Commun. 2008, 146, (9-10), 351-355. 16. Bolotin, K. I.; Ghahari, F.; Shulman, M. D.; Stormer, H. L.; Kim, P. Nature 2009, 462, (7270), 196-199. 17. Du, X.; Skachko, I.; Duerr, F.; Luican, A.; Andrei, E. Y. Nature 2009, 462, (7270), 192-195. 18. Allen, M. T.; Martin, J.; Yacoby, A. Nat. Commun. 2012, 3, 934. 19. Song, X. X.; Li, H. O.; You, J.; Han, T. Y.; Cao, G.; Tu, T.; Xiao, M.; Guo, G. C.; Jiang, H. W.; Guo, G. P. Sci. Rep. 2015, 5, 8142. 20. Tayari, V.; McRae, A. C.; Yigen, S.; Island, J. O.; Porter, J. M.; Champagne, A. R. Nano Lett. 2015, 15, (1), 114-119. 21. Bao, W. Z.; Miao, F.; Chen, Z.; Zhang, H.; Jang, W. Y.; Dames, C.; Lau, C. N. Nat. Nanotechnol. 2009, 4, (9), 562-566. 22. Huang, M.; Yan, H.; Chen, C.; Song, D.; Heinz, T. F.; Hone, J. PNAS 2009, 106, (18), 7304-8. 23. Guinea, F.; Katsnelson, M. I.; Geim, A. K. Nat. Phys. 2010, 6, 30-33. 24. Levy, N.; Burke, S. A.; Meaker, K. L.; Panlasigui, M.; Zettl, A.; Guinea, F.; Neto, A. H. C.; Crommie, M. F. Science 2010, 329, (5991), 544-547. 25. Huang, M.; Pascal, T. A.; Kim, H.; Goddard, W. A.; Greer, J. R. Nano Lett. 2011, 11, (3), 1241-6. 26. Benameur, M. M.; Gargiulo, F.; Manzeli, S.; Autes, G.; Tosun, M.; Yazyev, O. V.; Kis, A. Nat. Commun. 2015, 6, 8582. 13 / 15 27. Novoselov, K.-S.; Jiang, Z.; Zhang, Y.; Morozov, S.-V.; Stormer, H.-L.; Zeitler, U.; Maan, J.-C.; Boebinger, G.-S.; P., K.; Geim, A.-K. Science 2007, 315, 1379. 28. Barreiro, A.; van der Zant, H. S. J.; Vandersypen, L. M. K. Nano Lett. 2012, 12, 6096. 29. Bae, S.; Kim, H.; Lee, Y.; Xu, X.; Park, J.-S.; Zheng, Y.; Balakrishnan, J.; Lei, T.; Kim, R. R.; Song, Y. I.; Kim, Y.-J.; Kim, K. S.; Ozyilmaz, B.; J.-H., A.; Hong, B. H.; Lijima, S. Nat. Nanotechnol. 2010, 5, 574. 30. Suzeki, H.; Kaneko, T.; Shibuta, Y.; Ohno, M.; Maekawa, Y.; Kato, T. Nat. Commun. 2016, 7, 11797. 31. Armour, A. D.; Blencowe, M. P.; Zhang, Y. Phys. Rev. B 2004, 69, (12), 125313. 32. Armour, A. D. Phys. Rev. B 2004, 70, (16), 165315. 33. Rodrigues, D. A.; Armour, A. D. New J. Phys. 2005, 7, 251-251. 34. Bennett, S. D.; Cockins, L.; Miyahara, Y.; Grutter, P.; Clerk, A. A. Phys. Rev. Lett. 2010, 104, (1), 017203. 35. Pistolesi, F.; Fazio, R. Phys. Rev. Lett. 2005, 94, (3), 036806. 36. Johansson, J. R.; Mourokh, L. G.; Smirnov, A. Y.; Nori, F. Phys. Rev. B 2008, 77, (3), 035428. 37. Ouyang, S. H.; You, J. Q.; Nori, F. Phys. Rev. B 2009, 79, (7), 075304. 38. Aspelmeyer, M.; Schwab, K. New J. Phys. 2008, 10, 095001. 39. Xue, F.; Wang, Y. D.; Liu, Y.-x.; Nori, F. Phys. Rev. B 2007, 76, (20), 205302. 40. Knobel, R. G.; Cleland, A. N. Nature 2003, 424, 291-3. 41. LaHaye, M. D.; Buu, O.; Camarota, B.; Schwab, K. C. Science 2004, 304, 74. 42. Naik, A.; Buu, O.; LaHaye, M. D.; Armour, A. D.; Clerk, A. A.; Blencowe, M. P.; Schwab, K. C. Nature 2006, 443, (7108), 193-6. 43. Steele, G. A.; Huttel, A. K.; Witkamp, B.; Poot, M.; Meerwaldt, H. B.; Kouwenhoven, L. P.; Zant, H. S. J. v. d. Science 2009, 325, (5944), 1103. 44. Lassagne, B.; Tarakanov, Y.; Kinaret, J.; Garcia-Sanchez, D.; Bachtold, A. Science 2009, 325, 1107. 45. Deng, G. W.; Zhu, D.; Wang, X. H.; Zou, C. L.; Wang, J. T.; Li, H. O.; Cao, G.; Liu, D.; Li, Y.; Xiao, M.; Guo, G. C.; Jiang, K. L.; Dai, X. C.; Guo, G. P. Nano Lett. 2016, 16, (9), 5456-5462. 46. Li, S. X.; Zhu, D.; Wang, X. H.; Wang, J. T.; Deng, G. W.; Li, H. O.; Cao, G.; Xiao, M.; Guo, G. C.; Jiang, K. L.; Dai, X. C.; Guo, G. P. Nanoscale 2016, 8, (31), 14809-13. 47. Croy, A.; Midtvedt, D.; Isacsson, A.; Kinaret, M. Phys. Rev. B 2012, 86, 235435. 48. Voje, A.; Croy, A.; Isacsson, A. New J. Phys. 2013, 15, 053041. 49. Chen, C.-Y.; Deshpande, V. V.; Koshino, M.; Lee, S.; Gondarenko, A.; MacDonald, A. H.; Kim, P.; Hone, J. Nat. Phys. 2016, 12, 240. 50. Castellanos-Gomez, A.; Buscema, M.; Molenaar, R.; Singh, V.; Janssen, L.; van der Zant, H. S. J.; Steele, G. A. 2D Materials 2014, 1, (1), 011002. 51. Weber, P.; Guttinger, J.; Tsioutsios, I.; Chang, D. E.; Bachtold, A. Nano Lett. 2014, 14, (5), 2854-60. 52. Sillanpaa, M. A.; Sarkar, J.; Sulkko, J.; Muhonen, J.; Hakonen, P. J. Appl. Phys. Lett. 2009, 95, 011909. 53. Laird, E. A.; Pei, F.; Tang, W.; Steele, G. A.; Kouwenhoven, L. P. Nano Lett. 2012, 12, (1), 193-7. 54. Molitor, F.; Jacobsen, A.; Stampfer, C.; Güttinger, J.; Ihn, T.; Ensslin, K. Phys. Rev. B 2009, 79, (7), 075426. 55. Meerwaldt, H. B.; Labadze, G.; Schneider, B. H.; Taspinar, A.; Blanter, Y. M.; van der Zant, H. S. J.; Steele, G. A. Phys. Rev. B 2012, 86, (11), 115454. 56. Eichler, A.; Moser, J.; Chaste, J.; Zdrojek, M.; Wilson-Rae, I.; Bachtold, A. Nat. Nanotechnol. 2011, 6, 339-342. 57. Li, M.; Tang, H. X.; Roukes, M. L. Nat. Nanotechnol. 2007, 2, (2), 114-20. 58. Kenny, T. Ieee Sens. J. 2001, 1, (2), 148-157. 14 / 15  15 / 15 

1406.6823

2

1406

Scanning-gate-induced effects and spatial mapping of a cavity

Tailored electrostatic potentials are the foundation of scanning gate microscopy. We present several aspects of the tip-induced potential on the two-dimensional electron gas. First, we give methods on how to estimate the size of the tip-induced potential. Then, a ballistic cavity is formed and studied as a function of the bias-voltage of the metallic top gates and probed with the tip-induced potential. It is shown how the potential of the cavity changes by tuning the system to a regime where conductance quantization in the constrictions formed by the tip and the top gates occurs. This conductance quantization leads to a unprecedented rich fringe pattern over the entire structure. Finally, the effect of electrostatic screening of the metallic top gates is discussed.

cond-mat.mes-hall

cond-mat

Scanning-gate-induced effects and spatial mapping of a cavity R Steinacher, AA Kozikov, C Rossler, C Reichl, W Wegscheider, T Ihn, and K Ensslin Solid State Physics Laboratory, ETH Zurich, 8093 Zurich, Switzerland E-mail: richard.steinacher@phys.ethz.ch Abstract. Tailored electrostatic potentials are at the heart of semiconductor nanostructures. We present measurements of size and screening effects of the tip-induced potential in scanning gate microscopy on a two-dimensional electron gas. First, we show methods on how to estimate the size of the tip-induced potential. Second, a ballistic cavity is studied as a function of the bias-voltage of the metallic top gates and probed with the tip-induced potential. It is shown how the potential of the cavity changes by tuning the system to a regime where conductance quantization in the constrictions formed by the tip and the top gates occurs. This conductance quantization leads to a unprecedented rich fringe pattern over the entire structure. Third, the effect of electrostatic screening of the metallic top gates is discussed. 6 1 0 2 b e F 0 1 ] l l a h - s e m . t a m - d n o c [ 2 v 3 2 8 6 . 6 0 4 1 : v i X r a Scanning-gate-induced effects and spatial mapping of a cavity 2 1. Introduction Scanning gate microscopy (SGM) is a powerful method to investigate local transport properties of electronic nanostructures. Typically the biased tip of a scanning force microscope is used to locally deplete (in the case of AlGaAs heterostructures, [1 -- 14]) or change (in the case of graphene, [15 -- 18]) the carrier density below the tip. The conductance is monitored as a function of tip position resulting in so-called scanning gate images. Important ingredients for the interpretation of such images are the shape and size of the tip-induced potential in the landscape of the electronic nanostructure. In the literature numbers for the size of the tip-induced potential at the Fermi energy vary between a few tens of nm and more than 1 m depending on experimental setup and analysis procedure [1, 17, 19 -- 29]. Since many nanostructures, such as quantum point contacts (QPCs) and quantum dots, are formed by suitably biased top gates, the effective electronic landscape is a superposition of the gate-defined and tip-induced potential. In this paper we describe four different and complimentary methods which allow us to determine the effective size of the tip-induced potential at the Fermi energy. The sample is a gate-defined ballistic cavity. With the additional tip-induced potential we measure the positions of the conductance plateaus formed by the tip-gate constrictions and how they shift as a function of tip position. Equipped with this knowledge we also analyze the effects of gate screening on the detailed positions of observed features in scanning gate images. Our methods give a better understanding of the details of the potential landscape in complex gate geometries. Beyond that they are useful for scanning gate microscopy and in agreement with calculations. 2. Experimental setup The investigated 2DEG is formed in a molecular-beam-epitaxy-grown GaAs/AlGaAs heterostructure with a density of 1.5 × 1011 cm−2 and a mobility of 3.8 × 106 cm2/Vs at a temperature of 300 mK. It is buried 120 nm below the surface. The electrons have a Fermi wavelength of 65 nm and an elastic mean free path of about 50 µm. The sample under study is fabricated by etching a conventional Hall bar. On top Au/Ti gates [see figure 1a)] are placed using electron beam lithography to define the two cavities in the following measurements. The segmented design is intended to give flexibility in forming cavities with different diameters (d1 = 1.0 µm for cavity I with gates g1 − g5, d2 = 1.5 µm for cavity II g8 − g12). The lithographic width of the constrictions used as openings of cavity I (gates g1 and g4, g3 and g4, as seen in figure 1) is 0.62 µm. The constrictions used for cavity II (g8 and g9, and g11 and g12) are W = 0.4 µm wide. The experimental setup is a home-built AFM operated in a 3He cryostat [31] at a base temperature of 300 mK. A Pt/Ir wire, sharpened with chemical wet-etching and consecutive milling with a focussed ion beam is used as the tip. It is glued to a tuning fork sensor, which is controlled by a phase-locked loop [32, 33]. Scanning-gate-induced effects and spatial mapping of a cavity 3 Figure 1. a) Scanning electron micrograph of the sample used in most of the presented measurements. The bright parts correspond to the Ti/Au top gates gi placed on the GaAs surface (dark). b) Electric scheme of the measurement setup. The Hall bar is connected in a two-terminal configuration. The structure in the 2DEG is formed by applying negative voltages [Vg in figure 1b)] to the top gates thereby decreasing the charge carrier density below the gates. The gate pinch-off is determined to be −0.35 V. Biasing the tip (Vtip ≈ −3 .. − 8 V) 60 nm above the GaAs surface depletes the 2DEG underneath, and hence forms a movable gate. The transport measurements are carried out in a two-terminal configuration with a source- drain voltage (VSD) of 100 µV modulated at 27 Hz [figure 1b)]. The source-drain current (ISD) is measured by standard lock-in techniques. 3. Tip depletion size in the 2DEG Information on the tip-induced potential is needed in order to interpret SGM results. In the following we show four methods which allow us to estimate the radius Rtip of the tip-depleted region in the plane of the electron gas. In figure 2a) the conductance G of cavity II is shown as a function of tip position. The black lines correspond to the edges of the biased top gates (Vg8−12 = −0.4 V) which form the structure. The conductance decreases from approximately seven conductance quanta (7 × 2e2/h) for the tip at a position where it does not influence the cavity transmittance to zero in the vicinity of the two QPCs. The result are lens-shaped regions close to the two QPCs , labeled A and B in figure 2a) and b), similar as observed in [8]. In order to understand how we can read the approximate size of the tip-depleted region from this image, we first concentrate on the solid green sectional line in figure 2a) and b). Figure 2c) shows the conductance and its derivative along this line together with schematic drawings of the tip position relative to the constriction. It is evident from the data and the schematics that Rtip ≈ 0.28 µm. At the same time the electronic width of the constriction is seen to be Wel ≈ 0.26 µm in agreement with the lithographic size and the depletion width caused by the applied voltage. This crude estimate, which we call method I in the following, regards the tip-depleted region to be hard-wall, simplifies the detailed geometry, and neglects all screening and stray-capacitance effects caused by the surface gates. It should therefore be taken as an order of magnitude estimate. The oscillations in the derivative of the conductance, also seen as fringes in figure 2b), reflect quantized conductance plateaux in the constriction formed between the tip and Scanning-gate-induced effects and spatial mapping of a cavity 4 Figure 2. a) Conductance map of scanning the tip (Vtip = −4 V) above stadium II. The black lines indicate the outline of the biased top gates, grey ones the grounded gates, respectively. b) Numerical derivative dG/dx of the conductance map. c) Line cut along the green line in a) and b) of the conductance (blue line) and the derivative (red line) with respect to the cut direction r. The upper arrow indicates the diameter of the tip, the lower one the width of the constriction size Wel. Above the line cut the tip position of the marked positions are sketched. The long-range action of the tip is negelected by extrapolating the approximative linear decrease in conductance towards the unperturbed QPC conductance value. d) Geometry of the length l of the zero conductance region, the width Wel of the electronic constriction, and the gate width a. e) The blue dotted line follows the path of the onset of current-flow between tip and gate. one of the QPC gates [8]. Figure 2d) illustrates another geometric consideration for estimating Rtip from the extent of the lens-shaped region along the green dashed line (method II). One finds (cid:113) [Wel/2]2 + [(l − a)/2]2 ≈ 0.33 µm, Rtip = where the width of the QPC gate is taken to be a ≈ 0.15 µm, the extent of the lens- shaped region l ≈ 0.75 µm, and the electronic width of the constriction Wel ≈ 0.3 µm. This result is in agreement with the previous estimate. In figure 2b) we observe that the last fringe before depletion in the lens-shaped region can be followed into the interior as indicated by the blue dotted lines. These lines run at approximately constant distance from the edge of the gate directly indicating Scanning-gate-induced effects and spatial mapping of a cavity 5 Rtip ≈ 0.5 µm (method III) as illustrated in figure 2e). This estimate is an order of magnitude agreement with the previous ones, given the fact that the density in the cavity may be enhanced compared to the constrictions (although possibly reduced compared to the bulk), and given the distinct electrostatic environment formed by the surface gates. All previous estimates of Rtip neglected the long-range tails of the tip-induced potential. The long-range capacitive coupling of the tip to a QPC [9] can be used to determine this tail quantitatively. To this end the tip, kept at constant voltage, is placed at several positions along the transport axis of the QPC. At each point the QPC depletion gate-voltage is determined. Using finite-bias spectroscopy this gate-voltage can be calibrated to an energy scale [34]. The resulting data for another tip than the previous, is shown in figure 3a), where the horizontal axis represents the distance from the tip to the center of the QPC. We fit these data with a lorentzian line shape, since this was shown to be a reasonable approach [1, 20 -- 28] E(x; E0, A, x0, γ) = E0 + A (x − x0)2 + γ2 , where E0, A, x0, and γ are fitting parameters describing an energy offset, the peak amplitude, a position offset, and the line-width, respectively. The particular data shown in figure 3a) lead to E0 = (0.24 ± 0.03) meV, A = (0.372 ± 0.005) meVnm2, x0 = (−0.085 ± 0.003) nm, and γ = (0.160 ± 0.005) nm. The intersection point of this reconstructed particular tip-induced potential with the Fermi energy of the electron gas gives an estimate of Rtip ≈ (0.17 ± 0.08) µm (method IV). The largest contribution of the uncertainty of this estimate stems from the energy offset E0, because this quantity results from the QPC gate-voltage to energy conversion. Figure 3. a) Reconstruction of the tip induced potential (method IV), measured on the same sample as used in [8, 9], the tip-voltage is -4.5 V. The data points are converted from gate voltage to energy via a finite bias measurement and fitted with a Lorentzian. The dashed line indicates the Fermi energy level. b) Tip radii of four different tips, determined with different approaches shown in the text. In figure 3b) Rtip of four different tips is shown as a function of tip-voltage. The tip-surface separation and the depth of the 2DEG are the same for all measurements Scanning-gate-induced effects and spatial mapping of a cavity 6 shown (60 nm and 120 nm, respectively). The plot confirms that the different methods are consistent for a given tip. At a given tip-voltage different values of Rtip [compare the values of the tips used in figure 3a) and b)] are brought about by unintentional differences in tip fabrication and by modifications of the tip shape during topography scans [29]. The radius of the tip-depleted region is found to increase linearly with the tip voltage. This linear behavior is understandable since the Lorentzian is steep at the Fermi energy and thus can be approximated as a straight line within the given range of tip voltages. For voltages above -- 3 V this assumption is not justified since the 2DEG is no longer depleted. The change of Rtip with Vtip is approximately 80 nm/V for tips of any radius given the tip-surface and surface-2DEG separation of 60 nm and 120 nm, respectively. 4. Forming a cavity with the top gates The tip characterized by the measurements of figure 2 is now used to find the change of the depletion width at the borders of the gate-defined cavity I [see figure 1 a)] as a function of gate voltage. In order to get such spatial information, a set of 2d scans with the biased AFM tip and varying gate voltages is taken. For a first set of five scans the voltage on g4 is varied while the voltage on g1 and g3 is kept constant at −0.55 V. For the second set the roles of g4 and g1, g3 are interchanged. The first set, shown in figure 4b)-f), leads to a fringe pattern in dG/dx filling the whole cavity. In figure 4a) the conductance G(x, y) corresponding to figure 4b) is given. The origin of the fringes is the same as in figure 2b): a quantized constriction forms between the tip-depleted region and one of the cavity gates. There are two groups of fringes, group I/II related to gate g1 and g3, and group III/IV related to g4 [see labeling in figure 4 c)]. The sequence of images in figure 4b)-f) shows that the group III/IV fringes shift in space with changing Vg4, whereas group I/II stays in place. This shift contains the desired quantitative information about the change of the depletion width. The exact positions of the fringes can be extracted from the cuts [figure 5a)] along the green dashed line shown in figure 4b) for the five gate voltages applied to g4. The fringes are labeled starting from the center of the cavity. These positions are indicated by filled circles in figure 5a). In figure 5b) we plot these points and fit them with a linear function of Vg using li = αi∆Vg + l0,i, (1) where ∆Vg = Vpinch−off − Vg is the difference of gate voltage from the gate pinch-off (-0.35 V), and l0,i is an arbitrary length offset irrelevant for the determination of the αi, with i as the fringe number. In figure 5c) we show the αi determined from all scans. In addition, with the cavity divided into four regions I-IV [see figure 4c)], one characteristic cross-section is analyzed in each region for each scan. Points connected by solid lines refer to the situation where the constriction forms between the tip and the gate that is varied (case 1). Points Scanning-gate-induced effects and spatial mapping of a cavity 7 Figure 4. Set of images taken with varying gate voltage applied to the upper top gate g4 (solid purple line). The black solid line draws the biased gate with a fixed voltage throughout the whole set, the thin grey line belongs to grounded gates. a) Conductance map with Vg = −0.4 V. b)-f) Numerical derivatives with respect to x- direction. The black dashed lines gives a guide to the eye to follow the separation of the inner fringes. connected by dashed lines refer to the situation where the gate is varied whose action on the constriction is screened by the tip (case 2). In the latter case the values of αi are smaller, and they increase with fringe number (tip position) due to reduced screening of the gate voltage by the tip. The α1 parameter, which indicates the change of the depletion width with gate voltage, is of the same order of magnitude for all regions. At the same time the αi vary only very little within each region in case 1. 5. The origin of the fringe pattern shape The shape of the fringe pattern in figure figure 4 does not reflect the cavity gate outline. Instead, it can be divided into four regions indicated in figure 4c). An additional effect not considered so far altering the tip-induced potential must be involved, since the tip- depleted region was found to be symmetric around the constrictions in section 3 [see figure 2a) and b)]. The fringes in these regions surround the lens-shaped regions A' and B' in figure 4a). This suggests that the constrictions involving the tip form mainly Scanning-gate-induced effects and spatial mapping of a cavity 8 Figure 5. a) Cut of the dG/dx maps of the set of varying g4. The cut direction is the same as in figure 4a), but only for region IV, starting from the center. b) Position of the conductance plateaus (fringes in dG/dx) along the cut of the different gate voltages. c)All parameters characterizing the shift of the fringes of the different areas of both sets of measurements. Figure 6. Simulations of the electrostatic potential on the 2DEG induced by the biased tip. The gate configuration is the same as in the measurements above. The gate g2 (grey) is grounded, g1, g3, and g4 (black) are biased . The insets show a 2d map at the Fermi energy, thus the depletion of the 2DEG. a) The tip is placed in the center of the entrance of the cavity. b) The tip is moved towards the cavity center until it blocks the constriction. with the openings of the cavity, similar as discussed for figure 2. Additionally, the lens- shaped regions are shifted into the cavity from the geometric center of the constriction. A striking observation is made when the tip moves along the dotted line from point α to β in figure 4b). While we would naively expect the conductance to increase we observe a decrease. Tentatively we ascribe this effect to enhanced screening of the tip-induced potential by the surface gates. By moving the tip closer to the constriction, its distance to the surface gates decreases, the tip-induced potential gets increasingly screened, and the 2DEG is no longer depleted below the tip. Figure 6 shows electrostatic simulations supporting this interpretation. Calcula- tions were carried out with COMSOL [35] treating the 2DEG as a grounded metallic plane 100 nm below the metallic top gates. The GaAs material was modeled as a di- electric with
= 13. The tip, implemented as a metallic cone with a hemisphere with radius 50 nm at its end, is placed 70 nm above the surface. We determine the induced density in the 2DEG and consider regions to be depleted if the induced density exceeds the sheet density of the electron gas. Scanning-gate-induced effects and spatial mapping of a cavity 9 In figure 6a) the tip is placed in the opening of the cavity [position α(cid:48) in figure 6a)]. The screening of the induced potential by the gate g2 leads to an open conductance channel, as indicated in the inset showing the depletion area below the gates and the tip. In figure 6b) the tip is moved along the dotted line towards the cavity center until the constriction is closed (position β(cid:48)). These simulations show that the zero conductance regions A' and B' are shifted relative to the constriction center into the cavity in the presence of grounded gates close to the constriction. 6. Conclusion We have presented methods for estimating the size of the tip-induced depletion region in the 2DEG using a biased AFM tip and the investigation of the shape of a ballistic cavity. Even though most of the methods use simplified geometric assumptions their errors may play a minor role compared to electrostatic screening effects encountered in the experiments. But even with such limitations, fully quantized transport resulting in an unprecedented clear fringe pattern covering the entire stadium is observed. The findings are pointing towards the accessibility of the local density of the electronic states in ballistic cavities for optimized structures regarding tip potential screening by the top gates. 7. Acknowledgements We acknowledge financial support from the Swiss National Science Foundation, the NCCR Quantum science and Technology and ETH Zurich. References [1] Eriksson MA, Beck RG, Topinka M, Katine JA, Westervelt RM, Campman KL, and Gossard, AC 1996 Appl. Phys. Lett. 69 671 -- 673 [2] Topinka MA, LeRoy BJ, Shaw SEJ, Heller EJ, Westervelt RM, Maranowski KD, and Gossard AC 2000 Science 289 2323 -- 2326 [3] Topinka MA, LeRoy BJ, Westervelt RM, Shaw SEJ, Fleischmann R, Heller EJ, Maranowski KD, and Gossard AC 2001 Nature 410 183 -- 186 [4] LeRoy BJ, Topinka MA, Westervelt RM, Maranowski KD, and Gossard AC 2002 Appl. Phys. Lett. 80 4431 -- 4433 [5] Jura MP, Topinka MA, Urban L, Yazdani A, Shtrikman H, Pfeiffer LN, West KW, and Goldhaber- Gordon D 2007 Nat. Phys. 3 841 -- 845 [6] Jura MP, Topinka MA, Grobis M, Pfeiffer LN, West KW, and Goldhaber-Gordon D 2009 Phys. Rev. B 80 041303 [7] Jura MP, Grobis M, Topinka MA Pfeiffer LN, West KW, and Goldhaber-Gordon D 2010 Phys. Rev. B 82 155328 [8] Kozikov AA, Weinmann D, Rossler C, Ihn T, Ensslin K, Reichl C, and Wegscheider W 2013 New J. Phys. 15 083005 [9] Kozikov AA, Rossler C, Ihn T, Ensslin K, Reichl C, and Wegscheider W 2013 New J. Phys. 15 013056 Scanning-gate-induced effects and spatial mapping of a cavity 10 [10] Kozikov AA, Steinacher R, Rossler C, Ihn T, Ensslin K, Reichl C, and Wegscheider W 2014 New J. Phys. 16 053031 [11] Brun B, Martins F, Faniel S, Hackens B, Bachelier G, Cavanna A, Ulysse C, Ouerghi A, Gennser U, Mailly D, Huant S, Bayot V, Sanquer M, and Sellier H 2014 Nat. Commun. 5 4290 [12] Pascher N, Rossler C, Ihn T, Ensslin K, Reichl C, and Wegscheider W 2014 Phys. Rev. X 4 011014 [13] Paradiso N, Heun S, Roddaro S, Biasiol G, Sorba L, Venturelli D, Taddei F, Giovannetti V, and Beltram F 2012 Phys. Rev. B 86 085326 [14] Crook R, Smith CG, Graham AC, Farrer I, Beere HE and Ritchie DA 2003 Phys. Rev. Lett. 91 246803 [15] Berezovsky J, Borunda MF, Heller EJ, and Westervelt RM 2010 Nanotechnology 21 274013 [16] Schnez S, Guettinger J, Huefner M, Stampfer C, Ensslin K, and Ihn T 2010 Phys. Rev. B 82 165445 [17] Pascher N, Bischoff D, Ihn T, and Ensslin K 2012 Appl. Phys. Lett. 101 063101 [18] Garcia AGF, Koenig M, Goldhaber-Gordon D, and Todd K 2013 Phys. Rev. B 87 085446 [19] Crook R, Smith CG, Simmons MY, and Ritchie DA 2000 J. Phys.: Condens. Matter 12 L735 [20] Girard C, Joachim C, Chavy C, and Sautet Ph 1993 Surf. Sci 282 400 -- 410 [21] Atlan D, Gardet G, Binh VT, a N, and Saenz JJ 1992 Ultramicroscopy 42 154 -- 162 [22] Pala MG, Hackens B, Martins F, Sellier H, Bayot V, Huant S, and Ouisse T 2008 Phys. Rev. B 77 125310 [23] Pala MG, Baltazar S, Martins F, Hackens B, Sellier H, Ouisse T, Bayot V, and Huant S 2009 Nanotechnology 20 264021 [24] Chae J, Jung S, Woo S, Baek H, Ha J, Song YJ, Son YW, Zhitenev NB, Stroscio JA, and Kuk Y 2012 Nano Lett. 12 1839 -- 1844 [25] Martins F, Hackens B, Pala MG, Ouisse T, Sellier H, Wallart X, Bollaert S, Cappy A, Chevrier J, Bayot V, Huant S 2007 Phys. Rev. Lett. 99 136807 [26] Kicin S, Pioda A, Ihn T, Sigrist M, Fuhrer A, Ensslin, Reinwald M, and Wegscheider W 2005 New J. Phys. 7 185 [27] Pioda A, Kicin S, Brunner D, Ihn T, Sigrist M, Ensslin K, Reinwald M, and Wegscheider W 2007 Phys. Rev. B 75 045433 [28] Gildemeister AE, Ihn T, Sigrist M, Ensslin K, Driscoll DC, and Gossard AC 2007 Phys. Rev. B 75 195338 [29] Sellier H, Hackens B, Pala MG, Martins F, Baltazar S, Wallart X, Desplanque L, Bayot V, and Huant S 2011 Semicond. Sci. Technol. 26 064008 [30] Kolasi´nski, K and Szafran, B 2013 Phys. Rev. B 88 165306 [31] Ihn T 2004 Electronic quantum transport in mesoscopic semiconductor structures (Springer Tracts in Modern Physics vol 192)(Berlin: Springer) [32] Rychen J, Ihn T, Studerus P, Herrmann A, and Ensslin K 1999 Rev. Sci. Intrum. 70 2765 -- 2768 [33] Rychen J, Ihn T, Studerus P, Herrmann A, Ensslin K, Hug HJ, van Schendel PJA, and Gunterodt HJ 2000 Rev. Sci. Intrum. 71 1695 -- 1697 [34] Rossler C, Baer S, de Wiljes E, Ardelt PL, Ihn T, Ensslin K, Reichl C, and Wegscheider W 2011 New J. Phys. 13 113006 [35] COMSOL Multiphysics 4.4 

1811.02592

2

1811

Classification of crystalline topological insulators through K-theory

Topological phases for free fermions in systems with crystal symmetry are classified by the topology of the valence band viewed as a vector bundle over the Brillouin zone. Additional symmetries, such as crystal symmetries which act non-trivially on the Brillouin zone, or time-reversal symmetry, endow the vector bundle with extra structure. These vector bundles are classified by a suitable version of K-theory. While relatively easy to define, these K-theory groups are notoriously hard to compute in explicit examples. In this paper we describe in detail how one can compute these K-theory groups starting with a decomposition of the Brillouin zone in terms of simple submanifolds on which the symmetries act nicely. The main mathematical tool is the Atiyah-Hirzebruch spectral sequence associated to such a decomposition, which will not only yield the explicit result for several crystal symmetries, but also sheds light on the origin of the topological invariants. This extends results that have appeared in the literature so far. We also describe examples in which this approach fails to directly yield a conclusive answer, and discuss various open problems and directions for future research.

cond-mat.mes-hall

cond-mat

Classification of crystalline topological insulators through K -theory Luuk Stehouwer, ♣ Jan de Boer,♠ Jorrit Kruthoff,♠ Hessel Posthuma,♣ ♠ Institute for Theoretical Physics Amsterdam and Delta Institute for Theoretical Physics, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands ♣ Korteweg-de Vries Institute for Mathematics, University of Amsterdam, Science Park 105-107, 1098 XG Amsterdam, The Netherlands j.deboer@uva.nl, j.kruthoff@uva.nl, h.b.posthuma@uva.nl, luuk.stehouwer@gmail.com Abstract Topological phases for free fermions in systems with crystal symmetry are classified by the topology of the valence band viewed as a vector bundle over the Brillouin zone. Ad- ditional symmetries, such as crystal symmetries which act non-trivially on the Brillouin zone, or time-reversal symmetry, endow the vector bundle with extra structure. These vector bundles are classified by a suitable version of K-theory. While relatively easy to define, these K-theory groups are notoriously hard to compute in explicit examples. In this paper we describe in detail how one can compute these K-theory groups starting with a decomposition of the Brillouin zone in terms of simple submanifolds on which the symmetries act nicely. The main mathematical tool is the Atiyah-Hirzebruch spectral sequence associated to such a decomposition, which will not only yield the explicit result for several crystal symmetries, but also sheds light on the origin of the topological invari- ants. This extends results that have appeared in the literature so far. We also describe examples in which this approach fails to directly yield a conclusive answer, and discuss various open problems and directions for future research. Contents 1 Introduction 1.1 Outline, summary of results and comparison . . . . . . . . . . . . . . . . . . . 2 Time-reversal only 3 The spectral sequence and applications 3.1 A general method: the Atiyah-Hirzebruch spectral sequence . . . . . . . . . . 3.2 Time-reversal only: revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Time-reversal and a twofold rotation symmetry . . . . . . . . . . . . . . . . . 4 Generalizations 4.1 Other crystal symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Class AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Class A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Discussion A Appendix A.1 Freed & Moore K-theory and twists . . . . . . . . . . . . . . . . . . . . . . . A.2 Higher twisted representation rings and the K-theory of a point . . . . . . . . A.3 Construction of the spectral sequence . . . . . . . . . . . . . . . . . . . . . . 1 4 5 9 11 15 20 25 25 31 33 34 37 37 42 43 1 Introduction Topological phases of matter form an interesting playground for both experimental and the- oretical physics. These phases have the remarkable property to be resilient against external perturbations such as weak disorder or weak interactions. This emerges from a gap in the spectrum, either between the ground state and first excited state or around the Fermi level and most of the physics is contained in the states below the gap, i.e. in the degeneracy of the ground state or the occupied states below the Fermi energy. For systems consisting of free fermions moving in a crystal the occupied states form the valence bands and it is the topology of this part of the spectrum that makes a topological phase (of free fermions) topological. Using topological invariants one can capture this topology and, in fact, charac- terise the topological phase. The most well-known example of this is the Chern number that characterises the integer quantum Hall (IQHE) plateaus. 1 We mentioned that a topological phase is resilient against external perturbations, but once these perturbations become too strong so that they cause the (band) gap to close, the topological phase is destroyed, either by becoming an ordinary insulator or by going to another topological phase. Deforming one topological phase into another allows us to understand how many topological phases there are and how to classify them. For free fermion systems with or without time-reversal symmetry and/or particle-hole symmetry, this program was initiated in [1 -- 3]. In particular, Horava and Kitaev noticed an intricate relation between this classification and the classification of vector bundles using K-theory. It was not until the work of Freed and Moore [4] that a complete proposal was formulated to classify topological phases of free fermions by including not only time-reversal or particle-hole symmetry, but also the crystal symmetries that these fermions experience. The proposal of Freed and Moore involves the computation of a suitable twisted and equivariant K-theory. It captures all topological invariants present for a given symmetry class and crystal. These invariants describe both global and local information of valence bands. Although we can treat these invariants in a unifying way, from a physics point of view, global and local invariants describe different aspects of the valence bands, which is why we will now discuss them separately. The local invariants [5, 6] can be defined by carefully analyzing how crystal symmetries act on momentum space and the Hilbert space. They count the number of bands with a particular eigenvalue under the unbroken symmetry of the high-symmetry point in the Brillouin zone. For instance, when there is a fourfold rotation symmetry, the bands at the origin of the Brillouin zone are labelled by the four fourth roots of unity. The number of bands with a particular eigenvalue at the origin is a topological invariant, because changing it would either require closing the gap or breaking the symmetry. In particular this means that at the origin we already have four topological invariants, one for each eigenvalue. Repeating this procedure for other high-symmetry points of the Brillouin zone as well, yields other topological invariants, but these are not all independent. There are gluing conditions between representations associated to points and ones associated to other subspaces such as lines and planes. In other words, when going from a point with less symmetry to a point with more symmetry, the representations of the bigger stabilizer group have to restrict to the representations of the little group. This becomes especially visible when considering two- dimensional crystal groups with reflection symmetries that cause full circles to be fixed (or even 2-tori in three dimensions). On these circles there can then be special points at which the stabilizer group enhances. Implementing these constraints consistently on the full set of topological invariants on each fixed point determines the local invariants. To complete the classification we also have 2 to determine the global invariants. These invariants naturally live on circles or surfaces and at the same time use the global topology of the Brillouin zone in a non-trivial way. They generalize the more well-known invariants to the case where the phase is protected by additional crystal symmetries. The global invariants can be visualized in an intuitive way as follows. Let us start with a basic example, the Chern number, which is known to model the plateaus of the IQHE. While not really necessary for this case, we will for simplicity assume that the Brillouin zone is a two-dimensional sphere rather than a two-torus. With this assumption we are ignoring most of the topology of the Brillouin zone, with would correspond to a system which is neither protected by time-reversal symmetry nor by any other symmetry. The sphere can then for example be thought of as a compactified version of the entire two-dimensional space of momenta. We will frequently encounter spheres in what follows, which is why we choose this approximation here as well, but we emphasize once more that for this particular case the result does not depend on whether we take the Brillouin zone to be a two-sphere or a two-torus. For the two-sphere (or two-torus), it is well known that there exists an infinite family of non-trivial band structures labelled by the integral of the Berry curvature [7]. This integral is an integer and is known as the Chern number. Explicitly, it can be written as C = 1 2πZS 2 TrF ∈ Z (1.1) with F the Berry curvature two-form constructed out of a U (N ) Berry connection Aαβ i = hα∂iβi, where α = 1, . . . N , αi are the energy eigenstates of a Hamiltonian H and ∂i derivatives along the directions of the sphere. The curvature is F = dA + A ∧ A, which is again U (N )-valued. Let us focus on N = 1. In that case we have a U (1)-connection on the sphere. To see what type of connections correspond to non-trivial Chern numbers and hence to non-trivial topological invariants, we concentrate all the curvature on the north pole. This is possible as long as we do not change the integral. In fact, in the limit we simply have a delta function at the north pole with a coefficient such that the integral in 1.1 is an integer. On the level of the connection we can view this configuration as a vortex around the north pole. In local coordinates ϕ around that point, the connection will simply be A = dϕ. (1.2) We can thus conclude that a non-trivial Chern number corresponds to a vortex in the Berry connection. The location of these vortices can be moved (the north pole is not special) but 3 their vorticity is a topological invariant. The situation becomes more interesting when we consider topological insulators with time-reversal symmetry, which come with their associated Z2 invariants. Using the vortex picture sketched above, we can understand this invariant as follows. When there is a time- reversal symmetry present which squares to minus one, the band structure will always consist of an even number of bands. Thus the minimal Berry connection is at least U (2)-valued. However, the curvature F will always be zero, since time-reversal symmetry acts as an orientation reversing operation on the base space, i.e the sphere. Nevertheless, this does not mean that there is no other topological invariant. As shown by [7, 8], there still exists a Z2 invariant. If we focus on the U (2) case, the upshot of [8] is that there is still a way to define a Chern number associated to only one of the two energy eigenstates. Its parity then gives the Z2 invariant. The other energy eigenstate will then carry the opposite Chern number. In local coordinates around e.g. the north pole, the non-trivial connection will then look like A = dϕ 0 −dϕ! 0 (1.3) or any other odd vorticity in each block. The trivial connection instead has an even vorticity in both blocks. The non-trivial connection is thus one of a vortex-antivortex pair in the Bril- louin zone. Notice that their position can be changed, but due to time-reversal symmetry they are always at antipodal points. Of course this is a simplified picture, but it serves as an intuitive and physical interpretation of the invariant. In particular, when other crystal sym- metries are added, the vortices need to respect that symmetry too. This greatly constraints their position and in combination with representation theory, their vorticity [9]. 1.1 Outline, summary of results and comparison The objective of this paper is to formalise some of these ideas. In particular, in Section 2 we will introduce some basic terminology in algebraic topology by discussing an example of a crystal with only time-reversal symmetry. The next section, Section 3 contains the meat of the paper. We discuss in more detail what K-theory we want to compute to classify topological insulators with time-reversal and crystal symmetry. To compute these K-theories we describe the construction of an Atiyah-Hirzebruch spectral sequence and compute two examples in detail. Section 4 is devoted to various other examples we computed, for example, we compute, for the first time, the full classification of a two dimensional crystal with time- reversal in class AII and a four fold rotation symmetry. Furthermore, we also determine the twisted representation rings, which are needed in the spectral sequence, in an algorithmic 4 way. In Section 5 we mention various subtleties and future directions. Finally, in appendix A we have gathered various mathematical details on the spectral sequence construction, twists and twisted group algebras. Computing twisted equivariant K-theory groups using an Atiyah-Hirzebruch spectral sequence is not new. In previous works, [10, 11], an Atiyah-Hirzebruch spectral sequence was also proposed and used to compute the classification for certain symmetry groups and classes. In this work, we fill in certain gaps left open in these works and put the computation of the K-theory groups with an Atiyah-Hirzebruch spectral sequence on a firm mathematical footing. We have gathered most of these details in the appendix. The K-theory groups we have computed match with known results in the literature, but also agree with a set of heuristic arguments given in [5,9] in the cases where we have explicit results. In particular, for the Altland and Zirnbauer classes AI and AII with an order two symmetry in two dimensions, our results match with those in [12 -- 14]. These works extended the analysis by Kitaev in [2] to include additional order two symmetries such as a reflection or two fold rotation symmetry. The basis of this analysis is Clifford algebras, which allow for a straightforward implementation of order two symmetries, but for more complicated symmetries, such a procedure is more difficult. In those cases one has to resort to more sophisticated computational methodes of which we outline one in this paper. 2 Time-reversal only In this section we shall focus on topological insulators with only unbroken time-reversal sym- metry on a two-dimensional lattice without any additional rotation or reflection symmetries. Such topological phases belong to either symmetry class AI or AII [7]. In the former case the time-reversal operator T squares to the identity, whereas in the latter case it squares to minus the identity. To classify such topological insulators, we need to know how many topologically distinct insulators there are with this symmetry. As was explained in the introduction, with distinct we mean that upon going from one to the other phase, either the gap closes or the symmetry is broken. For a more formal definition, see [4]. The classification is most easily understood by translating the problem to momentum space, where discrete translations cause the momenta to only take values in a two-dimensional Brillouin torus T2. We visualize this torus as the square [−π, π] × [−π, π] with opposite sides identified. Due to time-reversal symmetry, a non-trivial group acts on this two-torus which sends k to −k, which is intuitively clear as reversing time should reverse the sign of momenta. Time-reversal symmetry can also easily be shown to be an anti-unitary symmetry. Another example of a possible anti-unitary symmetry (which we will not consider in this paper) is 5 particle-hole symmetry, which acts trivially on the torus. Since we will ignore interactions the momenta k are conserved quantities that can be used to label the states in our Hilbert space. The states with label k are exactly the momentum k Bloch waves. The collection of all these Hilbert spaces form a vector bundle. This vector bundle is the collection of all valence and conduction bands and since we are dealing with insulators here, there is a gap between them. In a (topological) insulator, only the valence bands are physically relevant. For the classification, we hence focus on this finite-dimensional sub-bundle. The classification of topological insulators has now been translated into a mathemat- ical question about the classification of vector bundles over the torus. In the absence of time-reversal symmetry, such a classification can be performed using standard (complex) K-theory. With time-reversal symmetry, things get a little more exotic, since time-reversal is an antiunitary operator which in particular anticommutes with the imaginary unit i. Nev- ertheless, Atiyah [15] generalized K-theory to incorporate this symmetry and dubbed it Real K-theory. Specifically, for class AI and AII, we need to compute KR−q(T2). Here the T2 is the two-dimensional Brillouin zone and the index q labels the various Altland-Zirnbauer classes [3]. In this situation we need q = 0, 4 as they indicate class AI and AII respectively. It is actually not too hard to compute these K-theory groups [2, 4]. The result is KR0(T2) = Z, KR−4(T2) = Z ⊕ Z2. (2.1) The conventional computation of these groups uses various basic properties of KR-theory, which cannot be generalized to include point group symmetries. Moreover, this computation is rather unsatisfactory as it gives no insight into what these invariants mean and where they come from. Part of the motivation of this work and of [5, 9] is to understand what the physical origin is of these invariants and what computational tool makes this physical origin manifest. In particular, we would like to see how the gluing of representations reveals itself in the computation. Looking ahead, we can interpret the result (2.1) as follows. The invariants Z are local in nature and just give the rank of the bundle, i.e. they represent the number of valence bands present. The more interesting invariant Z2 is a global two-dimensional strong topological invariant called the Fu-Kane-Mele invariant and is related to topologically protected edge states [16, 17]. In order to better understand the physical origin, we decompose the Brillouin zone into various parts that are easy for K-theory to handle. Within K-theory we have the freedom to consider a so-called stable equivalent space instead of the torus. Fortunately, there exists a nice space that is stably equivalent to the torus. This space is a certain wedge sum of one 6 and two-dimensional spheres1. Moreover, K-theory is additive under taking such wedge sums and hence we only have to compute the K-theories of spheres, see the end of Section 3.1 for a more precise statement. Physically, this means that we are looking at properties of the band structure insensitive to (part of) the discrete translation symmetry. The two-dimensional sphere just represents the (compactified) momentum space of a topological insulator without translation symmetry and the K-theory then gives the topological invariants associated with this Brillouin zone. For instance, the Chern number in the IQHE is just the complex K- theory of the 2-sphere and is known not to rely on translational symmetry. After computing the K-theories of all such pieces, one simply assembles all these pieces together by taking direct sums. In two dimensions, going from the torus to the sphere can be accomplished by identifying the boundary of the square [−π, π] × [−π, π] with a single point. Let us focus on this sphere for the moment. After this operation, the time-reversal action of Z2 on the Brillouin zone torus reduces to an action on the sphere that is still given by the formula k 7→ −k if we view the sphere as R2 ∪ ∞. Now suppose we have a Hilbert bundle H over the sphere with time-reversing operator T , i.e. a bundle map T : Hk → H−k, where Hk denote the fibers of the bundle H. There are two special points at k = 0 and k = ∞ under the action of time-reversal at which the Bloch states with momentum k are mapped to themselves. This gives vector space automorphisms on the corresponding fibers Hk. In class AI (so T 2 = 1), the operator T acts as an effective complex conjugation on the Bloch states of momenta k = 0 and k = ∞. In more mathematical jargon, there are canonical real structures on the vector spaces H0 and H∞. In class AII, when T squares to −1, we instead have canonical quaternionic structures at k = 0 and k = ∞. In particular, we deduce that the space of Bloch waves at these special points is even dimensional, which is a manifestation of Kramer's theorem. However, at a generic point on the sphere, the momenta are not preserved by T , so that the state spaces at these points do not admit any extra structure. We have now discussed how time-reversal acts on the Brillouin zone once the torus is reduced to a sphere. To include more complicated symmetries later on, it is convenient to view the sphere as being build up out of points, intervals and disks. We have chosen these particular building blocks because they are topologically trivial, i.e. contractible. Such a collection of building blocks is called a CW-complex. The building blocks themselves are called d-cells where d is the dimension of the block. When additional symmetries are present, such a CW-complex has to respect the symmetry. By this is meant that for each cell the symmetry must either fix it completely or map it to a different cell in the decomposition. 1The wedge sum of two spaces is the union of the two spaces but where one point of the first space is identified with one point of the second. 7 In the case of time-reversal symmetry for instance, such a CW-complex is given in Figure 1. In this figure, we also gave the cells an orientation that is preserved by the symmetry, which is visualized by the direction of the arrows on the 1-cells. Note that the north and south pole are fixed by the Z2 action and hence constitute the 0-cells. The 1-cells are a line from p0 to p∞ and its symmetry-related partner. The same is true for the 2-cells, which are the two hemispheres. This yields a practical setting to do the classification using K-theory, because we can simply classify the bundles over these d-cells and then glue them together consistently. Let us see how this works in more detail. To start, we consider the complex and Real K-theory of spheres. It is a known fact that the K-theory of a point in degree −p is equal to the (reduced) K-theory of a p-dimensional sphere. The results are given in Table 1. Bundles on the 0-cells, i.e the north and south pole in Figure 1, are classified by KR0(pt). In class AI, we have KR0(pt) = Z for each of the two fixed points. The Z now assigned to the north and south pole are simply given by the dimension of the fiber at those points. In class AII we get for each fixed point KR−4(pt) = Z, which is given by the quaternionic dimension of the fiber. On the two intervals there is no real or quaternionic structure. Hence we should assign the (reduced) complex K-theory of the interval, where the boundary points of the interval are identified with each other. This K-theory is equal to the (reduced) complex K-theory of the circle, which is zero. The precise reason for this assignment is addressed in detail in the appendix. Finally to the two hemispheres, we assign the (reduced) complex K-theory of a sphere, which is Z. As before, the sphere appears here because we are identifying the boundary of the disc to a point. If our Hilbert space of states is to be preserved by the time-reversal symmetry, the bundle over the two 2-cells should come in pairs that are mapped into each other by the action of time-reversal symmetry. It is thus enough to know the bundle on one such 2-cell and hence under T , the two copies of Z are identified. We thus have Z2 in zero dimensions (0-cells), a 0 in one dimension (1-cells) and Z in two dimensions (2-cells). To get to a complete classification of topological insulators, we have to make sure that our assignment of bundles to cells is consistent. This can be done by imposing constraints in successive dimensions. For dimension zero, this means that when the fibers above the 0-cells are all extended to the 1-cells, the result should be consistent. In our case this means that the state spaces at the points k = 0 and k = ∞ should have the same dimension, thereby reducing the Z2 we found before to Z. This approach is intuitively clear and can easily be generalized to include point group symmetries. However, as advocated in the beginning, the approach of assigning represen- tations to points is only part of the full classification. To get the other part, the global part, we should check consistency of assignments of bundles (not just representations) to 8 higher-dimensional cells. This becomes a lot more difficult and it is hard to understand for generic crystal symmetries. In the case without time-reversal symmetry, these invariants are most of the time first Chern numbers, but there are exceptions [10]. The invariants that can take any integer value can be understood by using the equivariant Chern character [18] or Segal's formula [19], which also has an extension to the twisted case [20]. However, for crystals invariant under time-reversal symmetry, the invariants are often torsion invariants and take only particular integer values. There is no systematic way of understanding them in the sense that there is no explicit formula for this piece. Instead, when assigning bundles to higher-dimensional cells, we have to check which bundles can be realized as a certain coho- mological boundary and quotient out by these. The result will indeed give the Z2-invariant of equation (2.1), but it requires some abstract mathematical theory to see this. Physically, however, there is a heuristic way of understanding these invariants as vortex-anti-vortex pairs in the connection on the bundle, which was presented in the beginning of this section. Below we will formalize the heuristic arguments given above and put them on a firm mathematical footing. The example we have seen in this section will be computed again using machinery that allows for a generalization to more complicated crystal symmetries. To illustrate this, we compute the full classification for topological insulators in class AII on a two dimensional lattice with a twofold rotation symmetry. 3 The spectral sequence and applications Now we come to the core of the paper. In the above we gave a heuristic classification of topological insulators with time-reversal symmetry. We will now make this classification precise and generalize to cases with non-trivial crystal symmetry. The strategy of this section will be to introduce all necessary tools. We will then reconsider the example without any crystal symmetry but with time-reversal symmetry. Whenever appropriate, we will mention the physical motivation and interpretation for these tools along the way. Let us consider topological insulators in d dimensions in class A, AI or AII, possibly with a point group symmetry. We denote the full classical symmetry group of the Brillouin zone by G. If present, G therefore contains time-reversal symmetries and point group symmetries but no translational symmetries. These are taken into account by the topology of our Brillouin zone torus. Let us denote by G the space group and G its (magnetic) point group that does not contain time-reversal symmetry, then the space group G is a group extension 1 → Zd → G → G → 1, (3.1) 9 where Zd represents the discrete lattice translations in d spatial dimensions. When this extension is split, the space group is called symmorphic and non-symmorphic otherwise. We will focus on the former from now on and comment on the non-symmorphic case in the discussion. We will assume that there are no other symmetries, such as gauge symmetries with which the time-reversal operator could mix. In order to classify topological phases in the sense of Freed and Moore [4], we have to compute a joint generalization of Real and equivariant K-theory. In particular, we want to take two additional things into account. First of all, we want to keep track of which elements in G act antiunitarily or not. For this we will use a map φ which sends an element of G to +1 if it is unitary and to −1 if it is antiunitary. Moreover, we want to know how elements of G acting on the Brillouin zone lift to elements acting on the fiber. This is most easily accounted for by a twist τ , a suitable group two-cocycle. This twist encodes the action of the symmetries on the quantum Hilbert space. For example, it prescribes whether T 2 = ±1. But it also provides the signs coming from taking the spin of particle into account. For example, an n-fold rotation operator R for spin− 1 2 particles satisfies Rn = −1. This minus sign is also encoded in τ . Let us for a moment describe the situation in more precise abstract mathematical terms. Assume we have the following data: (i) a finite group G acting on a space X (in our case X = Td, the Brillouin zone); (ii) a homomorphism φ : G → Z2; (iii) a group 2-cocycle τ ∈ Z 2(G, U (1)φ) with values in the circle group U (1) with G-action g · eiθ = eiφ(g)θ. Such a cocycle τ is a special case of the more general twists defined by Freed and Moore [4], called φ-twisted central extensions. Using such data, Freed and Moore [4] defined a version of twisted equivariant K-theory denoted by φK τ G(X), (3.2) which was further studied in [21]. It was also argued that this K-theory group classifies free fermion topological insulators protected by the quantum symmetry defined by G, φ and τ . To connect with more common language used in the physics literature, we describe the G, φ and τ that occur in the classification of crystalline topological insulators. Firstly, a class A topological insulator with point group G0 simply has G = G0 and φ and τ are both trivial. For class AI, G will instead be the magnetic point group, i.e. it will contain both point 10 symmetries and time-reversal symmetry. We will only consider magnetic point groups of the form G = G0 × ZT 2 the action of time-reversal symmetry on the Brillouin zone, even though the mathematical machinery developed here can handle more general point groups 2 , with ZT as well. For instance, one could also consider cases in which the time-reversal operator is a combination of the usual time-reversal operator with a lattice translation or point group symmetry in G0. For G = G0 × ZT 2 in class AI, φ : G → Z2 will simply be projection onto the second factor and τ will be trivial. Finally, for class AII, we again take G = G0 × ZT 2 to be the magnetic point group and φ the same projection, but now we pick τ in such a way that the twisted group action represents the desired action on the quantum Hilbert space. In particular, we pick τ so that time-reversal squares to −1, reflections square to −1 and rotation by 2π equals −1. To assure a consistent choice, a precise construction of τ for a given point group G0 is given at the end of appendix A.1. It is shown in [21, Thm 3.11] that the groups φK τ G(X) satisfy certain equivariant versions of the homotopy, excision, additivity and exactness axioms of Eilenberg and Steenrod. The fact that our twisting class is defined by a group cocycle implies that these axioms are exactly the axioms for an equivariant cohomology theory on the category of G-spaces as defined in Bredon [22, §I.2]. This is what makes the following computations mathematically sound; as explained in [22, §IV.4] the axioms guarantee the existence of the Atiyah -- Hirzebruch spectral sequence. Moreover, the orbifold point of view advocated in [21] allows us to change the group G and the space X as long as the quotient space remains the same and we keep the same stabilizer. This is useful in some computations, see the end of Section 3.3. For more details on how the K-theory we use is defined, see appendix A.1. We are therefore left with the task to compute the K-theory φK τ G(Td) of the Brillouin zone dressed with φ and τ . The technique to compute these groups goes along the lines that we have discussed in the previous section. We first decompose the Brillouin zone into cells and view them as a CW-complex. Non-trivial symmetries have to leave these complexes invariant. Such complexes are equivariant G-CW complexes, which is nothing more than an upgraded version of the unit cell in momentum space. After having found this G-CW complex, we use an Atiyah-Hirzebruch spectral sequence to compute φK τ G(Si), which are assembled to give φK τ G(Td). Let us now formalize this computational method. 3.1 A general method: the Atiyah-Hirzebruch spectral sequence The spectral sequence for the computation of the twisted equivariant K-theory of a space X is constructed by using a decomposition as a G-CW complex X 0 ⊆ · · · ⊆ X d = X, where the superscript on the cells indicates the dimension of the subspace. For the applications 11 considered in this paper, X is either a torus Td or a sphere Sd (we will remark on how to reduce the computation of the K-theory of the torus to the K-theory of a sphere at the end of this section). A spectral sequence is a successive approximation method converging to the desired answer in a number of steps. For us these steps will always be finite and in fact, most of the time only two steps are necessary. These steps are usually referred to as pages. The first page of the spectral sequence, just as in the last example, is given by equivariant assignments f of the K-theory of spheres to the cells of X. For the 0-cells X 0, this means that we assign to each k ∈ X 0 a twisted representation of the stabilizer group Hk of that point. These representations, which are twisted using τ and the map φ, are conveniently packaged in the twisted representation ring φRτ (Hk). These objects are actually not rings, but since they are equal to the usual representation ring of Hk in case φ and τ are trivial, we will keep on referring to them as twisted representation rings. Details about twists and twisted representations can be found in appendix A.1. So f maps k to an element of the twisted representation ring φRτ (Hk) of the corresponding stabilizer group Hk. By equivariance is meant that f preserves the symmetry in the following sense: f is required to map gk to the resulting conjugate representation in φRτ (gHkg−1) = φRτ (Hgk). More generally, we equivariantly assign higher representation rings φRτ −q(Hσ) (i.e. the higher degree twisted equivariant K-theory of a point, see appendix A.2 for details) to p-cells σ. These classify twisted Hσ-equivariant bundles over q-spheres, instead of over just a point. The grid of such assignments of representation rings for each p and q form the first page of the spectral sequence and is denoted by Ep,−q . Those assignments can be shown to be equivalent to Bredon p-cochains with values in the coefficient functor φRτ −q G . In appendix A.3 we define these coefficient functors and present a derivation of this result. Intuitively, the functor φRτ −q they restrict to each other. For Bredon p-cochains, this functor will pick out the stabilizer keeps track of both the (higher) representations at fixed loci and how G 1 group of the p-cells and assign degree −q twisted representation rings to the p-cells. It should be noted here that the action of group elements on the higher representation rings can be tedious to determine explicitly in certain examples, so that the equivariance of f can result in nontrivial results. One example of this is given in Section 3.2. To go to the next page of the spectral sequence, we have to take the cohomology of the first page with respect with the first differential, which in our case is known as the Bredon differential. In fact, the first differential maps Ep,−q of Bredon cohomology, which is and is given by the differential to Ep+1,−q 1 1 [µ : σ]f (µ)Gσ , (3.3) (df )(σ) = Xµ∈C p(X) 12 with C p(X) denoting the set of p-cells of X and f a Bredon p-cochain. Here f (µ)Gσ means that we take the higher twisted representation of Gµ that f assigns to µ and restrict it to a representation of Gσ. The notation [µ : σ] stands for an integer factor that tells us in which way µ intersects the boundary of the p + 1-cell σ. In general the behavior and computation of this number can be quite complicated, but if our CW-complex is sufficiently nice, this number is usually just a sign depending on a fixed orientation. For example, if we have a line (1-cell) ℓ oriented from the endpoint p0 to the other endpoint p1, i.e. ∂ℓ = p1 − p0, then we simply have [ℓ : p1] = 1, [ℓ : p0] = −1 (3.4) and of course [ℓ : pt] = 0 if pt is not an endpoint of ℓ. If instead σ = A is a 2-cell that lies in a disk surrounded by a couple of intervals ℓ1, . . . , ℓk, then [A : ℓi] = ±1 depending on whether the orientations of the line ℓi coincide with the orientation of A. In more general situations, where there is nontrivial gluing present, it can be computed as the degree of a certain map between spheres. This map is exactly the same as for the cellular boundary map in ordinary cellular homology, which can be found for example in Hatcher's book [23]. 1 given in (3.3). Mathematically speaking the second page entry (p, −q) The second page is the cohomology of the first page with respect to the differential 1 → Ep+1,−q d : Ep,−q therefore equals the degree p Bredon equivariant cohomology of X with coefficient functor φRτ −q G . For the third and higher order pages, we need to know the higher differentials, which are much more abstractly defined and no explicit form is known. Therefore, until more is known about this it is not possible to fully classify topological phases for general point groups using this method. It is however often the case in practice that we can arrive at a definite answer without knowing explicit expressions for the higher differentials. At least it is known that the rth differential is of bidegree (r, 1 − r), so dr : Ep,−q . Therefore, for d-dimensional spaces, the rth differential dr is zero for all r > d. For more details on the construction of the spectral sequence and explicit definitions, see appendix A.3. r → Ep+r,−q+1−r r But how do we construct the twisted equivariant K-theory of X from the data of the spectral sequence? After taking the cohomology with respect to the dth differential, we arrive at the final page, Ep,−q ∞ , so by equating p and q. In two dimensions, this means that there exist exact sequences ∞ . We can construct the K-theory by extensions out of Ep,−p 0 → E2,−2 0 → F → φK τ ∞ → F → E1,−1 G(X) → E0,0 ∞ → 0, ∞ → 0, (3.5) (3.6) see the final paragraph of appendix A.3 for the details. Unfortunately, these sequences do 13 not split in general. Therefore the K-theory is not always fully determined by the spectral sequence (unless of course we would explicitly determine the maps in these sequences, which is a tedious exercise). We will call this the problem of non-unique extensions, which unfor- tunately is intrinsic to our approach. An example of this phenomenon will be addressed in Section 3.3. Now that the spectral sequence is contained in our toolbox, we will explain how to reduce the computation of equivariant K-theory of the Brillouin torus Td to the computation of the K-theory of spheres. For this we use an equivariant stable homotopy equivalence that generalizes [4, Thm 11.8]. This equivalence adresses the decomposition of the Brillouin torus in terms of spheres. Indeed, if the action of G on Td = S1 × · · · × S1 can be realized as the restriction of an action of H d ⋊ Sd, where H acts on S1 and Sd permutes the copies of S1, then Td is equivariantly stably homotopy equivalent to a wedge of spheres. More explicitly, this means in two dimensions that there is an isomorphism φK τ G(T2) ∼= φK τ G(S2) ⊕ G(S1 ∨ S1). (3.7) φeK τ Here the tilde indicates the reduced K-theory and S1 ∨ S1 is a space that looks like the figure 8, which is nothing but the boundary of the Brillouin zone torus T2 seen as a square [−π, π] × [−π, π] with opposite sides identified. Note that the symmetry G could potentially interchange the two S1's of the figure eight, for example in case there is a fourfold rotation symmetry. If there is no group element permuting the two copies of the circle, the K-theory decomposes further as where we used that φK τ G(T2) ∼= φK τ G(S2) ⊕ G(S1) ⊕ G(S1), φeK τ φeK τ φeK τ φeK τ G(S1 ∨ S1) = G(S1) ⊕ G(S1). φeK τ (3.8) (3.9) A similar isomorphism as in (3.8) exists in three dimensions under the given assumptions. The relation between reduced and unreduced K-theory φK τ G(X) is φK τ G(X) = φK τ G(pt) ⊕ G(X). (3.10) When using the equivariant splittings (3.7) and (3.8), we can thus compute the unreduced K-theory and then strip of the φK τ G(pt)-part to obtain the reduced K-theory. Note that the assumption that the action of G comes from some action of H d ⋊ Sd does not always hold, so that we cannot always use (3.7). If for example three-fold rotations are present, we seem φeK τ 14 T A p∞ T ℓ ℓ p0 A Figure 1: A Z2-CW structure of the Z2-space S2 that is the one-point compactification of the two-dimensional representation of Z2 given by k 7→ −k. We have denoted this action by T . to be bound to applying the Atiyah-Hirzebruch spectral sequence to the Brillouin zone torus directly. Since the K-theory of a one-dimensional space X is easy to compute, the isomorphism (3.8) effectively reduces computations of the K-theory of a two-dimensional torus to a two- dimensional sphere. Indeed, for one-dimensional spaces all higher differentials vanish and E2,−2 Because the twisted representation ring φRτ (G) is torsion free (for q = 0), so is H 0(X,φ Rτ Hence the resulting sequence splits, giving us ∞ = 0, so that the exact sequences (3.5) and (3.6) reduce to a single exact sequence. G). φK τ G(X) ∼= H 0(X,φ Rτ G) ⊕ H 1(X,φ Rτ −1 G ). (3.11) Despite the absence of torsion in the first term, the second term can give rise to torsion of which we will see examples below. The torsion in H 1 was anticipated before in [24] for systems in class AII with a reflection symmetry in one dimension. 3.2 Time-reversal only: revisited Let us now illustrate how the Atiyah-Hirzebruch spectral sequence formalizes the intuitive approach of the last section. So again we will take X = S2 with the Z2-action k 7→ −k with the Z2-CW-structure as given in Figure 1. We will consider the classes AI (T 2 = 1) and AII (T 2 = −1) simultaneously and note the distinctions along the way. Mathematically, we distinguish between the two classes by picking the twist τ = τ0 to be trivial in class AI and τ = τ1 nontrivial in class AII. The higher twisted representation rings of Z2 and the trivial subgroup 1 ⊆ Z2 are given in Table 1. Note that the stabilizers of the 0-cells are both Z2, while for the other cells the stabilizer is trivial. 15 φRτ0−q(Z2) φRτ1−q(Z2) φRτ0−q(1) φRτ1−q(1) = = = = KR−q(pt) KR−q−4(pt) K −q(pt) K −q(pt) Z Z2 Z2 0 Z 0 0 0 Z q = 0 q = 1 q = 2 q = 3 q = 4 q = 5 q = 6 q = 7 q = 8 ... Z 0 0 0 Z Z2 Z2 0 Z Z 0 Z 0 Z 0 Z 0 Z Z 0 Z 0 Z 0 Z 0 Z Table 1: The twisted representation rings in the case of Gσ = Z2 and Gσ = 1. They are 8-periodic in the degree. We will start by computing all Bredon cohomology groups that are necessary for obtaining the K-theory group from the spectral sequence. These cohomology groups are the ones that ∞ , E1,−1 ∞ and E2,−2 correspond to second page entries of the spectral sequence which could possibly influence the three desired entries E0,0 ∞ of the final page occurring in the exact sequences of equation (3.5) and (3.6). Because the second differential (which is of bidegree (2, 1)) is the only possibly nonzero higher differential, we have E3 = E∞. Therefore we have to compute H p G(S2, φRτ −q First we have to find all necessary Bredon equivariant cochains, as they constitute the G ) for (p, q) equal to (0, 0), (0, 1), (1, 1), (2, 1) and (2, 2). first page Ep,−q . We start with p = q = 0. So we consider the equivariant 0-cochains with 1 values in φRτ G, which here are the equivariant maps from the set {p0, p∞} to φRτ (Z2) = Z for both twists. Because the 0-cells are completely fixed by the group, all 0-cochains are equivariant. Therefore the equivariant 0-cochains are spanned by two basis elements π0 and π∞ over Z: C 0 Z2(S2, φRτ G) = hπ0, π∞iZ = Z2. (3.12) Here π0 maps p0 to 1 ∈ φRτ (Z2) and p∞ to 0 ∈ φRτ (Z2). For π∞ the roles of p0 and p∞ are interchanged. In more basic terms: π0 assigns a state space of dimension one to p0 and a zero space to p∞, while π∞ assigns a zero space to p0 and a one-dimensional space to p∞. Going up to p = 1, q = 0, there is only one equivariant 1-cochain, so that C 1 Z2(S2, φRτ G) = hλiZ = Z. (3.13) 16 Indeed, from Table 1 it is clear the this cochain is an equivariant map from {ℓ, T ℓ} to Z. By equivariance, it is uniquely specified by specifying its value on ℓ, which we take to be 1 for λ. In case the reader is interested in the actual value of λ(T ℓ), simply note that the action of T on the representation ring is just complex conjugation. In case T acts on the representation ring of a nontrivial group this will result in the complex conjugation of nontrivial representations, which are in general not isomorphic to the original representation. However, a complex vector space is noncanonically isomorphic to its complex conjugation since it has the same dimension. Hence the automorphism that T induces on φRτ (1) is simply the identity, thus λ(T ℓ) = λ(ℓ) = 1. Along the way we will see that this automorphism is not always trivial and acts with minus the identity on the higher representation ring of degree −2. This is the heart of the matter, since it is the aspect that creates torsion in this example. For q = 1, the situation simplifies, since the degree −1 representation ring of the trivial group equals zero. Hence there are no equivariant 1-cochains or 2-cochains with values in φRτ −1 G . The degree −1 representation ring of Z2 depends on whether the twist τ is taken trivial (class AI) or nontrivial (class AII). For τ trivial it equals Z2 and for τ nontrivial it equals 0. In class AII, we therefore also have no equivariant 0-cochains with values in φRτ −1 G . In class AI instead, the equivariant 0-cochains are spanned by π0 and π∞, just as for q = 0. However, this time they are a basis over Z2: C 0 Z2(S2, φRτ −1 G ) = hπ0, π∞iZ2 = Z2 2. (3.14) Here π0 maps p0 to the nontrivial element of Z2 and p∞ to the trivial element, while for π∞ it is the other way around. Finally, for q = 2 there are some subtleties. The degree −2 representation ring of the trivial group is equal to Z. Analogously to q = 0, we get that the 1-cochains are spanned over Z by a single element λ with λ(ℓ) = 1. Similarly, the 2-cochains are spanned by a single element α with α(A) = 1. However, unlike for q = 0, we have that α(T A) = T α(A) = −α(A) = −1. (3.15) This is because the action of T on the degree −2 representation ring is −1, as can be shown by an explicit analysis using Clifford algebras, using the explicit definitions in appendix A.2. We can conclude that the relevant part of the first page of the spectral sequence for class AI and class AII respectively is given in the following table: 17 p = 0 G(S2, φRτ0 C 0 q = 0 G(S2, φRτ0−1 q = 1 C 0 q = 2 G ) = Z2 G ) = Z2 p = 1 G(S2, φRτ0 C 1 G(S2, φRτ0−1 2 C 1 G(S2, φRτ0−2 C 1 G ) = Z G ) = 0 C 2 G ) = Z C 2 G(S2, φRτ0−1 G(S2, φRτ0−2 G ) = 0 G ) = Z q = 0 C 0 q = 1 C 0 q = 2 p = 0 G(S2, φRτ1 G(S2, φRτ1−1 p = 1 G(S2, φRτ1 C 1 G ) = Z2 G(S2, φRτ1−1 G ) = 0 C 1 G(S2, φRτ1−2 C 1 G ) = Z G ) = 0 C 2 G ) = Z C 2 G(S2, φRτ1−1 G(S2, φRτ1−2 G ) = 0 G ) = Z p = 2 p = 2 With the information we have gathered now, we can construct the second page, consisting G). For this we need to of Bredon cohomologies. Let us start off by computing H 0 compute the kernel of the Bredon differential G(S2, φRτ d : Z2 = C 0 G(S2, φRτ G) → C 1 G(S2, φRτ G) = Z. (3.16) On the 0-cochain π∞ it acts as dπ∞(ℓ) = π∞(∂ℓ)1 = π∞(p∞ − p0)1 = π∞(p∞)1 − π∞(p0)1 = π∞(p∞)1, (3.17) where the symbol 1 denotes the restriction of the representation to the trivial group. In class AI, this restriction maps complex vector spaces with a real structure T to their underlying complex vector space. Since all complex vector spaces admit a real structure, this implies that the restriction map φRτ (Z2) → φRτ (1) is the identity. In class AII, where T 2 = −1, the restriction is multiplication by two, because only complex vector spaces of even dimension admit a quaternionic structure. Hence we get dπ∞ = λ 2λ if τ = τ0, if τ = τ1. (3.18) Using the orientation we analogously get that dπ0 = −dπ∞. In both class AI and AII, we see that the degree zero cohomology equals ker d = H 0 G(S2, φRτ G) ∼= Z. (3.19) In general, this cohomology group contains all local topological invariants. More precisely, the zeroth degree cohomology group is actually a mathematical formalization of the heuristic 18 method of consistently assigning representations to point sketched in Section 2 and used extensively in [5] and [9]. The next row of the second page is easily deduced from the first page. In class AII, all cochains vanish and therefore so do the cohomology groups. In class AI nontrivial 0-cochains exist, but not in higher degrees. Therefore the differential is necessarily zero and the second page equals the first page. The final relevant cohomology group is H 2 G(S2, φRτ −2 G ). In this case, T induced a non- trivial automorphism on φRτ −2(1) given by −1, so that dλ(A) = λ(∂A) = λ(ℓ) − T λ(ℓ) = 2λ(ℓ) = 2, (3.20) hence dλ = 2α for both τ = τ0 and τ1, so that im d = 2Z. The kernel of d acting on 2-cochains is Z, since we are in top degree. Thus H 2 G(S2, φRτ −2 G ) = Z2. Summarizing the results by filling in the second page of the spectral sequence, we get the following tables for T 2 = 1 and T 2 = −1 respectively: p = 0 G(S2, φRτ0 H 0 G(S2, φRτ0−1 G ) = Z G ) = Z2 q = 0 q = 1 H 0 q = 2 p = 1 p = 2 2 H 1 G(S2, φRτ0−1 G ) = 0 H 2 H 2 G(S2, φRτ0−1 G(S2, φRτ0−2 G ) = 0 G ) = Z2 p = 0 G(S2, φRτ1 G(S2, φRτ1−1 G ) = Z G ) = 0 H 1 q = 0 H 0 q = 1 H 0 q = 2 p = 1 p = 2 G(S2, φRτ1−1 G ) = 0 H 2 H 2 G(S2, φRτ1−1 G(S2, φRτ1−2 G ) = 0 G ) = Z2 When T 2 = −1, we immediately see that all higher differentials vanish. The spectral sequence thus collapses at E2 and the exact sequences (3.5) and (3.6) reduce to the single exact sequence 0 → Z2 → KR−4(S2) → Z → 0. (3.21) Since Z is a free group, the sequence splits. This gives us KR−4(S2) = Z ⊕ Z2. Moreover, (S1) = 0 (for an example of a computation of the K-theory of a one-dimensional space using the spectral sequence, see the using the spectral sequence it can easily be shown that gKR next section). By the equivariant splitting (3.8), the K-theory of the torus is thus −4 KR−4(T2) = Z ⊕ Z2, (3.22) 19 which confirms the result using a different approach, see equation (2.1). It is worth noting that the torsion invariant Z2 managed to appear because of the nontrivial action of T induced by complex conjugation and not by the torsion in the KR-theory of a point as it does when computing KR−4(S2) using the methods of for example Freed and Moore [4]. For T 2 = 1 another lesson is to be learned from this example. Namely, note that as long as we do not know any expression for the second differential d2 : Z2 2 → Z2, we cannot uniquely determine the K-theory group by the spectral sequence method. However, we know from other methods that KR0(S2) = Z so that this differential must be surjective. If in future research an explicit expression for the second differential is found, it would be interesting to compute it in this example. 3.3 Time-reversal and a twofold rotation symmetry For a more exciting example, we now also include a rotation R by π. So we take the symmetry group G = Z2 × Z2 = {1, R} × {1, T }. We twist the group so that the twisted group algebra satisfies the desirable physical situation on the quantum level, namely R2 = T 2 = −1 and T R = RT . This thus represents spinful fermions on a two-dimensional square lattice with twofold rotation symmetry and hence the wallpaper group is p2. On the Brillouin torus T2 = [−π, π]2/ ∼ these symmetries act as T k = −k and Rk = −k. Before the topological computations, we first have to compute the twisted Bredon coef- ficients, i.e. the representation rings and the relevant maps between them. Note that the only stabilizers that occur are G and H := {1, T R}, so we only have to compute twisted representations for these groups. Because of this exceptional role played by T R it is useful to set S := T R and forget about T for the moment. Note that in the twisted group algebra, Si = −iS, S2 = 1 and SR = RS. The twisted group algebras are abstractly isomorphic to matrix algebras: φCτ H = φCτ G = R[i, S] (i2 = −S2 = 1, iS = −Si) ∼= Cl1,1 ∼= M2(R) R[i, S, R] (i2 = R2 = −S2 = 1, iS = −Si, RS = SR, iR = Ri) ∼= M2(C), (3.23) (3.24) where the last isomorphism follows because the twisted group algebra is φCτ H ⊗R C. There- fore the twisted group algebra of H is Morita equivalent to the algebra R, while the twisted group algebra of G is Morita equivalent to the algebra C. The representation rings are therefore φRτ −q(H) ∼= KR−q(pt) (3.25) 20 φRτ −q(G) ∼= K −q(pt), (3.26) see appendix A.2 for details on higher degree representation rings. The restriction map in degree zero Z ∼= φRτ (G) → φRτ (H) ∼= Z (3.27) is just given by mapping a complex vector space to its underlying real space and hence it is given by multiplication by two. For q = 1, the restriction map can only be zero, since K −1(pt) = 0. For q = 2 restriction is a map Z ∼= φRτ −2(G) → φRτ −2(H) ∼= Z2, (3.28) so it can either be zero or reduction mod 2. It is possible to explicitly check which it is by using explicit Clifford modules, but it turns out that we do not need to know which one it is in order to compute the K-theory. The only remaining map between representation rings is the action of R on the representation ring. This action is given by conjugating modules over φCτ H with R. Since R is in the center of φCτ G, the automorphism on φRτ (H) resulting from this is trivial. On the two relevant higher degree representation rings φRτ −1(H) = φRτ −2(H) = Z2 (3.29) the action of R is trivial as well because Z2 has no nontrivial automorphisms. In order to compute the full twisted equivariant K-theory of the Brillioun zone torus, we first use the equivariant splitting method, giving the isomorphism 3.8. Secondly we apply the spectral sequence on the components. Note that the circles occuring in the isomorphism, i.e. kx = 0 and ky = 0 in the Brillouin zone, have identical group actions. Hence they give isomorphic K-theory groups and we only have to compute one. Next we have to decide on G-CW decompositions of our new spaces S2 and S1. Since the action of R is the same as the action of T , we can reuse the G-CW structure of the last example for S2 as given in Figure 1. For the circle we use the one-dimensional sub-G-CW complex of the G-CW structure on S2. Let us start by computing the twisted equivariant K-theory of the circle. We compute the K-theory by using 3.11 for one-dimensional spaces. The zeroth-cohomology H 0 G) is analogous to the example in the previous subsection. We can define Z-bases of equivariant 0-cochains π0, π∞ and 1-cochains λ. In contrast with the last example, we now have complex vector spaces on the fixed points and real vector spaces on the k-cells for k > 0. Recall that G(S1, φRτ the restriction map sends a complex vector space to its underlying real space and therefore 21 this map is given by multiplication by two. The Bredon differential is thus given by dπ∞(ℓ) = π∞(∂ℓ)H = 2 =⇒ dπ∞ = 2λ. (3.30) G(S1, φRτ G(S1, φRτ −1 G) ∼= Z. Notice that for the first cohomology group Similarly dπ0 = −2λ. Hence H 0 H 1 G ) the twisted representation ring of G vanishes in te corresponding degree, so that the differential equals zero. Hence this cohomology group is equal to the group of equivariant 1-cochains hλiZ2, which equals the twisted representation ring of H in degree −1. We conclude that H 1 G ) ∼= Z2. Via equation (3.11), we arrive at G(S1, φRτ −1 φK τ G(S1) = Z ⊕ Z2. (3.31) Since φK τ of the torus. These are precisely the invariants proposed by Lau et al. G(S1) = Z2 for both circles in the splitting in [24] and when G(pt) = φRτ (G) = Z, we see that non-trivial represent a Mobius twist in the Hilbert space of states along the invariant circles at kx = 0 and ky = 0. Our K-theory computation thus provides a mathematical proof of the existence of this invariant. φeK τ Now we turn to the computation of the twisted equivariant K-theory of the 2-sphere. We use the same bases of equivariant cochains {π0, π∞}, {λ} and {α} as in the last example. For the zeroth cohomology, H 0 G), the computation is equivalent to the one in the previous subsection, hence H 0 G) = Z. Going to q = 1, we see that there are no 0-cochains, since φRτ −1(G) = 0. The differential on 1-cochains gives G(S2, φRτ G(S2, φRτ dλ(A) = λ(∂A)H = λ(ℓ) − λ(Rℓ)H = λ(l)H − Rλ(ℓ)H = 0, (3.32) (3.33) (3.34) since R necessarily acts trivially on φRτ −1(H) = Z2. Hence the cohomology groups are equal to the cochain groups: H 0 G(S2, φRτ −1 G ) = 0, H 1 G(S2, φRτ −1 G ) = Z2, H 2 G(S2, φRτ −1 G ) = Z2. (3.35) Since for q = 2 the 1-cochains and 2-cochains are exactly the same, the above computation also applies to the computation of the cohomology in degree 2. Therefore it follows that H 2 G ) equals Z2 as well. The relevant part of the second page is thus conveniently summarized in the following table. G(S2, φRτ −2 22 p = 0 G(S2, φRτ G(S2, φRτ −1 G) = Z G ) = 0 H 1 q = 0 H 0 q = 1 H 0 q = 2 p = 1 p = 2 G(S2, φRτ −1 G ) = Z2 H 2 H 2 G(S2, φRτ −1 G(S2, φRτ −2 G ) = Z2 G ) = Z2 The second differential d2 : H 0 G ) is either zero or reduction modulo 2. Independent of this distinction, the kernel of d2 is abstractly isomorphic to Z. Hence the relevant part of the final page of the spectral sequence agrees with the diagonal G) → H 2 G(S2, φRτ −1 G(S2, φRτ p = q in the table above. The exact sequences (3.5) and (3.6) that follow from the spectral sequence now reduce to 0 → Z2 → F → Z2 → 0, G(S2) → Z → 0. 0 → F → φK τ (3.36) (3.37) Note that the second sequence splits. Unfortunately, the first exact sequence implies only G(S2) that F = Z2 is either Z ⊕ Z2 2 or Z ⊕ Z4, depending on whether the first exact sequence splits or not. We can conclude from equation 3.8 that 2 or F = Z4. Hence the Atiyah-Hirzebruch spectral sequence gives that φK τ φK τ G(T2) ∼= Z ⊕ Z4 2 or φK τ G(T2) ∼= Z ⊕ Z2 2 ⊕ Z4. (3.38) To determine which of these two is the correct one, we employ the equivariant Mayer- Vietoris exact sequence. We can focus on the sphere since two possibilities for the K-theory originated there. Take open G-neighbourhoods of the north and south pole as U1 = S2 \{p∞} and U2 = S2 \ {p0}. Now consider the following part of the equivariant Mayer-Vietoris exact sequence with respect to U1 and U2: · · · → φK τ −1 G (U1 ∩ U2) → φK τ G(S2) → φK τ G(U1) ⊕ φK τ G(U2) → . . . (3.39) Note that both U1 and U2 are G-contractible to a point. Hence φK τ G(U1) ⊕ φK τ G(U2) ∼= φK τ G(pt) ⊕ φK τ G(pt) ∼= Z2. Moreover, U1 ∩ U2 is G-homotopy equivalent to the equator S1. To compute the twisted equivariant K-theory of this S1, note that the action of the subgroup H ′ generated by R is free. This implies that we can quotient this subgroup without having to worry about orbifold- type singularities. Since there is a homeomorphism S1/H ′ ∼= S1, we arrive at a circle with 23 a trivial H-action. Because twisted K-theory of orbifolds is invariant under equivalence [21], we see that φK τ −1 G (U1 ∩ U2) ∼= φK τ −1 G (S1) ∼= φK τ −1 H (S1). (3.40) The new twist is simply the restriction of the old twist to H and since (T R)2 = 1, the twist results in nonequivariant KO-theory. Using suspensions and reduced KO-theory, we then arrive at φK τ −1 G (U1 ∩ U2) ∼= KO−1(S1) ∼= KO−1(pt) ⊕gKO ∼= KO−1(pt) ⊕ KO−2(pt) ∼= Z2 2. Hence the equivariant Mayer-Vietoris sequence takes on the form · · · −→ Z2 2 f1−→ φK τ G(S2) f2−→ Z2 −→ . . . −1 (S1) (3.41) (3.42) (3.43) G(S2) does not contain any 4-torsion. A simple diagram chasing argument now shows that φK τ Indeed, suppose that a ∈ φK τ G(S2) satisfies 4a = 0. We will show that this implies 2a = 0. First note that 4f2(a) = f2(4a) = 0. However, the image of f2 is torsion-free so we necessarily have that f2(a) = 0. We can conclude that a is in the kernel of f2. By exactness, this implies 2 such that f1(b) = a. Now it follows by the group structure of Z2 that there is some b ∈ Z2 2 that 2a = 2f1(b) = f1(2b) = f1(0) = 0 as desired. The twisted equivariant K-theory for p2 symmetry in class AII is thus, φK τ G(T2) ∼= Z ⊕ Z4 2, (3.44) which exactly agrees with the heuristic arguments proposed in [9]. In particular, there it is argued that each Z2 invariant comes from a vortex anti-vortex pair in the Berry connection stuck at the high-symmetry points. In this case, there are four of such point and hence four Z2 invariants. The part of this K-theory coming from the sphere was also computed in [13]. Our method now provides a rigorous mathematical proof of this computation. Notice that the spectral sequence also gives insight on the origin of the invariants. First of all, the Z factor is simply the rank of the bundle. Next we have two Z2's that are Mobius- type line invariants along the two cycles of the torus and are simply the invariants already found in [24]. On the Brillouin zone sphere remaining after the equivariant splitting, we have again one such Z2-line invariant. Finally, we have the fourth Z2-invariant, which is a Fu- Kane-Mele-type surface invariant on the Brillouin sphere. We discuss a possible connection between these invariants and the vortex picture in the Section 5. Also note that from the 24 equivariant Mayer-Vietoris exact sequence we could only determine the type of torsion and not the full K-theory group. It is the combination with the Atiyah-Hirzebruch spectral sequence that resulted in the final answer. 4 Generalizations We will now discuss various generalizations of the simple examples we studied in the pre- vious section. Furthermore, we will give an algorithmic method to compute the twisted representations rings. 4.1 Other crystal symmetries In the above we performed computations involving either no spatial symmetries or a twofold rotation. The computation of the last section can be generalized straightforwardly to a fourfold rotation. The only thing that requires extra attention is that even though the splitting (3.7) holds true, the K-theory does not split further according to equation (3.8). This is because the fourfold rotation interchanges the two circles of the figure eight. So one really has to compute the K-theory of the figure eight. The final result for a fourfold rotation in class AII is φK τ Z4×ZT 2 (T2) = Z3 2 ⊕ Z3. (4.1) Just as for a twofold rotation, there are two torsion invariants coming from the spherical Brillouin zone: one is a line invariant and one a surface invariant. However, there is only one Z2-invariant corresponding to the boundary figure eight, since the two line invariants of the last section are identified by the fourfold rotation. It is amusing to see that this computation exactly agrees with the more heuristic arguments presented in [9]. There are three high- symmetry points (not related by any symmetry operation) at which vortex anti-vortex pairs can be stuck, giving rise to three Z2 invariants. The free part of the K-theory is simply the consistent assignment of representations as discussed in previous subsections. It is interesting to do the same computation for wallpaper groups with reflections. For instance, let us consider a two-dimensional crystal with a single reflection symmetry. On the Brillouin torus this symmetry acts as t · k = (kx, −ky) for k ∈ T2. Just as for the twofold rotation symmetry, we have that on the fibers t2 = −1. Going through the spectral sequence analysis, we find that the spectral sequence method gives a unique answer by itself. The K-theory is given by φK τ Z2×ZT 2 (T2) = Z2 2 ⊕ Z, (4.2) in class AII. This result also agrees with [9]. The Z2 invariants come from the circles fixed 25 under the reflection symmetry. Note that one of these circles is part of the figure eight boundary, whereas the other comes from the circle of reflection on the sphere after the equivariant splitting (3.7). The part of the K-theory coming from the sphere (in particular exactly one of the two Z2 invariants) was also obtained in [12 -- 14]. For more complicated symmetries, the method becomes rather involved. For example, symmetries with elements of odd order can in two spatial dimensions only occur on a hexago- nal lattice, so one has to take the non-trivial identifications of the Brillouin zone into account. The Brillouin zone will still be a torus, but the equivariant splitting we used in the previous section becomes more difficult. Also we cannot make use of the real structure T R in the same way we did above; in case of an odd order rotation the stabilizer of a generic point will be trivial instead. Moreover, T R will anticommute with reflections in class AII, see the example at the end of this section. The basic difficulty as the symmetry group gets larger is computing the twisted group algebras and their representation rings; determining them using abstract algebra as we did in equations (3.23) and (3.24) quickly becomes tedious. One way forward is to give an alternative way of describing and constructing the twisted representation rings. Essentially, the only requirement for the construction of these twisted representation rings is knowing how many representations exist and which are real, complex and quaternionic. More precisely, if the number of real, complex and quaternionic irreducible (φ, τ )-twisted representations of G is denoted by nk with k = R, C, H, we have φRτ (G) = K 0(pt)nC ⊕ KR0(pt)nR ⊕ KR−4(pt)nH. (4.3) Once this data has been computed, the higher order rings follow from the Bott clock, see appendix A.2. Remember that although we refer to φRτ (G) as rings, they are actually not rings. The task we are thus left with is to determine the integers nk. In other words, we need an adequate representation theory for twisted groups. From a physics point of view, such a theory was outlined in [25] and we will now showcase this to make contact with the approaches to such problems by the crystallography community. It would also be interesting to see this point of view compared with the Wigner test, which is the appropriate generalization of the Frobenius-Schur indicator as given in [25]. Although we will not give a rigorous proof of the connection between the representation theory of space groups and twisted representation rings, we will give evidence for such a connection below and in appendix A.2. In the rest of this section, we will mention the procedure of [25] and illustrate it using a simple example. For this we will need to briefly change gears. We formulate twisted groups in terms of double covers and twisted representations as double-valued representations. To see how this formulation is related to the twists τ in the main text, the reader may wish to 26 consult the final paragraph of appendix A.1. Although this procedure works for both class AI and AII, let us focus on the latter. For simplicity we assume that G = G0 × ZT 2 , where ZT 2 represents the time-reversal action on the Brillouin zone and G0 consists of the other symmetries. We have written G in this way, because G0 will be lifted to a linear action on the Hilbert space, whereas time-reversal lifts to an anti-linear action. Let us focus first on G0. In class AII, we are dealing with fermions (class AI assumes the system is bosonic) and we have to consider a certain double cover of G0, which we denote by bG0. The representations of the double cover take the usual signs into account that come for example from rotations over 2π acting with a minus sign on the Hilbert space. In this sense they form the structure analogous to the twist τ in the rest of this paper. It is usually intuitively clear which double cover is desirable, but for a general point group G0 of a d-dimensional lattice it can be described by the following abstract mathematical construction. The double cover should be the pullback of the negative Pin group Pin−(d) covering the orthogonal group: 1 1 Z2 Z2 bG0 Pin−(d) G0 O(d) 1 1. The reason that it is not the spin group is that this group would only account for rotations, not for reflections. In other words, Spin(d) is only a double cover of SO(d). The reason that we pick the negative Pin group Pin−(d) instead of the other central extension Pin+(d) of O(d), is that in Pin+(d) reflections square to the identity instead of to the desired −12. as the dicyclic group of order 4n. Twisted representations of the group G0 can now be It is useful to note that if G0 = Dn is the symmetry group of an n-gon, then bG0 is known described as certain ordinary representations of the double cover group bG0. Let us call representations fermionic when the newly introduced subgroup Z2 ⊆ bG0 acts nontrivially. Representations where the Z2 acts trivially are called bosonic. This exactly means that 2π rotations (or double reflections) act by −1 on fermionic representations and trivially on bosonic representations. In general, the double cover will admit both fermionic and bosonic representations, but for systems in class AII the Hilbert space is organized in terms of fermionic representations only. Given these fermionic representations, we are now ready to add in time-reversal symmetry. This enlarges the group with another generator T that squares to minus one. Set-theoretically, we can write the full group acting on the Hilbert 2When other symmetries such as gauge or flavour symmetries are present or when there are interactions, the double cover could also be Pin−(d). Although it would be interesting to map out all possible choices, it is beyond to scope of the present work. 27 space as 3 and (4.4) (4.5) (4.6) bG = bG0 ⊔ AbG0, the non-trivial element in the double cover of the trivial group. Notice also that the choice takes A = T , but other forms of A are also possible. For example, in the example in the where A is an antiunitary symmetry operator, i.e. φ(Ag) = −1 for any g ∈ bG0. Notice that starting with any group bG and any nontrivial homomorphism φ : bG → Z2, we could have created such a decomposition by taking bG0 = ker φ and picking some A /∈ ker φ. Usually one previous subsection, for H := {1, T R} we would have A = T R and bG0 = {1, −1}, with −1 of A in (4.4) is a bit arbitrairy as A′ = Ag for g ∈ bG0 instead of A will give rise to unitarily Let us denote a fermionic representation of bG0 by ρ, which we can assume to be unitary. As shown in [25], the fermionic (matrix) representations D of bG are then given by equivalent representations. 0 D(g) = ρ(g) D(Ag) = 0 ¯ρ(g) 0 ¯ρ(A−1gA)! 0 ! ρ(AgA) for g ∈ bG0. Here ¯ρ denotes the complex conjugate of the representation ρ. When A is just T , we can commute it with g ∈ bG0 and the expressions become much simpler. In general we can always write A as T times something in G0 and then it will appear quadratically or not at all. Hence T can be factored out by using the fact that ρ is a homomorphism. Thus, the representation D will only depend on the sign of T 2, which for our case is minus one. For these representations the notion of reducibility is similar to ordinary representations, see [25]. Now there are three different cases to distinguish: a) Either time-reversal symmetry does nothing to the representation, b) two unitarily equivalent representations of dimension k form a new (irreducible) representation of dimension 2k, or c) complex conjugate irreducible representations of dimension l form a representation of dimension 2l. These three cases can be described as follows: a) In this case ρ(g) is unitarily equivalent to ¯ρ(A−1gA), i.e. ρ(g) = N ¯ρ(A−1gA)N −1 for some fixed unitary matrix N and g ∈ bG0. Moreover, N satisfies N ¯N = +ρ(A2), and 3Note that in the previous section, we used G to denote the group acting on the Brillouin zone. The group bG is the lift of that group to a group acting on the Hilbert space in which time-reversal, 2π rotation and double reflections square to minus one. Hence fermionic representations of bG are equivalent to twisted representations of G. 28 then D(g) = ρ(g) and D(Ag) = ±ρ(AgA−1)N 4. b) In this case ρ(g) is unitarily equivalent to ¯ρ(A−1gA), i.e. ρ(g) = N ¯ρ(A−1gA)N −1 for some fixed unitary matrix N and g ∈ bG0. However, N satisfies N ¯N = −ρ(A2), and then (4.7) D(g) = ρ(g) 0 0 ρ(g)! , D(Ag) = 0 −ρ(AgA−1)N ρ(AgA−1)N 0 ! . c) In this case ρ(g) is not unitarily equivalent to ¯ρ(A−1gA). The representations are then given by D(g) = ρ(g) 0 0 ¯ρ(A−1gA)! , D(Ag) = 0 ¯ρ(g) ρ(AgA) 0 ! . (4.8) It is clear that representations of type c) always correspond to complex representations. Type a) are the real representations and type b) are quaternionic representations. An important subtlety is when the unbroken symmetry group of the fixed point does not contain any antiunitary symmetries. In that case the representations remain complex, just as we saw in Table 1 in the case of a trivial stabilizer group. We will now study some simple examples to see how this works in practice. For a rotation symmetry Zn, the double cover is Z2n which has n complex fermionic representations for n even. For n odd, there are n − 1 complex fermionic representations and one quaternionic representation. Due to time-reversal symmetry the complex fermionic representations will pair up and hence the twisted representation ring of degree −q of G = Zn × ZT 2 is φRτ −q(G) =( K −q(pt)n/2 if n = even K −q(pt)(n−1)/2 ⊕ KR−q−4(pt) if n = odd . (4.9) This is exactly the same result as one would get by constructing the representation ring of the twisted group algebra, which is a more formal way of computing the twisted equivariant K-theory of a point. More details on that construction can be found in the appendix. A more non-trivial example is G0 = Z2 × Z2. This group is generated by t1 and t2, which represent reflections in the kx and ky axis respectively. The double cover of this group is Q8, the quaternion group. The action of G0 on the Brillouin zone torus T2 has four fixed 4In this case, the matrix representations of g and Ag for g ∈ bG0 given in (4.5) and (4.6) can both be made block diagonal and in fact the representation D is reducible. Consequently, ρ and D have the same dimensionality. The ± appearing for D(Ag) represents to unitary equivalent representations. See [25] for more details. 29 points (0, 0), (0, π), (π, 0) and (π, π) and four fixed circles (0, ky), (π, ky), (kx, 0) and (kx, π). The fixed points have stabilizer group G0. We will analyze the representations of the cover of this group first. The group Q8 has five representations of which only one is fermionic, since for all other representations −1 ∈ Q8 acts trivially. This fermionic representation is just its regular action on the quaternions H, which is a two-dimensional representation over the complex numbers. We denote the generators of Q8 by t1 and t2. The representation is concretely given by ρ(t1) = iσ1, ρ(t2) = iσ2 (4.10) where σi are the Pauli matrices. Note that since we are at fixed points whose stabilizer group is the full point group, we have A = T . To determine what time-reversal does with these representations, we have to find out whether ρ and ¯ρ are unitarily equivalent. Clearly this is the case, since N = iσ2 is an explicit unitary matrix that intertwines ρ with ¯ρ. To see this, note that it anticommutes with all purely imaginary matrices in this representation. Given this N , we have N ¯N = −σ2 2 = −1 = +T 2. (4.11) Thus we are in case a) and we have φRτ −q(G) = KR−q(pt). For the fixed circles the twisted representation theory of the stabilizer is unexpectedly interesting. Consider for example the circle ky = 0. This circle is fixed by H = {1, t2}, 2}. The full set of elements in the double cover (including time-reversal) that leave ky = 0 fixed is {1, t2, t2 2}, hence we pick A = T R, where R = t1t2 is the lift of the twofold rotation in the double group. Note that A2 = 1, but since A contains the rotation R it anticommutes with the reflections. which lifts to the double cover bH = {1, t2, t2 2} ⊔ T R{1, t2, t2 2, t3 2, t3 2, t3 Even though the fermionic representations ρ± of bH have complex eigenvalues ±i for the reflection, these two irreps nevertheless belong to case a). To see this, first note that since these representations are one-dimensional over the complex numbers, N drops out. Hence two such representations are unitarily equivalent if and only if they are equal. Now note that there is an extra minus sign that cancels the minus sign coming from complex conjugation. Indeed we have ¯ρ± = ρ∓ and hence ¯ρ±(A−1t2A) = ρ∓(−t2) = ρ±(t2). (4.12) For the other fixed circles the exact same argument holds. At the fixed circles we therefore have two real representations and hence the representation ring is φRτ −q(H × ZT 2 ) = KR−q(pt) ⊕ KR−q(pt). (4.13) 30 From this computation, one immediately sees that a lot of torsion will appear in the spectral sequence. We will not compute the full K-theory for this crystal group here, but we expect the exact sequences that result from the spectral sequence to split. The approach given in this section has a natural extension to non-symmorphic symme- tries, but the computations become more tedious. We will discuss these symmetries further in the discussion section. 4.2 Class AI Topological insulators in class AI satisfy T 2 = 1. The bundle therefore has a real structure given by T . In the absence of a point-group symmetry, the K-theory classifying topological insulators is Real K-theory KR, which we computed in (2.1). When the topological insulator has a non-trivial crystal symmetry, the twist τ is trivial. Therefore we have to compute equivariant Real K-theory, which for a group G is denoted by KRG. We can again use the spectral sequence method to compute the relevant K-theory groups. For a single reflection symmetry in class AI we find that KRZ2(T2) = Z3, (4.14) which is again in agreement with [12 -- 14]. The invariants are purely coming from the repre- sentations at the fixed points together with a non-trivial glueing condition. It turns out that for other symmetries, the K-theory for class AI is much harder to com- pute. In the examples we considered, the computations are plagued by the higher differentials of the spectral sequence. This problem appears especially in class AI because in two dimen- sions, the second differential can only make a nontrivial contribution in case the (twisted) stabilizer Hk of some zero-cell k admits a real representation. In contrast to class A (where all representations are complex) and in class AII (where real representations do appear some- times), real representations are the norm in class AI. This is because time-reversal is a real structure. In particular whenever Hk contains T , the trivial representation of Hk is always real. Therefore the second differential can often make contributions to the K-theory, but we have not been able to show what these contributions are for a general crystal symmetry. What we do know however is that for the analogous computation of the one done in 3.3 (i.e. time-reversal and a twofold rotation symmetry) the K-theory is one of the two groups KRZ2(T2) = Z5 or KRZ2(T2) = Z5 ⊕ Z2. (4.15) We were not able to show which of these two is the actual answer, but the analysis in [26] 31 suggests that there cannot be any torsion in this case. In this work a fourfold rotation symmetry is considered and it is shown that a Z2 invariant appears because the two complex representations form a two dimensional representation at the fixed points (0, 0) and (π, π) once time-reversal symmetry is taken into account. An effective time-reversal operator T R can be defined that squares to minus one and hence at these points the vector bundle has the same quaternionic structure that is found in class AII. This observation was crucial to show that there is a single Z2 for p4 in class AI. For a twofold rotation symmetry a similar construction does not work, which is why we believe KRZ2(T2) = Z5. From the point of view of the spectral sequence, this would be the case if the second differential d2 (of which no convenient explicit expression is known) is surjective. In [10], it is indeed argued that in this case d2 is surjective, just like we found in case when no point symmetries were present. We have also computed the KR-theory associated to p4 rotation symmetry in class AI, but we ran into the same problems as for p2. However, from the above discussion, we know that there should be a single Z2 invariant. If a closed expression for the second differential is derived, it would therefore be interesting to show rigorously that d2 is surjective for p2, but not surjective for p4. In case of more non-trivial crystal symmetries, the representation theory and hence its twisted rings can be computed using the technique outlined above. The difference is that now we are interested in the bosonic representations of G. In other words, we do not have to consider the double cover group, but instead we can work with the point group itself, thus (4.4) changes to G = G0 ⊔ AG0. If A happens to commute with all g ∈ G0, this means we just have to determine the real representation theory of G0. So, for example let us consider a point group D4. This group has five representations and all are realizable over the real numbers. These representations are thus in case a). The twisted representation ring is then φRτ −q(D4 × ZT 2 ) = KR−q(pt)5 (4.16) Again for a generic point on the torus, there is an unbroken symmetry group H = {1, T R2}, hence a smart choice would be to pick A = T R2. Notice that again (T R2)2 = 1 and T R2 commutes with all elements of G. For the subgroup H we have H0 = {1}, whose representation obviously belongs to case a), hence φRτ −q(H) = KR−q(pt). The geometric action of D4 on the torus has three other non-trivial stabilizer groups, two isomorphic to Z2 and one isomorphic to Z2 × Z2. The representation rings for these two groups also only consist of copies of KR−q(pt), since all representations are real. In fact, we have φRτ −q(Z2 × ZT 2 ) = KR−q(pt), (4.17) 32 φRτ −q(Z2 × Z2 × ZT 2 ) = KR−q(pt). (4.18) 4.3 Class A In class A the spectral sequence is well-known [27, 28] and the computations are a lot more tractable. There are no antiunitary operators and the K-theory is just Atiyah & Segal's complex equivariant K-theory [29], but possibly twisted. The twist now comes purely from non-symmorphic symmetries. Let us focus on symmorphic symmetries so that there is no twist. In this case, the relevant exact sequences in the spectral sequence constructed here will always split and so unlike in class AI and AII, we get a unique answer. One might wonder whether the higher differentials could give non-trivial contributions. Firstly, since R−q G = 0 for odd q, every other row on the second page of the spectral sequence is trivial so that the second differential always vanishes. Moreover, in two dimensions the third and higher differentials always vanish. Thus we can easily determine the equivariant K-theory exactly. Indeed, since in the complex case R−2 G = RG, one quickly sees that the exact sequences (3.5) and (3.6) imply that KG(X) ∼= H 0 G(X, RG) ⊕ H 2 G(X, RG). (4.19) This isomorphism also holds in three dimensions, since in that case the third differential gives no additional contribution. We will illustrate this fact by a short argument. If X is three-dimensional, the arguments above give us the isomorphism KG(X) ∼= ker(cid:0)d3 : H 0 G(X, RG) → H 3 G(X, RG)(cid:1) ⊕ H 2 G(X, RG). (4.20) Now note that in class A, the Bredon cochains map into ordinary representation rings of subgroups H ⊆ G. Since these representation rings are always torsion-free, so are all groups of Bredon cochains. Since H 0 G(X, RG) is a kernel of a Z-linear map between Bredon cochains, it must therefore also be torsion-free and hence so is ker d3. But by the equivariant Chern character isomorphism [18] KG(X) ⊗ C ∼= H 0 G(X, RG ⊗ C) ⊕ H 2 G(X, RG ⊗ C), (4.21) we already know that the formula (4.19) holds modulo torsion. Therefore, there must be an isomorphism ker d3 ∼= H 0 G(X, RG). Note that even though this argument does not imply that d3 vanishes, it still implies that it can be ignored in abstract computation. Let us reflect on the results we have just established in class A. First of all, in light of [5], 33 we see that there is indeed a clear distinction between the representations at fixed points and how they are glued to the representations at lines on the one hand, which are captured by H 0 G(X, RG), and higher-dimensional invariants, such as Chern numbers on the other hand, which are in H 2 G(X, RG). In two dimensions, one can check by explicit computation that H 2 G is torsion-free, but in three dimensions it is known that it contains torsion in certain examples [10]. Nevertheless, the torsion-free part is still captured by the proposed algorithm in [5]. The torsion is hard to understand systematically, but intuitively one expects it to arise from either non-symmorphic space groups or non-trivial identifications due to the crystal structure. Since the Bredon cohomology of a complex is something purely combinatorial, equation 4.19 provides an algorithmic approach to computing the K-theory for class A in full generality. It would be interesting to develop such an algorithm and compare it to the results of [10]. In fact, comparing with existing literature on the equivariant K-theory associated with the space group F 222, we see that [10] obtains an Z2 invariant in class A, whereas in [30] this K-theory is found to be torsion-less. We leave a detailed analysis of this discrepancy to future work. We also briefly mention a useful alternative method to determine the K-theory groups in class A in some cases. Namely, in the cases in which the equivariant splitting method applies, the classification is determined by the equivariant K-theory of representation spheres. The nontwisted complex K-theory of representation spheres is easily determined in terms of purely representation-theoric data as was described by Karoubi, see the survey paper [31]. It would be interesting to research whether pure representation-theoretic data could describe the K-theory of representation spheres in case time-reversal is included. 5 Discussion We have outlined a way to compute the K-theory that classifies topological insulators with or without time-reversal symmetry and with non-trivial crystal symmetry. Using an Atiyah- Hirzebruch spectral sequence, we computed these groups in a couple of examples and saw that often more work is needed to compute the exact answer. Nevertheless, it is noteworthy that the classification with K-theory matches with the rather heuristic arguments presented in [5, 9], at least for the examples that we computed. Moreover, with the techniques of crys- tallography, we were able to give an algorithmic way of computing the twisted representation rings in any degree. In our K-theory computations, we have also stumbled upon some difficulties that in general seem hard to overcome. Firstly, there is the fact that as of yet no explicit expression for the higher differentials of the spectral sequence is known, even in the simplest cases. For 34 nonequivariant complex K-theory, it is known that the second differential vanishes and the third differential is the extended third Steenrod square Sq3, which is the composition H p(X, Z) → H p(X, Z2) Sq2 → H p+2(X, Z2) β → H p+3(X, Z), (5.1) where Sq2 is the second Steenrod square and β is the Bockstein homomorphism associated to the exact sequence 0 → Z ×2→ Z → Z2 → 0. (5.2) This result has been generalized to twisted complex K-theory in [32], but even in nontwisted equivariant K-theory the situation is much more involved as is illustrated in [28]. Real K-theory on the other hand even introduces a second differential and is less studied in the literature. For KO-theory (the K-theory that classifies real vector bundles instead of complex ones, i.e. KR-theory with trivial involution) it has long been known that the second differential is the (appropriately extended) second Steenrod square [33]. Keeping the applications in mind, it would be interesting to work out explicit expressions of the higher differentials for small groups and CW-complexes from their abstract definition. Intuitively, a non-trivial rth differential represents obstructions of extending the vector bundle on a d dimensional subspace to an d+r dimensional subspace. This intuitive understanding was used in [10] to argue that a non-trivial rth differentials is an obstruction of smoothly extending (i.e. without gap closing) a topological insulator on a d-cell to an (d + r)-cell. Surprisingly, this allowed the authors to construct explicit expressions for the higher differentials in specific examples. It would be would be interesting to rigorously show that this construction works in general. The second and more fundamental difficulty in using the spectral sequence method is the problem of non-unique extensions. Indeed, since the exact sequences (3.5) and (3.6) are not always split, we cannot determine the K-theory uniquely unless we explicitly know the maps involved. In Section 3.3 we had to face this problem, since torsion groups appeared both in degree 1 and degree 2 simply because KR−1(pt) = KR−2(pt) = Z2. With just the Atiyah-Hirzebruch spectral sequence in our toolbox, this problem could only be solved by explicitly determining all maps involved in our exact sequences, which is tedious even in simple examples. In order to fully determine the (especially 2-)torsion invariants for general point groups, we will need a supplement to the spectral sequence. The supplement we used in Section 3.3 was the equivariant Mayer-Vietoris exact sequence. There are several other possibilities for such a supplement. One would be an Adams-type spectral sequence. Such spectral sequences are made precisely to measure the torsion part of groups of stable homotopy classes of maps between spaces. Another supplement, which was recently discussed 35 in [34], relates a K-theory with a non-unique extension problem to one which does have a unique extension through a notion of T-duality. Setting aside these difficulties, it would also be interesting to apply our method to topolog- ical superconductors and insulators with a chiral symmetry. These cases cover the remaining 7 Altland-Zirnbauer classes [3] and might also give many new invariants that can be studied experimentally. A simple example would be to study a topological superconductor with only particle hole symmetry C which can square to +1 or −1. Such systems are in class D and C, respectively. In two dimension the classification without any symmetry is just Z and it would be interesting to see how crystal symmetry changes this. However, even though chiral symmetries are incorporated in the framework of Freed & Moore's K-theory, it does not seem to be well-suited for this purpose. For example, for class AIII a short argument shows that the Freed-Moore K-theory group of S1 vanishes in case no other symmetries are present. This contradicts the ten-fold way, which says that in one dimension class AII topological insulators on a spherical Brillouin zone are classified by Z. As argued in [21, §3.5], this discrepancy results from the fact that the types of K-theory that include chiral symmetries are no longer realizable by finite-dimensional bundles as Freed and Moore assume. However, this seems to contradict the physical principle that our topological insulators only admit a finite number of bands. Assuming that the K-theory defined in [21] is the physically relevant type of K-theory, we know from this work that it satisfies the desired axioms for cohomology. In that case, the higher representation rings admit an obvious generalization to particle-hole reversing symmetries and a version of the spectral sequence similar to the one developed here therefore probably holds. Another very interesting class of symmetries, which we have not touched upon yet, are non-symmorphic symmetries. In two dimensions most symmetries are symmorphic, but in three dimensions there is a large class of crystals that exhibit some form of non- symmorphicity. The implementation of such symmetries in our recipe is mathematically challenging, because non-symmorphic crystals give twists that can vary throughout the Bril- louin zone. The known representation theory of non-symmorphic space groups seems to reveal that these non-trivial twists can result in a change in the type of representation at fixed loci and hence in different K-theory at different points. We leave a full understanding of these twisted representation rings and a rigorous construction of the spectral sequence for non-symmporphic space groups to future work. In the introduction we explained an intuitive picture of the Z2 invariants in class AII that was put forward in [9]. In particular, when crystal symmetries are present it was argued in [9] that in two dimensions the vortex anti-vortex pair is stuck on fixed points whenever there is a rotation symmetry and stuck on fixed circles whenever there is a reflection symmetry. 36 In string theory there is a analogous interpretation. Witten showed in [35] that there is a direct relation between the charges of D-branes on orbifolds and equivariant K-theory. Depending on what string theory is considered and whether there are involutions present, various versions of K-theory classify the corresponding D-brane charges. Moreover, at the orbifold singularity, say C2/Zn, only charge-n D-branes can be peeled off, other charges are stuck on the singularity, branes with these charges are called fractional branes. In K-theory this means that only certain vector bundles see the singularity. This is similar to the fact that only 2 vortex anti-vortex pairs can be moved away from a fixed point with n-fold rotation symmetry for topological insulators in class AII. Thus a single vortex anti-vortex pair is frozen on the fixed point. Although this frozen vortex picture gives the correct number of Z2 invariants for the example we computed, this interpretation is not immediately clear from our K-theory com- putations. One way to clarify this, is by using a localisation technique by Segal and Atiyah- Segal [29, 36]5. Originally, this is a result that applies to (untwisted) equivariant complex K-theory, KG(X), and uses the fact that KG(X) is an R(G)-module, with R(G) the rep- resentation ring of G. Generalisations to other K-theories have also been mentioned in the literature, [37, 38]. For instance, in the latter reference, this technique has been applied to the three-dimensional diamond structure in class A, which has a non-trivial twist. For our purposes we would have to be generalise the localisation technique to cases in which a time-reversing operator is present as well, which we hope to pursue in future work. Acknowledgement It is a pleasure to thank Gregory Moore, Peter Teichner, Bernardo Uribe and Jasper van Wezel for their engaging discussions. JK is supported by the Delta ITP consortium, a program of the Netherlands Organisation for Scientific Research (NWO) that is funded by the Dutch Ministry of Education, Culture and Science (OCW). A Appendix A.1 Freed & Moore K-theory and twists In order to illustrate our approach to computing these K-theory groups, let us first rigor- ously define the notions used in the text. While doing this, we also connect the technical mathematical language of Freed & Moore [4] to our more concrete setting. To motivate this, first consider a 2 + 1-dimensional square crystal and a finite classical symmetry group G 5We thank Gregory Moore for this suggestion. 37 consisting of a time-reversing symmetry T and a spatial symmetry R of rotation by π. To account for the fact that R acts unitarily and T acts antiunitarily, we define a homomorphism φ : G → Z2 by φ(R) = 1 and φ(T ) = −1. Although the classical group is G = Z2 × Z2, we know that for fermions, we have T 2 = R2 = −1 on the quantum level. Hence we are not interested in modules over the group algebra of G, but in modules over a twisted group algebra. To implement this fact mathematically, we twist the group G by a group 2-cocycle τ ∈ Z 2(G, U (1)), where U (1) is the circle group seen as a G-module by T eiθ = e−iθ and Reiθ = eiθ. This cocycle is given by τ (T, T ) = −1, τ (R, R) = −1, τ (T, R) = τ (R, T ) = 1. (A.1) We extend this definition to a cocycle on all of G by the cocycle relation and demanding that τ (g, 1) = τ (1, g) = 1 for all g ∈ G. On the quantum level of twisted representations, we want to impose equations such as T · T = τ (T, T ) instead of the equations holding in G. To implement this in a more general setting, suppose that we are given a finite classical symmetry group G consisting of point group symmetries and (possibly) time-reversal sym- metry. Let φ : G → Z2 be a homomorphism determining whether a group element acts unitarily or antiunitarily. Consider a group 2-cocycle τ ∈ Z 2(G, U (1)φ), where U (1)φ is the G-module g · eiθ := eφ(g)iθ. By the one-to-one correspondence between group extensions and group cohomology, the data of τ is equivalent to what is called a φ-twisted extension in Freed & Moore [4]. This is a group extension 1 → U (1) → Gτ π→ G → 1 (A.2) such that eiθg = geφ(π(g))iθ for all g ∈ Gτ . isomorphic if and only if the corresponding group 2-cocycles are cohomologous. Therefore only the cohomology class [τ ] ∈ H 2(G, U (1)φ) of the cocycle is relevant for the theory. In fact, two such φ-twisted extensions are Now we can define how to twist representations of G by φ and τ . Indeed, a (φ, τ )- projective action of G is a map ρ : G → GLR(V ) into the real linear automorphisms of a complex vector space V such that ρ(g) is complex linear if φ(g) = 1, complex antilinear if φ(g) = −1 and ρ(g)ρ(h) = τ (g, h)ρ(gh). (A.3) Note in particular that if τ is nontrivial, ρ is not a homomorphism of groups. Via the cor- respondence between group cocycles τ and extensions, such projective actions are exactly the same as (φ, τ )-twisted representations in the sense of Freed and Moore [4]. These are defined as genuine homomorphisms ρτ : Gτ → GLR(V ) into the real linear automorphisms 38 of a complex vector space V such that ρτ (g) is complex linear if φτ (g) = 1, complex an- tilinear if φτ (g) = −1 and ρτ (z) is just multiplication by z if z is in the circle subgroup U (1) ⊆ Gτ . Therefore we will use (φ, τ )-projective actions and (φ, τ )-twisted representations interchangeably. The twist τ can also be used to twist the group algebra as follows. We define the twisted group algebra φCτ G to be the 2 · #G-dimensional algebra over R generated by the symbols xg for every g ∈ G and a formal imaginary unit i with defining relations xgxh = τ (g, h)xgh, i2 = −1, xgi = φ(g)ixg. (A.4) If no confusion can arise we usually just write g for the symbol xg. Modules over the twisted group algebra are clearly equivalent to projective actions with cocycle τ and hence equivalent to (φ, τ )-twisted representations. For example, suppose we consider time-reversal symmetry T and n-fold rotation R in a system of spinful fermions. Then the symmetry group is G = Zn × Z2 and φ is projection on the second factor. It can be shown using basic techniques in group cohomology that H 2(G, U (1)φ) = Z2 2 if n is even, Z2 if n is odd. (A.5) For n even, the two Z2's correspond exactly to the choices of signs in Rn = ±1 and T 2 = ±1. For example, if n = 2, a representative τ of the cohomology class that assumes for both signs the negative is given in equation (A.1). For n odd however, the sign of R does not influence the isomorphism class of the twist. This is not very surprising from a representation- theoretic perspective. Indeed, if we redefine S := −R in the group algebra φCτ G then we get the group algebra with the twist chosen such that Sn = −1 and T has the same square as before. Note that this would not work for even n; we could have defined S := iR, but then S would not commute with T . However, if there would have been no time-reversal symmetry, this argument would have worked and the sign of Rn does not matter for the cohomology class. This simply resonates the fact that H 2(Zn, U (1)) = 0 in case of trivial φ. Conclusively, assuming that Rn = −1 in class A would not influence the classification of topological insulators and assuming that Rn = 1 in class AII would not influence the classification in case n is odd. The group cocycle τ can be used as a twisting for the Freed-Moore K-theory groups. The G-equivariant K-theory of the Brillouin zone torus twisted by τ then classifies topological phases protected by the twisted symmetry group (G, φ, τ ). In the abstract language of Freed 39 and Moore, to twist a G-space X means that we consider the following φ-twisted extension of the action groupoid (or orbifold) X//G. The line bundle is picked trivial and the cocycle on X//G is picked equal to the cocycle τ at every x ∈ X, see also [21, section 2]. Twisted equivariant bundles are then the bundle-theoretic analogue of (φ, τ )-twisted representations of G in the same sense that Atiyah & Segal's complex equivariant vector bundles are the bundle-theoretic analogue of ordinary group representations of G. More concretely, we define a (φ, τ )-twisted equivariant vector bundle over a G-space X to be a complex vector bundle E over X together with a family {ρ(g) : E → E : g ∈ G} of maps such that (i) ρ(g) covers the action of g on the base space; (ii) ρ(g) is complex linear if φ(g) = 1 and conjugate linear if φ(g) = −1; (iii) ρ(g)ρ(h) = τ (g, h)ρ(gh). The K-theory classifying such bundles is simply the Grothendieck completion of the monoid of isomorphism classes of such bundles. Written out in full, the resulting abelian group φK τ G(X) is called the (φ, τ )-twisted G-equivariant K-theory of X. Twisted equivariant K-theory of this form can be expanded to a contravariant functor from a category of sufficiently nice G- spaces to the category of abelian groups. By mimicking the technique of Segal [29], it can be shown that this theory extends to a Z-graded additive generalized equivariant cohomology theory in the sense of Bredon [22]. One can also show that the twisted equivariant K- theory above is equivalent to the K-theory of Freed & Moore (with our specific form of the twist τ ) under the correspondences described above c.f. [21]. The fact that Freed & Moore's K-theory satisfies the axioms desired for a cohomology theory of this kind also follows from [21]. Hence we can use the equivariant form of the usual cohomology axioms (suspension axiom, homotopy invariance, etc.) freely. The associated reduced cohomology theory is called reduced twisted equivariant K-theory. For a pointed G-space (X, x), where x ∈ X is a point that is completely fixed under the action, the reduced K-theory is defined as the kernel of the map given by restricting to the fiber over the base point: If Y ⊆ X is a subspace closed under the G-action, then the relative twisted equivariant G (X) →φ K τ +p G (x)(cid:17) . (A.6) G (X) := ker(cid:16)φK τ +p φeK τ +p 40 K-theory of the pair (X, Y ) is defined as φK τ +p G (X, Y ) := φeK τ +p G (X/Y ) (A.7) It should also be noted that that the K-theory twisted by a cocycle τ ′ cohomologous to τ is isomorphic to the K-theory twisted by τ . Finally we remark on how to formally construct the necessary τ for classifying class AII crystalline topological insulator. We do this by considering the double cover group sketched in Section 4.1. Before we can do this, we first have to sketch the relation between double cover groups and twists τ . In order to show this, suppose we have a group of the form G = G0 × ZT 2 with φ projection onto the second factor and G0 ֒→ O(d) a point group. As described using group extensions in Section 4.1, we start with the class in H 2(O(d), Z2) corresponding to the Pin−-double cover of O(d). Restricting the class to the subgroup G0 ⊆ O(d) then gives a class [τ1] ∈ H 2(G0, Z2), which exactly corresponds to the double cover group bG0. To get the right square of time-reversal, we then take the class [τ2] ∈ H 2(ZT 2 , Z2) = Z2 to be trivial in class AI and nontrivial in class AII. Next, in order to construct the desired total double cover group which covers G instead of G0, we now combine [τ1] and [τ2]. To do this we pull the classes back along the two projection maps p1 : G → G0 and p2 : G → ZT 2 and then take their product: [τ ′] = p∗ 1([τ1]) · p∗ 2([τ2]) ∈ H 2(G, Z2). (A.8) Note that this product is not the cup product, but simply the product on the level of the coefficients Z2. Finally, we get our desired twist [τ ] ∈ H 2(G, U (1)φ) by extending the coeffi- cients to U (1). More precisely, we consider the map H 2(G, Z2) → H 2(G, U (1)φ) induced by the G-module injection Z2 ֒→ U (1)φ. More generally, we could have taken group cocycles τ ∈ Z 2(G, C(X, U (1)φ)) that vary over the Brillouin zone X to account for nonsymmorphic crystal structures. These more general twists can be used as a φ-twisted extension of the action groupoid X//G in the Freed- Moore framework by picking the line bundle to be trivial and the cocycle to be equal to τ , now varying over space. This construction results in a well-defined K-theory group that classifies nonsymmorphic crystalline topological insulators. However, we can no longer define the relative K-theory of (X, Y ) to be the reduced K-theory of the quotient if τ varies along X p−1. For a definition of relative K-theory that holds in a more general setting, see [21]. 41 A.2 Higher twisted representation rings and the K-theory of a point Assume we are given a finite group G, a homomorphism φ : G → Z2 and a group 2-cocycle with values in the G-module g·eiθ = eiφ(g)θ. The basic building blocks of K-theory from which spectral sequences can be built are the higher degree K-theory groups of a point φK τ −q G (pt). It follows from the last section that for q = 0, the group φK τ G(pt) is the Grothendieck completion of the monoid of isomorphism classes of modules over the group algebra φCτ G twisted by τ . This is what is called the twisted representation ring of degree q = 0 in the main text. Remark that due to the twist τ , these abelian groups do not have a ring structure, but we will nevertheless refer to them as twisted representation rings in analogy with the nontwisted case. In higher degrees, the twisted equivariant K-theory of a point has a similar concrete description in terms of representation theory and Clifford algebras. This is expressed by using a generalized version of the Atiyah-Bott-Shapiro isomorphism [39]. Heuristically, this theorem asserts that going down a degree in K-theory (i.e. taking a suspension) corresponds to adding a Clifford algebra element on the algebraic level, at least for the K-theory of a point. More explicitly, let Clp,q for the real Clifford algebra of signature (p, q). Then Clp,q is an algebra over the real numbers with has a natural Z2-grading such that the standard generators γ1, . . . , γp+q ∈ Clp,q are odd. If we then take group elements to be even, we can give the tensor product φCτ G ⊗R Cl0,q the structure of a Z2-graded algebra as well. The (φ, τ )-twisted equivariant K-theory of a point can then be described by the isomorphism φK τ −q G (pt) ∼= φRτ −q(G) := {Z2-graded modules over φCτ G ⊗R Cl0,q} {modules that extend to a φCτ G ⊗R Cl1,q-module} , (A.9) and we call φRτ −q(G) the (φ, τ )-twisted (higher) representation ring of G in degree −q. This fact follows from the discussion in [21, §3.5]. This way of computing the complex and real equivariant K-theory of a point has been known for a long time, see the final section of [40]. Also see Donovan-Karoubi [41, §6.15] for a precise version of such a statement for a certain type of twisted equivariant K-theory of more general spaces. We again stress that the twisted higher representation rings as defined above are not rings themselves, because the tensor product of two (φ, τ )-twisted representations is a (φ, 2τ )-twisted representation. However, they are modules over the (φ, 1)-twisted representation ring. We could in theory make them into genuine rings by summing over all possible twists τ , but this would not be a very natural thing to do from the perspective of physics. Namely, using the language of Section 4.1, the product of two fermionic representations will be a bosonic one. To illustrate the definition of the twisted representation ring, we provide a few examples. 42 First of all, if φ and τ are trivial, it can be shown that the higher representation rings are equal to the representation ring of the group in even degree and vanish in odd degree, i.e. it gives the complex equivariant K-theory of a point as expected. For certain simple groups like the time-reversal group G = Z2 = {1, T } (with φ(T ) = −1), the twisted representation rings are also readily computed using basic Clifford algebra theory. They are simply the real (respectively quaternionic) K-theory of a point for a trivial (respectively nontrivial) twist τ . This results in the relevant representation rings for both class AI with trivial twist τ0 and class AII with twist τ1, as was summarized in Table 1. If we just take the real group algebra instead of the twisted group algebra, we get the equivariant KR-theory of a point as can be shown by comparing with the results in [40]. It is also worth noting that the higher twisted representation ring defined by (A.9) indeed equals the Grothendieck group of the monoid consisting of isomorphism classes of (φ, τ )-twisted representations of G in case q = 0. The higher representation rings of the twisted group algebra in general can be computed by decomposing the algebra into a direct sum of matrix rings over R, C and H. Since the twisted group algebra is semisimple, this can always be done. If the number of real, complex and quaternionic matrix rings occurring in this decomposition are nR, nC and nC respectively, then φRτ −q(G) = K −q(pt)nC ⊕ KR−q(pt)nR ⊕ KR−q−4(pt)nH. (A.10) This follows because the representation rings preserve direct sums and are independent of Morita equivalence. Therefore, to determine the representation rings, we only need to make an analogous twisted representation theory to the theory of real representations of finite groups. Such a theory already exists in the physics literature and is described in Section 4.1. A.3 Construction of the spectral sequence In this section some details on the existence of the spectral sequence will be outlined. For an introduction to the theory used in this section and throughout the paper, such as spec- tral sequences in general and the Atiyah-Hirzebruch spectral sequence for (nonequivariant) cohomology theories such as ordinary K-theory, we refer the reader to [42]. The construc- tion of the spectral sequence for the type of K-theory we are concerned with can be done using standard methods already described for general equivariant cohomology theories by Bredon [22]. As before, let G be a finite group, φ : G → Z2 a homomorphism and τ ∈ Z 2(G, U (1)φ) a group cocycle. We now start with a finite G-CW complex X 0 ⊆ · · · ⊆ X d = X and construct the Atiyah-Hirzebruch spectral sequence with respect to this G-CW-structure. In order to 43 make sure that the reduced K-theory of X is defined, we have to assume that there is at least one point 0 ∈ X 0 that is fixed by the whole group G. Note that this is always the case in practice; if X is the Brillouin zone torus then the point k = 0 is fixed by all symmetries. As stated in [22], the first page Ep,−q groups of the spectral sequence is given by the relative K-theory 1 Ep,−q 1 = φK τ +p−q G (X p, X p−1). (A.11) Let us now show that this is the group of Bredon equivariant cochains with a particular coefficient functor. Indeed first note that since τ is constant in space, the relative K-theory equals reduced K-theory of the quotient: (A.12) (A.13) φK τ +p−q G G (X p, X p−1) = φeKτ +p−q ∼= φeKτ +p−q G (X p/X p−1) _σ _gHσ Sp gHσ . Here the first wedge product is over all equivariant p-cells σ of X, which look like Sp σ × G/Hσ, where Hσ ⊆ G is the stabilizer of the cell σ. The second wedge product is over all ordinary cells Sp contained in the equivariant cell σ, i.e. over all cosets gHσ of the stabilizer group. The appearance of the wedge sums is a consequence of the quotient X p/X p−1. For example, if X = S2 is the sphere of figure 1 and p = 2, then X p/X p−1 is the space that results from pinching ℓ and T ℓ to a point, i.e. S2 ∨ S2. gHσ Now by additivity and the suspension axiom, φK τ +p−q G G Sp Sp gHσ _σ _gHσ gHσ _gHσ φeK τ +p−q φeK τ +p−q φeK τ −q G (G/Hσ ⊔ pt) G φK τ −q G (G/Hσ). G (X p, X p−1) = φeKτ +p−q ∼= Mσ p-cell = Mσ p-cell ∼= Mσ p-cell ∼= Mσ p-cell (Σp(G/Hσ ⊔ pt)) (A.14) (A.15) (A.16) (A.17) (A.18) Here ΣpY denotes the pth reduced suspension of the pointed G-space Y and ⊔pt is the disjoint union with an extra added basepoint. Similarly to nontwisted equivariant K-theory 44 we can then make use of the isomorphism φK τ −q G (G/H) ∼= φK τ −q H (pt) (A.19) induced by restricting bundles over G/H to the fiber lying over the trivial coset H. Note that on the right hand side, we have to restrict the twisting data φ, τ to H, but this is omitted in the notation. The first page of the spectral sequence can now be rewritten as a group of Bredon equiv- ariant cochains Mσ p-cell φK τ −q G (G/Hσ) ∼= Mσ p-cell ∼= Mσ p-cell φK τ −q Hσ (pt) φRτ −q(Hσ) ∼= C p G(X, φRτ −q G ), (A.20) (A.21) (A.22) where the coefficients φRτ −q G form a functor from the orbit category of G to the category of abelian groups. Topologically, this functor is just the restriction of the twisted equivariant K-theory functor to the orbit category, but an algebraic desciption is more enlightening. It sends the orbit space G/Gσ to the twisted representation ring φRτ −q(Gσ) defined in the previous section: φRτ −q G (G/Gσ) = φRτ −q(Gσ). (A.23) Thus although our cochains appear to have coefficients in a functor, once evaluated for specific cells in the CW complex of X, the coefficients are just the degree −q twisted representation ring of the stabilizer group of that cell. However, its action on morphisms is more complicated to describe algebraically. Quotient maps G/H → G/K are sent to restrictions of represen- tations φRτ −q(K) → φRτ −q(H) as expected, but conjugation maps G/H → G/gHg−1 can yield nontrivial results similar to the action of T as described around equation (3.15). As stated in the main text, it can be shown that the first differential is precisely the cellular Bredon differential d (df )(σ) = Xµ∈C p(X) [µ : σ]f (µ)Gσ , (A.24) see for example Bredon's work [22]. Summarizing, there exists a spectral sequence Ep,−q r associated to a finite pointed G-CW- 45 complex X converging to twisted equivariant K-theory: Ep,−q r =⇒ φK τ +p−q G (X) (A.25) is Bredon cohomology of degree p with coefficient functor G . To derive explicit computational tools from this fact, recall from the basic theory of such that the second page Ep,q 2 φRτ −q spectral sequences that this means that there is a filtration F p of φK τ G(X) 0 = F d+1 ⊆ F d ⊆ · · · ⊆ F 1 ⊆ F 0 = φK τ G(X) (A.26) such that the final page Ep,−p ∞ forms the associated graded space, i.e. Ep,−p ∞ ∼= F p F p+1 . (A.27) If d = 2, this results in the short exact sequences (3.5) and (3.6), where F = F 1. For computations for nonsymmorphic crystals, we would need to generalize these methods to include twisting data for K-theory that varies over X. Since τ is no longer constant in this setting, most arguments given in this section fail. First of all, the relative K-theory of (X p, X p−1) no longer seems to be equal to the K-theory of the quotient X p/X p−1 if τ varies along X p−1, so we have to preserve the information of the values of τ over X p−1. Moreover, it is not clear how to compute the twisted equivariant K-theory of a sphere for nonconstant twist, since there is no obvious generalization of the isomorphism (A.9) here. Ignoring these mathematical difficulties for the moment and assuming we have some kind of spectral sequence, we should at least arrive at a point where the coefficient functors φRτ −q are no longer constant. In particular, a Bredon cochain of say degree 1 could map different G points of a single 1-cell into different twisted representation rings. The second page is no longer ordinary Bredon cohomology and therefore it is not clear how to generalize this section to that setting. References [1] P. Horava, Stability of fermi surfaces and K-theory, Phys. Rev. Lett. 95, 016405, 2005. [2] A. Kitaev, Periodic table for topological insulators and superconductors, AIP Conference Proceedings 1134, 22 -- 30, 2009. [3] A. Altland and M. R. Zirnbauer, Nonstandard symmetry classes in mesoscopic normal-superconducting hybrid structures, Physical Review B 55, 1142, 1997. 46 [4] D. S. Freed and G. W. Moore, Twisted equivariant matter, Annales Henri Poincar´e 14, 1927 -- 2023, 2013. [5] J. Kruthoff, J. de Boer, J. van Wezel, C. L. Kane and R.-J. Slager, Topological classification of crystalline insulators through band structure combinatorics, Phys. Rev. X 7, 041069, 2017. [6] H. C. Po, A. Vishwanath and H. Watanabe, Symmetry-based indicators of band topology in the 230 space groups, 8, Nature Communications, 2017. [7] S. Ryu, A. P. Schnyder, A. Furusaki and A. W. W. Ludwig, Topological insulators and superconductors: tenfold way and dimensional hierarchy, New Journal of Physics 12, 065010, 2010. [8] G. De Nittis and K. Gomi, Classification of "quaternionic" Bloch-bundles, Communications in Mathematical Physics 339, 1 -- 55, 2015. [9] J. Kruthoff, J. de Boer and J. van Wezel, Topology in time-reversal symmetric crystals, 2017, [arXiv:1711.04769 [cond-mat.str-el]]. [10] K. Shiozaki, M. Sato and K. Gomi, Atiyah-Hirzebruch Spectral Sequence in Band Topology: General Formalism and Topological Invariants for 230 Space Groups, 2018, [arXiv:1802.06694 [cond-mat.str-el]]. [11] K. Shiozaki, C. Z. Xiong and K. Gomi, Generalized homology and Atiyah-Hirzebruch spectral sequence in crystalline symmetry protected topological phenomena, 2018, [arXiv:1810.00801 [cond-mat.str-el]]. [12] C.-K. Chiu, H. Yao and S. Ryu, Classification of topological insulators and superconductors in the presence of reflection symmetry, Phys. Rev. B 88, 075142, 2013. [13] K. Shiozaki and M. Sato, Topology of crystalline insulators and superconductors, Phys. Rev. B 90, 165114, 2014. [14] T. Morimoto and A. Furusaki, Topological classification with additional symmetries from clifford algebras, Phys. Rev. B 88, 125129, 2013. [15] M. F. Atiyah, K-theory and reality, The Quarterly Journal of Mathematics 17, 367 -- 386, 1966. 47 [16] L. Fu, C. L. Kane and E. J. Mele, Topological insulators in three dimensions, Phys. Rev. Lett. 98, 106803, 2007. [17] C. L. Kane and E. J. Mele, Quantum spin Hall effect in graphene, Phys. Rev. Lett. 95, 226801, 2005. [18] W. Luck and B. Oliver, Chern characters for the equivariant K-theory of proper G-CW-complexes, in Cohomological methods in homotopy theory, pp. 217 -- 247. Springer, 2001. [19] F. Hirzebruch and T. Hofer, On the Euler number of an orbifold, Mathematische Annalen 286, 255 -- 260, 1990. [20] A. Adem and Y.-b. Ruan, Twisted orbifold K theory, Commun. Math. Phys. 237, 533 -- 556, 2003, [arXiv:math/0107168 [math-at]]. [21] K. Gomi, Freed-moore K-theory, arXiv preprint arXiv:1705.09134, 2017. [22] G. E. Bredon, Equivariant cohomology theories, vol. 34. Springer, 2006. [23] A. Hatcher, Algebraic topology. 2002, 606, Cambridge UP, Cambridge, 2002. [24] A. Lau, J. van den Brink and C. Ortix, Topological mirror insulators in one dimension, Phys. Rev. B 94, 165164, 2016. [25] C. Bradley and A. Cracknell, The mathematical theory of symmetry in solids: representation theory for point groups and space groups. Clarendon Press, 1972. [26] L. Fu, Topological crystalline insulators, Phys. Rev. Lett. 106, 106802, 2011. [27] C. Dwyer, Twisted equivariant K-theory for proper actions of discrete groups, K-Theory 38, 95 -- 111, 2008. [28] N. B´arcenas, J. Espinoza, B. Uribe and M. Vel´asquez, Segal's spectral sequence in twisted equivariant K-theory for proper and discrete actions, Proceedings of the Edinburgh Mathematical Society 1 -- 30, 2018. [29] G. Segal, Equivariant K-theory, Publications math´ematiques de l'IH´ES 34, 129 -- 151, 1968. [30] E. A. McAlister, Noncommutative CW-complexes arising from crystallographic groups and their K-theory. PhD thesis, University of Colorado, 2005. 48 [31] J.-H. Cho, A survey on equivariant K-theory of representation spheres, Trends in Mathematics 4, 84 -- 89, 2001. [32] M. Atiyah and G. Segal, Twisted K-theory, arXiv preprint math/0407054, 2004. [33] J. F. Adams and S. Priddy, Uniqueness of BSO, in Mathematical Proceedings of the Cambridge Philosophical Society, vol. 80, pp. 475 -- 509, Cambridge University Press, 1976. [34] K. Gomi and G. C. Thiang, Crystallographic T-duality, 2018, [arXiv:1806.11385 [hep-th]]. [35] E. Witten, D-branes and K theory, JHEP 12, 019, 1998, [arXiv:hep-th/9810188 [hep-th]]. [36] M. F. Atiyah and G. B. Segal, The Index of Elliptic Operators: II, Annals of Mathematics 87, 531 -- 545, 1968. [37] J. Distler, D. S. Freed and G. W. Moore, Orientifold Precis, 2009, [arXiv:0906.0795 [hep-th]]. [38] G. W. Moore, "Quantum symmetries and K-theory." http://www.physics.rutgers.edu/~gmoore/PiTP-LecturesA.pdf, 2015. [39] M. F. Atiyah, R. Bott and A. Shapiro, Clifford modules, Topology 3, 3 -- 38, 1964. [40] M. F. Atiyah and G. B. Segal, Equivariant K-theory and completion, J. Differential Geometry 3, 9, 1969. [41] P. Donovan and M. Karoubi, Graded brauer groups and K-theory with local coefficients, Publications Math´ematiques de l'Institut des Hautes ´Etudes Scientifiques 38, 5 -- 25, 1970. [42] P. J. Hilton, General cohomology theory and K-theory, vol. 1. Cambridge University Press, 1971. 49 

1004.0787

1

1004

On the Phase Boundaries of the Integer Quantum Hall Effect. II

It is shown that the statements about the observation of the transitions between the insulating phase and the integer quantum Hall effect phases with the quantized Hall conductivity $\sigma_{xy}^{q}$ $\geq 3e^{2}/h$ made in a number of works are unjustified. In these works, the crossing points of the magnetic field dependences of the diagonal resistivity at different temperatures at $\omega_{c}\tau \approx 1$ have been misidentified as the critical points of the phase transitions. In fact, these crossing points are due to the sign change of the derivative $d\rho_{xx}/dT$ owing to the quantum corrections to the conductivity.

cond-mat.mes-hall

cond-mat

On the Phase Boundaries of the Integer Quantum Hall Effect. II Institute of Solid State Physics RAS, 142432, Chernogolovka, Moscow District, Russia S. S. Murzin It is shown that the statements about the observation of the transitions between the insulating xy ≥ 3e2/h phase and the integer quantum Hall effect phases with the quantized Hall conductivity σq made in a number of works are unjustified. In these works, the crossing points of the magnetic field dependences of the diagonal resistivity ρxx at different temperatures T at ω cτ ≈ 1 have been misidentified as the critical points of the phase transitions. In fact, these crossing points are due to the sign change of the derivative dρxx/dT owing to the quantum corrections to the conductivity. Here, ω c is the cyclotron frequency, τ is the transport relaxation time. PACS numbers: 73.43.Nq The phase diagram of two dimensional systems in the magnetic field has attracted the attention of both theo- rists and experimentalists for many years. Treating the integer quantum Hall effect (IQHE) in the context of the two parameter scaling theory [1], which is graphically represented as a flow diagram [2, 3], yields the solution of the problem disregarding the electron -- electron inter- action. The further development of the scaling theory showed that the electron -- electron interaction does not affect the position of the IQHE phase boundaries of a spin polarized electron system [4, 5]. According to the scaling theory, the boundary between two IQHE phases is possible only if the quantized val- ues of the Hall conductivity of these two phases differ by e2/h or (in the case of the spin degeneracy of the Landau levels) 2e2/h. However, a number of works re- ported on the observation of the transitions between the insulating phase (σq xy = 0) and the IQHE phases with xy ≥ 3e2/h. Song the quantized Hall conductivity σq [6] reported on the observation of the transition et al. xy = 3e2/h in two dimensional hole systems xy = 0 ↔ σq σq in a strained Ge quantum well. The observation of the xy ≥ 3e2/h in the two dimen- transitions σq sional hole systems in a strained Ge quantum well was also announced in [7]. Lee et al. [8] claimed the observa- tion of the transitions σq xy = xy = 8e2/h in doped AlGaAs/GaAs/AlGaAs quan- 0 ↔ σq tum wells. Huang et al. [9] reported on the observation of the transitions from the state with σq xy = 0 to the states xy = 6 − 16e2/h in GaAs/AlGaAs heterojunctions. with σq In all of these works [6 -- 9], the crossing points of the mag- netic field dependences of the diagonal resistivity ρxx at different temperatures T , at ωcτ ≈ 1 (ωc = eB/m is the cyclotron frequency, τ is the transport relaxation time, and m is the effective electron mass) were considered as the critical points Bc of the phase transitions (see Fig. 1). In this case, ρxx weakly depends on the magnetic field and temperature near Bc. xy = 6e2/h and σq xy = 0 ↔ σq xy = 0 ↔ σq In this work, it is shown that the above statements of the observation of the transitions between the insulat- ing phase and the IQHE phases with the quantized Hall conductivity σq In fact, the crossing point of the magnetic field dependences of the xy ≥ 3e2/h are unjustified. FIG. 1: The diagonal (ρxx) and Hall (ρxy) resistivities of the doped AlGaAs/GaAs/AlGaAs quantum well as a function of the magnetic field. The electron density is Ns = 1.04 × 1016 m−2 [8]. The temperatures for different ρxx- curves are 0.3, 0.5, 0.8, 1.2, 1.7, 2, and 3.2 K. The spin splitting is small and therefore invisible in ρxx and ρxy. The figure is taken from Ref.. diagonal resistivity ρxx(T ) at different temperatures is caused by the sign change of the derivative dρxx/dT ow- ing to the quantum corrections to the conductivity [10]. The classical diagonal and Hall conductivities in the magnetic field take the form σ0 xx = Nse2τ 1 m 1 + (ωcτ )2 (1) 2 ture and approaches the nearest quantized integer value σxy(Bi) = ie2/h with Bi < Bc at ωcτ < 1. Taking into account the quantum corrections, the diag- onal and Hall resistivities of the two dimensional electron system are given by the expressions ρxx(T ) = ρ0 xx + h(cid:0)ρ0 xy(cid:1)2 − (cid:0)ρ0 xx(cid:1)2i ∆σxx(T ) (3) and ρxy(T ) = ρ0 xy − 2ρ0 xxρ0 xy∆σxx(T ) took + (cid:0)σ0 xy, ρ0 Here, we into account that xy = σ0 xy/h(cid:0)σ0 xx(cid:1)2 xx(cid:1)2 xx, σ0 xy(cid:1)2i, ρ0 xx and ρ0 xx/h(cid:0)σ0 xy(cid:1)2i σ0 and σ0 xy are the bare (non- renormalized) values of the conductivity and resis- tivity, which correspond to the diffusion motion of electrons without the interference (localization) effects at distances longer than the diffusion step length. The derivative dρxx/dT changes its sign in the magnetic field B such that ρ0 xy(B). In the classical treatment xx(B) = ρ0 ρ0 At σxy ≫ e2/h and ωcτ < 1, the diagonal resistivity ρxx first increases with a decrease in the temperature, reaching the value xy(B) at ωcτ = 1. xx(B) = ρ0 (4) = ρ0 xx + (cid:0)σ0 t = 1 c ) h / 2 e ( x x 6 4 2 0 1 2 xy (e2/h) 3 4 ρxx,max = 1 2σ0 xy , (5) FIG. 2: Sketch of the scaling flow diagram for the quan- tum well with the parameters given in Fig. 1. The spin splitting is negligible. The solid lines are the separatrices of the diagram. The dashed line shows dependence σxx(σxy) at ωcτ < 1 for the sample with the zero-field bare conductivity σ0 = 7.52e2/h. The dotted lines are the scaling flow lines. The dash -- dotted straight line corresponds to ω cτ = 1. and σ0 xy = Nse2τ m ωcτ 1 + (ωcτ )2 . (2) xx ≪ σ0 where Ns is the electron or hole density. The quan- tum corrections to the diagonal conductivity ∆σxx(T ) = σxx(T ) − σ0 xx decrease with temperature. At T → 0, σxx(T ) → 0 except for the critical points, where σ0 xy = (i + 1/2)e2/h (see Fig. 2). Excluding the weak localization region (B . 1 T), the Hall conductiv- ity σxy is independent of the temperature down to the temperatures at which σxx ∼ e2/h. At lower temper- atures, the Hall conductivity depends on the tempera- and then decreases and vanishes at T → 0 excluding the critical magnetic fields in which σ0 xy = (i + 1/2)e2/h. Thus, the negative value of the derivative dρxx/dT within the experimental range does not imply that the electron system is in an insulating phase. Note that the magnetic field position of the IQHE phases at ωcτ . 1 is not determined by the filling factor xyh/e2. This ν. Rather, it is given by the magnitude of σ0 quantity is different from ν at ωcτ . 1. [11] At ν = 8 in Fig.1 σ0 xxh/e2 = 3.8 at T = 0.3 K. According scaling diagram presented on Fig2 at this ν xy = 4e2/h. quantized values of the Hall conductivity σq xyh/e2 = 3.8, σ0 Thus, we have shown that the statements [6 -- 9] of the observation of the transitions between the insulat- ing phase and the IQHE phases with the quantized Hall conductivity σq xy ≥ 3 are unjustified. In fact, the crossing point of the magnetic field dependences of the diagonal resistivity ρxx(T ) at different temperatures is caused by the sign change of the derivative dρxx/dT at ωcτ ≈ 1 owing to the quantum corrections to the conductivity. This work was supported by the Russian Foundation for Basic Research. [1] H. Levine, S. B. Libby, and A. M. M. Pruisken, Phys. [2] D. E. Khmel'nitskii, Pis'ma Zh. Eksp. Teor. Fiz. 38, 454 Rev. Lett. 51, 1915 (1983). w s s 3 (1983), [JETP Lett. 38, 552 (1984)]; Phys. Lett. A 106, 182 (1984); Helvetica Phys. Acta 65, 164 (1992). [3] A. M. M. Pruisken, in The Quantum Hall Effect, edited by R. E. Prange and S. M. Girvin, Springer-Verlag, 1990. [4] A. M. M. Pruisken, M. A. Baranov, I. S. Burmistrov, Pis'ma v ZhETF 82, 166 (2005) [JETP Lett. 82, 150 (2005)]. [5] A. M. M. Pruisken, I. S. Burmistrov, Pis'ma v ZhETF 87, 252 (2008) [JETP Lett. 87, 220 (2008)]. Xie, Phys. Rev. B 62, 6940 (2000). [8] C. H. Lee, Y. H. Chang, Y. W. Suen, and H. H. Lin, Phys. Rev. B 58, 10629 (1998). [9] C. F. Huang, Y. H. Chang, C. H. Lee, et al., Phys. Rev. B 65, 045303 (2001). [10] B. L. Al'tshuler and A. G. Aronov, in Electron-Electron Interaction in Disordered Systems, edited by A. L. Efros and M. Pollak, North-Holland, Amsterdam, 1987. [11] S.S. Murzin, Pis'ma v ZhETF 89, 347 (2009) [JETP Lett. [6] S.-H. Song, D. Shahar, D. C. Tsui, Y. H. Xie, and Don 89, 298 (2009)]. Monroe, Phys. Rev. Lett. 78, 2200 (1997). [7] M. Hilke, D. Shahar, S. H. Song, D. C. Tsui, and Y. H. 

1610.07425

2

1610

Quantum spin Hall effect in rutile-based oxide multilayers

Dirac points in two-dimensional electronic structures are a source for topological electronic states due to the $\pm \pi$ Berry phase that they sustain. Here we show that two rutile multilayers (namely (WO$_2$)$_2$/(ZrO$_2$)$_n$ and (PtO$_2$)$_2$/(ZrO$_2$)$_n$, where an active bilayer is sandwiched by a thick enough (n=6 is sufficient) band insulating substrate, show semi-metallic Dirac dispersions with a total of four Dirac cones along the $\Gamma-M$ direction. These become gapped upon the introduction of spin-orbit coupling, giving rise to an insulating ground state comprising four edge states. We discuss the origin of the lack of topological protection in terms of the valley spin-Chern numbers and the multiplicity of Dirac points. We show with a model Hamiltonian that mirror-symmetry breaking would be capable of creating a quantum phase transition to a strong topological insulator, with a single Kramers pair per edge.

cond-mat.mes-hall

cond-mat

a Quantum spin Hall effect in rutile-based oxide multilayers J. L. Lado,1 Daniel Guterding,2 Paolo Barone,3, 4 Roser Valent´ı,2 and V. Pardo5, 6 1QuantaLab, International Iberian Nanotechnology Laboratory, Braga, Portugal∗ 2Institut fur Theoretische Physik, Goethe-Universitat Frankfurt, Max-von-Laue-Strasse 1, 60438 Frankfurt am Main, Germany† 3Consiglio Nazionale delle Ricerche (CNR-SPIN), 67100 L'Aquila, Italy 4Graphene Labs, Istituto Italiano di Tecnologia, via Morego 30, 16163 Genova, Italy 5Departamento de F´ısica Aplicada, Universidade de Santiago de Compostela, E-15782 Campus Sur s/n, Santiago de Compostela, Spain 6Instituto de Investigaci´ons Tecnol´oxicas, Universidade de Santiago de Compostela, E-15782 Campus Sur s/n, Santiago de Compostela, Spain Dirac points in two-dimensional electronic structures are a source for topological electronic states due to the ±π Berry phase that they sustain. Here we show that two rutile multilayers (namely (WO2)2/(ZrO2)n and (PtO2)2/(ZrO2)n, where an active bilayer is sandwiched by a thick enough (n=6 is sufficient) band insulating substrate, show semi-metallic Dirac dispersions with a total of four Dirac cones along the Γ − M direction. These become gapped upon the introduction of spin- orbit coupling, giving rise to an insulating ground state comprising four edge states. We discuss the origin of the lack of topological protection in terms of the valley spin-Chern numbers and the multiplicity of Dirac points. We show with a model Hamiltonian that mirror-symmetry breaking would be capable of creating a quantum phase transition to a strong topological insulator, with a single Kramers pair per edge. I. INTRODUCTION Topological states of matter1,2 have been the source of tremendous excitement and have fostered a rich variety of new (and old) ideas in condensed matter physics. Quan- tum Hall effect, quantum spin Hall states,3 topological crystals and topological crystalline insulators are some examples in which the topology of the single-electron Hamiltonian translates into robust electronic transport and surface states, resilient to the typical perturbations that real samples will have, such as defects and im- purities. The different types of quantum Hall effects have provided us with a way to easily measure the quality of samples and even the charge of the electron, whereas quantum spin Hall states realize chiral wires,3 where momentum and spin are coupled together. Fur- thermore, their superconducting analogues, topological superconductors4 are known to give rise to Majorana bound states.5 Such many-body states can show non- abelian braiding properties, making them suitable to be- come building blocks for topological quantum comput- ing. Moreover, the nature of some of these topological states in superconductors still needs to be understood.6 A common way to design topological superconductors is precisely based on a quantum spin Hall state proximized to a conventional superconductor,4 turning quantum spin Hall states into a key ingredient not only in spintronics,7 but also in topological quantum computing. Two main mechanisms need to be understood in or- der to design quantum spin Hall states. The first one corresponds to band inversion, which relies on spin-orbit coupling (SOC) altering the order of the s−p orbital char- acters in a band structure, whose best known realization are HgTe/CdTe quantum wells.8 The modification can occur at the Γ point, so that SOC basically changes the parity of the highest occupied band. The second mecha- nism is the opening of a protected Dirac point by SOC.3 Dirac points host protected ±π Berry phases that be- come ±1/2 Chern numbers upon a gap opening, so that in the presence of time-reversal symmetry a total spin- Chern number gives rise to the quantum spin Hall state, whereas with broken time reversal symmetry a quantum anomalous Hall state is realized. The theoretical quest for new topological insulators has involved semiconductors9,10, transition metal oxides,11 metal organic frameworks12,13 or optical lattices.14 Among them, oxides offer a rich set of possibilities due to the variety of materials that can be synthesized in a chemically stable form, together with the ease of fabrica- tion in low-dimensional forms such as thin films or multi- layers. Predictions of both, quantum spin Hall and quan- tum anomalous Hall effect have appeared in various ox- ides based on perovskites,15 -- 17 pyrochlores,11,18 rutiles,19 corundum20 or doped Kagome structure.21 Many of the proposals rely on an underlying hexagonal lattice, where Dirac points are prone to appear at the K, K(cid:48) corners of the Brillouin zone. Rutiles, however, develop (semi) Dirac points, but due to their tetragonal unit cell, the Dirac points show up at a certain k-point in the Γ − X direction.22 Opening a gap at these points with a time- reversal symmetry breaking has been shown to give rise to a quantum anomalous Hall state19 with a very small gap of about 1 meV. Whether a quantum spin Hall state in a rutile-based structure can be engineered or not in such a way is still an open issue. In this manuscript we address whether a rutile ma- terial with time-reversal symmetry is able to develop a quantum spin Hall state by opening a gap at the Dirac points of its band structure by the SOC effect. For that sake, we have designed two different rutile multi- layers that show Dirac points in their band structure, i.e. (WO2)2/(ZrO2)4 and (PtO2)2/(ZrO2)4. Both het- erostructures become gapped upon introduction of SOC. We show by means of their topological invariants and the calculation of their edge states, that both systems realize a 2D crystalline topological insulating state, character- ized by four in-gap surface states. We discuss the origin of this insulating state in terms of the spin valley Chern numbers, whose multiplicity is determined by the sym- metry of the unit cell. We propose a model Hamiltonian to describe the system and we show that breaking an in- plane mirror symmetry can drive a transition towards a strong topological insulator by compensating two of the Dirac points and leaving just two of them with uncom- pensated spin valley Chern numbers. II. GAPPED DIRAC POINTS IN NON-MAGNETIC RUTILES The multilayers proposed (see Fig. 1a) are based on the rutile structure and grow along the (001) direction with the a, b lattice parameters fixed to those of the band insulating substrate and the c-axis and internal atomic positions fully relaxed. We use as insulating substrate rutile ZrO2 whose bulk lattice parameters were calcu- lated ab initio (see Methods section) yielding a= 4.93 A, which was the used value. The only requisite that the substrate should have is to provide a substantial gap where the d electrons of the active bilayer are allowed to form the Fermi surface without mixing with the states of the substrate. We have tried other substrates, e.g. TiO2 would be a good candidate in terms of the size of its band gap. However, the band bending it introduces will be so large that it would destroy the Dirac points. TiO2 with a slightly enlarged a parameter would be fine for our purposes, or even a very thin layer of ZrO2 on top of TiO2 would also do. WO2 on top of ZrO2 would be almost unstrained (a= 4.86 A), but PtO2 would undergo substantial strain since our calculations yields a= 4.59 A. We have also tried other insulating substrates, such as SnO2 and PbO2, whose lattice matching will be better than ZrO2, but their very narrow gap destroys the Dirac points by mixing substantially with the active d electrons of the bilayer. A. Bulk electronic structure In the absence of SOC, a single crossing takes place along the Γ − M direction in both systems. Inspection of the band dispersion around those points reveals that the low-energy states form a Dirac cone. When SOC is introduced, a gap opens up at the crossing points (Fig. 1c,d), giving rise to a bulk insulating state. The topological properties of the band structure can be characterized by calculating the topological invariant ν associated with a system with time-reversal symmetry, 2 FIG. 1. (a) Sketch of the multilayer structure, based on the rutile unit cell, consisting of a bilayer of WO2 or PtO2 sand- wiched between insulating ZrO2. In the absence of spin-orbit coupling, the low-energy electronic properties are dominated by four Dirac equations located along the Γ − M path, as shown in the sketch (b). When SOC is introduced, the band structure develops a gap for both the W-based (c) and Pt- based (d) multilayers. The solid lines in (c,d) correspond to the DFT band structure, whereas the dashed lines corre- spond to the Wannier interpolation. In (d), the anti-crossing between Γ and X is already present without SOC. by means of the Z2 invariant. In the case of crystals with inversion symmetry, it is known that the topological in- variant can be obtained simply by calculating the parities of the wavefunctions at the time reversal invariant mo- menta (TRIM).23 Nevertheless, the rutile structure pre- sented here does not posses global inversion symmetry, but only in-plane, so that we cannot apply such proce- dure to the whole band structure. To calculate topolog- ical invariants for systems without inversion symmetry several methods have been proposed: (i) direct numerical calculation of the Pfaffian,24 (ii) looking for an obstruc- tion for a time-reversal smooth gauge25 and (iii) Wannier charge center evolution.26,27 We have chosen this latter method, since it is numerically highly stable and spe- cially well suited for electronic structure calculations. It is based on following the flow of the Wannier charge cen- ters as one moves across half of the Brillouin zone. With this scheme, an odd number of crossings of the Wannier charge centers label a non-trivial system, whereas an even number implies a trivial one. From our calculations, we obtain that the charge cen- ters cross an even number of times, implying that there is an even number of Kramers pairs per edge. Never- theless, we observe that the charge centers can be sep- arated in two families, each one showing an odd num- ber of crossings, giving rise to a topological crystalline insulator.27,28 Every family of Wannier centers, whose existence is related to the mirror symmetry of the unit cell, will produce a pair of Kramers edge states, adding up to a total of four edge states. The gapless nature of these edge states is not totally protected against perturbations even when they respect time-reversal symmetry, because perturbations mixing the two families, such as chemical edge reconstruction, will be able to open up an edge gap This topological crystalline insulating phase is analogous to two coupled non-trivial quantum spin Hall states, as in bilayer graphene with spin-orbit coupling.29,30 Impor- tantly, topological insulating states, even if they are not protected against symmetry mixing perturbations, have been shown to be robust31 -- 36 against edge perturbations. B. Edge states Having a bulk gapped spectrum and a crystalline topo- logical invariant, the rutile structures are expected to show edge states when studied in a finite geometry. Us- ing the Wannier Hamiltonian derived from the electronic structure calculations, we calculate the surface spectral function by solving Dyson's equation G(kx, E) = (E − H0 − t†(kx)G(kx, E)t(kx))−1 (1) where H0 is the matrix with the intra-cell matrix elements, kx the vector parallel to the interface and t(kx) the hopping of the Bloch Hamiltonian in the di- rection perpendicular to the interface. From the Green's function, the surface spectral function is calculated as Γ(kx, E) = 1 π Im(G(kx, E)), and shown in Fig. 2 for the W- and Pt- based rutile multilayers. The surface spectral function shows that both systems develop surface states, a total of four of them. The ex- istence of four edge states is coherent with the inter- pretation that the electronic structure is equivalent to two coupled quantum spin Hall insulators. In a topo- logical quantum spin Hall insulator, the number of sur- face states is 2 (or 6,8..), so that the surface hosts a single Kramers pair, whose crossing cannot be avoided due to time-reversal symmetry. In our present case, due to the existence of two Kramers pairs, perturbations can gap out the surface states without breaking time-reversal symmetry. Kramers degeneracy holds at the TRIM (Γ and X), where the two branches of edge states are de- generate in both valence and conduction band. 3 FIG. 2. Edge (01) k-resolved density of states for the W- based (a) and Pt-based multilayer (b). A set of edge states appear within the gap, but due to the lack of a strong topo- logical index, their existence is not protected, developing a small gap. of Wannier charge centers, the origin can be traced back to the additional symmetries of the system, namely mir- ror symmetry, that impose that there are two families of Wannier centers, each one yielding a pair of Kramers edge states. In terms of effective low energy Dirac points, the C4 symmetry imposes that the system has four spin- less low energy Dirac points, yielding twice the number of conventional honeycomb lattices, and twice as many edge states. Therefore, the combination of time rever- sal, C4 and mirror symmetry, and low energy effective Dirac Hamiltonian, imposes that the system will show four edge states. In the following we will try to give an intuitive understanding of why this happens, as well as suggest a situation where a strong topological state can be obtained by modifying the symmetry of our material. III. TOPOLOGICAL INVARIANT IN RUTILES A. Origin of the topological insulating state While the opening of Dirac points is a well known route to engineer quantum spin Hall insulators, we have shown that in the present rutile multilayers quantum spin Hall states are realized that are not fully protected. In terms Starting with the situation without spin-orbit cou- pling, the band structure is characterized by the four Dirac crossings introduced at the beginning. Indepen- dently of whether those crosses are Dirac, anisotropic Dirac or semi-Dirac, the important feature for the present discussion is that they carry a ±π Berry phase. We have verified numerically that for the non-SOC calculations, the Wannier Hamiltonians obtained generate Dirac-like crosses with ±π Berry phase. The Dirac points are lo- cated along the Γ − M path (shown in Fig. 1b), where the position depends on the details of the electronic struc- ture, giving rise to a total of four non-equivalent Dirac points in the full Brillouin zone. In the presence of SOC, a spin-dependent mass term appears in the Hamiltonian. Due to the absence of in- version symmetry, the eigenvalues are not degenerate ex- cept at the TRIM points, where Kramers degeneracy is retained. This is reflected in the different masses for different Kramers states. It has been shown that the Z2 invariant can be related to the so-called spin Chern number, or simply to the Chern number C of one of the Kramers sectors. This can be easily understood in topo- logical states where sz is conserved, and the Z2 invariant is ν = C↑( mod 2) (2) which in the case of monolayer graphene gives CS = 2, C↑ = 1 and ν = 1, while for bilayer graphene it results in CS = 4, C↑ = 2 and ν = 0. In the case of rutile multilayers, due to the existence of four gapped Dirac equations for each Kramers manifold, the Chern number for a certain Kramers manifold will be a sum over the Chern numbers of four Dirac equations Ci, which due to the ±π Berry phase will be Ci = ±1/2. By labeling the Kramers manifold as ↑ in analogy to the spin conserving case, we have 4(cid:88) i=1 C↑ = si 1 2 (3) where si = ±1 depending on the sign of the SOC- induced Dirac masses. Due to the in-plane inversion sym- metry, the Chern number of the different Dirac equations around +(cid:126)k and −(cid:126)k will be the same. We will have that 1 C1 = C3 = s1 2 . Therefore, the Chern number for one of the Kramers sectors will be 1 2 and C2 = C4 = s2 4 single spin flavor is present at the Fermi level, the pre- vious argument will predict that the total Chern num- ber is 2. This has been explicitly calculated in a V- based magnetic rutile multilayer19 that shows four semi- Dirac points along Γ − M , obtaining that the system is an anomalous Hall insulator showing two co-propagating edge states.19 It is interesting to note that, contrary to other oxide systems,37 in this system magnetism brings about the topological protection for the edge states, as time-reversal symmetry breaking leads to protected topo- logical states. The crystalline topological state in the time rever- sal Pt- and W- multilayers is therefore a consequence of the existence of a total of four Dirac points in the non-relativistic band structure. This compares with the case of honeycomb lattices, where usually only two Dirac equations are present, and the Chern number spin fla- vor is C↑,↓ = ±1. In these rutile-based nanostructures, C4 symmetry imposes that a total of 4 Dirac equations appear between Γ and M . If such symmetry is broken, it could be possible to have an electronic structure with only two Dirac equations that could give rise to a strong topological insulating phase. A way to realize that will be to expand the cell along the (11) direction. With such distortion, two of the Dirac points will displace differently than the other in k-space. In particular, they can open up a trivial gap, whereas the other two remain gapless (or with a much smaller gap). In that situation, switching on SOC might be able to create band inversion in two of the Dirac points, but not in the other two, giving rise to C↑ = 1 and robust quantum spin Hall state ν = 1. B. Model Hamiltonian for spinless fermions In the following we will illustrate the different topologi- cal states of the rutile lattice by means of a model Hamil- tonian. We stress that the following model Hamiltonians do not intend to precisely capture the electronic struc- ture of the two multilayers presented above, their goal is just to provide an intuitive understanding of the topol- ogy of the low-energy electronic structure. We choose a variation from a model previously shown to develop anisotropic Dirac points38 in a similar system. (cid:18)
1((cid:126)k) T ((cid:126)k) T ∗((cid:126)k)
2((cid:126)k) (cid:19) C↑ = s1 + s2 = 0,±2 (4) H↑((cid:126)k) = (5) with With the previous Kramers Chern number, the topo- logical invariant will result into ν = 0, so that the system will not be a strong topological insulator in any case. For the case of C↑ = +2, time reversal symmetry (C = 0) imposes that the Chern number of the other sector is C↓ = −2, giving a total CS = 4, predicting four edge states. This last situation is precisely the one we ob- tained when calculating the edge spectra of the Pt- and W- multilayers. We would like to emphasize that in the case of hav- ing a similar system but with magnetic order, so that a
1(kx, ky) = −
+ 2t(cos kx + cos ky)
2(kx, ky) =
− 2t(cos kx + cos ky) T ((cid:126)k) = V ((cid:126)k) + W ((cid:126)k) + Z((cid:126)k) V (kx, ky) = 2t(cid:48)(cos kx − cos ky) W (kx, ky) = 2iα sin(kx − ky) Z(kx, ky) = 2iλ sin kx sin ky 5 instead of the anisotropic Dirac dispersion shown in Eq. 5. If we first switch on the mirror-symmetry-breaking term W , the spectrum consists of two gapped Dirac equa- tions in the (1,-1) direction with opposite Chern numbers, and two gapless Dirac equations in the (1,1) direction. If now time-reversal symmetry is broken by switching on Z (λ (cid:54)= 0), and provided λ is not large enough to invert the gaps in (1,-1), the Dirac equation in the (1,1) will open up a gap with the same Chern number. The band structure for this situation is shown in Fig. 3d, where it is observed that the calculated Berry curvatures (Fig. 3e) and Chern number (Fig. 3f) are in agreement with the previous argument. Therefore, the model proposed in Eq. 5 shows a phase with Chern number C = 1, provided mirror symmetry is broken. The previous phenomenology can be understood by expanding the Hamiltonian around the Dirac points. Hκ = −σzp1 + κ1κ2σxp2 − σymκ (6) where mκ is a mass term induced by W and Z, p1 = κ1px + κ2py, p2 = −κ2px + κ1py, with px,y the crystal momenta around the different valleys. The Chern num- ber for the previous valley Hamiltonian can be written as 1 2 Cκ = sign(κ1κ2mκ) (7) In the case of λ (cid:54)= 0 and α = 0, we have mκ = κ1κ2m so Cκ = 1/2, and when summing over the four valleys it gives C = 2. In the case of α (cid:54)= 0, two of the valley Chern numbers can yield opposite signs, so that the net Chern number will be C = 1. C. Model Hamiltonian for quantum spin Hall So far we have been studying a model for spinless fermions, showing that it leads to two different quan- tum anomalous Hall states. With the previous model a spinful time-reversal version can be built as (cid:18)H↑((cid:126)k) (cid:19) (cid:32) (cid:33) H((cid:126)k) = 0 = 0 H↓((cid:126)k) H↑((cid:126)k) 0 H∗ 0 ↑ (−(cid:126)k) (8) With the previous ansatz the net Chern number is zero as imposed by time reversal, and the parameter λ can be now understood as arising from SOC. For α = 0, follow- ing the discussion in Sec. III A, the spin Chern number yields CS = 4, giving rise to a crystalline topological insulator (Fig. 4a), compatible with the DFT results presented in Fig. 1. Switching on the mirror symmetry term α, the spin Chern number yields CS = 2, giving rise to a strong topological insulator (Fig. 4b). Therefore, a FIG. 3. Band structure (a) along the (1,1) direction for the model in Eq. 5, in the quantum anomalous (C = 2) regime (α = 0,λ (cid:54)= 0), Berry curvature along the path (b) and in the whole Brillouin zone (c). Dashed lines in (a) show the bands for (α = λ = 0), the gapless regime. Band structure (d), and Berry curvatures (e,f) in the QAH regime with α (cid:54)= 0, λ (cid:54)= 0, with total Chern number C = 1. Parameters are
= −2, t = 1, t(cid:48) = 2,λ = 0.5 and α = 2 The terms
1,
2, V correspond to a simplified version of anisotropic Dirac points.38 The term W breaks the equivalence between the (1, 1) and (1,−1) directions but conserves time reversal. The term Z breaks time-reversal symmetry, but maintains the equivalence between the (1,1) and (1,-1) directions. In the case of Z = W = 0, the previous Hamiltonian describes a band structure that shows four anisotropic Dirac points, located along the Γ − M directions. The Dirac nodes are equally spaced from Γ and occur at points (cid:126)k0 = (κ1, κ2)k0, with κ1,2 = ±1 We first focus on the case where W = 0 (α = 0), but with a non-zero Z (λ (cid:54)= 0) so that time-reversal symmetry is broken. In this situation, the band structure develops a gap (Fig. 3a), and the split Dirac points generate a non- zero Berry curvature (Fig. 3b). When integrated over the whole Brillouin zone, a Chern number C = 2 is obtained. This situation is analogous to the one observed19 in half- metallic V-based multilayers with SOC, a system that is also a QAH with C = 2, but with the difference that the low-energy electronic structure is of type II semi-Dirac19 6 with C = 2, breaking in-plane mirror symmetry will al- low to enter a state with C = 1. The extension of the previous model with mirror symmetry breaking to the spinful time-reversal case would give rise to a strong 2D topological insulator. ACKNOWLEDGMENTS J.L.L. acknowledges financial support from Marie- Curie-ITN 607904 SPINOGRAPH. D.G and R.V. thank the Deutsche Forschungsgemeinschaft for funding through SFB/TR 49. V.P. thanks the Xunta de Galicia for financial support under the Emerxentes Program via Project No. EM2013/037 and the MINECO via Project No. MAT2013-44673-R. V.P. acknowledges support from the MINECO of Spain via the Ramon y Cajal program under Grant No. RyC-2011-09024. P.B. acknowledges partial support from the European Unions Horizon 2020 research and innovation programme under Grant Agree- ment No. 696656 GrapheneCore. We thank GEFES2016 for providing the platform for this collaboration to suc- ceed. Appendix A: Methods We have carried out density functional theory (DFT) calculations with various codes: WIEN2k,39 and Quan- tum Espresso40 for the various cell and geometry relax- ations plus the electronic structure analysis (both codes yielding comparable results), and using the code FPLO41 for relativistic Wannierization calculations. Structural relaxations using both WIEN2k and Quantum Espresso were carried out with the PBE version of the general- ized gradient approximation42 as an exchange-correlation functional, using PAW pseudopotentials in the QE case and a full-potential calculation with WIEN2k, both with- out SOC. The construction of Wannier functions was per- formed within the full-relativistic version of FPLO us- ing a 6 × 6 × 3 k-point grid. For the (WO2)2/(ZrO2)n case we included all Zr 4d and W 5d states. For the (PtO2)2/(ZrO2)n case we included all Zr 4d, O 2p and Pt 5d states. FIG. 4. Sketch of the edge states for the Hamiltonian Eq. 8, showing the crystalline insulating state with spin Chern number CS = 4 (a), and the strong phase with CS = 2. The transition from (a) to (b) is driven by the mirror symmetry parameter α. In a real material, such transition can be in- duced by strain in the (11) direction. The DFT results of Fig. 2 correspond to the symmetric case (a). mirror-symmetry breaking term in our model Hamilto- nian is capable of creating a quantum phase transition from the original crystalline topological insulating state to a strong topological insulator. We finally clarify that in the real materials introduced in this manuscript, spin mixing terms would show up in their effective Hamilto- nian. IV. CONCLUSIONS We have shown that two rutile-based bilayers formed by an active 5d-electron system with 5d2 and 5d6 elec- tron count host a crystalline topological insulating state. The origin of the non robust topological state comes from having four Dirac equations in the absence of SOC, in comparison with honeycomb lattices (whether this is graphene or oxide-based) that show only two. This lim- itation can be traced to the Chern number per Kramers sector, that in the rutile structure is 2 whereas in the honeycomb lattice is 1. We have suggested that by re- moving undesired Dirac points, it would be possible to design strong quantum spin Hall insulators in the rutile lattice. We have shown with a toy model calculation for spinless fermions that apart from the insulating phase ∗ jose.luis.lado@gmail.com † guterding@itp.uni-frankfurt.de 1 M. Z. Hasan and C. L. Kane, "Colloquium : Topological insulators," Rev. Mod. Phys., vol. 82, pp. 3045 -- 3067, Nov 2010. 2 X.-L. Qi and S.-C. Zhang, "Topological insulators and su- perconductors," Rev. Mod. Phys., vol. 83, pp. 1057 -- 1110, Oct 2011. 3 C. L. Kane and E. J. Mele, "Quantum spin hall effect in graphene," Phys. Rev. Lett., vol. 95, p. 226801, Nov 2005. 4 L. Fu and C. L. Kane, "Superconducting proximity effect and majorana fermions at the surface of a topological in- sulator," Phys. Rev. Lett., vol. 100, p. 096407, Mar 2008. 5 S. R. Elliott and M. Franz, "Colloquium : Majorana fermions in nuclear, particle, and solid-state physics," Rev. Mod. Phys., vol. 87, pp. 137 -- 163, Feb 2015. 6 Z. Wang, H. Zhang, D. Liu, C. Liu, C. Tang, C. Song, Y. Zhong, J. Peng, F. Li, C. Nie, et al., "Topological edge states in a high-temperature superconductor fese/srtio3 (001) film," Nature Materials, 2016. 7 J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, "Spin hall effects," Rev. Mod. Phys., vol. 87, pp. 1213 -- 1260, Oct 2015. 8 B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, "Quantum spin hall effect and topological phase transition in hgte quantum wells," Science, vol. 314, no. 5806, pp. 1757 -- 1761, 2006. 9 M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, "Quantum spin hall insulator state in hgte quantum wells," Science, vol. 318, no. 5851, pp. 766 -- 770, 2007. 10 H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.-C. Zhang, "Topological insulators in bi2se3, bi2te3 and sb2te3 with a single dirac cone on the surface," Nature physics, vol. 5, no. 6, pp. 438 -- 442, 2009. 11 B.-J. Yang and Y. B. Kim, "Topological insulators and metal-insulator transition in the pyrochlore iridates," Phys. Rev. B, vol. 82, p. 085111, Aug 2010. 12 Z. Wang, N. Su, and F. Liu, "Prediction of a two- dimensional organic topological insulator," Nano letters, vol. 13, no. 6, pp. 2842 -- 2845, 2013. 13 L. Wei, X. Zhang, and M. Zhao, "Spin-polarized dirac cones and topological nontriviality in a metal -- organic framework ni 2 c 24 s 6 h 12," Physical Chemistry Chemical Physics, vol. 18, no. 11, pp. 8059 -- 8064, 2016. 14 X. Li, E. Zhao, and W. V. Liu, "Topological states in lattice containing ultracold atoms a ladder-like optical in higher orbital bands," Nature communications, vol. 4, p. 1523, 2013. 15 D. Xiao, W. Zhu, Y. Ran, N. Nagaosa, and S. Okamoto, "Interface engineering of quantum hall effects in digital transition metal oxide heterostructures," Nature commu- nications, vol. 2, p. 596, 2011. 16 J. L. Lado, V. Pardo, and D. Baldomir, "Ab ini- tio study of Z2 topological phases in perovskite (111) (srtio3)7/(sriro3)2 and (ktao3)7/(kpto3)2 multilayers," Phys. Rev. B, vol. 88, p. 155119, Oct 2013. 17 Q.-F. Liang, L.-H. Wu, and X. Hu, "Electrically tunable topological state in [111] perovskite materials with an an- tiferromagnetic exchange field," New Journal of Physics, vol. 15, no. 6, p. 063031, 2013. 18 H.-M. Guo and M. Franz, "Three-dimensional topologi- cal insulators on the pyrochlore lattice," Phys. Rev. Lett., vol. 103, p. 206805, Nov 2009. 19 H. Huang, Z. Liu, H. Zhang, W. Duan, and D. Vanderbilt, "Emergence of a chern-insulating state from a semi-dirac dispersion," Phys. Rev. B, vol. 92, p. 161115, Oct 2015. 20 J. F. Afonso and V. Pardo, "Ab initio study of nontrivial in corundum-structured (M2o3)/(Al2O3)5 multilayers," Phys. Rev. B, vol. 92, p. 235102, Dec 2015. topological phases 21 D. Guterding, H. O. Jeschke, and R. Valent´ı, "Prospect of quantum anomalous hall and quantum spin hall effect in doped kagome lattice mott insulators," Scientific reports, vol. 6, p. 25988, 2016. 22 V. Pardo and W. E. Pickett, "Half-metallic semi-dirac- point generated by quantum confinement in tio2/vo2 nanostructures," Phys. Rev. Lett., vol. 102, p. 166803, Apr 2009. 23 L. Fu and C. L. Kane, "Topological insulators with inver- sion symmetry," Phys. Rev. B, vol. 76, p. 045302, Jul 2007. 24 M. Wimmer, "Algorithm 923: efficient numerical computa- tion of the pfaffian for dense and banded skew-symmetric matrices," ACM Transactions on Mathematical Software 7 (TOMS), vol. 38, no. 4, p. 30, 2012. 25 T. Fukui and Y. Hatsugai, "Quantum spin hall effect in three dimensional materials: Lattice computation of z2 topological invariants and its application to bi and sb," Journal of the Physical Society of Japan, vol. 76, no. 5, p. 053702, 2007. 26 A. A. Soluyanov and D. Vanderbilt, "Computing topologi- cal invariants without inversion symmetry," Phys. Rev. B, vol. 83, p. 235401, Jun 2011. 27 D. Gresch, G. Autes, O. V. Yazyev, M. Troyer, D. Van- derbilt, B. A. Bernevig, and A. A. Soluyanov, "Z2pack: Numerical implementation of hybrid wannier centers identifying topological materials," arXiv preprint for arXiv:1610.08983, 2016. 28 L. Fu, "Topological crystalline insulators," Phys. Rev. Lett., vol. 106, p. 106802, Mar 2011. 29 E. Prada, P. San-Jose, L. Brey, and H. Fertig, "Band topol- ogy and the quantum spin hall effect in bilayer graphene," Solid State Communications, vol. 151, no. 16, pp. 1075 -- 1083, 2011. 30 N. A. Garc´ıa-Mart´ınez, J. L. Lado, and J. Fern´andez- in multilayer Rossier, graphene," Phys. Rev. B, vol. 91, p. 235451, Jun 2015. "Quantum spin hall phase 31 D. Xiao, W. Yao, and Q. Niu, "Valley-contrasting physics in graphene: Magnetic moment and topological trans- port," Phys. Rev. Lett., vol. 99, p. 236809, Dec 2007. 32 M. Sui, G. Chen, L. Ma, W.-Y. Shan, D. Tian, K. Watan- abe, T. Taniguchi, X. Jin, W. Yao, D. Xiao, et al., "Gate- tunable topological valley transport in bilayer graphene," Nature Physics, vol. 11, no. 12, pp. 1027 -- 1031, 2015. 33 Z. Qiao, J. Jung, Q. Niu, and A. H. MacDonald, "Elec- tronic highways in bilayer graphene," Nano letters, vol. 11, no. 8, pp. 3453 -- 3459, 2011. 34 J. Li, K. Wang, K. J. McFaul, Z. Zern, Y. Ren, K. Watan- abe, T. Taniguchi, Z. Qiao, and J. Zhu, "Gate-controlled topological conducting channels in bilayer graphene," Na- ture Nanotechnology, 2016. 35 J. Li, I. Martin, M. Buttiker, and A. F. Morpurgo, "Topo- logical origin of subgap conductance in insulating bilayer graphene," Nature Physics, vol. 7, no. 1, pp. 38 -- 42, 2011. 36 Z. Ringel, Y. E. Kraus, and A. Stern, "Strong side of weak topological insulators," Phys. Rev. B, vol. 86, p. 045102, Jul 2012. 37 S. Okamoto, W. Zhu, Y. Nomura, R. Arita, D. Xiao, and N. Nagaosa, "Correlation effects in (111) bilayers of per- ovskite transition-metal oxides," Phys. Rev. B, vol. 89, p. 195121, May 2014. 38 S. Banerjee, R. R. P. Singh, V. Pardo, and W. E. Pickett, "Tight-binding modeling and low-energy behavior of the semi-dirac point," Phys. Rev. Lett., vol. 103, p. 016402, Jul 2009. 39 K. Schwarz, P. Blaha, and G. Madsen, "Electronic struc- ture calculations of solids using the wien2k package for material sciences," Computer Physics Communications, vol. 147, no. 1, pp. 71 -- 76, 2002. 40 P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, et al., "Quantum espresso: a modular and open- source software project for quantum simulations of materi- als," Journal of physics: Condensed matter, vol. 21, no. 39, p. 395502, 2009. 41 K. Koepernik and H. Eschrig, "Full-potential nonorthog- onal local-orbital minimum-basis band-structure scheme," Phys. Rev. B, vol. 59, pp. 1743 -- 1757, Jan 1999. 42 J. P. Perdew, K. Burke, and M. Ernzerhof, "Generalized gradient approximation made simple," Phys. Rev. Lett., vol. 77, pp. 3865 -- 3868, Oct 1996. 8 

1009.0170

2

1009

Prediction of huge magnetic anisotropies of transition-metal dimer-benzene complexes

Based on numerically accurate density functional theory (DFT) calculations, we systematically investigate the ground-state structure and the spin and orbital magnetism including the magnetic anisotropy energy (MAE) of 3d- and 4d-transition-metal dimer benzene complexes (TM2Bz, TM = Fe, Co, Ni, Ru, Rh, Pd; Bz = C6H6). These systems are chosen to model TM-dimer adsorption on graphene or on graphite. We find that Fe2, Co2, Ni2, and Ru2 prefer the upright adsorption mode above the center of the benzene molecule, while Rh2 and Pd2 are adsorbed parallel to the benzene plane. The ground state of Co2Bz (with a dimer adsorption energy of about 1 eV) is well separated from other possible structures and spin states. In conjunction with similar results obtained by ab initio quantum chemical calculations, this implies that a stable Co2Bz complex with C6v symmetry is likely to exist. Chemical bonding to the carbon ring does not destroy the magnetic state and the characteristic level scheme of the cobalt dimer. Calculations including spin- orbit coupling show that the huge MAE of the free Co dimer is preserved in the Co2Bz structure. The MAE predicted for this structure is much larger than the MAE of other magnetic molecules known hitherto, making it an interesting candidate for high-density magnetic recording. Among all the other investigated complexes, only Ru2Bz shows a potential for strong-MAE applications, but it is not as stable as Co2Bz. The electronic structure of the complexes is analyzed and the magnitude of their MAE is explained by perturbation theory.

cond-mat.mes-hall

cond-mat

Prediction of huge magnetic anisotropies of transition-metal dimer−benzene complexes Ruijuan Xiao, Daniel Fritsch,∗ Michael D. Kuz'min, Klaus Koepernik, and Manuel Richter IFW Dresden e.V., PO Box 270116, D-01171 Dresden, Germany Knut Vietze and Gotthard Seifert Physikalische Chemie, Technische Universitat Dresden, D-01062 Dresden, Germany (Dated: October 29, 2018) Abstract Based on numerically accurate density functional theory (DFT) calculations, we systematically investigate the ground-state structure and the spin and orbital magnetism including the magnetic anisotropy energy (MAE) of 3d- and 4d-transition-metal dimer benzene complexes (TM2Bz, TM = Fe, Co, Ni, Ru, Rh, Pd; Bz = C6H6). These systems are chosen to model TM-dimer adsorption on graphene or on graphite. We find that Fe2, Co2, Ni2, and Ru2 prefer the upright adsorption mode above the center of the benzene molecule, while Rh2 and Pd2 are adsorbed parallel to the benzene plane. The ground state of Co2Bz (with a dimer adsorption energy of about 1 eV) is well separated from other possible structures and spin states. In conjunction with similar results obtained by ab initio quantum chemical calculations, this implies that a stable Co2Bz complex with C6v symmetry is likely to exist. Chemical bonding to the carbon ring does not destroy the magnetic state and the characteristic level scheme of the cobalt dimer. Calculations including spin- orbit coupling show that the huge MAE of the free Co dimer is preserved in the Co2Bz structure. The MAE predicted for this structure is much larger than the MAE of other magnetic molecules known hitherto, making it an interesting candidate for high-density magnetic recording. Among all the other investigated complexes, only Ru2Bz shows a potential for strong-MAE applications, but it is not as stable as Co2Bz. The electronic structure of the complexes is analyzed and the magnitude of their MAE is explained by perturbation theory. PACS numbers: 31.15.es, 75.30.Gw, 75.75.-c ∗ present address: School of Physics, Trinity College Dublin, Dublin 2, Ireland. 1 I. INTRODUCTION Motivated by the ongoing quest for yet higher-density magnetic data storage in the con- text of the rapid advance of information technology, there is a continued search for nanoscopic magnetic structures with a large magnetic anisotropy energy (MAE) density by different experimental1 -- 6 and theoretical7 -- 15 methods. The MAE is the energy needed to turn the saturated magnetization of a system from one direction, usually the ground-state orientation, to another high-symmetry direction. Talking for simplicity about a nanoscopic system, we mean a canonical statistical ensemble of such systems, i.e. a macroscopically large number of identical non-interacting systems at equilibrium with a thermal bath. Thus, the MAE describes the stability of the magnetization direction against differently oriented external magnetic fields. In the simplest case of uniaxial anisotropy with an easy axis, the MAE is the energy barrier between two opposite equivalent directions of magne- tization with the lowest energy. A well-known example for this situation is bulk hcp Co. Using the bi-stability of magnetic structures, a bit of information can be stored: for example, one stable direction of magnetization on a hard disc area may encode "0", the other stable direction "1". It is generally accepted that long-term data storage requires that the total MAE of each magnetic particle should exceed 40 kT ,16 where k is the Boltzmann constant and T is the temperature. Spin-orbit interaction in a magnetic state is the primary source of MAE.17 The size of the spin-orbit coupling parameter of a given shell is a merely atomic property that depends on the atomic number. Hence, the spin-orbit splitting of atomic, molecular, or band states is fully determined by the character of these states in terms of atomic orbitals. On the other hand, the MAE as an energy difference depends sensitively on the particular electronic structure and, thus, on the geometry of the system. For instance, bulk hcp Co shows a moderate MAE of 0.06 meV per atom. Larger MAE can be obtained in surface-supported structures. It can be considerably influenced by tuning the coordination and the hybridization through the choice of substrate and size of the deposited clusters. For instance, the deposition of single Co atoms on a Pt(111) surface yields a record MAE of 9 meV per Co atom.1 Further reduction of the dimensions leads into the realm of nano-particles and magnetic molecules. To give an example, Fe4 organometallic clusters with a propeller-like structure exhibit magnetic anisotropy barriers which can be tuned by altering the ligands and reach 2 up to 1.5 meV per cluster.2 A large anisotropy barrier of 7 meV per cluster, generated by a deliberate structural distortion of the magnetic core with the help of bulky organic ligands, was recently reported for a complex within the Mn6 family of single-molecule magnets.3 Isolated magnetic dimers are the smallest chemical objects that possess a magnetic anisotropy as their energy depends on the relative orientation between dimer axis and mag- netic moment. Huge MAE values of up to 100 meV per atom were predicted for several transition-metal dimers (Ti2, Fe2, Co2, Ni2, Zr2, Tc2, Rh2, Ir2, and Pt2).8,11,18,19 However, it is impossible to utilize the huge MAE of dimers technologically unless they are bound to some medium. Our recent studies demonstrate that carbon-based substrates are suitable for this purpose.14 Both benzene (Bz) and graphene are ideal support materials that do not spoil magnetism and the huge MAE of the Co dimer. The Co2Bz complex has a ground-state structure with C6v symmetry, in which the dimer is bound perpendicularly to the carbon plane. This hexagonal environment preserves the two-fold degenerate singly occupied high- est molecular orbital (HOMO) of the free Co dimer, which is responsible for the large dimer MAE.8 As a result, Co2Bz was predicted to show a magnetic anisotropy of the order of 100 meV per molecule.14 This finding may open a way to enhance the presently available area density of magnetic recording by 3 orders of magnitude. The present work has two aims. First, additional detailed results on Co2Bz will be presented that support the conclusions drawn in Ref. 14. Second, related results for a whole series of dimer-benzene complexes including the 3d and 4d dimers Fe2, (Co2,) Ni2, Ru2, Rh2, and Pd2 will be shown. Among these, only Co2Bz and Ru2Bz turn out to be interesting candidates for a potential strong-MAE application. Benzene-reacted metal dimers have been explored since the 1980s. Trevor et al.20 studied the reaction products of benzene with gas-phase platinum clusters of different size. The existence of Fe2Bz was verified with infrared spectroscopy by Ball et al.21 after the reac- tion of iron atoms with cyclic hydrocarbons in an argon matrix. More recently, Bowen's group performed a series of mass spectrometry and photoelectron spectroscopic studies on iron-benzene,22 iron-coronene,23 cobalt-benzene,24 cobalt-pyrene,25 cobalt-coronene,26 and nickel-benzene27 cluster anions. The electron affinities and the vertical electron detachment energies were extracted from experimental spectra for the above complexes. By compar- ing measured spectra with results of density functional theory (DFT) calculations,23,26 they proposed structure models for a part of these clusters. For example, a half-sandwich ground- 3 state structure (C6v symmetry) was postulated for Fe2Bz−.22 Their investigations suggest that carbon rings could be a suitable template to deposit small transition-metal clusters. The reaction of Rh+ n cations with benzene was studied by Berg et al.28 These authors found that for n = 2 small amounts of Rh2Bz+ were formed, while the main products were RhBz+ + Rh and Rh2C6H4 + H2. Luttgens et al.29 measured the photoelectron detachment spectra of M2Bz− (M = Pt, Pd, Pb) and resolved the electron affinities and ground-state vibra- tion energies of these complexes. By analyzing the vibration frequencies, they postulated a perpendicular arrangement of Pd and Pt dimers on Bz (C6v symmetry) and a parallel coordination between Pb2 and Bz (C2v symmetry). On the theoretical side, DFT calculations were performed on several transition-metal ben- zene systems. Considering reaction products of iron atoms and benzene in low-temperature matrices, Parker recently calculated the energy of a number of isomers and simulated the related infrared spectra.30 By comparison of calculated and measured spectra he concluded that Fe2Bz is formed at high iron concentrations. If the Fe dimer is assumed perpendicular to the benzene plane above the center of the carbon ring, the calculated infrared spectrum shows an excellent agreement with the experimental data, though the calculation finds a different ground-state isomer. The ground-state structures of Fe2-coronene,23 Co2-pyrene,25 and Co2-coronene26 were also studied. For Fe2-coronene, a ground state with a total spin S = 3 was found with three quasi-degenerate isomers, where the Fe dimer is oriented either parallel or perpendic- ular to the coronene plane.23 For both Co2-pyrene and Co2-coronene, the ground state was found to be S = 2 with perpendicular orientation of the Co dimer at a position above a peripheral C-C bridge.25,26 Earlier work in this field is due to Senapati et al.31 who studied neutral and cationic Fe2-coronene complexes but discussed only parallel adsorption modes. Also, Rao and Jena studied the geometries and magnetic moments of neutral and ionic NinBzm complexes.32,33 They predicted a parallel adsorption mode between the Ni dimer and benzene, but it is unclear whether a perpendicular geometry was considered or not. The interaction of benzene with Rh+ and with Rh+ 2 was investigated by Majumdar et al.34 For the physisorbed dimer cation, Rh2Bz+, the minimum energy geometry has C2v symme- try with the two rhodium atoms lying horizontally above the benzene at the C-C bridge sites. Further research activities were devoted to transition-metal dimers interacting with 4 graphene or with fullerenes. Interaction of silver and gold adatoms and dimers with graphite or graphene was studied,35,36 and a perpendicular orientation of gold dimers on graphene was predicted.36 Also, both structure and spin magnetic properties of 3d transition-metal adatoms and dimers on graphite were investigated by Duffy et al.,37 but a possible per- pendicular arrangement of the dimers was not considered. DFT calculations for palladium clusters supported on graphene38 and on C60 39 find that the two atoms of an adsorbed Pd dimer are located on bridge sites, i.e., on top of C-C bonds. Recently, two detailed theoret- ical studies of Fe, Co, and Ni adatoms and dimers adsorbed on graphene were published: Johll et al.40 found that the most stable structure for all considered dimers, Fe2, Co2, and Ni2, has a dimer axis oriented perpendicularly to the graphene plane and placed at the hole site. An enhancement of the magnetic moment for the atom farther from the graphene was predicted,40 compared with the free dimer. Cao et al.41 found the same ground-state geometry, if the density-gradient corrected functional according to Perdew, Burke, and Ernz- erhof42 was used, but they note that partly different results were obtained by using the local spin-density approximation (LSDA). DFT calculations do not only allow to predict the ground-state geometry and spin of a magnetic complex, but may also provide a basic understanding of the electronic structure and of the orbital magnetic properties like orbital magnetic moment and MAE. We are however not aware of any published calculations of the magnetic anisotropy of transition- metal dimers on carbon-based systems except our recent letter, Ref. 14. In the following, we will demonstrate by DFT calculations that carbon hexagons are suitable hosts, where adsorbed transition-metal dimers may preserve their exceptional magnetic anisotropy. The investigated TM2Bz complexes are meant to serve as model structures for the ad- sorption of transition-metal dimers on the surface of graphite, on graphene, or on other carbon structures including molecular systems. For this reason, we only consider so-called physisorption (adsorption without expelling other atoms, e.g., hydrogen) as opposed to chemisorption that includes the possibility of de-hydrogenation.34 Most probably, the pre- sented predictions can only be verified under ultra-high vacuum conditions. Any interaction of the transition-metal atoms with, e.g., oxygen may deteriorate the specific structure and the related magnetic state we are focusing on. The paper is organized as follows. In Sec. II, the calculation method and computational details are explained. Sec. III compiles all results and related discussion: structure opti- 5 mization and stability of the ground states, analysis in terms of the bonding mechanism, the spin and orbital moments, and the strength of the MAE. The origin of the huge MAE in some of these molecules is also explained. Finally, the paper is summarized in Sec. IV. The appendix contains a description of auxiliary calculations. II. METHOD AND COMPUTATIONAL DETAILS The DFT calculations were performed with an highly accurate all-electron full-potential local-orbital scheme (FPLO),43 release 8.00-31.44 The code is based on a linear combination of overlapping nonorthogonal orbitals with a compact support. The molecular mode of FPLO with free boundary conditions was used. The presented data were obtained using the generalized gradient approximation (GGA) with a parameterized exchange-correlation functional according to Perdew, Burke, and Ernzerhof.42 All results were checked against additional calculations using the LSDA in the parameterization by Perdew and Wang.45 The dimer adsorption energy calculated by GGA is in all cases about 1 eV smaller than the related LSDA energy, but both approaches find the same ground-state structure type. Also, the ground-state spin obtained with GGA or LSDA is the same for all systems except Rh2Bz, where LSDA yields a non-magnetic ground state and GGA yields an S = 1 ground state. In both cases, however, the energies of the S = 0 and S = 1 states are very close. The molecular levels were occupied according to a Fermi-Dirac distribution in order to ensure the convergence of the Kohn-Sham equations. The presented results were obtained with a broadening temperature of T = 100 K. The basis set comprised 3d-transition-metal (3s, 3p, 3d, 4s, 4p, 4d, 5s), 4d-transition-metal (4s, 4p, 4d, 5s, 5p, 5d, 6s), carbon (1s, 2s, 2p, 3s, 3p, 3d), and hydrogen (1s, 2s, 2p) states. Lower-lying states of the transition-metal atoms were treated as core states. Geometry optimization was carried out with a scalar relativistic scheme. To find out the lowest-energy geometry and spin magnetic state of each TM2Bz complex, three possible high-symmetry structures (Fig. 1) were optimized for S = 0, 1, 2, and 3 (total spin moment, µS = 0, 2, 4, and 6µB), and for different initial spin arrangements (ferro- and ferrimagnetic). The point group symmetry C6v was applied for the configuration shown in Fig. 1(a), while for the structures depicted in Figs. 1(b) and 1(c), C2v was used for the structure optimization. Previous theoretical investigations of metal-benzene systems have shown that the struc- 6 tural changes of the benzene plane due to the metal-benzene interaction are negligible.9,46 -- 48 Thus, in nearly all our calculations the positions of the C and H atoms were fixed with C-C bond length 1.40 A and C-H bond length 1.09 A. Exceptions from this strategy are reported below. The Co2Bz complex is of particular interest.14 To make sure that the correct ground state was found for this system, we utilized the pseudopotential code ESPRESSO-4.0.149 and cross-checked the above calculations by full optimization of all the atomic positions starting from 14 kinds of initial structures. The computational details and a brief description of the results are given in the appendix. A quantity used to judge the stability of the considered structures is the adsorption energy (Ead) for a dimer entity attached to a benzene molecule, which is defined as Ead = Etot(Bz) + Etot(TM2) − Etot(TM2Bz) . (1) Here Etot refers to the respective total energy of the species indicated in the parentheses. Negative values of Ead mean that the TM2Bz complex is unstable. To evaluate the orbital magnetic moment and the MAE, spin-orbit coupling has to be included in the calculation. However, standard (quasi-)local DFT approximations like LSDA or GGA do not include orbital-dependent exchange effects.50 Thus, the orbital moments and the MAE are usually underestimated by these approaches. The orbital polarization (OP) correction51 is a frequently applied method to cure this problem. As a matter of experience, the MAE evaluated with standard LSDA or GGA approximation gives a lower estimate to the expected MAE, while the value obtained by including the OP correction provides an upper estimate. Experimental values of the MAE are most probably located between these lower and upper estimates. This has been demonstrated, e.g., in Refs. 52 and 7 and also for the special case of Co atoms in different chemical and structural surroundings in Ref. 14, Fig. 3. The MAE and the orbital magnetic moment were calculated by means of self-consistent fully relativistic calculations using the bond lengths obtained in the scalar relativistic cal- culations. The MAE was defined as MAE = Etot[k] − Etot[⊥] , (2) where Etot[k] and Etot[⊥] denote total energies of states with magnetization direction parallel and perpendicular to the Bz plane, respectively. The choice of the direction parallel to the 7 plane is arbitrary, since the in-plane anisotropy is negligible on the scale of the considered energies. Results obtained with and without OP corrections are reported. In the former case, the spin-dependent OP correction53 was applied to the 3d-orbitals of Fe, Co, Ni and to the 4d-orbitals of Ru, respectively. In order to cross-check one of the most important details of the GGA calculations, the bonding behavior of the Co dimer with Bz, we also performed ab initio quantum chemical calculations at the level of second order Møller-Plesset perturbation theory (MP2). The MP2 results were obtained from the MP2 implementation of Gaussian03.54 For Co2Bz, the Co atoms were described with a scalar-relativistic effective core potential (ECP) replacing 10 core electrons (MDF10),55 with the corresponding (8s7p6d 1f )/[6s5p3d 1f ] GTO basis set of triple-zeta quality. Accordingly, for benzene the Dunning correlation-consistent basis sets of double- and triple-zeta quality (cc-pVDZ and cc-pVTZ, respectively)56 were used. Since all of these basis sets are rather large, all energies have been corrected for the basis set superposition error (BSSE) with respect to the dissociation of the Co dimer from the benzene ring, employing the counterpoise scheme proposed by Boys and Bernardi57,58 as implemented in Gaussian03. III. RESULTS AND DISCUSSION A. Structure and spin state Dimer adsorption energies for each considered total spin and symmetry are shown in Fig. 2. The optimized structure parameters, dimer adsorption energies, and related spin magnetic moments for the six ground-state structures are listed and compared with literature data in Table I. 1. 3d transition-metal complexes We find that the adsorption mode with the dimer axis perpendicular to the benzene plane results in the most stable structure for all investigated 3d systems, Fe2Bz, Co2Bz, and Ni2Bz. This ground-state geometry is consistent with other GGA results obtained for dimers on graphene.40,41 The only discrepancy occurs for the ground-state spin magnetic moment of Fe2Bz. Here, we found a reduction of the free dimer spin (S = 3) to S = 2, 8 whereas Johll et al. and Cao et al. had reported S = 3 for the marginally different situation of Fe2 on graphene.40,41 As the related energy difference in our calculation was very small (9 meV), we repeated the calculations with full optimization of the C-C and C-H distances (C6v symmetry). The full optimization inverted the order of the two considered states, the state with S = 3 now being 16 meV lower than the competing state with S = 2. Moreover, a ferrimagnetic state with S = 1 is found only about 40 meV higher in energy, Fig. 2. Such small energy differences cannot guarantee the stability of the Fe2Bz magnetic ground state and should give rise to strong spin fluctuations. Turning our attention to the ground-state geometry of Fe2Bz, we note that Parker30 pro- vided evidence of the same proposed geometry by obtaining an excellent agreement between calculated and experimental infrared spectra. Note, that in this geometry Fe2Bz complexes with S = 1, 2, and 3 give almost identical simulated spectra.30 Thus, the comparison cannot be used to distinguish the magnetic state. Our results add additional weight to Parker's arguments, who proposed a C6v geometry for Fe2Bz. One should note, however, that other structure types compete with the C6v geometry, see Fig. 2. Indeed, if the C and H coordi- nates are optimized as well, the ground state turns to kb, S = 3, almost degenerate with the states ⊥c, S = 2 and S = 3. In line with this finding, Cao et al. reported a ground state with the Fe dimer above a graphene hollow site, but not perpendicular to the plane.41 In the following discussion, we will disregard Fe2Bz structures different from C6v due to the mentioned experimental evidence of this structure type. For Co2Bz and Ni2Bz, we obtained both the magnetic moment and the ground-state structure in agreement with the results by Johll et al.40 and by Cao et al.41 The GGA dimer adsorption energy, Ead = 1.39 eV for Co2 on Bz from our calculation, and the related energies 0.92 eV from Johll et al. and 1.13 eV from Cao et al. for Co2 on graphene indicate a reasonable stability of this structure. Noteworthy, any other spin state considered in our calculations, including a ferrimagnetic solution with a total spin S = 1, has a much higher energy, at least 0.80 eV above the ground state. The ground state of Ni2Bz is also sufficiently separated from other states with a different spin and/or geometry, see Fig. 2. We also checked a further geometry with Ni atoms close to next nearest C-bridges (see Fig. A1 (iv)). Such a geometry was found by Rao and Jena33 to be lowest in energy. We could not confirm this finding and obtained, even with full relaxation, a 0.45 eV higher energy for this geometry (S = 1) than for the ⊥c geometry with S = 1. 9 To investigate the stability of the proposed perpendicular adsorption mode for the Co system in more detail, three further structures were considered, including two dissociated cases: (i) attachment of one Co atom on each side of the carbon ring, (ii) dissociation of one of the Co atoms resulting in one free Co atom and CoBz, and (iii) dissociation of both Co atoms resulting in two free Co atoms and one free benzene molecule. In the related scalar relativistic atom calculations, non-integer occupation of the open shells was admitted. All the three structures have higher energies, by 2.47 eV, 3.49 eV, and 5.05 eV, respectively, than the perpendicular arrangement. From the energy difference between state (ii) and state (iii), a value of 1.56 eV is found for the adsorption energy of a single Co adatom on benzene. We also can find a binding energy of 3.66 eV for the Co dimer from the third dissociated state. In comparison with the DFT data, the experimental binding or adsorption energies are considerably smaller: 0.34 eV59 (adsorption energy of a Co adatom to benzene) and 1.72 eV60 (binding energy of a Co dimer). This is in line with the known tendency of DFT calculations to overestimate the binding energies in many cases. Thus, to confirm the qualitative validity of the energies and structure sequence, we per- formed quantum chemical (MP2) calculations. In comparison with our previously published results,14 the MP2 calculations were improved by taking into account BSSE corrections and by extending the benzene basis. Fig. 3 compares the energy sequences of three possible high-symmetry structures and one dissociated configuration obtained by GGA with related MP2 results. All MP2 energies were evaluated by single point calculations using the GGA- derived geometries, except for the Co dimer. For the latter the interatomic distance was optimized at the MP2 level. It turned out slightly shorter (0.1909 nm) than the GGA result (0.1997 nm). For all structures, MP2 calculations were carried out for S = 0, 1, 2 and 3. In all cases a total spin of 2 was found to be most favorable. Importantly, the adsorption energy of the Co dimer to benzene was found to be yet higher than in the GGA calculation. It is obvious from Fig. 3 that the quantum chemical calculations confirm the main GGA result, that bonding of a Co dimer with a single molecule of benzene results in the structure depicted at the bottom of Fig. 3 with a total spin S = 2. Also, the sequence of the higher- energy structures is the same in GGA and in MP2, and the same spin magnetic moments are found with the exception of the dissociated state, where MP2 predicts S = 3 and GGA yields S = 2. When improving the benzene basis from a double- to a triple-zeta level, the BSSE-corrected adsorption energies rise consistently by about half an eV, reaching 2.36 eV 10 for the most stable structure. This result should serve as a valid proof that the Co dimer can be bound to the benzene ring. As a final check that the calculations described above provided the correct ground-state geometry, we performed a cross-check for Co2Bz with the pseudopotential code ESPRESSO- 4.01.49 We carried out a full optimization of all atomic positions starting from 14 kinds of initial structures. As before, GGA and a scalar relativistic mode were used. Other technical details are described in the appendix. The results confirm that the bonding of Co2 with a single molecule of benzene very likely results in the perpendicular configuration, which is separated from other possible arrangements by at least several hundred meV. Only a tiny distortion of the benzene plane is found in the full optimization results. This is a weak Jahn- Teller effect that splits the singly occupied two-fold degenerate HOMO state originating from Co2. This splitting is very small (<1 meV), since the original C∞ symmetry of the Co dimer where the HOMO resides is only weakly distorted by the hexagonal ligand. Thus, it will hardly have any influence on the magnetic properties of Co2Bz. In particular, if spin-orbit coupling is taken into account, the HOMO will be split by this interaction rather than by the Jahn-Teller effect, which is almost 2 orders of magnitude weaker than the spin-orbit coupling in the considered case. Only if the magnetization is oriented perpendicular to the dimer axis, the spin-orbit splitting vanishes in lowest order. In this case, the Jahn-Teller effect might marginally reduce the total energy and, thus, the MAE. We conclude that Co2Bz and Ni2Bz probably exhibit a C6v symmetry, like Fe2Bz. As distinct from the Fe system, Co2Bz and Ni2Bz have a stable magnetic ground state. Experi- mental evidence of these proposed structures seems however lacking at the moment. For the anion Co2Bz−, the observed photoelectron spectra24 allowed to exclude a structure where the two Co atoms are placed on both sides of the benzene plane. 2. 4d transition-metal complexes We find (Fig. 2) that among the investigated 4d dimers only Ru2 prefers an upright adsorption mode. It binds to the benzene as strongly as the cobalt dimer, but its magnetic ground state has a lower spin, S = 1, and lies only 0.25 eV below a zero spin state. The ground states of both Rh2Bz and Pd2Bz are found to be almost degenerate with respect to spin multiplicity (Rh2Bz) or geometry (Pd2Bz). 11 We are not aware of any published information about the geometry of the neutral com- plexes Ru2Bz and Rh2Bz. For the cation Rh2Bz+, the structure type kb (Fig. 1) was obtained as the lowest-energy structure by DFT calculations using the B3LYP functional.34 This is the same ground-state structure as we find for the neutral Rh2Bz. Luttgens et al.29 deduced the vibration energies of both Pd2Bz and Pd2Bz− from photo- electron detachment spectra. They postulated an orientation of the Pd dimer perpendicular to the benzene ring because the observed vibration frequency of Pd2Bz is close to that of the free Pd2. One should note that our calculated ground-state geometry of Pd2Bz contradicts this analysis. On the other hand, calculations by Cabria et al.38 using LSDA and GGA and by Loboda et al.39 using the B3LYP functional find a ground-state geometry with the two palladium atoms placed horizontally above the carbon ring, similar to our results. We performed a series of additional tests in order to clarify this discrepancy between experiment and theory. First, we checked the influence of spin-orbit interaction and found, that related total energy shifts do not exceed 0.1 eV. The parallel adsorption mode hence is still more stable than the perpendicular one. Second, we checked a possible asymmetric adsorption above a single bridge site. Indeed, the related total energy is about 0.5 eV lower than for the adsorption above the hollow site, if every symmetry constraint is released. The dimer axis then deviates from the initial perpendicular orientation and forms an angle of about 40 degrees with the benzene plane, while S = 0. Yet, the energy of this structure is still higher than that of the parallel configuration. Third, we calculated vibration frequencies of the Pd-Pd bond using the harmonic approximation for the three adsorption modes defined in Fig. 1. If S = 1, the three structures give rise to almost the same vibration energies of 21.9 meV (⊥c), 22.3 meV (kb) and 22.5 meV (kt). For the asymmetric bridge site configuration, a value of 20.6 meV is obtained. All of these energies are close to the vibration energy of the free dimer, 26 meV.61 This comparison shows that proximity of the vibration spectrum to that of the free dimer is no proof of the perpendicular geometry. For the ground state, S = 0, the parallel structures turn out to be much softer, 12.8 meV (kb) and 13.3 meV (kt), than the free dimer and also than the perpendicular geometry, 19.7 meV. On the whole, the calculated vibration spectra do not provide enough evidence in favor of any investigated structure. Finally, we optimized the structures of Pd2Bz− anions. The structure kb with S = 1/2 is again more stable than the structure ⊥c, by 0.45 eV. The structure with asymmetric bonding above a single bridge site is 0.13 eV lower in energy than the structure ⊥c, but it is 12 still 0.32 eV higher than the structure kb. Summarizing this point, the discrepancy between experimental and theoretical results on Pd2Bz persists. B. Electronic structure and bonding mechanism Free TM dimers have been discussed in detail recently.8,11,19,62 Analysis of their electronic structure reveals that a singly occupied HOMO which is two-fold degenerate in the absence of spin-orbit coupling is responsible for the giant magnetic anisotropy predicted in some of these dimers.8 For example, the most important feature in Co2 is a two-fold degenerate singly occupied 3d-δ∗ u state. It is split by spin-orbit interaction, if the magnetic moment is oriented along the dimer axis but stays degenerate if the moment is oriented perpendicular to the axis.8 Concerning the bonding between metal atoms and a benzene molecule, Mokrousov et al.9 reported a schematic analysis for V-Bz complexes and showed that the HOMO and the lowest unoccupied molecular orbital (LUMO) of benzene interact with the metal s and d orbitals of the same symmetry. To better understand the bonding mechanism between the TM dimer and the benzene molecule, we compare the levels of Co2Bz and of Fe2Bz with those of the related free dimers, Fig. 4. The third panel (from left) shows the textbook electronic structure of benzene and the leftmost panel refers to the free Co2, as recently discussed in Ref. 8. The other panels show the electronic structure evaluated for the ground-state geometries and spin multiplicities of Co2Bz, Fe2Bz, and Fe2 as well as for the low-lying S = 3 state of Fe2Bz. It turns out that bonding of Co2 on benzene does not lead to any deterioration of the magnetic properties of Co2: (i) the ground-state spin stays S = 2, as in the free dimer; (ii) the Co 3d-δ∗ u level is still two-fold degenerate in Co2Bz due to the C6v symmetry; (iii) this level is still the singly occupied HOMO. In this way, the key feature responsible for the giant MAE of the free Co dimer is preserved in the Co2Bz structure if the benzene molecule binds perpendicularly to Co2 in C6v symmetry. Yet, there is an important difference between free Co2 and Co2Bz. In the free dimer, the two Co atoms contribute equal weights to the minority spin 3d-δ∗ u state. At variance, the HOMO of Co2Bz receives 94% of its weight from that Co atom which is farther away from the benzene plane, see Fig. 5. One can note that the magnetic moment of free Co2 is also preserved in the parallel arrangements kb and kt (Fig. 2). However, the reduction of the symmetry to C2v causes a 13 split of all δ and π states (not shown here), resulting in non-degenerate HOMO and LUMO in these structures. The minority spin π∗ state near to the Fermi level is split by 0.57 eV in the kb structure and by 0.60 eV in the kt structure, while the δ∗ state in the minority spin channel is split yet more strongly, by 0.68 eV in kb and by 0.85 eV in kt. We find that, unlike in the case of Co2, the adsorption of Fe2 on benzene results in a change of both magnetic moment and electron configuration, as compared with the free Fe dimer. The ground-state level scheme of Fe2 is very similar to that of Co2, see Fig. 4, but two more holes are introduced in the minority spin channel. As a result, S = 3 (Fe2) instead of S = 2 (Co2) and the exchange splitting is enhanced, so that the majority spin 4sσ∗ level is again quasi-degenerate with the HOMO. The latter is now allocated to the singly occupied δ orbital instead of δ∗ in Co2. If Fe2 binds to benzene, the electron occupying the majority spin π∗ level of Fe2 moves into the minority spin δ level. The minority spin δ level becomes doubly occupied, while the majority spin π∗ level turns singly occupied and acts as the HOMO in the S = 2 ground state of Fe2Bz. The electronic structure of Fe2Bz is very similar to that of Co2Bz in the same spin state (S = 2). The covalent splittings of the d-states are somewhat larger in the Fe system than in the Co one, due to the larger extension of the Fe-d orbitals compared with the d-orbitals of Co. This yields a somewhat different orbital order. It will be recalled that in Fe2Bz a state with S = 3 is close in energy to the ground state; in this state the exchange splitting is larger and both holes enter the minority spin channel, like in the free Fe dimer. We included the corresponding level scheme for comparison in Fig. 4. The essential difference between Fe2 and excited Fe2Bz in the same spin state is that σ and δ in the minority spin channel interchange their positions. Thus, Fe2Bz in the state S = 3 has a fully occupied δ HOMO and, thus, a small magnetic anisotropy. The orbital characteristics at the Fermi level determine the main physical properties of the structure, while the stability of the complex depends on how the TM dimer binds to the benzene molecule. Fig. 5 shows the integrated density of states (IDOS) for the ground state of the Co2Bz structure and the orbital composition of each state. In an axial symmetry, the five d orbitals are split into three groups: dσ(dz2), two-fold degenerate dπ(dxz, dyz), and two-fold degenerate dδ(dxy, dx2−y2). The six benzene π orbitals can also be classified with respect to the same axis63 and include one Lσ(π1) orbital, two degenerate HOMO Lπ(π2, π3), two degenerate LUMO Lδ(π∗ 5), and one Lφ(π∗ 4, π∗ 6) orbital, 14 where L means ligand. The adsorption between the Co dimer and the benzene molecule is realized primarily by forming three types of chemical bonds, δ, π, and sσ, between carbon atoms and Co1 (the Co atom nearest to the benzene ring), while the the other Co atom, Co2, mainly binds with Co1 and scarcely contributes to the bonding with the benzene. For example, the minority spin δ state at -4.3 eV and the related π∗-dominated (Bz-Lδ) δ-state at -1.2 eV in Fig. 5 stem from the combination of the dxy, dx2−y2 orbitals of Co1 and the LUMO of benzene. The π states positioned near -7.8 eV in the minority spin channel and around -8.0 eV in the majority channel can be attributed to the dxz, dyz orbitals of Co1 and the HOMO of benzene. The hybridization between the 4s orbital of Co1 and the Lσ(π1) orbital of benzene forms the sσ orbital, located at around -10.2 eV. Formation of these three kinds of bonds lowers the energy of the Co2Bz complex compared with the dissociated state and stabilizes the perpendicular adsorption structure. A simple model for Co2Bz can be sketched from above analysis of the bond mechanism: the benzene molecule plays the role of a substrate to fix the Co dimer, the Co atom next to the benzene plane acts as "glue" to bind the dimer to the benzene, and the other Co atom mainly contributes to the states near the Fermi level and dominates the magnetic properties of the whole structure. Level schemes for the ground states of Ni2Bz, Ru2Bz and for the related free dimers/benzene are shown in Fig. 6. The electronic structure of Ni2 is almost unchanged upon adsorption. The spin remains S = 1 and π∗ of the minority spin channel is still the singly occupied HOMO. The situation is different for Ru2, where the total spin is reduced from S = 2 to S = 1 due to the adsorption. Together with the change of spin, an essential alteration of the orbital order close to the Fermi level takes place. In particular, majority spin σ∗ and π∗ levels interchange their positions. Thus, Ru2Bz has a singly occupied δ∗ HOMO like Co2Bz. C. Spin and orbital moments Site-resolved spin moments µS and orbital moments µL for the stable magnetic structures with perpendicular geometry (TM = Fe, Co, Ni, Ru) are listed in Table II. The calculations were carried out within the fully relativistic scheme. The moments are aligned perpendicu- larly (⊥) or parallelly (k) to the benzene plane, respectively. We checked that the in-plane 15 anisotropy of the magnetic moments is marginally small and can be safely neglected. The effect of the OP correction is also given for comparison. It is clear that the magnetic moment is mainly distributed between the two transition-metal atoms. The magnetic moment on the C sites is so small that it can be neglected (µS(C) < 0.05µB). Inspection of the µS data reveals that in all four systems the TM2 atom shows a much higher spin moment than the TM1 atom. The relatively high coordination number of TM1, seven, results in a considerable reduction of its spin moment. On the other hand, TM2 is only singly coordinated and thus behaves almost like a free atom. The spin moments of Fe, Co, Ni, and Ru free atoms amount to 4, 3, 2, and 4 µB, respectively. The calculated spin moments of the TM2 atoms in 3d-TM2Bz complexes are only 0.5 . . . 0.7 µB smaller than the corresponding atomic values. In the case of Ru, the TM2 carries only about half of the atomic spin moment. Spin magnetism of 4d atoms is in general less stable than that of isoelectronic 3d atoms, since the 4d intra-atomic exchange (Stoner) integrals are somewhat smaller than the related 3d integrals. Another feature is that the spin moments are nearly the same for both magnetization directions. When the moment orientation switches from ⊥ to k, the primary effect is a change of the orbital moment of the TM atoms. This fact indicates that the magnetic anisotropy of these systems is closely connected to the anisotropy of the orbital moments.64 It is worth noting that in both the Co2Bz and the Ru2Bz systems the TM2 atoms show very large orbital moments in the ⊥ orientation and relatively small values in the k orientation. This is a sign of a large magnetic anisotropy of these systems. As expected, the calculated moments in Table II show that the OP correction generally increases the orbital moment while scarcely affecting the spin moment. For example, when the OP correction is allowed for, the orbital moments of Co atoms with magnetic moments parallel to the Bz plane are about three times larger than those calculated without the OP correction. This is caused by the very construction of the OP correction scheme, where additional (exchange) energy is gained if the orbital moment is enhanced.50,51 In case of perpendicular orientation of the moments with respect to the Bz plane, the orbital moments are less influenced by the OP correction, since spin-orbit coupling alone already provides almost the maximum orbital moment allowed by the given electronic level sequence. The total orbital moments evaluated for the case when the magnetic moment is parallel to the dimer axis (i.e., perpendicular to the benzene plane) directly reflect the nature of the 16 HOMO. In the case of Fe2Bz, Fig. 4, the HOMO is a π∗ state in the majority spin channel. Spin-orbit coupling splits this state in such a way that the energy of the m = −1 sub- level is reduced (m denotes the magnetic quantum number). This sub-level is consequently occupied, while the m = +1 sub-level is empty, and the total orbital moment is close to −1 µB. In the case of Co2Bz, Fig. 4, the HOMO is a δ∗ state in the minority spin channel. Spin-orbit coupling splits this state so that the energy of the m = +2 sub-level is reduced. Accordingly, µL ≈ 2 µB. In Ni2Bz, the HOMO is a δ∗ state with both sub-levels occupied. Therefore, the orbital moment nearly vanishes. Finally, in Ru2Bz the HOMO is a δ state in the minority spin channel with µL ≈ 2 µB. D. MAE After analyzing the spin and the orbital moments in the stable ⊥c TM2Bz structures, we proceed to another important property of magnetic systems, the MAE. This quantity is in the main focus of the present investigation. In the perpendicular adsorption mode of TM2 on the benzene molecule, it is natural to consider the MAE as the energy difference between the states with magnetization direction parallel and perpendicular to the benzene plane, Eq. (2). The first two lines in Table III list our calculated MAE for the stable magnetic TM2Bz structures with TM = Fe, Co, Ni, Ru. The data for Etot were obtained by two self-consistent fully relativistic calculations with respective magnetization directions. The values calculated without OP correction should be considered as a lower estimate to the expected MAE, while an upper estimate is obtained by including the OP correction. Endowed with a large ground-state orbital moment as demonstrated in Sec. III C, Co2Bz and Ru2Bz show a huge MAE. The lower estimate to the MAE in Co2Bz is hardly changed in comparison with the free Co dimer.11 This is because the magnetic state and the important features of the electronic structure of Co2 are not changed by the adsorption. The upper estimate, 334 meV per Co2Bz molecule, is even higher than the related value for Co2 (188 meV per dimer). This is due to the almost complete localization of the HOMO on TM2 in the case of Co2Bz (Table III, lines three and four). While µ⊥ L(TM2) ≈ 2 µB in Co2Bz (Table II), it is only half as large in the free Co dimer, where by symmetry µL(TM1) = µL(TM2) ≈ 1 µB in the ground state. The OP correction energy is only half as large in the latter case compared 17 to the former one, since it is quadratic in the atom-projected µL. At variance, the spin-orbit coupling energy is linear in µL. The case of Ru2Bz is different. We find in the ground state of the free Ru dimer (with S = 2) a two-fold degenerate, completely occupied majority-spin 3d − π∗ HOMO and, thus, a small MAE. The spin moment is reduced by the adsorption of Ru2 on benzene. This causes a change of the electron configuration, resulting in a two-fold degenerate and singly occupied δ∗ HOMO with a related huge MAE. The iron system shows a somewhat smaller MAE, mainly due to the smaller value of m of the HOMO. Finally, the Ni system has a fully occupied HOMO that does not contribute to the MAE in lowest order. Nonetheless, the obtained MAE reaches or exceeds the highest known experimental values.1,3 We finally would like to understand the obtained numbers in terms of simple arguments based on perturbation theory,8 extended here to include the OP correction. Our consider- ation is limited to systems with singly occupied, two-fold degenerate HOMO (here, Fe2Bz, Co2Bz, and Ru2Bz). The MAE is approximated by the single-particle energy change of the occupied HOMO level upon changing the direction of magnetization. If the OP correc- tion is included, we call the magnetic anisotropy energy MAE(SO+OP), otherwise it is called MAE(SO). Thus, MAE(SO) ≈ m X i (C 2 m,d−TMi + C 2 −m,d−TMi) ξd/2 (3) in first-order perturbation theory. Here, ξd is the d-shell spin-orbit parameter and Cm,d−TMi is the projection of one of the HOMO orbitals on the d-orbital of atom TMi (i = 1, 2) with magnetic quantum number m (m = 1 or 2 for a HOMO of type π or δ, respectively). If OP corrections are taken into account, this first-order estimate changes to MAE(SO+OP) ≈ m X i (C 2 m,d−TMi + C 2 −m,d−TMi) (ξd/2 + B∆µL(TMi)/µB) . (4) Here, B denotes the TM-specific Racah parameters,65 evaluated from the related atomic orbitals, and ∆µL(TMi) = µ⊥ L(TMi) − µk L(TMi), (i = 1, 2), according to Tab. II. Table III shows the major contributions to the composition of the HOMO, (C 2 m,d−TMi + C 2 −m,d−TMi), the related magnetic quantum number m, the occupation of the HOMO, the spin-orbit parameters, and the Racah parameters. MAE values estimated by first-order perturbation theory are given in brackets following the self-consistently evaluated MAE data. We find that the self-consistent values are smaller than the estimates obtained by 18 perturbation theory. This has at least two reasons: (i) negative contributions from levels other than the HOMO and higher-order HOMO contributions, cf. the results for Ni2Bz where the HOMO does not contribute in first order, and (ii) charge relaxation reduces the effect. Nonetheless, the self-consistent MAE amounts to 60 . . . 80% of the first-order estimates. According to the above analysis, Co2Bz and Ru2Bz are interesting candidates for strong- MAE applications. It should be noted that in both cases the easy axis of magnetization is directed perpendicularly to the benzene ring. This fact is advantageous for conventional recording techniques. IV. SUMMARY We report a systematic DFT study of the ground-state structures, bonding mechanism, spin and orbital moments, and in particular of the MAE of TM2Bz complexes (TM=Fe, Co, Ni, Ru, Rh, Pd), using the full-potential local-orbital method FPLO. Upright adsorption modes with C6v symmetry of Fe2, Co2, Ni2, and Ru2 on benzene molecules are confirmed (Fe2Bz) or predicted (others). Huge MAE, stable geometry, and stable magnetic ground states are predicted for Co2Bz and for Ru2Bz. The main origin of the large anisotropy of these two systems is the large orbital moment of the TM atom which is farther away from the benzene plane. Analysis of the electronic states shows that bonding of the Co dimer on the benzene molecule does not lead to any deterioration of the magnetic properties of the dimer. Most important is that the two-fold degenerate singly occupied HOMO state of the free dimer is preserved, which allows the spin-orbit coupling to produce a large magnetic anisotropy. An important conclusion can be drawn from these results: robust and easy-to- prepare carbon-based substrates are well-suited to adsorb transition-metal dimers for the purpose of high-density magnetic recording. We hope that our predicted exceptionally large MAE of Co2Bz and Ru2Bz will motivate experimental investigations of transition-metal dimers on carbon-based substrates, like graphite or graphene. 19 Appendix A full optimization of all atomic positions was carried out for Co2Bz complexes by us- ing the ESPRESSO code.49 We used the pseudopotentials Co.pbe-nd-rrkjus.UPF, C.pbe- rrkjus.UPF and H.pbe-van_bm.UPF from the http://www.quantum-espresso.org distri- bution. A supercell of the size 20 A× 20 A× 20 A was used to make sure that there is virtually no interaction between the molecules of neighboring cells. The Brillouin zone sam- pling was performed only on the Γ point. The cutoffs used for the wave functions and for the charge density were 60 Ry and 300 Ry, respectively. The convergence in total energy was carefully checked. A Marzari-Vanderbilt cold smearing66 with 0.007 Ry was used to get the convergence in energy levels. The optimizations were done without any constraints on symmetry or spin moment. Fig. A1 shows the 14 initial structures used to start the geometry optimization. The optimization results in six types of final structures. Structures (i), (xi), (xii), and (xiii) converge to the perpendicular adsorption mode; structure (x) stays almost unchanged and leads to the parallel adsorption mode on the bridge site of the carbon ring; structures (v) and (ix) converge to the parallel adsorption mode on the top site of the carbon ring; structures (iii), (vi), and (viii) finally go over to a structure type similar to (iii); structures (ii) and (iv) become the structure type (iv); and the final structure of (xiv) is still the adsorption mode on both sides. The optimization for structure (vii) does not converge. Among these structures, the perpendicular adsorption mode shows the lowest total energy, which is 0.74 eV lower than the kb adsorption mode. The spin of this structure is S = 2, confirming the FPLO result. There, the kb adsorption mode was found 0.98 eV higher than the perpendicular adsorption mode. Full optimization of all atomic positions shows that there is only a tiny distortion of the benzene plane in the perpendicular adsorption mode. The length of the C-C bonds increases slightly, two of them changing from 1.40 A to 1.4175 A and four of them changing to 1.4169 A. The C-H bond-length changes from 1.09 A to 1.0889 A and 1.0887 A. The electronic structure shows that the two δ∗ states (now, HOMO and LUMO) are still nearly degenerate with a gap smaller than 1 meV. 20 (a) ⊥c (b) kb (c) kt FIG. 1: (Color online) Three possible high-symmetry structures for TM2Bz complexes. Hexagons and blue bullets indicate benzene rings and and transition-metal atoms, respectively. (a) ⊥c − the TM dimer is situated on the C6v symmetry axis perpendicularly to the benzene plane; (b) kb − the TM dimer is parallel to the benzene plane with TM atoms near the middle of opposite C-C bonds (C2v symmetry); (c) kt − the TM dimer is parallel to the benzene plane with TM atoms near the top of opposite C atoms (C2v symmetry). 21 1 0 -1 -2 1 0 -1 -2 ] V e [ y g r e n e n o i t p r o s d a r e m d i (a) Fe2Bz (d) Ru2Bz 0 2 4 6 1 0 -1 -2 1 0 -1 -2 (b) Co2Bz (e) Rh2Bz 0 6 fixed total spin moment [µΒ] 2 4 1 0 -1 -2 1 0 -1 -2 ⊥ c b t (c) Ni2Bz (f) Pd2Bz 0 2 4 6 FIG. 2: (Color online) Scalar-relativistic dimer adsorption energies Ead calculated for optimized structures of (a) Fe2Bz, (b) Co2Bz, (c) Ni2Bz, (d) Ru2Bz, (e) Rh2Bz, and (f) Pd2Bz complexes. For all systems, three initial structures illustrated in Fig. 1 were optimized with fixed C and H coordinates for the following values of the fixed total spin moment, µS=0, 2, 4, and 6µB. Both parallel and anti-parallel relative spin orientations were considered. Open (filled) symbols indicate that the moments of the two transition-metal atoms in the spin state with the highest adsorption energy are parallel (anti-parallel). 22 2 2 Bµ Bµ 3 3 Bµ Bµ 1.39 eV 1.04 eV 0.98 eV 0.00 eV GGA−PBE Bµ2 Bµ2 Bµ2 Bµ2 2.5 Bµ Bµ 1.5 2.2 Bµ Bµ 1.9 2.15 eV 1.97 eV 1.45 eV 2.36 eV 1.92 eV 1.38 eV 0.00 eV 0.00 eV MP−2, DZ MP−2, TZ FIG. 3: (Color online) Energies and magnetic moments of different Co2Bz configurations calculated by DFT (left column) and by MP2 (right columns). DZ and TZ abbreviate double-zeta and triple- zeta basis sets, respectively. The energies refer to the ground-state energy. For the uppermost, dissociated and for the lowermost, ground-state configuration, the spin moments in red (on the right-hand side of the Co dimer) refer to the MP2 calculations and the moments in black (on the left-hand side of the dimer) refer to the GGA calculations. For the other configurations, GGA and MP2 yield the same spin. Hexagons and blue bullets indicate Bz and Co, respectively. 23 Co2 ⊥ Bz 4sσ* π* Co2 4sσ* π* Bz *,π π * 5 4 Fe2 4sσ* π* ⊥ Bz (4µΒ) Fe2 4sσ* π* σ* δ* δ 4sσ π σ π* 4sσ σ* δ* δ π σ π σ π* σ* δ* 4sσ δ π σ π σ π* σ* δ* 4sσ π δ σ π σ π 2,π 3 σ π 1 π* π* σ* δ* 4sσ δ π σ π σ ⊥ Bz (6µΒ) 4sσ* π* π* 4sσ* π* σ* δ* σ δ π 4sσ π σ π π* σ* δ* 4sσ π δ σ π σ π ] V e [ y g r e n e n o r t c e l e - e l g n i s 0 -2 -4 -6 -8 -10 4sσ* δ* σ* π* δ 4sσ π σ Fe2 σ* π* δ* 4sσ* δ σ π 4sσ 4sσ* σ* π* δ* 4sσ δ π σ FIG. 4: (Color online) Scalar-relativistic single-particle levels of Co2 (left panel), Co2Bz (ground- state structure, second panel), benzene (third panel), Fe2Bz (ground-state structure, fourth panel and first spin-excited state, fifth panel), and Fe2 (right panel). All energies refer to a common vacuum level. Black lines denote occupied states, orange (gray) lines denote empty states, and thick blue (gray) lines indicate singly occupied two-fold degenerate states. With the exception of benzene, the levels are spin-split (S = 2, 2, 2(3), and 3 for Co2, Co2Bz, Fe2Bz, and Fe2, respectively). Majority states are indicated by up-arrows, minority states by down-arrows. Dimer- dominated states are labeled in black and benzene-dominated states are labeled in red (gray). 24 σ (sσ) π (π) π σ π δ σ* π* δ* 4sσ (δ) π* 4sσ* σ σ σ σ 1.0 0.5 0 0.5 ] t i n u . b r a [ n o i t s o p m o c l a t i b r o σ σ 1.0 σ -12 -12 π (sσ) -10 -10 π (π) -8 -8 σ -6 -6 Energy [eV] σ* π* π δ 4sσ δ* (δ) EF -4 -4 4sσ* π* (δ) -2 -2 0 0 45 30 15 0 15 30 45 I D O S [ s t a t e s ] Co2 3dx2-y2, 3dxy Co2 3dxz, 3dyz Co2 3dz2 Co2 4s Co1 3dx2-y2, 3dxy Co1 3dxz, 3dyz Co1 3dz2 Co1 4s C 2pz C 2px, 2py FIG. 5: The orbital composition of each state (left axis) and the integrated density of states (IDOS, right axis) for the ground-state structure and spin of Co2Bz. The upward (downward) arrow indicates majority (minority) spin states. All energies refer to a common vacuum level. The labels for each state are the same as those in Fig. 4. The three types of chemical bonds between the Co dimer and benzene are labelled blue in parentheses. The Co atom closer to the benzene plane is labelled Co1, the other one is labelled Co2. The position of the Fermi level (EF ) is indicated by a vertical line. Missing contributions to the orbital composition which should add up to unity are due to the omitted C-2s and H-states. 25 Ni2 ⊥ Bz Ni2 4sσ* π* δ* σ* δ π 4sσ σ 4sσ* π* δ δ* 4sσ σ* π* δ π σ π σ 4sσ* π* π* δ* σ* 4sσ π δ σ π σ Bz *,π π * 5 4 π 2,π 3 σ π 1 Ru2 5sσ* π* π* σ* 5sσ δ* δ π σ π σ ⊥ Bz 5sσ* π* π* σ* δ* 5sσ δ π σ π σ 5sσ* σ* π* δ* 5sσ δ π σ Ru2 5sσ* π* σ* δ* 5sσ δ π σ ] V e [ y g r e n e n o r t c e l e - e l g n i s 0 -2 -4 -6 -8 -10 4sσ* δ* σ* δ π* 4sσ π σ FIG. 6: (Color online) Scalar-relativistic single-particle levels of Ni2 (left panel), Ni2Bz (ground- state structure, second panel), benzene (third panel), Ru2Bz (ground-state structure, fourth panel), and Ru2 (right panel). All energies refer to a common vacuum level. Black lines denote occupied states, orange (gray) lines denote empty states, and thick blue (gray) lines indicate singly occupied two-fold degenerate states. With the exception of benzene, the levels are spin-split (S = 1 in all cases but Ru2, where S = 2). Majority states are indicated by up-arrows, minority states by down- arrows. Dimer-dominated states are labeled in black and benzene-dominated states are labeled in red (gray). 26 (i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi) (xii) (xiii) (xiv) FIG. A1: (Color online) Illustration of 14 initial structures optimized by the ESPRESSO code. From (i) to (x), top-views of possible parallel adsorption modes; from (xi) to (xiii), side-views of possible upright adsorption modes are shown; (xiv) is a case in which one Co atom is attached on each side of the carbon ring. 27 TABLE I: Dimer adsorption energies, Ead, total and atom-resolved spin magnetic moments, µS(total), µS(TM1), and µS(TM2) (TM1 refers to the atom closer to the benzene in case of per- pendicular bonding, TM2 to the other atom), the distance between the two transition-metal atoms dTM−TM, and the distance between benzene plane and TM1, dTM−Bz, for the ground-state struc- tures of TM2Bz (TM = Fe, Co, Ni, Ru, Rh, Pd) complexes. C-C and C-H bond lengths are fixed (1.40 A and 1.09 A, respectively). The structure type of each molecule is labelled according to the notation introduced in Fig. 1. Our present results are labelled "a", literature data (for dimers on graphene) are labelled "b", Ref. 40 and "c", Ref. 38. system Fe2Bz Co2Bz Ni2Bz Ru2Bz Rh2Bz Pd2Bz a ⊥c b ⊥c a ⊥c b ⊥c a ⊥c b ⊥c a ⊥c a kb a kb c kb 0.87 0.72 1.39 0.92 1.12 0.96 1.40 0.97 1.16 1.28 structure Ead(eV) µS(total)(µB) 4 6 µS(TM1)(µB) 0.75 2.76 µS(TM2)(µB) 3.35 3.48 dTM−TM(A) 2.04 2.08 dTM−Bz(A) 1.60 1.86 4 1.64 2.45 2.09 1.66 4 1.66 2.43 2.03 1.72 2 0.71 1.29 2.15 1.71 2 2 0.73 −0.10 1.29 2.14 1.73 2.13 2.22 1.77 2 1.01 1.01 2.49 2.08 0 0.00 0.00 2.80 2.14 0 0.00 0.00 2.75 2.15 28 TABLE II: Spin moments µS and orbital moments µL (in µB) for the ground-state structures of TM2Bz (TM = Fe, Co, Ni, Ru) complexes calculated within the fully relativistic scheme with mag- netization perpendicular (⊥) or parallel (k) to the benzene plane. The effect of the OP correction is also illustrated by comparing the values calculated without the OP correction (SO) and with the OP correction (SO+OP). Fe2Bz Co2Bz Ni2Bz Ru2Bz SO SO+OP SO SO+OP SO SO+OP SO SO+OP µ⊥ S(TM1) µk S(TM1) 0.75 0.75 µ⊥ L(TM1) −0.72 k L(TM1) 0.02 µ µ⊥ S(TM2) µ k S(TM2) 3.34 3.34 µ⊥ L(TM2) −0.17 k L(TM2) 0.11 µ µ⊥ L(total) −0.89 k L(total) 0.13 µ 0.73 0.75 −0.89 0.05 3.36 3.34 −0.17 0.25 −1.06 0.30 1.64 1.64 0.07 0.13 2.45 2.45 1.93 0.17 2.00 0.30 1.64 1.64 −0.10 0.39 2.46 2.45 2.12 0.54 2.02 0.93 0.71 0.71 0.01 0.04 1.28 1.28 0.01 0.37 0.02 0.41 0.71 0.67 0.02 −0.31 1.28 1.33 0.00 1.69 0.02 1.38 −0.09 −0.10 −0.09 −0.09 0.04 0.03 2.13 2.10 1.91 0.11 1.95 0.14 −0.04 0.02 2.13 2.11 2.01 0.21 1.97 0.23 29 TABLE III: The MAE (per molecule), calculated using Eq. (2) for the ground-state structures of TM2Bz (TM = Fe, Co, Ni, Ru). Positive values of MAE indicate that the easy axis of the system is perpendicular to the benzene plane, while negative values mean that the direction parallel to the benzene plane is the easy axis. Both a lower estimate of the MAE calculated without OP correction (MAE(SO)) and an upper estimate of the MAE obtained with OP correction (MAE(SO+OP)) are listed. Data in brackets indicate estimates obtained by first-order perturbation theory, see text. Further, the principal composition C 2 m,d−TMi + C 2 −m,d−TMi, the magnetic quantum number m, and the occupation of the HOMO are given as well as the spin-orbit coupling parameter ξd and the Racah parameter B. Fe2Bz Co2Bz Ni2Bz Ru2Bz −7 −96 21% 77% 2 2 96 154 +104 [+123] +279 [+403] 2% 94% 2 1 128 95 MAE(SO) (meV) +15 [+25] +51 [+74] MAE(SO+OP) (meV) +61 [+107] +334 [+519] C 2 C 2 m,d−TM1 + C 2 m,d−TM2 + C 2 −m,d−TM1 −m,d−TM2 m of the HOMO occupation of the HOMO ξd (meV) B (meV) 68% 15% 1 1 61 140 3% 94% 2 1 76 149 30 Acknowledgments Discussions with Helmut Eschrig and with Hway Chuan Kang are gratefully acknowl- edged. 31 1 P. Gambardella, S. Rusponi, M. Veronese, S. S. Dhesi, C. Grazioli, A. Dallmeyer, I. Cabria, R. Zeller, P. H. Dederichs, K. Kern, C. Carbone, and H. Brune, Science, 300, 1130 (2003). 2 S. Accorsi, A.-L. Barra, A. Caneschi, G. Chastanet, A. Cornia, A. C. Fabretti, D. Gatteschi, C. Mortal`o, E. Olivieri, F. Parenti, P. Rosa, R. Sessoli, L. Sorace, W. Wernsdorfer, and L. Zobbi, J. Am. Chem. Soc., 128, 4742 (2006). 3 C. J. Milios, A. Vinslava, W. Wernsdorfer, S. Moggach, S. Parsons, S. P. Perlepes, G. Christou, and E. K. Brechin, J. Am. Chem. Soc., 129, 2754 (2007). 4 M. Mannini, F. Pineider, P. Sainctavit, C. Danieli, E. Otero, C. Sciancalepore, A. M. Talarico, M.-A. Arrio, A. Cornia, D. Gatteschi, and R. Sessoli, Nature Mater., 8, 194 (2009). 5 P. Gambardella, S. Stepanow, A. Dmitriev, J. Honolka, F. M. F. de Groot, M. Lingenfelder, S. S. Gupta, D. D. Sarma, P. Bencok, S. Stanescu, S. Clair, S. Pons, N. Lin, A. P. Seitsonen, H. Brune, J. V. Barth, and K. Kern, Nature Mater., 8, 189 (2009). 6 H. Stillrich, C. Menk, R. Fromter, and H. P. Oepen, J. Appl. Phys., 105, 07C308 (2009). 7 P. Ravindran, A. Kjekshus, H. Fjellvag, P. James, L. Nordstrom, B. Johansson, and O. Eriksson, Phys. Rev. B, 63, 144409 (2001). 8 T. O. Strandberg, C. M. Canali, and A. H. MacDonald, Nature Mater., 6, 648 (2007). 9 Y. Mokrousov, N. Atodiresei, G. Bihlmayer, S. Heinze, and S. Blugel, Nanotechnology, 18, 495402 (2007). 10 A. Smogunov, A. D. Corso, A. Delin, R. Weht, and E. Tosatti, Nature Nanotechnology, 3, 22 (2008). 11 D. Fritsch, K. Koepernik, M. Richter, and H. Eschrig, J. Comp. Chem., 29, 2210 (2008). 12 M. E. Gruner, G. Rollmann, P. Entel, and M. Farle, Phys. Rev. Lett., 100, 087203 (2008). 13 A. M. Conte, S. Fabris, and S. Baroni, Phys. Rev. B, 78, 014416 (2008). 14 R. Xiao, D. Fritsch, M. D. Kuz'min, K. Koepernik, H. Eschrig, M. Richter, K. Vietze, and G. Seifert, Phys. Rev. Lett., 103, 187201 (2009). 15 H. Zhang, M. Richter, K. Koepernik, I. Opahle, F. Tasn´adi, and H. Eschrig, New J. Phys., 11, 043007 (2009). 16 S. H. Charap, P.-L. Lu, and Y. He, IEEE Trans. Magn., 33, 978 (1997). 17 H. Brooks, Phys. Rev., 58, 909 (1940). 32 18 L. Fern´andez-Seivane and J. Ferrer, Phys. Rev. Lett., 99, 183401 (2007). 19 T. O. Strandberg, C. M. Canali, and A. H. MacDonald, Phys. Rev. B, 77, 174416 (2008). 20 D. J. Trevor, R. L. Whetten, D. M. Cox, and A. Kaldor, J. Am. Chem. Soc., 107, 518 (1985). 21 D. W. Ball, Z. H. Kafafi, R. H. Hauge, and J. L. Margrave, J. Am. Chem. Soc., 108, 6621 (1986). 22 W. Zheng, S. N. Eustis, X. Li, J. M. Nilles, O. C. Thomas, K. H. Bowen, and A. K. Kandalam, Chem. Phys. Lett., 462, 35 (2008). 23 X. Li, S. Eustis, K. H. Bowen, A. K. Kandalam, and P. Jena, J. Chem. Phys., 129, 074313 (2008). 24 M. Gerhards, O. C. Thomas, J. M. Nilles, W.-J. Zheng, and K. H. Bowen, J. Chem. Phys., 116, 10247 (2002). 25 A. K. Kandalam, P. Jena, X. Li, S. N. Eustis, and K. H. Bowen, J. Chem. Phys., 129, 134308 (2008). 26 A. K. Kandalam, B. Kiran, P. Jena, X. Li, A. Grubisic, and K. H. Bowen, J. Chem. Phys., 126, 084306 (2007). 27 W. Zheng, J. M. Nilles, O. C. Thomas, and K. H. Bowen, J. Chem. Phys., 122, 044306 (2005). 28 C. Berg, M. Beyer, T. Schindler, G. Niedner-Schatteburg, and V. E. Bondybey, J. Chem. Phys., 104, 7940 (1996). 29 G. Luttgens, N. Pontius, C. Friedrich, R. Klingeler, P. S. Bechthold, M. Neeb, and W. Eber- hardt, J. Chem. Phys., 114, 8414 (2001). 30 S. F. Parker, J. Phys. Chem. A, 114, 1657 (2010). 31 L. Senapati, S. K. Nayak, B. K. Rao, and P. Jena, J. Chem. Phys., 118, 8671 (2003). 32 B. K. Rao and P. Jena, J. Chem. Phys., 117, 5234 (2002). 33 B. K. Rao and P. Jena, J. Chem. Phys., 116, 1343 (2002). 34 D. Majumdar, S. Roszak, and K. Balasubramanian, J. Chem. Phys., 107, 408 (1997). 35 G. M. Wang, J. J. BelBruno, S. D. Kenny, and R. Smith, Surf. Sci., 541, 91 (2003). 36 R. Varns and P. Strange, J. Phys.: Condens. Matter, 20, 225005 (2008). 37 D. M. Duffy and J. A. Blackman, Phys. Rev. B, 58, 7443 (1998). 38 I. Cabria, M. J. L´opez, and J. A. Alonso, Phys. Rev. B, 81, 035403 (2010). 39 O. Loboda, V. R. Jensen, and K. J. Børve, Fullerenes, Nanotubes, and Carbon Nanostructures, 14, 365 (2006). 33 40 H. Johll, H. C. Kang, and E. S. Tok, Phys. Rev. B, 79, 245416 (2009). 41 C. Cao, M. Wu, J. Jiang, and H.-P. Cheng, Phys. Rev. B, 81, 205424 (2010). 42 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett., 77, 3865 (1996). 43 K. Koepernik and H. Eschrig, Phys. Rev. B, 59, 1743 (1999). 44 http://www.fplo.de. 45 J. P. Perdew and Y. Wang, Phys. Rev. B, 45, 13244 (1992). 46 R. Pandey, B. K. Rao, P. Jena, and J. M. Newsam, Chem. Phys. Lett., 321, 142 (2000). 47 R. Pandey, B. K. Rao, P. Jena, and M. A. Blanco, J. Am. Chem. Soc., 123, 3799 (2001). 48 M. R. Philpott and Y. Kawazoe, Chem. Phys., 342, 223 (2007). 49 P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, A. D. Corso, S. de Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A. P. Seitsonen, A. Smogunov, P. Umari, and R. M. Wentzcovitch, J. Phys.: Condens. Matter, 21, 395502 (2009). 50 H. Eschrig, M. Sargolzaei, K. Koepernik, and M. Richter, Europhys. Lett., 72, 611 (2005). 51 O. Eriksson, M. S. S. Brooks, and B. Johansson, Phys. Rev. B, 41, 7311 (1990). 52 J. Trygg, B. Johansson, O. Eriksson, and J. M. Wills, Phys. Rev. Lett., 75, 2871 (1995). 53 L. Nordstrom, M. S. S. Brooks, and B. Johansson, J. Phys.: Condens. Matter, 4, 3261 (1992). 54 Gaussian 03, Revision C.02, M. J. Frisch et al., Gaussian, Inc., Wallingford CT, 2004. 55 M. Dolg, U. Wedig, H. Stoll, and H. Preuss, J. Chem. Phys., 86, 866 (1987). 56 T. H. Dunning, Jr., J. Chem. Phys., 90, 1007 (1989). 57 S. F. Boys and F. Bernardi, Mol. Phys., 19, 553 (1970). 58 S. Simon, M. Duran, and J. J. Dannenberg, J. Chem. Phys., 105, 11024 (1996). 59 T. Kurikawa, H. Takeda, M. Hirano, K. Judai, T. Arita, S. Nagao, A. Nakajima, and K. Kaya, Organometallics, 18, 1430 (1999). 60 J. R. Lombardi and B. Davis, Chem. Rev., 102, 2431 (2002). 61 J. Ho, M. L. Polak, K. M. Ervin, and W. C. Lineberger, J. Chem. Phys., 99, 8542 (1993). 62 P. B lo´nski and J. Hafner, Phys. Rev. B, 79, 224418 (2009). 63 T. Yasuike and S. Yabushita, J. Phys. Chem. A, 103, 4533 (1999). 64 P. Bruno, Phys. Rev. B, 39, 865 (1989). 34 65 B. R. Judd, Operator Techniques in Atomic Spectroscopy (McGraw-Hill, New York, 1963). 66 N. Marzari, D. Vanderbilt, A. De Vita, and M. C. Payne, Phys. Rev. Lett., 82, 3296 (1999). 35 

1903.03476

1

1903

Quantised conductance of one-dimensional strongly-correlated electrons in an oxide heterostructure

Oxide heterostructures are versatile platforms with which to research and create novel functional nanostructures. We successfully develop one-dimensional (1D) quantum-wire devices using quantum point contacts on MgZnO/ZnO heterostructures and observe ballistic electron transport with conductance quantised in units of 2e^{2}/h. Using DC-bias and in-plane field measurements, we find that the g-factor is enhanced to around 6.8, more than three times the value in bulk ZnO. We show that the effective mass m^{*} increases as the electron density decreases, resulting from the strong electron-electron interactions. In this strongly interacting 1D system we study features matching the 0.7 conductance anomalies up to the fifth subband. This paper demonstrates that high-mobility oxide heterostructures such as this can provide good alternatives to conventional III-V semiconductors in spintronics and quantum computing as they do not have their unavoidable dephasing from nuclear spins. This paves a way for the development of qubits benefiting from the low defects of an undoped heterostructure together with the long spin lifetimes achievable in silicon.

cond-mat.mes-hall

cond-mat

Quantised conductance of one-dimensional strongly-correlated electrons in an oxide heterostructure H. Hou,1 Y. Kozuka,2, 3, 4 Jun-Wei Liao,1 L. W. Smith,1 D. Kos,1 J. P. Griffiths,1 J. Falson,5 A. Tsukazaki,6 M. Kawasaki,2 and C. J. B. Ford1, ∗ 1Cavendish Laboratory, University of Cambridge, CB3 0HE, UK. 2Department of Applied Physics and Quantum-Phase Electronics Center (QPEC), The University of Tokyo, Tokyo 113-8656, Japan 3Research Center for Magnetic and Spintronic Materials, National Institute for Materials Science (NIMS), 1-2-1 Sengen, Tsukuba 305-0047, Japan 4JST, PRESTO, Kawaguchi, Saitama 332-0012, Japan 5Max Planck Institute for Solid State Research, D-70569 Stuttgart, Germany 6Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan Oxide heterostructures are versatile platforms with which to research and create novel functional nanostructures. We successfully develop one-dimensional (1D) quantum-wire devices using quantum point contacts on MgZnO/ZnO heterostructures and observe ballistic electron transport with con- ductance quantised in units of 2e2/h. Using DC-bias and in-plane field measurements, we find that the g-factor is enhanced to around 6.8, more than three times the value in bulk ZnO. We show that the effective mass m∗ increases as the electron density decreases, resulting from the strong electron- electron interactions. In this strongly interacting 1D system we study features matching the '0.7' conductance anomalies up to the fifth subband. This paper demonstrates that high-mobility oxide heterostructures such as this can provide good alternatives to conventional III-V semiconductors in spintronics and quantum computing as they do not have their unavoidable dephasing from nuclear spins. This paves a way for the development of qubits benefiting from the low defects of an undoped heterostructure together with the long spin lifetimes achievable in silicon. I. INTRODUCTION Physical phenomena in transition-metal oxides and their complex compounds have stimulated intense inter- est in research covering metallic, semiconducting, and insulating properties. At the heterointerface of two ox- ide layers, the symmetry breaking leads to novel proper- ties including quantum confinement of electrons, strong correlations, superconductivity, and ferromagnetism.1 In a MgZnO/ZnO heterostructure, the polarisation mis- match between MgZnO and ZnO originating from spon- taneous and piezoelectric contributions induces a two- dimensional electron gas (2DEG) at the interface.2 By engineering the strain via Mg composition and the MgZnO thickness, the 2DEG density can reach down to 1011 cm−2 with mobility over 106 cm2V−1s−1.3 Both integer and fractional quantum-Hall effects have been investigated.2,4,5 Furthermore, the weak spin-orbit in- teraction and the low concentration of nuclear spins in ZnO lead to a long spin-coherence time.6 These unique properties in the MgZnO/ZnO heterostructure create an excellent platform for investigating quantum physics be- yond the more conventional III-V alternatives.7,8 More recently, Andreev reflection has been demonstrated at the interface between a MgZnO/ZnO heterostructure and MoGe superconductor, and this could be a good system for investigating non-abelian quasiparticles, such as Ma- jorana fermions.9 Most low-dimensional devices use gates to define nanostructures such as one-dimensional (1D) quantum point contacts (QPCs)/quantum wires or quantum dots. However, gating oxide heterostructures is a challenge. In a MgZnO/ZnO heterostructure, we have found that us- ing standard insulators such as Al2O3 causes a reduction in mobility and strong parallel conduction. The latter may be because the hard Al2O3 deposited on the thin stressed MgZnO layer contributes to a mismatch of the spontaneous and piezoelectric polarisations, and induces another 2DEG at the Al2O3/MgZnO interface. So far it has been necessary to use one-off techniques to create nanostructures, such as conducting AFM lithography on LaAlO3/SrTiO3, which showed quantised conductance in units of e2/h in a strong magnetic field.10 -- 12 In our work, we have successfully prevented parallel conduction by re- placing hard Al2O3 insulator with soft parylene C. The zero-field quantisation of conductance in inte- ger multiples of 2e2/h is a signature of ballistic charge transport in 1D systems. The lateral electrostatic con- finement creates a series of 1D subbands, in which spin-up (↑) and spin-down (↓) subbands each con- tribute e2/h. This is already observed in many ma- including GaAs/AlGaAs,13,14 InGaAs/InAlAs terials, heterostructures,15 strained epitaxial germanium16 and carbon-based materials.17,18 An anomalous feature at conductance G = 0.7 × 2e2/h was first investigated by Thomas et al. and attributed to a possible spontaneous spin polarisation.19,20 Its origin has since been much debated,21 and other explanations proposed including quasi-bound-state formation and the Kondo effect.22,23 Recently, Bauer et al. used a smeared van Hove singular- ity to explain it and emphasised the important role that electron-electron interactions play in the 0.7 anomaly.24 Here we report ballistic transport in a high-quality MgZnO/ZnO heterostructure and show well-defined con- ii high m∗ (= 0.3 me in bulk ZnO, where me is the bare electron mass) and small dielectric constant ε = 8.3 compared to GaAs. The ratio rs = EC/EF can reach 10 in low-density wafers (EC is the average Coulomb energy per electron and EF is the Fermi energy). These strong interactions may be the source of 'N.7' structures in 1D subbands with higher index N apparent in our data, which behave like the 0.7 structure. Any such N.7 structures28 -- 31 are very weak in GaAs electron or hole systems. We confirm the importance of electron-electron interactions in the 0.7 anomaly with an independent measurement of the strength of the electron-electron interaction from the electron effective mass.24 II. EXPERIMENTAL DETAILS The MgZnO/ZnO heterostructures are grown by molecular-beam epitaxy.3 Devices A and B (C) have a 2DEG 92 nm (500 nm) below the surface, den- sity 3.2 × 1011 cm−2 (1.2 × 1012 cm−2) and mobility 3.5 × 105 cm2V−1s−1 (6 × 104 cm2V−1s−1), correspond- ing to an electron mean free path of le = 3.2 µm (1.1 µm). A scanning electron micrograph and schematic cross- section of a device are shown in Fig. 1(a). A Hall- bar mesa is patterned by Ar ion milling and Ti/Au is thermally evaporated to create Ohmic contacts without annealing. Ti/Au split gates of length L = 200 nm and width W = 300 nm are deposited on a 30 nm-thick parylene-C insulator layer, forming a quasi-1D wire in the 2DEG. We measure conductance through the QPC using a four-terminal lock-in technique with excitation voltage 10 µV at 77 Hz, at a temperature of ∼50 mK. For per- pendicular magnetic-field measurements we measure the diagonal resistance to obtain the number of transmitted Landau-level edge states.32 III. RESULTS AND DISCUSSION A. Quantised 1D conductance Figure 1(b) shows the 1D conductance vs split-gate voltage (VSG). At VSG = −1.5 V, electrons below the gate are depleted and a quasi-1D wire is defined in the gap [inset, Fig. 1(b)]. This definition voltage matches the expected value calculated using a parallel-plate ca- pacitor model with this 2DEG density, dielectric constant and thickness. 1D conductance plateaus appear as VSG is made more negative, quantised in units of ∆G ≈ 2e2/h. They are visible up to 14 e2/h (devices A and B). For G < 2e2/h, Coulomb-blockade (CB) peaks appear, prob- ably because dots form owing to the reduced electron screening increasing disorder, so we will discuss results from the second plateau and above. Figure 1(c) shows the transconductance dG/dVSG vs source-drain bias VDC (devices A and B). Dark FIG. 1. (a) Scanning electron micrograph of a QPC (source S and drain D), and schematic diagram of the device cross-section across the channel. The 2DEG forms at the MgZnO/ZnO interface (red line). Ti/Au gates are patterned using electron-beam lithography and deposited on parylene- C insulator. (b) Conductance G as a function of split-gate voltage VSG (devices A and B), showing plateaus at multiples of 2e2/h. Inset: G over the full VSG range. The drop cor- responds to depletion of electrons beneath the gates leaving a quasi-1D wire in the gap. (c) Transconductance dG/dVSG vs bias VDC and VSG (devices A and B). Dark (bright) re- gions correspond to plateaus (risers) in conductance. The energy difference between the third and fourth 1D subbands is e∆VDC, which is measured from the crossing point of adja- cent subbands as indicated in the plot. The numbers indicate heights of conductance plateaus (units e2/h). ductance quantisation. Using DC-bias spectroscopy25 and in-plane magnetic-field measurements, we find a g- factor in the 1D wire that is enhanced by a factor of ∼ 3 compared to the bulk and is fairly constant at low 1D subband index. Additionally, we show that the effective mass m∗ increases as the density decreases in our 1D sys- tem, as occurs for 2DEGs in similar heterostructures.26,27 system in MgZnO/ZnO heterostructures offers a novel plat- form to investigate interaction effects, particularly the 0.7 structure. These strong correlations arise from the correlated electron The strongly -4-3.5-3-2.5VSG(V)02468101214G(e2/h)Device ADevice B-4-2002000.7 anomalies?(a) (b)(c)iii (light) regions correspond to plateaus (transitions be- tween plateaus). The splitting of the source and drain chemical potentials causes the risers to split until quan- tised plateaus appear between them at odd-integer values G = M e2/h (M = 3, 5, 7 . . . ), giving diamond-shaped features in Fig. 1(c).25 The 1D subband energy spacing ∆E is given by the size of the diamond ∆VDC times e, and is around 0.4 meV here. Surprisingly, this remains reasonably constant as the 1D subband index decreases. Because of the high m∗ and hence relatively small sub- band spacing, the plateaus are strongly temperature de- pendent, disappearing for T (cid:38) 1 K (Fig. S1 in Supple- mental Material (SM)). B. Magnetic-field dependence & Enhancement of g-factor In an in-plane field, the quantised plateaus at even multiples of e2/h split due to the Zeeman energy EZ (Fig. 2(a)). At B(cid:107) = 0.5 T, plateaus occur at G = ne2/h, (n = 3, 4, 5 . . .), when a 1D subband of the lower-energy spin-polarisation direction becomes fully transmitted be- fore the subband with opposite spin. As B(cid:107) increases further to 1 T, the spin-split 1D subbands cross, leaving plateaus only at G = M e2/h (M = 3, 5, 7 . . .). From EZ and the subband spacing, we estimate the g-factor as25 g∗ = 1 µB δE δVSG δVSG δB = e µB ∆VDC ∆B , (1) with e the electronic charge and µB the Bohr magne- ton. We estimate g∗ ≈ 6.8 (6.4) for the second (fifth) subband, enhanced above the bulk value g∗ ≈ 2 for het- erostructure and bulk ZnO.6 An enhancement of g∗ by a similar factor occurs for electrons in GaAs QPCs, with values from 0.75 to 1.5,19,33,34 compared to 0.44 in the bulk. For electrons in GaAs QPCs, g∗ decreases fairly rapidly with subband index,28 whereas in our data g∗ is almost constant, as the Zeeman splittings of subbands are almost identical [Fig. 2(a)]; the gradient of subband edges and the field at which subbands cross are very sim- ilar for subbands 2 to 6. For a field applied perpendicular to the 2DEG (B⊥), the electron energy contains Zeeman (EZ = g∗µBB) and cyclotron-energy (Ec = ωc) terms: (cid:18) (cid:19) (cid:113) 1 2 EN(B) = N + ω2 0 + ω2 c ± 1 2 g∗µBB, (2) (N is subband index, ω0 subband spacing assuming a parabolic potential, and ωc = eB⊥/m∗ the cyclotron frequency).35 In the low-field regime (ω0 (cid:29) ωc), spin-degenerate plateaus at even multiples of G = e2/h are split by EZ, leading to additional plateaus at G = ne2/h (n = 3, 4, 5 . . .). These merge to odd multiples of G = e2/h with increasing B⊥ as adjacent spin-split subbands cross. However, for high B⊥, Landau-level formation leads to The differential conductance dG/dVSG (units FIG. 2. 2e2/h/V ) vs VSG at different (a) B(cid:107), and (b) and (c) B⊥ for devices as labelled. ∆B indicates the required B(cid:107) for the crossing of subbands 2↑ and 3↓. Ellipses in (b) indicate po- sitions of subband crossings. The calculated electron energy spectrum vs B⊥ with (d) constant and (e) increasing m∗. The numbers indicate the heights of conductance plateaus (units e2/h), and blue dots mark positions of crossings. the creation of hybrid magneto-electric subbands in the constriction,13,14,35,36 for which plateaus again occur at both even and odd integers. The onset of this regime can be quantified by the ratio κ = ωc/EZ = e/(g∗m∗µB). We estimate m∗ = (0.36 ± 0.02) me from temperature- dependent Shubnikov -- de-Haas measurements and the Dingle formula37 (Fig. S2 in SM). This gives κ < 1 for g∗ = 6.8. The smaller m∗ = 0.067 me and g-factor (≈ 1) in GaAs devices leads to κ (cid:29) 1, so odd-integer plateaus are not observed. Figures 2(b) and (c) show the evolution of subbands with B⊥. Bright regions correspond to risers in conduc- tance between plateaus, where subband edges cross the Fermi energy. Plateau heights are labelled. We model the subband energy vs B using Eq. 2. Results for con- stant m∗ = 0.4 me, g∗ and ω0 are plotted in Fig. 2(d), black and red lines representing spin-down (↓) and spin- up (↑) subbands, respectively. The pattern of plateaus matches the experiment well. However, the value of B⊥ at which subbands cross is independent of subband in- dex, which does not match the measurements. In Fig. 2(b), the crossing between N = 2↑ and N = 3↓ subbands occurs around B⊥ = 0.8 T, and between N = 3↑, 4↓ around B⊥ = 1.1 T. For subbands N > 4, the required B⊥ decreases. Given that m∗ in ZnO increases with decreasing density,26,27 we repeat the calculation with m∗ increas- ing from 0.4 to 1 me as the subband index decreases, (Fig. 2(e)). This reproduces the trend from experimental data that the value of B⊥ at which spin-split subbands cross initially increases, then decreases for higher sub- bands (although the model tends to overestimate B⊥, or perhaps experimental crossing points are underestimated due to energy blurring). We vary m∗ instead of g∗ since our data suggests that g∗ is reasonably independent of subband index, unlike in GaAs. In addition, DC-bias spectroscopy (Fig. 1(c)) indicates a reasonably constant subband spacing over this range. Experimentally, the precise points at which bands cross at low index cannot be determined, and some look more like anticrossings (la- belled γ, Fig. 2(c)). This may be because of a strong electron-electron interaction, and could be modelled by introducing an exchange interaction term in Eq. 2.38,39 C. Effective mass measurements 0 + ω2 subbands is ∆E = (cid:112)ω2 To investigate m∗ further we measure the DC-bias de- pendence at B⊥ = 1 T (device A) (Fig. 3(a)). Adjacent spin-split subbands cross near this value of B⊥, indicated by only odd plateaus being present for VDC = 0. Since the subband spacing is roughly constant, each pair of sub- bands 2↑/3↓, 3↑/4↓ etc is degenerate since spin ↑/↓ sub- bands are shifted by +/− 1 2 gµBB, respectively, cancelling out EZ. The energy difference between these pairs of c . In contrast to the B = 0 case, at B⊥ = 1 T the spacing between subband pairs de- creases at lower subband index N (Fig. 3(a)). This could be explained by increasing m∗ at lower density, leading to a smaller Ec = eB/m∗ for lower N . Fig. 3(b) shows the measurement repeated for device C with different n2D and device dimensions (L = 300 nm and W = 800 nm), in which more plateaus are evident. The same trend of increasing spacing with subband index occurs. Figure 3(c) shows m∗ vs plateau height (units e2/h), estimated using ∆E. At high conductance (∼ 13e2/h), m∗ = (0.31 ± 0.03) me, close to the bulk effective mass found above for this wafer. When G decreases to 5e2/h, m∗ increases to (0.96 ± 0.2) me, which is comparable to that of a 2D system at a lower density of 1 × 1011 cm−2 iv FIG. 3. Transconductance dG/dVSG as a function of VSG and VDC for (a) device A at B⊥ = 1 T, and (b) device C at B⊥ = 2 T. In the device C, the high electron density and thick MgZnO (500 nm) require strong negative voltages to define the 1D wire. The plateau conductances are indicated (units e2/h). (c) m∗ measured vs G in both devices, compared with the bulk value for ZnO. (0.8± 0.2) me.27 The large error bar is due to the blurred nature of this subband crossing. However, the trend of increasing m∗ as G decreases is clear. In previous studies of MgZnO/ZnO heterostructures, m∗ measured from temperature-dependent Shubnikov -- de-Haas oscillations increases as the 2DEG density de- creases, while m∗ from cyclotron resonance is roughly constant.26,27 This indicates that the increase in trans- port effective mass arises from electron-electron interac- tions, which are more significant at lower density. A sim- ilar effect is observed in other 2DEG systems,40,41 but much weaker (a factor of ∼ 1.4 rather than 3 as in these ZnO heterostructures). In 1D GaAs wires, an increase in m∗ by at most 30% was reported when the 1D density decreased from 2.6 × 108 m−1 to 1 × 108 m−1.42 However, the ratio of electron-electron interaction energy to kinetic energy in GaAs is relatively low, and non-parabolicity, disorder and electron-phonon interactions may also con- tribute significantly to this increase.43 D. 0.7 anomaly We now return to our discussion on the 0.7 anomaly. Fig. 1(b) shows several shoulder-like features below the 5791113G(e2/h)0.20.40.60.811.2m*/meDevice A: Lown2DDevice C: Highn2Dm*bulk= 0.3me(c)v prehensive, they give a strong indication that these struc- tures belong to the 0.7 family, and the great similari- ties with the complex magnetic-field behaviour in GaAs are striking. An additional test is the temperature de- pendence. Because of the small 1D subband spacing in our samples, any structure is rapidly smeared by tem- perature, disappearing by T > 1 K (Fig. S1 in SM). N.7 structures are more visible in QPCs in MgZnO/ZnO compared to GaAs heterostructures because of the large effective mass and small dielectric constant leading to strong electron-electron interactions. From the cyclotron energy, we estimated the strength of electron-electron interactions using the electron effective mass. The in- creasing interaction energy relative to the kinetic energy explains why N.7 structures become better defined as the 1D subband index decreases, which is consistent with the model.24 MgZnO/ZnO heterostructures show dilute ferromagnetic properties49 (see SM) that may help to enhance the local spin susceptibility in the channel at B = 0.24 IV. CONCLUSION To conclude, we have shown ballistic electron transport with conductance quantisation in 1D quantum wires de- fined on a MgZnO/ZnO heterostructure. We also find an increasing effective mass at lower density, consistent with measurements on 2D ZnO systems. Because of the large g∗ and m∗, a perpendicular field drives the system into a regime where the Zeeman energy is greater than the cyclotron energy, leading to only odd plateaus in the con- ductance. At zero field we see evidence of 0.7-like anoma- lies up to the fifth 1D subband. Such structures are rarely observed in GaAs, a key reason for which is probably the significantly higher interaction strength in ZnO. The bal- listic transport and the importance of interactions and spin, together with a long spin-coherence time owing to the low concentration of nuclear spins in ZnO, could make high-quality MgZnO/ZnO heterostructures an interest- ing alternative to III-V semiconductors as platforms for quantum information and spintronic technologies. ACKNOWLEDGMENTS We thank Stuart Holmes for helpful discussions. H. Hou acknowledges the Chinese Scholarship Council and Cambridge Trust for financial support. This work was partly supported by JST, PRESTO Grant Num- ber JPMJPR1763 and JST, CREST Grant Number JP- MJCR16F1, Japan. FIG. 4. Conductance G vs VSG at different B(cid:107) from 0 to 1 T in steps of 0.02 T. Each trace is shifted to the right by 0.01 V relative to the previous trace for clarity. Red dots illustrate how the shoulder features resembling the 0.7 anomaly evolve with B(cid:107) (same measurement data and labels as in Fig. 2(a)). An estimated series resistance of around 110 Ω has been sub- tracted. main plateaus. We test whether they behave similarly to the 0.7 anomaly or to CB-like resonances from impurities. i) Resonant peaks from CB should split with VDC at a rate determined by the size of the dot-like impurity and coupling to the gates, but DC-bias spectroscopy for our devices shows an orderly splitting above the second plateau typical of clean 1D devices,44,45 [Fig. 1(c)]. ii) In Figs. 2(a-c), even low B lifts spin-degeneracy. The edges of spin-split subbands do not meet at B = 0, showing clear gaps [for example, β0 in Fig. 2(a)] at the plateau, as seen for the 0.7 anomaly in GaAs.19 Fig. 2(a) also shows gaps at higher-order crossings (labelled α2, β1, β2 matching labels in Ref. 46), which is an important sign of the 0.7 analogue.46 The gaps can be explained as an effect of interactions.47 iii) The conductance sweeps in Fig. 2(a) are re-plotted (Fig. 4) as lines up to B(cid:107) =1 T. The N.7 plateaus (just below G = 2(N + 1)e2/h) appear to evolve smoothly to spin-polarised plateaus at G = 2(N +1/2)e2/h, then they split to form an extra plateau [indicated by * in Fig. 2(a) and Fig. 4]. This split before the crossing was observed in GaAs,46 but was much weaker. How interactions con- tribute needs further theoretical investigation. In Fig. 4, the N.7 structures strengthen and occur lower on the riser for lower subbands, consistent with more significant interactions24,48 due to the lower density. iv) Plateaus and N.7 structures stay reasonably con- stant as the wire is shifted laterally by asymmetric bias on the QPC gates (Fig. S3 in SM). The wire position should not significantly affect either quantisation or in- teraction effects inherent in thef 1D system such as the 0.7 structure. While tests i) to iv) described above are not fully com- vi the '1.7' structure, which is consistent with the behaviour of the 0.7 structure in GaAs, as shown by the arrows in Fig. 5. B. Measurement of electron effective mass FIG. 5. Conductance G as a function of split-gate voltage VSG for device B, as the temperature is increased from 0.05 K to 0.9 K (from left to right -- the red line indicates data obtained at the fridge base temperature of 0.05 K, and the black lines are for T from 0.1 K to 0.9 K in steps of 0.1 K). Traces are offset to the right for clarity. V. SUPPLEMENT MATERIAL A. Temperature-dependence of the conductance Fig. 5 shows the temperature dependence of the con- ductance G as a function of split-gate voltage VSG for device B. The step-like features in conductance are ther- mally smeared with increasing temperature. We first address the non-exact conductance quantisation at low T . Data are corrected for a constant series resistance (RS = 110 Ω) measured at zero gate voltage. This does not lead to a constant vertical spacing of 2e2/h between instead the spacing is less than 2e2/h near plateaus, pinch-off and increases to close to 2e2/h as G increases [also seen in Fig. 1(b) of main article]. This indicates that the series resistance increases as the conductance decreases. The correction overestimates the voltage drop across the device, and underestimates the device conduc- tance. Therefore, this simple correction does not manage to align all the quantised plateaus with exact integer mul- tiples of e2/h. The region near the wire will be affected by the gate voltage and its series resistance will increase as VSG goes more negative, reducing the electron den- sity and hence the elastic mean free path le because of reduced screening. le is lower than in GaAs because of the higher effective mass and may become short enough for collisions with impurities to occur within the region affected by the gate, making the series resistance change more than it does for electrons in GaAs. It is there- fore likely that the apparent suppression of the 4e2/h plateau is the result of an incorrect series resistance, and the plateau should be closer to 4e2/h. Moreover, le de- creases with increasing temperature, and this is likely to further increase the series resistance near the chan- nel, as seen in Fig. 5. These combined factors make temperature-dependent studies harder than in GaAs. Ig- noring the uncertainties in series-resistance calibration, the second plateau (N = 2) degrades much faster than FIG. 6. Temperature dependence of Shubnikov -- de-Haas os- cillations from T = 0.4 to 1.2 K. The inset shows the fitting of ln(∆Rxx/T ) as a function of T at fixed magnetic fields, and m∗ is estimated to be 0.36±0.02 me. We measure Shubnikov -- de-Haas oscillations in the 2DEG resistance Rxx in the Hall bar containing devices A and B, with all gates grounded. Then the amplitude of Shubnikov -- de-Haas oscillation ∆Rxx it fitted to the Dingle expression37 ∆Rxx ∝ χ sinh χ e−π/(ωcτq), (3) where χ = 2π2kBT /(ωc), ωc = eB/m∗, τq is the quan- tum scattering lifetime (which we assume is temperature- independent over this measurement range) and T is the temperature. The effective mass m∗ obtained from this fit is around 0.36 me in this MgZnO/ZnO heterostruc- ture, higher than it is in bulk ZnO (around 0.3 me), where me is the bare electron mass. In device C, which is fab- ricated from a MgZnO/ZnO heterostructure with high electron density, m∗ at VSG = 0 should be close to the bulk value. C. Asymmetric bias measurements Fig. 7 plots the conductance of the 1D wire in Device B with asymmetric bias on quantum point contacts. The quantised plateaus and N.7 structures stay reasonably constant as the wire is shifted laterally by an asymmetric bias on the QPC gates. In contrast, the resonant peaks due to CB below the first plateau vary significantly. CB- type features are not independent of position since the disorder potential changes as the wire moves relative to impurities. However, the wire position should not sig- nificantly affect either 1D quantisation or interaction ef- fects inherent in the 1D system such as the 0.7 structure -4-3.5-3VSG (V)0246810G (e2/h)0.05KDevice B0.9K00.20.40.60.81B (T)00.10.2Rxx (KΩ)T=0.4KT=0.5KT=0.7KT=0.9KT=1.2K0.40.81.2T (K)-20246ln(∆Rxx/T)m*= 0.36±0.02 mevii must come from a random impurity or dot, perhaps un- der a gate, as they are only very slightly dependent on B or VDC and so provide a conduction path in parallel to the 1D wire. (cid:82) VDC All DC-bias data shown in the main paper are cor- DC (1 − 0 GdVDC), where RS is the series resistance mea- SD is the DC bias ap- rected for series resistance using VDC = V appl RS sured at zero gate voltage, and V appl plied in the measurement.50 FIG. 7. Conductance G as function of VSG1 as an asymmetric bias is applied to the gates. The difference in voltage on the two split gates is defined by ∆VSG = VSG1 − VSG2 changes from 0.5 V to −0.5 V in steps of 50 mV (although small variations might occur if the asymmetry alters the wave-function localisation and therefore affects the strength of interactions and 0.7 structure24). D. Other details Note that straight lines in intensity plots in the main paper, such as the straight line at VSG = −2.9 V in Fig. 1(c) (device A) or near VSG = −2.6 V (device B) This MgZnO/ZnO heterostructure also shows dilute ferromagnetic properties with an anomalous Hall effect brought about by spin-dependent electron scattering off localised magnetic moments, which are likely to arise from point defects in epitaxial ZnO with localised un- paired electrons.49 This may increase the strength of 0.7-like structure51 because this generally appears to be strongly related to spin.19,24 However, the dilute ferro- magnetic moments cannot be strong enough to produce full spontaneous electron spin polarisation in the 1D wire, as only plateaus at even multiples of e2/h are observed at B = 0. Instead, the dilute ferromagnetism may possibly help to enhance the local spin susceptibility in the chan- nel at B = 0,24 making it easier for interactions to give rise to the shoulders on each plateau (the 'N .7' structure) seen, for example, as a pair of vertical lines in Fig. 2(b) in the main paper. ∗ cjbf@cam.ac.uk 1 P. Zubko, S. Gariglio, M. Gabay, P. Ghosez, and J.-M. Triscone, Annu. Rev. Condens. Matter Phys. 2, 141 (2011). 2 A. Tsukazaki, A. Ohtomo, T. Kita, Y. Ohno, H. Ohno, and M. Kawasaki, Science 315, 1388 (2007). 3 J. Falson, Y. Kozuka, M. Uchida, J. H. Smet, T.-h. Arima, A. Tsukazaki, and M. Kawasaki, Scientific reports 6, 26598 (2016). 4 A. Tsukazaki, S. Akasaka, K. Nakahara, Y. Ohno, H. Ohno, D. Maryenko, A. Ohtomo, and M. Kawasaki, Nature materials 9, 889 (2010). 5 J. Falson, D. Maryenko, B. Friess, D. Zhang, Y. Kozuka, and M. Kawasaki, Nature A. Tsukazaki, J. H. Smet, Physics 11, 347 (2015). 6 Y. Kozuka, S. Teraoka, J. Falson, A. Oiwa, A. Tsukazaki, S. Tarucha, and M. Kawasaki, Phys. Rev. B 87, 205411 (2013). 7 Y. Kozuka, A. Tsukazaki, and M. Kawasaki, Applied Physics Reviews 1, 011303 (2014). 8 J. Falson and M. Kawasaki, Reports on Progress in Physics 81, 056501 (2018). 9 Y. Kozuka, A. Sakaguchi, J. Falson, A. Tsukazaki, and et al., arXiv preprint arXiv:1611.05127 (2016). 13 D. A. Wharam, T. J. Thornton, R. Newbury, M. Pepper, H. Ahmed, J. E. F. Frost, D. G. Hasko, D. C. Peacock, D. A. Ritchie, and G. A. C. Jones, Journal of Physics C: solid state physics 21, L209 (1988). 14 B. J. van Wees, L. P. Kouwenhoven, H. van Houten, C. W. J. Beenakker, J. E. Mooij, C. T. Foxon, and J. J. Harris, Phys. Rev. B 38, 3625 (1988). 15 T. Bever, Y. Hirayama, and S. Tarucha, Japanese journal of applied physics 33, L800 (1994). 16 Y. Gul, S. N. Holmes, P. J. Newton, D. J. P. Ellis, C. Mor- rison, M. Pepper, C. H. W. Barnes, and M. Myronov, Appl. Phys. Lett. 111, 233512 (2017). 17 S. Frank, P. Poncharal, Z. L. Wang, and W. A. De Heer, science 280, 1744 (1998). 18 N. Tombros, A. Veligura, J. Junesch, M. H. Guimaraes, I. J. Vera-Marun, H. T. Jonkman, and B. J. Van Wees, Nature Physics 7, 697 (2011). 19 K. J. Thomas, J. T. Nicholls, M. Y. Simmons, M. Pepper, D. R. Mace, and D. A. Ritchie, Phys. Rev. Lett. 77, 135 (1996). 20 L. P. Rokhinson, L. N. Pfeiffer, and K. W. West, Phys. M. Kawasaki, preprint arXiv:1810.08198 (2018). Rev. Lett. 96, 156602 (2006). 10 C. Cen, S. Thiel, J. Mannhart, and J. Levy, Science 323, 21 A. P. Micolich, Journal of Physics: Condensed Matter 23, 1026 (2009). 443201 (2011). 11 M. Tomczyk, G. Cheng, H. Lee, S. Lu, A. Annadi, J. P. Veazey, M. Huang, P. Irvin, S. Ryu, C.-B. Eom, et al., Phys. Rev. Lett. 117, 096801 (2016). 12 A. Annadi, G. Cheng, S. Lu, H. Lee, J.-W. Lee, A. Tylan- Tyler, M. Briggeman, M. Tomczyk, M. Huang, D. Pekker, 22 S. M. Cronenwett, H. J. Lynch, D. Goldhaber-Gordon, L. P. Kouwenhoven, C. M. Marcus, K. Hirose, N. S. Wingreen, and V. Umansky, Phys. Rev. Lett. 88, 226805 (2002). 23 Y. Meir, K. Hirose, and N. S. Wingreen, Phys. Rev. Lett. -4-3.8-3.6-3.4-3.2-3VSG1 (V)02468G (e2/h)∆VSG=0.5VDevice B∆VSG=-0.5Vviii 47 K.-F. Berggren, P. Jaksch, and I. Yakimenko, Phys. Rev. B 71, 115303 (2005). 48 L. W. Smith, H. Al-Taie, A. A. J. Lesage, F. Sfigakis, P. See, J. P. Griffiths, H. E. Beere, G. A. C. Jones, D. A. Ritchie, A. R. Hamilton, M. J. Kelly, and C. G. Smith, Phys. Rev. B 91, 235402 (2015). 49 D. Maryenko, A. S. Mishchenko, M. S. Bahramy, A. Ernst, and J. Falson, Y. Kozuka, A. Tsukazaki, N. Nagaosa, M. Kawasaki, Nature Communications 8, 14777 (2017). 50 T.-M. Chen, A. Graham, M. Pepper, I. Farrer, and D. Ritchie, Appl. Phys. Lett. 93, 032102 (2008). 51 K. Aryanpour and J. E. Han, Phys. Rev. Lett. 102, 056805 (2009). 89, 196802 (2002). 24 F. Bauer, J. Heyder, E. Schubert, D. Borowsky, D. Taubert, B. Bruognolo, D. Schuh, W. Wegscheider, J. von Delft, and S. Ludwig, Nature 501, 73 (2013). 25 N. K. Patel, J. T. Nicholls, L. Martin-Moreno, M. Pepper, J. E. F. Frost, D. A. Ritchie, and G. A. C. Jones, Phys. Rev. B 44, 13549 (1991). 26 Y. Kasahara, Y. Oshima, J. Falson, Y. Kozuka, A. Tsukazaki, M. Kawasaki, and Y. Iwasa, Phys. Rev. Lett. 109, 246401 (2012). 27 J. Falson, Y. Kozuka, J. Smet, T. Arima, A. Tsukazaki, and M. Kawasaki, Appl. Phys. Lett. 107, 082102 (2015). 28 K. J. Thomas, J. T. Nicholls, N. J. Appleyard, M. Y. Sim- mons, M. Pepper, D. R. Mace, W. R. Tribe, and D. A. Ritchie, Phys. Rev. B 58, 4846 (1998). 29 K. S. Pyshkin, C. J. B. Ford, R. H. Harrell, M. Pepper, E. H. Linfield, and D. A. Ritchie, Phys. Rev. B 62, 15842 (2000). 30 E. J. Koop, A. I. Lerescu, J. Liu, B. J. van Wees, D. Reuter, A. D. Wieck, and C. H. van der Wal, Journal of Super- conductivity and Novel Magnetism 20, 433 (2007). 31 R. Danneau, O. Klochan, W. R. Clarke, L. H. Ho, A. P. Micolich, M. Y. Simmons, A. R. Hamilton, M. Pepper, and D. A. Ritchie, Phys. Rev. Lett. 100, 016403 (2008). 32 M. Buttiker, Phys. Rev. B 38, 9375 (1988). 33 N. K. Patel, J. T. Nicholls, L. Martin-Moreno, M. Pepper, J. E. F. Frost, D. A. Ritchie, and G. A. C. Jones, Phys. Rev. B 44, 10973 (1991). 34 S. M. Cronenwett, H. J. Lynch, D. Goldhaber-Gordon, L. P. Kouwenhoven, C. M. Marcus, K. Hirose, N. S. Wingreen, and V. Umansky, Phys. Rev. Lett. 88, 226805 (2002). 35 K. F. Berggren, T. J. Thornton, D. J. Newson, and M. Pepper, Phys. Rev. Lett. 57, 1769 (1986). 36 A. Srinivasan, L. A. Yeoh, O. Klochan, T. P. Martin, J. C. H. Chen, A. P. Micolich, A. R. Hamilton, D. Reuter, and A. D. Wieck, Nano Letters 13, 148 (2013). 37 P. T. Coleridge, Phys. Rev. B 44, 3793 (1991). 38 A. J. Daneshvar, C. J. B. Ford, M. Y. Simmons, A. V. Khaetskii, A. R. Hamilton, M. Pepper, and D. A. Ritchie, Physical Review Lletters 79, 4449 (1997). 39 A. J. Daneshvar, C. J. B. Ford, A. R. Hamilton, M. Y. Simmons, M. Pepper, and D. A. Ritchie, Physica B: Con- densed Matter 249, 166 (1998). 40 Y.-W. Tan, J. Zhu, H. L. Stormer, L. N. Pfeiffer, K. W. Baldwin, and K. W. West, Phys. Rev. Lett. 94, 016405 (2005). 41 V. M. Pudalov, M. E. Gershenson, H. Kojima, N. Butch, E. M. Dizhur, G. Brunthaler, A. Prinz, and G. Bauer, Phys. Rev. Lett. 88, 196404 (2002). 42 G. Ploner, J. Smoliner, G. Strasser, M. Hauser, and E. Gornik, Phys. Rev. B 57, 3966 (1998). 43 M. H. Degani and O. Hipolito, Solid state communications 65, 1185 (1988). 44 A. Kristensen, H. Bruus, A. E. Hansen, J. B. Jensen, P. E. Lindelof, C. J. Marckmann, J. Nygard, C. B. Sørensen, F. Beuscher, A. Forchel, and M. Michel, Phys. Rev. B 62, 10950 (2000). 45 T.-M. Chen, A. C. Graham, M. Pepper, I. Farrer, D. An- derson, G. A. C. Jones, and D. A. Ritchie, Nano letters 10, 2330 (2010). 46 A. C. Graham, K. J. Thomas, M. Pepper, N. R. Cooper, M. Y. Simmons, and D. A. Ritchie, Phys. Rev. Lett. 91, 136404 (2003). 

cond-mat/0605574

2

0605

Power dissipation in spintronic devices: A general perspective

Champions of spintronics often claim that spin based signal processing devices will vastly increase speed and/or reduce power dissipation compared to traditional charge based electronic devices. Yet, not a single spintronic device exists today that can lend credence to this claim. Here, I show that no spintronic device that clones conventional electronic devices, such as field effect transistors and bipolar junction transistors, is likely to reduce power dissipation significantly. For that to happen, spin-based devices must forsake the transistor paradigm of switching states by physical movement of charges, and instead, switch states by flipping spins of stationary charges. An embodiment of this approach is the single spin logic idea proposed more than 10 years ago. Here, I revisit that idea and present estimates of the switching speed and power dissipation. I show that the Single Spin Switch is far superior to the Spin Field Effect Transistor (or any of its clones) in terms of power dissipation. I also introduce the notion of matrix element engineering which will allow one to switch devices without raising and lowering energy barriers between logic states, thereby circumventing the kTln2 limit on energy dissipation. Finally, I briefly discuss single spin implementations of classical reversible (adiabatic) logic.

cond-mat.mes-hall

cond-mat

Energy dissipation in spintronic digital switches: A general perspective Supriyo Bandyopadhyay Department of Electrical and Computer Engineering Virginia Commonwealth University Richmond, VA 23284, USA Abstract: Champions of “spintronics” often claim that spin based signal processing devices will vastly increase speed and/or reduce power dissipation compared to traditional ‘charge based’ electronic devices. Yet, not a single spintronic device exists today that can lend credence to this claim. Here, I show that no spintronic device that clones conventional electronic devices, such as field effect transistors and bipolar junction transistors, is likely to reduce power dissipation significantly. For that to happen, spin-based devices must forsake the transistor paradigm of switching states by physical movement of charges, and instead, switch states by flipping spins of stationary charges. An embodiment of this approach is the “single spin logic” idea proposed more than 10 years ago. Here, I revisit that idea and present estimates of the switching speed and power dissipation. I show that the Single Spin Switch is far superior to the Spin Field Effect Transistor (or any of its clones) in terms of power dissipation. I also introduce the notion of “matrix element engineering” which will allow one to switch a binary “switch” without raising or lowering energy barriers between the two states of the switch, thereby reducing energy dissipation. Finally, I briefly discuss single spin implementations of classical reversible (conservative) logic. Keywords: Spintronics, single spin logic, power dissipation, reversible logic CONTENTS 1. Introduction…………………………………1 2. Spintronics…………………………….........2 3. Reading and writing spin……………………3 3.1 Writing spin……………………….3 3.2 Reading spin………………………4 4. Spin transistors……………………………...5 5. Single spin logic family……………….........8 5.1 The NAND gate…………………..8 5.2 The issue of unidirectionality........10 5.3 Energy and power dissipation…...11 5.4 Operating temperature……………11 5.5 What about the Landauer result?...12 5.6 Matrix element engineering……...13 5.7 Energy dissipation in the clock….14 5.8 Comparison with the SPINFET….15 6. Spintronic reversible (adiabatic) logic……..15 6.1 Toffoli-Fredkin gate……………..16 7. Conclusion……………………………........17 1. INTRODUCTION A binary switch is the primitive unit of every classical digital computer or signal processor. It processes digital information or signals by switching from one state to another in response to a digital input. It is a bistable component, one of whose states – say, the “on” state - encodes the logic bit 1 and the other – the “off’ state – encodes the logic bit 0. The best known electronic switch is a transistor. In the case of a “field effect transistor” (FET) or a “bipolar junction transistor” (BJT), the device is “on” when the active region (the channel of an FET or the base of a BJT) contains a large amount of mobile charges, and it is “off” when that region is depleted of mobile charges. Therefore, switching between logic bits can only be accomplished by physically moving charges in 1 and out of the active region with an external agency (such as the gate voltage in an FET or the base current in a BJT). This physical motion consumes considerable energy, which is ultimately dissipated as heat. The obvious way the dynamic to reduce dissipation during the switching event is to switch between states without moving charges. Unfortunately, this is virtually impossible in charge based electronics, where the difference in the amount of charge in the active region is used to demarcate logic levels. Charge is a scalar quantity and therefore it only has a magnitude. Thus, logic levels can be demarcated solely by a difference in the magnitude of the charge, or by the presence and absence of charge 1 . Therefore, in order to switch from one logic state to another, we must invariably move charges from one region of space to another, thereby causing current flow (I) and associated energy dissipation (I2Rt = Q2R/t), where Q is the amount of charge moved, R is the resistance in the path of the current and t is the switching delay. This energy dissipation is unavoidable and it is a fundamental shortcoming of charge based electronics. 2. SPINTRONICS Spin, unlike charge, is not a scalar. It has a magnitude and a “polarization”. It is easy to make the spin polarization a bistable quantity – and therefore use it to encode binary bits - by simply placing the electron (or hole) in a magnetic field. The Hamiltonian describing a single electron in a magnetic field is     B Aqp H m g 2/) ( ( B)2/ * 2     (1)  A where is the vector potential due to the  B , B is the Bohr magnetic flux density 1 It does not have to be “absolute” presence or absence. It has to be simply a “relative” presence or absence. For example 10 electrons could represent ‘logic 1’ and 2 electrons could represent ‘logic 0’. All we need is a ‘difference’ in the quantity of charge to demarcate logic levels. But in order to switch from one logic state to another, we must be able to alter this difference of 8 electrons, and therefore we must move 8 electrons in space. That consumes energy.  magneton, g is the Lande g-factor, and  is the Pauli spin matrix. If the magnetic field is ^  B = B ), then z directed in the z-direction ( diagonalization of the above Hamiltonian immediately produces two mutually orthogonal eigenspinors [1, 0] and [0, 1] which are the +z and –z-polarized spins, i.e. states whose spin quantization axes are parallel and anti-parallel to the z-directed magnetic field. Thus, the spin quantization axis (or spin polarization) can become a binary variable. The “down” (parallel) or “up” (anti-parallel) states can encode logic bits 0 and 1, respectively. These two states are not degenerate in energy, but that does not pose any problem with encoding logic bits. We could switch from logic bit 0 to 1, or vice versa, by simply flipping spin with an external agent such as a localized magnetic field 2 , without having to physically move charge and causing current flow. This should eliminate the I2Rt dissipation. Some energy would still be dissipated since the two states are not degenerate BgB , but this but separated by an amount could be made arbitrarily small by making the magnetic flux density B arbitrarily small. Furthermore, since there is no motion, the switching time is not limited by the transit time of charges, or the velocity of electrons, but instead is limited by the spin flip time which can be engineered to some extent (e.g. by introducing magnetic impurities in the vicinity of the spin). Spin also has another advantage; it does not couple easily to stray electric fields unless the host material has strong spin-orbit interaction. Thus, it has some natural noise immunity, unlike charge. How stable is the spin polarization? In InAs quantum dots, the single electron spin flip time has been measured to be about 15 ns at 10 K [1]. At room temperature, we expect the spin-flip time to decrease considerably because of spin- phonon coupling [2]. There is currently no published measurement of spin flip times in quantum dots at room temperature. We can - 2 We could also simply reverse the magnetic field to flip spin. 2 perhaps somewhat optimistically - estimate it to be about 1 ns in InAs quantum dots. If we assume a clock speed of 50 GHz (we will show later that this is a reasonable estimate), then the clock period is 20 ps, which is 1/50-th of the spin flip time. Therefore, the probability that an unintentional spin flip will occur between two 1 e 1 50 successive clock pulses is = 0.02, or 2%, which can be handled by modern error correction algorithms [3]. Handling an error rate this high requires considerable hardware overhead, but because the device density can be extremely large, this may not turn out to be taxing. If instead of InAs we choose silicon or InP, then the spin flip time can increases dramatically. Spin flip times of several microseconds have been measured for electrons bound to phosphorus donor atoms in silicon at 20 K [4] and spin flip times exceeding 100 s have been reported for InP dots at a temperature of 2 K [5]. Room temperature data is unavailable, but even if we assume that the spin flip time is 100 ns at room temperature in silicon or InP quantum dots, the probability of having a random bit flip two between successive clock cycles 1 e  1 5000 = 0.02% for 50 GHz clock rate, is which could be handled by error correction schemes without mammoth overhead. 3. READING AND WRITING SPIN Encoding binary logic bit information in single electron spin polarization is an interesting notion, but how does one “read” or “write” spin to manipulate or extract this information? In order to controllably orient and detect spin polarization of a single electron (i.e. read and write a bit), we have to first place it in an appropriately designed quantum dot and then place it in a magnetic field that defines the spin quantization axis. The quantum dot will be delineated electrostatically by split metal gates in a penta-layered structure consisting of a ferromagnet-insulator-semiconductor-insulator- ferromagnet combination as shown in Fig. 1(a). The wrap-around split Schottky gate is used for the “write” operation. The ferromagnetic layers also play a critical role in performing the “reading” and “writing” operations. x y z Fig. 1 (a) Structure of a gated quantum dot to host a single electron with a well defined spin orientation. The top figure shows the top view (in the x-y plane) and the bottom figure shows the cross-section (in the y-z plane). (b) The idealized conduction band energy diagram along the z-direction, perpendicular to the heterointerfaces. 3.1 Writing spin e- EF Fig. 2: The scheme for writing a spin bit. The solid black lines are the conduction band profile in the z-direction before applying the potential to the split gates to enlarge the dot. The dot is “enlarged” in the x-y plane. The broken purple lines are the conduction band profile in the z- direction after enlarging the dot. The lowest spin split state falls below the Fermi level and a single electron occupies the dot. The ferromagnet, insulator and semiconductor materials are so chosen that the conduction band energy diagram at equilibrium (in the direction normal to the heterointerfaces) is as shown in 3 Fig. 1(b). The lowest subband (broken lines) is spin split because of the magnetic field caused by the ferromagnetic contacts, plus any other external magnetic field. We will place the Fermi level below both spin split levels by appropriate choice of materials and doping. In that case, there will be no electron in the quantum dot (for this discussion, we will assume that the temperature is ~0 K). In order to “write” a spin bit, a positive potential is applied to the wrap-around gate which decreases the confinement and effectively makes the semiconductor dot larger, thereby pulling the lower spin-split level of the first subband below the Fermi level, while still keeping the higher level above the Fermi level. This is shown by the broken red lines in Fig. 2. A single electron now tunnels into the semiconductor dot from a ferromagnet. Because of Pauli Exclusion Principle, only a single electron can be hosted in the dot. This electron's spin is polarized along the magnetization of the ferromagnet and is therefore known apriori. Thus, we have successfully “written” a bit, or a pre-determined spin polarization. Let us say that this bit is logic ‘0’. Now, in order to guarantee that the injected spin is always polarized along the direction of the ferromagnet’s magnetization, we will really need a ferromagnet with 100% spin polarization. Of course, no known ferromagnet has a 100% spin polarization at room temperature, but it is possible to have rather high spin injection efficiency, as high as 70% at room temperature, even from non-ideal ferromagnets [6]. Thus, there can be ~ 30% error in writing a bit. This error can be corrected by introducing redundancy, i.e. writing the same bit in 3 or more cells and then using majority voting logic to determine threefold the correct bit (a redundancy reduces the error probability to 9% and a fivefold redundancy reduces it to 2.7%). It should be noted that the writing error is a one- time error that does not change randomly with time. Such errors are easier to correct than dynamic errors. The two insulating layers shown in Fig. 1 are important in this context. First, they provide the confinement of carriers in the semiconductor dot. Second, they increase the efficiency of spin injection from the ferromagnet into the semiconductor [7], thereby reducing the error probability during the write operation. So far, we have described how to write the logic bit 0. But what if we wanted to write logic bit ‘1’ instead? One might guess that we should simply reverse the magnetization of the contacts to do the trick. That would be fine except reversing the magnetization is not so easy. We could not have done it with a magnetic field since the latter would be difficult to confine to a single dot. Therefore, we have to find a different method. We will do is apply a differential potential between the two split Schottky gates in Fig. 1 to induce a Rashba interaction in the chosen dot [8]. The total spin-splitting energy  in the semiconductor layer will then be a mixture of the Rashba and the Zeeman spin splitting. It will be given by approximately [9] 42 2  q 16 2)2/ BBg yE xWxWf ' ( ,  44* xWxWcm (2) 2/1)]  [(2  ' Here Ey is the y-directed electric field due to the differential potential applied between the Schottky gates, Wx is the spatial width of the wavefunction along the x-direction in the lower Zeeman split spin state, and W’x is the spatial width in the upper spin state (they are different because the potential barriers confining the electron are of finite height) 3 . 3 This expression is based on a simplified treatment that ignores spatial variation of the spin orientation within the dot. A more accurate theory yields a significantly different quantitative result, but is avoided here since it does not yield analytical expressions [see D. Bhowmik and S. Bandyopadhyay, Physica E, 41, 587 (2009) for the x WWf , ( where the function (2cos xWxW '2/ )  1 ((2)1 xWxW / ' xWxW (( / ' xWxWf , ( ' 2) F    ) )' is x (2cos FxWxW 2/ ' )  2) 2))1  4  By adjusting Ey, we can make the total spin splitting energy  in the chosen dot resonant with a global ac magnetic field of frequency   We hold this resonance for a time duration such that Bac 2/h  (3) B acB is the amplitude of the global ac where magnetic field. This is a so-called  pulse. It flips the spin and will write the logic bit 1 only in the chosen dot. This is the well-known principle used in electron spin resonance. Thus, we can write the logic bit 1 as well. The magnitude of Ey is of the order of 100 kV/cm. Since the split-gate separation is typically ~ 1 m, the differential voltage that we have to apply between the two split gates is about 10 V. We can estimate the value of  by assuming a reasonable value of Bac = 0.01 Tesla. In that case,  = 18 ps. This tells us what clock speed we can have. The clock speed is limited to 1/(18 ps) = 55 GHz, which is plenty fast. A much simpler method for reversing the magnetization of nanoscale ferromagnetic contacts is the use of a spin polarized current to exert a spin transfer torque on the magnetization vector 4 . This method was not popular at the time the original manuscript was written, but since then has become immensely popular. However, this method dissipates a lot of energy unlike the previous method and is also much slower with ~ 1 ns.  3.2 Reading spin “Reading'” its ascertaining or spin, a polarization, is more difficult than writing spin. Single spin reading has been demonstrated with a variety of techniques [10], but most of them are difficult and slow. For electrical detection, we can use the technique of ref. [11] (which is more accurate treatment. This paper appeared two years after this manuscript was originally written.] 4 For a review, see D. C. Ralph and M. D. Stiles, J. Magn. Magn. Mater., 320, 1190 (2008). by no means unique and variations exist). This scheme requires the use of ferromagnetic contacts to determine the spin polarization of a target electron. Hence, the need for ferromagnetic contacts. 4. SPIN TRANSISTORS We have described the essential ingredients of a Single Spin Switch which acts as a binary logic device. Early research in spintronics however was not concerned with encoding data in single spin polarization. Instead, it concentrated on devising spin based analogs of conventional field effect transistors [12] and bipolar junction transistors [13]. Source L Gate Insulator L Drain Semiconductor Fig. 3: Schematic representation of a Spin Field Effect Transistor after ref. [12]. The source and drain contacts are ferromagnets that inject and detect spin of a particular polarization. The gate terminal applies a transverse electric field on the transistor channel (shown with a broken line) which induces a Rashba spin orbit interaction that precesses the spins at a rate determined by the magnitude of the gate voltage. Fig. 3 shows a schematic representation of the Spin Field Effect Transistor (SPINFET) proposed in the seminal work of ref. [12]. We will assume that the channel is strictly one- dimensional and only the lowest subband is occupied. This device looks exactly like a conventional metal-oxide-semiconductor field effect transistor (MOSFET) except that the source and drain contacts are ferromagnetic. The ferromagnetic source injects carriers with a particular spin polarization (the majority spins in the ferromagnet) into the channel. The gate 5 voltage induces a Rashba spin-orbit interaction which precesses the spins at a rate that depends on the magnitude of the gate voltage and the carrier momentum. The precession rate in time depends on the gate voltage and carrier momentum, but the precession rate in space can be shown to be independent of the carrier momentum and depend only on the gate voltage. If the gate voltage is such that the spins have precessed by an angle that is an odd multiple of  when they arrive at the drain, then they have a polarization anti-parallel to the drain’s magnetization. These spins are blocked by the drain and therefore the source to drain current falls to zero. Without a gate voltage, the spins do not precess 5 and the source to drain current is non-zero. Thus, the gate voltage causes current modulation and realizes transistor action. It should be obvious that in this device, “spin” actually plays a relatively minor role. Information is still encoded in charge which carries the current. The transistor is still switched by moving charges in space (because a current flows from source to drain). The role of spin is only to provide an alternate means of controlling the current. In ref. [14], we showed that this device does not yield any significant advantage over a conventional charge based transistor such as the MOSFET, unless materials with extremely strong spin-orbit interaction become available. We showed in ref. [14] that the ratio of the switching voltage of a SPINFET to that of a comparable MOSFET is given by  (4) eh V 2  SPINFET ELm V 2 *  MOSFET F where L is the channel length, m* is the carrier effective mass, e is the electronic charge, h is the Planck constant, EF is the Fermi energy in the channel, and  is the change in the Rashba spin orbit interaction strength per unit change in the gate voltage. The quantity  has been measured in InAs and reported in ref. [15]. If we 5 Even without the gate voltage there is obviously some Rashba interaction in the channel since the structure is not symmetric. We ignore that here. assume the measured value, then the ratio in Equation (4) will be less than unity only if the channel length exceeds 295 m in an one- dimensional InAs SPINFET with carrier concentration (and therefore EF) small enough to maintain single subband occupancy. In other words, the SPINFET can have a lower switching voltage (and therefore lower power dissipation) than the MOSFET only if it is impractically long. On the other hand, if we assume the theoretical maximum value of , then the SPINFET has a lower threshold voltage if the channel length is larger than 4.9 m. Thus, no sub-micron SPINFET, let alone a nanometer scale SPINFET, is likely to yield any advantage over a comparable MOSFET. This is because we are still switching the SPINFET by moving charges in space. As long as we are within that paradigm, we cannot expect any significant improvement in power dissipation over charge based devices. Many proposals have now appeared in the literature claiming to improve the design of the original SPINFET of ref. [12]. One of them [16] posits a device that can operate in the diffusive regime unlike the SPINFET of ref. [12] which normally requires ballistic transport (see footnote [2]). In this device, 100% spin polarized current is assumed to be injected from the ferromagnetic source contact. In the absence of any gate voltage, there is no spin flip scattering in the channel so that the injected spins arrive intact at the ferromagnetic drain. The drain (which is magnetized parallel to the source), transmits all of these spins and the current is maximum. When the gate voltage is turned on, it changes the Rashba spin orbit interaction in the channel which causes spin flip scatterings. As a result, some carriers flip spin. The carriers with flipped spins are blocked by the drain. At best 50% of the spins will be flipped, at which point, the spin polarization of the current becomes zero. Thus the current can drop to one-half of the maximum value when the gate voltage is turned on. It is obvious that this device can provide a maximum ratio of the “on”-conductance to the “off”-conductance = 2, instead of the 1000 6 required for most applications 6 . This ratio can improve if the device is modified so that the two ferromagnetic contacts are magnetized in anti- parallel directions, unlike the device of Fig. 3 where the two contacts are magnetized in parallel direction. If nearly perfect spin injection from the ferromagnetic contacts can be attained, the anti-parallel arrangement can cause dramatic improvement in the conductance ratio. In fact, 100% spin injection and spin extraction efficiencies would make the conductance ratio infinity. However, perfect spin injectors or spin detectors do not exist as yet and the maximum spin injection efficiency demonstrated so far is 90% from non-permanent ferromagnets [18] and 70% from permanent ferromagnets [6]. In the former case, 5% of the total injected current is due to minority spins, and in the latter case, 15% is due to minority spins. Consequently, even if the drain contact is a perfect spin detector, the leakage current flowing during the “off” state will be 5% or 15% of the injected current. The current flowing during the “on” state is 50% of the injected current at best (when the spin flip scattering completely destroys any net spin polarization in the current). Therefore, the on-to- off conductance ratio is 0.5/0.05 = 10, or 0.5/0.15 = 3.33, in the two cases. That is not a significant improvement over the factor of 2 attained when the two contacts have parallel magnetizations. Anti-parallel magnetizations however reduce the stray magnetic field the channel of a in SPINFET. This can improve the performance of the SPINEFT of ref. [12] in a number of ways, as discussed in refs. [19] and [20]. Other SPINFET proposals have recently appeared in the literature. One of them [21] is similar to the proposal of ref. [16] in that the gate voltage changes the spin flip scattering rate, and thus modulates the source-to-drain current. Because of this similarity, it inevitably suffers from the same drawback as ref. [16], namely small on-to-off conductance ratio, even for anti- parallel magnetizations, as long as we assume 6 Simulations [17] show that the actual ratio is even less than 2; it is only about 1.2. Therefore, it is clearly insufficient for device applications. realistic spin injection efficiencies. However, ref. [21] makes an important claim. It claims that “[this] spin transistor operation will be possible at a much lower threshold voltage than conventional CMOS technology”. Ref. [22] expands on this claim and asserts that the SPINFET will have a lower leakage current in the OFF-state than a MOSFET. We have recently shown that these claims are invalid [23]. The apparent basis of such claims is the authors’ belief that a small voltage can induce a large modulation of the Rashba interaction in the channel of quantum wells, so that a SPINFET could be switched with a smaller gate voltage than a MOSFET. This claim is in direct contradiction with experimental findings of ref. [15] which shows that the gate modulation of the Rashba interaction is actually rather weak. The quantity  was experimentally measured to be only 8 x 10-31 C m and even the maximum theoretical value is only 60 times larger [14]. Ref. [21] calculates that in a channel with Fermi energy EF = 30 meV, a gate voltage of 140 mV should reduce the spin lifetime to ~ 10 ps. It will be reasonable to assume that the carrier mobility at room temperature will be ~ 1 m2/V-sec in the InAs channel considered. Then, the spin diffusion length will be ~ 0.16 m at a gate voltage of 140 mV. Thus, a device with channel length  0.16 m can be turned on and off with a gate voltage of 140 mV. A shorter device will require a larger gate voltage. Based on the above, the switching voltage of a 0.16 m long SPINFET can be only 140 mV. That may be true, but if this same device were used as an ideal MOSFET instead of an ideal SPINFET, the voltage required to deplete the channel completely (and therefore switch the transistor) would have been simply EF/e = 30 mV. Therefore, this spin device will not have a lower switching voltage, but rather a higher switching voltage by a factor of 140/30 = 4.6. Note that in calculating the switching voltage, we have consistently neglected any voltage drop across the gate insulator, but that does not change our conclusion. 7 In ref. [23], we also showed that this type of SPINFET has a particularly leakage large current because of the small ratio of the on-to- off conductance. The leakage current is consequence of imperfect spin injection and detection. We showed that in order to have the off-current smaller than 0.1% of the on-current, one would need the spin injection efficiency at the ferromagnet/semiconductor interface to exceed 99.9%. That is a very tall order. We have also analyzed the spin bipolar junction transistors proposed in ref. [13] and found that too are not significantly better they than conventional (charge based) bipolar junction transistors [24]. Again, the reason is that “spin” plays a minor role in their operation and switching is still accomplished by moving charges in space. 5. SINGLE SPIN LOGIC FAMILY Significantly lower threshold voltages can only be achieved via a paradigm shift, namely by switching states without physically moving charges. Consistent with this viewpoint, we proposed the idea of “single spin logic” in 1994 where a single electron acts as a binary switch and its two orthogonal (anti-parallel) spin polarizations encode binary bits 0 and 1 [25]. Switching is accomplished by flipping the spin without moving charges. This is the first known logic family based on single electron spins. Recent advances in controlling and manipulating an electron at the single spin level [26-34] has now made it possible to make important strides towards the implementation of single spin logic. This motivates the present review. Many logic circuits, both combinational and sequential, have been designed and theoretically verified using the single spin idea [25, 35]. Here, I will repeat the design of a NAND gate since it is a universal gate. Any logic circuit can be realized with it. 5.1. The NAND gate Consider a linear chain of three electrons in three quantum dots as shown in Fig. 4. They are placed in a global (dc) magnetic field, as mentioned before, to make the spin polarization a bistable entity. We will assume that only nearest neighbor electrons interact via exchange since their wavefunctions overlap. Second nearest neighbor interactions are negligible since exchange interaction decays exponentially with distance. Fig. 4: Spin configurations in a 3-dot system with nearest neighbor exchange coupling. The peripheral spins are the two inputs and the central spin is the output. The four configurations correspond to the four entries in the truth table of a NAND gate. The B-field is necessary to make spin polarization a bistable entity and to resolve “ties”. We will also assume that the exchange energy (defined as the energy difference between the triplet and singlet states of two neighboring interacting electrons) is larger than the Zeeman splitting energy caused by the globally applied external (dc) magnetic field. In this case, the ground state of the linear array has anti- ferromagnetic ordering where nearest neighbors have opposite spin. In fact, the ground state of the array looks like as in Fig. 4(a). This was verified by quantum mechanical many-body calculation in ref. [36]. 8 Let us now regard the two peripheral spins as the two “inputs” to the logic gate and the central spin as the “output”. Assume the downspin state   [parallel to the global magnetic field]  is  represents logic 1, and the upspin state logic 0. Then, we find that when the two inputs are 0, the output is automatically 1 because the ground state has anti-ferromagnetic ordering. That is encouraging since it is one of the four entries in the truth table of a NAND gate. We now have to realize the other three entries in the truth table. If we change the two inputs to 1 from 0, using the technique embodied in Equation (3), then this takes the system to an excited state since the ordering is no longer anti- ferromagnetic. We then let the system relax. In order to regain anti-ferromagnetic ordering, either the central spin will flip down (e.g. by emitting a phonon), or else the two peripheral spins will flip down to their original state after the writing operation is complete. The former requires a single spin flip, while the latter requires two spin flips. However, the former process will take the system to a local metastable state, while the latter takes it to the global ground state. Note that the states in Figs. 4(a) and 4(b) are not degenerate in energy because of the global magnetic field, which favors 4(a) over 4(b). Whether the metastable state is reached, or the global ground state is reached depends on the energy landscape and the switching dynamics. If the time  in Equation (3) is much smaller than the spontaneous spin flip time, then we will likely reach the metastable state. If the metastable state is reached, then we will achieve the configuration shown in Fig. 4(b). This is the desired configuration because when the two inputs are 1, the output is 0. This is yet another entry in the truth table of a NAND gate. The metastable state must ultimately decay to the global spins flipping by state ground spontaneously. But since the spin flip time is typically much longer than the inverse of the input data rate (~ 50 GHz), we can ignore it. Finally, what happens if one input is 1 and the other 0? This situation seemingly causes a tie, but the global magnetic field resolves this situation in favor of the central dot having a down-spin configuration (parallel to the magnetic spin corresponding field). The arrangements are shown in Figs. 4(c) and 4(d). Note that these conform to the other two entries in the truth table of a NAND gate. Finally, we have realized the entire truth table: Output 1 1 1 0 Input 1 0 0 1 1 Input 2 0 1 0 1 In 1995 Molotkov and co-workers verified the entire truth table by carrying out a fully quantum mechanical calculation of the various spin configurations in a 3-dot system [36]. Further work in this area has been performed by Bychkov and co-workers [37]. Note that the global magnetic field serves dual purposes: (i) it defines the spin quantization axis while making spin a binary variable, and (ii) it resolves ‘ties’ when they arise. Without the global magnetic field, this paradigm would not work. A similar idea for implementing logic gates using single electron charges confined in ‘quantum dashes’ was proposed by Pradeep Bakshi and co-workers in 1991 [38]. There, logic bits were encoded in bistable charge polarizations of elongated quantum dots known as ‘quantum dashes’. Coulomb interaction between nearest neighbor quantum dashes makes the ground state charge configuration “anti-ferroelectric” just as exchange interaction the ground state spin in our case makes configuration anti-ferromagnetic. Three Coulomb coupled quantum dashes would realize a NAND gate in a way very similar to what was described here. Bakshi’s idea inspired a closely related idea known as ‘quantum cellular automata’ [39] 9 which uses a slightly different host, namely a 4- or 5-quantum dot “cell” instead of a quantum dash to store a bit. Here too the charge polarization of the cell is bistable and encodes the two logic bits. Coulomb interaction makes the configuration charge state ground ferroelectric. Logic gates are implemented in the usual fashion. The paradigm of ref. [38] and its clone [39] involve charge motion for switching. Thus, they are likely to be more dissipative than single spin logic which eliminates charge motion altogether. But more importantly, it is very difficult to enforce only nearest neighbor interactions because the Coulomb interaction is long range. If second nearest neighbor interactions are not very much weaker than the nearest neighbor ones, the ground state charge configuration is only marginally stable and not robust against noise. In this respect, our spin based approach has an advantage. Since exchange interaction is short range, it is much easier to make the second nearest neighbor interactions considerably weaker than the nearest neighbor interactions. Accordingly, is the anti-ferromagnetic state much more stable than the anti-ferroelectric (or ferroelectric) state against other spurious states. This, coupled with the fact that spin does not easily couple to stray electric fields unlike charge, makes the single spin logic gates superior. 5.2 The issue of unidirectionality There is, however, a serious problem with these types of logic gates which may not be apparent at first. There is no isolation between the input and output of the logic gate since exchange interaction is “bidirectional”. Consider just two exchange coupled spins in two neighboring quantum dots. They will form a singlet state and therefore act as a natural NOT gate. However, exchange interaction cannot distinguish between which spin is the input bit and which is the input and output are output. Since the ultimately becomes it indistinguishable, impossible for logic signal to flow unidirectionally from an input stage to an output stage and not the other way around. We have discussed this issue at length in various publications [25, 40] since it is vital. Because of this problem, the ideas of ref. [38] and [39], as proposed, could not work since Coulomb interaction is also bidirectional. It is of course possible to forcibly impose unidirectionality by holding the input cell in a fixed state until the desired output is produced. In that case, the input signal itself enforces unidirectionality because it is a symmetry- breaking influence. This approach was actually used to demonstrate a ‘magnetic cellular automaton’ where the input enforced unidirectionality and produced the correct output [41]. However, there are problems. First, this approach can only work for a small number of cells before the influence of the input dies out. Second, and more important, the input cannot be changed until the final output has been produced since otherwise the correct output may not be produced at all. That makes such architectures non-pipelined and therefore unacceptably slow. There may also be additional problems associated with random errors when this approach is employed. They have been discussed in ref. [42]. In 1994, when we first proposed the single spin logic gates, we thought of enforcing unidirectionality artificially by progressively increasing the distance between quantum dots, so that there is spatial symmetry breaking [25]. In 1996, we revised this idea and instead pointed out the possibility of imposing unidirectional flow of signal in time, rather than in space, by using clocking to activate successive stages sequentially in time [43]. The actual clocking can be done in the same way as is done in bucket brigade devices, such as charge-coupled-device (CCD) shift registers 7 , where a push clock and a drop clock are used to lower and raise barriers and thus steer a charge packet unidirectionally from one device to the next. In single spin 7 CCDs also have no inherent unidirectionality. There a push clock and a drop clock are used to steer charge packets from one device to the next. See, for example, D. K. Schroeder, Advanced MOS Devices, Modular Series in Solid State Devices, Eds. G. W. Neudeck and R. F. Pierret (Addison-Wesley, Reading, MA, 1987). 10 circuits, during the positive cycle, the clock signal will apply a positive potential to the barrier separating two neighboring quantum dots which will lower the barrier temporarily to exchange couple these two spins. This will result in spins assuming anti-parallel two the polarizations. Then during the negative clock cycle, the barrier is raised again to decouple the two spins. In this fashion, spin bits can be transferred unidirectionally from one dot to the next. Just as in CCD devices, a single phase clock cannot impose the required unidirectionality; a 3-phase clock is required to do this job [44]. The clocking circuit however introduces additional dissipation. More importantly, it takes away the most attractive feature of these circuits, namely the absence of interconnects (or “wires”) between successive devices. In Single Spin Logic, exchange interaction plays the role of physical wires transmit signal between to neighboring spins (switches), but in order to steer bits “unidirectionally” down a logic chain, we will need each switch or spin to be clocked individually and requires a physical that interconnect (clock pad) to be placed between every pair of quantum dots. The split gates in Fig. 1 can be used as the clock pads, but wires must be attached to them to ferry the clock signal. Therefore, the architecture is no longer wireless. More importantly, the lithographic burden is daunting. Neighboring dots should not be separated by more than a few nm in order to retain adequate exchange coupling and this mandates interposing a gate pad within a very narrow space (~few nm), which is extremely challenging. All this of course detracts from the appeal of a “wireless architecture”, or the so- called “quantum coupled architecture”, but this is the price one must pay when inter-device communication is via bidirectional exchange interaction. 5.3 Energy and power dissipation We are now ready to calculate the energy and power dissipation per bit flip in Single Spin Logic. When a bit flips, the energy released (or absorbed) is the energy difference E between the two spin states in a quantum dot which will depend on the spin orientations of its neighbors because of exchange interaction. Roughly speaking, this energy will be of the order of the exchange splitting energy. Reasonable upper estimates for the latter in today’s coupled dot systems is about 1 meV [45], so that E ~ 1 meV. Now, if the energy change in transitioning between the logic states is ~1 meV, then the energy dissipated in switching is also no more than 1 meV = 1.61 x 10-22 Joules. The corresponding power dissipation for a 50 GHz clock is only 8 pW. 5.4 Operating temperature The reader may wonder at this point what the temperature of operation will be if the energy difference between the two logic states is only 1 meV. The answer depends on whether the spin system is in thermal equilibrium with the surroundings. If it is, then the relative occupation probability of the two spin states will be governed by Boltzmann statistics and the probability of spontaneous occupation of the excited state – which is the error probability – E   p exp   kT   . With p = 10-9 [a reasonable E    p ln 1 error probability], kT will be 47.6 eV, or T = 0.55 K. This would doom us to cryogenic operation, making Single Spin Logic entirely impractical. However, a spin system does not need to be in thermal equilibrium with its environment. Equilibration takes place only if thermal perturbations can flip spin to make the spin distribution conform to Boltzmann or Fermi- Dirac statistics. In reality, fluctuations alone cannot flip spin. Otherwise, electron spin resonance experiments – where the spin splitting energies are usually less than 1 meV – could not be carried out at room temperature. Thermal fluctuations (phonons) may supply the energy required to cause a spin flip, but supplying the energy alone is not sufficient to initiate a spin yields kT  be  will . This 11 flip. As ref. [2] points out: “the phonon itself does not flip the spin but [merely] provides energy conservation”. Spin flips are caused by spin-orbit coupling, interactions with nuclear spins, magnetic impurities, etc. [2]. While the strengths of these interactions and the rate of spin flip can certainly go up with temperature, there is no connection between the thermal energy and the energy difference between the two spin states in causing unwanted bit flips. In other words, if the spin system does not equilibrate with the thermal bath, then it is very g B kT possible to have and yet have a B reasonable error since probability Bg B  p exp   kT   equilibrium with the thermal bath. As long as we can maintain the spins out of equilibrium, even room temperature operation is not out of the question. when the spins are not in  kT Fig. 5: The energy splitting between the spin levels may be much smaller than kT. Spin levels are not broadened significantly by coupling to phonons since the spin-phonon coupling is very weak. Another issue that often raises doubts is the following: how can one resolve an energy splitting of 1 meV at room temperature? Are not the levels broadened by ~ kT = 25 meV at room temperature which should swamp the levels? The answer is no. Spin does not couple strongly to phonons and therefore the level broadening is much less than kT which allows one to resolve states separated by 1 meV at room temperature. If this were not the case, then again electron spin resonance experiments could not be carried out at room temperature. The same weak coupling between spins and phonons also makes it much easier to maintain a spin system out of thermal equilibrium than a charge system since charge- phonon coupling is much stronger. In the end, this is another advantage of spin over charge. 5.5 What about the Landauer result? The reader will also likely have another doubt about room temperature operation. If the operating temperature is 300 K so that kT = 25 meV, then it seems implausible that the energy dissipation to switch can be only 1 meV since that seems to contradict a popular idea which states that the minimum energy dissipated in an irreversible logic operation is kTln2 [45]. At room temperature, kTln2 is 18.75 meV, which is nearly 19 times the energy dissipated! So, how is this possible? In reality, the kTln2 limit, known as the Landauer limit after its author, is valid only when the switch is in thermal equilibrium with its surroundings. If the switch is far out of equilibrium, then this limit has no relevance and does not apply. Even in thermal equilibrium, the “Landauer limit” is not straightforward and there are subtleties associated with it. Landauer showed that only if we switch in a complicated way with a very specific sequence of clocking [45], we will switch with 100% probability while dissipating kTln2 amount of energy. But if we switch in a straightforward fashion by simply tilting the potential profile of the switch to bias it towards the desired state, then the minimum   kT p , ln 1 energy dissipated in switching is where p is the static bit error probability. All this is of course valid only in equilibrium and has no validity for out-of-equilibrium systems. Logic 0 E1 Fig. 6: Energy landscape for a binary switch Logic 1 E2 Eb 12 Let us examine the basis of this result. A binary switch has two distinguishable states that encode logic bits 0 and 1. The potential energy landscape of such a switch is shown in Fig. 6 where the two potential wells represent the two stable states. Note that we have intentionally made the two states non-degenerate in energy (E1  E2) to correspond to our situation where they are separated by E. The rate of transition between these two states will be normally given by Fermi’s Golden Rule     )  2  2 (5) EEM EES E E )2,1( 2(  2,1 1  where ME1,E2 is the matrix element for transition between the states and  is the energy transfer. Now, we do not want spontaneous transitions to occur when the switch is not activated because that will result in a bit error. Therefore, we must EES zero when the switch is inactive )2,1( make and non-zero when the switch is active. There are two ways to achieve this. First, we can build an energy barrier between the two states so that one is not directly accessible from the other. This is shown by the broken line in Fig. 6. In that case, the only way spontaneous transitions can take place is if there is tunneling through the barrier (which is negligible if the barrier is tall enough or wide enough), or there is thermionic emission over the barrier. If the switch is in thermal equilibrium and the occupation probability of a state is given by Boltzmann statistics, then the probability of spontaneous thermionic emission over a barrier of energy Eb bE  0  e p kT bE p ln 1  , which yields . is  kT 0 Clearly, cause emission will thermionic spontaneous switching even when the switch is inactive. Hence, the static bit error probability is the probability of essentially thermionic tunneling). Therefore, emission (neglecting   p p bE kT p ln 1 and we get the result .  0 This result tells us that at a given temperature, the taller the barrier is the smaller will be the static bit error probability. 2 p  .   E b kT  ln 1 In order to active the switch and induce switching, we can tilt the potential profile in Fig. 6 towards the left to increase the energy difference E2 – E1 if the desired final state is E1 and the initial state is E2. To be safe, we should make E2 – E1 at least equal to Eb, which will ensure that the barrier separating the two states is completely eroded. In that case, the system will switch from state E2 to state E1 with very high probability since is no barrier there impeding the transition. The energy dissipated in switching is then E E  1 This is of course a “brute-force” method of switching which is more dissipative than necessary. There are more subtle and less dissipative switching strategies but those will entail exquisite timing synchronizations [45] that large dynamic errors during can introduce switching. There is however a second way to prevent unintentional switching when the switch is inactive and induce switching when the switch is activated. This is usually not considered, but we discuss it here. We can make the matrix element in Equation (5) nearly zero when the switch is inactive and large when the switch is active. Thus, we have to modulate the matrix element, making it vanishingly small during the inactive phase and large during the active phase. This “matrix element engineering” can replace “barrier engineering” and eliminate the need for the barrier. In other words, we switch not by tilting the potential profile to erode the barrier, but by modulating the matrix element. Since there is no tilting by an amount equal to or greater than the barrier height, the minimum energy dissipated during switching need not   p kT bE ln 1 / , but can be considerably be  less, or even zero. We illustrate this with a concrete example. 5.6 Matrix element engineering implement “matrix element to In order engineering” we have to ensure that the matrix element is zero when the switch is inactive and 13 non-zero when active. Let us consider the case of a single spin placed in a magnetic field of flux density B. The two allowed spin states are parallel and anti-parallel to the field and are and  . They are separated designated as Bg B . in energy by We will assume that there is no agent (no magnetic impurity, no hyperfine interaction, etc.) that can couple the two mutually anti- parallel spin states in the dot, so that no spin flip transition can occur and the matrix element for spin flip transition is zero. Hence, an excited spin state cannot decay to the ground state by emitting phonons, magnons, etc. In other words, the switch is inactive. In order to activate the switch, we apply a - pulse with an ac magnetic field whose frequency resonant with is the spin splitting,   Bg B i.e. the angular is , where  frequency of the field. This will flip the spin. The pulse makes the switch active temporarily and makes the matrix element temporarily non- zero (only over the duration of the pulse). Once the pulse is removed, the switch reverts to the inactive state. The pulse therefore implements matrix element engineering; its presence makes the matrix element non-zero and its absence makes the matrix element zero. Dynamical switching of a spin by this method does not dissipate any energy at all since the spin rotates by coherently exchanging photons with the electromagnetic wave producing the ac magnetic field. It will absorb a photon of energy E if the final state is E2 and emit a photon of energy E back to the field if the final state is E1. The only caveat is that we must be exquisitely precise with the pulse period and frequency since there is no tolerance for error here. The point to note is that we can deterministically switch without dissipating an equal energy amount of   . This g B kT E E p ln 1 to has     B 1 2 been made possible by matrix element engineering. This method of switching a spin is of course very well known and routinely used in electron spin resonance spectroscopy. We mention it here in the context of switching since it is very different from the traditional switching methods of raising and lowering barriers to switch. 5.7 Energy dissipation in the clock Finally, there is also the issue of how much power is dissipated in the clock that steers bits unidirectionally in Single Spin Logic. For square-wave clock pulses, the energy dissipated in each clock pad is CV2 where C is the capacitance of the pad and V is the voltage applied to lower the potential barrier between neighboring cells. We can of course reduce this energy dramatically if we do not switch abruptly, but switch slowly, but then that would tend to increase clock error rates and reduce bit propagation rate. We can also eliminate this dissipation altogether by using a resonant LRC circuit where the inductor and capacitor are in parallel and the resistance is in series, but this is ultimately cumbersome. Therefore, let us stick with a capacitor and abrupt clock pulse. If C = 1 aF and V = 100 mV (which are reasonable estimates), the energy dissipation in a clock pad is 10-20 Joules and the clock power dissipated with a 50 GHz clock frequency = 0.5 nW per pad. Thus, the clock dissipates 625 times more energy than the device itself. Once again, if this is unacceptable, we can always resort to adiabatic clocking and sacrifice bit propagation speed. Assuming that there are 1011 quantum dots and therefore 1011 clock pads/cm2, the total power dissipated in the chip is 50 Watts/cm2 in the worst case, assuming that all spins are flipping simultaneously. With a more reasonable 10% activity level, i.e. assuming only 1 in 10 spins is flipping at any given instant of time, the power dissipation will be 5 Watts/cm2. The Intel Pentium 4 chip of circa 2000 has a transistor density slightly less than 108/cm2 [48], which is 3 orders of magnitude less than what we assume for single spin based switches. The Pentium IV dissipates about 50 Watts/cm2 [49] 14 with somewhere in the vicinity of 10% activity level. Thus, with Single Spin Logic, one can potentially increase the bit density by 1000 times while keeping the power dissipation per unit area 10 times smaller. All this is of course theoretical is no there speculation since experimental report of anything approaching Single Spin Logic at the time of writing. 5.8 Comparison with the SPINFET We will now compare the Single Spin Switch with the Spin Field Effect Transistor, which, in some ways, is a fair comparison since neither has been demonstrated experimentally. The switching voltage of a SPINFET is h2/(2m*LIn our calculations, we willassume the theoretical maximum value of given in ref. [14], which is 5 x 10-29 C m. Since nm gate length CMOS devices are available and since Single Spin Switches can be easily manufactured with feature sizes less than 90 nm, we will assume that the SPINFET has a gate length L = 90 nm for a fair comparison. In that case, the switching voltage of an InAs SPINFET is Vswitch = 5.6 V. If we assume that the gate capacitor has a width of 90 nm and the gate insulator has a relative dielectric constant of 4 and a thickness of 10 nm, then the gate capacitance Cgate = 28 aF. Accordingly, the energy dissipated in switching = CgateV2 switch = 8.7 x 10-16 Joules, which is higher than what present day transistors dissipate [49] and nearly five orders of magnitude higher than what is dissipated in the clock of single spin logic. With a 30 GHz clock, the power dissipated will be 26 W/bit flip, which is higher than what transistors today dissipate [49]. A nanoscale SPINFET, therefore, is not a low- power device. SPINTRONIC 6. (ADIABATIC) LOGIC So far, we have discussed a spintronic logic family (Single Spin Logic) that dissipates very little energy during switching. But can we design logic gates that dissipate no energy at all? REVERSIBLE It is well known that such gates must be logically reversible [50], i.e. we should be able to infer the input state unambiguously from the output state. There is a vast body of literature on reversible computers, which dissipate no energy at all [50]. Quantum computing is a subset of reversible computing. In 1996, we proposed an idea to implement a quantum adiabatic inverter with single spins [51]. Just two exchange coupled spins, in two closely spaced quantum dots placed in a weak external magnetic field, make a quantum inverter. This gate is logically reversible since the input can always be inferred from the output (they are simply logic complements of each other). The inverter could be switched adiabatically without dissipating any energy at all and ref [51, 52] discuss some interesting results pertaining to the quantum mechanical evolution of this system in time deduced from the time-dependent Schrődinger equation. We showed that the switching time of the optimally designed adiabatic inverter is h t (6)  J 28 where J is the exchange energy, or the energy difference between the triplet and singlet configuration of the two spins. If J = 1 meV, t is less than 1 ps. Therefore, this gate switches quite fast and dissipates no energy. The problem with such adiabatic devices is that they have no fault tolerance and there is a “halting problem”. For example, the quantum inverter must be halted after precisely the time duration given in Equation (6), if we want the desired output. Otherwise, the system will continue to evolve with time and the output will continue to drift from the correct state. The system will periodically revisit the correct state since it is reversible and therefore obeys the principle of Poincare recurrence. However, the halting problem is a major inconvenience. These devices are interesting theoretical curiosities, but probably not very practical at this time. In ref. [51] we also implicitly posited the idea of using the spin of an electron in a quantum dot to represent a “qubit”, although we did not use the term “qubit” since it had not been coined yet. 15 This idea has now become a widely popular concept in the context of quantum computing after Loss and DiVincenzo showed that a universal quantum gate can be realized based on two exchange coupled spins [53]. 6.1 Toffoli-Fredkin gate The inverter designed in ref. [51] is not a universal adiabatic gate, since not any arbitrary adiabatic circuit can be realized with an inverter alone. The universal adiabatic gate is the Toffoli- Fredkin (T-F) gate which has three inputs A,B,C and three outputs A',B',C'. [54]. Input-output relationships for the T-F gate are A' = A, B' = B and C' = C A ● B, where  is the  EXCLUSIVE-OR operation and ● is the AND operation. The T-F gate requires that C toggle if, and only if, A and B are both logic 1; otherwise, nothing should happen. We can realize this gate with three spins placed in a global magnetic field with nearest neighbor exchange interaction (exactly the same configuration as the NAND gate described earlier). The two peripheral spins are the control bits A and B and the central spin is the bit C. There is only one difference with the NAND gate. This time, we will make the Zeeman interaction (due to the magnetic field) stronger than the exchange interaction. The Zeeman interaction makes the downspin state lower in energy than the upspin state in each dot. The exchange interaction, on the other hand, tries to make spins in neighboring dots anti-parallel. The interplay of these two effects realizes the T-F gate. If the spins in A and B are “down”, then the exchange energy will tend to keep the spin in C “up”, while the stronger Zeeman interaction still retains the downspin state (the state parallel to the magnetic field) as the lower energy state. In this case, the exchange interaction subtracts from the Zeeman splitting in the central dot and makes the total spin splitting energy in this dot less than the bare Zeeman splitting. If the spin in dot A is “up” and that in dot B “down”, then the exchange interaction effects due to A and B on the spin in dot C tend to cancel and the spin splitting in dot C is more or less the Zeeman energy. On the other hand, if the spins A and B are both “up”, then the exchange interactions due to them add to the Zeeman splitting in dot C, making the total spin splitting energy in C larger than the bare Zeeman splitting. In essence, the total spin splitting energy in C is larger when both A and B are in logic 1 state, than otherwise. If we denote the total spin splitting energy in dot C where  and  are the logic states in C as  dots A and B, then the following inequality holds:  C =  C   in the energy  C  C These situations are shown diagrams in Fig. 7. Fig. 7: The realization of a spintronic Toffoli- Fredkin gate. The spin splitting energies in the central dot are shown as a function of the spin orientations in the two peripheral dots. The two peripheral dots are the two control bits ‘A’ and ‘B’ and the central spin is the target bit ‘C’. To implement the dynamics of the T-F gate, the whole system is pulsed with a global ac magnetic field of amplitude Bac whose frequency C. The pulse duration is is resonant with  h/(2B Bac). Therefore, C will toggle if A and B are both in logic state 1. Otherwise, nothing will happen. Thus, we have realized the truth table of a Toffoli-Fredkin gate. This is the standard technique for realizing this gate (see ref. [55] for a similar idea). The Toffoli-Fredkin realization leads to an entire family of reversible logic gates 16 based on exchange coupled single spins in a global dc magnetic field. Note that in the case of the T-F gate, we assume that the -pulse induces coherent rotation of the spin in dot C. This is necessary to ensure adiabaticity. 7. CONCLUSION In this work, I have reviewed some examples where the spin polarization of a charge carrier is used to encode bits of information. Signal is processed without physically moving charges, thereby saving energy in the device. These devices are “classical” and therefore spin coherence is not an issue. They are considerably easier to implement than spin-based quantum computing. I have also introduced the concept of matrix element engineering that can be beneficially employed to switch binary switches with arbitrarily small dissipation. Admittedly this is not easy to implement and requires careful design of systems and peripherals. However, most paradigms for computation that claim to overcome the kTln2 limit are at least equally challenging. The Single Spin Logic idea also has some obvious drawbacks. The generators for the ac and dc magnetic fields are necessarily bulky and therefore this paradigm is not suitable for portable electronics such as laptops and cell phones. It is more suitable for desktops and mainframe platforms. The dc magnetic field may be eliminated by the use of permanent ferromagnetic contacts as illustrated in Fig. 1, as long as the field is of the order of ~ 1 Tesla. This is what is required for InAs dots. However, the ac magnetic field still requires microwave generators which are bulky and not portable. In the end, the advantage of Single Spin Logic is that it may be possible to increase device density 1000-fold (and therefore processing power 1000-fold) without any in power increase density on the chip. This is where, I believe, spintronics can carve out a niche. Acknowledgement This work is supported by the US Air Force Office of Scientific Research under grant FA9550-04-1-0261. I am indebted to Prof. Marc Cahay of the University of Cincinnati, Prof. Michael Forshaw of the University College, London, Prof. Vladimir Privman of Clarkson University, and Dr. Suman Datta of Intel Corporation for discussions. REFERENCES: 1. S. Cortez, et al., Phys. Rev. Lett., 89, 207401 (2002). 2. A. V. Khaetskii and Y. V. Nazarov, Phys. Rev. B, 61, 12639 (2000). 3. E. www.arXiv.org/quant- Knill, ph/0410199. 4. G. Feher and E. A. Gere, Phys. Rev., 114 1245 (1959); T. G. Kastner, Phys. Rev. Lett., 8, 13 (1962). 5. M. Ikezawa, B. Pal, Y. Masumoto, I. V. Ignatiev, S. Yu. Verbin and I. Ya. Gerlovin, Phys. Rev. B, 72, 153302 (2005). 6. G. Salis, R. Wang, X. Jiang, R. M. Shelby, S. S. P. Parkin, S. R. Bank and J. S. Harris, Appl. Phys. Lett., 87, 262503 (2005). 7. E. I. Rashba, Phys. Rev. B, 62, R16267 (2000). 8. E. I. Rashba, Sov. Phys. Semicond., 2, 1109 (1960). 9. S. Bandyopadhyay and M. Cahay, Superlat. Microstruct., 32, 171 (2002). 10. D. Rugar, R. Budakian, H. J. Mamin and B. W. Chui, Nature (London), 430, 329 (2004); J. M. Elzerman, et al., Nature (London), 430, 431 (2004); M. Xiao, et al., Nature (London), 430, 435 (2004). 11. S. Bandyopadhyay, Phys. Rev. B, 67, 193304 (2003). 12. S. Datta and B. Das, Appl. Phys. Lett., 56, 665 (1990). 13. See, for example, Jaroslav Fabian, Igor Zutic and S. Das Sarma, Appl. Phys. Lett., 84, 85 (2004); M. E. Faltte, Z. G. Yu, E. Johnston-Halperin and D. D. Awschalom, Appl. Phys. Lett., 82, 4740 (2003). 17 14. S. Bandyopadhyay and M. Cahay, Appl. Phys. Lett., 85, 1433 (2004). 15. J. Nitta, T. Takazaki, H. Takayanagi and T. Enoki, Phys. Rev. Lett., 78, 1335 (1997). 16. J. Schliemann, J. C. Egues and D. Loss, Phys. Rev. Lett., 90, 146801 (2003); X. Cartoixa, D. Z-Y Ting and Y-C Chang, Appl. Phys. Lett., 83, 1462 (2003). 17. Ehud Safir, Min Shen and Semion Siakin, Phys. Rev. B, 70, 241302(R), (2004). 18. R. Fiederling, M. Keim, G. Reuscher, W. Ossau, g. Schmidt, A. Waag and L. W. Molenkamp, Nature (London), 402, 787 (1999). 19. M. Cahay and S. Bandyopadhyay, Phys. Rev. B,, 68, 115316 (2003). 20. M. Cahay and S. Bandyopadhyay, Phys. Rev. B, 69, 045303 (2004). 21. K. C. Hall, Wayne H. Lau, K. Gundogdu. Michael E. Flatte and Thomas F. Boggess, Appl. Phys. Lett., 83, 2937 (2003); K. C. Hall, K. Gundogdu, J. L. Hicks, A. N. Kocbay, M. E. Flatte. T. F. Boggess, K. Holabird, A. Hunter, D. H. Chow and J. J. Zink, Appl. Phys. Lett., 86, 202114 (2005). 22. K. C. Hall and M. E. Flatte, Appl. Phys. Lett., 88, 162503 (2006). 23. S. Bandyopadhyay and M. Cahay, www.arXiv.org/cond-mat/0604532. 24. S. Bandyopadhyay and M. Cahay, Appl. Phys. Lett., 86, 133502 (2005). 25. S. Bandyopadhyay, B. Das and A. E. Miller, Nanotechnology, 5, 113 (1994). 26. M. Ciorga, A. S. Sachrajda, P. Hawrylak, C. Gould, P. Zawadzki, S. Jullian, Y. Feng, and Z. Wasilewski, Phys. Rev. B, 61, R16315 (2000). 27. M. Piero-Ladriere, M. Ciorga, J. Lapointe, M,. Zawadzki, P. Korukusinski, P. Hawrylak and A. S. Sachrajda, Phys. Rev. Lett., 91, 026803 (2003). 28. C. Livermore, C. H. Crouch, R. M. Westervelt, K. L. Campman and A. C. Gossard, Science, 274, 1332 (1996). 29. T. H. Oosterkamp, T. Fujisawa, W. G. van der Wiel, K. Ishibashi, R. V. Hijman, S. Tarucha and L. P. Kouwenhoven, Nature (London), 395, 873 (1998). 30. A. W. Holleitner, R. H. Blick, A. K. Huttel, K. Eberl and J. P. Kotthaus, Science, 297, 70 (2001). 31. N. J. Craig, J. M. Taylor, E. A. Lester, C. M. Marcus, M. P. Hanson and A. C. Gossard, Science, 304, 565 (2004). 32. R. Hanson, L. H. W. van Beveren, I. T. Vink, J. M. Elzerman, W. J. M. Naber, F. H. L. Koppens, L. P. Kouwenhoven and L. M. K. Vandersypen, Phys. Rev. Lett., 94, 196802 (2005). 33. R. Hanson, B. Witkamp, L. M. K. Vandersypen, L. H. W. van Beveren, J. M. Elzerman and L. P. Kouwenhoven, Phys. Rev. Lett., 91, 196802 (2003). 34. J. R. Petta, A. C. Johnson, J. M. Taylor, E. A. Laird, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson and A. C. Gossard, Science, 309, 2180 (2005). 35. S. K. Sarkar, T. Bose and S. Bandyopadhyay, Phys. Low Dim. Struct. (in press). 36. S. N. Molotkov and S. N. Nazin, JETP Lett., 62, 273 (1995). 37. A. M. Bychkov, L. A. Openov and I. A. Semenihin, JETP Lett., 66, 298 (1997). 38. P. Bakshi, private communication. P. Bakshi, D. Broido and K. Kempa, J. Appl. Phys., 70, 5150 (1991). 39. C. S. Lent, P. D. Tougaw, W. Porod and G. H. Bernstein, Nanotechnology, 4, 49 (1993). 40. S. Bandyopadhyay P. and V. Roychowdhury, Jpn. J. Appl. Phys., 35, Part 1, 3350 (1996). 41. R. P. Cowburn and M. E. Welland, Science, 287, 1466 (2000). 42. M. Anantram and V. P. Roychowdhury, J. Appl. Phys., 85, 1622 (1999). 43. S. Bandyopadhyay P. and V. Roychowdhury, Jpn. J. Appl. Phys., Pt. 1, 35, 3353 (1996). 44. S. Bandyopadhyay, Microstruct., 37, 77 (2005). 45. Jean-Pierre Leburton, communication. 46. R. W. Keyes and R. Landauer, IBM J. Res. Develop., 14, 152 (1970); R. Superlat. private 18 Landauer, Int. J. Theor. Phys., 21, 283 (1982). 47. Ralph K. Cavin, Victor V. Zhirnov, James A. Hutchby and George I. Bourianoff, Fluctuation and Noise Letters, 5, C29 (2005). 48. Patrick P. Gelsinger, Proc. of the IEEE International Solid State Circuits Conference (2001). 49. 2003 Technology International Roadmap for Semiconductors, published by the Semiconductor Industry Association, Austin, Texas. 50. See, for example, C. Bennett, IBM J. Res. Develop., 17, 525 (1973). 51. S. Bandyopadhyay P. and V. Proceedings of the Roychowdhury, Conference International on Superlattices and Microstructures, Liege, Belgium, 1996, Also in Superlat. Microstruct., 22, 411 (1997). 52. S. Bandyopadhyay, A. Balandin, V. P. Roychowdhury and F. Vatan, Superlat. Microstruct., 23, 445 (1998). 53. D. Loss and D. P. DiVincenzo, Phys. Rev. A, 57, 120 (1998). 54. E. Fredkin and T. Toffoli, Int. J. Theor. Phys., 21, 219 (1982). 55. S. Lloyd, Science, 261, 1569 (1993). 19 

1609.07529

1

1609

Chirality-driven orbital magnetic moments as a new probe for topological magnetic structures

When electrons are driven through unconventional magnetic structures, such as skyrmions, they experience emergent electromagnetic fields that originate several Hall effects. Independently, ground state emergent magnetic fields can also lead to orbital magnetism, even without the spin-orbit interaction. The close parallel between the geometric theories of the Hall effects and of the orbital magnetization raises the question: does a skyrmion display topological orbital magnetism? Here we first address the smallest systems with nonvanishing emergent magnetic field, trimers, characterizing the orbital magnetic properties from first-principles. Armed with this understanding, we study the orbital magnetism of skyrmions, and demonstrate that the contribution driven by the emergent magnetic field is topological. This means that the topological contribution to the orbital moment does not change under continous deformations of the magnetic structure. Furthermore, we use it to propose a new experimental protocol for the identification of topological magnetic structures, by soft x-ray spectroscopy.

cond-mat.mes-hall

cond-mat

Chirality-driven orbital magnetic moments as a new probe for topological magnetic structures Manuel dos Santos Dias,1, ∗ Juba Bouaziz,1 Mohammed Bouhassoune,1 Stefan Blugel,1 and Samir Lounis1, † 1Peter Grunberg Institut and Institute for Advanced Simulation, Forschungszentrum Julich & JARA, D-52425 Julich, Germany (Dated: September 4, 2018) Abstract When electrons are driven through unconventional magnetic structures, such as skyrmions, they experience emergent electromagnetic fields that originate several Hall effects. Independently, ground state emergent magnetic fields can also lead to orbital magnetism, even without the spin- orbit interaction. The close parallel between the geometric theories of the Hall effects and of the orbital magnetization raises the question: does a skyrmion display topological orbital magnetism? Here we first address the smallest systems with nonvanishing emergent magnetic field, trimers, characterizing the orbital magnetic properties from first-principles. Armed with this understand- ing, we study the orbital magnetism of skyrmions, and demonstrate that the contribution driven by the emergent magnetic field is topological. This means that the topological contribution to the orbital moment does not change under continous deformations of the magnetic structure. Fur- thermore, we use it to propose a new experimental protocol for the identification of topological magnetic structures, by soft x-ray spectroscopy. 6 1 0 2 p e S 3 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 2 5 7 0 . 9 0 6 1 : v i X r a 1 INTRODUCTION The magnetic moment has two contributions: the spin magnetic moment and the orbital magnetic moment, which are due to lifting of spin and orbital degeneracy, respectively. The familiar mechanism lifting the orbital degeneracy is the spin-orbit coupling (SOC), HSOC = ξ L · S, which leads to an orbital moment tied to the spin moment. A more general picture emerges by analogy with the classic orbital moment, a closed current loop. In quantum physics, ground states hosting bound currents also require lifting of orbital degeneracy. Such ground state currents have been proposed for magnetic structures where the magnetic moments do not all lie in the same plane, i.e. with nonvanishing scalar spin chirality, C123 = S1 ·(cid:0)S2 × S3 (cid:1) (cid:54)= 0 [1 -- 3]. Ground state magnetic structures with C123 (cid:54)= 0 can be stabilized by SOC-driven interactions, such as the magnetic anisotropy or the Dzyaloshinskii-Moriya coupling, or by interactions independent of SOC, such as frustrated bilinear exchange interactions [4] or higher-order exchange interactions, exemplified by the biquadratic and four-spin interactions [5]. Remarkably, electronic structure calculations have predicted orbital moments without SOC [6, 7], but the properties of these orbital moments and their usefulness are unexplained and unexplored. Electrons flowing through a magnetic system couple to emergent electromagnetic fields, leading to several Hall effects [8 -- 11]. Noncoplanar magnetic structures (C123 (cid:54)= 0) have a finite emergent magnetic field in their ground state [9], with the most well-known examples being skyrmions [12 -- 15]. The scalar spin chirality, C123, is closely related to the emergent magnetic field [10, 11], a natural concept in the geometric theories of the Hall effects and of the orbital magnetization [16]. The net flux of the emergent magnetic field in a skyrmion is quantized, and corresponds to the topological charge or skyrmion number of the magnetic structure [9]. Ref. 7 suggested a link between the emergent magnetic field and the orbital moments, but did not atttempt to investigate it. Possible consequences of the topological properties of the magnetic structure on the orbital magnetism remain open. Establishing the topological character of a given magnetic structure experimentally is challenging. The topological Hall effect, driven by the emergent magnetic field, is the most direct signature, but all other contributions to the Hall signal have to be unpicked and care- fully subtracted [8, 17, 18]. The topology can also be ascertained via full knowledge of the three-dimensional magnetic structure, which can be mapped via small-angle neutron scat- 2 tering [13], Lorentz force microscopy [14], scanning tunneling microscopy [19], and soft x-ray magnetic circular dichroism (XMCD) adapted for microscopy [20 -- 23]. All these techniques focus on the spin magnetism, ignoring the orbital aspect. In this work, we consider three aspects of the chirality-driven orbital moment physics. First, we characterize their properties using symmetry and the underlying electronic struc- ture, clearly separating the SOC-driven from the chirality-driven orbital moments, by per- forming first-principles calculations for magnetic trimers. Then, we show that the chirality- driven orbital moments inherit the topological properties of the magnetic structure that generates them, focusing on skyrmionic structures. Lastly, we exploit the distinct properties of the chirality-driven and SOC-driven orbital moments to propose a new experimental pro- tocol, based on XMCD, that can directly establish whether a given sample has a topological magnetic structure. 3 RESULTS Chirality-driven orbital moments in magnetic trimers A nonvanishing scalar spin chirality requires at least three magnetic atoms, so as proto- types we take magnetic trimers formed by Cr, Mn, Fe and Co atoms. They are supported on the Cu(111) surface (see Fig. 1(a) for the atomic structure), a common choice of sub- strate with weak SOC. The electronic structure of a target magnetic state is found using constrained density functional theory (DFT) calculations [24 -- 26] (see Methods and Supple- mentary Note S1). Three kinds of magnetic structures are considered: ferromagnetic (F), chiral right-handed (R) and chiral left-handed (L), Fig. 1(b -- d). If the orbital moments de- pend on the chirality of the magnetic state (C123 (cid:54)= 0), the R and L structures should show dissimilar behavior; the F structures serve as reference (C123 = 0). First we consider the Fe trimer. Varying the polar angle θ from 0 to 90◦ brings the spin moments from pointing normal to the surface to lie in the surface plane. The spin FIG. 1. Atomic and magnetic structures of trimers on Cu(111). (a) Atomic structure. The magnetic atoms are represented by golden spheres, and part of the Cu(111) substrate is represented by the blue spheres, indicating Cu atoms. The orientations of the spin moments are given by red arrows, and fixed as ni = (cos φi sin θ, sin φi sin θ, cos θ) for i = 1, 2, 3. The surface normal defines the z-axis and the polar angle θ. (b -- d) Magnetic structures, top view (θ = 45◦). The choice of azimuthal angles φi defines the (b) ferromagnetic (F: φ1 = φ2 = φ3 = 0◦), (c) chiral right-handed (R: φ1 = 0◦, φ2 = +120◦, φ3 = −120◦) and (d) chiral left-handed (L: φ1 = 0◦, φ2 = −120◦, φ3 = +120◦). Atom #1 (encircled) is chosen to have the same orientation of the spin moment in all structures. 4 1 (a) (b) (c) (d) 1 1 2 3 z θ F R L FIG. 2. Orbital magnetic moment for atom #1 of trimers on Cu(111). (a) Fe #1 in Fe trimer: variation of the z-component of the orbital moment with the direction of the spin moment, for the three magnetic structures in Fig. 1(b -- d) (black, red and blue curves). The green curve shows the average of the data for the two chiral magnetic structures. (b) Fe #1 in Fe trimer: difference between the orbital moments for the two chiral magnetic structures including SOC, and orbital moment for R excluding SOC. (c -- f) Nonvanishing components of the orbital moment for atom #1 in Cr, Mn, Fe and Co trimers, respectively, computed excluding the SOC, in the R structure, Fig. 1(c). moment of Fe #1 follows the same angular path for all three kinds of magnetic structures, as emphasized in Fig. 1(b -- d). In Fig. 2(a) we focus on the projection of its orbital moment on the z-axis. We find a nearly cos θ dependence for the F sweep, as the orbital moment follows the direction of the spin moment very closely. In contrast, the two chiral structures show opposite deviations from the F curve, with their average leading to a cos θ-like trend. Fig. 2(b) reveals that the difference between the R and L curves has a distinct angular dependence. Redoing the calculations without SOC, the orbital moment for the F structure vanishes, while the orbital moments for R and L are equal in magnitude and opposite in sign, reproducing the orbital moment difference computed with SOC. Our results demonstrate that the orbital moment comprises two parts: a SOC-driven part, dominated by the local atomic SOC; and a nonlocal chirality-driven part (C123), determined by the entire magnetic structure of the trimer. If the magnetic structure is chiral, i.e. C123 (cid:54)= 0, but SOC is absent, the orbital moments persist. 5 0.000.060.12Orbital moment mZ (µB)SOI FSOI RSOI LSOI (R+L)/20306090Polar angle θ (°)0.000.030.06SOI (R-L)/2No SOI R-0.008-0.0040Orbital moment (µB)00.020.04Orbital moment (µΒ)0306090Polar angle θ (°)-0.008-0.00400306090Polar angle θ (°)-0.04-0.020mxmz(c) (d) (f) (e) (a) (b) Cr Mn Fe Co Fig. 2(c -- f) shows the two nonvanishing components of the chiral orbital moment of atom #1, for the four different trimers. The properties of the chiral orbital moments depend on the electronic structure of the trimers (see Supplementary Note S1), in particular on which orbitals are near the Fermi energy. Fe3 and Co3 have partially filled d-states, with orbital degeneracies that can be easily lifted by the SOC or by the chiral magnetic structure, leading to sizeable orbital moments. In contrast, Cr3 and Mn3 are close to half-filling, and so the orbital moments are one order of magnitude smaller. The orientation of the chiral orbital moment depends additionally on the local site symmetry (here Cs; the global symmetry of the trimer is C3v). Counterintuitively, and in contrast to the SOC-driven orbital moment, the chiral orbital moment is independent of the absolute real-space orientation of the spin moments: if all spin moments are rotated together in a way that leaves C123 unchanged, the chiral orbital moment is also unchanged. 6 Chirality-driven topological orbital moments in skyrmions Armed with our understanding of the chirality-driven orbital moments provided by the study of the trimers, we now investigate the role of the topology of the spin structure. Skyrmions are the most well-known topological magnetic structures, making them a natural target. After Ref. 9, we define the spin structure by n(x, y) =(cid:0)cos(mφ + γ) sin θ(r) , sin(mφ + γ) sin θ(r) , cos θ(r)(cid:1) , (1) assuming the skyrmion center to be at the origin. In polar coordinates we have (x, y) = (r cos φ, r sin φ), m is the vorticity, γ is the helicity, and θ(r) is the radial polar angle profile. An integer skyrmion number is obtained by integrating the emergent magnetic flux [9, 10], (cid:90)(cid:90) (cid:90)(cid:90) (cid:19) (cid:18)∂n ∂x Nsk = 1 4π dx dy Bsk(x, y) = 1 4π dx dy n(x, y) · × ∂n ∂y = −m . (2) The rightmost integrand generalizes C123 to the continuum case. For chiral skyrmions Nsk = −1, with θ = π in the center. The skyrmions in MnSi are Bloch-like (γ = π/2) [9], while FIG. 3. Skyrmionic magnetic structures. (a) Hexagonal unit cell for the tight-binding model, containing 961 magnetic sites. It is 8.4 nm wide, using the lattice constant of the Pd/Fe/Ir(111) system. The cones show the orientation of the spin moments, and are colored by their polar angle. The skyrmion radius, Rsk, is defined by θ(Rsk) = 90◦. (b -- e) Top view of structures with different skyrmion numbers, Nsk = −2,−1, +1, +2, respectively (see Eq. (2); here γ = 0). The chiral skyrmions in Pd/Fe/Ir(111) correspond to Nsk = −1. Achiral skyrmions have Nsk = +1. 7 (e) (a) ← Rsk → (b) (c) (d) θ (º) Nsk = -2 Nsk = -1 Nsk = +1 Nsk = +2 the skyrmions in Pd/Fe/Ir(111) are N´eel-like (γ = 0) [19]. As the skyrmion number is independent of the helicity, see Eq. (2), we set γ = 0 in the following. The skyrmion profile θ(r) has been experimentally determined for the Pd/Fe/Ir(111) system [19], and depends on the applied magnetic field. The interplay between the magnetic and the electronic structure has attracted attention in connection to tunneling experiments [27, 28]. To analyze the orbital magnetism of skyrmions, we adopt mainly a tight-binding model, due to its transparency and ability to model large systems (see Methods and Supplementary Note S2). We consider a hexagonal unit cell, Fig. 3(a), with periodic boundary conditions. The meaning of Nsk in Eq. (2) is illustrated in Fig. 3(b -- e); its connection to the orbital moments will be examined in the following. For comparison, we also studied the largest skyrmions accessible with DFT [28, 29]. These comprise 73 Fe atoms within an embedding cluster of 211 atoms, see Methods. Fig. 4(a -- b) shows the emergent magnetic flux per site Bsk (Eq. (2)), which maps the local scalar spin chirality, for two skyrmion sizes. Fig. 4(c -- d) displays the corresponding orbital FIG. 4. Chirality-driven orbital magnetic moments in skyrmions of different sizes (Nsk = −1). (a -- b) Emergent magnetic flux per site, see Eq. (2), for the skyrmion profiles cor- responding to Rsk = 1.9 nm and to Rsk = 0.8 nm, corresponding to applied magnetic fields of B = 1.1 T and B = 2.7 T, respectively. (c -- d) orbital moment per site (in µB) generated by the respective emergent magnetic fluxes shown in (a -- b), without SOC, from the tight-binding model for Pd/Fe/Ir(111). 8 (a) (b) (c) (d) B = 1.1 T Rsk = 1.9 nm B = 2.7 T Rsk = 0.8 nm moment distributions, computed directly by leaving out the SOC. The orbital moments are concentrated in regions of large noncollinearity, signaled by large Bsk. This confirms that the emergent magnetic field is the source of the chirality-driven orbital moments. From Fig. 4(c -- d) the magnitude of the chirality-driven local orbital moments is seen to depend strongly on the skyrmion radius. However, the total chiral orbital moment does not, as shown in Fig. 5(a), both for chiral skyrmions (Nsk = −1) and for the other skyrmionic structures sketched in Fig. 3(b -- e). Strikingly, the magnitude of the total chiral orbital mo- ment is nearly constant, for Rsk > 0.5 nm. Fig. 4 exposed the connection between Bsk (see Eq. (2)) and the orbital moments, so the total chiral orbital moment must inherit the topo- logical properties of a skyrmionic structure, and thus be insensitive to deformations of the spin structure that preserve Nsk. Fig. 5(a) also shows that the orbital moments are integer multiples of the total chiral orbital moment of the Nsk = −1 skyrmion. First-principles calculations for the largest attainable skyrmions (73 Fe atoms) confirm the existence of chi- ral orbital moments of the same order of magnitude as those found with the tight-binding model, see two data points in Fig. 5(a). FIG. 5. Signatures of chirality-driven orbital magnetic moments in skyrmions. (a) Net chirality-driven orbital moments (computed without SOC) and (b) topological orbital magneti- zation ratio, TOMR, see Eq. (3) (computed with SOC), for the skyrmionic structures shown in Fig. 3(b -- e), using the tight-binding model with 961 magnetic sites. The skyrmion radius can be varied experimentally by changing the applied magnetic field. The TOMR can be detected in an XMCD experiment by comparing the skyrmion lattice phase with a reference ferromagnetic phase. Two additional datapoints in (a) mark the chirality-driven orbital moments from DFT calculations for the largest skyrmions first described in [28], for Pd/Fe/Ir(111) (×) and Pd2/Fe/Ir(111) (+), which consist of 73 Fe atoms. 9 0123Rsk (nm)-404TOMR (%)Nsk = -2Nsk = -1Nsk = +1Nsk = +20123Rsk (nm)-202morb (µB)Nsk = -2Nsk = -1Nsk = +1Nsk = +2(a) (b) Experimental signatures of topological magnetic structures The first-principles calculations for the magnetic trimers and the tight-binding model for a skyrmion lattice show that the SOC-driven and the chiral contribution to the atomic orbital moments depend on the spin structure in distinct ways. Our calculations confirm that the SOC contribution to the atomic orbital moment is mostly driven by the local SOC of the corresponding atom. If the chiral contribution is absent (e.g. the sample is ferromagnetic), there is a direct proportionality between the atomic orbital and spin moments, and thus between the net orbital and spin moments, Morb ∝ Mspin. This proportionality between the SOC-driven net orbital moment and the net spin moment should hold also for a noncollinear magnetic structure. An apparent deviation from this proportionality law is a signature of the presence of the chiral contribution to the orbital moment, and we propose that this can be exploited experimentally. The XMCD sum rules [30 -- 33] provide the net spin moment (Mspin) and orbital moment (Morb) separately. These measurements are usually performed under an applied external magnetic field, to ensure a net ferromagnetic component of the sample magnetization. The XMCD signal then leads to the average spin and orbital moments projected on the direction of incidence of the x-ray beam. All this applies equally well for a sample in a noncollinear magnetic state, as long as a net ferromagnetic component is present, which is ensured by the external field. The XMCD effect can also be used as a magnetic microscopy technique, as shown in Refs. [20 -- 23]. We propose the following experimental protocol to detect the topological chiral orbital moments in a skyrmion-hosting sample, for instance, Pd/Fe/Ir(111). Applying a sufficiently large external magnetic field, the sample is driven to the field-polarized or ferromagnetic (F) state. XMCD measurements then provide the net spin and orbital moments, Mspin(F) and Morb(F) = α Mspin(F), where the orbital moment is driven purely by SOC. Reducing the strength of the applied magnetic field, the sample enters the skyrmion phase (Sk), and the net orbital moment is now Morb(Sk) = Morb(SOC) + Morb(chiral) ≈ α Mspin(Sk) + Morb(chiral). Here the main approximation is assuming the constant of proportionality α to be independent of the magnetic structure. The topological nature of the skyrmion spin structure generates the topological chiral contribution. We then expect a nonvanishing 10 topological orbital magnetization ratio (TOMR) to be detected: TOMR = Morb(Sk) Morb(F) − Mspin(Sk) Mspin(F) ≈ Morb(chiral) Morb(F) ∝ Nsk . (3) An advantage of forming these ratios is that unknowns in the XMCD sum rules providing the net spin and orbital moments from the x-ray absorption intensities will mostly cancel out [30 -- 33]. Fig. 5(b) shows the expected behavior of the TOMR using the tight-binding model of a skyrmion lattice with SOC, and comparison with Fig. 5(a) verifies the topological signature. Varying the applied magnetic field in the skyrmion phase (thus changing Rsk) has a small impact on the TOMR, which further corroborates the topological origin. Thus, the detection of the TOMR requires no theoretical input, only the ability to drive a sample from a reference ferromagnetic state into a (possibly unknown) noncollinear state. If the TOMR is finite and also insensitive to changes in the noncollinear magnetic structure (for instance driven by the external magnetic field), it is a strong experimental sign of the topological character of the magnetic structure. Comparing Fig. 5(b) with Fig. 5(a) also shows that the magnitude and sign of the TOMR determined by XMCD can be used to discriminate between chiral (Nsk = −1) and achiral (Nsk = +1) skyrmions, or more complex skyrmionic structures. As there is no simple rule for predicting even the sign of the TOMR, theoretical input on the properties of the electronic structure of the sample is needed to allow for definite conclusions on this point. 11 DISCUSSION We have shown that orbital magnetic moments arise in magnetic materials not only from the spin-orbit interaction but also from the emergent magnetic field due to nonvan- ishing scalar spin chirality of the magnetic structure. The chirality-driven orbital magnetic moments have properties distinct from the SOC-driven ones, and have comparable mag- nitudes. The only requirement is a finite scalar spin chirality, so they should be present both in small clusters, wires, thin films and in bulk samples, as long as the spin structure is noncoplanar. For topological magnetic structures, the chirality-driven orbital magnetic moments inherit the underlying topology through the emergent magnetic flux that drives them, and so can be used to fingerprint skyrmionic magnetic structures. These topological chiral orbital moments do not change appreciably under deformations of the spin structure that leave its topology unchanged, in contrast to the usual SOC-driven ones. They present a new way to characterize and investigate candidate materials for skyrmionic applications, via experimental determination of their orbital magnetic properties, with soft x-ray or other appropriate optical measurements. From a different perspective, comparing the spin and orbital magnetic moment distributions yields a real-space map of the emergent magnetic field in topological magnetic structures. METHODS First-principles electronic structure calculations First-principles calculations for the trimers and skyrmions are performed within DFT, as implemented in the Korringa-Kohn-Rostoker Green function method [26, 29], employing the scalar-relativistic approximation and the local spin density approximation parametrized by Vosko, Wilk and Nusair [34]. The SOC potential around each atom, Vsoc(r) = ξ(r) L · σ, is self-consistently included when required. We make use of a real-space embedding procedure, where the trimers or the isolated magnetic skyrmions are embedded in a non-perturbed host. For the trimer case the host is the Cu(111) surface, while for skyrmions two types of hosts are considered: Pdn/Fe/Ir(111) (n = 1, 2) in their ferromagnetic phase [28]. The skyrmion spin structure is self-consistently determined with SOC, for an embedded cluster containing 73 Fe atoms and 211 atoms in total. To extract the chiral orbital moments for the skyrmions, one 12 single iteration is performed without the SOC. A brief discussion of the electronic structure calculations for trimers is given in Supplementary Note S1. Tight-binding electronic structure calculations for large skyrmions The tight-binding model for the skyrmionic structures is constructed using the density of states from ferromagnetic DFT calculations for Pd/Fe/Ir(111). We consider two orbitals, x2−y2(cid:105) and xy(cid:105), degenerate for the hexagonal lattice (C3v symmetry), and a Hamiltonian with local exchange coupling to prescribed spin directions and hopping to nearest-neighbors only. We employ the skyrmion profile extracted in Ref. [19]. A brief discussion of the rationale behind the model and its construction is given in the Supplementary Note S2. Data availability The data that support the findings of this study are available from the corresponding authors upon request. REFERENCES ∗ m.dos.santos.dias@fz-juelich.de † s.lounis@fz-juelich.de [1] G. Tatara and N. Garcia, "Quantum toys for quantum computing: Persistent currents con- trolled by the spin Josephson effect," Phys. Rev. Lett. 91, 076806 (2003). [2] R. Shindou and N. Nagaosa, "Orbital ferromagnetism and anomalous Hall effect in antiferro- magnets on the distorted fcc lattice," Phys. Rev. Lett. 87, 116801 (2001). [3] L. N. Bulaevskii, C. D. Batista, M. V. Mostovoy, and D. I. Khomskii, "Electronic orbital currents and polarization in Mott insulators," Phys. Rev. B 78, 024402 (2008). [4] S. Lounis, "Non-collinear magnetism induced by frustration in transition-metal nanostructures deposited on surfaces," Journal of Physics: Condensed Matter 26, 273201 (2014). 13 [5] Ph. Kurz, G. Bihlmayer, K. Hirai, and S. Blugel, "Three-dimensional spin structure on a two-dimensional lattice: Mn/Cu (111)," Phys. Rev. Lett. 86, 1106 (2001). [6] K. Nakamura, T. Ito, and A. J. Freeman, "Curling spin density and orbital structures in a magnetic vortex core of an Fe quantum dot," Phys. Rev. B 68, 180404 (2003). [7] M. Hoffmann, J. Weischenberg, B. Dup´e, F. Freimuth, P. Ferriani, Y. Mokrousov, and S. Heinze, "Topological orbital magnetization and emergent Hall effect of an atomic-scale spin lattice at a surface," Phys. Rev. B 92, 020401 (2015). [8] T. Schulz, R. Ritz, A. Bauer, M. Halder, M. Wagner, C. Franz, C. Pfleiderer, K. Everschor, M. Garst, and A. Rosch, "Emergent electrodynamics of skyrmions in a chiral magnet," Nat. Phys. 8, 301 -- 304 (2012). [9] N. Nagaosa and Y. Tokura, "Topological properties and dynamics of magnetic skyrmions," Nat. Nanotechnol. 8, 899 -- 911 (2013). [10] K. Everschor-Sitte and M. Sitte, "Real-space berry phases: Skyrmion soccer," J. Appl. Phys. 115, 172602 (2014). [11] C. Franz, F. Freimuth, A. Bauer, R. Ritz, C. Schnarr, C. Duvinage, T. Adams, S. Blugel, A. Rosch, Y. Mokrousov, and C. Pfleiderer, "Real-space and reciprocal-space berry phases in the Hall effect of Mn1−xFexSi," Phys. Rev. Lett. 112, 186601 (2014). [12] A. N. Bogdanov and D. A. Yablonskii, "Thermodynamically stable "vortices" in magnetically ordered crystals. the mixed state of magnets," Sov. Phys. JETP 68, 101 -- 103 (1989). [13] S. Muhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. Boni, "Skyrmion lattice in a chiral magnet," Science 323, 915 -- 919 (2009). [14] X. Z. Yu, Y. Onose, N. Kanazawa, J. H. Park, J. H. Han, Y. Matsui, N. Nagaosa, and Y. Tokura, "Real-space observation of a two-dimensional skyrmion crystal," Nature 465, 901 -- 904 (2010). [15] S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Kubetzka, R. Wiesendanger, G. Bihlmayer, and S. Blugel, "Spontaneous atomic-scale magnetic skyrmion lattice in two dimensions," Nat. Phys. 7, 713 -- 718 (2011). [16] D. Xiao, M. C. Chang, and Q. Niu, "Berry phase effects on electronic properties," Rev. Mod. Phys. 82, 1959 -- 2007 (2010). [17] A. Neubauer, C. Pfleiderer, B. Binz, A. Rosch, R. Ritz, P. G. Niklowitz, and P. Boni, "Topological Hall effect in the A phase of MnSi," Phys. Rev. Lett. 102, 186602 (2009). 14 [18] N. A. Porter, J. C. Gartside, and C. H. Marrows, "Scattering mechanisms in textured FeGe thin films: Magnetoresistance and the anomalous Hall effect," Phys. Rev. B 90, 024403 (2014). [19] N. Romming, A. Kubetzka, C. Hanneken, K. von Bergmann, and R. Wiesendanger, "Field- dependent size and shape of single magnetic skyrmions," Phys. Rev. Lett. 114, 177203 (2015). [20] J. Li, A. Tan, K.W. Moon, A. Doran, M.A. Marcus, A.T. Young, E. Arenholz, S. Ma, R.F. Yang, C. Hwang, and Z.Q. Qiu, "Tailoring the topology of an artificial magnetic skyrmion," Nat. Commun. 5, -- (2014). [21] O. Boulle, J. Vogel, H. Yang, S. Pizzini, D. de Souza Chaves, A. Locatelli, T. O. Mente¸s, A. Sala, L. D. Buda-Prejbeanu, O. Klein, M. Belmeguenai, Y. Roussign´e, A. Stashkevich, S. M. Ch´erif, L. Aballe, M. Foerster, M. Chshiev, S. Auffret, I. M. Miron, and G. Gaudin, "Room-temperature chiral magnetic skyrmions in ultrathin magnetic nanostructures," Nature Nanotech. 11, 449 -- 454 (2016). [22] C. Moreau-Luchaire, C. Moutafis, N. Reyren, J. Sampaio, C. A. F. Vaz, N. Van Horne, K. Bouzehouane, K. Garcia, C. Deranlot, P. Warnicke, P. Wohlhuter, J.-M. George, M. Weigand, J. Raabe, V. Cros, and A. Fert, "Additive interfacial chiral interaction in multi- layers for stabilization of small individual skyrmions at room temperature," Nature Nanotech. 11, 444 -- 448 (2016). [23] S. Woo, K. Litzius, B. Kruger, M.-Y. Im, L. Caretta, K. Richter, M. Mann, A. Krone, R. M. Reeve, M. Weigand, P. Agrawal, I. Lemesh, M.-A. Mawass, P. Fischer, M. Klaui, and G. S. D. Beach, "Observation of room-temperature magnetic skyrmions and their current-driven dy- namics in ultrathin metallic ferromagnets," Nat. Mater. 15, 501 -- 506 (2016). [24] P. H. Dederichs, S. Blugel, R. Zeller, and H. Akai, "Ground states of constrained systems: Application to cerium impurities," Phys. Rev. Lett. 53, 2512 -- 2515 (1984). [25] B. Ujfalussy, X. D. Wang, D. M. C. Nicholson, W.A. Shelton, G. M. Stocks, Y. Wang, and B. L. Gyorffy, "Constrained density functional theory for first principles spin dynamics," J. Appl. Phys. 85, 4824 -- 4826 (1999). [26] N. Papanikolaou, R. Zeller, and P. H. Dederichs, "Conceptual improvements of the KKR method," J. Phys.: Condens. Matter 14, 2799 (2002). [27] C. Hanneken, F. Otte, A. Kubetzka, B. Dup´e, N. Romming, K. von Bergmann, R. Wiesendan- ger, and S. Heinze, "Electrical detection of magnetic skyrmions by tunnelling non-collinear magnetoresistance," Nat. Nanotechnol. 10, 1039 -- 1042 (2015). 15 [28] D. M. Crum, M. Bouhassoune, J. Bouaziz, B. Schweflinghaus, S. Blugel, and S. Lounis, "Perpendicular reading of single confined magnetic skyrmions," Nat. Comms. 6, 8541 (2015). [29] D. S. G. Bauer, Development of a relativistic full-potential first-principles multiple scattering Green function method applied to complex magnetic textures of nano structures at surfaces, Ph.D. thesis, RWTH Aachen (2014). [30] B. T. Thole, P. Carra, F. Sette, and G. van der Laan, "X-ray circular dichroism as a probe of orbital magnetization," Phys. Rev. Lett. 68, 1943 -- 1946 (1992). [31] P. Carra, B. T. Thole, M. Altarelli, and X. Wang, "X-ray circular dichroism and local magnetic fields," Phys. Rev. Lett. 70, 694 -- 697 (1993). [32] C. T. Chen, Y. U. Idzerda, H.-J. Lin, N. V. Smith, G. Meigs, E. Chaban, G. H. Ho, E. Pellegrin, and F. Sette, "Experimental confirmation of the x-ray magnetic circular dichroism sum rules for iron and cobalt," Phys. Rev. Lett. 75, 152 -- 155 (1995). [33] P. Gambardella, S. Rusponi, M. Veronese, S. S. Dhesi, C. Grazioli, A. Dallmeyer, I. Cabria, R. Zeller, P. H. Dederichs, K. Kern, C. Carbone, and H. Brune, "Giant magnetic anisotropy of single cobalt atoms and nanoparticles," Science 300, 1130 -- 1133 (2003). [34] S. H. Vosko, L. Wilk, and M. Nusair, "Accurate spin-dependent electron liquid correlation energies for local spin density calculations: a critical analysis," Can. J. Phys. 58, 1200 -- 1211 (1980). END NOTES Acknowledgments MdSD would like to thank B. Dup´e, S. Heinze and Y. Mokrousov for insightful dis- cussions. This work is supported by the HGF-YIG Programme VH-NG-717 (Functional Nanoscale Structure and Probe Simulation Laboratory -- Funsilab) and the ERC Consolida- tor grant DYNASORE. S.B. acknowledges funding from the European Unions Horizon 2020 research and innovation programme under grant agreement number 665095 (FET-Open project MAGicSky). 16 Author contributions MdSD uncovered the chirality-driven orbital moments in DFT calculations for magnetic trimers, and developed and implemented the tight-binding model for skyrmions. JB and MB performed the DFT calculations for skyrmions and provided input for the construction of the tight-binding model. All authors discussed the results and helped writing the manuscript. Competing financial interests The authors declare no competing financial interests. 17 

1206.4566

1

1206

Shot noise thermometry of the quantum Hall edge states

We use the non-equilibrium bosonization technique to investigate effects of the Coulomb interaction on quantum Hall edge states at filing factor nu=2, partitioned by a quantum point contact (QPC). We find, that due to the integrability of charge dynamics, edge states evolve to a non-equilibrium stationary state with a number of specific features. In particular, the noise temperature of a weak backscattering current between edge channels is linear in voltage bias applied at the QPC, independently of the interaction strength. In addition, it is a non-analytical function of the QPC transparency T and scales as Tln(1/T) at T<< 1. Our predictions are confirmed by exact numerical calculations.

cond-mat.mes-hall

cond-mat

Shot noise thermometry of the quantum Hall edge states D´epartement de Physique Th´eorique, Universit´e de Gen`eve, CH-1211 Gen`eve 4, Switzerland Ivan P. Levkivskyi and Eugene V. Sukhorukov (Dated: October 11, 2018) We use the non-equilibrium bosonization technique to investigate effects of the Coulomb inter- action on quantum Hall edge states at filing factor ν = 2, partitioned by a quantum point contact (QPC). We find, that due to the integrability of charge dynamics, edge states evolve to a non- equilibrium stationary state with a number of specific features. In particular, the noise temperature Θ of a weak backscattering current between edge channels is linear in voltage bias applied at the QPC, independently of the interaction strength. In addition, it is a non-analytical function of the QPC transparency T and scales as Θ ∝ T ln(1/T ) at T ≪ 1. Our predictions are confirmed by exact numerical calculations. PACS numbers: 73.23.-b, 03.65.Yz, 85.35.Ds Rapid experimental progress in the field of the electron transport in one-dimensional systems has unveiled new exciting phenomena inherent in strong, non-perturbative interactions characteristic of such systems. The notable examples are the recent experiments on the energy relax- ation [1], and on non-equilibrium dephasing of quantum Hall (QH) edge states [2, 3]. These chiral electron states may be viewed as quantum analogs of classical skipping orbits arising at the edge of a two dimensional electron system exposed to a perpendicular magnetic field. The aforementioned experiments utilize QPCs to bring edge states of opposite chirality close to each other in order to mix them, thereby inducing electron backscattering. By applying a voltage bias between these edge states, one may create a non-equilibrium state with the electron dis- tribution function in the form of a "double-step" [1] (see the upper panel of Fig. 1). The double-step distribution is characteristic of the ef- fectively free-fermion behavior of electrons in metals [4]. Weak interactions leads to the equilibration of electrons in the long-time limit. At the QH edge, however, this distribution may evolve in a non-trivial way [5] and, in the weak injection regime, through several intermediate asymptotics [6], before reaching the equilibrium state. At the origin of this behavior are the non-perturbative inter- action effects: For the Landau level filling factor ν > 1, when several co-propagating channels coexist at the edge, the strong Coulomb interaction leads to the formation of collective excitations called edge magneto-plasmons [7] (see the lower panel of Fig. 1). Propagating with dif- ferent velocities, these excitations strongly redistribute electrons. We have shown earlier [8], that this process is also responsible for non-monotonic dephasing observed in the resent experiments [3]. Instead of determining directly the electron distribu- tion function, as in Ref. [1], one may investigate the ef- fects of interactions in a non-equilibrium state by measur- ing the effective noise temperature of a system [9]. One way of doing this in a QH system [10] is by attaching a cold Ohmic contact to the co-propagating edge channel, FIG. 1: (Color on-line) Fermionic and bosonic aspects of the edge states physics. Upper panel: At zero temperature, the electron distribution functions of the edge states arriv- ing at the biased QPC are f1(ǫ) = θ(ǫF − ǫ) and f2(ǫ) = θ(ǫF + ∆µ − ǫ). If the transparency T of the QPC is inde- pendent of the energy, then the distribution function of the outgoing electrons is f (ǫ) = (1 − T )f1(ǫ) + T f2(ǫ). Lower panel: Schematic illustration of the strong Coulomb interac- tion effect at the QH edge at filling factor ν = 2. The electron wave packet of the charge e, created in the outer edge chan- nel (lower, black line), decays into two eigenmodes of the edge Hamiltonian, the charged and dipole mode, which propagate with different speeds and carry the charge e/2 in the outer channel. Similar situation arises when an electron is injected in the inner channel (upper, blue line). via the second QPC, as shown in Fig. 2, and measuring the zero-frequency noise power Sbs of the backscattering current jbs: Sbs =Z dthjbs(t)jbs(0)i. (1) The important property of this measurement scheme is that in the absence of the interaction between the chan- nels, one should not expect any influence of the electron injection at the first, source, QPC on the noise at the second, detector, QPC. Therefore, by measuring Sbs as a function of the voltage bias ∆µ, the transparency T of the source QPC, and of the distance D between the QPCs, one may investigate interaction effects and the evolution of a non-equilibrium state initially prepared at the first QPC. In this Letter, we demonstrate that the strong in- teraction and the integrability of the charge dynamics at the QH edge lead to the formation of a non-equilibrium stationary state, which manifests itself in the singular, non-analytical behavior of the effective noise tempera- ture. Effective noise temperature. -- In the regime of weak tunneling at the second, detector QPC, one can write [9]: Sbs = GDZ dǫ{f (ǫ)[1 − fD(ǫ)] + fD(ǫ)[1 − f (ǫ)]}, (2) where GD is the conductance of the QPC, f (ǫ) is the electron distribution function in the inner channel, and fD(ǫ) is the equilibrium distribution in the detector's Ohmic contact. Assuming the Fermi distribution f (ǫ) = fF (ǫ − ǫF ) with the temperature Θeq in the inner edge channel, and with the zero temperature at the detector's Ohmic contact, fD(ǫ) = θ(ǫF − ǫ), one immediately finds that Sbs = (2 ln 2)GDΘeq. Therefore, away from equilib- rium, it is natural to define the effective noise tempera- ture Θ via the expression Sbs ≡ (2 ln 2)GDΘ. (3) On the other hand, since the inner and the outer edge channels are electrically isolated from each other, there is no average current contribution from the first QPC to the inner channel, which may be expressed asR dǫ[f (ǫ) − θ(ǫF − ǫ)] = 0. Combining this identity with the ex- pression (2), one obtains the simple expression for the effective noise temperature: Θ = (1/ln 2)Z ∞ ǫF dǫf (ǫ). (4) Facing strong interactions that cannot be accounted for perturbatively, one may choose to treat tunneling at the first QPC perturbatively with respect to its small transparency T . Recently, using this method, the Ref. [11] has found that the noise temperature Θ is linear in T , while non-perturbative interactions manifest them- selves in the non-trivial power-law dependence of Θ on the voltage bias ∆µ. However, it turns out that far from the injecting QPC, where a non-equilibrium stationary state arises, the perturbation theory fails to correctly de- scribe the behavior of Θ at small T . Very roughly, this happens because the weak partitioning noise at the first QPC generates a correction to the distribution function of the form f (ǫ) ∝ T ∆µ/(ǫ − ǫF ) [6], therefore the in- tegral in Eq. (4) has a logarithmic divergence. At the upper limit, this integral is cut at ǫ − ǫF ∼ ∆µ, since this is the maximum energy provided by the source. At the lower limit, the integral has to be cut at ǫ − ǫF ∼ T ∆µ, due to broadening of the distribution function induced by the noise. This leads to the behavior Θ ∝ T ln(1/T )∆µ at T ≪ 1, i.e., the noise temperature is singular in T and linear in ∆µ, contrary to the prediction of Ref. [11]. In the rest of the paper, we demonstrate this fact rigorously by resumming weak tunneling using the non-equilibrium bosonization technique [12], and investigate various phys- ical regimes in detail. 2 FIG. 2: (Color on-line) Schematics of the measurement of the effective noise temperature. The "double-step" distribu- tion is created at the left (x = 0) voltage-biased QPC of the arbitrary transparency T . The state propagates towards the right (x = D) QPC of the small transparency T ′ ≪ 1, connected to a cold Ohmic contact, and induces the zero- frequency backscattering current noise, Sbs. Thereby, the right QPC serves as a detector of the effective temperature of this noise, Θ ∝ Sbs. The notations for the boson fields describing each QH edge are shown near the corresponding edge channels: the index s = L, M, U enumerates the edges, while the index α = 1, 2 enumerates the edge channels at the same edge at filling factor ν = 2. Model and theoretical method. -- In an experiment, the applied voltage bias ∆µ is typically much smaller than the Fermi energy ǫF . Thus, it is appropriate to use the low-energy effective theory [13] describing edge states at filling factor ν = 2 as collective fluctuations of the charge density ρsα(x), where α = 1, 2 enumerates channels at the QH edge, and s = L, M, U denotes the lower, mid- dle, and upper edge (see Fig. 2). The charge density fields are expressed in terms of chiral boson fields, φsα(x), sat- isfying the commutation relations [φsα(x), φrβ (y)] = iπδsrδαβsgn(x − y), (5) namely, ρsα(x) = (1/2π)∂xφsα(x). The total Hamilto- nian of the system, H = H0 + (A + A′ + h.c.), contains the term describing the edge states H0 = 1 8π2 Xs,α,βZ dxdyVαβ (x − y)∂xφsα(x)∂yφsβ (y), (6) where the kernel, Vαβ(x−y) = 2πvF δαβδ(x−y)+Uαβ(x− y), includes the free-fermion contribution with the Fermi velocity vF , and the Coulomb interaction potential Uαβ. Vertex operators A = t eiφL1(0)−iφM 1(0), A′ = t′eiφM 2(D)−iφU 2(D) (7) describe electron tunneling between the edge channels at the QPCs. The right QPC, serving as a non-invasive detector, is in the weak tunneling regime. Therefore, we treat corresponding operator A′ perturbatively [14]. The backscattering current at the second QPC may be written as jbs = i(A′ − A′†) and, to the leading order in the tunneling operator A′, the noise power (1) of this current reads: Sbs =R dth{A′†(t), A′(0)}i. The relatively straightforward steps lead to the standard result (2), and to the effective noise temperature (4) with f (ǫ) ∝ Z dte−i(ǫ−ǫF )tK(t), K(t) = he−iφM 2(D,t)eiφM 2(D,0)i, (8a) (8b) where the normalization prefactor in (8a) is determined by the condition f (ǫ) = 1 at ǫ → −∞. The average in the definition of K(t) has to be taken with respect to the non-equilibrium state created by the source QPC. Therefore, we apply the non-equilibrium bosonization technique proposed in our earlier work [12]. The Hamiltonian (6), together with the commutation relations (5), generates equations of motion for the fields φsα that have to be accompanied with boundary condi- tions: ∂tφM α(x, t) = − 1 2πXβ Z dyVαβ(x − y)∂yφM β (y, t), (9a) ∂tφM α(0, t) = −2πjα(t). (9b) We place the boundary at the point x = 0, right after the source QPC. At low energies of interest, the charac- teristic length scales are much longer than the screening length d of the Coulomb potential Uαβ(x − y). There- fore, we can neglect its logarithmic dispersion and ap- proximate Uαβ(x − y) = Uαβδ(x − y), and consequently, Vαβ(x − y) = Vαβ δ(x − y). Then, Eqs. (9) acquire a form of first-order differential equations. We solve these equations by diagonalizing the matrix V ≡ Vαβ with the rotation V = S(θ)Λ S†(θ) by the angle θ de- fined as tan2θ = 2V12/(V11 − V22). Then, the spectrum of the collective charge excitations splits in two modes, Λ = diag(u, v), with the speeds u, v = (V11 + V22)/2 ± Imposing the boundary condi- p(V11 − V22)2/4 + V 2 12. tion (9b), we arrive at the solution φM2(x, t) = λ1Q1(tu) + λ2Q2(tu) − λ1Q1(tv) + λ′ 2Q2(tv), (10a) λ1 = π sin 2θ, λ2 = π(1 + cos 2θ), λ′ 2 = 2π − λ2, (10b) R t where we have introduced the injected charges Qα(t) = −∞ dt′jα(t′), and notations tu = t−x/u and tv = t−x/v. Since the edge state dynamics is chiral, and the screened Coulomb interaction is effectively short-range, the fields φM α do not influence fluctuations of the cur- rents jα at the QPC [8, 15]. As a consequence, the electron transport through a single QPC is not affected by the interaction, which seems to be an experimental fact [16]. Therefore, when finding the correlator (8b), one may utilize the free-fermion scattering theory for the statistics of injected charges Qα [9, 17]. Gaussian noise regime. -- It has been shown in Ref. [6] that a weak dispersion of plasmon modes suppresses 3 higher-order cumulants at large distances. Therefore, we first focus on the situation, where the fluctuations of the boson fields may be considered Gaussian. Then, the logarithm of the correlation function (8b) can be writ- ten as ln K(t) = −hφ2 M2(D, t) − 2φM2(D, t)φM2(D, 0) + φ2 M2(D, 0)i/2, where the linear terms in field φM2 vanish, since there is no contribution to the average current in the inner channel. Using Eqs. (10), we obtain: π (cid:17)2 ln K(t) = −2πZ dω sin2(cid:16) ωtD ω2 (1−eiωt)n(cid:16) λ1 sin2(cid:16) ωtD 2 (cid:17)iS2(ω)o, +h1 − λ2λ′ 2 π2 2 (cid:17)S1(ω) (11) where we have introduced the noise power, Sα(ω) = R dteiωthδjα(t)δjα(0)i, and the time delay between the wave packets, tD = D/v − D/u. Since the transport through the injecting QPC is not affected by interactions, the free-fermion scattering ap- proach [9] may be used to obtain Sα(ω) = Sq(ω) + Tα(1 − Tα)Sn(ω), (12) where Sq(ω) = ωθ(ω)/2π is the ground-state (Fermi sea) contribution, and Sn(ω) = P±[Sq(ω ± ∆µ) − Sq(ω)] is the non-equilibrium part. Therefore, in the expression (11) the ground-state and non-equilibrium contributions separate, ln K(t) = − ln ǫF t + ln Kn(t), and the noise temperature (4) may be presented as Θ = − 1 2π ln 2Z dt (t − iη)2 Kn(t), η → 0, (13) where the non-equilibrium contribution reads ln Kn(t) = −4T (1 − T )(λ1/π)2 ×Z 1 0 dx x2 (1 − x) sin2(cid:16) ∆µtx 2 (cid:17) sin2(cid:16) ∆µtDx 2 (14) (cid:17). We note, that the ground-state contribution to the cor- relator K is always Gaussian and is independent of the interactions, because the effect of the injecting QPC on the states below ǫ = ǫF is simply a unitary transforma- tion. Next, we focus on the weak injection regime, T ≪ 1, verify the validity of the perturbation approach with respect to weak tunneling, and show that it may fail. It turns out, that the expansion of Kn with respect to T as Kn = 1 + ln Kn + . . . is dangerous, because ln Kn diverges at large t and tD. More precisely, at distances D ≫ Dex, where Dex ≡ uv/[(u − v)∆µ] is the characteristic length of the energy exchange between edge channels [6], its asymptotic reads: ln Kn = −(λ2 1/2π)T ∆µ min(t, tD). Therefore, to leading order in tunneling at the first QPC, the time integral in Eq. (13) diverges logarithmically. At the short-time limit, this integral should be cut at t ∼ 1/∆µ, where it behaves regularly. At the upper limit, it is cut at either t ∼ 1/(T ∆µ), where ln Kn is not small, and perturbation approach fails, or at t ∼ tD, where ln Kn takes a constant value smaller than 1 if T ∆µtD ≪ 1. Thus, for T ≪ 1 the noise temperature reads 4 Θ ∆µ = λ2 1T 2π2 ln 2( ln(∆µtD), ln(1/T ), if Dex/T ≫ D ≫ Dex, if D ≫ Dex/T . (15) We recall the notations tD = D/v − D/u and Dex = uv/[(u − v)∆µ]. It remains to investigate the noise temperature at short distances, D ≪ Dex. In this case, we can replace sin2(∆µtDx/2) → (∆µtDx/2)2 in Eq. (14). It is more convenient to substitute ln Kn into Eq. (13) and first eval- uate the time integral, and then the integral over x. The result for the noise temperature reads: Θ = λ2 1T t2 D 24π2 ln 2 (∆µ)3, D ≪ Dex. (16) This regime can be viewed as perturbative both with respect to tunneling and interactions. Non-Gaussian noise: exact results. -- To complete our analysis, we investigate the situation, where even at long distances, D ≫ Dex/T , the fluctuations of the edge fields remain non-Gaussian. At such distances, two plasmon modes, arriving with the time delay tD longer than the correlation time 1/∆µ of boundary currents (see Fig. 1), separate the injected charges Qα in Eq. (10a) into un- correlated terms. Therefore, the correlation function K splits in the product of four terms K(t) = χ1(λ1, t)χ1(−λ1, t)χ2(λ2, t)χ2(λ′ 2, t), (17) each taking the form of the generator of full counting statistics (FCS) [17]: χα(λ, t) = heiλQα(t)e−iλQα(0)i. (18) The correlation function (17) is independent of D, i.e., in the limit D ≫ Dex/T electrons in the inner channel do indeed reach a non-trivial stationary state. We note, that the FCS generator of the inner channel at the edge M contains only the Gaussian contribution from the Fermi sea, ln χ2(λ, t) = −(λ2/4π2) ln ǫF t, while the FCS generator at the outer channel, being perturbed by a QPC, acquires additional non-Gaussian part from the transport electrons, ln χ1(λ, t) = −(λ2/4π2) ln ǫF t + ln χn(λ, t). This leads to the expression (13) for the effec- tive noise temperature with Kn(t) = χn(λ1, t)χn(−λ1, t). (19) We stress that in the limit ∆µ ≪ ǫF the non-equilibrium FCS generator χn depends on time only via the dimen- sionless combination ∆µt, which is the consequence of a free-fermion character of the electron transport through a single QPC. Therefore, at distances D ≫ Dex/T the FIG. 3: (Color on-line) The normalized effective noise tem- perature at the detector QPC is plotted as a function of the transparency T of the source QPC, generating a non-Gaussian noise. Solid, blue line: The exact value of Θ/∆µ for a non- equilibrium stationary state at D ≫ Dex/T , evaluated with the help of the determinant representation of the FCS gen- erator [17]. Dashed, blue line: Its asymptotic behavior (21) for T ≪ 1, and similar result for 1 − T ≪ 1. Dotted, red line: The temperature (20) of a locally equilibrium state at D → ∞, extracted from the energy flux in the inner edge channel. noise temperature is always linear in applied voltage bias ∆µ, independently of details of the interaction. In the following, we concentrate on the realistic case of a Coulomb interaction screened at distances d ≫ a, where a is the distance between edge channels. There- fore, one may approximate Uαβ = πu, where u/vF ∼ ln(d/a) ≫ 1, so that θ = π/4 and λ1 = π [8]. The dimensionless function χn(π, t) can be represented as a determinant of a single-particle operator [17] and cal- culated numerically [18]. The result for the normalized noise temperature Θ/∆µ as a function of transparency T of the injecting QPC is shown in Fig. 3. We also plot the normalized temperature Θeq/∆µ of an equilibrium distribution reached by electrons in the inner channel at D → ∞, Θeq/∆µ =p3T (1 − T )/2π2, (20) which is found by comparing the energy flux of electrons π2Θ2 eq/6 to the half of the heat flux ∆µ2T (1 − T )/2 in- jected at the first QPC. One can see in Fig. 3 a singular behavior of Θ at T → 0 and T → 1. In order to describe it analyti- cally, we recall the FCS generator for the tunneling pro- cess: ln χn(λ1, t) = (∆µt/2π)T (eiλ1 − 1) for ∆µt ≫ 1. Note, that this FCS generator is universal, i.e., it does not require an assumption of free-electron transport at the QPC, and reflects the simple fact that tunneling is a Poisson process with all the current cumulants equal to the average current. Substituting this expression into Eqs. (13) and (19), and setting λ1 = π, we find the noise temperature at T ≪ 1 in the non-Gaussian noise regime Θ/∆µ = (2/π2ln 2) T ln(1/T ), D ≫ Dex/T. (21) It differs from the one for the Gaussian noise, Eq. (15), only by a numerical prefactor. To summarize, we have investigated the effects of the integrability of the charge dynamics at QH edge at fill- ing factor ν = 2, where two chiral edge channels coex- ist. We have found that the double-step electron distri- bution, created in one of the channels with the help of a voltage-biased QPC, evolves via several intermediate regimes to a non-equilibrium stationary state. Measur- ing the backscattering current noise in the second, co- propagating channel reveals a non-trivial effect of the integrability and strong inter-channel Coulomb interac- tions: The effective noise temperature Θ of this station- ary state is a non-analytical function of the transparency T , which scales as Θ ∝ T ln(1/T ) at T ≪ 1. We acknowledge support from the Swiss NSF. [1] C. Altimiras et al., Nature Physics 6, 34 (2010); H. le Sueur et al., Phys. Rev. Lett. 105, 056803 (2010); C. Altimiras et al., Phys. Rev. Lett. 105, 226804 (2010). [2] Y. Ji et al., Nature (London) 422, 415 (2003). [3] I. Neder et al., Phys. Rev. Lett. 96, 016804 (2006); I. Neder et al., Nature Physics 3, 534 (2007); P. Roulleau et al., Phys. Rev. Lett. 100, 126802 (2008); L.V. Litvin et al., Phys. Rev. B 78, 075303 (2008). [4] H. Pothier, S. Gu´eron, N. O. Birge, D. Esteve, and M. 5 H. Devoret, Phys. Rev. Lett. 79, 3490 (1997). [5] A. M. Lunde, S. E. Nigg, and Markus Buttiker, Phys. Rev. B 81, 041311(R) (2010); P. Degiovanni, Ch. Gre- nier, G. F´eve, C. Altimiras, H. le Sueur, and F. Pierre, Phys. Rev. B 81, 121302(R) (2010). [6] I.P. Levkivskyi, E.V. Sukhorukov, Phys. Rev. B 85, 075309 (2012). [7] V.A. Volkov and S.A. Mikhailov, Sov. Phys. JETP 67, 1639 (1988); D.B. Chklovskii, B.I. Shklovskii, and L.I. Glazman, Phys. Rev. B 46, 4026 (1992); I.L. Aleiner and L.I. Glazman, Phys. Rev. Lett. 72, 2935 (1994). [8] I.P. Levkivskyi, E.V. Sukhorukov, Phys. Rev. B 78, 045322 (2008). [9] Y.M. Blanter, M. Buttiker, Phys. Rep. 336, 1 (2000). [10] M. Heiblum, private communication. [11] I. Neder, Phys. Rev. Lett. 108, 186404 (2012). [12] I.P. Levkivskyi, E.V. Sukhorukov, Phys. Rev. Lett. 103, 036801 (2009). [13] X.-G. Wen, Phys. Rev. B 41, 12838 (1990); J. Frohlich, A. Zee, Nucl. Phys. B364, 517 (1991). [14] For arbitrary T ′, the exact calculations show that the noise temperature is equal to 1−T ′ times its perturbative value. Therefore, to avoid unnecessary complications, we use the perturbative approach. [15] E.V. Sukhorukov, V.V. Cheianov, Phys. Rev. Lett. 99, 156801 (2007). [16] E. Bieri et al., Phys. Rev. B 79, 245324 (2009). [17] L.S. Levitov, H. Lee, and G.B. Lesovik, J. Math. Phys. 37, 4845 (1996). [18] I.P. Levkivskyi, E.V. Sukhorukov, unpublished. 

1710.04308

1

1710

Influence of thermal effects on stability of nanoscale films and filaments on thermally conductive substrates

We consider films and filaments of nanoscale thickness on thermally conductive substrates exposed to external heating. Particular focus is on metal films exposed to laser irradiation. Due to short length scales involved, the absorption of heat in the metal is directly coupled to the film evolution, since the absorption length and the film thickness are comparable. Such a setup requires self-consistent consideration of fluid mechanical and thermal effects. We approach the problem via Volume-of-Fluid based simulations that include destabilizing liquid metal-solid substrate interaction potentials. These simulations couple fluid dynamics directly with the spatio-temporal evolution of the temperature field both in the fluid and in the substrate. We focus on the influence of the temperature variation of material parameters, in particular of surface tension and viscosity. Regarding variation of surface tension with temperature, the main finding is that while Marangoni effect may not play a significant role in the considered setting, the temporal variation of surface tension (modifying normal stress balance) is significant and could lead to complex evolution including oscillatory evolution of the liquid metal-air interface. Temperature variation of film viscosity is also found to be relevant. Therefore, the variations of surface tensions and viscosity could both influence the emerging wavelengths in experiments. In contrast, the filament geometry is found to be much less sensitive to a variation of material parameters with temperature.

cond-mat.mes-hall

cond-mat

Influence of thermal effects on stability of nanoscale films and filaments on thermally conductive substrates Department of Mathematical Sciences, New Jersey Institute of Technology, Newark, NJ, USA Ivana Seric, Shahriar Afkhami, and Lou Kondic (Dated: July 26, 2021) We consider films and filaments of nanoscale thickness on thermally conductive substrates exposed to external heating. Particular focus is on metal films exposed to laser irradiation. Due to short length scales involved, the absorption of heat in the metal is directly coupled to the film evolution, since the absorption length and the film thickness are comparable. Such a setup requires self- consistent consideration of fluid mechanical and thermal effects. We approach the problem via Volume-of-Fluid based simulations that include destabilizing liquid metal-solid substrate interaction potentials. These simulations couple fluid dynamics directly with the spatio-temporal evolution of the temperature field both in the fluid and in the substrate. We focus on the influence of the temperature variation of material parameters, in particular of surface tension and viscosity. Regarding variation of surface tension with temperature, the main finding is that while Marangoni effect may not play a significant role in the considered setting, the temporal variation of surface tension (modifying normal stress balance) is significant and could lead to complex evolution including oscillatory evolution of the liquid metal-air interface. Temperature variation of film viscosity is also found to be relevant. Therefore, the variations of surface tensions and viscosity could both influence the emerging wavelengths in experiments. In contrast, the filament geometry is found to be much less sensitive to a variation of material parameters with temperature. I. INTRODUCTION particular) leads to satisfactory results, see, e.g. [17]. Metal films of nanoscale thickness are of interest in numerous applications including solar cells, plasmonics related applications, sensing and detection among oth- ers. These applications include various geometries: par- ticles, films, filaments and more complicated shapes. For a recent application-centered review, see [1, 2] for re- cent application-centered reviews. These films are typi- cally exposed to a heat source (pulsed nanosecond laser) and, while in the liquid phase, evolve on a time scale measured in nanoseconds. This evolution is typically unstable due to the presence of destabilizing forces, in particular involving liquid metal-solid substrate interac- tions [3]. In addition to its scientific interest, understand- ing these instabilities and the subsequent dynamics is further motivated by their potential to drive various self- and directed-assembly mechanisms in a variety of con- texts; only some examples are cited here [4 -- 9]. A significant progress has been reached in understand- ing the instability mechanisms by considering essentially isothermal models that assume films and other geome- tries under isothermal conditions [10 -- 14]. In particu- lar for the film geometry, the film-substrate interaction forces are crucial: a destabilizing force is needed for the instability to develop. In our earlier works, such a force has been included in the Navier-Stokes solver [15, 16], and the implemented approach is used in the present work as well. For filaments or other geometries that are characterized by limited spatial extent and the pres- ence of contact lines, capillary effects are known to be dominant. In particular, for the commonly considered filament geometry, it has been shown that simply consid- ering Rayleigh-Plateau instability mechanism with ap- propriately chosen material parameters (contact angle in Clearly, the evolution of metal films and other geome- tries exposed to laser radiation is more complicated than that of an isothermal film. A laser pulse leads to a sig- nificant variation of the temperature field both in the film and in the substrate, resulting in phase change, and in variation of material parameters. The influence of such variation of material parameters with temperature on the stability of films and other metal geometries has been considered only to the limited extend so far [7, 18 -- 21]. Furthermore, the existing studies, some of which we discuss next, have focused mostly on Marangoni effects, related to spatial variation of the film surface tension with temperature. We are not aware of any works in the present context that focus on the influence of temporal variation of surface tension (that is clearly relevant in the case of a time dependent laser pulse), or on the influence of variation of the film viscosity with temperature. These effects are among those discussed in the present work. The Marangoni effect results from the spatial varia- tion of tangential stresses due to temperature depen- In [19], Marangoni effect is dence of surface tension. claimed to be responsible for the change of the average distances between the drops that form during pulse irra- diation of metal films. For computing the temperature field, Ref. [19] used the model, that we reference in the remaining part of this paper as 'reduced' model, that includes two important assumptions (i) that the temper- ature field is slaved to the film thickness, meaning essen- tially that the temperature is completely defined by the current value of the film thickness, and (ii) that the heat flux in the plane of the substrate can be ignored. Next, in a recent modeling and computational study [21], the Marangoni effect was considered (within long wave limit) in a setup that relaxed the assumption (i) above, but still used the assumption (ii). In that work, it was found that the results were dramatically different compared to the ones obtained if the assumption (i) was used. In partic- ular, a regime characterized by an oscillatory instability development has been found. This is in contrast to usual monotonic nature of instability evolution. To summarize, the influence of thermal effects on the evolution of metal films and other geometries has not been studied extensively, and the results of the exist- ing works are not always consistent. This motivates the present paper that focuses on providing further insight by carrying out careful and self-consistent simulations of the evolution of metal films and filaments. The con- sidered approach is mainly computational and is based on a Volume-of-Fluid (VoF) method for solving the fluid mechanical problem, coupled with a thermal solver that computes the temperature field in the metal and in the substrate. Thin film (long wave) limit is used only for the purpose of obtaining basic insight via linear stabil- ity analysis, and the thermal problem is solved fully and self-consistently with the Navier-Stokes equations gov- erning thin film evolution. In particular, the thermal problem considers both the in-plane and out-of-the-plane heat transport: we will see that for the setup consid- ered, considering the in-plane heat diffusion is crucial. The VoF solver includes the interaction between metal and the substrate modeled via disjoining pressure ap- proach [16], as well as efficient calculation of tangential stresses and the resulting Marangoni effect that has been developed recently [22]. We note that in the present work, we discuss the phase change effects only indirectly, by limiting the metal film and filament evolution to the times for which metal temperature is above melting. Fur- thermore, we do not consider possible phase change of the substrate itself [23]. Inclusion of these effects is left for future work. The remainder of this paper is organized as follows. First, in Section II we formulate the governing equations coupling fluid mechanics with the thermal effects. We outline two temperature solutions used in the study of film breakup: first in Section II A 1, we present a reduced model that ignores the in-plane heat conduction, as well as temporal evolution of the film or filament (referred to as the 'reduced' model' from now on) [18]; and second in Section II A 3, we present the numerical 2D tempera- ture solution computed using Gerris (referred to as the 'complete' model'). In Section II A 2, we outline an ana- lytical solution for a flat film, which we use for validating and comparing the two models above. For definiteness, throughout the paper we use the parameters correspond- ing to nickel at the melting temperature, if not specified differently. Section III presents the main findings. First, in Sec- tion III A, using linear stability analysis (LSA) (in long wave limit), we find that the spatial temperature varia- tions in the film can have a stabilizing or destabilizing effect depending on the film thickness: for films thinner than a critical value hc, the temperature variations have 2 a stabilizing effect, and for the films of thickness larger than hc, the temperature variations have a destabilizing effect; such a critical value appears due to the nature of absorption of the laser energy by the film. Second, in Section III B, using the direct numerical simulations, we consider the influence of temperature variation of surface tension on stability of a film, for the reduced and com- plete temperature models. We find that for the reduced model the relevance of Marangoni effect is exaggerated due to the lack of in-plane heat conduction. Further- more, we find that, compared to the spatial variations leading to Marangoni effect, the temporal variation of sur- face tension has significantly stronger effect on the film stability; in other words, the balance of normal stresses (and its dependence on temperature) play much more important role than the variation of tangential stresses that lead to Marangoni effect. The temporal variation of the temperature is found also to influence the film evolution via temperature dependence of viscosity; this effect is discussed in Section III C. Section III D is de- voted to a brief discussion of the expected influence of temperature variation of material properties in physical experiments. Finally, in Section III E, we consider the influence of the thermal effects on the breakup of the liq- uid metal filaments. We find that the influence of thermal effects is weak compared to the capillary ones governing the Rayleigh-Plateau type of instability. The paper is concluded by Section III F, where we also discuss some directions for the future work. II. GOVERNING EQUATIONS AND NUMERICAL METHODS We represent the two-phase flow by the Navier-Stokes equations, where the material properties are phase de- pendent; additional energy equation is discussed later in this section. The surface forces at the interface between two fluids are represented by a body force using the Con- tinuum Surface Force (CSF) method [24 -- 27]. Hence, the equations governing the flow are ρ(∂tu + u · ∇u) = −∇p + ∇ · (2µD) + F, ∇ · u = 0, (1) (2) and the advection of the phase-dependent density ρ (χ) ∂tρ + (u · ∇)ρ = 0, (3) where u = (u, v, w) is the fluid velocity, p is the pressure, ρ(χ) = χρ1 +(1−χ)ρ2 and µ(χ) = χµ1 +(1−χ)µ2 are the phase dependent density and viscosity respectively. D is the rate of deformation tensor, D = (cid:0)∇u + ∇uT(cid:1) /2. Subscripts 1 and 2 correspond to the fluids 1 and 2, re- spectively (see Fig. 1). Here, χ is the characteristic func- tion, such that χ = 1 in the fluid 1, and χ = 0 in the fluid 2. Note that any body force can be included in F. The characteristic function is advected with the flow, thus ∂tχ + (u · ∇)χ = 0. (4) n t Fluid 2 ρ2, µ2 Fluid 1 , µ ρ 1 y z x [[n · T · n]] = σ (x) κ [[n · T · t]] = t · ∇ σ (x) FIG. 1: Schematic of a system with two immiscible fluids and the corresponding boundary conditions. Note that solving Eq. (4) is equivalent to solving Eq. (3). The presence of an interface gives rise to the stress boundary conditions, see Fig. 1. The normal stress boundary condition at the interface defines the stress jump [28, 29] Jn · T · nK = σ (x) κ, (5) where T = −pI + µ(cid:0)∇u + ∇uT(cid:1) is the total stress ten- sor, σ (x) is the surface tension, κ is the curvature of the interface, and n is the unit normal at the interface point- ing out of the fluid 1. The variation of surface tension results in the tangential stress jump at the interface Jn · T · tK = t · ∇σ (x) , (6) which drives the flow from the regions of low surface tension to the ones with high surface tension. Here, t is the unit tangent vector in two dimensions (2D); in three dimensions (3D) there are two linearly indepen- dent unit tangent vectors. Using the CSF method [30], the forces resulting from the normal and tangential stress jump at the interface can be included in the body force F = Fsn + Fst, defined as and Fsn = σ (x) κδs n, Fst = ∇sσ (x) δs, (7) (8) where δs is the Dirac delta function centered at the in- terface, δs n = ∇χ, and ∇s is the surface gradient. The details of the computations of the interfacial curvature and normals in the VoF method can be found in [27], and the implementation of the surface gradients of the surface tension in [22]. The destabilizing mechanism leading to breakup of nanoscale films is modeled by the fluid-solid interaction in the form of a disjoining pressure [14]. The disjoining pressure can be included in the Navier-Stokes equations (1) as a body force specified by y (cid:19)m −(cid:18) h∗ Fvdw (y) = Kπ(cid:20)(cid:18) h∗ σ0(1 − cos θeq) Kπ = y (cid:19)n(cid:21) nδs, , M h∗ n − m M = (m − 1)(n − 1) 3 (9) (10) (11) where σ0 is the surface tension of nickel at the melting temperature, and θeq = 70◦ is the prescribed equilibrium contact angle. We use the exponents m = 3 and n = 2 as in [14]; see also [11, 31] and the references therein for further discussion regarding disjoining pressure models for metal films. The equilibrium film thickness used is h∗ = 1.5 nm, comparable to the values discussed in [14]. Hence, the complete Navier-Stokes equations with the surface forces are as follows ρ(∂tu + u · ∇u) = −∇p + ∇ · (2µD) + σκδs n+ ∇sσδs + Fvdw (y) (12) The details of the implementation of the disjoining pres- sure in the VoF method can be found in [15]. A. Temperature Models The variations of the temperature of a liquid metal film exposed to a pulsed laser can be caused by spatial variability of the pulse itself. However, since the spatial scale of the pulse is typically few orders of magnitude larger than any other relevant length scale [11], we con- sider a spatially homogenous pulse where a temperature variation may result from variable film thickness. This is due to fact that the optical (and energy absorption) properties of the metal depend on the film thickness [32], as we will discuss in what follows. We start this section by considering a simplified ('re- duced') model that ignores various aspects of the heat flow, such as heat transport in the in-plane direction, as well as the convective effects. In Section II A 1 a numer- ical solution of such a model, that has been previously used in the context of metal films [18, 19] is discussed, and in Section II A 2, an analytical solution of the underlying one-dimensional model for heat transport is presented. Complete numerical solution of the energy equation is given in Section II A 3. The comparison of the results obtained using these approaches will allow to gain bet- ter insight into the most important factors determining temperature distribution. 1. Reduced Model for the Temperature of a Thin Film In this section, we outline the reduced model for the metal temperature; a version of such model is discussed 60 50 40 30 20 10 0 0 8 2 7 1 1 7 2 8 1728 1728 10 20 30 40 50 4 4000 3500 3000 2500 2000 1500 1000 500 More details regarding the derivation of the source term and the explanation of the parameters are given in Ap- pendix A; the parameters themselves are specified in Tab. I. The boundary conditions are as follows km ∂Tm ∂y ∂Tm ∂y = 0 at y = h(t, x), = ks ∂Ts ∂y Tm = Ts Ts → T0 at y = 0, at y = 0, as y → −∞, (17) (18) (19) (20) where y = h(x, t) corresponds to the film-air interface, and y = 0 is the film-substrate interface. m, FIG. 2: The average temperature of a metal film, T ∗ as a function of film thickness and time. The red highlighted curve represents the melting temperature of nickel, TM = 1728 K. The parameters used are specified in Tab. I. in [18]. The film -- substrate bilayer is assumed to be in- finitely wide in the in-plane directions. The substrate layer is assumed to be thick compared to the film thick- ness and it is modeled as a semi-infinite medium 0 ≤ y < −∞. Assuming that any variation of film thickness occurs on the scale which is much larger than the film thickness, one could argue that the heat conduction in the in -- plane direction in the film is negligible compared to the conduction in the out-of-plane direction. Although the same argument does not apply to a (thick) substrate, it is still assumed to hold. Hence, within this reduced model, the heat conduction in the bilayer is described by the one-dimensional heat equation in each layer, = km in the fluid, (13) (ρCp)m (ρCp)s ∂Tm ∂t ∂Ts ∂t ∂2Tm ∂y2 + S(y, t) ∂2Ts ∂y2 = ks in the substrate, (14) where Cp is the effective heat capacity and k is the ther- mal conductivity. The subscripts s and m correspond to substrate and metal, respectively. The source term can be written as S(y, t) = E0f (t) √2πσtp (cid:2)1 − r0(cid:0)1 − e−arh(cid:1)(cid:3) e−αa(y+h), where the first factor represents the incident energy from the laser source, second factor accounts for the reflectance of the metal film, and the last factor represents the energy absorbed by the film. In Eq. (15), E0 is the intensity of the incident radiation, σtp = tp(cid:16)2√2 ln 2(cid:17)−1 of the Gaussian laser pulse at half maximum, and f (t) gives the temporal profile of the laser fluence, is the width f (t) = exp(cid:2)−(t − tp)2/(2σ2 tp)(cid:3). (16) The spatial variation of the temperature in the metal film is expected to be small due to the small film thick- ness and high thermal conductivity of the metal. Hence, within this model, it can be assumed that the tempera- ture only weakly depends on the y-coordinate, and the temperature of the film-air interface can be approximated by the average film temperature, T ∗ 0 Tm dy. Inte- grating Eq. (13) with respect to y, from y = 0 to y = h, and using the boundary conditions (17) and (18), gives the equation for the average temperature of the film, T ∗ m hR h m = 1 ∂T ∗ m ∂t = S ∗ (h, t) − 1 h qs(t) (ρCp)m , (21) S ∗(h, t) = KSf (t) h (cid:2)1 − r0(cid:0)1 − e−arh(cid:1)(cid:3)(cid:2)1 − e−αah(cid:3) , (22) where and KS = E0/(cid:16)√2πσtp(ρCp)m(cid:17) . qs(t) = ∂Ts/∂yy=0 The heat equation for the substrate (14) can be solved using Green's functions or Laplace transform. Using the boundary conditions (18), (19) and (20), the average tem- perature of the film is found to be T ∗ m (h, t) = T0 + S ∗ e − 0 t2 p 2σ2 tp Z t exp − (t − u)2 2σ2 tp (t − u)(cid:19) eK 2u erfc(K√u) du, + tp σ2 tp (23) K (h) = p(ρCpk)s (ρCp)mh . Figure 2 shows the contour plot of the average temper- ature of the metal film, T ∗ m(h, t), as given by Eq. (23). The red highlighted curve represents the melting tem- perature of nickel (TM = 1728 K). We note that the energy absorption depends on the film thickness in a non-monotonic manner. For a small film thickness, h < hc ≈ 14.3 nm, only a part of the laser energy is absorbed, (15) where erfc(u) is the complementary error function and TABLE I: The values of the parameters used in simulations. The material parameters come from [33], the source term properties are as in [34], and the parameters related to disjoining pressure are the ones used in [14]. Description Notation Value/Expression 5 Density of the metal Density of the substrate Room temperature Melting temperature of the metal Viscosity of the metal at TM Surface tension Reference surface tension Change of σ with respect to temperature Conductivity of the metal Conductivity of the substrate Heat capacity of the metal Heat capacity of the substrate Laser fluence Time of maximum fluence Absorption length Fit parameter for reflectance Fit parameter for reflectance Equilibrium contact angle of metal with substrate ρm ρs T0 TM µm σ(T ) σ0 σT km ks 7900 kg/m3 2200 kg/m3 300 K 1728 K 4.61 × 10−3 Pa s σ0 + σT (T − TM ) 1.778 N/m −3.3 × 10−4 N/m K 90 W/m K 1.4 W/m K (Cp)m (Cp)s 0.44 × 103 J/kg K 0.712 × 103 J/kg K E0 tp αa r0 ar θeq 2500 J/m2 18 × 10−9 s 0.11688 × 10−9 m−1 0.459363 (8.0 × 10−9 m)−1 70◦ (2, 3) Exponents in in the disjoining pressure model (n, m) Precursor film thickness h∗ 1.5 × 10−9 m which leads to low film temperature. For film thicknesses h > hc the film absorbs most of the laser pulse energy. Hence, the film of thickness h ≈ hc reaches the high- est temperature, and for the film of thicknesses h > hc, the temperature decreases as h grows due to the larger amount of material that needs to be heated. Later in Sections III A and III B, we study the dynamics of the films with the film thickness either smaller or larger than hc. A known temperature at the interface which is ex- pressed as a function of the film thickness, h, and time, t only, is convenient for implementing the Marangoni force in the VoF solver. The surface gradients of the surface tension, ∇sσ, can be evaluated directly as ∇sσ = σT ∂h ∂x ∂T ∂h ds t, (24) where ds = q1 + (∂H/∂x)2 is the arc length, and H is the height function [22, 35]. As discussed earlier, T is m, and there- approximated by the average temperature T ∗ fore the gradient of the temperature with respect to the m/∂h that can be film thickness is approximated by ∂T ∗ computed analytically from Eq. (23) as follows ∂T ∗ m ∂h (h, t) = ∂S ∗ ∂h − e t2 p 2σ2 tp Z t 0 exp − (t − u)2 2σ2 tp + tp σ2 tp t2 p 2σ2 − 2S ∗ e dK dh tp Z t = KS(cid:26)(cid:2)−r0are−arh(cid:1)](cid:2)1 − e−αah(cid:3) 0 ∂S ∗ ∂h (t − u)! eK 2u erfc(K√u) du+ (t − u)2 2σ2 tp tp σ2 tp + exp − (t − u)!(cid:20)KeK 2u erfc(K√u) −r u π(cid:21) du (25) 1 h + (cid:2)1 − r0(cid:0)1 − e−arh(cid:1)(cid:3)(cid:20)(cid:2)αae−αah(cid:3) 1 h −(cid:2)1 − e−αah(cid:3) 1 h2(cid:21)(cid:27) . Note that ∂h/∂x in Eq. (24), is equivalent to ∂H/∂x, i.e., the derivative of the height function. For small film thicknesses, T ∗ m/∂h need to be carefully com- puted to ensure that the integrals in Eqs. (23) and (25) converge, see Appendix B. When used in our simulations, m and ∂T ∗ m and ∂T ∗ both T ∗ m/∂h are evaluated for an array of t and h values before the start of the simulations. During the simulation, we use bilinear interpolation to find the tem- perature at each interfacial cell and each time step. This makes the computations significantly faster, since we do not need to use the numerical integration to compute the integrals in Eqs. (23) and (25) for each interfacial cell at each time step. 2. Analytical Solution of the Heat Equation in a Film -- Substrate System The Eqs. (13) and (14), including the spatial variations in the y-direction in both the metal and the substrate, can be solved analytically using separation of variables, following the technique given in [36]. The analytical so- lution presented here is used for the verification of the reduced model given in Eq. (23) and the complete model is presented in Section II A 3. In contrast to the reduced model presented in Sec- tion II A 1, for computing the analytical solution, we as- sume that the substrate is of finite depth. However, when comparing the solutions resulting from different models, we use substrate depth large enough so that the tempera- ture solution is converged with increasing substrate thick- ness, and the solution is equivalent to that for the semi- infinite substrate (see Fig. 24 in Appendix C). Therefore, instead of Eq. (20), here we use the following boundary condition Ts = T0 at the bottom of the substrate y = −b, (26) where b is the substrate thickness. The solution can be found using separation of variables [36], and it can be written compactly in terms of Green's functions as S (ξ, τ ) dξdτ (27) S (ξ, τ ) dξdτ (28) Tm (y, t) =Z t 0 Z b 0 Z b Ts (y, t) =Z t a a G12 (y, t; ξ, τ ) Dm km G22 (y, t; ξ, τ ) Ds ks where G12 and G22 are given in Appendix C along with the details of the solution. Figure 3 shows the analytical temperature solution in the metal film (y > 0) and the substrate (y < 0). The temperature variation across the film thickness is small compared to the variation in the substrate. Therefore, ignoring temperature gradients across the film, as used in the reduced model, is justified. We note, however, that such a conclusion can be reached only for stationary flat films. As we will see later, using the reduced model for nonuniform films, or for the time dependent films, in general cannot be justified. On a different note, we point out that since there is no in-plane dependence in the source term, the 1D analytical solution presented here holds for 2D or 3D flat stationary films. 3. Computational Model for the Temperature of the Film -- Substrate System Next we consider the outlined problem via direct nu- merical simulations. We implement our numerical meth- 6 ods into the open source Gerris software [37]. We de- note the top subdomain containing metal and air by Ωf , and the bottom one containing the substrate by Ωs. In general, the temperature in Ωf , denoted Tf , satisfies the advection-diffusion equation, and the temperature in Ωs, Ts, satisfies the diffusion equation, ρCp [∂tTf +( u · ∇ )Tf ] = ∇ · (k∇Tf ) + Sn(x, t) (ρCp)s ∂tTs = ks∇2Ts in Ωs in Ωf (29) (30) where ρ = ρ (χ), Cp = Cp (χ) and k = k (χ) are the phase dependent density, heat capacity and the conduc- tivity of the metal and air, defined as the volume fraction weighted average of the metal and air properties. The simulation setup has to address the following is- sue: using Gerris we cannot solve for the temperature inside of the solid substrate directly, since except on the boundaries, the implementation of the solid entities does not contain the computational cells. Hence, in order to solve for the temperature in the fluid and the substrate using Gerris, we treat the solid domain -- the substrate -- as an immobile fluid. Furthermore, in order to impose the no-slip boundary condition on the metal-substrate boundary, we separate the two phases by a thin solid plate, see Fig. 4. Later in this section we show by direct comparison with the analytical solution that this setup, referred to as the complete model, produces correct re- sults. The coupling of the temperature solution between the liquid and solid domains is accomplished using Newton's law of cooling, which we impose on the top and bottom of the solid plate as follows km ks ∂Tf ∂y ∂Ts ∂y = α (Tf − Ts) = α (Tf − Ts) at the top of the solid, (31) at the bottom of the solid, (32) where α is the heat transfer coefficient. Note that the right hand sides of Eqs. (31) and (32) are equal, imply- ing the continuity of the flux between the liquid and sub- strate. Furthermore, in the limit α → ∞, the boundary conditions (31) and (32) both imply Tf = Ts. Hence, the boundary conditions specified by Eqs. (31) and (32) effec- tively encapsulate both the continuity of flux, Eq. (18), and the continuity of temperature, Eq. (19). Addition- ally, in Appendix C, see Fig. 25, we confirm that the analytical solution, given in Section II A 2, with the New- ton's cooling law boundary condition, converges for large α to the solution specified by requiring continuity of tem- perature. We note that the presented computational approach can be used for arbitrary metal-air interface shape. In the remainder of this section, we verify that the nu- merical solutions to Eqs. (29) and (30), along with the boundary conditions, Eqs. (31) and (32), converge with 7 4000 3000 2000 1000 4000 3000 2000 1000 0 -400 -300 -200 -100 0 0 -400 -300 -200 -100 0 (a) h0 = 10 nm (b) h0 = 20 nm FIG. 3: The analytical solution for the temperature of the film (y > 0) and the substrate (y < 0). The lines show different times; the time specified in the legends is in ns. Note very small variation of the temperature across the film. Figure 5 shows the temperature solution of the com- plete model for a flat film geometry and vanishing ve- locity in Eq. (29) compared to the analytical solution for increasing values of α. Clearly, as α increases, the results of the complete model approach the analytical solution. The largest relative difference in the tempera- ture between α = 50 W m−2K−1 and α = 70 W m−2K−1 is 0.7% for a h0 = 10 nm, and 0.6% for a h0 = 20 nm. Since larger α decreases the required time-step, we use α = 50 W m−2K−1 in order to reduce the computational time. Figure 6 shows the convergence of the solution of the complete model for the averaged temperature for in- creasing substrate size, b. The largest relative difference in the temperature between b = 400 nm and b = 600 nm is 0.12% for a h0 = 10 nm, and 0.08% for a h0 = 20 nm. We conclude that it is sufficient to use b = 400 nm. So far, we have shown that the solution to the com- plete model described in this section converges with the increased heat transfer coefficient, α, and the substrate thickness, b, to the analytical solution with continuity of temperature boundary condition at the fluid-substrate interface. Next, Fig. 7 shows the comparison of the av- erage temperature in the metal film obtained using the reduced model given in Section II A 1, the analytical so- lution outlined in Section II A 2, and the complete model described in this section. The average temperature of a flat metal film in both the reduced and the complete model agrees with the analytical solution. Hence, we are confident in the numerical implementation of the tem- perature models in our numerical solver. We point out this agreement between different models, in particular the reduced and complete model holds only for flat films for which the heat flow in the in-plane direc- tion is not relevant. For perturbed films, the two models produce different results, as we will discuss in the next section. FIG. 4: The fluid -- substrate setup used in the direct numerical simulations. the increasing heat transfer coefficient, α, and the sub- strate thickness, b. To do this, we consider the follow- ing test problem. Assume the metal-air interface is flat and the solution is x-independent. Hence, the temper- ature for the 2D problem satisfies the 1D heat equation at any fixed position x, and the 1D analytical solution, presented in Section II A 2, holds. We compare the tem- perature solution obtained from the complete model to the 1D analytical solution, where we average the tem- perature over the film thickness for the purpose of this comparison. 5000 4000 3000 2000 1000 5000 4000 3000 2000 1000 8 0 0 10 20 30 40 50 0 0 10 20 30 40 50 (a) h0 = 10 nm (b) h0 = 20 nm FIG. 5: The convergence of the average temperature in the metal film using the numerical solution of the complete model with increasing α. The units of α in the legend are in W m−2K−1. Note that the source term is a Gaussian centered at t = tp = 18 ns, and of the width σtp, see Eq. (16). 5000 4000 3000 2000 1000 5000 4000 3000 2000 1000 0 0 10 20 30 40 50 0 0 10 20 30 40 50 (a) h0 = 10 nm (b) h0 = 20 nm FIG. 6: The convergence of the average temperature in the metal film using the solution of the complete model with increasing substrate size, b. III. RESULTS AND DISCUSSION We are now ready to discuss the influence of thermal effects on the film stability. First in Sections III A - III D we focus on film geometry, and then consider filaments in Section III E. We start by discussing briefly in Sec- tion III A the results of linear stability analysis carried out within the long wave limit in a simplified setting (that assumes film temperature to be dependent only on the current value of the film thickness). Then, we follow in Section III B by presenting the results using the models for temperature computation outlined in the preceding section. We will see that the outcomes of the models considered differ significantly. The main finding is that the reduced model overestimates the Marangoni effect, and the particular reason for this is the omission of the in-plane heat conduction. We will also see that the tem- poral temperature variations lead to a change in the sur- face tension, which can in turn affect the stability of the perturbed interface during the evolution. As we will see, the interplay of the stabilizing/destabilizing Marangoni effect and temporal variation of surface tension leads to a complex form of film evolution. In Section III C we consider also temperature variation of film viscosity, and its influence on the film stability. Section III E discusses the influence of thermal effects on stability of metal fila- ments. For both films and for filaments, we focus on develop- ing basic understanding of the influence of thermal effects on their stability. Therefore, we focus on relatively simple computational domains and initial conditions - in partic- ular, in simulations we will consider films and filaments perturbed by a single wavelength, and defer considering more complex domains and initial conditions to future work. As we will see, even for simple setups the influ- ence of variation of material parameters (surface tension and viscosity) is rather complex, and the simplicity of the considered computational domains and initial conditions 5000 4000 3000 2000 1000 5000 4000 3000 2000 1000 9 0 0 10 20 30 40 50 0 0 10 20 30 40 50 (a) h0 = 10 nm (b) h0 = 20 nm FIG. 7: The comparison of the average temperatures of a flat film for the reduced, the analytical, and the complete model. Here, the substrate thickness is b = 400 nm, and the heat transfer coefficient is α = 50 W m−2K−1. helps in focusing on the basic questions. In addition, we focus on the question of stability versus instability of a given perturbation; for this reason in our simulations we choose initial conditions that are close to the critical ones where stability changes: such a choice helps to fur- ther simplify the problem considered and reach answers to the basic stability questions. A. Linear Stability Analysis of a Thin Film in Two Dimensions First, to gain basic insight, we present the results of lin- ear stability analysis (LSA) carried out within the long wave limit. While the LSA provides only approximate results since (i) it is valid only for early stages of insta- bility, and (ii) corresponds to the long wave limit, we still expect it to provide a useful insight. The long wave limit [38], for a Newtonian film with Marangoni effect and the fluid-substrate interaction in the form of the disjoining pressure [39] leads to the fol- lowing 4th order nonlinear partial differential equation 3µ ∂h ∂t + ∇ ·(cid:20)σ0h3∇∇2h + 3 2 h2∇σ(T ) + Kπh2∇(cid:18) hn hn − ∗ (33) ∗ hm hm(cid:19)(cid:21) = 0, where Kπ is given by Eq. (10). We assume the film thick- ness is perturbed around the equilibrium one, h0, as h(x, t) = h0(cid:0)1 + εeβt+ikx(cid:1) , (34) where ε is a small parameter, β is the growth rate of the perturbation and k is the wavenumber. Note that tem- perature is not an independent variable here, instead it is a function of h. An alternative approach is to consider both h and T as independent variables, and perturb each of them separately, however, we are not doing this here for simplicity. Keeping only the leading order terms in ε we obtain the dispersion relation: β = − h2 0 k2 3µ (cid:20)σ0h0k2 − 3 2 σT ∂T ∂h − Kπ(cid:18)n ∗ hn 0 − m hn ∗ hm hm 0 (cid:19)(cid:21) . (35) To illustrate the expected influence of Marangoni effect on stability, in Fig. 8 we plot the temperature gradient, m/∂h, for a fixed film thickness computed using the ∂T ∗ reduced model (see Eq. (25)). The temperature of the film, T ∗ m (see Eq. (23)), is plotted to show the film melting time. We assume that before the temperature of the film rises above T ∗ m, the film does not evolve. The value of the gradient changes as a function of time, and in order to obtain an estimate for the stability of a perturbed film using Eq. (35), we approximate ∂T /∂h by the largest m/∂h, during the time the film is absolute value of ∂T ∗ melted. Hence, the obtained dispersion curve provides the upper bound on the influence of the Marangoni effect. Figure 9 shows the dispersion curve for 10 and 20 nm thick films. For the h0 = 10 nm thick film, Fig. 9(a), the Marangoni effect is stabilizing, since ∂T /∂h > 0, for all the times at which the film is melted. Conversely, for h0 = 20 nm film, Fig. 9(b), the Marangoni effect is destabilizing for a significant period after melting, since ∂T /∂h < 0. In the rest of this section, we will use the outlined LSA results to rationalize the results computed using the re- duced and complete models for the flow of thermal en- ergy. B. Evolution of a Thin Film Interface in Two Dimensions In this section, we examine the stability of the films by solving the Navier-Stokes equations including the ther- mal effects. We also include the fluid-substrate interac- 5000 4000 3000 2000 TM 1000 0 0 80 60 40 20 0 10 20 30 40 -20 50 5000 4000 3000 2000 TM 1000 0 0 10 10 0 -10 -20 -30 10 20 30 40 -40 50 (a) h0 = 10 nm (b) h0 = 20 nm FIG. 8: The average temperature, T ∗ m (thick solid blue line) , and ∂T ∗ m/∂h (thick dashed orange line) of a metal film as a function of time, using the reduced model. The thin horizontal blue and orange dashed lines indicate the meting temperature, TM , and the line ∂T ∗ m/∂h = 0, respectively. ] 1 - s n [ 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 0 0.02 0.04 0.06 k [ nm -1 ] ] 1 - s n [ 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 0 0.01 0.02 k [ nm -1 ] 0.03 (a) h0 = 10 nm, ∂T /∂h = 76.7 K nm−1 (b) h0 = 20 nm, ∂T /∂h = −33.9 K nm−1 FIG. 9: The growth rate for a perturbed film resulting from the LSA, where ∂T /∂h is approximated by the specified maximum values, as explained in the text, with (blue solid) and without (orange dashed) Marangoni effect. tion in the form of a disjoining pressure (see Eq. (12)). We compare the influence of thermal effects on the film breakup using the temperature solution from the reduced and from the complete model, described in Sections II A 1 and II A 3, respectively. The initial geometry of the film in the simulations is as described by Eq. (34) where ε = 0.1. At the time t = 0, the metal is in solid state (the pulse maximum occurs at tp = 28 ns). For the simulations that use the reduced temperature model, we simulate only the times after the film is melted. In the case of simulations such that the complete temperature model is implemented, we keep the film stationary until melted (by putting the fluid velocity to zero). 1. Film of Thickness 10 nm; dT /dh > 0 Figure 10 shows the evolution of the interface for a 10 nm film. We use stable perturbation wavelength, λ = 100 nm, that is slightly smaller than the critical one, λc = 114 nm found from the LSA in Section III A. Hence, we expect the perturbation to be stable. Figure 10(a) shows the evolution of the interface with temper- ature solution from the reduced model. The perturba- tion of the interface decays, as expected. Figure 10(b) shows the evolution with the temperature solution from the complete model, where initially (see t = 20 ns) the perturbation decays, then grows for all following times, and the film eventually breaks into drops. Hence, the two temperature models, which agree for a flat film, produce different evolution for a perturbed interface. Understanding the difference in stability resulting from the models requires considering both normal and tan- gential stress balances at the liquid-air interface, that is, ] m n [ t i h g e H 11.0 10.5 10.0 9.5 9.0 20 15 10 5 ] m n [ t i h g e H Time [ns] 0 11.4 20.4 30.4 40.4 49.9 11 Time [ns] 0 10 20 30 40 50 0 25 50 x [nm] 75 100 0 25 50 x [nm] 75 100 (a) Reduced model (b) Complete model FIG. 10: The comparison of the evolution of the interface for the two models considered, for h0 = 10 nm and λ = 100 nm. The film is stable in (a) but goes through oscillatory instability in (b) (for early times, until t = 20 ns, the imposed perturbation decays slightly in (b)). Note also that for this and the following figures, the evolution starts at t ≈ 11.4 ns, the time at which the film temperature rises above the melting temperature, TM . Note different thickness scales in (a) and (b). both spatial gradients of the surface tension leading to Marangoni effect, and the temporal evolution of surface tension due to the evolution of temperature. We discuss both of these effects next, for both reduced and complete models. First, let us consider the influence of the temperature on the normal component of the surface force, and ig- nore the tangential component (therefore ignoring the Marangoni effect). Figure 11 shows the evolution of the interface using the complete model, for (a) constant sur- face tension, σ = σ0, and (b) temperature dependent, σ = σ(T ), but the Marangoni effect is not included. In the simulation that use constant surface tension the perturbation is stable, as expected from the dispersion relation, Eq. (35). However, when the surface tension dependence on the temperature is included, the pertur- bation initially decays (see t = 20 ns), but grows for all later times. Note in particular that the results shown in Figs. 10(b) and 11(b) are almost identical, showing that the Marangoni effect is essentially irrelevant in the present context. Therefore, the stability change is not due to the spatial variations of σ, but due to the change of the normal component of the surface tension force due to temporal change of σ. We still need to explain why they the results shown in the parts (a) and (b) of Fig. 10 differ. For this purpose, Fig. 12(a) shows the average temperature of the film in the reduced and complete models for the parameters as used in Fig. 10. According to both models, from the melting time at t = 11.4 ns, the temperature of the film increases to T & 4000 K, which corresponds to the de- crease in the surface tension from σ(TM ) = 17.78 N m−1 to σ(Tmax) = 10.15 N m−1. Figure 12(b) shows the dis- persion curve computed using Eq. (35), with σ0 = σ(TM ) and σ0 = σ(Tmax), and ∂T /∂h = 0. The change in σ0 shifts the critical wavenumber, kc, and the stable perturbation (λ = 100 nm which corresponds to k ≈ 0.0628 nm−1) becomes unstable. This explains why the stable mode in simulations in Figs. 10(b) and 11(b) be- comes unstable as the film temperature increases, even- tually leading to the film breakup. Note that we have already shown that Marangoni effect is not relevant for the simulations that use the complete model for temper- ature calculation. In contrast to the results obtained by implementing the complete temperature model shown in Figs. 10(b) and 11(b), the results obtained with the temperature from the reduced model, Fig. 10(a), show stability, despite the fact that the average temperature of the film increases simi- larly as in the complete model, see Fig. 12(a). To gain better understanding of this finding, we examine the dif- ference in the temperature solutions along the liquid-air interface for a perturbed stationary film, corresponding to the initial condition used in Figs. 10 and 11. (The motivation for considering a stationary film is the source term dependence on the film thickness - the film evolu- tion would affect the source term, and we prefer to avoid this effect for simplicity of the argument.) Figure 13(a) shows the differences of the computed temperature along the interface from the average temperature. The tem- perature varies significantly more for the reduced model compared to the complete one. Thus, the temperature gradients at the liquid-air interface are significantly larger for the reduced model, and the stabilizing Marangoni ef- fect prevents the interface in Fig. 10(a) from becoming unstable. This finding explains the different film evolu- tion between the two models. To summarize the results for the film thickness of h0 = 10 nm: If the temperature is computed using the complete model, the Marangoni effect turns out to be ] m n [ t i h g e H 11.0 10.5 10.0 9.5 9.0 20 15 10 ] m n [ t i h g e H 5 Time [ns] 0 11.4 20.4 30.4 40.4 12 Time [ns] 0 10 20 30 40 50 0 25 50 x [nm] (a) σ = σ(TM ) 75 100 0 25 50 x [nm] 75 100 (b) σ = σ(T ) FIG. 11: The evolution of the interface with the temperature solution from the complete model for h0 = 10 nm and λ = 100 nm, and for the fixed (a) and temperature dependent (b) surface tension as noted. Marangoni effect is not considered. Note different thickness scales in (a) and (b). 4000 3000 2000 1000 ] K [ e r u t a r e p m e T Complete model Reduced model 0 10 20 30 40 50 time [ns] (a) ] 1 - s n [ 0.6 0.4 0.2 0 -0.2 -0.4 0 0.06 0.08 0.02 0.04 k [ nm -1 ] (b) FIG. 12: (a) The average temperature of the metal film from Fig. 10. (b) Growth rate given by Eq. (35), for h0 = 10 nm and for σ at the melting temperature, TM , and at the maximum temperature predicted by the reduced model, Tmax. The vertical dashed line in part (b) shows the value of k used in the simulations. Marangoni effect is not considered. irrelevant; however the temporal change of surface ten- sion due to evolving laser pulse and film thickness may influence film stability, leading to instability in the case considered here. If the temperature is computed using the reduced model, then (stabilizing) Marangoni effect may compete with the destabilizing effect of the over- all surface tension decrease, leading to stability. There- fore, computing temperature carefully is crucial for un- derstanding the film stability. 2. Film of Thickness 20 nm; dT /dh < 0 Next, we consider thicker film, with the average thick- ness of h0 = 20 nm. Recall that for h0 = 20 nm, the reduced model predicts essentially the opposite direc- tion of the Marangoni effect relative to h0 = 10 nm, see Fig. 8. Here we impose a perturbation of the wave- length, λ = 250 nm, which is stable when Marangoni effect is ignored, see the dispersion relation, Eq. (35). Figure 14 shows the comparison of the evolution of the interface using the reduced and complete temperature models. Using the reduced model, Fig. 14(a), we find in- stability, and the perturbation grows until the film breaks into drops. This is not surprising since the reduced tem- perature model predicts ∂T ∗ m/∂h < 0 which destabilizes the film (see Fig. 9(b)). When the temperature is com- puted using the complete model however, see Fig. 14(b), the evolution is oscillatory: at t = 20 ns the perturba- tion decays; at t = 30 ns and t = 40 ns the perturbation grows; and at t = 50 ns the perturbation decays again. Similarly as for h0 = 10 nm, to explain these dynamics, 50 0 ] K [ T Δ −50 40 0 ] K [ T Δ −40 13 Time [ns] 0 10 20 30 40 0 25 50 x [nm] 75 100 0 50 100 150 200 250 x [nm] (a) h0 = 10 nm, λ = 100 nm (b) h0 = 20 nm, λ = 250 nm FIG. 13: The difference, ∆T , between the computed and average temperatures at the liquid metal-air interface for a static perturbed film showing the results for the reduced (dashed) and the complete (solid) models. Note that for the temperature scale shown, the temperature gradients for the complete model are almost invisible. 40 30 20 10 Time [ns] 0 11.8 20.8 30.8 40.8 48.8 23 22 21 20 19 18 17 Time [ns] 0 10 20 30 40 50 ] m n [ t i h g e H ] m n [ t i h g e H 0 50 100 x [nm] 150 200 250 0 50 100 150 200 250 x [nm] (a) Reduced model (b) Complete model FIG. 14: The comparison of the evolution of the interface for the two models considered, for h0 = 20 nm and λ = 250 nm. Here, the film is unstable in (a), but goes through oscillatory instability in (b). Note that the evolution starts at t ≈ 11.8 ns, the time at which the film temperature rises above the melting temperature, TM . Note also different thickness scales in (a) and (b). we examine the simulations without Marangoni effect, and investigate the influence of the temperature on the normal component of the surface force. Figure 15 shows the evolution of the interface when the surface tension is (a) constant, σ = σ0, and (b) tem- perature dependent, σ = σ(T ), but Marangoni effect is ignored. When constant surface tension is considered, the perturbation is stable, as expected from the LSA. However, for temperature dependent surface tension, we uncover the same dynamics as in Fig. 14(b). Thus, we see again that the Marangoni force is negligible. We show next that the changes in the stability in the complete model are due to the temporal changes of the surface tension. Figure 16(a) shows the average temperature of the film using the reduced and complete models for the param- eters as used in Fig. 14. From the melting time, at t = 11.8 ns, the temperature evolution (and the sur- face tension change) are similar as for the h0 = 10 nm films. Figure 16(b) shows the dispersion curve computed using Eq. (35), with σ0 = σ(TM ) and σ0 = σ(Tmax), and ∂T /∂h = 0. The change in σ0 shifts the critical wavenumber, kc, and the (linearly) stable perturbation (λ = 250 nm which corresponds to k ≈ 0.02512 nm−1) becomes unstable, as we see in Figures 14(b) and 15(b) after t = 20 ns. After t = 40 ns, the temperature de- creases again to T ≈ 3000 K, hence (see Fig. 16(b)), the perturbation becomes (linarly) stable again. In sum- mary, similarly to the h0 = 10 nm film, the stability of the interface using the complete model is governed by the temporal variations of σ. Similarly as for the h0 = 10 nm film, the temporal ] m n [ t i h g e H 22 21 20 19 18 22 20 18 ] m n [ t i h g e H Time [ns] 0 11.4 20.4 30.4 40.4 49.9 14 Time [ns] 0 10 20 30 40 50 0 50 100 x [nm] 150 200 250 0 50 100 150 200 250 x [nm] (a) σ = σ(TM ) (b) σ = σ(T ) FIG. 15: The evolution of the interface with the temperature solution from the complete model for h0 = 20 nm and λ = 250 nm, and for the two values of surface tension as noted. Marangoni effect is not considered. Note different thickness scales between (a) and (b). 4000 ] K [ e r u t a r e p m e T 3000 2000 1000 Complete model Reduced model 30 40 50 0 10 20 time [ns] (a) ] 1 - s n [ 0.1 0.05 0 -0.05 -0.1 -0.15 0 0.01 0.02 k [ nm -1 ] (b) 0.03 FIG. 16: (a) The average temperature of the metal film from Fig. 14. (b) Growth rate given by Eq. (35), for h0 = 20 nm, ∂T /∂h = 0, and σ at the melting temperature, TM , and the maximum temperature predicted by the reduced model, Tmax. The vertical dashed line in part (b) shows the wave number used in the simulations. changes of the surface tension do not explain the film in- stability for the temperature from the reduced model in Fig. 14(a). Therefore, we compare again the temperature at the liquid-air interface of a stationary film using the reduced and complete models. Figure 13(b) shows the deviation from the average temperature at the interface of a stationary film corresponding to the initial condition for the simulations shown in Fig. 14. We see once again that the effect of the Marangoni effect is augmented sig- nificantly by the reduced model. To summarize, we find that for both thin and thick films (relative to the critical thickness at which dT /dh changes sign), the complete and reduced model produce different results, showing clearly that careful computa- tion of heat flow is required to accurately describe the evolution. Using the reduced model, or in other words ignoring the heat conduction in the in-plane direction, leads to qualitatively different results compared to the ones obtained if this assumption is not made. C. The Influence of the Temperature Dependent Viscosity Here we focus the influence of the viscosity variations with temperature on the stability and breakup dynam- ics. During the metal heating and melting, the viscosity of the metal changes several orders of magnitude [33]. The viscosity of most metals can be modeled by an ex- ponential as RT(cid:19) µ (T ) = µ0 exp(cid:18) E (36) where µ0 = 0.1663 mN s m−2 and E = 50.2 kJ mol−1 are constants dependent on the material, and R = 8.3144 J K−1mol−1 is the gas constant [33]. Figure 17 shows the viscosity of nickel as a function of tempera- ture. The influence of the temperature dependent viscosity on the film breakup can be estimated based on the dis- persion relation specified by Eq. (35). This relations says that the stability of the film, and in particular the critical wave number, kc, do not depend on the viscosity. How- ever, the growth rate, β, is inversely proportional to µ and, as we will see, this may be sufficient to influence the stability. To study the influence of the variable viscosity on the breakup dynamics, we use the same initial geometry as in Section III B within the framework of the complete model described in Section II A 3. Here, surface tension is taken as temperature dependent, but for simplicity we do not include Marangoni effect (which is essentially ir- relevant for the complete model). Figure 18 shows the evolution of the film interface with temperature depen- dent viscosity compared to the evolution for constant vis- cosity, µ = µ(TM ). For the 10 nm film, the same evolu- tion dynamics are present in both cases: perturbations initially decay, but start growing as the film temperature rises (see the discussion related to Figs. 10(b) and 11(b)). As expected based on the LSA, the stability of the per- turbations is not affected by the variable viscosity, but the growth rate is faster, and therefore the breakup time occurs ≈ 5 ns faster with variable viscosity compared to the constant one (see Section III B). For the 20 nm film, again, the decay and the growth of the perturbations for the variable viscosity follows the same direction as the constant viscosity in Figures 14(b) and 15(b). During the time of the perturbation growth, as in t = 20 ns to t = 40 ns for µ = µ(TM ), the perturbation for µ = µ(T ) grows fast enough so that the film breaks. Recall that for µ = µ(TM ), the stability changes after t = 40 ns, and the film stabilizes due to the decrease of the film tem- perature. Therefore, inclusion of temperature dependent viscosity can strongly influence the film evolution. D. Discussion of the Influence of Thermal Effects on Film Stability At the beginning of Section III, we motivated the focus of this work on simple computational domains and initial conditions. In principle, more complex setups would be needed to reach precise understanding regarding the in- fluence of thermal effects in physical experiments where the relevant domains are large and the film and tem- perature perturbations are more complicated. However, based on the existing results, we are already in the posi- tion to develop a basic insight. To be specific, let us ask for the following question: What is the influence of the fact that the temperature of metal films rises significantly above melting temperature 15 on the film stability? Focusing first on the temperature dependence of surface tension, we start by noting that Marangoni effect is not relevant. However, the overall decrease of surface tension as the temperature of a film raises suggests that shorter wavelengths are expected, see the dispersion curves in Figs. 12(b) and 16(b). Possibly, such an increase of surface tension may be responsible for observation of shorter wavelengths in experiments [19]. Clearly, as the film temperature changes as a function of time due to time-dependent source term (laser pulse) and heat loss through the substrate, the effect of the surface tension change will be time dependent as well. Turn- ing now to temperature dependence of film viscosity, we note that viscosity influences the time scale of instabil- ity growth and could therefore influence the film stability strongly, as discussed in Section III C. To conclude this brief discussion, a variation of mate- rial parameters with temperature clearly influences film stability in a manner which may be complex in particular due to the fact that the relevant time scales, related to the source term and to the temporal evolution of the film itself, are comparable. In general, based on the results presented so far, one expects that heating of the films above melting will result in a decrease of the emerging wavelengths (such as the distance between drops that form eventually), compared to the ones expected if one assumes that the material parameters (surface tension, viscosity) are given by their values at the melting tem- perature. The details of the instability evolution how- ever may depend on the particular choice of metals, sub- strates, and laser pulse energy and duration. E. Breakup of Liquid Metal Filaments In a recent work [34], that included both experimen- tal and computational study, we considered concentra- tion Marangoni effect in a two-metal setup focusing on metal filament geometry. In that study, it was found that concentration Marangoni effect played a significant role. In particular, the Marangoni induced flow led to inver- sion of instability development, in the sense that initially thicker filament parts (nickel covered by a thin copper film) ended up thinning, while initially thinner filament parts increased in thickness due to concentration induced Marangoni flow. The question that we will consider in the present paper, is whether a similar effect could be observed for thermal Marangoni effect. To answer this question, we consider the following setup: the initial geometry is a flat filament with super- imposed rectangular perturbations (similarly as in [34] just with a single metal (nickel). Since we know from Section III B that the temperature gradients are small in the metal film, we increase the thickness of the rectan- gular perturbations compared to [34] in an attempt to increase the temperature gradients. Therefore we take the base filament thickness as h0 = 8 nm, and superim- pose the perturbations of ∆h = 8 nm (so that the film 10 7 10 ] 2 m / s N m [ 9 8 7 6 5 4 3 2 1 0 16 5.5 5 4.5 4 ] 2 m / s N m [ 3.5 3 2.5 2 1.5 1 0 1000 2000 3000 4000 0.5 1500 2000 Temperature [K] (a) 3500 4000 2500 3000 Temperature [K] (b) FIG. 17: Viscosity of nickel as a function of temperature given by Eq. (36), for temperature range starting from (a) room temperature and (b) melting temperature of nickel. 20 15 ] m n 50 40 ] m n 30 [ t i h g e H 10 5 [ t i h g e H Time [ns] 0 10 20 30 40 0 25 50 x [nm] 75 100 20 10 0 Time [ns] 0 10 20 30 35 0 50 100 x [nm] 150 200 250 (a) h0 = 10 nm (b) h0 = 20 nm FIG. 18: The comparison of the evolution of the film interface with constant (full line) and temperature dependent viscosity (dashed line) using the complete temperature model. thickness vary between 8 and 16 nm), and the width of the filament is w = 185 nm. The average filament thick- ness is kept at 12 nm as in [34]. We note that the filament simulations are carried out using the approach described in [34]; briefly, we do not consider disjoining pressure here but instead specify the contact angle (90◦ for sim- plicity), and use Navier-slip boundary condition with the slip length of 20 nm. The simulations consider one half of the perturbation wavelength and half of the filament width, and impose symmetry boundary conditions at all in-plane directions. The fluid is stationary until the melt- ing time of the filament (see Section III B). For early times, the initial geometry quickly evolves to a cylindrical filament on a substrate. The basic idea regarding the stability of such a filament could be reached by considering the Rayleigh-Plateau type of instability of a free standing cylindrical jet. Within this model, the growth rate of the perturbations, β, is given by [40, 41] β2 = σ ρR3 (cid:20)kR(cid:0)1 − k2R2(cid:1) I1(kR) I0(kR)(cid:21) (37) where R is the radius of the jet, and I0 and I1 are the modified Bessel functions. Hence the stability of a jet depends on its radius, R: the modes, k, for which kR < 1 are unstable and the modes for which kR > 1 are stable. Here, k is the wavenumber related to the perturbation wavelength, λ, by k = 2π/λ. The fastest growing mode In the present context, R corresponds to kmR ≈ 0.7. corresponds to the radius of a filament characterized by the equilibrium contact angle θ, and of the same cross- sectional area as the initial rectangular geometry of the thickness h and width w, i.e., R =r hw θ − cos θ sin θ . (38) 1 0.8 0.6 0.4 0.2 0 0 200 400 600 800 FIG. 19: Growth rate as a function of the wavelength for filament width w = 185 nm. Since the perturbation is large, the growth rate is computed both with h0 = 8 nm and h0 = 12 nm (average thickness). Figure 19 shows the growth rate for a filament. We con- sider both the filament thickness without the perturba- tion, h0 = 8 nm and the average filament thickness in- cluding the perturbations, h0 = 12 nm. The Rayliegh- Plateau stability curve gives us an approximation for the critical wavelength, λc ∼ 236 nm using h0 = 12 nm; in- cluding the presence of substrate is known to make λc slightly larger [42]. Our numerical results show consis- tently that λc value is in the range [240, 250] nm. In what follows, we show the results for two filaments: one with a stable and one with an unstable perturbation. We compare the results for simulations with and without the thermal effects. The temperature is governed by the complete model described in Section II A 3. Figure 20 shows the evolution of a linearly stable fila- ment, with the wavelength of the perturbations close to the critical one. The results are similar, independently of whether the surface tension is treated as a constant or temperature dependent. To understand this result, we note that the filament setup differs in a significant man- ner from the the film one: here, a change of the surface tension does not change the stability of the filament; it only modifies the growth rate. Hence, the thermal varia- tion of the surface tension does not change the qualitative behavior. Figure 21 shows the evolution of an unstable filament. Again, the thermocapillary force does not change the qualitative breakup dynamics. However, the breakup with temperature dependent surface tension happens about 10 ns slower compared to the constant surface ten- sion. This is expected, since the increase in the temper- ature leads to a decrease in the surface tension, which in turn leads to a decrease of the growth rate. An obvious question to ask is why concentration Marangoni effect, discussed in [34] is so much more prominent compared to the thermal Maranogni effect 17 discussed here. There are at least two sources of the difference: first, the thermal Marangoni effect is much weaker compared to the concentration one (one needs the temperature difference of ≈ 1, 600 degrees to produce the change of surface tension of nickel corresponding to the difference of the surface tensions of nickel and copper considered in [34], see Tab. I): and second, for the setup considered in the present work, the thermal Marangoni effect is induced: the filament needs to heat up for the thermal Marangoni effect to be established; in the setup considered in [34], the concentration Marangoni effect is present from the very beginning of the evolution. We note in passing that the present results also suggest that thermal Marangoni effect can be safely ignored in two (or multi) metal setting: concentration Marangoni effect is expected to play a dominant role. F. Conclusions In this paper we have studied the influence of the ther- mal effects on the evolution of thin metal films and fila- ments. For the films, we have shown that the dynamics of the evolution can change due to the surface tension de- pendence on temperature. Perhaps surprisingly, the in- fluence of the temperature is not manifested through the Marangoni effect, but through the capillary force (bal- ance of normal stresses). In other words, the thermal effects influence the interface evolution due to the time- dependent changes of the surface tension during a laser pulse. We have the reached the main conclusion outlined in the preceding paragraph by considering two models for computing the film temperature. The reduced 1D tem- perature model (Section II A 1) is found to overestimate the temperature gradients along the free interface, since the in-plane heat conduction is not considered. The com- plete model, based on the numerical computation of the temperature (Section II A 3) shows that the temperature gradients along the interface are in fact not strong enough to influence the breakup of the films. The changes in the viscosity during the metal film heating can accelerate the growth of the perturbations, leading to a breakup of films that would not break if a constant value of viscosity at the melting temperature were used. This finding is also to a certain degree unexpected since it is known (at least within the long wave limit) that viscosity influences only the growth rate, and does not modify the range of un- stable wave numbers. However, an interplay of the time scales responsible for heating and for instability devel- opment modifies the evolution in a nontrivial manner. Our results suggest that the fact that temperature rises significantly over melting will result in the emergence of shorter length-scales compared to the ones expected if the material parameters considered were assumed to be fixed at their values corresponding to the melting tem- perature. In the case of filaments, the temperature dependence 18 t = 0.0 ns t = 16.0 ns t = 49.0 ns t = 0.0 ns t = 16.0 ns (a) t = 49.0 ns (b) FIG. 20: Evolution of a stable filament with wavelength λ = 240 nm, (a) surface tension dependent on the temperature and (b) surface tension fixed at σ0. The color in part (a) represents the temperature at the interface in degrees Kelvin. of surface tension has only rather minor influence on the qualitative behavior of the breakup of the liquid metal fil- aments. At least for the parameters considered here, the stability of a (single metal) filament is not influenced by surface tension variation. This is in contrast to the two- metal filaments, considered in [34], where concentration dependence of surface tension can qualitatively change the dynamics. The presented results open new avenues of research. Some of the questions that one may ask are as follows: 19 t = 0.0 ns t = 16.0 ns t = 49.0 ns t = 0.0 ns t = 16.0 ns (a) t = 39.0 ns (b) FIG. 21: Evolution of an unstable filament with wavelength λ = 250 nm, (a) surface tension dependent on the temperature and (b) surface tension fixed at σ0. The color in part (a) represents the temperature at the interface in degrees Kelvin. Note slower breakup in (a) due to a decrease of surface tension with temperature. • What is the influence of the substrate thickness and thermal properties on the findings reported here? Presumably for sufficiently thin substrate, heat dif- fusion in the in-plane direction may be less impor- tant, modifying the results, and perhaps bringing the findings of the complete model (that includes the heat diffusion in the in-plane direction) and the reduced model (that does not include the in-plane heat diffusion) closer together. • The result presented here show in some cases oscil- latory instability, with the film evolution evolving in a non-monotonous manner. Can one understand the conditions required for such oscillatory evolu- tion more precisely? • The results of the present paper, combined with the analysis of the propagation of the melting front con- sidered in a stationary setup [23], should allow to analyze evolution of metal films on the substrates that may go through the phase transitions them- selves. What is the influence of substrate melting on the film stability? • How significant are thermal effects for the evolution and stability of multimetal films and alloys? We expect that the results presented here will serve as a basis for answering some of the outlined questions. ACKNOWLEDGMENTS We acknowledge many useful discussions regarding metal films with Ryan Allaire, William Batson III, Linda 20 Cummings, Javier Diez, Francesc Font, Jason Fowlkes, Alejandro Gonz`alez, Kyle Mahady and Philip Rack. This work was supported by the NSF grant No. CBET- 1604351. Appendix A: Laser Source Term The absorption, reflectance and transmittance of a thin metal film can be computed from Maxwell's equations with appropriate boundary conditions. The equations are greatly simplified when considering a single film layer on a transparent (non-absorbing) substrate. The simpli- fied expressions for computing reflectance and transmit- tance given in Heavens [32] are R1 = T1 = , t2 12 + u2 12 p2 12 + q2 12 n2 n0 e2γ1 + (g2 1 + h2 2)e−2γ1 + C cos(2γ2) + D sin(2γ2) ((1 + g1)2 + h2 1)(g2 2 + h2 1)((1 + g2)2 + h2 2) (A1) (A2) where the terms in Eqs. (A1) and (A2) are defined as γ1 = g1 = 2πk1h , λl 0 − n2 n2 γ2 = 1 − k2 (n0 + n1)2 + k2 1 1 2n0k1 h1 = (n0 + n1)2 + k2 1 C = 2(g1g2 − h1h2), p2 = eγ1 cos(γ2), q2 = eγ1 sin(γ2), 2πn1h λl , , g2 = 1 − n2 n2 2 + k2 1 (n1 + n2)2 + k2 1 −2n2k1 h2 = (n1 + n2)2 + k2 1 D = 2(g1h2 + g2h1) p12 = p2 + g1t2 − h1u2 q12 = q2 + h1t2 + g1u2 t2 = e−γ1(g2 cos(γ2) + h2 sin(γ2)), t12 = t2 + g1p2 − h1q2 u2 = e−γ1(h2 cos(γ2) − g2 sin(γ2)), u12 = u2 + h1p2 + g1q2 and h is the metal film thickness, λl is the wavelength of the incident radiation, n0 is the refractive index of air, n1 and k1 are the metal refractive index and the extinction coefficient respectively, and n2 is the refractive index of the substrate. Figure 22 shows a schematic of the laser energy absorption. The incident energy E0 is perpendic- ular to the film surface. One part of the energy, R(h), is reflected at the film surface, and the rest of the energy, denoted by E1, penetrates the surface. Then, a part of the laser energy, denoted E2 is transmitted through the metal. Hence the energy absorbed by the metal film is A = E0 [1 − T (h)] [1 − R(h)] . (A3) The expressions for R and T given in Eqs. (A1) and (A2), can be approximated by simpler functions, as it is done by Trice et al. [18] T2(h) = e−αah, R2(h) = r0(cid:0)1 − e−arh(cid:1) , (A4) where αa = 4πk1/λl, and r0 and ar can be found by fitting T2 and R2 to Eqs. (A1) and (A2). Figure 23 shows the comparison of the reflectance and transmit- tance given by Eqs. (A1) and (A2) and Eq. (A4). The parameters used here are λl = 248 nm [34], n0 = 1, n1 = 1.7167, k1 = 2.3067, n2 = 1.59157 [43], R0 = 0.4594 and a−1 r = 8 nm. In Section II A for simplicity we use Eqs. (A4) for com- puting the absorption of the laser energy by a metal film. 21 Appendix B: The Temperature Solution of the Reduced Model in the Limit of Small Film Thickness The solution to the reduced temperature model given in Section II A 1 contains integrals that pose numerical difficulties for small film thicknesses. Here we give the expressions that can be used for computing the temper- m, and the gradient of the tem- ature of the metal film, T ∗ perature with respect to the film thickness, ∂T ∗ m/∂h, to alleviate those difficulties. In the limit of small film thick- ness, T ∗ m/∂h, can be expanded using asymptotic series as m and ∂T ∗ FIG. 22: E0 is the intensity of the incident radiation. R(h) and T (h) are the thickness dependent reflectance and transmittance of the metal. T ∗ m (h → 0, t) = T0 + S ∗ e − t2 p 2σ2 tp Z t 0 exp − (t − u)2 2σ2 tp + tp σ2 tp (t − u)(cid:19)" h CK√π u − h3 2√π (CK√u)3 + ...# du, (B1) ∂T ∂h (h → 0, t) = ∂S ∗ ∂h − e t2 p 2σ2 tp Z t 0 − S ∗ e where exp − tp Z t 0 + (t − u)2 2σ2 tp exp − tp σ2 tp (t − u)2 2σ2 tp t2 p 2σ2 (t − u)!" h CK√π u − h3 2√π (CK√u)3 + ...# du+ + tp σ2 tp (t − u)!" 1 CK√π u − 3 h2 2√π (CK√u)3 + ...# du. (B2) S ∗ (h → 0, t) = αa(cid:20)1 −(cid:16) αa 2 + arr0(cid:17) h + 1 6(cid:0)α2 a + 3αaarr0 + 3a2 rr0(cid:1) h2 + ...(cid:21) , Hence, the integrals in Eqs. (23) and (25) are convergent as h → 0. In our simulations, we use the expression given here for small film thicknesses, since the direct evaluation of the integrals in Eqs. (23) and (25) is numerically dif- ficult. Appendix C: Analytical Temperature Solution Here we provide the details of the analytical temper- ature solution in the fluid-substrate domain specified in Section II A 2. Note that in order to simplify the presen- tation, we change the notation for the domain boundaries compared to Section II A, and denote the bottom of the substrate as y = 0, the fluid-substrate interface as y = a, and the fluid-air interface as y = b. The temperature in the fluid, Tm, and the temperature in the substrate, Ts, as h → 0. (B3) in 0 < y < a (C1) in a < y < b (C2) satisfy the diffusion equation ∂Ts ∂t ∂2Ts ∂y2 = Ds ∂2Tm ∂y2 + S (y, t) ∂Tm ∂t = Dm where ks Ds = , Dm = along with the boundary conditions ρsCef fs km , ρmCef fm Ts = T0 Ts = Tm ks ∂Ts ∂y ∂Tm ∂y km = km ∂Tm ∂y = 0 at y = 0, at y = a, at y = a, (C3) (C4) (C5) at y = b. (C6) R 0.5 0.4 0.3 0.2 0.1 0 0 22 T T 1 2 1 0.8 0.6 0.4 0.2 T R R 1 2 20 Height [nm] 40 60 0 0 20 Height [nm] 40 60 FIG. 23: The comparison of the reflectance and transmittance given by the Eqs. (A1) and (A2) and Eq. (A4). The source term, S(y, t) is given by Eq. (15). The solu- tion to the above equations can be written compactly in terms of Green's functions as given in Eqs. (27) and (28), where The eigenfunctions ψi,n satisfy the boundary conditions in Eqs. (C3) - (C6). Hence, it follows ψ1,n = 0 at y = 0 → B1,n = 0, A1,n = 1 without loss of generality, (C13) ψ1,n = ψ2,n at y = a Gi,j (y, t; ξ, τ ) = Nn = ∞ Xn=1 Ds Z a ks 0 e−β2 n(t−τ ) 1 Nn kj αj ψi,n (y) ψj,n (ξ) (C7) ψ2 2,ndξ . (C8) ψ2 1,ndξ + km Dm Z b a Here ψi,n and βn are eigenfunctions and eigenvalues com- puted using separation of variables, and ψi,n = Ai,nΦi,n(y) + Bi,nΘi,n(y) in yi < y < yi+1, (C9) y(cid:19), Φi,n(y) = sin(cid:18) βn √αi Θi,n = cos(cid:18) βn√αi y(cid:19), where y0 = 0, y1 = a and y2 = b. In order to simplify the notation, let b ks km ∂ψ1,n ∂ψ2,n → sin γ = A2,n sin(cid:16) a → K cos γ = A2,n cos(cid:16) a η(cid:17) + B2,n cos(cid:16) a η(cid:17), η(cid:17) − B2,n sin(cid:16) a at y = a ∂y ∂y = b b b η(cid:17), = 0 at y = b ∂ψ2 ∂y → A2,n cos η − B2,n sin η = 0. (C14) (C15) (C16) We can solve for the coefficients A2,n and B2,n using Eqs. (C14) and (C15) b η(cid:17) − K cos γ cos(cid:16) a η(cid:17)i , η(cid:17) − sin γ cos(cid:16) a η(cid:17)i , η(cid:17) = −1. b η(cid:17) − cos2(cid:16) a b (C17) (C18) (C19) (C10) where (C11) A2,n = B2,n = b b 1 1 ∆h− sin γ sin(cid:16) a ∆hK cos γ sin(cid:16) a ∆ = − sin2(cid:16) a (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) sin γ − sin(cid:0) a K cos γ − cos(cid:0) a − cos η 0 b In order to have a solution, we require vanishing deter- minant of the system of Eqs. (C14) - (C16) b η(cid:1) − cos(cid:0) a b η(cid:1) b η(cid:1) − sin(cid:0) a b η(cid:1) − sin η = 0. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) The equation above leads to the following equation for the eigenvalues, βn γ = aβn√Ds , η = bβn√Dm , K = ks kmrDm Ds . (C12) tan aβn√Ds tan(cid:18) βn√Dm (b − a)(cid:19) = K, (C20) 4500 4000 3500 3000 2500 2000 1500 1000 500 ] K [ t e r u a r e p m e T 0 0 4500 4000 3500 3000 2500 2000 1500 1000 500 ] K [ e r u t a r e p m e T 200 nm 400 nm 600 nm 10 20 30 40 50 time [ns] (a) 0 0 0
2 0 time [ns] 30 (b) 23 200 nm 400 nm 600 nm 0 4 0 5 FIG. 24: Convergence of the analytical solution with increased substrate depth for film thickness of (a) h0 = 10 nm and (b) h0 = 20 nm. which can be solved numerically. (C6) and (D1). Hence, it follows Figure 24 shows the average temperature in the metal as a function of time for different substrate thicknesses. We see that the solution converges as the substrate thick- ness increases. Hence, for a substrate thick enough, the solution is equivalent to the one obtained in the setup in- volving semi-infinite substrate. This result is used in Sec- tion II A to justify comparing the temperature obtained using the reduced model and semi-infinite substrate, with the one obtained by using the complete model and finite substrate thickness. ∂ψ1,n ψ1,n = 0 at y = 0 → B1,n = 1, A1,n = 1 without loss of generality, (D2) − ks = α (ψi,n − ψ2,n) at y = a → −ks cos γ = α sin γ − αA2,n sin(cid:16) a η(cid:17) − αB2,n cos(cid:16) a b (D3) ∂y b η(cid:17), ks ∂ψ1,n ∂y = km ∂ψ2,n ∂y at y = a η(cid:17) − B2,n sin(cid:16) a b → K cos γ = A2,n cos(cid:16) a = 0 at y = b ∂ψ2 ∂y → A2,n cos η − B2,n sin η = 0. b η(cid:17), (D4) (D5) Appendix D: Newton's Law of Cooling From the Eqs. (D3) and (D4), we can solve for the coef- ficients A2,n and B2,n We show that replacing the continuity of temperature boundary condition at the fluid-substrate interface with Newton's law of cooling yields an equivalent solution as long as the heat transfer coefficient is large enough. We replace the boundary condition (C4) by A2,n = B2,n = where 1 ∆h(−H cos γ − sin γ) sin(cid:16) a ∆hK cos γ sin(cid:16) a η(cid:17) − K cos γ cos(cid:16) a η(cid:17) + (−H cos γ − sin γ) cos(cid:16) a 1 b b b η(cid:17)i , η(cid:17)i , b (D6) (D7) ∂Ts ∂y − ks = α (Ts − Tm) at y = a. (D1) The condition for existence of a solution is vanishing de- terminant as follows H = ksβn α√Ds . (D8) (the notation is the same as in the preceding appendix). Then, the eigenfunctions of the same form as given in Eq. (C9) satisfy the boundary conditions (C3), (C5), The equation satisfied by the eigenvalues βn is now (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) −H cos γ − sin γ − sin(cid:0) a − cos(cid:0) a − cos η K cos γ b η(cid:1) − cos(cid:0) a b η(cid:1) b η(cid:1) − sin(cid:0) a b η(cid:1) − sin η 0 = 0. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:18)H + tan aβn√Ds(cid:19) tan(cid:18) βn√Dm (b − a)(cid:19) = K (D9) which can be solved numerically. 24 Figure 25 shows the average temperature in the metal as a function of time for different α. We see that the solu- tion with Newton's cooling law converges to the solution with continuity of temperature for large α. This result is used in Section II A to justify comparing the reduced model that implements continuity of temperature with the complete model that uses Newton's law of cooling. [1] R. A. Hughes, E. Menumerov, and S. Neretina, "When lithography meets self-assembly: a review of recent ad- vances in the directed assembly of complex metal nanos- tructures on planar and textured surfaces," Nanotechnol- ogy 28, 282002 (2017). [2] S. V. Makarov, V. A. Milichko, I. S. Mukhin, I. I. Shishkin, D. A. Zuev, A. M. Mozharov, A. E. Krasnok, and P. A. Belov, "Controllable femtosecond laser-induced dewetting for plasmonic applications," Laser Photonics Rev. 10, 91 (2016). [3] J. Bischof, D. Scherer, S. Herminghaus, and P. Leiderer, "Dewetting modes of thin metallic films: Nucleation of holes and spinodal dewetting," Phys. Rev. Lett. 77, 1536 (1996). [4] F. M. Ross, "Controlling nanowire structures through real time growth studies," Rep. Prog. Phys. 73, 114501 (2010). [5] C. Favazza, J. Trice, H. Krishna, and R. Kalyanara- man, "Laser induced short- and long-range orderings of Co nanoparticles on SiO2," Appl. Phys. Lett. 88, 153118 (2006). [6] Q. Xia and S.Y. Chou, "The fabrication of periodic metal nanodot arrays through pulsed laser melting in- duced fragmentation of metal nanogratings," Nanotech. 20, 285310 (2009). [7] H. Krishna, R. Sachan, J. Strader, C. Favazza, M. Khen- ner, and R. Kalyanaraman, "Thickness-dependent spon- taneous dewetting morphology of ultrathin Ag films," Nanotech. 21, 155601 (2010). [8] J. D. Fowlkes, N. A. Roberts, Y. Wu, J. A. Diez, A. G. Gonz´alez, C. Hartnett, K. Mahady, S. Afkhami, L. Kondic, and P. D. Rack, "Hierarchical nanoparti- cle ensembles synthesized by liquid phase directed self- assembly," Nano Letters 14, 774 (2014). [9] H. Reinhardt, H-C. Kim, and C. Pietzonka, "Self- organization of multifunctional surfaces - the fingerprints on a complex system," Advanced Mat. 25, 3313 (2013). [10] L. Kondic, J. A. Diez, P. D. Rack, Y. Guan, and J. D. Fowlkes, "Nanoparticle assembly via the dewetting of patterned thin metal lines: Understanding the instability mechanisms," Phys. Rev. E. 79, 1 (2009). [11] A. G. Gonzalez, J. D. Diez, Y. Wu, J.D. Fowlkes, P. D. Rack, and L. Kondic, "Instability of Liquid Cu Films on a SiO2 Substrate," Langmuir 13, 9378 (2013). [12] A. G. Gonzalez, J. D. Diez, and L. Kondic, "Stability of a liquid ring on a substrate," J. Fluid Mech. 718, 213 (2013). [13] L. Kondic, N. Dong, Y. Wu, J. D. Fowlkes, and P. D. Rack, "Instabilities of nanoscale patterned metal films," EPJST 224, 369 (2015). [14] K. Mahady, S. Afkhami, and L. Kondic, "A numerical approach for the direct computation of flows including fluid-solid interaction: Modeling contact angle, film rup- ture, and dewetting," Phys. Fluids 28, 062002 (2016). [15] K. Mahady, S. Afkhami, and L. Kondic, "A volume of fluid method for simulating fluid/fluid interfaces in con- tact with solid boundaries," J. Comput. Phys. 294, 243 (2015). [16] K. Mahady, S. Afkhami, and L. Kondic, "On the in- fluence of initial geometry on the evolution of fluid fila- ments," Phys. Fluids 27, 092104 (2015). [17] J. Diez, A. Gonz´alez, and L. Kondic, "On the breakup of fluid rivulets," Phys. Fluids 21, 082105 (2009). [18] J. Trice, D. Thomas, C. Favazza, R. Sureshkumar, and R. Kalyanaraman, "Pulsed-laser-induced dewetting in nanoscopic metal films: Theory and experiments," Phys. Rev. B 75, 235439 (2007). [19] J. Trice, D. Thomas, C. Favazza, R. Sureshkumar, and R. Kalyanaraman, "Novel Self-Organization Mechanism in Ultrathin Liquid Films: Theory and Experiment," Phys. Rev. Lett. 101, 017802 (2008). [20] A. Atena and M. Khenner, "Thermocapillary effects in driven dewetting and self assembly of pulsed-laser- irradiated metallic films," Phys. Rev. B 80, 075402 (2009). [21] N. Dong and L. Kondic, "Instability of nanometric fluid films on a thermally conductive substrate," Phys. Rev. Fluids 1, 063901 (2016). [22] I. Seric, S. Afkhami, and L. Kondic, "Direct numeri- cal simulation of variable surface tension flows using a Volume-of-Fluid method," to appear: J. Comput. Phys. (2017). [23] F. Font, S. Afkhami, and L. Kondic, "Substrate melting during laser heating of nanoscale metal films," Int. J. Heat Mass Transfer 113, 237 (2017). [24] J. U. Brackbill, D. B. Kothe, and C. Zemach, "A contin- uum method for modeling surface tension," J. Comput. Phys. 100, 335 (1992). [25] M. M. Francois, S. J. Cummins, E. D. Dendy, D. B. Kothe, J. M. Sicilian, and M. W. Williams, "A balanced- force algorithm for continuous and sharp interfacial sur- face tension models within a volume tracking frame- work," J. Comput. Phys. 213, 141 (2006). [26] S. Afkhami and M. Bussmann, "Height functions for ap- plying contact angles to 3D VOF simulations," Int. J. Numer. Meth. Fluids 61, 827 (2009). [27] S. Popinet, "An accurate adaptive solver for surface- tension-driven interfacial flows," J. Comput. Phys. 228, 5838 (2009). [28] L. D. Landau and E. M. Lifshitz, Fluid mechanics, 2nd, Vol. 6 (Pergamon Press, Oxford, 1987). [29] V. G. Levich and V. S. Krylov, "Surface-tension-driven phenomena," Annu. Rev. Fluid Mech. 1, 293 (1969). [30] J. U. Brackbill, D. B. Kothe, and C. Zemach, "A contin- 5000 4000 3000 2000 1000 ] K [ e r u t a r e p m e T 5000 4000 3000 2000 1000 ] K [ e r u t a r e p m e T 25 0 0 10 20 30 40 50 0 0 10 20 30 40 50 time [ns] (a) time [ns] (b) FIG. 25: Convergence of the analytical solution with increased α for film thickness of (a) h0 = 10 nm and (b) h0 = 20 nm. uum method for modeling surface tension," J. Comput. Phys. 100, 335 (1992). [31] V. S. Ajaev and D. A. Willis, "Thermocapillary flow and rupture in films of molten metal on a substrate," Phys. Fluids 15, 3144 (2003). [32] O. S. Heavens, Optical properties of thin solid films (New York: Dover Publications, 1991). [33] W. F. Gale and T. C. Totemeier, Smithells metals refer- ence book (Butterworth-Heinemann, 2003). [34] C. A. Hartnett, I. Seric, K. Mahady, L. Kondic, S. Afkhami, J. D. Fowlkes, and P. D. Rack, "Exploiting the Marangoni effect to initiate instabilities and direct the assembly of liquid metal filaments," Langmuir 33, 8123 (2017). [35] M. M. Francois and B. K. Swartz, "Interface curva- ture via volume fractions, heights, and mean values on nonuniform rectangular grids," J. Comput. Phys. 229, 527 (2010). [36] M. Necati Ozı¸sık, Boundary value problems of heat con- duction (Dover Publications, 1989). [37] S. Popinet, "Gerris Flow Solver," , (2013-12- http://gfs.sourceforge.net/wiki/index.php 06). [38] A. Oron, S. H. Davis, and S. G. Bankoff, "Long-scale evolution of thin liquid films," Rev. Mod. Phys. 69, 931 (1997). [39] J. A. Diez and L. Kondic, "On the breakup of fluid films of finite and infinite extent," Phys. Fluids 19, 072107 (2007). [40] L. Rayleigh, "On the instability of jets," Proc. London Math. Soc. 1, 4 (1878). [41] J. Eggers, "Nonlinear dynamics and breakup of free- surface flows," Rev. Mod. Phys. 69, 865 (1997). [42] J. A. Diez, A. G. Gonzalez, and L. Kondic, "On the breakup of fluid rivulets," Phys. Fluids 21 (2009). [43] P. B. Johnson and R. W. Christy, "Optical constants of transition metals: Ti, V, Cr, Mn, Fe, Co, Ni, and Pd," Phys. Rev. B 9, 5056 (1974). 

1610.02480

1

1610

Engineering the structural and electronic phases of MoTe2 through W substitution

MoTe$_2$ is an exfoliable transition metal dichalcogenide (TMD) which crystallizes in three symmetries, the semiconducting trigonal-prismatic $2H-$phase, the semimetallic $1T^{\prime}$ monoclinic phase, and the semimetallic orthorhombic $T_d$ structure. The $2H-$phase displays a band gap of $\sim 1$ eV making it appealing for flexible and transparent optoelectronics. The $T_d-$phase is predicted to possess unique topological properties which might lead to topologically protected non-dissipative transport channels. Recently, it was argued that it is possible to locally induce phase-transformations in TMDs, through chemical doping, local heating, or electric-field to achieve ohmic contacts or to induce useful functionalities such as electronic phase-change memory elements. The combination of semiconducting and topological elements based upon the same compound, might produce a new generation of high performance, low dissipation optoelectronic elements. Here, we show that it is possible to engineer the phases of MoTe$_2$ through W substitution by unveiling the phase-diagram of the Mo$_{1-x}$W$_x$Te$_2$ solid solution which displays a semiconducting to semimetallic transition as a function of $x$. We find that only $\sim 8$ \% of W stabilizes the $T_d-$phase at room temperature. Photoemission spectroscopy, indicates that this phase possesses a Fermi surface akin to that of WTe$_2$.

cond-mat.mes-hall

cond-mat

Engineering the structural and electronic phases of MoTe2 through W substitution D. Rhodes†,1, 2 D. A. Chenet†,3 B. E. Janicek,4 C. Nyby,5 Y. Lin,6 W. Jin,6 D. Edelberg,7 E. Mannebach,8 N. Finney,3 A. Antony,3 T. Schiros,9, 10 T. Klarr,11 A. Mazzoni,11 M. Chin,11 Y.-c Chiu,1, 12 W. Zheng,13, 12 Q. R. Zhang,13, 12 F. Ernst,14, 15 J. I. Dadap,16 X. Tong,17 J. Ma,18 R. Lou,19 S. Wang,19 T. Qian,18 H. Ding,18 R. M. Osgood, Jr,16, 6 D. W. Paley,20, 21 A. M. Lindenberg,8, 15, 22 P. Y. Huang,23 A. N. Pasupathy,7 M. Dubey,11 J. Hone,3 and L. Balicas1, ∗ 1National High Magnetic Field Laboratory, Florida State University, Tallahassee, FL 32310, USA 2Department of Physics, Florida State University, Tallahassee-FL 32306, USA 3Department of Mechanical Engineering, Columbia University, New York, NY 10027, USA 4Department of Materials Science and Engineering, University of Illinois UrbanaChampaign, Urbana, IL 61801, USA 5Department of Chemistry, Stanford University, Stanford University, Stanford, CA 94305-4401, USA 6Department of Applied Physics and Applied Mathematics, Columbia University, New York, NY 10027, USA 7Department of Physics, Columbia University, New York, NY 10027, USA 8Department of Materials Science and Engineering, Stanford University, Stanford, CA 94305, USA 9Columbia University, Materials Research Science and Engineering Center, NY, NY 10027 USA 10SUNY Fashion Institute of Technology, Department of Science and Mathematics, NY, NY 10001 USA 11Sensors and Electronic Devices Directorate US Army Research Laboratory Adelphi, MD 20723, USA 12Department of Physics and National High Magnetic Field Laboratory, Florida State University, Tallahassee, FL 32310, USA 13National High Magnetic Field Laboratory, Florida State University, Tallahassee-FL 32310, USA 14Department of Applied Physics, Stanford University, Stanford, CA 94305-4090, USA 15Stanford PULSE Institute, SLAC National Accelerator Laboratory, Menlo Park, CA 94025, USA 16Department of Electrical Engineering, Columbia University, New York, NY 10027, USA 17Center for Functional Nanomaterials, Brookhaven National Laboratory, Upton, NY 11973-5000, USA 18Beijing National Laboratory for Condensed Matter Physics, and Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China 19Department of Physics, Renmin University of China, Beijing 100872, China 20Department of Chemistry, Columbia University, New York, NY 10027, USA 21Columbia Nano Initiative, Columbia University, New York, NY 10027, USA 22Stanford Institute for Materials and Energy Sciences, SLAC National Accelerator Laboratory, Menlo Park, CA 94025, USA 23Department of Mechanical Science and Engineering, University of Illinois UrbanaChampaign, Urbana, IL 61801, USA MoTe2 is an exfoliable transition metal dichalcogenide (TMD) which crystallizes in three symmetries; the semiconducting trigonal-prismatic 2H−phase, the semimetal- lic 1T ′ monoclinic phase, and the semimetallic orthorhombic T d structure1–4. The 2H−phase displays a band gap of ∼ 1 eV5 making it appealing for flexible and transparent optoelectronics. The T d−phase is predicted to possess unique topolog- ical properties6–9 which might lead to topologically protected non-dissipative trans- port channels9. Recently, it was argued that it is possible to locally induce phase- transformations in TMDs3,10,11,14, through chemical doping12, local heating13, or electric-field14,15 to achieve ohmic contacts or to induce useful functionalities such as electronic phase-change memory elements11. The combination of semiconducting and topological elements based upon the same compound, might produce a new generation of high performance, low dissipation optoelectronic elements. Here, we show that it is possible to engineer the phases of MoTe2 through W substitution by unveiling the phase-diagram of the Mo1−xWxTe2 solid solution which displays a semiconducting to semimetallic transition as a function of x. We find that only ∼ 8 % of W stabilizes the T d−phase at room temperature. Photoemission spectroscopy, indicates that this phase possesses a Fermi surface akin to that of WTe2 16. 6 1 0 2 t c O 8 ] l l a h - s e m . t a m - d n o c [ 1 v 0 8 4 2 0 . 0 1 6 1 : v i X r a The properties of semiconducting and of semimetal- lic MoTe2 are of fundamental interest in their own right, but are also for their potential technological relevance. In the mono- or few-layer limit it is a direct-gap semicon- ductor, while the bulk has an indirect bandgap5,17,18 of ∼ 1 eV. The size of the gap is similar to that of Si, mak- ing 2H −MoTe2 particularly appealing for both purely electronic devices19,20 and optoelectronic applications21. Moreover, the existence of different phases opens up the possibility for many novel devices and architectures. For example, controlled conversion of the 1T ′−MoTe2 phase to the 2H −phase, as recently reported22, could enable circuits composed of a single material functioning as both semiconducting channels and metallic intercon- nects. More precise control of the phase change might also be used to minimize the metal-semiconductor Schot- tky barrier by continuous evolution of the electronic band structure, in order to overcome current limits on opto- electronic performance23. In fact, recent work has re- ported contact phase engineering by laser processing13 and chemical modification12. The ability to phase-engineer MoTe2 has many broader applications and potentially deeper implica- tions. For instance, doping, temperature, strain, and electric fields can be used to drive metal-to-insulator transitions10,14,24,25 for sensors and nonvolatile informa- tion storage. More fundamentally, the electronic struc- ture of mono-layers of semimetallic MoTe2 (and of WTe2) have been proposed to possess a Z2 topological invariant characteristic of a quantum spin Hall-effect ground-state which has a gap in the bulk and non-dissipative edge states9. If confirmed4, these edge states could be used for dissipation-free nano-interconnects between logical el- ements based on semiconducting 2H −MoTe2 for low- power electronics. More recent theoretical developments also claim that both orthorhombic MoTe2 and WTe2 would be candidates for a new type of Weyl semimetal- lic state characterized by linear touching points between hole- and electron-Fermi surfaces, where the Berry-phase would present topological singularities6–8,26–28. These singularities, which were recently claimed to have been 29, could observed in the orthorhombic phase of MoTe2 lead to unconventional transport properties. To fully control and utilize phase transitions in the two-dimensional (2D) tellurides, it is crucial to under- stand the phase diagram in detail. In particular, dop- ing of the lattice can be used to precisely tune the semiconducting-metallic phase transition, and in fact W doping is known to induce a phase transition24 from the 2H − to an orthorhombic structure, originally iden- tified as the 1T ′−phase, in Mo1−xWxTe2. Early studies identified a structural phase-transition from 2H to Td for x > 0.15, and with a zone of phase coexistence for 0.15 < x < 0.3430. However, given the renewed interest in this material, there is strong motivation to revisit the question of the precise evolution of the phases in the 2D tellurides with doping. Here, we synthesize bulk crystals of Mo1−xWxTe2 al- loys, and characterize their composition and structure through a combination of techniques including electron microscopy, x-ray diffraction, scanning tunneling mi- croscopy, and Raman spectroscopy. We find that W doping produces homogeneous alloys, with no phase co- existence as previously observed30. The structural phase transition from the semiconducting 2H −phase towards the orthorhombic and semimetallic Td−phase is sharp and occurs at a modest critical molar fraction xc ∼ 0.08. Since crystals with x . xc are likely to be susceptible to small perturbations such as strain or electric field, this opens the possibility of reversibly controlling the 2 structural, and therefore electronic properties, of the Mo1−xWxTe2 series. Additionally, we show through angle resolved photoemission spectroscopy that the ge- ometry of the Fermi surface of Td−Mo1−xWxTe2 is re- markably similar to that of WTe2, thus confirming its semimetallic character. Single crystals of the Mo1−xWxTe2 series were grown through a chemical vapor transport technique as de- scribed in Methods. Unless otherwise noted, samples were cooled slowly in order to obtain the equilibrium phase at room temperature. Their precise stoichiometry was determined through energy dispersive X −ray spec- troscopy (EDS) and photoelectron spectroscopy (XPS), see Methods as well as Supplementary Fig. S1 for pho- toelectron core level spectrum of a Mo1−xWxTe2 crystal and Supplementary Fig. S2 for details concerning the determination of the W content (x ± 0.01). Stoichiomet- ric MoTe2 (x = 0) and WTe2 (x = 1) were synthesized through a Te flux method. For MoTe2, samples were slowly cooled to yield the 2H −phase or quenched to room temperature to yield the metastable 1T ′−phase. Figure 1 shows structural analysis via single crystal x-ray diffraction (XRD), scanning transmission electron microscopy (STEM), and scanning tunneling microscopy (STM). For STEM, the crystals were exfoliated following a standard procedure and transferred onto a TEM grid, see Methods. Figures 1a and 1b display atomic reso- lution STEM images collected from two distinct multi- layered crystals with compositions of Mo0.93W0.07Te2 and Mo0.87W0.13Te2, respectively. These crystals display distinct crystallographic structures: x ≃ 0.07 shows the hexagonal pattern characteristic of the trigonal prismatic or the 2H − phase, while x ≃ 0.13 shows a striped pat- tern consistent with either the 1T ′− or the Td-phase. In Supplementary Fig. S3 we have included STEM and electron diffraction images for x ≃ 0.13 from whose anal- ysis we conclude that it crystallizes in the orthorhombic Td phase. Nevertheless, in Figs. 1c and 1d we also show single-crystal XRD patterns for x = 0.0 and x = 0.27, re- spectively. Analysis of these patterns confirms that crys- tals with x . 0.07 crystallize in the 2H −phase, whereas crystals with x > 0.07, in this case x = 0.27, display the orthorhombic Td−phase instead of the monoclinic 1T ′− one. Supplementary Fig. S4 shows X −ray diffrac- tion patterns for Mo0.91W0.09Te2 and Mo0.82W0.18Te2, also indicating the Td−phase for these concentrations. Our complete set of structural studies indicate that for all concentrations x > xc = 0.08, W doping stabilizes the semimetallic Td phase, confirming that the structural transition is sharp and occurs at a W doping level signifi- cantly lower than previously reported30. Figure 1e shows a larger-scale STEM image of Mo0.93W0.07Te2. In this image, bright dots surrounded by three additional dots (Te atoms) correspond to randomly distributed W atoms, as highlighted in Fig. 1f. Therefore, the STEM images clearly indicate that the Mo1−xWxTe2 series results from a homogeneous dilution of W atoms into a MoTe2 ma- trix and not from the coexistence of 2H −MoTe2 and 3 2H – Mo1-xWx (cid:100)(cid:3)0.07Te2 T – Mo1-xWx > 0.07Te2 d a b c d 5 Å 5 Å e f 5 Å g 5 Å h i j 1L x = 0.07 x = 0.07 x = 0.13 FIG. 1. Structural analysis of few-layered Mo1−xWxTe2 crystals. a, Scanning transmission electron microscopy image of a few-layered crystal of Mo1−xWxTe2 with x ≃ 0.07 % displaying the hexagonal pattern typical of the 2H− or trigonal prismatic phase. b, STEM image of a few-layered crystal of Mo1−xWxTe2 with x ≃ 0.13. Notice that its atomic arrangement is no longer consistent with the 2H−phase. All STEM images are lightly smoothed. c Single-crystal X -ray diffraction pattern for x = 0.0 (00l) indicating that it crystallizes in the 2H−phase. d, Single-crystal X -ray diffraction pattern for x = 0.27 (0kl) indicating that it crystallizes in the Td phase (orthorhombic P mn21). Powder X -ray diffraction indicates that for x > 0.08 the Mo1−xWxTe2 series crystallizes in the Td−phase. e, (8.25 nm)2 area STEM image of monolayer 2H−Mo1−xWxTe2. f, Brighter W atoms (indicated by red dots) are identifiable through their contrast with respect to the darker Mo atoms. Therefore, STEM indicates that these crystals are homogeneous solid-solutions containing Mo and W atoms. g, Scanning tunneling microscopy image of a Mo1−xWxTe2 single crystal with x = 0.07, corresponding to an area of (15 nm)2 and showing a clear hexagonal pattern as expected for the 2H−phase. The spatial modulation in intensity reflects the local coupling between the layers. h, Magnification of a local area of (2.5 nm)2 where one can detect a Te vacancy. i, STM image of a x = 0.13 single-crystal, also corresponding to an area of (15 nm)2, showing a pattern of parallel chains as expected for the orthorhombic Td−phase. j, Magnification of a local region of (2.5 nm)2 revealing the intra-chain structure and illustrating the crystallographic positions of transition metal (black dots) and Te (yellow dots) atoms, respectively. WTe2 domains. This lack of phase coexistence is fur- ther confirmed by room-temperature STM imaging of vacuum-cleaved crystals, as shown in Figs. 1g and 1h. For x = 0.07, see Fig. 1g, the equidistant distribution of Te atoms around the transition metal(s) forming an an- gle of θ = 120◦ among them, indicates unambiguously the trigonal prismatic coordination of the 2H −phase. In contrast, for x = 0.13, see Figs. 1i and 1j, rows of atoms indicate a change in symmetry from triangular to (nearly) rectangular at the surface. In amplified images, e.g. Fig. 1h, one can clearly discern Te vacancies (indi- cated by a multicolored dot). Therefore, we have enough resolution to observe vacancies, but we do not observe the coexistence of distinct crystallographic phases. Thus TEM, STM, and XRD analysis yield consistent results, namely a transition from the 2H phase to the Td phase at xc ∼ 0.08, with no phase coexistence even near the phase boundary. These observations stand in contrast to the early work in Ref. 30, which reported a higher critical W concentration and a region of phase coexistence near the boundary. This discrepancy is likely attributable to the difference between the methods of a b 4 2H-MoTe2 1T'-MoTe2 Td-WTe2 x = 0 x = 0.03 x = 0.06 x = 0.08 x = 0.09 x = 0.10 x = 0.11 x = 0.12 x = 0.21 x = 0.23 x = 0.27 x = 1 ) s t i n U . b r A ( y t i s n e n t I n a m a R Raman Shift (cm -1) Raman Shift (cm -1) FIG. 2. Raman spectra of the Mo1−xWxTe2 series as a function of W doping. a, Raman spectra of 2H−MoTe2 (black trace), 1T ′−MoTe2 (blue) and T d−WTe2 (red) at room temperature using an excitation wavelength of 532 nm. b, Raman spectra of Mo1−xWxTe2 for several values of x (fraction of W). Notice the change in the spectrum observed for x ≥ 0.09, indicating a structural phase-transition as a function of doping . synthesis used for each study. Having established the room temperature phase boundary between the 2H and the Td transition, we now turn to the temperature axis of the phase-diagram. The structural phase transition as a function of dop- ing is accompanied by changes in vibrational modes, as probed by Raman spectroscopy (see, Fig. 2). Figure 2a shows room-temperature Raman spectra obtained from the 2H- and 1T ′−MoTe2 phases, and from WTe2. The 2H − structure displays two main Raman peaks at 174 cm−1 and 235 cm−1 corresponding to the A1g and E1 g modes, respectively19,31. Reflecting its reduced sym- metry, the 1T ′ phase displays several peaks at lower wave-numbers. For this structure the main peaks oc- cur at 163 cm−1 and 260 cm−1 and have been indexed as the Bg and Ag modes4, respectively. WTe2 presents a spectrum having peaks occurring at 136 cm−1 and 165 cm−1 respectively, both of Ag character32, in addition to a peak at ∼ 210 cm−1 previously indexed as the A2 1 mode. In Supplementary Fig. S6 we show Raman spec- tra for Td−Mo0.88W0.12Te2 as the number of layers de- crease, indicating that the Td-phase is stable down to the single-layer limit despite its high Mo content. Fig- ure 2b shows Raman spectra for several stoichiometries of the Mo1−xWxTe2 series. Mo1−xWxTe2 crystallizes in the 2H −phase for concentrations up to x ∼ 0.08. For concen- trations beyond this value the spectra abruptly change, as indicated by the disappearance of the A1g and of the 2g peaks at 174 cm−1 and 235 cm−1 respectively, which E1 are observed when x . 0.08. These data support the con- clusions reached by the structural probes above, namely a phase transition around xc ∼ 0.08 with no evidence for phase coexistence. Interestingly, As the W concentration increases beyond x = 0.08 we see the emergence of peaks which, at the first glance, would seem to be related to the Bg and the Ag modes of the 1T ′-phase4. However, single crystal X −ray diffraction shown in Fig. 1d and in Sup- plementary Fig. S3 clearly indicate that the Raman spec- tra in Fig. 2b must be associated with the orthorhombic Tdphase, with certain peaks shifted with respect to those of WTe2 due to the high Mo content. An important observation is that Raman scattering yields nearly iden- tical spectra for 1T ′−MoTe2 and for Td−Mo1−xWxTe2, for reasons that will have to be clarified through theo- retical calculations. We note that this similarity might lead to misidentification of the 1T ′ phase if Raman spec- troscopy is the only method used to probe the crystal structure. In Supplementary Fig. S6, we show Raman scattering data as a function of the number of atomic layers for a crystal having x = 0.12, which is close to the critical concentration xc ≃ 0.08, indicating that it pre- serves its structure upon exfoliation despite its proximity to the phase-boundary. In Supplementary Fig. S7, we have include transport data, like the room temperature conductivity as a function of doping, which changes by orders of magnitude as one crosses the phase-boundary. Next, we investigated the electronic phase-transition accompanying the structural phase transition. In partic- ular, while the nature of the semiconducting 2H −phase is well understood, it is not known whether the Td phase in the W-doped material is a conventional, or a Weyl, semi-metallic system. Therefore, we investigated the b c 5 kz Z (cid:299) U X a kx T R S Y ky d e f g h k (Å-1) E - EF (eV) k (Å-1) k (Å-1) k (Å-1) Angle resolved photoemission spectroscopy (ARPES) of Mo1−xWxTe2. FIG. 3. a, Bulk Brillouin zone of Mo0.73W0.27Te1.99 indicating its high symmetry points. b, ARPES intensity plot at the Fermi level as a function of both the momentum and the photon energy. c, Topography of the Fermi surface at kz = 0. One electron- and one hole-like pocket is observed at either side of the Γ−point. The resolution of our ARPES measurements limits our ability to identify possible points of contact between the electron and the hole pockets. d, ARPES band structure along Y-Γ-Y direction acquired with a photon energy of 70 eV, i.e. corresponding to (kz = 0). e, Plot of the energy distribution curve of the low energy bands. Blue dotted line serves as a guide to the eye, indicating the positions of peaks for the electron-like band. f, Second derivative of the band structure collected along the Y-Γ-Y direction. g, ARPES band structure along T-Z-T direction acquired with a photon energy of 57 eV, i.e. corresponding to (kz = π). h, Second derivative of the band structure collected along the T-Z-T direction. electronic structure of heavily doped Mo1−xWxTe2 sin- gle crystals through angle-resolved photoemission spec- troscopy (ARPES), as shown in Fig. 3. The core level spectrum, shown in the Supplementary Fig. S1, displays the characteristic peaks of W and Te elements, confirm- ing that W is alloyed into the 1T ′−MoTe2 crystal. As seen in this figure, the W 4f core levels have one set of doublets at 31.4 eV and 33.6 eV (right inset in Supple- mentary Fig. S1) respectively, in perfect agreement with the values found in the literature33. Meanwhile the Te 4d5/2 and 4d3/2 doublets split into four peaks (left inset in Supplementary Fig. S1). This suggests that the Mo/W layer is sandwiched by the Te layers, making the Te layer the exposed surface. To investigate the electronic struc- ture along the kz-direction of the three-dimensional Bril- louin zone (BZ), which is depicted in Fig. 3a, we per- formed photon-energy-dependent ARPES measurements with energies ranging from 40 to 90 eV. Figure 3b shows the ARPES spectra at the Fermi level EF as a function of the momentum and photon energy from 55 to 75 eV. We extracted the positions of the Γ (kz = 0) and Z (kz = π) points from the dispersion as a function of kz, as shown. Figure 3c shows the Fermi surface of Mo0.73W0.27Te2 ac- quired at hν = 70 eV. The Fermi surface along the Y-Γ-Y direction shows two hole-pockets and two-electron pock- ets at either side of Γ which would seem to touch. This geometry for the Fermi surface of Mo0.73W0.27Te2 (as well as its overall electronic band-structure) is remark- ably similar to the one reported in Ref. 34 for WTe2 and therefore remarkably different from the one already reported29 for orthorhombic MoTe2. This difference is particularly striking given its considerably larger content of Mo relative to W. Notice that such a simple Fermi sur- face would be in broad agreement with our recent study36 on the quantum oscillatory phenomena observed in Td- MoTe2. ARPES band-maps along the high symmetry direc- tions of a Mo0.73W0.27Te2 single crystal, as well as the corresponding plots of their second derivatives, are shown in Figs. 3d through 3h. Figs. 3d and 3f show band-maps, and corresponding second derivative, acquired with a photon energy of 70 eV (kz = 0) along the Y-Γ-Y direc- ) s t i n U . b r A ( y t i s n e n t I 6 T = 833 oC T = 767 oC T = 733 oC T = 700 oC T = 667 oC T = 633 oC T = 600 oC T = 500 oC T = 400 oC T = 300 oC T = 100 oC T = 26 oC 25 26 2 (degrees) 27 FIG. 4. Powder X-ray diffraction as a function of angle for 2H-Mo.95W.05Te2. Notice the disappearance of the peaks associated with the 2H-phase and the emergence of new peaks, above T = 600 ◦C which can be ascribed to the 1T ′-phase. In a certain range of temperatures the coexistence of Bragg reflections associated to both phases results from the coexistence of domains and indicates a first-order phase-transition. tion. Figs. 3g and 3h correspond to band maps and sec- ond derivatives collected along the T-Z-T direction with a photon energy of 57 eV corresponding to (kz = π). The remarkable features near EF are the flat hole-like band crossing EF around kk ∼ 0.2 A−1, and an electron-like pocket in the vicinity of kk ∼ 0.4 A−1. The band con- necting the hole- and the electron-like pockets is assigned to a surface state, which have already been claimed to be topologically nontrivial28,35. When compared to the calculations in Ref. 28 the conduction band minimum is observed to be very close to the Fermi level, which makes this surface state not as easily detectable as one would expect from the calculations. The surface state is more clearly exposed in Supplementary Fig. S8. Notice that the bands near EF at kz = 0 have higher binding energies than those at kz = π. As a result, the electron pocket and the surface state become more apparent in Figs. 3d and 3f. Having established the room temperature boundary between the 2H − and Td− phases and explored the electronic structure of the latter phase, we now turn to the temperature axis of the phase-diagram. Fig- ure 4 shows powder XRD patterns for a sample with x ≃ 0.05, at different temperatures upon heating from room-temperature. Above T = 600 ◦C, the peaks as- sociated with the 2H −phase disappear and new peaks that can be ascribed to the 1T ′−phase appear. Similar studies for different compositions are shown in Supple- mentary Fig. S5. We find that the boundary is situated at T2H−1T ′ ∼ 650◦C with a large, sample dependent un- certainty of the order of ∼ 50◦C previously attributed to variations in the Te stoichiometry4. The variation of T2H−1T ′ as a function of x remains within this uncer- tainty, therefore the boundary should be considered as doping independent. We do not see evidence for an ex- tended region in temperature where both phases would coexist4. The 1T ′−MoTe2 phase continues to display a good degree of crystallinity at high T s indicating that the structural transition is not driven by an increase in the number of Te vacancies or material degradation. Monoclinic (1T ') ) Orthorhombic (Td 800 600 200 ) C 400 o ( T ) H 2 ( c i t a m s i r p l a n o g i r T 0 0.0 0.2 0.4 0.6 0.8 1.0 x FIG. 5. Resulting temperature (T ) as a function of the doping fraction, x, phase-diagram of the Mo1−xWxTe2 series. Overall phase-diagram based on the array of exper- imental techniques used for this study. The phase-boundary between the 2H−, 1T ′ and the Td phases were determined through powder X−ray diffraction measurements as a func- tion of T and x up to x = 0.13. Above 800 ◦C the samples decompose or lose their crystallinity. The proposed phase diagram, shown in Figure 5, de- picts a sharp phase-boundary between the 2H − and the Td phases at xc ∼ 0.08, and the boundary between the 2H − and 1T ′− phases at ∼ 650◦C. In Supplemen- tary Fig. S9 we compare X −ray powder diffraction data among samples crystallizing in the 2H − and in the Td−phases and the role of the temperature. The im- portant point is that, in contrast to the 2H −phase, and even for samples with a W concentration very close to the critical one, we could not detect a structural phase- transition as a function of T in samples crystallizing in the Td−phase. Given that the orthorhombic Td−phase becomes the ground state of 1T ′−MoTe2 and the larger area occupied by it in the phase diagram, one is led to conclude that it is thermodynamically more stable than the latter phase. The most remarkable feature of the phase diagram is the very small concentration in W required to stabilize the orthorhombic semi-metallic Td−phase, and not the coexistence of the 2H − and the 1T ′− phases as predicted by Ref. 24, through a sharp boundary situated at x = 0.08 ± 0.01. Such a sharp boundary points to a first-order phase-transition as a function of doping with the caveat that we could not detect phase coexistence. It is quite remarkable that a semiconducting band gap as large as ∼ 1 eV5 for 2H −MoTe2, can be entirely sup- pressed by substituting just xc ≃ 9 ± 1 % of Mo for W which stabilizes a semimetallic state, as clearly indicated by angle resolved photoemission experiments. Here, the situation bears a certain resemblance with the transi- tion metal oxides such as the cuprates, whose charge- or Mott-gap is estimated to be ∆ ∼ 2 eV, but where a small concentration of dopants, in the order of 5 %, is enough to stabilize a metallic state (albeit anoma- lous) and even superconductivity37. This clearly indi- cates that both the structural and the concomitant elec- tronic phases of MoTe2 are particularly susceptible to small perturbations. This suggests that it should be pos- sible to reversibly induce structural-transitions through the application of strain24 or an electric field14, particu- larly in 2H −Mo1−xWxTe2 crystals with x . xc. This would make the 2H −Mo1−xWxTe2 series particularly appealing for the development of phase-change memory devices10,14,15,24 or for a new generation of optoelectronic devices, whose metallic interconnects could be created or "erased" at will through the application of an electrical signal, instead of a chemical treatment. Finally, the fact that the Mo1−xWxTe2 series produce homogeneous al- loys, is not only a major result of this study, but opens the unique possibility of exploring the evolution of their predicted topological/electronic properties28, and of per- haps detecting topological phase-transitions in the bulk as well as in the surface state through the evolution of the Fermi surface. In effect, ARPES indicates that the Fermi surfaces of Td−MoTe2 (Ref. 29) and of Td−Mo1−xWxTe2 (this work) are remarkably different, a fact that can only be reconciled with an electronic/topological phase- transition as a function of W doping. In effect, since W 7 doping tends to stabilize the Td-phase, it is reasonable to expect that one can stabilize it also in samples contain- ing small amounts of W by quickly cooling the crystals to room temperature during the synthesis process. This would produce a phase-diagram not containing the re- gion originally occupied by the 2H −phase. This set of orthorhombic samples would allow us to explore the evo- lution of the Fermi surface as a function of W doping to understand, for example, how the large hole-pocket seen by ARPES at the center of the Brillouin zone29 dis- appears to originate the hole-pockets seen by us at ei- ther side of zone center. Such electronic phase-transition should lead to either the suppression or the displacement of the Weyl-points already seen by ARPES, or to a con- comitant topological phase-transition as a function of W doping. METHODS Sample synthesis For the synthesis of pure 2H − and 1T ′−MoTe2 as well as WTe2 Mo, 99.9999%, W, 99.9999 % and Te 99.9999 % were placed in a quartz ampoule in a ratio of 1:25 for growth in a Te flux. Subsequently, the material was heated up to 1050◦C and held there for 1 day. Then, the ampoule was slowly cooled down to 525◦C to yield ei- ther Td−WTe2 or 2H −MoTe2 and then centrifuged. To produce the 1T ′−MoTe2 phase crystals were centrifuged at 900◦C. The "as harvested" single crystals were subse- quently annealed for a few days at a temperature gradient to remove the excess Te. For Mo1−xWxTe2, single crys- tals were synthesized through a chemical vapor transport technique using iodine or TeCl4 as the transport agent. Samples were held at 750 ◦C with a 100 ◦C temperature gradient for 1 week, then subsequently cooled over 3 days to 400 ◦C and removed from the furnace. Each growth commonly yielded crystals of both structure types (Td− and 2H −), except for those crystals very rich in Mo, i.e. x < 0.05. Stoichiometry was determined by energy dispersive X −ray spectroscopy (EDS) analysis through a field-emission scanning electron microscope (FEI Nova 400). A more precise stoichiometric determination was achieved using X-ray photoelectron spectroscopy (XPS) either at the Shanghai Synchrotron Radiation Facility or at the Stanford Synchrotron Radiation Lightsource. Scanning transmission electron microscopy For scanning transmission electron microscopy imag- ing we used a JEOL 2200FS spherical aberration cor- rected tool operated under 200 kV. When using a 25.6 mrad convergence angle our probe size was 0.9 A. Al- though 200 kV is most likely above the sample damage threshold, we used limited acquisition times and beam exposure to minimize the possible changes to the sam- ple structure. Micro exfoliated few-layered samples were transferred onto TEM grids via a dry transfer method using polypropylene carbonate. X-ray diffraction as a function of the temperature 8 Scanning tunneling microscopy Scanning tunneling microscopy (STM) measurements were performed with a home built variable temperature, ultra high vacuum STM system at T = 82 K. Single crys- talline Mo1−xWxTe2 was mounted onto metallic sample holders using a vacuum safe silver paste. Samples were transferred into the STM chamber and cleaved in-situ to expose a clean surface on which measurements were per- formed. The Pt-Ir STM tip was cleaned and calibrated against a gold (111) single crystal prior to the measure- ments. Powder samples of Mo1−xWxTe2 were prepared by sonicating chemical vapor transport grown bulk crystals in hexane. The Mo1−xWxTe2 dispersion was then drop cast onto c-axis sapphire substrates. This preparation led to highly textured powders, with the c-axis of the sample roughly aligned with the substrate surface normal. Heat- ing X-ray diffraction measurements were carried out in a PANalytical XPert 2 diffractometer with an Anton-Paar domed hot stage that was purged with ultra pure nitro- gen. X-rays were generated from a copper target, with the Cu Kβ radiation removed by using a nickel filter. Several samples were also prepared via exfoliation of bulk crystals onto sapphire substrates. The phase-transition temperature was found to be independent on the method used. Angle resolved photoemission spectroscopy ARPES measurements were performed at the Dream- line beamline of the Shanghai Synchrotron Radiation Fa- cility with a Scienta D80 analyzer. The energy and angu- lar resolutions were set to 15 meV and 0.2◦, respectively. The ARPES data were collected using horizontally- polarized light with a vertical analyzer slit. The sam- ples were cleaved in situ and measured at T = 40 K in a vacuum better than 5 × 10−11 Torr. The cleaved sur- faces are observed to be flat at a scale > 100 µm while the beam spot size of the incident light is 30 × 20 µm2, therefore the electronic structure probed by us is from a single domain. ∗ balicas@magnet.fsu.edu 1 Revolins, E. & Beerntse D. J. Electrical properties of α- and β-MoTe2 as affected by stoichiometry and preparation temperature. J. Phys. Chem. Solids 27, 523-526 (1966). 2 Vellinga, M. B., de Jonge, R., & Haas, C. Semiconductor to metal transition in MoTe2 J. Solid State Chem. 2, 299-302 (1970). 3 Voiry, D. Mohite, A. & Chhowalla, M. Phase engineering of transition metal dichalcogenides. Chem. Soc. Rev. 44, 2702-2712 (2015). 4 Keum, D. H., et al. Bandgap opening in few-layered mon- oclinic MoTe2. Nature Phys. 11, 482-486 (2015). 5 Ruppert, C. Aslan, O. B. & Heinz, T. F. Optical Proper- ties and Band Gap of Single- and Few-Layer MoTe2 Crys- tals. Nano Lett. 14, 62316236 (2014). 6 Soluyanov, A. A. et al. A New Type of Weyl Semimetals. Nature 527, 495-498 (2015). 7 Sun, Y. Wu, S.C. Ali, M. N. Felser, C. & Yan, B. Pre- diction of the Weyl semimetal in the orthorhombic MoTe2. Phys. Rev. B 92, 161107 (2015). 8 Wang, Z. et al. MoTe2: A type-II Weyl Topological Metal. Phys. Rev. Lett. 117, 056805 (2016). 9 Qian, X. F. Liu, J. W. Fu, L. & Li, J. Quantum spin Hall effect in two-dimensional transition metal dichalcogenides. Science 346, 1344 (2014). 10 Duerloo, K. A. N. Li, Y. Reed, E. J. Structural phase transitions in two-dimensional Mo- and W-dichalcogenide monolayers. Nat. Commun. 5, 4214 (2014). 11 Duerloo, K. A. N. & Reed, E. J. Structural Phase Transi- tions by Design in Monolayer Alloys. ACS Nano 10, 289- 297 (2016). 12 Kappera, R. et al. Phase-engineered low-resistance con- tacts for ultrathin MoS2 transistors. Nature Mater. 13, 1128-1134 (2014). 13 Cho, S. et al. Phase patterning for ohmic homojunction contact in MoTe2. Science 349, 625-628 (2015). 14 Li, Y., Duerloo, K.-A. N., Wauson, K. & Reed, E. J. Struc- tural semiconductor-to-semimetal phase transition in two- dimensional materials induced by electrostatic gating. Nat. Commun. 7, 10671 (2016). 15 Zhang, C. et al. Charge Mediated Reversible MetalInsula- tor Transition in Monolayer MoTe2 and WxMo1xTe2 Alloy. ACS Nano 10, 73707375 (2016). 16 Pletikosi´c, I., Ali, M. N., Fedorov, A. V. Cava, R. J. & Valla, T. Electronic Structure Basis for the Extraordinary Magnetoresistance in WTe2, Phys. Rev. Lett. 113, 216601 (2014). 17 Lezama, I. G. Ubaldini, A. Longobardi, M. Giannini, E. Renner, C. Kuzmenko, A. B. Morpurgo, A. F. Surface transport and band gap structure of exfoliated 2H-MoTe2 crystals. 2D Materials 1, 021002 (2014). 18 Lezama, I. G. Arora, A. Ubaldini, A. Barreteau, C. Gian- nini, E. Potemski, M. Morpurgo, A. F. Indirect-to-Direct Band Gap Crossover in Few-Layer MoTe2. Nano Lett. 15, 2336-2342 (2015). 19 Pradhan, N. R. et al. Field-Effect Transistors Based on Few-Layered alpha-MoTe2. ACS Nano 8, 5911-5920 (2014). 20 Lin, Y. F. et al. Ambipolar MoTe2 Transistors and Their Applications in Logic Circuits. Adv. Mater. 26, 32633269 (2014). 21 Mak, K. F. & Shan, J. Photonics and optoelectronics of 2D semiconductor transition metal dichalcogenides. Nature Photon. 10, 216-226 (2016). 22 Park, J. C. et al. Phase-Engineered Synthesis of Centimeter-Scale 1T ′- and 2H-Molybdenum Ditelluride Thin Films. ACS Nano 9, 6548-6554 (2015). 23 Zhang, W. Chiu, M. H. Chen, C. H. Chen, W. Li, L. J. & Wee, A. T. S. Role of Metal Contacts in High-Performance Phototransistors Based on WSe2 Monolayers. ACS Nano 8, 8653-8661 (2014). 24 Alexander, K.-, Duerloo N. & Reed, E. J. Structural Phase Transitions by Design in Monolayer Alloys. ACS Nano 10, 289-297 (2016). 25 Zhang, C., KC, S., Nie, Y., Liang, C., Vandenberghe, W. G., Longo, R. C., Zheng, Y., Kong, F., Hong, S., Wal- lace, R. M. & Cho, K. Charge Mediated Reversible Metal- Insulator Transition in Monolayer MoTe2 and WxMo1xTe2 Alloy, ACS Nano 10, 7370-7375 (2016). 26 Weng, H. M. Fang, C. Fang, Z. Bernevig, B. A. & Dai, X. Weyl Semimetal Phase in Noncentrosymmetric Transition- Metal Monophosphides. Phys. Rev. X 5, 011029 (2015) 27 Xu, S. Y. et al. Discovery of a Weyl fermion semimetal and topological Fermi arcs. Science 349, 613-617 (2015). 28 Chang, T. R. et al. Prediction of an arc-tunable Weyl Fermion metallic state in MoxW1−xTe2, Nat. Commun. 7, 10639 (2016). 29 Huang, L. et al. Spectroscopic evidence for a type semimetallic state in MoTe2. Nat. Mater. II Weyl doi:10.1038/nmat4685 (2016). 30 Champion, J. A. Some propertities of (Mo,W)(Se,Te)2. Brit. J. Appl. Phys. 16, 1035 (1965). 31 Yamamoto, M. et al. Strong Enhancement of Raman Scat- tering from a Bulk-Inactive Vibrational Mode in Few-Layer MoTe2. ACS Nano 8, 3895-3903 (2014). 32 Jiang, Y. C. et al. Raman fingerprint for semimetal WTe2 evolving from bulk to monolayer. Sci. Rep. 6, 19624 (2016). 33 Chastain, J., & King, R. C. Eds. Handbook of X-ray pho- toelectron spectroscopy: a reference book of standard spec- tra for identification and interpretation of XPS data. Eden Prairie, MN: Physical Electronics, 1995. 34 Pletikosic, I. Ali, M. N., Fedorov, A. V., Cava, R. J., & Valla, T. Electronic Structure Basis for the Extraordinary Magnetoresistance in WTe2. Phys. Rev. Lett. 113, 216601 (2014). 35 Belopolski, I. et al. Fermi arc electronic structure and Chern numbers in the type-II Weyl semimetal candidate MoxW1xTe2. Phys. Rev. B 94, 085127 (2016). 36 Rhodes, D. et al. Impurity dependent superconductivity, Berry phase and bulk Fermi surface of the Weyl type-II semimetal candidate MoTe2. arXiv:1605.09065 (2016). 37 Lee, P. A., Nagaosa, N. & Wen, X.-G. Doping a Mott In- sulator: Physics of High Temperature Superconductivity. Rev. Mod. Phys. 78, 17-85 (2006). 9 ACKNOWLEDGMENTS †These authors contributed equally to this work. The subsequent order of authorship do not reflect the relative importance among the contributions from the different authors and groups. Their contributions to this work should be considered of equal relevance. L. B. is sup- ported by the U.S. Army Research Office MURI Grant W911NF-11-1-0362. This work was supported in part by the Molecular and Electronic Nanostructures theme of the Beckman Institute at UIUC. Electron microscopy work was performed at the Frederick Seitz Materials Re- search Laboratory Central Research Facilities, University of Illinois. Single-crystal X-ray diffraction was performed in the Shared Materials Characterization Laboratory at Columbia University. AML acknowledges support by the U.S. Department of Energy, Basic Energy Sciences, Materials Sciences and Engineering Division. The work of R.M.O., J.I.D., W.J., and Y.L. was financially sup- ported by the U.S. Department of Energy under Contract No. DE-FG 02-04-ER-46157. F.E. gratefully acknowl- edges Grant LPDS 2013-13 from the German National Academy of Sciences Leopoldina. This work was also supported by the DOE-BES, Materials Sciences and En- gineering Division, under Contract DE-AC02-76SF00515 and by the W. M. Keck Foundation and the Gordon and Betty Moore Foundations EPiQS Initiative through Grant No. GBMF4545. D.C., N.F., A.A. and J.H. acknowledge support from AFOSR grant FA9550-14-1- 0268. R.L. and S.C.W. were supported by the National Natural Science Foundation of China (No. 11274381). J.Z.M., T.Q., and H.D. were supported by the Ministry of Science and Technology of China (No. 2015CB921300, No. 2013CB921700), the National Natural Science Foun- dation of China (No. 11474340, No. 11234014), and the Chinese Academy of Sciences (No. XDB07000000). STM work is supported by AFOSR (FA9550-11-1-0010, DE) and NSF (DMR-1610110, ANP). This research used resources of (XPS at) the Center for Functional Nano- materials, which is a U.S. DOE Office of Science Facil- ity, at Brookhaven National Laboratory under Contract No. de-sc0012704. The NHMFL is supported by NSF through NSF-DMR-1157490 and the State of Florida. AUTHOR CONTRIBUTIONS D.R. synthesized and characterized the single crystals through electron dispersive spectroscopy, Raman scat- tering and transport measurements. Y.-c.C., W.Z., and Q.R.Z. were directly involved in the synthesis and in the preliminary characterization of the crystals. D.C., T.K., A.M., M.C., M.D., N.F., A.A., and J.H. moti- vated the project, performed Raman, photoluminescense and device characterization. B.E.J. and P.Y.H. per- formed atomic resolution STEM transmission electron microscopy. W.J., J.Z.M., R.L., T.S., S.C.W., T.Q., H.D., J.I.D., and R.M.O. performed angle-resolved pho- mation is available online at www.nature.com/reprints. Correspondence and requests for materials should be ad- dressed to L.B. 10 toemission spectroscopy measurements. Y.L., X.T., T.S., J.I.D., and R.M.O. performed x-ray photoemission spec- troscopy measurements. D.E. and A.N.P. performed scanning tunneling spectroscopy measurements. C.N., E.M., F.E., and A.M.L. performed powder X-ray and electron diffraction measurements as function of the temperature. D.W.P. Performed single crystal X −ray diffraction. D. R., J. H. and L. B. wrote the manuscript with the input of all co-authors. ADDITIONAL INFORMATION COMPETING FINANCIAL INTERESTS Supplementary information is available in the online version of the paper. Reprints and permissions infor- The authors declare no competing financial interests. 

1109.6142

1

1109

Electronic States and Local Density of States in Graphene with a Corner Edge Structure

We study electronic states of semi-infinite graphene with a corner edge, focusing on the stability of edge localized states at zero energy. The 60{\deg}, 90{\deg}, 120{\deg} and 150{\deg} corner edges are examined. The 60{\deg} and 120{\deg} corner edges consist of two zigzag edges, while 90{\deg} and 150{\deg} corner edges consist of one zigzag edge and one armchair edge. We numerically obtain the local density of states (LDOS) on the basis of a nearest-neighbor tight-binding model by using Haydock's recursion method. We show that edge localized states appear along a zigzag edge of each corner edge structure except for the 120{\deg} case. To provide insight into this behavior, we analyze electronic states at zero energy within the framework of an effective mass equation. The result of this analysis is consistent with the behavior of the LDOS.

cond-mat.mes-hall

cond-mat

Typeset with jpsj3.cls <ver.1.0> Full Paper Electronic States and Local Density of States in Graphene with a Corner Edge Structure Yuji Shimomura, Yositake Takane, and Katsunori Wakabayashi1,2 Department of Quantum Matter, Graduate School of Advanced Sciences of Matter, Hiroshima University, Higashihiroshima, Hiroshima 739-8530, Japan 1International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science (NIMS), Namiki, Tsukuba 305-0044, Japan 2PRESTO, Japan Science and Technology Agency, Kawaguchi 332-0012, Japan (Received ) We study electronic states of semi-infinite graphene with a corner edge, focusing on the stability of edge localized states at zero energy. The 60◦, 90◦, 120◦ and 150◦ corner edges are examined. The 60◦ and 120◦ corner edges consist of two zigzag edges, while 90◦ and 150◦ corner edges consist of one zigzag edge and one armchair edge. We numerically obtain the local density of states (LDOS) on the basis of a nearest-neighbor tight-binding model by using Haydock’s recursion method. We show that edge localized states appear along a zigzag edge of each corner edge structure except for the 120◦ case. To provide insight into this behavior, we analyze electronic states at zero energy within the framework of an effective mass equation. The result of this analysis is consistent with the behavior of the LDOS. KEYWORDS: graphene corner edge, localized state, zigzag edge, armchair edge, tunneling spectroscopy 1. Introduction The realization of a monolayer graphene sheet1, 2) has triggered extensive studies on its unusual electronic properties arising from the two-dimensional honeycomb structure of carbon atoms.3) Since the unit cell of the honeycomb lattice contains two nonequivalent sites which form two sublattices A and B, the low-energy electronic states of graphene near the Fermi energy are described by a 2 × 2 matrix form which is equivalent to the massless Dirac equa- tion.4) Thus, electrons in graphene are called massless Dirac fermions. The band structure of massless Dirac fermions has a unique character, since they have linear energy dispersion in the vicinity of two nonequivalent symmetric points, called K + and K− points, in the Bril- louin zone, where the conduction and valence bands conically touch.5) This structure is called Dirac cone. We hereafter set the electron energy at the band touching point as ε = 0. The unique energy band structure provide a number of intriguing physical properties such as the half-integer quantum Hall effect,2, 6) the absence of backward scattering associated with the Berry’s phase by π7) and Klein tunneling.8) The presence of edges makes an strong impact on the Dirac fermions in graphene near 1/21 J. Phys. Soc. Jpn. Full Paper (a) (b) (c) (d) (e) Fig. 1. Typical corner edge structures with coner angles of (a)30◦, (b)60◦, (c)90◦, (d)120◦ and (e)150◦. the Fermi energy. As stressed by Fujita et al., the electronic states near the graphene edge strongly depends on its edge orientation.9) Typical straight edges of graphene are classified into two structures: one is zigzag (zz) edge and the other is armchair (ac) edge. Fujita et al. analyzed electronic states in graphene with an infinitely long straight edge on the basis of a nearest-neighbor tight-binding model, and showed that highly degenerate edge localized states appear at ε = 0 along a zz edge.9) These states at ε = 0 result in a sharp zero-energy peak structure in the local density of states (LDOS) near a straight zz edge of graphene. The edge localized states have a characteristic feature that their probability amplitude is finite only on one sublattice including edge sites and completely vanishes on the other sublattice. No such localized states appear along an ac edge. The presence of edge localized states along a zz edge has been confirmed by using scanning tunneling microscopy and scanning tunneling spectroscopy.10, 11) Theoretically, the presence or absence of zero-energy localized states has been well under- stood for infinitely long straight edges. However, actual edges of graphene samples are never straight nor infinitely long, and are much more complex than ideal ones. An actual edge line consists of several zz and/or ac segments, and a corner edge inevitably appears at the bound- ary of two adjacent segments. Typical corner edge structures are shown in Fig. 1. Hereafter each corner edge is referred to according to its corner angle. The 30◦, 90◦, and 150◦ corner edges consist of one zz edge and one ac edge, while the 60◦ and 120◦ corner edges consist of two zz edges. There arises a natural question: Do edge localized states exist at ε = 0 along a bent edge of these corner edge structures? In this paper we study electronic states in the cor- ner edge structures to answer this question. We adopt a nearest-neighbor tight-binding model and numerically obtain the LDOS by using Haydock’s recursion method. We find that edge 2/21 J. Phys. Soc. Jpn. Full Paper localized states appear along a zz edge of each corner edge structure with an exception of the 120◦ corner edge. In the 120◦ case, edge localized states locally disappear near the corner but emerge with increasing the distance from the corner. To provide insight into these unexpected behaviors, we analyze electronic states at ε = 0 within the framework of an effective mass equation. The result of this analysis is consistent with the behavior of the LDOS. 2. Formulations for Numerical Analysis 2.1 Model of graphene corner edges We describe π electrons in graphene with a corner edge structure by using a tight-binding model on a honeycomb lattice. The Hamiltonian of this model is represented as H = −t X<i,j> iihj +Xi wiiihi, (1) where t is the nearest neighbor hopping integral and wi is a site-dependent potential. If wi = 0 for any i, this model corresponds to a bulk graphene sheet. The site-dependent potential wi is introduced for a techinical reason. For practical application of our numerical approach, it is convenient to treat a lattice system being infinite in both the longitudinal and transverse directions. However, such a system contains lattice sites which are irrelevant for a corner edge structure. To model a corner edge on this infinite system, we put a large on-site potential on each irrelevant site to prevent electrons arriving on it. Therefore, we set wi = w with a sufficiently large w if the ith site is irrelevent for a corner edge structure while wi = 0 otherwise. We consider four corner edges having corner angles differ from each other. The angles are 60◦, 90◦, 120◦ and 150◦. We particularly focus on corner edges including one or two zz edges. 2.2 Haydock’s recursion method The LDOS can be calculated with Haydock’s recursion method12–15) which in applicable to systems having no translational symmetry such as graphene with a corner edge. By applying this method, we can obtain the LDOS at an arbitrary site. We outline the method to obtain the LDOS at an ith site. To start with, we transform our model to a one-dimensional chain model. We first introduce the coefficient a0 given by with l0i ≡ ii, and define l1i and b1 in terms of a0 = hl0Hl0i b1l1i = (H − a0)l0i with hl1l1i ≡ 1. The coefficient b1 is obtained as b1 =phl0(H − a0)(H − a0)l0i. 3/21 (2) (3) (4) J. Phys. Soc. Jpn. We next introduce a1 given by Full Paper a1 = hl1Hl1i, and define l2i and b2 in terms of b2l2i = (H − a1)l1i − b1l0i with hl2l2i ≡ 1. The coefficient b2 is obtained as Repeating this n times, we obtain b2 =p{hl1(H − a1) − hl0b1}{(H − a1)l1i − b1l0i}. with bn+1ln+1i = (H − an)lni − bnln−1i, an = hlnHlni, (5) (6) (7) (8) (9) (10)  b2 a2 b3 b3 a3 . . .  bn+1 =p{hln(H − an) − hln−1bn}{(H − an)lni − bnln−1i}. This manipulation with the reccurence equation, eq. (8), is equivalent to a transformation of the original electron system to a one-dimentinal chain model. {l0i,l1i,l2i, . . .} stands for the orthonormal basis set of the chain model. Here, lni involves neighboring sites of ii up to the nth nearest neighbors. On this basis, H can be rewritten with real coefficients {a0, a1, . . .} and {b1, b2, . . .} as a tridiagonal matrix  a0  b1 b1 a1 b2 H = . (11) With the coefficients {a0, a1, . . .} and {b1, b2, . . .}, the Green’s function Gi(E) for the ith site can be represented as a continued fraction, Gi(E) = 1 E − a0 − . b2 2 ... b2 1 E−a1− (12) Practically, we need to terminate this continued fraction at a sufficiently large n. If it is terminated at n = N , we obtain the approximate expression of Gi(E) as where Gi(E) = 1 E − a0 − E−a1− , b2 1 b2 2... E−aN −t(E) t(E) = E − aN 2b2 N (cid:20)1 − {1 − 4b2 N (E − aN )2} 1 2(cid:21) . 4/21 (13) (14) J. Phys. Soc. Jpn. Gi(E) gives the LDOS at the ith site in terms of the relation Ni(E) = 1 π ImGi(E − iδ), Full Paper (15) where δ is a positive infinitesimal. In actual numerical calculations, we treat δ as a sufficiently small but finite constant. 3. The LDOS 3.1 The LDOS in the presence of a single edge To confirm the validity of our approach using the recursion method, we calculate the LDOS in the presence of an ideal single zz or ac edge. We set N = 1000, w/t = 300 and δ/t = 0.01 throughout this paper. We first consider the case with a single zz edge. The site indices in the unit cell are given in Fig. 2(a). We display the LDOS at the sites 1, 2, 3, and 4 in Fig. 2(b)-(e). A peak at ε = 0 exists at the site 1 on the zz edge. The LDOS also possesses a zero-energy peak at sites on the sublattice which includes the site 1. We see that the peak decays with increasing the distance from the edge. At the sites belonging to the other sublattice, such as the site 2, a peak does not appear at ε = 0. These results are consistent with the presence of edge states at ε = 0. The decay of the zero-energy peak reflects the fact that an edge state has a finite penetration depth. We next consider the case with a single ac edge. Figure 3 shows the LDOS in the presence of a single ac edge. We do not observe a peak of the LDOS at ε = 0. This is consistent with the absence of edge state in the single ac edge case. 3.2 The LDOS in the presence of a corner edge (i) 60◦ corner edge.- Figure 4 shows the LDOS at several sites in the presence of the 60◦ corner edge consisting of two zz edges. From this figure, we can see the appearance of edge states at ε = 0. As shown in Fig. 4(b) and (e), a zero-energy peak exists at the sites 7 and 10 belonging to a same sublattice. Let us compare the LDOS at the site 10 (Fig. 4(e)) with that at the site 4 in the single zz edge case (Fig. 2(e)). Note that the distance from the zz edge to the site of our interest is equivalent in both the cases. We observe that the peak of the LDOS at the site 10 is higher than that at the site 4 in the single zz edge case. We consider that this enhancement of the zero-energy peak at the site 10 is caused by a superposition of edge states at one zz edge and those at the other edge, i.e. constructive interference between two edge states. (ii) 90◦ corner edge.- Figure 5 shows the LDOS at several sites in the presence of the 90◦ corner edge. From this figure, we see that edge states appear at ε = 0. As shown in Fig. 5(b), (d), and (e), a zero-energy peak exists at the sites 11, 13 and 14 belonging to a same sublattice. Thus, the LDOS near the corner possesses both the character of the LDOS in the single zz edge case and that in the single ac edge case. Let us focus on the LDOS at the site 5/21 J. Phys. Soc. Jpn. Full Paper (a) 4 1 3 2 (b) 1 (d) 3 (c) 2 (e) 4 Fig. 2. (a) The strcture of a single zz edge. A broken line represents a unit cell. (b), (c), (d) and (e) display the LDOS at the site 1, site 2, site 3, and site 4, respectively. The number indicated above each graph represents the site number defined in (a). 13 for example. The site 13 corresponds to the site 3 in the zz edge case (Fig. 2(d)) and the site 5 in the single ac edge case (Fig. 3(b)). Roughly speaking, we can regard that the LDOS at the site 13 (Fig. 5(d)) is a mixture of the LDOS at the site 3 (Fig. 2(d)) and the site 5 (Fig. 3(b)). The nature similar to this is also observed at other sites. There is no enhancement of the zero-energy peak of LDOS in contrast to the 60◦ case. (iii) 120◦ corner edge.- Figure 6 shows the LDOS at several sites in the presence of the 120◦ 6/21 J. Phys. Soc. Jpn. Full Paper (a) 5 5 6 6 (b) 5 (c) 6 Fig. 3. (a) The strcture of a single ac edge. A broken line represents a unit cell. (b) and (c) display the LDOS at the site 5 and site 6, respectively. The number indicated above each graph represents the site number defined in (a). corner consisting of two zz edges. In this case, peculiar features arise. The LDOS at the site 15 (Fig. 6(b)) is quite different from that of the site 1 in the single zz edge case (Fig. 2(b)). As seen from Fig. 6(b), (c), and (g), there is no zero-energy peak at the sites near the corner and hence edge states locally disappear. However, edge states appear at the sites away from the corner. Indeed we observe a broad peak at the site 17 (Fig. 6(d)), and the LDOS at the site 18 (Fig. 6(e)) shows a sharp peak. (iv) 150◦ corner edge.- Figure 7 shows the LDOS at several sites in the presence of the 150◦ corner edge. As shown in Fig. 7(b), (d), and (e), a zero-energy peak exists at the sites 21, 23 and 24 belonging to a same sublattice. This indicates the existence of edge states. As in the 90◦ case, the LDOS near the corner possesses both the character of the LDOS in the single zz edge case and that in the single ac edge case. For example, we can regard that the LDOS 7/21 J. Phys. Soc. Jpn. Full Paper (a) 10 7 8 9 7 9 (b) 7 (d) 9 (c) 8 (e) 10 Fig. 4. (a) The strcture of the 60◦ corner edge. (b), (c), (d) and (e) display the LDOS at the site 7, site 8, site 9, and site 10, respectively. The number indicated above each graph represents the site number defined in (a). at the site 24 (Fig. 7(e)) is a mixture of the LDOS at the site 4 on the single zz edge (Fig. 2(e)) and that at the site 6 on the single ac edge (Fig. 3(c)). The peculiarity of the 150◦ case is that the zero-energy peak of the site 21 is quite smaller than that at the site 1 on the single zz edge (Fig. 2(b)). 8/21 J. Phys. Soc. Jpn. Full Paper (a) 14 11 13 12 (b) 11 (d) 13 (c) 12 (e) 14 Fig. 5. (a) The strcture of the 90◦ corner edge. (b), (c), (d) and (e) display the LDOS at the site 11, site 12, site 13, and site 14, respectively. The number indicated above each graph represents the site number defined in (a). 9/21 J. Phys. Soc. Jpn. Full Paper (a) 16 19 15 20 15 20 19 16 17 18 17 (b) 15 (d) 17 (f) 19 18 (c) 16 (e) 18 (g) 20 Fig. 6. (a) The strcture of the 120◦ corner edge. (b), (c), (d), (e), (f), and (g) display the LDOS at the site 15, site 16, site 17, site 18, site 19, and site 20, respectively. The number indicated above each graph represents the site number defined in (a). 10/21 J. Phys. Soc. Jpn. Full Paper (a) 24 23 22 21 (b) 21 (d) 23 (c) 22 (e) 24 Fig. 7. (a) The strcture of the 150◦ corner edge. (b), (c), (d), and (e) display the LDOS at the site 21, site 22, site 23, and site 24, respectively. The number indicated above each graph represents the site number defined in (a). 11/21 J. Phys. Soc. Jpn. Full Paper (a) (b) (c) Fig. 8. The LDOS in the presence of the (a) 60◦, (b) 90◦ and (c) 150◦ corner edges at ε = 0. The radius of open circles indicates the magnitude of the LDOS. Figure 8 represents the spatial dependance of the LDOS at ε = 0 in the presence of the (a) 60◦, (b) 90◦, and (c) 150◦ corner edges. A radius of each open circle indicates the magnitude of the LDOS. In these figures, the LDOS has a finite value only on the sublattice involving zz edge sites but vanishes on the other sublattice. We observe that the LDOS localizes near zz edges, indicating the presence of edge localized states. However, special emphasis is placed on the case of the 60◦ corner edge, where the magnitude of the LDOS at inner sites is larger than that in the other two cases. This reflects the fact that edge localized states are present at both the two zz edges. The overlap of these edge localized states enhances the magnitude of the LDOS, i.e. constructive interference. The other corner edge structures with the angle 90◦ or 150◦ consist of one zz edge and one ac edge. Note the LDOS in the single ac edge system vanishes at ε = 0 on any sites. In these corner edge structures, the LDOS becomes finite even at ε = 0 due to the presence of a zz edge. Even at sites on the ac edge, the LDOS can have a 12/21 J. Phys. Soc. Jpn. Full Paper Fig. 9. The LDOS in the presence of the 120◦ corner edge at ε = 0. The radius of open circles indicates the magnitude of the LDOS. finite value. Figure 9 represents the spatial dependance of the LDOS at ε = 0 in the presence of the 120◦ corner edge. In this figure, we observe that the LDOS vanishes at the sites near the corner, indicating local disappearance of edge states, i.e. destructive interference. In spite of the fact that the 120◦ corner edge consists of two zz edges, the edge states are not fully stabilized in contrast to the case of the 60◦ corner edge. Note that in the 120◦ case, edge sites on one zz edge and those on the other zz edge belong to different sublattices, while all edge sites in the 60◦ case belong to a same sublattice. As we discuss in the next section, this is the reason for the qualitative difference between the two cases. 4. Analytical Treatment Our study on the LDOS reveals that edge localized states are stabilized in corner edge structures except for the 120◦ case. To provide insight into this behavior, we analyze edge localized states in corner edge structures by using an effective mass description, which is applicable to low-energy states in the vicinity of the K± point. The K + and K− points are characterized by K+ = (−4π/3a, 0) and K− = (4π/3a, 0), respectively. Here, a is lattice constant. As shown in Fig. 10 (a), the unit cell of graphene has two non-equivalent carbon atoms A and B which form A sublattice and B sublattice, respectively. We represent the wave function ψA(r) for A sublattice and the wave function ψB(r) for B sublattice as ψA(r) = eiK+·rF + A (r) + eiK −·rF −A (r), ψB(r) = eiK+·rF −B (r) − eiK −·rF −B (r), 13/21 (16) (17) J. Phys. Soc. Jpn. Full Paper B A − K k y + K k x y x a (a) + K Γ − K − K + K (b) Fig. 10. (a) Honeycomb structure of graphene. The region enclosed in a broken line is a unit cell. A and B are non-equivalent sites which form sublattices. (b) The first Brillouin zone of graphene, where K +, K − and Γ are symmetric points. The K + point is located at (−4π/3a, 0) and (2π/3a,±2π/√3a) while the K − point is located at (4π/3a, 0) and (−2π/3a,±2π/√3a). where F ± are envelope functions near the K± point. The envelope functions at energy ε satisfy kx + iky γ 0 0 0   kx − iky 0 0 0 0 0 0 kx − iky 0 0 kx + iky 0     F + A (r) F + B (r) F −A (r) F −B (r)   F + A (r) F + B (r) F −A (r) F −B (r)     = ε , (18) where γ is a band parameter, kx = −i∂/∂x, and ky = −i∂/∂y. This is called k·p equation,4, 16) which is an effective mass equation for graphene systems. For later convenience, we present the envelope functions for edge states at ε = 0.17) Let us consider a semi-infinite graphene which occupies the region of y > 0, and has a zigzag edge at y = 0. Assumng that edge sites belong to A sublattice, we adopt the boundary condition of F ±B (r)y=0 = 0. The envelope functions for edge states are given as F ±A (r) F ±B (r)! = C e±ikxxe−kxy 0 ! , (19) where C is a normalization constant. The absolute value of F ±A in eq. (19) has a maximum value at y = 0 and exponentially decays with increaseing y. This represents edge states local- ized along the zz edge. We construct zero-energy wave functions which satisfy the boundary condition of corner edges by using eq. (19). 14/21 J. Phys. Soc. Jpn. Full Paper Fig. 11. The boundary condition for the 60◦ corner edge requires that wave functions vanish at sites marked with a triangle. A site marked with an open triangle belongs to A sublattice, and sites marked with a filled triangle belongs to B sublattice. (i) 60◦ corner edge.- We first consider wave functions in the presence of the 60◦ corner edge as shown in Fig. 11. We attempt to construct wave functions near the K + point in terms of two edge localized wave functions. One is the wave function for the 0◦ zz edge C e−iKxeikx(x+iy) 0 ! , and the other is the wave function for the 60◦ zz edge C e−iKxeikx(x+iy)ei 2  0 3 π   , where K ≡ 4π/3a. Here and hereafter we refer to zz edge intersecting the x axis with angle θ degree as θ◦ zz edge. We adopt their linear combination (20) (21) (22) (24) ψB(r)! = e−iKx(C1eikx(x+iy) + C2eikx(x+iy)ei 2 ψA(r)  0 3 π )   as a trial wave function in the presence of the 60◦ corner edge. The boundary condition requires that the wave function vanishes at sites marked with triangles in Fig. 11. Because ψB(r) = 0, we need to consider only the boundary condition for ψA(r). Only the site at the corner with an open triangle belongs to A sulattice. We define this site as the origin of the coordinate. Hence, the boundary condition for ψA(r) is simply given by ψA(0, 0) = 0, (23) yielding C2 = −C1. We obtain the wave function in the presence of the 60◦ corner edge as ψB(r)! = C ψA(r)  e−iKx(eikx(x+iy) − eikx(x+iy)ei 2 3 π ) 0   . This indicates the existence of edge states in the 60◦ corner edge.18) (ii) 90◦ corner edge.- Secondly we consider the 90◦ corner edge as shown in Fig. 12. In this 15/21 J. Phys. Soc. Jpn. Full Paper Fig. 12. The boundary condition for 90◦ corner edge requires that wave functions vanish at sites marked with a triangle. Sites marked with an open triangle belongs to A sublattice, and sites marked with a filled triangle belongs to B sublattice. case, states near K + and K− points are mixed due to the presence of an ac edge. We construct zero-energy wave functions by using edge localized wave function near the K + point, and that near the K− point, C e−iKxeikx(x+iy) 0 ! , C eiKxe−ikx(x−iy) 0 ! . We adopt their linear combination ψA(r) ψB(r)! = C3e−iKxeikx(x+iy) + C4eiKxe−ikx(x−iy) 0 ! (25) (26) (27) as a trial wave function. This must vanishes at sites marked with triangles in Fig. 12. Because ψB(r) = 0, we need to consider only the boundary condition for ψA(r). The sites marked with open triangles belong to A sulattice. We define the site at the corner with an open triangle as the origin. The coordinates of the open triangles are (x, y) = (0,√3a×m) with m = 0, 1, 2, . . .. Hence, the boundary condition for ψA(r) reads ψA(0,√3a × m) = 0 (m = 0, 1, 2, . . .). (28) Imposing this condition to ψA in eq. (27), we obtain C4 = −C3. We obtain the wave function for the 90◦ corner edge as ψA(r) ψB(r)! = C e−iKxeikx(x+iy) − eiKxe−ikx(x−iy) 0 ! . (29) This indicates the existence of edge states in the 90◦ corner edge. (iii) 150◦ corner edge.- Thirdly we consider the 150◦ corner edge as shown in Fig. 13. We obtain zero-energy wave functions using a conformal mapping technique.19) In terms of the 16/21 J. Phys. Soc. Jpn. Full Paper Fig. 13. The boundary condition for the 150◦ corner edge requires that wave functions vanish at sites marked with a triangle. Sites marked with an open triangle belongs to A sublattice, and sites marked with a filled triangle belongs to B sublattice. complex variable z ≡ x + iy, eq. (19) is rewritten as F + A (z) F + B (z) F −A (z) F −B (z)     Ceikxz 0 C′e−ikxz∗ 0     = , (30) where z∗ is the complex conjugate of z. Here we introduce the transformation of w = z3/5. This transformation maps a 150◦ corner on z plane to a 90◦ corner on w plane and vice versa. Thus, the wave function for the 90◦ corner edge on w plane ψB(w)! = C e−iKxeikxw − eiKxe−ikw∗ ψA(w) 0 ! is mapped to (31) (32) (33) ψB(r)! = C ψA(r)  = C  e−iKxeikxz 3 5 − eiKxe−ikxz∗ 3 5 0 3   e−iKxeikx(x+iy) 5 − eiKxe−ikx(x−iy) 0 3 5   on z plane. The boundary condition requires that the wave function vanishes at sites marked with triangles in Fig. 13. Again, we need to consider only the boundary condition for ψA(r). The sites marked with open triangles belong to A sulattice. We define the site at the corner with an open triangle as the origin. The coordinates of the open triangles are (x, y) = (− 3 2 a× √3 2 a × m) with m = 0, 1, 2, . . .. Hence, the boundary condition for ψA(r) is given by m, ψA(− 3 2 a × m, √3 2 a × m) = 0 (m = 0, 1, 2, . . .). (34) The wave funciton ψA in eq. (33) satisfies this condition. Therefore eq. (33) can be considered 17/21 J. Phys. Soc. Jpn. Full Paper Fig. 14. The boundary condition for the 120◦ corner edge requires that wave functions vanish at sites marked with a triangle. Sites marked with an open triangle belongs to A sublattice, and sites marked with a filled triangle belongs to B sublattice. as a wave function in the presence of the 150◦ corner edge. This indicates the existence of edge states. The envelope funcions eikx(x+iy) 5 are diffrent from ordinary envelope functions given in eq. (19), but both satisfy the k · p equation given in eq. (18). (iv) 120◦ corner edge.- Lastly we consider the 120◦ corner edge as shown in Fig. 14. The boundary condition requires that ψB(r) vanishes at sites marked with a filled triangle and 5 and e−ikx(x−iy) 3 3 ψA(r) vanishes at sites marked with an open triangle. Therefore, both ψA(r) and ψB(r) are subjected to the boundary condition, in contrast to the 60◦ case where one component is free from the boundary condition. This crucially affects zero-energy states in the 120◦ case as we see below. Similar to the treatment for the 60◦ case, we first adopt a linear combination of the wave function for the 0◦ zz edge and that for the 120◦ zz edge as a trial wave function. The wave function for the 0◦ zz edge near the K + point is C e−iKxeikx(x+iy) 0 ! and the wave function for the 120◦ zz edge near the K + point is C 0 3! . e−iKxe−ikx(x−iy)e−i π (35) (36) The former has only the A-sublattice component, while the latter has only the B-sublattice component. Obviously, their linear combination does not satisfy the boundary condition for both the A-sublattice and B-sulattice components. We next consider a linear combination of the 0◦ edge wave functions near the K + and K− points, ψA(r) ψB(r)! = C5e−iKxeikx(x+iy) + C6eiKxe−ikx(x−iy) 0 ! . (37) This is equivalent to eq. (27). Though ψB of eq. (37) satisfies the boundary condition, ψA cannot satisfy the boundary condition for arbitrary C5 and C6. Finally, we consider a linear 18/21 J. Phys. Soc. Jpn. Full Paper combination of the 0◦ zz edge wave function and an arbitrary evanescent wave function near the K± point given by C e∓iKxe±ip(x±iy)e±iθ 0 ! . (38) This wave function, reducing to the 0◦ zz edge wave function when θ → 0, satisfies eq. (18) and is bounded for 0 ≤ θ ≤ π 3 in the 120◦ case. Their linear combination ψB(r)! = C7e−iKxeikx(x+iy) + C8e∓iKxe±ip(x±iy)e±iθ ψA(r) 0 ! . (39) does not satisfy the A-sublattice boundary condition for arbitrary C7, C8, p, and θ as long as p is sufficiently small. We failed to construct zero-energy wave functions in the 120◦ case in the form of a linear combination of the edge states, in striking contrast to the 60◦ case. It is considered that this corresponds to the disappearance of the LDOS peak at ε = 0 near the corner observed in the numerical result. We suppose that correct zero-energy states consist of zz edge states and complex scattered waves. We point out that the sublattice configuration of two zz edges plays a crucial role in the qualitative difference between the 60◦ and 120◦ cases. In the remaining of this section we briefly consider the behavior of the LDOS shown in Fig. 8, on the basis of the wave functions obtained above. Figure 8 shows the spatial dependence of the LDOS at ε = 0 in the 60◦, 90◦ and 150◦ cases. We observe that the LDOS on a zz edge is slightly suppressed in the close vicinity of a corner. This should be distinguished from the strong suppression of the LDOS observed near a 120◦ corner, and is simply accounted for on the basis of the wave functions for zero-energy edge localized states presented in eqs. (24), (29) and (33). We see that due to destructive interference, the amplitude of these wave functions is suppressed in the close vicinity of a corner located at (x, y) = (0, 0) for a sufficiently small kx. This accounts for the slight suppression of the LDOS. 5. Summary We have studied electronic states in semi-infinite graphene with a corner edge, focusing on the stability of edge localized states. The 60◦, 90◦, 120◦ and 150◦ corner edges are examined. The 90◦ and 150◦ corner edges consist of one zz edge and one ac edge, while the 60◦ and 120◦ corner edges consist of two zz edges. We have numerically obtained the local density of states on the basis of a nearest-neighbor tight-binding model by using Haydock’s recursion method. We have shown that edge localized states appear along a zz edge of each corner edge structure except for the 120◦ case. In the 120◦ case, we have also shown that edge localized states locally disappear near the corner but emerge with increasing the distance from the corner along each zz edge. To provide insight into these behaviors, we have analyzed electronic states at ε = 0 within the framework of an effective mass equation. Except for the 120◦ case, we 19/21 J. Phys. Soc. Jpn. Full Paper have succeeded to obtain eigenstates of the effective mass equation by forming a superposition of pair of edge localized wave functions for an infinitely long straight zz edge. This indicates the existence of edge localized states, and is consistent with the behavior of the local density of states. Contrastingly, no eigenstate has been obtained in such a simple form in the 120◦ case. This suggests a possibility that the local disappearance of edge localized states in the 120◦ case is beyond the effective mass description. Note that although both the 60◦ and 120◦ corner edges consist of two zz edges, zero-energy eigenstates of the effective mass equation are obtained only in the former case. We have pointed out that this reflects the fact that two zz edges belong to a same sublattice in the former case while they belong to different sublattices in the latter case. Acknowledgment This work was supported in part by a Grant-in-Aid for Scientific Research (C) (No. 21540389) from the Japan Society for the Promotion of Science, and by a Grant-in-Aid for Specially promoted Research (No. 20001006) from the Ministry of Education, Culture, Sports, Science and Technology. 20/21 J. Phys. Soc. Jpn. References Full Paper 1) K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubons, I. V. Grigoriva, and A. A. Firsov: Science 306 (2004) 666. 2) K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos and A. A. Firsov Nature 438, (2005) 197. 3) See, for a review, A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim: Rev. Mod. Phys. 81 (2009) 109. 4) J. W. McClure: Phys. Rev. 104 (1956) 666. 5) P. R. Wallace: Phys. Rev. 71 (1947) 622. 6) Y. Zhang, Y.-W. Tan, H. L. Stormer, P. Kim: Nature 438 (2005) 201. 7) T. Ando, T. Nakanishi and R. Saito: J. Phys. Soc. Jpn. 67 (1998) 2857. 8) M. I. Katsnelson, K. S. Novoselov, A. K. Geim: Nat. Phys. 2 (2006) 620. 9) M. Fujita, K. Wakabayashi, K. Nakada, and K. Kusakabe: J. Phys. Soc. Jpn. 65 (1996) 1920. 10) Y. Kobayashi, K. Fukui, T. Enoki, K. Kusakabe, and Y. Kaburagi: Phys. Rev. B 71 (2005) 193406. 11) Y. Niimi, T. Matsui, H. Kambara, K. Tagami, M. Tsukada, and H. Fukuyama: Phys. Rev. B 73 (2006) 085421. 12) R. Haydock: in Solid State Physics , edited by H. Ehrenreich, F. Seitz and D. Turnbull (Academic, New York, 1980), Vol. 35, p. 216. 13) M. J. Kelly: in Solid State Physics , edited by H. Ehrenreich, F. Seitz and D. Turnbull (Academic, New York, 1980), Vol. 35, p. 296. 14) L. C. Davis: Phys. Rev. B 28 (1983) 6961. 15) S. Wu, L. Jing, Q. Li, Q. W. Shi, J. Chen, H. Su, X. Wang, and J. Yang: Phys. Rev. B 77 (2008) 195411. 16) J. C. Slonczewski, P. R. Weiss: Phys. Rev. 109 (1957) 272. 17) K. Wakabayashi: Ph. D. Thesis, University of Tsukuba (2000). 18) M. Ezawa: Phys. Rev. B 81 (2010) 201402(R). 19) C. Iniotakis, S. Graser, T.Dahm, and N. Schopohl: Phys. Rev. B 71 (2005) 214508. 21/21 

1806.09212

1

1806

Single-photon controlled thermospin transport in a resonant ring-cavity system

Cavity-coupled nanoelectric devices hold great promise for quantum technology based on coupling between electron-spins and photons. In this study, we approach the description of these effects through the modeling of a nanodevice using a quantum master equation. We assume a quantum ring is coupled to two external leads with different temperatures and embedded in a cavity with a single photon mode. Thermospin transport of the ring-cavity system is investigated by tuning the Rashba coupling constant and the electron-photon coupling strength. In the absence of the cavity, the temperature gradient of the leads causes a generation of a thermospin transport in the ring system. It is observed that the induced spin polarization has a maximum value at the critical value of the Rashba coupling constant corresponding to the Aharonov-Casher destructive interference, where the thermospin current is efficiently suppressed. Embedded in a photon cavity with the photon energy close to a resonance with the energy spacing between lowest states of the quantum ring, a Rabi splitting in the energy spectrum is observed. Furthermore, photon replica states are formed leading to a reduction in the thermospin current.

cond-mat.mes-hall

cond-mat

Single-photon controlled thermospin transport in a resonant ring-cavity system Nzar Rauf Abdullaha,b,c, Vidar Gudmundssonc aPhysics Department, College of Science, University of Sulaimani, Kurdistan Region, Iraq bKomar Research Center, Komar University of Science and Technology, Sulaimani City, Iraq cScience Institute, University of Iceland, Dunhaga 3, IS-107 Reykjavik, Iceland 8 1 0 2 n u J 4 2 ] l l a h - s e m . t a m - d n o c [ 1 v 2 1 2 9 0 . 6 0 8 1 : v i X r a Abstract Cavity-coupled nanoelectric devices hold great promise for quantum technology based on coupling between electron- spins and photons. In this study, we approach the description of these effects through the modeling of a nanodevice using a quantum master equation. We assume a quantum ring is coupled to two external leads with different temperatures and embedded in a cavity with a single photon mode. Thermospin transport of the ring-cavity system is investigated by tuning the Rashba coupling constant and the electron-photon coupling strength. In the absence of the cavity, the temperature gradient of the leads causes a generation of a thermospin transport in the ring system. It is observed that the induced spin polarization has a maximum value at the critical value of the Rashba coupling constant corresponding to the Aharonov-Casher destructive interference, where the thermospin current is efficiently suppressed. Embedded in a photon cavity with the photon energy close to a resonance with the energy spacing between lowest states of the quantum ring, a Rabi splitting in the energy spectrum is observed. Furthermore, photon replica states are formed leading to a reduction in the thermospin current. Keywords: Thermo-optic effects, Electronic transport in mesoscopic systems, Cavity quantum electrodynamics, Electro-optical effects PACS: 78.20.N-, 73.23.-b, 42.50.Pq, 78.20.Jq 1. Introduction Thermoeletric properties have been so far investigated mainly in nanoscale systems to achieve high thermoelectric efficiency that would be useful for energy harvesting [1, 2]. To obtain a high thermoelectric efficiency, thermoelectri- cally active materials are used. These materials should have high electrical conductivity and low thermal conduc- tivity. High electrical conductivity can be obtained by increasing the carrier mobility or their concentration that can be influenced in quantum structures. Consequently, the figure of merit and Seebeck effect can be enhanced in nanodevices [3, 4, 5]. Generally, there is a challenge in the conventional thermoelectric material because if the elec- trical conductivity is enhanced the thermal conductivity is increased as well. As a result, the Seebeck effect and the device efficiency are decreased. An advantageous method relying on the Spin Seebeck ef- fect has been used to decouple electrical conductivity from thermal conductivity. So the electrical and thermal con- ductivity can be separately controlled concurrently [6, 7]. In 2008, Saitoh and et. al. discovered the spin Seebeck effect when heat is applied to a magnetized metal. In a magnetically active material electrons reconfigure them- selves according to their spin. In this way, unlike in a Email addresses: nzar.r.abdullah@gmail.com (Nzar Rauf Abdullah), vidar@hi.is (Vidar Gudmundsson) conventional electron transport, this rearrangement does not create heat as a waste product. The spin Seebeck ef- fect can lead the way to the growth of smaller, faster and more energy-efficient microchips as well as spintronics de- vices [8, 9, 10]. On the other hand, the influences of light on thermo- electric effects have been investigated and shown that the thermoelectric power can be enhanced by increasing the intensity of light [11]. It was also found that a polarized light can induce a Fano-like resonance in the thermal con- ductance [12]. Therefore, The thermopower and the figure of merit may be enhanced near a Fano-like resonance. In addition, the polarized light and the increase of magnetic polarization may lead to a better thermoelectric perfor- mance, especially, a significant increase of the spin thermal efficiency may be obtained. Influenced by the aforementioned studies, we try to ex- plore the influences of a quantized photon field on ther- mospin transport through a quantum ring including the Rashba spin-orbit coupling. We model a quantum ring system coupled to two leads with different temperatures. The ring system is embedded in a cavity with a linearly polarized photon field. In our previous publications, we have seen that both thermoelectric and heat currents can be controlled by a polarized photon field [13, 14, 15, 16]. The aim of our study here is twofold. First, we induce a thermospin current through a multi-level quantum ring Preprint submitted to Elsevier September 3, 2018 and see the influences of the Rashba spin-orbit coupling on thermospin current using a quantum master equation. Second, we show how the spin-polarization and thermo- spin current can be controlled by a single photon in the cavity. The paper is organized as follows: In Sec. 2, we present the model describing a quantum ring coupled to a photon cavity. Section 3 shows the numerical results and discus- sion. Concluding remarks are addressed in Sec. 4. 2. Model and Theory In this section, we first present the Hamiltonian of the system, and subsequently apply the general master equa- tion formalism to calculate the thermospin-polarized cur- rent. 2.1. Hamiltonian of the system The Hamiltonian of a quantum ring system coupled to a cavity can be expressed as H = He + He−γ + Hγ, (1) where the Hamiltonian of the electronic part He and the electron-photon interaction He−γ together can be defined as He + He−γ = Z d2r Ψ†(r)(cid:20)(cid:18) p2 + HR(r)i Ψ(r) + Hee, 2m∗ + Vr(r)(cid:19) + HZ (2) and the Hamiltonian of the free photon field in the cavity is Hγ = ωγ a†a. (3) Herein, Ψ(r) is the spinor vector [15], and p is the mo- mentum operator of the quantum ring system coupled to the photon cavity which can be written as p(r) =  i ∇ + e c h A(r) + Aγ(r)i , (4) where the vector potential of the external perpendicular magnetic field is A(r) = −By x with B = Bz, and the vector potential of the photons in the cavity is Aγ(r) which can be introduced in terms of the photon creation (a†) and annihilation (a) operators as Aγ = A(ea + e∗a†), (5) with e = ex for the x-polarized and e = ey for the y- polarized photon field [17]. The potential that forms the quantum ring is Vr(r) and HZ is the Zeeman Hamilto- nian [15]. The strength of the vector potential of the pho- tons A is defined by the electron-photon coupling constant gγ = eAΩwaw/c, where Ωw = (ω2 2 is the effective characteristic frequency with Ω0 the frequency of the con- fined electron in the y-direction, ωc is the cyclotron fre- quency and aw the effective magnetic length. c + Ω2 0) 1 Furthermore, HR(r) is the Rashba-spin orbit interaction HR(r) = α  (σx py(r) − σy px(r)) , (6) with α the Rashba spin-orbit (RSO) coupling constant that can be tuned by an external electric field, and σx and σy are the Pauli matrices. The last term of Eq. (2) is Hee which accounts for the electron-electron interaction of the quantum ring system [17, 18]. The Hamiltonian pre- sented in Eq. (1) is used to obtain the energy spectrum of the quantum ring-cavity system using a numerically exact diagonalization technique [19, 20]. 2.2. Transport Formalism To describe the transient electron transport through the quantum ring system, we use a time-convolution-less gen- eralized master equation (TCL-GME) [21, 22, 23]. The TCL-GME is local in time and satisfies the positivity for the many-body state occupation described by the reduced density operator (RDO). The RDO of the system quantum ring system ρS, in terms of the total density matrix ρT , is defined as ρS(t) = Trl(cid:2)ρT (t)(cid:3), (7) where l ∈ {L, R} indicates the two electron reservoirs, the left (L) and the right (R) leads, respectively. In our study, we integrate the GME to the point in time t = 220 ps, late in the transient regime when the total system is approaching the steady state. The RDO is utilized to calculate the spin-polarization and thermospin current. We define the spin polarization Si of the quantum ring system in i = x, y, z direction. Thus, the spin polarization operator is Si = Z d2rni(r), (8) with ni(r) the spin polarization density operator for the spin polarization Si [24]. In addition, the top local ther- mospin current (I th,i ) through the upper arm (y > 0) of the quantum ring system can be introduced as t I th,i t (t) = Z ∞ 0 dy jth,i x (x = 0, y, t) (9) and the bottom local thermospin polarization current (I th,i ) through the lower arm (y < 0) of the quantum ring system b I th,i b (t) = Z 0 −∞ dy jth,i x (x = 0, y, t), (10) where the spin polarization current density is jth,i(r, t) = (cid:18)jth,i x jth,i y (r, t) (r, t)(cid:19) = Trhρs(t)jth,i(r)i (11) calculated by the expectation value of the spin polarization current density operator [24]. 2 Finally, the total local (TL) thermospin polarization current is obtained from the top and the bottom ther- mospin current polarization I th,i tl (t) = I th,i t (t) + I th,i b (t). (12) The TL-thermospin current is related to non-vanishing spin-polarization sources, and the circular local (CL) ther- mospin polarization current is I th,i cl (t) = 1 2hI th,i b (t) − I th,i t (t)i. (13) In the result section, we present the main results on the thermospin transport in the quantum ring system and the influence of the photon field on the quantum ring system. 3. Results In this section the numerical results are shown for a ring-cavity system including the Rashba spin-orbit inter- action and the electron-photon interaction. The quantum ring and the leads are made of a GaAs-based material with electron effective mass m∗ = 0.067me and the relative di- electric material κ = 12.4. An external perpendicular magnetic field with strength B = 10−5 T is applied to the total system including the leads. We assume a very weak external magnetic field to avoid spin degeneracy, and have it weak enough to pre- vent creating a circular motion due to a Lorentz force in the quantum ring system. The main goal of the study here is to show the "real" circular local (CL) thermospin cur- rent due to the Rashba effect in the quantum ring system. Therefore we choose the low strength of the external mag- netic field and tune the Rashba coupling constant. This assumed external magnetic field is out of the Aharonov- Bohm (AB) regime because the area of the ring structure is A = πa2 ≈ 2 × 104 nm2 leading to a magnetic field B0 = φ0/A ≈ 0.2 T corresponding to one flux quantum φ0 = hc/e [22, 25]. The ring system is parabolically confined in the y- direction with characteristic energy Ω0 = 1.0 meV. It gives a broad quantum ring. However, the broad ring ge- ometry together with the spin degree of freedom, and the spin-orbit interactions require a substantial computational effort. Transport properties linked to spin-orbit interac- tions can be clearly realized in a broad quantum ring. It is assumed that the cavity initially contains one pho- ton with linear polarization. It is also considered that the left and the right leads have the same chemical potential (µL = µR = µ), but are at different temperatures, which induce a thermal transport in the quantum ring system. 3.1. Energy spectrum The potential defining the quantum ring system and its energy spectrum are demonstrated here. Figure 1 shows the potential of the ring where the top arm is located in the positive y-axis and the bottom arm is in the negative y- axis. It should be mentioned that the electrons are mainly transferred through the ring system in the x-direction by the thermal bias. Vr (meV) Vr (meV) 15 10 5 0 -4.5 4.5 3 1.5 0 -1.5 y/aw -3 -1.5 0 x/aw 1.5 3 -3 4.5-4.5 Figure 1: (Color online) The potential Vr(r) defining the central ring system that will be coupled diametrically to the semi-infinite left and right leads in the x-direction. The energy spectrum of the quantum ring system as a function of the photon energy is plotted in Fig. 2 where the electron-photon coupling strength is gγ = 0.05 meV and the photon is linearly polarized in the x-direction. The Rashba coupling constant is fixed at α = 14.0 meV nm, which is a critical value of the Rashba coupling constant described later. The energy states around 0.885 meV and 1.112 meV are the one-electron ground state (GS) and the first-excited state (FES), respectively. The aforemen- tioned states are double states due to the spin-orbit inter- action including both spin-up and spin-down states. The energy of the second-excited state (SES) is monotonically increased with increasing photon energy. The energy value of the SES is 1.4 meV at photon energy ωγ = 0.3 meV and it is enhanced to 1.9 meV at ωγ = 1.0 meV. In addi- tion, the SES is in resonant with the first photon replica of the ground state (γGS). The energy value of γGS is 1.2 meV at ωγ = 0.3 meV and it becomes 1.49 meV at ωγ = 1.0 meV. The resonant states, the SES and the γGS, form a Rabi-splitting (RS) in the energy spectrum. The strongest Rabi-effect is recorded at the photon energy ≈ 0.55 meV (black arrows). We study the thermospin transport in the strong Rabi effect regime. Therefore, we fix the photon energy at ωγ = 0.55 meV in our calculations from now on, and investigate the properties of the thermospin transport of the system. The many-electron (ME) energy (a) and the Many-Body (MB) energy (b) for the photon dressed elec- tron states of the quantum ring system versus the RSO- coupling are shown in Fig. 3. In the absence of the cavity (Fig. 3(a)), the energy of the one-electron states decreases with increasing RSO-coupling and crossings of the one- electron states are formed at α ≈ [10 − 15] meV nm. The crossing of the states corresponds to the Aharonov-Casher (AC) destructive phase interference in the quantum ring 3  2 1.8 1.6 1.4 1.2 1 ) V e m ( m u r t c e p s y g r e n e - B M RS SES γGS FES GS 0.8 0.3 0.5 0.7 0.9 Photon energy (meV) Figure 2: (Color online) Many-Body (MB) energy spectrum of the ring system versus the photon energy ωγ . The green rectangles are the zero-electron states (0ES), the red circles indicate the one- electron states (1ES), and the two electron states are located at the upper part of the energy spectrum (now shown). The black arrow indicates the Rabi-splitting (R-S) between the second-excited state and the one-photon replica of the ground state. The Rashba coupling constant is fixed at α = 14.0 meV, the electron-photon coupling strength is gγ = 0.05 meV, and the photon is linearly polarized in the x-direction. The magnetic field is B = 10−5 T, and Ω0 = 1.0 meV. system. In our study, we focus on the three lowest degen- erate energy states which are the the components of the GS at EGS ⋍ 0.88 meV, FES at EFES ⋍ 1.112 meV, and SES at ESES ⋍ 1.47 meV, respectively. The MB-energy spectrum of the quantum ring system in the presence of the cavity is shown in 3(b) where the photon energy is ωγ = 0.55 meV and electron-photon coupling strength gγ = 0.05 meV. Comparing to the ME- energy shown in Fig. 3(a), in addition to the degenerate states, the photon replica states are formed. The en- ergy spacing between the photon replicas is approximately equal to the photon energy at low electron-photon cou- pling strength gγ = 0.05 meV. For instance, the γGS is formed near the SES and the energy spacing between GS and SES is approximately equal to the photon energy ESES − EGS ≈ ωγ = 0.55 meV. Under this condition, the quantum ring system is resonant with the photon field [15]. In addition, we should mention that the first subband en- ergy of the semi-infinite leads is located at 1.0 meV (not shown). Therefore, the GS energy of the quantum ring system does not play an important role in the transport. In addition, the range of RSO is α = [0 : 24] meV nm which is a reasonable and applicable range for GaAs ma- terials. The parameter α depends on the electric field. An electric field can be generated in a heterostructure with two layers either by the intrinsic potential at the interface or as an external field. The Rashba parameter depends on that electric field, and thus it can be varied in the selected 4 2.2 2 1.8 1.6 1.4 1.2 1 0.8 0.6 ) V e m ( m u r t c e p s y g r e n e - E M (a) 2.2 2 1.8 1.6 1.4 1.2 1 0.8 0.6 ) V e m ( m u r t c e p s y g r e n e - B M (b) SES FES GS SES γGS FES GS 0 0.5 1 1.5 2 0 0.5 1 1.5 2 RSO-coupling (meV nm) RSO-coupling (meV nm) Figure 3: (Color online) Many-Electron (ME) energy spectrum (a) and Many-Body (MB) energy spectrum (b) of the ring system versus the Rashba spin-orbit (RSO) coupling constant (α). The green rect- angles are the zero-electron states (0ES), the brown circles indicate the one-electron states (1ES), and the two electron states are located at the upper part of the energy spectrum (now shown). The photon energy is ωγ = 0.55 meV, the electron-photon coupling strength is gγ = 0.05 meV, and the photon in the cavity is linearly polar- ized in the x-direction. The magnetic field is B = 10−5 T, and Ω0 = 1.0 meV. range here [24, 26]. 3.2. Transport properties in the absence of the cavity In this section we investigate the transport properties of the quantum ring system without the photon field. To calculate the spin-polarization and the thermospin current for the three lowest energy states, the chemical potential of the leads are fixed at µL = µR = 0.91, 1.112 and 1.47 meV for the GS, FES and SES calculations, respectively. The spin polarization, Sx (a), Sy (b) and Sz (c), of the elec- trons versus the RSO-coupling constant is shown in Fig. 4 for the three lowest energy states of the quantum ring system without the photon field. The non-vanishing spin- polarization in the range of α = [10 − 15] meV nm corre- sponds to the location of degenerate energy states (cross- ing energy states) shown in Fig. 3(a) and the destructive AC interference. Furthermore, it appears that the spin- polarization in both x- and z-directions is much smaller than that of y-direction (∼ 10 times smaller) in the se- lected range of the Rashba coupling constant [10−15] meV nm. The reason is that the main transport and canonical momentum are along the x-direction, and thus the effec- tive magnetic field of Rashba effect should be parallel to the y-direction. As a result, a higher spin-polarization in the y-direction (Sy) is induced in the system [22, 24]. We note that the spin-polarization of the GS (blue rect- angles) are very small comparing to the FES and SES which is due to the position of the GS located below the first subband of the leads. In addition, the spin- polarization of the FES is higher than that of the SES. Since the Sy is dominant in the quantum ring system comparing to both the Sx and Sz, we only focus on the GS FES SES 0.0001 0 -0.0001 -0.0002 -0.0003 -0.0004 0.006 0.004 0.002 0 ) 2 / -h ( x S ) 2 / -h ( y S (a) (b) -0.002 0.00016 0.00012 ) 2 / -h ( z S 8x10-5 4x10-5 0 -4x10-5 (c) 0 5 10 15 20 RSO-coupling (meV nm) Figure 4: (Color online) Spin polarization Sx (a), Sy (b) and Sz (c) of the quantum ring system without the photon field versus the RSO- coupling constant. The temperature of the left (right) lead is fixed at TL = 0.41 K (TR = 0.01 K) implying a thermal energy kBTL = 0.35 meV (kB TR = 0.00086 meV), respectively. The magnetic field is B = 10−5 T, and Ω0 = 1.0 meV. properties of the thermospin current in the y-direction. The TL-Thermospin current (a) and CL-Thermospin cur- rent (b) as a function of the RSO-coupling are presented in Fig. 5 for the three lowest states of the quantum ring without the cavity. It is clearly seen that the GS thermo- spin currents up to α ≃ 15.0 meV nm are almost zero due to the position of the GS located below the first subband of the leads. For the same range of α, the FES and SES are further active in the transport and we notice that the TL-Thermospin current of the FES and SES has a pro- nounced minimum value and the CL-Thermospin current has a maximum value at the RSO-coupling α ≃ 14.0 meV nm corresponding to a destructive AC interference. We should mention that the CL-Thermospin currents after α = 15 meV nm is slowly increasing due to the contri- bution of the FES to the transport at higher RSO-coupling (α > 15). The aforementioned explanations indicates that both FES and SES are active in the transport but with different weights. 3.3. Transport properties in the presence of the cavity We now assume the quantum ring system is coupled to the photon field. The energy spectrum of the ring-cavity system was shown in Fig. 3(b) where the photon field is lin- early polarized in the x-direction. The cavity forms photon replica states influencing the thermospin transport prop- erties. Since the FES and SES are the most active state 5 GS FES SES 2 0 -2 -4 -6 ) A p ( t n e r r u c n i p s o m r e h T - L T (a) -8 1.6x10-3 1.2x10-3 8x10-4 4x10-4 ) A p ( t n e r r u c n i p s o m r e h t - L C 0 -2x10-4 (b) 0 5 10 15 20 RSO-coupling (meV nm) Figure 5: (Color online) (a) TL-Thermospin polarization current, and (b) CL-Thermospin polarization current in the y-direction ver- sus the RSO-coupling for three lowest energy states, GS (blue rect- angles), FES (red circles) and SES (green diamonds) in the ab- sence of the photon field. The temperature of the left (right) lead is fixed at TL = 0.41 K (TR = 0.01 K) implying thermal energy kBTL = 0.35 meV (kB TR = 0.00086 meV), respectively. The mag- netic field is B = 105 T, and Ω0 = 1.0 meV. in transport, we focus only on the transport properties of these two states here. We fix the chemical potential of the leads at µL = µR = 1.112 meV for the FES and 1.47 meV for the SES calculations. The photon energy is fixed at 0.55 meV which is approximately equal to the energy spacing between the GS and the SES. We can thus say that the SES of the quantum ring system is resonant with the photon cavity while the FES is off-resonant. Fig- ure 6 shows the Sy spin polarization of the quantum ring system without (w/o ph) and with (w ph) photon field. We have mentioned, that in the absence of the photon, the Rashba effective magnetic field should be parallel to the y- direction and induce a spin polarization in the y-direction. In the presence of the photon field, a kinetic momentum in the x-direction is added to the electrons (see Eq. (4)), therefore, the Sy spin polarization should increase with the x-polarized photon field. This can be clearly seen in the Sy spin polarization of the FES, off-resonant regime. But in the on-resonant regime, the Sy spin polarization of the SES is slightly decreased in the presence of the photon field which is a direct consequence of the Rabi effect. Similar effect has been observed for the thermoelectric current in a quantum wire [13]. Figure 7 indicates the TL-Thermospin current (a) and CL-Thermospin current (b) of the quantum ring sys- tem versus the RSO-coupling for both FES and SES. The photon field causes to decrease the CL-Thermospin current while the TL-Thermospin current is almost un- changed. The suppression of CL-Thermospin of the FES, off-resonant regime, is due to the enhancement of the spin- polarization shown in Fig. 6 especially in the ranges of the degenerate energy states (α = [10 − 15] meV nm). It is interesting to see the resonant regime. The spin FES, w/o ph FES. w ph SES, w/o ph SES, w ph ) 2 / -h ( y S 0.014 0.012 0.01 0.008 0.006 0.004 0.002 0 -0.002 -0.004 0 5 10 15 20 RSO-coupling (meV nm) Figure 6: (Color online) Shows the spin polarization Sy versus the RSO-coupling of the quantum ring system without photon field (w/o ph) for the FES (red circles) and the SES (green diamonds), and with photon field (w ph) for the FES (blue circles) and SES (magenta diamonds). The photon energy is ωγ = 0.55 meV and the electron- photon coupling strength is gγ = 0.05 meV. The temperature of the left (right) lead is fixed at TL = 0.41 K (TR = 0.01 K) implying a thermal energy kBTL = 0.35 meV (kB TR = 0.00086 meV), respec- tively. The magnetic field is B = 10−5 T, and Ω0 = 1.0 meV. thermospin transport, we present Fig. 8 which shows the TL-Thermospin current (a) and CL-Thermospin cur- rent (b) of the SES (on-resonance regime) as a function of the RSO-coupling for different values of the electron- photon coupling strength gγ. Both the TL-Thermospin and CL-Thermospin currents are suppressed with increas- ing electron-photon coupling strength. This reduction in the TL- and CL-Thermospin currents is a direct conse- quence of the Rabi-splitting of the energy levels of the quantum ring system in which the energy spacing between the SES and γGS shown in Fig. 2 and Fig. 3(b) is increased at high electron-photon coupling strength [13]. Therefore, the TL- and CL-Thermospin currents decrease. w/o ph w ph, gγ = 0.05 w ph, gγ = 0.1 w ph, gγ = 0.15 (a) ) A p ( t n e r r u c 2 0 -2 -4 -6 2x10-3 n i p s o m r e h T - L T polarization of the SES is decreased in the presence of the cavity as is shown in Fig. 6, the CL-Thermospin current of the SES is also decreased. In addition to the Sy spin polarization that is used to explain the characteristics of the thermospin current, the photon replica states play an important role in the properties of the thermospin current. In the resonant regime, the γGS together with a SES form a Rabi-split pair (see Fig. 2) and contribute to the trans- port. As a result, the participation of the γGS to the transport decreases the CL-Thermospin current. ) A p ( t n e r r u c n i p s o m r e h T - L C 1x10-3 0 -1x10-3 -2x10-3 (b) 0 5 10 15 20 RSO-coupling (meV nm) FES, w/o ph FES, w ph SES, w/o ph SES, w ph (a) ) A p ( t n e r r u c n i p s o m r e h T - L T 2 0 -2 -4 -6 -8 1.6x10-3 1.2x10-3 8x10-4 4x10-4 0 -2x10-4 (b) 0 5 10 15 20 RSO-coupling (meV nm) ) A p ( t n e r r u c n i p s o m r e h T - L C Figure 7: (Color online) (a) TL-Thermospin current and (b) CL- Thermospin current versus the RSO-coupling are plotted for the SES of the quantum ring system without photon field (w/o ph) (green circles) and with photon field (w ph) for three different values of the electron-photon coupling strength gγ = 0.05 (magenta diamonds), 0.1 (red rectangles) and 0.15 meV (blue triangles) The photon energy is ωγ = 0.55 meV. The temperature of the left (right) lead is fixed at TL = 0.41 K (TR = 0.01 K) implying a thermal energy kBTL = 0.35 meV (kB TR = 0.00086 meV), respectively. The magnetic field is B = 10−5 T, and Ω0 = 1.0 meV. To further show the influences of the photon field on Figure 8: (Color online) (a) TL-Thermospin current and (b) CL- Thermospin current versus the RSO-coupling are plotted for the SES (on-resonant regime) of the quantum ring system without photon field (w/o ph) (green circles) and with photon field (w ph) for the electron-photon coupling strength gγ = 0.05 (magenta diamonds), 0.1 (red rectangles), and 0.15 meV (blue triangles). The photon en- ergy is ωγ = 0.55 meV. The temperature of the left (right) lead is fixed at TL = 0.41 K (TR = 0.01 K) implying a thermal energy kBTL = 0.35 meV (kBTR = 0.00086 meV), respectively. The mag- netic field is B = 10−5 T, and Ω0 = 1.0 meV. 4. Conclusions In summary, we have demonstrated properties of a ther- mospin transport through a quantum ring coupled to a photon cavity. In the absence of the photon field, ther- mospin current is induced at a low temperature gradient of the reservoirs that are connected to the quantum ring system. Tuning the Rashba spin-orbit coupling, degen- erate energy states are formed. It is observed that spin- polarization is maximum at the point of degenerate energy states corresponding to the AC destructive interference. In the presence of the photon field, the thermospin trans- port can be controlled using a single photon mode in the cavity. Two regimes, off- and on-resonant regimes, are studied. In the resonant regime, when the photon energy is approximately equal to the two lowest energy state of the quantum ring system, photon replica states are formed 6 [17] N. R. Abdullah, Magnetically and Photonically Tunable Dou- ble Waveguide Inverter, IEEE Journal of Quantum Electronics 52 (12) (2016) 1 -- 6. doi:10.1109/JQE.2016.2626080 . [18] N. R. Abdullah, C. S. Tang, A. Manolescu, V. Gudmundsson, Coherent transient transport of interacting electrons through a quantum waveguide switch, Journal of physics: Condensed matter 27 (2015) 015301. [19] N. R. Abdullah, C.-S. Tang, V. Gudmundsson, Phys. Rev. B 82 (2010) 195325. doi:10.1103/PhysRevB.82.195325. [20] N. R. Abdullah, T. Arnold, C.-S. Tang, A. Manolescu, Con- URL V. densed Matter 145303. http://stacks.iop.org/0953-8984/30/i=14/a=145303 Gudmundsson, of (2018) Physics: Journal (14) 30 [21] T. Arnold, C.-S. Tang, A. Manolescu, V. Gudmundsson, Magnetic-field-influenced nonequilibrium transport through a quantum ring with correlated electrons in a photon cavity, Phys. Rev. B 87 (2013) 035314. [22] T. Arnold, C.-S. Tang, A. Manolescu, V. Gudmunds- son, The European Physical Journal B 87 (5) (2014) 113. doi:10.1140/epjb/e2014-50144-y. [23] T. Arnold, The influences of cavity photons on the transient transport of correlated electrons through a quantum ring with magnetic field and spin-orbit interaction, Ph.D. thesis (School of Engineering and Natural Science, University of Iceland, Reyk- javik, 2014). [24] T. Arnold, Physica E: Low-dimensional C.-S. Tang, A. Manolescu, V. Gud- and 170 -- 182. (Supplement C) Systems (2014) mundsson, Nanostructures doi:10.1016/j.physe.2014.02.024. 60 [25] N. R. Abdullah, C.-S. Tang, A. Manolescu, V. Gudmundsson, Journal of Physics: Condensed Matter 28 (37) (2016) 375301. URL http://stacks.iop.org/0953-8984/28/i=37/a=375301 [26] D. V. Tuan, F. Ortmann, D. S. O. riano, Pseudospin-driven spin relaxation mechanism in?graphene, Nature Physics 10 (2014) 857 EP -- , article. URL http://dx.doi.org/10.1038/nphys3083 Valenzuela, S. So- Roche, and the spin polarization is sufficiently enhanced. Tuning the electron-photon coupling strength, the energy spacing between the states is increased leading to a suppression of the thermospin transport which is a direct consequence of the Rabi-splitting. 5. Acknowledgment This work was financially supported by the Research Fund of the University of Iceland, the Icelandic Research Fund, grant no. 163082-051, and the Icelandic Infrastruc- ture Fund. We acknowledge the University of Sulaimani, and Komar Research Center, Komar University of Science and Technology, Sulaimani City, Iraq References [1] T. Ji, X. Zhang, W. Li, Construction and Build- (2016) 576 -- 581. ing Materials 115 (Supplement C) doi:https://doi.org/10.1016/j.conbuildmat.2016.04.035. [2] J. Wei, L. Zhao, Q. Zhang, Z. Nie, L. Hao, En- ergy and Buildings 159 (Supplement C) (2018) 66 -- 74. doi:https://doi.org/10.1016/j.enbuild.2017.10.032 . [3] Mao Jun, Liu Zihang, Ren Zhifeng, Size effect in thermoelectric materials 1 (2016) 16028. doi:10.1038/npjquantmats.2016.28. [4] J. P. Heremans, V. Jovovic, E. S. Toberer, A. Sara- S. Yamanaka, 554 -- 557. mat, K. Kurosaki, A. Charoenphakdee, G. doi:10.1126/science.1159725. Snyder, Science (2008) (5888) 321 J. [5] A. I. Hochbaum, R. Chen, R. D. Delgado, W. Liang, E. C. Garnett, M. Najarian, A. Majumdar, P. Yang, Enhanced ther- moelectric performance of rough silicon nanowires, Nature 455 (2008) 778 -- 781. [6] Uchida K., Takahashi S., Harii K., Ieda J., Koshibae W., Ando K., Maekawa S., Saitoh E., Observation of the spin See- beck effect, Nature 455 (7214) (2008) 778 -- 781. doi:10.1038/ nature07321. [7] W. Zheng, D. Yang, W. Wei, F. Liu, X. Tang, J. Shi, Z. Wang, R. Xiong, Applied Physics Letters 107 (20) (2015) 203901. doi:10.1063/1.4936123 [8] H. Adachi, K. ichi Uchida, E. Saitoh, Theory of the spin seebeck effect, Reports on Progress Physics 76 (3) (2013) 036501. S. Maekawa, in [9] K. i. Uchida, H. Adachi, T. Kikkawa, A. Kirihara, M. Ishida, S. Yorozu, S. Maekawa, E. Saitoh, Thermoelectric generation based on spin seebeck effects, Proceedings of the IEEE 104 (10) (2016) 1946 -- 1973. doi:10.1109/JPROC.2016.2535167. [10] Y. Liu, X. Hong, J. Feng, X. Yang, Nanoscale Research Letters 6 (1) (2011) 618. doi:10.1186/1556-276X-6-618 . [11] K. Ghatak, S. Bhattacharya, S. Pahari, D. De, S. Ghosh, (2008) 195 -- 220. M. Mitra, Annalen der Physik 17 (4) doi:10.1002/andp.200710287. [12] L. Bai, S.-W. Ye, R. Zhang, crostructures 190 doi:https://doi.org/10.1016/j.spmi.2014.03.018. (Supplement C) 71 Superlattices and Mi- 202. (2014) -- [13] N. R. Abdullah, C.-S. Tang, A. Manolescu, V. Gud- 249 -- 254. (2016) mundsson, doi:10.1021/acsphotonics.5b00532, . ACS Photonics (2) 3 [14] N. R. Abdullah, C.-S. Tang, A. Manolescu, V. Gudmundsson, Physica E: Low-dimensional Systems and Nanostructures 95, (2017) 102-107. doi:10.1016/j.physe.2017.09.011. [15] N. R. Abdullah, C.-S. Tang, A. Manolescu, V. Gud- mundsson, Physics Letters A 382 (4) (2018) 199 -- 204. doi:https://doi.org/10.1016/j.physleta.2017.11.007. [16] N. R. Abdullah, Physics Letters A 382, (2018) 1432-1436. doi:https://doi.org/10.1016/j.physleta.2018.03.042. 7 

1710.10250

1

1710

Spin transfer and spin pumping in disordered normal metal-antiferromagnetic insulator systems

We consider an antiferromagnetic insulator that is in contact with a metal. Spin accumulation in the metal can induce spin-transfer torques on the staggered field and on the magnetization in the antiferromagnet. These torques relate to spin pumping: the emission of spin currents into the metal by a precessing antiferromagnet. We investigate how the various components of the spin-transfer torque are affected by spin-independent disorder and spin-flip scattering in the metal. Spin-conserving disorder reduces the coupling between the spins in the antiferromagnet and the itinerant spins in the metal in a manner similar to Ohm's law. Spin-flip scattering leads to spin-memory loss with a reduced spin-transfer torque. We discuss the concept of a staggered spin current and argue that it is not a conserved quantity. Away from the interface, the staggered spin current varies around a zero mean in an irregular manner. A network model explains the rapid decay of the staggered spin current.

cond-mat.mes-hall

cond-mat

Spin transfer and spin pumping in disordered normal metal-antiferromagnetic insulator systems Sverre A. Gulbrandsen and Arne Brataas Center for Quantum Spintronics, Department of Physics, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway (Dated: July 23, 2018) We consider an antiferromagnetic insulator that is in contact with a metal. Spin accumulation in the metal can induce spin-transfer torques on the staggered field and on the magnetization in the antiferromagnet. These torques relate to spin pumping: the emission of spin currents into the metal by a precessing antiferromagnet. We investigate how the various components of the spin-transfer torque are affected by spin-independent disorder and spin-flip scattering in the metal. Spin-conserving disorder reduces the coupling between the spins in the antiferromagnet and the itinerant spins in the metal in a manner similar to Ohm's law. Spin-flip scattering leads to spin- memory loss with a reduced spin-transfer torque. We discuss the concept of a staggered spin current and argue that it is not a conserved quantity. Away from the interface, the staggered spin current varies around a zero mean in an irregular manner. A network model explains the rapid decay of the staggered spin current. I. INTRODUCTION Charge currents cannot flow through antiferromagnetic insulators (AFIs). However, recent experiments have demonstrated that typical AFIs such as NiO and CoO are good spin conductors.1 -- 5 One of the origins of these features is that spin pumping and spin transfer are as efficient across AFI-normal metal (NM) interfaces as in ferromagnet-NM systems.6 This potent spin transfer can empower low-dissipation high-frequency spin circuits in AFIs.7 In insulators, lossy itinerant charge carriers do not contribute to dissipation. These developments in insulators increase the potential applicability of antiferromagnetic spintronics. Antiferro- magnets have no stray fields,8,9 and this feature might en- able denser antiferromagnetic elements in future devices. However, the notable advantage of antiferromagnets com- pared to ferromagnets is that they can operate at speeds that are a hundred times faster7,10,11. By using antifer- romagnets, we can envision circuits that function in the largely unexplored spintronics THz regime. For instance, one can envision THz spin-torque oscillators.7,12 -- 14 Fur- thermore, achieving such rapid spin dynamics is possible even in the absence of external magnetic fields. Many of the phenomena that occur in ferromagnets also occur in antiferromagnets.15 -- 17 In antiferromagnetic metals and junctions involving antiferromagnets, there is a significant anisotropic magnetoresistance.8,18 -- 21 This magnetoresistance enables the detection of the staggered field.18 There are also strong spin-orbit torques that can function to reorient the staggered field,21 -- 23 as demon- strated in recent seminal experiments.24,25 In AFIs, spin waves can also reorient the staggered field.11,26,27 In addition to spin-wave transport, spin su- perfluidity is also possible.28,29 Spin superfluidity is anal- ogous to superfluidity in helium-4 and could offer a low- dissipation route to spin transport.30 -- 33 In many of the phenomena that involve electrical con- trol of spin transport in AFIs, spin transfer and spin pumping across AFI-metal interfaces are essential.34 -- 37 Seminal papers have computed this interface spin cou- pling in the ideal case of no disorder and no spin-flip scat- tering within the metal6,29. The purpose of the present paper is to elucidate in further detail how the torques on the staggered field and on the magnetization occur and how these torques are influenced by various types of disorder. To this end, we first reformulate the previous theo- ries in a quantum language. We compute the rate of change of the spins at the A and B sublattices. In three- dimensional systems, where the number of involved spins is large, we evaluate the quantum mechanical rates of change in the classical limits of large spins and recover the previous results. Subsequently, we consider how the coupling between the spins in the AFI and the itinerant spins in the NM are affected by spin-conserving disorder and spin-flip-inducing disorder. The remainder of our manuscript is organized as fol- lows. In section (II), we present our model for the elec- trons in the NM, the spins in the AFI, and the coupling between the two sub-systems. In section (III), we present our numerical results of the spin-transfer torques (STTs) on the AFI and the spin currents in the NM. Finally, we discuss and summarize our findings in section (IV). II. SPIN TRANSFER AND SPIN PUMPING Dynamic antiferromagnets can pump spins into adja- cent conductors even when they are insulating. Antifer- romagnets are also affected by STTs arising from spin accumulations in neighboring metals. We will investi- gate how spin-conserving impurity scattering and spin- flip scattering influence these processes. Using a dynamic gauge transformation, we will relate spin pumping to spin transfer, similar to the Larmor theorem for ferromagnets. 7 1 0 2 t c O 7 2 ] l l a h - s e m . t a m - d n o c [ 1 v 0 5 2 0 1 . 0 1 7 1 : v i X r a A. Spin-transfer torques The AFI is located next to a NM (see Fig. 1). A tight- binding model describes the itinerant electrons. At the interface, we assume that the AFI spins are exchange coupled to the itinerant spins in the metal. We assume that the spin dynamics in the AFI is spatially homoge- neous. The spins behave as macrospins, albeit on two different sublattices. Excluding the spins in the bulk of the AFI, the Hamiltonian that involves the electron de- grees of freedom is H = HNM + HAF,N , (1) where HNM describes the NM and HAF,N characterizes the coupling between the itinerant spins and the localized AFI spins. The Hamiltonian for the itinerant electrons in the NM (cid:88) (cid:88) r reads as (cid:88) r =(cid:0) c (cid:104)r,r(cid:48)(cid:105) † rcr(cid:48) + 6t c † c rcr + † c rVrcr , (2) HNM = −t r (cid:1) is in terms of the creation (anni- † † where c† r↑ , c r↓ hilation) operator c† rs (crs) of an electron at site r with spin projection s =↑ or s =↓ along the spin-quantization axis z. In Eq. (2), the sum ((cid:104)r, r(cid:48) (cid:105)) is over nearest neigh- bors and t is the hopping energy. The onsite potential consists of a spatially constant contribution, 6t, and a random potential, Vr, that models disorder. We will specify the statistical properties of the random potential below. We set the lattice constant as a = 1. Operators and unit vectors are denoted by a hat (). FIG. 1. An antiferromagnetic insulator (AFI) is in contact with a normal metal (NM). The NM is connected to a reser- voir with a spin accumulation, µS. 2 scattering matrix S. In our geometry, there is only one lead. The scattering matrix thus only includes reflection matrices. Our focus is on the spin information transferred be- tween the AFI and the NM. In the NM, the itinerant spin density operator at position r is sr =  2 † c rσcr , (3) in terms of the 2 × 2 spin-space vector of Pauli matrices, σ = (σx, σy, σz). We separate the longitudinal, x, and transverse, r⊥ = (y, z), coordinates, r → (x, r⊥). Fur- thermore, we use a discrete index to label the lattice po- sition along the x-direction such that x → 0, 1, . . .. For instance, s1,r⊥ denotes the electron spins in the trans- verse layer x = a. The interface exchange coupling be- tween the itinerant electron spins at the interface to the adjacent localized AFI spins is (see Fig. 2) HAF,N = J S 0,r⊥ · s1,r⊥ , (4) (cid:88) r⊥ where J is the interface exchange coupling. S 0,r⊥ is the operator associated with the localized spin in the AFI next to the NM. FIG. 2. Sections of the AFI and the NM around their inter- face. In the AFI (x ≤ 0), the localized spins residing on the red (blue) nodes are at sublattice A (B). The localized spins at the interface (x = 0) are exchange coupled to adjacent itinerant electron spins (x = 1), as indicated by semi-long dashes. In the NM (x ≥ 1), we use a different convention for the sublattices. In the NM, the brown (black) sites reside on sublattice A (B). The solid black lines illustrate the itinerant electron hopping. The spin current in the longitudinal (x) direction is j±x r = it 2 † † rσcr±δx r±δx σcr − c c . (5) (cid:16) (cid:17) Even in antiferromagnets where the spins precess rapidly at THz frequencies, the spin dynamics are slower than the electron dynamics. The STT on the AFI can therefore be computed as a scattering problem that keeps the localized spins static. We assume that the NM cou- ples to a large reservoir where there may be spin and charge accumulations. The out-of-equilibrium distribu- tion of the incoming electrons is therefore assumed to be known in the reservoirs. The out-of-equilibrium spin and charge densities are governed by the time-independent In Eq. (5), the superscript +(−) classifies spin currents to the right (left) when electrons hop from x to x + a (x to x − a). δx = ax equals one lattice constant in the x-direction. Similarly, we define the spin currents in the two transverse directions: j±y r . Correspondingly, the hoppings are δy and δz in the y- and z-directions, respectively. r and j±z We consider a bipartite AFI. A localized spin on sub- lattice A has six nearest-neighbor spins on sublattice B and vice versa. The lattice is cubic with lattice constant µSreservoirNMAFInxyzx=0x=1A similar expression is obtained for the spin current in the transverse direction z, I ∆z , by substituting y → z. x,r⊥ 3 2a for sublattice A. Sublattice B is displaced by ax with respect to sublattice A. For convenience, we use a different set of sublattices in the NM, but we also label these sublattices as sublattices A and B. This notation describes the (staggered) spin currents in the most transparent way. In our coordinate system, sublattice A is spanned by the vectors {rA} = {(a, 0, 0), (0, a, a), (0, a,−a)}. Sublattice B is displaced by ay with respect to sublattice A, as shown in Fig. 2. The volume of a unit cell in sublattice A or B is then 2a3. We separate the currents into terms that are associated with the rate of change of the spin at the A and B sub- lattices. The spin current in the longitudinal direction associated with lattice sites A in layer x is (cid:40)(cid:80) (cid:80) (cid:16) ∈A ∈A j +x x,r⊥ 1 2 r⊥ r⊥ I s,A x = j +x x,r⊥ − j−x x,r⊥ (cid:17) x = a , x > a . (6) We define a similar relation for the spin current in the longitudinal direction associated with lattice sites B, I s,B x . The total longitudinal spin current at layer x is the total spin current carried at both sublattices, I s x = I s,A x + I s,B x . FIG. 4. (a) The staggered spin current I ss 1 at the AFI-NM interface (x = 1), which is shown by black arrows. The torque τN on the Neel order parameter is determined by I ss 1 . (b) The staggered spin current I ss x at layer x > 1 includes spin currents between lattices A and B within the layer at x (the interlayer staggered transverse current I AB x ) and longitudinal staggered spin currents between the layers at x− 1 and x and between the layers at x and x + 1. Note the factor of 2 in front of I AB in the definition of the staggered spin current I ss x . x A staggered spin current involves the transfer of spin angular momentum between the two sublattices within each transverse layer. We define the interlayer staggered transverse current operator as I AB x = (cid:88) (cid:88) (cid:16) (cid:17) (cid:16) (cid:17) I ∆y x,r⊥ + I ∆z x,r⊥ I ∆y x,r⊥ + I ∆z x,r⊥ , − (8) as illustrated in Fig. 4. The staggered spin current at layer x is then ∈A r⊥ ∈B r⊥ FIG. 3. (a) The spin current I s 1 at the interface (x = 1), indi- cated by black arrows, determines the torque on the magneti- zation τM. (b) The spin current I s x in the NM bulk (x > 1) is the average of the spin current between the layers at x − 1 and x and between the layers at x and x + 1. The NM has finite lengths Nya and Nza in the trans- verse directions y and z, where Ny and Nz are the number of lattice sites, respectively. In the transverse direction y, the spin currents (between the sublattices) are j +y r j +y r − j−y −j−y r r I ∆y x,r⊥ = y = 1 , 1 < y < Ny , y = Ny . (7) I ss x = I s,A x − I s,B x + 2 I AB x , (9) which consists of a longitudinal part and a transverse part. We are interested in the average torque per spin in the AFI. The total number of localized spins is NAF = N⊥N(cid:107), where N⊥ is the number of spins in one transverse layer and N(cid:107) is the number of transverse layers in the longitudinal direction. The total volume of the AFI is then NAFa3. We define the total spin of the AFI as (cid:88) r (cid:104) S r(cid:105) , SM = (10) where the brackets denote the expectation value. xyzx=0x=1x=2x−1xx+1(a)(b)xyzx=0x=1x=2x−1xx+1(a)(b)and B is SN = The difference between the total spin on sublattices A (cid:88) r∈A (cid:88) r∈B (cid:104) S r(cid:105) − (cid:104) S r(cid:105) . (11) The total spin SM and the spin difference SN are re- lated to the total magnetization and the Neel order pa- rameter in the AFI, respectively. We define the cor- responding torques per spin τM = ∂tSM/NAF and τN = ∂tSN /NAF. In the steady state, the itinerant spins do not vary in time, (cid:104)∂tsr(cid:105) = 0. The rate of change of the localized spins can thus be expressed in terms of steady-state spin cur- rents in the NM via the spin continuity equations (Eqs. (A1) and (A5)). The torques on the AFI are When there is no spin-flip scattering, the reflection ma- trix in Eq. (13) simplifies to 4 (r 1 2 1 2 ↓ mn) + ↑ mn + r rβα mn = δαβ ↑ mn − r (r ↓ mn)n · σβα , (15) in terms of reflection coefficients for spin up (down), r↑ (r↓ mn mn), along the direction of the Neel order parameter. The scattering states constitute a current-normalized and complete set of states. Therefore, the reflection ma- trix is unitary, and particle current is conserved. The electron field operator (inside the lead) is expressed in terms of the scattering states as cr(t) = dEψnEα(r)cnα(E)e −iEt/ , (16) N⊥(cid:88) (cid:88) (cid:90) ∞ n=1 α=↑,↓ ε⊥n τM = − τN = − I s 1 , I ss 1 , 1 NAF 1 NAF in terms of the expectation values of the spin current and the staggered spin current at the interface (layer x = 1). (12a) (12b) where we define the operator cnα(E) (c† nα(E)), which an- nihilates (creates) the scattering state ψnEα originating from the reservoir.38 The distribution function in the reservoir determines the expectation values of the operators. We allow for a small spin accumulation in the reservoir. Then, (cid:10)c † nα(E)cmβ(E (cid:48) )(cid:11) = δnmδ(E − E (cid:48) )fβα(E) . (17) B. Umklapp scattering The distribution function is In the lead, scattering states coupled to out-of- equilibrium reservoirs determine the torques on the AFI. Energy is conserved and the transport is elastic. The scattering states are labeled by the incoming (transverse) mode n, energy E, and spin α: ψnEα(r) = −ik (cid:107) nx ξαϕn(r⊥)e 1 N⊥(cid:88) √hvn + (cid:88) 1 √hvm mnξβϕm(r⊥)eik rβα (cid:107) mx , (13) m=1 † ↑ = ( 1 0 ), and ξ β=↑,↓ (cid:107) where vn = ∂E/∂k n is the group velocity of the prop- agating wave and h = 2π is Planck's constant. ξα is a † ↓ = ( 0 1 ). The transverse spinor, ξ wavefunction ϕn(r⊥) is an eigenfunction of the transverse part of HNM with eigenenergy ε⊥n ∈ [−4t, 4t]. The lon- (cid:107) nx gitudinal wavenumber is positive, k propagates to the left (-) or to the right (+). The scat- tering state has a total energy of (cid:107) n > 0; thus, e∓ik (cid:16) (cid:17) E = ε⊥ n + 2t 3 − cos k fαβ(E) = δαβfFD(E − µ) + σαβ · fS(E) , (18) in terms of the Fermi-Dirac function fFD(E − µ) with chemical potential µ and temperature T . The func- tion fS(E) describes the spin accumulation µS = (cid:82) dETrs{σ(σ · fS(E))} by taking the trace in spin space.39 It is illustrative to first consider the ideal case of no disorder and no spin-flip scattering, as considered previously6,29. For simplicity, we also assume periodic boundary conditions in the transverse directions. There are N lattice sites in both transverse directions (y and z). The orthonormal transverse wavefunctions are ϕn(r⊥) → ϕny (y)ϕnz (z), where ϕny (y) = 1 √N e2πiyny/N a , (19) with discrete values ny = 1, 2, ..., N and similarly for the wave function ϕnz (z) that is characterized by nz. The resulting scattering matrix consists of normal scattering (cid:107) na . (14) 1 2 ↑ mymz,nynz + r ↓ mymz,nynz ) = Anynz δmyny δmznz , (20) (r The total number of transverse modes is N⊥. A mode is propagating if its energy E lies within its band, i.e., if ε⊥n +4t < E < ε⊥n +8t. The number of propagating modes is then Np < N⊥. We will consider a NM at half-filling such that the Fermi energy is EF = 6t. The reflection amplitudes rβα mn describe reflection from incoming mode n to outgoing mode m and from spin α to β. and spin-dependent umklapp scattering 1 2 ↑ mymz,nynz − r (r ↓ mymz,nynz ) = Bnynz δmy ¯ny δmz ¯nz , (21) with coefficients Anynz and Bnynz defined in the Ap- pendix in Eqs. (D1a) and (D1b). Umklapp scattering changes the incoming transverse modes from (ny, nz) 2 , nz ± N 2 ) or (¯ny, ¯nz) = (ny ± N to (¯ny, ¯nz) with four possible combinations: (¯ny, ¯nz) = (ny ± N 2 , nz ∓ N 2 ). When the incoming mode ny is in the range from 1 to N 2 , then ¯ny = ny + N 2 , and when ny is in the range from N 2 + 1 to N , then ¯ny = ny − N 2 . There are similar considerations for nz. A finite magnetization may induce spin precession of the conduction electrons. To separate the different con- tributions to the torques in a transparent manner, we focus on AFIs with a vanishingly small magnetization. In clean systems, we compute that the torques on an AFI with volume VAF = NAFa3 are τM = τN = 1 NAF 1 NAF 1 4π 1 4π g⊥ mn × (µS × n) , g⊥ n(µS × n) , (22a) (22b) in terms of the dimensionless spin-mixing conductances g⊥ m = g⊥ n = 8λ2N⊥ (1 + λ2)2 4λN⊥ (1 + λ2)2 sin2 k (cid:107) nynz a , (23a) (cid:107) nynz a) cos 2k (cid:107) nynz a , (23b) (− sin k (cid:88) (cid:88) nynz nynz 3 (cid:107) where the longitudinal wavenumber k nynz is evaluated at the Fermi energy EF. The summations are over the fraction of the Brillouin zone (BZ) where the modes are propagating. λ = JS/t parametrizes the exchange cou- pling at the interface, where S is the localized spin. The torque τM obtains its maximum for λ = ±1, whereas τN has the highest value at λ = ± 1√ . Without the λ-dependent fractions in Eqs. (23a) and (23b), the BZ summations of the trigonometric functions in g⊥m and g⊥n yield values on the order of unity (numerical values of ≈ 4.57862 and ≈ 3.163, respectively). mn(r↓ (cid:80) mn r↑ in the clean limit that g↑↓ = (cid:80) spin-mixing conductance39 g↑↓ = Np − as the real part: g⊥m = Re(cid:8)g↑↓(cid:9). The spin-mixing con- Note that there is an important difference in the defini- tions of the spin-mixing conductances g⊥m and g⊥n for the AFI (Eq. (23)) compared to the usual expression for the mn)∗ for ferromagnets. For the AFI discussed above, we find nynz 2Bnynz2 in terms of the amplitude for umklapp scattering Bnynz from the scattering matrix (Eqs. (21) and (D1b)). The spin- mixing conductance g⊥m can be expressed in terms of g↑↓ ductance g⊥n is not related to g↑↓ in a simple way. Instead, g⊥n has to be calculated directly from the wavefunctions at the interface. Consequently, we find that g⊥n arises from an expression that is a combination of normal and umklapp scattering coefficients, Anynz and Bnynz , which is then reduced to Eq. (23b). 5 C. Spin pumping from an antiferromagnetic insulator We now consider the dynamic situation where the lo- calized spins in the AFI rotate uniformly. Similar to how precessing spins in an FM pump a spin current into an adjacent metal, we show that a dynamical AFI pumps both a spin current and a staggered spin current. To demonstrate this property, we perform a time- dependent gauge transformation. The spin axes rotate and transform the time-dependent Hamiltonian H(t) in the lab frame to a static Hamiltonian H with an effec- tive magnetic field in the rotating frame. The same gauge transformation relates the spin currents in the lab frame to the spin currents in the rotating frame. Static scat- tering states determine the latter spin currents. In the classical limit of many AF spins, the coupling † (t) · c 1,r⊥ σc1,r⊥ , (24) (cid:88) between the AFI and the metal is  2 HAF,N(t) = J S 0,r⊥ where the dynamic spins S 0,r⊥ r⊥ (t) = ±Sn(t) are parallel or anti-parallel to the direction of the Neel order param- eter n(t). The gauge transformation rotates the spin space. In the rotating frame, the field operator is cr = U (t)cr , (25) where U (t) is a time-dependent, unitary 2 × 2 matrix, U† = U−1. The evolution of cr(t) is governed by the Hamiltonian H = U HU † − iU ∂U† ∂t , (26) which adds the gauge potential Veff = −iU ∂tU†, as shown in App. C. The time-dependent problem with the Hamiltonian H(t) is transformed into a static problem by finding a unitary matrix U (t) that makes the terms U (n(t) · σ)U† and Veff both independent of time. A time-dependent gauge transformation can describe spin pumping, both in a ferromagnet40,41 and in an an- tiferromagnet, when the spins precess around a static precession axis or if the precession axis changes slowly compared to the precession frequency. We now illus- trate this property by considering a Neel order param- eter n(t) = x cos ω0t sin θ0 + y sin ω0t sin θ0 + z cos θ0 that rotates around the z-axis. The spins precess with angu- lar frequency ω0 and a constant polar angle θ0, as shown in Fig 5. The spin-rotation matrix that yields a static Hamiltonian in the rotating frame is then U (t) = eiσyθ0/2eiσzω0t/2 . (27) U (t) diagonalizes U (n(t) · σ)U† = σz. The resulting gauge potential is Veff = ω0 2 (sin θ0σx − cos θ0σz) . (28) 6 In other words, the spin current I s 1 and the staggered spin current I ss 1 pumped from a precessing Neel order parameter can be calculated by considering a static prob- lem, where the Neel order parameter points along the z- axis with a spin accumulation ω0 sin θ0σx x in the reser- voir. D. Decay of a staggered current The staggered spin current I ss 1 has a finite value at the AFI-NM interface, which induces a torque on the Neel order parameter. However, in the NM, the staggered spin current I ss x is expected to decay rapidly away from the interface because it is not conserved inside the NM. We now illustrate in a heuristic manner how this can be understood from a network circuit model. FIG. 6. (a) A section of the network circuit model. There are l transverse layers with associated potentials Vi for i = 1, .., l. Black (red) nodes in layer i have potential Vi (−Vi). To the left, layer 1 is connected to nodes where the potential is staggered: V0 or −V0. To the right, layer l is connected to nodes with zero potential. Dotted lines indicate where layers 2, 3, ..., l − 1 are located. The connections between the nodes have the same conductance, shown by solid lines. (b) The leftmost section of the network circuit. The currents and their directions in the circuit are indicated by the arrows. The current between a node in layer i − 1 and a node in layer i is Ii−1,i. For the current between two nodes within the same layer i we label the current as Ii,i. We consider a network circuit with l transverse layers of nodes, as shown in Fig. 6. The network is cubic to resemble the tight-binding model. The nodes are con- nected to adjacent nodes via identical conductances. As a boundary condition, layer 1 is connected to a layer of nodes where the potentials are V0 or −V0 in a staggered configuration. For simplicity, we use periodic boundary FIG. 5. (a) Laboratory frame of reference, where the AFI spins precess around the z-axis with frequency ω0. The Neel vector n(t) is at an angle θ0 with respect to the z-axis. There is no spin accumulation in the lab frame. (b) In the rotating spin frame, the Neel vector points in the z-direction. The gauge potential Veff induces a finite spin accumulation µS in the NM, including the reservoir. A Zeeman-like gauge potential in the NM reservoir splits the population of spins and is equivalent to a spin accu- mulation µS = ω0(sin θ0 x − cos θ0 z) . (29) We now consider the spin currents in the rotating frame. Here, the static AFI spins point along z. In linear re- sponse, we can calculate the spin currents by disregard- ing the gauge-transformation-induced magnetic field in the scattering region while retaining a spin accumulation in the form of µS in the reservoir. Only the component µS · x = ω0 sin θ0 results in the non-zero spin current I s ω0 sin θ0(−x) and the staggered spin cur- 1 = (1/4π)g⊥r rent I ss ω0 sin θ0 y. Spin currents in the ro- tating frame (denoted with a tilde) are defined similar to Eq. (5) with cr → cr. When transforming back to the lab frame, we use that † ω0 sin θ0U (−σx)U = (n × n) · σ , and ω0 sin θ0U†σyU =  n·σ, such that the spin currents 1 = (1/4π)g⊥i (30) in the lab frame are I s 1 = I ss 1 = 1 4π 1 4π g⊥ r n × n , g⊥ i n . (31a) (31b) The expressions for the pumped spin currents in Eqs. (31a) and (31b) are consistent with the results obtained from a time-dependent scattering-matrix approach6 in the limit of a vanishing magnetization, m → 0. µS=0(a)Labframe:n(t)µS(b)Rotatingframe:n=zω0θ0θ0xyzV0−V0V1−V0V00Vl−V1−VlLayer2,3,...,l−1Layer1000(a)(b)I0,1I0,1I1,1I1,2I1,2I2,2I2,3I2,3conditions in the two transverse directions. Then, the potentials at the nodes within one transverse layer have the same absolute value but opposite (staggered) signs because of symmetry, in the same way as the leftmost layer in Fig. 6. We associate a potential V1, V2, ..., Vl to each layer 1, 2, ..., l, respectively. Layer l is connected to layer l − 1 on one side and to a layer of nodes with zero potential on the other side. We define currents Ii−1,i be- tween adjacent nodes in layer i− 1 and layer i. Similarly, there are currents Ii,i between adjacent nodes within the same layer i. The directions of these currents are indi- cated in Fig. 6. From current conservation at each node, we can calculate all potentials and currents in the circuit. Kirchoff's law determines the potentials Vi for layer i from l linear equations,  −1 10 1 −1 10 ... ... 1 −1 10 0 ... 0 0 0 −1 10 1 0 0 −1 10 ... ... ... −1 10 0 0 0 0 ... −1 10 1 1 −1 10   V1 V2 V3 ... Vl−1 Vl  =   , V0 0 0 ... 0 0 (32) which consists of a uniform, tridiagonal l × l matrix. The analytical solution of Eq. (32) for a large integer l is cumbersome. Instead, we present the resulting currents Ii−1,i between nodes in layers i − 1 and i when l = 10 in Fig. 7. We have verified that the results are similar for a range of different l. FIG. 7. Logarithmic plot of the current Ii−1,i between ad- jacent nodes in layer i − 1 and layer i. There are l = 10 transverse layers in the middle of the circuit. The currents are normalized by the current I0,1, which is the current be- tween the nodes with potentials V0 and V1. The currents Ii,i+1 between layers i and i + 1 are re- duced by a factor on the order of 10−1 compared to the currents Ii−1,i. Similarly, the currents Ii,i between nodes in the same layer i are reduced compared to the currents Ii−1,i−1 in the preceding layer i− 1. Based on this result, the classical analogue of a staggered current, in the form of Eq. (9), decays within few transverse layers. 7 III. NUMERICAL RESULTS We will now consider the effects of spin-conserving and magnetic disorder on the torques. Disorder is modeled using a static random potential Vr. The elastic poten- tial consists of spin-conserving and magnetic-impurity- induced spin-flip scattering parts: Vr = V c r 1 + V m r · σ . (33) We will consider the effects of spin-conserving and spin- flip scatterings separately. (cid:17) r = V m r The values of V c 1 − ζ 2 cos φ, 1 − ζ 2 sin φ, ζ (cid:2)0, ηm (cid:68) (cid:3), ζ ∈ (cid:2) r )2(cid:69) (cid:16)(cid:112) (cid:112) (cid:2)0, 2π(cid:1). In this way, the − 1, 1(cid:3) and φ ∈ r )2(cid:11) r (cid:105) = 0, while the variances (cid:10)(V c r are random in a uniform distribu- tion within the energy range between −η and η. In the case of spin-flip scattering, the magnetic impuri- ties V m , where the parameters are uniformly distributed within V m r ∈ direction of the magnetic impurities V m r is uniformly dis- tributed on the unit sphere. The disorder then has zero mean, (cid:104)V c (V m and m, respectively. The presence of magnetic disorder modifies the spin continu- ity equations (Eq. (A5)) by introducing additional spin- orbit-like terms c† r ) cr, which we also include in our considerations. are given by η2 and η2 r (cid:105) = (cid:104)V m r (σ × V m We utilize the python package KWANT42 to solve the scattering problem with disorder. In this way, we obtain the wavefunction at all lattice sites. Based on this wave- function, we compute the spin currents and the torques. The i'th component of the torques on the total spin and the spin difference depends on the spin accumulation in the NM: τ iM =A τ iN =A M ij µj S , N ij µj S , (34a) (34b) where summation over repeated indices is implied. The real-valued second-rank tensors AM ij and AN ij are (cid:88) (cid:88) (cid:88) M ij = A NAF 1 (cid:104) × − ψ j r⊥ α1α2ss(cid:48) n∈prop (cid:48) i σα2α1 σss t 4i † (cid:48) nEFα1(x = 2, r⊥, s)ψnEFα2 (x = 1, r⊥, s ψ † (cid:48) nEFα1 (x = 1, r⊥, s)ψnEFα2(x = 2, r⊥, s ) ) (cid:105) , (35) 12345678910i10-910-810-710-610-510-410-310-210-1100101logIi−1,iI0,1and N ij = A (cid:88) (cid:88) (cid:88) (cid:88) NAF (cid:48) j 1 (cid:104) ) − ( r⊥ t 4i † (cid:48) nEFα1 (x = 2, r⊥, s)ψnEFα2(x = 1, r⊥, s i σα2α1 σss n∈prop α1α2ss(cid:48) ∈B ∈A r⊥ ψ (cid:16) × † (cid:48) nEFα1 (x = 1, r⊥, s)ψnEFα2(x = 2, r⊥, s ψ + 2 × † (cid:48) − ψ nEFα1(x = 1, r⊥, s)ψnEFα2(x = 1, r⊥ + δ, s † (cid:48) nEFα1 (x = 1, r⊥ + δ, s)ψnEFα2 (x = 1, r⊥, s (cid:17)(cid:105) )− ψ ) ) , (36) and 8(d), which indicates that the specific disorder con- figurations become important when the strength of the impurities may be large. 8 ) where the sum is over all propagating modes at the Fermi energy, and we sum terms with δ = ay and δ = az. For finite systems of length N a in the transverse directions, the end point of the transverse sums with transverse hop- pings is N − 1. The dominant components of the STTs τN and τM follow from µS × n and n × (µS × n), respectively, even for disordered configurations. We show how the relevant STTs are affected by spin- conserving disorder (ηm = 0) in Fig. 8. In these cases, the AFI has an order parameter n = z, which is transverse to the direction of transport. In the adjacent metal reser- voir, a transverse spin accumulation µS = µS y induces the STTs. The exchange coupling was set to JS/t = 1. There are 40 × 40 sites in the two transverse directions. Here, the concentration of the impurities in the NM is constant at 0.125. The STTs are compared to the elec- trical conductances G of a two-terminal system with the same system parameters as the (one-terminal) systems considered for the STTs, as shown in Fig. 8. G0 is the Sharvin conductance. We normalize the STTs by the torques τ xN ,0 and τ yM,0, which are the torques calculated for systems without disorder. All the values presented in Fig. 8 are calculated by averaging over 15 impurity configurations, and the standard deviations are shown as error bars. In Figs. 8(a) and 8(b), we consider different lengths L of the disordered region while keeping the impurity- associated parameter η = t fixed. In Fig. 8(a), we find that spin-conserving disorder enhances the torque τ xN on n compared to the torque without disorder, and the ef- fect is most prominent for disorder close to the interface (L/a less than 62). As the length of the conductor in- creases, the torque τ xN /τ xN ,0 decreases with a slope similar to the reduction in the conductances G/G0. As shown in Fig. 8(b), the normalized values of τ yM follow the con- ductances very closely as the length of the conductor is changed. In Figs. 8(c) and 8(d), we fix the length L/a = 40 of the metal and show the torques and conductances for different impurity-associated energies η. The torque on n is enhanced (Fig. 8(c)) for η ∼ t, whereas it decreases for larger η. For the torque on the magnetization, the mean values τ yM/τ yM,0 follow the conductance G/G0 (Fig. 8(d)). The standard deviations are large in Figs. 8(c) FIG. 8. STTs (blue diamonds) for AFI-NM systems with spin-conserving disorder (ηm = 0). The Neel order parameter points in n = z, and the torques are induced by a spin accu- mulation µS = µS y. There are 40 × 40 sites in the transverse directions, and the concentration of the impurities is fixed at 0.125. In (a) and (b), we vary the length L of the NM while keeping the disorder-associated potential energy at η = t. In (c) and (d), we vary the disorder strength η for a system with length L/a = 40. The red squares show the electrical conduc- tance of a two-terminal system with the same parameters as in the STT cases. The torques are normalized with respect to their corresponding torques, τ xN ,0 and τ yM,0, when η = 0, i.e., without disorder. G0 is the Sharvin conductance. All values are averaged over 15 impurity configurations. The error bars show the standard deviation in the average torques. We now consider the STTs on the AFI when all the impurities are magnetic (η = 0), as shown in Fig. 9. Many of the system parameters are the same as the ones considered in the previous paragraphs: Neel order pa- rameter n = z, spin accumulation µS = µS y, exchange parameter JS/t = 1, transverse lengths of 40 sites each and an impurity concentration of 0.125. In Figs. 9(a) and 9(b), we vary the conductor length L when ηm = t. The torques τ xN (Fig. 9(a)) and τ yM (Fig. 9(b)) consid- erably decrease as the length of the conductor increases. Spin loss reduces both the spin current and the staggered spin current, resulting in smaller torques on m and n, respectively. In Figs. 9(c) and 9(d), we change the pa- rameter ηm for a disordered region with length L/a = 40. The mean values of the torques τ xN (Fig. 9(c)) and τ yM (Fig. 9(d)) quickly decrease to zero as the parameter ηm increases. We have also considered the torques on n and m for different scattering region sizes and impurity-associated energies when the Neel order parameter points parallel to the transport direction (n = x) and the spin accumu- 020406080100L/a0.00.20.40.60.81.01.21.41.6τxNτxN,00.00.20.40.60.81.01.21.41.6GG0(a)020406080100L/a0.00.20.40.60.81.01.21.41.6τyMτyM,00.00.20.40.60.81.01.21.41.6GG0(b)01234567η/t210123τxNτxN,0210123GG0(c)01234567η/t210123τyMτyM,0210123GG0(d)9 FIG. 9. STTs (blue diamonds) for AFI-NM systems subject to magnetic impurities (η = 0). The static magnetic impuri- ties point in arbitrary directions, with a concentration of 0.125 in the NM. A spin accumulation µS = µS y induces STTs on n = z and m. The transverse size is 40 × 40 lattice sites. In (a) and (b), the conductor length L is varied for a fixed value ηm = t. In (c) and (d), the length of the disordered region is L/a = 40, while the parameter ηm varies. The values are normalized with respect to the STTs τ xN ,0 and τ yM,0 obtained for the systems without disorder. The error bars show the standard deviation of the mean values of the STTs, which are averaged over 15 configurations of the impurities. x FIG. 10. The staggered spin current I ss,x as a function of po- sition x inside the NM from the left (near the interface) to the right (in the NM bulk). The length of the NM is L = 30a, and there are 40 sites in the two transverse directions. The posi- tion x/a = 1 is directly at the AFI-NM interface. We separate three cases: (a) no disorder, (b) only spin-conserving disor- der (η = t, ηm = 0) and (c) only magnetic disorder (η = 0, ηm = t). The staggered spin currents are normalized to their value at the interface, I ss,x x=1. In (b) and (c), the concentrations of the impurities are 0.125. A spin accumulation µS = µS y in the reservoir is assumed. We also investigated how the spin current I s lation is transverse (along y). The torques −τ zN and τ yM (when n = x and µS = µS y) show very similar results compared to the torques τ xN and τ yM (when n = z and µS = µS y), respectively. The STTs are therefore similar for a Neel order parameter that is parallel (x) or trans- verse (z) to the transport direction. This result is rea- sonable because we average over impurity configurations where the impurity spins point in arbitrary directions. x and the staggered spin current I ss x (Eq. (9)) vary in the transport direction inside the NM. In the presence of magnetic im- purities and/or spin-orbit coupling, the spin is not con- served microscopically. The spin current can then be defined in several ways. Our definition (Eq. (5)) only includes electron hoppings. As expected, we find that spin-flip-inducing impurities in the NM reduce the spin current (toward the interface) with position x. The spin current I s x is unaffected by spin-conserving disorder in the NM. In the network circuit model, a staggered current van- ishes within few lattice constants, as shown in Fig. 7 from section II D. However, in the coherent and finite-size regime, the staggered spin current I ss x = I s,A x + 2I AB inside the NM bulk exhibits a slightly different be- havior. In Fig. 10, the staggered spin current I ss x obtains irregular values around 0, on the order of plus/minus I ss x=1, which is the value measured at the interface. At x = 2, there is also a peak in I ss x , which is generally x − I s,B x much larger than the staggered spin current directly at the interface, x = 1, even with or without disorder. The system considered in Fig. 10 has length L/a = 30, n = z, µS = µS y and 40 sites in the two transverse directions. Here, we separated three cases where there is no disorder (Fig. 10(a)), spin-conserving disorder (Fig. 10(b)) and magnetic disorder (Fig. 10(c)). Even for the cases with- out disorder, there are non-zero irregular and fluctuating values for the staggered spin current. IV. CONCLUSIONS The microscopic spin currents at the AFI-metal inter- face determine the STTs. The torques on the magnetiza- tion m and the Neel order parameter n are proportional to the spin current I s 1 and the staggered spin current I ss 1 , respectively. In a reciprocal way, a precessing AF Neel order parameter pumps both a spin current and a stag- gered spin current. In a manner similar to spin pumping in ferromagnets, one can calculate the AF-pumped spin currents by transforming to a rotating spin frame of ref- erence. The STTs on the AFI are affected by spin-conserving and/or spin-flip-inducing disorder in the NM. These ef- fects are particularly prominent when the disorder is very close to the AFI-NM interface. Spin-conserving disorder reduces the torques on n and m, in a manner similar 020406080100L/a0.00.20.40.60.81.01.21.41.6τxNτxN,0(a)020406080100L/a0.00.20.40.60.81.01.21.41.6τyMτyM,0(b)01234567ηm/t2101234τxNτxN,0(c)01234567ηm/t2101234τyMτyM,0(d)051015202530100102030405060Iss,xxIss,xx=1(a)η=0ηm=0051015202530505101520253035Iss,xxIss,xx=1(b)η=tηm=0051015202530x/a505101520253035Iss,xxIss,xx=1(c)η=0ηm=tto Ohm's law. Magnetic impurities result in spin loss and reduced torques. However, for intermediate-strength non-magnetic impurities close to the interface, we find that the torque on the Neel order parameter is enhanced compared with the corresponding torque without impu- rities. For a finite tight-binding system, the staggered spin current I ss x rapidly becomes small and irregular away from the AFI-NM interface, both with or without dis- order. The staggered spin current is not a conserved quantity, even when the spin is conserved. This behavior is well captured by a classical circuit model, where the staggered current quickly decays away from an interface with staggered potentials. The research leading to these results has received fund- ing from the European Research Council via Advanced Grant number 669442 "Insulatronics" as well as the Re- search Council of Norway via Grant no. 239926 and through its Centres of Excellence "QuSpin". Appendix A: Rate of change of the spins The rate of change of an operator O(t) = is calculated via its commutator with ei Ht/ Oe−i Ht/ the Hamiltonian, i.e., ∂t O(t) = i(cid:2) H, O(t)(cid:3). We use (cid:2) S (j) that the localized spins in the magnetic insulator obey , where summation over spatial directions, j = x, y or z for j, j(cid:48), j(cid:48)(cid:48), is implied (
jj(cid:48)j(cid:48)(cid:48) is the Levi-Civita tensor). Then, the rate of change of the localized AFI spins at the interface is r(cid:48) (cid:3) = δr(cid:48),ri
jj(cid:48)j(cid:48)(cid:48) S (j r , S (j (cid:48)(cid:48) r ) ) (cid:48) ∂ S 0,r⊥(t) ∂t =Js1,r⊥ (t) × S 0,r⊥ (t) , (A1) in terms of the itinerant spins s1,r⊥ at the interface (in the layer at x = 1). To find spin continuity equations for the itinerant spins, we write the kinetic part Hhop of the Hamiltonian in Eq. (2), with the notation r = (x, y, z), as z=1 y=1 (cid:0)c Nz(cid:88) Ny(cid:88) ∞(cid:88) r+δz cr(1 − δz,Nz ) + h.c.(cid:1) , rscrs(cid:48)(cid:3) = c† Hhop = − t x=1 † + c † † r+δy cr(1 − δy,Ny ) r+δx cr + c (A2) where h.c. is the Hermitian conjugate. Expressions for the spin currents are found by commuting Hhop with the itinerant spin density. The electron operators obey the r1s1 crs(cid:48)δr2,rδs2,s − c† rscr2s2 δr1,rδs1,s(cid:48), when writing the spin indices (s and so on) explicitly. As an example, the rate of change of spin density in the x-direction follows from commutator (cid:2)c† r1s1 cr2s2 , c† (cid:104) Nz(cid:88) Ny(cid:88) ∞(cid:88) + (1 − δ1,x)(cid:0)c y(cid:48)=1 x(cid:48)=1 z(cid:48)=1 † † r+δx σcr − c rσcr+δx =c † † r−δx σcr − c rσcr−δx (cid:1) , (A3) † † † r(cid:48) cr(cid:48) +δx, c r(cid:48) +δx cr(cid:48) + c c rσcr (cid:105) 10 which via the spin continuity equation (Eq. (A5)) yields our definition of the spin current in Eq. (5). Similarly, we find j±y r and j±z r . We also use that [σj, σj(cid:48)] = 2iεjj(cid:48)j(cid:48)(cid:48) σj(cid:48)(cid:48) to calculate HAF,N,  2 † c rσcr =δ1,xJ S 0,r⊥ × s1,r⊥ , (A4) (cid:20) i  (cid:21) which relates the rate of change of the localized AFI spins to the itinerant spins in the NM (Eqs. (A1) and (A4)). In total, the spin continuity equation is 0 = + j +x ∂sr ∂t + (1 − δ1,y)j−y † r (σ × V m + c where the terms c† purities. r + (1 − δ1,x)j−x r + (1 − δNy,y)j +y r r + (1 − δ1,z)j−z r , (A5) r + (1 − δNz,z)j +z r ) cr + δ1,xJs1,r⊥ × S 0,r⊥ r (σ × V m r ) cr arise from magnetic im- Appendix B: Expectation values 2i (cid:16) rσcr+δ (cid:17)(cid:69) (cid:68) t † r+δσcr − c† c To calculate the STTs, we evaluate spin currents be- tween lattice sites, similar to the expression in Eq. (5), in all three spatial directions. In general, we need to cal- culate the expectation value of the quantity J (r, δ) = , where δ = {±x,±y,±z}. The amplitude at a node inside the scattering region de- pends on the scattering states incoming from the NM reservoir. In this way, the expectation value of quantities inside the scattering region can be found via the dis- tribution function in the reservoir fαβ(E). For a static scattering problem, the field operator inside the scatter- ing region can be expanded in wavefunctions similar to Eq. (16). For the disordered cases, we find all ψnEα(r) inside the scattering region by using the python pack- age KWANT42, which solves the scattering problem by matching of the wave functions. In linear response, one can write the part of the distri- bution function (from Eq. (18)) that describes the spin accumulation as fS(E) = µS 2µS (fFD(E − µ↑) − fFD(E − µ↓)) , (B1) by defining chemical potentials µ↑ (µ↓) for spin up (down) along the direction of the spin accumulation µS = µ↑ − µ↓. By Taylor expansion of the FD dis- tributions to first order around the equilibrium chemical potential and by using −∂fFD/∂Eµ=EF ≈ δ(E − EF) in the low-temperature regime, then fS(E) ≈ (1/2)µSδ(E− EF), which we use in the following. We now write the spin projections s explicitly with the notation ψnEα(r, s) and use Eqs. (16), (17) and (18) to express the quantity J (r, δ) as N⊥(cid:88) (cid:88) (cid:90) ∞ J (r, δ) = t 2i (cid:16) ψ − ψ dE σss(cid:48)fα2α1(E) α1α2ss(cid:48) ε⊥n n=1 † (cid:48) nEα1(r + δ, s)ψnEα2(r, s ) † (cid:48) nEα1 (r, s)ψnEα2(r + δ, s (cid:17) ) , (B2) which we separate in two terms: J (r, δ) ≡ J1(r, δ) + J2(r, δ). The term J1(r, δ) is independent of the spin ac- cumulation and can be evaluated in the low-temperature limit by approximating fFD(E − µ) ≈ Θ(EF − E) as a Heaviside step function: N⊥(cid:88) (cid:88) (cid:90) EF (cid:16) J1(r, δ) = t 2i dE ψ † (cid:48) nEα(r + δ, s)ψnEα(r, s ) αss(cid:48) ε⊥n n=1 † (cid:48) nEα(r, s)ψnEα(r + δ, s ) − ψ (cid:48) σss . (B3) (cid:17) The spin-accumulation-induced term J2(r, δ) is calcu- lated with the distribution from Eq. (B1) in linear re- sponse, which yields J2(r, δ) = t 4i (cid:88) (cid:88) (cid:48) σss (cid:16) n∈prop α1α2ss(cid:48) (σα2α1 · µS) (cid:17) † (cid:48) ψ nEFα1 (r + δ, s)ψnEFα2(r, s † (cid:48) nEFα1 (r, s)ψnEFα2(r + δ, s ) ) × − ψ Appendix C: Gauge transformation 11 The gauge transformation in Sec. II C is explained in detail in the following. When the AFI spins precess, the evolution of the electron operators cr(t) are gov- erned by the Hamiltonian H(t) = HNM + HAF,N(t), where these two terms are given by Eqs. (2) and (24), respectively. To find the spin currents pumped at the AFI-NM interface, we consider that there are no mag- netic impurities in HNM. The classical AFI spin vec- tors at the interface, S 0,r⊥ (t) = ±Sn(t), with length S, are parametrized by the Neel order parameter n(t) = (sin θ0 cos ω0t, sin θ0 sin ω0t, cos θ0) in terms of precession frequency ω0 and the angle θ0 with respect to the z- axis. The sign + (-) of S 0,r⊥ (t) is determined by which sublattice A (B) the spins reside on. In the frame where the spin quantization axis rotates together with n(t), the zation without a hat (), i.e., H =(cid:80) evolution of cr(t) = U (t)cr(t) is described by the Hamil- tonian H. We denote Hamiltonians in the first quanti- similarly, H =(cid:80) rHr,r(cid:48) cr(cid:48), and L = i(cid:88) In a Lagrangian approach, the time evolution of the operators cr(t) is described by the Lagrangian † r(t)∂tcr(t) − c † r(t)Hr,r(cid:48) cr(t) , c (cid:88) Hr,r(cid:48) cr(cid:48). rr(cid:48) c† rr(cid:48) c† (C1) r r rr(cid:48) and the Euler-Lagrange equation , (B4) d dt ∂L † r) ∂(∂tc = ∂L † ∂c r . (C2) where the sum is over the modes that are propagating at the Fermi energy. The torques on the total spin and the spin difference in the AFI are then found by sum- ming currents in the form of J (r, δ) in terms of the total spin current and staggered spin current at the interface, respectively. From Eqs. (12a) and (12b), The evolution of the operators cr(t) is described in a similar way by changing cr → cr in Eq. (C2). We use the unitarity of U (t) and insert cr(t) = U†(t)cr(t) and its Hermitian conjugate into the Lagrangian in Eq. (C1), which yields (cid:88) (cid:88) r⊥ τM = τN = 1 NAF 1 NAF (cid:88) (cid:16) J (x = 1, r⊥, δ = x) , (B5a) ( r⊥ ∈A − ∈B r⊥ ) J (x = 1, r⊥, δ = x) + 2 (J (x = 1, r⊥, δ = y) + J (x = 1, r⊥, δ = z)) (cid:17) , L =i(cid:88) (cid:88) =i(cid:88) − rr(cid:48) r r † c r(t)U U † † r(t)U Hr,r(cid:48)U c ∂tcr(t) + i(cid:88) (cid:88) cr(cid:48)(t) † r † c r(t)∂tcr(t) − rr(cid:48) † c r(t)U ∂U† ∂t cr(t) † r(t) Hr,r(cid:48) cr(cid:48)(t) , c (C3) (B5b) in terms of where the transverse sums with transverse hoppings stop at N−1 for finite lengths N a in the transverse directions. The response coefficients AM ij in Eqs. (34a) and (34b) then follow by evaluating J2(r, δ) in Eqs. (B5a) and (B5b). ij and AN We have numerically verified for some relevant cases that the contributions from J1(r, δ) to the torques vanish when the disorder is spin conserving. However, for some special configurations with magnetic impurities, there may be small but non-zero torques, even in equilibrium (µS = 0). Such equilibrium torques represent an addi- tional anisotropy in the magnetic system. Hr,r(cid:48) = U Hr,r(cid:48)U † − iU ∂U† ∂t δrr(cid:48) . (C4) The Hamiltonian H then includes the gauge potential −iU ∂U ∂t , which is present at all sites. † Appendix D: Umklapp scattering To calculate the torques in Eq. (22), we used peri- odic boundary conditions in the two transverse direc- (cid:107) nynz (E) of tions. Then, the longitudinal momentum k (cid:107) nynz a = 3− E/(2t)− cos( 2πny the propagating states is determined via the energy dis- N )− cos( 2πnz N ). persion cos k We solve the scattering problem via wavefunction matching with the ansatz that the scattering coeffi- cients include both normal and umklapp scattering: 2 (r↑ 1 mymz,nynz ) = Anynz δmy,ny δmz,nz and 2 (r↑ In mymz,nynz ) = Bnynz δmy,¯ny δmz,¯nz . terms of the parameter λ = JS/t, the normal and um- klapp scattering reflection coefficients are mymz,nynz + r↓ mymz,nynz − r↓ 1 Anynz (E) =−1 + λ2ei(k (cid:114) v¯ny ¯nz 1 − λ2ei(k Bnynz (E) =λ vnynz , (cid:107) ny nz )a (cid:107) ¯ny ¯nz−k (cid:107) (cid:107) ny nz )a ¯ny ¯nz +k (cid:107) 2i sin k nynz a (cid:107) (cid:107) ny nz )a ¯ny ¯nz +k 1 − λ2ei(k (D1a) , (D1b) respectively, with quantum numbers ¯ny and ¯nz as defined in the main text. The wavefunction within the scattering region is ψnynzEα(x = 1, r⊥, s) = 12 . (D2) (cid:107) nynz a (cid:107) ¯ny ¯nz +k (cid:107) ny nz )a (−2i) sin k 1(cid:112)hvnynz (cid:104) (cid:105) × − λ(n · σsα)ϕ¯ny ¯nz (r⊥) 1 − λ2ei(k δα,sϕnynz (r⊥) These solutions are such that in the limits JS/t → 0 and JS/t → ±∞, the scattering problem is reduced to hard wall scattering at x = 1 and x = 2, respectively. Ex- pressions for the spin currents are found by evaluating Anynz (E) and Bnynz (E) at the Fermi energy EF = 6t (half-filling). In this case, the umklapp-scattered longi- (cid:107) tudinal momentum is k nynz (EF) and v¯ny ¯nz (EF) = vnynz (EF). The staggered spin current is calculated by using the wavefunctions in Eq. (D2) di- rectly. The spin current can be calculated either from the wavefunctions at the interface or as the real part of mn)∗, where the transverse quan- nynz and tum labels are n → (ny, nz) such that(cid:80) (cid:107) ¯ny ¯nz (EF) = π/a− k (cid:80) mn r↑ g↑↓ = Np − mn(r↓ (cid:80) n → similarly for m. 1 C. Hahn, G. de Loubens, V. V. Naletov, J. B. Youssef, and M. Viret, Europhys. Lett. 108, 57005 O. Klein, (2014). 16 E. Gomonay and V. Loktev, Low Temp. Phys. 40, 17 (2014). 17 V. Baltz, A. Manchon, M. Tsoi, T. Moriyama, T. Ono, 2 H. Wang, C. Du, P. C. Hammel, and F. Yang, Phys. Rev. and Y. Tserkovnyak, arXiv:1606.04284v2 . Lett. 113, 097202 (2014). 3 H. Wang, C. Du, P. C. Hammel, and F. Yang, Phys. Rev. B 91, 220410 (2015). 4 T. Moriyama, S. Takei, M. Nagata, Y. Yoshimura, N. Mat- suzaki, and T. Terashima, Appl. Phys. Lett. 106, 162406 (2015). 5 W. Lin, K. Chen, S. Zhang, and C. L. Chien, Phys. Rev. Lett. 116, 186601 (2016). 6 R. Cheng, J. Xiao, Q. Niu, and A. Brataas, Phys. Rev. Lett. 113, 057601 (2014). 7 R. Cheng, D. Xiao, and A. Brataas, Phys. Rev. Lett. 116, 207603 (2016). 8 X. Mart´ı, I. Fina, C. Frontera, J. Liu, P. Wadley, Q. He, R. J. Paull, J. D. Clarkson, J. Kudrnovsk´y, I. Turek, et al., Nat. Mater. 13, 367 (2014). 9 X. Mart´ı, I. Fina, and T. Jungwirth, IEEE Trans. Magn. 51, 2900104 (2015). 10 H. V. Gomonay and V. M. Loktev, Phys. Rev. B 81, 144427 (2010). 11 O. Gomonay, T. Jungwirth, and J. Sinova, Phys. Rev. Lett. 117, 017202 (2016). 12 H. Gomonay and V. Loktev, Cond. Mat. Phys. 15, 43703 (2012). 13 R. Khymyn, I. Lisenkov, V. Tiberkevich, B. A. Ivanov, and A. Slavin, Sci. Rep. 7, 43705 (2017). 14 O. R. Sulymenko, O. Prokopenko, V. S. Tiberke- and R. Khymyn, vich, A. N. Slavin, B. A. Ivanov, arXiv:1707.07491v1 . 15 T. Jungwirth, X. Mart´ı, P. Wadley, and J. Wunderlich, Nat. Nanotechnol. 11, 231 (2016). 18 X. Mart´ı, B. G. Park, J. Wunderlich, H. Reichlov´a, Y. Kurosaki, M. Yamada, H. Yamamoto, A. Nishide, J. Hayakawa, H. Takahashi, and T. Jungwirth, Phys. Rev. Lett. 108, 017201 (2012). 19 X. Jia, H. Tang, S. Wang, and M. Qin, Phys. Rev. B 95, 064402 (2017). 20 B. G. Park, J. Wunderlich, X. Mart´ı, V. Hol´y, Y. Kurosaki, M. Yamada, H. Yamamoto, A. Nishide, J. Hayakawa, H. Takahashi, et al., Nat. Mater. 10, 347 (2011). 21 L. Smejkal, J. Zelezn´y, J. Sinova, and T. Jungwirth, Phys. Rev. Lett. 118, 106402 (2017). 22 J. Zelezn´y, H. Gao, K. V´yborn´y, J. Zemen, J. Masek, A. Manchon, J. Wunderlich, J. Sinova, and T. Jungwirth, Phys. Rev. Lett. 113, 157201 (2014). 23 J. Zelezn´y, H. Gao, A. Manchon, F. Freimuth, and Y. Mokrousov, J. Zemen, J. Masek, J. Sinova, T. Jungwirth, Phys. Rev. B 95, 014403 (2017). 24 P. Wadley, B. Howells, J. Zelezn`y, C. Andrews, V. Hills, R. P. Campion, V. Nov´ak, K. Olejn´ık, F. Maccherozzi, S. Dhesi, et al., Science 351, 587 (2016). 25 M. J. Grzybowski, P. Wadley, K. W. Edmonds, R. Beard- sley, V. Hills, R. P. Campion, B. L. Gallagher, J. S. Chauhan, V. Novak, T. Jungwirth, et al., Phys. Rev. Lett. 118, 057701 (2017). 26 E. G. Tveten, A. Qaiumzadeh, and A. Brataas, Phys. Rev. Lett. 112, 147204 (2014). 27 S. A. Bender, H. Skarsvag, A. Brataas, and R. A. Duine, Phys. Rev. Lett. 119, 056804 (2017). 28 E. B. Sonin, Adv. Phys. 59, 181 (2010). 29 S. Takei, B. I. Halperin, A. Yacoby, and Y. Tserkovnyak, Phys. Rev. B 90, 094408 (2014). 30 Y. M. Bunkov, E. M. Alakshin, R. R. Gazizulin, A. V. Klochkov, V. V. Kuzmin, V. S. L'vov, and M. S. Tagirov, Phys. Rev. Lett. 108, 177002 (2012). 31 S. Takei and Y. Tserkovnyak, Phys. Rev. Lett. 112, 227201 (2014). 32 H. Skarsvag, C. Holmqvist, and A. Brataas, Phys. Rev. 35 Ø. Johansen and A. Brataas, Phys. Rev. B 95, 220408 (2017). 36 A. Kamra and W. Belzig, arXiv:1706.07118v1 . 37 E. L. Fjaerbu, N. Rohling, and A. Brataas, Phys. Rev. B 95, 144408 (2017). 38 M. Buttiker, Phys. Rev. B 46, 12485 (1992). 39 A. Brataas, Y. V. Nazarov, and G. E. W. Bauer, Phys. 13 Lett. 115, 237201 (2015). 33 A. Qaiumzadeh, H. Skarsvag, C. Holmqvist, A. Brataas, Phys. Rev. Lett. 118, 137201 (2017). 34 Y. Tserkovnyak and H. Ochoa, Phys. Rev. B 96, 100402 (2017). Rev. Lett. 84, 2481 (2000). and 40 Y. Tserkovnyak, A. Brataas, G. E. W. Bauer, and B. I. Halperin, Rev. Mod. Phys. 77, 1375 (2005). 41 G. Tatara, Phys. Rev. B 94, 224412 (2016). 42 C. W. Groth, M. Wimmer, A. R. Akhmerov, and X. Wain- tal, New J. Phys. 16, 063065 (2014). 

1511.06539

1

1511

Temperature dependence of a graphene growth on a stepped iridium surface

We have used scanning tunneling microscopy to study the growth of graphene on a periodically stepped Ir(332) substrate surface, which is a promising route for modification of graphene properties. We have found that graphene continuously extends over iridium terraces and steps. Moreover, new distinctive mesoscopic features of the underlying surface are formed involving large, flat terraces accompanied by groups of narrower steps. The distribution of the newly formed terraces is sensitive to the preparation temperature and only below 800{\deg}C terrace width distribution closer to the intrinsic distribution of clean Ir(332) are found. We propose that the microscopic shape of steps found after graphene formation is strongly influenced by the orientation of graphene domains, where graphene rotated by 30{\deg} with respect to the substrate has a prominent role in surface structuring.

cond-mat.mes-hall

cond-mat

Temperature dependence of a graphene growth on a stepped iridium surface Iva Šrut*, Vesna Mikšić Trontl, Petar Pervan, Marko Kralj Institut za fiziku, Bijenička 46, HR-10000 Zagreb, Croatia Abstract We have used scanning tunneling microscopy to study the growth of graphene on a periodically stepped Ir(332) substrate surface, which is a promising route for modification of graphene properties. We have found that graphene continuously extends over iridium terraces and steps. Moreover, new distinctive mesoscopic features of the underlying surface are formed involving large, flat terraces accompanied by groups of narrower steps. The distribution of the newly formed terraces is sensitive to the preparation temperature and only below 800°C terrace width distribution closer to the intrinsic distribution of clean Ir(332) are found. We propose that the microscopic shape of steps found after graphene formation is strongly influenced by the orientation of graphene domains, where graphene rotated by 30° with respect to the substrate has a prominent role in surface structuring. * Tel/Fax: +385 1469 8889, E-mail address: isrut@ifs.hr, 1 1. Introduction Graphene is a promising material for future application in nano-electronic devices [1, 2]. The extensive research of epitaxial graphene on different substrates is an open route to the large-scale applications of graphene. Graphene can be grown on transition metals by a chemical vapor deposition (CVD) technique [3], and it can be transferred from a metal to other supports, e.g. to polymer sheets to make a flexible transparent touch screen display, where graphene has a role of a transparent electrode [4]. Possibility of such manipulation makes CVD grown graphene attractive for a number of applications. Many of the desirable properties of graphene are related to its electronic structure with conical , * bands (Dirac cone) of vertices touching in a single point at the Fermi level. This makes graphene a gapless semiconductor which is a limiting factor for some applications of graphene in electronics where a sizeable band gap at the Fermi energy is required. For that reason, a lot of effort is directed towards manipulation of the electronic structure of graphene around the Fermi energy. One possible route, which is subject of this work, relies on altering the graphene band structure by means of an additional periodic potential, i.e. graphene superlattice. Calculations indicate that the lateral superlattice structures may lead to unexpected and potentially useful charge carrier behavior, e.g. gap openings or Fermi velocity anisotropy [5]. Graphene superlattices have 2 been realized on well defined single crystal substrate surfaces as a consequence of the mismatch of graphene and substrate lattices which leads to the so called moiré superstructures. For graphene on Ir(111) it has been shown that the moiré superstructure adds a long range superperiodic potential to graphene, which opens minigaps in the Dirac cone [6]. For the same system it was demonstrated that the adsorption of hexagonal array of metal-clusters gives rise to highly anisotropic Dirac cones [7]. A one dimensional (1D) type of graphene superlattice can be achieved by adding a 1D periodic potential to graphene, in which context it is appealing to consider a periodically stepped metal surface as a substrate for the growth of graphene. Motivated by this consideration, in our work we used scanning tunneling microscopy (STM) to explore structures of graphene prepared under various growth conditions on a stepped Ir(332) surface. Iridium was chosen as a substrate because Ir(111) is known to support growth of single layer graphene which can be tuned to uniform orientation over the entire sample surface [8, 9], extending across step edges [10]. For many metals the issues of uniform thickness and varying graphene rotation with respect to the substrate are often limiting factors for graphene quality at large scales. The uniform orientation cannot be achieved e.g. on foils due to their polycrystallinity [11] or on a single crystal Pt(111) due to very weak binding of graphene 3 [12]. On Ir(111), macroscopically aligned graphene is achieved only by the careful optimization of the growth procedures [6, 8], whereby the substrate temperature turns to be the main growth parameter. The temperatures above 1230°C are needed to grow graphene aligned to the substrate with perfectly matched graphene and Ir directions, which is referred to as the R0 [13], implying 0° rotation. For this phase, approximately 10% difference in lattice parameters of graphene and Ir results in an incommensurate hexagonal (9.32×9.32) moiré superstructure with a repeat distance of 2.53 nm [6]. Thus the R0 moiré unit cell consists of about 200 carbon atoms positioned differently with respect to atoms of an Ir lattice. With respect to the carbon atom stacking, the moiré cell is prominently divided into atop, fcc, and hcp regions [14]. The atop-type region, where the iridium atom lies right in the centre of the carbon ring, defines the corners of the moiré cell [14]. For lower growth temperatures, graphene on Ir(111) grows in many minor different orientations where specific orientations of 14°, 18.5° and 30° appear along side with 0° [13]. Most frequent minor contribution is of 30° oriented phase, i.e. with 30° rotated graphene and Ir directions, which is referred to as the R30, and has a moiré repeat distance of 0.5 nm [13, 9]. It should be noted that besides lattice mismatch, which is fixed by the choice of the substrate, the relative rotation of graphene with respect to the substrate leads to 4 02111100211110different sizes of moiré unit cells, with the R0 having the largest repeat distance. Our study indicates that the growth of graphene on a stepped Ir(332) surface alters its original narrow terrace width distribution (TWD). In terms of surface morphology, our result show the formation of two different areas: (i) compact areas with narrower step spacing (one to three atoms wide) wide up to 15 nm, so called step bunches, and (ii) areas consisting of single flat (111) terraces up to 19 nm wide. We find that graphene extends across step edges in the shape of domains of different orientations and sizes, ranging up to 100 nm. The detailed analysis indicates that graphene domains consisting of the R30 graphene result in a morphology with pronouncedly straight Ir step edges. 2. Experimental The experiments were performed in two ultra high vacuum (UHV) setups. Both UHV systems were operated at a base pressure of 5×10-10 mbar. One setup was used for low energy electron diffraction (LEED) measurements, and the other for the STM characterization of clean and graphene covered Ir(332) surface. The STM measurements were performed at room temperature. The STM was calibrated by measurements on the HOPG sample and the STM images were processed by a WSxM software [15]. The applied bias voltage 5 to the sample was in the range of 130 mV to 300 mV, while the tunneling current was around 0.5 nA for all images except for the atomically resolved STM image in Fig. 2a (It=2.01 nA). Same Ir(332) sample was used in both setups, a crystal 6 mm in diameter, of 99.99% purity, polished with roughness <0.03 μm and with an orientation accuracy <0.1°. In both setups the sample was heated by e-beam heating using a hot filament at negative potential near to the grounded sample. In the system for LEED measurements a C type thermocouple was directly connected to the sample, whereas, in the STM system, sample temperature was calibrated by the C type thermocouple attached to the sample with respect to the heating time and power used and, additionally, a values measured by a K type thermocouple attached at the sample plate. The C type thermocouple was then removed from the sample to enable multiple graphene preparations and transfer from the sample preparation manipulator to STM. 2.1 Substrate cleaning Ir(332) sample was cleaned by using cycles consisting of 50 min of 0.8 keV argon-ion sputtering while the sample was kept at 200°C, followed by 10-25 min of annealing in oxygen partial pressure of 1×10-7 mbar at 800°C, and a final 5-15 min annealing in UHV at 850°C. Each cycle was finished by uniformly cooling the sample to room temperature over a period of 30-40 min. After five such cycles typically a sharp 6 LEED pattern and large scale STM images confirmed clean surface with ordered array of steps. 2.2 Graphene growth Graphene preparation methods used in this work involved two well known procedures. One is a room- temperature hydrocarbon adsorption followed by decomposition at a fixed, elevated temperature, referred as a temperature-programmed growth (TPG). Another procedure, CVD, consists of exposing the hot substrate to hydrocarbon gas. Growth conditions (temperatures, pressures, hydrocarbon doses) described below were used by taking known conditions for preparation of graphene on Ir(111), but not crossing 1120°C to avoid irreversible structural transformation of the nominal stepped surface. Note that this condition limits us to temperatures below 1230°C needed to achieve single domain of the R0 graphene [13, 14]. We have prepared graphene either by (i) CVD or by (ii) one TPG cycle followed by CVD. In this work, we have focused to the correlation of a temperature used during CVD cycle and a resulting graphene structures. We did not notice any significant influence of the hydrocarbon pressure parameter in the range 5×10-9 to 5×10-8 mbar, used in our work. Preparation (i) was performed at two sample temperatures, 775°C and 880°C, with corresponding ethylene pressures and duration presented in Table 1. Preparation (ii) involved dosing 7 the ethylene at the pressure of 2×10-8 mbar for 30 s, followed by annealing to high temperatures which corresponded to the ones of subsequent CVD cycles. The CVD was performed at three different sample temperatures, 770°C, 820°C and 930°C (see Table 1). Graphene growth was finalized by short annealing at the CVD temperature and slow cool down (30-45 min) to room temperature. The combined TPG+CVD preparation was motivated by findings that such procedure produces a uniform orientational order of graphene on flat Ir(111) surface [8]. preparation temperature method CVD TPG+CVD [°C] 880 775 770 820 930 pressure [mbar] 5x10-8 1.5x10-8 3x10-8 2x10-8 5x10-9 time [min] corresponding figure 21 12 25 20 90 1a 1c 3a 3b 3c Table 1. Overview of preparation parameters used in this work with a reference to figures with STM images. 3. Results 3.1 Clean Ir(332) Structural model of an ideal Ir(332) surface is displayed in Fig. 1a. The model shows (111) terraces 1.25 nm wide, consisting in width of atoms, and (111) steps with edges parallel to the high-symmetry direction. The step height of 0.22 nm corresponds to the interlayer distance between the 8 315110(111) planes of iridium. The LEED pattern of clean Ir(332) surface, Fig. 1b, reproduces a hexagonal atomic arrangement of the surface. Also the additional splitting of diffraction spots is visible. This, so called, spot splitting reflects the periodicity of ordered array of steps corresponding to (1.3±0.1) nm, matching the expected value. Figure 1. a) Structural model of a (332) fcc-crystal surface generated by Surface Explorer software1. b) LEED pattern taken at the electron energy of 64 eV, c) STM image of clean Ir(332) with an inset showing 1.5 times magnified region marked by a white dashed rectangle, and d) TWD of clean Ir(332). Fig. 1c, which shows a large scale STM image, confirms that the Ir(332) surface was well ordered and clean. Inset of Fig. 1c illustrates that small local deviation in terrace widths may be 1 http://surfexp.fhi-berlin.mpg.de 9 seen after cleaning process. Analysis of several large scale STM images taken at different surface positions gives a narrow TWD with a mean value of (1.1±0.5) nm (Fig. 1d). TWD is shown as the probability P(w) of finding terrace of the width w. Note a small discrepancy between the mean width values obtained by STM and LEED. This can be attributed to the local variation of the terrace width e.g. due to the finite processing accuracy in crystal orientation of <0.1°. Since STM is a local technique it will be more susceptible to local variations than LEED, which averages over macroscopic sample scales. 3.2 Graphene growth and morphology Graphene covered areas were easily identified by observing the honeycomb carbon lattice (with unit cell parameter of 2.45 Å) and, at larger scales, the moiré structure. Both structural features are seen in Fig. 2a, where a moiré unit cell is marked by a white rhombus. Carbon rows of graphene are rotated by 41.2° with respect to the moiré lattice, as marked in Fig. 2a. This rotation serves as a „magnifying lense“ to determine graphene rotation with respect to iridium, and by using formalism developed in Ref. [14] we calculate it in this case be 5.7°±0.1°. This is in a good agreement with the measured moiré unit cell of (1.80±0.05) nm which corresponds to the rotation of 5.8°±0.4° [14]. 10  Figure 2. a) STM image of graphene area showing atomic and moiré resolution. b) STM image of a Ir(332) surface partially covered by graphene. Panels below are schematic models of graphene covered and bare Ir(332). TWD of c) graphene grown by CVD method at 775°C and d) graphene grown by TPG+CVD method at 770°C. Fig. 2b shows an STM image of Ir(332) partially covered with graphene. Graphene covered areas obviously have different morphology compared to clean Ir(332), which is readily seen in Fig. 2b and schematically illustrated in the panels below it. Graphene growth leads to the formation of a number of wide (111) terraces followed by groups of narrower 11 steps (one to three atoms wide), i.e. a step bunch. Graphene grows continuously over terraces and step bunches. This can be confirmed by analyzing the orientation and the moiré repeat distance of graphene which extends over several terraces and step bunches. In Fig. 3c inset one such analysis was done for moiré with a repeat distance of 1.36 nm. In this work we have observed that TPG+CVD growth can lead to a narrower TWD compared to growing graphene just by CVD, with the mean terrace width closer to the value of the clean Ir(332) surface (cf. Fig. 2c, 2d). Note that our TWD data in Figs. 2 and 3 include only the widths of the wide (111) terraces, and not of the very narrow steps in step bunches. 3.3 Temperature dependence of graphene growth Since the motivation for our work is the modulation of the electronic structure of graphene by uniaxial periodic structure-induced potential, in the remaining part of the paper we focus to TPG+CVD growth procedures with narrower TWD’s. In particular, we focus to several different temperatures during the CVD cycle: 770°C, 820°C and 930°C, presented by large scale STM images in Fig. 3a, 3b and 3c, respectively. One can observe a general feature that the (111) terraces are getting wider as the preparation temperature increases. 12  Figure 3. STM images of graphene covered Ir(332) for different preparation temperatures (a-c), and corresponding TWD’s (d-f). Inset in c) shows 1.5 times magnified region marked by white dashed rectangle. The black line corresponds to the diagonal direction across the moiré unit cell. 13 In Figs. 3d, 3e, 3f, this is clearly seen from TWD’s for corresponding temperatures. The mean terrace width shifts from 2.03 nm for 770°C to 7.91 nm for 930°C. In addition to this, TWD data also reveal significant smearing of distributions for higher preparation temperature. The smearing effect enables us only to compare TWD’s in terms of overall range and mean terrace widths. Pronounced peaks slightly below and around 1 nm are seen in TWD’s for all preparation temperatures and are due to a large number of narrow terraces, examples of which are marked by arrows on Fig. 3c. 3.3 Analysis of rotational domains and moiré structures The LEED pattern in Fig. 4a corresponds to the sample prepared under the same conditions as the one shown in STM image in Fig. 3c. The diffraction pattern reveals that the surface is covered by graphene in several different orientations. Dominant carbon diffraction intensities are grouped around 0° orientation (R0) with an angular spread of ±6°, and around 30° orientation (R30) in a similar angular spread but with distinctive spots at symmetric 26° and 34° (R26) positions. In addition, a faint diffraction ring is visible, suggesting also any other possible orientation of graphene. The diffraction maxima corresponding to Ir(111) are now visible in a hexagonal pattern but without characteristic spot splitting of uniformly ordered stepped surface. This is consistent with rather wide TWD from Fig. 3f. 14  Figure 4. a) LEED pattern taken at the electron energy of 59 eV, after growing graphene by TPG+CVD method at temperature of 930°C. b) STM image of two connected rotational domains of graphene corresponding to R0 and R30 groups. c) STM image of the R30 graphene showing atomic resolution at terraces and step bunches where three kinks are marked by blue lines. d) STM image showing several domains within the R0 group graphene with clearly visible moiré patterns and four kinks marked by blue lines. Similar to 930°C, the LEED pattern for preparation temperature of 1120°C (not shown) reveals moiré pattern around the R0 15 graphene spot and a faint but visible contribution at the R30 and the R26. LEED patterns provide us with relevant information on the surface structure averaged over macroscopic areas. In addition, STM enables us to characterize microscopic areas of certain orientation. The analysis of the moiré pattern is especially helpful in that sense (cf. Fig. 2a). Moreover, STM images can be used to characterize the appearance of underlying Ir steps and step edges. Figure 4b shows the STM image of graphene covered surface with two distinct graphene domains (separated by blue dashed line). The upper section of the image shows graphene domain orientated within the R30 group, while the lower section exhibit graphene domain orientated within the R0 group. It is obvious that graphene in the upper part of the image shows pronouncedly strait step edges with only few kinks, while the lower graphene domain shows distinctly different, less regular step edges with frequent step branching and kinks. The observation that graphene domains around the R30 form straighter step edges, is additionally corroborated by the analysis of other large scale STM images. Figure 4c shows a single R30 domain, and Fig. 4d shows several domains with orientations close to the R0. On both images, the kinks (marked by blue lines) are found at step edges. The kinks on a single graphene domain (cf Fig. 4c) are all of the same size, in a direction perpendicular to the steps. 16 Additionally, the moiré superstructure is clearly visible in Fig. 4d, where the bright protrusions correspond to the atop regions that corner moiré cells (marked by black net of rhombuses). From Fig. 4d it is evident that the step edges are terminated by these atop regions, and further, that kinks as well occur at the atop positions. 4. Discussion In this work it is demonstrated that it is possible to cover the entire Ir(332) surface by a monolayer graphene. LEED patterns showed spots that reveal the presence of graphene with several different orientations, grouped around the R0 and the R30 rotations (cf. Fig. 4a). LEED patterns also revealed a loss of uniform order of pristine Ir(332) surface through deterioration of characteristic diffraction spot splitting. STM images with atomic and moiré resolution accordingly corroborated the presence of graphene in various orientations that spread over several terraces (cf. Fig. 3c). Moreover, STM characterization clearly revealed mesoscopic restructuring of the surface upon graphene formation accompanied by two new structural features: (i) wide (111) terraces, and (ii) groups of narrower steps - step bunches, consisting of steps one to three atoms wide. Such step bunches do not occur on perfectly clean Ir(332). 17 The mesoscopic restructuring of the supporting surface does not come as a surprise for it is known from the studies of graphene growth on flat Ir(111) that terrace edges play an active role in graphene nucleation and growth and that, in fact, graphene islands reshape Ir steps during the growth [16]. Such substrate step-edge mobility seems to be general phenomenon for graphene growth on metal surfaces. For graphene which strongly binds to Ru(0001) surface, a creation of wider terraces and multisteps (facets) on substrate has been observed during graphene growth for the appropriate growth conditions [17]. Graphene covered wide (0001) terraces but did not grow over facets, which was explained by thermodynamically stable state of the surface consisting of large graphene covered areas and facets [17]. Note that compared to faceting on Ru(0001), graphene on Ir(332) exhibits formation of groups of several narrow steps, which are, in contrast to Ru facets, all covered by graphene. This difference might be due to the binding character of graphene to Ir and Ru substrates, i.e. it is weak van der Waals-like binding to Ir and strong chemisorption binding to Ru [18, 19]. Observed mesoscopic structure of graphene on Ir(332) could be generally described as an interplay of several parameters representing energy costs or benefits for the system. On the one hand, the major energy benefit in the restructured system comes from the van der Waals binding of graphene to 18 created, large, flat (111) terraces. On the other hand, energy costs for the system include the energy needed to bend graphene over step bunches and to cause substrate restructuring, which is driven by temperature and ethylene decomposition chemistry. Although our results indicate that graphene growth on Ir(332) leads to substrate restructuring, it does not necessarily mean that it is not possible to obtain graphene with very narrow TWD’s over limited scales. Our results show that the growth temperature can be used to tune the TWD. We found that lower growth temperatures lead to comparably narrower TWD, while high growth temperatures of 930°C induce broad terraces with no obviously preferable terrace width (cf. Fig 1d and Fig. 3). Intuitively, this is because higher temperatures lead to higher mobility of surface atoms, and consequently, surface restructuring. For the lowest growth temperature of 770°C, TWD was centered around 2.03 nm and in the range between 0.5nm and 5.5nm. However, additional lowering of the growth temperature below 700°C is expected to lead to higher contribution of the dehydrogenated carboneous species [20], which is not suitable for our goal. To get more insight on how lower synthesis temperatures produce graphene with more favorable TWD, we turned our attention to the analysis of moiré structures. LEED patterns obtained for graphene grown at 930°C (Fig. 4a), and 19 1120°C indicate significant contribution of the R30 graphene. The R30 contribution seems to be higher than is to be expected for similar growth preparation conditions on the Ir(111) surface [13, 16]. We believe this is due to much larger number of steps on Ir(332) which play a role in graphene nucleation, and which on flat (111) surfaces are not as nearly abundant [16]. Moreover, our STM data indicate that on Ir(332) the R30 and close rotations become much more abounded at lower temperatures which as a trend was also observed for graphene on Ir(111) [9]. From the analysis of STM images, we suggest that the structure of step edges, found at the surface, is on a microscopic scale related to rotational domains. Our STM images (cf. Fig. 4d) showed that the terraces, as well as the structural details at the terrace edges, such as kinks, are defined by the atop regions of the moiré structure of graphene. This is consistent with previous STM studies which showed that graphene covered Ir(111) terraces are always terminated by an integer number of atop regions [10], and that graphene islands and flakes as well consist of the integer number of atop regions which corner moiré cells. [21, 22]. Consequently, the moiré unit cell can be regarded as the smallest „building block“ of graphene. The size of that building block is tuned by the rotation of graphene on iridium. The R30 rotation of graphene has substantially smaller moiré parameter than the R0 and the 20 structural details, such as kinks, are therefore much smaller and less pronounced for the R30 graphene. Hence, one might expect that the size of the building block plays a role in an appearance of step edges, with seemingly straighter edges formed for smaller building blocks, for example for the R30 graphene. It is indeed distinctive observation that the R30 graphene and close to it are accompanied by pronouncedly long and straight step edges, compared to the graphene orientations grouped around the R0 (cf. Fig. 4b). The average R30 domain has less step branching, and fewer, smaller kinks than the R0 domain. Correspondingly, these well defined terraces are a desirable feature in an attempt to impose an anisotropic periodic step potential which should affect the electronic band structure of graphene. All this suggest that using lower growth temperatures and forcing the growth of the R30 graphene on stepped Ir is a promising route to induce periodic modulation in graphene. 5. Conclusion We have studied the growth on graphene on Ir(332) at different preparation temperatures using STM. Our results show that growing graphene at 770°C produces sample with relatively narrow TWD, where on average terraces are two times wider than that on an ideal Ir(332). Moreover, as a result 21 of a rather low growth temperature, the areal contribution of the R30 graphene is substantial. As our analysis of the R30 areas indicates, graphene not only extends across Ir steps but also forms more uniform terrace edges. All this suggests that low growth temperatures (below 800°C) are advantageous for controlling the periodic structuring of graphene on Ir(332). Although single R30 domains, characterized in this work, are not extended over millimeter scales, which would be an optimal sample apply area-averaging techniques, areas of periodic features that we have formed should induce effects of a one dimensional periodic potential in the electronic band structure of graphene at well defined microscopic scales. This calls for further spectroscopic (scanning tunneling spectroscopy and μ- angle resolved photoemission spectroscopy) characterizations of this system in search for characteristic anisotropic effects in the band structure of graphene. Acknowledgements We thank M. Petrović and I. Delač Marion for useful discussion and help with the measurements, and T. Michely for valuable comments and suggestions. We gratefully acknowledge financial supports by the UKF (grant No. 66/10) and the MZOS (project No. 035-0352828-2840). 22 References: [1] Geim K, Novoselov KS. The rise of graphene. Nat Mater 2007; 6:183-91. [2] Avouris P, Chen Z, Perebeinos V. Carbon-based electronics. Nat Nanotec 2007; 2(1):605-15. [3] Wintterlin J, Bocquet ML. Graphene on metal surfaces. Surf Sci 2009; 603:1841–52. [4] Bae S, Kim H, Lee Y, Xu X, Park JS, Zheng Y, et al. Roll- to-roll production of 30-inch graphene films for transparent electrodes. Nat Nanotec 2010; 5:574-8. [5] Park CH, Yang L, Son YW, Cohen ML, Louie SG. Anisotropic behaviours of massless Dirac fermions in graphene under periodic potentials. Nature Phys 2008; 4:213-7. [6] Pletikosic I, Kralj M, Pervan P, Brako R, Coraux J, N'Diaye AT, et al. Dirac Cones and Minigaps for Graphene on Ir(111). Phys Rev Lett 2009; 102(5):056808-4. [7] Rusponi S, Papagno M, Moras P, Vlaic S, Etzkorn M, Sheverdyaeva PM, et al. Highly Anisotropic Dirac Cones in Epitaxial Graphene Modulated by an Island Superlattice. Phys Rev Lett 2010; 105(24):246803-4 [8] van Gastel R, N’Diaye AT, Wall D, Coraux J, Busse C, Buckanie NM, et al. Selecting a single orientation for millimeter sized graphene sheets. Appl Phys Lett 2009; 95:121901-4. 23  [9] Hattab H, N’Diaye AT, Wall D, Jnawali G, Coraux J, Busse C, et al. Growth temperature dependent graphene alignment on Ir(111). Appl Phys Lett 2011; 98:141903-6. [10] Coraux J, N'Diaye AT, Busse C, Michely T. Structural coherency of graphene on Ir(111). Nano Lett 2008; 8(2):565- 70. [11 ] Li A, Cai W, An J, Kim S, Nah J, Yang D, et al. Large- area synthesis of high-quality and uniform graphene films on copper foils. Science 2009; 324:1312-4. [12] Preobrajenski B, Ng ML, Vinogradov AS, Mårtensson N. Controlling graphene corrugation on lattice-mismatched substrates. Phys Rev B 2008; 78(7):073401-4. [13] Loginova E, Nie S, Thürmer K, Bartelt NC, McCarty KF. Defects of graphene on Ir(111): Rotational domains and ridges. Phys Rev B 2009; 80(8):085430-8. [14] N'Diaye AT, Coraux J, Plasa TN, Busse C, Michely T. Structure of epitaxial graphene on Ir(111). New J Phys 2008; 10:043033-16. [15] Horcas I, Fernandez R, Gomez-Rodriguez JM, Colchero J, Gomez-Herrero J, Baro AM. WSXM: A software for scanning probe microscopy and a tool for nanotechnology. Rev Sci Instrum 2007; 78:013705-8. 24  [16] Coraux J, N'Diaye AT, Engler M, Busse C, Wall D, Buckenie N, et al. Growth of graphene on Ir(111). New J Phys 2009; 11:023006-22. [17] Günther S, Dänhardt S, Wang B, Bocquet ML, Schmitt S, Wintterlin J. Single terrace growth of graphene on a metal surface. Nano Lett 2011; 11:1895-1900. [18] Busse C, Lazić P, Djemour R, Coraux J, Gerber T, Atodirese N, et al. Graphene on Ir(111): Physisorption with Chemical Modulation. Phys Rev Lett 2011; 107(3):036101-4. [19] Wang B, Bocquet ML, Marchini S, Günther S, Wintterlin J. Chemical origin of a graphene moiré overlayer on Ru(0001). Phys Chem Chem Phys 2008; 10:3530-4. [20] Lizzit S, Baraldi A. High-resolution fast X-ray photoelectron spectroscopy study of ethylene: From chemisorption to dissociation and graphene formation. Catal Today 2010; 154:68-74. [21] Phark S, Borme J, Vanegas AL,Corbetta M, Sander D, Kirschner J. Atomic structure and spectroscopy of graphene edges on Ir(111). Phys Rev B 2012; 86(4):045442-4. [22] Lu J, Yeo PSE, Gan CK, Wu P, Loh KP. Transforming C60 molecules into graphene quantum dots. Nat Nanotechnol 2011; 6:247-52. 25  List of captions for Figures and Tables: Table 1. Overview of preparation parameters used in this work with a reference to figures with STM images. Figure 1. a) Structural model of a (332) fcc-crystal surface generated by Surface Explorer software1. b) LEED pattern taken at the electron energy of 64 eV, c) STM image of clean Ir(332) with an inset showing 1.5 times magnified region marked by a white dashed rectangle, and d) TWD of clean Ir(332). Figure 2. a) STM image of graphene area showing atomic and moiré resolution. b) STM image of a Ir(332) surface partially covered by graphene. Panels below are schematic models of graphene covered and bare Ir(332). TWD of c) graphene grown by CVD method at 775°C and d) graphene grown by TPG+CVD method at 770°C. Figure 3. STM images of graphene covered Ir(332) for different preparation temperatures (a-c), and corresponding TWD’s (d-f). Inset in c) shows 1.5 times magnified region marked by white dashed rectangle. The black line corresponds to the diagonal direction across the moiré unit cell. Figure 4. a) LEED pattern taken at the electron energy of 59 eV, after growing graphene by TPG+CVD method at temperature of 930°C. b) STM image of two connected rotational domains of graphene corresponding to R0 and R30 groups. c) STM image of the R30 graphene showing atomic 26  resolution at terraces and step bunches where three kinks are marked by blue lines. d) STM image showing several domains within the R0 group graphene with clearly visible moiré patterns and four kinks marked by blue lines. 27 

1609.04619

1

1609

Spin pumping and torque statistics in the quantum noise limit

We analyze the statistics of charge and energy currents and spin torque in a metallic nanomagnet coupled to a large magnetic metal via a tunnel contact. We derive a Keldysh action for the tunnel barrier, describing the stochastic currents in the presence of a magnetization precessing with the rate $\Omega$. In contrast to some earlier approaches, we include the geometric phases that affect the counting statistics. We illustrate the use of the action by deriving spintronic fluctuation relations, the quantum limit of pumped current noise, and consider the fluctuations in two specific cases: the situation with a stable precession of magnetization driven by spin transfer torque, and the torque-induced switching between the minima of a magnetic anisotropy. The quantum corrections are relevant when the precession rate exceeds the temperature $T$, i.e., for $\hbar \Omega \gtrsim k_B T$.

cond-mat.mes-hall

cond-mat

a Spin pumping and torque statistics in the quantum noise limit P. Virtanen1, 2 and T.T. Heikkila2 1NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, I-56127 Pisa, Italy 2University of Jyvaskyla, Department of Physics and Nanoscience Center, P.O. Box 35, 40014 University of Jyvaskyla, FINLAND We analyze the statistics of charge and energy currents and spin torque in a metallic nanomagnet coupled to a large magnetic metal via a tunnel contact. We derive a Keldysh action for the tunnel barrier, describing the stochastic currents in the presence of a magnetization precessing with the rate Ω. In contrast to some earlier approaches, we include the geometric phases that affect the counting statistics. We illustrate the use of the action by deriving spintronic fluctuation relations, the quantum limit of pumped current noise, and consider the fluctuations in two specific cases: the situation with a stable precession of magnetization driven by spin transfer torque, and the torque- induced switching between the minima of a magnetic anisotropy. The quantum corrections are relevant when the precession rate exceeds the temperature T , i.e., for Ω (cid:38) kBT . Spin transfer torque, angular momentum contributed by electrons entering a magnet, can be used to control magnetization dynamics via electrical means, as demon- strated in many experiments. [1 -- 3] Often the effect can be described by considering the ensemble average mag- netization dynamics, or taking only thermal noise into account. [4] The spin transfer torque is in general also a stochastic process, but at bias voltages large enough to drive the magnetization, it is not necessarily Gaussian nor thermal, [5] especially at cryogenic temperatures. The statistical distribution of electron transfer and the associated torque in magnetic tunnel junctions can be de- scribed by counting statistics, [6] via a joint probability distribution of charge, energy, and spin transferred into the magnet during time t0, Pt0(δn, δE, δs). The distribu- tion is conditional on the magnetization dynamics during time t0, which necessitates consideration of back-action effects. Here we construct a theory describing the probabil- ity distribution for electron transfer via a Keldysh action (Eq. (2)) describing a metallic magnet with magnetiza- tion M , coupled to a fermionic reservoir (another ferro- magnetic metal), illustrated in Fig. 1. In the presence of a bias voltage in the reservoir, this coupling may lead to a stochastic spin transfer torque affecting the magneti- zation dynamics. Unlike some of the earlier discussions of counting and spin torque statistics [7 -- 9], we follow the approach of Ref. [10] and retain geometric phase factors in the derivation of the generating function. This be- comes relevant in the quantum limit Ω > kBT where the precession rate Ω is large compared to the tempera- ture T . To study the implications, we suggest two specific set- tings (Fig. 1b,c), characterized by opposite regimes of the external field Hext and anisotropy field Han. When Hext (cid:29) Han, a suitably chosen voltage drives the magnet into a stationary precession with rate Ω around the direc- tion of Hext. [1, 11 -- 13] This precession pumps charge [14] and heat into the reservoir, along with the direct charge and heat currents due to the applied voltage. The noise FIG. 1. (a) Tunnel junction between magnetic materials with free (F (cid:48)) and fixed (F ) magnetizations. The total spin S = VM /γ in F (cid:48) precesses at angular frequency Ω around the z-axis. As described by Eq. (3), the motion pumps charge, spin and heat currents through the junction, and the back- action spin transfer torque τ drives a change in the tilt angle θ. (b -- c) Schematic of effective magnetic potential energy, in the presence of an external field Hext and large spin transfer torque, or, in the presence of a magnetic anisotropy Ωan = γHan. of these currents depends on the intrinsic noise of the pumped current and, at low frequencies, also on the fluc- tuations of the magnetization, driven by the spin torque noise. The opposite limit Han (cid:29) Hext is the one rele- vant for memory applications, as the spin transfer torque can be used to switch between the two stable magnetiza- tion directions [15, 16]. Our approach allows finding the switching rate at any temperature and voltage, also for kBT (cid:28) Ω. Besides the average currents and noise, the Keldysh action allows us to calculate the full probability distri- bution Pt0 (δn, δE, δs) of transmitted charge δn, energy δE, or change δMz = Sγδs/V of the z-component of magnetization in a nanomagnet with volume V and spin S (cid:29) 1, within a long measurement time t0. Here γ is the gyromagnetic ratio. The precise distribution depends on the exact driving conditions and the parameters of the setup. However, symmetries constrain the probability distribution, leading to a spintronic fluctuation relation (here and below, kB =  = e = 1) Pt0 (δn, δE, δs) = −1 F −T eV δn/TF eδE(T −1 F (cid:48) )e−Ωδs/TF P (cid:48) t0 (−δn,−δE, δs), (1) t0 where P (cid:48) corresponds to the case with reversed mag- netizations. As in fluctuation relations presented ear- lier [7, 17 -- 21], this allows for a direct derivation of Onsager symmetries, thermodynamical constraints, and fluctuation-dissipation relations, valid for the coupled charge-spin-energy dynamics (see Appendix). and the to tunnelling, Generating function. Consider a magnetic tun- junction depicted in Fig. 1. The spin transfer nel torque due corresponding counting statistics can be described by a Keldysh action obtained by integrating out conduction elec- trons in F and F (cid:48). [8, 10] We apply the approach of Ref. 10 to the characteristic function Z(χ, ξ) = (cid:104)ei[NF (t0)χ0+(HF (t0)−µ)ξ0]e−i[NF (0)χ0+(HF (0)−µ)ξ0](cid:105) describing the change in particle number N and internal energy HF in the ferromagnetic lead F . t0 (cid:29) 1/T, 1/V , [20, 22] this results to the action S = S0 + ST , where the Berry phase for total spin S = VM/γ. Moreover, the tunneling action is −∞ dt [−2 ψq −(cid:80)±(± φcl + φq) cos(θcl ± θq)] is (cid:90) ∞ S0 = S(cid:82) ∞ In the long-time limit, ST = iW2 dt dt(cid:48) d
2π −∞ Tr P (t) GF (cid:48)(t − t(cid:48)) P (t(cid:48))† GF (
) , (2) where P (t) = ei(
−V )tei[χ(t)+(
−µ)ξ(t)]γx/2 R(t) contains the bias voltage V , and the charge and energy count- ing fields χ(t) and ξ(t). The rotation matrix R(t) = e−i φ(t)σz/2e−iθ(t)σy/2e−i ψ(t)σz/2 describes the direction of the magnetization S = (cos φ sin θ, sin φ sin θ, cos θ)S in terms of Euler angles θ and φ. Keldysh fields are in the basis [23] φ = φcl + φq γx, where γx is a Pauli matrix. Below, we fix the gauge [10] so that ψq = −φq cos θcl, ψcl = − φcl cos θcl. We assume a spin and momentum in- dependent tunneling matrix element W . The conduction electrons are described by Keldysh -- Green functions G, with the exchange field of F (cid:48) always parallel to z in the rotating frame, GR F (cid:48)(
, k) = [
− ξk + hF (cid:48)σz]−1. Consider now the situation depicted in Fig. 1a, where S precesses around z due to an external magnetic field and/or magnetic anisotropy contributing poten- tial energy EM . The corresponding action is Sext = (cid:82) dt(cid:80)± ±EM [S±] = 2(cid:82) dt Sq · zΩ, with Ω = Ωext + Ωan[cos θ]cl. Separating out the fast motion φcl(t) = Ωt + φcl(t), the dynamics of θ, φ are driven only by the spin transfer torque. We assume this dynamics is slow, and evaluate Eq. (2) under a time scale separation ∼ t−1 0 ,W2/S (cid:28) T, Ω: [24] (cid:88) ST (cid:39) −i (cid:90) ∞ dt d
−∞ σσ(cid:48)α=± 2 Γσσ(cid:48)α(
)(eiαησσ(cid:48) (
) − 1) . (3) Here, ησσ(cid:48)(
) = χ(t) + (
− µ + V + Ωσσ(cid:48))ξ(t)− 2 Ωσσ(cid:48) Ω φq(t) and Ωσσ(cid:48) = [σΩ − σ(cid:48)Ω cos θcl(t)]/2. The transition rates per energy are Γσσ(cid:48)α(
) = ¯Gσσ(cid:48) 1 + σσ(cid:48) cos θcl(t) 2 (cid:40) Λα(
, V + Ωσσ(cid:48)) , (4) Λα(
, V ) = fF (cid:48)(
)[1 − fF (
+ V )] , α = + , fF (
+ V )[1 − fF (cid:48)(
)] , α = − . (5) 2 2 1+σPF z 1+σ(cid:48)PF (cid:48) Here, fF/F (cid:48)(
) = 1/[e(
−µ)/TF /F (cid:48) + 1] are Fermi distri- bution functions, and the time-averaged conductance is where G0 = 2πW2(νF↑ + ¯Gσσ(cid:48) = G0 νF↓)(νF (cid:48)↑ + νF (cid:48)↓), the polarizations are defined as P = (ν↑ − ν↓)/(ν↑ + ν↓), and PF z = PF cos θF is the polar- ization of the fixed magnet projected onto the precession axis. The densities of states ν↑/↓ of majority/minority spins are given at the Fermi level. The resulting ST is independent of φcl, i.e. its dynamics decouples, which constrains θq = 0 (see Appendix). The result describes Poissonian transport events, each associated with a back-action on θ due to the spin trans- fer torque, as described by the dependence on φq. The rates are proportional to the averaged densities of states and squared spin overlaps (cid:104)σσ(cid:48)(cid:105)2 = [1 + σσ(cid:48) cos θ]/2, in the frame rotating with the magnetic precession. The transferred energy V + Ωσσ(cid:48) consists of the voltage bias and the difference ±Ω/2 − (±Ω cos θ)/2 of energy shifts on the right and left sides of the junction in the rotating frame [2, 25]. The relation of this additional dependence on θ to geometric phases is discussed in Ref. 10. It also separates Eq. (3) from the result of Ref. 7 for tunneling through a ferromagnetic insulator barrier, where such an- gular dependencies are not included. Equation (3) is a main result of this work, as the knowl- edge of ST allows access to the statistics of charge, en- ergy and spin transfer in the generic case depicted in Fig. 1a. Below, we describe some applications. First, we can identify the following spintronic fluctuation relation (see Appendix) ST (χ, ξ, φq) = S(cid:48) T (−χ + ,−ξ + iV TF i TF − i TF (cid:48) , φq + iΩ 2TF ) , (6) where the prime denotes inverting the magnetizations and the sign of the precession. Identifying the conjugate fields of χ, ξ, and φq to the number of charges δn, change of energy δE and transfer of spin angular momentum δs, this relation is equivalent with Eq. (1). This relation also implies the Onsager relation dI/dΩ = (sin2 θ)dτ /dV relating the pumped current to the torque τ sin2 θ ≡ 3 FIG. 3. (a) Normalized variance of the magnetization z- component in the steady state around the fixed point s∗, as a function of bias voltage and temperature for PF (cid:48) = 1/2, PF z = 1/2. Dotted lines indicate results where the energy shifts in the spin torque noise are neglected. (b) Switch- ing exponent ∆sw, for PF (cid:48) = 1, PF z = 1/4, and different temperatures and voltages. Dotted lines indicate the range −Ωan/8 < V < Ωan where ∆sw = ∞ at T = 0. the currents, (cid:88) (cid:88) σσ(cid:48) SI = S QF = σσ(cid:48) ¯Gσσ(cid:48) ¯Gσσ(cid:48) 1 + σσ(cid:48) cos θ 2 1 + σσ(cid:48) cos θ 6 Vσσ(cid:48) coth Vσσ(cid:48) 2T , (π2T 2 + V 2 σσ(cid:48))Vσσ(cid:48) coth (10) , Vσσ(cid:48) 2T (11) = 2 ¯GL0T 3, where ¯G = dI where Vσσ(cid:48) = V +Ωσσ(cid:48) and TF = TF (cid:48). In the classical lin- ear regime V, Ω < T , the results reduce to a form dictated by the fluctuation-dissipation theorem and Wiedemann- Franz law, SI = 2 ¯GT , S QF dV is the electrical dc conductance of the magnetic tunnel junc- tion, [28] and L0 the Lorenz number. The presence of the angle-dependent frequencies is revealed in the quantum noise regime Ω > T . The noise in the pumped current for V = 0 is plotted in Fig. 2 -- the location of the quantum -- classical crossover is pushed up to higher pre- cession frequencies as the tilt angle approaches θ = 0. Spin torque induced fluctuating precession. The above results are conditional on a specific value of θ. For the full probability distribution, the distribution P (θ) would need to be known. assume S (cid:29) 1 and take a semiclassical approximation. Defining sz = cos θ and p = 2iSφq, the action reads To find P (θ) = (cid:82) D[θcl, φq] eiSχ=ξ=0δ(θcl(0) − θ), we iSχ=ξ=0 =(cid:82) dt [p sz − H(p, sz)] where H = −iST is real for real sz, p. The problem can then be analyzed as in Hamiltonian mechanics, [23] In a time-sliced discretization of the path integral, the δ restriction specifying the exact measured value adds a boundary condition sz(0) = sz0 that removes one of the integration variables and saddle point equations. This p = −∂sz H. sz = ∂pH, FIG. 2. (a) Noise in pumped charge current, for different tilt angles θ and precession speeds Ω, for PF (cid:48) = PF z = 0.9. (b) Semiclassical trajectories for V = −Ωext, PF (cid:48) = 1, PF z = 1/2, χ = ξ = 0, T = 0. Shown are the H = 0 lines AB and the fixed point s∗ = PF(cid:48) Ω (black). Measurement trajectories C χ for χ = 1 (red) and χ = −1 (blue) are also shown. (c) Trajectories with anharmonicity, Ω = Ωansz, V = 1.5Ωan, PF z = 0.1, PF (cid:48) = 1, T = 0. PF z PF(cid:48) + 2V 1 dST /d(2φq) acting on the angle θ [26]. This and fur- ther details of the fluctuation relation are discussed in the Appendix. The average dynamics follows the θ component of the Landau-Lifshitz-Slonczewski equation, [1] here obtained from stationarity of S vs φq, S θ = − sin(θ)τ (θ), τ (θ) = Ωα(θ) + Isz , (7) 4 G0PF zV and damping 8 G0[1 − PF zPF (cid:48) cos θ] [27] have been discussed where the spin current Isz = 1 α(θ) = 1 in Ref. 8. The equation describes motion of cos θ in an effective potential −(cid:82) cos θ d(cos θ(cid:48)) τ (θ(cid:48)) defined by Ω(θ) and the spin torque, illustrated in Fig. 1b. In certain parameter ranges, a fixed point τ (θ∗) = 0 appears -- it can be either attractive or repulsive. This can correspond to a stable but fluctuating precession around the angle θ∗ (Fig. 1b), induced by spin torque, or spin torque-induced switching between two energy minima (Fig. 1c). Average current and noise. For fast measurements, t0 (cid:28) 1/ θ, we can assume θ remains fixed, and find the average currents, I = QF = 1 2 1 2 G0[1 + PF (cid:48)PF z cos θ]V + IV + 1 2 τ (θ)Ω sin2 θ , 1 4 G0PF zΩ sin2 θ , (8) (9) where the pumped charge current (second term in Eq. (8)) is that found in Ref. 25. The heat current is a sum of the Joule heat and the magnetic energy lost 2 ΩanS2 z ], due to the spin torque, dissipated equally in F and F (cid:48). In contrast to the aver- age values, the energy shifts Ωσσ(cid:48) remain in the noise of EM = −∂t[ΩextSz + 1 051015/T1.01.52.02.53.03.54.0SI/(2GeT)(a)0/43/8/2101sz21012p/SABC1C1(b)s*101sz21012p/Spsw(c)s*0.71.01.5V/ext0.00.51.01.52.02.53.0(coss*)2/00.250.50.751T/(a)1012V/an0123456sw/(b)T/an10.104 shown in Fig. 2a. For simplicity, we consider the limit T (cid:28) Ω,V with full polarization of the free magnet PF (cid:48) = 1. Then, close to s∗, H (cid:39) eα χ[1 − s2∗][ S 2 p2] − Γ(sz)(eα χ − 1) , S p − D(s∗) τ (sz) (15) (cid:17) (cid:16) where α = sgn V and Γ(sz) = Γ++α + Γ−+α. For quadratic H, the Hamiltonian equations can be solved exactly (see Appendix). From this approach, we find the current noise: 1 − 1 − e−t0/τm , SI = Γ(s∗) + 4Γ(cid:48)(s∗)2σ2 s τm t0/τm (16) where τm = −S/[(1 − s2∗)τ(cid:48)(s∗)] is the slow time scale associated with the spin transfer torque and σ2 s = −D(s∗)/(2Sτ(cid:48)(s∗)) the variance of the magnetization z- component in Eq. (14). The first term Γ(s∗) in Eq. (16) is the Poissonian shot noise (10), and the second term orig- inates from magnetization fluctuations. The dependence on the measurement time is shown in Fig. 4. The current m (cid:28) Ω can be used to probe noise at frequencies ω ∼ τ−1 the dynamics and distribution of the magnetization. Spin torque induced stochastic switching. Magnetic anisotropy field Han results to an effective magnetic po- tential with two minima (see Fig. 1c), and the spin torque can induce switching between the two. Here, we take Hext = 0, and Ω = γHansz ≡ Ωansz. The corre- sponding semiclassical Hamiltonian picture is shown in [1 − (1 + Fig. 2c. An unstable fixed point s∗ = an )1/2] separates the two stable fixed points 8PF (cid:48)P 2 sz = ±1. The leading exponent of the rate of switching from sz = −1 to sz = 1 is, [7] (cid:82) s∗ Γsw ∝ e−∆sw = e −1 dsz psw(sz) , H(sz, psw(sz)) = 0 , F zV Ω−1 2PF(cid:48) PF z 1 (17) 2 8 where psw(sz) is shown in Fig. 2c. The switching occurs deterministically (∆sw → 0) if PF zV > 1+PF (cid:48) PF z Ωan as sz = −1 becomes unstable. At lower voltages, the switching is stochastic. Numerically computed results are shown in Fig. 3b. At zero temperature, the switching is blocked [7] at − Ωan 8 < V < Ωan for PF (cid:48) > PF z and −Ωan < V < Ωan otherwise. This occurs because the transition rates Γσσ(cid:48)α vanish for α(V + Ωσσ(cid:48)) ≤ 0, and because the back-action ∝ Ωσσ(cid:48) vanishes for σ = −σ(cid:48), sz → −1. [32] The latter constraint is due to the addi- tional angle dependence in the spin torque, which traces back to the geometric phases [10] in the spin dynamics. Discussion. In conclusion, we have derived a Keldysh action (3), describing the stochastic charge and energy currents affected by a precessing magnetization. We ob- tain a fluctuation relation for the transferred charge, en- ergy, and magnetization. The noise in the current at low temperatures displays features related to geomet- ric phases, and its low frequency component reflects the FIG. 4. Current noise SI as a function of the measurement bandwidth 1/t0, for PF (cid:48) = 1, PF z = 1/2, T = 0. allows for a discontinuity of p at t = 0, cf. Refs. [29, 30]. The other boundary conditions are p(t → ±∞) = 0, so that relevant paths have integration constant H = 0. 1 Consider now fluctuations close to an attractive fixed point τ (θ∗) = 0 (cf. Fig. 1b). For dynamics driven by an external field, it is located at sz = s∗ = , PF(cid:48) Ωext and it is attractive if τ(cid:48)(s∗) = −ΩPF zPF (cid:48) < 0. The phase space picture is shown in Fig. 2b. Expanding around p = 0 in terms of the torque τ and torque noise correlator D, + 2V PF(cid:48) PF z D(sz) = H (cid:39) [1 − s2 (cid:88) z][S−1τ (sz)p − S−2D(sz)p2] , where Γσσ(cid:48)α = (cid:82) ∞ −∞ d
Γσσ(cid:48)α(
) = ¯Gσσ(cid:48) 1+σσ(cid:48)sz (1 − σσ(cid:48)sz)2 α(V + Ωσσ(cid:48))/[1 − e−α(V +Ωσσ(cid:48) )/T ] for TF = TF (cid:48) = T . The fluc- tuation contribution comes from following path A from (s∗, 0) to (sz0,Sτ (sz0)/D(sz0)): (12) (13) 1 − s2 σσ(cid:48)α=± Γσσ(cid:48)α , 1 8 2 z P (cos θ) (cid:39) N e S(cid:82) cos θ s∗ dsz τ (sz ) D(sz ) (cid:39) N e S τ(cid:48) (s∗ ) D(s∗ ) (cos θ−s∗)2 , (14) where N is a normalization constant. This agrees with Ref. 8 in the semiclassical limit S (cid:29) 1, except for the presence of the energy shifts ∝ Ωσσ(cid:48) [10] in the spin torque noise correlator D, which are relevant in the quan- tum limit Ω ∼ V (cid:29) T . The variance is plotted in Fig. 3a. Long measurement times. For t0 (cid:38) 1/ θ, the slow fluc- tuation of the magnetization contributes low-frequency noise to observables. This contribution is not small in 1/S: the typical excursion from the average position is small, δsz ∝ S−1/2, but it lasts for a long time τm ∝ S, generating low-frequency noise SI ∼ ( dI δsz)2τm. The situation is similar to noise induced in tunneling currents by temperature fluctuations on small islands. [31] dsz We now find the result within the semiclassical ap- proximation. The counting fields are switched on in the interval 0 < t < t0, e.g. χ(t) = θ(t)θ(t0 − t)i χ. They make the semiclassical path to transition from branch A to B in the time interval 0 < t < t0 following a trajec- tory C χ, ξ of constant H χ,ξ. Two such trajectories are 104103102101100101102e2/(G0t0)110SI/eIV/0.60.81.01.21.4magnetization fluctuations. Information about the spin torque noise is also contained in the switching probability of anisotropic magnets. Our predictions are readily ac- cessible in experiments probing spin pumping at low tem- peratures T < Ω/kB. Precession frequencies in 10 GHz range have been achieved, [12, 13] which translates to T (cid:46) 1 K. We thank B. Nikolic and S. van Dijken for discus- sions. This work was supported by the MIUR-FIRB2013 - Project Coca (Grant No. RBFR1379UX), the Academy of Finland Centre of Excellence program (Project No. 284594) and the European Research Council (Grant No. 240362-Heattronics). [1] J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). [2] Y. Tserkovnyak, A. Brataas, G. E. W. Bauer, and B. I. 5 [25] Y. Tserkovnyak, T. Moriyama, and J. Q. Xiao, Phys. Rev. B 78, 020401 (2008). [26] A. Brataas, Y. Tserkovnyak, G. E. Bauer, and P. J. Kelly, "Spin current," (Oxford University Press, 2011) Chap. 8. [27] For simplicity, we assume here that spin torque dom- inates magnetization damping. The presence of extra damping would lead to an additional term in α(θ). [28] D. Huertas-Hernando, Y. V. Nazarov, and W. Belzig, Phys. Rev. Lett. 88, 047003 (2002). [29] S. Pilgram, A. N. Jordan, E. V. Sukhorukov, M. Buttiker, Phys. Rev. Lett. 90, 206801 (2003). and [30] T. T. Heikkila and Y. V. Nazarov, Phys. Rev. Lett. 102, 130605 (2009). [31] M. A. Laakso, T. T. Heikkila, and Y. V. Nazarov, Phys. Rev. Lett. 104, 196805 (2010). [32] The blocking is due to non-Gaussianity of the spin torque. The result neglects the exact quantization of spin and ignores e.g. quantum tunneling. [33] A. G. Abanov and A. Abanov, Phys. Rev. B 65, 184407 (2002). Halperin, Rev. Mod. Phys. 77, 1375 (2005). [34] M. Kindermann and S. Pilgram, Phys. Rev. B 69, 155334 [3] D. Ralph and M. Stiles, J. Magn. Magn. Mater. 320, (2004). 1190 (2008). [4] W. F. Brown, Phys. Rev. 130, 1677 (1963). [5] J. Foros, A. Brataas, Y. Tserkovnyak, and G. E. W. [35] S. Pilgram, K. E. Nagaev, and M. Buttiker, Phys. Rev. B 70, 45304 (2004). [36] D. Andrieux and P. Gaspard, J. Chem. Phys. 121, 6167 Bauer, Phys. Rev. Lett. 95, 016601 (2005). (2004). [6] L. S. Levitov and G. B. Lesovik, JETP Lett. 58, 230 (1993). [7] Y. Utsumi and T. Taniguchi, Phys. Rev. Lett. 114, 186601 (2015). [8] A. L. Chudnovskiy, J. Swiebodzinski, and A. Kamenev, Phys. Rev. Lett. 101, 066601 (2008). [9] G.-M. Tang and J. Wang, Phys. Rev. B 90, 195422 (2014). [10] A. Shnirman, Y. Gefen, A. Saha, I. S. Burmistrov, M. N. Kiselev, and A. Altland, Phys. Rev. Lett. 114, 176806 (2015). [11] L. Berger, Phys. Rev. B 54, 9353 (1996). [12] S. I. Kiselev, J. C. Sankey, I. N. Krivorotov, N. C. Emley, and D. C. Ralph, R. J. Schoelkopf, R. A. Buhrman, Nature 425, 380 (2003). [13] W. H. Rippard, M. R. Pufall, S. Kaka, S. E. Russek, and T. J. Silva, Phys. Rev. Lett. 92, 027201 (2004). [14] Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer, Phys. Rev. Lett. 88, 117601 (2002). [15] R. H. Koch, J. A. Katine, and J. Z. Sun, Phys. Rev. Lett. 92, 088302 (2004). [16] S. Yakata, H. Kubota, T. Sugano, T. Seki, K. Yakushiji, A. Fukushima, S. Yuasa, and K. Ando, Appl. Phys. Lett. 95, 242504 (2009), 10.1063/1.3275753. [17] G. E. Crooks, Phys. Rev. E 60, 2721 (1999). [18] G. E. Crooks, Phys. Rev. E 61, 2361 (2000). [19] U. Seifert, Rep. Prog. Phys. 75, 126001 (2012). [20] M. Esposito, U. Harbola, and S. Mukamel, Rev. Mod. Phys. 81, 1665 (2009). [21] J. Tobiska and Y. V. Nazarov, Phys. Rev. B 72, 235328 (2005). [22] M. Campisi, P. Hanggi, and P. Talkner, Rev. Mod. Phys. 83, 771 (2011). [23] A. Kamenev and A. Levchenko, Adv. Phys. 58, 197 [24] Here and below, we choose ψ(t) as in 10, and only con- (2010). sider ψq(±∞) = 0. Appendix: Details of derivation of the generating function We consider a tunneling Hamiltonian model for the ferromagnet/nanomagnet junction, † † σk(HF (cid:48))σk,σ(cid:48)k(cid:48)cσ(cid:48)k(cid:48) + d σk(HF )σk,σ(cid:48)k(cid:48)dσ(cid:48)k(cid:48)] + HT + Hext , [c H = kk(cid:48)σσ(cid:48) (HF (cid:48))σk,σ(cid:48)k(cid:48) = δk,k(cid:48)[1
F (cid:48),k + gS · σ]σσ(cid:48) , (HF )σk,σ(cid:48)k(cid:48) = δk,k(cid:48)
F,σσ(cid:48)k , † Wσk,σ(cid:48)k(cid:48)c σkdσ(cid:48)k(cid:48) + h.c. , HT = Hext = JHext · S . σσ(cid:48)kk(cid:48) (cid:88) (cid:88) (cid:90) ∞ −∞ 6 (18) (19) (20) (21) (22) (23) (24) (25) (26) Above, c (d) are conduction electrons in the free (fixed) magnet, S is the magnetization in the free (single-domain) magnet, Hext an externally applied field, and g and J coupling constants. Moreover,
F (cid:48)/F describe the noninteracting energy dispersions. The Keldysh action corresponding to H is, dt(cid:0)¯cT [i∂t − HF (cid:48)]c + ¯dT [i∂t − HF ]d + ¯cT W d + ¯dT W †c(cid:1) , S = S0[S] + where S0 is the standard spin action [33]. We also include source terms in the generating function Z [34], (cid:90) (cid:90) ∞ (cid:90) ∞ 1 2 −∞ −∞ Z[χ, ξ, ζ] = Sc[χ, ξ] = Sc2[ζ] = dt ζ∂tScl z D[S, ¯c, c, ¯d, d] eiS+iSc+iSc2 , dt ¯dT γx[ χ + (HF − µ) ξ]d , (cid:90) so that derivatives vs. χ and ξ produce cumulants of the charge and energy transfer, and ζ characterizes Sz. The terms added in Sc can be eliminated with a change of variables d(t) (cid:55)→ e−iγx[(HF −µ)ξ(t)+χ(t)]/2d(t) and conversely for ¯d, provided χ(±∞) = ξ(±∞) = 0. This results to addition of phase factors in W , via W (cid:55)→ W e−iγx[(HF −µ)ξ(t)+χ(t)]/2. Here and below, we use Larkin-Ovchinnikov rotated Keldysh basis [23]: fermion fields have Keldysh structure d = (d1, d2), where d1/2 = (d+ ± d−)/ 2 are related to the fields d± on the Keldysh branches. Real fields are split similarly, as X cl/q = (X+ ± X−)/2. γx is a Pauli matrix in the Keldysh space. √ 2, ¯d1/2 = ( ¯d+ ∓ ¯d−)/ √ We integrate out the conduction electrons, and expand in small W , S [8, 10]. The resulting tunneling action can be obtained via the same route as in Ref. 10, dt dt(cid:48) Tr[R(t)GF (cid:48)(t, t(cid:48))R(t(cid:48))†W e−iγx[HF ξ(t(cid:48))+χ(t(cid:48))]/2GF (t(cid:48), t)eiγx[HF ξ(t)+χ(t)]/2W †] . ST [χ, ξ, S] = i F (cid:48) = [
± i0+ − ξF (cid:48) − gSσz]−1, GR/A (27) = [
± i0+ − ξN ]−1, and the unitary matrices R are defined by Here, GR/A RσzR† = S · σ and the gauge degree of freedom ψ in R = R(cid:48)eiψσz is fixed [10] so that (R†∂tR)cl z is proportional to time derivatives of classical field components. The conduction electrons c may also contribute other terms than ST , for example change (or generate) the total spin in S0 [33]. However, here we are mainly interested in spin torque and pumping, and therefore concetrate on dynamics implied by ST and assume any other effects are absorbed to changes in parameters or phenomenological damping terms. form Tr[eiξ0O(t0)/2e−iξ0O(0)/2ρe−iξ0O(0)/2eiξ0O(t0)/2], t0 → ∞, which characterize a two-measurement protocol [20, 22]. They correspond to choices ξ(t) = θ(t)θ(t0 − t)ξ0, and χ(t) = θ(t)θ(t0 − t)χ0. We can write (see below) z = 0 and (R†∂tR)q correlation long-time functions consider limit us now of the Let the F e−iγxHN ξ(t)GF (t, t(cid:48))eiγxHN ξ(t(cid:48)) = (28) where the correction term a(t, t(cid:48)) is zero for t − t(cid:48) ≥ ξ(t) − ξ(t(cid:48)). As discussed below in more detail, it can be neglected in the long-time limit [34]. The tunneling action then reads −∞ e−i
(t−t(cid:48))e−i
γxξ(t)GF (
)ei
γxξ(t(cid:48)) + a(t, t(cid:48)) , d
2π (cid:90) ∞ Tr[P (t,
)GF (cid:48)(t, t(cid:48))P (t(cid:48),
)†W GF (
)W †] , d
2π (29) (cid:90) ∞ dt −∞ (cid:90) ∞ −∞ ST (cid:39) i where P (t,
) = ei(
−V )te−iγxχ(t)/2e−iγx
ξ(t)/2R(t), and X(t, t(cid:48)) =(cid:82) ∞ In the case considered in the main text, φ = Ωt + We have −∞ dt(cid:48) X(t + t(cid:48)/2, t − t(cid:48)/2). φ, and dynamics of θ and φ arises from the spin transfer torque. R(t) = e−i[Ωt+ψ0(t)]σz/2e−iσz[ where ψ0(t) = −(cid:82) t dt(cid:48) Ω cos θcl, ψ0(t + q) (cid:39) ψ0(t) − Ωq cos θcl(t). Keeping only the non-oscillating parts of Eq. (29) + e−i[Ωt−ψ0(t)]σz/2e−iσz[ φ− ψ]/2(−iσy) sin ψ]/2 cos (30) φ+ , θ 2 θ 2 and taking the leading term of the gradient expansion vs. θ, φ, we can write the time average: 7 (31) ei
(t−t(cid:48))P (t,
)GF (cid:48)(t, t(cid:48))P (t(cid:48),
)† (cid:39) (cid:90) ∞ −∞ d
(cid:48) (cid:88) (cid:16) σs tr s=↑/↓ θ(t) δ(
−
(cid:48) − V − sΩ−(t))eiη−,s(
,t)γx/2 cos 2 + δ(
−
(cid:48) − V − sΩ+(t))eiη+,s(
,t)γx/2 sin G1,s(
(cid:48)) cos θ(t) 2 e−iη−,s(
,t)γx/2 θ(t) 2 where η±,s(
, t) = χ(t) + [
− µ]ξ(t) − s[ φq(t) ∓ ψq(t)] = χ(t) + [
− µ]ξ(t) − 2s Ω±(t) φq(t) and Ω± = Ω[1 ± cos θcl]/2. To this order, ST is independent of φcl. Provided no source fields measuring the statistics of φcl are added, the only 2 ∂t[cos(θcl + θq)− cos(θcl − θq)], which implies a constraint cos(θcl + θq)− cos(θcl − θq) = const. and we set θq = 0. The slow part of the dynamics of the polar φ angle decouples from the rest of the problem. part dependent on it is S0 = . . . +S(cid:82) dt φcl 1 (cid:82) dt dt(cid:48) (. . .)GA = (3) F (t(cid:48), t) = 0 and neglecting terms unimportant in the long-time limit introduced by (cid:82) dt dt(cid:48) (. . .)GR e−iη+,s(
,t)γx/2(cid:17) θ(t) 2 G1,−s(
(cid:48)) sin F (cid:48)(t, t(cid:48))GR F (cid:48)(t, t(cid:48))GA now follows, F (t(cid:48), t) result Eq. the main noting text The in Ω , Eq. (28). In the eigenbasis of the single-particle operator HN , we can write Energy counting (cid:88) GF (
) = + GF,j(
) = j P
−
j GF,j(
)j(cid:105)(cid:104)j , + iπδ(
−
j) (cid:18)1 2 tanh
−µ 2T (cid:19) , (32) (33) 0 −1 = = 2π (cid:90) d
(cid:90) d
e−i
[t+γxξ(t)] GF (
)ei
[t(cid:48)+γxξ(t(cid:48))] + where P is the Cauchy principal value. Straightforward calculation now yields eiHF γxξ(t) GF (t, t(cid:48))e−iHF γxξ(t(cid:48)) (cid:90) d
(cid:88) (cid:88) e−i
j (t−t(cid:48)+γxξ(t)−γxξ(t(cid:48)))j(cid:105)(cid:104)j−i 4 a1(t, t(cid:48)) = 2 sgn(t − t(cid:48)) − sgn(t − t(cid:48) + ξ(t) − ξ(t(cid:48))) − sgn(t − t(cid:48) − ξ(t) + ξ(t(cid:48))) , a2(t, t(cid:48)) = sgn(t − t(cid:48) + ξ(t) − ξ(t(cid:48))) − sgn(t − t(cid:48) − ξ(t) + ξ(t(cid:48))) . The correction term is zero for t − t(cid:48) ≥ ξ(t) − ξ(t(cid:48)), and consequently gives negligible contribution in the long-time limit where ξ(t) = ξ(t(cid:48)) except near the ends of the measurement interval. While neglecting it is not necessary in principle, this simplifies the approach. Appearance of such "time-energy uncertainty" was also noted in Ref. [34]. e−i
(t−t(cid:48))ei
j γx[ξ(t)−ξ(t(cid:48))][1 − ei(
−
j )γx[ξ(t)−ξ(t(cid:48))]] e−i
[t+γxξ(t)] GF (
)ei
[t(cid:48)+γxξ(t(cid:48))] + P
−
j [a1 + a2γx] j(cid:105)(cid:104)j (34) 2π 2π j j Let us consider a system where the nanomagnet is coupled to a ferromagnetic electrode via a tunnel barrier with spin-flip conductance GT and polarization P , and via an ohmic contact to a normal metal. We disregard Fluctuation theorem the magnetization damping caused by the normal metal, and assume that the voltage completely drops across the tunnel junction. Below, we denote the temperature of the normal metal and the nanomagnet by TF (cid:48), and that of the ferromagnetic electrode by TF . We disregard charge and energy pile-up effects, limiting ourselves to time scales long compared to the charge and energy relaxation time of the system. In this case we can specify the probability dt(cid:48) Q(t(cid:48)) tunneling through the tunnel contact, and a change in the z component of magnetization δMz = Sγδsz/ν, δsz = δ[cos(θcl)] in a time t0. It reads distribution of charge δn =(cid:82) t+t0 t dt(cid:48)I(t(cid:48))) and energy δE =(cid:82) t+t0 (cid:110) (cid:104) i(cid:82) Sζ(t) t (cid:105) (cid:111) 8 (35) (cid:90) dζ (cid:90) dζ Pt0 (δn, δE, δsz) dξ0 2π dξ0 2π dχ0 2π dχ0 2π 2π 2π = = e−iχ0δn−iξ0δEDφqDsze DφqDsze−iχ0δn−iξ0δE−iζδsz ei{S0[φq(t)+ζ(t)/2]+ST (χ,ξ,φq;V,TF ,TF(cid:48) )}, ∂tsz(t)/V− δsz t0 dt ei[S0(φq)+ST (χ,ξ,φq;V,TF ,TF (cid:48) )] where χ(t) = χ0θ(t)θ(t0 − t), ξ(t) = ξ0θ(t)θ(t0 − t) and ζ(t) = ζ0θ(t)θ(t0 − t) specify the two-measurement protocol. Since ST is independent of the slow component of the φ-coordinate, this component does not affect the statistics of the other parameters and can be integrated out. In the above equation, we assume that the measurement time t0 is long compared to charge relaxation times of the island, but it can be short compared to the time scale of magnetization relaxation. The presence of the spin action S0 ensures the conservation of the total angular momentum, analogous to the other conservation laws explained in [35]. We can use the Fermi function identities f (2µ −
) = f (
)eβ(
−µ) = 1 − f (
) to get Λ+(
, V ) = e(
−µ)/TF(cid:48) e−(
−µ+V )/TF Λ−(
, V ), Λ+(2µ −
,−V ) = Λ−(
, V ). Applying these to the action ST yields the symmetries ST (χ, ξ, φq; V, Ω, TF , TF (cid:48)) = ST (−χ, ξ,−φq;−V,−Ω, TF , TF (cid:48)) ST (χ, ξ, φq; V, Ω, TF , TF (cid:48)) = ST (−χ − ξV,−ξ,−φq − ξΩ/2;−V,−Ω, TF (cid:48), TF ) ST (χ, ξ, φq; V, Ω, TF , TF (cid:48)) = ST (−χ + iV /TF ,−ξ + i(T −1 F − T −1 F (cid:48) ),−φq + iΩ/(2TF ); V, Ω, TF , TF (cid:48)) (36) (37a) (37b) (37c) In addition, one more relation can be obtained by reversing the magnetizations of both systems, or changing the signs of the polarizations PF z and PF (cid:48). This sign change can be balanced by replacing Ω (cid:55)→ −Ω and φq (cid:55)→ −φq. Denoting the magnetization reversal and reversal of the sign of precession with a prime hence yields the symmetry ST (χ, ξ, φq; V, Ω, TF , TF (cid:48)) = S(cid:48) T (χ, ξ,−φq; V, Ω, TF , TF (cid:48)). (38) These relations together with the definition (35) allow us to find various symmetries of the probability distribution. For example, combining (37a) with (37b) yield Pt0(δn, δE, δsz; TF , TF (cid:48)) = Pt0(δn,−δE + V δn − δszΩ, δsz, TF (cid:48), TF ). (39) This relates the probabilities of charge and energy transfer and change of magnetization upon the interchange of the two temperatures. The detailed form is a result of the particle-hole symmetry of our model (no thermoelectric effects are included). For TF = TF (cid:48) this implies the first law of thermodynamics for the processes. Namely, it implies 2(cid:104)δE(cid:105) − V (cid:104)δn(cid:105) + Ω(cid:104)δsz(cid:105) = 0, (40) i.e., for an arbitrary nonequilibrium state, the expectation value of the internal energy increase in the two terminals (when TF = TF (cid:48), (cid:104)δEF(cid:105) = (cid:104)δEF (cid:48)(cid:105) ≡ (cid:104)δE(cid:105)) equals the sum of the dissipated Joule heat and the work done by the change in the magnetization direction. This result is also reflected in the average heat current in Eq. (9) of the main text. On the other hand, combining (37c) with (38) yields Pt0(δn, δE, δsz; TF , TF (cid:48)) = eV δn/TF eδE(T −1 F −T −1 F(cid:48) )e−δszΩ/TF P (cid:48) t0 (−δn,−δE, δsz; TF , T (cid:48) F ). (41) This is the spintronic fluctuation relation for the setup, and it in particular cases yields those presented in [7, 17 -- 19, 21]. Note that in contrast to [7], this presents the probability statistics of change of magnetization, rather than that of the spin current, as the latter as such is difficult to measure directly. One direct consequence of fluctuation theorems is the Onsager symmetry of linear response coefficients characterizing nonequilibrium observables [20, 36]. In particular, we can define the electrical and energy currents and spin transfer torque via 9 Ic = Q = sin2(θ)τ = ∂ST ∂χ ∂ST ∂ξ ∂ST ∂ζ χ=ξ=φq=0 χ=ξ=φq=0 χ=ξ=φq=0, or denoting λ1,2,3 ∈ {χ, ξ, φq} and generalized current as Ji ≡ ∂ST ∂λi λ=0. (42a) (42b) (42c) (43) Let us consider f1,2,3 ∈ {V, TF (T −1 F − T −1 F (cid:48) ), Ω/2} as generalized forces. The linear response coefficients are thus defined via Lij ≡ ∂Ji ∂fj f =0 = ∂2ST ∂λi∂fj λ=f =0. In terms of the generalized forces and currents, Eq. (37c) can be expressed as ST (λ; f ) = ST (−λ + if /TF ; f ). Differentiating this with respect to fj and λi and setting λ = f = 0 then gives (now TF = TF (cid:48) ≡ T ) 2T Lij ≡ 2T ∂ST ∂λi∂fj (0; 0) = −i ∂2ST ∂λi∂λj (0; 0). (44) (45) (46) This hence yields the Onsager reciprocity relations, accoring to which Lij = Lji is a symmetric matrix. Note that Eq. (46) is the (zero-frequency) fluctuation-dissipation theorem: The left hand side of this equation yields the zero- frequency autocorrelation function (cid:104)δJiδJj(cid:105). Semiclassical equations In the main text we analyze the slow stochastic dynamics of the magnetization under the fluctuating spin torque within the semiclassical approximation. As a result, the z component of the magnetization of the free magnet, sz = cos(θ), and a term proportional to the counting field φq, p ≡ 2iSφq become conjugate variables, whose dynamics follows the effective Hamiltonian H = −iST . Let us consider such Hamiltonian mechanics generated by a quadratic Hamiltonian, such as that in Eq. (15) in the main text, H = −Aqp − Bp2 + C + Dq + Eq2 with initial and final conditions Bp(0) + Aq(0) = 0 and p(t0) = 0. The integration constant is H0 = C + Dq(0) + Eq(0)2, which also implies q(t0) = q(0). √ Define r = p + αq where α = A+β 2B and β = A2 + 4BE. Then, H = βqr − Br2 + C + Dq , r = −βr − D . Taking the initial and final conditions into account, we find the starting point of the trajectory so that q(0) = −D αβ + A/B eβt0−1 , r(t) = e−βt β − A 2B q(0) − D β [1 − e−βt] . (47) (48) (49) The action of the trajectory now reads iSC = (cid:90) t0 0 dt [p q − H] = −t0H0 + (cid:90) t0 dt p q . The second term is (A, B > 0) (cid:90) t0 (cid:90) t0 (cid:104) 0 dt p q = −p(0)q(0) − 0 = Aq(0)2 B − B 2β2 0 (cid:90) r(t0) r(0) dr dt q[ r − α q] = Aq(0)2 B + (βr(t0) − 2D)r(t0) − (βr(0) − 2D)r(0) H0 − C + Br2 (cid:105) D + βr + t0[H0 − C + 10 (50) (51) (52) BD2 β2 ] For βt0 (cid:29) 1, the last term is ∝ t0e−βt0 , and the others approach a constant as t0 → ∞. Therefore, for t0 → ∞ we have iSC (cid:39) − t0H0. Let us furthermore expand the action in a counting field λ to the second (Gaussian) order. Given expansions A = A0 + A1λ + A2λ2 + . . ., of A, B, C, D, E, with C0 = D0 = E0 = 0, we find the leading terms iSC (cid:39) −C1λt0 − C2λ2t0 + λ2 B0D2 A2 0 1 (cid:104) t0 − 1 − e−2A0t0 dt p qC=D=E=0 = − Aq(0)2 (cid:39) −λ2 B0D2 1 2A3 2B 0 2A0 iSA = (cid:90) 0 −∞ (cid:105) , (1 − e−A0t0)2 . (53) (54) Adding iS = iSA + iSC, taking λ = α χ, and substituting in the values corresponding to Eq. (15) of the main text, A0 = −[1 − s2∗]τ(cid:48)(s∗)/S, B0 = [1 − s2∗]D(s∗)/S 2, C1 = −α χΓ(s∗), C2 = − 1 2 χ2Γ(s∗), D1 = −Γ(cid:48)(s∗)α χ, we arrive at Eq. (16) in the main text. 

1606.05430

3

1606

Mechanical Actuation of Magnetic Domain-Wall Motion

We theoretically study the motion of a magnetic domain wall induced by transverse elastic waves in a one-dimensional magnetic wire, which respects both rotational and translational symmetries. By invoking the conservation of the associated total angular and linear momenta, we are able to derive the torque and the force on the domain wall exerted by the waves. We then show how ferromagnetic and antiferromagnetic domain walls can be driven by circularly- and linear-polarized waves, respectively. We envision that elastic waves may provide effective means to drive the dynamics of magnetic solitons in insulators.

cond-mat.mes-hall

cond-mat

a Mechanical Actuation of Magnetic Domain-Wall Motion Department of Physics and Astronomy, University of California, Los Angeles, California 90095, USA Se Kwon Kim, Daniel Hill, and Yaroslav Tserkovnyak (Dated: November 5, 2018) We theoretically study the motion of a magnetic domain wall induced by transverse elastic waves in a one-dimensional magnetic wire, which respects both rotational and translational symmetries. By invoking the conservation of the associated total angular and linear momenta, we are able to derive the torque and the force on the domain wall exerted by the waves. We then show how ferromagnetic and antiferromagnetic domain walls can be driven by circularly- and linear-polarized waves, respectively. We envision that elastic waves may provide effective means to drive the dynamics of magnetic solitons in insulators. PACS numbers: 75.78.-n, 75.60.Ch, 62.30.+d, 63.20.-e Introduction. -- Phonons, quanta of elastic vibrations, are ubiquitous in condensed matter systems including magnets. Owing to their gapless nature, they can eas- ily absorb energy from excited spins, thereby engender- ing the damping term in the description of spin dynamics [1]. Apart from this passive role, the idea of actively using phonons to induce magnetic dynamics has been recently gaining attention in spintronics. It has been experimen- tally demonstrated that acoustic pulses can induce coher- ent magnetization precession [2] via spin-lattice coupling [3]. Also excitation of elastic waves can generate spin currents [4] and thereby drive magnetic bubbles [5]. A domain wall in an easy-axis magnet is one of the sim- plest and well-studied topological solitons [6], which has practical importance exemplified by the racetrack mem- ory [7]. They can be driven by various means: a magnetic field [8], an electric field [9], a spin-polarized electric cur- rent [10], a temperature gradient [11], or a spin wave [12]. Moving domain walls have been known to generate and drag phonons, which in turn gives rise to the damping force on the walls [13]. This force increases as the domain wall approaches the speed of sound, which was pointed out as the origin of the plateau in the dependence of the domain-wall speed on an external field [14]. In this Letter, we study the reciprocal problem: actua- tion of the magnetic domain-wall motion via the phonon current, which can be injected by mechanical means. The stress-induced motion of a domain wall has been previ- ously studied in Ref. [15], in which the domain wall is energetically driven by the axial stress gradient gener- ated by the static voltage profile in piezoelectric mate- rials. Differing from that, we focus on the effects of the dynamic phonon current on the domain wall via scat- tering. Specifically, we consider a one-dimensional mag- netic wire with a coaxial easy-axis anisotropy, which can be realized by a single-crystalline iron nanowire embed- ded in a carbon nanotube [16]. It respects the rotational and translational symmetries and thus conserves the to- tal angular and linear momenta. A magnetic domain wall breaks both symmetries, which opens channels for the exchange of both momenta with phonons. See Fig. 1 FIG. 1. A schematic illustration of a ferromagnetic wire with a magnetic domain wall (shown by blue thick arrows) and a transverse elastic deformation (which is exaggerated for il- lustrative purposes). The yellow circles represent incoming phonons, quanta of elastic waves; the yellow diamonds rep- resents transmitted and reflected phonons. The red (blue) arrows on circles represent phonons' angular momentum in the positive (negative) z direction. Phonons are injected from the left; some of them are reflected by the domain wall and thereby exert the force on it; some of the transmitted and re- flected phonons change their angular momentum and thereby exert the torque on the wall. for an illustration of a domain-wall configuration for a ferromagnetic system. We show that the domain wall is birefringent for transverse waves and can thus act as a waveplate that alters the circular polarization -- and thus the angular momentum -- of phonons traveling through it. This change of phonons' angular momentum applies the torque on the domain wall. Reflection of phonons by the domain wall gives rise to the force acting on it. We study the domain-wall motion induced by the phononic torque and the force in ferromagnets and antiferromagnets. Main Results. -- Our model system is a one-dimensional magnetic wire stretched along the global z axis by an external tension, in which the magnetic order parame- ter tends to align with the local orientation of the wire. The order parameter is the local spin angular-momentum density for ferromagnets and the local N´eel order for anti- ferromagnets. For temperatures well below the ordering temperature, the local order parameter has the saturated magnitude and thus can be represented by the unit vector ⦿xzyn(ζ, t) pointing along its direction. Here, ζ is the inter- nal coordinate of the lattice atoms along the wire. In this Letter, we are interested in the interaction between the magnetic soliton -- domain wall -- and the transverse vi- brations of the wire, which are represented by u(ζ, t) and v(ζ, t) for the displacements of the atom at ζ in the lab- frame x and y direction, respectively [17]. We shall focus on small displacements by working to the quadratic order in u and v. In studying the dynamics of elastic waves, we shall assume that the dynamics of the magnetization is slow enough to be treated as static in the equations of motion for elastic waves, which would be valid if the speed of sound is much larger than that of magnons. The potential energy that involves the magnetic order parameter is given by Um =(cid:90) dζ (cid:2)An(cid:48)2 + K{1 − (n · t)2}(cid:3) /2 , (1) where the positive constants A and K are the exchange and anisotropy coefficients, respectively [18]. Here, (cid:48) is the derivative with respect to the intrinsic coordinate ζ; t(ζ, t) ≡ (u(cid:48), v(cid:48),√1 − u(cid:48)2 − v(cid:48)2) is the unit tangent vector of the wire. The magnetic anisotropy can be rooted in either the magneto-crystalline anisotropy or the shape anisotropy induced by dipolar interactions. When the wire is straight along the z axis, u ≡ v ≡ 0, there are two ground states: n ≡ z and n ≡ −z. A domain wall is a stationary solution of δnUm = 0 that interpolates two ground states n(ζ = ±∞) = ∓z. It is given by [8] nx(ζ) = sech[(ζ − Z)/λ] cos Φ , ny(ζ) = sech[(ζ − Z)/λ] sin Φ , nz(ζ) = − tanh [(ζ − Z)/λ] . (2a) (2b) (2c) Here, Z and Φ are the position and the azimuthal angle of the domain wall, respectively; λ ≡(cid:112)A/K is the char- acteristic length scale of the problem, corresponding to the domain-wall width. Z and Φ parametrize two zero modes of the domain wall, which are associated with the breaking of the translational and spin-rotational symme- tries. The dynamics of the position Z induced by the wire's transverse vibrations is of our main interest. The linearized dynamics of the transverse displace- ments of the stretched wire can be described by the La- grangian [19] Le =(cid:90) dζ (cid:2)µ( u2 + v2) − T (u(cid:48)2 + v(cid:48)2)(cid:3) /2 , (3) where the positive constants µ and T are the mass den- sity of the wire and the applied tension, respectively [20]. The equations of motion for u and v, that are derived from the Lagrangian Le in conjunction with the poten- tial energy Um [Eq. (1)], are given by µu −(cid:2)(cid:8)T + K(n2 µv −(cid:2)(cid:8)T + K(n2 z − n2 z − n2 x)(cid:9) u(cid:48)(cid:3)(cid:48) = −K(nznx)(cid:48) , y)(cid:9) v(cid:48)(cid:3)(cid:48) = −K(nzny)(cid:48) . (4a) (4b) 2 FIG. 2. The torque τ on the domain wall by the circularly- polarized waves as a function of the wavenumber kλ for the parameters κ = 0.2 and a = λ/10. The solid line is obtained with the analytical expression for τ in Eq. (5); the dots are ob- tained with τ in Eq. (13) calculated from numerical solutions of the differential equations (7). The inset shows a zoom-in at small wave vectors kλ < 1. tion is given by ω = ±v0k with the speed v0 ≡(cid:112)Tκ/µ. For the uniform ground states, n ≡ ±z, the right-hand sides vanish and the tension is effectively increased from T to Tκ ≡ (1 + κ)T with κ ≡ K/T . The dispersion rela- Using the propagating-wave solutions to the above equa- tions in the presence of the domain wall, details of which will be shown later, we can derive the phononic torque and the force on the wall by invoking the conservation of the angular and linear momenta. The induced domain- wall speed is quadratic in the amplitude of waves, which allows us to assume that the domain wall is static in Eqs. (4) to the linear order in the amplitude [21]. First, circularly-polarized waves incoming from the left, u(ζ, t) = a cos(kζ−ωt) and v(ζ, t) = −a sin(kζ−ωt), exert the torque (i.e., the transfer of angular momentum) on the domain wall, τ (cid:39) Tκa2k[1 − cos{kλ ln(1 − κ)}] , (5) for high-energy waves kλ (cid:29) 1, which is obtained by the subtraction of the angular momentum current of the transmitted wave, Tκa2k cos{kλ ln(1 − κ)}, from that of the incoming wave, Tκa2k. The physical ori- gin of the torque can be understood as follows. From Eqs. (4), the domain wall locally modifies the tension for the u and v displacements by K[1 − 2sech2(ζ/λ)] and K[1 − sech2(ζ/λ)], respectively. The v component thus propagates faster than the u component within the domain wall, which acts as a birefringent medium that can alter the polarization of the wave. The argument of the cosine function in Eq. (5) is the relative phase shift of u and v components of the transmitted wave, φu,t − φv,t (cid:39) kλ ln(1 − κ). Figure 2 shows the torque τ as a function of the wavenumber kλ. Note that it oscil- lates as a function of kλ with the period of 2π/ ln(1− κ). This torque by the elastic waves can drive ferromagnetic 020406000.5100.51: numerical solution: analytical solutionk02040600.51⌧/T000.0003013 on solutions for v henceforth, from which we can ob- tain solutions for u by replacing κ by 2κ. For the given incoming-wave component, the solution far away from the wall can be characterized by four real numbers: the amplitude t > 0 and the phase-shift φt of the transmitted component and the amplitude r > 0 and the phase-shift φr of the reflected component: v(ζ) ∝(cid:40)eikζ + re−ikζ+iφr , teikζ+iφt , for ζ (cid:28) −λ for ζ (cid:29) λ . (8) The equation can be transformed into a quantum- mechanic scattering problem by introducing a new co- ordinate η satisfying dζ/dη = 1 − κ sech2(ζ/λ): λ (cid:21) v = k2v . dη2 + k2κ sech2 ζ(η) (cid:20)− (9) d2 We obtain approximate solutions in the two ex- treme energy regimes. First, in the high-energy limit, kλ (cid:29) 1, we use the Wentzel-Kramers-Brillouin ap- proximation [23], within which the solution is v(ζ) ∝ dinate ζ with the transmission amplitude t = 1. The phase shift of the transmitted wave is given by exp(cid:2)ik(cid:82) dζ{1 − κ sech2(ζ/λ)}−1/2(cid:3) in the original coor- φt = k(cid:90) ∞ = −kλ ln(1 − κ) . (10) −∞ dζ (cid:113)1 − κ sech2(ζ/λ) In the low-energy limit, we approximate the poten- tial by the delta-function barrier with the height hk ≡ the potential. After solving the scattering problem and going back to the original coordinate ζ, we obtain 2k2(cid:112)κ/(1 − κ) arcsin√κ that is the spatial integral of r = κkλ , φr = −π/2, t = 1 , φt = κkλ , (11) to the first order in kλ. Note that for a small anisotropy, κ (cid:28) 1, the phase shifts of the transmitted wave in the two regimes coincide: −kλ ln(1 − κ) (cid:39) κkλ. Torque and force. -- In the uniform state, n(ζ) ≡ z, the effective Lagrangian density for the waves which includes the effect of the anisotropy is given by L = µ( u2 + v2) − Tκ(u(cid:48)2 + v(cid:48)2) . Axial symmetry of the Lagrangian implies conservation of the corresponding angular momentum. The temporal and spatial components of the associated Nother current [22, 24] are given by ρs = µ(u v − v u) and I s = −Tκ(uv(cid:48) − vu(cid:48)), which are, respectively, the density and the current of the (orbital) angular momentum [25]. We obtain the linear momentum density T 10 = −µ uu(cid:48) + (u → v) and the current T 11 = (µ u2 +Tκu(cid:48)2)/2+(u → v) from the stress-energy tensor, T αβ ≡ ∂αu[∂L/∂(∂βu)] + (u → v)−δαβL [24]. For monochromatic waves, u(ζ, t) = u0 cos(kζ − ωt) and v(ζ, t) = v0 cos(kζ − ωt + ∆φ), the angular and linear momentum currents are given by I s = Tκu0v0 sin(∆φ)k , T 11 = Tκ(u2 0 + v2 0)k2/2 . (12) FIG. 3. The force F on the domain wall exerted by the linearly-polarized waves perpendicular to the wall plane for the parameters κ = 0.2 and a = λ/10. The solid line is ob- tained with the analytical expression for F in Eq. (6); the dots are obtained with F in Eq. (14) calculated from numer- ical solutions of the differential equations (7). domain walls, analogous to the torque of spin waves [12]. The steady-state speed of the ferromagnetic domain wall is V = τ /2s [Eq. (17)] in the absence of damping, where s ≡ S/V is the saturated spin density (V is the volume per spin). Secondly, linearly-polarized waves incoming from the left, v(ζ, t) = a cos(kζ − ωt) and u(ζ, t) ≡ 0, exert no torque, but a finite force (i.e., the transfer of linear mo- mentum) on the domain wall due to the reflection, F (cid:39) Tκκ2λ2a2k4 , (6) for low-energy waves kλ (cid:28) 1. It is the product of the pressure (i.e., the linear momentum current) of the in- coming wave, Tκa2k2/2, and twice the reflection prob- ability, 2κ2λ2k2. The reflection probability is exponen- tially small for high-energy waves kλ (cid:29) 1, and so is the force. Figure 3 shows the force F as a function of the wavenumber kλ. This force of the elastic waves can drive antiferromagnetic domain walls, analogous to the force of spin waves [12, 22]. The steady-state speed of the antifer- romagnetic domain wall is V = λF/2αs [Eq. (18)], where α is the Gilbert damping constant. Transverse waves. -- Let us solve the differential equa- tions (4) for u and v in the presence of the static domain wall given by Eqs. (2). A general solution is composed of static and dynamic components. A static one is deter- mined by the right-hand sides of the equations [13, 22], whereas dynamic components, which are of interest to us, are the propagating waves, for which we can neglect the right-hand sides. For the monochromatic solutions, i.e., ∝ exp(−iωt), the equations are given by with k2 ≡ ω2/v2 0 and κ ≡ κ/(1 + κ). We shall focus (cid:2)(cid:8)1 − 2κ sech2(ζ/λ))(cid:9) u(cid:48)(cid:3)(cid:48) = −k2u , (cid:2)(cid:8)1 − κ sech2(ζ/λ)(cid:9) v(cid:48)(cid:3)(cid:48) = −k2v , (7a) (7b) 012340110000015000001100000150000: numerical solution: analytical solution012k12F/T⇥10534Let us now derive the torque and the force on the do- main wall exerted by elastic waves. The torque is the dif- ference of the angular momentum current I s between the far left and far right of the domain wall; the force is the difference of the linear momentum current T 11 between them. First, for the circularly-polarized incoming wave, u(ζ, t) = a cos(kζ − ωt) and v(ζ, t) = −a sin(kζ − ωt), the time-averaged torque and force are given by τ = Tκa2(1 − tutv cos ∆φt − rurv cos ∆φr)k , F = Tκa2(r2 u + r2 v)k2 , (13) (14) where tu and tv are respectively the transmission ampli- tudes of the u and v components (and similarly ru and rv for the reflection amplitudes) and ∆φt is the relative phase shift of the transmitted u and v components (and similarly ∆φr for the reflected wave). Equation (5) for τ is the reflectionless limit of Eq. (13), corresponding to kλ (cid:29) 1. Secondly, for the linearly-polarized incoming wave, v(ζ, t) = a cos(kζ − ωt), the torque vanishes and the force is given by F = Tκa2r2 vk2 . (15) Ferromagnetic domain wall. -- The dynamics of the fer- romagnet is described by the Lagrangian [26], L = s(cid:82) dζ a(n) · n − U [n], where a(n) is a vector potential of a magnetic monopole, ∇n × a = n. The angular and linear momenta of the domain wall [Eqs. (2)] are given by, respectively, J = J0 + 2sZ and P = P0 − 2sΦ, where J0 and P0 are arbitrary [27, 28]. Viscous losses can be represented by the Rayleigh dissipation function [24] R = αs(cid:82) dζ n2/2, where α is Gilbert's damping con- stant [1]. By plugging the domain-wall solution, we ob- tain R = αs(λ Φ2 + X 2/λ). The conservations of the total angular and linear momenta yield τ = J + 2αsλ Φ , F = P + 2αs Z/λ . (16) The steady-state velocity V = Z is given by V = τ + αλF 2(1 + α2)s . (17) Antiferromagnetic domain wall. -- The dynamics of the antiferromagnet is described by the Lagrangian, L = der parameter [29]. The Rayleigh dissipation function is χ(cid:82) dζ n2/2 − U [n], where χ quantifies inertia of the or- R = αs(cid:82) dζ n2/2 [30]. For slow dynamics, the angular and linear momenta of the domain wall are respectively given by J = I Φ and P = M Z, where I ≡ 2χλ and M ≡ 2χ/λ are the moment of inertia and the mass of a static domain wall [22]. Their equations of motion are same as Eqs. (16). The steady-state velocity Z(t) → V is given by V = λF 2αs . (18) 4 Discussion. -- For experiments, linearly polarized elas- tic waves can be coherently excited by attaching a piezo- electric transducer to the magnetic wire, as done in Refs. [31] to probe the magnetoacoustic Faraday effect [3]. Coherent excitation of circularly-polarized waves can be generated, for example, by exciting two linearly po- larized modes with the fixed relative phase of π/2. Let us make a quantitative estimate for the speed of the domain-wall in ferromagnets, V = τ /2s [Eq. (17)] in the zero-damping limit α = 0 at the maximum effi- ciency of the phononic torque, i.e., τ = 2Tκa2k [Eq. (5)]. We take the parameters of iron for the magnet, the sat- uration magnetization Ms = 2 × 106 A/m and the mass density ρ = 7 × 103 kg/m3 [32], and the parameter of lead zirconate titanate for the piezoelectric strain con- stant d = 10−10 m/V [15]. For the iron wire of the cross- sectional area A = 20 nm2 [16] subjected to the tension T = 10−3 N, the application of the electric field E = 1 V/mm rotating at the frequency 10 MHz across the piezo- electric transducer of length L = 100 nm yields the speed V ≈ 40 m/s (assuming the perfect coupling of stress be- tween the transducer and the wire), which is comparable to the domain-wall speed by the spin-polarized electric current [7] and by the magnon current [12]. We have neglected the magnetoacoustic Faraday effect [31] as well as its inverse effect [33], which are absent in the static treatment of the magnetization according to the energy Um (1) that is even under magnetization reversal. The effects might be present when the magne- tization is made dynamic, but it is suppressed when the magnetic and the acoustic resonances are significantly mismatched [3]. These effects, in principle, can influence the domain-wall motion. In particular, via the inverse effect, the circularly polarized elastic waves can induce the effective magnetic field along the axial direction and drive the ferromagnetic domain wall by contributing to the force F in Eq. (17) [8]. In the zero-damping limit α = 0, however, this contribution can be neglected. We are grateful for discussions with H´ector Ochoa and Ricardo Zarzuela, who helped us to formulate the prob- lem. We also thank the anonymous referees, whose com- ments and questions led to the significant improvement of the Letter. This work was supported by the Army Research Office under Contract No. 911NF-14-1-0016. [1] T. Gilbert, IEEE Trans. Magn. 40, 3443 (2004). [2] A. V. Scherbakov, A. S. Salasyuk, A. V. Akimov, X. Liu, M. Bombeck, C. Bruggemann, D. R. Yakovlev, V. F. Sapega, J. K. Furdyna, and M. Bayer, Phys. Rev. Lett. 105, 117204 (2010); J.-W. Kim, M. Vomir, and J.-Y. Bigot, Phys. Rev. Lett. 109, 166601 (2012). [3] C. Kittel, Phys. Rev. 110, 836 (1958). [4] K. Uchida, H. Adachi, T. An, T. Ota, M. Toda, B. Hille- brands, S. Maekawa, and E. Saitoh, Nat. Mater. 10, 737 (2011); M. Weiler, H. Huebl, F. S. Goerg, F. D. Czeschka, R. Gross, and S. T. B. Goennenwein, Phys. Rev. Lett. 108, 176601 (2012). [5] N. Ogawa, W. Koshibae, A. J. Beekman, N. Nagaosa, M. Kubota, M. Kawasaki, and Y. Tokura, Proc. Natl. Acad. Sci. 112, 8977 (2015). [6] A. Kosevich, B. Ivanov, and A. Kovalev, Phys. Rep. 194, 117 (1990), and references therein. [7] S. S. P. Parkin, M. Hayashi, and L. Thomas, Science 320, 190 (2008). [8] N. L. Schryer and L. R. Walker, J. Appl. Phys. 45, 5406 (1974). [9] A. P. Pyatakov, A. S. Sergeev, E. P. Nikolaeva, T. B. Kosykh, A. V. Nikolaev, K. A. Zvezdin, and A. K. Zvezdin, Phys. Usp. 58, 981 (2015), and references therein. [10] J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996); L. Berger, Phys. Rev. B 54, 9353 (1996); A. C. Swav- ing and R. A. Duine, Phys. Rev. B 83, 054428 (2011); K. M. D. Hals, Y. Tserkovnyak, and A. Brataas, Phys. Rev. Lett. 106, 107206 (2011). [11] A. A. Kovalev and Y. Tserkovnyak, Europhys. Lett. 97, 67002 (2012); 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). [12] P. Yan, X. S. Wang, and X. R. Wang, Phys. Rev. Lett. 107, 177207 (2011); E. G. Tveten, A. Qaiumzadeh, and A. Brataas, Phys. Rev. Lett. 112, 147204 (2014). [13] V. G. Bar'yakhtar, B. A. Ivanov, and M. V. Chetkin, Sov. Phys. Usp. 28, 563 (1985), and references therein. [14] S. O. Demokritov, A. I. Kirilyuk, N. M. Kreines, V. I. Kudinov, V. B. Smirnov, and M. V. Chetkin, J. Magn. Magn. Mater. 102, 339 (1991). [15] M. T. Bryan, J. Dean, and D. A. Allwood, Phys. Rev. B 85, 144411 (2012). [16] K. Lipert, S. Bahr, F. Wolny, P. Atkinson, U. Weissker, T. Muhl, O. G. Schmidt, B. Buchner, and R. Klingeler, App. Phys. Lett. 97, 212503 (2010). [17] By assuming that the Young's modulus is much larger than the applied tension, E (cid:29) T , we shall neglect lon- gitudinal displacements by focusing on low-energy trans- verse modes [19, 24? ]. 5 [18] Transverse elastic deformations may affect the magnetic potential energy by modifying the geometry of the wire, which can be captured by an additional potential-energy term δUm =(cid:82) dζ (u(cid:48)2 + v(cid:48)2)[ξAn(cid:48)2 + νK{1 − (n · t)2}]/2, where ξ and ν are the dimensionless parameters. This term modifies the strength of the potential for elastic waves in Eq. (7), 2κ (cid:55)→ [2 − (ξ + ν)]κ for u and κ (cid:55)→ [1 − (ξ + ν)]κ for v, which does not change the elastic- wave-induced motion of the domain wall qualitatively. versity Press, Cambridge, 2007). [19] M. Kardar, Statistical Physics of Fields (Cambridge Uni- [20] We neglect the bending energy ∝ (u(cid:48)(cid:48))2 [? ] by assuming that the tension is strong so that the potential energy ∝ T in Le dominates over the bending energy. [21] In the quantum regime, the domain wall can be consid- ered static only when its effective inertia is much larger than that of incoming phonons. [22] S. K. Kim, Y. Tserkovnyak, and O. Tchernyshyov, Phys. Rev. B 90, 104406 (2014); A. K. Zvezdin and A. A. Mukhin, Sov. Phys. JETP 75, 306 (1992). [23] L. D. Landau and E. M. Lifshitz, Quantum Mechanics, 3rd ed. (Butterworth-Heinemann, Oxford, 1976). [24] H. Goldstein, C. Poole, and J. Safko, Classical Mechan- ics, 3rd ed. (Addison Wesley, 2002). [25] ? ] showed that phonons in magnetic crystals can have nonzero orbital angular momentum in equilibrium. [26] A. Altland and B. Simons, Condensed Matter Field The- ory (Cambridge University Press, Cambridge, 2006). [27] P. Yan, A. Kamra, Y. Cao, and G. E. W. Bauer, Phys. Rev. B 88, 144413 (2013). [28] O. Tchernyshyov, Ann. Phys. 363, 98 (2015). [29] I. V. Bar'yakhtar and B. A. Ivanov, Fiz. Nizk. Temp. 5, 759 (1979); A. F. Andreev and V. I. Marchenko, Sov. Phys. Usp. 23, 21 (1980). [30] H. V. Gomonay and V. M. Loktev, Phys. Rev. B 81, 144427 (2010). [31] H. Matthews and R. C. LeCraw, Phys. Rev. Lett. 8, 397 (1962); A. Sytcheva, U. Low, S. Yasin, J. Wosnitza, S. Zherlitsyn, P. Thalmeier, T. Goto, P. Wyder, and B. Luthi, Phys. Rev. B 81, 214415 (2010). [32] D. Jiles, Introduction to Magnetism and Magnetic Mate- rials (Chapman and Hall, New York, 1991). [33] D. I. Tokman and I. V. Pozdnyakova, Eur. Phys. J. B 86, 207 (2013). 

1411.5532

4

1411

A universal explanation of tunneling conductance in exotic superconductors

A longstanding mystery in understanding cuprate superconductors is the inconsistency between the experimental data measured by scanning tunneling spectroscopy (STS) and angle-resolved photoemission spectroscopy (ARPES). In particular, the gap between prominent side peaks observed in STS is much bigger than the superconducting gap observed by ARPES measurements. Here, we reconcile the two experimental techniques by generalising a theory which was previously applied to zero-dimensional mesoscopic Kondo systems to strongly correlated two-dimensional (2D) exotic superconductors. We show that the side peaks observed in tunneling conductance measurements in all these materials have a universal origin: They are formed by coherence-mediated tunneling under bias and do not directly reflect the underlying density of states (DOS) of the sample. We obtain theoretical predictions of the tunneling conductance and the density of states of the sample simultaneously and show that for cuprate and pnictide superconductors, the extracted sample DOS is consistent with the superconducting gap measured by ARPES.

cond-mat.mes-hall

cond-mat

A universal explanation of tunneling conductance in exotic superconductors Jongbae Hong1 and D. S. L. Abergel∗2, 3 1Center for Theoretical Physics of Complex Systems, Institute for Basic Science, Daejeon 305-811, Korea 2Nordita, KTH Royal Institute of Technology and Stockholm University, Roslagstullsbacken 23, SE-106 91 Stockholm, Sweden 3Center for Quantum Materials, KTH and Nordita, Roslagstullsbacken 11, SE-106 91 Stockholm, Sweden 6 1 0 2 g u A 5 1 ] l l a h - s e m . t a m - d n o c [ 4 v 2 3 5 5 . 1 1 4 1 : v i X r a 1 A longstanding mystery in understanding cuprate superconductors is the inconsistency between the experimental data measured by scanning tunneling spectroscopy (STS) and angle-resolved photoemission spectroscopy (ARPES). In particular, the gap between prominent side peaks observed in STS is much bigger than the superconducting gap observed by ARPES measurements. Here, we reconcile the two experimental techniques by generalising a theory which was previously applied to zero-dimensional mesoscopic Kondo systems to strongly correlated two-dimensional (2D) exotic superconductors. We show that the side peaks observed in tunneling conductance measurements in all these materials have a universal origin: They are formed by coherence-mediated tunneling under bias and do not directly reflect the underlying density of states (DOS) of the sample. We obtain theoretical predictions of the tunneling conductance and the density of states of the sample simultaneously and show that for cuprate and pnictide superconductors, the extracted sample DOS is consistent with the superconducting gap measured by ARPES. Tunneling conductance of non-strongly-correlated materials (where the electron -- electron interactions do not have a fundamental effect on the physics) is simply understood because the dI/dV as a function of V (where V is the voltage difference between the sample and the tip and I is the measured current) is known to correspond to the DOS of the sample. However, for strongly correlated materials (SCMs) this is not true because the strong on-site interactions fundamentally change the tunneling mechanism by which electrons move from the current source to the sample itself. Nevertheless, this interpretation is frequently used when discussing SCMs, which can lead to confusing inconsistencies. For example, a sharp peak indicating the superconducting gap of cuprate superconductors is observed in the ARPES data near the nodal region1,2, but in the STS data, two prominent side peaks occur at much higher energy3 -- 5. Also, STS data for strongly correlated pnictide superconductors display two prominent side peaks which are much bigger than the energy scale of the superconducting transition temperature6. Therefore, explaining the structure of tunneling conductance of SCMs is one of the most challenging and urgent subjects in condensed matter physics. This situation will remain until a comprehensive theory taking into account the non-equilibrium nature of the STS experiment and the strong 2 a (cid:115)(cid:140)(cid:141)(cid:155) (cid:147)(cid:140)(cid:136)(cid:139) (cid:121)(cid:144)(cid:142)(cid:143)(cid:155) (cid:147)(cid:140)(cid:136)(cid:139) b . U V V =+5V top T=60mK FIG. 1. A schematic of the mesoscopic Kondo system and typical experimental dI/dV data. a, An entangled state is composed of a linear combination of singlets (the solid and dashed magenta loops) and coherent up- and down-spins denoted by different colored dots. b, Tunneling conductance for a QPC. Experimental data (red) reported in Ref. 8, theoretical results (blue) obtained with parameters in the first line of Tab. I. electron -- electron interactions has been developed. The main purpose of this study is to prove that the two prominent side peaks in the STS data are produced by the interplay between strong electron correlations in the sample and the non-equilibrium situation imposed by the experimental setup. An additional highlight of this study, just as important as the interpretation of the side peaks, is that we can effectively find the DOS of the specific 2D SCM under study within the fitting procedure in the theory. We find that the DOS predicted for cuprate and pnictide superconductors are consistent with the data given by ARPES2,7. The non-equilibrium calculation of dI/dV for SCMs is achieved by generalizing a theory for the non-equilibrium Kondo effect in zero-dimensional mesoscopic systems9,10 to the extended 2D situation. Some mathematical details are presented in the Methods section and in the Supplementary Materials, but in summary, the non-equilibrium tunneling dynamics are encoded within the non-equilibrium many-body Green's function for the mediating site (MS, indicated by the subscript "d"), ρd(ω), which is derived from the Liouville operator. Once this is known, the tunneling conductance can be computed from the formula in equation (2). The advantage of using the Liouvillian approach instead of the Hamiltonian approach is the availability of a complete set of basis vectors, which is not possible in the latter case. The difficulty with the Liouvillian approach arises in the calculation of Liouville matrix elements for a specific model system. This problem can be resolved by making phenomenological assumptions or treating the unknowns as fitting 3 a LR cγ b LR dγ c L exchγ d L hopγ (cid:122)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:143)(cid:150)(cid:151)(cid:97) (cid:122)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:143)(cid:150)(cid:151)(cid:97) (cid:122)(cid:151)(cid:144)(cid:149) (cid:140)(cid:159)(cid:138)(cid:143)(cid:136)(cid:149)(cid:142)(cid:140)(cid:97) (cid:122)(cid:151)(cid:144)(cid:149) (cid:140)(cid:159)(cid:138)(cid:143)(cid:136)(cid:149)(cid:142)(cid:140)(cid:97) (cid:115)(cid:140)(cid:141)(cid:155) (cid:154)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:141)(cid:150)(cid:153)(cid:148)(cid:136)(cid:155)(cid:144)(cid:150)(cid:149)(cid:97) (cid:121)(cid:144)(cid:142)(cid:143)(cid:155) (cid:154)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:141)(cid:150)(cid:153)(cid:148)(cid:136)(cid:155)(cid:144)(cid:150)(cid:149)(cid:97) (cid:115)(cid:140)(cid:141)(cid:155) (cid:154)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:141)(cid:150)(cid:153)(cid:148)(cid:136)(cid:155)(cid:144)(cid:150)(cid:149)(cid:97) (cid:121)(cid:144)(cid:142)(cid:143)(cid:155) (cid:154)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:141)(cid:150)(cid:153)(cid:148)(cid:136)(cid:155)(cid:144)(cid:150)(cid:149)(cid:97) (cid:122)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:143)(cid:150)(cid:151)(cid:97) (cid:122)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:143)(cid:150)(cid:151)(cid:97) (cid:122)(cid:151)(cid:144)(cid:149) (cid:140)(cid:159)(cid:138)(cid:143)(cid:136)(cid:149)(cid:142)(cid:140)(cid:97) (cid:122)(cid:144)(cid:149)(cid:142)(cid:147)(cid:140)(cid:155) (cid:143)(cid:150)(cid:151)(cid:97) FIG. 2. Coherent hybridization dynamics of γ. The first row indicates the initial entangled states before performing hybridization dynamics (subsequent rows). The black dotted line is added to represent an entangled state. The dashed line indicates the state after exchange or singlet hopping, and the red solid line indicates the singlet formation for next hybridization process. The yellow arrows indicate transferred spins. a,b, Finite-bias coherent current formation in γLR c and γLR d exclusively by singlet hopping. c, The exchange process contribution to γL. d, Low-bias coherent current forming dynamics in γL which is given by the combination of spin exchange and singlet hopping. parameters. However, for the Hamiltonian approach, one encounters a fundamental difficulty in obtaining a complete set of basis vectors at the beginning of the calculation and therefore progress is blocked. The DOS in the leads are captured by the lead DOS functions ΓL and ΓR which enter via Γ(ω) and ρd(ω) in equation (2). Hence, these functions are inputs to the theory, and can be chosen to represent various different scenarios. We now describe the coherent tunneling processes which produce the dominant features of the experimental dI/dV curves. Our central claim is that many-body singlet states form between electrons on the MS and in both leads (as shown by the ellipses in figures 1, 2, and 3). When a bias voltage is applied, the coherent singlet cotunnels unidirectionally through the MS. In this context, coherence implies that the singlet may perform spin exchange or change its partner in the leads without any energy cost. A mesoscopic Kondo system is sketched in figure 1a and consists of two leads with constant DOS and a central zero-dimensional MS. This system can be described by a two-reservoir Anderson impurity 4 model11. To understand the coherent tunneling mechanism of the entangled state, we must examine the elements of the Liouville operator shown in equation (1). These are illustrated schematically in figure 2 and their full operator expressions are given in equation (S6) in the Supplementary Information. To put this discussion in context, figure 1b shows an archetypal plot of the tunneling conductance as a function of bias voltage V for a quantum point contact (QPC) with symmetric coupling to the leads on both sides8. Notice there are three coherent peaks: one "zero bias peak", and two "side peaks" which cannot be Coulomb peaks because of the small energy scale. We shall demonstrate which of the γ elements contribute to the formation of these peaks. We first discuss the formation of the side peaks. Figure 2a and its mirror image 2b show coherent current formed solely from singlet hopping. In this process, a left singlet performs a singlet hop to the right (second sketch) before partner exchange on the left (third sketch) enables a second singlet hop to the right which returns the MS to its original state (fourth sketch). This coherent process allows two electrons (highlighted in yellow in the sketch) to tunnel from the left lead to the right lead without paying the on-site Coulomb repulsion energy cost U. Matrix elements A in equation (1) are given by symmetric and antisymmetric combinations and γLR γLR S S = γLR A = γLR γLR applied across the system), γLR c + γLR and γLR d c d c − γLR and γLR . It is clear that at equilibrium (i.e., when no bias is d will have the same magnitude because the singlet hops will occur with equal probability in each direction, so that γLR A = 0 and γLR S 6= 0, indicating that γLR S contributes to the zero bias peak. However, out of equilibrium, either or γLR γLR c so that γLR c d will be suppressed because it must work against the external bias in the system 6= γLR d and as a result, γLR A 6= 0. This contributes an additional channel for current at finite bias, and this manifests as the side peaks in dI/dV away from equilibrium. Therefore, it is the inclusion of left-right antisymmetric coherent superpositions γLR A which allow us to explain the existence of the side peaks in the tunneling conductance. Additionally, γL = γL exch + γL the Kondo coupling strength on the left side, and γL hop, where γL hop (figure 2d) describes the current exch (figure 2c) represents flow forming the zero bias peak, in which spin exchange is involved. However, the zero bias peak is suppressed in the 2D SCMs because of the interaction-induced gap at the Fermi level and so we do not focus on it in this study. We briefly mention that the γ elements are not strongly temperature dependent unless the temperature gets so high that thermal fluctuations destroy the coherence. In fact, the main thermal effects will be driven by the 5 a O b tip tip U U U U U U Non-interacting Quasiparticles FIG. 3. Schematics of STS experiments on 2D cuprate superconductors. a, A schematic of an STS experiment for a real cuprate superconductor. All Cu atoms have on-site repulsion U and the MS is located just below the tip. b A schematic of the model from the many-body point of view. There is a single MS with on-site repulsion U , and the lower part is a non-interacting system of Bogoliubov quasiparticles excited by applied bias, whose DOS is that of a correlated superconductor. changes in the Fermi-Dirac distribution of the leads, and changes in the sample DOS itself. In real STS experiments for cuprate superconductors, every copper atom has strong on-site repulsion U (shown in figure 3a). We set up our model for 2D cuprate superconductors by stating that the tunneling current coming from the STS tip only enters through one copper atom. Then, the tunneling mechanism depends on the strong U at that site only so that the rest of the sample can then be modeled as a system of non-interacting Bogoliubov quasiparticles1,12,13 excited by applied bias, as sketched in figure 3b. Here, an "entangled state" comprising a linear combination of two singlets (solid and dashed purple loops) accompanied by other coherent spins in the tip (blue and red dots) and the sample is formed. The Hamiltonian for the left lead is chosen to be appropriate for the metallic tip so that ΓL is constant, the Hamiltonian for the right lead is taken to be a system of Bogoliubov quasiparticles1,12,13 whose frequency-dependent DOS characterizes the superconductor. Therefore, the DOS of the superconductor becomes an input to the theory [ΓR(ω) in equations (1) and (2)] and hence we can use it as a fitting function to distinguish between various theoretical propoals for the superconductor DOS. This allows us to separate the physics of the tunneling mechanism by coherent co-tunneling of singlet states through the MS from the physics of the superconductor in the bulk of the sample. Now, we apply the theory described above to correlated superconductors to show that it gives accurate predictions for published STS data and that the features in the sample DOS 6 lead to the fine structure in the low-bias region. We require a phenomenological frequency-dependent sample DOS to use as ΓR(ω). We find that a DOS characteristic of a d-wave superconducting order parameter for the under-doped (UD) cuprate and a DOS composed of an s-wave gap and a sharp DOS barrier for the pnictide give a well-fitting tunneling conductance. In both cases, we use flat DOS in the high frequency region to emphasize that the side peaks are not created by features in the sample DOS. These DOS functions are shown as the green lines (right-hand axis) in figures 4a and 4b. Choosing the parameters given in the second and third row of Tab. I gives the fit (blue line) to the experimental data (red line). The quantitative agreement up to the side peaks is almost exact, and we reproduce all the low-bias features. The peaks of the DOS are located at ∆p = 17.8meV (figure 4a) and ∆p = 2.3meV (figure 4b), which are consistent with the superconducting gap reported in recent ARPES data2,7 and correspond to the low-energy shoulders in the dI/dV . The result is natural since the Bogoliubov quasiparticles introduced in figure 3b describe the coherent superconducting state1. We also show the dI/dV and the DOS for optimally doped (OP) and highly overdoped (HOD) cases in the insets of figure 4a. It is noteworthy that the two side peaks in the UD and OP cases are formed in the region of flat DOS. This demonstrates that the side peaks are not connected to the superconducting gap, and they must be reinterpreted as a coherent peak of the non-equilibrium transport mechanism. Their sustained presence when T > Tc signifies the existence of pre-formed Cooper pairs12 up to T ∗ of cuprate superconductors. Since a big gap observed in the anti-nodal region in the ARPES measurements is not a coherent gap1,2, the anti-nodal gap is not included in this study. A big difference occurs in the HOD case where strong correlation no longer persists. We obtain the HOD line shape using vanishing γ elements for a very weak effective Coulomb interaction. Note that two peaks occur within the d-wave region. In conclusion, we have generalised a non-equilibrium quantum transport theory which accurately models tunneling conductance measurements for extended 2D strongly correlated systems. This theory is therefore crucial in understanding experimental data in a wide variety of different contexts. A particularly interesting point is that our theoretical tunneling conductance and sample DOS for cuprate and pnictide superconductors are consistent with the results obtained by two different experimental techniques. This fact eliminates the temptation to think that the dI/dV curve is simply a reflection of the 7 a 6 ) . . u a ( 4 2 0 0 Δ =177.5 meV Δ =17.8 meV p TC = 45 K UD b 2 1 k m 2 ρ s ( ω ) / Δ (cid:17) OP -100 HOD 100 0 E (meV) 0 Δ =13.4 meV Δ =2.32 meV p TC = 20 K 36 k m 24 2 ρ s ( ω ) / Δ (cid:17) 1.5 0.5 Sample Bias [mV] FIG. 4. Comparison with experiment: Correlated superconductors. In each case, the red line is experimental data, the blue line is the prediction of our theory, and the green line (right axis) is the phenomenological sample DOS used to obtain the fit. The vertical dashed line indicates the edge of the shoulder corresponding to the peak in the sample DOS. The grey shading denotes the region of flat sample DOS. a, STS data of cuprate superconductor Bi2Sr2CaCu2O8+δ given in Ref. 3. Insets: OP (left) and HOD (right) cases. b, STS data of pnictide superconductor LiFeAs given in Ref. 6. sample DOS (as it would be in the weakly interacting case), and therefore to interpret the two side peaks as evidence of a correlated band gap in the sample DOS. We demonstrate that the correct interpretation is that the side peaks are evidence of the coherent non-equilibrium tunneling mechanism. METHODS. After the appropriate modelling described in the text, the STS experiment may be described by the following Hamiltonian: H = HL + HM S + HR + HV where HL and HR correspond to the left and right leads, which are taken to be a metallic tip and a system of non-interacting Bogoliubov quasiparticles (described in detail in Section III of the Supplementary Information) excited by the applied bias at T = 0K, respectively. The Hamiltonian for the MS is HM S, and HV describes the tunneling 8 Figure γL γR γLR A , γLR S Re[U α j−] Re[U L j+] Re[U R j+] 1b 4a 4b 0.53 0.53 0.68 0.01 0.5 0.01 0.5 0.5 0.5 1.0 1.75 3.5 1.17 0.35 0.7 1.17 2.45 3.85 Im[U α] ΓL ∆0 0 0.8 0.99 0.1 0.24 118 0 0.24 13.4 β11 = β15 = β55 = 0.252, β12 = β14 = β25 = β45 = 0.254, β22 = β24 = β44 = 0.258, β33 = 1, and β13 = β23 = β43 = β53 = 0. TABLE I. Values of matrix elements and β. The values of the matrix elements appearing in equation (1) used to create the figures. All values are in units of ∆0, except ∆0 (meV). Note that the value in the ΓL column for figure 1c is actually Γ. between the MS and the leads. As described in the Supplementary Information, sections IB and IC, the Liouville operator describing the dynamics of the MS is 0 −γL γL −U L 0 −U L γLR A γLR S S j− γLR j+ γLR U R∗ A   iLr = 0 U L∗ j− −γLR U L∗ j+ S −γLR A −U R j+ j− γR S −U R A −γLR −γLR j+ U R∗ j− 0 −γR 0   + iΣ. (1) where iΣ denotes the self-energy matrix with elements iΣpq(ω) = βpq[ΓL + ΓR(ω)]/2. The coefficients βpq come from the process of matrix reduction which generates equation (1). Each matrix element has its own unique role in determining dI/dV . The middle row and column of the matrix are the incoherent couplings of the MS to the leads which represent double occupancy at the MS via the fluctuations of incoherent spins from the tip or the sample. Therefore, these elements represent the effective Coulomb interaction. The doping effect is given by the imaginary part of these elements shown in equation (S7). The subscript j − indicates the current operator on side α, whereas j+ indicates the sum of the left and right movements. Within this formalism, the tunneling conductance at zero temperature can be obtained as discussed in the Supplementary Information, sections IB and IE: dI dV = e2 ¯h Γ(ω)ρd(ω)(cid:12)(cid:12)(cid:12)(cid:12)¯hω=eV 9 (2) where Γ(ω) is the effective coupling of the left lead, the MS, and the right lead and is given by Γ(ω) = ΓLΓR(ω) ΓL + ΓR(ω) (3) for 2D SCMs. The quantity ρd(ω) = 1 π Pσ Re[M−1]dd, where the matrix M is given by the elements Mpq = −iωδp,q + (iLr)pq, is the non-equilibrium local DOS at the MS (which is distinct from the sample DOS) and depends on the strong interactions. It is computed from the Liouville matrix in equation (1) as described in the Supplementary Information. 1 Lee, W. S. et al. Abrupt onset of a second energy gap at the superconducting transition of underdoped Bi2212. Nature 450, 81 -- 84; DOI:10.1038/nature06219 (2007). 2 Tanaka, K. et al. Distinct Fermi-momentum-dependent energy gaps in deeply underdoped Bi2212. Science 314, 1910 -- 1913; DOI:10.1126/science.1133411 (2006). 3 Lawler, M. J. et al. Intra-unit-cell electronic nematicity of the high-Tc copper-oxide pseudogap states. Nature 466, 347 -- 351; DOI:10.1038/nature09169 (2010). 4 Pushp, A. et al. Extending universal nodal excitations optimizes superconductivity in Bi2Sr2CaCu2O8+δ. Science 324, 1689 -- 1693; DOI:10.1126/science.1174338 (2009). 5 Fischer, O., Kugler, M., Maggio-Aprile, I., Berthod, C. & Renner, C. Scanning tunneling spectroscopy of high-temperature superconductors. Rev. Mod. Phys. 79, 353 -- 419; DOI:10.1103/RevModPhys.79.353 (2007). 6 Chi, S. et al. Scanning tunneling spectroscopy of superconducting LiFeAs single crystals: Evidence for two nodeless energy gaps and coupling to a bosonic mode. Phys. Rev. Lett. 109, 087002; DOI:10.1103/PhysRevLett.109.087002 (2012). 7 Umezawa, K. et al. Unconventional anisotropic s-wave superconducting gaps of the LiFeAs iron-pnictide superconductor. Phys. Rev. Lett. 108, 037002; DOI:10.1103/PhysRevLett.108.037002 (2012). 8 Sarkozy, S. et al. Zero-bias anomaly in quantum wires. Phys. Rev. B 79, 161307; DOI:10.1103/PhysRevB.79.161307 (2009). 9 Hong, J. A complete set of basis vectors of the Anderson model and its Kondo dynamics. J. Phys: Conden. Matt. 23, 225601; DOI:10.1088/0953-8984/23/22/225601 (2011). 10 10 Hong, J. Kondo dynamics of quasiparticle tunneling in a two-reservoir Anderson model. J. Phys: Conden. Matt. 23, 275602; DOI:10.1088/0953-8984/23/27/275602 (2011). 11 Anderson, P. W. Localized magnetic states in metals. Phys. Rev. 124, 41 -- 53; DOI:10.1103/PhysRev.124.41 (1961). 12 Yang, H.-B. et al. Emergence of preformed Cooper pairs from the doped Mott insulating state in Bi2Sr2CaCu2O8+δ. Nature 456, 77-80; DOI:10.1038/nature07400 (2008). 13 Matsui, H. et al. BCS-like Bogoliubov quasiparticles in high-Tc superconductors observed by angle-resolved photoemission spectroscopy. Phys. Rev. Lett. 90, 217002; DOI:10.1103/PhysRevLett.90.217002 (2003). ACKNOWLEDGEMENTS. This work was supported by Project Code (IBS-R024-D1), the Basic Science Research Program via the NRF Korea (2012R1A1A2005220), Nordita, ERC project DM-321031, and partially supported by a KIAS grant funded by MSIP. JH thanks Piers Coleman and Peter Fulde for helpful discussions. AUTHOR CONTRIBUTIONS J.H. designed the study, performed the calculations, and wrote the manuscript. D.S.L.A designed the study and wrote the manuscript. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 11 

1608.04368

2

1608

Numerical calculation of the Casimir-Polder interaction between a graphene sheet with vacancies and an atom

In this work the Casimir{Polder interaction energy between a rubidium atom and a disordered graphene sheet is investigated beyond the Dirac cone approximation by means of accurate real-space calculations. As a model of defected graphene, we consider a tight-binding model of \Pi-electrons on a honeycomb lattice with a small concentration of point defects. The optical response of the graphene sheet is evaluated with full spectral resolution by means of exact Chebyshev polynomial expansions of the Kubo formula in large lattices with in excess of ten million atoms. At low temperatures, the optical response of defected graphene is found to display two qualitatively distinct behavior with a clear transition around non-zero Fermi energy, \mu~\mu*. In the vicinity of the Dirac point, the imaginary part of optical conductivity is negative for low frequencies while the real part is strongly suppressed. On the other hand, for high doping, it has the same features found in the Drude model within the Dirac cone approximation, namely, a Drude peak at small frequencies and a change of sign in the imaginary part above the interband threshold \omega > 2\mu. These characteristics translate into a non-monotonic behavior of the Casimir{Polder interaction energy with very small variation with doping in the vicinity of the neutrality point while having the same form of the interaction calculated with Drude's model at high electronic density.

cond-mat.mes-hall

cond-mat

Numerical calculation of the Casimir -- Polder interaction between a graphene sheet with vacancies and an atom T. P. Cysne and T. G. Rappoport Instituto de F´ısica, Universidade Federal do Rio de Janeiro, Caixa Postal 68528, Rio de Janeiro 21941-972, RJ, Brazil Department of Physics, University of York, York YO10 5DD, United Kingdom Aires Ferreira J. M. Viana Parente Lopes Centro de Fisica and Departamento de Fisica, Universidade do Minho, Campus de Gualtar, Braga 4710-057, Portugal and Departamento de F´ısica e Astronomia,Faculdade de Ciencias da Universidade do Porto, Rua do Campo Alegre 687, 4169-007 Porto, Portugal Centro de Fisica and Departamento de Fisica, Universidade do Minho, Campus de Gualtar, Braga 4710-057, Portugal N. M. R. Peres In this work the Casimir -- Polder interaction energy between a rubidium atom and a disordered graphene sheet is investigated beyond the Dirac cone approximation by means of accurate real- space tight-binding calculations. As a model of defected graphene, we consider a tight-binding model of π-electrons on a honeycomb lattice with a small concentration of vacancies. The optical response of the graphene sheet is evaluated with full spectral resolution by means of exact Chebyshev polynomial expansions of the Kubo formula in large lattices with in excess of ten million atoms. At low temperatures, the optical response of defected graphene is found to display two qualitatively distinct behavior with a clear transition around finite (non-zero) Fermi energy. In the vicinity of the Dirac point, the imaginary part of optical conductivity is negative for low frequencies while the real part is strongly suppressed. On the other hand, for high doping, it has the same features found in the Drude model within the Dirac cone approximation, namely, a Drude peak at small frequencies and a change of sign in the imaginary part above the interband threshold. These characteristics translate into a non-monotonic behavior of the Casimir -- Polder interaction energy with very small variation with doping in the vicinity of the neutrality point while having the same form of the interaction calculated with Drude's model at high electronic density. I. INTRODUCTION Dispersive forces -- including van der Waals, Casimir, and Casimir -- Polder types -- are interactions between neu- tral, but polarizable objects, and have their origin in fluc- tuations of the vacuum electromagnetic field [1]. The Casimir force involves interactions between macroscopic objects [2], such as plates, while Casimir -- Polder forces act between a macroscopic object and a microscopic par- ticle [3]. Van der Waals forces act between objects in the short range regime, where effects of retardation can be neglected. These forces are dominant on nano and micro scales and their control and manipulation are important to applications, such as nano-electromechanical systems, among others [4, 5]. Dispersive forces are strongly influ- enced by the shape and material composition, as well as the dielectric and magnetic responses of the objects they act upon. It is possible to tailor the sign and magnitude of dispersive forces by tuning, for example, the dielectric response of the plate. As a result, the correct modeling of dispersive forces from a materials science perspective becomes important [6, 7]. Since its isolation, graphene has attracted great at- tention, owing to its unconventional low-energy physics described by the Dirac -- Weyl equation for massless exci- tations in two spatial dimensions, and a number of de- sirable physical properties, including superior mechan- ical strength, high charge carrier mobilities, and gate- tunable optical response [8 -- 10]. Intense theoretical ef- fort has been devoted to the study of Casimir [11 -- 23] and Casimir -- Polder [23 -- 29] interactions in graphene and related systems [30, 31]. The Casimir -- Polder energy of different atoms on single layer has been considered in Refs. [24, 26], and on multilayer graphene in Ref. [29]. The tunability of interactions have been demonstrated in atom-on-graphene [28] and in graphene bilayers [20] using external magnetic fields, and in a graphene -- metal system by tuning the chemical potential [21]. In gen- eral, a Dirac cone approximation is considered where the reflection coefficients of graphene are calculated ei- ther within the hydrodynamic model or the polariza- tion tensor with a Drude model approach. The weak- ness of dispersive interactions on graphene systems is ex- perimentally challenging, and most of theoretical predic- tions point to an enhancement of interactions by charge doping [21, 22]. Recently, the control of the interaction between graphene and naphtalene molecule at short dis- tances (van der Waals regime) has been achieved exploit- ing the high tunability of the chemical potential [32]. Recent advances in the understanding of dispersive in- teractions involving graphene and other low-dimensional systems have shown the importance of a detailed char- acterization the electrical response of the layers for the control and tailoring of dispersive forces. Although ab initio methods have been used to model van der Waals forces [30], a more materials oriented approach to study Casimir interactions is still needed. In this article, we consider a realistic model of a large graphene sheet with vacancies. To determine the Fermi energy dependence of its optical conductivity, we employ an accurate large- scale quantum transport approach based on an exact polynomial representation of disordered Green functions recently introduced in Ref. [33]. The large number of expansion moments in the numerical evaluation of the Kubo formula in large graphene lattices allows us to determine the optical response with fine spectral reso- lution. This information is then used to compute the Casimir -- Polder force between a defected graphene sheet and an atom in function of the charge doping, and com- pare it with the force calculated using the Drude model. Far from the Dirac point, the Casimir -- Polder force varies linearly with the chemical potential. The Drude model is found in accord with numerical calculations in that regime, as expected, but fails to capture the behavior of the Casimir -- Polder force close to the Dirac point. Furthermore, we find that the strength of the interac- tion is reduced in the vicinity of the Dirac point, follow- ing the trend of the dc conductivity [33], and increases again above a certain Fermi energy scale µ∗ > 0, in con- trast with the monotonic enhancement of interactions predicted by calculations based on perfect graphene mod- els. This article is organized as follows. Section II out- lines the real space quantum transport methodology used to extract the optical conductivity of large disordered graphene lattices. In Sec. II A, we apply the method- ology to a graphene lattice with a dilute concentration of vacancies. In section III we describe the calculation of the Casimir -- Polder force between the graphene layer and a rubidium atom and present our results. Finally, section IV summarizes the main findings of our work. II. METHODOLOGY The graphene sheet is modeled by a tight-binding Hamiltonian of π electrons defined on a honeycomb lat- tice H = −t † i (a bj + H.c.) , (1) (cid:88) (cid:104)i,j(cid:105) 2 Figure 1. Schematic of a graphene lattice with vacancy de- fects. A(B) sublattices are represented by filled (open) circles. Shaded area shows a vacancy. The numerical simulations in this work have a computational domain of size 3200 × 3200, with periodic boundary conditions on both directions (torus). † where the operator a i creates an electron at site ri = (xi, yi) on sublattice A (an equivalent definition holds for sublattice B), and t = 2.7 eV is the nearest-neighbor hopping integral [34]. The point defects are introduced by removing sites in any sublattice at random (compen- sated vacancies). The defect concentration is ni = Nd/D, where Nd is the number of missing carbon atoms and D is the number of sites in the pristine lattice (see Fig. 1). The real part of the diagonal optical conductivity at zero temperature and finite frequency is given by [35] µ (cid:60) σ(ω) = π ω Ω where Jx = (ite/)(cid:80)(cid:104)i,j(cid:105)(xi − xj)(a µ−ω d
Tr(cid:104) Jx A(
) Jx A(
+ ω)(cid:105)c , (2) † i bj − H.c.) is the x-component of the current density operator, and A(
) = − 1 π (cid:61) 1
− H + iη , (3) is the spectral operator of the system. The symbol (cid:104)...(cid:105)c denotes configurational average, Ω is the area of the lat- tice, µ is the chemical potential, and η is a small broaden- ing parameter required for numerical convergence. Phys- ically, the broadening η = /τi mimics the effect of un- correlated inelastic scattering processes with lifetime τi (e.g., due to phonons), and can be viewed as an energy uncertainty due to coupling of electrons to a bath [36, 37]. The response functions of large tight-binding systems can be assessed numerically by means of specialized spec- tral methods [38 -- 43]. A particularly convenient approach is the kernel polynomial method [44], in which spectral operators are approximated by accurate matrix polyno- mial expansions. The coefficients of the polynomial ex- xypansion are computed recursively thereby bypassing ma- trix inversion that limits systems sizes in exact diagonal- ization schemes. The kernel polynomial method has been applied intensively to study the electronic properties of disordered graphene [45 -- 48]. Here, we make use of an exact Chebyshev polynomial representation of the resol- vent operator recently obtained in Ref. [33], in order to perform numerically exact large-scale calculations of the optical conductivity. The starting point in our approach n=0 an(z)Tn(h) , where z = (
+ iη)/W , h is the rescaled Hamiltonian of disordered graphene h = H/W (here W = 3t is half bandwidth), and Tn(h) are matrix Chebyshev polynomi- als of first kind (see Appendix). Using this expansion, the spectral operator [Eq. (3)] can be recast into the form is the operator identity (z − h)−1 =(cid:80)∞ A = − 1 πW (cid:61)[an(z)]Tn(h) , (4) ∞(cid:88) n=0 whose action on a given basis set can be computed iter- atively by standard Chebyshev recursion [44]. In a numerical implementation, the sum in Eq. (4) is truncated when convergence to a given desired accuracy is achieved. The N th-order approximation to the optical conductivity is therefore given by N−1(cid:88) n,m=0 (cid:60) σ(N )(ω) = π ω Ω where σnm Anm(µ, ω) , (5) σnm = Tr (cid:104) Jx Tn(h) Jx Tm(h)(cid:105)c , µ 1 Anm(µ, ω) = π2W 2 µ−ω d
αn(
)αm(
+ ω) , (6) (7) and αn(
) is a shorthand for (cid:61)[an ((
+ iη)/W )]. Clearly, the problem boils down to the evaluation of σnm, which contains the relevant dynamical information. Once the expansion moments have been determined, the optical conductivity can be quickly retrieved using Eq. (5). For a recent review on the application of Chebyshev expansions in the context of disordered graphene, we refer the reader to Ref. [45]. A. Optical Conductivity of Disordered Graphene With the approach described in the previous section we can study, in a numerically rigorous way, the optical conductivity of graphene in the presence of strong disor- der -- for instance, that created by vacancies, or strongly adsorbed atoms for the same purpose [46]. As a model system of disordered graphene, we have simulated a large lattice of size 3200× 3200 (atoms) with a dilute vacancy concentration, ni = 0.4% (atomic ratio). 3 Figure 2. The optical conductivity of graphene with a dilute vacancy concentration ni = 0.4% at selected values of the chemical potential µ with η ≈ 8 meV. The spectrum of graphene with vacancies is particle-hole symmetric, and hence for simplicity we assume µ ≥ 0 in what follows. Owing to the large system size, it suffices to consider a single disorder realization when performing configurational averages. The optical conductivity for a typical broadening parameter is shown in Fig. 2. To en- sure convergence of the optical conductivity to a good precision [Eq. (5)], we have computed a very large num- ber of Chebyshev iterations N 2 = 80002. Finally, the trace in Eq. (6) has been performed by means of stochas- tic trace evaluation (STE) technique [44]. We have used 5000 random vectors in the STE to enable determination of σnm with accuracy better than 1%. Roughly speaking, we expect that disorder should play a role at low frequencies, ω (cid:28) µ. This is the case if the Fermi energy is not too small. Indeed, we see in Fig. 2 that for a Fermi energy of 0.5 eV there is a well defined step at twice the Fermi energy. A calculation of the opti- cal conductivity based on the Boltzmann equation, given by σD(ω, T, µ) = σD,0(T, µ) (8) 1 1 − iωτ (cid:104) 2 cosh(cid:2) µ , (cid:3)(cid:105) π log 2kB T where σD,0(T, µ) = 2e2τ kB T , predicts the onset of intraband transitions forming a well defined Drude peak [49, 50]. This becomes clear in Fig. 3 where our large-scale numerical calculations for µ = 0.5 eV can be well fit by the Drude model of Eq. (8) with a sin- gle adjustable parameter /τ ≈ 0.07 eV and the spec- tral weight σD,0(T, µ). However, as the Fermi energy decreases, the Fermi step becomes progressively less well defined (compare, for example, the curves in Fig. 2 for 0.3 eV and 0.20 eV; in the latter there is no trace of the Fermi step). In Eq. (8), the intensity of the Drude peak is proportional, at low temperatures, to µ. Therefore, it is no surprise that the curves in Fig. 2 for Fermi energies of 0.5 eV, 0.3 eV, and 0.2 eV show a progressively smaller 00.511.5 ω (eV)01234σ(ω) / σ0σ(ω) / σ00.00 eV0.10 eV0.20 eV0.30 eV0.50 eV00.511.501234ω(eV)4 Figure 4. Conductivity in the imaginary frequency axis at selected values of the Fermi energy. The inset shows the same curves over a wider range of frequencies. that for the cases where the Drude peak was been washed out the curves of the Casimir -- Polder interaction should bunch, whereas for the case where the Drude peak is well defined such bunching should not occur. This is because, in the former case, all conductivity curves essentially co- alesce among themselves. The two regimes discussed above, that is µ < µ∗ and µ > µ∗, become quite clear when the optical conductiv- ity is represented in terms of Matsubara frequencies, as shown in Fig. 4. In this figure, the regime where a Drude peak is well define is characterized by an optical conduc- tivity that presents a positive curvature, whereas in the opposite case the curvature is negative. Therefore, this way of representing the optical conductivity data is an effective tool for separating the two regimes. III. COMPUTATION OF CASIMIR -- POLDER INTERACTION Here we compute the Casimir -- Polder (CP) energy be- tween an atom and a graphene sheet with vacancies and discuss the changes in the CP energy with doping. The optical properties of graphene, necessary for the calcula- tions, can be well described by the numerical results pre- sented in the previous section. We consider a rubidium atom placed at a distance z above a suspended graphene sheet with chemical potential µ. The whole system is assumed to be in thermal equilibrium at sufficiently low temperature T , such that one can use the conductivity numerical calculations carried out at T = 0 K. We choose the rubidium atom due to existence of experimental data of its electric polarizability for wide range of frequencies [53]. The CP energy interaction is calculated within the Figure 3. Fit of numerical optical conductivity data with the Drude Model for µ = 0.5 eVwhere/τ = 0.07 eV is the adjustable parameter. The spectral weight is given by σD,0(T, µ) = 2e2τ kB T 2 cosh(cid:2) µ (cid:3)(cid:105) (cid:104) π log 2kB T intensity of the Drude peak. Very disordered graphene layers might present a renormalized spectral weight, as observed experimentally in CVD graphene [51]. However, for smaller Fermi energy the Drude peak is completely washed out by disorder. In our simulations with a dilute vacancy concentration (see Fig. 2), the crit- ical Fermi energy reads µc ≈ 0.15 eV. We note that, in a realistic scenario, the precise value for µc will depend on the types and strength of disorder present in the sample. The drastic change of behavior in the real part of the optical conductivity has its counterpart in the imaginary component of this quantity as ensured by causality. In- deed, for the Fermi energies where the real part has a well defined Drude peak, one sees in the inset to Fig. 2 that the imaginary part of the conductivity changes from neg- ative to positive as the frequency decreases. This behav- ior is well known for the optical conductivity of graphene and signals the dominance of intraband transitions. On the contrary, for values of the Fermi energy where the Drude peak is supressed, the imaginary part of the conductivity is always negative in the entire frequency range (see Fig. 2). This fact has profound consequences in the interaction of graphene with electromagnetic radia- tion. Just to give an example, when (cid:61)σ(ω) < 0, graphene does not support p-polarized surface waves. On the con- trary, for the case of a well-defined Drude peak both p− and s−polarized waves are supported, albeit in different frequency ranges [52]. The reflection coefficients of a graphene sheet are de- termined by its optical response. Therefore, we expect that the behavior of the Casimir -- Polder interaction to be strongly dependent on the details of the optical conduc- tivity, as those discussed above. Specifically, we expect 00.10.20.30.402468σ(ω) / σ0DrudeNumerical00.10.20.30.4ω(eV)01234σ(ω) / σ000.20.40.60.81ω (eV)0.511.522.53σ(iω) / σ00123456ω (eV)123σ(iω) / σ00.00 eV0.10 eV0.20 eV0.30 eV0.50 eVscattering approach [54] ∞(cid:88) (cid:48) l=0 ξ2 l α(iξl) (cid:32) d2k (2π)2 (cid:33) 5 e−2κlz 2κl (cid:35) UT (z) = kBT ε0c2 × (cid:34) κl = (cid:112)ξ2 rs,s(k, iξl, µ) − 1 + 2c2k2 ξ2 l rp,p(k, iξl, µ) , (9) where ξl = 2πlkBT / are bosonic Matsubara frequencies, l /c2 + k2, α(iξ) is the electric polarizability of rubidium, and rs,s(k, iξ, µ), rp,p(k, iξ, µ) are the diago- nal reflection coefficients associated with graphene. In Eq. (9), the prime indicates that the first term of the summation (l = 0) is halved. By modeling graphene as a two-dimensional material with a surface density current K = σ · Ez=0, and apply- ing the appropriate boundary conditions to the electro- magnetic field, the reflections coefficients are calculated as rs,s(k, iξ, µ) = 2σxx(iξ, µ)Z h + η2 0σxx(iξ, µ)2 −∆(k, iξ, µ) rp,p(k, iξ, µ) = 2σxx(iξ, µ)Z e + η2 ∆(k, iξ, µ) 0σxx(iξ, µ)2 , , (10) (11) ∆(k, iξ, µ) = [2 + Z hσxx(iξ, µ)][2 + Z eσxx(iξ, µ)],(12) where Z h = ξµ0/κ, Z e = κ/(ξ
0), η2 0 = µ0/
0, and σxx(iξ, µ) is the longitudinal optical conductivity of graphene [20, 55]. In the absence of an external magnetic field, the transverse optical conductivity of graphene with vacancies vanishes. A key point for the computation of the CP energy is the correct modelling of the material surface. In our ap- proach, the characteristics of the material are incorpo- rated in the longitudinal optical conductivity σxx(ω). Far from the Dirac point, Drude's model is expected to work for frequencies smaller than the chemical potential. How- ever, for low values of µ, a more accurate calculation must be carried out to capture the detailed physics of graphene and the effects of disorder. In our approach, we use the optical conductivity calculated numerically from a tight- binding Hamiltonian of graphene with vacancies. For that purpose, we first use the Kramers-Kronig relations to obtain the conductivities in the imaginary frequency axis. As shown in Fig. 4, in that case, the separation be- tween the two regimes becomes more clear with different characteristic curve inflections for each regime at for low frequencies. Using equations (9-12) and the optical conductivities shown in Fig. 4, we calculate the CP interaction energy between the graphene sheet and a rubidium atom. Fig- ure 5 presents the CP energy normalized by the inter- action between an atom and a perfect metallic surface −3cα(0) (UCP (d) = 32π2
0d4 ), as a function of the distance at selected values of µ and T =10K. For µ < µ∗, graphene behaves basically as a dieletric and there is a bunching of Figure 5. Casimir Polder energy between a rubidium atom and a graphene sheet with 0.4% of randomly distributed va- cancies normalized by the CP energy between an atom and a metal plate as function of the distance z between the atom and the graphene layer for different values of the Fermi en- ergy. The lower panel shows the comparison between the energy obtained with the numerical calculation (solid lines) and Drude's model (dashed Lines). Figure 6. CP energy for finite µ normalized by CP energy for µ = 0 at z = 2µm as function of chemical potential. the CP energy curves. In the opposite regime, the curves are well separated, as expected from the simple Drude model [Eq. (8)]. The lower panel shows a comparison be- tween the CP energy calculated using the numerical data and the Drude model for graphene. Although the optical conductivity curves present a Drude peak for µ > µ∗, the Drude model does not fit well the numerical results for the CP energy, overestimating the CP force for ex- perimentally accessible distances. Saying it differently, our calculations put stringent constrains on the values of the Fermi energy needed to observe the CP effect on U10K (z) / UCP (z)0.00 eV0.10 eV0.15 eV0.20 eV0.30 eV0.10.51510z(m)0.050.100.150.20U10K (z) / UCP (z)0.20 eV0.30 eV0.50 eVμ0.10.515100.040.200.000.050.100.150.200.25(eV)1.001.051.101.15 0.00.10.20.30.40.5(eV)1.01.11.21.31.41.5μU / =0μUμU / =0μUμμgraphene. If these are too small the force is also small and one may not be able to measure the effect. We show in Fig. 6 the variation of CP energy as func- tion of the Fermi energy for z = 2µm. The strong effect of the vacancies, reliably captured by the numerical cal- culation, results in an almost constant CP energy for a large range of µ around the Dirac point. For larger val- ues of µ, graphene behaves as a Dirac metal, leading to the linear increase of CP energy as a function of µ (see inset). For µ=0.50 eV, the CP energy is increased by 50% if compared to the neutrality point. It is clear from our results that the dependency of CP force with the Fermi energy can be tailored by consider- ing different types of disorder like adatoms and clusters or a higher concentration of vacancies and can become a route to manipulate the behavior of dispersive interac- tions. IV. CONCLUSION In this work we have performed realistic large-scale calculations of the optical conductivity of graphene, re- vealing the role of disorder for "small" Fermi energies. Our calculations show that in the latter regime, the Drude peak is washed out by disorder and the applica- tion of the Drude conductivity for describing the intra- band optical conductivity of graphene becomes unjus- tifiable. This is an important result, as experiments have been conducted with Fermi energies around 0.2 eV, where our calculations show that the Drude model is no longer valid. As expected, this behavior has im- portant consequences on the Casimir -- Polder effect. For large Fermi energies -- µ ∼ 0.5 eV -- the optical conduc- tivity of graphene is well described by a Drude model at low frequencies. However, at "small" Fermi energies Drude model breaks down and one cannot distinguished the Casimir -- Polder interaction energies for varying Fermi energies. Furthermore, the Drude model predicts a larger shift of the interaction energy relative to that of a perfect metallic plane that what will happen in a real situation. Therefore, the forces experienced by the atom will be necessarily smaller than that predicted by the idealized Drude model and may become difficult to measure. Thus for a meaningful experiment our study reveals that highly doped graphene is required. ACKNOWLEDGEMENTS T.G.R. acknowledges support from the Newton Fund and the Royal Society (U.K.) through the Newton Ad- vanced Fellowship scheme (ref. NA150043). A.F. grate- fully acknowledges the financial support of the Royal Society (U.K.) through a Royal Society University Re- search Fellowship. N. M. R. Peres acknowledges the hos- 6 pitality of UFRJ where this work has started and finan- cial support from the European Commission through the project "Graphene-Driven Revolutions in ICT and Be- yond" (Ref. No. 696656). T. G. R, N. M. R. P, J. M. V. P. L. thank Brazil Science without Borders program and CNPq for financial support. APPENDIX ∞(cid:88) The resolvent operator admits an exact representation in terms of Chebyshev polynomials [33] (z − h)−1 = an(z)Tn(h) , (13) n=0 where z is a complex energy variable with (cid:61) z > 0, h is a compact Hamiltonian operator satisfying h ≤ 1, and Tn(h) are Chebyshev polynomials of first kind defined by the recursion relations: T0(h) = 1 ,T1(h) = h, and Tn+1(h) = 2hTn(h) − Tn−1(h) . (14) The expansion coefficients are given by (cid:0)z − i 1 − z2(cid:1)n √ √ 1 − z2 an(z) = 2i−1 1 + δn,0 . (15) These results allow us to express the spectral operator in the form given in main text [Eq. (4)]. [1] M. Bordag, U. Mohideen, and V.M. Mostepanenko, Phys. Rep. 353, 1 (2001); K.A. Mil-ton, J. Phys. A 24, R209 (2004); S.K. Lamoreaux, Rep. Prog. Phys. 68, 201 (2005). [2] H. B. G. Casimir, Proc. K. Ned. Akad. Wet. 51, 793 (1948). [3] H. B. G. Casimir and D. Polder, Nature (London) 158, 787 (1946). [4] X. M. H. Huang, C. A. Zorman, M. Mehregany, and M. L. Roukes, Nature 421, 496 (2003). [5] M. Aspelmeyer, T. J. Kippenberg, and F. Marquardt, Rev. Mod. Phys. 86, 13911452 (2014). [6] L. M. Woods, D. A. R. Dalvit, A. Tkatchenko, P. Rodriguez-Lopez, A. W. Rodriguez, and R. Podgornik, (2015), arXiv:1509.03338 [cond-mat.mtrl-sci]. [7] G. L. Klimchitskaya, U. Mohideen, V. M. Mostepanenko, Rev. Mod. Phys. 81, 1827 (2009) [8] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [9] N. M. R. Peres, Rev. Mod. Phys. 82, 2673 (2010). [10] K. S. Novoselov, V. I. Falko, L. Colombo, P. R. Gellert, M. G. Schwab, K. Kim, Nature 490, 192 (2012). [11] M. Bordag, I.V. Fialkovsky, D.M. Gitman and D.V. Vas- silevich, Phys. Rev. B 80, 245406 (2009). [12] G. G´omez-Santos, Phys. Rev. B 80, 245424 (2009). 7 [13] V. Svetovoy, Z. Moktadir, M. Elwenspoek and H. Mizuta, (2009). Europhys. Lett. 96, 14006 (2011). [35] G. D. Mahan, Many Particle Physics (Plenum Press, New [14] I.V. Fialkovsky, V.N. Marachevsky and D.V. Vassilevich, York, 1981). Phys. Rev. B 84, 035446 (2011). [36] D. J. Thouless and S. Kirkpatrick, J. Phys. C 14, 235 [15] D. Drosdoff and L.M. Wooks, Phys. Rev. A 84, 062501 (1981). (2011). [37] Y. Imry, Introduction to mesoscopic physics, 2nd edition [16] B.E. Sernelius, Phys. Rev. B 85, 195427 (2012). [17] M. Bordag, G.L. Klimchitskaya and V.M. Mostepanenko, (Oxford University Press, New York, 2002). [38] H. Tal-Ezer and R. Kosloff, J. Chem. Phys. 81, 3967 Phys. Rev. B 86, 165429 (2012). (1984). [18] G.L. Klimchitskaya and V.M. Mostepanenko, Phys. Rev. [39] S. Roche, and D. Mayou, Phys. Rev. Lett. 79, 2518 B 87, 075439 (2013). (1997). [19] G. L. Klimchitskaya and V. M. Mostepanenko, Phys. Rev. B 89, 035407 (2014). [20] W.-K. Tse and A. H. MacDonald, Phys. Rev. Lett. 109, [40] H. Tanaka, Phys. Rev. B 57, 2168 (1998). [41] A. Weisse, Eur. Phys. J. B 40, 125 (2004). [42] S. Yuan, H. De Raedt, and M. I. Katsnelson, Phys. Rev. 236806 (2012). [21] M. Bordag, I. Fialkovskiy and D. Vassilevich, Phys. Rev. B 93, 075414 (2016) B 82, 115448 (2010). [43] A. Holzner et al., Phys. Rev. B 83, 195115 (2011). [44] A. Weisse, G. Wellein, A. Alvermann, and H. Fehske, [22] D. Drosdoff, A. D.Phan, L. M. Woods, I.V.Bondarev, Rev. Mod. Phys. 78, 275 (2006). and J.F. Dobson, Eur. Phys. J. B 85, 365 (2012). [45] N. Leconte, A. Ferreira, and J. Jung, 2D Materials, El- [23] G. L. Klimchitskaya, U. Mohideen, and V. M. Mostepa- sevier, Vol. 95, 35 (2016). nenko, Phys. Rev. B 89, 115419 (2014). [24] Yu V. Churkin, A. B. Fedortsov, G. L. Klimchitskaya and V. A. Yurova, Phys. Rev. B 82, 165433 (2010). [25] T. E. Judd, R.G. Scott, A.M. Martin, B. Kaczmarek and T.M. Frohold, New J. Phys.13, 083020 (2011). [26] M. Chaichian, G. L. Klimchitskaya, V. M. Mostepanenko and A. Tureanu, Phys. Rev. A 86, 012515 (2012). [27] S. Ribeiro and S. Scheel, Phys. Rev. A 88, 042519 (2013); Erratum: Phys. Rev. A 89, 039904 (2014). [28] T. Cysne, W. J. M. Kort-Kamp, D. Oliver, F. A. Pin- heiro, F. S. S. Rosa , and C. Farina, Phys. Rev. A. 90, 052511 (2014) [29] N. Khusnutdinov, R. Kashapov, and L. M. Woods, (2016), arXiv:1602.08443v1 [cond-mat.mes-hall] [30] A. Ambrosetti, N. Ferri, R. A. DiStasio Jr., and A. Tkatchenko, Science 351, 1171 (2016). [31] P. R.-Lopez, and A. G. Grushin, Phys. Rev. Lett. 112, 056804 (2014). [32] F. Huttmann, et al. Phys. Rev. Lett. 115, 236101 (2015) [33] A. Ferreira, and E. Mucciolo, Phys. Rev. Lett. 115, 106601 (2015). [34] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 [46] A. Ferreira et al., Phys. Rev. B 83, 165402 (2011). [47] Z. Fan, A. Uppstu, and A. Harju, Phys. Rev. B 89, 245422 (2014) [48] J. H. Garc´ıa, L. Covaci, and T. G. Rappoport, Phys. Rev. Lett. 114, 116602 (2015). [49] T. Stauber, N. M. R. Peres, and A. K. Geim Phys. Rev. B 78, 085432 (2008). [50] N. M. R. Peres, J. M. B. Lopes dos Santos, and T. Stauber, Phys. Rev. B 76, 073412 (2007). [51] Jason Horng, Chi-Fan Chen, Baisong Geng, Caglar Girit, Yuanbo Zhang, Zhao Hao, Hans A. Bechtel, Michael Martin, Alex Zettl, Michael F. Crommie, Y. Ron Shen, and Feng Wang Phys. Rev. B 83, 165113 (2011). [52] Y. V. Bludov, A Ferreira, N. M. R Peres, M. I. Vasilevskiy, Int. J. Mod. Phys. B 27, 1341001 (2013). [53] A. Derevianko, S. G. Porsev, and J. F. Babb, At. Data Nucl. Data Tables 96, 323 (2010). [54] A. M. Contreras-Reyes, R. Guı¿oerout, P. A. Maia Neto, D. A. R. Dalvit, A. Lambrecht, and S. Reynaud, Phys. Rev. A 82, 052517 (2010). [55] P.A.D. Gon¸calves and N. M. R. Peres, An Introduction to Graphene Plasmonics, (World Scientific, 2016, Singa- pore). 

1610.07358

1

1610

Electrically tunable artificial gauge potential for polaritons

Neutral particles subject to artificial gauge potentials can behave as charged particles in magnetic fields. This fascinating premise has led to demonstrations of one-way waveguides, topologically protected edge states and Landau levels for photons. In ultracold neutral atoms effective gauge fields have allowed the emulation of matter under strong magnetic fields leading to realization of Harper-Hofstadter and Haldane models. Here we show that application of perpendicular electric and magnetic fields effects a tuneable artificial gauge potential for two-dimensional microcavity exciton polaritons. For verification, we perform interferometric measurement of the associated phase accumulated during coherent polariton transport. Since the gauge potential originates from the magnetoelectric Stark effect, it can be realized for photons strongly coupled to excitations in any polarizable medium. Together with strong polariton- polariton interactions and engineered polariton lattices, artificial gauge fields could play a key role in investigation of non-equilibrium dynamics of strongly correlated photons.

cond-mat.mes-hall

cond-mat

Electrically tunable artificial gauge potential for polaritons Hyang-Tag Lim∗, Emre Togan∗, Martin Kroner, Javier Miguel-Sanchez, and Atac Imamoglu Institute of Quantum Electronics, ETH Zurich, CH-8093 Zurich, Switzerland. ∗These authors contributed equally to this work. Neutral particles subject to artificial gauge potentials can behave as charged parti- cles in magnetic fields. This fascinating premise has led to demonstrations of one-way waveguides, topologically protected edge states and Landau levels for photons. In ultracold neutral atoms effective gauge fields have allowed the emulation of matter under strong magnetic fields leading to realization of Harper-Hofstadter and Haldane models. Here we show that application of perpendicular electric and magnetic fields effects a tuneable artificial gauge potential for two-dimensional microcavity exciton po- laritons. For verification, we perform interferometric measurement of the associated phase accumulated during coherent polariton transport. Since the gauge potential originates from the magnetoelectric Stark effect, it can be realized for photons strongly coupled to excitations in any polarizable medium. Together with strong polariton- polariton interactions and engineered polariton lattices, artificial gauge fields could play a key role in investigation of non-equilibrium dynamics of strongly correlated photons. Synthesis of artificial gauge fields for photons have been demonstrated in a number of different optical systems. In most cases, the implementation is achieved through the design of the optical system [1 -- 4], leaving little or no room for fast control of the magnitude of the effected gauge field after sample fabrication is completed. For many applications on the other hand, it is essential to be able to tune or adjust the strength of the gauge field during the experiment [5 -- 7]; this is particularly the case for nanophotonic structures [8] where fast local control of the gauge field strength may open up new possibilities for investigation of many-body physics of light [9]. The realization we describe here has the potential to fulfill this premise since the strength and direction of the effected gauge potential is controlled electrically. Moreover, since our scheme relies on the magnetoelectric Stark effect, time reversal symmetry [10 -- 12] of the optical excitations is broken. Cavity-polaritons are hybrid light-matter quasi-particles arising from non-perturbative coupling between quantum well (QW) excitons and cavity photons. A magnetic field Bz applied along the growth direction influences polaritons through their excitonic nature and leads to a diamagnetic shift, Zeeman splitting of circularly polarized modes and enhancement of exciton-photon coupling strength [13]. For excitons with a non-zero momentum, the applied Bz also induces an electric 2 dipole moment [14 -- 17]. In a classical picture, this dipole moment is due to the Lorentz force that creates an effective electric field Eeff [18] that is proportional to (cid:126)k × (cid:126)B. For excitons with small momentum and polarizability α, this Eeff causes an induced dipole moment ((cid:126)d ∝ α(cid:126)k × (cid:126)B). An additional external electric field (cid:126)Eext in the QW plane then alters the dispersion of excitons as it leads to energy changes proportional to k: −(cid:126)d · (cid:126)Eext ∝ −α minimum is no longer at (cid:126)k = 0 as would be the case for a free particle with an effective mass but at finite (cid:126)k. This simple modification of the dispersion relation is equivalent to an effective gauge (cid:17) · (cid:126)Eext, so that the energy (cid:16)(cid:126)k × (cid:126)B potential Aeff for excitons; due to their partly excitonic character, the hamiltonian describing the dynamics of polaritons also contains a term proportional to Aeff . We first demonstrate the presence of an effective electric field whose magnitude and direction depends on the direction of propagation for polaritons subjected to an out-of-plane magnetic field. The cavity polariton sample we use is illustrated in Figure 1a. Our demonstration relies on the fact that it is possible to excite polaritons with a well defined in-plane wavevector (cid:126)k by appropriately choosing the angle and energy of the excitation beam. As illustrated in Figure 1a we choose (cid:126)k = ky y. Two metal gates deposited 30 µm apart allow us to apply an electric field in the x direction such that for polaritons propagating with a wavevector along the y direction, (cid:126)Eext will add or cancel Eeff due to the Lorentz force. By recording changes in the transition energy of polaritons as a function of Eext we can determine the strength of Eeff . In our experiments, we only address the lower polariton branch. Changes in the reflected intensity of a laser beam probing polaritons at Bz = 0 T with k+ y = 2.7 µm−1 as a function of Eext is shown in Figure 1d. For each Eext the reflected intensity shows a dip at an energy corresponding to the polariton resonance (Figure 1e). The spectral center of the dip shifts to lower energies with Eext in a way that is well described by a second order polynomial. The expected behaviour for a neutral polarizable quasi-particle (i.e. exciton or polariton) that is subject to Eext, would be to have a dc-Stark shift proportional to −αEext2 [19]; here, the coefficient α is the quasi-particle polarizability. Therefore the electric field that yields the maximum lower polariton energy identifies Eext = Eext that exactly cancels any internal or effective electric fields 3 Eeff ( Eext = −Eeff ). To quantify Eeff due to the Lorentz force described above, we extract the y = −2.9 µm−1. difference of Eext at a fixed Bz for polaritons excited with k+ This approach also allows us to exclude influences of built in electric fields. At Bz = 0 T we find that the difference of VG values that correspond to the maximum energy with k± V, indicating that Eeff is negligible at 0 T (Figure 1f). y = 2.7 µm−1 and k− y excitation is 0.02 In stark contrast, for Bz = 5 T the energy shift of polaritons with Eext displays a significant y and k− difference between Eext for k+ difference of Eext for k± with Bz, demonstrating Eeff due to the Lorentz force. y , as illustrated in Figure 2a. Magnetic field dependence of the y , Figure 2b, shows a behaviour that is well described by a linear increase The shape of the parabolas also contain further information about the Bz induced changes to the polaritons. With increasing Bz, the polarizability decreases and the polariton energy at Eext = 0 increases, as illustrated in Figure 2c. The polariton energy at Eext = 0 and the polarizability can be extracted from a fit to a second order polynomial:
0 − (cid:126)d · (cid:126)Eext − αE2 of the polynomial in turn, yields the induced dipole moment of the polaritons (excitons). The difference of the first order coefficient for ±ky excitation beams (twice the induced dipole moment) are shown in Figure 2d: we observe that the induced dipole moment first increases with Bz and then starts to slowly decrease for Bz ≥ 4 T. We expect that for the high magnetic field regime, the induced dipole moment decreases as (cid:12)(cid:12)(cid:12) ∼ 1/Bz [15]. We find very good agreement with a ext. The first order coefficient (cid:12)(cid:12)(cid:12) (cid:126)d theoretical model of the polaritons (shown as solid lines in Figure 2) that allows us to identify the physical mechanism underlying the overall Bz and Eext dependence (see Supplementary Information for the detailed information about the model). The change in the lower polariton energy is due to an interplay between a diamagnetic blue shift of the exciton transition energy and a red-shift due to an increase in the cavity-exciton coupling strength [13]. Concurrently, the polarizability decreases with Bz, as would be expected in a classical picture as the size of the polarizable particle decreases. We emphasize that due to the change in the exciton energy and cavity-exciton coupling strength the exciton content of the polaritons changes as Bz is varied. We also find that the energy shift of polaritons with Eext is influenced by the decrease in the electron hole overlap due to Eext induced polarization, which leads to a reduction of the cavity-exciton coupling strength. In the Eext range that we probe the decrease in the cavity-exciton coupling leads to an effective smaller polarizability for polaritons as compared to excitons, due to a reduction on the excitonic character of the polaritons. Moreover, the model also captures the presence of Eeff due to the center of 4 mass motion of polaritons and confirms that even with changes in exciton content with Bz, its expected dependence on Bz is still linear. Finally, we emphasize that while Eeff arises from exciton physics, the strong coupling between excitons and photons has to be taken into account to obtain a quantitative agreement between the model and our experiments. Remarkably, our findings demonstrate that polariton energy under perpendicular Bz and Eext depends not only on the magnitude of ky but also on its direction: this non-reciprocal flow of light is an indication of the presence of a gauge field for polaritons. The dispersion relation for free excitons propagating in the QW plane with wavevector ky y in the presence of an electric field (cid:126)E = (  M kyBz + Eext)x is:
exc (ky) =
exc(0) + (cid:18)  M 2k2 y 2M − α (cid:19)2 =
(cid:48) exc + kyBz + Eext (ky − qAeff )2 (cid:48) 2M , (1) where M is the total exciton mass, and
exc(0) is the exciton energy at ky = 0 and Eext = 0. ext− (qAeff )2 qAeff = M(cid:48) and M . The strong coupling to the cavity further changes the dispersion M 2αBzEext is an effective vector potential for excitons, with
(cid:48) = M exc =
exc(0)−αE2 (cid:17)−1 (cid:16) (cid:48) 2M (cid:48) 1 − αB2 z 2M relation, but, polaritons in a narrow energy-momentum range can still be treated as free particles under the influence of qA, whose value depends on the detuning between the excitons and polaritons as well as their effective masses (see Supplementary Information). For our experiments both Bz and Eext are uniform over the area of our experiment yielding a constant gauge potential qA. Since physical observables are gauge invariant, we might expect our measurements to be independent of qA. In practice, our experiments involve propagation of photons between two regions with different constant gauge potentials; when calculating the magnitude of qA above, we fixed a gauge by choosing the constant gauge potential for photons outside the cavity to be qA = 0. Conversely, qA is observable in our experiments due to the presence of an effective (sheet) magnetic field at the interface between the cavity and vacuum that imparts kicks to the canonical momentum as photons/polaritons pass through the interface [20]. To demonstrate the validity of the description of photon/polariton dynamics as determined by a gauge potential, we perform an interference experiment where we measure the phase accumulated by polaritons propagating for a length l inside the sample relative to a fixed phase reference (see Figure 3a and Methods for detailed information about the interference experiment). An interference image obtained at Eext = 0 is shown in Figure 3b. Changes in the interference pattern with y at two different Eext are illustrated in Figure 3c. A plot of the phase change of polaritons with Eext for different direction of excitation at Bz = 6 T is shown in Figure 3d. The data shows that at a fixed Eext and at the same excitation frequency, the phase accumulated for polaritons propagating along ±y directions differ, and that this phase 5 difference depends on the sign of Eext and Bz. The in-plane momentum of the polaritons with a fixed energy is modified by the absence or presence of a constant vector potential so that the additional accumulated phase for traveling a distance l is given by q  difference between the phase accumulated when exciting with k± = 2 q due to the Stark effect and to directly obtain ∆φk+ − ∆φk− y y (cid:82) (cid:126)A · dl. At a fixed Eext the y allows us to avoid phase changes − ∆φk−  Al. The fact that ∆φk+ y y depends both on the sign of the electric field as well as the magnetic field is another manifestation of the validity of the description with Aeff . Since the phase shift as a function of Eext can be modeled with a second order polynomial in Eext, the dc-Stark effect is the dominant underlying mechanism for the phase shift. Figure 3d shows that polaritons acquire a phase shift of ∼ 0.25 radians due to qA as they propagate an average distance l = 9 µm. Hence, the parameters of our experiment already ensure significant phase accumulation over the lattice constant (≥ 2.4 µm) of state of the art polariton lattices [8] or average seperation (∼ 4 µm) of coupled polariton microcavities [21]. Choosing detunings that yield predominantly exciton-like lower polaritons will further increase qA. In addition, employing structures with longer polariton lifetimes [22] will allow the realization of polariton lattices with larger lattice constants. With these advances, it should be possible to design structures where accumulated polariton phase due to qA in a unit cell could be on the order of π. Together with strong photon-photon interactions, realization of tunable artificial gauge fields is key for ongoing research aimed at observation of topological order in driven-dissipative photonic systems [23, 24]. Cavity-polaritons constitute particularly promising candidates in this endeavour since their partly exciton nature enhances interactions [25 -- 27] and ensures time-reversal-symmetry breaking by external magnetic fields [28, 29]. Our work demonstrates that the same polariton system also allows for realization of tunable gauge fields, if an in-plane electric-field gradient is introduced in the QW. While such gradients are manifest in most gated structures, creation of a constant artificial magnetic field over a region of several µm should be possible in specially designed structures. We envision that lattices of dipolaritons [30] would provide a very promising avenue towards this endeavour: the use of an in-plane magnetic field in this case would effect a gauge potential that is substantially enhanced by the large dipolariton-polarizability and strong external electric fields along the growth direction [16, 17]. The gradients of the latter in turn could be used to generate fluxes approaching magnetic flux quantum in a plaquette of area ∼ 5× 5µm2. (Di)Polaritons subject to complex electric field distributions enabling large, tunable and inhomogeneous effective gauge field profiles [31] would constitute a novel feature in exploration of topologically non-trivial non-equilibrium quantum systems. 6 I. METHODS In-plane electric field. As described in Figure 1a, we apply an electric potential VG be- tween two gates to create Eext. We find a good agreement between our data and a numerical calculation for the polariton behavior (see Supplementary Information) with the electric field Eext = 0.84 VG/(30 µm). Since the two quantities are linearly related we use Eext and VG in- terchangeably. For all experiments that involve a non-zero in-plane electric field, we acquire data in a pulse sequence (see Supplementary Information). This sequence is applied in order to avoid charge accumulation induced variation in the actual electric field. Polariton interference measurement. We use the setup depicted in Figure 3a to create two beams that are incident with different in plane momenta (k± The laser energy is chosen to be nearly resonant with the polariton modes with k± y and k = 0) and are separated by l. y . Thanks to the steep polariton dispersion, at the same energy, k = 0 beam is off resonance from the polariton modes; the photons in this beam are therefore reflected by the top mirror without being subject to Aeff . Photons in the k± y beam are converted into a propagating polariton cloud. As polaritons propagate towards the position of the k = 0 beam, they accrue an additional phase due to Aeff . Propagating polaritons continue to emit photons out of the sample, and when these photons overlap with those from the k = 0 beam they interfere, allowing us to measure changes in the accumulated phase. We model the light intensity in the region where the interference occurs (interference region) as the sum of a Gaussian (k = 0 beam) and a plane wave (polaritons propagating with ky): b0e−(y−y0)2/σ2 y + a0eiφe−ikyy2 + b1. We independently determine b0, σy, and y0 using images obtained at very high applied electric field for which the polariton term is negligible (a0 (cid:28) b0), and fix ky. For each Eext we fit for a0, b1, and φ values. 7 Since our excitation beams have a finite size, they have a finite width in ky (0.45 µm−1), therefore the exact ky value of the polaritons that are excited is determined by the energy of the excitation beam. For a fixed excitation laser energy, changes in Eext will shift the dispersion relation so that ky + ∆ky is resonant with the excitation laser. These changes (due to Aeff or Stark effect) in the dispersion relation show up as phase changes in our experiments, ∆φ = l∆ky where l is the mean distance from the excitation spot to the interference region. Since we are interested in how the phase changes with Eext and the absolute phase between the two beams is not determined, we set the phase at Eext = 0 to 0 radian. [1] Rechtsman, M. C. et al. Strain-induced pseudomagnetic field and photonic Landau levels in dielectric structures. Nature Photonics 7, 153 -- 158 (2012). [2] Rechtsman, M. C. et al. Photonic Floquet topological insulators. Nature 496, 196 -- 200 (2013). [3] Hafezi, M., Mittal, S., Fan, J., Migdall, A. & Taylor, J. M. Imaging topological edge states in silicon photonics. Nature Photonics 7, 1001 -- 1005 (2013). [4] Schine, N., Ryou, A., Gromov, A., Sommer, A. & Simon, J. Synthetic Landau levels for photons. Nature 534, 671 -- 675 (2016). [5] Aidelsburger, M. et al. Realization of the Hofstadter hamiltonian with ultracold atoms in optical lattices. Physical Review Letters 111, 185301 (2013). [6] Jotzu, G. et al. Experimental realization of the topological Haldane model with ultracold fermions. Nature 515, 237 -- 240 (2014). [7] Dalibard, J., Gerbier, F., Juzelinas, G. & Ohberg, P. Colloquium: Artificial gauge potentials for neutral atoms. Reviews of Modern Physics 83, 1523 -- 1543 (2011). [8] Jacqmin, T. et al. Direct observation of Dirac cones and a flatband in a honeycomb lattice for polaritons. Physical Review Letters 112, 116402 (2014). [9] Carusotto, I. & Ciuti, C. Quantum fluids of light. Reviews of Modern Physics 85, 299 -- 366 (2013). [10] Wang, Z., Chong, Y., Joannopoulos, J. D. & Soljacic, M. Observation of unidirectional backscattering- immune topological electromagnetic states. Nature 461, 772 -- 775 (2009). [11] Fang, K., Yu, Z. & Fan, S. Realizing effective magnetic field for photons by controlling the phase of dynamic modulation. Nature Photonics 6, 782 -- 787 (2012). [12] Tzuang, L. D., Fang, K., Nussenzveig, P., Fan, S. & Lipson, M. Non-reciprocal phase shift induced by an effective magnetic flux for light. Nature Photonics 8, 701 -- 705 (2014). [13] Pietka, B. et al. Magnetic field tuning of exciton-polaritons in a semiconductor microcavity. Physical 8 Review B 91, 075309 (2015). [14] Kallin, C. & Halperin, B. I. Excitations from a filled Landau level in the two-dimensional electron gas. Physical Review B 30, 5655 -- 5668 (1984). [15] Paquet, D., Rice, T. M. & Ueda, K. Two-dimensional electron-hole fluid in a strong perpendicular magnetic field: Exciton Bose condensate or maximum density two-dimensional droplet. Physical Review B 32, 5208 -- 5221 (1985). [16] Lozovik, Y. E., Ovchinnikov, I. V., Volkov, S. Y., Butov, L. V. & Chemla, D. S. Quasi-two-dimensional excitons in finite magnetic fields. Physical Review B 65, 235304 (2002). [17] Butov, L. V. et al. Observation of magnetically induced effective-mass enhancement of quasi-2d excitons. Physical Review Letters 87, 216804 (2001). [18] Thomas, D. G. & Hopfield, J. J. A magneto-Stark effect and exciton motion in CdS. Physical Review 124, 657 -- 665 (1961). [19] Miller, D. A. B. et al. Electric field dependence of optical absorption near the band gap of quantum-well structures. Physical Review B 32, 1043 -- 1060 (1985). [20] Fang, K. & Fan, S. Controlling the flow of light using the inhomogeneous effective gauge field that emerges from dynamic modulation. Physical Review Letters 111, 203901 (2013). [21] Rodriguez, S. R. K. et al. Interaction-induced hopping phase in driven-dissipative coupled photonic microcavities. Nature Communications 7, 11887 (2016). [22] Steger, M., Gautham, C., Snoke, D. W., Pfeiffer, L. & West, K. Slow reflection and two-photon generation of microcavity exciton -- polaritons. Optica 2, 1 -- 5 (2015). [23] Hafezi, M., Lukin, M. D. & Taylor, J. M. Non-equilibrium fractional quantum Hall state of light. New Journal of Physics 15, 063001 (2013). [24] Umucalilar, R. O. & Carusotto, I. Fractional quantum Hall states of photons in an array of dissipative coupled cavities. Physical Review Letters 108, 206809 (2012). [25] Amo, A. et al. Superfluidity of polaritons in semiconductor microcavities. Nature Physics 5, 805 -- 810 (2009). [26] Ferrier, L. et al. Polariton parametric oscillation in a single micropillar cavity. Applied Physics Letters 97, 031105 (2010). [27] Ferrier, L. et al. Interactions in confined polariton condensates. Physical Review Letters 106, 126401 (2011). [28] Nalitov, A. V., Solnyshkov, D. D. & Malpuech, G. Polariton z topological insulator. Physical Review Letters 114, 116401 (2015). [29] Karzig, T., Bardyn, C.-E., Lindner, N. H. & Refael, G. Topological polaritons. Physical Review X 5, 031001 (2015). [30] Cristofolini, P. et al. Coupling quantum tunneling with cavity photons. Science 336, 704 -- 707 (2012). 9 [31] Imamoglu, A. Inhibition of spontaneous emission from quantum-well magnetoexcitons. Physical Review B 54, R14285 -- R14288 (1996). Acknowledgements The Authors acknowledge many insightful discussions with Iacopo Caru- sotto. This work is supported by SIQURO, NCCR QSIT and an ERC Advanced investigator grant (POLTDES). 10 Figure 1: Description of the sample and characterization experiments at Bz = 0 T. a. The sample, held at 4 K in a helium bath cryostat, contains three 9.6 nm-thick In0.04Ga0.96As QWs located at an antinode of a cavity formed by two distributed Bragg reflectors (DBR). On the sample surface two metal gates are deposited with a 30 µm gap (see Supplementary Information). An electric potential VG applied to the gates creates an electric field in the x direction. Inset: With a magnetic field in the z direction polaritons excited with in plane wavevector ky exhibit a dipole moment dx. b. ky resolved photoluminescence spectra at Bz = 0 T. The position on the sample is chosen such that for (cid:126)k ∼ 0 the detuning of the cavity mode energy from the exciton resonance is less than 0.09 meV (Figure 1b) which is small compared to the exciton cavity coupling strength 5.2 meV. Gray dashed lines show the extracted bare cavity and exciton dispersions, black lines show polariton dispersions (see Supplementary Information for parameters), orange and blue lines correspond to ky = k+ y and ky = k− y , respectively. c. Experimental setup used in reflection measurements. Reflection of a linearly polarized laser beam (< 100 pW) with in-plane wavevector ky is coupled to a single mode fiber (SMF) and its intensity is detected with an avalanche-photodiode (APD). d. Changes in the reflected intensity of a laser beam exciting polaritons with in-plane wavevector k+ y as a function of laser energy and VG. Yellow dots indicate the extracted polariton resonance energies. Note that the polariton signal is missing when large VG is applied (VG > 8 V) due to the ionization of the exciton. e. Line cut of the reflection spectrum at VG = 0 V. Gray points are the measured reflection data, black solid line shows the fitted lineshape (see Supplementary Information). Dashed yellow line indicates the extracted polariton resonance energy. f. Change of polariton energy from VG = 0 V, as a function of VG for polaritons y (yellow) and k− excited with k+ 1.9 × 10−6 VG − 5.9 × 10−6 V 2 extracted from fits. y (blue) at Bz = 0 T, solid lines are fits to second order polynomials (Blue: G, Yellow: 2.2 × 10−6 VG − 5.8 × 10−6 V 2 G). Error bars are confidence intervals xyzVGdxkyBzLaserBeamSplitterNA = 0.68LensSampleAPDa.b.c.d.e.f.SMFVG = 0 V11 Figure 2: Magnetic field dependence. a. Shift of polariton energies with applied voltage for polaritons y (yellow) and k− excited with k+ y (blue) at Bz = 5 T; solid lines are fits to second order polynomials (−dV VG − αV V 2 G, Blue: dV = 5.2 × 10−6 eV/V, αV = 1.9 × 10−6 eV/V2, Yellow: dV = −3.0 × 10−6 eV/V, αV = 1.8× 10−6 eV/V2). Error bars are confidence intervals extracted from fits to the reflection line shape. Gray dashed lines indicate the voltage at which maximum energy occurs for the two parabolas, the gray arrow indicates the difference between them (∆V ). b. Using data such as a at different Bz, difference between the extracted voltage (∆V ) at which maximum energy occurs for k− y excitations. c. Change y and k+ in αV (polarizability) with Bz (filled squares, left axis, and gray curve). Change of lower polariton resonant energy with Bz at VG = 0 V (filled circles, right axis, and red curve) from the value at Bz = 0 T. Yellow and blue data points are data for k+ y and k− y , respectively. d. Difference between extracted dV values (electric dipole moment) for polaritons propagating with k+ y . For b -- d, error bars are the estimated standard y and k− deviations of the mean for three repetitions of the experiment, and the solid lines are results of a numerical calculation (see text and the Supplemental information). a.b.c.d.ΔV12 Figure 3: Demonstration of effective gauge potential for propagating polaritons. a. A single laser beam is split into two using a birefringent beam displacer, ensuring the relative phase between the two beams do not fluctuate during the experiment. Resulting two linearly polarized beams (total power 40 nW) are incident at different positions on the high numerical aperture (NA = 0.68) lens. One beam that is 1.2 mm off from the center of the lens is incident on the sample with k± y . The second beam passes through the center of the lens and thus has vanishing in-plane momentum (k = 0 beam). The sample -- lens distance is less than the focal length of the lens to ensure that the two beams are incident on the sample with their centers displaced by l ∼ 9 µm. To detect polariton flow we use an additional linear polarizer between the camera and sample that transmits light that is nearly orthogonally polarized to the incident beams. b. An interference image obtained at Bz = −6 T, VG = 0 V. Polaritons excited by the k+ propagate in the direction indicated by the white arrow. In the spatial region where photons emitted by y beam at 1.46467 eV propagating polaritons overlap with the reflected k = 0 beam an interference pattern is observed. c. In green, sum of the detected intensity of the interference pattern shown in b in the range x = [8.5 − 9.4] µm. In yellow, same as for green but shifted in intensity, for an image obtained at VG = 14.4 V. Solid lines are fits to the model described in Methods and show a difference in the extracted phase. d. Change in the extracted phase at Bz = 6 T as a function of VG for polaritons excited with k+ second order polynomial fits to the phase change as a function of VG (Blue: −0.016 VG + 0.0072 V 2 y (blue). Solid lines show y (yellow) and k− G, Yellow: LaserNA4=40.68LensSampleCameraPolarizerBeam4DisplacerPolarizera.b.d.e.Bz4=4+64TBz4=4-64Tc.VG4=404VVG4=414.44Vkyky+ beaminterference4regionlky+ beamk4=404beam

1502.03760

1

1502

Nanoscale wear and kinetic friction between atomically smooth surfaces sliding at high speeds

The kinetic friction and wear at high sliding speeds is investigated using the head-disk interface of hard disk drives, wherein, the head and the disk are less than 10 nm apart and move at sliding speeds of 5-10 m/s relative to each other. While the spacing between the sliding surfaces is of the same order of magnitude as various AFM based fundamental studies on friction, the sliding speed is nearly six orders of magnitude larger, allowing a unique set-up for a systematic study of nanoscale wear at high sliding speeds. In a hard disk drive, the physical contact between the head and the disk leads to friction, wear and degradation of the head overcoat material (typically diamond like carbon). In this work, strain gauge based friction measurements are performed; the friction coefficient as well as the adhering shear strength at the head-disk interface are extracted; and an experimental set-up for studying friction between high speed sliding surfaces is exemplified.

cond-mat.mes-hall

cond-mat

Nanoscale wear and kinetic friction between atomically smooth surfaces sliding at high speeds Sukumar Rajauria,1, ∗ Sripathi V. Canchi,1, † Erhard Schreck,1 and Bruno Marchon1 1HGST, a Western Digital Company, San Jose Research Center, San Jose, CA 95135 USA. (Dated: May 11, 2019) Abstract The kinetic friction and wear at high sliding speeds is investigated using the head-disk interface of hard disk drives, wherein, the head and the disk are less than 10 nm apart and move at sliding speeds of 5-10 m/s relative to each other. While the spacing between the sliding surfaces is of the same order of magnitude as various AFM based fundamental studies on friction, the sliding speed is nearly six orders of magnitude larger, allowing a unique set-up for a systematic study of nanoscale wear at high sliding speeds. In a hard disk drive, the physical contact between the head and the disk leads to friction, wear and degradation of the head overcoat material (typically diamond like carbon). In this work, strain gauge based friction measurements are performed; the friction coefficient as well as the adhering shear strength at the head-disk interface are extracted; and an experimental set-up for studying friction between high speed sliding surfaces is exemplified. 5 1 0 2 b e F 2 1 ] l l a h - s e m . t a m - d n o c [ 1 v 0 6 7 3 0 . 2 0 5 1 : v i X r a 1 Tribology at nano and micro scales is a field of growing interest. An in-depth under- standing of friction and its properties is essential for the design of miniaturized devices (micro-electro mechanical systems) [1]. Phenomenological laws of dry friction were estab- lished by Amontons and Coulomb, around 200 years ago, and it was empirically determined that the frictional force between two macroscopic bodies is linearly proportional to the ap- plied load, and is independent of the macroscopic 'apparent' contact area [2]. Later, Bowden and Tabor proposed that at the microscopic level, the friction force arises due to the shear strength of the contacting junction, which in turn is proportional to the 'real' contact area, and verified it through various experimental and theoretical studies [3 -- 7]. For adhering sur- faces the friction force is a sum of both the load dependent and real contact area dependent forces [3]. However, most of these frictional studies at nanometer scale are limited to sliding speeds of less than few tens of micrometers per seconds [8 -- 10]. Understanding friction at high sliding speeds, and at the nanometer scale has been par- ticularly challenging [11, 12]. It is desirable to have surfaces with sub-nanometer roughness that sustain minimal wear during high speed contact. The head-disk interface of hard disk drives provides a unique platform for such studies. In a hard disk drive, every intentional or unintentional contact between the head and the disk leads to friction and physical wear of the head overcoat layer, which typically is a diamond like carbon [15]. The head has a local microscale heater which sets both the sub-nanometer vertical clearance and a few tens of square microns of 'apparent' contact area with the disk, which is sliding at a speed of 5-10 m/s relative to the head. These sliding speeds are six orders of magnitude higher than AFM based studies [13, 14]. While the current experimental results are motivated and illustrated using the head disk interface and the associated drive components, this unique setup may be extended to perform fundamental studies on nanoscale friction at high sliding speeds using other material interfaces. In this letter, a detailed experimental study of friction between the head and the disk and the resulting head overcoat wear is presented. The disk is fabricated by depositing a magnetic multilayer structure onto a glass substrate, and coated with ≈ 3 nm amorphous nitrogenated carbon (protective overcoat layer), and finally covered with a molecular layer of perfluoropolyether polymer lubricant (≈ 1 nm thick). The head is coated with 1.4 nm of diamond like carbon on top of a 0.3 nm silicon based adhesive layer on an alumina coated substrate. The typical roughness Ra of the disk and the head surfaces are both 0.4 2 nm. Thus, the head disk interface is a smooth carbon-lubricant-carbon high speed sliding interface. The head surface is carefully etched such that an air lift force keeps it passively afloat at a fixed clearance over the disk [16]. The initial nominal clearance (physical gap) between the head and the disk is typically 10 nm. Further, clearance is controlled using an embedded micro-heater in the head (inset Figure 1). The micro-heater generates a localized protrusion on the head surface, thus bringing it in contact with the disk. Contact between the head and the disk is detected using an acoustic emission (AE) sensor, which detects elastic propagating waves generated during the head-disk contact events [17]. A calibrated strain gauge is instrumented to measure the friction force at the head disk interface along the sliding direction. Figure 1 shows a typical contact and friction measurement. As the micro-heater power increases (abscissa), a protrusion develops on the head (lateral dimensions ≈ 5 µm × 2 µm and radius of curvature of around 20 mm), bringing it closer to the disk. The contact is detected as a sharp increase in the AE signal. Further increase in the micro-heater power increases the applied normal load at the interface. Finite element based protrusion simulation is used to calculate the normal load L at the interface [18]. Simulations agree with the experiments within to 20% for various measurable quantities like initial flying height and protrusion geometry, and estimates that the normal load changes by 0.25 mN per milliwatt of heater power beyond contact. Subsequent to contact, the data shows a linear increase in the friction force with increasing normal load at the interface, the slope being the friction coefficient. In most AFM based friction and wear studies, the wear is estimated ex-situ using either the AFM or a electron microscopy of the worn track, which limits the measurement of the wear depth/volume versus sliding distance to a few data points [19 -- 21]. Ideally, to gain better insight into the wear process an in-situ monitoring of the wear is desired. In this aspect, the head-disk interface has a unique feature with the micro-heater power calibrated precisely to measure the wear depth/volume in a continous manner during the experiment. Figure 2a shows both the wear depth and wear volume on the head overcoat as a function of the sliding distance traveled in contact with the disk at a constant normal load of 2.5 mN . The head overcoat wears rapidly by 0.5 nm during the first 10 m of traveled distance. Subsequently, the wear occurs at a steady rate of 0.0028 nm/m. The initial rapid head wear may be attributed to the rapid burnishing of outlying asperities on the head, as well 3 as to the sp2 enriched content on topical overcoat layer. After remaining in contact with disk for 300 m, the protective carbon overcoat is worn out completely, and the wear rate increases significantly. This result demonstrates the superior wear resistance of the carbon overcoat, and the protection it offers to the critical structures underneath. Figure 2b shows the friction force between the head and the disk for the experiment presented in Figure 2a. It shows that the friction force increases gradually with travelled distance, and rises sharply once the carbon overcoat is worn out. Figure 2c shows the friction force as a function of the overcoat wear depth. Based on the slopes, the wear is divided into three distinct 'stages': stage-1 for the first 1.4 nm (matching the thickness of the carbon overcoat); stage-2 extends for the next 0.3 nm (matching the thickness of the Silicon based adhesive thin-film), and stage-3 for the wear of the alumina substrate material. It is interesting to note that the three stages, demarcated visually on the figure based on a change in the friction force, correspond excellently to the actual thickness of the carbon overcoat and the underlayer materials. It is noteworthy that stage-1 has the least slope, which qualitatively exemplifies that the wear resistance properties of the carbon overcoat are superior to those of the underlying materials. Figure 3a shows the friction force as a function of the overcoat wear on three different heads, where different normal load conditions are used for each case. For the range of normal load, the frictional heat is expected to increase the interface temperature by around 30-60 degrees Celsius. The experiment is restricted to stage-1, wherein only head carbon overcoat wear occurs. The friction force increases, both, with the overcoat wear depth and with increasing applied normal load condition (of 1.25, 2.5 and 3.75 mN presented here). To further analyze the data, Figure 3b shows the friction force as a function of normal load, at a fixed wear depth of 1 nm as obtained from Figure 3a, and the linear fit is also shown. The aforementioned experimental data formalizes into the form, F = µL + FA [3], where F is the total friction at the interface, µ the coefficient of friction, L the applied normal load, and FA the adhering component of friction. The slope of the linear fit in Figure 3a gives the coefficient of friction, and the ordinate intercept corresponds to the adhering friction at 'no/negligible' applied normal load. This component of friction at 'no/negligible' applied normal load may be attributed to the adhesive interaction between the atomically smooth interacting surfaces and is expected to scale with the real area of contact. Figure 4a shows the friction coefficient (similarly as in Figure 3b) at different wear depth conditions, and it shows that the coefficient of friction changes little with increasing head 4 wear. The friction coefficient obtained for this high sliding speed interface is similar to that reported in literature (ranging between 0.1-0.3), which were measured using an AFM for about six orders of magnitude slower sliding speed [22, 23]. Figure 4b shows the adhering component of the friction force FA as a function of the wear depth. In our micro-heater geometry the real area of contact increases linearly with the overcoat wear depth. This is confirmed by a linear increase in the electric current between the head and disk as a function of the wear depth (see supplementary material) for a similar head-disk interface [24]. The data supports an excellent linear behavior between the adhering friction force and the wear depth. The non-zero offset at a wear depth of 0 nm is due to the already existing contact area between the head and disk under pristine condition. This result exemplifies that the adhering component of frictional force FA takes the form, FA = σA0, where σ is the shear strength of the interface, and A0 the real area of contact at 'no/negligible' applied normal load conditions. Consequently, the slope of linear fit in Figure 4 gives the shear strength σ of the head disk interface. Estimating A0 for a high sliding speed interface, which is wearing with the sliding distance is non-trivial. Therefore, the shear strength is estimated from the measured apparent area of contact. The apparent area of contact is estimated by observing the burnish pattern on the head using an AFM. It is measured to be around ≈ 18 µm × 5 µm, which gives a shear strength of around 9 M P a, which is in the range of measured values reported in literature [4, 25]. It is noted that the apparent area of contact estimated from AFM agrees with the simulation to within 20%. In summary, a systematic study of friction and wear is performed for a high sliding speed interface using the experimental apparatus of hard disk drives. It is shown that the frictional force follows the dry frictional model even at high sliding speeds. In accordance with this model, a method to extract the coefficient of friction as well as the shear strength of the interface is outlined. This information and methodology is expected to be of great importance in understanding and improving the wear properties of nanoscale devices, which experience high sliding speeds. With particular focus on the head disk interface of hard disk drives, a thorough under- standing of the head wear process impacts the understanding of drive reliability, and hence holds great economic value to the industry. The results presented in this work demon- strate the superior wear resistance properties of the carbon overcoat and its importance in 5 protecting the critical head elements underneath. The results also show that the adher- ing component of the friction force is important, increases with head wear, and can be a significant contributor to friction during contact events. ∗ email: sukumar.rajauria@hgst.com † email: sripathi.canchi@hgst.com [1] D. Dowsin, History of Tribology (Longman, London, 1979). [2] Amontons, G. De la resistance causee dans les machniines. Mem. Acad. R. A. 275-282 (1699). [3] Berman, A., Drummond, C. and Israelachvili, J. Amontons' law at the molecular level. Tri- bology Letters 4, 95-101 (1998). [4] Homola, A. M., Israelachvili, J. N., McGuiggan P. M. and Gee M. L. Fundamental experi- mental studies in tribology: The transition from interfacial friction of undamaged molecularly smooth surfaces to normal friction with wear. Wear 136, 6583 (1990). [5] Mo, Y. F., Turner, K. T. and Szlufarska, I. Friction laws at the nanoscale. Nature 457, 1116-1119 (2009). [6] Bowden, F.P. and Tabor D. The friction and Lubrication of Solids (Claredon, 1950). [7] Carpick, R. W., Ogletree, D. F. and Salmeron, M. A general equation for fitting contact area and friction vs load measurements. Journal of Colloid and Interface Science 211, 395-400 (1999). [8] Zworner, O., Holscher, H., Schwarz, U. D. and Wiesendanger, R. The velocity dependence of frictional forces in point-contact friction. Applied Physics A-Materials Science and Processing 66, S263-S267 (1998). [9] Gnecco, E., Bennewitz, R., Gyalog, T., Loppacher, C., Bammerlin, M., Meyer, E. and Gun- therodt, H. J. Velocity dependence of atomic friction. Physical Reviw Letters 84, 1172-1175 (2000). [10] Bhushan, B. Nano- to microscale wear and mechanical characterization using scanning probe microscopy. Wear 251, 1105-1123 (2001). [11] Israelachvili, J., Min, Y., Akbulut, M., Alig, A., Carver, G., Greene, W., Kristiansen, K., Meyer, E., Pesika, N., Rosenberg, K. and Zeng, H. Recent advances in the surface forces apparatus (SFA) technique. Rep. Prog. Phys. 73, 036601 (2010). 6 [12] Lowrey, D. D., Tasaka, K., Kindt, J. H., Banquy, X., Belman, N., Min, Y., Pesika, N. S., Mordukhovich, G. and Israelachvili, J. N. High-Speed Friction Measurements Using a Modified Surface Forces Apparatus. Tribology Letters 42, 117-127 (2011). [13] Suh, A. Y., Mate, C. M., Payne, R. N. and Polycarpou, A. A. Experimental and theoretical evaluation of friction at contacting magnetic storage slider-disk interfaces. Tribology Letters 23, 177-190 (2006). [14] Tang, H., Wang, L. P., Gui, J. and Kuo, D. A study of dynamic friction at the headdisk interface. Journal of Applied Physics 87, 6152-6154 (2000). [15] Dai, Q., Yen, B. K., White, R. L., Peterson, P. J. and Marchon, B. Toward an understanding of overcoat corrosion protection. IEEE Transactions on Magnetics 39, 2450-2452 (2003). [16] Juang, J. Y., Chen, D. and Bogy, D. B. Alternate air bearing slider designs for areal density of 1 Tb/in2. IEEE Transactions on Magnetics 42, 241-246 (2006). [17] Canchi, S. V., Bogy, D. V., Wang, R. H. and Murthy A. N. Parametric Investigations at the Head-Disk Interface of Thermal Fly-Height Control Sliders in Contact. Advances in Tribology 2012, 303071 (2012). [18] Zheng, J. and Bogy, D.B. Investigation of Flying-Height Stability of Thermal Fly-Height Control Sliders in Lubricant or Solid Contact with Roughness. Tribology Letters 38, 283-289 (2010). [19] Chung, K. H. and Kim, D. E. Fundamental Investigation of Micro Wear Rate Using an Atomic Force Microscope. Tribology Letters 15, 135-144 (2003). [20] Khurshudov, A. G., Kato, K. and Koide H. Nano-wear of the diamond AFM probing tip under scratching of silicon, studied by AFM. Tribology Letters 2, 345-354 (1996). [21] Bhushan, B. and Kwak K. J. Velocity dependence of nanoscale wear in atomic force microscopy. Applied Physics Letters 91, 163113 (2007). [22] Hayward, I.P. d, Singer, I.L. and Seitzma L.E. Effect of roughness on the friction of diamond on cvd diamond coatings. Wear 157, 215 (1992). [23] Konicek, A. R., Grierson, D. S., Sumant, A. V., Friedmann, T. A., Sullivan, J. P., Gilbert, P. U. P. A., Sawyer, W. G. and Carpick, R. W. Influence of surface passivation on the friction and wear behavior of ultrananocrystalline diamond and tetrahedral amorphous carbon thin films. Phys. Rev. B. 85, 155448 (2012). [24] See supplementary material at http://scitation.aip.org/content/aip/journal/apl for the over- 7 coat wear as a fuction of current on a similar head-disk interface. [25] Israelachvili, J.N. , McGuiggan, P.M. and Homola, A.M. Dynamic properties of molecularly thin liquid films. Science 240, 189-191 (1988). 8 Figures 9 FIG. 1: Main: Normalized acoustic emission signal and the friction force between the head and the disk. Inset shows the schematic of high sliding speed head-disk interface. Vertical clearance and the normal load between the head and the disk is set using the microscale heater embedded inside the head. 10 FIG. 2: (a) Wear depth and wear volume of the head overcoat as a function of distance traveled in contact at a normal load of 2.5 mN. (b) Friction force between the head and the disk as a function of distance traveled. (c) Friction force as a function of head wear depth. 11 FIG. 3: (a) Friction force between the head and the disk at different normal load condition of 1.25, 2.5 and 3.75 mN respectively. (b) Friction force as a function of normal load, at a contant wear depth of 1 nm. 12 FIG. 4: (a) Shows the friction coefficient at different wear depth condition. (b) shows the adhering component of the friction force (FA) under no load condition (L=0) as a function of wear depth. Red dashed line shows the linear fit to the data. 13 

1001.5375

1

1001

Laser-induced Field Emission from Tungsten Tip: Optical Control of Emission Sites and Emission Process

Field-emission patterns from a clean tungsten tip apex induced by femtosecond laser pulses have been investigated. Strongly asymmetric field-emission intensity distributions are observed depending on three parameters: (1) the polarization of the light, (2) the azimuthal and (3) the polar orientation of the tip apex relative to the laser incidence direction. In effect, we have realized an ultrafast pulsed field-emission source with site selectivity of a few tens of nanometers. Simulations of local fields on the tip apex and of electron emission patterns based on photo-excited nonequilibrium electron distributions explain our observations quantitatively. Electron emission processes are found to depend on laser power and tip voltage. At relatively low laser power and high tip voltage, field-emission after two-photon photo-excitation is the dominant process. At relatively low laser power and low tip voltage, photoemission processes are dominant. As the laser power increases, photoemission from the tip shank becomes noticeable.

cond-mat.mes-hall

cond-mat

Laser-induced Field Emission from Tungsten Tip: Optical Control of Emission Sites and Emission Process Hirofumi Yanagisawa1,∗ Christian Hafner2, Patrick Don´a1, Martin Klockner1, Dominik Leuenberger1, Thomas Greber1, Jurg Osterwalder1, and Matthias Hengsberger1 1Physik Institut, Universitat Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland 2Laboratory for Electromagnetic Fields and Microwave Electronics, ETH Zurich, Gloriastrasse 35, CH-8092 Zurich, Swizerland (Dated: August 7, 2018) Field-emission patterns from a clean tungsten tip apex induced by femtosecond laser pulses have been investigated. Strongly asymmetric field-emission intensity distributions are observed depending on three parameters: (1) the polarization of the light, (2) the azimuthal and (3) the polar orientation of the tip apex relative to the laser incidence direction. In effect, we have realized an ultrafast pulsed field-emission source with site selectivity of a few tens of nanometers. Simulations of local fields on the tip apex and of electron emission patterns based on photo-excited nonequilibrium electron distributions explain our observations quantitatively. Electron emission processes are found to depend on laser power and tip voltage. At relatively low laser power and high tip voltage, field-emission after two-photon photo-excitation is the dominant process. At relatively low laser power and low tip voltage, photoemission processes are dominant. As the laser power increases, photoemission from the tip shank becomes noticeable. PACS numbers: 79.70.+q, 73.20.Mf, 78.47.J-, 78.67.Bf I. INTRODUCTION Field emission from metallic tips with nanometer sharpness has been introduced some time ago as highly bright and coherent electron source [1 -- 8]. Only recently, pulsed electron sources with high spatio-temporal reso- lution were realized by laser-induced field emission from such tips [9 -- 12]. Potentially, spatio-temporal resolution down to the single atom and the attosecond range ap- pears to be possible [7, 8, 10]. Such electron sources would be very attractive for applications in time-resolved electron microscopy or scanning probe microscopy. How- ever, the interaction of the laser pulses with the sharp tip and the electron emission mechanism are not yet fully understood [9, 11 -- 13]. When a focused laser pulse illuminates the tip, optical electric fields are modified at the tip apex due to the exci- tation of surface electromagnetic (EM) waves that couple with collective surface charge excitations to form e.g. sur- face plasmon polaritons. Interference effects of the result- ing surface EM waves can lead to local field enhancement [14]. Depending on the field strength, different electron emission processes become dominant [9]. For relatively weak fields, single electron excitations by single- or multi- photon absorption are dominant, and photo-excited elec- trons are tunneling through the surface potential barrier; such processes are termed photo-field emission. On the other hand, very strong local fields largely modify the tunneling barrier and prompt the field emission directly, leading to optical field emission. So far, the different emission processes were disputed in the literature, while the local field enhancement was treated as a static effect ∗Electronic address: hirofumi@physik.uzh.ch such as the lightening rod effect [9, 10, 15 -- 17]. Hence, local fields on the tip apex are considered to be symmet- ric with respect to the tip axis. However, when the tip size is larger than approximately a quarter wavelength, dynamical effects are predicted to occur [18]. Here, we used a tip whose apex was approximately a quarter wavelength and we investigated laser-induced field-emission patterns. We have found that dynami- cal effects substantially influence the symmetries of lo- cal field distributions and thereby field-emission intensity distributions. Varying the following three parameters changes these distributions substantially: (1) the laser polarization, (2) the azimuthal and (3) the polar orienta- tion of tip apex relative to the laser incidence direction. These are effects that had not been observed in earlier experiments [9, 19]. At the same time, we realized an ultrafast pulsed field-emission source with emission site selectivity on the scale of a few tens of nanometers. In our previous paper, simulations confirm that the photo- field emission process is dominant in laser-induced field emission [20]. Here, we further discuss electron emis- sion processes and their dependence on laser power and tip voltage by investigating electron emission patterns, Fowler-Nordheim plots, and calculated electron energy distributions. At relatively low laser power and high tip voltage, field emission after two-photon photo-excitation is the dominant process. At still relatively low laser power and low tip voltage, multiphoton photoemission over the surface barrier is dominant. As laser power in- creases, photoemission from the tip shank contributes. This manuscript consists of four main sections. In sec- tion II, we explain our experimental setup and our theo- retical model. In section III, we discuss the optical con- trol of field-emission sites and the emission mechanism based on simulations of local fields on the tip apex and laser-induced field-emission microscopy (LFEM) images. In section IV, we discuss the emission processes for vary- ing laser power and tip voltage based on experimental and calculated results. In the last section, we present conclusions and proposals for future experiments. II. METHODOLOGY A. Experimental setup Fig. 1(a) schematically illustrates our experimental setup. A tungsten tip is mounted inside a vacuum cham- ber (3 · 10−10 mbar). Laser pulses are generated in a Ti:sapphire oscillator (center wavelength: 800 nm; rep- etition rate: 76 MHz; pulse width: 55 fs) and intro- duced into the vacuum chamber. An aspherical lens (fo- cal length: f = 18 mm) is mounted on a holder that is movable along the y direction and located just next to the tip to focus the laser onto the tip apex; the diam- eter of the focused beam is approximately 4 µm (1/e2 radius) measured by a method with a razor blade [21]. Linearly polarized laser light was used. The polarization vector can be changed within the transversal (x, z) plane by using a λ/2 plate [22]. As shown in the inset, where the laser propagates towards the reader's eye as denoted by the circled dot, the polarization angle θP is defined by the angle between the tip axis and the polarization vector. The tip can be heated to clean the apex and also nega- tively biased for field emission. Although the tip is poly- crystalline tungsten, heating to around 2500◦C leads to the tip apex being crystallized and oriented towards the (011) direction [4, 23]. A position sensitive detector with a Chevron-type double-channelplate amplifier in front of the tip is used to record the emission patterns. The spa- tial resolution of field emission microscopy (FEM) is ap- proximately 3 nm [4]. The tip holder can move along three linear axes (x, y, z) and has two rotational axes for azimuthal (ϕ, around the tip axis) and polar (θ, around the z axis) angles [24]. The laser propagates parallel to the horizontal y axis within an error of ± 1 degree. θ is set so that the tip axis is orthogonal to the laser propaga- tion axis. The orthogonal angle in θ was determined by plotting positions of the tip in (x, y) coordinates while keeping the tip apex in the focus of the laser as schemat- ically shown in Fig. 1(b). The maximum position in x gives the orthogonal angle. Experimental data are shown in Fig. 1(c). The plots were taken in 0.5 degree steps. The data clearly show the maximum in the x position. We defined the corresponding angle as θ = 0 for conve- nience. The precision is estimated to be ± 1 degree. In these experiments, the base line of the rectangular detec- tor is approximately 20◦ off from the horizontal (y axis) incidence direction, which means that the laser propaga- tion axis is inclined by 20◦ from the horizontal line in the observed laser-induced FEM images (see dashed red arrow in Fig. 3a). All the measurements were done at room temperature. 2 FIG. 1: (color online). Schematic diagram of the experimental setup (a). A tungsten tip is mounted inside a vacuum cham- ber. Laser pulses are generated outside the vacuum chamber. An aspherical lens is located just next to the tip to focus the laser onto the tip apex. Emitted electrons are detected by a position-sensitive detector in front of the tip. The po- larization angle θP is defined in the inset, where the laser beam propagates towards the reader's eye (see text for fur- ther description). (b) shows a schematic diagram defining the orthogonal angle between the laser propagation direction and the tip axis. The right angle is where x is maximum. (c) shows tip positions in (x, y) coordinates for different θ, found while the tip apex is kept in the focus of the laser. The angle which gives maximum x is defined as θ = 0 for convenience. B. Theoretical model Although the field emission is a quantum mechanical phenomenon, the interaction between the optical fields and the tungsten tip apex can be treated classically by solving Maxwell equations. Such an interaction can be understood by a mechanistic picture as shown in Fig. 2(a). When a laser pulse illuminates the metallic tip, surface EM waves are excited, which propagate around the tip apex. As a result of the interference among the excited waves, the optical fields are modulated. To sim- ulate the superposition of surface EM waves and the re- sulting local field distributions on the tip apex, we used the Multiple Multipole Program (MMP) [25 -- 27], which is a highly accurate semi-analytic Maxwell solver, avail- able in the package MaX-1 [28] and in the open source project OpenMaX [29, 30]. A droplet-like shape was employed as a model tip as shown in Fig. 2(b), with a radius of curvature of the tip apex of 100 nm, which is a typical value for a clean tungsten tip. Atomic structures were not included in the model becasue the tip apex can be regarded as a smooth surface on this length scale given by the tip dimensions and the wavelength of the laser field. The dielectric func- tion ǫ of tungsten at 800 nm was used, i.e. a real part Re(ǫ) = 5.2 and an imaginary part Im(ǫ) = 19.4 [31]. Note that accuracy of the dielectric functions does not affect our conclusion, which will be demonstrated in sec- FIG. 2: (color online). Schematic illustration of the exci- tation and interference of surface EM waves (a) and of the model tip (b). The radius of curvature of the tip apex is 100 nm, and the length is 700 nm. (c) represents the field distribution of the focused beam used in the simulation at a certain time. The beam waist is 1 µm and the wavelength is 800 nm. Small arrows indicate the field direction, and field strength is represented by a linear color scale: highest field values are represented in yellow (brightest color). The calcu- lated time-averaged field distribution around the model tip is shown in (e) for θP = 0◦ together with a magnified picture in the vicinity of the tip apex. (e) shows the time-averaged field distribution around a longer model tip together with a mag- nified picture in the vicinity of the tip apex; the tip length is twice as large as that of (d). (f) represents the time-averaged field distributions around the model tip simulated by incident laser light with wavelengths of 750 nm, 800 nm and 850 nm. 3 tion III B by comparing with local fields on a gold tip. A focused laser with a beam waist of 1 µm and a wave- length of 800 nm was used as shown in Fig. 2(c). The model tip was set so that its apex is at the center of the focus. By using different droplet sizes it was verified that the model tip is long enough so as to mimick the infinite length of the real tip; the fields at the truncated side of the tip are substantially weaker so that the excited surface EM waves propagating arond the whole tip do not affect the induced field distribution at the tip apex. Figure 2(d) shows the calculated time-averaged field dis- tribution around the model tip of Fig. 2(b). Fig. 2(e) shows the same calculated field distribution for a longer tip with the same radius of curvature of the tip apex of 100 nm. In both cases, the laser is propagating from left to right, where the polarization vector has been chosen parallel to the tip axis (θP = 0◦). The magnified pictures around the tip apex of Fig. 2(d) and 2(e) show that the local field distributions of the two are basically the same, indicating the length of the model tip in Fig. 2(b) to be long enough. In the simulations, we only used a center wavelength of 800 nm, even though the laser pulse has a spectral width of ∆ = 25 nm with respect to the center wavelength. Jus- tification of the use of only a center wavelength is done by simulating the local electric fields with laser light of wavelengths of 750 nm and 850 nm. Fig. 2(f) shows the time-averaged field distributions around the model tip obtained by excitation with these three wavelengths. They are almost the same, which indicates that the sub- stantial spectral width of the light around 800 nm would not affect the position of the maximum local electric fields simulated with the wavelength of 800 nm. III. OPTICAL CONTROL OF FIELD-EMISSION SITES AND EMISSION MECHANISM A. Experimental Results The field emission pattern from the clean tungsten tip apex which orients towards the (011) direction is shown in Fig. 3(a). The most intense electron emission is ob- served around the (310)-type facets, and weaker emission from (111)-type facets. These regions are highlighted by green areas with white edges on the schematic front view of the tip apex in the inset of Fig. 3(a). The intensity map roughly represents a work function map of the tip apex: the lower the work function is, the more electrons are emitted. The relatively high work functions of (011)- and (001)-type facets [32] suppress the field emission from those regions. The laser-induced FEM (LFEM) image in Fig. 3(b), taken with the light polarization oriented parallel to the tip axis (θP = 0), shows a striking difference in symmetry compared to that of the FEM image in Fig. 3(a). Emis- sion sites are the same in both cases, but the emission pattern becomes strongly asymmetric with respect to the 4 shadow (right) and exposed (left) sides to the laser inci- dence direction. The most intense emission is observed on the shadow side as illustrated in the inset of Fig. 3(b). Figs. 3(c) and 3(d) give the same comparison for a dif- ferent azimuthal orientation of the tip as shown in the inset of Fig. 3(c). In the magnified image of Fig. 3(c), two emission sites can be identified which are separated by approximately 30 nm. The strong emission asymme- try is observed even over such short distances, as shown in Fig. 3(d). Actually, the laser pulses arrive at an an- gle of 20◦ off the horizontal line in both LFEM images, as indicated by the dashed red arrow in the inset of Fig. 3(a). This oblique incidence slightly affects the symmetry with respect to the central horizontal line in the observed LFEM images (see below). The asymmetry in LFEM images can be controlled fur- ther by changing θ. In Fig. 4, the θ-dependence of LFEM images at [ϕ = 0◦, θP = 0◦] is shown, which were taken at Vtip = -1500 V and PL = 20 mW. θ is varied from θ = −12◦ to θ = 12◦ by 4 degree steps. Schematic di- agrams for the experimental configuration are shown at the top, in which red arrows indicate the laser propaga- tion direction. The corresponding FEM images, which were taken at Vtip = -2200 V, are also shown. As θ in- creases, the asymmetry of the LFEM images becomes clearly enhanced. At θ = 12◦, electrons are emitted al- most only from right-side emission sites. Among these θ FIG. 3: (color online). Electron emission patterns for two orthogonal azimuthal orientations (ϕ) of the tip without laser [ϕ = 0◦ (a) and ϕ = 90◦ (c)], and with laser irradiation [ϕ = 0◦ (b) and ϕ = 90◦ (d)]. Vtip indicates the DC potential applied to the tip and PL indicates the laser power measured outside the vacuum chamber. The insets of (a) and (c) show the front view of the atomic structure of a tip apex with a cur- vature radius of 100 nm, based on a ball model, in which green areas with white edges indicate the field emission sites and the red arrow indicates the laser propagation direction. The inset of (b) shows a schematic side view of the laser-induced field emission geometry, in which green vectors indicate intensities of electron emission and the white arrow indicates the laser propagation direction. A dashed white line denotes a mirror symmetry line of the atomic structure in each picture. In (c) and (d) specific regions of interest, marked by dashed red rectangles, are blown up on the right hand side. FIG. 4: (color online). θ-dependence of LFEM images at [ϕ = 0◦, θP = 0◦], which were taken at Vtip = -1500 V and PL = 20 mW. θ is varied from θ = −12◦ to θ = 12◦ by 4◦ steps. Schematic diagrams for the experimental configuration are shown at the top, in which red arrows indicate the laser propagation direction. The corresponding FEM images are also shown below, which were taken at Vtip = -2200 V. The white dashed lines in the pictures denotes a mirror symmetry line of the atomic structure. The total yield Sright from right side of each image and the total yield from left side Slef t with respect to the white dashed line were taken. The ratio of Sright to Slef t are plotted in the graph. Blue circles are for LFEM and black squres are for FEM. 5 FIG. 5: (color online). Comparison of measured and simulated laser-induced FEM (LFEM) images for different light polarization angles θP and for different azimuthal orientations ϕ of the tip. The leftmost column gives the FEM images without laser irradiation for four different azimuthal angles (Vtip = -2250 V). For the same azimuthal angles, observed LFEM images are shown as a function of polarization angle θP in 30◦ steps (Vtip ≈ -1500 V and PL = 20 mW). The simulated LFEM images from the photo-field emission model, in which Vtip and PL were set as in the corresponding experiments, are shown on the right-hand side of the observed LFEM images. The color scale and laser propagation direction are the same as in Fig. 3. values, the symmetry of the FEM images changed only slightly due to a change of the DC field distribution on the tip apex. To distinguish between the contributions of DC and laser field distributions to the asymmetry of the LFEM images, we evaluated the change in symme- try quantitatively. The total yield Sright from the right side of each image and the total yield from the left side Slef t were obtained from each image with respect to the white dashed line. Then the ratios of Sright to Slef t are plotted in the graph: high values indicate large asym- metries. The asymmetry is clearly enhanced in LFEM with respect to FEM, which indicates that the laser fields mainly contribute to enhance the asymmetry for higher angles θ. We also found experimentally a strong dependence of the electron emission patterns on the laser polarization direction and on the azimuthal tip orientation. Fig. 4 shows the LFEM patterns for different values of θP in 30◦ steps, and for four different azimuthal orientations ϕ of the tip. The corresponding FEM images are also shown in the left-most column; they show simply the azimuthal rotation of the low work function facets around the tip axis. Throughout the whole image series, the emission sites do not change, but intensities are strongly modulated resulting in highly asymmetric features. For instance, for [ϕ = 0◦, θP = 0◦] the intense emission sites are located on the right-hand (shadow) side of the tip, but for [ϕ = 0◦, θP = 60◦] the emission sites on the left- hand side become dominant. LFEM images recorded for θP = 180 (not shown) are exactly the same as those for θP = 0◦, and all the LFEM images are well reproducible. B. Simulations of Local fields When a laser pulse illuminates the metallic tip, surface EM waves are excited, which propagate around the tip apex as illustrated schematically in Fig. 2(a). Due to the resulting interference pattern, the electric fields show an asymmetric distribution over the tip apex, depending also on the laser polarization. Fig. 6(a) shows the time evolution of laser fields at 800 nm wavelength over a cross section of the model tip while propagating through the tip apex from left to right, where the polarization vector has been chosen parallel to the tip axis (θP = 0◦). It can be seen that a surface EM wave is propagating around the tip apex indicated by white arrows and enhanced at the tip apex. The calculated time-averaged field distribution around the tip apex is shown in Fig. 6(b). The field distribution is clearly asymmetric with respect to the tip axis, with a maximum on the shadow side of the tip. The field enhancement factor of the maximum point is 2.5 with respect to the maximum field value of the incident laser, and 1.7 for the counterpart of the maximum point on the side exposed to the laser. This is consistent with our observations in Fig. 3(b) and 3(d) where the field emission is enhanced on the shadow side. Additionally, we would like to note that a similar asym- metric distribution can also be seen even for a metal with a dielectric function which is largely different from that of tungsten. For example, we performed a simulation for gold using a real part Re(ǫ) = −24 and an imaginary part Im(ǫ) = 1.5 [33]. Fig. 6(c) shows the time-averaged field distribution on the gold tip apex. The field distribu- tion shows a similar asymmetry as for tungsten. It should also be mentioned that surface EM waves are classified in terms of the dielectric functions of the interacting mate- rial [34] though some authours do not distinguish. If the real part of the dielectric function is negative, then the surface EM waves are proper surface plasmon polaritons. On the other hand, if the real part is positive and the imaginary part is large, the term Zenneck waves is more appropriate. The dielectric functions of tungsten and gold between 700 nm and 900 nm are plotted by black 6 4 where the most asymmetric emission is observed at θ = 12◦. Fig. 6(f) shows, in a front view, time-averaged field distribution maps from the white dashed line region of the model tip in Fig. 6(b). This area corresponds roughly to the observed area in our experiments. The red arrows indicate the laser propagation direction which has been set to the same as in our experimental situation. The field distribution changes strongly depending on the polarization angle. While the maximum field is located directly on the shadow side of the tip for θP = 0◦, the maximum moves towards the lower side of the graphs in concert with the polarization vector for θP = 30◦ and θP = 60◦, and reappears on the upper side for θP = 120◦ and θP = 150◦. For θP = 90◦ the polarization vector is perpendicular to the tip axis and produces two sym- metric field lobes away from the tip apex. In general the observed LFEM images show the same intensity modu- lations (Fig. 5): each LFEM pattern at θP = 30◦ and 60◦ shows pronounced emission at the lower side of the image, while each LFEM pattern at θP = 120◦ and 150◦ has maximum emission at the upper side of the image. C. Simulations of LFEM by the photo-field emission model From the calculated local fields, we further simulated the LFEM images by considering the photo-field emission mechanism. The current density jcalc of field emission can be described in the Fowler-Nordheim theory based on the free-electron model as follows [4, 5, 35, 36], jcalc = em 2π23 Z ∞ −Wa Z W =E −Wa D(W, Φ, F )f (E) dW dE. (1) Here, e is the electron charge and m the electron mass, −Wa is the effective constant potential energy inside the metal, W is the normal energy with respect to the sur- face, and E is the total energy. Important factors are D(W, Φ, F ) and f (E). D(W, Φ, F ) is the probability that an electron with the normal energy W penetrates the sur- face barrier. It depends exponentially on the triangular- shaped potential barrier above W, as indicated by the cross-hatched area in Fig. 7(a) where field emission with energy W occurs. The cross-hatched area is determined by the work function Φ and the electric field F just out- side the surface. f (E) is an electron distribution func- tion. In the case of field emission we have F = FDC , where FDC is the applied DC electric field, and f (E) is the Fermi-Dirac distribution at 300 K as shown in Fig. 7(a). In the photo-field emission model, F still equals FDC , but the electron distribution is strongly modified by the electron-hole pair excitations due to single- and multi-photon absorption, resulting in a nonequilibrium distribution characterized by a steplike profile, as illus- trated in Fig. 7(c) [13, 37]. For example, one-photon ab- sorption creates a step of height S1 from EF to EF +hν by exciting electrons from occupied states between EF − hν and EF . Absorption of a second photon creates a step FIG. 6: (color online). Time evolution of laser fields over a cross section of the model tip while propagating through the tip apex from left to right (a). The polarization vector is parallel to the tip axis (θP = 0◦). Small black arrows indicate the field direction, and field strength is represented by a linear color scale: highest field values are represented in yellow (brightest color). The time-averaged field distribution for tungsten and gold tips are shown in (b) and (c) where the model tip of Fig. 2(b) is employed for both. The dielectric function of tungsten and gold for the wavelengths between 700 nm and 900 nm are plotted by black dots in (d) and the values at 800 nm are highlighted by red circles. (e) shows the time-averaged field distributions around the tungsten tip for three different polar angles: θ = −12◦, θ = 0◦, and θ = 12◦. In (f) the time-averaged field distributions are given in a front view of the model tip for different polarization directions θP (θ = 0◦). The laser propagation direction is indicated by red arrows, and is the same as in the experiment. dots in Fig. 6(d), where the values at 800 nm are high- lighted by red circles. From Fig. 6(d), strictly speaking, the excited surface EM waves on tungsten are Zenneck waves and those on gold are surface plasmon polaritions. Figs. 6(b) and (c) also indicate that different kinds of surface EM waves do not show substantial difference in the resulting field distribution. The asymmetric local field distribution can be con- trolled by changing the polar angle θ and the laser po- larization angle θP . Fig. 6(e) shows time-averaged field distributions on the tungsten tip apex with different laser incidence directions relative to the polar orientation of the tip apex. As θ increases, the asymmetry becomes stronger. This is consistent with our observation in Fig. 7 FIG. 7: (color online). A schematic diagram of field emission from a Fermi-Dirac distribution (a), where an electron with a normal energy W is emitted. The surface barrier above W is shown by a cross-hatched area. (b) shows the logic diagram for obtaining work function and absolute DC field maps. (c) and (d) show schematic diagrams of photo-field emission from a nonequilibrium electron distribution and optical field emis- sion from a Fermi-Dirac distribution, respectively. of height S2 from EF + hν to EF + 2hν, where S2 ≈ S2 1 . We included absorption of up to four photons. The step height S1 is proportional to the light intensity I. In the vicinity of the tip we have I ∝ F 2 laser where Flaser is the enhanced optical electric field that varies over the tip apex as illustrated in Fig. 6(f) [38]. There are three adjustable parameters in our calcu- lations of jcalc: Φ, FDC , and S1. They are all func- tions of position on the tip apex. Φ and FDC maps on the tip apex were obtained from the measured FEM im- ages by following the logic diagram shown in Fig. 7(b). The measured FEM images, which were symmetrized to have the ideal two-fold symmetry, represent the current density jexp as a function of position on the tip apex, TABLE I: Table of work functions of tungsten for several faces. An error in experimental values can be as much as ± 0.3 eV [40]. Facet type Simulation Experiment (011) 4.9 eV 5.25 eVa (001) 4.6 eV 4.63 eVa (031) 4.45 eV 4.35 eVb aReference [32]. bReference [39]. cReference [40]. FIG. 8: (color online). The calculated DC field distribution around the model tip is shown in (a). A wire-like shape was employed for the simulation of relative DC fields. The color scale is the same as in Fig. 6. The relative DC field distribu- tion at the tip apex of (a) is shown as a function of angle θc, which is defined in (a). The obtained work function profile along a (001)-(011)-(010) curve is shown in (c) as a function of θc. because the electrons follow closely the field lines from the tip apex to the position sensitive detector. In prac- tice we assumed a radius of curvature of the tip apex of 100 nm and used the FEM image at ϕ = 45◦ shown in Fig. 5. Second, a relative DC field FDC relative distribu- tion was generated by MaX-1. We used a more wire-like tip shape for this purpose as shown in Fig. 8(a), and a grounded plate was set 1 cm away from the tip, which is close to that in our experimental setup. The simulated FDC relative is shown in Fig. 8(b), which is normalized by the value at tip apex. Going away from the tip apex FDC relative decreases. A scaling factor α is introduced, which determines FDC by FDC = α · FDC relative. We then obtained the Φ map by inserting FDC into Eq. (1) and postulating jexp − jcalc = 0. The resulting Φ map was compared to known values for several surface facets of tungsten. The scaling factor α was changed and the procedure was iterated until reasonable work functions were obtained. Thus, a full Φ map and absolute values for FDC were determined. A line profile of the resulting Φ map along the (001)- (011)-(010) curve is shown in Fig. 8(c). The work func- tion has local curve maxima at the (011)- and (001)- type facets and local minima at (310)- type facets, which is in line with the observed field emission intensity distribu- tion seen in Fig. 3(a); the higher the work function, the lower the intensity. The resulting Φ values are summa- rized for several facets and compared with known experi- mental values in Table 1. They are in fair agreement with each other. A field strength FDC of 2.15 V/nm results at the tip apex center for the FEM image taken with Vtip = −2250 V, which is a typical value for FEM. The LFEM experiments were carried out with a reduced tip voltage Vtip ≈ −1500 V. Therefore we used a down-scaled value of 1.43 V/nm in the LFEM simulations. Note that the uncertainty in the Φ values is not important for our conclusions which will be discussed below: we have also checked the whole discussion in this section with a differ- ent work function map using 4.6 eV, 4.32 eV, 4.20 eV for (011), (001) and (310) facets, respectively, but the main outcome does not change. Substituting the obtained Φ and F distribution maps into Eq. (1), and using a nonequilibrium electron distri- bution f (E), the absolute values of S1 over the tip apex were determined by fitting the measured total current from the (310) facet on the right-hand side of the LFEM image in Fig. 3(b). The resulting maximum value for S1 was 1.6 · 10−6. By substituting all the adjusted pa- rameters into Eq. (1), we could simulate all the LFEM images. The calculated current densities on the tip apex were projected to the flat screen by following the static field lines. The simulated images can now directly be compared to the experimental images (Fig. 5): they are in excellent agreement in every detail. This comparison clearly demonstrates that the observed strongly asym- metric features originate from the modulation of the local photo-fields. D. Simulations of LFEM by the optical field emission model We also simulated the LFEM images for the optical field emission process and compared the resulting in- tensity distributions to those of the photo-field emission model. In the optical field emission model schematically shown in Fig. 7(d), the Fermi-Dirac distribution is not modified, but instead the electric field F in Eq. (1) is expressed as F = FDC + F ⊥ laser is the nor- mal component of Flaser at each point of the tip surface. The absolute values for F ⊥ laser on the tip apex were de- termined in the same way as described above for S1. The resulting maximum value for F ⊥ laser where F ⊥ laser was 0.71 V/nm. Fig. 9(a) shows the LFEM image for the optical field emission process together with experimentally obtained LFEM and the simulated LFEM image based on the photo-field emission model at [ϕ = 0◦, θP = 0◦]. The op- tical field emission model results in an even more strongly asymmetric pattern as compared to the photo-field emis- sion model. This contrasts with the experimental data. Fig. 9(b) shows line profiles extracted from the observed FEM and LFEM images, and from the corresponding simulations for both models, which are all normalized to their maximum value. The measured LFEM profile clearly shows the asymmetric feature observed in the re- gions B and D. The photo-field emission model catches this asymmetry much more quantitatively than the op- tical field emission model, as can be best seen in region D. Therefore, the local fields in our experiment are still 8 FIG. 9: (color online). (a) experimentally obtained LFEM image and simulated LFEM image at [ϕ = 0◦, θP = 0◦] for both photo-field emission and optical field emission models. (b) shows line profiles extracted from the observed FEM im- ages (red line with squares), LFEM images (blue line with circles), and from LFEM images simulated by the photo-field emission model (green solid line) and the optical field emission model (black dashed line) at [ϕ = 0◦, θP = 0◦]. The whole scanned line corresponds to the unfolded rectangle indicated by the dashed blue line in the FEM figures above, and the corresponding sides are indicated by white arrows. Each line profile has been normalized by the maximum value. weak enough such that the photo-field emission process is the dominant one. IV. VOLTAGE- AND POWER- DEPENDENCE OF EMISSION PROCESSES A. Lower laser power In this section, we futher discuss the details of elec- tron emission processes and their dependence on laser power and tip voltage by investigating electron emission patterns and Fowler-Nordheim plots, and by simulating electron energy distributions. Figure 10 shows the de- pendence of LFEM images at [ϕ = 0◦, θP = 0◦] on the average laser power PL and the tip voltage Vtip applied to the tip. Throughout the whole seriese of pictures, the intensity of each image was normalized by the maximum intensity. Normally, total yields decrease as either laser power or tip voltage decrease. As can be seen, the left- right asymmetry is present in all images below a laser power of 60 mW except when PL = 0 mW (FEM). Nev- ertheless, the images show a trend: the outlines of emis- sion facets become diffuse in the lower tip voltage region, see e.g. the 20 mW row surrounded by a white dashed line. In these experiments, electron emission is consid- ered to be a concerted action of photoemission and field 9 FIG. 10: (color online). Laser power and tip voltage dependence of the electron emission patterns at [ϕ = 0◦, θP = 0◦]. On the vertical axis laser power, on the horizontal axis tip voltage are plotted. The inset shows the electron emission patterns where the laser beam is displaced from the tip apex downwards by a distance of 1 µm and 16 µm, taken at the 90 mW laser power. The definition of distance d is also shown on the right-hand side of the inset. The time-averaged field distribution around the tip when the laser beam is displaced by a distance of 0.5 µm is also shown in the inset, where the longer model tip shown in Fig. 2(e) was used. The green dashed circles highlight the left-side electron-emission sites (see section IV B). emission. In the case of field emission, the emitted elec- trons strongly feel the work function corrugation on the nanometer scale, which generates a sharp contrasts at the border of each emission facet. In the case of photoemis- sion, the excited electrons encounter a much narrower surface barrier and appear thus to be less sensitive to the work function corrugation, hence showing a smeared contour of the emission sites. At lower tip voltage, mul- tiphoton processes will be enhanced since field emission is suppressed. Eventually, photoemission from 3PPE or 4PPE will contribute, with energies above the vacuum level. Therefore, the outline of each emission facet be- comes diffuse in the lower tip voltage region. This is confirmed by simulations of energy distribution curves and emission patterns. We have simulated the energy distribution of field- emitted electrons with the parameters representing the emission site where maximum intensity can be seen in the simulated image in Fig. 5 at [ϕ = 0◦, θP = 0◦]: the work function at this point is 4.45 eV, the DC field is 1.32 V/nm and S1 is 1.6·10−6, which should correspond to the conditions where the most intense point can be seen in 10 FIG. 11: (color online). Simulated electron energy distribu- tion curves (a). The spectrum at the top of (a) shows the sim- ulated electron energy distribution of field-emitted electrons with the parameters representing the point where maximum laser intensity can be observed for ϕ = 0◦, θP = 0◦. The parameters are the following: the work function is 4.45 eV , a DC field of 1.32 V/nm corresponds to a tip voltage of -1500 V and S1 is 1.6 · 10−6 for 20 mW laser power. The energy distri- bution for lower DC fields but the same laser power are also shown in (a). The vacuum level Evac is defined as 0 eV, and the Fermi level EF is 4.45 eV below Evac. The energies cor- responding to one-, two- and three-photon excitations from EF (1PPE, 2PPE and 3PPE) are also indicated by vertical dashed lines. The simulated LFEM images at corresponding tip voltage and laser power are shown in (b). the image at 20 mW laser power and -1500 tip voltage in Fig. 10. Figure 11(a) shows the corresponding simulated energy spectrum at the top, and underneath, spectra for various lower tip voltages. We find that field emission from two-photon processes is strongly dominant at the higher DC fields. For lower DC voltage, the calculated energy distributions clearly show that field-emission from the 2PPE line is suppressed and photoemission from the 3PPE line is enhanced. The simulated LFEM images at corresponding tip voltage are shown in Fig. 11(b). As the DC voltages decrease, the outlines of emission sites be- come diffuse due to the fact that photoemission processes become dominant. This is in line with the experimental observations described above. Experimental Fowler-Nordheim (FN) plots give further suppports for the suggested emission mechanism. Fig. 12 shows FN plots of FEM and LFEM, where the elec- tron count rate i devided by V 2 tip is displayed on a log scale versus the inverse of Vtip. The count rate was taken by integrating the electron emission from the right-hand (310) type facet in Fig. 10. According to the FN the- ory, the linearity of such plots indicates that the electrons are emitted through field emission processes [4]. The FN plots of FEM data in Fig. 12 clearly show linear be- havior, and linearity can be seen also for LFEM at 10 mW and 20 mW, which are shown together with approx- imated exponential functions (black solid lines). The FN FIG. 12: (color online). Fowler-Nordheim (FN) plots for FEM and LFEM (PL = 10 mW, 20 mW, 30 mW, 40 mW, 50 mW). The vertical axis is signal i over the V 2 tip on a logarithmic scale. The signal i is the total yield of electron emission from the right hand (310) type facet of each image in Fig. 10. The horisontal axis is 1/Vtip. plots for 20 mW show non-linear behavior at low bias voltage. This indicates that photoemission processes be- come dominant at low voltage as discussed above. Sim- ilar behavior is also observed in the case of higher laser power, 30 mW, 40 mW and 50 mW. The slope of the straight sections is proportional to Φ3/2 [4]. From this fact, we can estimate the effec- tive barrier height ΦLF EM(10mW ) and ΦLF EM(20mW ) at PL = 10 mW and 20 mW, respectively, which an emitted electron feels in the case of LFEM. First, the barrier height ratios were derived from the propor- tionality constants: ΦLF EM(10mW )/ΦF EM = 0.24 and ΦLF EM(20mW )/ΦF EM = 0.2. Taking the work func- tion of (310) type facets of 4.35 eV for ΦF EM , thus ΦLF EM(10mW ) = 1.05 eV and ΦLF EM(20mW ) = 0.85 eV were obtained. The energy difference between ΦLF EM and ΦF EM should corresponds to the energy of emit- ted electrons measured from the Fermi level in LFEM. Here we obtain ΦLF EM(10mW ) - ΦF EM = 3.3 eV and ΦLF EM(20mW ) - ΦF EM = 3.5 eV, which is close to the electron energy after two-photon excitation, i.e. 3.1 eV. These values corroborate the two-photon photo-field emission processes, which is consistent with the simu- lated electron energy distributions for higher voltage in Fig. 11(a). B. Higher laser power For this discussion, we would like to point out the electron emission from the shank side of the tip in the higher laser power range. As in the previous sections, the electron emission from the tip apex is dominant in the emission pattern at low laser power because of the local field enhancement at the tip apex even though the surface area exposed to the laser beam is much larger for the shank than for the apex. At the higher laser power, however, the electron emission from the shank side be- comes noticiable because of the nonlinear dependence of the electron emission intensities on the laser power. In the column of Vtip = −100V of Fig. 10, the left-side electron-emission features highlighted by green dashed circles becomes suddenly very strong for laser powers ex- ceeding 70 mW. At 90 mW, the intensity of left-side elec- tron emission is comparable to that on the right-side. It remains even when the position of the tip in the beam waist is varied. The insets of Fig. 10 show electron emis- sion patterns at 90 mW laser power where the laser beam is displaced from the tip apex downwards by distance of 1 µm and 16 µm. In the two images, right-side emission sites disappear, but the left-side emission remains, indi- cating that it originates from the shank side of the tip. Such an electron emission should be dominated by pho- toemission over the surface barrier because DC fields on the shank side are significantly weak with respect to the tip apex. Since the laser pulses arrive at an angle of 20◦ off the horizontal line in both LFEM images, the position of the electron emission from the shank is also deviated from the horizontal line. In the inset, we also show the time-averaged field distribution around the tip when the laser beam is displaced downwards from the tip apex by a distance of 0.5 µm: the longer model tip shown in Fig. 2(e) was used. The maximum field can be observed at the side exposed to laser, which is consistent with our observations. 11 tion and the laser incidence direction relative to both azimuthal and polar orientation of the tip apex, we have demonstrated the realization of an ultrafast pulsed field- emission source with convenient control of nanometer sized emission sites. These experimental observations are quantitatively reproduced by using simulated local fields for the photo-field emission model. We discussed the emission processes further and found field-emission after two photon photo-excitation to be the dominant process in laser-induced field emission. From experimen- tal data and simulations, the dependence of the emission processes on laser power and tip voltage could be under- stood. This type of electron source is potentially useful for many applications such as time-resolved electron mi- croscopy, spatio-temporal spectroscopy, near-field imag- ing techniques, surface-enhanced Raman spectroscopy, or coherent chemical reaction control. Maybe the most in- teresting applications will arise when two laser pulses with different polarizations or paths are used for the emis- sion of two independent electron beams from two differ- ent sites on the tip, spaced only a few tens of nanometers apart, and with an adjustable time delay between the two electron pulses. Since field emission electron sources pro- duce highly coherent electron beams due to their inher- ently small source size, comparable to the finite spatial extent of electron wave packets inside the source [2, 3], we could expect two spatially and temporally coherent elec- tron beams to be available within the coherence time. This should create new opportunities for addressing fun- damental questions in quantum mechanics such as anti- correlation of electron waves in vacuum [41], or for new directions in electron holography [42]. V. CONCLUSIONS Acknowledgments We have observed laser-induced modulations of field emission intensity distributions resulting in strong asym- metries, which originate from the laser-induced local fields on the tip apex. By varying the laser polariza- We acknowledge many useful discussions with Prof. H. W. Fink, Dr. C. Escher, Dr. T. Ishikawa, and Dr. K. Kamide. This work was supported in part by the Japan Society for the Promotion of Science (JSPS), and the Swiss National Science Foundation (SNSF). [1] K. Nagaoka, T. Yamashita, S. Uchiyama, M. Yamada, H. Fujii, and C. Oshima, Nature (London). 396, 557 (1998). [2] C. Oshima, K. Mastuda, T. Kona, Y. Mogami, M. Ko- maki, Y. Murata, T. Yamashita, T. Kuzumaki, and Y. Horiike, Phys. Rev. Lett. 88, 038301 (2002). Lett. 65, 1204 (1990). [7] H. W. Fink, IBM. J. Res. Develop. 30, 460 (1986). [8] T-Y. Fu, L-C. Cheng, C. -H. Nien, and T. T. Tsong, Phys. Rev. B 64, 113401 (2001). [9] P. Hommelhoff, Y. Sortais, A. Aghajani-Talesh, and M. [3] B. Cho, T. Ichimura, R. Shimizu, and C. Oshima, Phys. A. Kasevich, Phys. Rev. Lett. 96, 077401 (2006). Rev. Lett. 92, 246103 (2004). [10] P. Hommelhoff, C. Kealhofer, and M. A. Kasevich, Phys. [4] R. Gomer, "Field Emission and Field Ionization", (Amer- Rev. Lett. 97, 247402 (2006). ican Institute of Physics, New York, 1993). [11] C. Ropers, D. R. Solli, C. P. Schulz, C. Lienau, and T. [5] G. Fursey, "Field Emission in Vacuum Microelectron- ics", (Kluwer Academic / Plenum Publishers, New York, 2003). [6] H. -W. Fink, W. Stocker, and H. Schmid, Phys. Rev. Elsaesser, Phys. Rev. Lett. 98, 043907 (2007). [12] B. Barwick, C. Corder, J. Strohaber, N. Chandler-Smith, C. Uiterwaal, and H. Batelaan, New Journal of Physics 9, 142 (2007). 12 [13] L. Wu, and L. K. Ang, Phys. Rev. B 78, 224112 (2008). [14] M. Aeschlimann et al., Nature 446, 301 (2007). [15] L. Novotny, R. X. Bian, and X. S. Xie, Phys. Rev. Lett. 79, 645 (1997). [16] L. Novotny, J. Am. Ceram. Soc. 85, 1057 (2002). [17] B. Hecht, L. Novotny, "Principles of Nano-Optics", (Cambridge University Press, Cambridge, 2005). 1999). [27] http://alphard.ethz.ch/. [28] http://MaX-1.ethz.ch [29] C. Hafner, "MaX-1: A Visual Electromagnetics Platform for PCs" (John Wiley & Sons, Chichester, 1998). [30] http://OpenMaX.ethz.ch [31] CRC Handbook of Chemistry and Physics, edited by D. [18] Y. C. Martin, H. F. Hamann, and H. K. Wickramasinghe, R. Lide (CRC Press, Boca Raton, FL, 2009), 90th ed. J. Appl. Phys. 89, 5774 (2001). [19] Y. Gao, and R. Reifenberger, Phys. Rev. B 35, 4284 [32] H. B. Michaelson, J. Appl. Phys. 48, 4729 (1977). [33] P. B. Johnson, and R. W. Christy, Phys. Rev. B 6, 4370 (1987). (1972). [20] H. Yanagisawa, C. Hafner, P. Don´a, M. Klockner, D. Leuenberger, T. Greber, M. Hengsberger, and J. Oster- walder, Phys. Rev. Lett. 103, 257603 (2009). [34] F. Yang, J. R. Sambles, and G. W. Bradberry, Phys. Rev. B 44, 5855 (1991). [35] E. L. Murphy, and R. H. Good Jr., Phys. Rev. 102, 1464 [21] A. H. Firester, M. H. Heller, and P. Sheng, Appl. Opt. (1956). 16, 1971 (1977). [22] G. R. Fowles, "Introduction to Modern Optics" (Dover Publications, New York, ed. 2, 1989). [23] M. Sato, Phys. Rev. Lett. 45, 1856 (1980). [24] T. Greber, O. Raetzo, T. J. Kreutz, P. Schwaller, W. Deichmann, E. Wetli, and J. Osterwalder, Rev. Sci. In- strum. 68 4549 (1997). [25] C. Hafner, "The Generalized Multipole Technique for Computational Electromagnetics" (Artech House Books, Boston, 1990). [26] C. Hafner, "Post-modern Electromagnetics: Using Intel- ligent MaXwell Solvers" (John Wiley & Sons, Chichester, [36] R. D. Young, Phys. Rev. 113, 110 (1959). [37] M. Lisowski et al., Appl. Phys. A 78, 165 (2004). [38] M. Merschdorf, C. Kennerknecht, and W. Pfeiffer, Phys. Rev. B 70, 193401 (2004). [39] C. E. Mendenhall, and C. F. DeVoe, Phys. Rev. 51, 346 (1937). [40] S. Hufner, Photoelectron Spectroscopy: Principles and Applications (Springer-Werlag, Berlin, 1995). [41] H. Klesel, A, Renz, and F. Hasselbach, Nature 418, 392 (2002). [42] A. Tonomura, Rev. Mod. Phys. 59, 639 (1987). 

1802.01291

1

1802

Plasmonic physics of 2D crystalline materials

Collective modes of doped two-dimensional crystalline materials, namely graphene, MoS$_2$ and phosphorene, both monolayer and bilayer structures, are explored using the density functional theory simulations together with the random phase approximation. The many-body dielectric functions of the materials are calculated using an {\it ab initio} based model involving material-realistic physical properties. Having calculated the electron energy-loss, we calculate the collective modes of each material considering the in-phase and out-of-phase modes for bilayer structures. Furthermore, owing to many band structures and intreband transitions, we also find high-energy excitations in the systems. We explain that the material-specific dielectric function considering the polarizability of the crystalline material such as MoS$_2$ are needed to obtain realistic plasmon dispersions. For each material studied here, we find different collective modes and describe their physical origins.

cond-mat.mes-hall

cond-mat

Article Plasmonic physics of 2D crystalline materials Zahra Torbatian 1 and Reza Asgari 1,2* 1 School of Nano Science, Institute for Research in Fundamental Sciences (IPM), Tehran 19395-5531, Iran ; z.torbatian@ipm.ir 2 School of Physics, Institute for Research in Fundamental Sciences (IPM), Tehran 19395-5531, Iran ; asgari@theory.ipm.ac.ir * Correspondence: asgari@ipm.ir; Tel.: +98-212-283-5061 Academic Editor: name Version February 6, 2018 submitted to MDPI Abstract: Collective modes of doped two-dimensional crystalline materials, namely graphene, MoS2 and phosphorene, both monolayer and bilayer structures, are explored using the density functional theory simulations together with the random phase approximation. The many-body dielectric functions of the materials are calculated using an ab initio based model involving material-realistic physical properties. Having calculated the electron energy-loss, we calculate the collective modes of each material considering the in-phase and out-of-phase modes for bilayer structures. Furthermore, owing to many band structures and intreband transitions, we also find high-energy excitations in the systems. We explain that the material-specific dielectric function considering the polarizability of the crystalline material such as MoS2 are needed to obtain realistic plasmon dispersions. For each material studied here, we find different collective modes and describe their physical origins. Keywords: Collective modes; density functional theory; random phase approximation; dielectric function 1. Introduction In a groundbreaking theoretical concept in the early 1950s, Bohm and Pines [1] proved that excitations in long-range Coulomb interacting systems can be decomposed into two separate sectors, namely the high-energy collective excitations, so-called longitudinal bulk plasmons, and the low-energy single-electron excitations. Well-defined plasmon oscillations exist if the momentum of carriers in the system is much smaller than the Thomas–Fermi wavevector. Moreover, the plasmon branch enters the single-electron excitation region where at this point, the collective energy of the plasmon dissipates into single electron-hole excitations. This process is known as the Landau damping [2]. The Bohm–Pines result is consistent with the classical plasma picture and was the first demonstration of the idea of the renormalization group theory in physics. Plasmonics is based on interaction processes between electromagnetic radiation and itinerant charges (electrons or holes) at metallic or doped semiconductor interfaces or in small metallic nanostructures. Although it is well known that there are two main ingredients of plasmonics, namely surface plasmon polaritons and localized surface plasmons, it is often far from trivial to appreciate the interlinked nature of many of the phenomena and applications of this field. This is compounded by the fact that throughout the 20th Century, surface plasmon polaritons have been rediscovered in a variety of different contexts. Accordingly, the science of plasmonics is dealing with generation, manipulation and detection of surface plasmon polaritons. Surface plasmon-polaritons are electromagnetic surface waves coupled to plasmon modes of the itinerant charges (electrons or holes), propagating along the interface between a dielectric and a conductor. Therefore, surface plasmon-polaritons are bound modes whose fields decay exponentially Submitted to MDPI , pages 1 – 30 www.mdpi.com/journal/notspecified Version February 6, 2018 submitted to MDPI 2 of 30 away from the interface, and therefore, plasmonics opened the possibility for manipulating light and controlling light-matter interactions at scales below the diffraction limit. Propagation of electromagnetic waves at the interfaces of plasmas with other dielectrics depends strongly on the interface geometry. A surface plasmon was predicted by Ritchie in 1957 [3]. The plasmon mode shows dispersions of the various plasmon modes at the metal-vacuum surface. The two dispersion branches of the surface plasmon, the dispersion-less branch of the longitudinal bulk plasmon (ωp) and the dispersion of the so-called multipole surface plasmon, can be found along with a linear dispersion for an electromagnetic wave, ω = ck where c is the speed of light in vacuum, parallel to its surface stemming from the coupling of a photon and a plasmon at the interface. The coupling between the light and the longitudinal bulk plasmon leads to a splitting of the (ω − k) dispersion curves for the excitations, which form a photon dispersion and the bulk plasma mode as the joint of the photon mode and surface plasmon mode. For small wave vectors, dispersion of the surface plasmon mode asymptotically approaches the light-line. For the large wavevector, in the local approximation, this surface plasmon approaches asymptotically a constant frequency, which for metals is ω = ωp/ √ 2. Let us focus on the interaction of metals with electromagnetic fields using a classical approach based on Maxwell's equations. Small sizes of metallic nanostructures on the order of a few nanometers could be qualitatively described by semiclassical mechanics. The reason for that is the high density of free carriers results in minute spacings of the electron energy levels compared to thermal excitations of energy kBT at room temperature. Moreover, the optical response of metals clearly depends on the frequency, and therefore, we have to take into account the non-locality in time and space by generalizing the linear relationships to: (cid:90) dt(cid:48)dr(cid:48) (cid:90) ε(r − r(cid:48), t − t(cid:48))E(r(cid:48), t(cid:48)) dt(cid:48)dr(cid:48) σ(r − r(cid:48), t − t(cid:48))E(r(cid:48), t(cid:48)) D(r, t) =
0 J(r, t) = (1) It should be noticed that we have implicitly assumed that all length scales are significantly larger than the lattice spacing of the material, i.e., the impulse response functions do not depend on absolute spatial and temporal coordinates, but only on their differences. A fundamental relationship between the dielectric function and the conductivity is given by: ε(k, ω) = 1 + iσ(k, ω)
0ω (2) The general form of the dielectric response ε(k, ω), in the interaction between the light and metals, can be simplified to the limit of a spatially local response through ε(k, ω) = ε(ω). The simplification is valid as long as the wavelength in the material is significantly longer than all characteristic dimensions such as the size of the unit cell or the mean free path of the electrons. In general, ε(k, ω) is a complex valued function, and the imaginary part of the dielectric function determines the amount of absorption inside the medium. Most importantly, including quantum interlayer contributions leads to increasing the imaginary part of the dielectric function, and it turns out that the effects of quantum mechanics are very vital in systems in which interlayer transitions play an important role. It is basically known that the traveling-wave solutions of Maxwell's equations in the absence of external stimuli is given by: k2E − k(k · E) = ε(k, ω) ω2 c2 E (3) Two cases, depending on the polarization direction of the electric field vector, need to be distinguished. For transverse waves, where k · E = 0, yielding the generic dispersion relation and more intriguingly for longitudinal waves, where k(k · E) − k2E = 0, it requires the particular condition Version February 6, 2018 submitted to MDPI 3 of 30 where ε(k, ω) = 0, signifying that longitudinal collective oscillations can only occur at frequencies corresponding to zeros of the dielectric function. Plasmon modes in doped graphene, which are obtained from the condition in which ε(k, ω) = 0, show many special properties, and some of them are listed here. charge carriers) [5,6]; (a) Their momentum is larger than the light momentum with the same energy [4]; (b) They can be actively tuned through chemical doping or electrical gating in real time (tuning (c) They illustrate higher levels of confinement (λSPP/λlight (cid:39) 0.025 in the normal case) [4]; (d) They have a longer lifetime and propagating lengths (τ (cid:39) 500 fs) [7]; (e) They occur in the terahertz and mid-infrared modes, which are absent in normal metals [4]; (f) They can be coupled with quasiparticles (for instance, generating plasmarons) [8]; (g) They are used in the quasi-zero dimension as emitters [9,10]; (h) They show very particular properties in hybrid structures (combining graphene with other 2D crystalline materials) [11,12]; (k) At long wavelength limits, they behave like as n1/4 [4]. (cid:113) n1/2q, which is proportional to the charge career Those properties of plasmon modes in doped graphene have been measured by using the near-field optical microscopic technique with nanometer resolution. Based on this technique, a scanning near-field optical microscope with the aperture radius much smaller than the wavelength of incident light can be used. The near-field evanescent components of light coming out from the microscope provide the required in-plane momenta. Two independent research groups performed experiments on graphene plasmonics using a similar concept [5,6] where they detected graphene surface plasmon modes. Moreover, monolayer graphene supports transverse-electric modes, which are absent in normal metals. The reason that we have this mode in doped graphene lies in the fact that the imaginary part of the conductivity is negative due to the interband transitions. Moreover, the transverse-electric modes' dispersion does not depart much from the light line, which means poor confinement, and thus, its effect is negligible. We briefly look at another physical concept, which is the energy of the electro-magnetic field inside dispersive media, since the dielectric function is a complex-valued function. Since the amount of field localization is often quantified in terms of the electromagnetic energy distribution, a careful consideration of the effects of dispersion is necessary. In metals, the dielectric function is complex and frequency-dependent owing to the dispersion. For a field consisting of monochromatic components, Landau and Lifshitz have shown that the conservation law can be well described by an effective electric energy density ueff, defined as [13]: ueff =
0 4 (ε1 + 2ωε2 γ )E2 (4) where γ = 1/τ in which the relaxation time of the free electron gas is τ and the dielectric function is given by ε = ε1 + iε2. It is worth mentioning that the real part of the dielectric function or the imaginary part of the conductivity describes the reflection of light (an elastic process); however, the imaginary part of the dielectric function describes the absorption of light (inelastic process). Basically, plasmons are observed when the real part of the dielectric function is negative (metallic behavior). In order to give some numbers, the plasmon modes occur at ultraviolet frequencies for aluminum and other materials, at ultraviolet frequencies for zero- and one-dimensional carbon structures, at visible-near infrared frequencies for noble materials (Ag, Cu, Au), at terahertz and mid-infrared frequencies for two-dimensional (2D) carbon structures (graphene) and at amplitude modulation radio frequencies for the ionosphere. With these general properties, let us now discuss in more detail the organization of the article. The scope of this paper is collective modes in pristine doped two-dimensional crystalline systems, Version February 6, 2018 submitted to MDPI 4 of 30 where the real part of their dielectric functions is basically negative, with the emphasis on fundamental physics from theoretical and experimental viewpoints. Details of the band structure properties are covered in some stand stemming from the density functional theory. Phonon scattering, the effect of impurity or strain and corrugation and optical conductivity in those materials are not covered. Detailed reviews of the plasmon modes in general are available in [14,15] and in particular for graphene in [16–20]. Our ultimate goal is to facilitate the reader's independent study of original papers on the plasmonics of crystalline two-dimensional materials using the density functional theory. In this article, we are using a recently-proposed theoretical formulation based on ab initio density functional theory (DFT) together with the random-phase approximation (RPA) to investigate the electronic excitation spectrum of doped graphene, MoS2 and phosphorene in monolayer and bilayer structures. To commence, the electronic ground-state of the periodically-repeated slab of each material is first determined, and then a Dyson-like equation is solved within the RPA to calculate the density-density response function. Having calculated the density-density response function for each structure, we therefore can calculate the macroscopic dielectric function whose imaginary part gives the optical absorption spectrum, and the collective modes are established by the zero in the real part of the macroscopic dielectric function. The theoretical dielectric function is related to the electron energy-loss function, and it provides useful information about the optical properties of the system. Here, we are just interested in the low-energy excitations for investigating the collective modes. 2. Theoretical Framework 2.1. Density Functional Theory Density functional theory (DFT) has long been the pillar of ground-state energy and density profile calculations in condensed matter science, widely used both by physicists, chemists and material researchers to study theoretically various properties of many-body systems, and in particular, it is an approach for the description of ground-state properties of metals, semiconductors and insulators. The success of density functional theory not only encompasses standard bulk materials, but also complex materials, such as proteins and nanostructures [21,22]. In 1965, Kohn and Sham proposed a practical way to implement DFT and made a significant breakthrough when they showed that the problem of many weakly-interacting electrons in an external potential can be mapped exactly to a set of non-interacting electrons in an effective external potential. The effective potential in this non-interacting particle system (the Kohn–Sham system) can be shown to be completely determined by the electron density of the interacting system and is for this reason called a density functional theory [23]. The Kohn–Sham (KS) equation looks like a simple one-particle Schrödinger equation, and it can be described by the following equation: [− ¯h2∇2 2m + Vext(r) + VHartree + Vxc(r)]Φj(r) =
jΦj(r) where Φj(r) is KS wave functions and Vext is the external potential acting on the interacting system. Furthermore, VHartree is the Hartree part of the Coulomb electron-electron interaction. The exchange-correlation potential, Vxc(r), which stems from the many-body effects, describes the effects of the Pauli principle and the Coulomb potential beyond a pure classical electrostatic interaction of the electrons. Possessing the exact exchange-correlation potential means that we solve the weakly many-body problem exactly. A common approximation is the so-called local density approximation (LDA), which locally substitutes the exchange-correlation energy density of an inhomogeneous system by that of an electron gas evaluated at the local density. Version February 6, 2018 submitted to MDPI 5 of 30 2.1.1. Computational Method There are several DFT packages that are available in the world, and each one mainly uses different basis sets. PWscf, a core component of the Quantum ESPRESSO distribution [24], performs many different kinds of self-consistent calculations of electronic structure properties such as ground-state energy and one-electron Kohn–Sham orbitals; within density functional theory, using a plane-wave basis set and pseudopotentials. The expression of the Kohn–Sham orbitals in the plane waves basis has the form: Φnk(r) = 1 Ω ∑ G Cnk(G)ei(k+G)·r where Ω represents the crystal volume, G is the reciprocal lattice vectors and k is the quasi-wavevectors of the first Brillouin zone (BZ). The coefficients Cnk(G) are obtained by solving the LDA-KS equations self-consistently. Since the core electrons of an atom are highly localized, it would be difficult to implement them using the plane waves basis sets. Actually, a very large number of plane waves is required to expand their wave functions. Furthermore, the contribution of the core electrons to bonding compared to those of the valence electrons is usually negligible. Therefore, it is practically desirable to replace the atomic potential owing to the core electrons with a pseudopotential that has the same effect on the valence electrons [25]. There are mainly two kinds of pseudopotentials, norm-conserving soft pseudopotentials [26] and Vanderbilt ultra-soft pseudopotentials [27]. The first part of our calculations includes determining the KS ground-state of pristine 2D crystalline materials and the corresponding wave functions and energies. We carry out the first-principles simulations based on the DFT simulations implemented in the QUANTUM ESPRESSO code. The calculation of the density-density response functions (χ(q, ω)) is performed employing our own code. The computation of this quantity will be discussed in the next section. In this study, the electronic structures of 2D materials are computed using the Perdew–Zunger local-density approximation [28], unless otherwise stated. Furthermore, we use this throughout the norm-conserving pseudopotentials and the plane wave basis. The energy convergence criteria for electronic and ionic iterations are set to be 10−5 eV and 10−4 eV, respectively. Careful testing of the effect of the cutoff energy on the total energy can be implemented to determine a suitable cutoff energy. The cutoff energy is required to obtain a particular convergence precision. Well-converged results are found with a kinetic energy cutoff equal to 50 Ry. The k-point grid is another calculated parameter that must be considered, and it is used to approximate integrals of some property, e.g., the electron density over the entire unit cell. Notice that the integration is performed in reciprocal space (in the first Brillouin zone) for convenience and efficiency. We thus use a Monkhorst–Pack [29] k-point grid, which is essentially a uniformly-spaced grid in the Brillouin zone. Geometry optimization and ground-state calculations are carried out on the irreducible part of the first BZ, using a Γ-centered and unshifted Monkhorst–Pack grid of 60 × 60 × 1 k-points for graphene and MoS2 and 30 × 40 × 1 k-points for phosphorene. The converged electron density is then used to calculate the KS electronic structure, i.e., the single-particle energies and orbitals on a denser k-point mesh. We perform calculations to check the convergence of the plasmon spectra with respect to the k-point sampling to obtain reliable spectra, and these are listed in Table 1. The equilibrium distance between two layers in bilayer materials is also determined by varying the interlayer distance, while keeping the in-plane lattice constant fixed at the monolayer value. In all bilayer systems, studied in this paper, the van der Waals interaction is included in order to obtain accurate results. In the 2D case, we consider a system that is infinite and periodic only in the basal (x − y) plane, but confined along the third (z) direction. A so-called supercell approach is commonly used to treat 2D systems, in which the system is modeled by repeated 2D slabs, separated by a large vacuum region Version February 6, 2018 submitted to MDPI 6 of 30 along the z direction. We use a vacuum region of at least 20 Å to avoid spurious interaction between the periodic images. We have discovered that increasing these separations would not affect the band structure of the system. The structural parameters, band gap and sampling of the reciprocal space BZ to calculate the density-density response function are listed in Table 1. Table 1. The lattice constants (a, b) and band gap (in eV) of different 2D crystalline materials. Sampling of the reciprocal space Brillouin zone (BZ) is done by a Monkhorst–Pack (MP) grid for the considered 2D materials. 2D Structures Lattice a (Å) graphene MoS2 phosphorene Hexagonal Hexagonal Rectangular 2.46 3.15 4.62 b (Å) − − 3.30 Gap (eV) MP Monolayer Bilayer Monolayer 101 × 101 × 1 101 × 101 × 1 60 × 80 × 1 0.00 1.80 0.98 0.00 1.05 0.63 Bilayer 201 × 201 × 1 151 × 151 × 1 120 × 160 × 1 2.2. Density-Density Response Function A central quantity in the theoretical formulation of the many-body effects in electronic systems is the non-interacting dynamical polarizability function [30] χ(0)(q, ω) for a finite chemical potential, µ. Here, we would like to emphasize that we have calculated the non-interacting polarization function for doped graphene, MoS2 and phosphorene, both monolayer and bilayer structures based on DFT simulations. The expression of the non-interacting density-density response function of a three-dimensional periodic electrons in the reciprocal space is: χ0 GG(cid:48) (q, ω) = 2 Ω ∑ k,ν,ν(cid:48) fν(k) − fν(cid:48) (k + q) ¯hω + εn(k) − εm(k + q) + iη kq νν(cid:48) (G)ρ ρ kq νν(cid:48) (G(cid:48))∗ (5) which is obtained from the Adler–Wiser periodic system [31]. Here, εn(k) and εm(k + q) denote the empty and filled bands and fn(k) = θ(EF − εn(k)) is the Fermi-Dirac distribution of the charge carrier with energy εn(k) at temperature zero. Furthermore, G and G(cid:48) are the three-dimensional (3D) reciprocal lattice vectors, and ω is the frequency. In this theory, the linear combination of plane-waves is used to determine the KS single-particle orbitals of the DFT. The KS wave functions are normalized to unity in the crystal volume Ω. The sum is over a special set of k vectors and energy bands (ν and ν(cid:48)). In Table 1, the sampling k point in the BZ is indicated for different 2D materials in order to fully converge the results. The factor of two accounts for the spin degeneracy, and η is a small (positive) lifetime broadening parameter. The matrix elements of Equation (5) have the form: νke−i(q+G)·rΦ kq νν(cid:48) (G) =< Φ ρ (6) where q is the momentum transfer vector parallel to the x − y plane. Wave functions Φnk(r) are the KS electron wave functions and when expanded in the plane-wave basis have the form: ν(cid:48)k+q >Ω Φnk(r) = 1 Ω ∑ G Cnk(G)ei(k+G)·r (7) where the coefficients Cnk(G) are obtained by solving the LDA-KS equations self-consistently. The exact interacting density-response function can be obtained in the framework of the DFT, as follows [32]: χGG(cid:48) = χ0 GG(cid:48) + ∑ G1G2 χ0 GG1 νG1G2 χG2G(cid:48) (8) Version February 6, 2018 submitted to MDPI 7 of 30 where νGG(cid:48) represents the Fourier coefficients of an effective electron-electron interaction. In the GG(cid:48) = 4πe2δGG(cid:48)/
q + G2 where
is the electron liquid, the bare Coulomb interaction is given by ν0 average dielectric constant of the environment. In all our numerical results, we consider
= 1. The RPA procedure, an approximation valid in the high-density limit, takes into account electron interaction only to the extent required to produce the screening field, and thus, the response to the screened field is measured by χ0. The RPA follows from a microscopic approach whose main assumption is that the electrons respond not to the bare Coulomb potential, but to an effective potential resulting from the dynamical rearrangement of charges in response to the Coulomb potential. The long-range behavior of the Coulomb interaction allows non-negligible interactions between repeated planar arrays even at a large distance. This unphysical phenomenon can be removed by replacing νGG(cid:48) with the truncated Fourier integral over the cutoff plane axis (z) [33,34], and thus, we have: ν0 GG(cid:48) = 2πe2δgg(cid:48) q + g (cid:90) L/2 −L/2 (cid:90) L/2 −L/2 dz dz(cid:48)ei(Gzz−G(cid:48) zz(cid:48))−q+gz+z(cid:48) (9) where the g and Gz denote the in-plane and out-plane components of G, and we assume that q is never zero owing to a uniform background of positive charge. In the framework of the linear response theory, the inelastic cross-section corresponding to a process where the external perturbation creates an excitation of energy ¯hω and wavevector q + G is related to the diagonal elements of the dielectric function in the level of the RPA: εGG(cid:48) = δGG(cid:48) − ∑ G1 ν0 GG1 χ0 G1G(cid:48) (10) and the plasmon modes are established by the zero in the real part of the macroscopic dielectric function given by: ε(q, ω) = 1 (ε−1)GG(cid:48) G=G(cid:48)=0 (11) as long as there is no damping. The electron-energy loss (EEL) is proportional to the imaginary part of the inverse dielectric function, which is given by: EEEL(q, ω) = −(cid:61)m[1/ε(q, ω)] (12) It is worth mentioning that the nonlocal field effects are included in EEL through the off-diagonal elements of the general χGG(cid:48) [35] function. In addition, one may use the non-local dynamical conductivity to describe the electronic processes and light-matter interactions. This can be simply achieved by writing the longitudinal conductivity in terms of the density-density response function through the relation: σ(q, ω) = ie2ω q2 χ(q, ω) (13) We note that Equations (11)–(13) are the fundamental physics describing the optoelectronic interactions in 2D crystalline materials. 3. Result and Discussion 3.1. Monolayer Graphene Graphene is a 2D layer of carbon atoms arranged in a honeycomb lattice with Dirac cones, i.e., massless Dirac fermion at K point, where the π and π∗ bands show a linear energy dispersion [36–38]. Version February 6, 2018 submitted to MDPI 8 of 30 In Figure 1, we illustrate the atomic structure and also the electronic energy band structure of graphene based on our DFT simulations. Research on collective electronic excitations (plasmons) in graphene has attracted enormous interest both from theoretical and experimental viewpoints [4,7,39–46]. Three kinds of the collective excitations of electrons can be considered in graphene that unroll on a wide range of energy. The first kind is attributed to finite electron doping, originating from the intraband transitions of Dirac fermions in the vicinity of the K and K(cid:48)points of the BZ at low energies (0–2 eV), and it can be regarded as intraband plasmon modes. The second kind of plasmons in the monolayer graphene is the intrinsic π plasmons, arising from the collective excitations of electrons from the π to π∗ bands at energies of about 4–15 eV. At higher energies, σ bands start to contribute, and the mixture of the π → π∗ and σ transitions leads to another kind of plasmon excitation of graphene, usually denoted as π + σ plasmons. It is worthwhile mentioning here that the calculation of the π + σ plasmon would require including high-lying bands; hence, the ab initio approach would be the more appropriate way to capture them [47]. Figure 1. (a,b) Top and side view of monolayer graphene. (c) The band structure of graphene along the high symmetry Γ − M − K − Γ directions and the associated Brillouin zone. The zero in the energy axis is set at the Fermi level as shown by the solid line. The plasmon spectrum of a pristine single-layer graphene was investigated in [48] using ab initio calculations. They observed the π and π + σ plasmon modes in freestanding single sheets at 4.7 and 14.6 eV, which were red shifted in comparison to the corresponding modes in the bulk graphite. Our attention is now focused on the intraband and π plasmons of the spectrum (below 10 eV), and it appears to be owed to the intraband and interband transitions, respectively, as mentioned above. In order to better perceive the electronic excitations of monolayer graphene, we calculate the non-interacting density-density response function, χ0(q, ω). The non-interacting density-density response function can be decomposed into two parts where the first part can be considered by only including transitions within a band and the second part contains all transitions between separate bands. An example of the non-interacting density-density response function is shown in Figure 2, for q = 0.076 Å−1 and EF = 0.8 eV (the Fermi level is shifted upward by 0.8 eV above the Dirac KMΓ-10-50510h_ω (eV)ΓΓMK(a)(b)(c)Version February 6, 2018 submitted to MDPI 9 of 30 cone, and this corresponds to an electron doping level of 7.4 × 1013 cm−2). The limq→0 χ0(q, ω = 0) is finite and equal to the density of states at the Fermi energy, N(0) a measure of the number of excited states. This figure shows that intraband transitions contribute more at low-energy, while the interband transitions dominate at high-energy. In Figure 3, the EEL functions for various momentum transfers and the plasmon spectrum of electron doped monolayer graphene (n = 7.4 × 1013 cm−2) are computed using the linear response DFT-RPA approach. The effect of doping on the band structures of 2D materials is ignorable [49]. In the long-wavelength limit, plasmons can be viewed as a center-of-mass oscillation of the electron gas as a whole. The physical origin of plasmons is described as follows. When electrons in free space move to screen a charge inhomogeneity, they tend to overshoot the mark. They are then pulled back toward the charge disturbance and overshoot again, setting up a weakly-damped oscillation. The restoring force responsible for the oscillation is the average self-consistent field created by all the electrons. √ q As expected, the dispersion behavior of the plasmon mode at the low-energies shows a standard dispersion predicted in the 2D electron gas system. To be more precise, the plasmon mode at the long wavelength limit in graphene is given by: (cid:34) (cid:114) 2D0
ωp(q) = q 1 + 12 − N2 ee − 8(1 − κ0/κ) f α2 16 q qTF + ... (cid:35) (14) where D0 = vFkFe2/¯h is the Drude weight in graphene. For ordinary parabolic-band fermions with mass mb, the Drude weight is given by D0 = πne2/mb. Graphene's fine-structure constant is αee = e2/¯hvF
; NF = 4 is the flavor number in graphene; the Thomas–Fermi screening wave number is defined by qTF = NFαeekF; and the electron compressibility of interacting and non-interacting graphene is κ and κ0, respectively [7] . The second term in the square brackets refers to the quantum non-local effects. Notice that in the classical picture, the long-wavelength plasmon mode behaves like ωp(q) =
m q, which is totally different from the one we have in graphene. (cid:113) 2πne2 Figure 2. The real (solid line) and imaginary (dashed line) parts of the non-interacting density response function of monolayer graphene in units of the Fermi-level density of states as a function of ¯hω for q = 0.076 Å−1. Our numerical results show that the general behavior of the plasmonic dispersion agrees with the results from [47,49]. The dispersion relation of the π plasmon is presented, and it has a quasi-linear behavior, in good agreement with the earlier theoretical study [47]. In this figure, the electron-hole continuum is indicated with a black line, and it can be obtained at any specific momentum transfer q by the difference between system energy at kF and kF + q. We would like to compare the plasmon mode obtained here with that calculated within a semiclassical approach. We suppose that the graphene sheet, along the x and y directions, is 0246810h_ω (eV)-1012χ0(q,ω)024681000.10.20.30.4InterbandIntrabandN(0)Version February 6, 2018 submitted to MDPI 10 of 30 located between two semi-infinite dielectric media with the same dielectric constant,
, and consider a solution of Maxwell's equations for a transverse magnetic wave. By using the proper boundary conditions, we arrive at [19]:
(cid:112)q2 − ω2
/c2 2 = −i σ(ω) ω
0 which describes the plasmon mode, ω(q), of graphene with conductivity σ(ω). It should be noted that this expression implements every 2D crystalline material with its σ(ω). This equation has solutions, if the imaginary part of the conductivity is positive and its real part vanishes. The conductivity of graphene from the terahertz to mid-infrared regime is dominated by a Drude term and given by: (15) (16) (17) σ(ω) = e2 π¯h iEF ¯hω It turns out that at the long wavelength limit, the plasmon mode is given by: (cid:115) ¯hωp = 2e2
0 EFq which is the same expression that we have in the quantum many-body framework. We also include the semiclassical plasmon mode in Figure 3b, and our numerical results show that they are the same at long wavelength regimes; however, at large momenta, the semiclassical plasmon mode deviates from that calculated by the many-body approach, especially at lower electron density. Version February 6, 2018 submitted to MDPI 11 of 30 Figure 3. (a) The electron-energy loss (EEL) function for different momentum transfers and (b) plasmon dispersions of electron-doped graphene along the Γ − M direction. In this case, the Fermi level is shifted upward by 0.8 eV above the Dirac cone, and this corresponds to a doping level of 7.4 × 1013 cm−2. The plasmon mode based on the semiclassical approach given by Equation (15) is plotted. The boundary of the electron-hole continuum is indicated with a black line. 3.2. Bilayer Graphene Bilayer graphene displaces the simplest possible system where graphene sheets are brought together to create a new nanostructure whose physical properties show remarkable similarities and differences as compared to monolayer graphene. Bilayer graphene, like single-layer graphene, is a zero-gap semimetal that consists of two coupled monolayers of carbon atoms stacked as in natural graphite (AB stacking or Bernal-stacked form where half of the atoms lie directly over the center of a hexagon in the lower graphene sheet, and half of the atoms lie over an atom) yielding a unit cell of four atoms (Figure 4) [50–52]. 0246810h_ω(eV)00.10.20.30.4q (Å-1)012h_ω (eV)DFT45670.0290.0590.0880.1180.1470.1770.2060.235q (Å-1)IntrabandInterbandEq.15π plasmon(a)(b)Version February 6, 2018 submitted to MDPI 12 of 30 Figure 4. (a,b) Top and side view of the atomic structure of bilayer graphene (BLG). (c) The band structure of BLG along the high symmetry Γ − K − M − Γ directions in reciprocal space. The optimized interplane distance (d) is calculated to be 3.45 Å. We calculate the band structure of bilayer graphene through DFT simulations, and this is illustrated in Figure 4. Bilayer graphene has four electronic bands (a pair of conduction bands and a pair of valence bands) with pz symmetry, namely: π1, π2, π∗ 2. The dispersion of these bands is parabolic near the K point, and their occupation depends on the doping values. In undoped bilayer graphene, the π1 and π2 bands are fully occupied, while the π∗ 2 bands are fully unoccupied. 2 bands along K − M leads to an anisotropic dispersion at Furthermore, an overlap of the π∗ energies from 1–1.5 eV. 1 and π∗ 1 and π∗ 1 and π∗ Research on the collective electronic excitations (plasmons) in freestanding multilayer graphene [53–55] shows that , similar to single-layer graphene and graphite, the spectra of bilayer graphene feature two characteristic high-energy plasmon peaks, i.e., the π plasmon below 10 eV and the π + σ plasmon above 15 eV, but these plasmons are red shifted with respect to graphite. In the low-energy range, a conventional 2D plasmon was predicted to exist originating from the intraband transitions. It is noteworthy to mention that, as reported in [56], when the two layers are near each other (separated by a distance d in the z direction with the 2D layers in the x − y plane), the 2D plasmons are coupled by the interlayer Coulomb interaction leading to the formation of in-phase and out-of-phase interlayer density fluctuation modes: an in-phase optical plasmon mode, where the densities in the two layers fluctuate in phase with the usual 2D plasma dispersion (ω ∼ √ q) and an out-of-phase acoustic plasmon mode, where the densities in the two layers fluctuate out-of-phase with a linear wavelength dispersion (ω ∼ q). In a recent study, the plasmon modes of undoped (intrinsic) and doped (extrinsic) bilayer graphene were calculated and analyzed carefully based on density-functional theory in an energy range from a few meV to ∼30 eV, along the inequivalent Γ − K and Γ − M directions [33]. In that paper, they found an acoustic plasmon mode for momenta along the Γ − M direction for a positive shift in the Fermi level of 1 eV. Although they have claimed this acoustic plasmon is an undamped collective excitation, an overdamped acoustic plasmon was predicted by Das Sarma et al. [56]. To investigate the origin of different plasmon modes in bilayer graphene, we calculate the interband and intraband parts of the non-interacting response function of bilayer graphene, and we present our results in Figure 5 for only q = 0.084 Å−1. In this case, the Fermi energy is shifted upward by 0.7 eV towards the bottom of the conduction bands in bilayer graphene, corresponding to n = 9.5 × 1013 cm−2. Our numerical results show that the non-interacting response function of -15-10-50510h_ω (eV)ΓΓKM(a)(b)(c)dVersion February 6, 2018 submitted to MDPI 13 of 30 bilayer graphene is similar to its monolayer, but there is an extra contribution of the interband part at low-energies that is obviously absent in monolayer graphene. Figure 5. The inter- and intra-band terms of the non-interacting response function (χ0(q, ω)) of doped bilayer graphene in units of the Fermi-level density of states as a function of ¯hω for q = 0.084 Å−1 along the Γ − M direction. The real and imaginary parts of this function are indicated by solid and dashed lines, respectively. In Figure 6, we illustrate the loss spectra for various amounts of q and also electronic excitations for bilayer graphene along the Γ − M direction for n = 9.5 × 1013 cm−2. The loss spectra involves both intraband and interband excitations at energies below 10 eV, but we neglect to show π plasmon in Figure 6b. (a) The loss spectra for different values of q and (b) the plasmon dispersion of doped Figure 6. bilayer graphene with EF = 0.7 eV corresponding to n = 9.5 × 1013 cm−2 along the Γ − M direction. The diamond symbols refer to the acoustic plasmon mode. 0246810h_ω (eV)-1012χ0(q,ω)0246810-0.200.20.40.6InterbandIntrabandN(0)0246810h_ω(eV)00.10.20.3q (Å-1)00.511.52h_ω (eV)IntrabandInterband0.0390.0590.0980.1360.1560.1950.2340.273q (Å-1)(a)(b)Version February 6, 2018 submitted to MDPI 14 of 30 It is clear that the plasmon dispersion of bilayer graphene follows a general q dispersion at low energies, which is also seen in monolayer graphene. These results are in reasonable agreement with earlier work reported by Pisarra and coauthors [33]. Our ab initio calculations show the acoustic plasmon mode, which is highly damped in bilayer graphene structures. √ Most importantly, by breaking the inversion symmetric of the two layers, a non-zero band gap can be induced. The potential of a continuously tunable band gap through a gate voltage applied perpendicularly to the sample is very interesting [51,57,58]. The realization of a widely tunable electronic band gap in electrically-gated bilayer graphene has been experimentally demonstrated [59]. They showed a gate-controlled, continuously tunable band gap of up to 250 meV by using dual-gate bilayer graphene field-effect transistor infrared micro-spectroscopy. Moreover, this electrostatic band gap control suggests nanoelectronic and nanophotonic device applications based on graphene. Notice that the band gap can be observed in photoemission, magneto transport, infrared spectroscopy and scanning tunneling spectroscopy. The low-energy effective model Hamiltonian for a gated bilayer graphene can be written as: Heff = − 1 2m ((cid:126)σ · (cid:126)p)σx((cid:126)σ · (cid:126)p) + ∆σz (18) where (cid:126)σ are Pauli matrices and ∆ is the gated energy. In bilayer graphene, changing the applied gate voltages will turn into controlling the electron density, n, and the interlayer asymmetry different potential energies, ∆. In other words, the asymptotic energy ∆V is related to layer density, and the layer densities depend on ∆, ultimately [60]. Here, we ignore this effect and assume that the band gap is independent of the electron density. We have examined first-principle calculations to investigate the plasmon modes of AB stacked bilayer graphene in the presence of a perpendicular applied electric field. Figure 7 shows the gate dependence of the optical plasmon modes of bilayer graphene for ∆ = 1.1 and 2.5 eV. Clearly, the gap reduces the plasmon mode and softens the collective excitation modes, especially at the long wavelength limit. In this regime, the plasmon mode behaves slightly different from that obtained for nq a system with ∆ = 0 given by ωp(q) = m [61]. Moreover, the interband contribution decreases with the gated energy ∆; however, the intraband contribution increases with the bias owing to the fact that the energy dispersion leads to an enhancement of the density of states near the Fermi energy. (cid:113) e2 4
0 Figure 7. The gate voltage dependence of the plasmon mode in bilayer graphene. The values of the gate voltages are 1.1 V/nm and 2.5 V/nm related to band gaps of about 0.1 and 0.2 eV, respectively. 3.3. Monolayer MoS2 The lack of a natural band gap makes graphene unsuitable for developing optoelectronic and photovoltaic devices [62]. A particularly interesting class of 2D materials is the transition metal 00.10.2q (Å-1)00.511.5h_ω (eV)0.0 V/nm1.1 V/nm2.5 V/nmVersion February 6, 2018 submitted to MDPI 15 of 30 dichalcogenides (TMDC) whose electronic properties range from semiconducting to metallic and even superconducting. Among them, molybdenum disulfide (MoS2) with a natural band gap is gaining increasing interest [63]. MoS2 has a hexagonal structure that consists of two planes of hexagonally-arranged S atoms bonded through covalent bonds to central layer Mo atoms. In agreement with previous reports [64–67], the MoS2 monolayer is a direct band gap semiconductor with a maximum of the valence and the minimum of the conduction band located at the Kpoint of the Brillouin zone (Figure 8). The bands around the energy gap are relatively flat, which are as expected from the d character of the electron states at these energies. The states around the Fermi energy are mainly due to the d orbitals of Mo, while strong hybridization between d orbitals of Mo and p orbitals of S atoms below the Fermi energy has been observed [30]. Figure 8. (a,b) Top and side view of the atomic structure of molybdenum disulfide (MoS2). (c) The band structure of MoS2 along the high symmetry Γ − M − K − Γ directions and the associated Brillouin zone. The Fermi level is set at 0 eV. The blue balls are molybdenum atoms, and the yellow ones are sulfur atoms. The GW approximation, is an approximation made in order to calculate the self-energy of a many-body system of electrons, has been shown to provide very reliable descriptions of the electronic and dielectric properties for many semiconductors and insulators [68]. The recent quasiparticle self-consistent GW calculations have reported that MoS2 is a direct gap semiconductor at both the LDA and GW levels, and the GW gap is 2.78 eV at both the K and K(cid:48)points [69]. In the following, we will explore the dispersion behavior of plasmon modes of monolayer MoS2 by using our DFT-RPA code. The intraband plasmons in metallic single-layer transition metal dichalcogenides (TMDCs) have been studied using density functional theory in the random phase approximation [70]. They have q behavior of found that at very small momentum transfer, the plasmon energy follows the classical free electrons in 2D. For larger momentum transfer, the plasmon energy is significantly red shifted due to screening by interband transitions. The non-interacting response function of MoS2 with n = 5.6 × 1013 cm−2 and for q = 0.069 Å−1 is plotted in Figure 9 and shows that the intraband term of χ0(q, ω) is dominated at energies below 1.0 eV, and the interband term is inconsiderable. √ KMΓ-4-202h_ω (eV)ΓMKΓ(a)(b)(c)Version February 6, 2018 submitted to MDPI 16 of 30 Figure 9. The interband and intraband of non-interacting response function (χ0(q, ω)) of doped MoS2 in units of the Fermi-level density of states as a function of ¯hω for q = 0.069 Å−1. In Figure 10a, the EEL function of MoS2 for different values of q with electron doping of n = 5.6 × 1013 cm−2 is depicted and the plasmon dispersion of system plotted in Figure 10b (black dots). In the following, we compare the resulting plasmon dispersion calculated by our ab initio model to that obtained by a model Hamiltonian. (a) The EEL functions of MoS2 for different q and n∼5.6 × 1013 cm−2 for
= 1. Figure 10. (b) Our optical plasmon mode of MoS2 (dots), obtained from the peaks in the EEL functions in the (a), in comparison with a model Hamiltonian by Scholz et al. [71] (blue line). The red line is obtained using the modified bare potential given by Equation (22) by a model Hamiltonian. The black line shows our ab initio calculation with the modified bare potential (see the text for more details). We use Eq.11 in the paper by Scholz, et al. [71], where the long wavelength behavior of the plasmon frequency of monolayer MoS2 can be obtained from: (cid:115) (cid:115) ω0 q = e2q 2π
0
r (2EF − ∆)[EF(∆ + 2EF) − λ2] 4E2 F − λ2 00.20.40.60.81h_ω (eV)00.51χ0(q,ω)00.20.40.60.8100.05InterbandIntrabandN(0)00.20.40.60.81h_ω (eV)00.050.1q (Å-1)00.40.81.2h_ ω (eV)DFT.Eq.90.1380.0230.0460.0690.0920.1150.1610.1840.207q (Å-1)(a)(b)ScholzScholz (corrected)DFT.Eq.22Version February 6, 2018 submitted to MDPI 17 of 30 Here, ∆ is the energy gap and λ = 80 meV. In this case, we consider
r = 1 because it can be compared to our ab initio results. It is obvious that the plasmon mode obtained within DFT-RPA approach for monolayer MoS2 differs significantly from with that calculated using the low-energy model Hamiltonian. The similar comparison was performed for hole doping in the case of MoS2 in [72], and they observed strongly reduced plasmon energies and concluded that neglecting the material-specific dielectric function
αβ(q) within the minimal three-band model is a severe approximation leading to unrealistic plasmonic properties. This discrepancy could be understood by looking at both the bare potential and the Lindhard function of the MoS2. In order to perceive the aforementioned discrepancy, we look at the structure of the MoS2 again, which basically consists of three atomic layers. In ordinary 3D materials, the effect of lattice screening is simply a re-scaling of the interaction strength by a dielectric constant. In quasi-2D materials, however, the interaction is modified by the polarizability of the crystalline material originated from the induced charge, nind = −∇ · P, where P is the polarizability [73]. Therefore, the Poisson equation for the potential of the external point charge takes the form: ∇2V(r, a) = −4πeδ(z) − 4πa∇2 (19) where r = (ρ, z) and ρ = (x, y), and we use the fact that P = −a∇ρV(ρ, z = 0, a)δ(z). In order to solve this equation, it is convenient to take the Fourier transformation of the equation to obtain V(q, a). Afterwards, by taking the inverse Fourier transformation of V(q, a), we can find the effective potential in real space, which is no longer e2/r, and it yields: ρV(ρ, z = 0, a)δ(z) V(r, a) = e2π[−Y0(r/a) + H0(r/a)] 2a where the Bessel function of the second kind is defined by: Yn(x) = Jn(x) cos(nx) − J−n(x) sin(nx) (20) (21) where Jn(x) is the Bessel function of the first kind. For n an integer, this formula should be understood as a limit. The Struve function, Hn(x), solves the inhomogeneous Bessel equation. It would be worthwhile to mention that this potential was proposed in the Keldysh model [74], which is based on a slab of constant dielectric value, the potential between two charges in a slab of thickness d. In this model, a = d
(cid:107)/2 where
(cid:107) is the in-plane dielectric constant of the bulk material. Let us look at the Fourier transformation of the modified bare potential, which is given by: V(q, a) = 2πe2
(q + aq2) = 2πe2
q 1 εlattice(q) (22) where
is the average dielectric constant of the environment, a might depend on the thickness of the 2D crystalline material and εlattice(q) plays the role as a lattice local field factor. In the case of MoS2, Qiu et al. [69] fitted the Keldysh model to their ab initio effective dielectric function at small q and obtained an effective screening length of a = 35 Åor the slab thickness of d = 6 Å. Apparently, the exact value of the thickness of monolayer MoS2 is just d (cid:39) 6.3 Å. It is interesting to note that when we consider the corrected coefficient of (1 + aq) with a = 35 Å in the bare Coulomb potential, we find better agreement between plasmon dispersion of the low-energy model Hamiltonian of MoS2 with that obtained using the DFT-RPA approach, although it is not yet coincident with our result. This difference can be related to the non-interacting density-density response function in two methods. Notice again that the χ0 in the DFT scheme consists of many occupied bands together with the structure of the many orbitals, and as has been discussed [75,76], those play important roles in the physical properties of MoS2. This means that the χ0 calculated within DFT-RPA contains some effects of the exchange-correlation of the system. These results also show that for quasi-two-dimensional materials, the Coulomb potential Version February 6, 2018 submitted to MDPI 18 of 30 should be modified properly. At the end of this section, it is worth noting that when the potential in Equation (22) is used to calculate plasmon modes in MoS2 using DFT simulations, we obtain very similar results, which are shown with a black line in Figure 10b. 3.4. Bilayer MoS2 The AA' stacking, in which the layers are exactly aligned, with Mo over S is the stablest stacking for the MoS2 bilayer, and it is shown in Figure 11a,b. From the band structure plotted in Figure 11c, based on our DFT simulations, along the high symmetry Γ − M − K − Γ directions, it can be seen that bilayer MoS2 has an indirect band gap in contrast to the direct band gap of the corresponding monolayer. In fact, the conduction band minima are located between the Γ and K high symmetry points (Q point), while the valence band maxima are located at the Γ point of the BZ, revealing the indirect band gap, and this is in good agreement with previous calculations [77–79]. Figure 11. (a,b) Top and side view of bilayer molybdenum disulfide (BL.MoS2). (c) The band structure of BL.MoS2 along the high symmetry Γ − K − M − Γ directions. The Fermi level is set at 0 eV, and interlayer separation d is 5.89 Å. More analysis of the band structure of MoS2 can show why this material experiences an indirect to direct band gap transition when its bulk or multilayers are replaced by a monolayer. In fact, the states originating from mixing of Mo(dz2) orbitals and the S(pz) orbitals at Γ are fairly delocalized and have an antibonding nature. With increasing the separation between consecutive MoS2 layers, the layer-layer interaction decreases and lowers the energy of the antibonding states, and consequently, the valence band maximum shifts downward. The states at K, which are of the dxy − dx2−y2 character, are mostly unaffected by interlayer spacing. Thus, in the limit of widely-separated planes, i.e., monolayer MoS2, the material becomes a direct gap semiconductor [80]. The non-interacting response function of bilayer MoS2 for q = 0.084 Å−1 with the charge carrier concentration of n = 6.9 × 1013 cm−2 shown in Figure 12. We can see that in bilayer MoS2, both interband and intraband contributions of non-interacting response function are considerable. The intraband contribution of this quantity is quite strong and refers to the main collective mode, although its interband term is weak, and this can lead to new modes in the system; thus, in the following, we will explain it further. In Figure 13, we display the EEL function and plasmon spectrum of bilayer MoS2 for the electron doping with n = 6.9× 1013 cm−2 and for energies below 1 eV and momenta q along the high-symmetry Γ − K direction. We observe that plasmonic features in electron-doped bilayer MoS2 are mainly characterized by a square root mode in small q. In the paper by Andersen, plasmons in bilayer NbS2 were studied [70], and they obtained very similar plasmon dispersions. -4-2024h_ω (eV)ΓMKΓd(a)(b)(c)Version February 6, 2018 submitted to MDPI 19 of 30 Figure 12. The non-interacting response function in units of the Fermi-level density of states as a function of ¯hω of bilayer MoS2 for q = 0.084 Å−1. Figure 13. (a) The loss spectra of bilayer MoS2 for various amounts of q along the Γ − K direction with EF = 0.05 eV corresponding to n = 6.9 × 1013 cm−2. (b) The same as (a) for two different energy regions. (c) Three different plasmon modes of bilayer MoS2; the black dots refer to the optical plasmon mode. The results of the acoustic and high energy modes are shown by red diamond and triangle symbols, respectively. The black line shows the boundary of the electron-hole continuum. 00.20.40.60.81h_ω (eV)-0.500.51χ0(q,ω)00.20.40.600.020.04InterbandIntrabandN(0)0.10.2h_ ω (eV)q (Å-1)0.40.50.60.700.10.20.30.40.50.60.7h_ω (eV)q(Å-1)00.050.10.150.2q (Å-1)00.20.40.6h_ω (eV)(c)0.0790.1190.1190.1320.1450.1320.1720.145(b)0.1720.0790.1850.1060.1590.1060.1850.159(a)Version February 6, 2018 submitted to MDPI 20 of 30 As previously mentioned, in the two-layer systems, we can find two plasmon modes, which can be regarded as symmetric and antisymmetric combinations of two unperturbed monolayer plasmons. One of them has a nearly linear dispersion, referred to as the acoustic plasmon mode. In this case, we find that the low-energy dynamical dielectric function in bilayer MoS2 is controlled by both interband and intraband contributions, leading to an extra collective mode. Also, there are a damped high energy mode and a highly damped acoustic mode, which originates from the interband transitions. More details of these calculations can be found in [81]. In Figure 14, we display the damping parameter of plasmon modes in bilayer MoS2 for the electron doping value n = 6.9 × 1013 cm−2 for both the acoustic and higher plasmon modes. Since the (cid:61)mε(q, ω) at the position of the acoustic and high plasmon modes are finite, it turns out that those collective modes are damped, and the damping parameter, γ, as a function of the momentum would be small. In this case, we should verify the condition where ε(q, ωp − iγ) = 0 is satisfied. This equation leads to two separate equations in which the γ(q) is given by γ(q) = (cid:61)mχ0(q, ω)/∂(cid:60)eχ0(q, ω)/∂ω at ω = ωp and the collective mode is obtained by 1 − νq(cid:60)eχ0(q, ω) − γνq∂(cid:61)mχ0(q, ω)/∂ω = 0 again at ω = ωp. Note that a condition required to define those plasmon modes are γ(q)/ωp(q) << 1. Results shown in Figure 14 indicate that the acoustic mode is highly damped, and the high plasmon mode is slightly damped. Figure 14. The damping parameter of plasmon modes in bilayer MoS2 as a function of the momentum for n = 6.9 × 1013 cm−2. Note that in order to have a well-defined plasmon mode, γ(q)/ωp(q) << 1. 3.5. Monolayer Phosphorene Although two-dimensional materials such as monolayer graphene and monolayer transition metal dichalcogenide, the most common component MoS2, have attracted intensive research interest owing to their fascinating electronic, mechanical, optical and thermal properties, the lack of the band gap in graphene and low carrier mobility in MoS2 limits its wide applications in electronic devices and nanophotonics [82,83]. 00.050.10.150.2q (Å-1)00.20.40.60.8 γ (q)00.050.10.150.2q (Å-1)00.020.040.060.080.1 γ (q)acoustic modehigh energy modeVersion February 6, 2018 submitted to MDPI 21 of 30 Recently, isolated two-dimensional black phosphorus (BP), known as phosphorene, with a nearly direct band gap and high carrier mobility has received enormous interest owing to its extraordinary electronic and optical properties in engineering application [84,85]. Our DFT calculations for mono- and bi-layer phosphorene were carried out using the Perdew–Burke–Ernzerhof exchange-correlation functional [86] coupled with the DFT-vdw method. Figure 15a,b shows that phosphorene is a puckered honeycomb structure with each phosphorus atom covalently bonded with three neighboring atoms within a rectangular unit cell, and the two lattice constants are ax = 4.62 Å and ay = 3.30 Å along the armchair and zigzag directions, respectively [87]. As a result of the puckered structure, each single-layer phosphorene contains two atomic layers where the distances between the two nearest atoms (2.22 Å) and the distance between the top and bottom atoms (2.24 Å) are slightly different. It should be mentioned that monolayer phosphorene is a nearly direct band gap semiconductor because the exact top of the valence band is slightly away from the Γ point. However, they are sufficiently close to be considered as a direct band gap as previous first-principles calculations mention [88]. Our DFT calculations predict that the band gap of monolayer phosphorene is 0.98 eV, and the self-energy correction enlarges the band gap to 2.0 eV, which is ideal for potentially broad electronic applications [89]. The optical gap of phosphorene was reported to be 1.6 eV by using the GW and Bethe-Salpeter equation method, and it is much lower than the electronic band gap of 2.15 eV, which shows significant excitonic effects in phosphorene [90]. The electronic band structure of monolayer phosphorene is plotted in Figure 15c, and it shows the high anisotropic band structure with very different effective masses along the armchair and zigzag directions for both electrons and holes. This anisotropic electronic structure is useful for thermoelectric materials, because for the direction with a smaller effective mass, the carrier mobility and thus the electrical conductivity can be high, while the larger effective mass along the other direction contributes to an overall large density of states that improves the Seebeck coefficient [84,91]. Figure 15. (a,b) Top and side view of the atomic structure of monolayer phosphorene. (c) The band structure of monolayer phosphorene along the high symmetry X − Γ − Y directions and the associated Brillouin zone. The Fermi level is set at 0 eV. In Figure 16, the plasmon spectrum of monolayer phosphorene with EF = 0.07 eV along the armchair direction based on our DFT simulations is compared to the result recently reported in [92], and they show an acceptable agreement. ΓXYyx (armchair)(zigzag)x z -4-3-2-1012h_ω (eV)XΓY(a)(b)(c)Version February 6, 2018 submitted to MDPI 22 of 30 Figure 16. Our calculated plasmon dispersion (dots) of monolayer phosphorene with EF = 0.07 eV along the armchair direction, in comparison with results obtained by Ghosh et al. for monolayer phosphorene and with the same concentration along the armchair direction (open symbols) [92]. To know the origin of the electronic excitations in the monolayer phosphorene, we obtain the different contributions (interband and intraband) of the non-interacting response function, and they are shown in Figure 17 for q = 0.046 Å−1 and n = 3.9 × 1013 cm−2 along the armchair direction. These results show that the intraband term of this quantity plays an important role, and its interband term is negligible. The loss spectra for different values of the q and dispersion curve of the intraband plasmons of phosphorene for charge carrier concentration of n = 3.9 × 1013 cm−2 along the armchair and zigzag directions are plotted in Figure 18. The anisotropic band structure of monolayer phosphorene along the zigzag and armchair directions makes anisotropic features in the collective plasmon excitations, q dependence at the long wavelength limit. This is due to the although they follow a low energy paraboloidal band dispersion of phosphorene at low-energies, but the plasmon modes in the armchair direction have higher plasmon energy compared to the zigzag direction at the same momenta [93,94]. √ Figure 17. The real (solid line) and imaginary (dashed line) parts of interband and intraband contributions of the non-interacting response function of monolayer phosphorene in units of the Fermi-level density of states as a function of ¯hω for q = 0.046 Å−1 along the armchair direction. 3.6. Bilayer Phosphorene The stable stacking order of bilayer phosphorene is of the AB type such that the bottom layer is shifted half the lattice period along the x or y directions, and this is in agreement with previous studies [95,96]. The crystal structure of AB-stacked bilayer phosphorene is shown in Figure 19a,b. 00.050.10.150.2q (Å-1)00.20.40.60.8h_ω (eV)This workRef. 9200.20.40.60.81h_ω (eV)-0.500.51χ0(q,ω)00.20.40.60.81-0.200.2InterbandIntrabandN(0)Version February 6, 2018 submitted to MDPI 23 of 30 Figure 18. The electron-energy loss function of monolayer phosphorene for different values of q along the ΓX and ΓY directions (upper panel). In the bottom panel, the plasmon modes corresponding to the peaks in the EEL function in the upper panel are plotted. In this case, the Fermi level is shifted up by 0.1 eV, corresponding to n = 3.9 × 1013 cm−2. Figure 19. (a,b) Top and side view of the atomic structure of bilayer phosphorene (BLP). (c) The band structure of bilayer phosphorene along the high symmetry X − Γ − Y directions. The Fermi level is set at 0 eV. The optimized interlayer separation between the closest phosphorus atoms is indicated with d, being 3.56 Å. When two monolayers are combined to create a bilayer, the gap reduces and two additional bands emerge around the gap at the Γ point [97]. Bilayer phosphorene possesses a direct band gap of 0.63 eV, in agreement with the band gap recently reported in [98]. Importantly, the inclusion of many-body effects using GW increases the band gap to 1.45 eV. The electronic band structure of AB-stacked bilayer phosphorene is illustrated in Figure 19c based on our DFT simulations. We find that the valence band maximum is contributed by the localized states of P atoms, while the conduction band minimum is partially contributed from the delocalized states, especially in the interfacial area between the top and bottom layers [99]. In order to investigate more, we illustrate the real and imaginary parts of interband and intraband contributions of the non-interacting response function of bilayer phosphorene in Figure 20 for a specific 0.20.40.60.81h_ω (eV)00.050.10.150.20.2q (Å-1)00.20.40.60.8h_ω (eV)00.10.20.30.40.5h_ω (eV)00.050.10.15q (Å-1)00.20.40.60.8h_ω (eV)0.0230.0460.0690.1040.1160.1620.1390.0920.058ΓYΓX0.0240.0480.0710.0950.1190.1420.1660.1900.213x (armchair)(zigzag)yx z -3-2-1012h_ω (eV)XΓY(a)(c)(b)dVersion February 6, 2018 submitted to MDPI 24 of 30 momentum transfer in the amount of 0.046 Å−1 and EF = 0.05 eV. Our ab initio calculations [100] show that the collective electronic excitations in bilayer phosphorene originated from intraband term of χ0(q, ω), and the other one is not be important enough to be considered. Figure 20. The non-interacting response function in units of the Fermi-level density of states as a function of ¯hω of bilayer phosphorene for q = 0.046 Å−1. We illustrate the loss functions for different amounts of q and also the plasmon modes of bilayer phosphorene along the armchair direction in Figure 21. For this system, we have shifted the Fermi level up by 0.05 eV to imitate the effect of finite doping. As expected, bilayer phosphorene involves two q behavior as the optical mode, as seen in its monolayer. plasmon modes. One of them follows the The other one is a damped acoustic mode with a linear dispersion at the long wavelength limit. √ 00.20.40.60.81h_ω (eV)-0.500.511.5χ0(q,ω)00.20.40.6-0.2-0.100.1InterbandIntrabandN(0)Version February 6, 2018 submitted to MDPI 25 of 30 Figure 21. The optical (black dots) and acoustic plasmon modes (red diamond) of bilayer phosphorene that correspond to the peaks in the EEL functions in the upper panel for n = 3.2 × 1013 cm−2 (EF = 0.05 eV) along the armchair direction. The black line shows the boundary of the electron-hole continuum. 4. Conclusions In this article, we have focused on collective modes of some pristine two-dimensional crystalline materials including graphene, MoS2 and phosphorene in monolayer and bilayer structures. Our studies were based on density-functional theory simulations together with random phase approximation. In the DFT approach, we have considered all electron band structures, and therefore, the non-interacting density-density response function could bring some effects beyond the normal Lindhard function. Furthermore, the Kohn–Sham wave functions in the vicinity of the Fermi energy are quite similar to the real and exact wave functions of the system when the system is doped. Therefore, for the doped 2D crystalline materials, the density-density response function is calculated by the Kohn–Sham wave functions and can describe the many-body effects of the system well enough to use these to explore the physical quantities of the system. In addition, the definition of the dielectric constant given by Equation (11) provides some many-body effects, which could describe why the ab initio-RPA calculations provide a better dispersion relation of the plasmon mode in electronic systems. It is essential to emphasize that the material-specific dielectric function considering the multi-orbital and multiband structures (or quasi-two-dimensional polarization) such as MoS2 are needed to obtain realistic plasmon dispersions. We have used a full DFT simulations together with RPA analysis to calculate the band structure, non-interacting density-density response function, the energy loss functions and, finally, plasmon dispersions of the extrinsic crystalline materials. For each material studied here, we have found different collective modes and described their physical origins. In all studied materials, the in-phase mode, which refers to the optical mode, the plasmon dispersion is q originating from low-momentum carrier scattering. An acoustic mode on some displayed as a √ 00.050.10.150.2q (Å -1)00.20.40.60.8h_ω (eV)00.20.40.60.8h_ω (eV)q (Å-1)0.0230.0460.0690.0920.1150.1360.1630.1850.208(b)(a)Version February 6, 2018 submitted to MDPI 26 of 30 systems is observed; however, this mode is damped. In bilayer MoS2, we have observed that the plasmon modes of the electron and hole doping are not equivalent, and the discrepancy is owed to the fact that the Kohn–Sham band dispersions are not symmetric for energies above or below the zero Fermi level. The anisotropic band structure of monolayer phosphorene along the zigzag and armchair directions make anisotropic features in the collective plasmon excitations, and the plasmon mode in the armchair direction has higher energy compared to the zigzag direction at the same momenta. Acknowledgments: Z.T.would like to thank the Iran National Science Foundation for their support through a research grant. This work is also partially supported by the Iran Science Elites Federation. Author Contributions: R. A conceived and designed the paper; Z. T and R. A analyzed the data and wrote the paper . Conflicts of Interest: The authors declare no conflicts of interest. References 1. 2. 3. 4. 5. 6. 7. 8. 9. Phys. Rev. 1957, 106, 874, Bohm, D.; Pines, D. A Collective Description of Electron Interactions: III. Coulomb Interactions in a Degenerate Electron Gas. Phys. Rev. 1953, 92, 609. doi:10.1103/PhysRev.92.609. Landau, L. On the vibrations of the electronic plasma. J. Phys. (USSR) 1946, 10, 25. Ritchie, R.H. Plasma Losses by Fast Electrons in Thin Films. doi:10.1103/PhysRev.106.874. Koppens, F.H.L.; Chang, D.E.; de Abajo, F.J.G. Graphene Plasmonics: A Platform for Strong Light-Matter Interactions. Nano Lett. 2011, 11, 3370–3377, doi:10.1021/nl201771h. Fer, Z.; Rodin, A.S.; Andreev, G.O.; Bao, W.; McLeod, A.S.; Wagner, M.; Zhang, L.M.; Zhao, Z.; Thiemens, M.; Dominguez, G.; et al. Gate-tuning of graphene plasmons reveald by infrared nano-imaging. Nature 2012, 487, 82–85, doi:10.1038/nature11253. Chen, J.; Badioli, M.; Alonso-González, P.; Thongrattanasiri, S.; Huth, F.; Osmond, J.; Spasenovi´c, M.; Centeno, A.; Pesquera, A.; Godignon, P.; et al. Optical nano-imaging of gate-tunable graphene plasmons. Nature 2012, 487, 77–81, doi:10.1038/nature11254. Lundeberg, M.; Gao, Y.; Asgari, R.; Tan, C.; Duppen, B.V.; Autore, M.; Alonso-Gonzalez, P.; Woessner, A.; Watanabe, K.; Taniguchi, T.; et al. Tuning quantum nonlocal effects in graphene plasmonics. Science 2017, 357, 187–191, doi:10.1126/science.aan2735. Bostwick, A.; Speck, F.; Seyller, T.; Horn, K.; Polini, M.; Asgari, R.; MacDonald, A.H.; Rotenberg, E. Observation of Plasmarons in Quasi-Free-Standing Doped Graphene. Science 2010, 328, 999–1002. Kasry, A.; Ardakani, A.A.; Tulevski, G.S. Menges, B.; Copel, M.; Vyklicky, L. Highly efficient fluorescence quenching with graphene. J. Phys. Chem. C 2012, 116, 2858–2862. 10. Gaudreau, L.; Tielrooij, K.J.; Prawiroatmodjo, G.E.D.K.; Osmond, J.; de Abajo, F.J.G.; Koppens, F.H.L. Universal distance-scaling of nonradiative enegy transfer to graphener. Nano Lett. 2013, 13, 2030–2035. 11. Principi, A.; Carrega, M.; Asgari, R.; Pellegrini, V.; Polini, M. Plasmons and Coulomb Drag in 12. Dirac/Schroedinger Hybrid Electron Systems. Phys. Rev. B 2012, 86, 085421. Faridi, A.; Asgari, R. Plasmons at the LaAlO3/SrTiO3 interface and Graphene-LaAlO3/SrTiO3 double layer. Phys. Rev. B 2017, 95, 165419. 13. Loudon, R. The propagation of electromagnetic energy through an absorbing dielectric. J. Phys. A 1970, 3, 233, doi:10.1088/0305-4470/3/4/515. 14. Maier, S.A. Plasmonics: Fundamentals and Applications; Springer, 2007. 15. Wang, Y.; Plummer, E.W.; Kempa, K. Foundations of Plasmonics. Adv. Phys. 2011, 60, 799–898, doi:10.1080/00018732.2011.621320. 16. Garcia de Abajo, F.J. Graphene Plasmonics: Challenges and Opportunities. ACS Photonics 2014, 1, 135–152, 17. doi:10.1021/ph400147y. Jablan, M.; Sljaci´c, M.; Buljan, H. Plasmons in Graphene: Fundamental Properties and Potential Applications. Proc. IEEE 2013, 101, 1689–1704, doi:10.1109/JPROC.2013.2260115. Version February 6, 2018 submitted to MDPI 27 of 30 18. Yan, H.; Low, T.; Zhu, W.; Wu, Y.; Freitag, M.; Li, X.; Guinea, F.; Avouris, P.; Xia, F. Damping pathways of mid-infrared plasmons in graphene nanostructures. Nat. Photonics 2013, 7, 394–399, doi:10.1038/nphoton.2013.57. 19. Goncalves, P.A.D.; Peres, N.M.R. An Introduction to Graphene Plasmonics; World Scientific Publishing: Singapore, 2016. 20. Tame, M.S.; McEnery, K.R.; Özdemir, S.K.; Lee, J.; Maier, S.A.; Kim, M.S. Quantum plasmonics. Nat. Phys. 21. 2013, 9, 329–340, doi:10.1038/nphys2615. Sholl, D.S.; Steckel, J.A. Density Functional Theory: A Practical Introduction; Wiley-Interscience, New York, 2009. 22. Parr, R.G.; Yang, W. Density-Functional Theory of Atoms and Molecules; Oxford, New York, 1989. 23. Kohn, W.; Sham, L.J. Self-Consistent Equations Including Exchange and Correlation Effects. Phys. Rev. 1965, 140, A1133, doi:10.1103/PhysRev.140.A1133. 24. Giannozzi, P.; Baroni, S.; Bonini, N.; Calandra, M.; Car, R.; Cavazzoni, C.; Ceresoli, D.; Chiarotti, G.L.; I.; et al. QUANTUM ESPRESSO: A modular and open-source software Condens. Matter 2009, 21, 395502, for quantum simulations of materials. J. Phys. Cococcioni, M.; Dabo, project doi:10.1088/0953-8984/21/39/395502. 25. Troullier, N.; Martins, J.L. Efficient pseudopotentials for plane-wave calculations. Phys. Rev. B 1991, 43, 1993–2006. 26. Vanderbilt, D. Soft self-consistent pseudopotentials in a generalized eigenvalue formalism. Phys. Rev. B 1990, 41, 7892–7895. 27. Moroni, E.G.; Kresse, G.; Hafner, J.; Furthmüller, J. Ultrasoft pseudopotentials applied to magnetic Fe, Co, and Ni: From atoms to solids. Phys. Rev. B 1997, 56, 15629, doi:10.1103/PhysRevB.56.15629. 28. Perdew, J.P.; Zunger, A. Self-interaction correction to density-functional approximations for many-electron systems. Phys. Rev. B 1981, 23, 5048, doi:10.1103/PhysRevB.23.5048. 29. Chadi, D.J.; Cohen, M.L. Special Points in the Brillouin Zone. Phys. Rev. B 1973, 8, 5747, doi:10.1103/PhysRevB.8.5747. 30. Giuliani, G.F.; Vignale, G. Quantum Theory of the Electron Liquid; Cambridge University Press: Cambridge, UK, 2005. 31. Adler, S.L. Quantum theory of the dielectric constant in real solids. Phys. Rev. 1962, 126, 413, doi:10.1103/PhysRev.126.413. 32. Petersilka, M.; Gossmann, U.J.; Gross, E.K.U. Excitation energies from time-dependent density-functional theory. Phys. Rev. Lett. 1996, 76, 1212, doi:10.1103/PhysRevLett.76.1212. 33. Pisarra, M.; Sindona, A.; Gravina, M.; Silkin, V.M.; Pitarke, J.M. Dielectric screening and plasmon resonances in bilayer graphene. Phys. Rev. B 2016, 93, 035440, doi:10.1103/PhysRevB.93.035440. 34. Gomez, C.V.; Pisarra, M.; Gravina, M.; Pitarke, J.M.; Sindona, A. Plasmon Modes of Graphene Nanoribbons with Periodic Planar Arrangements. Phys. Rev. Lett 2016, 117, 116801, doi:10.1103/PhysRevLett.117.116801. 35. Kramberger, C.; Hambach, R.; Giorgetti, C.; Rümmeli, M.H.; Knupfer, M.; Fink, J.; Büchner, B.; Reining, L.; Einarsson, E.; Maruyama, S.; et al. Linear plasmon dispersion in single-wall carbon nanotubes and the collective excitation spectrum of graphene. Phys. Rev. Lett 2008, 100, 196803, doi:10.1103/PhysRevLett.100.196803. 36. Rostami, H.; Asgari, R. Electronic ground-state properties of strained graphene. Phys. Rev. B 2012, 86, 155435, doi:10.1103/PhysRevB.86.155435. 37. Polini, M.; Tomadin, A.; Asgari, R.; MacDonald, A.H. Density functional theory of graphene sheets. Phys. Rev. B 2008, 78, 115426, doi:10.1103/PhysRevB.78.115426. 38. Neto, A.H.C.; Guinea, F.; Peres, N.M.R.; Novoselov, K.S.; Geim, A.K. The electronic properties of graphene. Rev. Mod. Phys. 2009, 81, 109, doi:10.1103/RevModPhys.81.109. 39. Despoja, V.; Dekani´c, K.; Šunji´c, M.; Maruši´c, L. Ab initio study of energy loss and wake potential in the vicinity of a graphene monolayer. Phys. Rev. B 2012, 86, 165419, doi:10.1103/PhysRevB.86.165419. 40. Yan, J.; Thygesen, K.S.; Jacobsen, K.W. Nonlocal Screening of Plasmons in Graphene by Semiconducting 2011, 106, 146803, Phys. Lett. and Metallic Substrates: First-Principles Calculations. doi:10.1103/PhysRevLett.106.146803. Rev. 41. Wunsch, B.; Stauber, T.; Sols, F.; Guinea, F. Dynamical polarization of graphene at finite doping. New J. Phys. 2006, 8, 318, doi:10.1088/1367-2630/8/12/318. Version February 6, 2018 submitted to MDPI 28 of 30 42. Jablan, M.; Buljan, H.; Soljaci´c, M. Plasmonics in graphene at infrared frequencies. Phys. Rev. B 2009, 80, 245435, doi:10.1103/PhysRevB.80.245435. 43. Wachsmuth, P.; Hambach, R.; Kinyanjui, M.K.; Guzzo, M.; Benner, G.; Kaiser, U. High-energy collective electronic excitations in free-standing single-layer graphene. Phys. Rev. B 2013, 88, 075433, doi:10.1103/PhysRevB.88.075433. 44. Gao, Y.; Yuan, Z. Anisotropic low-energy plasmon excitations in doped graphene: An ab initio study. Solid State Commun. 2011, 151, 1009–1013, doi:10.1016/j.ssc.2011.05.001. 45. Novko, D.; Despoja, V.; Šunji´c, M. Changing character of electronic transitions in graphene: From 46. single-particle excitations to plasmons. Phys. Rev. B 2015, 91, 195407, doi: 10.1103/PhysRevB.91.195407. Stauber, T. Plasmonics in Dirac systems: From graphene to topological insulators. J. Phys. Condens. Matter 2014, 26, 123201, doi:10.1088/0953-8984/26/12/123201. 47. Li, P.; Ren, X.; He, L. First-principles calculations and model analysis of plasmon excitations in graphene and graphene/hBN heterostructure. Phys. Rev. B 2017, 96, 165417, doi:10.1103/PhysRevB.96.165417. 48. Eberlein, T.; Bangert, U.; Nair, R.R.; Jones, R.; Gass, M.; Bleloch, A.L.; Novoselov, K.S.; Geim, A.; Briddon, P.R. Plasmon spectroscopy of free-standing graphene films. Phys. Rev. B 2008, 77, 233406, doi:10.1103/PhysRevB.77.233406. 49. Despoja, V.; Novko, D.; Dekani´c, K.; Šunji´c, M.; Maruši´c, L. Two-dimensional and π plasmon spectra in pristine and doped graphene. Phys. Rev. B 2013, 87, 075447, doi:10.1103/PhysRevB.87.075447. 50. McCann, E.; Koshino, M. The electronic properties of bilayer graphene. Rep. Prog. Phys. 2013, 76, 056503, doi:10.1088/0034-4885/76/5/056503. 51. Min, H.; Sahu, B.; Banerjee, S.K.; MacDonald, A.H. Ab initio theory of gate induced gaps in graphene bilayers. Phys. Rev. B 2007, 75, 155115, doi:10.1103/PhysRevB.75.155115. 52. Ohta, T.; Bostwick, A.; Seyller T.; Horn, K.; Rotenberg, E. Controlling the electronic structure of bilayer graphene. Science 2006, 313, 951–954, doi:10.1126/science.1130681. 53. Wachsmuth, P.; Hambach, R.; Benner, G.; Kaiser, U. Plasmon bands in multilayer graphene. Phys. Rev. B 2014, 90, 235434, doi:10.1103/PhysRevB.90.235434. 54. Borghi, G.; Polini, M.; Asgari, R.; MacDonald, A.H. Dynamical response functions and collective modes of bilayer graphene. Phys. Rev. B 2009, 80, 241402, doi:10.1103/PhysRevB.80.241402. 55. Profumo, R.E.V.; Asgari, R.; Polini, M.; MacDonald, A.H. Double-layer graphene and topological insulator 56. thin-film plasmons. Phys. Rev. B 2012, 85, 085443, doi:10.1103/PhysRevB.85.085443. Sarma, S.D.; Wang, E.H. Plasmons in Coupled Bilayer Structures. Phys. Rev. Lett. 1998, 81, 4216, doi:10.1103/PhysRevLett.81.4216. 57. McCann, E. Ab initio theory of gate induced gaps in graphene bilayers. Phys. Rev. B 2006, 74, 161403. 58. Guinea, F.; Neto, A.H.C.; Peres, N.M.R. Electronic states and Landau levels graphene stacks. Phys. Rev. B 2005, 73, 245426. 59. Zhang, Y.; Tang, T.T.; Girit, C.; Hao, Z.; Martin, M.C.; Zettl, A.; Crommie, M.F.; Shen, Y.B.; Wang, F. Direct 60. observation of a widely tunable bandgap in bilayer graphene. Nature 2009, 11, 820. Skakalova, V.; Kaiser, A.B. Graphene, Properties, Preparation, Characterisation and Devices; Woodhead Publishing of Elsevier: Cambridge, UK, 2014. 61. Wang, X.F.; Chakraborty, T. Coulomb screening and collective excitations in biased bilayer graphene. Phys. Rev. B 2010, 81, 081402. 62. Bonaccorso, F.; Sun, Z.; Hasan, T.; Ferrari, A.C. Graphene photonics and optoelectronics. Nat. Photonics 2010, 4, 611, doi:10.1038/nphoton.2010.186. 63. Rukelj, Z.; Strkalj, A.; Despoja, V. Optical absorption and transmission in a molybdenum disulfide monolayer. Phys. Rev. B 2016, 94, 115428, doi:10.1103/PhysRevB.94.115428. 64. Kadantsev, E.S.; Hawrylak, P. Electronic structure of a single MoS2 monolayer. Solid State Commun. 2012, 152, 909–913, doi: 10.1016/j.ssc.2012.02.005. 65. Kumara, A.; Ahluwalia, P.K. Electronic structure of transition metal dichalcogenides monolayers 1H-MX2 (M = Mo, W; X = S, Se, Te) from ab-initio theory: New direct band gap semiconductors. Eur. Phys. J. B 2012, 85, 186, doi:10.1140/epjb/e2012-30070-x. 66. Rostami, H.; Roldan, R.; Cappelluti, E.; Asgari, R.; Guinea, F. Theory of strain in single-layer transition metal dichalcogenides. Phys. Rev. B 2015, 92, 195402, doi:10.1103/PhysRevB.92.195402. Version February 6, 2018 submitted to MDPI 29 of 30 67. Rostami, H.; Asgari, R. Valley Zeeman effect and spin-valley polarized conductance in monolayer MoS2 in a 68. perpendicular magnetic field. Phys. Rev. B 2015, 91, 075433, doi:10.1103/PhysRevB.91.075433. Shishkin, M.; Kresse, G. Self-consistent GW calculations for semiconductors and insulators. Phys. Rev. B 2007, 75, 235102, doi:10.1103/PhysRevB.75.235102. 69. Qiu, D.Y.; da Jornada, F.H.; Louie, S.G. Optical Spectrum of MoS2: Many-Body Effects and Diversity of Exciton States. Phys. Rev. Lett. 2013, 111, 216805, doi:10.1103/PhysRevLett.111.216805. 70. Andersen, K.; Thygesen, K.S. Plasmons in metallic monolayer and bilayer transition metal dichalcogenides. 71. Phys. Rev. B 2013, 88, 155128, doi:10.1103/PhysRevB.88.155128. Scholz, A.; Stauber, T.; Schliemann, J. Plasmons and screening in a monolayer of MoS2. Phys. Rev. B 2008, 88, 035135, doi:10.1103/PhysRevB.88.035135. 72. Groenewald, R.E.; Rösner, M.; Schönhoff, G.; Haas, S.; Wehling, T.O. Valley plasmonics in transition metal dichalcogenides. Phys. Rev. B 2016, 93, 205145, doi:10.1103/PhysRevB.93.205145. 73. Cudazzo, P.; Tokatly, I.V.; Rubio, A. Dielectric screening in two-dimensional insulators: Implications for excitonic and impurity states in graphane. Phys. Rev. B 2011, 84, 085406, doi:10.1103/PhysRevB.84.085406. 74. Keldysh, L.V. Coulomb interaction in thin semiconductor and semimetal films. JETP Lett. 1979, 29, 658. 75. Rubio, G.; Stauber, G.; G´'omez-Santos, G.; Asgari, R.; Guinea, F. Orbital magnetic susceptibility of graphene and MoS2. Phys. Rev. B 2016, 93, 085133. 76. Rostami, H.; Asgari, R. Valley Zeeman effect and spin-valley polarized conductance in monolayer MoS2 in a perpendicular magnetic field. Phys. Rev. B 2015, 91, 075433. 77. Cheiwchanchamnangij, T.; Lambrecht, W.R.L. Quasiparticle band structure calculation of monolayer, bilayer, and bulk MoS2. Phys. Rev. B 2012, 85, 205302, doi:10.1103/PhysRevB.85.205302. 78. Debbichi, L.; Eriksson, O.; Lebegue, S. Electronic structure of two-dimensional transition metal dichalcogenide bilayers from ab initio theory. Phys. Rev. B 2014, 89, 205311, doi:10.1103/PhysRevB.89.205311. 79. Komsa, H.-P.; Krasheninnikov, A.V. Effects of confinement and environment on the electronic structure B 2012, 86, 241201, and exciton binding energy of MoS2 from first principles. doi:10.1103/PhysRevB.86.241201. Phys. Rev. 80. Ramasubramaniam, A.; Naveh, D.; Towe, E. Tunable band gaps in bilayer transition-metal dichalcogenides. Phys. Rev. B 2011, 84, 205325, doi:10.1103/PhysRevB.84.205325. 81. Torbatian, Z.; Asgari, R. Plasmon modes of bilayer molybdenum disulfide: A density functional study. J. Phys. Condens. Matter 2017, 29, 465701, doi:10.1088/1361-648X/aa86b9. Schwierz, F. Graphene transistors. Nat. Nanotech. 2010, 5, 487–496, doi:10.1038/nnano.2010.89. 82. 83. Mak, K.F.; Lee, C.; Hone, J.; Shan, J.; Heinz, T.F. Atomically Thin MoS2: A New Direct-Gap Semiconductor. Phys. Rev. Lett. 2010, 105, 136805, doi:10.1103/PhysRevLett.105.136805. 84. Zare, M.; Rameshti, B.Z.; Ghamsari, F.G.; Asgari, R. Thermoelectric transport in monolayer phosphorene. Phys. Rev. B 2017, 95, 045422, doi:10.1103/PhysRevB.95.045422. 85. Wang, V.; Kawazoe, Y.; Geng, W.T. Native point defects in few-layer phosphorene. Phys. Rev. B 2015, 91, 045433, doi:10.1103/PhysRevB.91.045433. 86. Perdew, J.P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865, doi:10.1103/PhysRevLett.77.3865. 87. Elahi, M.; Khaliji, K.; Tabatabaei, S. M.; Pourfath, M.; Asgari, R. Modulation of electronic and mechanical properties of phosphorene through strain. Phys. Rev. B 2015, 91, 115412, doi:10.1103/PhysRevB.91.115412. 88. Tran, V.; Soklaski, R.; Liang, Y.; Yang, L. Layer-controlled band gap and anisotropic excitons in few-layer black phosphorus. Phys. Rev. B 2014, 89, 235319. doi:10.1103/PhysRevB.89.235319. 89. Radisavljevic, B.; Radenovic, A.; Brivio, J.; Giacometti, V.; Kis, A. Single-layer MoS2 transistors. Nat. Nanotechnol. 2011, 6, 147–150, doi:10.1038/nnano.2010.279. 90. Ganesan, V.D.S.O.; Linghu, J.; Zhang, C.; Feng, Y.P.; Shen, L. Heterostructures of phosphorene and transition metal dichalcogenides for excitonic solar cells: A first-principles study. Appl. Phys. Lett. 2016, 108, 122105. doi:10.1063/1.4944642. Fei, R.; Faghaninia, A.; Soklaski, R.; Yan, J.-A.; Lo, C.; Yang, L. Enhanced Thermoelectric Efficiency via Orthogonal Electrical and Thermal Conductances in Phosphorene. Nano Lett. 2014, 14, 6393–6399, doi:10.1021/nl502865s. 91. Version February 6, 2018 submitted to MDPI 30 of 30 92. Ghosh, B.; Kumar, P.; Thakur, A.; Singh Chauhan, Y.; Bhowmick, S.; Agarwal, A. Anisotropic plasmons, excitons, and electron energy loss spectroscopy of phosphorene. Phys. Rev. B 2017, 96, 035422, doi:10.1103/PhysRevB.96.035422. Jin, F.; Roldán, R.; Katsnelson, M.; Yuan, S. Screening and plasmons in pure and disordered single- and bilayer black phosphorus. Phys. Rev. B 2015, 92, 115440, doi:10.1103/PhysRevB.92.115440. 93. 94. Low, T.; Roldán, R.; Wang, H.; Xia, F.; Avouris, P.; Moreno, L.M.; Guinea, F. Plasmons and 2014, 113, 106802, Screening in Monolayer and Multilayer Black Phosphorus. Phys. Rev. Lett. doi:10.1103/PhysRevLett.113.106802. 95. Dai, J.; Zeng, X.C. Bilayer Phosphorene: Effect of Stacking Order on Bandgap and Its Potential Applications 96. in Thin-Film Solar Cells. J. Phys. Chem. Lett. 2014 5, 1289–1293, doi:10.1021/jz500409m. Jhun, B.; Park, C.-H. Electronic structure of charged bilayer and trilayer phosphorene. Phys. Rev. B 2017, 96, 085412, doi:10.1103/PhysRevB.96.085412. 97. Qiao, J.; Kong, X.; Hu, Z.-X.; Yang, F.; Ji, W.H. High-mobility transport anisotropy and linear dichroism in 98. few-layer black phosphorus. Nat. Commun. 2014, 5, 4475, doi:10.1038/ncomms5475. Jin, Z.; Mullen, J.T.; Kim, K.W. Highly anisotropic electronic transport properties of monolayer and bilayer phosphorene from first principles. Appl. Phys. Lett. 2016, 109, 053108, doi:10.1063/1.4960526. 99. Caklr, D.; Sevik, C.; Peeters, F.M. Significant effect of stacking on the electronic and optical properties of few-layer black phosphorus. Phys. Rev. B 2015, 92, 165406, doi:10.1103/PhysRevB.92.165406. 100. Torbatian, Z.; Asgari, R. Collective modes in few layer phosphorous. To be submitted 2018, In preparation. c(cid:13) 2018 by the authors. Submitted to MDPI for possible open access publication under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). 

1202.5606

2

1202

All-electrical injection and detection of a spin polarized current using 1D conductors

All-electrical control of spin transport in nanostructures has been the central interest and chal- lenge of spin physics and spintronics. Here we demonstrate on-chip spin polarizing/filtering actions by driving the gate-defined one dimensional (1D) conductor, one of the simplest geometries for integrated quantum devices, away from the conventional Ohmic regime. Direct measurement of the spin polarization of the emitted current was performed when the momentum degeneracy was lifted, wherein both the 1D polarizer for spin injection and the analyzer for spin detection were demonstrated. The results showed that a configuration of gates and applied voltages can give rise to a tunable spin polarization, which has implications for the development of spintronic devices and future quantum information processing.

cond-mat.mes-hall

cond-mat

All-electrical injection and detection of a spin polarized current using 1D conductors T.-M. Chen,1, 2, ∗ M. Pepper,3 I. Farrer,2 G. A. C. Jones,2 and D. A. Ritchie2 1Department of Physics, National Cheng Kung University, Tainan 701, Taiwan 2Cavendish Laboratory, J J Thomson Avenue, Cambridge CB3 0HE, United Kingdom 3Department of Electronic and Electrical Engineering, University College London, London, WC1E 7JE, United Kingdom All-electrical control of spin transport in nanostructures has been the central interest and chal- lenge of spin physics and spintronics. Here we demonstrate on-chip spin polarizing/filtering actions by driving the gate-defined one dimensional (1D) conductor, one of the simplest geometries for integrated quantum devices, away from the conventional Ohmic regime. Direct measurement of the spin polarization of the emitted current was performed when the momentum degeneracy was lifted, wherein both the 1D polarizer for spin injection and the analyzer for spin detection were demonstrated. The results showed that a configuration of gates and applied voltages can give rise to a tunable spin polarization, which has implications for the development of spintronic devices and future quantum information processing. PACS numbers: There is considerable interest in being able to control spin dynamics, particularly in mesoscopic and nanoscale semiconductor devices[1, 2] as this could lead to the de- velopment of a range of electronic functions not presently available. In order to develop successfully such concepts it is necessary to controllably generate, manipulate, and detect spin currents by electrical means and so minimize, or eliminate, the use of ferromagnetic contacts or external magnetic fields. Most research towards the implemen- tation of this electrical approach has focussed on using the spin-orbit interaction to induce spin polarized trans- port, as reported in various nanostructures[3 -- 6] includ- ing one-dimensional (1D) conductors[7, 8]. However, it is essential to develop a more general approach in which materials with a strong intrinsic spin-orbit coupling are no longer necessary, and consequently a longer spin de- phasing (relaxation) time will be obtained of crucial im- portance for quantum information processing. In theory, it is possible to produce transition from anti- ferromagnetic to ferromagnetic behaviour by controlling the exchange interaction, although this can be difficult to achieve in practice. If such a mechanism could be suc- cessfully utilized for on-chip spin injection, the problems associated with conventional methods of spin injection -- such as the impedance mismatch, which drastically limits the spin polarization (spin pumping efficiency) of the injected current[9] -- can be avoided. Furthermore, the fast-gating technique, which has been well developed in conventional microelectronics, allows it to be used for rapid control of the spin content. Studies of quasi one-dimensional conduction[10] have been of interest for a considerable time due to its strong electron-electron interaction, much of this work has been with reference to the spin properties[11, 12]. The varia- ∗Electronic address: tmchen@mail.ncku.edu.tw tion of the current with the dc source-drain voltage has been shown to be particularly useful in providing quanti- tative measurements on the energies of the 1D subbands in the channel. For ballistic transport this voltage is dropped at the two ends of the channel and lifts the mo- mentum degeneracy, and has been used, for example, to derive the value of the Lande g factor by measuring the spin splitting in a magnetic field[13, 14]. It has also been used to show that there is a spontaneous lifting of the spin degeneracy in the absence of a magnetic field, which is related to the 0.7 structure[14]. Furthermore, there is a feature which appears as a plateau, or structure, with increasing dc source-drain voltage at, or near, the value of 0.25(2e2/h) in the dif- ferential conductance. Although this feature was appar- ent in early work on one-dimensionality[15 -- 17], it was in general regarded as a spin degenerate state with a de- creased differential conductance[18 -- 20]. However, it was recently proposed that the 0.25(2e2/h) feature could be a consequence of a lifting of both momentum and spin degeneracy[21]. The loss of the momentum degeneracy on its own producing a value of e2/h and an absence of spin degeneracy accounting for the remaining factor of 1/2. This is a very surprising result of increasing the source-drain voltage, and in order to substantiate this conclusion it is crucial to provide direct evidence of spin polarization which does not rely on an inference from con- ductance plateaux, particularly because it has been sug- gested that it is possible for the differential conductance value to be reduced in the non-Ohmic regime[18 -- 20]. In this work, we have utilized a technique of electron focusing[22 -- 25] to directly measure the degree of spin polarization of the current. The focusing device geom- etry is shown in Figure 1(a), wherein a small perpen- dicular magnetic field B⊥ is applied to bend and inject ballistic electrons from an emitter, a short one dimen- sional region formed by split gates (quantum point con- tact), which acts as a spin polarizer in this work. The electrons pass through the two-dimensional base region, which is grounded, into the collector which is an identical device to the emitter; in the context of this experiment the collector acts as a spin analyzer. With current flow- ing into the device from the emitter, and with the base connected to ground, the collector-base voltage shows pe- riodic peaks as a function of B⊥ which is due to the focus- ing of electrons into the collector. These focusing peaks occur whenever an integer multiple of the cyclotron di- ameter, 2m∗vF /eB⊥, where m∗ is the electron effective mass and vF is the Fermi velocity, equals the distance, L, between emitter and collector. As the collector is not connected to ground, a volt- age Vc = Ic/Gc develops between the collector and base, where Ic is the current flowing through the collector which has conductance Gc. Both the conductance and current can be further written as Gc = e2/h(T↓ + T↑) and Ic = αIe(T↓+T↑) where the arrows represent the electron spins, Ie = I↓+I↑ is the current injected from the emitter, T↓ (T↑) is down-spin (up-spin) transmission of the collec- tor and α is a parameter, accounting for spin-independent imperfections during the focusing process[22]. This situation has been considered by Potok et al.[22], who have shown that a simple derivation gives the mag- nitude of the height of the peaks in collector-base voltage. This can be written in terms of the degree of spin polar- ization induced by the emitter Pe = (I↓−I↑)/(I↓+I↑) and the spin selectivity of the collector Pc = (T↓ − T↑)/(T↓ + T↑). They found the following relation Vc = α h 2e2 Ie(1 + PePc), (1) which was confirmed by inducing a Zeeman spin split- ting with a strong in-plane magnetic field[22, 23]. Con- sequently, if both emitter and collector are spin polarized the collector voltage is doubled compared to when either emitter or collector allows spin degeneracy. Here we investigated the spin balance in the focusing stream as the conductances of both emitter and collec- tor were varied in the absence of a magnetic field (ex- cept for the small focusing field B⊥). The particular objective was to clarify the spin content of the current when the differential conductance was in the region of the 0.25 plateau. This work used samples comprising a high-mobility two-dimensional electron gas formed at the interface of GaAs/Al0.33Ga0.67As heterostructures. The low temperature mobility was 2.3 × 106 cm2/Vs at a car- rier density 1.17 × 1011 cm2 giving a mean free path for momentum relaxation ∼ 13 µm. This is much longer than the focusing path, although we note that the small angle scattering length is much less and may contribute to a broadening of the focusing peak. Measurements were performed at a temperature of 80 mK, the electrical connections are shown in Fig- ure 1(a). Two devices were measured and gave similar and reproducible results. Simultaneous lock-in measure- ments of the emitter and collector conductances and the 2 1 μm B Vsd = -1.5 mV (b) 2 ) h / 2 e 2 ( G 1 V E C Vsd = 0 mV Vexc Iexc 0 10 G e (2e2/h) : G c -0.9 Gate Voltage (V) -0.8 -0.7 V sd = 0 ; I sd = 0 8 6 4 2 0 8 6 4 2 0 0.00 G e (2e2/h) : G c V sd = -1.5 mV ; I sd = 30 nA 0.05 0.10 0.15 Magnetic Field (T) (a) (c) ) V μ ( e g a t l o V r o t c e l l o C (d) ) V μ ( e g a t l o V r o t c e l l o C FIG. 1: (a) Micrograph and electric circuit showing the emitter-collector configuration used in the experiment. Elec- trons are focused by a small perpendicular magnetic field and travel from the emitter (E), through the 2D base region (B) into the collector (C). (b) The differential conductance of the emitter at B = 0 with application of a source-drain bias from Vsd = 0 to −1.5 mV in steps of −0.3 mV. The con- ductance anomaly at around G = 0.25(2e2/h) starts when Vsd = −0.9 mV. Data are offset for clarity. (c) The collector voltage (focusing signal) for Vsd = Isd = 0; the focusing volt- age is nearly independent of the conductance of both emitter and collector from 2e2/h to 0.25(2e2/h). (d) A substantial rise in the focusing signal appears when both emitter and col- lector were set to G = 0.25(2e2/h) and the dc source-drain biases Vsd = −1.5 mV and Isd = 30 nA were applied respec- tively. focusing signal were performed by applying two indepen- dent excitation sources of (i) a 77 Hz ac voltage 20 µV with a dc bias Vsd applied to the emitter and (ii) 31 Hz ac current 1 nA with a dc bias Isd applied to the collec- tor. It was verified that the focusing signal Vc was linear with current Ie; for clarity, all the data presented here was rescaled for Ie = 1 nA. The current-bias excitation, i.e., source (ii), is required to prevent the collector from sinking injected current, as well as increasing the bias across the collector pushing it into the 0.25 regime. Both the emitter and collector show one-dimensional conductance quantisation and a source-drain voltage in- duced plateau at 0.25(2e2/h) at B = 0[21], as shown in Figure 1(b). The measured focusing peaks are shown in Figure 1(c) when the emitter and collector are set at the described values of conductance for Vsd = Isd = 0. Focusing peaks appear periodically, at intervals of B⊥ = 0.05 T, which is consistent with the cyclotron motion B⊥ = 2m∗vF /eL calculated from the two- dimensional electron concentration. The height of the focusing peaks, as anticipated, barely changes with de- creasing conductance of both emitter and collector from 2e2/h to 0.25(2e2/h), indicating that there is no change in their spin polarization, i.e., Pe and/or Pc = 0. As expected the peak height is independent of Ge and Gc for constant current Ie injected from the emitter point contact. When a dc bias is applied across the emitter and col- lector the focusing peak exhibits very different behaviour to that previously observed at zero bias. In Figure 1(d), the focusing peaks are shown for various Gc when Ge is fixed at 0.25(2e2/h), and when the dc biases across the emitter and collector were set at Vsd = −1.5 mV and Isd = 30 nA, respectively. It was observed that the peak height barely changes with decreasing Gc from 1.5(2e2/h) to 0.5(2e2/h), but rises considerably when this approaches 0.25(2e2/h), i.e., where the anomalous plateau is found as shown in Figure 1(b). This substan- tial rise is predicted by Equation (1), if there is an in- creasing degree of spin polarization in both the emitter and collector. Figure 2(a) shows the height of the first peak as both the dc biases and the collector conductance were varied with the emitter conductance locked at Ge = 0.25(2e2h); this peak was chosen for investigation because of its robust structure and is seen to stay fairly constant at ∼ 3 µV, essentially independent of both Vsd and Isd, when Gc ∼ 2e2/h. However, in the low conductance region when Gc ≤ 0.25(2e2/h), and Isd = 30 nA, the focusing peaks increase as the dc bias is increased, neg- atively, from Vsd = 0 and then saturates when Vsd is near −0.9 mV. The focusing peaks at Vsd ≥ 0.9 mV are approximately twice the value of those at Vsd = 0 for every individual value of collector conductance below ∼ 0.25(2e2/h). This, according to Eq. (1), implies that both emitter and collector are fully spin polarized, i.e., Pe = Pc = 1. The saturation of the peak height is also consistent with the fact that both Pe and Pc cannot be larger than 1. To further verify this bias-induced spin polarization, Isd was decreased from 30 nA to 0 with Vsd still at −1.5 mV. Figure 2(a) shows that the height of the focus- ing peaks drops back to almost the same value obtained when Vsd = 0 and Isd = 30 nA as well as when both Vsd and Isd are zero. This is again expected when ei- ther polarizer or analyzer are spin degenerate (i.e., either Pe or Pc equals 0). Finally, it is important to note that the value of source-drain bias Vsd = −0.9 mV at which the focusing peak height Vc saturates is consistent with the bias at which the 0.25 anomaly appears, as shown in Vsd (mV) : Isd (nA) 0 : 30 -0.3 : 30 -0.6 : 30 -0.9 : 30 -1.2 : 30 -1.5 : 30 -1.5 : 0 0 : 0 GC at Isd = 30 nA GC at Isd = 0 nA Vsd (mV) : Isd (nA) 0 : 30 -0.3 : 30 -0.6 : 30 -0.9 : 30 -1.2 : 30 -1.5 : 30 GC at Isd = 30 nA 3 G ( 2 e 2 / h ) G ( 2 e 2 / h ) 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 -0.91 -0.90 -0.89 -0.88 -0.87 -0.86 -0.85 -0.84 Gate Voltage (V) (a) 10 ) V μ ( e g a t l o V r o t c e l l o C (b) o i t a R k a e P 8 6 4 2 2.0 1.5 1.0 0.5 FIG. 2: (a) Voltage of the first focusing peak, left, and the corresponding conductance of the collector, right, as a func- tion of gate voltage, for various values of dc source-drain bias Vsd and Isd. Enhancement of focusing peaks only occurs near the 0.25 conductance region when dc source-drain biases are applied across both the emitter and collector. (b) Normal- ized peak ratio and corresponding conductance as a function of gate voltage, with Isd set to 30 nA and Vsd swept from 0 to −1.5 mV. The peak ratio rises and then saturates when both conductance and dc source-drain bias are at the appropriate values for the appearance of the 0.25(2e2/h) plateau. Figure 1(b). The evolution of the focusing peaks as a function of conductance[26] is also shown in Figure 2(a), the focus- ing peak rises as Gc is reduced below 2e2/h, but the manner of the increase varies for different dc source-drain biases. At Vsd = 0 and Isd = 30 nA, the peak voltage barely increases until Gc is reduced below ∼ 0.15(2e2/h), in the near-pinch-off region, whereas at Vsd < 0 and Isd = 30 nA the peak voltage starts to increase once Gc is reduced below ∼ 0.6(2e2/h). The near-pinch-off in- crease in peak voltage with the reduced value of Gc could be attributed to an α-dependent enhancement; this has been suggested previously when Gc is low[22] although the origin is not clear. To remove nonspin related effects from the focusing peak, all the peak voltages are normalized by the val- ues at Vsd = 0 and Isd = 30 nA[27]. Figure 2(b) shows the normalized peak ratio, proportional to (1 + PePc), and the corresponding conductance as a function of gate voltage, with Isd set to 30 nA and Vsd swept from 0 to −1.5 mV. As seen the peak values rise with reducing con- ductance and then saturate when Gc reaches the region 0 1 = 2e2/h G c 20 V sd = -1.5 mV Ge (2e2/h) 2 § * 3 4 20 15 10 5 C o l l e c t o r V o l t a g e ( μ V ) 0 0.15 (a) ) t i n u . b r a ( e g a t l o v e a g t t i l p S (b) ) V μ ( e g a t l o V r o t c e l l o C 15 10 5 0 8 6 4 2 -0.05 0.00 0.05 0.10 B (T) Isd = 30 nA; split peak 1 Isd = 30 nA; split peak 2 Isd = 30 nA; unsplit peak (0.25 regime) Isd = 0 nA; split peak 1 Isd = 0 nA; split peak 2 Isd = 0 nA; unsplit peak (0.25 regime) = 0.25(2e2/h) G c V sd = -1.5 mV 0 0.00 0.25 0.50 0.75 1.00 Ge (2e2/h) FIG. 3: (a) The focusing peak voltage (right half) versus B- field for several emitter conductance Ge from 0 to 1.5(2e2/h), along with its corresponding value of Ge (left half) when the source-drain bias Vsd = -1.5 mV is applied across the emitter. The focusing voltage is offset for clarification. The focusing peak starts to split in two when the Ge starts to rise from the 0.25(2e2/h) plateau. (b) Voltage of the first focusing peak versus Ge when Gc is locked at the 0.25 plateau. Both the split peak 1, the peak marked by § in (a), and the split peak 2, the peak marked by ⋆ in (a), barely change with Isd and Ge. Enhancement of focusing peaks only occurs when both Ge and Gc are at the 0.25 regime. of the 0.25 plateau, suggesting that Pc has reached its maximum value of 1. Similarly, the peak ratio in the 0.25 regime rises with increasing source-drain bias ap- plied across the emitter and then saturates. This reaches a value of ∼ 2, at Vsd = −0.9 mV when the 0.25 feature appears in the emitter conductance. Such focusing peak enhancement was further verified by another set of measurement where Gc was locked at 0.25(2e2/h) as both the dc biases and Ge were varied. Figure 3(a) shows that the focusing peak splits when the emitter conductance Ge is above the 0.25(2e2/h) plateau, indicating that both the source and the drain chemical potential have reached a 1D subband. The two split peaks represent focusing electrons with the source and the drain potential, respectively, whereas in the 0.25 and zero-bias regions no splitting occurs because there is only one potential across the 1D subband. Indeed the peak enhancement, which is the evidence of spin polarization, occurs only when both Ge and Gc are in/below the bias-induced 0.25 plateau. Figure 3(b) clearly shows that the split focusing peaks barely changes 4 with Ge [only when Ge > 0.25(2e2/h)] and Isd when Gc is locked at 0.25(2e2/h). The enhancement only occurs when Ge = 0.25(2e2/h) and the source-drain bias Isd is applied. In addition, it is noteworthy that the spin- polarized 0.25(2e2/h) plateau is very robust, nearly in- dependent of temperatures up to 4.2 K. These results show that the emitter is functioning as a spin polarizer and the collector as a spin analyzer, demonstrating that a manipulation of the degree of polar- ized spin current can be achieved by tuning the source- drain bias at low values of conductance. For instance, Figure 2(b) shows that the spin polarization of the in- jecting current Pe was ∼ 30% when the emitter was set to Vsd = −0.3 mV and Ge = 0.25(2e2/h), whereas Pe reaches 60 ∼ 70% for Vsd = −0.6 mV. We note that in the region of the 0.7 anomaly, which is found in the ab- sence of bias, the enhancement of the peak height will be ∼ 10% and difficult to observe unambiguously. The effects observed here indicate that the non- equilibrium electron energy distribution and the spin co- herence are maintained during the focusing transit into the collector. The transit time is sufficiently short (ap- proximately 20 picoseconds) that phonon emission is not occurring to any significant degree, so allowing all the emitted electrons to enter the collector. The spin co- herence length exceeds the path length so that the spin polarization is maintained during the focusing which au- gurs well for applications of this phenomenon. Our experiments establish a link between spin and mo- mentum which is unusual in the system with weak spin- orbit coupling. It seems most likely that the cause of the 0.25 is that a spin polarized stream of electrons is the low- est energy configuration; this configuration is retained as there is only one direction of momentum and an absence of spin scattering by electrons with the opposite momen- tum. A physical mechanism based on exchange inter- action has recently been proposed for the 0.25 anomaly which explains the lifting of the spin degeneracy[28] in the regime of non-equilibrium transport. How such exchange induced spin polarization is retained, or enhanced, by an absence of momentum degeneracy is puzzling. However, for practical applications, it is now possible to vary the degree of spin polarization in a way not previously pos- sible. A complex arrangement of gates and applied volt- ages can be utilized for on-chip spin manipulation with applications in spintronics and quantum information pro- cessing. Acknowledgements We thank L.W. Smith, K.-F. Berggren, and C.-T. Liang for useful discussions. For technical assistance, we thank D. Anderson and S.-J. Ho. This work was supported by the Engineering and Physical Sciences Re- search Council (UK), EU Spinmet, and the National Sci- ence Council (Taiwan). I.F. acknowledges support from Toshiba Research Europe. 5 [1] I. Zutic, J. Fabian, S. Das Sarma, Rev. Mod. Phys. 76, (2009). 323-410 (2004). [2] A. Fert, Rev. Mod. Phys. 80, 15171530 (2008). [3] Y. K. Kato, R. C. Myers, A. C. Gossard, D. D. [21] T.-M. Chen et al., Appl. Phys. Lett. 93, 032102 (2008). [22] R. M. Potok, J. A. Folk, C. M. Marcus, V. Umansky, Phys. Rev. Lett. 89, 266602 (2002). Awschalom, Science 306, 19101913 (2004) [23] J. A. Folk, R. M. Potok, C. M. Marcus, V. Umansky, [4] M. Konig et al., Science 318, 766770 (2007). [5] D. Hsieh et al., Nature 452, 970974 (2008). [6] S. M. Frolov et al., Nature 458, 868871 (2009). [7] P. Debray et al., Nature Nanotechnology 4, 759 (2009). [8] C. H. L. Qua et al., Nature Physics 6, 336 (2010). [9] G. Schmidt et al., Phys. Rev. B 62, R4790 (2000). [10] T. J. Thornton, M. Pepper, H. Ahmed, D. Andrews & G. J. Davies, Phys.Rev.Lett. 56, 1198 (1986). [11] K. J. Thomas et al., Phys. Rev. Lett. 77, 135 (1996). [12] A. P. Micolich, J. Phys.: Condens. Matter 23, 443201 (2011). [13] N. K. Patel et al., Phys. Rev. B 44, 10973 (1991). [14] T.-M. Chen et al., Nano Lett. 10, 2330-2334 (2010). [15] N. K. Patel et al., Phys. Rev. B 44, 13549 (1991). [16] K. J. Thomas et al., Phil. Mag. B 77, 1213 (1998). [17] A. Kristensen et al., Phys. Rev. B 62, 10950 (2000). [18] R. de Picciotto, L. N. Pfeiffer, K. W. Baldwin, K. W. Science 299, 679-682 (2003). [24] L. P. Rokhinson, V. Larkina, Y. B. Lyanda-Geller, L. N. Pfeiffer, K. W. West, Phys. Rev. Lett. 93, 146601 (2004). [25] L. P. Rokhinson, L. N. Pfeiffer, K. W. West, Phys. Rev. Lett. 96, 156602 (2006). [26] Differential conductance of the collector was measured at a constant current.This prevents the one-dimensional channel from being fully pinched off. When the dc bias current is high (Isd = 30 nA) the lowest conductance is 0.25(2e2/h) and consequently a 0.25 plateau is not ob- served although the spin polarization is identical. Also note that at a fixed dc current bias, Isd, the correspond- ing dc voltage bias varies with the conductance. [27] The collector differential conductance at a particular gate voltage is dependent on source-drain bias and, therefore, normalization is not possible for different Isd. [28] H. Lind, I. I. Yakimenko, & K. F. Berggren, Phys. Rev. West, Phys. Rev. Lett. 92, 036805 (2004). B 83, 075308 (2011). [19] H. Kothari et al., J. Appl. Phys. 103, 013701 (2008) [20] S. Ihnatsenka, I. V. Zozoulenko, Phys. Rev. B 79, 235313 

1306.5876

1

1306

Electromechanical Piezoresistive Sensing in Suspended Graphene Membranes

Monolayer graphene exhibits exceptional electronic and mechanical properties, making it a very promising material for nanoelectromechanical (NEMS) devices. Here, we conclusively demonstrate the piezoresistive effect in graphene in a nano-electromechanical membrane configuration that provides direct electrical readout of pressure to strain transduction. This makes it highly relevant for an important class of nano-electromechanical system (NEMS) transducers. This demonstration is consistent with our simulations and previously reported gauge factors and simulation values. The membrane in our experiment acts as a strain gauge independent of crystallographic orientation and allows for aggressive size scalability. When compared with conventional pressure sensors, the sensors have orders of magnitude higher sensitivity per unit area.

cond-mat.mes-hall

cond-mat

This document is the unedited Author’s version of a Submitted Work that was subsequently accepted for publication in Nano Letters, copyright © American Chemical Society after peer review. To access the final edited and published work see http://pubs.acs.org/articlesonrequest/AOR-kGn5XaTPGiC7mYUQfih3. Electromechanical Piezoresistive Sensing in Suspended Graphene Membranes A.D. Smith1, F. Niklaus1, A. Paussa2, S. Vaziri1, A.C. Fischer1, M. Sterner1, F. Forsberg1, A. Delin1, D. Esseni2, P. Palestri2, M. Östling1, M.C. Lemme1,3,* 1 KTH Royal Institute of Technology, Isafjordsgatan 22, 16440 Kista, Sweden, 2 DIEGM, University of Udine, Via delle Scienze 206, 33100 Udine, Italy 3University of Siegen, Hölderlinstr. 3, 57076 Siegen, Germany                                                                                                                           *  Corresponding  author:  max.lemme@uni-­‐siegen.de   Abstract Monolayer graphene exhibits exceptional electronic and mechanical properties, making it a very promising material for nanoelectromechanical (NEMS) devices. Here, we conclusively demonstrate the piezoresistive effect in graphene in a nano- electromechanical membrane configuration that provides direct electrical readout of pressure to strain transduction. This makes it highly relevant for an important class of nano-electromechanical system (NEMS) transducers. This demonstration is consistent with our simulations and previously reported gauge factors and simulation values. The membrane in our experiment acts as a strain gauge independent of crystallographic orientation and allows for aggressive size scalability. When compared with conventional pressure sensors, the sensors have orders of magnitude higher sensitivity per unit area. Keywords: graphene, pressure sensor, piezoresistive effect, nanoelectromechanical systems (NEMS), MEMS   2   Graphene is an interesting material for nanoelectromechanical systems (NEMS) due to its extraordinary thinness (one atom thick), high carrier mobility 1,2, a high Young’s modulus of about 1 TPa for both pristine (exfoliated) and chemical vapor deposited (CVD) graphene.3,4 Graphene is further stretchable up to approximately 20%.5 In addition, it shows strong adhesion to SiO2 substrates6 and is nearly impermeable for gases, including helium.7 In this article, we demonstrate piezoresistive pressure sensors based on suspended graphene membranes with direct electrical signal read-out. We utilize a piezoresistive effect induced by mechanical strain in the graphene, which changes the electronic band structure8 and exploits the fact that the sensitivity of membrane-based electromechanical transducers strongly correlates with membrane thickness9. While graphene has been used as a piezoresistive strain gauge on silicon nitride10 and polymer membranes11, we extend the use of the graphene to both membrane and electromechanical transduction simultaneously with an average gauge factor of 2.92. The sensitivity per unit area of our graphene sensor is about 20 to 100s of times higher than that of conventional piezoresistive pressure sensors. The piezoresistive effect is nearly independent of crystallographic orientation. In our experiments, graphene membranes made from CVD graphene are suspended over cavities etched into a SiO2 film on a silicon substrate. The graphene is electrically contacted and the devices are wire-bonded into a chip package. Process schematics are shown in Fig. 1a through c, while details of the fabrication process are described in the methods section. A scanning electron microscope image of a wire- bonded device and a photograph of a packaged device are shown in Fig. 1d. If a pressure difference is present between the inside and the outside of the cavity   3   (compare Fig. 1c), the graphene membrane that is sealing the cavity is deflected and thus strained. This leads to a change of device resistivity due to the piezoresistive effect in the graphene. Measurements were performed in an argon environment in order to reduce the effects of adsorbates. If air is used instead of argon for the experiments, adsorption of non-inert gases and/or molecules on the graphene will affect the resistivity (see details in supporting information). In the experiments, the packaged devices are placed inside a vacuum chamber. The chamber is then evacuated from atmospheric pressure down to 200 mbar, and then vented back to 1000 mbar. Thus the air sealed inside the cavity presses against the graphene membrane with a force proportional to the chamber pressure. The resistance of the graphene sensor is measured in a Wheatstone bridge (see supporting information), where the graphene membrane is one of the resistors in the bridge. The Wheatstone bridge is balanced at atmospheric pressure by adjusting a potentiometer to the same resistance value as the graphene membrane. The bridge is biased with 200mV square wave pulses with durations of 500 µs. These values were chosen to avoid excessive heating of the graphene device. The voltage output signal from the Wheatstone bridge is amplified and low pass filtered before being sampled with an analog-to-digital converter and converted into its corresponding resistance value. The experimental conditions were chosen to remain within the expected tearing limits of the graphene membrane.6 The suspended membrane sensors were first compared to devices with identical dimensions fabricated in parallel, but without cavities. This was done in order to verify that it is indeed the presence of the cavity and the resulting mechanical bending and   4   straining of the membrane that causes the pressure dependence of the resistance. Fig. 2a shows the amplified voltage (with an amplification factor of 870) versus pressure curve for two devices, one with a cavity and one without. In contrast to the reference device without the cavity (red hollow circles), the sensor device with the suspended graphene membrane (blue squares) shows a strong correlation of the resistance with respect to pressure. The graph includes data of six measurement cycles, where each cycle represents one pump-down or one venting of the chamber. Fig. 2b shows the average change in voltage of three cavity devices in comparison to two non-cavity devices. As can be seen, there is a very strong correlation between the devices’ sensitivity to pressure and the presence of a cavity. Finally, A device was held at constant pressures in order to investigate potential drift in the sensor signal (Fig. 2c). While there is a noticeable drift at several pressures, the resistance values generally follow the pressure. Nevertheless, further studies regarding stability are required. The sensitivity of piezoresitive membrane-based pressure sensors is given by Eq. 1, where S is the sensitivity, R is the resistance, V is the voltage, I is the current, and P is 𝑆 = ∆!! ∙! = ∆!!!! ∙! = ∆!! ∙! the pressure difference acting on the membrane.12 If the current is held constant, then the sensitivity based on voltage (1) measurements can be directly compared to the sensitivity based on the maximum change in resistance for a change in pressure of 477 mbar. The sensitivity of the piezoresitive graphene pressure sensor in Fig. 2 is measured to be 3.95 µV/V/mmHg.   5   The graphene membrane-based pressure sensor, though much smaller than conventional piezoresitive pressure sensors, outperforms conventional piezoresitive Si- based and carbon nanotube (CNT) based pressure sensors reported in literature.13-16 (see Table 2 in the supporting information for details). In general, the sensitivity S of membrane-based piezoresistve pressure sensors is dependent on the membrane material characteristics, the membrane thickness and the membrane area (see supporting information).3,12,17,18 When normalizing the sensitivity of the pressure sensors from Table 2 of the supporting information to a standard membrane area, the sensitivity of our graphene sensor is about 20 to 100s of times higher than the other sensors (Fig. 2d). This and the fact that the graphene sensor is already smaller in area than any of the other sensors indicate great potential for further size-reduction of graphene membrane-based sensors. In order to estimate the piezoresistive gauge factor of the graphene transducer in our sensor, the change in resistance of the cavity region must be determined. A finite element analysis of the deflection was performed using COMSOL multiphysics and calibrated using literature data of graphene membrane deflection obtained by atomic force microscopy.6,7 Material parameters taken from literature were used in the COMSOL model such as the elastic constant Et = 347 N/m, where E is the Young's modulus, t is the membrane thickness (t = 0.335 nm) and Poisson's ratio ν = 0.16. A comparison between the model and measured literature values is shown in Table 2 of the supporting information. Good agreement is noted both with the measurements of a 2.3 µm radius circular membrane in Koenig et al.6 and with the measurement in Bunch et al. on a square 4.75 µm x 4.75 µm membrane at a pressure difference of 930 mbar.7   6   The derived model was then applied to the 6 µm by 64 µm cavity used in the current experiment to estimate the deflection of the membrane. At a pressure difference of 477 mbar, the deflection of the membrane is calculated to be 202 nm (Fig. 3a), which results in an average strain of 0.290% across the membrane. An electrically equivalent circuit of the sensor is schematically shown in Fig. 3b !!!! !!!! !!! + 𝑅! and is described in Eq. 2 𝑅!"! = 𝑅! + ! The total resistance Rtot is taken from resistance measurements at chamber pressures of 1000 mbar and 523 mbar (Fig. 2a). The resistances R1 through R5 correspond to the resistances of the regions shown in Figure 3b. R2 represents the (2) resistance of the graphene membrane over the cavity and we assume that only R2 changes as a function of pressure. Using this method (details in supporting information), R2 is determined to be 0.191 kΩ at 1000 mbar, and the percent resistance change of the graphene membrane patch (R2) is determined to be 0.59%. The intrinsic graphene gauge factor in our sensor was then calculated as the percent change in resistance divided by the percent change in strain to be 3.67. Gauge factors vary depending on the pressure range measured with a maximum value of 4.33 and an average value of 2.92. Previous literature, by comparison, reports gauge factors of 1.9 for suspended graphene beams19, about 150 for graphene on SiO2 20 and nearly 18000 for graphene on a silicon nitride membrane.10   7   Simulations of the change of the graphene pressure sensor resistance due to strain were carried out in order to interpret the experimental results. For a low electric (𝐸! = 0 eV), the resistance R2 of the graphene foil suspended over the cavity is field in the transport direction and considering a Fermi level close to the Dirac energy 𝑅! = 𝜌 𝐿′𝑊! ′ = 1 (1 + 𝜀!! )𝐿 (1 + 𝜀!! )𝑊! ,             (3)   2𝑞𝑁! 𝜀 𝜇! 𝜀 expressed as where 𝜌 is the resistivity of the suspended graphene sheet, 𝑁! and 𝜇! are respectively as the hole density and hole mobility, respectively, since we set EF = 0 eV), 𝜀!! and 𝜀!! the electron density and the corresponding mobility (that are assumed to be the same normal to the transport, and 𝑞 is the positive electron charge. Note that the strain are the components of the strain tensor respectively in the direction of the transport and induced in graphene by the pressure difference between the cavity and the chamber Δ𝑝 influences both the terms 𝐿′ and 𝑊! ′ related to the geometry as well as the resistivity  𝜌. of the transport is dominant (𝜀!! ≫ 𝜀!! ). If ρ is not modified by strain, the change of 𝐿 The induced strain can be considered quasi-uniaxial since its component in the direction and 𝑊! alone is not enough to explain the experimental resistance change with strain. Then, the influence of strain on 𝜌 has been analyzed by starting with the effect of strain on the electron density 𝑁! , simulated by employing the strained graphene bandstructure stemming from the Tight-Binding (TB) Hamiltonian presented in Pereira et al.8, supporting information). 𝑁! increases with the strain, which, in contrast with the recalibrated to accurately reproduce DFT results reported by Huang et al.19 (see experiments (see Fig. 2a), would lead to a decrease of the resistivity (see Eq. 3 and supporting information). Hence the changes in the graphene charge are not sufficient to   8   explain the observed change of the resistance R2 with the strain. The effect of capacitive coupling was also explored through simulation and, though present, is found For this reason the effect of the strain on the mobility 𝜇! is simulated by solving to cause changes in resistance much lower than those observed experimentally. the Linearized Boltzmann Transport Equation (LBTE)21. This approach gives the exact solution of the LBTE even in the presence of anisotropic and non-monotonic energy dispersion relation and anisotropic scattering rates. In these calculations Neutral mobility decreases with increasing Δ𝑝 (see supporting information). Such mobility Defects (ND)22 are considered, which are dominant in CVD graphene23. The electron degradation more than compensates the 𝑁! enhancement, so that the calculations lead strain  is also compared (Fig. 3c). As can be seen the simulation results do not critically to an overall R2 increase. The simulated versus measured R2 modulation versus depend on whether the strain is uniaxial or biaxial and the overall agreement with experiments is reasonably good. Fig. 3d compares the corresponding calculated and measured gauge factors; the biaxial or uniaxial nature of the strain has a modest influence on the gauge factor. Simulation data from Huang et al. are also included for reference.19 Note also that, due to the flexibility of the approach, the relative variation of R2 is independent of the graphene orientation with respect to the direction of the transport (Fig. 3e and 3f), of the defect concentration, and of the considered Fermi level (i.e. the carrier density). This is an important aspect of our work because it means that the effect is independent of random crystallographic alignment and multiple grain graphene flakes.   9   The piezoresistive effect in graphene was demonstrated in graphene–membrane pressure sensors. The sensitivity of piezoresistive graphene sensors is superior to silicon and CNT-based sensors and orders of magnitude more sensitive when normalized for membrane dimensions. This is in line with theoretical considerations that indicate such a decisive advantage due to graphene’s extraordinary thinness. A finite element simulation is derived to describe the deflection of graphene membranes over sealed cavities as a function of pressure and verified with literature data. The estimated maximum gauge factor for graphene based on this model is 4.33 with values averaging at 2.92. Tight binding calculations support the experimental data, including only a small dependence of the observed effect on crystal orientation. This work demonstrates that thin graphene membranes can be efficiently implemented as piezoresistive transducer elements for emerging NEMS sensors.   10   Methods Devices are fabricated on p-type silicon substrates with a thermally grown silicon dioxide (SiO2) layer of 1.5 µm. Rectangular cavities of 6 µm by 64 µm are etched 650 nm deep into the SiO2 using a resist mask and an Ar and CHF3-based reactive ion etching (RIE) process at 200mW and 40 mTorr to provide vertical etch profiles. Next, contact areas are defined by lithography and etched 640 nm into the SiO2 layer using again an RIE process. The contact cavities are then filled with a 160 nm layer of titanium followed by a 500 nm layer of gold using metal evaporation so that the contacts are raised about 20 nm above the surface of the SiO2. The contacts are buried to prevent wire bonding from ripping the contacts off of the substrate. This has the added advantage of allowing the graphene to be transferred in a later step, which improves the cleanliness of the process and reduces the risk of rupturing the graphene membranes during processing. Also, the graphene-metal contacts are not degraded by polymer residues in this way. Commercially available chemical vapor deposited (CVD) monolayer graphene films on copper foils are used. The graphene on one side of the copper is spin-coated with either a poly(methyl methacrylate) (PMMA) or poly(Bisphenol A) carbonate (PC) layer in order to act as a mediator between the initial and final substrate 24-28. The graphene on the backside of the foil is etched using O2 plasma and the copper foil is subsequently wet etched in FeCl3 and then transferred into de-ionized water. The bottom left of Fig. 2a shows a contrast enhanced image of graphene with a polymer coating floating in a solution of FeCl3 after the copper is etched away. The PMMA/graphene film is picked up with the chip and dried on a hotplate. After drying, the chip is placed into a solution of Chloroform overnight in order to etch the PC polymer   11   layer. Next, a photoresist layer is applied and exposed in order to pattern the graphene. Finally, the graphene is etched into the desired shape using an O2 plasma etch and the photoresist is removed in acetone. Once the devices are fabricated, the chips are placed into a chip housing and gold wires are bonded from the housing to the contact pads. The layout of the contacts is shown schematically in Fig. 1c and the wire bonded device is shown in a scanning electron micrograph in Fig. 1d, Raman spectroscopy and electrical measurements were performed to verify the presence of graphene (see supporting information). ACKNOWLEDGMENTS Support from the European Commission through an ERC Advanced Investigator Grant (OSIRIS, No. 228229) and two Starting Grants (M&M’s, No. 277879 and InteGraDe, No. 307311) as well as the German Research Foundation (DFG, LE 2440/1-1) and the Italian MIUR through the Cooperlink project (CII11AVUBF) is gratefully acknowledged.   12    Figure 1: a) The graphene transfer process. A Layer of PMMA or PC is applied to one side of chemical vapor deposited (CVD) graphene on copper foil. Graphene is then etched from the back side of the copper foil using O2 plasma. Finally, the copper is etched using FeCl3. b) Fabrication sequence of the pressure sensor and the corresponding transfer of graphene onto the substrate. Once the graphene is transferred to the chip, the polymer layer is removed and the graphene is etched (c). After fabrication of the devices, they are packaged and wire bonded (d).   13    Figure 2:. a) Pressure versus voltage measurements of a device with a cavity (blue squares) and a device without a cavity (red hollow circles). There is a clear dependence in the case of the device with a cavity, where the pressure difference leads to bending and strain in the graphene membrane. This dependence is not observed in the unsuspended device. b) Average rate of change of the voltage relative to the pressure for the cavity devices compared to the non-cavity devices. Error bars show their respective standard deviation. c) Resistance of the same cavity device (black squares) compared to the pressure (red line). The pressure was held constant at different levels. d) Comparison of sensitivity. Normalized sensitivity per unit area for the graphene pressure sensors in this   14   paper compared to silicon and carbon nanotube-based sensors.13-16 The graphene sensor is roughly 20 to 100 times more sensitive per unit area than the conventional MEMS sensors showing the potential for aggressive scaling. Tabulated sensitivity values are shown in the supporting information.13-16   15    Figure 3: a) Model of the cavity deflection for the membrane dimensions in the experiment. The plot shows one half of the symmetric 6 µm by 64 µm membrane in the deflected state at a pressure difference of 477 mbar giving a total deflection of 202 nm. b) Different components of the resistor model that was used in order to calculate a gauge factor based on the experimental results. Simulated (lines) and experimental (triangles) relative variation of R2 (c) and the relative gauge   16   factor (6d) versus strain. Simulation results by Huang et al. added for reference.19 e) Schematic of the definition of the crystallographic angles used for the calculations in this work. f) Simulation showing that the strain effect on resistivity is independent of the strain angle.   17   Supporting Information Available The supporting information contains Raman and electrical data to demonstrate the presence of graphene in the devices as well as details of the Wheatstone bridge setup. It further contains experiments in various gaseous environments to explain how their influence was eliminated and additional measurements of the piezoresistive effect. The material includes a table with the calibration of the COMSOL model with literature data, a detailed explanation of the resistor model we applied and a table comparing the sensitivity of our devices with literature data. Finally, it contains additional figures with results from the tight-binding model. This material is available free of charge via the Internet at http://pubs.acs.org.   18   References 1.   Bolotin,  K.  I.;  Sikes,  K.;  Jiang,  Z.;  Klima,  M.;  Fudenberg,  G.;  Hone,  J.;  Kim,  P.;  Stormer,  H.,   Ultrahigh  electron  mobility  in  suspended  graphene.  Solid  State  Communications  2008,  146  (9),  351-­‐355.   2.   Morozov,  S.;  Novoselov,  K.;  Katsnelson,  M.;  Schedin,  F.;  Elias,  D.;  Jaszczak,  J.;  Geim,  A.,  Giant   intrinsic  carrier  mobilities  in  graphene  and  its  bilayer.  Physical  Review  Letters  2008,  100  (1),  16602.   3.   Lee,  C.;  Wei,  X.;  Kysar,  J.  W.;  Hone,  J.,  Measurement  of  the  elastic  properties  and  intrinsic   strength  of  monolayer  graphene.  Science  2008,  321  (5887),  385-­‐388.   4.   Lee,  G.;  Cooper,  R.  C.;  An,  S.;  Lee,  S.;  van  der  Zande,  A.;  Petrone,  N.;  Hammerberg,  A.  G.;  Lee,  C.;   Crawford,  B.;  Oliver,  W.;  Kysar,  J.  W.;  Hone,  J.;  High-­‐Strength  Chemical-­‐Vapor–Deposited  Graphene  and   Grain  Boundaries.  Science  2013,  340  (6136),  1073-­‐1076.   5.   Tomori,  H.;  Kanda,  A.;  Goto,  H.;  Ootuka,  Y.;  Tsukagoshi,  K.;  Moriyama,  S.;  Watanabe,  E.;  Tsuya,   D.,  Introducing  Nonuniform  Strain  to  Graphene  Using  Dielectric  Nanopillars.  Appl.  Phys.  Express  4,   075102    2011.   6.   Koenig,  S.  P.;  Boddeti,  N.  G.;  Dunn,  M.  L.;  Bunch,  J.  S.,  Ultrastrong  adhesion  of  graphene   membranes.  Nature  nanotechnology  2011,  6  (9),  543-­‐546.   7.   Bunch,  J.  S.;  Verbridge,  S.  S.;  Alden,  J.  S.;  Van  Der  Zande,  A.  M.;  Parpia,  J.  M.;  Craighead,  H.  G.;   McEuen,  P.  L.,  Impermeable  atomic  membranes  from  graphene  sheets.  Nano  letters  2008,  8  (8),  2458-­‐ 2462.   8.   Pereira,  V.  M.;  Neto,  A.  C.;  Peres,  N.,  Tight-­‐binding  approach  to  uniaxial  strain  in  graphene.   Physical  Review  2009,  80  (4),  045401.   Gong,  S.-­‐C.;  Lee,  C.,  Analytical  solutions  of  sensitivity  for  pressure  microsensors.  Sensors  Journal,   9.   IEEE  2001,  1  (4),  340-­‐344.   Hosseinzadegan,  H.;  Todd,  C.;  Lal,  A.;  Pandey,  M.;  Levendorf,  M.;  Park,  J.  In  Graphene  has  ultra   10.   high  piezoresistive  gauge  factor,  Micro  Electro  Mechanical  Systems  (MEMS),  IEEE  25th  International   Conference  2012,  611-­‐614.   11.   Kim,  K.  S.;  Zhao,  Y.;  Jang,  H.;  Lee,  S.  Y.;  Kim,  J.  M.;  Kim,  K.  S.;  Ahn,  J.-­‐H.;  Kim,  P.;  Choi,  J.-­‐Y.;  Hong,   B.  H.,  Large-­‐scale  pattern  growth  of  graphene  films  for  stretchable  transparent  electrodes.  Nature  2009,   457  (7230),  706-­‐710.   12.   Melvås,  P.  Ultraminiaturized  Pressure  Sensor  for  Catheter  Based  Applications.  Ph.D.  thesis,  KTH   Royal  Institute  of  Technology,  2002.   13.   Hierold,  C.;  Jungen,  A.;  Stampfer,  C.;  Helbling,  T.,  Nano  electromechanical  sensors  based  on   carbon  nanotubes.  Sensors  and  Actuators  A:  Physical  2007,  136  (1),  51-­‐61.   Fung,  C.  K.;  Zhang,  M.  Q.;  Chan,  R.  H.;  Li,  W.  J.  In  A  PMMA-­‐based  micro  pressure  sensor  chip   14.   using  carbon  nanotubes  as  sensing  elements,  Micro  Electro  Mechanical  Systems  2005.  MEMS  18th  IEEE   International  Conference,  251-­‐254.   Christel,  L.;  Petersen,  K.  In  A  catheter  pressure  sensor  with  side  vent  using  multiple  silicon  fusion   15.   bonding,  Proc.  Int.  Conf.  Solid-­‐State  Sensors  and  Actuators  (Trandsucers)  1993;  620-­‐623.   16.   Kalvesten,  E.;  Smith,  L.;  Tenerz,  L.;  Stemme,  G.  The  first  surface  micromachined  pressure  sensor   for  cardiovascular  pressure  measurements,  Micro  Electro  Mechanical  Systems  1998.  MEMS  98.   Proceedings.,  The  Eleventh  Annual  International  Workshop,  574-­‐579.   17.   Clark,  S.  K.;  Wise,  K.  D.;  Wise  K.  D.,  Pressure  sensitivity  in  anisotropically  etched  thin  diaphragm   pressure  sensors.  IEEE  Transactions  on  Electron  Devices,  26  (12):1887–1896,  1979.   Timoshenko,  S.;  Woinowsky-­‐Krieger,  S.,  Theory  of  Plates  and  Shells.  McGraw-­‐Hill  Book  Company,   18.   1959.       19   19.   Huang,  M.;  Pascal,  T.  A.;  Kim,  H.;  Goddard  III,  W.  A.;  Greer,  J.  R.,  Electronic-­‐mechanical  coupling   in  graphene  from  in  situ  nanoindentation  experiments  and  multiscale  atomistic  simulations.  Nano  letters   2011,  11  (3),  1241-­‐1246.   20.   Chen,  X.;  Zheng,  X.;  Kim,  J.-­‐K.;  Li,  X.;  Lee,  D.-­‐W.,  Investigation  of  graphene  piezoresistors  for  use   as  strain  gauge  sensors.  Journal  of  Vacuum  Science  &  Technology  B:  Microelectronics  and  Nanometer   Structures  2011,  29  (6),  06FE01-­‐06FE01-­‐5.   21.   Paussa,  A.;  Esseni,  D.,  An  exact  solution  of  the  linearized  Boltzmann  transport  equation  and  its   application  to  mobility  calculations  in  graphene  bilayers.  Journal  of  Applied  Physics  2013,  113,  093702.   22.   Bresciani,  M.;  Paussa,  A.;  Palestri,  P.;  Esseni,  D.;  Selmi,  L.  Low-­‐field  mobility  and  high-­‐field  drift   velocity  in  graphene  nanoribbons  and  graphene  bilayers,  Electron  Devices  Meeting  (IEDM),  2010  IEEE   International  2010,  32.1.  1-­‐32.1.  4.   23.   Iyechika,  Y.,  Application  of  Graphene  to  High-­‐Speed  Transistors:  Expectations  and  Challenges.   Science  and  technology  trends  2010,  37,  76-­‐92.   24.   Reina,  A.;  Son,  H.;  Jiao,  L.;  Fan,  B.;  Dresselhaus,  M.  S.;  Liu,  Z.;  Kong,  J.,  Transferring  and   identification  of  single-­‐and  few-­‐layer  graphene  on  arbitrary  substrates.  The  Journal  of  Physical  Chemistry   C  2008,  112  (46),  17741-­‐17744.   25.   Li,  X.;  Zhu,  Y.;  Cai,  W.;  Borysiak,  M.;  Han,  B.;  Chen,  D.;  Piner,  R.  D.;  Colombo,  L.;  Ruoff,  R.  S.,   Transfer  of  large-­‐area  graphene  films  for  high-­‐performance  transparent  conductive  electrodes.  Nano   letters  2009,  9  (12),  4359-­‐4363.   26.   Jiao,  L.;  Fan,  B.;  Xian,  X.;  Wu,  Z.;  Zhang,  J.;  Liu,  Z.,  Creation  of  nanostructures  with  poly  (methyl   methacrylate)-­‐mediated  nanotransfer  printing.  Journal  of  the  American  Chemical  Society  2008,  130  (38),   12612-­‐12613.   27.   Park,  H.  J.;  Meyer,  J.;  Roth,  S.;  Skákalová,  V.,  Growth  and  properties  of  few-­‐layer  graphene   prepared  by  chemical  vapor  deposition.  Carbon  2010,  48  (4),  1088-­‐1094.   28.   Lin,  Y.-­‐C.;  Jin,  C.;  Lee,  J.-­‐C.;  Jen,  S.-­‐F.;  Suenaga,  K.;  Chiu,  P.-­‐W.,  Clean  transfer  of  graphene  for   isolation  and  suspension.  ACS  Nano  2011,  2362-­‐2368.               20   

1307.1159

4

1307

Tunability of microwave transitions as a signature of coherent parity mixing effects in the Majorana-Transmon qubit

Coupling Majorana fermion excitations to coherent external fields is an important stage towards their manipulation and detection. We analyse the charge and transmon regimes of a topological nano-wire embedded within a Cooper-Pair-Box, where the superconducting phase difference is coupled to the zero energy parity states that arise from Majorana quasi-particles. We show that at special gate bias points, the photon-qubit coupling can be switched off via quantum interference, and in other points it is exponentially dependent on the control parameter $E_J/E_C$. As well as a probe for topological-superconductor excitations, we propose that this type of device could be used to realise a tunable high coherence four-level system in the superconducting circuits architecture.

cond-mat.mes-hall

cond-mat

Tunability of microwave transitions as a signature of coherent parity mixing effects in the Majorana-Transmon qubit Eran Ginossar1 and Eytan Grosfeld2 1Advanced Technology Institute and Department of Physics, University of Surrey, Guildford, GU2 7XH, United Kingdom 2Department of Physics, Ben-Gurion University of the Negev, Be’er-Sheva 84105, Israel (Dated: May 7, 2014) Coupling Majorana fermion excitations to coherent external fields is an important stage towards their manipulation and detection. We analyse the charge and transmon regimes of a topological nano-wire embedded within a Cooper-Pair-Box, where the superconducting phase difference is cou- pled to the zero energy parity states that arise from Majorana quasi-particles. We show that at special gate bias points, the photon-qubit coupling can be switched off via quantum interference, and in other points it is exponentially dependent on the control parameter EJ /EC . As well as a probe for topological-superconductor excitations, we propose that this type of device could be used to realise a tunable high coherence four-level system in the superconducting circuits architecture. Introduction.— Photons are used to control and mea- sure qubits in a diverse range of qubit systems from natural atoms to semiconducting and superconducting solid-state architectures [1–8]. The properties of solid- state qubits and resonators and coupling to photons can be engineered to some degree by design and this en- ables the combination of several different subsystems to form hybrid devices that take advantage of their relative strengths [9, 10]. In turn, this leads to new methods of probing physical systems and, where highly quantum co- herent subsystems are involved, to establishing control over their quantum variables. This approach is of particular interest in the con- text of topological Josephson junctions, where theo- retical predictions [11–13] supported by experimental progress [14, 15] indicate the presence of the highly sought-after Majorana zero-energy modes [16, 17] local- ized on the wire around its endpoints. To probe these excitations with microwave photons requires a hybrid de- vice formed by bridging a Josephson junction using a suitably prepared nanowire. The minimal quantum de- scription of the device involves an effective parity fermion degree of freedom, that arises from the Majorana quasi- particles on the wire, and represents the parity of the total charge. This fermion is coupled to the phase and charge degrees of freedom of the superconducting circuit. Anticipating experimental progress in these directions, a central question that needs to be addressed is what is the unique spectroscopic signature of the charge-neutral Majorana zero modes in such devices? Also, can one es- tablish control over the parity fermion, demonstrating its relevance for quantum information processing? To answer these questions, in this Letter we explore the properties of a Majorana-Cooper-Pair-Box parity-charge qubit hybrid in the charge [18–23] and transmon [24] regimes. We find two key differences when compared with the traditional Cooper-Pair-Box. First, a remark- able spectroscopic pattern consisting of two to four res- onant frequencies whose number and intensities are tun- able via a gate offset ng (see Fig. 1); the pattern is very different from the traditional transmon that admits ex- actly two resonant frequencies with non-tunable intensi- ties. Second, an excitation spectrum consisting of multi- ple doublets inherited from the fermion parity states (see Fig. 2c), whose lowest four levels can be used to form a double-Λ system with a highly coherent almost degener- ate doublet ground state. In the rest of this Letter we FIG. 1. (Color online) Spectroscopy of a Majorana-Transmon qubit device. The microwave transition frequencies (∆f ) of the system as function of the offset charge ng for the different transition paths within the lower part of the spectrum (see Fig. 2b). The upper (A) and lower (C) lines are associated with the usual transmon frequencies while the central lines (B) describe parity mixing effects and are a unique feature of the Majorana modes. At ng = 1/4 the transition amplitudes in each parity manifold interfere destructively and this gives rise to a ‘spectral hole’ in transition (A,C). In contrast to the traditional transmon, all spectral lines are gate tunable, and the pattern can be shifted between 4 (left) and 2 (center) res- onant frequencies. The frequencies are plotted relative to the average and in units of the line width κ/2π = 50KHz, taking EC /2π = 400MHz, EJ /EC = 27 and EM /2π = 0.5MHz. 2 tion of a Cooper-Pair-Box (CPB) (see Fig. 2), the latter being a prototypical charge qubit (realized as either a 2D or 3D transmon). A combination of strong spin-orbit coupling and a Zeeman gap can be used to push the wire into its topological state, provided that the chemical po- tential is tuned within the wire’s gap. In this phase, Majorana zero modes with operators γ2 and γ3 are lo- calized near the junction. Two additional distant Majo- rana modes are present within the superconductors, γ1 † i = γi and and γ4 (see Fig. 2a). The operators satisfy γ {γi, γj} = δij. The properties of the device are deter- mined by the interplay of the Josephson coupling EJ , the charging energy EC, the coupling EM between the Majorana excitations and the superconducting phase dif- ference ϕ [21, 23, 25–29]. The Hamiltonian for the hybrid junction is H = H0 + HM where ∂ϕ − ng (cid:19)2 − EJ cos ϕ, H0[ng] = 4EC (cid:18) 1 i HM = iEM γ2γ3 cos(ϕ/2). (1) Here ng describes the total offset charge that represents an external electrostatic gate control. The anomalous term HM is generated by coherent single-electron tun- nelling processes between the two superconductors fa- cilitated by the presence of the zero-energy Majorana modes. Its presence was predicted to lead to observ- able effects [22, 25, 26, 30–34], of which some experi- mental signatures were recently observed [35]. In our context, it allows coherent Rabi oscillations of unpaired electrons. We focus on the Majorana-Transmon (MT) regime, which we define as EM (cid:28) EC (cid:28) EJ , where the influence of charge noise is exponentially suppressed, and explore the dependence of the eigen-states and eigen- energies spectrum on ng. When EM is non-zero, we find a spectrum that is composed of closely spaced doublets of transmon-like energy levels with a periodicity of e com- pared to 2e for the transmon. The microwave photons couple to the charge operator and we find that even when EM /EC (cid:28) 1 the charge matrix element (cid:104)inj(cid:105) that cou- ples to the photon field is strongly dependent on ng and EJ /EC for the allowed GHz-range transitions since the bare states of the system become strongly hybridized. For completeness, we would like to mention other sce- narios that were recently discussed in the literature: the EM = 0 and EM = EC = 0 limits in [36, 37]; the flux qubit in [38]; circuit QED extensions in [39, 40]; and, photon-induced long-distance Majorana coupling in [41]. In contrast, the present work offers an effective model capturing the multi-level spectrum of the charge-qubit device. Due to a particular set of useful features that we now elaborate on, this minimal scenario may also have considerable appeal as an alternative to transport based experiments. Solution of the effective model.— In the topological phase, the nano-wire excitations carry a single zero en- ergy fermion state on either side of the junction, whose FIG. 2. (Color online) The experimental setup considered in the text (a). Two superconductors (grey) form a Josephson junction and are capacitively coupled to a gate (orange). A topological nanowire lying in close proximity (yellow) carries four Majorana zero modes. Two of the zero modes hybridize (dashed line) owing to single electron tunneling processes across the junction. The eigenstates, denoted 0/1;±(cid:105) are coherent superpositions of the opposite fermion parity trans- mon states. Panel (b) shows the tunable microwave couplings scheme between the eigenstates, realising a double-Λ system. The various optical transition strengths can be controlled by varying either EJ or ng. (c) Representative probability den- sities of the two parity doublets (±) are plotted against the background of the dominant Josephson energy potential, for EC /h = 0.4 GHz, EJ /EC = 25, EM /EC = 5 × 10−4, ng = 0.25. The doublet energy splittings are drawn exaggerated for visibility in panels (b,c). shall refer to this ground state doublet as the “Majorana- Transmon qubit”. The combination of these properties allows: (i) Spectroscopic detection of the parity fermion strengthening currently available transport signatures of Majorana fermions in a setting that avoids complications stemming from the need to attach leads; and, (ii), co- herent and tunable control over the parity fermion made possible by the highly anharmonic level structure and the tunability of transitions. As we now explain, the coupling strengths for the different transitions are determined by a quantum interference effect that arises from the presence of the coherently mixed parity fermions and reveal their presence. In particular, the parity-conserving-like transi- tions (green), see Fig. 2b, can be tuned to exactly zero by the gate offset ng or exponentially suppressed by varying the ratio EJ /EC (Fig. 3b-d). Such tunable control allows state manipulation of the Majorana-Transmon qubit via the upper levels. These properties are highly desirable for quantum information processing in the superconduct- ing qubit architecture which is currently based on a very weakly anharmonic level structure that limits the speed and fidelity of gate operations. Proposed setup and description of the effective model.— The proposed hybrid device contains a nano- wire which is placed in proximity to the Josephson junc- occupation becomes locked to the parity of the fermion number on the same side [21, 23, 42]. By projecting on the states labelled by these fermion numbers, and focus- ing on a single total parity state, the Hamiltonian ac- quires a matrix structure (cid:18) H0[ng] H = EM cos(ϕ/2) H0[ng] EM cos(ϕ/2) , (2) (cid:19) which we now diagonalize by solving the eigenvalue equa- tion Hχ = Eχ where χ = (f (ϕ), g(ϕ))T . A crucial point is that one should take into account the requirement on the Hilbert space that f (ϕ) is periodic in ϕ with a pe- riodicity 2π, while g(ϕ) is anti-periodic. Alternatively one can use a basis composed of solely periodic func- tions, but the Hamiltonian gets modified according to H → H(cid:48) = U HU†, with U = diag{1, eiϕ/2} 2 (1 + e−iϕ) EM 2 (1 + eiϕ) H0[ng + 1/2] (cid:18) H0[ng] H(cid:48) = (cid:19) (3) EM . The eigen-energies of this Hamiltonian were calculated numerically employing charge eigenstates and are pre- sented in Fig. S1 as function of ng. Further insight into the effect of the coupling HM on the system can be gained from diagonalising the Hamil- tonian in the basis of the transmon eigenfunctions [24] which will be denoted as Ψk(ng, ϕ) and which diago- nalise H0[ng] and H0[ng + 1/2], respectively. In this basis H(cid:48) can be shown to be approximately 2 × 2 block- diagonal since HM mostly couples the same band pairs {Ψk(ng, ϕ), Ψk(ng + 1/2, ϕ)}. This coupling results in an energy splitting of the order EM developing around ng = 1/4 (see Fig. S1b). Importantly, the periodicity of the energy levels is halved, and this will affect the be- havior of the system with respect to tunneling of non- equilibrium quasi-particles. Further, in the transmon regime EJ /EC (cid:29) 1 the dispersion of the transmon levels is exponentially suppressed such that the EM dominates the spectral gap and flattens it further, see Fig. S1(c-d), and hence improving its famous resilience against dephas- ing caused by charge fluctuations. Light-matter interaction.— Electromagnetic fields in- fluence the dynamics of the MT by coupling via the dipole operator, given by D = ide∂ϕ, where d is the distance be- tween the two superconductors. In the transformed basis D(cid:48) = U DU† the dipole operator acquires the form (cid:18) i∂ϕ D(cid:48) = ed (cid:19) i∂ϕ + 1 2 . (4) The parity-phase correlation has important consequences for the microwave and RF coupling between the states. The matrix elements of the dipole operator Gk,π→k(cid:48),π(cid:48)(ng) = (ed)−1 dϕ(cid:104)k, πD(cid:48)k(cid:48), π(cid:48)(cid:105) = (cid:90) 2π 0 k,πfk(cid:48),π(cid:48)Ψ∗ k(ng; ϕ)i∂ϕΨk(cid:48)(ng; ϕ) k(ng + 1/2; ϕ)(i∂ϕ + 1/2)Ψk(cid:48)(ng + 1/2; ϕ)(cid:3) (5) dϕ(cid:2)f∗ (cid:90) 2π k,πgk(cid:48),π(cid:48)Ψ∗ +g∗ 0 3 where π, π(cid:48) = ±, yields the coherent sum of two ma- trix elements resulting from the different fermion parities. Thus, the transition amplitude between states of same or different π, π(cid:48) is given by the coherent addition of the ma- trix elements (see Fig. 2b). Usually, the integrals need to be evaluated numerically, but some observations can be made on general grounds. At the degeneracy point ng = 1/4 the parity-spinor amplitudes have equal weight of uncoupled MT components and the matrix elements for the transitions 0; π(cid:105) → 1; π(cid:48)(cid:105) become G0,π→1,π(cid:48)(ng = 1/4) = i 2 0 Ψ∗ 0(ng = 3/4, ϕ)∂ϕΨ1(ng = 3/4, ϕ)] dϕ [Ψ∗ 0(ng = 1/4, ϕ)∂ϕΨ1(ng = 1/4, ϕ)± (6) where the relative sign ± depends on the choice of pair of states. At this degeneracy point, for the transition corre- sponding to (−), a full destructive interference between the dipole matrix elements [43] takes place. As we move away from the special point ng = 1/4 the amplitudes (cid:90) 2π FIG. 3. (Color online) (a) The fermion tunneling term HM leads to a correlation between the superconducting phase dif- ference (ϕ) and the parity-spinor. The x − y projection of the Bloch vector representation of the parity-spinor is plot- ted as a function of ϕ at the point of maximal hybridiza- tion ng = 1/4. (b) Shows the dependence of the opti- cal transition strength Gk,π→k,π(cid:48) as function of EJ /EC with EC /h = 0.4 GHz, ng = 0.1 and EM /EC = 1/400, and the logarithm of this dependence (inset, same x-axis) shows that the suppression is exponential. (c) Shows the dependence of the dipole on the effective gate charge ng for EJ /EC = 40 with EC /h = 0.4 GHz, and for EJ /EC = 20 (d). Three curves are drawn in each panel (c,d) to show the dependence on EM /EC = 2.5 × 10−3(dotted), 2.5 × 10−4(dashed), 2.5 × 10−5(solid). Strong parity-phase correlations arise near the avoided crossing point ng = 1/4 which results in a vanish- ing of the coupling between the 0 → 1;±(cid:105) states. Deeper in the transmon regime (EJ /EC = 40) these correlations per- sist across the whole range of ng even for small values of the Majorana coupling EM which suppresses the optical coupling matrix element, see dotted line in (c). change and the dipole coupling returns to a finite value on a scale that depends on EM , see Fig. 3. In contrast, the ‘intra-band’ dipole coupling Gk,π→k,π(cid:48) can be shown to be relatively small for all value of ng. As ng is varied it should be possible to see the different transitions lines of Fig. 2(b) appear and disappear as their transition rates vary. This spectroscopic pattern depends on the Majo- rana coupling EM and the latter may be estimated from it. The sensitivity of the dipole coupling to parameters also extends to an exponential dependence on the ratio of EJ /EC since the latter controls the relative effective- ness of the hybridization due to HM . A strong on-off control of the coupling, see Fig. 3(b) could be achieved with a split-CPB, albeit here with a qualitatively differ- ent behavior from a recent implementation that is based on circuit QED architecture [44]. Qubit realization and control.— A device that realises this Hamiltonian could have substantial advantages for quantum information processing. In addition to the gate tunability of the direct microwave coupling between transmon qubit states, the two lowest MT states form a parity-energy doublet, see Fig. 2(b), which constitutes a highly anharmonic MT qubit. The high anharmonicity is considered a benefit for qubit state manipulation as it reduces the probability of higher states being excited and shortens the duration of control pulses. The two almost degenerate ground states can be regarded as part of a Λ type system and thus optically manipulated via transi- tions involving the higher doublet states. The doublet states can be described as spinors in the odd-even parity space k,±(cid:105) = (fk,±, gk,±)T where k = 0, 1 denotes the transmon band index and ± denotes symmetric and anti- symmetric combinations. Due to the interaction HM the effective parity and phase (ϕ) degrees of freedom become correlated, showing equal weight hybridization close to the crossing point ng = 1/4, i.e. f2−g2 = 0 ((cid:104)σz(cid:105) = 0). Close to this point the resulting spinor nature of the wave function can be depicted by taking the partial trace in the pseudospin space ((cid:104)σx(cid:105),(cid:104)σy(cid:105),(cid:104)σz(cid:105))ϕ,± = Tr(ρk,±(ϕ)(cid:126)σ), see Fig. 3(a). The ϕ-dependent spinor lies on the equator of the Bloch sphere due to the equal weighting. It points predominantly in the σx direction since the parity of the transmon wave functions with respect to ϕ yields real ma- trix elements for HM . The sensitivity of the states to the parameter ng leads to adiabatic control, which by tuning ng from 0 to 1/2 will take a system prepared for example in a state pointing predominantly in the k; 0, 0(cid:105) direc- tion to a state close to k; 1, 1(cid:105). Microscopically, this in- volves splitting a Cooper-pair adiabatically between the two sides of the junction. For intermediate values the system maintains a coherent superposition of these two parity states. Alternative scenarios.— Microscopic two level systems (TLS) that interact with the charge qubit generally give rise to spurious spectroscopic signatures. However, these can be differentiated from the case of Majorana fermions. 4 In the case of a TLS that is resonantly coupled to the transmon there is only one ground state and therefore only two spectroscopic transitions to the one-excitation manifold. Correspondingly, there are only two transi- tion frequencies that can be observed compared to up to four in the Majorana case. In contrast, if the TLS is close to degenerate in energy, the full system may pos- sess a similar spectrum of two doublets. However, the crucial difference arises from the absence of direct cou- pling between the two quasi-degenerate ground states, in contrast to the case of Majorana fermions. This strongly suppresses any additional features in the spectroscopic pattern. Another possible scenario is a single electron state trapped within the oxide barrier, with a typical en- ergy E0, that can exchange electrons with the two super- conductors. Again, the phenomenology is very different since there will always be an extra energetic cost associ- ated with an unpaired electron within the superconduc- tors, leading to a very high energy scale E0 + ∆ in which such processes appear in the spectroscopic pattern (here ∆ is the superconductor gap). Generally speaking, in- dependently of the competing microscopic scenario, the Majorana fermions always satisfy: (i) the periodicity of the spectrum is coherently halved compared to the reg- ular transmon owing to a rigidity in the relative shift ng → ng + 1/2 of the two levels that hybridize to form the doublet; and, (ii), the spectroscopic signatures do not depend on the particular realization of randomly occur- ring TLSs in the oxide. To conclude, we have presented an analysis of a Majorana-Transmon qubit hamiltonian and discussed its unique spectroscopic signatures. It was shown that since the parity and transmon states are entangled, the dipole transitions become strongly dependent on the system pa- rameters ng and EJ /EC. The energy scale EM control- ling the Majorana coupling can be extracted from the spectroscopic pattern as discussed above. Such spectro- scopic signatures of the Majorana fermions can be differ- entiated from dominant impurity scenarios in the oxide barrier. We believe that these results will play an im- portant role in the analysis of experiments which involve the embedding of topological superconductors within the superconducting qubit architecture. In addition, the set of optical selection rules that we find support the realisa- tion of a dual-Λ system (see Fig. 2) with dipole allowed transitions to the higher transmon-like states of the sys- tem which can be externally controlled. This enables the preparation and manipulation of quantum states using transitions through the upper levels. We acknowledge useful discussions with L. Di- Carlo. E. Ginossar acknowleges support from EPSRC (EP/I026231/1). E. Grosfeld acknowledges support from the Israel Science Foundation (Grant No. 401/12) and the European Union’s Seventh Framework Programme (FP7/2007-2013) under Grant No. 303742. Both au- thors acknowledge support from the Royal Society In- 5 plex Systems 37, 349 (2004). [26] L. Fu and C. L. Kane, Phys. Rev. B 79, 161408 (2009). [27] E. Grosfeld and A. Stern, PNAS 108, 11810 (2011). [28] C. Mora and K. Le Hur, ArXiv e-prints (2012), 1212.0650. [29] P. Dutt, T. L. Schmidt, C. Mora, and K. Le Hur, ArXiv e-prints (2013), 1301.3434. [30] J. Nilsson, A. R. Akhmerov, and C. W. J. Beenakker, Phys. Rev. Lett. 101, 120403 (2008). [31] Y. Tanaka, T. Yokoyama, and N. Nagaosa, Phys. Rev. Lett. 103, 107002 (2009). [32] L. Jiang, D. Pekker, J. Alicea, G. Refael, Y. Oreg, and F. von Oppen, Phys. Rev. Lett. 107, 236401 (2011). [33] K. T. Law and P. A. Lee, Phys. Rev. B 84, 081304 (2011). [34] P. A. Ioselevich and M. V. Feigelman, Phys. Rev. Lett. 106, 077003 (2011). [35] L. P. Rokhinson, X. Liu, and J. K. Furdyna, Nature Physics 8, 795 (2012), 1204.4212. [36] F. Hassler, A. R. Akhmerov, and C. W. J. Beenakker, New Journal of Physics 13, 095004 (2011). [37] P. Bonderson and R. M. Lutchyn, Phys. Rev. Lett. 106, 130505 (2011). [38] D. Pekker, C.-Y. Hou, V. Manucharyan, and E. Demler, ArXiv e-prints (2013), 1301.3161. [39] C. Muller, J. Bourassa, and A. Blais, arXiv:1306.1539 (2013). [40] A. Cottet, T. Kontos, and B. Dou¸cot, Phys. Rev B 88, 195415 (2013). [41] T. L. Schmidt, A. Nunnenkamp, and C. Bruder, Phys. Rev. Lett. 110, 107006 (2013), 1211.2201. [42] A. Zazunov, A. L. Yeyati, and R. Egger, Phys. Rev. B 84, 165440 (2011), 1108.4308. [43] At this point it can be shown that Ψk(ng = 3/4; ϕ) = eiϕΨ∗ k(ng = 1/4; ϕ) which leads to a partial cancellation of terms in Eq. 6 for the transition corresponding to (−) and the remaining terms are anti-symmetric. [44] J. M. Gambetta, A. A. Houck, and A. Blais, Physical Review Letter 106, 030502 (2011). ternational Exchanges program. [1] T. Pellizzari, S. A. Gardiner, J. I. Cirac, and P. Zoller, Phys. Rev. Lett. 75, 3788 (1995). [2] A. Rauschenbeutel, G. Nogues, S. Osnaghi, P. Bertet, M. Brune, J. Raimond, and S. Haroche, Phys. Rev. Lett. 83, 5166 (1999). [3] A. Imamoglu, D. D. Awschalom, G. Burkard, D. P. Di- Vincenzo, D. Loss, M. Sherwin, and A. Small, Phys. Rev. Lett. 83, 4204 (1999). [4] D. Jaksch, J. I. Cirac, P. Zoller, S. L. Rolston, R. Cot´e, and M. D. Lukin, Phys. Rev. Lett. 85, 2208 (2000). [5] J. M. Raimond, M. Brune, and S. Haroche, Rev. of Mod. Phys. 73, 565 (2001). [6] A. Wallraff, D. I. Schuster, A. Blais, L. Frunzio, R. S. Huang, J. Majer, S. Kumar, S. M. Girvin, and R. J. Schoelkopf, Nature 431, 162 (2004). [7] E. Grosfeld, N. R. Cooper, A. Stern, and R. Ilan, Phys. Rev. B 76, 104516 (2007). [8] A. A. Houck, J. A. Schreier, B. R. Johnson, J. M. Chow, J. Koch, J. Gambetta, D. I. Schuster, L. Frunzio, M. H. Devoret, S. M. Girvin, et al., Phys. Rev. Lett. 101, 080502 (2008). [9] Y. Kubo, F. R. Ong, P. Bertet, D. Vion, V. Jacques, D. Zheng, A. Drau, J.-F. Roch, A. Auffeves, F. Jelezko, et al., Phys. Rev. Lett. 105, 140502 (2010). [10] D. I. Schuster, A. P. Sears, E. Ginossar, L. DiCarlo, L. Frunzio, J. J. L. Morton, H. Wu, G. A. D. Briggs, D. D. Awschalom, and R. J. Schoelkopf, Phys. Rev. Lett. 105, 140501 (2010). [11] A. Y. Kitaev, Physics-Uspekhi 44, 131 (2007). [12] R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. 105, 077001 (2010). [13] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105, 177002 (2010). [14] V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Science 336, 1003 (2012). [15] A. Das, Y. Ronen, Y. Most, Y. Oreg, M. Heiblum, and H. Shtrikman, Nature Physics 8, 887 (2012). [16] N. Read and D. Green, Physical Review B 61, 10267 (2000). [17] G. Moore and N. Read, Nuclear Physics B 360, 362 (1991). [18] A. Stern and B. I. Halperin, Phys. Rev. Lett. 96, 016802 (2006). [19] R. Ilan, E. Grosfeld, and A. Stern, Phys. Rev. Lett. 100, 086803 (2008). [20] R. Ilan, E. Grosfeld, K. Schoutens, and A. Stern, Phys. Rev. B 79, 245305 (2009). [21] L. Fu, Phys. Rev. Lett. 104, 056402 (2010). [22] B. van Heck, F. Hassler, A. R. Akhmerov, and C. W. J. Beenakker, Phys. Rev. B 84, 180502 (2011). [23] A. Golub and E. Grosfeld, Phys. Rev. B 86, 241105 (2012), 1206.0958. [24] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Phys. Rev. A 76, 042319 (2007). [25] H. J. Kwon, K. Sengupta, and V. M. Yakovenko, The Eu- ropean Physical Journal B-Condensed Matter and Com- SUPPLEMENTARY MATERIAL 2 2 (γ1 − iγ2), Γ2 = 1√ The mesoscopic nature of the Majorana-CPB pre- scribes certain intricacies in the diagonalization proce- dure which we now elaborate on. It is a common prac- tice to compose non-local Dirac fermion zero modes (γ3 − iγ4) and similarly Γ1 = 1√ † † 2 through hermitian conjugation. This set of oper- Γ 1, Γ ators satisfies the canonical anti-commutation relations † j} = δij. In addition we introduce a for fermions, {Γi, Γ number operator for each superconductor which we de- note by ni (i = 1, 2), which counts the number of Cooper pairs in units of 1, and which is conjugate to the super- conducting phase ϕi, [ni, ϕi] = −i. Importantly, in topo- logical superconductors ni can assume both integer and half-integer eigenvalues. The latter are associated with the presence of an unpaired electron, counted as half a Cooper pair. 6 be matched with the Majorana zero mode operator which also changes sign, leading to admissible operators (the spectrum is necessarily 2π periodic although some parity constrained states may exhibit 4π periodicity [11]). The single electron tunneling term is thus written in terms of the relative phase and the Majorana zero modes as † 2 + Γ2Γ1 in Eq. (1), with 2iγ2γ3 = Γ fully shuffling the state of the zero mode occupations: N1, N2(cid:105) ↔ ¯N1, ¯N2(cid:105). In addition, the phase dependent part of the electron tunneling operator ensures that the eigenvalue of n changes by ±1/2. Hence states of the form (S1) can only couple to states of the form (S2). By projecting on the orthogonal states N1, N2(cid:105) and ¯N1, ¯N2(cid:105) the Hamiltonian acquires a matrix structure, see Eq. 2, or in the alternative basis discussed in the main text, Eq. 3. † 1Γ2 + Γ † 2Γ1 + Γ † 1Γ We initialize the system by considering, in the absence of tunneling across the junction, a definite parity of the electron number within each superconductor. The occu- † pation of the Dirac zero mode Ni = Γ i Γi is thus set by the parity of the electron number, Ni = 2ni(mod 2). Let us denote the initial state of the two Dirac zero modes within one doublet by N1, N2(cid:105). For simplicity we also assume that initially n1 = n2 (different choices lead to slightly different formulations but the physical results re- main the same up to an overall shift of the gate charge). Next we turn on the couplings EM and EJ associated with single-electron tunneling and Cooper-pair tunneling respectively. We construct basis states for the even and odd parity sectors and relative number of Cooper pairs n = 1 2 (n1 − n2)(cid:8)eiϕnN1, N2(cid:105), n ∈ Z(cid:9) , (cid:26) (cid:27) (S1) (S2) , eiϕn ¯N1, ¯N2(cid:105), n ∈ Z + 1 2 where ¯Ni = 1− Ni and ϕ = ϕ1 − ϕ2 is the relative phase satisfying [n, ϕ] = −i. The first (second) set of wave functions is periodic (anti-periodic) under a change of ϕ by 2π. The Cooper-pair tunneling operator modifies the eigenvalue of n by ±1, hence only couples states inter- nally within the two subspaces (S1) and (S2). In con- trast, the single-electron tunneling operator intermixes the two subspaces. To model the latter we introduce electron raising and lowering operators e±iϕi/2 satifying [ni, e±iϕi/2] = ± 1 2 e±iϕi/2. These operators are double- valued as they change sign when ϕi → ϕi+2π and need to FIG. S1. (Color online) The different effects of the Majorana interaction on the spectrum. (a) Spectrum of the Majorana- Cooper-Pair-Box spectrum with EJ /h = 10 GHz, EC /h = 0.4 GHz for EJ /EC = 25 and EM /EC = 1/400 as a func- tion of the charge offset ng. An avoided crossing develops (b) between the states of different fermion parity and the period- icity of the spectrum is halved, to repeat when ng → ng +1/2. The ground state pair is shown for EM /EC = 1/1400 (black), EM /EC = 1/140 (blue) and EM /EC = 1/70 (green) for EJ /EC =7.14 (with EC /h = 1.4 GHz). (c) For EJ /EC = 25 the parity states for EM = 0 are crossing but exponen- tially close in energy with a dispersion of
0+(EM = 0)/h = 2 × 10−5 GHz (see inset graph of the gray dashed area, with including the Majorana coupling same x-axis). However, EM > 0 effectively removes the degeneracy and determines the energy splitting in the whole range of ng. (d) In the transmon regime the residual dispersion is further suppressed by the Majorana interaction HM , as the plot of dispersion ratio
1+(EM )/
1+(0) shows. 

1509.09108

1

1509

Magnetic anisotropy in Shiba bound states across a quantum phase transition

The exchange coupling between magnetic adsorbates and a superconducting substrate leads to Shiba states inside the superconducting energy gap and a Kondo resonance outside the gap. The exchange coupling strength determines whether the quantum many-body ground state is a Kondo singlet or a singlet of the paired superconducting quasiparticles. Here, we use scanning tunneling spectroscopy to identify the different quantum ground states of Manganese phthalocyanine on Pb(111). We observe Shiba states, which are split into triplets by magnetocrystalline anisotropy. Their characteristic spectral weight yields an unambiguous proof of the nature of the quantum ground state.

cond-mat.mes-hall

cond-mat

a Magnetic anisotropy in Shiba bound states across a quantum phase transition Nino Hatter,1 Benjamin W. Heinrich,1 Michael Ruby,1 Jose I. Pascual,1, 2 and Katharina J. Franke1 1Fachbereich Physik, Freie Universitat Berlin, Arnimallee 14, 14195 Berlin, Germany. 2CIC nanoGUNE and Ikerbasque, Basque Foundation for Science, Tolosa Hiribidea 78, Donostia-San Sebastian 20018, Spain. (Dated: September 29, 2018) The exchange coupling between magnetic adsorbates and a superconducting substrate leads to Shiba states inside the superconducting energy gap and a Kondo resonance outside the gap. The exchange coupling strength determines whether the quantum many-body ground state is a Kondo singlet or a singlet of the paired superconducting quasiparticles. Here, we use scanning tunnel- ing spectroscopy to identify the different quantum ground states of Manganese phthalocyanine on Pb(111). We observe Shiba states, which are split into triplets by magnetocrystalline anisotropy. Their characteristic spectral weight yields an unambiguous proof of the nature of the quantum ground state. Magnetic adsorbates on a superconductor create a magnetic scattering potential for the quasi-particles of the superconductor. A single spin gives rise to so-called Yu-Shiba-Rusinov (Shiba) states [1 -- 3]. Recently, it was argued that hybridization of the Shiba states can lead to Shiba bands with nontrivial topological character [4 -- 6]. This is essential for the formation of Majorana modes, which have been detected in ferromagnetic chains of Fe atoms on a Pb(110) surface [5]. If not only one but sev- eral Shiba states are present, the hybridization will lead to a more complex band structure. Different origins of multiple Shiba states are discussed theoretically. These include different angular momentum scattering contribu- tions, individual d orbitals acting as separate scatter- ing potentials, or low-energy excitations due to magnetic anisotropy or vibrations [7 -- 12]. However, experimentally, the origin of multiple Shiba states is often difficult to de- termine [13, 14]. [10, 15 -- 17]. Concomitantly with the formation of a Shiba state, the exchange coupling between the magnetic adsorbate and the substrate also drives the formation of a singlet Kondo state If the Kondo energy scale kBTK is much larger than the superconducting pairing energy ∆, the "Kondo screened" state is the ground state. On the other hand, if kBTK (cid:28) ∆, an unscreened "free spin" state with S > 0 is the ground state. A Shiba resonance as fingerprint of this magnetic interac- tion can be found in the quasiparticle excitation spectra within the superconducting energy gap if kBTK ∼ ∆ [18]. This resonance corresponds to a transient state, in which an electron is added/removed to/from the ground state. Thus, the electron occupation changes and the spin is altered by ∆S = ±1/2. In the "Kondo screened" case, the Shiba resonance is found with a binding energy Eb below the Fermi level EF. The weaker the exchange coupling and Kondo screening, the closer is the Shiba state to the Fermi level. The crossing of the Shiba state through EF marks the quantum phase transition from the "Kondo screened" S = 0 to the "free spin" state on the superconductor, i.e., S > 0. This transition oc- curs at kBTK ∼ 0.3∆ and has been described theoreti- cally [15, 17, 19] and experimentally [16, 20, 21]. Spin-1/2 systems feature one pair of Shiba resonances at ±Eb [18]. If the adsorbate carries a higher spin, mul- tiple Shiba states may appear inside the gap as discussed theoretically by Zitko and co-workers [10]. They argue that such systems may involve multiple Kondo screen- ing channels with different coupling strengths Jk and, hence, multiple Shiba states. Furthermore, a mutual cou- pling of the spins may lead to a splitting of the peaks in the excitation spectra in the presence of magnetic anisotropy. Here, we resolve triplets of Shiba states on single paramagnetic molecules. A multitude of differ- ent adsorption sites provides access to a large range of magnetic coupling strengths with the substrate. We de- tect the splitting of the Shiba states even throughout the quantum phase transition from a "Kondo screened" to a "free spin" ground state. The intensities of the Shiba resonances yield an unambiguous proof of the na- ture of the spin states and their splitting by magnetic anisotropy. The basic understanding of the influence of magnetic anisotropy on many-body interactions is crucial for the design of quantum states with controlled proper- ties. Furthermore, its knowledge may provide interesting approaches for creating and addressing Majorana states in proximity coupled magnetic nanostructures [12]. RESULTS Detection of Shiba states The effects of magnetic anisotropy on Shiba states can be explored experimentally by bringing a metal-organic molecule into contact with a superconductor. The or- ganic ligand is responsible for the splitting of the spin states with S ≥ 1 of the transition metal core [22, 23]. Manganese phthalocyanine (MnPc) has a spin S = 3/2 in gas phase and retains a magnetic moment on metal sur- faces [24 -- 26]. In particular, its magnetic moment inter- 2 asymmetric in intensity [Fig. 1(b)]. In the limit of small tunneling rates -- as in our experiment -- the asymmetric intensity is an expression of the different hole and elec- tron components of the Shiba wavefunctions [14]. The different weights arise from the particle-hole asymmetry in the normal state and an on-site Coulomb potential at the scattering site [9, 17, 27]. The triplets consist of very sharp peaks (50 to 100 µeV full width at half maximum), which are separated by up to 400 µeV [e.g., Fig. 1(c)]. To observe such narrow peaks at 1.2 K, a superconducting tip is required, because then the resolution is not limited by the Fermi-Dirac distribu- tion anymore [28, 29]. The varying coupling strength within the Moir´e-like structure leads to different bound state energies [16]. We use this property to further investigate the origin of the splitting of the Shiba resonances and perform tun- neling spectroscopy on more than 130 molecules. All spectra exhibit two triplets of peaks, one in the bias voltage window −2∆/e < Vbias < −∆/e and one in ∆/e < Vbias < 2∆/e. The additional resonances in the energy interval [−∆tip, ∆tip] are due to tunneling into/out of thermally excited Shiba states (Supplemen- tary Fig. 1, Supplementary Note 1). In Fig. 2(a), we ordered the spectra according to the energy of the most intense Shiba resonance. The false color plot shows a collective "shift" of the Shiba triplets through the su- perconducting gap. The spectra can be categorized into three different regimes. For each of these, we plot a spec- trum in Fig. 2(b). In spectrum I, the intensity of the triplet is larger for tunneling out of the occupied states; in spectrum II, the peaks are close to EF, which hinders a clear distinction of the triplet; spectrum III exhibits larger intensity of the triplet when tunneling into unoc- cupied states. The energetic position of the higher in- tensity subgap peaks corresponds to the binding energy Eb of the Shiba states [15 -- 17]. The order of the spectra from top to bottom thus represents a decreasing coupling strength J with the substrate, which comes along with different adsorption sites. The three spectra are represen- tative for the "Kondo screened" case [Eb < 0, spectrum I in Fig. 2(b)], the "free spin" case (Eb > 0, spectrum III), and a case close to the quantum phase transition (Eb ≈ 0, spectrum II). Possible origins of a Shiba state splitting The collective shift of the Shiba states through a wide range of the gap (and even the quantum phase transi- tion) suggests a correlated origin of the peaks within the triplet. In principle, three different scenarios may ac- count for the occurrence of multiple Shiba bound states in a type I superconductor: (i) different angular momen- tum scattering channels [7 -- 9, 12, 13], (ii) independent scattering at spins in different d orbitals [10], and (iii) FIG. 1. Manganese-phthalocyanine (MnPc) on Pb(111): (a) STM topography of a MnPc monolayer island on Pb(111) (V = 50 mV, I = 200 pA), scale bar is 2 nm. (b) dI/dV spec- tra on the pristine Pb(111) surface (top, offset for clarity) and on the center of a MnPc molecule (bottom) inside a molecular island (opening feedback loop at: V = 5 mV, I = 200 pA). (c) Zoom on the subgap excitations below EF in the MnPc spectrum in (b). The inset shows a MnPc structure model. acts with the superconductor Pb and shows single Shiba states when measured at 4.5 K [16]. Deposition of MnPc molecules at room temperature re- sults in self-assembled, densely-packed monolayer islands [see Fig. 1(a)]. The molecules appear clover-shaped with four lobes around a central protrusion of the Mn ion. The nearly square lattice of the molecular adlayer accom- modates many different adsorption sites of the Mn core on the hexagonal Pb lattice. Consequently, this Moir´e- like pattern involves variations in the electronic and mag- netic coupling strength between adsorbate and substrate [16, 25]. This rich system allows us to identify different quantum ground states with distinct fingerprints of their magnetic excitations. We use tunneling spectroscopy with a superconducting Pb tip at 1.2 K to detect fingerprints of magnetic inter- action of MnPc with the superconducting substrate. As a reference, we plot the differential conductance (dI/dV ) spectrum of the bare Pb surface in Fig. 1(b). A re- gion of zero conductance around the Fermi energy (EF), i.e., the superconducting gap, is framed by quasiparticle resonances at eV = ±(∆sample + ∆tip) = ± 2.63 meV. The doubling of the size of the SC gap is due to the superconductor-superconductor tunneling geometry. The observed presence of the two quasiparticle reso- nances at each side of the gap is explained by the two- band superconductivity of the Pb single crystal as de- scribed recently [28]. Interestingly, the spectra on the MnPc molecules show two triplet sets of peaks inside the superconducting en- ergy gap, which are symmetric in energy around EF, but MnPcPb(111)2∆tip+2∆sample(a)(b)(c)-3.0-2.5-2.001 dI/dV (μS)Sample bias (mV)Sample bias (mV)20-2dI/dV (μS)2013 FIG. 2. Differential conductance spectra of MnPc in Moir´e-like pattern: (a) False color plot of dI/dV spectra of 137 MnPc molecules ordered by the energy of the most intense Shiba resonance (feedback: V = 5 mV, I = 200 pA). (b) Spectra of three MnPc molecules with bound state energies in three different coupling regimes. (c) Spectral intensity obtained by deconvolution of the spectra shown in (b). bound state excitations coupled to other low-energy ex- citations, such as spin excitations or vibrations [10, 11]. Bound states, which originate from higher angular mo- mentum scattering channels (l = 1, 2, etc.), always reside close to the gap edge, while the l = 0 channel may shift through the superconducting energy gap depending on the coupling strength J [9, 12]. This is in contrast to our observation, and, hence, we can rule out (i) as possible origin for the split bound states. In the case of inde- pendent scattering of spins in different d orbitals (ii), a similar shift of all Ebi is unexpected, because d orbitals exhibit different symmetries and interactions with the surface. For the coupling to other degrees of freedom at a sim- ilar energy scale (iii), a collective shift is expected. As we will show in the following, a detailed analysis of the intensities of the Shiba states allows us to unambiguously identify a magnetocrystalline origin of the splitting of the Shiba states as predicted by Zitko and co-workers [10] for S ≥ 1 systems. Shiba intensities as fingerprint of a quantum phase transition The evaluation of the Shiba intensities also sustains the assigned regimes of "Kondo screened" and "free spin" ground states. Such an analysis requires the spectral density of the molecule-substrate system, which can be directly related to the relative weight in the tunneling processes. For this, we remove the effect of the supercon- ducting density of states of the tip by numerical decon- volution of the spectra as described in the Supplemen- tary Note 2 (see Supplementary Fig. 2 for fit quality) [Fig. 2(c)]. We observe a distinct change in the relative peak ar- eas within the triplets when crossing the quantum phase transition. In the "Kondo screened" regime, i.e., for Shiba states with negative binding energies, the individ- ual peaks within a triplet exhibit equal areas [Fig. 3(a)]. Thus, their relative areas Ai/Σ3 j=1Aj are close to 1/3 [left part in Fig. 3(c)]. In contrast, in the "free spin" regime, i.e., positive Shiba binding energies, the areas are considerably different [Fig. 3(b) and (c)]. Here, the relative area of the subgap excitations decreases from the outermost, Eb3, to the innermost, Eb1 [30]. The ratio of peak areas decreases with increasing energy separation [Fig. 3(d)]. Such a behavior is reminiscent of a Boltz- mann distribution, indicating a thermal occupation of a split many-body ground state. The characteristic change in the relative peak areas at the point of the quantum phase transition is a direct fin- gerprint of the origin of the splitting. Independent bound states of different d orbitals (ii) would not change their relative weight at the phase transition. Furthermore, the phase transition should not occur simultaneously for all scattering channels. Additional vibronic resonances (iii) would appear as satellite peaks at higher absolute en- ergy for both ground states. Their spectral weight should scale with the electron-phonon coupling strength, albeit being substantially lower than the weight of the main resonance [11]. This is clearly not the case in our data. Hence, both scenarios can safely be ruled out. Assignment of scattering channels and anisotropy splitting of Shiba states To conclude on the correct model for the description of the multiple subgap resonances, we summarize the essen- tial properties of the Shiba states: On the one hand, the equal peak areas for the "Kondo screened" ground state reflect that our system is characterized by a single level with three possible excitation levels of equal probability. On the other hand, the Boltzmann-like distribution of the areas in the "free spin" case indicate a triplet-split (b)IIIIII0213-202dI/dV (arb. units)Sample bias (mV)3120Spectral intensity (arb. units)Energy (meV)0-11IIIIII(c)Eb2Sample bias (mV)-330IIIIIIdI/dVlowhigh(a)Eb3Eb12∆tip2∆tip4 FIG. 4. Schematic representation of the many-body ground and excited states and the corresponding energy level dia- grams. (a) In the "Kondo screened" ground state the spin (white) in scattering channel k = 1 is screened and the tunnel- ing electron can enter with its spin (blue) parallel to the spin in k = 2, increasing the excited state's total spin to S∗ = 1. The excitation scheme including the anisotropy-split excited state is shown on the right. (b) In the "free spin" ground state, the spin in k = 1 (red) is only partially screened. The tunneling electron has to enter this state in an anti-parallel alignment, obeying the Pauli exclusion principle and reducing the spin to S∗ = 1/2. Red lines in the excitation scheme sym- bolize the thermal occupation of the anisotropy-split ground state. ent molecules by these spin states and their interactions: The coupling strength J1 of channel k = 1 depends on the adsorption site of MnPc on the Pb(111) surface. In the case of strong coupling, the spin is totally Kondo screened [white arrow, Eb < 0, Fig. 4(a)], but k = 2 remains un- screened (red arrow). Hence, the effective total spin is reduced to S = 1/2 by Kondo screening. Tunneling into the Shiba state reflects the excitation to S∗ = 1. In the case of weak coupling in k = 1 [left red arrow, Eb > 0, Fig. 4(b)], the total spin in the ground state multiplet is S = 1. The excited state probed by the Shiba reso- nance is a S∗ = 1/2 state, because the electron attached to k = 1 must obey Pauli's exclusion principle and align anti-parallel. Both spin-1 states, i.e., S∗ = 1 and S = 1, can be split by magnetic anisotropy. A breaking of spherical symmetry of the Mn orbitals by the organic ligand and the adsorption on a substrate leads to a splitting of these spin states [22, 23]. The corresponding Spin Hamilto- nian HS = DS2 spin operators in Cartesian coordinates, accounts for the axial anisotropy and additional rhombicity with the pa- rameters D and E, respectively. This yields a new set of spin eigenstates 0(cid:105), +(cid:105), and −(cid:105), with the latter two being linear combinations of the former eigenstates with ms = 1 and ms = −1 [23]. Schemes of the excitations ob- served in the tunneling spectra are shown in Fig. 4(a,b). In the strongly coupled regime ("Kondo screened"), the excitation of the system yields three excited states with the energy splittings being related to the anisotropy pa- rameters D and E. It should be noted, that their values (cid:1), where the Si are the z + E(cid:0)S2 x − S2 y the sum of all three subgap peaks (Ai/(cid:80)3 the mean energy (Ebi/(cid:80)3 FIG. 3. Shiba state analysis: (a) Zoom on the framed part of the deconvoluted spectrum I in Fig. 2(c). We model the spec- trum with three Lorentzian peaks of different width (shaded in blue, red, and gray) and a broadened step function at the gap edge (see Supplementary Material for details). (b) As (a), but on spectrum III in Fig. 2(c). (c) Peak areas relative to i=1 Ai) plotted vs. i=1 Ebi) of the respective set of sub- gap peaks. (d) Peak area ratios plotted vs. their respective energy splitting from all spectra in the "free spin" state. A Boltzmann distribution for T = 1.2 K is sketched as dashed line. ground state with one excited state. We can correlate these levels to the magnetic interaction channels of the MnPc molecule with the substrate. The MnPc molecule carries a spin S = 3/2 in gas phase [31]. Theory predicts that on Pb(111), the spin in the dxz,yz forms a singlet with the organic ligand states, which reduces the effective spin seen by the substrate's quasiparticles to S = 1 [32]. The unpaired spin in the dz2 orbital is subject to strong coupling with the electronic states of the substrate lead- ing to sizable Kondo screening, whereas the spin in the dxy orbital is not expected to show a significant coupling with the substrate [24, 32, 33]. The occurrence of the Shiba states in tunneling spectra is thus linked to the interaction of the dz2 orbital with the substrate. We la- bel this scattering channel as k = 1 (sketches in Fig. 4). The unscreened spin in the dxy orbital (which we label as k = 2) does not give rise to an observable Shiba state in agreement with the theoretical predictions [32]. How- ever, it couples to the spin in the dz2 orbital and leads to an anisotropy splitting of the Shiba state with k = 1 as discussed below. We can describe the whole set of spectra on the differ- Mean Shiba energy (meV)Ai/(A1+A2+A3)0-10.20.40.60(a)(c)(b)Energy (meV)00.40.8Spectral intensity (arb. u.)102-0.6-1-1.40Energy (meV)123Spectral intensity (arb. u.)Energy separation (meV)0.10.201peak area ratio(d) A1/A3 A2/A3 pBoltzmann at 1.2K A1 A2 A3-0.50.5(a)(b)k = 1k = 2 k = 2k = 1S*=1S=1/20〉−〉+〉D+ED-ES=1S*=1/20〉−〉+〉5 are not a direct measure of the anisotropy energies of the molecule on the surface, but rather represent a renor- malized value due to the many-body interactions with the substrate [10]. Since the energy separation between Eb1 and Eb2 is smaller than between Eb2 and Eb3, the anisotropy parameter D is negative, which means easy- axis anisotropy. For the weakly coupled system, i.e., the "free spin" case, the ground state is split by anisotropy. The excitation spectra reflect transitions from the states −(cid:105), +(cid:105) and 0(cid:105) into the excited S∗ = 1/2 state. The characteristic variations of the peak areas are also well captured in this scenario of anisotropy-split Shiba states: All three spin excitations in the "Kondo screened" regime are equally probable, therefore leading to the same relative peak area [Fig. 3(c)]. On the other hand, a splitting of the ground state, as it is found in the "free spin" regime, leads to a Boltzmann occupa- tion of the levels. In the zero temperature limit, which is discussed in Ref. [10], only one excitation would be detected in the "free spin" regime. At finite temper- ature, the levels are occupied according to Boltzmann statistics. The excitation probabilities are proportional to the state occupation and should thus directly reflect this distribution. Our data in Fig. 3(d) is in agreement with a Boltzmann distribution at 1.2 K or slightly higher, which reflects a temperature-induced population of the split ground state. superconducting gap, one may expect that a split Shiba band structure affects the number of crossing bands and the topological gap width, which is also in the order of 100 µeV [34]. METHODS The experiments were carried out in a commercial Specs JT-STM operating at a base temperature of 1.2 K and a base pressure below 10−10 mbar. The Pb(111) sin- gle crystal surface was cleaned by repeated cycles of Ne+ sputtering and annealing to 430 K until a clean, atomi- cally flat, and SC surface was obtained. From a Knudsen cell held at 673 K, MnPc was thermally evaporated onto the clean surface kept at room temperature, which then was directly transferred into the precooled STM. To gain energy resolution beyond the Fermi-Dirac limit of a normal metal tip, we indented the chemically-etched W tip into the clean SC surface applying 100 V tip bias until a Pb covered, superconducting tip was obtained (energy resolution better than 45 µeV ) [28]. The result- ing spectrum as acquired on the clean Pb(111) surface is shown in Fig. 1(b) top. We acquired dI/dV spectra as a function of sample bias under open-feedback conditions using conventional lock-in technique with a bias modula- tion of 15 µVrms at an oscillation frequency of 912 Hz. DISCUSSION ACKNOWLEDGMENTS Our study shows the importance of anisotropy effects on the subgap excitations which determine the elec- tron transport properties. They provide an unambigu- ous fingerprint of the nature of the ground and excited state throughout the whole range of magnetic interaction strengths, which drive the phase transition between the two quantum ground states. Although the anisotropy energies are renormalized by the coupling to the sub- strates quasiparticles, this method can be used to extract knowledge about magnetic anisotropy and, hence, about spin-orbit coupling of magnetic adsorbates on supercon- ductors. Peculiar consequences of the split Shiba states may occur for coupled magnetic impurities, which lead to the formation of extended Shiba bands. If the exchange cou- pling strength is similar to or smaller than the anisotropy energy, the Shiba bands are expected to reflect the split- ting of the individual Shiba states. Recently, subgap bands have gained particular importance in the search of topological phases and Majorana states in ferromagnetic chains coupled to an s-wave superconductor [5]. If an odd number of spin-polarized bands crosses the Fermi level, Majorana end states can form in the presence of an in- duced (non-trivial) topological gap. Considering that the energy level splitting amounts to about one third of the We thank F. von Oppen for fruitful discussions. We gratefully acknowledge funding by the Deutsche Forschungsgemeinschaft through grant FR2726/4 and by the European Research Council through grant "NanoSpin". [1] Yu, L. Bound state in superconductors with paramag- netic impurities. Acta Phys. Sin. 21, 75 -- 91 (1965). [2] Shiba, H. Classical Spins in Superconductors. Prog. Theor. Phys. 40, 435 -- 451, (1968). [3] Rusinov, A.I. On the theory of gapless superconductivity in alloys containing paramagnetic impurities. Zh. Eksp. Teor. Fiz. 56, 2047 -- 2056, (1969) [Sov. Phys. JETP 29, 1101 -- 1106 (1969)]. [4] Pientka, F., Glazman, L. I., von Oppen, F. Topological superconducting phase in helical Shiba chains. Phys. Rev. B 88, 155420 (2013). [5] Nadj-Perge, S., Drozdov, I. K., Li, J., Chen, H., Jeon, S., Seo, J., MacDonald, A. H., Bernevig, B. A., Yazdani, A. Observation of Majorana fermions in ferromagnetic atomic chains on a superconductor. Science 346, 602 -- 607 (2014). [6] Rontynen, J. & Ojanen, T. Topological Superconductiv- ity and High Chern Numbers in 2D Ferromagnetic Shiba Lattices. Phys. Rev. Lett. 114, 236803 (2015). [7] Ginsberg, D. M. Consequences of Shiba's theory of mag- netic impurities in superconductors, beyond s-wave scat- tering. Phys. Rev. B 20, 960 -- 962 (1979). [8] Kunz, A. B. & Ginsberg, D. M. Band calculation of the effect of magnetic impurity atoms on the properties of superconductors. Phys. Rev. B 22, 3165 -- 3172 (1980). [9] Flatt´e, M. E. & Byers, J. M. Local Electronic Structure of a Single Magnetic Impurity in a Superconductor. Phys. Rev. Lett. 78, 3761 -- 3764 (1997). [10] Zitko, R., Bodensiek, O., Pruschke, T. Effects of mag- netic anisotropy on the subgap excitations induced by quantum impurities in a superconducting host. Phys. Rev. B 83, 054512 (2011). [11] Golez, D., Bonca, J., Zitko, R. Vibrational Andreev bound states in magnetic molecules. Phys. Rev. B 86, 085142 (2012). [12] Kim, Y., Zhang, J., Rossi, E., Lutchyn, R. M. Impurity- Induced Bound States in Superconductors with Spin- Orbit Coupling. Phys. Rev. Lett. 114, 2368040 (2015). [13] Ji, S.H., Zhang, T., Fu, Y.S., Chen, X., Ma, X.C., Li, J., Duan, W.H., Jia, J.F., Xue, Q.K. High-Resolution Scanning Tunneling Spectroscopy of Magnetic Impurity Induced Bound States in the Superconducting Gap of Pb Thin Films. Phys. Rev. Lett. 100, 226801 (2008). [14] Ruby, M., Pientka, F., Peng, Y., von Oppen, F., Hein- rich, B. W., Franke, K. J. Tunneling processes into local- ized subgap states in superconductors. Phys. Rev. Lett. 115, 087001 (2015). [15] Matsuura, T. The Effects of Impurities on Superconduc- tors with Kondo Effect. Prog. Theor. Phys. 57, 1823 -- 1835 (1977). [16] Franke, K.J., Schulze, G., Pascual, J.I. Competition of Superconducting Phenomena and Kondo Screening at the Nanoscale. Science 332, 940 -- 944 (2011). [17] Bauer, J., Pascual, J. I., Franke, K. J. Microscopic reso- lution of the interplay of Kondo screening and supercon- ducting pairing: Mn-phthalocyanine molecules adsorbed on superconducting Pb(111). Phys. Rev. B 87, 075125 (2013). [18] Yazdani, A., Jones, B. A., Lutz, C. P. Crommie, M. F., Eigler, D. M. Probing the Local Effects of Magnetic Im- purities on Superconductivity Science, 275, 1767 -- 1770 (1997). [19] Balatsky, A.V., Vekhter, I., Zhu, J.-X. Impurity-induced states in conventional and unconventional superconduc- tors. Rev. Mod. Phys. 78, 373 -- 433 (2006). [20] Deacon, R. S., Tanaka, Y., Oiwa, A., Sakano, R., Yoshida, K., Shibata, K., Hirakawa, K., Tarucha, S. Tun- neling Spectroscopy of Andreev Energy Levels in a Quan- tum Dot Coupled to a Superconductor. Phys. Rev. Lett. 104, 076805 (2010). 6 [21] Lee, E. J. H., Jiang, X., Houzet, M., Aguado, R. Lieber, C. M., De Franceschi, S. Spin-resolved Andreev levels and parity crossings in hybrid superconductor-semiconductor nanostructures. Nature Nanotech. 9, 79 -- 84 (2014). [22] Gatteschi, D., Sessoli, R. & Villain, J. in Molecular Nano- magnets (Oxford University Press, USA, 2006). [23] Tsukahara, N. et al. Adsorption-Induced Switching of Magnetic Anisotropy in a Single Iron(II) Phthalocyanine Molecule on an Oxidized Cu(110) Surface. Phys. Rev. Lett. 102, 167203 -- 167206 (2009). [24] Kugel, J., Karolak, M., Senkpiel, J., Hsu, P.-J., San- giovanni, G., Bode, M. Relevance of Hybridization and Filling of 3d Orbitals for the Kondo Effect in Transi- tion Metal Phthalocyanines. Nano Lett. 14, 3895 -- 3902 (2014). [25] Ji, S.H., Fu, Y.S., Zhang, T., Chen, X., Jia, J.F., Xue, Q.K., Ma, X.C. Kondo Effect in Self-Assembled Man- ganese Phthalocyanine Monolayer on Pb Islands. Chin. Phys. Lett. 27, 087202 (2010). [26] Str´ozecka, A., Soriano, M., Pascual, J. I., Palacios, J. J. Reversible Change of the Spin State in a Manganese Phthalocyanine by Coordination of CO Molecule. Phys. Rev. Lett. 109, 147202 (2012). [27] Salkola, M. I., Balatsky, A. V., Schrieffer, J. R. Spec- tral properties of quasiparticle excitations induced by magnetic moments in superconductors Phys. Rev. B 55, 12648 -- 12661 (1997). [28] Ruby, M., Heinrich, B. W., Pascual, J. I., Franke, K. J. Experimental Demonstration of a Two-Band Supercon- ducting State for Lead Using Scanning Tunneling Spec- troscopy. Phys. Rev. Lett. 114, 157001 (2015). [29] Yet, a finite temperature leads to finite lifetime effects, which broaden the superconducting coherence peaks and subgap states. Hence, already at 4.5 K, the triplet peaks overlap and give rise to a single peak [16]. [30] Close to the quantum phase transition, i.e., close to EF , the Shiba resonances overlap with their thermal excita- tions. This prohibits a clear distinction of the peak areas and they are therefore omitted in the data plot. [31] Liao, M.-S., Watts, J.D., Huang, M.-J. DFT Study of Unligated and Ligated ManganeseII Porphyrins and Ph- thalocyanines. Inorg. Chem. 44, 1941 -- 1949 (2005). [32] Jacob, D., Soriano, M., Palacios, J. J. Kondo effect and spin quenching in high-spin molecules on metal sub- strates. Phys. Rev. B 88, 134417 (2013). [33] Kugel, J., Karolak, M., Kronlein, A., Senkpiel, J., Hsu, P.-J., Sangiovanni, G. Bode, M. State identification and tunable Kondo effect of MnPc on Ag(001). Phys. Rev. B 91, 235130 (2015). [34] Ruby, M., Pientka, F., Peng, Y., von Oppen, F., Hein- rich, B. W., Franke, K. J., End states and subgap struc- ture in proximity-coupled chains of magnetic adatoms. arXiv:1507.03104 (2015). 

1211.0298

1

1211

A Nanoscale Parametric Feedback Oscillator

We describe and demonstrate a new oscillator topology, the parametric feedback oscillator (PFO). The PFO paradigm is applicable to a wide variety of nanoscale devices, and opens the possibility of new classes of oscillators employing innovative frequency-determining elements, like such as nanoelectromechanical systems (NEMS), facilitating integration with circuitry, and reduction in cost and system size reduction. We show that the PFO topology can also improve nanoscale oscillator performance by circumventing detrimental effects that are otherwise imposed by the strong device nonlinearity in this size regime.

cond-mat.mes-hall

cond-mat

A Nanoscale Parametric Feedback Oscillator L. Guillermo Villanueva, Rassul B. Karabalin, Matthew H. Matheny, Eyal Kenig, Michael C. Cross, Michael L. Roukes* Kavli Nanoscience Institute, California Institute of Technology, Pasadena, CA, 91125 RECEIVED DATE: 2011-09-07 Email: roukes@caltech.edu TITLE RUNNING HEAD: A Nanoscale Parametric Feedback Oscillator ABSTRACT We describe and demonstrate a new oscillator topology – the parametric feedback oscillator (PFO). The PFO paradigm is applicable to a wide variety of nanoscale devices, and opens the possibility of new classes of oscillators employing innovative frequency-determining elements, like such as nanoelectromechanical systems (NEMS), facilitating integration with circuitry, and reduction in cost and system size reduction. We show that the PFO topology can also improve nanoscale oscillator performance by circumventing detrimental effects that are otherwise imposed by the strong device nonlinearity in this size regime. KEYWORDS: Self-sustained oscillator, Parametric, Nanoelectromechanical systems (NEMS), improved frequency stability. 1  MANUSCRIPT TEXT Frequency stability is essential for self-sustained oscillators that are at the heart of many current technologies spanning communication, computation, and geolocation1-2. An oscillator’s essential elements are a frequency-determining element, typically a mechanical or electrical resonator with a high quality factor response (high Q); and a feedback loop, usually composed of a linear amplification stage, a signal limiter, and a phase-delay element. Here, by the term resonator, we refer to a passive device that requires AC power in order to be driven into motion. In an oscillator, the feedback signal provides sufficient drive to overcome the resonator’s damping and, thereby, to sustain continuous vibrations. The most common type of frequency-determining element for oscillators is a macroscopic quartz crystal3. Quartz-crystal-based oscillators have been the prevalent standard for almost a century4, and have remained such despite the semiconductor microelectronics revolution. Recent advances in micro- and nano- fabrication permit miniaturization of semiconductor-based mechanical resonators, potentially facilitating their on-chip integration with electronic components. Initial steps towards scaling such integrated mechanical resonators downward to the realm of microelectromechanical systems (MEMS) shows promise5-6. Yet full compatibility with very-large-scale integration (VLSI) would require that a resonant element their dimensions should ultimately be reduced even further, to the domain of nanoelectromechanical systems (NEMS7), so as to become directly compatible in size with individual transistors. A complementary motivation for pursuing low-noise nanoscale oscillators emerges from the realm of sensing. NEMS resonators are increasingly being employed for sensing applications ranging from detection of mass8-9, gas10-11, biomolecules12-13 and force14-15 − and provide unprecedented resolution. The most responsive sensing modalities typically employ frequency-shift sensing, a configuration where the stimulus to be measured induces a change in the resonant frequency of the NEMS. A stable NEMS oscillator element, co-integrated with a few active transistors, thus enables an especially compact sensing “pixel”. 2 Whether for frequency control or sensing, implementing nanomechanical oscillators has proven to be extremely challenging, mainly due to the small magnitude of the motional signal generated by the NEMS in comparison to the parasitic cross-talk from the drive16. As the dimensions of a NEMS element shrink, so do both the electrical signal produced by its mechanical motion and its onset of nonlinearity17. This makes it exceptionally difficult to harness and control the motional signal produced by the NEMS in the in the face of the unavoidable stray reactances that, generally, cause overwhelmingly large background signals. As has been discussed previously16, in certain cases these deleterious phenomena can be circumvented by implementing carefully constructed bridging and filtering circuits, but these solutions are not universally applicable nor are they easily integrated on-chip. In this Letter, we present an alternative to the canonical oscillator topology, namely, a new architecture based on non-resonant parametric feedback that can be applied to a wide variety of nanoscale resonators. We use a feedback loop possessing a quadratic transfer function to apply parametric excitation at twice the resonant frequency (as opposed to “direct drive” at its resonant frequency). This parametric drive sets the resonator in motion by dynamically modulating one of its physical parameters18-19. A complete mathematical analysis of parametric feedback oscillators is provided in the Supporting Information. As we show, there are many advantages of this generalized parametric feedback technique, both from fundamental and technological points of view. Among them are: (a) it becomes possible to circumvent the need to satisfy the “Barkhausen criteria” that govern conventional oscillators20. This makes it possible to harness a wide variety of nanoscale resonators that would otherwise be impossible to employ for oscillator circuits. (b) An unprecedented level of control of the resonator’s nonlinear characteristics is afforded, enabling access to higher amplitudes of operation. (c) A wide range of frequency tunability becomes achievable. (d) Substantial improvement in frequency stability of the oscillator, compared with that of conventional direct-drive implementations, becomes possible. 3 Here we demonstrate the PFO concept using piezoelectric NEMS. We pattern doubly-clamped beam NEMS resonators from a four-layer stack of aluminum nitride (AlN)-molybdenum (Mo)-AlN-Mo, having a total thickness of 210 nm, a width of 470 nm and a length of 9 µm (Fig. 1a). Our fabrication process is described in detail in the Section I of the Supporting Information. We use this materials combination because it enables the fabrication of NEMS resonators with easily accessible and analytically predictable nonlinear behavior17 that can be easily excited directly21 and parametrically22 by means of the piezoelectric effect. Such piezoelectric NEMS are promising candidates for future co- integration with chip-based electronic circuitry given their small size and compatibility with CMOS processes. We detect the out-of-plane resonator motion using the time-varying, strain-induced resistance changes in a piezometallic (Mo) loop patterned at one end of the beam (Fig. 1b). Actuation is obtained by applying an AC voltage to an electrode that covers most of the beam’s length; this induces longitudinal strain by means of the inverse piezoelectric effect. This time-varying strain can be used to actuate the beam either directly21,23 or parametrically22,24. By driving the beam directly, we determine the natural frequency (f0 = 14.305 MHz) and quality factor (Q = 1220) of the specific device used in these studies by fitting its driven resonant response to a Lorentzian peak (Fig. 1c-bottom). By separately measuring the thermomechanical noise without any drive we calibrate the piezoresistive response to absolute displacement; from this we deduce the transduction responsivity of 8.7 nm/mV for a constant bias voltage of 200 mV across the piezoresistor (Fig. 1c-top). Our doubly-clamped beam devices exhibit stiffening behavior, characteristic of a Duffing nonlinearity, at large drive levels; we use the deduced transduction responsivity to ascertain that the critical amplitude characterizing the onset of nonlinearity is 9.6 nm. This agrees with the predictions of analytical calculations17 (Fig. 1d-top). The resonance frequency can be tuned by application of a voltage to the actuation electrode (Fig. 1d-inset). We find a tuning sensitivity of 35 kHz/V for this device. This significant tunability readily enables parametric excitation; we subsequently characterize this by sweeping the drive frequency and monitoring the 4 amplitude of vibration at half the applied drive frequency (Fig. 1d-bottom). Especially noteworthy is the plot at the bottom of Fig. 1d that shows both amplitude and frequency detuning grow faster with drive when the device is actuated parametrically at 2f as compared with results when directly driven at f (Fig. 1d-top). For example, at 120 mV drive the amplitude and detuning are higher for direct drive at f, but when the drive levels exceed 130 mV the situation is reversed. Accordingly, motion amplitudes for the same driving voltage can be much higher in the parametric case, and this is of special significance when building a low noise oscillator. The schematic of our implementation of parametric feedback oscillator topology is shown in Fig. 2a. Out-of-plane mechanical motion is transduced by providing a constant DC bias voltage across the piezoresistor. The motional signal is then amplified and filtered to suppress high frequency noise and higher harmonics (>f). Subsequently, the signal is delayed by a voltage controlled phase shifter (φ), then passed through a nonlinear element optimized to generate a 2f signal with amplitude proportional to the square of the resonator motion (see Supporting Information Section II and VI). This frequency doubled waveform is subsequently passed through a highly selective bandpass filter to ensure strong suppression of undesired harmonic content (f, 3f, 4f, etc.). This ensures that the drive signal fed-back to the resonator is purely sinusoidal at 2f, which prevents the induction of undesired motional response. Pure modulation at 2f induces a resonator response at f – that is, parametric oscillation – provided the 2f drive level exceeds the parametric threshold. This threshold can be surpassed in any resonator that has sufficient susceptibility to parametric tuning to permit its resonance frequency to be shifted by more than twice its linewidth22. We have evaluated the feasibility of achieving the parametric threshold for a variety of state-of-the-art resonators in Section III of the Supporting Information. Elements scaled down to the nanoscale in all dimensions attain high frequencies with low force constants; this proves ideal for attaining a low parametric threshold. Unlike behavior in traditional oscillators, the zero-amplitude state is a stable solution for our feedback system. This makes it necessary to initiate resonator motion by an external “start-up” source. After 5 oscillations commence, this start-up drive can be removed and stable oscillations at f will persist, sustained only by parametric feedback at 2f. In steady state, the parametrically-driven resonator acts as a frequency divider in the circuit. Given that the frequency of the feedback (2f) and output signal (f) are well separated for the parametric feedback oscillator, their crosstalk is minimized. This eliminates a traditional obstacle for oscillators based on small mechanical devices; for small electromechanical resonators, the output electrical signal is usually strongly dominated by the feed-through of the actuation voltage. In such conditions it is very challenging to attain an oscillation that uses the mechanical resonator as the frequency determining element. The equation of motion of our doubly-clamped beam PFO system can be written as: x  + f 2 π Q 0 x  + (2 ) π 2 f 2 0 [1 + ( ) t , ζ φ ] x ax + 3 2 x x G t ( ) + = η  . (1) Here x represents the displacement of the resonator; α is the nonlinear spring constant (also called the Duffing parameter); η is the coefficient of nonlinear damping; G(t) is an external drive signal (G = 0 when the system is in self-sustained parametrically fed-back oscillation); and ζ(t,φ) is the feedback function, which depends on the resonator displacement and the externally controlled phase delay (as shall be described below and in Section II of the Supporting Information). Detailed analysis of equation (1) shows that by varying the two parameters characterizing the parametric feedback, its phase delay φ and gain, control of both the resonator’s effective nonlinear stiffness (the Duffing coefficient, proportional to 3x ) and its nonlinear damping (proportional to 2x x ) becomes possible (see Section II of Supporting Information). We demonstrate this experimentally by measuring the driven resonant response with parametric feedback below the oscillation threshold, for different values of φ. Sweeps of the driven amplitude, as shown in Fig. 2b-d, display induced changes in the resonator’s nonlinear coefficient. This evolves from negative (shown in Fig. 2b) to positive values (shown in Fig. 2d). At an intermediate feedback phase (shown in Fig. 2c), the effective nonlinear Duffing constant vanishes and effective nonlinear damping becomes apparent; this is reflected in the 6 increased peak widths at higher drive levels. This control of the nonlinear properties of the system opens possibilities for combined operation, using both parametric (2f0) and direct-drive feedback (f0), to increase the system’s dynamic range and improve its frequency stability. Rotation of the external parametric feedback phase leads to a direct reduction of the non-linear damping. When this reduction is sufficient, the aforementioned parametric oscillation criterion is satisfied (see Section IV in Supporting Information) and oscillations ensue. We characterize the resulting parametric oscillations by capturing their power spectrum, and compare this to the open-loop resonator frequency response (Fig. 3a). The implementation presented here is one of the few examples of a NEMS oscillator reported to date16,25-26 and, we believe, represents the first realization of a parametric feedback oscillator in any system. For our prototype PFO, we deduce an effective quality factor of 99,000 from its power spectrum; this is more than eighty times larger than the Q of the NEMS resonator itself when operating in its linear regime. We now analyze the oscillator behavior as a function of the phase shift φ. In Fig. 3b we plot the spectral response of the oscillator for three different values of φ, each incremented by 10 degrees, which results in a frequency shift increment of about 140 kHz (≈14 kHz/deg). Fig. 3b also shows, for comparison, the open-loop resonator response at 20 mV drive. Our theoretical analysis (Supporting Information, Section II) predicts that oscillation frequency and amplitude should both display a strong dependence on φ. We verify this prediction by experimentally monitoring the oscillation frequency while quasistatically changing φ (using a voltage-controlled phase shifter). Fig. 3c shows the large tuning range obtained, which is almost 18% (from 14.35 to 16.9 MHz). This wide tuning range should prove useful for applications requiring voltage controlled oscillators27, and for the potential synchronization of coupled oscillators28. The extended tuning range is a direct consequence of using parametric feedback: as was shown in Fig. 1, frequency pulling induced by the parametric drive (2f) is much more efficient than that obtained from a direct drive (at f). The phase shift range accessible in our experiments is about 1400º, and enables our observation of parametric oscillations on three adjacent 7 branches of the phase response, each separated by 360º (Fig. 3c). Excellent quantitative agreement between theory and experiment is evident. Detailed inspection of our experimental results reveals a flattening of the frequency versus phase data in one specific region, which deviates from our initial theoretical model. This flattening occurs near 16.2 MHz for our device, and is observed in all three of the branches displayed (Fig. 3c and inset). These features arise from the coupling of the fundamental out-of-plane vibrational mode and its first in-plane mode (~32.4 MHz). By modeling an interaction between these two modes29 we obtain refined predictions that qualitatively match the experimental findings (Supporting Information, Section V). To assess the performance of our parametric feedback oscillator as a frequency source, we measure its frequency stability. We measure the oscillator’s phase noise, which represents the sideband power spectral density at a given offset frequency, normalized by the oscillator's signal power16. To provide a baseline for comparison, we separately construct a conventional feedback loop with direct drive at f, using the same resonant element and active components. Frequency stability comparisons between this direct-drive oscillator and our prototype PFO, for operation at identical energies, are shown in Fig. 3d. We observe that the frequency stability of our PFO is significantly improved compared to that of the traditional oscillator topology. This provides direct evidence of suppression that the PFO topology can suppress the effects of phase noise in the feedback electronics. In the inset of Fig. 3d, which shows the PFO phase noise at 1 kHz offset, the relative improvement of the PFO’s frequency stability is seen to remain relatively constant over the full range of φ. However, very striking enhancement is observed in the region where mode-coupling occurs; in this regime the oscillator’s phase noise is reduced by an additional 15 dB. This enhanced noise suppression is consistent with the fact that the frequency instability of our system is dominated by phase fluctuations in the fed back signal. Since phase noise is proportional to the slope of the phase tuning data, there is less noise associated with such “flattened” regions (see Section II in Supporting Information). We anticipate that substantial improvement in frequency stability, ultimately down to the fundamental thermal noise limit, 8 should become possible with optimal engineering of the frequency-phase dependence by such means. In fact, it should be possible to suppress essentially any noise mechanism that originates within the feedback loop itself. Thus, the PFO topology offers a means for resolving the long-standing challenge of attaining ultimate thermodynamic limits of performance in oscillators. In this work, we describe and demonstrate a novel oscillator circuit topology that employs a nonlinear, parametrically-actuated NEMS doubly-clamped beam, with feedback characterized by a nonlinear, square-law dependence on resonator signal. The advantages of this architecture, which include elimination of cross-talk, control of non-linear properties, large frequency tunability, and significant phase noise reduction, are evident from the experimental results we demonstrate. Since the requirements for realizing a PFO rely solely on the presence of sufficient frequency tunability, the PFO architecture offers wide applicability and outstanding frequency scalability across a variety of possible implementations. This opens new avenues for realizing miniaturized micro- and nano-scale mechanical oscillators based on resonator technologies ranging from MEMS electrostatic disc to graphene NEMS, and it should facilitate very large scale integration of such oscillators with state-of-the-art electronic circuitry. ACKNOWLEDGMENTS This work was supported by the Defense Advanced Research Projects Agency Microsystems Technology Office, Dynamic Enabled Frequency Sources Program (DEFYS) through Department of Interior (FA8650-10-1-7029). We thank R. Lifshitz and X.L. Feng for discussions, and P. Ivaldi, E. Defaÿ and S. Hentz for providing us with the AlN/SOI material. L.G.V. acknowledges financial support from Prof. A. Boisen and the European Commission (PIOF-GA-2008-220682). SUPPORTING INFORMATION PARAGRAPH Supporting Information Available: Piezoelectric NEMS fabrication process flow, PFO applicability criterion, theoretical considerations (derivation of PFO amplitude equation, PFO oscillation condition, 9 modeling of inter-mode coupling feature), thermomechanical limit for phase noise. This material is available free of charge via the Internet at http://pubs.acs.org. 10  Figure 1 Nanomechanical resonator characteristics. a, Colored SEM micrograph of the suspended mechanical device used to demonstrate the generalized feedback oscillator. The metal electrode covering most of the beam’s length is used for actuation, whereas the loop on the opposite side is used for detection. b, Detail of the piezometallic loop used to transduce the motion of the resonator. Scale bars: 500 nm. c, Top: Voltage spectral density showing the background (system) noise and the thermomechanical peak of the resonator. Detection efficiency (responsivity) of the system is estimated to be 8.7 nm/mV and sensitivity is 0.52 pm/Hz1/2. Bottom: Linear resonant response of the resonator in the vicinity of its characteristic resonant frequency. A Lorentzian fit reveals Q=1220. d, Top: Direct drive of the resonator. Curves show the amplitude response of the resonator around its natural frequency for different driving forces (from 20 mV to 160 mV in steps of 20 mV). A characteristic stiffening effect can be seen and fitted to a Duffing model, to obtain a critical amplitude of about 9.6 nm and a nonlinear dissipation coefficient of 0.01520. Bottom: Parametric excitation of the resonator. Curves show the amplitude response of the resonator as a function of half the driving frequency for different driving amplitudes (from 120 mV to 133 mV in 3 mV steps) showing the parametric excitation of the resonator. (Inset) Tunability of the characteristic frequency of the resonator versus DC voltage applied to the actuation electrode (35 kHz/V). 11  Figure 2 Feedback system and results below oscillation threshold. a, Schematic diagram of our generalized-feedback system. The signal from the resonator is amplified and filtered at high frequencies to eliminate higher harmonics and noise. After an externally controlled (φ) phase delay is applied, the signal is passed through a nonlinear element followed by a bandpass filter to ultimately generate a signal at 2f. This signal is applied to the actuation port of the beam through a power combiner that allows simultaneous feedback and a direct drive with an external source (G), which is necessary to initiate the oscillations. Once the self-sustaining state has been reached, the source can be disconnected and the motion persists. b-d, Driven resonant response of the system with feedback below oscillation threshold. Amplitude of motion is plotted as a function of drive frequency for 10 different drive values (from 10 to 100 mV in steps of 10 mV). b-d, are for different values of φ (in increments of 45o). The resonator’s effective nonlinear coefficient tunes from negative b, to positive d. In c, we show that the nonlinear Duffing coefficient vanishes and the effective nonlinear damping is more apparent. 12  Figure 3 Parametric Feedback Oscillator. a, Normalized comparison between the spectral power of a PFO (orange) and the linear resonant response of the open-loop system (purple). The compression ratio is approximately 82. b, Comparison between the open loop response (purple) and the PFO power spectrum (orange). Three PFO traces are shown, each for a different value of φ (separated by 10o). The tunability in this frequency range is around 14 kHz/o. c, Dependence of PFO frequency on φ. Three sets of data (orange) are experimental measurements corresponding to three different solution branches, separated by 360o. Theoretical predictions (purple) showing stable (solid lines) and unstable (dash lines) solutions show remarkable agreement with experiment. A flattening of the tunability curve close to 16.2 MHz appears in all three branches, showing interaction of the oscillator with a different mechanical mode in the beam. Inset: Detail of such flattening for the third branch. d, Inset: phase noise at 1kHz offset for our PFO as a function of φ. Little dependence is observed except in the proximity of the flattening feature shown in c. d, Phase noise measurements for our PFO in both a standard case and at the optimum phase value, showing a reduction of the noise. For comparison, the phase noise of a standard direct-drive oscillator is shown for the same oscillator energy, indicating higher phase noise than PFO over most of the frequency range. 13 REFERENCES 1. Razavi, B., Study of phase noise in CMOS oscillators. IEEE J. Solid-St. Circ. 1996, 31 (3), 331- 343. 2. Razavi, B., Design of integrated circuits for optical communications. McGraw-Hill: Boston, 2003; p 370. 3. Cady, W. G., The piezo-electric resonator. Proc. IRE 1922, 10 (2), 83-114. 4. Marrison, W. A., The Evolution of the Quartz Crystal Clock. Bell. Syst. Tech. J. 1948, 27 (3), 510-588. 5. Nguyen, C. T. C.; Howe, R. T., An integrated CMOS micromechanical resonator high-Q oscillator. IEEE J. Solid-St. Circ. 1999, 34 (4), 440-455. 6. Zuo, C. J.; Sinha, N.; Van der Spiegel, J.; Piazza, G., Multifrequency Pierce Oscillators Based on Piezoelectric AlN Contour-Mode MEMS Technology. J Microelectromech S 2010, 19 (3), 570-580. 7. Roukes, M., Nanoelectromechanical systems face the future. Physics World 2001, 14 (2), 25-31. 8. Naik, A. K.; Hanay, M. S.; Hiebert, W. K.; Feng, X. L.; Roukes, M. L., Towards single- molecule nanomechanical mass spectrometry. Nat. Nanotechnol. 2009, 4 (7), 445-450. 9. Bockrath, M.; Chiu, H. Y.; Hung, P.; Postma, H. W. C., Atomic-Scale Mass Sensing Using Carbon Nanotube Resonators. Nano Letters 2008, 8 (12), 4342-4346. 10. Roukes, M. L.; Li, M.; Myers, E. B.; Tang, H. X.; Aldridge, S. J.; McCaig, H. C.; Whiting, J. J.; Simonson, R. J.; Lewis, N. S., Nanoelectromechanical Resonator Arrays for Ultrafast, Gas-Phase Chromatographic Chemical Analysis. Nano Letters 2010, 10 (10), 3899-3903. 11. Ivaldi, P.; Abergel, J.; Matheny, M. H.; Villanueva, L. G.; Karabalin, R. B.; Roukes, M. L.; Andreucci, P.; Hentz, S.; Defay, E., 50 nm thick AlN film-based piezoelectric cantilevers for gravimetric detection. J Micromech Microeng 2011, 21 (8). 14 12. Fritz, J.; Baller, M. K.; Lang, H. P.; Rothuizen, H.; Vettiger, P.; Meyer, E.; Guntherodt, H. J.; Gerber, C.; Gimzewski, J. K., Translating biomolecular recognition into nanomechanics. Science 2000, 288 (5464), 316-318. 13. Arlett, J. L.; Myers, E. B.; Roukes, M. L., Comparative advantages of mechanical biosensors. Nat. Nanotechnol. 2011, 6 (4), 203-215. 14. Rugar, D.; Budakian, R.; Mamin, H. J.; Chui, B. W., Single spin detection by magnetic resonance force microscopy. Nature 2004, 430 (6997), 329-332. 15. Roukes, M. L.; Arlett, J. L.; Maloney, J. R.; Gudlewski, B.; Muluneh, M., Self-sensing micro- and nanocantilevers with attonewton-scale force resolution. Nano Letters 2006, 6 (5), 1000-1006. 16. Feng, X. L.; White, C. J.; Hajimiri, A.; Roukes, M. L., A self-sustaining ultrahigh-frequency nanoelectromechanical oscillator. Nat. Nanotechnol. 2008, 3 (6), 342-346. 17. Lifshitz, R.; Cross, M. C., Nonlinear Dynamics of Nanomechanical and Micromechanical Resonators. In Reviews of nonlinear dynamics and complexity, Schuster, H. G., Ed. Wiley-VCH: Weinheim, 2008; Vol. 1. 18. Rugar, D.; Grutter, P., Mechanical Parametric Amplification and Thermomechanical Noise Squeezing. Phys. Rev. Lett. 1991, 67 (6), 699-702. 19. Turner, K. L.; Miller, S. A.; Hartwell, P. G.; MacDonald, N. C.; Strogatz, S. H.; Adams, S. G., Five parametric resonances in a microelectromechanical system. Nature 1998, 396 (6707), 149-152. 20. Hajimiri, A.; Lee, T. H., The design of low noise oscillators. Kluwer Academic Publishers: Boston, 1999; p 208. 21. Karabalin, R. B.; Matheny, M. H.; Feng, X. L.; Defay, E.; Le Rhun, G.; Marcoux, C.; Hentz, S.; Andreucci, P.; Roukes, M. L., Piezoelectric nanoelectromechanical resonators based on aluminum nitride thin films. Applied Physics Letters 2009, 95 (10), 103111. 15 22. Karabalin, R. B.; Masmanidis, S. C.; Roukes, M. L., Efficient parametric amplification in high and very high frequency piezoelectric nanoelectromechanical systems. Applied Physics Letters 2010, 97 (18), 183101. 23. Masmanidis, S. C.; Karabalin, R. B.; De Vlaminck, I.; Borghs, G.; Freeman, M. R.; Roukes, M. L., Multifunctional nanomechanical systems via tunably coupled piezoelectric actuation. Science 2007, 317 (5839), 780-783. 24. Mahboob, I.; Yamaguchi, H., Piezoelectrically pumped parametric amplification and Q enhancement in an electromechanical oscillator. Applied Physics Letters 2008, 92 (17), 173109. 25. Ayari, A.; Vincent, P.; Perisanu, S.; Choueib, M.; Gouttenoire, V.; Bechelany, M.; Cornu, D.; Purcell, S. T., Self-oscillations in field emission nanowire mechanical resonators: A nanometric dc-ac conversion. Nano Letters 2007, 7 (8), 2252-2257. 26. Verd, J.; Uranga, A.; Abadal, G.; Teva, J. L.; Torres, F.; Lopez, J.; Perez-Murano, F.; Esteve, J.; Barniol, N., Monolithic CMOS MEMS oscillator circuit for sensing in the attogram range. Ieee Electr Device L 2008, 29 (2), 146-148. 27. Razavi, B., A study of injection locking and pulling in oscillators. IEEE J. Solid-St. Circ. 2004, 39 (9), 1415-1424. 28. Cross, M. C.; Zumdieck, A.; Lifshitz, R.; Rogers, J. L., Synchronization by nonlinear frequency pulling. Phys. Rev. Lett. 2004, 93 (22), 224101. 29. Westra, H.; Poot, M.; van der Zant, H.; Venstra, W., Nonlinear Modal Interactions in Clamped- Clamped Mechanical Resonators. Phys. Rev. Lett. 2010, 105 (11), 117205. 16  SYNOPSIS TOC 17 

1504.04952

2

1504

Tuning equilibration of quantum Hall edge states in graphene - role of crossed electric and magnetic fields

We probe quantum Hall effect in a tunable 1-D lateral superlattice (SL) in graphene created using electrostatic gates. Lack of equilibration is observed along edge states formed by electrostatic gates inside the superlattice. We create strong local electric field at the interface of regions of different charge densities. Crossed electric and magnetic fields modify the wavefunction of the Landau Levels (LLs) - a phenomenon unique to graphene. In the region of copropagating electrons and holes at the interface, the electric field is high enough to modify the Landau levels resulting in increased scattering that tunes equilibration of edge states and this results in large longitudinal resistance.

cond-mat.mes-hall

cond-mat

Tuning equilibration of quantum Hall edge states in graphene role of crossed electric and magnetic fields Sudipta Dubey∗ and Mandar M. Deshmukh† Department of Condensed Matter Physics and Materials Science, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400005, India (Dated: September 21, 2018) We probe quantum Hall effect in a tunable 1-D lateral superlattice (SL) in graphene created using electrostatic gates. Lack of equilibration is observed along edge states formed by electrostatic gates inside the superlattice. We create strong local electric field at the interface of regions of different charge densities. Crossed electric and magnetic fields modify the wavefunction of the Landau Levels (LLs) - a phenomenon unique to graphene. In the region of copropagating electrons and holes at the interface, the electric field is high enough to modify the Landau levels resulting in increased scattering that tunes equilibration of edge states and this results in large longitudinal resistance. Magnetotransport across one-dimensional superlattice (SL) had been studied in two-dimensional electron gas in semiconductor heterostructures (2DEGS) [1 -- 6], report- ing dissipationless transport across high potential bar- riers [1] and magnetic commensurability oscillations in longitudinal resistance [3]. The motivation was to study various competing length scales and energy scales be- tween tunable SL potential and quantum Hall system. Graphene offers the advantage of large cyclotron gap al- lowing quantum Hall effect to be observed at room tem- perature [7 -- 9]. Substrate induced SL in graphene in the presence of magnetic field led to the experimental ob- servation of Hofstadter butterfly physics [10 -- 12]. The ability to create abrupt (∼ 10 nm) tunable barriers in graphene allows new aspects to be explored. In addition, new physics, due to the role of crossed electric and mag- netic field, that cannot be seen in conventional 2DEGS can be studied in SL structures based on graphene. In this letter, we study magneto transport in an elec- trostatically defined 1D lateral SL in graphene [13]. In our device we apply a perpendicular magnetic field and periodically modulate the charge carrier density in ad- jacent "ribbons" of graphene, tuning from an array of p-p' (or n-n' ) to an array of p-n' junctions. Changing the magnetic field allows us to vary lB relative to λ; and changing the gate voltage allows us to tune the SL poten- tial strength relative to LL spacing. The relative abrupt- ness, bipolarity of charge carriers, large modulation and unequally spaced LLs distinguishes the present work from the previous work on 1D SL using 2DEGS systems [1 -- 4, 14]. Apart from the length scales, we also study the energy scales involved. The competition between SL amplitude (V0) and LL spacing (¯hωc, where ¯h = h/2π, h being the Planck's constant, and ωc is the cyclotron frequency) gives rise to three regimes. When V0 >> ¯hωc, SL effect dominates giving rise to extra Dirac points [15]. In the other extreme when V0 << ¯hωc, quantum Hall effect in ∗ sudipta.tifr@gmail.com † deshmukh@tifr.res.in graphene is restored [15]. However, the situation is more complex and little explored when V0 and ¯hωc have com- parable contribution, and we have experimentally probed this regime in graphene. The goal of our work is to extend quantum Hall stud- ies beyond single top-gate in graphene [16 -- 18]. Our work is the first experimental report on magnetotransport in multiple top-gates on graphene and we probe the physics of equilibration along the narrow region in graphene de- fined electrostatically. Our main observation is that when V0 is comparatively small in the unipolar region, the edge states do not equilibrate along this narrow region de- fined electrostatically. The extent of equilibration can be tuned in the bipolar region where the electric field is relatively large. In this regime, electric field significantly modifies the Landau level wavefunctions, increasing scat- tering, which is reflected in increased equilibration and large longitudinal resistance. We create a 1D tunable SL, of period λ, by fabri- cating an array of thin finger gates on graphene. The schematic of a device is shown in Figure 1(a), and Fig- ure 1(b) shows false colored scanning electron microscope image (details of fabrication in Section I of Supplemental Material). The geometric width of each top-gates is ∼ 30 nm and they have a period of λ = 150 nm. The effec- tive electrostatic width of the top-gates felt by the charge carriers in graphene is larger due to the finite thickness of the top-gate dielectric [13] (details in Section VII of Supplemental Material). In our device, graphene consists of two alternating re- gions - one where the charge carrier density is controlled only by the back-gate (BG region); and the other where the charge carrier density is controlled by both the top- gate and the back-gate (TG region). The difference in charge carrier density between BG and TG regions gives rise to a SL whose amplitude (V0) is controlled by Vbg and e − sgn(CtgVtg + )[13], where Vtg (Vbg) is the top- gate (back-gate) voltage, Ctg (Cbg) is the capacitance per unit area of top-gate (back-gate), e is the electronic charge and vF is the Fermi velocity.) (Details of calcu- Vtg. (V0 = √π¯hvF (sgn(CbgVbg)(cid:113)CbgVbg CbgVbg)(cid:113)CtgVtg+CbgVbg e 2 four-probe resistance (Rxx) plateau is given by Rxx =(cid:40) h e2 h e2 Nνtg−νbg N (νtg+νbg) νtgνbg νtgνbg νtgνbg > 0 νtgνbg < 0 (I) (II) where N is the number of top-gates which is 37 in our device. (Landau-Buttiker formalism to obtain Rxx for multiple top-gates in Section V of Supplemental Mate- rial.) Figure 2(a) shows the colorscale plot of zero bias four- probe resistance as a function of Vtg and Vbg at 14 T. We observe diamond shaped regions in the parameter space that represent integer filling factors in adjacent regions set by Vtg and Vbg. The filling factors in the two alter- nating regions are indicated as (νtg,νbg). Figure 2(b, c) show line slices of Rxx as function of νbg at νtg = 2, 6. (Line slices at νtg = 10 and 14 are presented in Section IV of Supplemental Material.) We do not observe large re- sistance as predicted by Equation I and II. The Rxx is relatively high, and does not show a plateau, in the bipo- lar regime, but not as high as predicted by Equation II. In the line slices (Figure 2(b, c)), the green curve is the experimental data. The black dashed lines correspond to the calculated plateau for N = 1 in Equation I and II. We find that in the unipolar regime, denoted by the blue region in Figure 1(d), the potential V0 is small and νtg−νbg the plateaus are well described by Rxx = h ; νtgνbg e2 there is good agreement between measured experimental data and expected plateau values for a single top-gate in Figure 2(b,c). Our experimental observation show plateaus corre- sponding to a single top-gate and not 37 top-gate, and so we do not have equilibration in our device inside the su- perlattice. The equilibration occurs only at the extreme edge of the top-gates near the voltage probes as illus- trated in the schematic in Figure 2(d). In the schematic, the yellow probes are the real probes and the gray probes are the virtual probes denoting equilibration at that edge. The virtual voltage probes are used to calculate resis- tance using Landauer-Buttiker formalism [22]. One pos- sible reason for the lack of equilibration is that the edge states under the top-gate are defined electrostatically, where the potential varies smoothly due to the finite thickness of the top-gate dielectric [23]. Equilibration requires inter edge state scattering or ohmic contacts so that all the edge states are at the same chemical poten- tial, and this does not happen due to the short length along the physical edge of graphene [24 -- 27]. In the bipolar region, plateau coincides with that ex- pected for a single top-gate when νtg = 2(-2) and νbg = -2(2). When we have one edge state for both electrons and holes circulating in adjacent regions, the resistance plateau is seen at h/e2 (25.8 kΩ). The plateau at (νtg,νbg) = (2,-2) results not only because the E is low in this state compared to any other state in the bipolar region but also because this is a special state where the charge carriers belong to the same LL (n = 0) and that the LL gap is maximum between n = 0 and n = 1 LL. Narrow TG and FIG. 1. Device geometry and different regimes of edge state transport. (a) Schematic of a device. (b) False colored scan- ning electron microscope image of a device with a zoomed-in image of the finger like top gates. (c) Depiction of the peri- odic 1D potential and the influence of the two gates in BG and TG regions. (d) Parameter space of Vtg and Vbg showing the different type of charge carriers in adjacent regions. (e) Line plot of resistance as function of Vbg when Vtg is biased in charge neutral region at 0 T and 2 K. (f) Line plot of re- sistance at 0 T and 2 K when TG regions are n-doped. (g) Schematic depicting edge state transport in the unipolar re- gion with νbg greater than νtg, and the bipolar region. The number of top-gates is four in this schematic. The BG region is denoted in purple and the TG region is shaded in blue. lation and plot of V0 as a function of Vbg and Vtg is in Section II of Supplemental Material.) (lB =(cid:113) ¯h We measure zero-bias four-probe longitudinal resis- tance (Rxx) while varying gate voltages at different mag- netic fields (B) at a temperature of 2 K. The charge neu- tral point is at Vtg = -0.1 V and Vbg = -2 V (Figure 1(e)) suggesting low unintentional doping. The mean free path in our device is ∼ 70 nm and phase coherence length is ∼ 600 nm at 2 K [19]. As B increases, the magnetic length eB ) decreases and the charge carriers encounter smaller periods of SL until they are confined within a sin- gle BG or TG region. Well resolved LLs start to appear only beyond 2 T (see Supplemental Material Section III for complete LL fan diagram). In this work we look at quantum Hall effect in 1D SL at λ = 22 and vary V0 lB upto ∼ 375 meV. The resistance depends on the filling factor ν = nh/Be in the adjacent regions [16, 17, 20, 21]. At a constant magnetic field B, the filling factor in the adjacent regions is determined by Vbg and Vtg. Depending on Vbg and Vtg at a given B, we have same type of edge states in the unipolar region, and in the bipolar region, electron and hole edge states co-propagate along the junction (Fig- ure 1(g)). In this device, we have 37 top-gates. So, if the edge states are formed under all the top-gates and they equilibrate along all the edges of the top-gates, then the 3 FIG. 2. Longitudinal resistance (Rxx) at B = 14 T when LLs are well resolved in three different case of edge state trans- port. (a) Rxx as function of Vtg and Vbg at 14 T at a tem- perature of 2 K. The white lines denote line of constant νtg. Rxx as function of νbg for (b) νtg = 2 and (c) νtg = 6. The green curves in (b), (c) denote experimental value with the black dashed lines representing the resistance plateau in case of equilibration of edge states for a single top-gate. ((νtg,νbg) values are denoted within the parenthesis.) (d) Schematic depicting equilibration along the extreme edges in unipolar region. The yellow probes are the real contacts and the gray probes are the virtual voltage probes. BG regions inside the superlattice leads to leaky barrier and transmission of edge states resulting in a plateau corresponding to a single top-gate [1]. However, when neighboring regions have other edge states, that is, (νtg × νbg) < 0 and νtg or νbg is greater than 2, we find resistance significantly larger than h/e2 (for example, (2,-6) in Figure 2(b), (6,-2) in Figure 2(c)); a feature not generally seen in single p-n'-p junction in the quantum Hall state. The feature of plateaus in Rxx corresponding to equilibration for a single top-gate in the unipolar region and deviation from this picture in the bipolar region, is quite robust and is also observed in another device with 35 top-gates (details in Section X of Supplemental Material). The large resistance with maximum Rxx ∼ 200 kΩ at 14 T, is seen along the diagonal direction in Figure 2(a); it is precisely in this diagonal direction of the parameter space that V0 increases. In the bipolar region, though the value of Rxx is higher than that for a single top- gate, it is not 37 times the value for a single top-gate, and thus there is no full equilibration. In this region the array of top-gates does not behave as a single top-gate. The resistance value depends on the gate voltages and the magnetic field, implying that the equilibration can be modified by electric field. We now try to understand the reason for this obser- vation and note that V0 is relatively large in this re- gion. Figure 3(a) shows a LL diagram along the length FIG. 3. Effect of E on equilibration (a) Variation of LLs along the length of device when (νtg,νbg) = (-2,6). The brown region in the schematic denotes the junction between TG and BG region where E exists. (b) Spatial variation of potential and E along the length of graphene in bipolar region (Vbg = - 30 V and Vtg = 3 V). Yellow regions denote geometric position of top-gates. (c) Effective magnetic length as function of E at B = 14 T. (d) Contour of maximum E at 1.2×107 V/m and 1.4×107 V/m as function of Vbg and Vtg overlaid on measured Rxx in bipolar regime at 14 T. of the device when (νtg, νbg) = (−2, 6); where we have co-propagating electron and hole edge states at the junc- tion. At the interface of these regions (shaded brown in Figure 3(a)) there is an electric field (E), due to the SL, and it has a significant effect on the LL wavefunctions. To get an accurate idea of the magnitude of E in our devices we performed numerical simulation of the electro- statics using finite element method. At a given Vtg and Vbg, the charge carrier density induced along the length of graphene is calculated, from which potential and E is obtained. (Details of calculation in Section VII of Sup- plemental Material.) Spatial variation of potential and E along length of graphene in bipolar region is shown in Figure 3(b). At a given Vtg and Vbg, the maximum E is obtained which is higher in the bipolar region compared to the unipolar region. It has been shown by Lukose et al. We note that in the bipolar regime, V0 created is much larger than ¯hωC. Large V0 leads to large E in the re- gion between BG and TG region which modifies the LLs locally. [28] and [18] for the case of a top- later extended by Gu et al. gate geometry that the LL spectrum and the wavefunc- tions in crossed E and B are fundamentally modified in graphene, an aspect that is not observable in conven- tional 2DEGS semiconductors. LL wavefunction is mod- ified in two ways. Effective magnetic length in the pres- ence of E can be written as l(cid:48) B = lB/(1 − (E/vF B)2)1/4 l(cid:48) [28]. B increases with increasing E and rises rapidly when E approaches vF B as seen in Figure 3(c). Sec- ondly, the E mixes the Landau levels. Contours of maximum E at 1.2×107 V/m and 1.4×107 V/m are overlaid on the measured data at 14 T as shown in Figure 3(d). Figure 3(d) shows that the con- tours lie along the region where we have high resistance state and departure from the value for a single top-gate. We argue that in our device geometry, E created is high enough to modify LLs which is reflected in charge trans- port measurements. The spatial extent of the wavefunc- tion (l(cid:48) B) increases with increasing E and approaches the width of TG (or BG) region. This leads to increased spatial overlap of the wavefunction within TG (or BG) region of the superlattice resulting in enhanced scatter- ing. Secondly, as function of E, there is LL mixing [28] which results in a non zero matrix element essential to cause scattering and equilibration. Let us now examine the various lengthscales of our system that support this scenario. From Figure 3(b) we find that the effective electrostatic width of the top-gate for this configuration is ∼50 nm and the extent of the region with high electric field (between 1.2 - 1.4×107 V/m) is ∼15 nm. In addi- tion, the magnetic length in the presence of transverse electric in this region is ∼11 nm (see Figure 3(c)). If one considers the scenario of (6,-2) state, we find that with three edge states with width 11 nm will have significant overlap with two adjacent regions and equilibration will be enhanced due to scattering. In the presence of high electric field, the resistance de- pends on E and B applied. For example, resistance val- ues at (νtg,νbg) = (6,-2) and (2,-6) at 14 T are different as the maximum E is different in the two regions (Plot of maximum E as a function of gate voltages in Section VII of Supplemental Material) -- this strongly suggests elec- tric field tuning of equilibration. We think disorder does not play an important role. 4 Because, in our sample, the disorder potential, estimated from the FWHM of the Dirac peak, is 71 meV [13], and thus is smaller than V0 in the bipolar region. So, the tuning of equilibration of edge states in our sample, which is seen at higher V0, is due to the E at the interface of TG and BG regions. However, recent work of Kumada et al suggests that the disorder along the length of p- n junction could play an important role in equilibration [29, 30]. Further experimental and theoretical studies need to be carried out to probe the role of disorder in periodically modulated structures. Details of LL modification in unipolar region at lower B of 3.5 T in Section IX of Supplemental Material. Our experiments with tunable superlattices suggest that tuning of the equilibration of edge states in graphene can be done using E at interfaces, which cannot be real- ized in conventional 2DEGS. In addition, the nature of the state that emerges after the collapse of the LLs is little understood and possibility of existence of correla- tions has been speculated [28, 31]. The close proximity of co-propagating electron and hole edge states can be used to construct large class of topological states [32] and also offers an opportunity to study excitonic effects, this has been recently explored in bilayer quantum Hall systems [33]. There have been predictions of correlated states in ν = 0 LL of graphene and LL mixing can enable explo- ration of such phases [34]. We thank Marcin Mucha-Kruczy´nski, G.Baskaran, R.Shankar, Jainendra Jain, Vibhor Singh, Shamashis Sengupta and K.Sengupta for discussions and comments on the manuscript. We acknowledge Swarnajayanthi Fel- lowship of Department of Science and Technology and Department of Atomic Energy of Government of India for support. [1] G. Muller, D. Weiss, K. von Klitzing, P. Steda, G. Weimann, Physical Review B 51, 10236 (1995). and [2] M. Tornow, D. Weiss, A. Manolescu, R. Menne, K. v. Klitzing, and G. Weimann, Physical Review B 54, 16397 (1996). [3] P. D. Ye, D. Weiss, R. R. Gerhardts, K. von Klitzing, K. Eberl, and H. Nickel, Surface Science Proceedings of the Eleventh International Conference on the Electronic Properties of Two-Dimensional Systems, 361/362, 337 (1996). [4] H. L. Stormer, L. N. Pfeiffer, K. W. Baldwin, K. W. West, and J. Spector, Applied Physics Letters 58, 726 (1991). [5] A. Endo, M. Kawamura, S. Katsumoto, and Y. Iye, Physical Review B 63, 113310 (2001). [6] C. L. Yang, J. Zhang, R. R. Du, J. A. Simmons, and J. L. Reno, Physical Review Letters 89, 076801 (2002). [7] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Reviews of Modern Physics 81, 109 (2009). Reviews of Modern Physics 83, 407 (2011). [9] M. O. Goerbig, Reviews of Modern Physics 83, 1193 (2011). [10] C. R. Dean, L. Wang, P. Maher, C. Forsythe, F. Ghahari, Y. Gao, J. Katoch, M. Ishigami, P. Moon, M. Koshino, T. Taniguchi, K. Watanabe, K. L. Shepard, J. Hone, and P. Kim, Nature 497, 598 (2013). [11] L. A. Ponomarenko, R. V. Gorbachev, G. L. Yu, D. C. Elias, R. Jalil, A. A. Patel, A. Mishchenko, A. S. Mayorov, C. R. Woods, J. R. Wallbank, M. Mucha- Kruczynski, B. A. Piot, M. Potemski, I. V. Grigorieva, K. S. Novoselov, F. Guinea, V. I. Fal'ko, and A. K. Geim, Nature 497, 594 (2013). [12] 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). [13] S. Dubey, V. Singh, A. K. Bhat, P. Parikh, S. Grover, R. Sensarma, V. Tripathi, K. Sengupta, and M. M. Desh- mukh, Nano Letters 13, 3990 (2013). [8] S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, [14] M. Kawamura, A. Endo, S. Katsumoto, Y. Iye, C. Ter- akura, and S. Uji, Physica B: Condensed Matter Inter- national Conference on High Magnetic Fields in Semi- conductors, 298, 48 (2001). [15] M. Killi, S. Wu, and A. Paramekanti, International Jour- nal of Modern Physics B 26, 1242007 (2012). [16] B. Ozyilmaz, P. Jarillo-Herrero, D. Efetov, D. Abanin, L. Levitov, and P. Kim, Physical Review Letters 99, 166804 (2007). [17] D.-K. Ki and H.-J. Lee, Physical Review B 79, 195327 (2009). [18] N. Gu, M. Rudner, A. Young, P. Kim, and L. Levitov, Physical Review Letters 106, 066601 (2011). [19] F. V. Tikhonenko, D. W. Horsell, R. V. Gorbachev, and A. K. Savchenko, Physical Review Letters 100, 056802 (2008). [20] D. A. Abanin and L. S. Levitov, Science 317, 641 (2007). [21] J. R. Williams, L. DiCarlo, and C. M. Marcus, Science 317, 638 (2007). [22] M. Buttiker, Physical Review B 38, 9375 (1988). [23] F. Amet, J. Williams, K. Watanabe, T. Taniguchi, and D. Goldhaber-Gordon, Physical Review Letters 112, 196601 (2014). [24] C. L. Kane and M. P. A. Fisher, Physical Review B 52, 17393 (1995). [25] B. W. Alphenaar, P. L. McEuen, R. G. Wheeler, and R. N. Sacks, Physica B: Condensed Matter Analogies in Optics and Micro-Electronics, 175, 235 (1991). [26] K. Ensslin, Superlattices and Microstructures 33, 425 (2003). [27] R. J. Haug, Semiconductor science and technology 8, 131 (1993). [28] V. Lukose, R. Shankar, and G. Baskaran, Physical Re- view Letters 98, 116802 (2007). [29] N. Kumada, F. D. Parmentier, H. Hibino, D. C. Glattli, and P. Roulleau, Nature Communications 6, 8068 (2015). [30] W. Long, Q.-f. Sun, and J. Wang, Physical Review Let- ters 101, 166806 (2008). [31] P. Carmier, C. Lewenkopf, and D. Ullmo, Physical Re- view B 84, 195428 (2011). [32] E. Sagi and Y. Oreg, Physical Review B 90, 201102(R) (2014). [33] D. Nandi, A. D. K. Finck, J. P. Eisenstein, L. N. Pfeiffer, and K. W. West, Nature 488, 481 (2012). [34] M. Kharitonov, Physical Review B 85, 155439 (2012). SupplementalMaterial:QuantumHalleffectintunable1-Dlateralsuperlatticeingraphene -- roleofcrossedelectricandmagneticfieldsSudiptaDubey1,a)andMandarMDeshmukh1,b)DepartmentofCondensedMatterPhysicsandMaterialsScience,TataInstituteofFundamentalResearch,HomiBhabhaRoad,Mumbai400005,Indiaa)sudipta.tifr@gmail.comb)deshmukh@tifr.res.in12I.DEVICEFABRICATIONGrapheneismechanicallyexfoliatedondegeneratelydopedsilicon(Si++)substratehaving300nmofsilicondioxide.Si++actsasglobalback-gate,thatis,ittunesthechargecarrierdensityoftheentiregrapheneflake.Standarde-beamlithographyisusedtopatternCr/AucontactsinHallbargeometry.ThedeviceisthenthermallyannealedtoremoveresidualPMMAfromgraphene.3nmofAl2O3isthendepositedusinge-beamlithographywhichactsasseedlayerforsubsequentdepositionof20nmAl2O3usingatomiclayerdeposition.Al2O3actsastop-gatedielectricabovewhichnarrow(∼27nmwide),periodicfingersofPdarepatternedusinge-beamlithography.Allthefingersoftop-gateareconnectedtooneelectrodeoutsidethegrapheneflakesothatallthefingersareatthesamepotential.Theperiodofthesefingersis150nm.II.VARIATIONOFAMPLITUDEOFSUPERLATTICEPOTENTIALWITHBACK-GATEANDTOP-GATEVOLTAGEInourdevicetherearetwoalternatingregionsingraphene-onewithoutatop-gate(BGregion),wherethechargecarrierdensityiscontrolledonlybytheback-gate;andtheotherunderatop-gate(TGregion),wherethechargecarrierdensityiscontrolledbytheback-gateandthetop-gate.Theamplitudeofsuperlatticepotential(V0)createdisgivenbythedifferenceofdopingintheBGandTGregion.Incaseofnounintentionaldoping,thenumberdensityofchargecarriersintheBGregionisgivenasnbg=CbgVbge;andintheTGregion,thenetnumberdensityofchargecarriersisthealgebraicsumofthedensityofcarriersinducedbythetwogatesntg=CtgVtg+CbgVbge;whereCbg(Ctg)isthecapacitanceperunitareabetweengrapheneandback-gate(top-gate)andVbg(Vtg)istheback-gate(top-gate)voltage.UsingparallelplatecapacitorgeometryCbgisgivenbyϵ0ϵr/d,whereϵ0=8.85×1012F/m,ϵr=3.9andd=300nm.CtgisobtainedfromtheslopeofchargeneutrallineintheplotofresistanceasfunctionofVbgandVtgatzeromagneticfield(FigureS1(a)).Fromtheslope,wegetCtg=18×Cbg.Incaseofgraphene,duetoitslineardispersion,EF=√πvF√n,wherevFistheFermivelocityandEFistheFermienergy,withall3energiesmeasuredrelativetothechargeneutralitypoint1.ThusV0isgivenbyV0=√πvF(sgn(CbgVbg)√CbgVbge−sgn(CtgVtg+CbgVbg)√CtgVtg+CbgVbge)(1)MagnitudeofV0isreferredasV0sinceitrepresentstheamplitudeofsuperlatticepotential.FigureS1(b)showsthevariationofV0withVbgandVtg.-40-2002040420-2-4Vbg (V)Vtg (V)0.33.5Rxx (kΩ)-40-2002040-4-20240375Vbg (V)Vtg (V)V0 (meV)(b)(a)FIG.S1.Amplitudeofsuperlatticepotential(a)ResistanceasfunctionofVtgandVbgatzeromagneticfieldattemperatureof2K.Whitearrowpointstothechargeneutralline.(b)ContourplotofV0asafunctionofVbgandVtg.V0ishigherinthebipolarregionwherethereisaseriesofp-n'junctionscomparetotheunipolarregion.III.QUANTUMHALLINTHEABSENCEOFSUPERLATTICEPOTENTIALFigureS2(a)showsthemagnitudeofHallresistanceasfunctionofVbgandmagneticfield(B)whenVtgisbiasedatchargeneutralpoint(Vtg=0.1).Thefandiagramcorrespondstomonolayergraphene.FigureS2(b)isalineplotat10Tshowingplateauatν=2(≡12.9kΩ),ν=6(≡4.3kΩ)andν=10(≡2.5kΩ),whereνisfillingfactor.WecanobservetheLandaulevelsfrom2T.IV.OBSERVEDRESISTANCEFORDIFFERENTFILLINGFACTORSINTGANDBGREGIONThefillingfactorisgivenbyν=nh/Be,wherenisthechargecarrierdensity,histhePlanck'sconstant,Bisthemagneticfieldandeistheelectroniccharge.ForconstantB,4-40-20020401086420Vbg (V)B (T)036912Hall resistance (kΩ)Vbg (V)Hall resistance (kΩ)(a)(b)ν = 2ν = 10ν = 6-40-20020401471013FIG.S2.(a)MagnitudeofHallresistanceasafunctionofVbgandB(b)LineplotofmagnitudeofHallresistanceasfunctionofVbgat10Tshowingplateausatν=±2,±6,+10.fillingfactorinBGregion(νbg)iscontrolledbyVbg,andfillingfactorinTGregion(νtg)iscontrolledbybothVtgandVbg.Lineslicesshowinglongitudinalresistance(Rxx)asfunctionofνbgforνtg=2and6areshowninFigure2inmaintext.Forνtg=10and14,wemeasuredRxxasfunctionofVtgandVbg(FigureS3(a))atalowerfieldof8T.FigureS3(b,c)showsRxxasfunctionofνbgatνtg=10and14.Inthelineslices(FigureS3(b,c)),thegreencurveistheexperimentaldatawiththeblackdashedlinemarkingthecalculatedplateauincaseofedgestateequilibrationforasingletop-gate.Plateauscoincidewiththeblackdashedlinesintheunipolarregionbutdeviatesfromtheminthebipolarregion.-40-2002040420-2-40255075100Rxx (kΩ)Vbg (V)Vtg (V)5040302010151050-5-10νtg = 10(8 T)νbgRxx (kΩ)40302010151050-5-10νbgRxx (kΩ)νtg = 14(8 T)(a)(b)(c)B = 8 T(10,6)(10,14)(10,2)(10,-2)(10,-6)(10,-10)(14,2)(14,-2)(14,-6)(14,6)(14,10)(νtg,νbg) (νtg,νbg) FIG.S3.(a)Longitudinalresistance(Rxx)asfunctionofVtgandVbgat8T.Rxxasfunctionofνbgfor(b)νtg=10and(c)νtg=14.Thegreencurvesin(b),(c)denoteexperimentalvaluewiththeblackdashedlinesrepresentingtheresistanceplateauincaseofequilibrationofedgestatesforasingletop-gate.((νtg,νbg)valuesaredenotedwithintheparenthesis.)5V.RESISTANCEINCASEOFEQUILIBRATIONOFEDGESTATESTheschematicofourdevicegeometryisshowninFigureS4.Theyellowprobesdenoterealcontactsusedtomeasuretheresistances.TheTGregionsaredenotedinblueandBGregionsinpurple.BGTGBGTGBGTGBGTGBGTGBGTGBG123456FIG.S4.Schematicshowingourdevicegeometrywithmultipletop-gates.Inthisschematic,sixtop-gatesareshown.Yellowprobesdenoterealcontacts.Longitudinalresistanceismeasuredbetweenvoltageterminal2and3,andHallresistancebetweenvoltageterminal2and6.WenowdescribetheresistanceobservedwhenballisticconductionoccursviaedgestatesusingtheLandau-Buttikerformalism2.Wepresentthecalculationfortwotop-gatessincethematrixsizeincreaseswithincreasingnumberoftop-gates.TheprocedurecanbegeneralizedforNtop-gates,whereNis37inourdevicegeometry.Theschematic(FigureS5)(a)showstwotop-gatesandweassumeequilibrationamongedgestates.Theassumptionofequilibrationisincorporatedinourcalculationusingvirtualvoltageprobewhichenforcesalltheedgestatestobeatthesamechemicalpotentialinthatterminal.Therealvoltageprobesaremarkedinyellow.Thegrayprobesdenotevirtualvoltageterminalwhichenforcesequilibrationalongtheelectrostaticedgeofthetop-gates.νbgisgreaterthanνtgandtheadjacentregionhavesamepolarityofchargecarriers.Bycountingthenumberofcurrentcarryingchannelsthatstartfromterminalpandendinterminalq,transmissionfunctionTpqisobtained.TheconductancematrixGpqisgivenbymultiplyingthefollowingmatrixwithe2/h,61214798345613121110νbg(BG)νbg(BG)νbg(BG)νtg(TG)νtg(TG)1214798345613121110(b)νbg(BG)νbg(BG)νbg(BG)νtg(TG)νtg(TG)(a)FIG.S5.Schematicshowingtwotop-gateswiththeyellowprobesastheactualcontactsusedtomeasureRxxandRxy.Grayprobesarethevirtualvoltageterminalsusedtoenforceequilibrationoftheedgestatesalongtheelectrostaticedgeofthetop-gate.(a)Equilibrationwhenνbgisgreaterthanνtgintheunipolarregion.(b)Equilibrationalongalltheelectrostaticedgesofthetwotop-gates,withνbgandνtghavingoppositepolarityinthebipolarregion.7AAAApq123456789101112131410000000000000νbg2νbg000000000000030νbg000000000000400νtg00000000νbg−νtg005000νbg000000000060000νtg0000νbg−νtg0000700000νbg000000008000000νbg000000090000000νbg0000001000000000νbg00000110000νbg−νtg0000νtg0000120000000000νbg0001300νbg−νtg00000000νtg0014000000000000νbg0(2)ThecurrentataterminalpisgivenbyIp=∑qGpq(Vp−Vq).Alloftheseequationsarenotindependent,so,voltageatthedrainterminal(terminal8)ischosentobezero,andweomittherowandthecolumncorrespondingtothisterminal.Invertingtheresultingmatrix,voltagesattheterminalsareobtained.Also,thecurrentatthevoltageterminalsarezero.Thetwoprobevoltage(V2−probe)isgivenby(V1−V8),thefour-probelongitudinalvoltage(Vxx)isgivenby(V2−V7)andtheHallvoltage(Vxy)isgivenby(V2−V14).DividingbycurrentI1,theresistancesareobtained.Usingtheaboveformalism,R2−probe=e2h2νbg−νtgνbgνtg,Rxx=2e2hνbg−νtgνbgνtg,Rxy=e2h1νbg.(3)Theabovecalculationisdonefortwotop-gates.Wehavenumericallycalculatedresis-tancesforNtop-gatesusingtheaboveformalismandobtained,R2−probe=e2hNνbg−(N−1)νtgνbgνtg,Rxx=Ne2hνbg−νtgνbgνtg,Rxy=e2h1νbg.(4)8Similarly,forνtggreaterthanνbgforNtop-gates,R2−probe=e2hNνtg−(N−1)νbgνbgνtg,Rxx=Ne2hνtg−νbgνbgνtg,Rxy=e2h1νbg;(5)Theschematicforelectronedgestatesandholeedgestatesintheadjacentregion,thatis,νtgandνbghavingoppositepolarityfortwotop-gatesisshowninFigureS5(b).Thevirtualgrayterminalensuresequilibrationamongelectronandholeedgestates.Theabovementionedprocedurecanbefollowedtoobtaintheresistanceas,R2−probe=e2h(N+1)νtg+Nνbgνbgνtg,Rxx=Ne2hνtg+νbgνbgνtg,Rxy=e2h1νbg,(6)whereNisthenumberoftop-gates.Intheunipolarregion,theobservedRxxfor37top-gatescorrespondstotheequilibrationincaseofasingletop-gate.Onepossiblereasonisthenarrowregioningraphenedefinedelectrostaticallywhichfacilitatesthetransmissionofedgestatesalongthelengthofgraphenewithoutequilibrationwithlocallycirculatingedges.TheschematicofthiscaseisshowninFigureS6wherethegrayprobesalongtheextremeedgesdenoteequilibrationalongthosetwoparticularedge.Anotherreasoncouldbethattheextremeedgesofthetop-gatearenear(∼300nm)theactualvoltageprobewhichhelpsintheequilibrationalongthoseedges.127935610νbgνbg48FIG.S6.Schematicwithyellowprobesastherealcontactsandthegrayprobesasthevirtualvoltageterminalsdenotingequilibrationofedgestatesalongthatedge.Schematicshowingequili-brationalongtheextremeedgesofthetop-gateintheunipolarregionwhereνbgisgreaterthanνbg.9VI.VARIATIONOFHALLRESISTANCEWITHGATEVOLTAGESLackofequilibrationordeviationfromLandau-ButtikerpictureisalsoobservedintheHallresistance(Rxy).Inourdevicegeometry,theelectrodesmeasuringRxyisoutsidethearrayoffingergatesandhencetheedgestateequilibrationgivesRxytobeafunctionofonlyνbgasseenintheprevioussection.FigureS7(a)showsRxymeasuredasfunctionofVbgandVtgat14T.Ahorizontallinesliceshowingplateausatνbg=±2,±6isshowninFigureS7(b).FigureS7(c,d)showsRxyasfunctionofVtgforνbg=2,6.RxybeingafunctionofonlyνbgshouldbeindependentofVtg.Weobservethatwhentherearesametypeofchargecarrierintheadjacentregion,thatis,forpositiveνbg(positiveVtg),Rxyisalmostconstant.However,whenwehaveelectronsandholesintheadjacentregion(positiveνbgandnegativeVtg),Rxydoesnotremainconstant.Aswegofromp-p'(orn-n')top-n',Rxy,fromitsconstantvalueshowsadipandthenincreasestowardtheconstantvalue.Thisfurthersupportsourpicturethatthereisenhancedscatteringintheelementsofthejunction.VII.NUMERICALSIMULATIONOFELECTROSTATICSUSINGCOMSOLWehavemodeledourdevicegeometryinCOMSOLsoftwareandobtainedtheelectricfieldfromnumericalsimulationoftheelectrostatics.Inthesimulation,weused15top-gatesalongthelengthofgrapheneandobtainedtheinducedchargecarrierdensityingrapheneforagivenback-gateandtop-gatevoltage.Thisinducedchargecarrierdensityisperiodicalongthelengthofgraphenefromwhichthepotentialisobtained.Theelectricfieldintheplaneofgrapheneisobtainedbytakingthenumericalderivativeofthepotential.ThepotentialandtheelectricfieldvariationalonglengthofgrapheneforVbg=-30VandVtg=3VinthebipolarregionareshowninFigureS8(a,b).Inthesimulation,thegeometricalwidthofthetop-gateelectrodeis30nm,andthetop-gatesare150nmapart,similartoourdevicegeometry.Duetothefinitethicknessoftop-gatedielectric,thereisaspreadoftheelectricfieldlinesfromthetop-gates.So,electrostatically,theeffectivewidthofthetop-gatedregioningrapheneislargerthan30nm.Forclarity,magnifiedimageswith3top-gatesalonglengthofgrapheneareshowninunipolarregion(FigureS8(c))andbipolar10-4-20241011121314Rxy (kΩ)Vtg (V)νbg = 2Vbg = 9.6 V-4-202423456Vtg (V)Rxy (kΩ)νbg = 6Vbg = 27.5 V-40-2002040420-2-41471013Vbg (V)Vtg (V)Rxy (kΩ)(a)(b)(c)-40-20020401471013Vbg (V)Rxy (kΩ)(d)νbg = 2νbg = 6νbg = -6νbg = -2Vtg = -0.1 VFIG.S7.Transverseresistance(Rxy)at14Tatatemperatureof2K.(a)MagnitudeofRxyasafunctionofVtgandVbg.(b)MagnitudeofRxyasafunctionofVbgwhenVtg=-0.1V.Plateausatνbg=±2and±6areobserved.RxyasfunctionofVtgat(c)νbg=2and(d)νbg=6.Inourdevicegeometry,electrodesmeasuringRxyisoutsidethearrayoffingergatesandhenceRxyshouldbeafunctionofonlyνbg,thatis,itshouldbeconstantwithVtg.However,RxydecreasesfromitsconstantvaluewhenVtgissuchthatthereisoppositetypeofchargecarrierinadjacentregion.region(FigureS8(d)).Weobservethemagnitudeofelectricfieldtobelargerinbipolarregioncomparedtotheunipolarregion.Anotherinterestingthingtonoteisthatnotonlythemagnitudebutalsotheshapeofthepotentialisdifferentinthetwocases.Numericalsimulationofspatialvariationofelectricfieldalonglengthandwidthofgrapheneinbipolarregion(Vbg=-30VandVtg=3V)isshowninFigureS9(a).Electricfieldvariesperiodicallyalongthelengthofgraphenewithmaximaandminimaattheedgeofthetop-gate,andisconstantalongthewidthofgraphene.Fromthesimulation,themax-11-0.15-0.10-0.050.000.050.102000150010005000Length of graphene (nm)Potential (eV)151050-5-10-152000150010005000Length of graphene (nm)Electric field (x 106 V/m)0.280.260.240.220.2013001200110010009008003210-1-2Length of graphene (nm)Electric field (x 106 V/m)Potential (eV)-0.15-0.10-0.050.000.050.101300120011001000900800151050-5-10-15Potential (eV)Electric field (x 106 V/m)Length of graphene (nm)(a)(b)(c)(d)FIG.S8.Numericalsimulationwith15top-gatesalonglengthofgrapheneusingCOMSOL.(a)VariationofpotentialalonglengthofgraphenewhenVtg=3VandVbg=-30V.(b)SpatialvariationofelectricfieldwhenVtg=3VandVbg=-30V.Magnifiedimageofspatialvariationofpotentialandelectricfieldin(c)unipolarregionwhenVtg=3VandVbg=30Vand(d)bipolarregionwhenVtg=3VandVbg=-30V.Yellowregionsdenotegeometricpositionoftop-gates.imumelectricfieldforagivenVbgandVtgisobtained.FigureS9(b)showsthemaximumvalueofelectricfieldasafunctionofVbgandVtginthebipolarregion.VIII.EFFECTOFELECTRICFIELDONEQUILIBRATIONAsseenintheprevioussection,thepotentialandtheelectricfieldishighinbipolarregionwherewehaveaseriesofp-n'junctionscomparedtounipolarregimewherethereisseriesofn-n'(orp-p')junctions.Intheunipolarregion,thereislackofequilibrationofedgestatesalongtheelectrostaticedgeandthearrayoftop-gatesbehaveasasingletop-gate.Theresistanceplateausdependsonlyonthecombinationoffillingfactors,thatis,resistanceplateauat(νtg,νbg)=(2,6)and(6,2)aresame.Inthebipolarregion,theelectricfieldmodifiesthislackofequilibrationalongtheelectrostaticedgesuchthattheresistanceplateaudoesnotdependonlyonthecombinationoffillingfactorintheadjacentregion.TheelectricfieldcausesmodificationofLandaulevelwavefunction.Withincreasingelectricfield120300600900120010008006004002000Length along graphene (nm)Width of graphene (nm)-101E (x 107 V/m)(a)-40-30-20-104.03.02.01.0Vbg (V)Vtg (V)161116E (x 106 V/m)(b)TGBGFIG.S9.Numericalsimulationofelectricfieldinourdeviceinthebipolarregion.(a)SpatialvariationoftheelectricfieldalongthelengthandwidthofgraphenewhenVtg=3VandVbg=-30V.TheeffectiveelectrostaticwidthsoftheTGandtheBGregionsforthisgatevoltagearemarkedinwhite.(b)MaximumelectricfieldasfunctionofVtgandVbg.thespreadofthewavefunctionincreasessuchthatitisoftheorderofthewidthofTGorBGregion,resultinginenhancedscatteringacrossLandaulevels.ThiseffectismanifestedinchargetransportwhenelectricfieldbecomescomparabletomagneticfieldtimestheFermivelocity.Variationofeffectivemagneticlengthat8TasafunctionofgatevoltagesisshowninFigureS10.-40-27-14-14.02.71.40.1Vbg (V)Vtg (V)1016222834lB' (nm)FIG.S10.Effectivemagneticlengthat8TasafunctionofVtgandVbg.Theeffectivemagneticlengthiscalculatedfromthemaximumelectricfieldobtainedfromthesimulationasshownintheprevioussection.Inthebipolarregion,thevalueofresistancedependsontheelectricandmagneticfieldinthatregion.Forexample,resistancevalueat(νtg,νbg)=(2,-6)and(6,-2)aredifferentas13thepotentialandtheelectricfieldaredifferentinthesetworegions.Similarlyresistancevaluefor(νtg,νbg)=(6,-2)at14Tand8TaredifferentasshowninFigureS11(a,b).(14 T)5040302010086420-2-4νbgRxx (kΩ)νtg = 6(6,-2)(6,2)(νtg,νbg)50403020100151050-5-10νtg = 6(8 T)(νtg,νbg) (6,2)(6,10)(6,-2)(6,-6)(6,-10)νbgRxx (kΩ)(a)(b)FIG.S11.Dependenceofmagnetotransportonelectricfield.Rxxasfunctionofνbgforνtg=6at(a)14Tand(b)8T.Intheabsenceofequilibrationandhighelectricfield,resistancevaluefor(νtg,νbg)=(6,-2)isdifferentatdifferentmagneticfield.IX.EFFECTOFELECTRICFIELDONMAGNETOTRANSPORTINUNIPOLARREGIONIntheunipolarregion,theelectricfieldislowcomparedtothebipolarregion.So,toobservethemodificationofLandaulevelsbyelectricfieldintheunipolarregion,wemeasuredRxxasafunctionofgatevoltagesfrom2Tto3.5T.AtaconstantVbgof20V,RxxasafunctionofVtgshowsSdHoscillationsasmagneticfieldisincreasedabove3TasseeninFigureS12(a).WemeasuredRxxasafunctionofVtgandVbgat3.5T(FigureS12(b))intheunipolarregion.LineplotofRxxvsVbgatVtg=1VisshownbythegreencurveinFigureS12(c).AtaconstantBof3.5T,weobserveRxxtooscillateasfunctionofVbgforagivenVtg.Theoscillationsariseduetoperiodicdecreaseofback-scatteringastheFermienergymovesthroughdifferentLandaulevelsinBGregion.AtVtg=2.8V,theelectricfieldishighercomparedtoVtg=1V.SowetooklinesliceofRxxvsVbgatVtg=2.8V(orangecurveinFigureS12(c))andobservedthefadingofSdHoscillations.Sowithincreasingelectricfield,theLandaulevelscomeclosetoeachotherresultinginadecreaseinthemodulationinresistance.ThusthedecreaseinamplitudeofoscillationisconsistentwiththemodificationofLandaulevelswithincreasingelectricfield.14010203040432100.52.0Vbg (V)Vtg (V)Rxx (kΩ)0102030400.51.01.52.0Rxx (kΩ)Vbg (V)Vtg = 1 VVtg = 2.8 V(b)(c)7006005004003210 3.5 T 3.0 T 2.5 T 2.0 TRxx (Ω)Vtg (V)(a)FIG.S12.Effectofelectricfieldonresistanceoscillationsinunipolarregion(a)RxxasafunctionofVtgatVbg=20Vatdifferentmagneticfield.SdHoscillationsareobservedwhenmagneticfieldisincreasedabove3T.Theplotsatdifferentmagneticfieldsareoffsetforclarity.(b)RxxasafunctionofVtgandVbgat3.5Tintheunipolarregion.(c)LineplotofRxxasafunctionofVbgatVtg=1V(greencurve)and2.8V(orangecurve).AmplitudeofoscillationsdecreaseswithincreasingVtgandhenceincreasingelectricfield.X.MEASUREMENTSFROMANOTHERDEVICEInthissectionwepresentmeasurementsofanotherdeviceat14Twhichshowedsimilarmagnetotransportbehavior.Thisdevicehasalowmobilityof2009cm2/Vs,andtheratiooftop-gatecapacitancetoback-gatecapacitanceis11.Thenumberoftop-gatesinthisdeviceis35andtheperiodis150nm.FigureS13(a)showsthevariationofRxxasafunctionofVtgandVbgat14T.Alltheplateausarenotwellresolvedduetolowmobility.Dissipationlesstransportisobservedfor(νtg,νbg)=(2,2),(-2,-2)and(-6,-6).Themaximumresistanceinthisdevice,observedinthebipolarregion,is58kΩ.FigureS13(b)showslinesliceofRxxasafunctionofνbgatνtgof-2.Theblackdashedlinescorrespondtoedgestateequilibrationforasingletop-gate.Theexperimentalcurve(ingreen)coincideswiththeblackdashedlinesat(νtg,νbg)=(-2,-6).Thusthisdevicehaving35top-gatesshowstheplateauinRxxcorrespondingtoasingletop-gateintheunipolarregion.15-40-2002040420-2-451525354555Rxx (kΩ)Vbg (V)Vtg (V)2520151050-10-8-6-4-2024Rxx (kΩ)νbgνtg = -2(a)(b)(-2,-6)(-2,2)(νtg,νbg)(14 T)FIG.S13.Chargetransportmeasurementofanotherdeviceinthepresenceofamagneticfieldof14T.(a)RxxasafunctionofVtgandVbg.(b)Rxxasafunctionofνbgforνtg=−2.Resistanceplateausat(νtg,νbg)=(-2,-6)and(-2,2)forasingletop-gatearemarkedbyblackdashedlines.REFERENCES1A.H.CastroNeto,F.Guinea,N.M.R.Peres,K.S.Novoselov,andA.K.Geim,ReviewsofModernPhysics81,109(2009).2M.Buttiker,PhysicalReviewB38,9375(1988).

1605.06730

3

1605

Short-Channel Field Effect Transistors with 9-Atom and 13-Atom wide Graphene Nanoribbons

Bottom-up synthesized GNRs and GNR heterostructures have promising electronic properties for high performance field effect transistors (FETs) and ultra-low power devices such as tunnelling FETs. However, the short length and wide band gap of these GNRs have prevented the fabrication of devices with the desired performance and switching behaviour. Here, by fabricating short channel (Lch ~20 nm) devices with a thin, high-k gate dielectric and a 9-atom wide (0.95 nm) armchair GNR as the channel material, we demonstrate FETs with high on-current (Ion >1 uA at Vd = -1 V) and high Ion/Ioff ~10^5 at room temperature. We find that the performance of these devices is limited by tunnelling through the Schottky barrier (SB) at the contacts and we observe an increase in the transparency of the barrier by increasing the gate field near the contacts. Our results thus demonstrate successful fabrication of high performance short-channel FETs with bottom-up synthesized armchair GNRs.

cond-mat.mes-hall

cond-mat

Short-channel field effect transistors with 9-atom and 13-atom wide graphene nanoribbons Juan Pablo Llinas1, Andrew Fairbrother2, Gabriela Borin Barin2, Wu Shi4, Kyunghoon Lee1, Shuang Wu1, Byung Yong Choi1,3, Rohit Braganza1, Jordan Lear1, Nicholas Kau4, Wonwoo Choi4, Chen Chen4, Zahra Pedramrazi4, Tim Dumslaff6, Akimitsu Narita6, Xinliang Feng7, Klaus Müllen6, Felix Fischer5,8, Alex Zettl4,8, Pascal Ruffieux2, Eli Yablonovitch1,8, Michael Crommie4,8, Roman Fasel2,9, Jeffrey Bokor1* 1Dept. of Electrical Engineering and Computer Sciences, UC Berkeley, Berkeley, CA, USA 2Empa, Swiss Federal Laboratories for Materials Science and Technology, Dübendorf, CH 3Flash PA Team, Semiconductor Memory Business, Samsung Electronics Co. Ltd., Korea 4Dept. of Physics, UC Berkeley, Berkeley, CA, USA 5Dept. of Chemistry, UC Berkeley, Berkeley, CA, USA 6Max Planck Institute for Polymer Research, Mainz, DE 7Center for Advancing Electronics Dresden, TU Dresden, Dresden, DE 8Kavli Energy NanoSciences Institute at the University of California, Berkeley and the Lawrence Berkeley National Laboratory, Berkeley, CA, USA 8Dept. of Chemistry and Biochemistry, University of Bern, Freiestrasse 3, 3012 Bern, CH 1 *jbokor@eecs.berkeley.edu Bottom-up synthesized GNRs1–8 and GNR heterostructures9,10 have promising electronic properties for high performance field effect transistors (FETs)11 and ultra-low power devices such as tunnelling FETs12. However, the short length and wide band gap of these GNRs have prevented the fabrication of devices with the desired performance and switching behaviour13–16. Here, by fabricating short channel (Lch ~20 nm) devices with a thin, high-gate dielectric and a 9-atom wide (0.95 nm) armchair GNR as the channel material, we demonstrate FETs with high on-current (Ion >1 A at Vd = -1 V) and high Ion/Ioff ~105 at room temperature. We find that the performance of these devices is limited by tunnelling through the Schottky barrier (SB) at the contacts and we observe an increase in the transparency of the barrier by increasing the gate field near the contacts. Our results thus demonstrate successful fabrication of high performance short-channel FETs with bottom-up synthesized armchair GNRs. The electronic, optical and magnetic properties of GNRs can be engineered by varying their width and edge structure17–19. However, traditional methods to pattern GNRs, such as unzipping carbon nanotubes or lithographically defining GNRs from bulk graphene, yield GNRs with rough edges that degrade electronic transport20. Recent experiments have demonstrated bottom- up chemical synthesis of GNRs with uniform width and atomically-precise edges, in which the width and edge structure of the GNR is determined by the oligophenylene used in the polymerization step1,2,5,9,10. This synthetic uniformity produces GNRs with high structural and electronic homogeneity, which is required for integration of GNRFETs into large-scale digital circuits21. 2 To create high performance GNRFETs, we used 9-atom and 13-atom wide armchair GNRs (9AGNRs and 13AGNRS, respectively). With a predicted band gap of 2.10 eV for the isolated 9AGNR and 2.35 eV for the 13AGNR18, these are the narrowest band gap GNRs that have been synthesized on a surface with useful length for device fabrication (5AGNRs have a smaller band gap but are only ~10 nm long with current synthetic methods7). To synthesize the GNRs, the requisite monomer was evaporated onto a Au(111) surface under ultra-high vacuum and heated until it polymerized. Heating the substrate further causes individual polymers to planarize into GNRs (cyclodehydrogenation). The high quality of the GNRs is verified by high-resolution scanning tunnelling microscope (STM)1,2,4 imaging as shown in Fig. 1a,b. Fabrication of GNRFETs requires the transfer of GNRs from the Au growth surface onto an insulating surface and subsequent device fabrication steps, as shown in Fig. 1c and as described in the Methods. Unfortunately, standard imaging techniques (atomic force microscopy, scanning electron microscopy, transmission electron microscopy, etc.) were not useful in imaging single GNRs on insulating surfaces due to the GNR's small dimensions (~30 nm long, ~1 nm wide and <1 nm thick). Instead, we used Raman spectroscopy in order to verify that the structural integrity of the GNRs is maintained throughout the transfer and device fabrication process. As shown in Fig. 2a, the Raman spectrum with 785 nm wavelength excitation of the processed 9AGNRs looks identical to the spectrum taken of the as-grown 9AGNRs on Au. The presence of the radial breathing-like mode (RBLM) peak (311.5 cm-1) is evidence that the GNR width and edge structure is intact throughout device processing22,23. Unlike 9AGNRs, the RBLM is not visible for the 13AGNR spectrum for either 532 nm or 785 nm excitation wavelengths due to off- resonance of the excitation (Fig. 2b and Fig. S2). Still, the 13AGNRFETs were processed with 3 the same fabrication steps as the 9AGNRFETs and both types of devices exhibit similar transport characteristics (Fig. 3). First, we fabricated devices with a nominal 20 nm channel length and a 50 nm SiO2 gate dielectric as illustrated in Fig. 3a. Using the same fabrication methods we made two different types of samples: one with 9AGNRs and one with 13AGNRs. After patterning ~300 pairs of electrodes in the transferred GNR area, each defined channel was biased and tested for gate modulation of the current to find devices bridged by a GNR. Of the 300 devices, 28 devices and 29 devices were successfully fabricated for 9AGNR and 13AGNRs, respectively. This ~10% ratio of bridged contacts to open devices indicates that almost all of the devices found contain one GNR in the channel as demonstrated by Fig. S1. These 9AGNRFETs and 13AGNRFETs, as shown in Fig. 3 and Fig. S3, showed similar electrical behaviour due to their similar band gap. The presence of a SB at the Pd-GNR interface is evident by the non-linear behaviour at low bias in the Id –Vd characteristics, shown in Fig. 3a,b. To determine the contributions of thermionic vs tunnelling current across the SB, we measured the devices in vacuum at 77 K, 140 K, 210 K and 300 K. As demonstrated in Fig. 3 c,d, there is no significant change in the characteristics at these different temperatures for either 13AGNRFETs or 9AGNRFETs and the off-state current is at the gate leakage level (Fig. S4). The weak temperature dependence in the current-voltage characteristics suggests that the limiting transport mechanism is tunnelling through the barrier as opposed to thermionic emission over the barrier at the contacts. Furthermore, the ambipolar behaviour observed at low temperatures is only realistically possible with tunnelling contacts, since thermally activated current is suppressed for electrons in a semiconductor with a band gap of >2 eV. Tunnelling contacts with weak temperature dependence have been observed for carbon nanotube FETs and 4 other low-dimensional materials and verified via simulations24–26. Yet, the Ion = ~100 nA in the devices shown in Fig. 3 is too low for high-performance applications, so the transmission through the SBs must be enhanced to improve the current. Ionic liquid (IL) gating has been previously used to improve the transparency of the SBs in MoS2 27. Thus, we used the IL N,N-diethyl-N-(2-methoxyethyl)-N-methylammonium bis(trifluoromethylsulphonyl-imide (DEME-TFSI) to improve the electrostatic coupling between the gate electrode and the GNR channel, increase the field at the Pd-GNR interface and improve the transmission through the barriers. The Id –Vg behavior of a 9AGNRFET with IL gating is shown in Fig. 4b. This device shows clear enhancement in the on-current to ~200 nA at -0.2 V drain bias (as opposed to 3 nA at -0.4 V for the 50 nm SiO2 dielectric device presented in Fig 4a). The transistor also switches at smaller gate voltages due to the high gate efficiency of the IL. Since solid dielectric gates must be used for logic device applications, we fabricated scaled 9AGNR devices with a thin HfO2 gate dielectric (effective oxide thickness of around 1.5 nm) as shown in Fig. 5. Resembling the IL device, the local HfO2 back gate is more efficient at improving transmission through the SB than the thick SiO2 global back gate28. As demonstrated by the Id vs Vg shown in Fig. 5, the device exhibits excellent switching characteristics, Ion/Ioff ~105, and a high Ion ~1 A at Vd = -1 V. This corresponds to a GNR-width (0.95 nm) normalized current drive of ~1000 A/µm at -1 V drain bias, superior to previously reported top-down GNR devices29–31. Therefore, the scaled device structure with the improved gate efficiency allows for ultra-narrow bottom-up GNRs to outperform the narrow band gap top-down GNRs by mitigating the impact of the SBs on the contact resistance. 5 We thus successfully demonstrate high performance short-channel FETs with bottom-up synthesized armchair GNRs. These GNRFETs have excellent switching behaviour and on-state performance after aggressively scaling the gate dielectric. Bottom-up GNR devices are therefore good candidates for high-performance logic applications, especially with advances in densely aligned GNR synthesis32 as well as narrow band gap GNR growth7. Our methodology can be applied to other exotic device structures as well, such as tunnel FETs, which incorporate atomically precise GNR heterostructures9,10,12. 6 METHODS 9AGNR growth. 9AGNRs are synthesized from 3',6'-dibromo-1,1':2',1''-terphenyl precursor monomers.4 First, the Au(111)/mica substrate (200 nm Au; PHASIS, Geneva, Switzerland) is cleaned in ultra-high vacuum by two sputtering/annealing cycles : 1 kV Ar+ for ten minutes followed by a 470 °C anneal for ten minutes. Next, the monomer is sublimed onto the Au(111) surface at a temperature of 60-70 °C, with the substrate held at 180 °C. After 2 minutes of deposition (resulting in approximately half monolayer coverage), the substrate temperature is increased to 200 °C for ten minutes to induce polymerization, followed by annealing at 410 °C for ten minutes in order to cyclodehydrogenate the polymers and form 9-AGNRs. 13AGNR growth. 13AGNRs were synthesized using 2,2'-Di((1,1'-biphenyls)-2-yl)-10,10'- dibromo-9,9'-bianthracene building blocks.2 Similar to the 9AGNR substrate, the Au(111)/mica substrate (200 nm Au; PHASIS, Geneva, Switzerland) is cleaned in ultra-high vacuum by two sputtering/annealing cycles : 1 kV Ar+ for ten minutes followed by a 450 °C anneal for ten minutes. The monomer was sublimed at 222 °C onto the clean substrate held at room temperature. The sample was then slowly annealed stepwise to 340 °C to form 13AGNRs. Preparation of 50 nm SiO2 back gates. Using dry oxidation, 50 nm SiO2 was grown on heavily doped 150 mm silicon wafers. Alignment markers and large pads for electrical probing were patterned using standard photolithography and lift-off patterning of 3 nm Cr and 25 nm Pt. The wafer was then diced and individual chips were used for GNR transfer and further device processing. Preparation of 6.5 nm HfO2 local back gates. Using dry oxidation, 100 nm SiO2 was grown on 150 mm silicon wafers. The local back gates were lithographically patterned and dry etched into 7 the SiO2 followed by lift-off of 3nm Ti and 17 nm Pt.33 6.5 nm HfO2 was grown in an atomic layer deposition system at 135 °C. Alignment markers and large pads for electrical probing were patterned using standard photolithography and lift-off of 3 nm Cr and 25 nm Pt. The wafer was then diced and individual chips were used for GNR transfer and further device processing. GNR transfer and patterning of source-drain electrodes. GNR/Au/mica was floated in 38% HCl in water, which caused the mica to delaminate with the Au film remaining floating on the surface of the acid.10 The floating gold film was picked up with the target substrate, with the GNRs facing the dielectric surface. Subsequent gold etching in KI/I2 yielded isolated, randomly distributed GNRs with sub-monolayer coverage on the target substrate. After the GNR transfer, poly-methyl methacrylate (PMMA, molecular weight 950K) was spun on the chips at 4 Krpm and followed by a 10 min bake at 180 °C. ~300 source drain electrodes (100 nm wide, 20 nm gaps) were patterned using a JEOL 6300-FS 100 kV e-beam lithography system and subsequently developed in 3:1 IPA:MIBK at 5 °C. 10 nm Pd was deposited using e-beam evaporation and lift-off was completed in Remover PG at 80 °C. Raman characterization. Raman characterization of the 9AGNR was performed with a Bruker SENTERRA Raman microscope using a 785 nm diode laser with 10 mW power and a 50x objective lens, resulting in a 1-2 micrometer spot size. No thermal effects were observed under these measurement conditions and an average of 3 spectra from different points was made for each sample. Raman measurements of the 13AGNR were made with a Horiba Jobin Yvon LabRAM ARAMIS Raman microscope using 532 nm and 785 nm diode lasers with 10 mW power each and a 50x objective lens, resulting in a 1-2 micrometer spot size. No thermal effects were observed under these measurement conditions and an average of 5 spectra from different points was made for each sample. 8 Electrical characterization. Devices were first screened in ambient conditions using a cascade probe station and an Agilent B1500A parameter analyser. Vacuum and variable temperature measurements were then performed in a Lakeshore probe station. Ionic liquid devices were measured with a Vg sweep speed of 50 mV/s. 9 REFERENCES 1. Cai, J. et al. Atomically precise bottom-up fabrication of graphene nanoribbons. Nature 466, 470–473 (2010). 2. Chen, Y.-C. et al. Tuning the Band Gap of Graphene Nanoribbons Synthesized from Molecular Precursors. ACS Nano 7, 6123–6128 (2013). 3. Narita, A. et al. Synthesis of structurally well-defined and liquid-phase-processable graphene nanoribbons. Nat. Chem. 6, 126–132 (2014). 5. Talirz, L. et al. On-Surface Synthesis and Characterization of 9-atom wide Armchair Graphene Nanoribbons. submitted. 5. Narita, A., Wang, X.-Y., Feng, X. & Müllen, K. New advances in nanographene chemistry. Chem. Soc. Rev. 44, 6616–6643 (2015). 6. Ruffieux, P. et al. On-surface synthesis of graphene nanoribbons with zigzag edge topology. Nature 531, 489–492 (2016). 7. Kimouche, A. et al. Ultra-narrow metallic armchair graphene nanoribbons. Nat. Commun. 6, 10177 (2015). 8. Talirz, L., Ruffieux, P. & Fasel, R. On-Surface Synthesis of Atomically Precise Graphene Nanoribbons. Adv. Mater. 6222–6231 (2016). doi:10.1002/adma.201505738 9. Chen, Y.-C. et al. Molecular bandgap engineering of bottom-up synthesized graphene nanoribbon heterojunctions. Nat. Nanotechnol. 10, 156–160 (2015). 10. Cai, J. et al. Graphene nanoribbon heterojunctions. Nat. Nanotechnol. 9, 896–900 (2014). 11. Luisier, M., Lundstrom, M., Antoniadis, D. A. & Bokor, J. Ultimate device scaling: Intrinsic performance comparisons of carbon-based, InGaAs, and Si field-effect transistors for 5 nm 10 gate length. in Electron Devices Meeting (IEDM), 2011 IEEE International 11.2.1-11.2.4 (2011). 12. Zhao, P., Chauhan, J. & Guo, J. Computational Study of Tunneling Transistor Based on Graphene Nanoribbon. Nano Lett. 9, 684–688 (2009). 13. Bennett, P. B. et al. Bottom-up graphene nanoribbon field-effect transistors. Appl. Phys. Lett. 103, 253114 (2013). 14. Abbas, A. N. et al. Deposition, Characterization, and Thin-Film-Based Chemical Sensing of Ultra-long Chemically Synthesized Graphene Nanoribbons. J. Am. Chem. Soc. 136, 7555– 7558 (2014). 15. Gao, J. et al. Ambipolar Transport in Solution-Synthesized Graphene Nanoribbons. ACS Nano 10, 4847–4856 (2016). 16. Abbas, A. N. et al. Vapor-Phase Transport Deposition, Characterization, and Applications of Large Nanographenes. J. Am. Chem. Soc. 137, 4453–4459 (2015). 17. Nakada, K., Fujita, M., Dresselhaus, G. & Dresselhaus, M. S. Edge state in graphene ribbons: Nanometer size effect and edge shape dependence. Phys. Rev. B 54, 17954–17961 (1996). 18. Yang, L., Park, C.-H., Son, Y.-W., Cohen, M. L. & Louie, S. G. Quasiparticle Energies and Band Gaps in Graphene Nanoribbons. Phys. Rev. Lett. 99, 186801 (2007). 19. Hsu, H. & Reichl, L. E. Selection rule for the optical absorption of graphene nanoribbons. Phys. Rev. B 76, 45418 (2007). 20. Yoon, Y. & Guo, J. Effect of edge roughness in graphene nanoribbon transistors. Appl. Phys. Lett. 91, 73103 (2007). 11 21. Franklin, A. D. Electronics: The road to carbon nanotube transistors. Nature 498, 443–444 (2013). 22. Verzhbitskiy, I. A. et al. Raman Fingerprints of Atomically Precise Graphene Nanoribbons. Nano Lett. 16, 3442–3447 (2016). 23. Vandescuren, M., Hermet, P., Meunier, V., Henrard, L. & Lambin, P. Theoretical study of the vibrational edge modes in graphene nanoribbons. Phys. Rev. B 78, 195401 (2008). 24. Chen, Z., Appenzeller, J., Knoch, J., Lin, Y. & Avouris, P. The Role of Metal−Nanotube Contact in the Performance of Carbon Nanotube Field-Effect Transistors. Nano Lett. 5, 1497–1502 (2005). 25. Appenzeller, J., Radosavljević, M., Knoch, J. & Avouris, P. Tunneling Versus Thermionic Emission in One-Dimensional Semiconductors. Phys. Rev. Lett. 92, 48301 (2004). 26. Perebeinos, V., Tersoff, J. & Haensch, W. Schottky-to-Ohmic Crossover in Carbon Nanotube Transistor Contacts. Phys. Rev. Lett. 111, 236802 (2013). 27. Zhang, Y., Ye, J., Matsuhashi, Y. & Iwasa, Y. Ambipolar MoS2 Thin Flake Transistors. Nano Lett. 12, 1136–1140 (2012). 28. Franklin, A. D. et al. Sub-10 nm Carbon Nanotube Transistor. Nano Lett. 12, 758–762 (2012). 29. Jiao, L., Zhang, L., Wang, X., Diankov, G. & Dai, H. Narrow graphene nanoribbons from carbon nanotubes. Nature 458, 877–880 (2009). 30. Li, X., Wang, X., Zhang, L., Lee, S. & Dai, H. Chemically Derived, Ultrasmooth Graphene Nanoribbon Semiconductors. Science 319, 1229–1232 (2008). 31. Wang, X. & Dai, H. Etching and narrowing of graphene from the edges. Nat. Chem. 2, 661– 665 (2010). 12 32. Linden, S. et al. Electronic Structure of Spatially Aligned Graphene Nanoribbons on Au(788). Phys. Rev. Lett. 108, 216801 (2012). 33. Franklin, A. D. & Chen, Z. Length scaling of carbon nanotube transistors. Nat. Nanotechnol. 5, 858–862 (2010). ACKNOWLEDGMENT This work was supported in part by the Office of Naval Research BRC program under Grant N00014-16-1-2229, DARPA, the U. S. Army Research Laboratory and the U. S. Army Research Office under contract/grant number W911NF-15-1-0237, the Swiss National Science Foundation, the DFG Priority Program SPP 1459 and Graphene Flagship (No. CNECT-ICT- 604391). Work at the Molecular Foundry was supported by the Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02- 05CH11231. Additional support was received from the Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences and Engineering Division, of the U.S. Department of Energy under Contract No. DE-AC02-05-CH11231, within the sp2-Bonded Materials Program (KC2207), which provided for development of the IL gating method. J.P.L. is supported by the Berkeley Fellowship for Graduate Studies and by the NSF Graduate Fellowship Program. AUTHOR CONTRIBUTIONS J.P.L., B.Y.C., R.B. and J.B. fabricated and measured the devices on SiO2. J.P.L., J.L., K.L., S.W., J.B. and E.Y. fabricated and measured the devices on HfO2. W.S. and A.Z. performed the IL gating experiments. A.F, G.B.B, P.R. and R.F. performed growth, transfer, STM, and Raman spectroscopy of 9AGNRs. N.K., W.C., C.C., Z.P. and M.C. performed growth and STM measurements of 13AGNRs. J.P.L. and J.B. transferred and performed Raman spectroscopy of 13AGNRs. F.F. synthesized the 13AGNR precursor molecule. T.D., A.N., X.F. and K.M. synthesized the 9AGNR precursor molecule. All the authors discussed and wrote the paper. 13  Figure 1 High resolution STM GNR characterization and FET structure (a) STM image of synthesized 9AGNR on Au (Vs = 1 V, It = 0.3 nA). Inset: High resolution STM image of 9AGNR on Au (Vs = 1 V, It = 0.5 nA) with a scale bar of 1 nm (b) High resolution STM image of 13AGNR on Au (Vs = -0.7 V, It = 7 nA). (c) Schematic of the short channel GNRFET with a 9AGNR channel and Pd source-drain electrodes (d) Scanning electron micrograph of the fabricated Pd source-drain electrodes. 14 Figure 2 Raman spectra of as-grown GNRs on Au and GNRs after transfer and device processing. Raman spectra of (a) 9AGNRs and (b) 13AGNRs on the Au(111) growth substrate and after device fabrication shows that the GNRs remain intact. Since the excitation is off- resonance with the 13AGNR absorption, the Raman signal is weak on Au and the RBLM is not visible. 15 Figure 3 Transport characteristics of 9AGNRs and 13AGNRs gated with 50 nm SiO2 gate oxide. The presence of a SB is confirmed by non-linear current behaviour at low drain bias and lack of current saturation at high drain bias for both (a) 9AGNRs and (b) 13AGNRs. The weak temperature dependence in the Id-Vg behaviour in (c) 9AGNRs and (d) 13AGNRs indicates that tunnelling through the Pd-GNR SBs is the limiting transport mechanism of the device. 16 Figure 4 Ionic liquid gating of a 9AGNRFET at room temperature. (a) Id-Vg characteristics of the device gated by the thick 50 nm SiO2 gate oxide (b) Id-Vg characteristics of the device gated with the ionic liquid which shows clear ambipolar behaviour and improved on-state performance. Inset: ionic liquid (DEME-TFSI) gated 9AGNRFET device schematic. 17 Figure 5 Transport characteristics of a scaled, high performance 9AGNRFET at room temperature. (a) Id-Vd characteristics of the scaled 9AGNRFET. (b) Id-Vg of the devices show high Ion >1 A for a 0.95 nm wide 9AGNR and high Ion/Ioff ~105. Inset: scaled 9AGNRFET schematic. 18 Supplemental Information Extraction of number of GNRs in the channel We used a Monte Carlo simulation to estimate the number of GNRs in our device channels based on our device yield. Assuming a uniform spatial distribution of GNRs, we simulate the expected device yield and distribution of number of GNRs in the channel. The input parameters of the simulation were the GNR number density on the surface and GNR length. We varied these parameters to generate Fig. S1. With the experimentally obtained yield of ~10%, the percentage of devices with more than 1 GNR in the channel goes as high as 8% for higher surface density and 4% for low surface density. Out of the devices with multiple GNRs, an insignificant percentage has more than 2 GNRs/channel. Thus, we estimate that only 1-3 devices out of ~30 fabricated devices have 2 GNRs in the channel. However, it is unclear whether these devices would account of the high tail end of the on-current distribution since both GNRs would have to have good contact length under the Pd contacts to improve conduction over a single GNR channel with a large contact length. Figure S3. Simulated % yield of working devices as a function of GNR length and number density. The P(1GNR) values denote the probability of a yielded device to have a single GNR in 19 the channel. With our ~10% experimental yield, we estimate that only 1-3 devices out of 30 contain 2 GNRs in the channel. Figure S2. Raman spectra of 13AGNRs after device fabrication using 532 nm and 785 nm wavelength excitation. The RBLM is not detectable under these excitation conditions due to excitation off-resonance effects. Figure S3. Cumulative distribution function (CDF) of Ion in 13AGNRFETs and 9AGNRFETs with 50 nm SiO2 gate dielectrics. The CDF is defined as the total fraction of devices with on- current greater than the given value of Ion. Both types of devices have similar behavior due to the similar band gap and variations in on-state performance are most likely due to variations in the overlap length between the Pd and GNR and variations in the channel length. 20 Figure S4. Id-Vbg characteristics of the 9AGNRFET shown in Fig 3 in the main text which shows that the gate leakage limits the off-current. 21 

1307.3670

1

1307

Quantum fluctuations in modulated nonlinear oscillators

With a modulated oscillator, we study several effects of quantum fluctuations far from thermal equilibrium. One of them is quantum heating, where quantum fluctuations lead to a finite-width distribution of a resonantly modulated oscillator over its quasienergy (Floquet) states. We also analyze large rare fluctuations responsible for the tail of the quasienergy distribution and switching between the states of forced vibrations. We find an observable characteristic of these fluctuations, the most probable paths followed by the quasienergy in rare events, and in particular in switching. We also explore the discontinuous change of the most probable switching path where the detailed balance condition is broken. For oscillators modulated by a nonresonant field, we compare different mechanisms of the field-induced cooling and heating of the oscillator.

cond-mat.mes-hall

cond-mat

Quantum fluctuations in modulated nonlinear oscillators Vittorio Peano1 and M I Dykman2 1Institute for Theoretical Physics II, University of Erlangen-Nuremberg, 91058 Erlangen, Germany 2Department of Physics and Astronomy, Michigan State University, East Lansing, MI 48824, USA Abstract. With a modulated oscillator, we study several effects of quantum fluctuations far from thermal equilibrium. One of them is quantum heating, where quantum fluctuations lead to a finite-width distribution of a resonantly modulated oscillator over its quasienergy (Floquet) states. We also analyze large rare fluctuations responsible for the tail of the quasienergy distribution and switching between the states of forced vibrations. We find an observable characteristic of these fluctuations, the most probable paths followed by the quasienergy in rare events, and in particular in switching. We also explore the discontinuous change of the most probable switching path where the detailed balance condition is broken. For oscillators modulated by a nonresonant field, we compare different mechanisms of the field-induced cooling and heating of the oscillator. PACS numbers: 05.30.-d, 03.65.Yz, 74.50.+r, 85.25.Cp Submitted to: New J. Phys. 1. Introduction The last few years have seen an upsurge in the interest in the dynamics of modulated nonlinear oscillators [1]. There have emerged several new areas of research where this dynamics plays a central role, such as nanomechanics, cavity optomechanics, and circuit quantum electrodynamics. The vibrational systems of the new generation are mesoscopic. On the one hand, they can be individually accessed, similar to macroscopic systems, and are well-characterized. On the other hand, since they are small, they experience comparatively strong fluctuations of thermal and quantum origin. This makes their dynamics interesting on its own and also enables using modulated oscillators to address a number of fundamental problems of physics far from thermal equilibrium. Many nontrivial aspects of the oscillator dynamics are related to the nonlinearity. Essentially all currently studied mesoscopic vibrational systems display nonlinearity. For weak damping, even small nonlinearity becomes important. It makes the frequencies of transitions between adjacent oscillator energy levels different. Where several levels are Quantum fluctuations in modulated nonlinear oscillators 2 occupied, the dynamics strongly depends on the interrelation between the width of the ensued frequency comb and the oscillator decay rate. An important consequence of the nonlinearity is that, when an oscillator is resonantly modulated, it can have coexisting states of forced vibrations, i.e., display bistability [2]. One of the general physics problems addressed with modulated nonlinear oscillators is fluctuation-induced switching in systems that lack detailed balance, see [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] for the classical and [15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26] for the quantum regime. A remarkable property of the switching rate in the quantum regime is fragility. The rate Wsw calculated for T = 0, where the system has detailed balance [27], is exponentially different from the rate calculated for T > 0, where the detailed balance is broken [15, 18]. Recently the effect of fragility of the rates of rare events was also found in the problem of population dynamics [28]. There, too, a small change of the control parameter (infinitesimal, in the semiclassical limit) leads to an exponentially strong rate change. The nature of the dynamics and the sources of fluctuations in a quantum oscillator and in population dynamics are totally different, and it is important to understand how it happens that they display common singular features. An important source of quantum fluctuations is the coupling of the oscillator to a thermal bath. It leads to oscillator relaxation via emission of excitations in the bath accompanied by transitions between the oscillator energy levels. The transitions lead to relaxation only on average, in fact they happen at random, giving rise to a peculiar quantum noise. For a resonantly modulated oscillator, the noise causes diffusion over the oscillator quantum states in the external field, which are the quasienergy (Floquet) states. As a result, even where the bath temperature is T = 0, the distribution over the states has a finite width, the effect of quantum heating [29]. We discuss quantum heating for a resonantly modulated oscillator and compare the predictions with the recent experiment [30] where the effect was observed. The spectral manifestation of quantum heating is considered, with the focus on the influence of dissipation on the oscillator spectral characteristic of interest for sideband spectroscopy, the technique which was nicely implemented in the experiment [30] using a microwave cavity with an embedded qubit. We also study switching between the stable states of forced vibrations of an oscillator modulated close to its eigenfrequency. As quantum heating, switching occurs because of the quantum-noise induced diffusion over the oscillator states. It reminds switching of a classical Brownian particle over the potential barrier due to diffusion over energy [31] and therefore is called quantum activation. Generally, the rate of quantum activation largely exceeds the rate of switching via quantum tunneling. We develop an approach to calculating the rate of quantum activation, which naturally connects to the conventional formulation of the rare events theory in chemical and biological reaction systems and in population dynamics [32, 33]. This approach provides a new insight into the fragility of the switching rate of the oscillator. The dynamics of a periodically modulated harmonic quantum oscillator coupled to a thermal reservoir is one of exactly solvable problems of physical kinetics [34, 35, 36]. Quantum fluctuations in modulated nonlinear oscillators 3 However, the solution disregards the fact that nonresonant modulation can open a new channel for oscillator relaxation, where a transition between the oscillator energy levels is accompanied by an emission (absorption) of an excitation in the medium, while the energy deficit is compensated by the modulation. Alternatively, the role of the medium can be played by another mode with the relaxation rate higher than that of the oscillator [37]. This mechanism underlies the cooling studied in cavity optomechanics. We provide a brief comment in order to unify various mechanisms of the change of the quantum distribution of the oscillator by nonresonant modulation. 2. Quasienergy spectrum and the master equation 2.1. Hamiltonian in the rotating frame A major type of the internal oscillator nonlinearity of interest for the effects we will discuss is the Duffing nonlinearity, where the potential energy has a term quartic in the oscillator displacement q; in quantum optics, it corresponds to the Kerr nonlinearity. The simplest types of resonant modulation that lead to the bistability of the oscillator are additive modulation at frequency ωF close to the oscillator eigenfrequency ω0 and parametric modulation (modulation of the oscillator frequency) at frequency ≈ 2ω0 [2]. The analysis of quantum fluctuations in these two systems has much in common [38], and the method that we will develop here applies to the both types of systems. For concreteness, we will consider here additive modulation. The oscillator Hamiltonian is H0 = 1 2 p2 + 1 2 ω2 0q2 + 1 4 γq4 + HF (t), HF = −qA cos ωF t, (1) where q and p are the oscillator coordinate and momentum, the mass is set equal to one, γ is the anharmonicity parameter, and A is the modulation amplitude. We assume that the the modulation is resonant and not too strong, so that δω ≪ ω0, δω = ωF − ω0; γhq2i ≪ ω2 0. (2) A periodically modulated oscillator is described by the Floquet, or quasienergy states Ψε(t). They provide a solution of the Schrodinger equation i∂tΨ = H0(t)Ψ that satisfies the condition Ψε(t + tF ) = exp(−iεtF /)Ψε(t), where tF = 2π/ωF . This expression defines quasienergy ε. To find quasienergies and to describe the oscillator dynamics it is convenient to change to the rotating frame. This is done by the standard canonical transformation U(t) = exp(cid:0)−ia†a ωF t(cid:1), where a† and a are the raising and lowering operators of the oscillator. We introduce slowly varying dimensionless coordinate Q and momentum P in the rotating frame, U †(t)qU(t) = C(Q cos ϕ + P sin ϕ), U †(t)pU(t) = −CωF (Q sin ϕ − P cos ϕ). Here, ϕ = ωF t and the scaling factor is C = 8ωF (ωF − ω0)/3γ1/2. The commutation relation between P and Q has the form [P, Q] = −iλ, λ = /(ωF C 2) ≡ 3γ/8ω2 F ωF − ω0. (3) Quantum fluctuations in modulated nonlinear oscillators 4 Parameter λ ∝  plays the role of the Planck constant in the quantum dynamics in the rotating frame. It is determined by the oscillator nonlinearity, λ ∝ γ. For concreteness we assume that γ, δω > 0; the oscillator displays bistability for γδω > 0. In the range (2) the oscillator dynamics can be studied in the rotating wave approximation (RWA). The RWA Hamiltonian is 3 8 Eslg, − β1/2Q, β = 3γA2/32ω3 F ωF − ω03, (4) H0 = U †H0U − iU † U ≈ g(Q, P ) = 1 4 (cid:0)P 2 + Q2 − 1(cid:1)2 where β is the scaled modulation intensity and Esl = γC 4 is the characteristic energy of motion in the rotating frame. This motion is slow on the time scale ω−1 F . Note that Esl ≪ ω2 0hq2i ∼ ω2 0C 2. Operator g = g(Q, P ) is the dimensionless Hamiltonian in the rotating frame. In the RWA, the Schrodinger equation for the RWA wave function ψ(Q) in dimensionless slow time τ reads iλ ψ ≡ iλ∂τ ψ = gψ, τ = tδω ≡ (λEsl/)t. (5) Operator g is independent of time and has a discrete spectrum, gni = gnni. The eigenvalues gn give the quasienergies in the RWA, εn = (3Esl/8)gn (we are using an extended ε-axis rather than limiting ε to the analog of the first Brillouin zone 0 ≤ ε < ωF ). Function g(Q, P ) has the shape of a tilted Mexican hat and is shown in Fig. 1(a); the quasienergy levels are shown in Fig. 1(b) . (a) The Hamiltonian function in the rotating frame in the RWA. The Figure 1. extrema of g(Q, P ) correspond to the stable vibrational states in the limit of weak damping. (b) The cross-section g(Q, 0) and the quasienergy levels of the states localized about the extrema of g(Q, P ). Points Qmin, Qmax and QS indicate the positions of the minimum, the local maximum, and the saddle point of g(Q, P ), respectively. The plots refer to β = 0.01 and λ = 0.041. (c) The transitions between the Fock states of the oscillator with energies EN ≈ ω0(N + 1/2) accompanied by emission of excitations in the bath, e.g., photons. Some of the corresponding transitions between quasienergy states are shown by small arrows in (b). The stationary state of the oscillator is formed on balance between relaxation and excitation by periodic modulation F (t). Quantum fluctuations in modulated nonlinear oscillators 5 In contrast to the Hamiltonian H0, g is not a sum of the kinetic and potential energies. As seen from Fig. 1, the eigenstates localized near the local maximum of g(Q, P ) correspond to semiclassical orbits on the surface of the "inner dome" of g(Q, P ); these states become stronger localized as gn increases toward the local maximum of g(Q, P ). The quasienergy level spacing ∝ λEsl is small compared to the distance between the oscillator energy levels in the absence of modulation, εn − εn+1 ∼ λEsl ≪ ω0. 2.2. Master equation for linear coupling to the bath The analysis of the oscillator dynamics is often done assuming that the oscillator is coupled to a thermal bath in such a way that the coupling energy is linear in the oscillator coordinate q and thus in the oscillator ladder operators a, a† [34]. In this case the coupling energy Hi and the typical relaxation rate Γ are of the form Hi = ahb + H.c., Γ ≡ Γ(ω0) = −2Re Z ∞ 0 dth[h† b(t), hb(0)]ibeiω0t, (6) where hb depends on the bath variables only and h. . .ib denotes thermal averaging over the bath states. Relaxation (6) corresponds to transitions between neighboring energy levels of the oscillator in the lab frame, with energy transferred to bath excitations, see Fig. 1(c). The renormalization of the oscillator parameters due to the coupling is assumed to have been incorporated. For a smooth density of states of the bath, resonant modulation does not change the decay rate parameter, Γ(ωF ) ≈ Γ(ω0). However, it excites the oscillator, as sketched in Fig. 1(c). In a stationary state of forced vibrations (in the lab frame) the energy provided by the modulation is balanced by the relaxation. To the second order in the interaction (6), the master equation for the oscillator density matrix ρ in dimensionless time τ reads ρ ≡ ∂τ ρ = iλ−1[ρ, g] − κρ, κρ = κ(¯n + 1)(a†aρ − 2aρa† + ρa†a) + κ¯n(aa†ρ − 2a†ρa + ρaa†), κ = Γ/ωF − ω0. (7) Here, the term ∝ [ρ, g] describes dissipation-free motion, cf. (5). Operator κρ describes dissipation and has the same form as in the absence of oscillator modulation [39, 40]; κ is the dimensionless decay rate, and ¯n is the oscillator Planck number, a = (2λ)−1/2(Q + iP ), ¯n ≡ ¯n(ω0) = [exp(ω0/kBT ) − 1]−1. (8) In the classical limit λ → 0 the oscillator described by (7) can have one or two stable states of forced vibrations. Their positions in the rotating frame (Qa, Pa) are given by the stable stationary solutions of the classical equations of motion of the oscillator Q = ∂P g − κQ, P = −∂Qg − κP. (9) Equations (9) are, essentially, the mean-field equations for the moments Tr(Qρ), Tr(P ρ) for λ → 0. For small damping Qa and Pa are close to the extrema of g(Q, P ). Quantum fluctuations in modulated nonlinear oscillators 6 3. Quantum heating 3.1. Balance equation We will concentrate on the oscillator dynamics in the case where the oscillator is strongly underdamped and its motion is semiclassical, λ ≪ 1, κ ≪ 1. (10) In this case the number of quasienergy states localized about the extrema of g(Q, P ) is large, ∝ 1/λ [the scaled quasienergies of such states gn lie between the value of g at the corresponding extremum and the saddle point value gS of g(Q, P ) in Fig. 1]. Also, the spacing between the levels is large compared to their width, gn − gn±1 ≫ λκ. Where the latter condition is met, the off-diagonal matrix elements of the density matrix on the quasienergy states ρnm ≡ hnρmi (n 6= m) are small. To the lowest order in κ the oscillator dynamics can be described by the balance equation for the populations ρnn of quasienergy states. From (7) ρnn = Xm (Wmnρmm − Wnmρnn) , Wmn = 2κ(cid:2)(¯n + 1)anm2 + ¯namn2(cid:3) , (11) where anm ≡ hnami. We disregard tunneling when defining functions ni ≡ ψn(Q), i.e., we use the wave functions localized about the extrema of g(Q, P ); the effect of tunneling is exponentially small for λ ≪ 1. We count the localized states off from the corresponding extremum, i.e., for a given extremum the state with n = 0 has gn closest to g(Q, P ) at the extremum. An important feature of the rates of interstate transitions Wmn is that, even for T = 0 (and thus ¯n = 0), there are transitions both toward and away from the extrema of g(Q, P ). This is because the wave functions ni are linear combinations of the wave functions of the oscillator Fock states, see Fig. 1(c). Therefore, even though relaxation corresponds to transitions down in the oscillator energy in Fig. 1(c), the transitions up and down the quasienergy have nonzero rates. One can show that, for the both extrema of g(Q, P ), the rates of transitions toward an extremum are larger than away from it. Therefore, depending on where the system was prepared initially, it would most likely move to one or the other extremum of g(Q, P ). This is why the extrema correspond to the stable states of forced vibrations of the modulated oscillator in the classical limit of a large number of localized states. For small effective Planck constant λ ≪ 1, the rates Wmn can be calculated in an explicit form by finding the matrix elements amn in the WKB approximation [15, 41]. The problem is then related to that of classical conservative motion with Hamiltonian g(Q, P ) and with equations of motion of the form Q = ∂P g, P = −∂Qg. A significant simplification comes from the fact that the classical trajectories Q(τ ; g) are described by the Jacobi elliptic functions. As a result, Q(τ ; g) is double-periodic on the complex- τ plane, with real period τ (1) p (g). For m − n ≪ λ−1 the matrix element amn is given by the Fourier m − n component of the function a(τ ; gn) = (2λ)−1/2[Q(τ ; gn) + iP (τ ; gn)][42]. It can be calculated along an appropriately p (g) and complex period τ (2) Quantum fluctuations in modulated nonlinear oscillators 7 chosen closed contour on the complex τ -plane and is determined by the pole of a(τ ; gn). In particular, for the states localized about the local maximum of g(Q, P ) we obtain an+k n2 = k2ν4 n 2βλ p )] exp[kνn Im (2τ∗ − τ (2) p /2]2 p (gn) [Im τ∗, Im τ (2) sinh[ikνn τ (2) p ≡ τ (2) Here, τ∗ ≡ τ∗(gn) and τ (2) p > 0]; τ∗(g) is the pole of Q(τ ; g) closest to the real axis; ν(g) is the dimensionless frequency of vibrations in the rotating frame with quasienergy g. To the leading order in λ, we have Wn n+k = Wn−k n for n, n ± k ≫ 1. , νn ≡ ν(gn) = 2π/τ (1) p (gn).(12) 3.2. Effective temperature of vibrations about a stable state Equation (12) has to be modified for states very close to the extrema of g(Q, P ). Near these extrema the classical motion of the oscillator in the rotating frame is harmonic vibrations. One can introduce raising and lowering operators b and b† for these vibrations (via squeezing transformation) and expand g(Q, P ) near an extremum as Q − Qa + iP = (2λ)1/2(b cosh ϕ∗ − b† sinh ϕ∗), g ≈ g(Qa, 0) + λν0(cid:0)b†b + 1/2(cid:1) sgn∂2 Qg, Qg∂2 ν0 = (cid:12)(cid:12)∂2 1/2 P g(cid:12)(cid:12) (13) [(Qa, P = 0) is the position of the considered extremum; it is given by equation ∂Qg = Q(Q2 − 1) − β1/2 = 0]. The derivatives of g in (13) are evaluated at (Qa, P = 0). Parameter φ∗ is given by equation tanh ϕ∗ = (∂2 P g1/2). P g1/2)/(∂2 Qg1/2 − ∂2 Qg1/2 + ∂2 From (11) and (13), near an extremum of g we have Wm+1 m = 2κ(m + 1)(¯ne + 1), ¯ne = ¯n cosh2 φ∗ + (¯n + 1) sinh2 φ∗, Wm m+1 = 2κ(m + 1)¯ne, (14) whereas Wm m+k = 0 for k > 1. Equation (14) is a familiar expression for the transition rates between the states of a harmonic oscillator coupled to a thermal bath, with ¯ne being the Planck number of the excitations of this fictitious bath at the frequency of vibrations in the rotating frame ν0δω. From (14), the stationary distribution of the modulated oscillator over its quasienergy states near an extremum of g(Q, P ) is of the Boltzmann type, with effective temperature Te = λν0/ ln[(¯ne + 1)/¯ne] [18, 38]. In agreement with the qualitative picture discussed above, this temperature is nonzero even where the temperature of the true bath is T = 0. This is the effect of quantum heating due to quantum fluctuations in a nonequilibrium system. Quantum heating of a resonantly modulated oscillator was recently observed in an elegant experiment [30] using a mode of a microwave cavity with an embedded Josephson junction [43]. The occupation of the excited quasienergy states was revealed using a two-level system (a transmon qubit) as a probe. As seen from Fig. 2, the results of the experiment are in a qualitative agreement with the above theory. The agreement improves for larger scaled field intensity β (4), where the ratio κ/ν0 is smaller. It is in the range of small κ/ν0 that the quantum temperature is a good Quantum fluctuations in modulated nonlinear oscillators 8
b
Fnn 8 6 4 2 0 - 3 - 2 - 1 0 1 2 3
Ω-ΩF

∆Ω (a) The effective Planck number ¯ne of the vibrations about the large- Figure 2. amplitude state of the modulated oscillator, which corresponds to the minimum of g(Q, P ) in the small-damping limit; β is the scaled driving field intensity (4). Squares: experimental data [30]. Solid line: equation (14) for ¯n = 0 (see also [38]). Triangles: the estimate of the experimentally measured parameter discussed in Appendix A. (b) The scaled power spectra of the oscillator occupation number n = a†a for vibrations about the large-amplitude stable state for δω/κ = 3.9 (the value used in [30]). The black and red curves correspond to β = 0.17 and 0.8. The triangles in (a) are determined from the ratio rΦ of the heights of the lower and higher peaks of Φnn as rΦ/(1 − rΦ). characteristic of the distribution over quasienergy states, as the lifetime of these states largely exceeds the reciprocal level spacing (scaled by ). Still, even for not too small κ/ν0, the technique developed in [30] makes it possible to reveal the broadening of the stationary distribution of the modulated oscillator due to quantum fluctuations far from equilibrium. A characteristic of this effect is discussed in Appendix A and the corresponding results are shown in Fig. 2. 4. Switching between the stable states The effect of diffusion over quasienergy states due to quantum fluctuations is not limited to the quantum heating described above. Along with small fluctuations, which lead to comparatively small deviations of quasienergy from its value at an extremum of g(Q, P ), there occasionally occur large fluctuations. They push the oscillator far away from the initially occupied extremum. It is clear that, if as a result of such fluctuation, the oscillator goes "over the quasienergy barrier" to states localized about the other extremum, with probability ≈ 1 it will then approach this other extremum. Such transition corresponds to switching between the stable states of forced vibrations via the quantum activation mechanism. As seen from Fig. 1(a), with an accuracy to a factor ∼ 1/2 the switching rate Wsw is determined by the probability to reach the saddle-point value gS of g(Q.P ). The switching rate is small, as switching requires that the oscillator makes many interlevel transitions with rates Wmn smaller than the rates of transitions in the opposite direction, Wnm. Therefore, before the oscillator switches, there is formed a quasistationary distribution over its states localized about the initially occupied extremum of g(Q, P ). This is similar to what happens in thermally activated switching Quantum fluctuations in modulated nonlinear oscillators 9 over a high barrier [31]. However, in contrast to systems in thermal equilibrium, a modulated oscillator generally does not have detailed balance. Its statistical distribution has a simple Boltzmann form with temperature Te only for small damping and only close to the extrema of g(Q, P ). Therefore the standard technique developed for finding the switching rate in quantum equilibrium systems [44, 45, 46, 47] does not apply. Also, even for T → 0 an oscillator modulated close to its eigenfrequency generally does not switch via tunneling (see [16, 19, 21, 48, 49, 50, 51, 52] for the theory of tunneling switching for additive and parametric modulation). Switching via quantum activation is exponentially more probable. 4.1. Relation to chemical kinetics and population dynamics For small scaled decay rate κ the switching rate Wsw can be obtained from the balance equation (11). An approach to solving this equation was discussed earlier [15, 18]. Here we provide a formulation that gives an insight into how the oscillator actually moves in switching and also makes a direct connection with the technique developed in chemical kinetics and population dynamics. The balance equation is broadly used in these areas. It describes chemical or biochemical reactions in stirred reactors (no spatial nonuniformity). The reactions can be thought of as resulting from molecular collisions in which molecules change, and if the collision duration is small compared to the reciprocal collision rate the kinetics is described by a Markov equation [53] ρ(X, τ ) = Xr [W (X − r, r)ρ(X − r, τ ) − W (X, r)ρ(X, τ )]. (15) Here, X = (X1, X2, ...) is the vector that gives the numbers of molecules Xi of different types i, and ρ is the probability for the system to be in a state with given X; W (X, r) is the rate of a reaction in which the number of molecules changes from X to X + r. Typically, Xi are large, Xi ∝ N ≫ 1, where N is the total number of molecules. In contrast, the change of the number of molecules in an elementary collision is r ∼ 1, because it is unlikely that many molecules would collide at a time. Equation (15) is also often used in population dynamics, including epidemic models, cf. [54]. In this case the components of X give populations of different species. Since the number of molecules (population) is large, N ≫ 1, fluctuations are small on average. Disregarding fluctuations corresponds to the mean-field approximation. In this approximation one can multiply (15) by X and sum over X while assuming that the width of the distribution ρ(X) is small. This gives the equation of motion for the scaled mean number of molecules (population) x = Xr rw(x, r), x = X/N, w(x, r) = W (X, r)/N. (16) Stable solutions of (16) give the stable states of chemical (population) systems. There may be also unstable stationary or periodic states. In population dynamics, an unstable stationary solution of (16) can be the state where one of the species goes extinct. Quantum fluctuations in modulated nonlinear oscillators 10 Equation (15) describes diffusion in the space of variables x. Along with small (∝ N −1/2) fluctuations around the stable states, this diffusion leads to rare large deviations (∼ O(N −1) in x-space) and to switching between the stable states. There is an obvious similarity between diffusion over the number of molecules and diffusion over quasienergy states of a modulated oscillator, but there are also some subtle differences, which we discuss below. There is also an obvious difference, with profound consequences: in the case of an oscillator the transition rates Wmn (11), (12) are not limited to m − n ∼ 1. 4.2. The eikonal approximation The role of the large number of molecules (population) in a modulated oscillator is played by the reciprocal effective Planck constant λ−1, which determines the number of states localized about the extrema of g(Q, P ), cf. Fig. 1. For λ ≪ 1 it is convenient to switch from the state number n to the classical mechanical action I for the Hamiltonian orbits Q(τ ; g), P (τ ; g), which are described by equations Q = ∂P g(Q, P ), P = −∂Qg(Q, P ), I = I(g) = (2π)−1Z 2π/ν(g) 0 P (τ ; g) Q(τ ; g)dτ, ∂gI = ν−1(g), (17) where ν(g) is the vibration frequency for given g [2πI gives the area of the cross-section of the surface g(Q, P ) in Fig. 1(a) by plane g =const]. One can show that, in spite of the nonstandard form of g(Q, P ), the semiclassical quantization condition has the familiar form In ≡ I(gn) = λ(n + 1/2). In the semiclassical approximation the rates of transitions between quasienergy states Wmn become functions of the quasicontinuous variable I and can be written as Wmn = W (Im, n − m). The dependence of W on I is smooth, as seen from (11) and (12), W (Im, n − m) ≈ W (In, n − m) for typical n − m ≪ 1/λ. Similar to (16), in the neglect of quantum fluctuations the equation for I = Pn Inρn has a simple form I = Xr rw(I, r), w(I, r) = λW (I, r). (18) This equation shows how the oscillator is most likely to evolve. Using that the matrix element amn in the expression (11) for the rate Wmn is the (n − m)th Fourier component of function (2λ)−1/2[Q(τ ; gm)+iP (τ ; gm)], one can show by invoking the Stokes' theorem that the time evolution of I is extremely simple, I = −2κI, I < IS, (19) where IS is the value of I(g) for g approaching the saddle-point value gS from the side of the extremum of g(Q, P ) of interest; the values of IS are different on the opposite sides of gS. Equation (19) coincides with the result for the evolution of g for a classical modulated oscillator [3]. We note that the semiclassical approximation breaks down very close to the saddle point (in particular, the relation W (Im, n − m) ≈ W (In, n − m) Quantum fluctuations in modulated nonlinear oscillators 11 clearly ceases to apply), but the width of the corresponding range of I goes to zero as λ → 0. We now consider the quasistationary distribution ρn about the initially occupied stable state. It is formed on times (κδω)−1 ≪ t ≪ W −1 sw . To find ρn far from the stable state we use the eikonal approximation [15, 18], but in the form similar to that used in chemical kinetics and population dynamics [33]. We set ρn = exp[−R(In)/λ] (20) and assume that ∂IR ≪ λ−1. Then ρn+r ≈ ρn exp[−r∂I R] for r ≪ λ−1 and, to the leading order in λ, the balance equation (11) becomes ∂τ R = −H(I, ∂IR), H(I, pI) = Xr w(I, r)[exp(rpI) − 1]. (21) Equation (21) has the form of the Hamilton-Jacobi equation for an auxiliary system with coordinate I, momentum pI, and action variable R(I) [2]. It thus maps the problem of finding the distribution of the modulated oscillator, which is formed by quantum fluctuations, onto the problem of classical mechanics. The quasistationary distribution is determined by the stationary solution of (21), i.e., by the solution of equation H(I, ∂IR) = 0. If there are several solutions, of physical interest is the solution with the minimal R(I), as it gives the leading-order term in ln ρn. 4.3. Optimal switching trajectory An advantageous feature of the formulation (21) is that it provides an insight into how the quantum oscillator evolves in large fluctuations that lead to occupation of quasienergy states far from the initially occupied extremum of g(Q, P ). Even though the diffusion over quasienergy states is a random process and different sequences of interstate transitions can bring the system to the given quasienergy state, the probabilities of such sequences are strongly different. Of physical interest is the most probable sequence, known as the optimal fluctuation. For classical fluctuating systems it has been understood theoretically and shown in experiment and simulations [33, 55, 56, 57, 58] that the evolution of the system in the optimal fluctuation, i.e., the optimal fluctuational trajectory is given by the classical trajectory of the auxiliary Hamiltonian system, which in the present case is described by equation I = ∂H(I, pI)/∂pI , pI = −∂H(I, pI )/∂I; R(I) = Z I 0 pIdI. (22) The concept of the optimal fluctuational trajectory can be extended to the quantum oscillator. Such trajectory for the action variable I is well-defined, since any information of the oscillator phase is automatically erased and the range of the I values largely exceeds the quantum uncertainty in I, which is ∝ λ. Therefore the optimal fluctuational trajectory I(t) can be measured in the experiment in the same way as it is done in classical systems. Quantum fluctuations in modulated nonlinear oscillators 12 In (22) we have set R(0) = 0 and thus ignored the normalization factor (an analog of the reciprocal partition function) in the expression (20) for ρn. This factor leads to a correction ∝ λ to R(0). Since, to logarithmic accuracy, the switching rate is determined by the probability of approaching the saddle-point value IS of I, we have Wsw ∼ κδω exp(−RA/λ), RA = R(IS). (23) Parameter RA plays the role of the effective activation energy for switching via quantum activation, with the effective Planck constant λ replacing the temperature in the conventional expression for thermally activated switching. Optimal fluctuations away from the extremum of g(Q, P ) are described by optimal trajectories that emanate from I = 0, which is reflected in (22). The value of the momentum pI ≡ ∂I R for I → 0 on the trajectory can be found by noticing that the distribution over quasienergy near the extremum of g(Q, P ) is of the form of the Boltzmann distribution with effective temperature Te, and thus R ∝ I/Te; from (14) I = 2κI on the optimal trajectory for pI = ln[(¯ne + 1)/¯ne] for I → 0. Then from (21), I → 0. As expected, the system moves along the optimal fluctuational trajectory away from the stable state of fluctuation-free dynamics. The facts that pI 6= 0 at the starting point of the optimal trajectory and that the state I = 0 lies on the boundary of the available values of I are connected with each other and present a distinctive feature of the oscillator dynamics. In chemical kinetics and population dynamics usually stable states lie in the middle of the space of dynamical variables X. The probability distribution has a Gaussian maximum at such X, and then the momentum on the optimal trajectory is equal to zero [28, 33]. The states (I = 0, pI = 0) and (I = 0, pI = ln[(¯ne + 1)/¯ne]) are stationary points of the Hamiltonian H(I, pI). From (14), I = pI = 0 at these points. The motion of the system near these points is exponential in time and is shown in Fig. 3. Figure 3(a) shows the mean-field (fluctuation-free) trajectory I(t) and the optimal trajectory I(t) obtained numerically from equations (19) and (22), respectively. An interesting feature of the considered model of the modulated quantum oscillator is that it satisfies the detailed balance condition for T = 0 and thus ¯n = [exp(ω0/kBT )−1]−1 = 0 [27]. This is seen from the explicit expression for the rates (11) and (12), as for ¯n = 0 they meet the familiar detailed balance condition Wn n+k/Wn+k n = exp(−k/ξn) [the explicit form of ξn ≡ ξ(In) follows from (12)]. Therefore pI = 1/ξ(I), and one can show from (22) that I = 2κI. As a consequence, the optimal fluctuational trajectory I(t) is the time-reversed mean-field trajectory I(t). This is a generic feature of classical systems with detailed balance, see [55]. Our results show that the symmetry also holds in quantum systems provided the notion of a trajectory is well-defined. Of special interest is the vicinity of the saddle-point value of the action variable IS, see Fig. 3. In a dramatic distinction from chemical kinetics, there is no slowing down of I(t) near IS. The quantity IS is a boundary value of I for states localized about a given extremum of g(Q, P ) in Fig. 1. Functions I and I are discontinuous there. This is an artifact of the balance equation approximation, which applies in the weak Quantum fluctuations in modulated nonlinear oscillators 13 Figure 3. (a) The mean-field (fluctuation free) and optimal fluctuational trajectories of the action variables. Because the system has detailed balance for ¯n = 0, the optimal trajectory in this case is the time-reversed mean-field trajectory. The data refer to the trajectories for the local maximum of g(Q, P ) in Fig. 1 for β = 0.035. The shape of the trajectory changes discontinuously where ¯n becomes nonzero; the trajectories for ¯n = 0 and ¯n → 0 coincide only for I < I¯n. The limit ¯n → 0 is taken with the constraint ¯n ≫ λ3/2. (b) The phase portrait of the auxiliary Hamiltonian system that describes large fluctuations of the oscillator in the small-damping limit. The real-time instantons in (a) correspond to the trajectories in phase space of the same color. The gray area shows the region where H(I, pI ) remains finite for ¯n 6= 0; for ¯n = 0, H remains finite in the whole region shown in the figure. (c) The logarithm of the probability distribution R(In) ≈ −λ ln ρn for ¯n = 0 and ¯n > 0. damping limit where the dimensionless frequency ν(g) ≫ κ. For g → gS the frequency ν(g) → 0, and the approximation breaks down. With account taken of decay, in the region of bistability the oscillator has a "true" unstable stationary state in the neglect of fluctuations. Both the mean-field trajectory and the optimal trajectory in phase space are moving away/approaching this state exponentially in time, cf. [15], but the region of I where it happens is very narrow for small κ. 4.4. Fragility in the problem of large rare fluctuations A striking feature of optimal fluctuational trajectories obvious from Fig. 3 is that these trajectories have different shapes depending on whether the oscillator Planck number is ¯n = 0 or ¯n > 0. The discontinuous with respect to ¯n change of the trajectories and the associated change of the logarithm of the distribution R(I) and of the activation energy for switching RA show the fragility of the detailed-balance solution for ¯n = 0 [15, 18]. It has been found that the fragility also emerges in a very different type of problem, the problem of population dynamics described by equation (15) [28]. In particular, the well-known result for the rate of disease extinction in the presence of detailed balance [59, 60, 61, 62] can change discontinuously with the varying elementary rates W (X, r) as the detailed balance is broken. We now show that the condition for the onset of fragility proposed in [28] applies also to the modulated oscillator, even though the divergence it reveals shows up in a different fashion. The condition relies on the expression for the switching exponent. Similar to how it was done for the oscillator, this exponent can be found by seeking Quantum fluctuations in modulated nonlinear oscillators 14 the solution of the master equation (15) in the eikonal form ρ(X) = exp[−N R(x)]. To the leading order in 1/N, the problem is then mapped onto Hamiltonian dynamics of an auxiliary system with mechanical action R(x). From (15), the Hamiltonian of the auxiliary system is H(x, p) = Xr w(x, r)[exp(rp) − 1], p = ∂x R (24) [as before, we use that W (X − r, r) ≈ W (X, r)]. If the system is initially near a stable state xa [a stable solution of (16)], R(x) is determined by the Hamiltonian trajectories that emanate from xa. From (24), R(x) = R x pdx. The rate of switching from xa (or extinction, in the extinction problem) is Wsw ∝ exp(−N RA). Similar to the quantum oscillator, xa RA = Z xS xa pdx = Z dtp(t) x(t)dt. (25) Here xS is the saddle point of the deterministic dynamics (16); it can be shown that it is the Hamiltonian trajectory that goes to the saddle point that provides the switching or extinction exponent RA, cf. [3, 33, 63]. Both xa and xS are stationary points of the Hamiltonian H, and the integral over time in (25) goes from −∞ to ∞. This is a significant distinction from the modulated oscillator problem; there equations (22) and (23) for the activation exponent can be written as RA = Z 0 −∞ dtpI(t) I(t), where we set the instant where I(t) reaches IS on the optimal trajectory to be t = 0. A small change of the reaction rates W (X, r) → W (X, r) + ǫW (1)(X, r) (ǫ ≪ 1) leads to the linear in ǫ change of the Hamiltonian, H → H + ǫH(1), as seen from (24). The action is then also changed. To the first order in ǫ, RA → RA +ǫ R(1) A . The correction term is given by a simple expression familiar from the Hamiltonian mechanics [2], ǫ R(1) A = −ǫZ dtH(1)(cid:16)x(t), p(t)(cid:17), H(1) = Xr w(1)(x, r)(epr − 1), (26) where the integral is calculated along the unperturbed trajectory x(t), p(t). In the extinction problem the integral (26) can diverge at the upper limit, t → ∞. This is because in this problem p(t) remains finite for t → ∞, and therefore if w(1)(xS, r) is nonzero, H(1) 6= 0 for t → ∞. The divergence indicates the breakdown of the perturbation theory; in the particular example studied in [28], for ǫ → 0 the change of RA was ∼ RA. For the modulated oscillator, the role of the small parameter ǫ is played by If w(0)(I, r) is the transition rate for ¯n = 0, then from the Planck number ¯n. (11) the thermally induced term in the transition rate has the form ¯nw(1)(I, r) = ¯n[w(0)(I, r) + w(0)(I, −r)]. Where the perturbation theory applies, the correction to the effective activation energy of switching reads ¯nR(1) A = −¯nZ 0 −∞ dtXr w(1)(cid:16)I(t), r(cid:17){exp[rpI(t)] − 1}. (27) Quantum fluctuations in modulated nonlinear oscillators 15 As we saw, in contrast to reaction/population systems, pI 6= 0 for t → −∞. However, w(1)(I, r) ∝ I ∝ exp(2κt) for t → −∞, therefore (27) does not diverge for t → −∞. There is also no accumulation of perturbation for large t, as the integral goes to t = 0. Therefore the cause of the fragility should be different from that in population dynamics/reaction systems. As mentioned earlier, in contrast to reaction systems, for the oscillator the values of r in (27) can be large. Then the correction R(1) A can diverge because of the divergence of the sum over r. This happens if on the optimal trajectory w(1)(I, r) decays with r slower than exp(rpI). From (12), w(1)(I, r) decays with r exponentially; in particular, w(1)(I, r) ∝ exp[−2rν(g)τ∗(g)] for r ≫ 1. The region of the values of pI where Pr w(1)(I, r) exp(rpI) remains finite is shown in Fig. 3(b). As seen from this figure, the value of pI on the ¯n = 0-trajectory can be too large for the sum over r to converge. Then the perturbation theory becomes inapplicable. The trajectory followed in switching changes discontinuously where ¯n changes from ¯n = 0 to ¯n > 0. The probability distribution also changes discontinuously. We note that pI ∼ 1 ≪ λ−1 on the optimal fluctuational trajectory, which justifies the approximation (21) that underlies the above analysis. It is clear that the optimal fluctuational trajectory I(t) corresponds to the optimal fluctuational trajectory of the quasienergy g(t), since the I and g variables are related by ∂gI = ν−1(g). The instanton approximation relies on the assumption that the mean square fluctuations provide the smallest scale in the problem, similar to the wavelength in the WKB approximation [42]. If the system is perturbed and the perturbation is small, it can be incorporated into the prefactor of the rate of rare large fluctuations. If the perturbation is still small but exceeds the small parameter of the theory, it can be incorporated into the instanton Hamiltonian and leads to a correction to the exponent of the rare event rates. This correction is generically linear in the perturbation. However, this is apparently not a universal behavior, as the unperturbed solution can be fragile with respect to a perturbation. So far the fragility has been found in cases where the perturbation breaks the time-reversal symmetry. An important problem is the crossover between the instanton solutions without and in the presence of the perturbation. For a modulated quantum oscillator it was recently addressed in [41] (but the most probable fluctuational trajectories were not studied in this paper). The analysis [41] shows that the very instanton approximation breaks down by thermal fluctuations, function ∂I R is not smooth for ¯n > 0, rather it displays a kink. The threshold for the onset of this behavior is exponentially low in ¯n, with ln ¯n . λ−1. It corresponds to the regime where the rate of transitions between oscillator states induced by absorption of thermal excitations, which is ∝ ¯n, becomes comparable with the switching rate Wsw calculated for ¯n = 0. The region where the instanton approximation is inapplicable extends to ¯n . λ3/2. This is why we indicate that the optimal trajectories in Fig. 3 for ¯n → 0 correspond to vanishingly small ¯n compared to the small parameter of the theory λ, yet ¯n & λ3/2. Quantum fluctuations in modulated nonlinear oscillators 16 5. Nonresonant modulation: a brief summary Much attention has attracted recently the possibility of cooling mesoscopic oscillators, and the whole new area, the cavity optomechanics, has emerged, see [64] for a recent review. The cooling is performed by nonresonant modulation, with frequency ωF significantly different from the oscillator frequency ω0. The very idea of cooling different types of quantum systems by a high-frequency field goes back to the mid-70s [37, 65, 66], about the same time when the laser cooling of atomic motion was proposed [67, 68]. The change of the distribution can be understood from Fig. 4 [40, 65]. It refers to a system coupled by the modulating field to another system, which can be a thermal bath or a mode with a relaxation time much shorter than that of the system of interest, so that it serves effectively as a narrow-band thermal reservoir. The modulation provides a new channel of relaxation for the relatively slowly relaxing system of interest. Figure 4 indicates possible transitions between the states of the system accompanied by energy exchange with the thermal reservoir. For example in (a), a transition of the system from the excited to the ground state is accompanied by a transition of the reservoir to the excited state with energy ωb = (ω0 + ωF ), with the energy deficit compensated by the modulation. On the other hand, a transition of the system from the ground to the excited state requires absorbing an excitation in the thermal reservoir, which is possible only when such excitation is present in the first place. The ratio of the state populations of the system is determined by the ratio of the rates of transitions up and down in energy, and thus by the population of the excited states of the thermal reservoir with energy ωb. If the corresponding process is the leading relaxation process, the effective temperature of the system becomes T ∗ = (ω0/ωb)T . It means there occurs effective cooling for ω0 ≪ ωb. Similarly, for ω0 ≫ ωb the modulation leads to heating of the system, see Fig. 4(b). In the case sketched in Fig. 4 (c), the induced transitions from the ground to the excited state of the system are more probable then from the excited to the ground state, which leads to a negative effective temperature for strong modulation. In the case of an oscillator, the system has many levels, but the above picture still applies. The goal of this section is to outline and compare different microscopic mechanisms of the coupling of the oscillator to the modulation and the bath. The unexpected feature is that the distribution of the oscillator over its Fock states can be of the Boltzmann form with an effective temperature determined by the strength and frequency of the modulation [37]. However, this is the case only provided the major mechanism of oscillator relaxation in the absence of modulation is the conventional mechanism (6), which in a phenomenological classical description of oscillator dynamics corresponds to a friction force proportional to the oscillator velocity. A simple model of the modulation-induced dissipation is where the external field parametrically modulates the coupling of the oscillator to a thermal bath. The coupling Hamiltonian is H (F ) i = −qh(F ) b A cos ωF t. (28) Quantum fluctuations in modulated nonlinear oscillators 17
              Figure 4. Modulation-induced relaxation processes leading to cooling (a), heating (b), and population inversion (c); ω0, ωF , and ωb are the frequency of the system (the oscillator, in the present case), the modulation frequency, and the frequency of the mode (or a thermal bath excitation) to which the oscillator is coupled by the modulation, respectively; the relaxation time of the mode is much shorter than that of the oscillator. Strong modulation imposes on the oscillator the probability distribution of the fast-decaying mode in (a) and (b) and leads to population inversion in (c). If, in the absence of modulation, oscillator relaxation is described by the standard model (6), (7), the distribution over the Fock states in the presence of modulation is of the form of the Boltzmann distribution [37]; in (c) the distribution over low-energy Fock states is described by negative temperature and the oscillator vibrates close to its eigenfrequency. b Here h(F ) depends on the variables of a thermal bath, or it can be the coordinate of a comparatively quickly decaying mode coupled to a thermal bath. The interaction (28) has the same structure as the interaction (6), except that it can lead to decay processes with the energy transfer (ω0 ± ωF ), cf. Fig. 4(a) and (b). Therefore the structure of the master equation for the oscillator should not change, but the decay parameters and the Planck numbers of excitations created in decay should change appropriately. The interaction can also lead to decay processes with energy transfer ωF − ω0, for the appropriate modulation frequencies. In this case absorption of bath excitations is accompanied by oscillator transitions down in energy. Respectively, in the master equation (7) in the expression for the rates of transitions due to excitation absorption one has to formally replace ¯n(ω0) → ¯n(ω0 − ωF ) = −¯n(ωF − ω0) − 1, which means that the friction coefficient becomes negative. The above qualitative arguments can be confirmed by a formal analysis similar to that in [37]. It shows that in the RWA the master equation for the oscillator with account taken of the modulation-induced relaxation processes has the form (7) with the relaxation parameter Γ and the Planck number ¯n replaced by ΓF = Γ + Γ+ + Γ− − Γinv and ¯nF , ∂tρ = −ΓF (¯nF + 1)(a†aρ−2aρa† + ρa†a) −ΓF ¯nF (aa†ρ−2a†ρa+ ρaa†), (29) where Γ±,inv = Re Z ∞ 0 A2 8ω0 (cid:12)(cid:12)(cid:12)(cid:12) dth[h(2) b (t), h(2) b (0)]ibei(ω0±ωF )t(cid:12)(cid:12)(cid:12)(cid:12) , ¯nF = {Γ¯n(ω0) + Γ+¯n(ω0 + ωF ) + Γ−¯n(ω0 − ωF ) + Γinv [¯n(ωF − ω0) + 1]} /ΓF . (30) Here, Γ± give the rates of transitions at frequencies ω0 ± ωF , which correspond to the processes sketched in Fig. 4(a) and (b); Γinv gives the rate of processes sketched in Quantum fluctuations in modulated nonlinear oscillators 18 Fig. 4(c), where excitation of the oscillator is accompanied by excitation of the thermal bath. If these processes dominate, they lead to vibrations of the oscillator at frequency ≈ ω0, with amplitude determined by other mechanisms of losses [40]. Parameters Γ− and Γinv in (30) refer to the cases where ω0 > ωF and ω0 < ωF , respectively. From (29) and (30), the probability distribution of the oscillator is characterized by effective temperature TF = ωF /kB ln[(¯nF + 1)/¯nF ]. i = q2h(2) A similar behavior occurs if the modulation is performed by an additive force A cos ωF t, but the interaction with the narrow-band thermal reservoir is nonlinear in the oscillator coordinate, H (2) b . This case was considered in [37]. It reduces to the above formulation if one makes a canonical transformation U(t) = exp[v∗(t)a − v(t)a†] with v(t) = Aosc(2ω0)−1/2(ω0 cos ωF t + iωF sin ωF t), where Aosc = A/(ω2 F ) is the amplitude of forced vibrations of the oscillator. Indeed, as a result of this transformation H (2) in which the field amplitude A is replaced with −2Aosc and i h(F ) b transforms into H (F ) i is replaced with h(2) b . The analysis of cooling of a vibrating mirror in an optical cavity can be also often mapped onto the analysis for the interaction (28). A quantum theory in this case was developed in [69, 70]. It considers an oscillator (the mirror) coupled to a cavity mode driven by external radiation. If the radiation is classical, in the appropriately scaled variables the coupling and modulation are described by Hamiltonians H (m) and H (m) F , respectively, 0 − ω2 i H (m) i = cbqq2 b, H (m) F = −qbA cos ωF t, (31) where q and qb are the coordinates of the mirror and the mode. In cavity optomechanics one usually writes H (m) bab; the following discussion immediately extends to this form of the interaction. i = cbqa† In the absence of coupling to the mirror, the cavity mode is a linear system, hence qb(t) = qb 0(t) + [χb(ωF ) exp(−iωF t) + c.c.]A/2, where qb 0(t) is the mode coordinate in the absence of modulation and χb(ω) is the susceptibility of the mode [70]. The coupling H (m) in the interaction representation then has a cross-term ∝ q0(t)qb 0(t) exp(±iωF t). Since the cavity mode serves as a thermal bath for the mirror, this term is fully analogous to H (F ) , with qb 0 playing the role of h(F ) b . i i 6. Conclusions It follows from the results of this paper that a modulated nonlinear oscillator displays a number of quantum fluctuation phenomena that have no analog in systems in thermal equilibrium. Oscillator relaxation is accompanied by a nonequilibrium quantum noise. It leads to a finite-width distribution of the oscillator over its quasienergy states even for the bath temperature T → 0. For resonant modulation, the distribution is Boltzmann- like near the maximum. We have discussed the recent experiment that confirmed this prediction and the effect of oscillator damping on the outcome of a sideband- spectroscopy based measurement of the distribution. Quantum fluctuations in modulated nonlinear oscillators 19 The quantum noise also leads to large rare events that form the far tail of the distribution of the oscillator over quasienergy and to switching between the coexisting states of forced vibrations. We have developed an approach to the analysis of the distribution tail and the switching rate, which makes a direct connection with the analysis of the corresponding problems in chemical and biological systems and in population dynamics. We show that, in a large deviation, the quasienergy of an underdamped oscillator most likely follow a well-defined real trajectory in real time. This trajectory is accessible to measurement. For T = 0, where the oscillator has detailed balance, the most probable fluctuational trajectory is the time-reversed trajectory of the fluctuation-free (mean-field) relaxation of the oscillator to the stable state. Thermal fluctuations break the detailed balance condition and, even where the thermal Planck number ¯n is small compared to the effective Planck constant, lead to an ¯n-independent change of the most probable fluctuational trajectory. We show that the criterion of the fragility, i.e., of a discontinuous change of the optimal fluctuation trajectory with the varying parameter can be formulated in a general form, that applies both to reaction systems with classical fluctuations and to the modulated quantum oscillator. An interesting effect of nonresonant modulation of the oscillator is that it can significantly change the oscillator distribution over the Fock states, leading to heating, cooling, or excitation of self-sustained vibrations depending on the modulation frequency and the coupling to the thermal bath or a mode with a relaxation time shorter than that of the oscillator. We show that different coupling and modulation mechanisms can be described in a similar way and derive explicit expressions for the effective decay rate and temperature of a modulated oscillator. Acknowledgments We are grateful to P. Bertet and V. N. Smelyanskiy for the discussion. This research was supported in part by the ARO, grant W911NF-12-1-0235, and the Dynamics Enabled Frequency Sources program of DARPA. VP also acknowledges support from the ERC Starting Grant OPTOMECH. Appendix A. Power spectrum of the occupation number of a modulated oscillator For some applications, including the experiment [30], of interest is the power spectrum hhn, niiω of the occupation number n = a†a of a modulated oscillator, where dteiωthhK(t)L(0)ii hhK, Liiω = Z ∞ 2π Z 2π/ωF ωF 0 0 hhK(t)L(0)ii = dtih[K(t + ti) − hK(t + ti)i][L(ti) − hL(ti)i]i (A.1) (we provide the definition of the relevant correlator for arbitrary operators K, L). The major contribution to this power spectrum comes from small-amplitude fluctuations Quantum fluctuations in modulated nonlinear oscillators 20 about the stable states of forced vibrations; in the limit of weak damping these states correspond to the extrema of function g(Q, P ). We will disregard fluctuation-induced transitions between the stable states and calculate the correlator (A.1) for each of these states separately; the averaging h...i then means averaging over small-amplitude fluctuations about the corresponding state. However, we will not limit ourselves to small damping; moreover, we will assume that the scaled damping rate κ ≫ λ, so that the spectra do not display the fine structure related to the nonequidistance of the levels gn near the stable states [29, 38]. Operator n(t) smoothly depends on time: it does not have fast-oscillating factors ∝ exp(±iω0t). However, for small κ, n(t) contains terms which oscillate at frequencies ∼ ν0δω and ∼ 2ν0δω with ν0 being the dimensionless frequency of vibrations about the considered stable state (13). To see this, we first note that classical motion about the stable state (Qa, Pa) is described by linearized equations (9) for δQ = Q − Qa, δP = P − Pa. For small κ this motion is decayed vibrations [3] with dimensionless frequency (the imaginary part of the eigenvalues of the equations for δQ, δP ) νa = (cid:0)κ2 + 3r4 a − 4r2 a + 1(cid:1)1/2 From (13) and (A.2), νa → ν0 for κ → 0. , r2 a = Q2 a + P 2 a . (A.2) We now write the operators of the oscillator as a = aa + δa, δa = (2λ)−1/2(δQ + iδP ), aa = (2λ)−1/2(Qa + iPa), n ≈ a† aaa + a† aδa + δa†aa + δa†δa. (A.3) This immediately shows that, indeed, n oscillates at dimensional frequencies νaδω and, with smaller amplitude (quadratic in δQ, δP ), 2νaδω. From (A.1) and (A.3), the leading term in the power spectrum of n is Re hhn, niiω ≈ (r2 Φnn(ω) = δω Re [hhδa, δa†iiω + hhδa†, δaiiω] a/2λ)δω−1Φnn(ω), (A.4) Functions hhδa, δa†iiω and hhδa†, δaiiω were found earlier [71] (see also [38] where the current notations were adopted). Using their explicit form, one can show that, for small decay rate, κ ≪ νa, function Φnn(ω) has two Lorentzian peaks. They are located at ≈ ±νaδω and have halfwidth κδω. Examples of the spectra Φnn(ω) are shown in Fig. 2(b). For the chosen parameters the spectra have two well-resolved peaks. A straightforward but somewhat tedious calculation shows that the ratio of the heights of the peaks of Φnn(ω) approaches ¯ne/(¯ne + 1) for κ/νa → 0, where ¯ne is the effective Planck number for vibrations about the stable state (14). Therefore by measuring this ratio one can reveal and quantitatively characterize the effect of quantum heating. However, for the parameters in Fig. 2(b), even though the peaks are well resolved, the ratio of their heights is different from ¯ne/(¯ne + 1). This ratio as a function of the scaled intensity of the modulating field β is shown by triangles in Fig. 2(a). In the experiment [30] the occupation of excited quasienergy states of the oscillator was detected by attaching the oscillator to a two-level system (qubit). There was Quantum fluctuations in modulated nonlinear oscillators 21 applied an extra field ∝ Fq cos ωqt at frequency ωq close to the transition frequency of the qubit ωge, and then the resulting population of the excited qubit state was measured. The frequencies were chosen in such a way that ω0 − ωq ≫ ωq − ωge, therefore transitions between the qubit states accompanied by transitions between the Fock states of the oscillator have negligible probability. However, one can think that the oscillator modulates the coupling of the qubit to the field Fq. Then phenomenologically one can write the effective Hamiltonian of the driven qubit as Hq = 1 2 ωgeσz − 1 4 [Fq exp(iωqt)σ−(1 + αq n) + H.c] , (A.5) where σz, σ± = σx ± iσy are Pauli matrices. We emphasize that this is a phenomenological "toy" model is a phenomenological parameter; a discussion of the microscopic model will be performed by the authors of Ref [30], it is beyond the scope of this paper. in which we keep only resonant terms, αq Our toy model captures the possibility of transitions between the qubit states induced by the field ∝ Fq and accompanied by transitions between the oscillator quasienergy levels. We will assume that the decay rate of the oscillator is larger than the decay rate of the qubit. Then, to the second order in αq, the contribution to the rate of qubit excitation ↓i → ↑i due to the qubit-oscillator coupling is given by (2)−1αqFq2Rehhn, niiω with ω = ωq − ωge. The coupling-induced contribution to the rate of qubit transitions ↑i → ↓i is given by the same expression with the correlator evaluated for ω = ωge − ωq. However, transitions ↑i → ↓i are more likely to be dominated by spontaneous processes or induced by the n-independent term in (A.5). On the other hand, for ωq − ωge near the peaks of the correlator Φnn(ω), the coupling- induced transitions ↓i → ↑i can have a substantial relative probability and determine the resulting population of the excited qubit state. Then the ratio of the heights of the peaks of Φnn(ω) is given by the ratio of the populations of the excited qubit state for the corresponding frequencies, which was measured in the experiment [30]. We note that our toy model leads also to the occurrence of a small peak in the population of the excited state of the qubit for ωq − ωge close to 2νaδω, which was reported in [30] This peak can be related to the peak in the power spectrum of hhn, niiω for ω ∼ 2νaδω, which results from the term ∝ δa†δa in n in (A.3); we note that a contribution to this peak comes also from the anharmonicity of the oscillator vibrations about the stable state, i.e., from the higher-order terms in the expansion of g(Q, P ) in Q − Qa, P − Pa [72]. References [1] Dykman M I (ed) 2012 Fluctuating Nonlinear Oscillators: from Nanomechanics to Quantum Superconducting Circuits (Oxford: OUP, Oxford) [2] Landau L D and Lifshitz E M 2004 Mechanics 3rd ed (Elsevier, Amsterdam) [3] Dykman M I and Krivoglaz M A 1979 Zh. Eksp. Teor. Fiz. 77 60 -- 73 [4] Dmitriev A P and Dyakonov M I 1986 Zh. Eksp. Teor. Fiz. 90 1430 -- 1440 [5] Kautz R L 1988 Phys. Rev. A 38 2066 -- 2080 Quantum fluctuations in modulated nonlinear oscillators 22 [6] Vogel K and Risken H 1990 Phys. Rev. A 42 627 -- 638 [7] Dykman M I, Maloney C M, Smelyanskiy V N and Silverstein M 1998 Phys. Rev. E 57 5202 -- 5212 [8] Lapidus L J, Enzer D and Gabrielse G 1999 Phys. Rev. Lett. 83 899 -- 902 [9] Siddiqi I, Vijay R, Pierre F, Wilson C M, Frunzio L, Metcalfe M, Rigetti C, Schoelkopf R J, Devoret M H, Vion D and Esteve D 2005 Phys. Rev. Lett. 94 027005 [10] Kim K, Heo M S, Lee K H, Ha H J, Jang K, Noh H R and Jhe W 2005 Phys. Rev. A 72 053402 [11] Aldridge J S and Cleland A N 2005 Phys. Rev. Lett. 94 156403 [12] Stambaugh C and Chan H B 2006 Phys. Rev. B 73 172302 [13] Almog R, Zaitsev S, Shtempluck O and Buks E 2007 Appl. Phys. Lett. 90 013508 [14] Mahboob I, Froitier C and Yamaguchi H 2010 Applied Physics Letters 96 213103 [15] Dykman M I and Smelyanskii V N 1988 Zh. Eksp. Teor. Fiz. 94 61 -- 74 [16] Vogel K and Risken H 1988 Phys. Rev. A 38 2409 -- 2422 [17] Kinsler P and Drummond P D 1991 Phys. Rev. A 43 6194 -- 6208 [18] Marthaler M and Dykman M I 2006 Phys. Rev. A 73 042108 [19] Peano V and Thorwart M 2006 Chem. Phys. 322 135 -- 143 [20] Katz I, Retzker A, Straub R and Lifshitz R 2007 Phys. Rev. Lett. 99 040404 [21] Serban I and Wilhelm F K 2007 Phys. Rev. Lett. 99 137001 [22] Vijay R, Devoret M H and Siddiqi I 2009 Rev. Sci. Instr. 80 111101 [23] Mallet F, Ong F R, Palacios-Laloy A, Nguyen F, Bertet P, Vion D and Esteve D 2009 Nature Physics 5 791 -- 795 [24] Peano V and Thorwart M 2010 EPL 89 17008 [25] Wilson C M, Duty T, Sandberg M, Persson F, Shumeiko V and Delsing P 2010 Phys. Rev. Lett. 105 233907 [26] Verso A and Ankerhold J 2010 Phys. Rev. E 82 051116 [27] Drummond P D and Walls D F 1980 J. Phys. A 13 725 -- 741 [28] Khasin M and Dykman M I 2009 Phys. Rev. Lett. 103 068101 [29] Dykman M I, Marthaler M and Peano V 2011 Phys. Rev. A 83 052115 [30] Ong F R, Boissonneault M, Mallet F, Doherty A C, Blais A, Vion D, Esteve D and Bertet P 2013 Phys. Rev. Lett. 110(4) 047001 [31] Kramers H 1940 Physica (Utrecht) 7 284 -- 304 [32] Touchette H 2009 Phys. Rep. 478 1 -- 69 [33] Kamenev A 2011 Field theory of non-equilibrium systems Cambridge) (Cambridge University Press, [34] Schwinger J 1961 J. Math. Phys. 2 407 [35] Zeldovich B Y, Perelomov A M and Popov V S 1969 JETP 28 308 [36] Zeldovich B Y, Perelomov A M and Popov V S 1970 JETP 30 111 [37] Dykman M I 1978 Sov. Phys. Solid State 20 1306 [38] Dykman M I 2012, in Fluctuating Nonlinear Oscillators: from Nanomechanics to Quantum Superconducting Circuits ed Dykman M I (OUP, Oxford) pp 165 -- 197 [39] Mandel L and Wolf E Camridge, 1995 Optical Coherence and Quantum Optics (Cambirdge University Press) [40] Dykman M I and Krivoglaz M A 1984 Sov. Phys. Reviews vol 5 ed Khalatnikov I M (Harwood Academic, New York) pp 265 -- 441 [41] Guo L, Peano V, Marthaler M and Dykman M I 2013 Phys. Rev. A 87 062117 [42] Landau L D and Lifshitz E M 1997 Quantum mechanics. Non-relativistic theory 3rd ed (Butterworth-Heinemann, Oxford) [43] Bertet P, Ong F R, Boissonneault M, Bolduc A, Mallet F, Doherty A C, Blais A, Vion D and Esteve D 2012, in Fluctuating nonlinear oscillators: from nanomechanics to quantum superconducting circuits ed Dykman M I (Oxford University Press, Oxford) pp 1 -- 31 [44] Langer J S 1967 Ann. Phys. 41 108 -- 157 [45] Coleman S 1977 Phys. Rev. D 15 2929 -- 2936 Quantum fluctuations in modulated nonlinear oscillators 23 [46] Affleck I 1981 Phys. Rev. Lett. 46 388 -- 391 [47] Caldeira A O and Leggett A J 1983 Ann. Phys. (N.Y.) 149 374 -- 456 [48] Larsen D M and Bloembergen N 1976 Opt. Commun. 17 254 -- 258 [49] Sazonov V N and Finkelstein V I 1976 Doklady Akad. Nauk SSSR 231 78 -- 81 [50] Dmitriev A P and Dyakonov M I 1986 JETP Lett. 44 84 -- 87 [51] Wielinga B and Milburn G J 1993 Phys. Rev. A 48 2494 -- 2496 [52] Marthaler M and Dykman M I 2007 Phys. Rev. A 76 010102R [53] Van Kampen N G 2007 Stochastic Processes in Physics and Chemistry 3rd ed (Amsterdam: Elsevier) [54] Anderson R M and May R M 1991 Infectious Diseases of Humans:Dynamics and Control (Oxford University Press, Oxford) [55] Luchinsky D G and McClintock P V E 1997 Nature 389 463 -- 466 [56] Hales J, Zhukov A, Roy R and Dykman M I 2000 Phys. Rev. Lett. 85 78 -- 81 [57] Ray W, Lam W S, Guzdar P N and Roy R 2006 Phys. Rev. E 73 026219 [58] Chan H B, Dykman M I and Stambaugh C 2008 Phys. Rev. Lett. 100 130602 [59] Weiss G H and Dishon M 1971 Math. Biosci. 11 261 -- 265 [60] Leigh E G J 1981 J. Theor. Biology 90 213 -- 239 [61] Jacquez J A and Simon C P 1993 Math. Biosc. 117 77 -- 125 [62] Doering C R, Sargsyan K V and Smereka P 2005 Phys. Lett. A 344 149 -- 155 [63] van Herwaarden O A and Grasman J 1995 J. Math. Biol. 33 581 -- 601 [64] Aspelmeyer M, Kippenberg T J and Marquardt F 2013 ArXiv e-prints 1303.0733 (Preprint 1303.0733) [65] Zeldovich Y B 1974 JETP Lett. 19 74 -- 75 [66] Shapiro V E 1976 Zh. Eksp..Teor. Fiz. 70 1463 [67] Hansch T W and Schawlow A L 1975 Opt. Commun. 13 68 [68] Wineland D and Dehmelt H 1975 Bull. Am. Phys. Soc. 20 637 [69] Wilson-Rae I, Nooshi N, Zwerger W and Kippenberg T J 2007 Phys. Rev. Lett. 99 093901 [70] Marquardt F, Chen J P, Clerk A A and Girvin S M 2007 Phys. Rev. Lett. 99 093902 [71] Serban I, Dykman M I and Wilhelm F K 2010 Phys. Rev. A 81 022305 [72] Andr´e S, Guo L, Peano V, Marthaler M and Schon G 2012 Phys. Rev. A 85(5) 053825 

1312.5593

1

1312

Electronic States of Wires and Slabs of Topological Insulators: Quantum Hall Effects and Edge Transport

We develop a simple model of surface states for topological insulators, developing matching relations for states on surfaces of different orientations. The model allows one to write simple Dirac Hamiltonians for each surface, and to determine how perturbations that couple to electron spin impact them. We then study two specific realizations of such systems: quantum wires of rectangular cross-section and a rectangular slab in a magnetic field. In the former case we find a gap at zero energy due to the finite size of the system. This can be removed by application of exchange fields on the top and bottom surfaces, which lead to gapless chiral states appearing on the lateral surfaces. In the presence of a magnetic field, we examine how Landau level states on surfaces perpendicular to the field join onto confined states of the lateral surfaces. We show that an imbalance in the number of states propagating in each direction on the lateral surface is sufficient to stabilize a quantized Hall effect if there are processes that equilibrate the distribution of current among these channels.

cond-mat.mes-hall

cond-mat

Electronic States of Wires and Slabs of Topological Insulators: Quantum Hall Effects and Edge Transport. 1 Instituto de Ciencia de Materiales de Madrid, (CSIC), Cantoblanco, 28049 Madrid, Spain L. Brey1 and H.A.Fertig2 2Department of Physics, Indiana University, Bloomington, IN 47405 (Dated: August 27, 2018) We develop a simple model of surface states for topological insulators, developing matching re- lations for states on surfaces of different orientations. The model allows one to write simple Dirac Hamiltonians for each surface, and to determine how perturbations that couple to electron spin im- pact them. We then study two specific realizations of such systems: quantum wires of rectangular cross-section and a rectangular slab in a magnetic field. In the former case we find a gap at zero energy due to the finite size of the system. This can be removed by application of exchange fields on the top and bottom surfaces, which lead to gapless chiral states appearing on the lateral surfaces. In the presence of a magnetic field, we examine how Landau level states on surfaces perpendicular to the field join onto confined states of the lateral surfaces. We show that an imbalance in the number of states propagating in each direction on the lateral surface is sufficient to stabilize a quantized Hall effect if there are processes that equilibrate the distribution of current among these channels. PACS numbers: 72.25.Dc,73.20.-r,73.50.-h I. INTRODUCTION. quantized to Topological insulators (TI) are materials with insulat- ing bulks and conducting surfaces. These materials typi- cally are gapped band insulators, where strong spin-orbit coupling has inverted the usual energetic ordering of the bands. Near an interface with the vacuum the bands revert to their usual order, inducing two-dimensional metallic states on their surfaces [1 -- 3]. These surface states have a conical dispersion, and are described by a two dimensional massless Dirac equation centered at a time reversal invariant point in momentum space. The metallic character of these states is protected by time- reversal symmetry, so that a gap can only be opened by perturbations which break this symmetry, e.g. mag- netic or exchange fields. Angle-resolved photoemission spectroscopy has confirmed the existence of these surface states in certain materials [4]. A magnetic field B applied perpendicular to the sur- face of the topological insulator quantizes the orbital mo- tion of the electrons and reorganizes the energy spectrum into Landau levels with energies [5 -- 9] √ E± = ± 2nωc, (1) (cid:113) c where ωc = vF /(cid:96), vF is the speed of the carriers at the Dirac points in the absence of the field, (cid:96) = eB is the magnetic length, and n = 0, 1, 2, .... If the Fermi energy of the system passes through the Dirac point for B = 0, the corresponding Fermi energy for B (cid:54)= 0 is pinned in the n = 0 Landau level, such that half the states in that level are occupied [10]. Particle-hole symmetry suggests that the Hall conductivity vanishes in this situation. If the Fermi level is raised from this zero point, as it passes between the n − th and the (n + 1)-th Landau levels the Hall conductivity from a single surface Dirac point is then σxy = (n + 1 2 ) e2 h . (2) In general, the integer quantized Hall effect is asso- ciated with current-carrying chiral edge modes. Each mode contributes ±e2/h to the Hall conductivity when the Fermi energy passes through it, so one expects only integer quantization is possible, in contrast to the bulk result suggested by Eq. 2. The resolution of this discrep- ancy is that in actual samples, whatever the geometry may be, different surfaces are always connected, and may share edge modes [11 -- 14]. Contacts used to measure the Hall conductance will thus inevitably probe Dirac cones on more than one surface, so that the observed σxy is always quantized in integer units of e2/h. A Hall effect may occur in a system whenever there is time reversal symmetry-breaking, so that it may be in- duced in principle without an external applied magnetic field. Doping the system with magnetic impurities, or placing a surface in proximity to a ferromagnetic insula- tor which can exchange-couple to the TI surface, offer two non-standard methods to induce σxy (cid:54)= 0. In such sys- tems, the effective surface Hamiltonian in the absence of such perturbations has the form H = vF (σxky − σykx), in which the Pauli matrices σx, σy represent electron spin operators. In the presence of an exchange field pointing in the z direction a gap opens in the surface spectrum, and a Chern number associated with the states on such a surface has the value ±1/2e2/h, with sign determined by the direction of the magnetization [15, 16]. This leads to an anomalous half integer contribution of the surface to the Hall conductivity, σxy = ±e2/2h [17 -- 19]. As in the usual quantum Hall effect, the existence of multiple sur- faces in any real geometry prevents a direct observation of a half-integral quantized Hall conductance in such sys- tems: measured Hall conductances are always integrally quantized. An effect much like this has recently been ob- served in thin films of (Bi,Sb)2Te2 doped with Cr atoms [20]. In this work we introduce a simplified approach to an- alyzing transport in TI surfaces which are nominally flat, but where surfaces of different orientations may be con- nected. We show that these simplified surface states give a good accounting of their dispersion within the bulk gap, and develop matching conditions for surfaces of dif- ferent orientations. We then apply the formalism first to the problem of a quantum wire with rectangular cross- section, examining the effect of different exchange fields on opposite surfaces. In particular, we find for equal exchange fields that the quantum wire supports gapless chiral states, but when oriented oppositely these states vanish and there is a gap in the spectrum. In a second application, we consider the quantum Hall problem for a rectangular slab geometry, with field ori- ented perpendicular to one pair of surfaces. We show that the bulk Landau levels couple surface states on sides parallel to the magnetic field, arriving at results very similar to those of Ref. 13. We then analyze the effect of phase-breaking processes by contacting the system to equilibrating voltage probes, and argue that in this cir- cumstance the system should support a quantized Hall effect. This paper is organized as follows. In Section II we introduce the three dimensional Hamiltonian that de- scribes the low energy properties of a prototypical TI such as Bi2Se3. In Section III we obtain Dirac-like Hamil- tonians that describe the various surface states. Section IV is devoted to a discussion of the matching conditions joining states on different surfaces. In Section V we dis- cuss the energy spectrum of the rectangular quantum wire, and Section VI discusses what happens to this when exchange fields are introduced on two parallel surfaces. In Section VII we consider the case of TI surfaces in a magnetic field, discuss the energy spectrum in the quan- tum Hall regime, and then analyze transport in this sys- tem in a multi-terminal geometry with phase-breaking leads. We conclude with a summary in Section VIII. II. BULK HAMILTONIAN The properties of three dimensional topological insu- lators in the Bi2Se3 family of materials can be described by a four band Hamiltonian introduced by Zhang et al. [21]. In the k · p approximation, states near zero energy are controlled by an effective continuum Hamiltonian of the form  M(k) A1kz 0 A1kz −M(k) A2k− 0 A2k− A2k+ M(k) −A1kz −A1kz −M(k) 0 0  , A2k+ where M(k)=M0 − B2(k2 and E(k)=C + D1k2 x + k2 y) − B1k2 x + k2 (3) z, k±=kx ± iky y). The basis states z + D2(k2 H 3D = E(k) + z ,↑>,3 >=p1+ 2 z ,↑>, for which this Hamiltonian is written are 1 >=p1+ z ,↓>, 2 >=−ip2− z ,↓>, and 4 >= ip2− which are hybridized states of the Se p orbitals and the Bi p orbitals, with even (+) and odd (−) parities, and spin up (↑) and down (↓). The Hamiltonian param- eters for a particular material can be obtained by fit- ting to ab initio band structure calculations [22]. In the case of Bi2Se3 the relevant parameters are M0=0.28eV , A1=2.2eV A, A2=4.1eV A, B1=10eV A2, B2=56.6eV A2, C=-0.0068eV , D1=1.3eV A2 and D2=19.6eV A2. In Fig. 1 we plot the band structure of a thick slab of topological insulator with these parameters. The electronic structure is obtained by diagonalizing Eq. 3 with kz → −i∂z, for fixed kx and ky, using basis states in which the wave- functions to vanish at the surfaces of the slab [23]. The system is rotationally invariant in the x-y plane, so that in Fig.1 k represents the magnitude of the in-plane mo- mentum. In what follows, we will be interested in coupling the spin degree of freedom to effective magnetic fields, cre- ated by exchange coupling to magnetic insulators or mag- netized impurities. To do this we need the spin operators in the basis of bulk states. These are [24] Sx = Sy = 0 0 0 −1 1 0 0 0 0 −1 0 0  0 0 1 0  0 0 −i 0  1 0 0 0 0 0 i i 0 0 0 0 −i 0 0 0 0 1 0 0 0 0 −1 0 0 0 0 −1 ,  ,  . and Sz = (4) We will project these operators onto surface states, yield- ing operators which depend on the orientation of the sur- face with respect to the bulk k axes. It is important to take into account their precise form when analyzing the influence of an effective Zeeman field at such surfaces. III. SURFACE HAMILTONIANS An important feature of H 3D is that, due to its non- trivial topology, when a surface is introduced one finds states in the gap which can be represented by Dirac Hamiltonians. In this section we will write down explicit forms for these surface states, following an approach in- troduced by Silvestrov and coworkers [24], albeit in a simplified form which allows an introduction of simple matching conditions between different surfaces. Because of the strong anisotropy of these layered materials, for 3 (cid:113) M0 ± vF compare the dispersion obtained from the Dirac Hamil- tonian Eq.5, ε = D1 y, with the ex- B1 act result obtained from the diagonalization of the 3D Hamiltonian in a thick slab geometry. In the region near the Dirac point, where the dispersion is linear, the Dirac Hamiltonian yields a good description of the surface band structure. k2 x + k2 The two states resulting from the solution of the sur- face problem, which are the envelope functions used in the basis of Eq.5, are (cid:113) (cid:113) ∓i  1+ D1 B1 1− D1 0 0 B1  , v±z= 1√ 2  0 0(cid:113) (cid:113) 1+ D1 B1 1− D1 B1 ±i  . u±z= 1√ 2 (6) For this surface orientation the electron spin operators, formed by projecting the the full spin operators (Eq. 4) onto the two surface states, coincide with the Pauli spin matrices, Sx = σx , Sy = σy and Sz = σz. (7) Thus, magnetic impurities or the proximity of ferromag- netic insulators will open a gap in the Dirac spectrum only when their magnetization has a component in the z direction. As mentioned in the Introduction, Eq. 5 in this case picks up a Dirac mass term. The integral of the Berry's curvature in the vicinity of the (now gapped) Dirac point then becomes half integral, and the resulting contribution to the Hall conductivity of electrons in these states is half-integral [15, 16]. B. ±x Surface The Hamiltonian describing the electrons moving in the surface perpendicular to the ±y direction has the form H±x = M0 ∓ D2 B2 1 − D2 2 B2 2 iA1kz −iA1kz −A2ky . (8) The envelope states forming the basis of this Hamiltonian are (cid:19) (cid:115) (cid:18) A2ky   , v±x= 1√ 2  (cid:113) (cid:113) ∓i 1+ D2 B2 0 0 1− D2 B2 u±x= 1√ 2  . (cid:113) (cid:113) ∓i 0 1− D2 1+ D2 B2 B2 0 FIG. 1: (Color online) Band structure of a thick TI slab, normal to the z direction, of obtained from diagonalizing the Hamiltonian in Eq. 3. Shadow region represents the bulk band structure. The states in the gap correspond to surface states. Dotted lines represent the surface states obtained from the Dirac Hamiltonian Eq. 5. a given surface the states depend on the orientation of that surface. We confine our analysis to surfaces of high symmetry (x, y, z), and define the surface orientation by normal vectors n = ±x, n = ±y and n = ±z. Gener- ally speaking, the strategy is to find states which vanish in all its components on the surface, are evanescent as one moves into the bulk of the system, and are constant on planes of constant depth into the bulk. Such states have energies within the bulk gap, and generically one finds two such states which are degenerate. Following the k · p approximation, one assumes a good approxima- tion to states near this energy can be formed out of lin- ear combinations of these bound surface states (envelope functions) multiplying plane wave states with wavevec- tor parallel to the surface, and projects the bulk Hamil- tonian, Eq. 3, into the space of these two states. This results in a 2 × 2 Hamiltonian, with a Dirac spectrum in the absence of other perturbations. Appendix A details, as an example, how one obtains the surface Hamiltonian for nz. In the following subsections we present the results of such calculations for the three orientations. A. ±z Surface The Hamiltonian describing the electrons moving on the surface with n = ±z has the form (cid:115) (cid:18) (cid:19) H±z = M0±A2 D1 B1 1− D2 1 B2 1 0 −ikx +ky ikx +ky 0 . (5) (9) For this surface orientation the projection of the spin operators in Eq. 4 become This Hamiltonian describes fermions with velocity vF = A2 two dimensional Dirac In Fig.1 we 1− D2 . 1 B2 1 Sx = D2 B2 σx , Sy = σy , Sz = D2 B2 σz. (10) (cid:113) k(nm-1)-0.6-0.4-0.20.00.20.40.6Energy(eV)-0.3-0.2-0.10.00.10.20.3SS Exact Diag.SS Dirac Eq.The non-integral coefficients in Sx and Sz arise because the envelope states, Eq. 9, have non-zero amplitudes for microscopic orbitals with different spin orientations. As in the case of the ±z surface, magnetic impurities polarized with a component in the normal direction to the surface open a gap in the Dirac spectrum. Note that for this surface, Sx ∝ σx because of the diagonal term E(k) in Eq. 3; for D2 = 0, on this surface the component of spin in the x direction will always be zero. C. ±y Surface The projected states and resulting Hamiltonian for the ±y surfaces are qualitatively very similar to those of the ±x surfaces. The Hamiltonian has the form (cid:115) H±y = M0 ± D2 B2 1 − D2 2 B2 2 The envelope states for the states in which the Hamilto- nian, Eq. 11, is expressed are (cid:19) . (11) −A1kz −A2kx (cid:18) A2kx −A1kz   , v±y= ∓(cid:113) (cid:113) 1√ 2 0 1− D2 1+ D2 B2 B2 0  .  (cid:113) ±(cid:113) 1+ D2 B2 0 0 1− D2 B2 u±y= 1√ 2 (12) Finally, the projections of the spin operators, Eq. 4, are in this case Sx = σx , Sy = D2 B2 σy and Sz = D2 B2 σz. (13) IV. MATCHING CONDITIONS As discussed in the Introduction, in many situations one cannot understand the transport properties of a TI based on individual surfaces in isolation; it is neces- sary to understand how these surfaces connect. Towards this end, in this section we develop a simple approach to matching wavefunctions on a line junction separating two perpendicular surfaces, labeled 1 and 2, of a three- dimensional TI. We assume the Fermi energy is in a bulk gap and focus on how these matching conditions impact the surface state spectra and associated conduction prop- erties. Our method uses a general approach [25], in which wavefunctions of a system are matched along some chosen surface that divides the system into disparate pieces, each having different transverse modes into which it is natural to decompose wavefunctions. In principle the matching can be carried out precisely by considering overlaps along the chosen surface of all the transverse modes. In prac- tice it is usually necessary to truncate the number of modes kept, allowing one to obtain approximate results for wavefunctions in some energy interval. This general 4 approach has been quite successful for treating semicon- ductor nanostructures [25], and recently has been use- ful for understanding transport through graphene nanos- tructures [26]. In the present context we are interested in understand- ing spectra and transport when there are only a small number of modes crossing the Fermi energy, with wave- functions confined to the surfaces, while all other trans- verse modes (associated with the bulk) represent states well above or below the Fermi energy. These latter states are incorporated as evanescent states which do not di- rectly contribute to the current in the system, although they quantitatively affect the scattering among conduct- ing modes. The simplest approximation in this situation is to ignore the evanescent modes entirely, leading to an "open mode approximation" [25]. However, this does not define the approximation scheme uniquely, as one may choose the matching surface to obtain the best results. Below we demonstrate that demanding that the trun- cated Hamiltonian be Hermitian effectively singles out a specific set of matching conditions within the open mode approximation. A. Open Mode Approximation n eik(n) n χ(z) z zc(x) x xc(z) n eik(n) ten in the form Ψ(z) = eikyy(cid:80) for the second slab Ψ(x) = eikyy(cid:80) To motivate our matching conditions, it is useful to consider two slabs of the TI system, with surface normals perpendicular to one another, joined through a perpen- dicular junction. Fig. 2 illustrates the corner of such a junction, emphasizing the role of the surface states, which are most important when the Fermi energy is in the bulk gap. For concreteness we assume one of these has hori- zontal surfaces, perpendicular to z, and the other vertical surfaces perpendicular to x. Assuming that the system is uniform along the y direction so that ky is a good quantum number, states of the first slab can be writ- n (z), and n χ(x) n (x). In these expressions χ(z,x) are transverse wavefunctions for the slabs, among which are the surface states dis- cussed in the last section [27] and for this problem are four component vectors; the wavevectors k(n) x,z are deter- mined by the energy of the state, and in most cases are actually complex (i.e., represent evanescent states) if the Fermi energy is in the bulk band gap, and the coefficients c(x,z) are weights which must be related by appropriate n matching conditions. This last requirement in principle can be implemented by matching all components of the wavefunctions on some surface along which the two slabs are joined together. In principle one may choose any convenient surface, and parameterize it as (xλ, zλ), with 0 ≤ λ ≤ 1. For a given set of coefficients on one side of the junction, say {c(z) n }, the coefficients on the other side can in principle be found [25] by a matrix multiplication, n m < x, nz, m > c(z) m with 5 c(x) n =(cid:80) (cid:90) λ=1 < x, nz, m >= dsλe−ik(n) λ=0 z zλ+ik(m) x xλ < χ(x) n (xλ)χ(z) m (zλ) > . Here dsλ is the differential arc length along the joining n (x)χ(z) surface, and < χ(x) m (z) > is the dot product of the four component vectors. In general, the challenge in carrying out this matching is that the full matrix < x, nz, m > is difficult to com- pute. Moreover, typically most of the transverse modes are "closed" -- i.e., they host evanescent states -- and do not contribute directly to current across the junction. In the "open mode" approximation one simply ignores the closed modes and retains only those that are current- carrying at the Fermi energy. In the present context this is particularly simplifying since only the surface modes are open when the Fermi energy is in the bulk gap. FIG. 2: (Color online) Junction between two perpendicular surfaces, 1 and 2 of a three-dimensional TI. The direction of the line junction is µ. xν indicates the effective one dimen- sional coordinate in the surfaces 1 and 2. sλ parameterizes a curve perpendicular to µ (here a straight line) on the surface joining the two slabs at a corner junction. with µ, ν = x, y, z, and Mz,x = M−z,−x = Cxz M−z,x = Mz,−x = Czx Mz,y = M−z,−y = Cyz Mz,−y = M−z,y = Czy M±x,±y = Cxy −γ− γ+ (cid:18) γ+ γ− (cid:19) (cid:18) γ+ −γ− (cid:19) (cid:18) γ+ iγ− (cid:19) (cid:18) γ+ −iγ− (cid:19) (cid:18) 1 0 −iγ− γ+ iγ− γ+ γ− γ+ . 0 1 (cid:19) (17) In the present case it is then natural to retain only the u and v modes detailed in the last section. If one further assumes that the penetration depths of the sur- face states (λ1 and λ2 in the Appendix) are short, such that the phase factor e−ik(n) x xλ has a negligi- ble variation on the joining surface in the region where < χ(x) m (zλ) > is significantly different than zero, the resulting connection between coefficients takes the simple form n (xλ)χ(z) z zλ+ik(m) (cid:19) (cid:18) cx u cx v (cid:19) (cid:18) cz u cz v = Mx,z , (14) where we have written the two open channel coefficients n with n → u, v. The matrix Mx,z has for each surface cµ the form (cid:18) γ+ γ− −γ− γ+ (cid:19) Mx,z = Cxz γ± =(cid:112)[(1 ± D1/B1)(1 ± D2/B2)], with and Cxz =(cid:82) dsλ[eλx 1 xλ − eλx 2 xλ][eλz 1 zλ − eλz 2 zλ], with λ(x,z) (1,2) the corresponding λ(1,2) constants in the Appendix. At this level of approximation, the only relevant in- formation about the joining surface is contained in the constant Cxz. We thus will ultimately choose this con- stant -- implicitly, by choosing the joining surface -- to ob- tain the best approximation, which we will argue below leaves the projected Hamiltonian Hermitian; this choice uniquely fixes the value of the constant. Before turning to this, we summarize the results of the open mode ap- proximation for other possible 90◦ corner junctions with In surfaces normal to principal axes of the structure. general, we write(cid:18) cµ (cid:19) u cµ v (cid:19) (cid:18) cν u cν v = Mµ,ν , (16) (15) B. Hermitian Effective Hamiltonian As discussed above, we would like to choose the Cµ,ν coefficients to optimize the approximation. In particular, in order to obtain sensible results within the approxima- tion scheme, the projected Hamitonian of the full sys- tem should be Hermitian. This guarantees among other things that current will be conserved across the junc- tions. We now show that this requirement uniquely fixes the coefficients Cµ,ν. As a concrete example we return to the geometry il- lustrated in Fig. 2. The system is invariant along the y direction, so that we can consider the system for each ky as one-dimensional, with a single coordinate along the surface, running perpendicular to the line junction. The corner can be "flattened" by taking x < 0 to represent the z surface, and x > 0 to represent the x surface, which we refer to respectively as the 1 and 2 surfaces in what follows. In this notation, the portion of the low energy Hamiltonian which represents the problem has the form hx = iA(x)σy∂x . (18) µ xν xν 1
2
Sλ A(x) is piecewise constant but jumps at x = 0. Poten- tially this leads to problems because matrix elements be- tween arbitrary two-component wavefunctions φ1(x) and φ2(x) may not obey (cid:82) dxφ∗ 1hxφ2 = (cid:82) dx(hxφ1)∗φ2 due to a surface term at x = 0 from integration by parts. In particular [13], the Hamiltonian is only Hermitian if A+φ −† 1 σyφ− 2 = A−φ+† 1 σyφ+ 2 , (19) 1,2 ≡ φ1,2(0±). From Eq. where A± ≡ A(x = 0±) and φ± 14, this means A+φ 2 = A−φ+† −† 1 σyφ− from which we read off 1 σy = A−φ+† −† A+φ 1 σyMxzφ− 2 , 1 σyMxz. Taking the Hermitian conjugate of this yields and since φ+ xzσyφ+ 1 , 1 = A−M† 1 , we arrive at the relation A+σyφ− 1 = Mxzφ− A+ A− = σyM† xzσyMxz. (20) From the form of Eq. 15 we see Mx,z = Cxz(γ+ + iγ−σy), and plugging this into Eq. 20 above, we arrive at the condition C2 xz = A+ A− (γ2 + + γ2−)−1. (21) For the xz line junction, A+ = A1 and A− = A2. Eq. 21 uniquely specifies the matching condition we should use in an open mode approximation to get physi- cally sensible results. It is interesting to note that if one sets φ1 = φ2 in Eq. 19, the resulting condition is pre- cisely what is needed to get current conservation across the junction. Finally, generalizing this result to other corner junctions, we find + + γ2−)−1 (γ2 + + γ2−)−1 (γ2 + + γ2−)−1 (γ2 C2 A2 zx = A1 C2 A1 yx = A2 C2 A2 zy = A1 Cxy = 1. (22) V. TOPOLOGICAL INSULATOR QUANTUM WIRE As a first example of how these matching conditions can be used, we analyze the electronic structure of the surface states of a quantum wire (QW) with rectangular cross section. For this example we neglect the diagonal 6 FIG. 3: (Color online.) Top panel, band structure of a TI QW with Lz=Lx=10nm, as obtained from Eq. 24. Bottom panel, dependence of the ky=0 energy levels on the dimensions of the TI QW. term E(k) in Hamiltonian Eq. 3, which breaks electron- hole symmetry. This allows us to obtain analytical results which are easily understood. The dimensions of the QW are Lx and Lz along the x- axis and z-axis respectively, and it is infinitely long in the y-direction, so that ky is a good quantum number. Given the momentum ky and the energy E, for each surface of the QW (±x,±z) one may find the corresponding elec- tron wavefunctions (Ψ±x,±z), each as a linear combina- tion of the two solutions of the corresponding Dirac-like surface Hamiltonians. Thus, there are eight coefficients that determine the QW wavefunction, which obey four equations of the form in Eq. 16. Explicitly, we define two-component wavefunctions of the form Lx=Lz=10nmky(nm-1)-0.4-0.20.00.20.4Energy(meV)-200-1000100200Lx=LzLx(nm)010203040Gap(meV)020406080100(cid:32) (cid:32) (cid:32) (cid:32) ϕ(+z)(x) = ϕ(+x)(z) = ϕ(−z)(x) = ϕ(−x)(z) = c(+z) u c(+z) v u c(+x) u c(+x) v c(−z) c(−z) c(−x) c(−x) u v v (cid:33) (cid:33) (cid:33) (cid:33) +kx +kz +kx +kz (cid:32) (cid:32) (cid:32) (cid:32) A(+z)eikxx + A(+x)eikzz + A(−z)eikxx + A(−x)eikzz + c(+z) u c(+z) v u c(+x) u c(+x) v c(−z) c(−z) c(−x) c(−x) u v v (cid:33) (cid:33) (cid:33) (cid:33) −kx −kz −kx −kz 7 B(+z)e−ikxx B(+x)e−ikzz B(−z)e−ikxx B(−x)e−ikzz, where kx is the value that, when substituted into Eq. 5, yields a particular energy eigenvalue E, kz is the analo- gous value for Eq. 8, (c(µ)∗ are the normalized eigenvectors of these Hamiltonians, and A(±µ), B(±µ) are coefficients which must be determined by matching at the corners. These matching conditions are † ) ±kν c(µ)∗ u v ϕ(+z)(x = 0)= Mz,x ϕ(+x)(z = 0) ϕ(−z)(x = 0)= M−z,x ϕ(+x)(z =−Lz) ϕ(−z)(x =−Lx)= M−z,−x ϕ(−x)(z =−Lz) ϕ(+z)(x =−Lx)= Mz,−x ϕ(−x)(z = 0). (23) For a given momentum ky, the matching conditions can only all be met at particular energies E =
n,ky that define the QW band structure. If we neglect E(k) in Eq. 3, it is possible after some algebra to find these energies analytically, with the result (cid:115)
n,ky = ± (cid:18) (A2ky)2 + π A1A2 A1Lz + A2Lx (n − 1 2 ) (24) with n=1,2,3..., and we have made the further simplifying assumption that B1 = B2 = D1 = D2 = 0. This band structure is spin degenerate. Eq. 24 can be easily rationalized with a geometrical argument. When a carrier moves in a closed loop around the quantum wire, the matching of the wavefunctions yields the quantization condition 2(kxLx + kzLz) + π = 2πn . (25) The phase π in the left part of the quantization equa- tion appears because of the helical nature of the carri- ers: when the electrons encircle the QW, the expecta- tion value of the Pauli matrices that appear in the Dirac Hamiltonians rotates by 2π, so that the wavefunction ac- quires a Berry phase of π. (An analogous accumulation of phases occurs in graphene hexagonal quantum rings with discrete 120◦ corners [28].) Moreover, the wavevectors kz (cid:19)2 , and kx are related to the energy by (cid:113) (cid:113) E = (A1kz)2 + (A2ky)2 = (A2kx)2 + (A2ky)2 . (26) Combining Eqs. 25 and 26, one obtains the band struc- ture Eq. 24. In Fig. 3 we plot the band structure as a function of the momentum ky for a QW with Lz=Lx=10nm, and the dependence of the ky=0 energy levels on the dimensions of the the QW. Analogous band structures for a TI QW have been obtained in a cylindrical geometry [29]. VI. EXCHANGE FIELDS AND ANOMALOUS QUANTUM HALL EFFECT Interesting physics can be induced in these types of systems by the introduction of time-reversal symmetry breaking perturbations on the surfaces. As discussed in the Introduction, this can be accomplished by thin film ferromagnets, exchange coupled to one or more surfaces of the system. In particular these can couple to the spin of the electrons without introducing orbital magnetic flux into the Hamiltonian. For a single such surface, for example with normal in the ±z direction, an exchange field (cid:126)∆ parallel to this opens an energy gap (see Eqs. 5-7.) By contrast, if the surface has normal perpendicular to (cid:126)∆, the spectrum re- mains gapless (see Eqs. 8-10.) An isolated gapped sur- face appears to support an anomalous half integer Hall e2 conductivity σxy=sign(∆) 1 h . As discussed in the In- 2 troduction, in real geometries for which there must be a top and bottom surface, the Hall conductivity becomes integrally quantized. In this Section we analyze a TI quantum wire of rectan- gular section in presence of a z-polarized exchange field. ∆ enters as a mass term in the Dirac Hamiltonians for the ±z surfaces, but does not qualitatively modify the Hamiltonians corresponding to the ±x and ±y surfaces. Again, in order to simplify the discussion, we neglect in the Hamiltonian terms proportional to D1 and D2. 8 FIG. 4: (Color online.) Energy bands of a TI QW in presence of equal exchange fields ∆T =∆B=90meV in top and bottom surfaces. Blue circle points are not chiral states, while red squares correspond to chiral states. Dimensions of the QW are Lx=20nm and Lz=5nm (10nm), in the top (bottom) panel. FIG. 5: (Color online.) Energy bands of a TI QW in presence of exchange fields of opposite sign in the the top and bottom surfaces, ∆B=-∆T = 90meV. In this configuration the total Chern number is zero and there are no chiral states. Dimen- sions of the QW are Lx=20nm and Lz=5nm (10nm), in the top (bottom) panel (cid:113) ∆2 + A2 (i) For energies smaller than Fig. 4 illustrates the energy spectrum of a TI QW with lateral dimensions Lz=20nm and Lz=5nm (top panel) and Lz=10nm (bottom panel) in the presence of an ex- change field of magnitude 90meV. There are three kinds 2k2 of states. y, there are states confined to the lateral surfaces, with en- ergies below the gap for states on the exchange-coupled surfaces. The energies of these states depend on the lat- eral size Lz of the wire. For the values of Lz and ∆ illustrated in Fig.4, tunneling between states on oppo- site lateral surfaces is essentially negligible, so that these states are nearly doubly degenerate; deviations from this are only apparent at energies very close to ∆. (ii) At energies larger than ∆ the states extend along the entire perimeter of the TI QW. Because time reversal symme- try is broken by the exchange field, these states are not degenerate. (iii) Finally, there gapless modes with linear dispersion ±A2ky. These describe chiral states moving in opposite directions on opposite lateral surfaces. States of type (i) and (ii) are not chiral: for each state there is a counter-propagating state on the same surface. Impurities can induce backscattering among these states and lead to localization. Chiral states moving in oppo- site directions reside on opposite surfaces, and for a wide enough system, backscattering is negligible. Magneti- cally gapped top and bottom surfaces are always con- nected by surfaces with these chiral states, so that the anomalous Hall conductivity of the system as a whole is e2/h. To gain more insight into the nature of the chiral states, we look for a criterion that determines when they are present. Consider a system in which the top and lay- ers are perturbed by exchange fields ∆T and ∆B, respec- tively. We look for wavefunctions on a single lateral sur- face with momentum kz=0, and energy E=sA2ky, with s=±1. In this geometry the top and bottom wavefunc- tions (extending into the x− y plane with x < 0) and the lateral wavefunction (at x = 0) have the form ky(nm-1)-0.4-0.20.00.20.4Energy(meV)-200-1000100200ky(nm-1)-0.4-0.20.00.20.4Energy(meV)-200-1000100200Lx=20nmLz=10nmΔΤ=ΔB=90meVLx=20nmLz=5nmΔΤ=ΔB=90meVky(nm-1)-0.4-0.20.00.20.4Energy(meV)-200-1000100200ky(nm-1)-0.4-0.20.00.20.4Energy(meV)-200-1000100200Lx=20nmLz=10nmΔΤ= -ΔB=90meVLx=20nmLz=5nmΔΤ= -ΔB=90meV(cid:32) (cid:32) (cid:18) 1 − s 1 + s 1 sA2ky−∆T A2ky+∆T sA2ky−∆B A2ky+∆B −1 (cid:19) + β e∆T x, (cid:33) (cid:33) (cid:18) 1 − s 1 + s e∆T x, (cid:19) ϕ(+z) = C ϕ(−z) = C(cid:48) ϕ(x) = α . (27) Solutions with energy ±sA2ky will exist if these wave- functions satisfy the boundary conditions Eq.16 at the matching points (x=0,z=0) and (x=0,z=−Lz), (cid:32) (cid:32) C C(cid:48) 1 sA2ky−∆T A2ky+∆T −1 sA2ky−∆B A2ky+∆B (cid:33) (cid:33) (cid:18) α + β (cid:19) (cid:18) −(α + β)s s(α + β) α + β = = , (cid:19) . (28) For top and bottom exchange fields with the same sign, the boundary conditions are only satisfied for s=-1. Therefore in the lateral surface (normal to x-direction) there is a chiral state where the electrons move in the -y-direction with speed A2. Similar equations can be written for the opposite lateral surface, normal to the -x-direction, where the band dispersion is A2ky, and the chiral carriers also move in the y-direction with speed A2, albeit in the opposite direction. Finally, it is interesting to see what happens to this picture when the exchange fields on the top and bottom surfaces point in opposite directions, (cid:126)∆T · (cid:126)∆B < 0. In this case Eqs. 28 have no solutions, and chiral states are not present in the system. Fig. 5 illustrates a full solution of the problem as described in the last section, corrob- orating this structure. This is consistent with general considerations in terms of the surface Chern numbers: the top and bottom surfaces have Chern number ±1/2, so that the system as a whole has Chern number zero. In this situation (and in the absence of gapless lateral states) the system does not exhibit an anomalous quan- tized Hall effect. VII. LANDAU LEVELS, EDGE STATES AND QUANTUM HALL EFFECT IN A TI SLAB. A. Energy spectrum In this section we study the electronic band structure of a TI slab in the presence of a perpendicular magnetic field B. The magnetic field points in the z-direction and does not affect the motion of electrons on surfaces where this is in the plane. We choose the Landau gauge A = (0,−Bx, 0), which does not depend on the coor- dinate y, so that the wavevector ky is a good quantum number. In what follows we again neglect the diagonal 9 FIG. 6: (Color online.) Schematic diagram of a semi-infinite thin slab of TI in presence of a perpendicular magnetic field. The slab, of thickness Lz, is perpendicular to the z-direction, invariant in the y direction and it is defined for x < 0. Carri- ers in the top and bottom surfaces are in the same magnetic field, whereas electrons in the lateral surface are not affected by it. When the guiding center of the electron motion is lo- cated away from the edge of the sample, the electronic wave- functions of the top and bottom surfaces are those of bulk Landau levels. When a guiding center approaches the edge, wavefunctions on top and bottom surfaces become coupled through lateral surface plane waves states. terms involving E(k) in the three dimensional Hamilto- nian Eq. 3. Adopting A2(cid:96)−1 as our unit of energy and (cid:96) =(cid:112)c/eB as our unit of length, the Hamiltonian Eq. 5 in the presence of the magnetic field takes the form (cid:33) −√ 2∂z + z√ 2 0 , (29) H±z = ± √ √ + 0 2∂z + z√ 2 (cid:32) 2(ky − x). The eigenvectors (φ± with z = 2 ) of Eq. 29 are obtained by squaring the eigenvalue equation H±zφ = Eφ, yielding (cid:18)√ z − z2 ∂2 4 (cid:19) (cid:19) 1 , φ± φ1 = 0 E2 2 (cid:18) (30) + φ1 = ±φ2 . (31) 2∂z + + 1 2 z√ 2 Solutions of the above equations that do not diverge at x → −∞ are (cid:19) (cid:18) φ± 1 φ± 2 (cid:32) = α √ 2(ky − x)) √ ( 2 −1( 2(ky − x)) D E2 2 D E2 ± E√ 2 (cid:33) , (32) where Dp(z) are parabolic cylinder functions [30]. Car- riers moving on the lateral surface x are not affected by the magnetic field so that the wavefunctions are eigen- states of the Hamiltonian in Eq. 8. By matching of these z


y
x
B
Lz
10 FIG. 8: (Color online). Energy spectrum of a TI slab of thickness Lz=2(cid:96) with an orbital B-field as in ?? and a Zeeman field coupling with energy Ez=0.2A2(cid:96)−1. center approaches the junction with the lateral surface, coupling between them becomes important and they form bonding and anti-bonding states with the accompanying level repulsion. For ky(cid:96)2 well inside the lateral surface, one finds bound states due to its finite width in the z direction; the energy spacing between these states scales as 1/Lz [13]. Increasing the width of the lateral surface (i.e., the separation between top and bottom surfaces) generates more lateral bound states, but does not affect the separation between Landau levels. The discussion above neglects the coupling of the elec- tron spin to the magnetic field. In the ±z surfaces there is an additional Zeeman coupling, so the Landau level energies become (2A2(cid:96)−1n)2 + E2 Z, (33) E± = ±(cid:113) FIG. 7: (Color online). Energy spectrum of a TI slab of thick- ness Lz=2(cid:96) (top) and Lz=5(cid:96) (bottom). Red square points correspond to wavefunctions that are mainly located in the top and bottom layers and decay exponentially in the lateral surfaces. For ky(cid:96) (cid:28) −1 these states evolve into bulk Landau levels. For large values of -ky(cid:96) the top and bottom surfaces Landau levels are degenerate. For ky(cid:96) (cid:38) −l, top and bot- tom Landau levels couple through the lateral states and the degeneracy is lifted. Black dot points correspond to states confined in the lateral surfaces. The energy spacing of these states scale as 1/Lz[13]. lateral wavefunctions with those on the top and bottom surfaces we obtain the band structure of a semi-infinite TI slab in presence of the magnetic field. The geometry is illustrated schematically in Fig. 6. gies ±√ For the nth Landau level, when −ky(cid:96) ∼(cid:112)2n, this cou- In Fig. 7 we plot the results of such a calculation [31]. For large and negative momentum ky, the guiding center of the electron orbits, ky(cid:96)2, is located well inside the top and bottom surfaces where there is a uniform magnetic field, and the coupling to the lateral surface is very small. The spectrum then consists of double degenerate Landau levels, one each for the top and bottom layers, with ener- 2nA2(cid:96)−1. As ky increases, approaching zero from below, the Landau level wavefunctions approach and ac- quire non-negligible coupling to the lateral surface states. pling becomes important and, for n > 0, the absolute value of the energy decreases. This occurs because the wavefunction penetrates into the (zero-field) lateral sur- face, where the carriers can have a smaller kinetic energy than in the presence of the field [32]. The n = 0 Landau levels of the top and bottom surfaces behave differently because they carry no kinetic energy. When the guiding where EZ = gµBB/2, with µB the Bohr magneton and g the effective Land´e factor. Note that in some topolog- ical insulators this last quantity can be as much as fifty times larger than for free electrons [33]. In Fig. 8 we plot the band structure for a Zeeman coupling Ez=0.2A2(cid:96)−1. The main effects of the Zeeman coupling are to break the electron-hole symmetry, shift the energies of the Landau levels, and to lift the degeneracy between the n = 0 Lan- dau levels. B. Quantum Hall effect The presence of many counterpropagating channels on the lateral surfaces can have important consequences for the quantization of the Hall conductance in this system. When the chemical potential is between Landau levels, for example as in the spectra illustrated in Fig. 7, it is apparent that the number of left- and right-moving chan- nels crossing the Fermi energy are not equal. If there are N + M channels propagating in one direction and N channels propagating in the other on each lateral surface, Lz=2ky-4-2024Energy(A2-1)-4-2024Lz=5ky-4-2024Energy(A2-1)-4-2024EZ=0.2A2l-1Lz=2lkyl-4-2024Energy(A2l-1)-4-202411 such a model [34], we ignore these possibilities to focus on phase-breaking effects. Defining left-moving and right-moving currents be- tween leads n − 1 an n as J (L) n−1/2, respec- tively, the current entering reservoir n has the form [J (R) n+1/2]Γ. Current not absorbed from a channel continues onto the next interval between reservoirs. With the assumption that the reservoir returns all the current it absorbs back into the lateral surface with equal prob- ability among the outgoing channels, one finds n−1/2 and J (R) n−1/2 +J (L) n+1/2 = (1 − Γ)J (R) J (R) n−1/2 = (1 − Γ)J (L) J (L) n−1/2 + n+1/2 + N 2N + M N + M 2N + M Γ(J (L) n+1/2 + J (R) n−1/2) Γ(J (L) n+1/2 + J (R) n−1/2). (34) Note this set of equation guarantees that the net current n+1/2−J (L) J (R) n+1/2 will be the same for all intervals n. They (cid:32) can be recast into a recursion relation of the form (cid:33) (cid:32) (cid:33) (35) (36) (cid:19) , J (R) n+1/2 J (L) n+1/2 = T J (R) n−1/2 J (L) n−1/2 with T = 1 1 − γN (cid:18) 1 − γ(2N + M ) γN −γ(N + M ) 1 where γ = Γ/(2N + M ). This recursion relation allows one to determine the cur- rents on the top edge, as illustrated in Fig. 9, anywhere down the length of the sample, provided the current far to the left of the system is known. An analogous relation can be written for the bottom edge, whose currents de- pend on a boundary condition on the right. These bound- ary conditions are met at current-injecting contacts (not shown in the figure, on lateral surfaces perpendicular to the one illustrated), and determine how the net current down the Hall bar divides between the top and bottom lateral surfaces. 0,t , J (L) n+1/2 , J (L) Because the transfer matrix T is independent of n, one may straightforwardly determine the distribution of current in the left-moving and right-moving chan- n+1/2)†, by expressing these nels, (cid:126)Jn+1/2 = (J (R) in terms of the eigenvectors of T , (cid:126)J0,t ≡ (J (R) 0,t )†, where the corresponding eigenvalues are λ0 = 1 and λt = 1 − γM/(1 − γN ): (cid:126)Jn+1/2 = a0λn (cid:126)Jt. The amplitudes a0,t are determined by the boundary condi- tions mentioned above. Since λt < 1, the component of currents associated with (cid:126)Jt decay away exponentially, representing a transient current distribution that relaxes exponentially as one moves away from the current con- tacts. The eigenvector (cid:126)J0 dictates the current distribu- tion inside the bulk. Solving for its explicit form, one finds J (R) 0 = N/(N + M ): the ratio of currents is proportional to ratio of the number of channels. As (cid:126)J0 + atλn t 0 /J (L) 0 FIG. 9: (Color online). Model of TI slab edge connected to equilibrating leads. Sample edge supports N right-moving channels and N + M left-moving channels. Left (right)- moving current between reservoirs n − 1 and n is labeled as J (n−1/2) . L(R) when the transport on these surfaces is ballistic, the Hall conductance does not turn out to be simply M e2/h [13]. Moreover, the longitudinal resistance on one of the lat- eral surfaces does not vanish. In these circumstances the system does not exhibit a quantized Hall effect. The absence of a quantized Hall effect in these circum- stances can be understood as due to the fact that when current is injected into the system from an ideal lead, since only channels with current directed away from the lead can absorb this current, the distribution of currents among the channels is out of equilibrium. This suggests that the system will support a quantized Hall effect if there are current-conserving mechanisms by which this distribution can relax. Generically this will be the case in real systems. As a simple model, we consider a geometry as il- lustrated in Fig. 9, which uses phase-breaking volt- age probes to equilibrate the populations at the edges [34, 35]. The voltage probes absorb current from each channel with probability Γ, taken to be the same for all the channels. Since voltage probes do not change the current down the lateral surface, the total current ab- sorbed by each probe is also emitted from a reservoir at some chemical potential µn. The current is assumed to be injected into each of the lateral channels with equal probability. In this way the voltage probes have the effect of relaxing the current into an equilibrium distribution. Although one can include backscattering among channels at the edge due the leads as well as quantum interference among the various edge and voltage probe channels in N+MNReservoirnReservoirn-1Reservoirn+1ܬோ(௡ିଵଶ)ܬ௅(௡ିଵଶ)ܬ௅(௡ାଵଶ)ܬோ(௡ାଵଶ)expected, the current relaxes into an equilibrium distri- bution in which the current carried by each channel at an edge is equal. That the system exhibits a quantized Hall effect can easily be seen from this last result. For any two volt- age contacts n1,2 which are both far from the current contacts, the transient part of the current distribution is negligibly small [35], so that (cid:126)Jn1+1/2 = (cid:126)Jn2+1/2 = (cid:126)Jn1−1/2 = (cid:126)Jn2−1/2. It follows that the chemical po- tentials in these voltage probes must be the same, so that the measured longitudinal resistance will vanish. To find the Hall resistance, we define J C T (B) as the difference in current carried by each channel on the top (bottom) edge due to currents injected/removed by the contacts far to the right and left of the system. On the top edge, the resulting extra current into a voltage contact is then Γ(2N + M )J C T . An equal current must then exit from the voltage reservoir back into the system. Assuming the reservoir also has 2N + M channels, detailed balance re- quires the probability of tunneling from a reservoir chan- nel back into an edge channel is also Γ. The extra current per channel exiting the reservoir must then be J C T , which fixes the change in chemical potential in the reservoir, δµT = hJ C T /e2. Analogous reasoning fixes the chemical potential change for voltage probes well inside the Hall B /e2. Finally, bar along the bottom edge to be δµB = hJ C recognizing that the net current down the length of the h (δµT − δµB), we Hall bar is I = M (J C arrive at a quantized Hall conductance of M e2 h . B ) = M e2 T − J C VIII. SUMMARY In this article we have studied a simplified model of surface states in topological insulators. The model al- lowed us to develop straightforward matching conditions for states on different surfaces, opening the possibility to understand the surface spectra of a variety of mesoscopic systems. Two systems were analyzed in this formalism in detail: a quantum wire of rectangular cross-section, and a slab geometry in the quantum Hall regime. For the rectangular wire, one finds transverse states with a quantization condition that reflects the helicity of the wavefunctions: an effective two component spinor fol- lows a circular trajectory as one moves around a closed path around the wire, inducing a phase that prevents gapless modes from appearing in the spectrum. The re- sulting gap vanishes only as the wire cross-sectional area becomes very large. The application of exchange fields on two of the surfaces changes the topological character of the surface states by inducing a non-vanishing Chern number. This results in chiral states on lateral surfaces which are gapless. In contrast, if the exchange fields on the two surfaces are directed antiparallel, the Chern num- ber vanishes, and chiral states are absent from the spec- trum. We also considered the surface spectrum of a slab in 12 a magnetic field. Landau level states appear on the surfaces perpendicular to the field, which are continu- ously connected to zero field states on the lateral surfaces [13]. The lateral states have unequal numbers of chan- nels propagating in opposite directions along the slab, in direct analogy with what expects of edge states in the quantized Hall effect. The large number of counterprop- agating edge channels spoils the quantum Hall effect in this system in the ballistic regime. This is due to the ex- istence of unequilibrated populations of the channels on a surface due to the injection of current in the system. We found that processes which restore the populations of the edge modes into local equilibrium, as modeled by float- ing voltage contacts along the edge, will lead to a quan- tum Hall effect in the system if voltage measurements are made sufficiently far from the current contacts. IX. ACKNOWLEDGMENTS LB acknowledges fruitful discussions with Alfredo Levy Yeyati. Funding for this work was provided by MEC- Spain via grant FIS2012-33521. HAF acknowledges sup- port by the US-NSF through Grant No. DMR-1005035, and by the US-Israel Binational Science Foundation through Grant No. 2012120. X. APPENDIX A: DIRAC HAMILTONIAN AND METALLIC STATES IN THE z SURFACE In this Appendix we outline how, starting from the three dimensional Hamiltonian Eq. 3, one obtains the Dirac Hamiltonian describing the surface states of a TI. For concreteness we discuss the z-surface, but similar derivations can be carried out for other surface orienta- tions. In this surface the system is invariant in the x and y direction, so that kx and ky are good quantum num- bers. States localized in the surface with energy in the bulk gap must decay exponentially in the bulk. More- over, we adopt a vanishing boundary condition [24] right at the surface, z=0. We thus look for wavefunctions of the form u(kx, ky, λ1,2) ei(kxx+kyy)(cid:0)eλ1z − eλ2z(cid:1) , (37) where Re λ1,2 > 0 and u is the spinor eigenstate of Hamil- tonian Eq. 3 corresponding to kx, ky and kz = −iλ1,2. Note one needs to find two different values of λ1,2 with the same such spinor [24], in order for all four compo- nents of the wavefunction to vanish at z = 0. We are in- terested in the surface Hamiltonian to lowest non-trivial order in the wavevector; therefore, in the spirit of the k · p approximation, we first obtain the eigenstates for kx=ky=0, and then write the finite wavevector Hamilto- nian in this basis. Our approach differs from that of Ref. 24 in dropping the kx and ky dependence of the basis states, which introduces higher order corrections in the wavevectors. This turns out to be a considerable sim- plification which allows us to develop relatively simple matching conditions, as well as to introduce a magnetic field in a straightforward way. Substituting iλ for kz in Eq. 3, the equation det[H3D− EI] = 0, where I is the 4×4 unit matrix, fixes the inverse decay length λ. Again ignoring the diagonal term in H3D, this yields the biquadratic equation (E + D1λ2)2 = (M0 + B1λ2)2 − A2 1λ2 . (38) which fixes possible values of λ. For each energy it is pos- sible to obtain two solutions λ1,2 with Reλ1,2 > 0. The required energy value is found by imposing the condition u(λ1)=u(λ2),(cid:0)H 3D(λ1) − H 3D(λ2)(cid:1) u(λ1) = 0, which implies a further relation 13 This, together with Eq. 38, yields an energy eigenvalue E = D1 M0. Each of the two allowed values of λ further- B1 more admit two eigenvectors of H3D, (cid:113) (cid:113) −i  1+ D1 B1 1− D1 0 0 B1  , v+z= 1√ 2  i 0 0(cid:113) (cid:113) 1+ D1 B1 1− D1 B1  . u+z= 1√ 2 (41) Finally, we project the three dimensional Hamiltonian Eq. 3 in basis states of the form in Eq. 37, using the spinors defined in Eq. 41. To lowest order in kx and ky, this results in the Dirac Hamiltonian (cid:115) (cid:18) (cid:19) (39) H z = D1 B1 M0 +A2 1− D2 1 B2 1 0 −ikx +ky ikx +ky 0 (42) (D2 1 − B2 1 )(λ2 + λ1)2 − A2 1 = 0 . (40) as given in the text. [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [2] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [3] Y. Ando, ArXiv e-prints (2013), 1304.5693. [4] D. Hsieh, et al., Nature 452, 970 (2008). [5] J. W. McClure, Phys. Rev. 104, 666 (1956), URL http: //link.aps.org/doi/10.1103/PhysRev.104.666. [6] P. Cheng, C. Song, T. Zhang, Y. Zhang, Y. Wang, J.-F. Jia, J. Wang, Y. Wang, B.-F. Zhu, X. Chen, et al., Phys. Rev. Lett. 105, 076801 (2010), URL http://link.aps. org/doi/10.1103/PhysRevLett.105.076801. [7] T. Hanaguri, K. Igarashi, M. Kawamura, H. Takagi, and T. Sasagawa, Phys. Rev. B 82, 081305 (2010), URL http://link.aps.org/doi/10.1103/PhysRevB. 82.081305. [8] Y. Jiang, Y. Wang, M. Chen, Z. Li, C. Song, K. He, L. Wang, X. Chen, X. Ma, and Q.-K. Xue, Phys. Rev. Lett. 108, 016401 (2012), URL http://link.aps.org/ doi/10.1103/PhysRevLett.108.016401. [9] A. A. Taskin, S. Sasaki, K. Segawa, and Y. Ando, Phys. Rev. Lett. 109, 066803 (2012), URL http://link.aps. org/doi/10.1103/PhysRevLett.109.066803. [10] R. Jackiw, Phys. Rev. D 29, 2375 (1984), URL http: //link.aps.org/doi/10.1103/PhysRevD.29.2375. [11] L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007), http://link.aps.org/doi/10.1103/PhysRevB. URL 76.045302. [12] D.-H. Lee, Phys. Rev. Lett. 103, 196804 (2009), URL http://link.aps.org/doi/10.1103/PhysRevLett.103. 196804. 81, 121401 (2010), URL http://link.aps.org/doi/10. 1103/PhysRevB.81.121401. [16] I. Garate and M. Franz, Phys. Rev. Lett. 104, 146802 (2010), URL http://link.aps.org/doi/10. 1103/PhysRevLett.104.146802. [17] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). [18] X.-L. Qi et al. , Science 323, 1184 (2009). [19] R. Yu et al. , Science 329, 61 (2010). [20] C.-Z. Chang, J. Zhang, X. Feng, Shen, Z. Zhang, M. Guo, K. Li, Y. Ou, P. Wei, L.-L. Wang, (2013), http://www.sciencemag.org/content/340/6129/167.full.pdf, URL http://www.sciencemag.org/content/340/6129/ 167.abstract. Science 340, 167 al., et J. [21] H. Zhang, et al., Nature Phys. 5, 438 (2009). [22] C.-X. Liu, X.-L. Qi, H. Zhang, X. Dai, Z. Fang, and S.-C. Zhang, Phys. Rev. B 82, 045122 (2010). [23] M. Lasia and L. Brey, Phys. Rev. B 86, 045317 (2012), http://link.aps.org/doi/10.1103/PhysRevB. URL 86.045317. [24] P. G. Silvestrov, P. W. Brouwer, and E. G. Mishchenko, Phys. Rev. B 86, 075302 (2012), URL http://link.aps. org/doi/10.1103/PhysRevB.86.075302. [25] J. Londergan, J. Carini, and D. Murdock, Binding and Scattering in Two-Dimensional Systems (Springer- Verlag, New York, 1999). [26] A. Iyengar, T. Luo, H. A. Fertig, and L. Brey, Phys.Rev.B 78, 235411 (2008). [27] In general the surface states can depend on the in-plane momentum, but this will plays no role in what follows. [13] O. Vafek, Phys. Rev. B 84, 245417 (2011), URL http: [28] T. Luo, A. Iyengar, H. A. Fertig, and L. Brey, //link.aps.org/doi/10.1103/PhysRevB.84.245417. Phys.Rev.B 80, 165310 (2009). [14] Y.-Y. Zhang, X.-R. Wang, and X. C. Xie, Journal of Physics: Condensed Matter 24, 015004 (2012), URL http://stacks.iop.org/0953-8984/24/i=1/a=015004. [15] T. Yokoyama, Y. Tanaka, and N. Nagaosa, Phys. Rev. B [29] R. Egger, A. Zazunov, and A. L. Yeyati, Phys. Rev. Lett. 105, 136403 (2010), URL http://link.aps.org/doi/ 10.1103/PhysRevLett.105.136403. [30] I.S.Gradshteyn and I.M.Ryzhik, Table of Integrals, Se- ries, and Products.- 5th ed. (Academic Press, San Diego Ca., 1994). [31] Similar results may be found in Ref. 13. [32] M. Ramezani Masir, P. Vasilopoulos, A. Matulis, and F. M. Peeters, Phys. Rev. B 77, 235443 (2008), URL http://link.aps.org/doi/10.1103/PhysRevB.77. 235443. [33] J.G.Analytis et al., Nature Phys. 6, 960 (2010). [34] M. Buttiker, Phys. Rev. B 32, 1846 (1985). [35] J. Wang, B. Lian, H. Zhang, and S.-C. Zhang, Phys. Rev. Lett. 111, 086803 (2013), URL http://link.aps.org/ doi/10.1103/PhysRevLett.111.086803. 14 

1303.5628

2

1303

Thermal balance and quantum heat transport in nanostructures thermalized by local Langevin heat baths

Modeling of thermal transport in practical nanostructures requires making trade-offs between the size of the system and the completeness of the model. We study quantum heat transfer in a self-consistent thermal bath setup consisting of two lead regions connected by a center region. Atoms both in the leads and in the center region are coupled to quantum Langevin heat baths that mimic the damping and dephasing of phonon waves by anharmonic scattering. This approach treats the leads and the center region on same footing and thereby allows for a simple and physically transparent thermalization of the system, enabling also perfect acoustic matching between the leads and the center region. Increasing the strength of the coupling reduces the mean free path of phonons and gradually shifts phonon transport from ballistic regime to diffusive regime. In the center region, the bath temperatures are determined self-consistently from the requirement of zero net energy exchange between the local heat bath and each atom. By solving the stochastic equations of motion in frequency space and averaging over noise using the general fluctuation-dissipation relation derived by Dhar and Roy [J. Stat. Phys. 125, 801 (2006)], we derive the formula for thermal current, which contains the Caroli formula for phonon transmission function and reduces to the Landauer-B\"uttiker formula in the limit of vanishing coupling to local heat baths.

cond-mat.mes-hall

cond-mat

Thermal balance and quantum heat transport in nanostructures thermalized by local Langevin heat baths K. Sääskilahti,∗ J. Oksanen, and J. Tulkki Department of Biomedical Engineering and Computational Science, Aalto University, FI-00076 AALTO, FINLAND (Dated: February 13, 2018) Modeling of thermal transport in practical nanostructures requires making trade-offs between the size of the system and the completeness of the model. We study quantum heat transfer in a self- consistent thermal bath setup consisting of two lead regions connected by a center region. Atoms both in the leads and in the center region are coupled to quantum Langevin heat baths that mimic the damping and dephasing of phonon waves by anharmonic scattering. This approach treats the leads and the center region on same footing and thereby allows for a simple and physically trans- parent thermalization of the system, enabling also perfect acoustic matching between the leads and the center region. Increasing the strength of the coupling reduces the mean free path of phonons and gradually shifts phonon transport from ballistic regime to diffusive regime. In the center region, the bath temperatures are determined self-consistently from the requirement of zero net energy ex- change between the local heat bath and each atom. By solving the stochastic equations of motion in frequency space and averaging over noise using the general fluctuation-dissipation relation derived by Dhar and Roy [J. Stat. Phys. 125, 801 (2006)], we derive the formula for thermal current, which contains the Caroli formula for phonon transmission function and reduces to the Landauer-Büttiker formula in the limit of vanishing coupling to local heat baths. We prove that the bath tempera- tures measure local kinetic energy and can, therefore, be interpreted as true atomic temperatures. In a setup where phonon reflections are eliminated, Boltzmann transport equation under gray ap- proximation with full phonon dispersion is shown to be equivalent to the self-consistent heat bath model. We also study thermal transport through two-dimensional constrictions in square lattice and graphene and discuss the differences between the exact solution and linear approximations. PACS numbers: 05.60.Gg, 63.22.-m, 44.10.+i I. INTRODUCTION Recent theoretical and experimental studies of ther- mal properties of materials have demonstrated many ex- otic phononic phenomena such as ballistic and anoma- lous transport [1–3], conductance quantization [4, 5], and phonon tunneling [6, 7]. These discoveries suggest that the ability to manipulate heat flow at microscopic level and to better understand phonon transfer in nanoscale may lead to important technological breakthroughs rang- ing from new materials for thermoelectric conversion [8, 9] to improved thermal management in future elec- tronics [10] and even information processing by phonons [11–13]. Modeling of thermal transport in practical nanostruc- tures typically requires making trade-offs between the size of the system and the completeness of the the model. Consequently, the commonly used models such as Boltz- mann transport equation (BTE) [14], molecular dynam- ics (MD), Landauer-Büttiker (LB) formalism [15, 16] for phonon transfer [4, 17] and full non-equilibrium Green's function (NEGF) method [18, 19] each have dis- tinct strengths and weaknesses. For instance, BTE for phonons is a powerful method that is applicable even for macroscopic systems, but it does not apply well to micro- scopic systems where wave effects such as diffraction are ∗ kimmo.saaskilahti@aalto.fi important. MD can be applied to phonon transport in microscopic systems and accounts, e.g., for wave effects and phonon-phonon scattering due to anharmonicity of the interatomic potential, but it becomes computation- ally heavy for large systems and cannot strictly account for quantum statistics. The LB and NEGF models can fully account for the quantum statistics, but LB assumes ballistic phonon transfer and NEGF is computationally extremely demanding and therefore limited so far to very small systems [20]. As a consequence of the above limitations, none of the above models are well suited for modeling phonon trans- fer in typical nanostructures consisting of a relatively large number of atoms. A very interesting compromise between system size and model completeness is provided by the self-consistent thermal bath (SCTB) model sug- gested by Bolsterli, Rich and Visscher [21]. In the SCTB model, the phonon scattering is mimicked by coupling the atoms to local heat reservoirs whose temperatures are determined from the condition that, in the steady state, there is no net energy transfer between an atom and the corresponding local heat reservoir. The concept was first used to show that for a classical system with bath temperatures equal to the local kinetic temperatures the thermal conductivity of a harmonic one-dimensional chain was rendered finite by the bath couplings. Later it was shown rigorously that in an infinite one-dimensional chain in a non-equilibrium steady state, the system is at local thermal equilibrium [22] and that local heat current is proportional to the thermal gradient, i.e. heat transfer is diffusive [22–25]. SCTB model has also been applied to investigating quantum effects in non-ballistic heat trans- fer [24–27], effects of additional anharmonicity [23, 28–30] and unequal masses [31–33] on heat conduction and the necessary ingredients of thermal rectification [27, 34–38]. Self-consistent heat baths are closely related to Büt- tiker's self-consistent voltage probes [39, 40], which are employed in electron transport as models for dephasing and dissipation caused by inelastic scattering. To ac- count for the inelastic effects using a microscopic model for the voltage probe, D'Amato and Pastawski modeled the probes by one-dimensional tight-binding chains [41] and were able to demonstrate a transition from coherent to diffusive transport. Their work was recently extended by Roy and Dhar [42] to cover simultaneous charge and heat transfer in the presence of a chemical potential and temperature gradient. Momentum-conserving scatterers have also been proposed [43]. In this paper, we extend the SCTB models beyond one-dimensional chains and study the heat transfer and the use of SCTB models in describing quantum thermal transfer in one-dimensional and two-dimensional struc- tures that exhibit geometric as well as phonon-phonon scattering. To describe the dissipative effects in the whole infinite system consisting of two leads and the center re- gion, atoms both in the leads and in the self-consistent center region are coupled to Langevin heat baths. This makes our setup different from the situation considered by Dhar and Shastry in Ref. [44] and Dhar and Roy in Ref. [24], describing purely ballistic phonon trans- port in the leads. We compare the predictions of SCTB with Landauer-Büttiker (LB) formalism and Boltzmann transport equation (BTE). In contrast to LB formalism, phonon transport in the SCTB model is not purely bal- listic due to the interaction with the local heat baths, but we show that the SCTB model reduces to the con- ventional LB model in the limit of vanishing coupling to local heat baths. In a setup where wave effects can be neglected, SCTB is shown to be equivalent to BTE under gray approximation. We demonstrate how the lo- cal bath temperatures are intuitively related to the local energy densities. We compare the exact self-consistent temperature profiles to linear and classical approxima- tions and thereby extend the work by Bandyopadhyay and Segal [27], who, in contrast to the semi-infinite leads studied here, considered purely Ohmic lead couplings. We also extend their work on one-dimensional chains by comparing the quantum and classical temperature pro- files in higher-dimensional structures. The paper is organized as follows. II, we present the computational setup and derive the for- mula for heat currents flowing to the leads and to self- consistent heat baths. This is achieved by first solving the Heisenberg-Langevin equations of motion in Sec. II A and then specifying the statistical properties of noise II B. Sections II C and II D are devoted terms in Sec. to calculating the average heat flow to the baths and presenting physical interpretation for the self-consistent In Sec. 2 FIG. 1. (Color online) (a) A schematic illustration of the system under study. The structure is divided into the left lead, the center region and the right lead. All atoms are cou- pled to spatially uncorrelated quantum Langevin heat baths, which are shown explicitly for one cross-section in (b). In the left and right lead, the temperatures of the local heat baths have prescribed values TL and TR, respectively. In the center region, on the other hand, temperature varies between TL and TR and the bath temperatures are determined self- consistently using the requirement that the average thermal current to each bath vanishes. The leads can contain an in- finite number of atoms, but the center region is finite. Two- dimensional square lattice with nearest neighbor interactions is shown for illustrative purposes, but the basic principle can be applied to any geometry. bath temperatures. For comparison purposes, we also solve the Boltzmann transport equation under gray ap- proximation for the one-dimensional chain in Sec. II E. In Sec. III, we discuss how to solve the self-consistent equations either by iterative means or by linearization. We also present a physically intuitive method of solving the equations, which can be interpreted as describing the transient behavior of the system. As an application of the formalism, we study in Sec. IV A heat transfer in a one-dimensional chain coupled to semi-infinite chains, in the so-called Rubin-Greer setup [45], and highlight the connection to Boltzmann transport equation. We then study thermal transport through two-dimensional con- strictions both for square lattice and the more practical case of graphene in Secs. IV B and IV C. The methods for solving the self-consistent non-linear equations are com- pared in Sec. IV D. Conclusions are finally given in Sec. V. II. THEORY In the theory and results of this paper, we mainly focus on a system that essentially consists of the left lead region, center region and right lead region as shown schematically in Fig. 1. All atoms within the leads are coupled to local Langevin heat baths set to prescribed values TL and TR. The atoms in the center region are coupled to local heat baths whose temperatures are de- termined self-consistently from the requirement of lo- cal current conservation. The coupling to the Langevin heat baths effectively mimics thermalizing events such as phonon-phonon scattering. It is important to stress that in contrast to Landauer-Büttiker model, phonon trans- port is not assumed to be ballistic either in the leads or in the center region. Although our approach of inte- grating out the leads, detailed below, is inspired by the work of Dhar and Shastry [44] and Dhar and Roy [24], the thermalization in the leads in our setup takes place through a coupling to heat baths instead of thermaliza- tion by Ford-Kac-Mazur formalism [46]. This method is physically transparent, since no difference is made be- tween the leads and the center region (except for the bath temperatures). The method also allows to include dissipative effects in the leads, thereby enabling perfect acoustic matching between the leads and the center re- gion. In the following, we first solve the equations of motion in Sec. II A, then specify the statistical properties of the noise in Sec. II B and finally derive the formula for heat currents in Sec. II C. A. Equations of motion The time evolution of atoms in the setup of Fig. 1 consists of two parts. The first part is deterministic and is specified by the system Hamiltonian H and Heisen- berg equations of motion. The second part consists of a stochastic force and friction due to the interaction with the local heat bath and cannot be directly derived from a Hamiltonian [47]. i = (i/)[H, uα The Hamiltonian time evolution of the atomic displace- ment uα i of atom i along direction α ∈ {x, y, z} and corresponding conjugate momentum pα is determined by i the Hamiltonian H and the Heisenberg equations of mo- tion uα i ]. Here [A, B] = AB − BA denotes the commutator and the i = qα atomic displacement uα is defined as the vari- from the equilibrium position qα0 ation of position qα . i The Hamiltonian of the system is, in the harmonic ap- proximation, i = (i/)[H, pα i ] and pα i i − qα0 i H = 1 2 XI (cid:20) p2 I m + uT I KI uI(cid:21) + 1 2 XI XJ6=I uT I VIJ uJ , (1) where index I ∈ {C, L, R} labels the region: C stands for center region, and L and R for the left and right leads, respectively. The displacement and momentum vectors uI and pI contain the displacements and momenta of all particles in region I and we assume the masses m of all atoms to be equal for notational simplicity. The spring constant matrix KI and the coupling matrices VIJ are the block components of the full spring constant matrix K divided into blocks as K =   0 KL VLC VCL KC VCR 0 VRC KR 3 (2)   , where we assumed that the leads do not interact, so VLR = VT RL = 0. The elements of K are obtained from the second derivative of the interatomic interaction en- ergy V as [48] K αβ ij = ∂2V i ∂qβ j ∂qα . 0 (3) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)q=q The equilibrium positions are defined by the condition of zero force ∂V ∂qα i (cid:12)(cid:12)(cid:12)(cid:12)q=q = 0, 0 (4) which must be satisfied for all atoms i and components α. The Heisenberg equations of motion that follow from the quadratic Hamiltonian (1) coincide with the classi- cal equations of motion. Accompanied with the non- Hamiltonian time-evolution arising from the interaction with the heat bath, the equations of motion become muI = −KI uI − XJ6=I VIJ uJ − mγI uI + ξI . (5) The last two terms are Langevin friction and noise terms that turn the Heisenberg equation of motion into a quan- tum Langevin equation [24, 47, 49]. The stochastic force ξI is a vector whose i'th component is the fluctuating force at site i due to the interaction with the local heat bath. The statistical properties of the Langevin terms are discussed in the next section. Focusing on the steady-state behavior enables solv- ing Eq. (5) by Fourier transformation defined, as usual, transforms into by f (ω) = R dteiωtf (t) with the corresponding inverse transformation f (t) = R (dω/2π)e−iωt f (ω). Equation (5) − mω2 uI = −KI uI −XJ6=I VIJ uJ + imγIω uI + ξI . (6) Rearrangement of Eq. (6) gives uI (ω) = gI(ω) XJ6=I VIJ uJ (ω) − ξI (ω)  , (7) where the uncoupled Green's function is defined as gI(ω) = (cid:2)mω2 − KI + imγIω(cid:3)−1 . (8) The uncoupled Green's function includes damping self- energy imγI ω due to coupling to the heat baths. Substituting Eq. (7) for I = L, R to (6) for I = C gives −mω2 uC(ω) VCI gI(ω) = −KC uC (ω) − XI=L,R ×hVIC uC (ω) − ξI (ω)i + imγC ω uC(ω) + ξC (ω) = −KC uC (ω) − XI=L,R [ΣI (ω)uC (ω) − ηI (ω)] (9) + imγC ω uC(ω) + ξC (ω). In the second line, we defined the lead self-energies ΣI (ω) = VCI gI(ω)VIC and the lead-coupled Langevin noise terms ηI (ω) = VCI gI (ω) ξI (ω). The solution to Eq. (10) is uC (ω) = −G(ω) ξC (ω) + XI=L,R ηI (ω)  , (10) (11) (12) (13) where the full Green's function of the center region is G(ω) =  mω2 − KC + imγCω − XI=L,R −1 . ΣI (ω)  (14) Equations (12) and (13) state that thermal fluctua- tions in the leads can propagate to the center region as described by the Green's function gI(ω) and coupling matrix VCI and thereby introduce additional noise terms ηL and ηR in the center region. The self-energies ΣL and ΣR appearing in the Green's function (14) describe the energy shift and broadening of the phonon energy levels in the center region due to the leaking of phonons into the leads. The self-energies of the semi-infinite leads can be determined by using, e.g., the recursive decimation routine by Lopez and Sancho [50]. By integrating out the leads, we have effectively re- placed the lead coordinates by the noise terms ηI and the accompanying self-energies ΣI . In the next section, we derive the fluctuation-dissipation relation connecting the statistical properties of ηI to Im[ΣI (ω)]. B. Noise power spectra The Langevin noise operators ξI appearing in Eq. (5) act as stochastic sources of thermal fluctuations due to coupling to the local heat bath [24, 47, 49]. They are Gaussian random variables with zero mean and covari- ance related to bath temperature. To calculate the sta- tistical averages of observables such as heat current in 4 the center region, we need the covariances of both the bare Langevin noises ξC and the lead noise contributions ηI = VCI gI ξI . Since the bath temperatures in the center region depend on position i, we handle the noise terms originating from the center region and leads separately. In the following, we assume for notational convenience that each atom only has a single degree of freedom corre- sponding to displacement along, say, x-direction. In the general case, local bath at site i will be coupled to dis- placements (ux i ) in different coordinate directions with a single temperature Ti, making the notation a bit more cumbersome but analogous. i , uy i , uz The covariance of the noises produced by the local heat baths at sites i and j in the center region is, for  = kB = 1, [24, 49] h ξCi(ω) ξCj(ω′)i = 2πδ(ω + ω′)Γij (ω) [fB(ω, Ti) + 1] , (15) where the coupling function for Ohmic friction is Γij(ω) = 2mγCωδij. This corresponds to the memoryless friction assumed in Eq. (5), but more general couplings could be straightforwardly included as well. The friction parameter γC determines the strength of coupling to the heat baths and can be interpreted as phonon decay rate [51]. The corresponding scattering time is τC = γ−1 C , which is independent of frequency for Ohmic baths. The term in braces, where the bath temperature Ti appears in the Bose-Einstein function fB(ω, Ti) = [exp(ω/Ti) − 1]−1, can be written in the more trans- parent form fB(ω, Ti) + 1/2 + 1/2, where the first term is the thermal phonon occupation number, the second term comes from zero-point fluctuations and the last term reflects the non-commutative quantum nature of the Langevin operators [52]. One can write the term as the sum of odd and even functions in ω as fB(ω, Ti)+1 = coth(ω/2Ti)/2 + 1/2, and it turns out that the additional factor of 1/2 cancels in all integrals over frequency after proper symmetrization. For the noise operators in the leads, the temperatures of the baths have prescribed values TL and TR, which do not depend on position. Therefore, we can write the covariance directly in matrix form as (I ∈ {L, R}) h ξI (ω) ξI (ω′)T i = 2πδ(ω + ω′)ΓI (ω) [fB(ω, TI ) + 1] , (16) which is useful in the following calculations. Here, the coupling function matrix ΓI is a diagonal matrix with elements ΓI ij (ω) = 2mγI ωδij. Using h ξI (ω)i) = 0 and Eqs. (12) and (16), we see that the noise terms ηI originating from the leads satisfy hηI (ω)i = 0 and hηI (ω)ηI (ω′)T i = 2πδ(ω + ω′)II (ω), (17) with the power spectrum II (ω) = VCI gI (ω)ΓI (ω)gI (−ω)VIC [fB(ω, TI ) + 1] , (18) where we noted that the Green's function gI (ω) is sym- metric, since the spring matrix KI is symmetric. A straightforward calculation shows that gI (ω)ΓI (ω)gI (−ω) = i[gI(ω) − gI (ω)∗], (19) In Eq. (25) and from now on, we drop the index C de- scribing the center region, since the lead coordinates do not appear anymore. Using the equation of motion (5), the symmetrized time derivative taking into account the non-commutativity of ui and pi can be calculated to be 5 so Eq. (18) becomes hi = II (ω) = iVCI (gI (ω) − gI(ω)∗) VIC [fB(ω, TI) + 1] mo + 1 4 Xj Kij ({ ui, uj} + {ui, uj}) (26) 1 1 pi 2 n pi, 4 Xj −(cid:18)mγ u2 = −2VCIIm [gI (ω)] VIC [fB(ω, TI) + 1] = −2Im[ΣI (ω)] [fB(ω, TI) + 1] , (20) (21) (22) where we used the definition (11). Defining the lead cou- pling function ΓI (ω) = −2Im[ΣI (ω)], (23) we see that the power spectrum of the noise caused by the leads can be written as hηI (ω)ηI (ω′)T i = 2πδ(ω + ω′)ΓI (ω) [fB(ω, TI) + 1] . (24) Equation (24) is analogous to Eqs. (15) and (16) except for the form of the coupling matrix ΓI (ω), now defined using the self-energy of the lead as shown in Eq. (23). Equation (24) is one of the main results of this paper, showing that an atomic reservoir (lead) coupled to local heat baths at prescribed temperature can be represented by noise and dissipation terms related by a fluctuation- dissipation relation. In contrast to previous works [24, 44], our model assumes from the beginning that there is damping everywhere in the system. This results, e.g., in a lead Green's function (8) that includes an additional self-energy term imγω, in contrast to the ballistic lead Green's function defined, e.g., below Eq. (2.5) in [24]. Simply adding the damping to the Green's function used in Ref. [24] would not give a consistent mathematical picture of the situation, since the presence of damping also introduces thermal noise through the fluctuation- dissipation relation and thereby modifies the equations of motion. Solution (13) combined with the noise correlations (15) and (24) allows us to calculate the thermal averages of all observables of interest. C. Heat flow to baths In the Heisenberg-Langevin equation of motion (10), the friction and stochastic force terms induce energy ex- change with the heat bath. The energy exchange rate can be calculated from the time derivative of local en- ergy [53]. A natural definition for the local Hamiltonian of atom i in the center region is hi = p2 i 2m + 1 2 Xj uiKijuj. (25) = − Kij ({ ui, uj} − {ui, uj}) i − { ui, ξi}(cid:19) . 1 2 (27) Here {A, B} = AB + BA is the anti-commutator and for simplicity, we have assumed that the particle is not at the boundary so that it is not directly coupled to the leads. The term inside the sum in Eq. (27) is the heat current flowing from site i to site j and the second term in parentheses is the heat current Qi = mγ u2 i − 1 2 [ uiξi + ξi ui] (28) flowing to the local heat bath at site i. As shown in the appendix, the statistical average of the heat current is formed as a sum over the contributions of the left (J = L) and right (J = R) leads and each local heat bath (J ∈ {1, 2, . . . , NC}) as hQii = Z ∞ 0 dω 2π 2γmω2XJ (cid:2)G(ω)ΓJ (ω)G(−ω)T(cid:3)ii × [fB(ω, TJ ) − fB(ω, Ti)]. (29) Here the sum over bath index J ∈ {L, R, 1, 2, . . . , NC } separately accounts for the contribution of each individ- ually treated heat bath to the thermal balance at site i as detailed in the appendix. The coupling matrix ΓJ is defined by Eq. (23) for the lead heat baths (J = L or J = R). For the local heat baths (J ∈ {1, 2, . . . , NC}), the only non-zero element of the coupling matrix ΓJ is JJ = 2γmω. The term [GΓJ G†]ii describes the ther- ΓJ mal coupling between bath J and site i. In the following, we refer to the bath at site i simply as bath i. Using the definition of Γi for local heat baths, we can write Eq. (29) for the heat flow to bath i in the general form hQii = Z ∞ 0 dω 2π ωXJ TiJ (ω) [fB(ω, TJ ) − fB(ω, Ti)] , (30) where the transmission function between baths i and J is TiJ (ω) = Tr(cid:2)Γi(ω)G(ω)ΓJ (ω)G(−ω)(cid:3) . (31) Equation (30) is also valid for the currents flowing to the leads, i.e. for the substitution i → L or i → R, and the derivation proceeds analogously. In this case, one should use the equation of motion (10) in the pi term of (27) to calculate the heat in-flow to center region by the noise term ηI and out-flow by the force term ΣI u(ω). Equation (30) is the multiprobe Landauer-Büttiker for- mula [16] for thermal transfer between several heat baths. Equation (31) is the Caroli formula [54] for phonon trans- mission function, first derived by Mingo and Yang [17] from the mode picture and by Yamamoto and Watanabe [55] using Keldysh formalism. We have rederived the for- mula using local Langevin heat baths and thereby also included dissipative effects in the leads. We point out that although the average heat current hQii vanishes in the self-consistent temperature configu- ration for all local heat baths in the center region, the spectral heat current h Qi(ω)i = 2γmω2XJ (cid:2)G(ω)ΓJ (ω)G(−ω)T(cid:3)ii × [fB(ω, TJ ) − fB(ω, Ti)] (32) to a local heat bath is generally non-zero. For example, a bath may have a net in-flow of high-energy phonons, but then there must be a corresponding net out-flow of low- energy phonons. These non-zero spectral currents lead to the redistribution of phonon energies inside the structure, similarly to the full non-equilibrium Green's function for- malism where generally hQL(ω)i 6= −hQR(ω)i [18]. In the limit of vanishing couplings to local heat baths, γ → 0+, γI → 0+, the lead and center region Green's functions (8) and (14) region reduce to their ballistic counterparts and the only non-zero transmission function is TLR(ω). Equation (30) reduces to hQRi = Z ∞ 0 dω 2π ωTLR(ω) [fB(ω, TL) − fB(ω, TR)] (33) for the current flowing to the right reservoir. This is the two-probe Landauer-Büttiker formula for ballis- tic phonon transfer, derived earlier by various methods [4, 17, 44, 55–57]. D. Physical interpretation of the bath temperatures For the classical self-consistent thermal bath models, the requirement of zero net energy exchange with the local baths was enforced by requiring the bath tempera- tures to be equal to the local kinetic temperatures [21]. The present model allows finding a fully quantum inter- pretation for the self-consistent bath temperatures. To this end, we first note that the heat current flowing to a self-consistent reservoir at site i can be written in the form hQii = γ(cid:26)2hekin i i −Z ∞ 0 dω 2π ωDi(ω)(cid:20)fB(ω, Ti) + 1 2(cid:21)(cid:27) , (34) where the local kinetic energy is ekin i /2 and the local density of states (LDOS) is defined as Di(ω) = i = m u2 6 −4ωmIm[Gii(ω)] [58]. This form results from noting that the first term of Eq. (28) is γm u2 , and the sec- ond term follows from Eq. (A.10) by using the definition of LDOS and dropping the odd term that cancels out in the integration. The self-consistency criterion hQii = 0 then reduces to the requirement i = 2γekin i 2hekin i i = Z ∞ 0 dω 2π ωDi(ω)(cid:20)fB(ω, Ti) + 1 2(cid:21) . (35) The left-hand side of Eq. (35) can be interpreted as the total energy at site i consisting of the kinetic and elastic energies, which are equal in a statistical-mechanical sys- tem according to the virial theorem [59]. Virial theorem is, of course, rigorously valid only for the total kinetic and interaction energies at thermal equilibrium. The right- hand side is the total vibrational energy of an oscillator at temperature Ti. Equation (35) gives a very natural interpretation to the self-consistent bath temperature Ti as a measure of energy located at site i. Note that for Ohmic baths, the integrals in Eqs. (34) and (35) actually diverge, because the density of states scales as D(ω) ∼ −ωIm[1/(ω2 + iγmω)] ∼ ω−2 for ω → ∞, resulting in a logarithmic divergence in the zero- point term. The divergence is, however, cancelled by an identical term in hekin i, making Ti well-defined. i In the classical limit, Eq. (35) reduces to hekin i i = Ti 2 Z ∞ 0 dω 2π Di(ω) = 1 2 Ti, (36) where we used the sum rule R ∞ 0 (dω/2π)Di(ω) = 1. This sum rule has been proven for the electronic case [60] and the proof for phonons is analogous. Equation (36) can be interpreted as the local equipartition theorem analo- gous to the statistical mechanical equipartition theorem hekini = Nf T /2, where Nf is the number of degrees of freedom in the system. Relation (36) is routinely used as the definition of local temperature in classical molec- ular dynamics simulations [61]. Equation (35) suggests a similar definition for quantum systems. E. Solution of the Boltzmann transport equation in 1D chain In a sense, self-consistent thermal bath model (SCTB) can be thought of as the fully wave enabled extension of the gray approximation [14, 62] to the Boltzmann trans- port equation (BTE). Therefore it is instructive to com- pare the results obtained from BTE and SCTB under conditions where the wave-effects are negligible and there are no reflections between the chain and the reservoirs. For the simple one-dimensional string of length L, BTE in the continuum approximation reads v(ω) −v(ω) ∂n+(x, ω) ∂x ∂n−(x, ω) ∂x = − = − n+(x, ω) − n0(x, ω) τ n−(x, ω) − n0(x, ω) τ (37a) , (37b) where n+(x, ω) and n−(x, ω) are the distribution func- tions for states with positive and negative group veloci- ties, respectively. The thermal boundary conditions are n+(0, ω) = fB(ω, TL) and n−(L, ω) = fB(ω, TR). Distri- bution functions relax towards the average distribution n0 = (n+ + n−)/2 with relaxation time τ . Mode disper- sion in a one-dimensional chain is ω(q) = 2ω0 sin(qa/2), where a is the lattice constant. Note that the disper- sion of the discrete chain is used in Eqs. (37a) and [63]). Mode velocity is (37b) as usual (see, e.g., Ref. v(ω) ≡ dω/dq = aω0p1 − (ω/2ω0)2 and the density of states is D(ω) ≡ dq/dω = v(ω)−1. The solution of BTE is n+(x, ω) = f (ω, TL) − C(ω) x 2Λ(ω) n−(x, ω) = f (ω, TL) − C(ω)(cid:20)1 + x 2Λ(ω)(cid:21) , (38a) (38b) where Λ(ω) = τ v(ω) is the scattering length and C(ω) = [fB(ω, TL) − fB(ω, TR)]/[1 + L/2Λ(ω)]. The solution re- sults in the heat current (again for  = kB = 1) 0 Q = Z 2ω0 = Z 2ω0 0 dω 2π dω 2π ω ω v(ω)D(ω) [n+(x, ω) − n−(x, ω)] (39) 1 {z 1 } 1 + L/2Λ(ω) [fB(ω, TL) − fB(ω, TR)], (40) which can be interpreted as Landauer-Büttiker current with the effective transmission function Tef f (ω) = 1 1 + L/2Λ(ω) . (41) QL For L ≫ Λ(ω) and classical statistics (T ≫ ω0), the thermal conductivity κ = lim∆T →0 ∆T derived from Eq. (40) coincides with the expression obtained for the clas- sical self-consistent heat bath model [24, 51] when the BTE relaxation time τ is identified with the inverse of the bath coupling constant γ. This shows the similarity of SCTB and BTE models in infinite classical systems. III. SOLVING THE SELF-CONSISTENT EQUATIONS The bath temperatures in the center region are deter- mined by demanding that the average heat current hQii to the local heat baths i ∈ {1, 2, . . . , NC } given by Eq. (29) vanishes. Since the bath temperatures appear in the Bose-Einstein functions, the equations are non-linear and the temperatures must be solved by using iterative methods or by resorting to linearizing approximations. We use both approaches and compare solutions obtained from the full non-linear equations with linear and classi- cal approximations. Solutions of the non-linear equations are calculated using the Newton-Raphson method and in 7 some cases an integration method based on the existence of a steady state towards which the system evolves. The Newton-Raphson method has previously been used to solve the SCTB equations for 1D chains [27] and is quite an efficient and reliable method for solving more general problems as well, especially when the linearized solution is used as the initial guess. However, each itera- tion of the Newton-Raphson method requires evaluating N 2 C frequency integrals, which makes the method heavy for large systems. A slightly different and potentially better-scaling method for solving the equations can be found by writing a set of equations for the time evolution of the temper- atures of the local baths and letting the system evolve towards the steady state. If the reservoir is imagined to have heat capacity Ci, the temperature Ti of the reservoir changes due to the in-flow or out-flow of thermal current and obeys the differential equation dTi(t) dt = 1 Ci hQi(T1, . . . , TNC )i, (42) where the time t is now macroscopic time such that any fluctuations in Qi vanish in the timescale of interest. This is an ordinary differential equation (ODE) of first or- der that evolves toward a steady-state where the bath temperatures satisfy the self-consistent temperature con- dition hQi(T1, . . . , TNC )i = 0 ∀i ∈ {1, 2, . . . , NC }. We call the method of integrating Eq. (42) the ODE method. In addition to having an intuitive physical interpretation as transient time evolution of the heat bath tempera- tures, ODE method has the advantage that at each time step, one only needs to calculate NC frequency integrals to calculate the time-derivative (dT1/dt, . . . , dTNC /dt). For large systems, the method could therefore provide a good alternative to the Newton-Raphson method. The heat capacity Ci simply affects the time-scaling in Eq. (42) and can be included in the time variable t. The full solution of the nonlinear equations can be avoided by two common approximations that provide a linear set of equations for the bath temperatures [35]. Linear response approximation is based on the assump- tion that temperature differences are small, allowing one to make in Eq. (29) the substitution fB(ω, TJ ) − fB(ω, Ti) → ∂fB ∂T (ω, Tm)(TJ − Ti), (43) where the mean temperature is Tm = (TL + TR)/2. The linearization typically produces too low bath tempera- tures compared to the exact results [27]. By considering a single-site model, we have traced this feature back to the fact that the second derivative of the Bose-Einstein function with respect to temperature is strictly positive. Unlike the linear-response approximation, the classical approximation fB(ω, TJ) − fB(ω, Ti) → 1 ω (TJ − Ti). (44) 8 tries, nearest neighbors are assumed to be connected by harmonic springs with spring constant k = mω2 0, where m is the mass of the atoms. Each atom has only a sin- gle degree of freedom corresponding to, e.g., the atomic displacement in the out-of-plane direction. Unless otherwise stated, we set ω0 = 1 in the follow- ing so that dimensionless temperatures are in units of ω0/kB and thermal currents in units of ω2 0. The di- mensionless friction parameter is then in units of ω0. The friction parameter in the leads is set equal to the fric- tion in the central region, γC = γL = γR = γ. In Secs. IV A, IV B and IV C, all exact self-consistent temperature configurations are calculated using the Newton-Raphson iteration with the linear response temperatures used as the initial guess. Newton-Raphson and ODE methods are compared in Sec. IV D. A. Rubin-Greer chain Due to the lack of geometric scattering in the Rubin- Greer setup of Fig. 2(a), the setup serves as an ideal sim- plified model to compare the basic differences and simi- larities between the exact and approximate solutions of the self consistent problem. Figure 3 compares the self- consistent quantum exact, quantum linear and classical bath temperature profiles in a chain of length N = 5 with friction parameters (a) γ = 10−3 and (b) γ = 0.1. The lead temperatures are set to TL = 0.2, TR = 0.1. In the nearly ballistic system of Fig. 3(a), all temperature profiles are nearly constants as a function of position, because coupling to baths is too weak for efficient ther- malization. For increased damping in Fig. 3(b), there is a clear temperature gradient due to interaction with the heat baths. The temperature gradients of the quantum exact and quantum linear response models are approx- imately the same, but the classical gradient is clearly larger. The most prominent feature in Figs. 3(a) and 3(b) is, however, that the quantum exact temperature is higher than the temperatures obtained in linear approx- imations, as noted also earlier for Ohmic leads [27]. To highlight the difference of the Rubin-Greer setup to the Ohmic reservoirs studied in Ref. [27], we compare in Fig. 4 the temperature profiles for the two setups. In the low-frequency limit ω → 0, the self-energy of the semi-infinite Rubin-Greer chain is ΣRG(ω) ≈ −1 − iω [64]. The real part effectively means that the ends of the chain are free and not coupled to fixed particles as in Ref. [27]. The imaginary part of the low frequency approxi- mation of the Rubin-Greer chain self-energy can then be imitated by an Ohmic self-energy Σo = −iγoω by choos- ing γo = 1. Since only low-frequency phonons are excited at low temperature, the low frequency approximation is fairly accurate at low temperature and the temperature profiles are then expected to agree closely, as verified by Fig. 4 for end temperatures TL = 0.2, TR = 0.1. For γo = 2, the temperature profile is steeper due to the stronger coupling to the external baths at the ends of FIG. 2. (Color online) Illustration of the systems studied in Secs. IV A and IV B: (a) a chain of length N connected to two semi-infinite chains, (b) a constriction of width w and length l between two leads of width W . L layers of atoms in the leads are included in the self-consistent calculation to account for the gradual temperature drop near the constriction. makes the self-consistent equations linear also in lead temperatures, so the scaling of lead temperatures by a constant simply scales the self-consistent bath tempera- tures by the same factor. Both linearizations exclude any non-linear effects such as thermal rectification [35] and produce more symmetric temperature profiles than the non-linear equations due to the equivalence of mapping Ti → TL + TR − Ti and the spatial reflection of the structure. IV. NUMERICAL RESULTS To highlight the pertinent physics and the properties of the exact and approximate solutions of the self-consistent equations, we study in more detail the thermal conduc- tion and temperature profiles in two structures shown in Fig. 2. In the one-dimensional setup of Fig. 2(a), the temperatures are determined self-consistently in the cen- ter region consisting of a chain of N atoms. The chain is connected to two semi-infinite chains interacting with heat baths at constant temperatures TL and TR so that there is no geometric scattering and phonon flow is re- duced only by interactions with the local heat baths. The setup reduces to the Rubin-Greer geometry [45] if the heat baths are removed. In the two-dimensional constriction geometry of Fig. 2(b), two wide leads are connected by a narrow constric- tion. The center region includes not only the constriction but also L layers of lead atoms to account for the effects of temperature drop near the constriction. In both geome- 9 FIG. 4. (Color online) Comparison of temperature profiles in a chain of length N = 5 for two kinds of external reservoirs: Rubin-Greer leads [Fig. 2(a)] and Ohmic reservoirs studied in Ref. [27]. In contrast to Ref. [27], we also couple the par- ticles at the ends of the chain to self-consistent baths. The reservoirs are at temperatures TL = 0.2 and TR = 0.1 and the coupling constant to self-consistent baths inside the chain is γ = 0.1. In the Rubin-Greer setup, the coupling constant in the leads is γL = γR = 0.1. In the case of Ohmic external reservoirs, coupling constant at the ends of the chain is de- noted by γo. Choosing γo = 1 reproduces the low-frequency self-energy of the Rubin-Greer chain, leading to similar tem- perature profiles at low temperatures. The geometries are shown in insets. damping caused by the heat baths. With both weak and strong friction, the classical approximation strongly overestimates thermal current, but the linear response approximation is valid up to TL . 0.6. The classical ap- proximation also makes the current response fully linear. Figure 6 compares the thermal currents given by the self-consistent thermal bath (SCTB) solution and Boltz- mann transport equation (BTE) solution (40) as a func- tion of chain length N . In the SCTB model, the friction parameter in the center region is γ = 0.1 and in the leads γL = γR = 0.1 or γL = γR = 0.001. The string length L in BTE is set to L = (N + 1)a to correspond to a chain of N atoms and the relaxation time in the chain is τ = γ−1 = 10. The temperatures of the baths are set to TL = 0.2 and TR = 0.1 so that the system is in the non- linear low-temperature regime. As expected, the current decreases in both models as a function of chain length due to phonon decay in the chain. The BTE result matches the exact SCTB result perfectly, when the lead friction parameters γL and γR are small, i.e., the leads are as- sumed to be nearly ballistic and the center region friction parameter is tied by the relation γ = τ −1. The require- ment of ballistic leads is natural, since we assumed that the phonon occupation in the leads is given by the Bose- Einstein distribution, which in SCTB model is exactly valid only in the limit of zero broadening, γR = γL → 0+. We have verified that the BTE and SCTB heat currents FIG. 3. (Color online) Bath temperature profiles in a chain of length N = 5 sandwiched between two semi-infinite leads at temperatures TL = 0.2 and TR = 0.1. Friction parameters are (a) γ = 10−3, i.e. the system is nearly ballistic and (b) γ = 0.1. Symmetry requires that T3 = 0.15 for the linearized models, but the quantum exact temperature is higher due to the non-linearity of the Bose-Einstein function. the chain, which also introduces more dissipation in the system. At higher temperature, high-frequency phonons in the non-linear range of self-energy are excited as well, and the Ohmic coupling with γo = 1 cannot reproduce the temperature profile as closely any more (not shown). Figure 5 shows the thermal current Q ≡ hQRi through the chain as a function of left lead temperature TL for fixed right lead temperature TR = 0.2. Friction parame- ters are set to (a) weak friction γ = 10−3 and (b) strong friction γ = 0.1. The length of the self-consistently mod- eled chain is N = 10. In the ballistic limit of Fig. 5(a), the current flowing through the chain at low bias TL ≈ TR is equal to Q = GQ(TL −TR), where the quantum of ther- mal conductance [4] is GQ = πk2 BT /6, which reduces to GQ = πT /6 in present units. When friction is in- creased in Fig. 5(b), currents decrease due to the phonon 10 FIG. 6. (Color online) Exact thermal current as a function of chain length N. The friction parameter in the center region is γ = 0.1. The BTE current (40) matches the self-consistent current if the leads are nearly ballistic, γL = γR = 0.001. The bath temperatures in the left and right semi-infinite chains are TL = 0.2 and TR = 0.1, but the currents also agree at other temperatures for ballistic leads. model and BTE would have been very cumbersome to highlight, if the leads had been described by Ohmic reservoirs as in earlier works [27] instead of Rubin-Greer chains. Because Ohmic baths at the ends of the chain would reflect some of the phonons back to the chain, the thermal boundary conditions for the distribution func- tions of right and left-moving phonons in the BTE for- mulation would have been different from simple Bose- Einstein functions. B. Constriction in two-dimensional lattice In real systems, phonon transport is more complicated than in a one-dimensional chain due to, e.g., phonon re- flections from boundaries. The new features arising from mode mismatch at contacts and other geometric factors will be studied in the constriction geometry of Fig. 2(b), where the atoms are set in a square lattice such that a constriction of width w and length l connects two leads of width W . To account for the effects of temperature drop near the junction, L atom layers in the leads closest to the constriction are also included in the self-consistent calculation. The constriction geometry has been studied earlier using molecular dynamics [65, 66], but in contrast to molecular dynamics, the present methodology allows to include full quantum statistics in the phonon popula- tions. From application point of view, constrictions are interesting due to their ability to act as thermal insu- lators, as noted in a recent experiment in GaAs point contacts [67]. Although the present square lattice model FIG. 5. (Color online) Thermal current Q as a function of lead temperature TL for fixed TR = 0.2. Chain length is N = 10 and the friction parameters are (a) γ = 10−3 and (b) γ = 0.1. With both weak and strong friction, the linear response approximation reproduces the exact current up to very high values of bias. In (a), current Q = GQ(TL − TR) corresponding to the quantum of thermal conductance GQ = πT /6 with T = 0.2 is also shown (black dashed). for small γR = γL agree also at other temperatures. Fig- ure 6 also shows that increasing γL and γR in the leads, which increases scattering, slightly reduces the thermal current flowing through the center region. Despite the similarities between the predictions of BTE and SCTB for the simple 1D geometry, the mod- els are not equivalent. For more complex geometries, the Green's function method, which contains full atom- istic dynamics, wave effects and geometric scattering, is a drastic improvement over solving BTE under gray ap- proximation. The agreement of thermal currents between the SCTB 11 sults in a thermal population whose temperature is higher than the average temperature. In the classical case of Fig. 7(b), on the other hand, symmetry of the self-consistent model requires that the temperature profile is symmetric with respect to spatial reflection and mapping Ti → TL +TR −Ti. Therefore, the central part of the junction is at temperature 0.15. In ad- dition, the temperature profile in the bulk parts exhibits directional features at 45 degree angles with respect to the junction. These features have been observed also ear- lier for similar geometry in classical molecular dynamics simulations [66]. In the quantum profile, these diagonal directional features are absent and the temperature pro- files are more directed straight towards the leads. This feature is even more prominent for narrower constrictions (not shown). The difference between the quantum and classical profiles is most likely related to the transmis- sion properties of high-energy phonons (ω & T ), whose populations are overestimated by classical statistics. An- other major difference between quantum and classical statistics is that the currents flowing through the struc- ture are Q = 84.1 × 10−4 for quantum statistics and Q = 138 × 10−3 for classical statistics, i.e., current is very strongly overestimated by the classical statistics in the low-temperature regime, as noted also for 1D chain [Fig. 5]. Our results indicate that the diagonal temperature pat- terns observed in Fig. 7(b) and in the classical molecular dynamics simulations of Ref. [66] may be washed out by the quantum effects at low temperature. At higher temperature, quantum effects are reduced and the di- agonal features reappear, but only if phonon transport remains close to ballistic. Increasing the temperature also increases phonon-phonon scattering [48], so finding a temperature regime where classical statistics prevail but phonon transport is sufficiently ballistic can be problem- atic. The self-consistently modeled center region of Fig. 7 contains 4873 atoms. For this size of system and tem- perature range, the determination of the quantum linear response temperature profile, which was used as the ini- tial guess for Newton-Raphson iteration, took approxi- mately five hours wall-time with 12 CPU cores. Newton- Raphson iteration converged after three iterations and took approximately 14 hours wall-time. The calculation of the classical temperature profile took approximately 14 hours wall-time as well. The solution of the classical temperature profile is computationally more demanding than calculating the linear response profile, since the pop- ulation functions appearing in the equations decay more slowly and need more integration time. Note that even though the system is smaller than the mean free path of long-wavelength phonons, the use of non-reflecting boundary conditions (i.e., semi-infinite leads) ensures that phonons are not reflected from the boundaries between the center region and the leads back to the junction. If the ends had been thermalized with Ohmic heat baths, reflections from the baths could skew FIG. 7. (Color online) Bath temperature profiles in a w = 5, l = 9 constriction coupled to leads of width W = 71 and L = 35 [see Fig. 2(b)]. Lead temperatures are TL = 0.2 and TR = 0.1. Figures show (a) quantum exact and (b) classical self-consistent bath temperature profiles. Friction parameter is γ = 0.01. The separation of isolines is 0.05 and four contour lines are labeled for convenience. is too primitive to accurately handle the experimental situation, our model could be used to gain insight into the local temperature profiles and diffusive effects inside the constriction. Figure 7 shows the (a) quantum exact and (b) classi- cal temperature profiles in a w = 5, l = 9 constriction coupled to leads of size W = 71 and L = 35 in the low-temperature (TL = 0.2, TR = 0.1) and nearly bal- listic regime (γ = 0.01). The asymmetry arising from the non-linearity of the self-consistency equations is very prominent in the quantum exact profile of Fig. 7(a), as the temperature profile patterns in the left and right sides are visibly different. The junction temperature is approximately 0.17, which is notably higher than the av- erage temperature 0.15. This is a similar effect as noticed in the previous section for the 1D chain: the mixing of the statistics of phonons at hot and cold temperatures re- 12 method also allows to include diffusive effects, which would become important in large systems where the mean-free path is comparable to device dimensions. Each atom now has three degrees of freedom, which are all cou- pled to a single local Langevin heat bath. We set the tem- perature range close to the room temperature, TL = 300 K and TR = 280 K, because the acoustic phonon life- time τ at room temperature is known to be of the or- der of τ = 1 ps [69], suggesting that the bath coupling constant is γ = τ −1 = 1012 s−1. Carbon-carbon inter- actions are modeled by the fourth-nearest-neighbor force constant model [70] with the parameters of Ref. [71], which reproduce the bulk ab-initio phonon spectrum of graphene very accurately, at least for the acoustic modes. The optical modes are not active at room temperature, since they are populated only at temperatures close to T ≈ 1000 K, but are fully included in the model in any case. We have verified the correct implementation of the force constant model by comparing ballistic thermal con- ductances of pure nanoribbons to the results of Ref. [72]. Figures 8(b) and 8(c) show the quantum and clas- sical temperature profiles close to the room tempera- ture. The temperature profiles agree quite closely, which is unexpected since the phonon populations originat- ing from the classical and Bose-Einstein distributions at room temperature are quite different: the highest-lying vibrational energies of graphene correspond to temper- atures of TD ≈ 2300 K. The agreement of temperature profiles therefore suggests that only the low-frequency modes close to the Γ point, for which quantum and clas- sical statistics agree, contribute to the transport and de- tailed temperature profile. On the other hand, the heat flow through the structure is still quite strongly overesti- mated by the classical approximation: quantum current is Q ≈ 2.1 × 10−8 W and the classical Q ≈ 5.1 × 10−8 W. No directional features appear in the studied geometry at room temperature, but lowering the temperature and in- creasing the phonon relaxation time could produce more complex temperature profiles. These studies, as well as investigation of different geometries and their influence on temperature profiles, are left for future work. Note also that approximately only half of the total tempera- ture drop takes place in the constriction. FIG. 8. (Color online) (a) Graphene nanoconstriction. The leads extend infinitely to the left and right, but the tempera- tures are determined self-consistently only for the gray atoms in the shown center region. (b) and (c) Self-consistent bath temperature profiles (K), in (b) quantum exact case and (c) classical approximation. The semi-infinite leads are at tem- peratures TL = 300 K and TR = 280 K. The relaxation time τ = 1/γ is set to 1 ps. the temperature profiles. D. Comparison of the solution methods C. Thermal transport in a graphene constriction The two-dimensional square lattice model is easily ex- tended to real materials such as graphene. It is interest- ing to see, for example, if the directional features ob- served in the square lattice remain for more complex lattice geometries. The example geometry is shown in Fig. 8(a). The junction geometry in graphene has also been studied earlier [68], but our methodology gives ac- cess to local temperature profiles in the constriction. The As a final example of the numerics of the solutions, we compare the Newton-Raphson iteration and the or- dinary differential equation (ODE) method. The differ- ential equation (42) is integrated using the MATLAB R(cid:13) [73] implementation of an explicit Runge-Kutta formula with the Dormand-Prince pair [74] and adaptive step- size. Using an adaptive step-size integrator is neces- sary to avoid slowing down as integration approaches the self-consistent temperature configuration. Integration is stopped when the maximum heat current flowing to the bath is less than 10−5γ. 13 1 + Q2 trative purposes. The contour lines of the target function f (T1, T2) = Q2 2 are also shown. The function f is defined such that the self-consistent temperature config- uration is the global minimum and zero of f . The self- consistent temperatures are ( T1, T2) = (0.1590, 0.1585) and ( T1, T2) = (0.1616, 0.1574) for the cases of Figs. 9(a) and 9(b), respectively. For both weak and strong friction, the Newton- Raphson iteration proceeds similarly: The first iteration step of the Newton-Raphson method slightly misses the solution, but the second iteration already takes temper- atures very close to the self-consistent temperature con- figuration. ODE method, on the other hand, proceeds approximately along the direction of steepest descent in the target function. Since the gradient ∇f = 2JT Q, J being the Jacobian matrix of Q, is not necessarily paral- lel to Q, the path taken by the ODE method is generally not strictly along the steepest descent. For the case of larger γ in Fig. 9(b), the contour lines are elongated forming a canyon-like shape and the heat exchange between the local baths starts to dominate over the heat exchange with the leads. In this case, ODE method does not proceed directly towards the solution. We have noted that such cases can be very difficult to handle for the ODE method, since the residual time inte- gration along the canyon requires a very small step size. Newton-Raphson iteration, on the other hand, always seems to find the solution with only a few iterations. In future, it would be interesting to study how well phonon damping and dephasing induced by the self- consistent heat baths mimics true anharmonic effects. Comparisons could be carried out, for example, by com- paring the classical approximation of self-consistent equa- tions with classical molecular dynamics (MD) simula- tions. Knowing now that the bath temperatures cor- (36)], respond to the local kinetic temperature [Eq. the local bath temperatures could be meaningfully com- pared to the local kinetic energy densities obtained from MD. It would also be worth investigating whether non- diffusive transport effects such as anomalous heat con- duction in one dimension [75] could be reproduced by using a frequency-dependent bath coupling constant. If that is the case, one could study quantum effects in anomalous transport using the Green's function method. V. CONCLUSIONS We studied quantum heat transport in nanostructures using the Green's function method combined with self- consistent heat baths. Semi-infinite leads acting as ther- mal reservoirs were reduced to sources of noise and dissipation in the boundaries of the scattering region. In the scattering region, the temperatures of the heat baths mimicking anharmonic effects were determined self-consistently from the requirement of heat current conservation. The self-consistent bath temperatures were shown to measure local energy density, thereby giving FIG. 9. (Color online) The search for the self-consistent bath temperatures T1 and T2 that satisfy Q1(T1, T2) = Q2(T1, T2) = 0 for an N = 2 chain. The lead tempera- tures are TL = 0.2 and TR = 0.1 and the friction parame- ters are (a) γ = 10−2 and (b) γ = 1. The two methods used for the search are Runge-Kutta-Fehlberg integration of Eq. (42) (dots connected by dashed lines) and Newton-Raphson iteration (crosses connected by dash-dotted lines). The con- tours belong to the "target function" f (T1, T2) that is defined to be the squared sum of the currents flowing to the self- consistent reservoirs, f (T1, T2) = Q1(T1, T2)2 + Q2(T1, T2)2. The self-consistent temperature configuration ( T1, T2) satis- fies f ( T1, T2) = 0 and is also the global minimum of f. The initial guess is T1 = 0.2, T2 = 0.1. Figure 9 shows the comparison of the two methods in the search for self-consistent solution. The setup is the Rubin-Greer setup of Fig. 2(a) with N = 2 and the val- ues of the friction parameter are (a) γ = 0.01, and (b) γ = 1. Although it is best to use, e.g., the linear response approximation temperatures as initial guess for iteration, we use now T1 = TL = 0.2 and T2 = TR = 0.1 for illus- them a meaningful physical interpretation. In the classi- cal limit, local kinetic temperature is equal to the bath temperature. By coupling one-dimensional chain to semi-infinite chains, thereby eliminating contact resistance, we demon- strated the equivalence of thermal currents obtained by the self-consistent thermal bath model and the Boltz- mann transport equation under gray approximation with full phonon dispersion. Self-consistent thermal bath model is, therefore, a physically meaningful method to introduce phonon relaxation to ballistic quantum trans- port models. As an application of the formalism, we presented tem- perature profiles in two-dimensional constrictions and showed that quantum statistics plays a vital role in how directional patterns of temperature emanate from the junction. In a graphene constriction at room temper- ature, the bath temperature profile obtained by classi- cal approximation agreed very closely with the quantum temperature profile, suggesting that quantum effects are not strong and molecular dynamics simulations could be justified under those assumptions. In more general cases, we expect, however, that the Green's function method combined with self-consistent thermal baths is a very use- ful tool in studying quantum heat transfer in the ballistic, diffusive and crossover regimes of phonon transport due to the good balance of complexity, insight and predictiv- ity it offers. We acknowledge the Finnish IT Center for Science and the Aalto Science-IT project for computer time. The work is in part funded by the MIDE and AEF re- search programs of Aalto University and by the Graduate School in Electronics, Telecommunications and Automa- tion (GETA). Appendix: Derivation of Eq. (29) In this appendix, we derive Eq. (29) for thermal cur- rent flowing to a local heat bath. For notational sim- plicity, we combine the lead noises ηL and ηR and center region local bath noises ξC to a single vector variable sJ , where the index J ∈ {L, R, 1, 2, . . . , NC} is now a general index for either a lead bath (J ∈ {L, R}) or a self-consistent local bath (J ∈ {1, 2, . . . , NC}). NC is the number of atoms in the center region. Explicitly, sL = ηL, sR = ηR and si is a vector whose only non-zero component is si i = ξCi, the ith component of ξC . The noise covariances are then hsJ (ω)sJ ′ (ω′)i = 2πδ(ω + ω′)ΓJ (ω) [fB(ω, TJ) + 1] δJJ ′ . (A.1) If index J ∈ {L, R}, the coupling function ΓJ is defined by the self-energy of the lead, Eq. (23). For J = i, the VI. ACKNOWLEDGEMENTS The average of the second term is 14 only non-zero matrix element in the coupling function is Γi ii(ω) = 2mγω (we write γC = γ in this section). Equation (13) can be written (dropping the index C for center region) u(ω) = −G(ω)XJ sJ (ω). (A.2) The heat flowing to an Ohmic bath is obtained by cal- culating the statistical average of the symmetrized heat current Qi = γm u2 i − 1 2 [ uiξi + ξi ui]. (A.3) We proceed term by term. The statistical average of the first term is hγm u2 i i (−iω)ui(ω)e−iωtZ dω′ 2π (−iω′)ui(ω′)e−iω′t(cid:29) 2π = γm(cid:28)Z dω = Z dω ×DsJ = Z dω dω′ 2π 2π γm(−ωω′)e−i(ω+ω′)tXJJ ′ Xjk k (ω′)E j (ω)sJ ′ (A.4) Gij (ω)Gik(ω′) (A.5) γmω2XJ (cid:2)G(ω)ΓJ (ω)G(−ω)T(cid:3)ii [fB(ω, TJ ) + 1] . 2π (A.6) (A.7) (A.8) −h uiξii = −(cid:28)Z dω 2π (−iω)ui(ω)e−iωtZ dω′ 2π ξi(ω′)e−iω′t(cid:29) dω′ 2π (−iω)e−i(ω+ω′)tXJ Xj Gij(ω) 2π = Z dω ×DsJ j (ω) ξi(ω′)E . The only term surviving the sum over baths J is the one corresponding to the local heat bath at site i, so −h uiξii = −2Z dω 2π iGii(ω)γmω2 [fB(ω, Ti) + 1] . (A.9) Combined with the symmetrizing term, one gets − h uiξi + ξi uii = −Z dω 2π 1 2 i[Gii(ω) − Gii(−ω)]γmω2 × [fB(ω, Ti) + 1] (A.10) = −Z dω 2π XJ [G(ω)ΓJ (ω)G(−ω)]ii × γmω2 [fB(ω, Ti) + 1] , (A.11) where we used Eq. (19) with the replacements gI → G and ΓI → PJ ΓJ . Combining Eqs. (A.6) and (A.11), we 15 get hQii = Z dω 2π γmω2XJ (cid:2)G(ω)ΓJ (ω)G(−ω)T(cid:3)ii × [fB(ω, TJ ) − fB(ω, Ti)] . (A.12) Noting that the integrand is an even function finally gives Eq. (29). [1] S. Berber, Y.-K. Kwon, and D. Tománek, Phys. Rev. [30] E. Pereira, R. Falcao, and H. C. F. Lemos, Phys. Rev. E Lett. 84, 4613 (2000). 87, 032158 (2013). [2] P. Kim, L. Shi, A. Majumdar, and P. L. McEuen, Phys. [31] F. Barros, H. C. F. Lemos, and E. Pereira, Phys. Rev. E Rev. Lett. 87, 215502 (2001). 74, 052102 (2006). [3] C. W. Chang, D. Okawa, H. Garcia, A. Majumdar, and [32] A. F. Neto, H. C. F. Lemos, and E. Pereira, Phys. Rev. A. Zettl, Phys. Rev. Lett. 101, 075903 (2008). E 76, 031116 (2007). [4] L. G. C. Rego and G. Kirczenow, Phys. Rev. Lett. 81, [33] L. M. Santana and E. Pereira, Phys. Rev. E 86, 032105 232 (1998). (2012). [5] K. Schwab, E. A. Henriksen, J. M. Worlock, and M. L. [34] E. Pereira and H. C. F. Lemos, Phys. Rev. E 78, 031108 Roukes, Nature 404, 974 (2000). [6] M. Prunnila and J. Meltaus, Phys. Rev. Lett. 105, 125501 (2010). [7] I. Altfeder, A. A. Voevodin, and A. K. Roy, Phys. Rev. Lett. 105, 166101 (2010). [8] A. Majumdar, Science 303, 777 (2004). [9] Y. Dubi and M. Di Ventra, Rev. Mod. Phys. 83, 131 (2011). [10] E. Pop, Nano Res. 3, 147 (2010). [11] M. Terraneo, M. Peyrard, and G. Casati, Phys. Rev. Lett. 88, 094302 (2002). [12] B. Li, L. Wang, and G. Casati, Appl. Phys. Lett. 88, 143501 (2006). (2008). [35] D. Segal, Phys. Rev. E 79, 012103 (2009). [36] E. Pereira, Phys. Lett. A 374, 1933 (2010). [37] E. Pereira, H. C. F. Lemos, and R. R. Ávila, Phys. Rev. E 84, 061135 (2011). [38] R. R. Ávila and E. Pereira, J. Phys. A: Math. and Theor. 46, 055002 (2013). [39] M. Büttiker, Phys. Rev. B 32, 1846 (1985). [40] M. Büttiker, Phys. Rev. B 33, 3020 (1986). [41] J. L. D'Amato and H. M. Pastawski, Phys. Rev. B 41, 7411 (1990). [42] D. Roy and A. Dhar, Phys. Rev. B 75, 195110 (2007). [43] R. Golizadeh-Mojarad and S. Datta, Phys. Rev. B 75, [13] C. W. Chang, D. Okawa, A. Majumdar, and A. Zettl, 081301 (2007). Science 314, 1121 (2006). [44] A. Dhar and B. S. Shastry, Phys. Rev. B 67, 195405 [14] J. Y. Murthy et al., Int. J. Multiscale Comp. Eng. 3, 5 (2003). (2005). [45] R. J. Rubin and W. L. Greer, J. Math. Phys. 12, 1686 [15] R. Landauer, Philos. Mag. 21, 863 (1970). [16] M. Büttiker, Phys. Rev. B 46, 12485 (1992). [17] N. Mingo and L. Yang, Phys. Rev. B 68, 245406 (2003). [18] N. Mingo, Phys. Rev. B 74, 125402 (2006). [19] J.-S. Wang, J. Wang, and N. Zeng, Phys. Rev. B 74, 033408 (2006). [20] J.-S. Wang, J.Wang, and J. Lü, Eur. Phys. J. B 62, 381 (2008). [21] M. Bolsterli, M. Rich, and W. M. Visscher, Phys. Rev. A 1, 1086 (1970). [22] F. Bonetto, J. L. Lebowitz, and J. Lukkarinen, J. Stat. Phys. 116, 783 (2004). [23] E. Pereira and R. Falcao, Phys. Rev. E 70, 046105 (2004). [24] A. Dhar and D. Roy, J. Stat. Phys. 125, 801 (2006). [25] D. Roy, Phys. Rev. E 77, 062102 (2008). [26] W. M. Visscher and M. Rich, Phys. Rev. A 12, 675 (1975). (1971). [46] G. W. Ford, M. Kac, and P. Mazur, J. Math. Phys. 6, 504 (1965). [47] U. Weiss, Quantum Dissipative Systems, 3rd ed. (World Scientific, Singapore, 2008). [48] J. Ziman, Electrons and Phonons: The Theory of Trans- port Phenomena in Solids (Oxford University Press, USA, 2001). [49] G. W. Ford, J. T. Lewis, and R. F. O'Connell, Phys. Rev. A 37, 4419 (1988). [50] M. P. Lopez Sancho, J. M. Lopez Sancho, J. M. L. San- cho, and J. Rubio, J. Phys. F: Met. Phys. 15, 851 (1985). [51] N. Li and B. Li, J. Phys. Soc. Jpn. 78, 044001 (2009). [52] J.-S. Wang, Phys. Rev. Lett. 99, 160601 (2007). [53] R. J. Hardy, Phys. Rev. 132, 168 (1963). [54] C. Caroli, R. Combescot, P. Nozieres, and D. Saint- James, J. Phys. C: Solid State Phys. 4, 916 (1971). [55] T. Yamamoto and K. Watanabe, Phys. Rev. Lett. 96, [27] M. Bandyopadhyay and D. Segal, Phys. Rev. E 84, 255503 (2006). 011151 (2011). [56] D. E. Angelescu, M. C. Cross, and M. L. Roukes, Super- [28] R. Falcao, A. Neto, and E. Pereira, Theoretical and lattices Microstruct. 23, 673 (1998). Mathematical Physics 156, 1081 (2008). [57] D. Segal, A. Nitzan, and P. Hänggi, J. Chem. Phys. 119, [29] F. Bonetto, J. Lebowitz, J. Lukkarinen, and S. Olla, J. 6840 (2003). Stat. Phys. 134, 1097 (2009). [58] W. Zhang, T. S. Fisher, and N. Mingo, Numer. Heat 16 Transfer, Part B 51, 333 (2007). [68] Y. Xu, X. Chen, J.-S. Wang, B.-L. Gu, and W. Duan, [59] H. Goldstein, Classical Mechanics (Addison-Wesley Pub- Phys. Rev. B 81, 195425 (2010). lishing Company, Inc., 1971). [69] N. Bonini, J. Garg, and N. Marzari, Nano Lett. 12, 2673 [60] G. D. Mahan, Many Particle Physics, 3rd ed. (Kluwer (2012). Academic/Plenum Publishers, New York, 2010). [61] S. Lepri, R. Livi, and A. Politi, Phys. Rep. 377, 1 (2003). [62] A. Majumdar, J. Heat Transfer 115, 7 (1993). [63] A. J. Minnich, G. Chen, S. Mansoor, and B. S. Yilbas, [70] R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Physical Properties Of Carbon Nanotubes (Imperial College Press, 1998). [71] L. Wirtz and A. Rubio, Solid State Comm. 131, 141 Phys. Rev. B 84, 235207 (2011). [64] P. E. Hopkins, P. M. Norris, M. S. Tsegaye, and A. W. Ghosh, Journal of Applied Physics 106, 063503 (2009). [65] S. K. Saha and L. Shi, J. Appl. Phys. 101, 074304 (2007). [66] K. Sääskilahti, J. Oksanen, R. P. Linna, and J. Tulkki, (2004). [72] Z. Huang, T. S. Fisher, and J. Y. Murthy, J. Appl. Phys. 108, 094319 (2010). [73] MATLAB, 8.0.0.783 (R2012b) (The MathWorks Inc., Natick, Massachusetts, 2012). Phys. Rev. E 86, 031107 (2012). [74] J. Dormand and P. Prince, J. Comp. Appl. Math. 6, 19 [67] T. Bartsch, M. Schmidt, C. Heyn, and W. Hansen, Phys. (1980). Rev. Lett. 108, 075901 (2012). [75] A. Dhar, Adv. Phys. 57, 457 (2008). 

1311.7491

1

1311

The influence of the surface roughness on dielectric function of two-dimensional electron gas

Low field response function calculations have been performed on a two-dimensional electron gas with well-defined electron-surface roughness scattering. The Lindhard model was employed to compute the response function. In particular, detailed investigations were made on the system searching for an interplay between surface roughness with well-defined correlation function, (characterizes by asperity height and correlation length) spatial confinement and the dielectric function. We analyze to what extent the normal behavior and functionality of dielectric function of two-dimensional devices are modified by random scattering events caused by the contribution from the surface roughness. Results of the current work indicate that contribution of the surface roughness on scattering and absorption process could not be considered as an underestimating effect. We find, however, that functionality of the dielectric function seems to be quite independent of the particular roughness features.

cond-mat.mes-hall

cond-mat

Noname manuscript No. (will be inserted by the editor) The influence of the surface roughness on dielectric function of two-dimensional electron gas A Lindhard model approach A. Phirouznia · L. Javadian · J. Poursamad Bonab · K. Jamshidi-Ghaleh Received: date / Accepted: date Abstract Low field response function calculations have been performed on a two- dimensional electron gas with well-defined electron-surface roughness scattering. The Lindhard model was employed to compute the response function. In partic- ular, detailed investigations were made on the system searching for an interplay between surface roughness with well-defined correlation function, (characterizes by asperity height and correlation length) spatial confinement and the dielectric function. We analyze to what extent the normal behavior and functionality of dielectric function of two-dimensional devices are modified by random scattering events caused by the contribution from the surface roughness. Results of the cur- rent work indicate that contribution of the surface roughness on scattering and absorption process could not be considered as an underestimating effect. We find, however, that functionality of the dielectric function seems to be quite independent of the particular roughness features. Keywords dielectric function · Surface roughness · Lindhard model 1 Introduction Physics of low-dimensional systems has become an intense research field in the last decades. Advances in material fabrication, submicrometer technology and ul- trathin film manufacturing have opened a new field in understanding the physical processes. Considering the remarkable progress in empirical manufacturing of low- dimensional systems and nano-structures and very high applications of this type A. Phirouznia Department of Physics, Azarbaijan Shahid Madani University, 53714-161, Tabriz, Iran E-mail: Phirouznia@azaruniv.ac.ir L. Javadian Department of Laser and Optical Engineering University of Bonab, 5551761167 Bonab, Iran. J. Poursamad Bonab Department of Laser and Optical Engineering University of Bonab, 5551761167 Bonab, Iran. K. Jamshidi-Ghaleh Department of Physics, Azarbaijan Shahid Madani University, 53714-161, Tabriz, Iran 3 1 0 2 v o N 9 2 ] l l a h - s e m . t a m - d n o c [ 1 v 1 9 4 7 . 1 1 3 1 : v i X r a 2 A. Phirouznia et al. of systems in electronic and optical devices, a vast area has been developed for physics of low-dimensional systems and nano-structure. Meanwhile some impor- tant physical effects, such as quantum Hall effect, shows that this research field could be very rich for fundamental studies [1, 2, 3, 4, 5, 6, 7, 8]. The low-dimensional objects can be utilized as components of the electronic de- vices. These low-dimensional structures can provide new functionalities for new generation of electronic devices. In addition to the applicable aspects, these sys- tems could be considered an accurate test for quantum mechanical properties. Such structures were first grown using molecular beam epitaxy (MBE) technique [9], meanwhile various techniques have been recently developed for low-dimensional system fabrication. Today, non-homogeneous low-dimensional structures might be grown from any substance at interfaces. These structures form the basis of quantum well devices. In these structures, electron wave vector is quantized along the confining electric field. Certain experiments have shown that the quantization of the vertical com- ponent of wave vector really happens [10]. In an experiment, it was discovered by measuring optical absorption in a multi-quantum well (MQW) that absorption rate increased in specific wavelengths. These results demonstrated that vertical wave vector is quantized and empirical characteristics of absorption can be ex- plained by step-like density of states of two-dimensional electron gas. Due to the unavoidable surface roughness of any low dimensional system, it is es- sential to understand the influence of boundary scattering in physical phenomena. Formation of transverse modes, quantization of electron momentum in the sys- tems with reduced dimensions include some aspects of physics in which surface roughness can effectively modify the optical response and electronic transport of the system. Some of the effects caused by roughness of quantum wells on electronic features have been formerly calculated [11]. In the current work, this was done about electron-photon interaction and optical features by considering successful models for the roughness. Calculations have been performed for confined electrons in a two-dimensional rough plane by introducing an appropriate correlation function for the roughness. 2 Theory and approach We have assumed an electronic two-dimensional system in which the carriers have been confined in x-y plane of an area Lx × Ly = S. This type of structures can be realized by a semiconductor quantum well. The system has been assumed to be subjected to an external field characterizes by a vector potential, A(r). This system can be described by the following Hamiltonian H = 1 2m (P − e c A)2 + Vconf (z) + ∆Vr(r), (1) where P is the momentum operator, Vconf (z) is the confining electric potential along the z axis, ∆Vr is the potential introduced by roughness and m is the electron mass. Then the Hamiltonian of the system in the Coulomb gauge, ∂iAi = 0, reads H = P2 2m + Vconf (z) − e 2mc A.P + e2 2mc2 A2 + ∆Vr(r), (2) The influence of the surface roughness on dielectric function The vector potential is given as follows A(r, t) = A0eei(q.r−ωt) + c.c, ei(q.r−ωt)e.P + Hep = eA0 mc −i(q.r−ωt)e.P, e eA0 mc 3 (3) in which e is the polarization vector and A0 denotes the amplitude of the vector potential associated with an external electromagnetic field. For this two-dimensional electron gas (2DEG) system in x-y plane, we can assume an average thickness Lz with a random rough boundary at z = Lz [12] z = 0, z = Lz + ∆(r), (4) where the roughness of the system characterizes by ∆(r) which denotes deviation from the perfect two-dimensional plane at r = xi + yj. If we choose the simple quantum box transverse modes in z direction given by Enz = (¯hπnz)2/(2mL2 z), where the eigenvalues of the H0 = P2/(2m) + Vconf (z) can be written as Ekn = Enz + ¯h2k2/(2m). Keeping only terms up to linear in A we arrive at the following Hamiltonian H = H0 − e/(2mc)A.P + ∆Vr(r) = H0 + Hep + ∆Vr(r), (5) (6) in which Hepdenotes the electron photon interaction. Then at the limit of ∆(r)/Lz (cid:28) 1 and for a given transverse band, nz, the rough- ness potential may then be written as ∆Vr(r) = Enz (Lz + ∆(r)) − Enz (Lz) (cid:39) Enz (Lz) 2∆(r) Lz (7) Regarding the fact that ∆(r) has been assigned randomly, this deviation should satisfy the requirement < ∆(r) >= 0 which results in < ∆Vr(r) >= 0 in which < ... > denotes the spatial average of a typical quantity. Gaussian type correlation functions are generally employed for roughness fluctu- ations. Meanwhile, the exponential correlation functions lead to a better fit with experimental results [12, 13]. Due to this fact an exponential correlation function, has been employed as follows (cid:10)∆Vr(r)∆Vr(r (cid:48) )(cid:11) = Enz (cid:10)∆(r)∆(r (cid:48) )(cid:11) 2 2 4 Lz 2 4 Lz = Enz −r−r(cid:48)/Λ. 2 ∆2 0e (8) In the presence of the above mentioned relaxation mechanisms i.e. the external field and the surface roughness total scattering rate in the system is given by W (q, ω) = (cid:104)ψi (−e/(2mc)A.P + ∆Vr(r))ψj(cid:105)2 4π ¯h ×δ(Ej − Ei − ¯hω) [f (Ei) − f (Ej)] , i,j (9) (cid:88) 4 A. Phirouznia et al. where ψi(cid:105) is the eigen-state of the unperturbed Hamiltonian, H0, and f (E) is the Fermi-Dirac distribution function. The matrix elements of the relaxation couplings can be easily find to be (cid:12)(cid:12)(cid:12)2 z ep,knz + = + 2Re((H k(cid:48)n(cid:48) (cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)∆V k(cid:48)n(cid:48) (cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)H k(cid:48)n(cid:48) (cid:11)(cid:12)(cid:12) ≈ (cid:104)knz ∆Vr(r)(cid:12)(cid:12)k (cid:11) = 0. ∗ × ∆V k(cid:48)n(cid:48) z ep,knz z r,knz )) (cid:48) (cid:48) z z r,knz (10) ). The third term of the above expression could be neglected since In which ∆Vr(r) =< ∆Vr(r) >. Meanwhile the second term, (cid:104)knz ∆Vr(r)(cid:12)(cid:12)k(cid:48)n(cid:48) (cid:12)(cid:12)(cid:12)∆V k(cid:48)n(cid:48) can be approximated as z r,knz n n (cid:48) z (11) (cid:11)2 = z (cid:48) z r,knz z ep,knz + ∆V k(cid:48)n(cid:48) (cid:12)(cid:12)(cid:12)H k(cid:48)n(cid:48) (cid:12)(cid:12)(cid:104)knz ∆Vr(r)(cid:12)(cid:12)k (cid:90) (cid:90) (cid:12)(cid:12)(cid:12)2 (cid:90) (cid:90) (cid:90) (cid:90) S (cid:39) δnz ,n(cid:48) S δnz ,n(cid:48) = z z δnz ,n(cid:48) z = S (cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)∆V k(cid:48)n(cid:48) z r,knz −i(k−k(cid:48)).(r−r(cid:48))∆Vr(r)∆Vr(r e (cid:48) )d2rd2r (cid:48) −i(k−k(cid:48)).(r−r(cid:48))(cid:104)∆Vr(r)∆Vr(r e eir−r(cid:48)k−k(cid:48) cos θe −r−r(cid:48)/Λd2rd2r (cid:48) (cid:48) )(cid:105)d2rd2r (cid:48) = 4 Lz 2 Enz 2∆2 0 2πΛ2 (1 + q2Λ2) 3 2 δnz ,n(cid:48) z . Accordingly the transition rate can be decomposed as W1(q, ω) = in which and W(q, ω) = W1(q, ω) + W2(q, ω), (cid:18) eA0 (cid:19)2 (cid:88) (cid:12)(cid:12)(cid:104)knz eiq.re.P(cid:12)(cid:12)k (cid:48)(cid:11)(cid:12)(cid:12)2 (cid:48) n 2 2π ¯h ×δ(Ek(cid:48) − Ek − ¯hω) [f (Ek) − f (Ek(cid:48) )] knz ,k(cid:48)n(cid:48) mc z W2(q, ω) = (cid:88) (cid:104)knz ∆Vr(r)(cid:12)(cid:12)k 2 2π ¯h ×δ(Ek(cid:48) − Ek − ¯hω) [f (Ek) − f (Ek(cid:48) )] . knz ,k(cid:48)n(cid:48) n z z (cid:48) (cid:48) (cid:11)2 (12) (13) (14) (15) Therefore the contribution of the surface roughness in the dielectric function is determined by W2(q, ω). If we assume e = x then (cid:104)knz eiq.rPx (cid:48) (cid:48) n z (cid:48) xδnz n(cid:48) z δ(k (cid:48) + q − k). (16) (cid:12)(cid:12)k (cid:11) = ¯hk The real part of the conductivity is in the framework of the Lindhard approach is then given by [14] σre(q, ω) = c2 2V ¯hωW (q, ω) ω2A2 0 . (17) The influence of the surface roughness on dielectric function 5 Fig. 1 Imaginary part of the dielectric function as a function of the photon energy at different correlation lengths. The imaginary dielectric function is given by σ1(q, ω)
im(q, ω) = = (cid:20) 4π ω Γ S (¯hω)2 W1 + 2π( (cid:21) . E1∆0Λ LzS ) 2 W2 γ (18) (19) f π2) and γ = e2A0 In which Γ = e2/(E2 2/(2mc2). These relations (Equations (17)- (19)) indicate that all of the mechanisms which contribute to the conductivity of the system can contribute in the amount of the dielectric function as well. Similarly the real part of the dielectric function is given by the Kramers-Kronig relation (cid:90) +∞ −∞
im(q, ω(cid:48)) ω(cid:48) − ω (cid:48) . dω (20)
re(q, ω) = 1 + 1 π P 3 Result and Discussion As mentioned in the previous section we have employed Lindhard model to for- mulate the influence of the surface roughness on the dielectric function of a two dimensional electron gas. Results of the current work have summarized in the fol- lowing figures. As depicted in Figure 1 by increasing the correlation length of the surface rough- ness the imaginary part of the dielectric function increases. Mean while increment of the correlation function preserves the typical functionality of the imaginary di- electric function. Numerical results show a similar effect for the real part of the dielectric function as shown in Figure 2. This can be inferred by analyzing the physical meaning of the correlation function. The effective potential of a single local roughness, varies in spatial scale characterizes by the correlation length in the real space. High cor- relation length corresponds to relatively smooth systems when ∆/Λ (cid:28) 1. In this 6 A. Phirouznia et al. Fig. 2 Real part of the dielectric function as a function of the photon energy at different correlation lengths. √ 2/q < Λ. √ 2/q and decreases when case the local rough domains have a considerable overlap. Increasing the correlation length, decreases the scattering potential of the rough- ness as given in Equation 12. Therefore it seems that by increasing the correlation length the imaginary part of the dielectric function (which measures the optical absorption) should be decreased (limΛ→∞ W2 = 0), however it should be noted that the W2 is not a monotonic function of Λ. In fact W2 can be increased by increasing the Λ when 0 < Λ < Meanwhile the energy conservation rule, where enforced by the Dirac delta (Equa- tion (15)) function, finally determines the transferred momwntum, q, and therefore the scattering rate and the effective contribution of the surface roughness in the absorption process. Since the transferred momentum during a single scattering process is limited to the range of q < kf therefore overall scattering rate increases by increasing the correlation length of the surface roughness. In this limit of the transferred momentum the scattering rate, could be increased by the increasing the correlation length. Therefore the imaginary part of the dielectric function which measures the optical absorption increases by increasing the correlation length. This fact will be reflected on real part of the dielectric function, as well this fact was traced back to the Kramers, Kronig relations. It should be noted that the functionality of the real part dielectric function is not influenced by increasing the correlation length however the value of the real dielectric function effectively changes up to several order of magnitudes (Figure 2). Accordingly since the functionality of the real part of the dielectric function will be preserved, therefore it seems that plasmon modes of the system could not sig- nificantly be affected by the magnitude of the correlation length. Meanwhile the The influence of the surface roughness on dielectric function 7 Fig. 3 Imaginary part of the dielectric function as a function of the photon energy at different asperity heights. screening length of the local charged impurities and Friedel oscillations could be influenced by the roughness parameters. Scattering rate of the system increases by increasing the asperity height and this increment is independent of the range of momentum transfer. In this case scat- tering rate is a monotonic increasing function of the asperity height. Therefore as reasonably expected, imaginary part of the dielectric function increases by in- creasing the asperity height (Figure 3). 4 Conclusion In present work we have shown that the surface roughness significantly contributes on scattering and absorption process and dielectric function. Dielectric function of the system increases by both asperity height and correlation length of the rough- ness. References 1. R. E. Prange and S. M. Girvin, The Quantum Hall Effect Springer, New York (1987) 2. R. Landauer, IBM J. Res. Dev. 1, 233 (1957) 3. M. Buttiker, Phys. Rev. Lett.57, 1761 (1986) 4. Hong-Kang Zhao and Jian Wang, Eur. Phys. J. B 44, 93100 (2005) 5. Y.Z. He and C.G. Bao, Eur. Phys. J. B 62, 465470 (2008) 6. K. J. Thomas, J. T. Nicholls, M. Pepper, W. R. Tribe, M. Y. Simmons, and D. A. Ritchie, Phys. Rev. B 61, R13365 (2000) 7. A. A. Starikov, I. I. Yakimenko, and K.-F. Berggren, Phys. Rev. B 67, 235319 (2003) 8 A. Phirouznia et al. 8. N. T. Bagraev, I. A. Shelykh, V. K. Ivanov, and L. E. Klyachkin, Phys. Rev. B 70, 155315 (2004) 9. Smith. S, Chiu. L. C, Margalit. S, Yariva. A, Cho. A. Y, Infrared phys, 23, 93 (1983) 10. Grondin. R. O, Porod. W, Ho. J, Ferry .D. K. Iafrate. G. J, Superlattices Microstruct, 1, 183 (1985) 11. B. R, Nag, Semicond. Si. Tecnol. 19, 162-166 (2004) 12. A. E. Meyerovich, I. V. Ponomarev Phys. Rev B, 65, 155413 (2002) 13. S. M. Goodnick, D. K. Ferry, C. W. Wilmsen, Z. Lilliental, D. Fathy, and O. L. Krivanek, Phys. Rev. B 32, 8171 (1985) 14. G. Grosso and G.P. Parravicini, Solid State Physics 255 Academic Press, London (2005) 

1309.2307

2

1309

Wigner crystal of a two-dimensional electron gas with a strong spin-orbit interaction

The Wigner-crystal phase of two-dimensional electrons interacting via the Coulomb repulsion and subject to a strong Rashba spin-orbit coupling is investigated. For low enough electronic densities the spin-orbit band splitting can be larger than the zero-point energy of the lattice vibrations. Then the degeneracy of the lower subband results in a spontaneous symmetry breaking of the vibrational ground state. The $60^{\circ}-$rotational symmetry of the triangular (spin-orbit coupling free) structure is lost, and the unit cell of the new lattice contains two electrons. Breaking the rotational symmetry also leads to a (slight) squeezing of the underlying triangular lattice.

cond-mat.mes-hall

cond-mat

Wigner crystal of a two-dimensional electron gas with a strong spin-orbit interaction P. G. Silvestrov1, 2 and O. Entin-Wohlman3, 4 1Physics Department and Dahlem Center for Complex Quantum Systems, Freie Universitat Berlin, 14195 Berlin, Germany 2Institute for Mathematical Physics, TU Braunschweig, 38106 Braunschweig, Germany 3Physics Department, Ben Gurion University, Beer Sheva 84105, Israel 4Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel (Dated: June 5, 2018) The Wigner-crystal phase of two-dimensional electrons interacting via the Coulomb repulsion and subject to a strong Rashba spin-orbit coupling is investigated. For low enough electronic densities the spin-orbit band splitting can be larger than the zero-point energy of the lattice vibrations. Then the degeneracy of the lower subband results in a spontaneous symmetry breaking of the vibrational ground state. The 60◦ −rotational symmetry of the triangular (spin-orbit coupling free) structure is lost, and the unit cell of the new lattice contains two electrons. Breaking the rotational symmetry also leads to a (slight) squeezing of the underlying triangular lattice. PACS numbers: 73.20.Qt,75.70.Tj 1. Introduction. The Wigner crystal [1], the insu- lating companion of a two-dimensional metal, is pre- dicted to appear in an electron gas of ultra-low den- sities formed in semiconductor heterostructures when the Coulomb repulsion-induced ordering wins over the zero-point quantum fluctuations [2, 3]. Low densities amount to very clean samples. That is why experimen- tally Wigner crystals were observed either in naturally clean systems, like electrons on the surface of liquid He- lium [4], or in two-dimensional semiconductors when the kinetic energy is suppressed by a strong magnetic field [5], or due to a large mass of the charge carriers [6]. However, lowering the electronic density not only enhances the electronic correlations, but also tends to increase the rel- ative importance of the spin-orbit interaction, generically present in low-dimensional systems [7]. Thus attempting to increase the role of electronic repulsion may lead one into a regime in which quantum fluctuations around the classical equilibrium sites of the Wigner crystal are dom- inated by the spin-orbit interaction. Such crystals, as we show, demonstrate a number of unexpected properties, having no analogue hitherto. The structure of a crystal is usually determined by the interaction between particles which oscillate slightly around their static equilibrium positions. For electrons subject to the Rashba spin-orbit interaction [8], which we consider in this paper, this picture is modified, since in this case even the notion of the "static particle posi- tion" requires a clarification. The spin-orbit interaction splits the electron spectrum, leading to a Mexican hat shaped lower subband with a circle of degenerate minima. However, in an interacting quantum-mechanical system, different minima of the single-particle spectrum are not equivalent. Picking up one of these minima breaks the ro- tational symmetry in momentum space, and then, via the uncertainty relation, the spatial motion and the Coulomb ∆ px ring of minima p=m λ p y p x y p ∆ b. y x d. a. c. FIG. 1: Schematic visualization of the electronic density in a two-dimensional electron crystal. a - the ring of minima in momentum space for the lower Rashba subband Ep− = (p − mλ)2/(2m). The colored ellipse shows the electronic density (for structure b). The density is centred around a particular minimum, but is strongly elongated along the line of minima, ∆px ≪ ∆py ≪ mλ, which helps to lower the interaction energy. b, c, d - three possible periodic configura- tions (see text). The dark ellipses indicate the directions of electrons' oscillations. Structure c has the lowest zero-point energy of the lattice vibrations. The arrows in b indicate the in-plane spin orientations. interaction of the electrons are affected. It has been even recently shown [9] that the Rashba spin-orbit inter- action stabilizes a strongly asymmetric Wigner crystal when the interaction potential is short-range, V ∼ 1/rα with α > 2. Such a crystal would have been otherwise unstable for any small electronic density. In this paper we analyze the crystal created by electrons interacting via the unscreened Coulomb repulsion, V = e2/r. Then the triangular lattice of the crystal [10] remains stable on the classical level. However, breaking the symmetry in momentum plane changes drastically the fluctuation properties and the electronic density distribution. We consider strong spin-orbit interactions, such that the effective Hilbert space is reduced to include only elec- tronic states of momenta close to the ring of minima in the lower subband (see Fig. 1a). Each electron in the crystal picks up only one of those minima. However, since the electron's displacement in momentum space along the line of minima costs no kinetic energy, one may effectively freeze its spatial vibrations in this direction. Freezing the spatial vibrational mode reduces the average poten- tial energy, since now the electron never leaves its classi- cal equilibrium location in this direction. The vibrations along the direction perpendicular to the ring of minima have the same effective mass as in the absence of the spin-orbit interaction. Hence the electrons' fluctuations when the spin-orbit coupling is strong enough are equiva- lent to vibrations of particles having anisotropic masses. Here though, the light and the (infinitely) heavy masses directions are chosen for each electron individually. As different minima of the Rashba Hamiltonian are classically equivalent, the proper choice of the electrons' configuration should be the one minimizing the zero- point fluctuation energy of the crystal. Finding the min- imum of the zero-point energy for general directions of the oscillations is a difficult task which will not be fully accomplished in this paper. Instead, we adopt a step-by- step approach, considering a series of configurations de- pending on 1, 2, 3, . . ., angles with respect to which one would minimize the energy functional. The first three steps in this scheme are illustrated in Figs. 1b, c, and d. The dark ellipses there indicate the directions along which the electrons vibrate. First, one requires all elec- trons to vibrate in unison, and minimizes the energy with respect to a single angle. Figure 1b shows the best con- figuration of such a one-parameter family. At the next step one allows two neighboring electrons on the trian- gular lattice to vibrate along independent directions and then repeats this configuration periodically. The unit cell now contains two electrons. Figure 1c shows the config- uration realizing the minimum of the vibrational energy for such a two-parameter family of crystals. One may consider similarly a lattice with more independent vibra- tional directions, see Fig. 1d. Among all crystal config- urations which we have analyzed, the one in Fig. 1c has the smallest vibrational energy. Different crystal con- figurations have also a very different vibrational spectra [Fig. 2 below] which can be exploited to distinguish them experimentally. tonian of the system under consideration is e2/Rij + rij , H =Xi H0i +Xi<j 2 (1) with Rij ≡ Ri − Rj. At equilibrium, the electrons are located at sites Ri on a triangular lattice [10] of spac- ing a. The oscillations around those sites are described by expanding the interaction in the small displacements ri ≪ a up to second order [11]. This expansion yields a single-electron harmonic potential and a (bi)linear in the displacement electron-electron interaction [see Eq. (7) below]. The former allows to introduce the frequency [12] ω0 =pγe2/(ma3) , (2) with γ =Pi6=0 a3/(2R3 i ) = 5.5171 and m being the effec- tive mass. The single-electron part of the Hamiltonian (1) reads H0 = p2/(2m) + λ(σxpy − σypx) + mλ2/2 , (3) where λ denotes the Rashba spin-orbit coupling strength and σx,y are the Pauli matrices. The spectrum of H0 consists of two subbands, Ep± = (p ± mλ)2/(2m), which correspond to electrons with in-plane spins directed nor- mal to the momentum, and pointing to its left or right, respectively. We focus on the regime where the spin-orbit energy exceeds the one due to the zero-point motion of the electrons around their equilibrium sites, mλ2 ≫ ω0 ∼pe22/(ma3) , (4) s = 1/(na2 which means that the electrons are always confined to the lowest subband. The relative strength of the Coulomb in- teraction compared to the kinetic energy is characterized by the dimensionless parameter rs, related to the elec- tronic density n and the Bohr radius aB = 2/(me2) as πr2 B) [13]. Obviously, in the case of a strong Rashba spin-orbit interaction the large value of the same parameter ensures the existence of the Wigner crystal: rs ≫ 1 here means that the gain in the Coulomb energy per electron in the ordered phase exceeds the energy rise due to the zero-point fluctuations of electrons in the lat- tice, e2/a ≫ ω0. (Quantum Monte-Carlo simulations indicate that in the absence of spin-orbit coupling the Wigner crystal exists at rs > 35 [2, 3].) Once the electrons reside in the lower Rashba subband, their (ground-state) momenta lie within a narrow ring in momentum plane of radius mλ and width ∆p, deter- mined by the zero-point energy (p − mλ)2 . ∆p2 ∼pe22m/a3 ≪ (mλ)2 . However, in the crystal different parts of this ring are not equivalent and each electron may choose its own small sector. Imagine a much-elongated wave packet built from the lower Rashba subband solutions, such that (5) 2. Wigner crystal with spin-orbit coupling. The Hamil- px − mλ . ∆px, py . ∆py, ∆px ≪ ∆py ≪ mλ, (6) as in Fig. 1a. The two spatial dimensions of this wave packet are very different, ∆x ≫ ∆y, and hence the minute displacement along the y−direction gives a neg- ligible correction to the interaction energy. The expec- tation value of H0, which has an eigenvalue E− = 0 at p = mλ, is determined by the smaller momentum ex- tension of the packet, hH0i ∼ ∆p2 x/(2m). This implies that the system may choose an anisotropic pattern where around each lattice site the density forms a narrow el- lipse of length ∆x ∼ [a32/(me2)]1/4. The orientations of these ellipses will be determined by the zero-point en- ergy of the vibrations pertaining to a specific pattern. 3. Vibration spectrum. We begin with the config- uration shown in Fig. 1b. In this structure the elec- trons oscillate along x, and (since the Hilbert space is reduced to that of the lower Rashba subband) are strongly spin-polarized along y, hσyi ≃ 1. To find the excitation spectrum around this particular structure, we shift the x−component of the momentum, px → mλ + px, multiplying the many-electron wave function by exp[(i/)Pi mλxi]. The reduced Hamiltonian con- tains only the term quadratic in px, while py appears at higher orders in py/(mλ) and may be discarded. The uncertainty principle then imposes no restrictions on the displacements along the y−direction, allowing to choose yi ≡ 0. This results in a single-coordinate effective Hamiltonian, Heff =Xi i /2] + e2Xi<j where uij = 1/R3 ij, Xij is the x−component of Rij and ω0 is defined in Eq. (2). Hamiltonian Eq. (7) contains half of the degrees of freedom of the original one, since it allows for a single vibrational direction per elec- tron. In the regime given by Eq. (4), the missing degrees of freedom pertain to low energy non-phonon excitations, whose analysis is beyond the scope of this paper. xi/(2m) + mω2 ij − 3X 2 xixjuij , (7) ij/R5 0x2 [p2 Exploiting the Bogoliubov transformation ck ≡Xi e−ik·Ri mω0xi + ipxi p2N mω0  = coshukdk + sinhukd† −k, (8) where N is the number of lattice sites, transforms the Hamiltonian into 1 2 ) , ωk = ω0p1 + 2v(k) . (9) kdk + ωk(d† Heff =Xk Pi6=0 eik·R0i[a3/(2γR3 Here, tanh 2uk = −v(k)/[1 + v(k)] and v(k) = In particular, 0i/R2 0i)](cid:0)1 − 3X 2 0i(cid:1). at small wavevectors the dispersion law becomes ωk = [4πe2/(√3ma2)]1/2kx/√k . (10) The √k−dependence of the low frequencies is expected for a two-dimensional plasmon gas. The striking feature 3 ) ω ( ρ 3.5 3 2.5 2 1.5 1 0.5 0 0 0.2 0.4 0.6 0.8 ω/ω 0 1 1.2 1.4 1.6 FIG. 2: The vibration density of states as a function of the frequency (scaled by ω0) for the Wigner-crystal configuration of Fig. 1b (dashed line) and of Fig. 1c (solid line). Though the spectrum extends all the way down to ω = 0 for 1b the average frequency turns out to be smaller for 1c. Note the linear vanishing of the density (characteristic of a Dirac cone) in the middle of spectrum for 1c and especially the inverse square- root divergence of the density at the lower edge, ω ≈ 0.36ω0. though is the angular dependence of the dispersion law (10), with vanishing frequency along the y−direction. It signals a spontaneous symmetry-breaking caused by the degeneracy at the bottom of the lower Rashba subband. The vibration spectra for the structures with several electrons per unit cell, like in Figs. 1c and d are found similarly. Details of this calculations are given in Ref. 14. 4. Minimizing the zero-point energy. We have in- vestigated numerically all possible configurations of the Wigner crystal with vibrational directions depending on 1, 2, and 3 angles, as described in the introduction. Im- portant examples representing these families of configura- tions are shown in Figs. 1b, 1c, and 1d. The ground-state energy is determined by the frequency ωk averaged over the Brillouin zone, which in the case of Fig. 1 gives hωb ki ω0 = 0.951 , hωc ki ω0 = 0.939 , hωd ki ω0 = 0.971 . (11) Quite unexpectedly, among the examined configurations the one with two electrons per unit cell shown in Fig. 1c has the smallest zero-point energy. Configuration 1b has the smallest hωki for the families with 1 and 3 electrons per unit cell, but is not the global minimum of the vi- brational energy. The highly-symmetric configuration 1d has the largest zero-point energy in the family with 3 independent vibrational directions. All phonon frequencies coincide independent of the di- rection of vibration, ωk ≡ ω0, upon exploiting the Ein- stein approximation, in which each electron is confined to a harmonic potential created when the locations of all other electrons are frozen [11]. In all considered crystal configurations the average frequency was always close to the Einstein approximation, like in Eq. (11). One way to probe experimentally the structure of a crystal is to measure the vibration spectrum. Figure 2 shows the density of states (DOS) as a function of the frequency for the structures in Figs. 1b and 1c. Several interesting features can be observed there (but unfortu- nately, none explains which structure is energetically fa- vorable, and why the resulting hωki's are so close to each other). For the configuration 1b the DOS is finite all the way down to zero frequency due to the plasmon mode Eq. (10). The spectrum has a step singularity at the high-frequency end, and two logarithmic van Hove singu- larities. Though the modes describing global translations for the full Hamiltonian Eq. (1) always have zero energy, it is only for the structure 1b that one of these modes is reproduced by the effective Hamiltonian Eq. (7). The vibration spectrum of the crystal phase shown in Fig. 1c is gapped (the solid line in Fig. 2). The DOS has several step singularities and logarithmic di- vergences. The linear vanishing of the DOS at ω ≈ ω0 corresponds to a conical crossing of the two bands (a Dirac point). Surprisingly enough, we observe an almost perfectly inverted square-root singularity at low frequen- cies, ρ(ω) ∼ [ω − 0.36ω0]−1/2; such a divergence usually characterizes one-dimensional systems. Careful numer- ical investigation shows that the dispersion law around these frequencies has the form [ωc k/ω0]kx≈0 ≈ 0.3584 + 0.564 × 10−3 cos(√3aky) + f (ky)(akx)2 , (12) with f (ky) ∼ 1. These modes correspond to a horizontal displacement of every second row in the configuration of 1c. The small coefficient of the second term in Eq. (12) implies that different electron rows "see" each other as a continuous charge lines. This is surprising, since the distance between the rows is only √3-times larger than the interval between electrons within each row. 5. Squeezing the crystal. The crystal structures of Figs. 1b and 1c have a preferential axis, violating the 60◦ symmetry of the triangular lattice. This opens the possibility for a (slight) squeezing of the lattice, caused by an interplay of the classical Coulomb repulsion and the quantum vibrations (cf. the strong squeezing in the case of a short-range interaction [9]). The 'density preserving' squeezing is defined as Ri → Ri = [(1 + α)Xi, Yi/(1 + α)], where α is a small but finite parameter. We may also write the energy per electron in a crystal as a power series in the Plank's constant, E(α) = ǫ0(α) +  ǫ1(α) + ··· . The first term is the elec- tron's electrostatic energy, ǫ0(α) = (e2/a)(c0 + αc1 + α2c2 + . . .) , (13) and the second comes from the average zero-point energy, ǫ1(α) = ω0(d0 − αd1 + . . .) , (14) 4 where ci and di are numerical coefficients. The coeffi- cients c0 and d0 [the latter is found in Eq. (11)] are of no interest. Since the triangular lattice is the minimum of the electronic Coulomb energy [10], one has c1 ≡ 0 and c2 > 0. The linear in α term in the quantum correction to the energy, d1, would also vanish due to the crys- tal symmetry in the absence of the spin-orbit coupling, or e.g. for the 120◦−rotational symmetric configuration Fig. 1d, but is allowed for the configurations Figs. 1b and c. Numerically we found c2 ≈ 0.527, db 1 ≈ 0.245 and dc 1 ≈ 0.425. Minimizing the energy, E(α), results in αc = ω0 2e2/a d1 c2 =r γaB 4a d1 c2 ∼r aB a . (15) The value of the squeezing parameter, αc, grows with the density as a−1/2 ∼ n1/4. This increase is limited, however, by the inequality (4), 6. Summary. As we have shown, the possibility of the electrons to occupy different minima in momentum space leads to a complicated ground-state of the Wigner crystal. We found the ground state by considering an ef- fective Hamiltonian, which accounts for a single energetic vibrational mode per electron, leaving untouched a num- ber of non-phonon excitations from the low-energy sector of the full problem. These soft modes would correspond to a small displacement of the electron wave packet in momentum space along the circle of degenerate minima (see Fig. 1a), or to spin flips associated with the 180◦ jumps of the electron to the opposite side of the circle of minima (see Refs. [12] and [15] for a discussion of the spin structure of a Wigner crystal.) Although we expect the true ground state of the crystal to be the structure in Fig. 1c, other configurations may exist as metastable states. If the configuration 1b would be realized experimentally, one will be able to probe the angular-dependent plasmon modes, Eq. (10). s Finally, the existence of the spin-orbit dominated phase of the Wigner crystal described in this paper re- quires the validity of the inequality (4). This may be rewritten as m∗λ2 ≫ Ry/r3/2 , where Ry = m∗e4/22ǫ2 and m∗ and ǫ are the effective mass and dielectric con- stant, respectively. For InAs [16], m∗λ2 ≈ 0.2meV and Ry ≈ 2.5eV. Assuming that in the presence of spin-orbit interactions the crystal stability still requires the large value of rs ∼ 35, we expect our results to be always ap- plicable for Wigner crystals in such materials. We thank Y. Imry, P. W. Brouwer, E. Bergholtz and M. Schneider for helpful discussions. The hospitality of the Institute for Advanced Studies at the Hebrew Univer- sity and the Albert Einstein Minerva Center for Theoret- ical Physics at the Weizmann Institute are gratefully ac- knowledged. This work was supported by the Alexander von Humboldt Foundation, the DFG grant RE 2978/1-1, the Israeli Science Foundation (ISF), and the US-Israel Binational Science Foundation (BSF). [1] E. Wigner, Phys. Rev. 46, 1002 (1934). [2] B. Tanatar and D. M. Ceperley, Phys. Rev. B 39, 5005 (1989). [3] N. D. Drummond and R. J. Needs, Phys. Rev. Lett. 102, 126402 (2009). [4] C. C. Grimes and G. Adams, Phys. Rev. Lett. 42, 795 (1979). [5] E. Y. Andrei, G. Deville, D. C. Glattli, and F. I. B. Williams, Phys. Rev. Lett. 60, 2765 (1988); V. J. Gold- man, M. Santos, M. Shayegan, and J. E. Cunningham, ibid. 65, 2189 (1990); F. I. B. Williams, P. A. Wright, R. G. Clark, E. Y. Andrei, G. Deville, D. C. Glattli, O. Probst, B. Etienne, C. Dorin, C. T. Foxon, and J. J. Har- ris, ibid. 66, 3285 (1991); M. A. Paalanen, R. L. Willett, R. R. Ruel, P. B. Littlewood, K. W. West, and L. N. Pfeiffer, Phys. Rev. B 45, 13784 (1992). [6] J. Yoon, C. C. Li, D. Shahar, D. C. Tsui, and M. Shayegan, Phys. Rev. Lett. 82, 1744 (1999). 5 [7] R. Winkler, Spin-Orbit Coupling Effects in Two- Dimensional Electron and Hole systems, Springer-Verlag Berlin Heidelberg, 2003. [8] E. I. Rashba, Fiz. Tverd. Tela (Leningrad) 2, 1224 (1960) [Sov. Phys. Solid State 2, 1109 (1960)]. [9] E. Berg, M. S. Rudner, and S. A. Kivelson, Phys. Rev. B 85, 035116 (2012). [10] L. Bonsall and A. A. Maradudin, Phys. Rev. B 15, 1959 (1977). [11] Technically, this approximation amounts to discarding the cross terms (ri · rj ) and (ri · Rij )(rj · Rij ) in the expansion of Eq. (1). [12] V. V. Flambaum, I. V. Ponomarev, and O. P. Sushkov, Phys. Rev. B 59, 4163 (1999). [13] or alternatively a/aB = rs(q2π/√3) [14] See supplementary material. [15] S. Gangadharaiah, J. Sun, and O. A. Starykh, Phys. Rev. Lett. 100, 156402 (2008). [16] D. Grundler, Phys. Rev. Lett. 84, 6074 (2000). APPENDIX: VIBRATION SPECTRUM FOR SEVERAL ELECTRONS PER UNIT CELL We first show how to obtain the phonon spectrum belonging to configuration 1c (Fig. 1 in the paper). This is an example of a lattice with two electrons per unit cell (which turns out to be our best candidate for the ground state). We then sketch briefly the derivation of the spectrum in the case of three electrons per unit cell, vibrating along three arbitrary directions. When there are two electrons within each unit cell, it is convenient to introduce two sublattices, a and b. The effective vibration Hamiltonian for the configuration of Fig. 1c is then The configuration of Fig. 1c xi 2m Heff =Xi∈a(cid:18) p2 +Xi∈b p2 yi 2m + + mω2 mω2 0x2 i 2 (cid:19) + e2 Xi<j∈a 2 ! + e2 Xi<j∈b 0y2 i xixj R3 ij 1 − 3 ij 1 − 3 yiyj R3 X 2 ij R2 ij! ij! − 3e2 Xi∈a,j∈b Y 2 ij R2 xiyj XijYij R5 ij . Introduction of the creation/annihilation operators a =r mω0 2 (x + i mω0 px), b =r mω0 2 (y + i mω0 py) , leads to the Hamiltonian Heff = ω0 Xi∈a (a† i ai + 1 2 ) + Xi<j∈a ij (ai + a† vx i )(aj + a† j) (b† i bi + +Xi∈b 1 2 ) + Xi<j∈b vy ij (bi + b† i )(bj + b† j) + Xi∈a,j∈b wij (ai + a† i )(bj + b† j)  , where vx ij = a3 2γR3 ij (1 − 3X 2 ij/R2 ij) , vy ij = a3 2γR3 ij (1 − 3Y 2 ij /R2 ij) , wij = −3 a3XijYij 2γR5 ij . (16) (17) (18) (19) 6 y x a 2 A 1 A 2 a 1 FIG. 3: Definition of the basis vectors for the original triangular lattice, a1 and a2, and for the superlattice with three electrons per unit cell, A1 and A2. Electrons from the three different sublattices are shown by different colors. In Fourier space, this Hamiltonian takes the form Heff = ω0Xk [(a† kak + 1/2) + vx(k) 2 (ak + a† −k)(a−k + a† k) vy(k) 2 (bk + b† −k)(b−k + b† k) + w(k)(ak + a† −k)(b−k + b† k)] , + (b† kbk + 1/2) + with vx(k) = vx(−k) = X0∈a,i6=0∈a vx 0ieik·Ri , vy(k) = vy(−k) = X0∈a,i6=0∈a vy 0ieik·Ri , w(k) = w(−k) = X0∈a,i∈b w0ieik·Ri . An important point to note here is that vy(k) 6= vx(k). In the next step one decouples the two vibration polarizations, (22) ck = cos(τk)ak + sin(τk)bk , dk = − sin(τk)ak + cos(τk)bk , with tan(2τk) = 2w(k)/[vx(k) − vy(k)], to obtain Heff = ω0Xk [(c† kck + 1/2) + v1(k) 2 (ck + c† −k)(c−k + c† k) + (d† kdk + 1/2) + v2(k) 2 (dk + d† −k)(d−k + d† k)] , where v1,2(k) = 1 2 Each polarization can be now Bogoliubov transformed exactly as is carried out in the main text. [vx(k) + vy(k) ±p(vx(k) − vy(k))2 + 4w2(k)] . (20) (21) (23) (24) Phonon spectrum for superlattices with three electrons per unit cell 7 The triangular lattice of a usual Wigner crystal is defined by two lattice vectors, a1 = (a, 0) and a2 = (a/2,√3a/2). The superlattice with three atoms per unit cell is defined by the lattice vectors A1 = (3a/2,√3a/2) and A2 = (3a/2,−√3a/2), such that A1 = A2 = √3a, as is shown in the figure. The electrons on this lattice naturally form three sublattices whose sites coordinates are iA1 + jA2, (a, 0) + iA1 + jA2, and (−a, 0) + iA1 + jA2. Here i and j are two arbitrary integer numbers. Let the displacement vectors for the three sublattices be ui, vi, and wi, where u = unu, v = vnv, and w = wnw, with the three unit vectors nu, nv, and nw pointing along the directions of the allowed vibrations for each sublattice. For example, for the electronic configuration shown in Fig. 1d of the main text one has nu = (0, 1), nv = (√3/2,−1/2), and nw = (−√3/2,−1/2). However, our derivation below does not assume any specific orientation of nu, nv, and nw. The Hamiltonian takes the form where, for example, H = hu + hv + hw + Vu + Vv + Vw + Wuv + Wuw + Wvw , and Let hu =Xi∈u(cid:18) p2 uiuj 1 uivj 1 Vu = e2 Xi<j∈u Wuv = e2 Xi∈u,j∈v ui 2m + mω2u2 2 (cid:19) , ij − 3 R3 (nu · Rij)2 R5 ij ! , ij − 3 R3 (nu · Rij )(nv · Rij) R5 ij ! . ui =r 3 N Xk eik·Riuk , vi =r 3 N Xk eik·Rivk , wi =r 3 N Xk eik·Riwk , where N is the total number of electrons and Rj is the true coordinate of the corresponding N -electron lattice site. Similarly we introduce the Fourier transformed momenta puk, pvk, pwk , so that [pk, xq] = −iδk+q. It follows that mω2u−kuk + 2m huk =Xk (cid:18) pu−kpuk u−kukVu(k), Vu(k) =Xj 2 (cid:19) , eik·R0j 1 0j − 3 R3 (nu · R0j)2 R5 0j ! , Vu → V (k) u = e2 2 Xk and Wuv → W (k) uv = e2Xk Wuv(k) =Xj u−kvkWuv(k) , eik·R0j (nu · nv) 0j − 3 R3 (nu · R0j)(nv · R0j) R5 0j ! . Next we perform a unitary rotation of the coordinates uk, vk, and wk, in order to diagonalize the matrix (25) (26) (27) (28) (29) (30) (31) (32) e2 2   Vu(k) Wuv(k) Wuw(k) Wvu(k) Vv(k) Wvw(k) Wwu(k) Wwv(k) Vw(k)   , which we do numerically. After that the Bogoliubov transformation for each polarization is carried out similar to the way it is done in the main text. 

1211.0491

2

1211

Demonstrating a Driven Reset Protocol of a Superconducting Qubit

Qubit reset is crucial at the start of and during quantum information algorithms. We present the experimental demonstration of a practical method to force qubits into their ground state, based on driving certain qubit and cavity transitions. Our protocol, called the double drive reset of population is tested on a superconducting transmon qubit in a three-dimensional cavity. Using a new method for measuring population, we show that we can prepare the ground state with a fidelity of at least 99.5 % in less than 3 microseconds; faster times and higher fidelity are predicted upon parameter optimization.

cond-mat.mes-hall

cond-mat

Demonstrating a Driven Reset Protocol for a Superconducting Qubit K. Geerlings,1 Z. Leghtas,2 I.M. Pop,1 S. Shankar,1 L. Frunzio,1 R.J. Schoelkopf,1 M. Mirrahimi,1, 2 and M.H. Devoret1 1Department of Applied Physics, Yale University, New Haven, Connecticut 06520-8284, USA 2 INRIA Paris-Rocquencourt, Domaine de Voluceau, B.P. 105, 78153 Le Chesnay cedex, France (Dated: October 19, 2012) Qubit reset is crucial at the start of and during quantum information algorithms. We present the experimental demonstration of a practical method to force qubits into their ground state, based on driving appropriate qubit and cavity transitions. Our protocol, called the double drive reset of population, is tested on a superconducting transmon qubit in a three-dimensional cavity. Using a new method for measuring population, we show that we can prepare the ground state with a fidelity of at least 99.5 % in less than 3 µs; faster times and higher fidelity are predicted upon parameter optimization. A method for qubit initialization is one of the funda- mental requirements of quantum information processing laid out by DiVincenzo [1]. Due to recent advancements in extending superconducting qubit relaxation times to the 100 µs range [2], active ground state preparation (qubit reset), other than by passively waiting for equi- libration with a cold bath, is becoming a necessity. The main use for a fast, high-fidelity reset is to place the qubit into a known pure state either before or during an algo- rithm. Active reset is preferred over passive reset when (a) the qubit thermal environment is hot on the scale of the transition frequency, and (b) rapid evacuation of en- tropy from the system is necessary, as in implementations of quantum error correction [3, 4]. The ancestor of active qubit reset is dynamical cooling of nuclear spins using paramagnetic impurities [5]. Su- perconducting qubits are analogous to single spins in a controlled environment, and it is therefore possible to de- sign similar dynamical cooling methods to achieve reset times much faster than the relaxation time T1 . While several methods [6–13] for reset and dynamical cooling have been demonstrated in superconducting qubits, they each require either qubit tunability or some form of feed- back and high-fidelity readout. We present a practical dynamical cooling protocol without these requirements. This protocol is related to dissipation engineering [14], as we use the dissipation through the cavity to stabilize the qubit ground state. Double Drive Reset of Population (DDROP) is tested on a transmon qubit [15] in a three- dimensional cavity [2] but can be applied to any circuit QED system. DDROP consists of a pulse sequence that manipulates the transition landscape of the qubit-cavity system in or- der to quickly drive the qubit to the ground state. The protocol relies on the number splitting property of the strong dispersive regime [16] of circuit QED, where the dispersive shift χ of the cavity due to a qubit excita- tion is larger than twice the cavity linewidth κ and qubit linewidth 1/T2 . Thus the cavity frequency depends on the state of excitation of the qubit, and the qubit fre- quency depends on the number of excitations in the cav- ity. Another requirement is needed: κ must be much larger than Γup = Pe /T1 , where Pe is the equilibrium ex- cited state population. This condition is easy to satisfy with the recent advances in extending T1 . Apart from special cases where it is desirable to have small κ, most transmons and other qubits read by a superconducting cavity are candidates for this type of reset. In the DDROP protocol, shown graphically in Fig. 1, two microwave drives are applied simultaneously for a duration of order 10 κ−1 in order to reach a steady state. The first drive frequency, f 0 ge , is chosen in order to Rabi drive the qubit if the cavity has zero photons. The ampli- tude of this drive is quantified by ΩR , the Rabi frequency. The second frequency, f g c , is chosen to populate the cav- ity with photons if and only if the qubit is in the ground state. The role of the cavity drive is to lift the popula- tion of g , 0(cid:105) to the coherent state g , α(cid:105), where α2 = ¯n, the steady state average photon number in the cavity. Due to number splitting, the qubit transition frequencies when the cavity is in state α(cid:105) differ sufficiently from f 0 that the Rabi drive does not excite g , α(cid:105). The only way ge for the system to leave g , α(cid:105) is through a spontaneous excitation happening at a rate Γup , which is slow com- pared to all other rates in the system. Once in e, α(cid:105), the system rapidly falls back to e, 0(cid:105) in a time of order κ−1 . The role of the Rabi drive, with Rabi frequency of order κ, is to speed up the transition between e, 0(cid:105) and g , 0(cid:105), thus allowing a fast return to g , α(cid:105). With both drives on, the system will be driven to g , α(cid:105) at a rate of order κ regardless of initial state, while the rate Γup away from this state is slow. Eventually, to prepare g , 0(cid:105) instead of g , α(cid:105), one must turn off the drives and wait for the photons to decay in a time of several κ−1 . Since the cavity is in a coherent state, this waiting time could be avoided by using a displacement pulse, which is easier to calibrate with cavities with higher quality factor. The ratio κ/Γup determines the fidelity of the ground state preparation, and must therefore be much greater than 1. The measurements presented here were performed in a standard circuit QED setup on an aluminum trans- mon qubit, fabricated using a bridgeless double-angle evaporation technique [17, 18], inside a three-dimensional copper cavity, thermally anchored to the mixing cham- 2 FIG. 1: Level structure of the transmon qubit coupled dis- persively to a single resonator mode. The qubit excitations are spanned vertically while the resonator photon numbers are spanned horizontally. The arrows show the transitions in- volved in the DDROP procedure along with their rates, with Γup (cid:28) κ ≈ ΩR < χ/2. The double arrows are driven transi- tions, while single arrows are spontaneous. Qubit transitions are represented by straight lines while cavity transitions are wavy lines. The steady-state equilibrium qubit-cavity joint state is the coherent state g , α(cid:105). For visualization, the state g , m(cid:105) is highlighted, where m is the closest integer to the steady-state average number of photons in the cavity. ber of a dilution refrigerator with a base temperature of 17 mK. The cavity was mounted inside a copper shield coated with infrared-absorbing material on the inside. A high-frequency filter similar to that of Ref. [19] and a microwave 12 GHz low-pass filter were placed on each input and output microwave line. Two 8-12 GHz circu- lators were installed between the cavity and the HEMT amplifier. System parameters were measured to be: f g c = 9.1 GHz, f 0 ge = 5.0 GHz, κ/2 π = 3 MHz, χ/2 π = 7 MHz, T1 = 37 µs, T Ramsey = 20 µs, T E cho = 40 µs, equilibrium Pe = 9 %, Γup/2π ≈ 400 Hz. Both require- 2 2 ments for the reset mechanism are achieved, with χ/κ = 2.3 and κ/Γup (cid:39) 8,000. The effect of the DDROP protocol on this qubit is shown in Fig. 2, where the y axis is the measured ex- cited state population and the x axis is the duration of the reset pulses (or delay time). Each data point is taken after waiting 1 µs (20 κ−1 ) after the end of the DDROP pulse to allow the system adequate time to decay from g , α(cid:105) to g , 0(cid:105). The two solid, nearly horizontal curves are the pre-reset ground and excited qubit states with- out DDROP pulse. The pre-reset is itself a 5 µs DDROP sequence done before all other pulses in order to suppress the initial excited state population. The slight downward trend in the excited state curve, due to the finite value of T1 , is barely noticeable on this scale. The other two solid curves correspond to the same preparation, but show the effect of a DDROP pulse whose duration is varied across the x axis. At short pulse duration, both initial pop- ulations tend towards 50% excitation, due to the Rabi drive. As the duration is increased, the population tends quickly towards the pre-reset ground state. The four FIG. 2: Measured excited state population after reset pulse of varying duration, for four different initial preparations, mea- sured after intervals of 40 ns. The solid lines include a pre- reset while the dashed lines begin with the steady state 9% ex- cited population. The ‘w/π pulse’ curve shows a slight down- ward trend due to the finite T1 . The curves with DDROP show that, regardless of initial state, the qubit is driven to the ground state for pulse durations less than 2 µs. For this measurement, ΩR ≈ 0.8 κ and ¯n = 8 dashed curves represent an identical set of data taken without the pre-reset, thus showing the effect of initial equilibrium population. Note that regardless of the ini- tial state, DDROP forces the population to the ground state in less than 3 µs (including the 1 µs decay from g , α(cid:105) to g , 0(cid:105)). This is a factor of 60 improvement over the standard protocol of waiting 5 T1 , which would give a comparable reduction of excited state population in a cold qubit environment. In order to benchmark our DDROP reset procedure, we had to carefully measure the resulting ensemble- averaged excited state population. A measurement of the ratio of the heights of the two spectroscopic peaks corre- sponding to the g(cid:105) to e(cid:105) and e(cid:105) to f (cid:105) qubit transitions, usually assesses the excited state population. However, this method does not take into account the variation of readout efficiency with qubit state, and is therefore not quantitative without further corrections. We introduce a method called the Rabi population measurement (RPM) that circumvents these problems. The basic idea of RPM is to measure two Rabi oscilla- tions whose amplitude ratio corresponds directly to the ratio of initial excited state (Pe ) to ground state popula- tion (Pg ). This method is similar to, but different from, techniques previously used in phase qubits [20, 21]. Note that for the cases treated in this paper, populations of states above e(cid:105) are negligible, so Pe + Pg = 1. The RPM is performed by applying two sequences of qubit pulses as shown in Fig. 3. The first sequence consists of a pulse performing a rotation around X on the e(cid:105) to f (cid:105) 3 cavity, while choosing the initial state to be the cavity in vacuum and an equilibrium state for the qubit. The dependence of F on ΩR for fixed ¯n was found to be weak, and fidelities above 99% were found for ΩR /κ > 0.3. Our numerical simulations show that F increases monotoni- cally with ¯n for a fixed ΩR and that with a higher ΩR , higher ¯n is required to reach the same fidelity, as shown by the contours of constant fidelity in Fig. 4(a). The simulations did not account for self-Kerr effects that will reduce the fidelity at photon numbers much higher than the range shown. While 99% fidelity is reached with a wide range of parameters, reset time is optimized when ΩR (cid:39) κ, yielding reset times comparable with those of two-qubit gates in the cQED architecture [22]. The re- set time for the ground state population to reach 99% is shown by the colored pixels of Fig. 4(a). With the guidance provided by these simulations and using RPM to experimentally quantify the fidelity, we have studied DDROP for a wide range of ΩR and ¯n. The pulse duration was kept fixed at the value 5 µs, chosen from simulation, to ensure DDROP has reached equilib- rium in all conditions. Fidelities greater than 99% were achieved for ΩR as low as 0.3 κ and as high as 1.0 κ, for 8 ≤ ¯n ≤ 50. For fixed ΩR = 0.8 κ, Fig. 4(b) shows measurement (markers) vs simulation (line) of remaining excited state population vs ¯n. Excited state population drops monotonically with ¯n, in good agreement with nu- merical simulation. On the other hand, above approxi- mately ¯n = 50 (data not shown), the reset excited state population increased significantly. This is understood to be due to the breakdown of the dispersive approximation. Overall, both drive amplitude parameters ΩR and ¯n have a wide range for which DDROP works well, making it a very reliable and stable protocol. As mentioned before, all of the DDROP characteri- zation measurements included a 1 µs (20 κ−1 ) wait be- tween drive pulses and the RPM measurement, to allow the cavity photons to decay. Therefore, the qubit ex- cited state population begins returning to its equilibrium value as soon as the reset drives are turned off. This re- equilibration should occur on a timescale given by the mixing time T1 , and this is what is found experimentally. DDROP is not the first demonstrated qubit reset mechanism to work on superconducting qubits; several distinct methods have been shown previously, includ- ing: sideband cooling through higher energy levels [6], sweeping the qubit frequency into resonance with a low- Q cavity [8, 9], a feedback loop with conditional coherent driving [10], and strong pro jective measurements [11–13]. However, DDROP has many advantages when compared to each of these processes. First, there is no need to tune in real time the qubit frequency, which means DDROP will still work with fixed-frequency qubits. There is no need for fast external feedback of any kind, thus sim- plifying the required setup. There is also no need for high-fidelity, single-shot readouts or in fact a low-noise FIG. 3: Upper panel: pulse sequences used to perform qubit population measurement (RPM, see text), each producing an oscillation whose amplitude is proportional to initial excited (a) and ground (b) state population. Circle radii indicate pop- ulation in each state, vertical bars separate the two extrema in Rabi oscillations. Lower panel (c): example normalized data for measurement of 7% excited state population. transition with varying angle θ ∈ [0, 2π ], followed by a π pulse on the g(cid:105) to e(cid:105) transition. Measuring the popula- tion of the g(cid:105) state results in a Rabi oscillation Ae cos(θ) with an amplitude Ae proportional to Pe . The second sequence differs only by the insertion of a π pulse to first invert the population of the g(cid:105) and e(cid:105) states, yielding a Rabi oscillation Ag cos(θ) with an amplitude Ag pro- portional to Pg . The proportionality constants between the Rabi oscillation amplitudes and the corresponding populations are equal since the same transition is used in both sequences, thus avoiding readout efficiency vari- ations. From the two oscillation amplitudes, an estimate of the population and its associated standard deviation can be calculated from Pe = Ae /(Ae + Ag ). The RPM protocol is self-calibrating and accesses smaller ampli- tudes than crude population measurements since it relies on the amplitude of an oscillation instead of just one value, in a lock-in fashion. The minimum measurable value of Pe was approximately 0.5%, limited for technical reasons by the characteristics of our readout amplifica- tion chain. In order to optimize the ground state preparation fi- delity of DDROP, we performed numerical simulations of the expected fidelity F versus qubit drive amplitude and average cavity excitation, ΩR and ¯n, respectively. We numerically simulated the Lindblad master equation obeyed by the qubit-cavity density operator, including the two drives and decoherence for both the qubit and 4 √ [29] the stabilized state is (g(cid:105)+ e(cid:105))/ 2, which requires a well-calibrated π/2 pulse to prepare g(cid:105). Interestingly, c − χ, by simply changing the cavity drive frequency to f g the stabilized state of DDROP becomes e, α(cid:105) instead of g , α(cid:105). Alternate reservoir engineering schemes can be used to stabilize Bell states of a two-qubit system or to perform an autonomous bit flip quantum error correc- tion in a three-qubit system. The agreement of DDROP measurements with numerical predictions provides a con- firmation of the validity of the basic methods of reservoir engineering and opens the door to many interesting au- tonomous feedback experiments. In conclusion, the DDROP protocol for qubit reset has been experimentally demonstrated on a transmon in a three-dimensional cavity to produce a fast, high-fidelity ground state preparation. This process satisfies the de- mand for qubit reset as part of an algorithm, and can also be used to improve the speed and fidelity of ground state preparation over that given by a return to equilib- rium. We have evaluated the performance of the DDROP protocol by using a new method (RPM) for quantifying the excited to ground state population ratio. The use of DDROP allowed experiments on this qubit to repeat at a rate 60 times faster than waiting 5 T1 . Regardless of initial state, a ground state preparation fidelity of 99.5% was achieved in less than 3 µs. Simulation predicts higher fidelities are possible; for example, simply reducing Pe from 9% to 1% and using ¯n = 25, simulations predict a fidelity of 99.99%. The requirements and constraints of DDROP are fewer than other forms of reset; neither feedback, high-fidelity readout nor qubit tunability are necessary. DDROP is readily applicable and practically useful for most cQED systems. This research was supported by IARPA under Grant No. W911NF-09-1-0369, ARO under Grant No. W911NF-09-1-0514 and NSF under grants Grant No. DMR-1006060 and No. DMR-0653377. Z.L. and M.M. acknowledge partial support from French Agence Na- tionale de la Recherche under the pro ject EPOQ2, No. ANR-09-JCJC-0070. Facilities used were supported by the Yale Institute for Nanoscience and Quantum Engi- neering and NSF Grant No. MRSEC DMR 1119826. [1] D. Divincenzo, Fortschritte der Physik 48, 771 (2000). [2] H. Paik, D. I. Schuster, L. S. Bishop, G. Kirchmair, G. Catelani, A. P. Sears, B. R. Johnson, M. J. Reagor, L. Frunzio, L. I. Glazman, et al., Phys. Rev. Lett. 107, 240501 (2011). [3] P. Schindler, J. T. Barreiro, T. Monz, V. Nebendahl, D. Nigg, M. Chwalla, M. Hennrich, and R. Blatt, Science 332, 1059 (2011). [4] M. D. Reed, L. DiCarlo, S. E. Nigg, L. Sun, L. Frunzio, S. M. Girvin, and R. J. Schoelkopf, Nature 482, 382 (2012), ISSN 0028-0836. FIG. 4: (a) Contours of 90, 95, and 99% predicted ground state preparation fidelity from numerical simulations vs two Rabi drive amplitudes expressed as ΩR /κ and ¯n. For fideli- ties greater than 99%, the shaded area indicates reset time. (b) Measured excited state population from RPM method (crosses with error bars) compared to numerical simulation (solid line) vs ¯n for ΩR /κ = 0.8. This population decreases monotonically with ¯n. amplifier at all. Finally, the decisive qualitative advan- tage is that the sensitivity to the drive amplitudes is low, and there is no need for accurate pulse timing or shapes; DDROP can be quickly tuned to near-optimum parame- ters. While a qubit reset is a fundamental primitive neces- sary for quantum information algorithms, DDROP ad- ditionally deals with “hot” qubits, which are often ob- served [23]. While usually unintentional, high qubit tem- peratures may be beneficial if loss is dominated by di- electrics [24] or if lower transition frequencies are found to be needed. A discussion of the nuance between cooling and reset is now in order. Qubit reset is ground state preparation with a minimum required fidelity in the shortest possi- ble time, whereas qubit cooling reduces the excited state population below that produced by contact with the ex- ternal bath. As shown in this Letter, DDROP satisfies both definitions, yet it differs significantly from other dy- namical cooling procedures. These methods, inherited from their counterpart in atomic physics [25], have been recently demonstrated in both nanomechanical systems [26–28] and superconducting qubits [6, 7, 29, 30]. As mentioned earlier, the DDROP protocol is one particular implementation of a wide class of procedures called reservoir engineering or autonomous feedback. In general, reservoir engineering involves designing the de- coherence landscape seen by the qubit with the goal of stabilizing a particular state or manifold. In the case of DDROP, the stabilized state is g , α(cid:105), whereas in Ref. [5] A. Abragam, The principles of nuclear magnetism (Clarendon Press, Oxford, 1961), 25 cm. The Interna- tional series of monographs on physics. [6] S. O. Valenzuela, W. D. Oliver, D. M. Berns, K. K. Berggren, L. S. Levitov, and T. P. Orlando, Science 314, 1589 (2006). [7] M. Gra jcar, S. H. W. van der Ploeg, A. Izmalkov, E. Il’ichev, H.-G. Meyer, A. Fedorov, A. Shnirman, and G. Schon, Nat Phys 4, 612 (2008). [8] M. D. Reed, B. R. Johnson, A. A. Houck, L. DiCarlo, J. M. Chow, D. I. Schuster, L. Frunzio, and R. J. Schoelkopf, Appl. Phys. Lett. 96, 203110 (2010). [9] M. Mariantoni, H. Wang, T. Yamamoto, M. Neeley, R. C. Bialczak, Y. Chen, M. Lenander, E. Lucero, A. D. OCon- nell, D. Sank, et al., Science 334, 61 (2011). [10] D. Rist`e, C. C. Bultink, K. W. Lehnert, and L. DiCarlo, Phys. Rev. Lett. 109, 240502 (2012). [11] J. E. Johnson, C. Macklin, D. H. Slichter, R. Vijay, E. B. Weingarten, J. Clarke, and I. Siddiqi, Phys. Rev. Lett. 109, 050506 (2012). [12] D. Rist`e, J. G. van Leeuwen, H.-S. Ku, K. W. Lehnert, and L. DiCarlo, Phys. Rev. Lett. 109, 050507 (2012). [13] P. Campagne-Ibarcq, E. Flurin, N. Roch, D. Darson, P. Morfin, M. Mirrahimi, M. H. Devoret, F. Mallet, and B. Huard, arXiv 1301, 6095 (2013). [14] J. F. Poyatos, J. I. Cirac, and P. Zoller, Phys. Rev. Lett. 77, 4728 (1996). [15] J. A. Schreier, A. A. Houck, J. Koch, D. I. Schuster, B. R. Johnson, J. M. Chow, J. M. Gambetta, J. Ma- jer, L. Frunzio, M. H. Devoret, et al., Phys. Rev. B 77, 180502 (2008). [16] D. I. Schuster, A. A. Houck, J. A. Schreier, A. Wallraff, J. M. Gambetta, A. Blais, L. Frunzio, J. Ma jer, B. John- son, M. H. Devoret, et al., Nature 445, 515 (2007), ISSN 0028-0836. [17] C. Rigetti, Ph.D. thesis, Yale University (2009). 5 [18] F. Lecocq, I. M. Pop, Z. Peng, I. Matei, T. Crozes, T. Fournier, C. Naud, W. Guichard, and O. Buisson, Nanotechnology 22, 315302 (2011), ISSN 0957-4484. [19] D. F. Santavicca and D. E. Prober, Measurement Science and Technology 19, 087001 (2008), ISSN 0957-0233. [20] E. Lucero, M. Hofheinz, M. Ansmann, R. C. Bialczak, N. Katz, M. Neeley, A. D. OConnell, H. Wang, A. N. Cle- land, and J. M. Martinis, Phys. Rev. Lett. 100, 247001 (2008). [21] M. Neeley, M. Ansmann, R. C. Bialczak, M. Hofheinz, E. Lucero, A. D. O’Connell, D. Sank, H. Wang, J. Wen- ner, A. N. Cleland, et al., Science 325, 722 (2009). [22] H. Paik and R. J. Schoelkopf, in Preparation (2012). [23] A. D. Corcoles, J. M. Chow, J. M. Gambetta, C. Rigetti, J. R. Rozen, G. A. Keefe, M. B. Rothwell, M. B. Ketchen, and M. Steffen, Appl. Phys. Lett. 99, 181906 (2011). [24] R. Barends, H. L. Hortensius, T. Zijlstra, J. J. A. Basel- mans, S. J. C. Yates, J. R. Gao, and T. M. Klapwijk, Appl. Phys. Lett. 92, 223502 (2008). [25] D. Leibfried, R. Blatt, C. Monroe, and D. Wineland, Rev. Mod. Phys. 75, 281 (2003). [26] J. Chan, T. P. M. Alegre, A. H. Safavi-Naeini, J. T. Hill, A. Krause, S. Groblacher, M. Aspelmeyer, and O. Painter, Nature 478, 89 (2011). [27] R. Rivi`ere, S. Del´eglise, S. Weis, E. Gavartin, O. Arcizet, A. Schliesser, and T. J. Kippenberg, Phys. Rev. A 83, 063835 (2011). [28] J. D. Teufel, T. Donner, D. Li, J. W. Harlow, M. S. Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and R. W. Simmonds, Nature 475, 359 (2011). [29] K. W. Murch, U. Vool, D. Zhou, S. J. Weber, S. M. Girvin, and I. Siddiqi, Phys. Rev. Lett. 109, 183602 (2012). [30] N. Masluk, Ph.D. thesis, Yale University (2012). 

1305.5172

1

1305

Integral Equation Analysis of Plane Wave Scattering by Coplanar Graphene-Strip Gratings in the THz Range

The plane wave scattering and absorption by finite and infinite gratings of free-space standing infinitely long graphene strips are studied in the THz range. A novel numerical approach, based on graphene surface impedance, hyper-singular integral equations, and the Nystrom method, is proposed. This technique guarantees fast convergence and controlled accuracy of computations. Reflectance, transmittance, and absorbance are carefully studied as a function of graphene and grating parameters, revealing the presence of surface plasmon resonances. Specifically, larger graphene relaxation times increases the number of resonances in the THz range, leading to higher wave transmittance due to the reduced losses; on the other hand an increase of graphene chemical potential up-shifts the frequency of plasmon resonances. It is also shown that a relatively low number of graphene strips (>10) are able to reproduce Rayleigh anomalies. These features make graphene strips good candidates for many applications, including tunable absorbers and frequency selective surfaces.

cond-mat.mes-hall

cond-mat

1 Integral Equation Analysis of Plane Wave Scattering by Coplanar Graphene-Strip Gratings in the THz Range Olga V. Shapoval, Juan Sebastian Gomez-Diaz, Julien Perruisseau-Carrier, Juan R. Mosig, and Alexander I. Nosich Abstract— The plane wave scattering and absorption by finite and infinite gratings of free-space standing infinitely long graphene strips are studied in the THz range. A novel numerical approach, based on graphene surface impedance, hyper-singular integral equations, and the Nystrom method, is proposed. This technique guarantees fast convergence and controlled accuracy of computations. Reflectance, transmittance, and absorbance are carefully studied as a function of graphene and grating parameters, revealing the presence of surface plasmon resonances. Specifically, larger graphene relaxation times increases the number of resonances in the THz range, leading to higher wave transmittance due to the reduced losses; on the other hand an increase of graphene chemical potential up-shifts the frequency of plasmon resonances. It is also shown that a relatively low number of graphene strips (>10) are able to reproduce Rayleigh anomalies. These features make graphene strips good candidates for many applications, including tunable absorbers and frequency selective surfaces. Index Terms—graphene surface plasmon strips, resonances, Nystrom-type algorithm, singular and hyper- singular integral equations I. INTRODUCTION G RAPHENE electrically are monolayers infinitesimal thin (single-atom) and display a rather This work was supported by the National Academy of Sciences of Ukraine via the State Target Program “Nanotechnologies and Nanomaterials,” the European Science Foundation via the Research Networking Programme “Newfocus,” the Swiss National Science Foundation via grant 133583, and the EU FP7 via Marie-Curie IEF grant “Marconi”, with ref. 300966. O. V. Shapoval and A. I. Nosich are with the Laboratory of Micro and Nano-Optics, Institute of Radio-Physics and Electronics of the National Academy of Sciences of Ukraine, Kharkiv 61085, Ukraine (e-mail: olga.v.shapoval@gmail.com, anosich@yahoo.com). J. R. Mosig is with the Laboratory of Electromagnetics and Acoustics, Ecole Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland (e-mail:juan.mosig@epfl.ch) J. Perruisseau-Carrier and, and J. S. Gomez-Diaz are with the group for Adaptive MicroNanoWave Systems, LEMA/Nanolab, École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland (e-mail: julien.perruisseau-carrier@epfl.ch). good electron conductivity σ that mainly depends on frequency, temperature, electron relaxation time and chemical doping [1-10]. In addition, one of the most promising features of graphene as compared with metals is the opportunity to modify its conductivity by applying and external electrostatic biasing field, which modify graphene chemical potential [1-2]. This can be easily implemented for instance by including an extremely thin polysilicon layer below the dielectric which supports graphene, and applying a DC bias between these two layers (see [1,2,8]). Graphene is very interesting in view of possibility of strong interaction with electromagnetic waves in the THz frequency range [3,4]. Indeed, it is able to support delocalized surface-plasmon waves at frequencies two orders of magnitude lower than the noble metals [3]. Graphene σ(ω, µc, τ, T) y  k  incE  incH n 1S 1NS − … NS θ x p d Fig. 1. Free-standing finite periodic grating of N infinitely long (along the z-axis) graphene strips of the same width d and period p. Some of the THz applications of graphene include giant Faraday rotation [5], fixed and reconfigurable periodic frequency selective surfaces [10,11], plasmonic waveguides, switches and lenses [6,7] or absorbers and cloaks [8,11]. In addition, there has been much interest in patterned graphene materials to control their interaction with electromagnetic waves, that has led to novel antennas [9, 12]. Graphene strips have already attracted attention in the THz science community as attractive and easily manufactured components of plasmonic waveguides, antennas [12] and sensors. Single strip scattering has been studied in [13] and its wave-guiding properties in [14]. Plasmon-assisted resonances in the scattering and  2 absorption by an infinite grating of coplanar graphene strips under normal incidence have been analyzed in [15] using the Fourier expansion method. In this context, this paper proposes a rigorous study of the transmittance, absorption and scattering of THz waves by a coplanar graphene-strip grating. Compared with [15], we present a novel integral equation approach able to very efficiently analyze the problem under study. This allows us to greatly extend the results shown in [15], including two wave polarizations, different angles of the incoming waves, the presence of finite or an infinite number of strips, and different graphene parameters. In particular, we investigate the surface plasmon resonances, which depend on each strip conductivity and width, and the build-up of the Rayleigh , the wavelengths anomalies at 1 θ − p mλ ( sin ) = ± ,.. caused by the grating periodicity and m = 1, 2 depending on the angle of incidence θ and the number of strips in the grating. The very efficient analysis of the problems is achieved via the combined use of (i) integral equations (IEs) obtained using the surface-impedance (sometimes also called resistive-sheet) boundary conditions and (ii) interpolation Nystrom-type discretization that uses polynomials and appropriate quadrature formulas accounting for the kernel-function singularities and edge behavior. As known [1,2,10,16], graphene monolayer can be electromagnetically characterized with the aid of the following boundary conditions on a zero-thickness boundary: (1 / 2)[  E + tg  E − tg + ] = σ − 1   n H [ × + tg −  H − tg ],  E + tg  E −= tg (1) imposed on the limiting values of the field components tangential (tg) to the top (+) and bottom (-) sides of layer. Note that the jump in magnetic field is the electric is called surface current and the quantity Z σ= 1 / surface impedance. Conditions (1) were widely used some time ago in the analysis of the scattering by the gratings made of so-called resistive strips [17-19], with applications to microwave absorbers. The Nystrom methods have already been used in the modeling of the wave scattering by perfectly electrically conducting strips [20-22], thin strips of conventional dielectrics [23] and finite periodic silver strip gratings in the optical band [24]. Their main merit is numerical efficiency and guaranteed convergence that entails easily controlled accuracy of computations. It allows rapid simulation of scatterers consisting of hundreds of micron-size graphene strips. The paper is structured as follows. In Section II we present the scattering problem formulation, reduce it to a set of hyper-singular integral equations (IEs), and briefly explain the discretization scheme. Section III shows the actual rate of the algorithm convergence and validates the proposed approach using data from literature and commercially available full-wave software. Section IV presents the scattering and the dependences of absorption of THz waves by a finite and infinite graphene strip gratings on the frequency and geometrical parameters. Conclusions are summarized in Section IV. A brief review of graphene conductivity characterization and details of the numerical treatment are presented in Appendix. We assume that the electromagnetic field is and omit this dependence. time-harmonic ~ i t e ω− II. FORMULATION, INTEGRAL EQUATIONS, AND NYSTROM-TYPE ALGORITHM A. Formulation and Boundary Conditions Consider at first the two-dimensional (2-D) scattering  and absorption of the H-polarized (vector E is across the strips) plane wave by a finite periodic grating made of N identical coplanar graphene strips, in the THz frequency range. The corresponding freestanding geometry and the problem notations are shown in Fig. 1. The strips are assumed infinite along the z-axis, have zero thickness, width d and period p. They are with characterized complex-valued graphene calculated via Kubo’s conductivity T ) , ( , , σωµ Γ c formalism [16, 25] (see Appendix), where ω is the cµ is the chemical potential, frequency, radian is the phenomenological scattering rate that is (2 )τ − Γ = 1 assumed to be independent of energy (τ is the relaxation time of charge carriers), and T is the temperature [1,2,11]. As known, in a 2-D scattering problem one has to find zH r that is the scattered magnetic a scalar function sc ( ) field z-component [23]. Here, the total field function   = , must satisfy the H r H r e ( ) ( ) = + sc ik x y ( cos sin ) − θ θ + r x y ( , ) z z , Helmholtz equation off the strip surface S S= 1U N = j j jb are the j ja and , and where a b S x x {( , 0) : [ , ]} = ∈ j j j S∈ take the strips endpoints. The conditions (1) at r form as   H r ( ) / ∂ z  n ik ∂ = − Z ( σ 0 ) [ 1 −    + − H r H r ( )], ( ) − z z (2) is the free-space wavenumber (c is the where k cω= / space velocity) and is the free-space Z µ ε ( / ) = 1/ 2 0 0 0 impedance.  3 For solution uniqueness, the formulation of the problem must be completed with the local integrability of power (edge condition) and the radiation condition at infinity [23,24]. B. Hyper-Singular Integral Equations To satisfy Helmholtz equation and radiation condition, we seek the scattered field as a sum of double-layer potentials, ')  dr ' , (3)  H r ( ) sc z =  w r ( ') j N ∑ ∫ j 1 = S j   G r r ( , ∂  r ' ∂     is the Green where r H k r G r r i ( , ') ( / 4) ( ' ) = − (1) 0 function. Note that the unknown functions are electric     currents induced on the strips, + − H r H r w r ( ) ( ) ( ), = − z z j S∈ , . r j = 1, ..., N j Using condition (2) and the properties of the limit values of the double-layer potentials, we obtain the following set of N coupled IEs for where jw x ( )j , x a b , ] [ ∈ j j j Z ) 4( σ − 1 0 w x ( i i 0 ) + b j N ∑ ∫ j 1 = a j w x ( j j ) H k x ( (1) 1 x − j j − x i 0 x i 0 ) j dx = f x ( i 0 ), (4) θ i ikx 0 cos where Note that e N i S f x x ). 1, ..., ( , 4 sin ) ( = ∈ = θ − i i i 0 0 these IEs are fully equivalent to the original boundary- value problem. The integrals in (4) are understood in the sense of finite part of Hadamard, and unknowns can be products represented as , where is jw x [ ( ) w x x j j w x x p p d 1) ] ( ] ( ( ) ( )[ 1) − − − = + − 1/ 2 1/ 2   j j smooth and non-singular at the S j endpoints. C. Nystrom-Type Discretization It is easy to see that the kernel functions in (5) display x→ , both hyper-type and logarithmic singularities, if x 0 because H k x ( − (1) 1 x x − 0 x 0 ) ln  ik x 0 x − π − 2 k x − π x 0 2 (5) In fact, it is this hyper-singularity, together with the step-like dependence of conductivity σ on x in the 0σ = out of strips), which makes the grating plane ( projection of (4) on the Floquet basis functions (this is Fourier-expansion method) a divergent algorithm. In contrast, we discretize the IEs (4) using Nystrom-type method based on the Gauss-Chebyshev quadrature formulas of interpolation type of the n -th order (with Chebyshev weight). As the discretization and collocations nodes, we choose the the nulls of Chebyshev polynomials of the second kind. This leads to N N× block-type matrix equation with n n× -sized it numerically we obtain blocks. After solving the form of N approximate solution of IEs in interpolations polynomials for the unknown surface currents. Then the field (3) is easily reconstructed in the near and far zone of the strip grating. Our meshless numerical algorithm is efficient and reliable and has theoretically guaranteed convergence (at least as 1 / n as follows from the convergence of quadratures) and controlled accuracy of computations. For more mathematical details about the Nystrom-type discretizations see [20-24]. III. NYSTROM-TYPE ALGORITHM CONVERGENCE AND METHOD VALIDATION A. Scattering and Absorption Characteristics To study the features of the plane wave scattering and absorption by a finite grating of N graphene strips, the figures-of-merit are not the reflectance, transmittance and absorbance usually introduced in infinite-grating scattering problems, but the total scattering cross section (TSCS) and the absorption cross-section (ACS) [26], p.13. Besides, the monostatic radar cross-section (RCS) is known to characterize the power that is reflected back to the source [27]. As the scattered field in the far zone ( r → ∞ ) takes the form , where zH i kr eπ (2 / ) ( ) ϕΦ sc ikr 1/ 2  the radiation pattern is found as = Φ ∑ ∫ N (6) k w x e dxϕ ( ) ( sin / 4) ( ) , ϕ ϕ j j 1 = S j the following expressions are obtained for TSCS and RCS: cos − ikx 2 π Φ∫ (7) k d k (2 / ) ( ) , 4 ( ) / ϕ ϕ π σ = = Φ θ σ 2 2 tsc rsc 0 The ACS is obtained from the near-field integration, = ∑ ∫ N (8) Z w x dx Re ( ) ( ) . σ − σ 1 2 abs j 0 j 1 = S j and the optical theorem (power conservation law) states that . k (4 / ) Re ( ) σ σ + = − ϕ θ Φ + abs tsc B. Algorithm Convergence As convergence the general is guaranteed by theorems, the accuracy can be, in principle, at the level of machine precision. For the demonstration of the  4 actual rate of convergence, we have computed the root- − of the mean-square deviations n n 2 e n θ θ ( ) / 1 = w w w the surface current functions, uniform norms of = ∑ ∫ 1 N , versus the order of discretization w t dt ( ) θ 2 2 j w j 1 = 1 − polynomial, n. The results are shown in Fig. 2. As visible, the proposed Nystrom method ensures algebraic convergence of the approximate solution to the accurate one with increasing discretization order n. For instance, to achieve 4-digit accuracy in the near field analysis one can take n = 55 in a frequency range up to 10 THz with a strip width of 20 µm. d = 20 µm f = 10 THz f = 5 THz d = 10 µm f = 10 THz f = 5 THz 100 10-2 10-4 10-6 1 - n w 2 θ / nw θ 0 20 60 40 n, discretization order 80 100 Fig. 2. Computation error e w (n) as a function of the discretization order n for a stand-alone graphene strip of the width d = 10 and 20 µm at f = 5 and 10 THz under the normal incidence of a H-wave; graphene parameters are τ = 1 ps, µ c = 0.13 eV and T = 300 K. m µ , S C A 0.8 0.6 0.4 0.2 0.0 0 Proposed Nystrom-type algorithm Direct numerical simulations of [13] 1 2 Frequency, THz 3 4 Fig. 3. ACS versus the frequency in THz range for a normally incident H-wave scattered by a stand-alone graphene strip of d = 5 µm calculated using proposed Nystrom-type algorithm (blue curve) and numerical simulations of [13] (red curve); graphene parameters are τ = 0.1 ps, µ c = 0.0 eV and T = 300 K. C. Algorithm Validation the proposed The ACS results obtained using Nystrom-type algorithm provides very good agreement with the data of [13] for a single infinitely long graphene strip of d = 5 µm suspended in free space (see Fig. 3). A comparison with the transmittance and absorbance of an infinite graphene-strip grating computed using the Fourier–Floquet expansion method of [15] has shown good qualitative agreement, especially in the location of resonances. However, it is known that this approach may suffer of the lack of convergence – see discussion in Section II C. Besides, we have compared our results for a stand- alone graphene strip with the data generated by the commercial software HFSS. In Fig. 4, we present the graphs of the monostatic RCS versus frequency for a strip of d = 20 µm, τ = 1 ps and µc = 0.2 eV under normal ( incidence the ) calculated with / 2 θ π= Nystrom-type algorithm (blue curve) and HFSS (red curve). HFSS results agree well with the proposed method for low frequencies; however this is not so above 2.2 THz where the mesh becomes insufficiently dense. Making it denser leads to prohibitively large simulation times. Overall obtaining the presented curve with HFSS requires about 9 hours of computation time, while modeling a similar problem using proposed Nystrom-type algorithm takes only 40 seconds with a 10- 3 THz step, in the wider frequency range. It should be noted that computing the TSCS or ACS with HFSS would lead to completely unrealistic simulation times, and therefore we provide only the RCS comparison. 102 m µ , S C R 100 10-2 Proposed Nystrom-type algorithm HSFF 10-4 0 8 2 4 6 Frequency, THz Fig. 4. RCS versus the frequency for a normally incident H-wave scattered by a stand-alone graphene strip of d = 20 µm calculated using Nystrom-type algorithm (blue curve) and HFSS (red curve); graphene parameters are τ = 1 ps, µ c = 0.2 eV and T = 300 K. 10  5 IV. NUMERICAL RESULTS AND DISCUSSION A. Stand-Alone Microsize Graphene Strip Stand-alone microsize graphene strip illuminated by the H-polarized wave in the THz frequency range demonstrates a variety of surface plasmon resonances. In Fig. 5 (a) and (b) we show the TSCS and ACS as a function of frequency for several strips of different widths d and two incidence angles, (solid / 2 θ π= (dotted curves). Here, the chemical curves) and / 4 θ π= potential is µ c = 0.13 eV, and the room 1 ps τ = temperature (T = 300 K) is assumed. As can be observed, the TSCS and ACS spectra strongly depend on the strip width. The wider strips demonstrate larger numbers of resonances on the localized surface plasmons, shifting-down in frequencies. We also observe little dependence of these parameters on the incident angle of the incoming waves. θ = π/2(solid curves) θ = π/4(dotted curves) N=1 d=20 µm d=10 µm d=5 µm 2.5 5.0 Frequency, THz (a) 7.5 10.0 β = π/2(solid curves) β = π/4(dotted curves) N=1 d=20 µm d=10 µm d=5 µm 101 m µ , S C S T 10-1 10-3 0.0 102 100 m µ , S C A 10-2 0.0 2.5 5.0 Frequency, THz (b) 7.5 10.0 Fig. 5. TSCS (a) and ACS (b) versus the frequency for the normally (solid curves) and inclined (dotted curves) incident H-wave scattering by a stand-alone graphene strip of varying width d; graphene parameters are τ = 1 ps, µ c = 0.13 eV and T = 300 K. d / y 2 d / y 2 1 0 -1 -2 2 1 0 -1 -2 2 1 0 -1 -2 2 1 0 -1 -2 d / y 2 d / y 2 zH Figs. 6 (a)-(b) display the near-field patterns of at the first four surface-plasmon resonances H m for the scattering by a graphene strip of d = 20 µm under the normal (a) and inclined (b) incidence, / 2 / 4 θ π= θ π= respectively. Here only odd-index resonances are excited at the normal incidence because of their symmetry across the y-axis, and both odd and even ones appear under the inclined incidence. It should be noted that each Hm resonance is formed as a Fabry-Perot like standing wave because of the reflections of the surface-plasmon natural wave of a graphene layer from the strip edges. The complex- valued propagation constant of this wave can be obtained analytically from (1), (9) , 2 1/ 2 Z k [1 ( / 2) ] = β σ − pl 0 and then the associated resonance frequencies satisfy (10) pl d m m Re , 1, 2, ... π≈ = β We have also analyzed the spectral response of a free-standing graphene strip for different values of chemical potential µc and relaxation time τ. 2 2 0,03500 0,03500 d / y 2 1 0 -1 -2 -2 -1 3.07 THz 0 1 2 2x/d 3 4 -2 -1 1.54 THz 0 1 2 2x/d 3 4 -2 -1 3 4 -2 -1 4.04 THz 0 1 2 2x/d 4.82 THz 0 2 1 2x/d 3 4 d / y 2 2 1 0 -1 -2 -2 -1 2.44 THz 0 1 2 2x/d 3 4 -2 -1 1.53 THz 0 1 2 2x/d 3 4 -2 -1 3 4 -2 -1 3.07 THz 0 1 2 2x.d 3.59 THz 1 2 2x/d 0 3 4 Fig. 6. Near-field patterns for a stand-alone graphene strip under the normal (a) and inclined incidence (b) for the four / 2 / 4 θ π= θ π= lower resonant frequencies. 2 1 0 -1 -2 d / y 2 (a) 0,000 0,2270 0,4190 0,6110 0,8030 0,9950 0,7800 0,8240 0,8680 0,9120 0,9560 0,9950 0,2010 0,4020 0,6030 0,8040 1,000 0,4820 0,5840 0,6860 0,7880 0,8900 0,9880 2 1 0 -1 -2 d / y 2 (b) 0,2270 0,4190 0,6110 0,8030 0,9950 0,5500 0,6380 0,7260 0,8140 0,9020 0,9900 0,9950 0,8735 0,8985 0,9235 0,9485 0,9735 0,9975 0,8735 0,8984 0,9233 0,9482 0,9731 0,9975  6 Fig. 7 shows that the magnitudes of the surface- plasmon resonances are quite sensitive to the relaxation time changes. The peaks of TSCS and ACS became more pronounced if τ increases due to the smaller dissipation losses (see Appendix). Unlike ACS, all in TSCS except of H 1 demonstrate resonances asymmetric Fano shapes. τ=0.25 ps τ=0.5 ps τ=0.75 ps τ=1 ps 0 2 8 4 6 Frequency, THz (a) 10 101 100 10-1 m µ , S C S T . 102 101 100 10-1 m µ , S C A τ=0.25 ps τ=0.5 ps τ=0.75 ps τ=1 ps 7.5 10.0 0.0 2.5 5.0 Frequency, THz (b) Fig. 7. Same study as in Fig. 5 but for a normally incident H-wave and a strip width d = 20 µm for different values of relaxation time, τ = 0.25, 0.5, 0.75 and 1 ps and fixed chemical potential µc = 0.13 eV. In turn, Fig. 8 shows that an increase of µc up-shifts the resonance frequencies and decreases losses in graphene. This is in agreement with the surface impedance of graphene (see Appendix) and explains the behavior of TCS and ACS in Fig. 8, where significant enhancement of the absorption coefficient is obtained with a slight change in the chemical potential µc . 101 100 10-1 10-2 m µ , S C S T 10-3 0.0 102 101 100 10-1 m µ , S C A 2.5 5.0 Frequency, THz (a) 0.0 2.5 5.0 Frequency, THz (b) µc=0.0 eV µc=0.13 eV µc=0.26 eV µc=0.39 eV 7.5 10.0 µc=0.0 eV µc=0.13 eV µc=0.26 eV µc=0.39 eV 7.5 10.0 Fig. 8. Same study as in Fig. 7 but for a fixed relaxation time of τ = 1 ps and different values of the chemical potential µc = 0.0, 0.13, 0.26 and 0.39 eV, T = 300 K. B. Resonance Prediction According to the Strip Width In order to obtain a better insight into single-strip resonances, Fig. 9 presents the dependences of the resonance frequencies of the first to the fourth-order plasmon resonances versus the strip width. For comparison, the stars show the same values obtained by solving the approximate equation (10). This result demonstrates that a wider free-standing graphene strip present more the in resonances considered THz range, with the lowest of them down- shifted in frequencies.  1st order SPR 2nd order SPR 3rd order SPR 4th order SPR 102 101 100 N / S C S T 15 20 10-1 0.0 102 101 100 N / S C A 7 N=1: d=60 µm N=3: d=20 µm g=0.01 µm g=0.1 µm g=1 µm g=7 µm g=30 µm 2.5 5.0 7.5 Frequency, THz (a) 10.0 N=1: d=60 µm N=3: d=20 µm g=0.01 µm g=0.1 µm g=1 µm g=7 µm g=30 µm 10.0 10.0 7.5 5.0 2.5 z H T , y c n e u q e r f n o m s a l P 0.0 0 5 10 Strip width, µm in Finite Graphene Strip Fig. 9. Dependence of the 1st, 2nd, 3rd and 4th order plasmon resonance frequencies versus graphene strip width; graphene parameters are T = 300 K, τ = 1 ps and µ c = 0.13 eV. C. Gap Size Effects Gratings In Fig. 10, we present the plots of TSCS (a) and ACS (b) of the grating of three graphene strips of widths d = 20 µm, focusing on the influence of the gap size, g. These results show that, if the gaps between the strips are large, the resonances keep their shapes and positions and the scattering and absorption spectra per one strip show the same behavior as for a stand-alone strip. Narrower gaps provide, however, a stronger wave coupling between the strips so that the whole terahertz- range spectral response of the three 20-µm strip configuration gradually transforms into the response of one 60-µm strip. In Fig. 11, we consider a larger grating of N = 10 strips of widths d varying from 10 to 60 µm and fixed period p = 70 µm. As far as the airgaps get smaller, the interaction between the strips becomes larger and the plasmon resonances shift down in frequency. D. Comparison between Finite and Infinite Gratings for H- and E-wave Scattering It is interesting to compare the THz properties of infinite and finite gratings made of graphene strips. However it is not obvious how to select a common figure-of-merit. We have found that the reflectance of a plane wave by a finite grating can be introduced as the part of TSCS associated with the power scattered into the upper half-space. The transmittance of finite grating can be derived in similar manner however with account of the optical theorem. In addition, we normalize these values by the strip electric width kd, the number of strips N and , that leads us to expressions /d p ζ = 10-1 0.0 2.5 5.0 7.5 Frequency, THz (b) Fig. 10. Normalized TSCS (a) and ACS (b) versus the frequency for a normally incident H-wave scattered by a grating of N = 3 strips of width d = 20 µm for the different gaps between strips g; graphene parameters are T = 300 K, τ = 1 ps and µc = 0.39 eV. finR finT 1 = + = Nkd 2 / ( ζ π 2 ζ Nkd π 2 π ∫ π Φ ) π Φ∫ 0 2 d ( ) ϕ ϕ + 2 d , ( ) ϕ ϕ 4 ζ Nkd (11) Re ( ), ϕ θ Φ + (12) ) fin fin + R while the absorbance can be found as T A 1 ( = − fin that follows from the conservation of power. These quantities can be conveniently compared to the reflectance, transmittance and absorbance of infinite grating. In Fig. 12, we present such comparison for the H- and E-wave scattering by finite (N = 10 and 50) and infinite strip gratings with d = 20 µm and p = 70 µm, at the normal incidence. Note that the alternative case of the E-  polarized wave scattering (vector E is parallel to the strips) reduces to a set of N coupled IEs with logarithmic singularities in the kernels. They are solved similarly to section II using Nystrom-type discretization with Gauss- Legendre quadrature formulas [21, 23]. This is because the currents tend to finite values at the edges of resistive  102 101 100 10-1 N / S C S T N =10: p=70 µm d=20 µm d=40 µm d=60 µm 8 from 0.1 to 10 THz, except for the narrow bands around the Rayleigh anomalies. Interestingly, either of polarizations does not display high-quality grating resonances that have been recently reported for the silver nanostrip gratings in the visible range [28] in the H-polarization case. 2 R.A. 0.02 0.01 p i r t s r e p e c n a t c e l f e R H-case 3.8 4.0 4.4 4.2 Frequency, THz (b) 4.6 10-1 10-3 p i r t s r e p e c n a t c e l f e R 100 10-2 p i r t s r e p e c n a b r o s b A 10-4 0.0 2.5 7.5 10.0 5.0 Frequency, THz (a) N=10: p=70 µm d=20 µm d=40 µm d=60 µm 101 N / S C A 100 10-1 0.0 2.5 5.0 7.5 Frequency, THz (b) 10.0 Fig. 11. Same study as in Fig. 7 but for a normally incident H- wave scattered by a grating of N = 10 strips with different values of width d and fixed period p = 70 µm; graphene parameters are τ = 1 ps, µ c = 0.13 eV and T = 300 K. strips and hence no weight is needed In the case of infinite strip grating, we have used the same algorithm as explained in Section II where Green’s function of the free space is replaced with the pseudo- periodic Green’s function (having the phase factor of the same ikp θ ). As both functions have exp( cos ) logarithmic singularity, this does not lead to new quadrature formulas. As one can see, the H-polarization case demonstrates plasmon resonances in the THz range and a gradual build-up of the Rayleigh anomalies at the associated , . As expected, in the E- wavelengths m = /p m 1, 2 λ= polarization case no plasmon resonances appear, however more pronounced Rayleigh anomalies are build and show enhanced transmission: one where the wavelength equals to period, f = 4.283 THz and the other where it makes a half of period, f = 8.565 THz. As visible, even 10 strips provide normalized reflectance and absorbance values very close to the infinite grating values in the whole band of frequencies d=20 µm, p=70 µm H-polarization N=infinity N=50 N=10 E-polarization N=infinity N=50 N=10 8 10 4 6 Frequency, THz (a) 10-1 10-3 p i r t s r e p e c n a t c e l f e R 10-5 3.5 R.A. E-case 4.0 4.5 Frequency, THz (c) 5.0 d=20 µm, p=70 µm H-polarization N=infinity N=50 N=10 E-polarization N=infinity N=50 N=10 8 2 10 4 6 Frequency, THz (d) Fig. 12. Reflectance (a) and absorbance (d) per one strip for the normally incident plane waves of two polarizations scattered by the infinite and N-strip gratings of d = 20 µm and p = 70 µm. Panels (b) and (c) show zoomed reflectance near the Rayleigh anomaly wavelength in H-case and E-case, respectively; graphene parameters are τ = 1 ps, µc = 0.39 eV and T = 300 K.  9 Fig. 13 shows the near E-field in the vicinity of the Rayleigh anomaly at f = 4.274 µm for a N = 50 strip grating (around the strips 25 to 28) with width d = 20 µm and period p = 70 µm, demonstrating good transparency. The corresponding near H-field patterns for the H-wave scattering at the first-order plasmon f = 2.6 THz (b) and at the first Rayleigh resonance anomaly frequency f = 4.27 THz (c) demonstrate high reflectivity of the incident plane wave and slightly increased transparency, respectively. 0,2850 0,4950 0,7050 0,9150 1,125 1,335 1,545 1,665 0,000 0,6000 1,200 1,800 2,400 3,000 3,600 3,960 0,7460 0,8180 0,8900 0,9620 1,034 1,106 1,178 1,226 Application of the method shows the presence of surface plasmon resonances in the THz range in the case of H-polarization, and a gradual build-up of the Rayleigh anomalies as the strip number gets larger in the both polarization cases. We have also investigated the tunability of the mentioned resonances with respect to the graphene chemical potential, relaxation time and the grating configuration. These effects can be potentially exploited in the design of tunable THz range filters, frequency selective surfaces, for panels absorbing ultrathin and electromagnetic compatibility. ACKNOWLEDGMENT We are grateful to A.Y. Nikitin and L. Martin-Moreno for valuable discussions. APPENDIX: GRAPHENE CONDUCTIVITY Graphene conductivity is characterized applying the Kubo formula [1,2,11,16], ( , σωµ c , Γ , T ) = i 2 ) [( + Γ ξ ω 2 − ∞  ∫ ε   0 f ( ) ε d 4( / ε  − ∞ ∫ 0 f ( ) ε − − d i ( 2 ) ω + Γ − 2 f ( ) ∂ ε d ∂ ε f ∂ − ( ) − ε d ∂ ε    ∂ ε ] ε ∂ , (A1) 2 ) , and ε is the energy, where eiq i T ( 2 )( ) π − ω ξ = − + 2 2 1  the is eq the Fermi-Dirac distribution, is df ε ( ) elementary charge, and  is the reduced Plank constant. The first term in (A1) relates to intraband contributions of graphene, which usually dominate in the low THz range, and interband to term relates the second contributions of graphene, which become more important at higher frequencies. From the engineering point of view, it is useful to study the surface impedance (sometimes also called complex resistivity) of a graphene monolayer defined as . Fig. A1 shows the real and imaginary parts of Z σ= 1 / this quantity versus different values of the chemical potential, considering the room temperature T = 300 K and the relaxation time τ = 1 ps. Note that μ c can be easily varied using an external electrostatic field. Analysis shows that graphene behaves almost as a frequency-independent resistor, with a significant purely inductive reactance. It is observed that an increase of the chemical potential μ c leads to the lower losses and an up-shift of the frequencies where graphene presents large inductive behavior. The latter feature makes this material appropriate for the propagation of delocalized surface plasmons, which are transverse magnetic waves d / y 2 10 5 0 -5 -10 165 170 175 10 5 0 -5 -10 d / y 2 d / y 2 10 5 0 -5 -10 165 165 170 175 170 175 185 190 195 185 190 195 180 2x/d (a) 180 2x/d (b) 185 190 195 180 2x/d (c) Fig.13. Near E-field pattern around the strips 25 to 28 for a E-wave normally incident at the grating of N = 50 graphene strips of d = 20 µm and p = 70 µm near Rayleigh anomaly frequency f = 4.274 THz (a); H-field patterns for the H-wave incidence in the plasmon resonance at f = 2.6 THz (b) and at Rayleigh anomaly (c). V. CONCLUSIONS In summary, we have presented a numerically efficient and accurate analysis of the scattering and absorption of plane waves by finite and infinite coplanar graphene-strip gratings in the THz frequency range. Unlike standard full-wave commercial software, the developed meshless algorithm is based on the singular IEs and Nystrom-type discretizations, which provides a theoretically guaranteed convergence up to machine precision.  Ω , Z e R Ω , Z m I - (a) (b) Fig. A1. Real (a) and imaginary (b) parts of the graphene impedance Z σ= in the THz range calculated at the room temperature T = 1 / 300 K and τ =1 ps versus the chemical potential µc . traveling along the interface between graphene and dielectric [1-3]. REFERENCES [1] K. Geim and K. S. Novoselov, “The rise of graphene,” Nature Mater. , vol. 6, pp. 183–191, 2007. [2] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, “Magnetooptical conductivity in graphene,” J. Physics: Condensed Matter, vol. 19, no. 2, pp. 026222, 2007. [3] M. Jablan, H. Buljan, and M. Soljacic, “Plasmonics in graphene at infrared frequencies,” Phys. Rev. B, vol. 80, pp. 245435, 2009. [4] T. Otsuji, S. A. Boubanga Tombet, A. Satou, H. Fukidome, M. Suemitsu, E. Sano, V. Popov, M. Ryzhii, and V. Ryzhii, “Graphene-based devices in terahertz science and technology,” J. Phys. D: Appl. Phys., vol. 45, pp. 303001-(9), 2012. [5] I. Crassee, J. Levallois, A. L. Walter, M. Ostler, A. Bostwick, E. Rotenberg, T. Seyller, D. van der Marel, and A. B. Kuzmenko, “Giant Faraday rotation in single- and multilayer graphene,” Nature Phys., vol. 7, pp. 48-51, 2011. [6] Y. V. Bludov, M. I. Vasilevskiy, and N. M. R. Peres, “Mechanism for graphene-based optoelectronic switches by tuning surface plasmon-polaritons in monolayer graphene,” European Phys. Lett., vol. 92, pp. 68001, 2010. 10 [7] A. Vakil and N. Engheta “Transformation optics using graphene,” Science, vol. 332, no 10, pp. 1291-1294, 2012. [8] P. Y. Chen and A. Alu, “Atomically thin surface cloak using graphene monolayers,” ACS Nano, vol. 5, no 7, pp. 5855-5863, 2011. [9] I. Llatser, C. Kremers, A. Cabellos-Aparicio, J. M. Jornet, E. Alarcón and D. N. Chigrin, “Graphene-based nano-patch antenna for terahertz radiation,” Photon Nanostruct: Fundam. Appl, vol. 10, no 4, pp. 353-358, 2012. [10] S. Thongrattanasiri, F. H. L. Koppens, and F. J. Garcia de Abajo, “Complete optical absorption in periodically patterned graphene,” Phys. Rev. Lett. vol. 108, no. 4, pp. 047401-(5), 2012. [11] M. Tamagnone, J. S. Gómez-Díaz, J. R. Mosig, and J. Perruisseau-Carrier, “Reconfigurable THz plasmonic antenna concept using a graphene stack,” Appl. Phys. Lett, vol. 101, 214102, 2012. [12] A. Fallahi and J. Perruisseau-Carrier, “Design of tunable biperiodic graphene metasurfaces,“ Phys. Rev. B, vol. 86, pp. 195408-(9), 2012. [13] I. Llatser, C. Kremers, A. Cabellos-Aparicio, J. M. Jornet, E. Alarcon and D. Chigrin, “Scattering of terahertz radiation on a Int. Workshop nano-antenna,” Proc. graphene-based Theoretical Computational Nanophotonics (TaCoNa—2011), AIP Conf. Proc., vol. 1398, pp. 144-146, 2011. [14] A. Y. Nikitin, F. Guinea, F. J. Garcia-Vidal, and L. Martin-Moreno, “Edge and waveguide terahertz surface plasmon modes in graphene microribbons,” Phys. Rev. B., vol. 84, pp. 161407(R)-(4), 2011. [15] A. Y. Nikitin, F. Guinea, F. J. Garcia-Vidal, and L. Martin-Moreno, “Surface plasmon enhanced absorption and suppressed transmission in periodic arrays of graphene ribbons,” Phys. Rev. B, vol. 85, pp. 081405(R)- (4), 2012. [16] G. W. Hanson, “Dyadic Green’s functions for an anisotropic, non-local model of biased graphene,” IEEE Trans. Antennas Propagat., vol. 56, no 3, pp. 747-757, 2008. [17] R.C. Hall and R. Mittra, “Scatteing frm aperiodic array of resistive strips,” IEEE Trans. Antennas Propagat., vol. 33, no 9, pp. 1009-1011, 1985. [18] J.L. Volakis, Y.C. Lin, and H. Anastassiou, “TE characterization of resistive strip gratings on a dielectric slab using a single-mode expansion,” IEEE Trans. Antennas Propagat., vol. 42, no 2, pp. 205-213, 1994. [19] T. L. Zinenko, A. Matsushima, and Y. Okuno, Scattering and absorption of electromagnetic plane waves by a multilayered resistive strip grating embedded in a dielectric slab, Trans. IEICE Electronics, vol. E82-C, no 12, pp. 2255-2264, 1999. [20] A. A. Nosich and Y. V. Gandel, “Numerical analysis of quasioptical multi-reflector antennas in 2-D with the method of discrete singularities: E-wave case,” IEEE Trans. Antennas Propagat., vol. 55, no.2, pp. 399-406, 2007. [21] J. L. Tsalamengas, “Exponentially converging Nystrom methods in scattering from infinite curved smooth strips. Part 1: TM-Case” IEEE Trans. Antennas Propagat. vol. 58, no 10, pp 3275-3273, 2010. [22] J. L. Tsalamengas, “Exponentially converging Nystrom methods in scattering from infinite curved smooth strips. Part 2: TH-Case” IEEE Trans. Antennas Propagat., vol. 58, no 10, pp 3274-3281, 2010. [23] O. V. Shapoval, R. Sauleau, and A. I. Nosich. “Scattering and absorption of waves by flat material strips analyzed using generalized boundary conditions and Nystrom-type algorithm,”  11 IEEE Trans. Antennas Propag., vol. 59, no 9, pp. 3339-3346, 2011. [24] M. V. Balaban, E.I. Smotrova, O. V. Shapoval, V. S. Bulygin, and A. I. Nosich, "Nystrom-type techniques for solving electromagnetics integral equations with smooth and singular kernels," Int. J. Numerical Modeling: Electronic Networks, Devices and Fields, vol. 25, no 5, pp. 490-511, 2012. [25] V. P. Gusynin, S. G. Sharapov, and J. B. Carbotte, “On the universal AC optical background in graphene,” New. J. Phys., vol. 11, pp. 095013, 2009. [26] H. C. van de Hulst, Light Scattering by Small Particles, New York: Dover Publ., 1981. [27] G.T. Ruck, Radar Cross Section Handbook, New York, Plenum Press, 1970. [28] T. L. Zinenko, M. Marciniak, and A.I. Nosich, “Accurate analysis of light scattering and absorption by an infinite flat grating of thin silver nanostrips in free space using the method of analytical regularization,” IEEE J. Selected Topics in Quantum Electronics, vol. 19, no 3, 2013. 

1601.07098

2

1601

Electrode Design for Antiparallel Magnetization Alignment in Nanogap Devices

The ability to manipulate the relative magnetization alignment between ferromagnetic source and drain electrodes attached to a molecule or small quantum dot is a prerequisite for a number of spintronic device applications. The influence of electrode shape and field orientation on pair-wise magnetization reversal mechanisms in nanogap and point-contact structures is investigated here using micromagnetic simulations. A favorable device geometry and setup are identified for enabling planar, monodomain source and drain electrodes with a magnetization alignment that may be controllably switched between a parallel and anti-parallel configuration.

cond-mat.mes-hall

cond-mat

Electrode Design for Antiparallel Magnetization Alignment in Nanogap Devices G. D. Scott1, a) Bell Laboratories, Nokia, 600 Mountain Ave, Murray Hill, NJ 07974 (Dated: 24 March 2021) The ability to manipulate the relative magnetization alignment between ferromagnetic source and drain elec- trodes attached to a molecule or small quantum dot is a prerequisite for a number of spintronic device applications. The influence of electrode shape and field orientation on pair-wise magnetization reversal mech- anisms in nanogap and point-contact structures is investigated here using micromagnetic simulations. A favorable device geometry and setup are identified for enabling planar, monodomain source and drain elec- trodes with a magnetization alignment that may be controllably switched between a parallel and anti-parallel configuration. I. INTRODUCTION Understanding and controlling the magnetization of micro- and nano-scale structures has been an essential component of spintronics research for purposes such as data storage, information processing, and magnetic field sensing.1,2 Many studies have focused on the magnetiza- tion reversal processes in both isolated and ordered ar- rays of magnetic nanoelements,2 -- 8 but there are a num- ber of novel applications which require control of the magnetization alignment between isolated pairs of pla- nar electrodes attached to a molecule or quantum dot. These include spin state detection of individual magnetic atoms or molecules,9 nanoelectromechanical single elec- tron shuttle devices,10,11 and tools for the investigation of quantum critical phenomena relevant to heavy fermion materials.12 Reliable methods have been established to control the relative orientation of magnetization between stacked magnetic layers separated by non-magnetic material, as in spin-valve devices.13 On the other hand, controlling the magnetization orientation between two adjacent thin film structures with nanometer-scale separation has proven to be more difficult. Experimental efforts with break junction devices have found limited success by producing what are likely to be in-plane flux-closure (vortex) magnetization distributions. Here, micromag- netic modeling is used to analyze the impact of geometry and magnetic field orientation on the magnetization reversal mechanisms in closely spaced source-drain electrode-pairs. The goal is to find a realistic design and device setup for obtaining electrodes with nanometer separation and uniform, monodomain magnetization distributions that may be reliably switched between two states of relative alignment: parallel (P) and anti-parallel (AP). II. NUMERICAL SIMULATION age, N mag.14 The source and drain electrodes are ini- tialized into a P configuration with an in-plane exter- nal magnetic field, Hext, roughly equal to Hs, the field required to saturate the magnetization in a particular direction. A small field component (1%) orthogonal to Hext is added to break the symmetry. The Landau- Lifshitz-Gilbert equation is integrated numerically over time14,15 at incrementally changing values of external field as Hext is swept from Hs to -Hs and then back to Hs. Permalloy (Ni80Fe20) was chosen as a test ma- terial due to its lack of magnetocrystalline anisotropy and common usage among research and industrial appli- cations. The material parameters used were saturation magnetization Ms = 7.958 ×105 A/m, exchange coupling strength C = 13 × 10−12 J/m, and damping parameter α = 0.5 or a more realistic value of α = 0.05 in selected cases. Earlier efforts experimental to create magnetic nanocontact devices that could be alternated between P and AP configurations focused on shape anisotropy, shapes,16 -- 18 utilizing two electrodes with different like the triangle-rectangle (T-R) geometry in Fig. 1, connected via atomic-scale contacts or tunneling gaps formed with standard break junction techniques. Shape anisotropy accounts for the relationship between the mean magnetization direction and the geometrical form of a magnetic element leading to a demagnetizing field magnitude dependent upon the direction of an applied external field. It was presumed that as Hext was swept from Hs through Hext = 0 A/m, the different coercivities associated with the mismatched geome- tries would enable the magnetization of one electrode to reverse polarity prior to the other, leading to an AP magnetization alignment. However, this premise ignores the magnetostatic coupling between the two electrodes, which becomes increasingly relevant as the inter-electrode separation decreases.4 The magnetization reversal process is investigated here with the finite-element micromagnetic simulation pack- III. RESULTS AND DISCUSSION a)Electronic mail: gavin.scott@nokia.com The T-R model shown in Fig. 1 was designed with 20 nm thickness, and length scales that approximated the dimensions used in Refs. 16 -- 18. Results of the simu- 2 FIG. 1. Snapshots of simulated magnetiza- tion reversal in the T-R geometry as Hext is reduced from positive saturation (a), at re- manence (b), immediately before (c) and af- ter (d) the jump at Hc, and further past the jump (e). Colorbar indicates normalized x- component of magnetization, Mx. (f) To- tal Mx vs. Hext calculated as Hext is swept down from Hs (magenta circles) and back up (green squares). Arrows indicate sweep direction. (g)-(i) Close up of nanogap re- gion corresponding to (c)-(e), respectively. Cones denote local magnetization direction. the individual elements. In a design used in Refs. 19,20 a pair of nominally identical elliptical contacts are uti- lized as source and drain electrodes. The long axis of the ellipses are aligned parallel to one another and per- pendicular to both the bridge connecting them and the direction of current flow. The external field, Hext, is aligned parallel to the long axis of the ellipses. To simulate the evolving magnetization distribution in devices akin to those in Refs. 19,20, we used a 15 nm thick base model with lateral dimensions as shown in Fig. 2(a). Simulations were alternately performed using four ba- sic design alternatives (Figs. 2(b)-(d)), each tested with Hext aligned along the x-axis and y-axis. The unbroken version consists of the two ellipses connected with the a small constriction, or bridge, of finite width (4 − 12 nm). The point-contact has a continuous bridge constricting down to a single point. The nanogap includes a tapered bridge possessing a gap between the left and right ends of the constriction from 6 nm down to 2 nm - an approx- imate lower bound on the length scale, below which the validity of the calculations may be ambiguous. Lastly, the ideal set of leads consists of the ellipses only (i.e. no constriction). This allowed for the effects of shape anisotropy and dipole-dipole interaction on the ellipses to be deduced both with and without the influence of the extraneous bridge material. Simulations using an easy axis field orientation for both the ideal and nanogap versions of this device model demonstrate an abrupt switch between a P state in which the magnetization of both electrodes points almost en- tirely in the same direction (+y) to an intermediate con- figuration with both ellipses exhibiting displaced single vortex states (Fig. 3(a),(b)). This results in a local AP FIG. 2. (a) Example of ellipse-pair geometry with the ideal model version. Enlarged view of constriction region, corre- sponding to red dashed box in (a), highlights variations be- tween the (b) unbroken, (c) point-contact, and (d) nanogap variations of the model. lations with this model validate that it will not accurately produce a uniform AP configuration between source and drain contacts. The magnetization reversal may be char- acterized here in part with snapshots of the longitudinal (x -axis) magnetization component magnitudes, denoted by colormap, and the transient spin configuration, de- noted by vector field, acquired at incrementally changing values of Hext. As Hext is reduced from saturation in the +x direction a vortex core nucleates near an edge in each lead and propagates roughly along the y-axis (Fig. 1(a)- (e)). The reversal process is further delineated with a graph of the sum of longitudinal magnetization vector components plotted as a function of Hext (Fig. 1(f)). This produces a hysteresis loop as the magnetization dis- tribution within the electrodes undergoes an abrupt shift at some critical field, Hc. An AP configuration would be expected at values of Hext for which ΣMx → 0. When the field is swept past −Hc two vortices with opposite chirality exist in each lead. As Hext is further reduced the vortices move to the edges and are ultimately an- nihilated. The evolving spin configuration in the T-R geometry (Fig. 1(g)-(i)) enables the magnetization of the electrodes to transition from a P state to a non-collinear configuration in which neighboring vortex states may ef- fectively lead to a local AP magnetization orientation between the domains immediately on the left and right sides of the tunneling gap. Here the remainder of the electrodes are in a mixed state of buckling modes and vortex-propagation modes, as is common for the magne- tization reversal in ferromagnetic thin film structures.15 Simulations were also run on other T-R pairs with var- ious scales, length/width ratios, and gap sizes and were found to produce qualitatively similar results (i.e. mixed state modes and the possibility of a local AP alignment at the break junction site). Magnetization reversals have been studied with vari- ously shaped elements, but ellipses have received partic- ularly close attention partially due to their well defined uniaxial shape anisotropy. The long (short) diameter of an ellipse corresponds to its easy (hard) axis, along which the demagnetizing field and magnetostatic energy are minimized (maximized). Magnetization reversals with closely-spaced ellipses rely on their magnetostatic dipole- dipole interaction in addition to the shape anisotropy of 3 along the hard axis. Representative snapshots of the sim- ulated magnetization distribution using the ideal device model with a longitudinal field orientation are presented in Fig. 3(d)-(f). For all versions of this device model, as Hext is reduced from Hs the magnetization in the regions along the outer edges of the ellipses begins to rotate in order to reduce the magnetostatic energy. In the example shown in Fig. 3(d), the magnetization of the left electrode turns upward and the magnetization of the right elec- trode turns downward. At the same time the constric- tion region (when present) remains homogenously mag- netized. As the field is further reduced, a buckling-type configuration arises which encompasses both ellipses such that they exhibit a joint magnetization distribution pat- tern with an inverted U-shape (Fig. 3(e)). For the ideal and nanogap models, this distribution pattern persists as Hext is swept past zero until an abrupt re-distribution occurs at the critical field, resulting in the configuration shown in Fig. 3(f) and Fig. 3(g) inset, respectively. Here the two electrodes exhibit nearly homogenous magnetic moments with opposite polarity, which is precisely the target magnetization configuration. For the unbroken and point-contact models this AP state is never reached, thus the presence of a tunneling gap is critical, consistent with the fact that the exchange energy would oppose a large rotation of magnetic moments between two adja- cent atoms in a chain. The magnetization distributions closest to the AP configuration can be seen in the insets of Fig. 3(h). Further simulations examining the toler- ance on the field angle for obtaining the AP configuration showed that this intermediate state is extremely robust with respect to an out-of-plane component of Hext, but is quite sensitive to the inclusion of a y-axis component, such that it may fail to form for in-plane field angles deviating beyond a small range (∼ 2◦) from the x -axis. Analogous simulations were also run with different variations of the ellipse-pair base model. These included versions of the above geometry with uniform in-plane scaling (×0.75, ×1.2 and ×1.5), modified ellipse eccen- tricity (height/width ratios of 1.5, 2.0, and 2.5), and in- creased material thickness (20 nm, 25 nm, and 30 nm) and ellipse spacing (30 nm and 40 nm). Every geometri- cal alteration affects the magnetostatic energy of a device and can impact its magnetic response to an external field. In general, smaller, thinner ellipses with closer spacing and larger eccentricity exhibited qualitatively similar re- versal processes, but with increased ranges of Hext over which the monodomain-like AP configuration was main- tained, which may be a beneficial attribute. However, advantages of reduced dimensions will be constrained as the electrodes become more difficult to create and contact with existing fabrication techniques. The ellipse-pair ge- ometry in Fig. 3 presents a practical solution (i.e. easily within fabrication limitations) with a magnetization re- versal process representative of the various models tested, but it does not preclude a more optimal design for a given application. Another variation was designed to more closely FIG. 3. Snapshots of simulated magnetization distribution us- ing the ideal ellipse-pair geometry with a transverse (y-axis) field orientation at remanence (a), and at H1 ext (b) immedi- ately after the jump at -Hc as labeled in (c). Colorbar (same for all of Fig. 3) indicates local My magnitude and black cones denote local magnetization direction. Intermediate states lead to vortices in both ellipses, but the AP configuration with uniform magnetization is never achieved. (c) Calculated hys- teresis loop of normalized My vs. Hext, from same simulation as data in (a) and (b). (d) Snapshots using the same model with a longitudinal (x-axis) field orientation as Hext is ini- tially reduced from saturation (d), shortly past remanence (e), and immediately after the jump at -Hc (f). (g) Calcu- lated hysteresis loop of Myvs.Hext, using the nanogap (3 nm gap) device variation. Inset: Monodomain-like AP magneti- zation alignment still possible when a gapped constriction is included in the simulation geometry. (h) Same as (g), but for the point-contact model. Insets: top left and bottom right show the transient spin configuration before and after, re- spectively, the jump at -Hc. The uniform AP configuration is never achieved. configuration between the closest points on the ellipses for the ideal model and on opposing sides of the tunnel- ing barrier in the nanogap model. The colorbar indicates the magnitude of the transverse magnetization compo- nent and the cones indicate local magnetization orienta- tion. When the point-contact and unbroken versions of this device model are used in the simulation, comparable intermediate states are not reached. Instead the abrupt transition at the coercive field takes the system from a P state in which the magnetization of both electrodes is nearly uniform in the +y-direction to the same P state in the -y-direction. A distinctly different reversal process is demonstrated by simulations performed using the same electrode ge- ometry and finite element mesh, but with Hext applied mimic the dimensions of the permalloy ellipses used in Refs. 19,20, which were deposited partially on top of pre- existing gold contacts to avoid the formation of an oxide barrier, thereby creating a step near the middle of each ellipse. Simulations performed using this non-planar geometry indicate that for the hard axis field orienta- tion, the magnetization distribution does not exhibit the jump to the AP configuration. Instead a transformation occurs resulting in single vortex states in each ellipse, much like the intermediate state of planar ellipses with an easy axis field orientation shown in Fig 3(b). For an applied field along the easy axis of the non-planar model a mixed state is reached in which one ellipse is homogeneously magnetized along the y-axis while the opposing electrode contains a single displaced vortex. IV. CONCLUSION The numerical simulations performed here indicate that the ellipse-pair electrode design is a favorable ge- ometry for obtaining the desired AP magnetization con- figuration in break junction devices. When an external magnetic field is oriented along the hard axis of ellipses with nanometer-scale separation, the system may be con- trollably tuned from a P configuration to a state in which the source and drain contacts exhibit monodomain-like magnetization distributions with AP alignment, whereas an easy axis field orientation will only result in vortex states within each ellipse. Furthermore, we find that the magnetization distribution will be influenced by the non- planar shape produced as a result of a fabrication process that involves partially overlapping the magnetic elements on top of other thin films of comparable thickness. Over- coming fabrication obstacles may provide further insights into the magnetization reversal mechanisms in the struc- tures addressed here by allowing for a direct comparison between simulated and experimental data. For instance, measurements of magnetoresistance in break junction de- vices with the geometries described above could be ex- 4 amined to substantiate the accuracy of these numerical results. 1S. D. Bader, Rev. Mod. Phys. 78, 1 (2006). 2R. P. Cowburn, Journal of Physics D: Applied Physics 33, R1 (2000). 3M.-F. Lai, C.-R. Chang, J. Wu, Z.-H. Wei, J. Kuo, and J.-Y. Lai, Magnetics, IEEE Transactions on 38, 2550 (2002), ISSN 0018-9464. 4V. Novosad, M. Grimsditch, J. Darrouzet, J. Pearson, S. D. Bader, V. Metlushko, K. Guslienko, Y. Otani, H. Shima, and K. Fukamichi, Applied Physics Letters 82, 3716 (2003). 5P. Vavassori, N. Zaluzec, V. Metlushko, V. Novosad, B. Ilic, and M. Grimsditch, Phys. Rev. B 69, 214404 (2004). 6F. Carace, P. Vavassori, G. Gubbiotti, S. Tacchi, M. Madami, G. Carlotti, and T. Okuno, Thin Solid Films 515, 727 (2006). 7J. W. Lau, M. Beleggia, and Y. Zhu, Journal of Applied Physics 102, 043906 (2007). 8X. Yin, S. H. Liou, A. O. Adeyeye, S. Jain, and B. Han, Journal of Applied Physics 109, 07D354 (2011). 9J. Chen, Y. Hu, K. Xia, and Z. Ma, Applied Physics Letters 92, 182104 (2008). 10S. I. Kulinich, L. Y. Gorelik, A. N. Kalinenko, I. V. Krive, R. I. Shekhter, Y. W. Park, and M. Jonson, Phys. Rev. Lett. 112, 117206 (2014). 11O. A. Ilinskaya, S. I. Kulinich, I. V. Krive, R. I. Shekhter, Y. W. Park, and M. Jonson, arXiv [cond-mat.mes-hall] (2015). 12S. Kirchner, L. Zhu, Q. Si, and D. Natelson, P.N.A.S. 102, 18824 (2005). 13S. Parkin, X. Jiang, C. Kaiser, A. Panchula, K. Roche, and M. Samant, Proceedings of the IEEE 91, 661 (2003), ISSN 0018- 9219. 14T. Fischbacher, M. Franchin, G. Bordignon, and H. Fangohr, IEEE Transactions on Magnetics 43, 2896 (2007). 15Z.-H. Wei and M.-F. Lai, Journal of Applied Physics 101, 09F515 (2007). 16A. N. Pasupathy, R. C. Bialczak, J. Martinek, J. E. Grose, L. A. K. Donev, P. L. McEuen, and D. C. Ralph, Science 306, 86 (2004). 17Z. K. Keane, L. H. Yu, and D. Natelson, Applied Physics Letters 88, 062514 (2006). 18K. Yoshida, I. Hamada, S. Sakata, A. Umeno, M. Tsukada, and K. Hirakawa, Nano Letters 13, 481 (2013), pMID: 23327475. 19K. I. Bolotin, F. Kuemmeth, A. N. Pasupathy, and D. C. Ralph, Nano Letters 6, 123 (2006), pMID: 16402799. 20S.-F. Shi, K. I. Bolotin, F. Kuemmeth, and D. C. Ralph, Phys. Rev. B 76, 184438 (2007). 

1502.07122

2

1502

Radiation comb generation with extended Josephson junctions

We propose the implementation of a Josephson radiation comb generator (JRCG) based on an extended Josephson junction subject to a time dependent magnetic field. The junction critical current shows known diffraction patterns and determines the position of the critical nodes when it vanishes. When the magnetic flux passes through one of such critical nodes, the superconducting phase must undergo a $\pi$-jump to minimize the Josephson energy. Correspondingly a voltage pulse is generated at the extremes of the junction. Under periodic driving this allows us to produce a comb-like voltage pulses sequence. In the frequency domain it is possible to generate up to hundreds of harmonics of the fundamental driving frequency, thus mimicking the frequency comb used in optics and metrology. We discuss several implementations through a rectangular, cylindrical and annular junction geometries, allowing us to generate different radiation spectra and to produce an output power up to $10$~pW at $50$~GHz for a driving frequency of $100$~MHz.

cond-mat.mes-hall

cond-mat

Radiation comb generation with extended Josephson junctions P. Solinas,1, ∗ R. Bosisio,1, 2, † and F. Giazotto2, ‡ 1SPIN-CNR, Via Dodecaneso 33, 16146 Genova, Italy 2NEST, Instituto Nanoscienze-CNR and Scuola Normale Superiore, I-56127 Pisa, Italy We propose the implementation of a Josephson radiation comb generator (JRCG) based on an extended Josephson junction subject to a time dependent magnetic field. The junction critical current shows known diffraction patterns and determines the position of the critical nodes when it vanishes. When the magnetic flux passes through one of such critical nodes, the superconducting phase must undergo a π-jump to minimize the Josephson energy. Correspondingly a voltage pulse is generated at the extremes of the junction. Under periodic driving this allows us to produce a comb-like voltage pulses sequence. In the frequency domain it is possible to generate up to hundreds of harmonics of the fundamental driving frequency, thus mimicking the frequency comb used in optics and metrology. We discuss several implementations through a rectangular, cylindrical and annular junction geometries, allowing us to generate different radiation spectra and to produce an output power up to 10 pW at 50 GHz for a driving frequency of 100 MHz. PACS numbers: 74.50.+r, 74.81.Fa, 06.20.fb, 04.40.Nr I. INTRODUCTION The field of optical combs has seen a growing interest in re- cent years1,2. The atomic clocks are extremely stable and have sharp resonances; for these reasons, they are used as time and frequency reference standards. However, their working range is limited to the radio frequency region. This limitation has been overtaken only a decade ago. A combination of techni- cal and conceptual improvements has allowed to extend this accuracy to higher frequency up to the optical region. The key phenomenon is simple: the atoms are manipulated to emit periodic sharp energy pulses which, in the frequency domain, correspond to a comb signal with the harmonics of the funda- mental frequency. Since the generated harmonics show very sharp resonances, they can be used as a frequency standard in the optical region. The realization of the optical frequency comb has paved the way to important applications in optical metrology3, high precision spectroscopy4,5 and telecommuni- cation technologies1,6. Recently, it has been shown that a similar frequency comb can be generated with a dc superconducting quantum interfer- ence device (SQUID) subject to a time-dependent magnetic field7. The driving induces π−jumps of the superconducting phase which are associated to voltage pulses generated at the extremes of the device. The voltage pulses sequence in the time domain translates, upon Fourier transforming, into a ra- diation comb in the frequency domain with up to hundreds of harmonics of the fundamental driving frequency. Here, we show how similar effect can be obtained in an ex- tended Josephson junction. The underlying physics is similar to that discussed in Ref. 7, but the details of the implementa- tion are different. A setup involving extended junctions opens up the possibility for different geometries and, therefore, for various power spectra of the emitted radiation. In particu- lar, since the generated radiation comb structure depends on the current-magnetic flux relation of the junction, this latter can be properly engineered in order to obtain a desired radia- tion power spectrum. We discuss the rectangular, cylindrical and annular junction designs with their different strengths and weaknesses as prototypical examples of extended junctions. FIG. 1. (Color online) (a) Cross section of an extended Josephson tunnel junction in the presence of an in-plane magnetic field H gen- erating a flux Φ piercing the junction in the x direction. t and λ represent the thickness and London penetration depth of the super- conductors S, respectively, whereas d is the insulator thickness. Pro- totypical junctions with rectangular, circular, and annular geometry are shown in (b)-(d). L, W , R and r are their geometrical parameters. The paper is structured as follows: in Section ?? we intro- duce the physical mechanism leading to the radiation genera- tion using extended Josephson junctions. The performance of such devices are investigated numerically in Section II, where the emitted radiation spectra of junctions with different ge- ometries are compared. Finally, we gather our conclusions in Section III. Our system is sketched in Fig. 1(a), and consists of an ex- tended Josephson tunnel junction composed of two identical superconducting electrodes8–12. We denote by λ and t the London penetration depth and the thickness of the supercon- ductors S, respectively, which satisfy the condition t > λ . Fur- thermore, d is the insulator thickness, whereas tH = 2λ + d is the magnetic penetration thickness. For the sake of clarity we focus on a junction in the short limit, i.e., with lateral dimensions much smaller than the Josephson penetration depth. In such a case the self-field gen- 5 1 0 2 p e S 5 1 ] l l a h - s e m . t a m - d n o c [ 2 v 2 2 1 7 0 . 2 0 5 1 : v i X r a (c) circular R (b) rectangular L (d) annular r S S t t y z H d (a) x y z H R Φ   insulator W IJ (Φ) erated by the Josephson current in the weak-link can be ne- glected with respect to the externally applied magnetic field and no traveling solitons are originated8,13. We choose a co- ordinate system such that the applied magnetic field (H), di- rected along the x direction, is parallel to a symmetry axis of the junction whose electrodes planes lies in the xy plane [see Fig. 1(a)]. For the sake of presentation, we focus on a rectangular junc- tion as in Fig. 1(b), keeping in mind that a similar discus- sion can be extended to junctions with different geometries [Figs. 1(c)-(d)]. In the limit of short junctions, the approx- imate behavior of the local phase is φ (y) = κy + ϕ where ϕ13,14 is the superconducting phase at the center of the junc- tion, κ = 2πΦ/(Φ0L) (Φ0 (cid:39) 2× 10−15 Wb is the flux quan- tum), L is the length of the junction whereas Φ = µ0HtHL is the magnetic flux through the junction and µ0 is the vacuum permeability. By integrating the Josephson current density per unit length13 Ic(y) over the junction length we obtain Irect J (Φ) = dy Ic(y)sinφ (y) = I+ f (t)sinϕ, (1) (cid:90) L/2 −L/2 (cid:90) J where f (t) = sinc(πΦ/Φ0), and I+ is the maximal critical current of the junction. The measurable critical current of the junction as a function of the magnetic flux is Irect (Φ) = maxϕIrect (Φ). It displays the celebrated Fraunhofer pattern which vanishes at the diffraction nodes appearing at Φ = nΦ0, where n is an integer [Fig. 2(a)]. The phase jump phenomenon we are interested in can be easily described in energetic terms. Assuming that there is no bias current, the energy associated to the Josephson current is c EJ(t) = J V (t)dt = −EJ0 f (t)cosϕ, Irect (2) where f (t) is the same defined above and EJ0 = Φ0I+/(2π). Notice that in the last step of Eq. (2) we have used the second Josephson relation ϕ = (2e/¯h)V (t). Initially f (t = 0) = 1 and the minima of the potential energy are found at ϕ = 2πk (with k integer). When the magnetic flux reaches the diffraction node at Φ = Φ0, EJ vanishes, becoming negative for Φ > Φ0. To remain in a minimum energy state, cosϕ must change sign, which implies that the superconducting phase must undergo a π jump. The original prediction of π jumps8 has also been indirectly confirmed via measurements in heat transport ex- periments performed in temperature-biased Josephson tunnel junctions15,16. Indeed, as it is explained in Ref. 8, the fact that the coherent component of the heat current remains posi- tive when crossing a diffraction node can be understood only if the superconducting phase undergoes a π-jump. Possible applications of this phenomenon for SQUID devices has been extensively discussed in Ref. 7. To determine the details of the voltage pulses, such as their shape and amplitude, the above discussed energetic picture is not sufficient. We rely on the so-called resistively and ca- pacitively shunted Josephson junction (RCSJ) model13,14 in which the Josephson junction is modelized as a circuit with a capacitor C, a resistor R, and a non-linear (Josephson) induc- tance LJ arranged in a parallel configuration. We consider a 2 sinusoidally-driven magnetic flux with frequency ν and am- plitude ε, centered in the first node of the interference pattern [Fig. 2(a)], so that Φ(t) = Φ0 2 [1− ε cos(2πνt)]. (3) integrated phase (cid:82) L/2 As a result, the magnetic flux crosses the nodes of the interfer- ence pattern at t = (2k + 1)/4ν, with k integer. Starting from the RCSJ model we can write an equation of motion for the −L/2 dyφ (y). Because of the symmetry of the problem, this reduces to a RCSJ equation for the phase at the center of the junction ϕ14: ¯hC 2e ϕ + ¯h 2eR ϕ − I+ f (t)sinϕ = IB. (4) We rescale the above equation in terms of adimensional time τ = 2πνt and, using ¯h/(2e) = Φ0/2π, we obtain7 c d2ϕ dτ2 + dϕ dτ − α[ f (τ)sinϕ − δ ] = 0, (5) where δ = IB/I+, c = 2πRCν and α = I+R/(Φ0ν). The bias current is supposed to be small (δ (cid:28) 1) and its effect is to impose a preferred direction to the π jumps of the phase. Fur- thermore, we focus on the limits c (cid:28) 1 (overdamped regime) and I+R (cid:29) 1, as these two conditions maximize the JRCG performance7. II. NUMERICAL RESULTS In the following, all the numerical simulations we discuss have been performed for rectangular Nb/AlOx/Nb Joseph- son junctions subject to an oscillating magnetic flux with fre- quency ν = 100 MHz and amplitude ε = 0.9. As the junction parameters we have assumed17 R = 20 Ohm and I+ = 0.2 mA, while the bias current has been set to IB = 10−3I+. The rela- tively low frequency we have chosen, besides allowing us to achieve an important up-conversion in frequency (see Fig. 4 and related discussion), also assures that the small capacitance of the junction C = 10 fF (corresponding roughly to a junction area of18 (cid:39) 0.2 µm2) has a negligible effect on its dynamics. As the critical current crosses the diffraction node at Φ = Φ0, the phase experiences a π jump and a voltage pulse is generated across the junction. The shape of the pulse is deter- mined by the product I+R: the larger I+R, the sharper the voltage pulse. Differently to what happens in the SQUID implementation7, for a rectangular junction the pulse ampli- tudes are not the same but show an alternating pattern of lower and higher peaks. This is due to the asymmetry in the diffrac- tion pattern of the critical current near the diffraction node [see Fig. 2(a)]. This real-time voltage comb could be used in several ways. One is as a generator of equally spaced voltage pulses to be used in electronics19. A second one is as a high precision con- troller. In fact, because of the Josephson relation, the time 3 Finally, the power P is calculated by integrating the PSD around the resonances kν (where ν is the monochromatic driving frequency) and dividing for a standard load resistance RL of 50 Ohm. This is the power we would measure at a given resonance frequency with a bandwidth exceeding the linewidth of the resonance. To increase the output power, we have considered a linear array of N identical junctions connected together via a super- conducting wire as done for the metrological standard for volt- age based on the Josephson effect7,22–24. The coupling among different junctions has been neglected: This condition can be realized in practice by a suitable design choice which reduces the cross capacitance and the inductance between neighbor junctions. In this case the current conservation through any i-th junction leads immediately to a set of decoupled RCSJ equations of the form (5)7. Therefore, the dynamics of the junctions are independent and the voltage at the extremes of the array is found by summing up the voltages of the sin- gle junctions. Under the hypothesis that all the junctions in the chain generate the same voltage V (t), the total voltage drop across the device is simply Vtot(t) = N V (t). Accord- ingly the intrinsic power, that is, the power delivered to an ideal load, would scale as N2. On the other hand, the extrin- sic power is less trivial and depends on the detection system used? . The N2-scaling of Josephson junctions performance is also a widely studied topic in the context of phase-locked Josephson junctions arrays25–27. In order to get an estimate of the device performance, in the following we have calculated the emitted power for arrays of N = 103 junctions by dividing the total voltage by a standard load resistance RL = 50 Ohm as mentioned above. Limitations to this simplified analysis can arise if we must take into account the effects of propagation of the emitted radiation along the chain7. In fact, in our model we have considered the device as a lumped element and this assump- tion breaks down as the length of the chain approaches the wavelength of the emitted radiation. As a quantitative es- timate, the minimum wavelength λmin (emitted at 50 GHz) must satisfy the relation 2λmin ≥ L where L is the total length of the device7. Considering a packing density of the junc- tions of 5 µm, the above condition is satisfied for N = 103 junctions. Despite being detrimental for the device perfor- mance, the propagation effects can be taken into account and corrected. With a careful design of the device, they could also be exploited to amplify the output power at specific work- ing frequencies. Another limiting factor may be an intrin- sic property of the device, such as the flux flow through the superconductor28. This could perturb the external magnetic flux that we use to induce the π-jumps of the phase. Closely related parasitic effects have also been studied recently29 for a similar system based on SQUIDs. Despite being beyond the scope of the present work, this remains an interesting issue that would require further investigation. Figure 4(a) shows the emitted radiation power spec- trum for a chain of N = 103 Nb/AlOx/Nb rectangular junctions14,17,30,31 driven by a 100 MHz oscillating magnetic field. As we can see, the device is able to provide a power of about 10 pW at 50 GHz (corresponding to the 500-th har- FIG. 2. (Color online) Diffraction pattern of the critical current for different geometries: (a) rectangular junction, (b) circular junction and (c) annular junction with different numbers n of trapped fluxons: n = 1 (solid blue line) and n = 2 (dashed green line). The red line shows Φ(t)/Φ0 oscillating with driving frequency ν and amplitude ε around different diffraction nodes. average of a voltage pulse is actually quantized as a conse- quence of the π jump of the phase: dtV (t) = ∆ϕ = π, (6) (cid:90) t2 t1 2e ¯h where t1 and t2 are the times in which the jump begins and ends, respectively. The phase jumps do not depend on the dy- namics or on the speed of the node crossing. The only condi- tion to be satisfied is the crossing of the diffraction node. This makes the pulse generation robust against imperfection in the dynamics of the junction and the driving. A possible applica- tion could be the high precise control of a quantum logic gate for superconducting based qubits20,21. The voltage pulse sequence in Fig. 3 has even more interesting applications as a radiation generator. In fact, in the frequency domain it corresponds to a frequency comb similar to the ones used in optics1. To test this possible implementation we have calculated the power spectrum P vs frequency Ω. We first compute the Fourier transform of the voltage (cid:90) T 0 V (Ω) = dteiΩtV (t). The power spectral density (PSD) is then PSD(Ω) = V (Ω)2. 1 T (7) (8) Rectangularjunction2\x{FFFF}0.0.51.Icrect\x{FFFF}\x{FFFF}\x{FFFF}\x{FFFF}I\x{FFFF}Circularjunction2\x{FFFF}0.0.51.Iccirc\x{FFFF}\x{FFFF}\x{FFFF}\x{FFFF}I\x{FFFF}2\x{FFFF}Annularjunction\x{FFFF}3\x{FFFF}2\x{FFFF}101230.0.51.\x{FFFF}\x{FFFF}\x{FFFF}0Icann\x{FFFF}\x{FFFF}\x{FFFF}\x{FFFF}I\x{FFFF}!"#$!%#$!&#$4 FIG. 4. (Color online) Power spectrum of the Josephson radiation comb generator over a 50 Ohm transmission line. The output power is shown in W (left) and dBm (decibel-milliwatt) (right). The cal- culation is performed for a N = 103 chain of Nb/AlOx/Nb extended Josephson junctions subject to a ν = 100 MHz driving. The cyan regions correspond to the insets. The panels refer to different ge- ometries: (a) rectangular junction, (b) cylindrical junction, (c) annu- lar junction with one fluxon trapped and (d) annular junction with two fluxons trapped. Blue and red points indicate the even and odd harmonics, respectively. In Figs. (a) and (c) only the even harmonics are present because of the comb-like shape of the voltage pulses. The junction parameters were set as those in Fig. 3. Particularly relevant is the annular junction case. Here, we have an additional controllable parameter: the number n of fluxons trapped in the junction. The most interesting situation is when there is one fluxon trapped (n = 1) [see Fig. 4(c)]. In this case, it is possible to modulate the magnetic flux near the vanishing point [see Fig. 2(c)] making the flux driving easier. In addition, the diffraction pattern is highly symmetric near Φ = 0. This allows one to generate a very precise voltage pulse patterns that is eventually reflected in a stronger power output at high frequency, as shown in Fig. 4(c) (0.1 nW at (Color online) Time dependence of the phase ϕ (dashed FIG. 3. line, left axis) and of the voltage V (solid line, right axis) across a rectangular Josephson junction. One-directional jumps can be re- alized by applying a suitable IB. The voltage pulses are generated when the interference node is crossed. The parameters chosen for the calculations are those typical of a Nb/AlOx/Nb junction17 with R = 20 Ohm, I+ = 0.2 mA, and we set ε = 0.9. The typical junction capacitance is about C = 10 fF and it has been neglected as it does not affect the junction dynamics. monic of the driving frequency). The implementation with extended junctions opens the way also for a geometric optimization. Choosing a different junc- tion geometry affects the critical current of the junction and, therefore, the position and the form of the nodes. For the cir- cular geometry the Josephson current exhibits the known Airy diffraction pattern, Icirc J (Φ) = I+ J1(πΦ/Φ0) (πΦ/2Φ0) sinϕ, (9) where J1(y) is the Bessel function of the first kind, Φ = 2µ0HRtH and R is the junction radius. For the annular junction32,33, the Josephson current takes the form (cid:90) 1 α Iann J (Φ) = I+ 2 1− α2 dxxJn(xπΦ/Φ0)sinϕ, (10) c J (Φ) = maxϕIann J (Φ) = maxϕIcirc where Φ = 2µ0HRtH, α = r/R, Jn(y) is the nth Bessel func- tion of integer order, R (r) is the external (internal) radius, and n = 0,1,2, ... is the number of trapped fluxons in the junction barrier. The critical currents for these two junc- tion geometries are defined as Icirc (Φ) and Iann (Φ), respectively, and they are shown in c Fig. 2 (b) and (c), respectively. We note that the position of the diffraction nodes is different from the rectangular geom- etry case: The critical currents vanish for non-integer values of the ratio Φ/Φ0. Moreover, in the annular case, we see that the slope of Iann (Φ) at Φ = 0 differs if the number of fluxons n changes. The parameter n can thus be seen as an additional degree of freedom which has important effects on the shape of the emitted radiation power spectrum [see Figs. 4(c) and (d)]. The driving is assumed to have the same periodic behavior, oscillating around the diffraction nodes as shown in Fig. 2. The power spectrum of the circular junction [Fig. 4(b)] is similar to the rectangular junction one. It generates smaller output power at high frequency reaching 2 pW at 50 GHz. c 05101501234040t
ns
(cid:106)
ΠV
ΜV
49.85050.2015.GHzP\x{FFFF}pW\x{FFFF}Ν\x{FFFF}100MHzRectangularjunction\x{FFFF}50.\x{FFFF}25.25.50.10\x{FFFF}11 10\x{FFFF}10 10\x{FFFF}9 10\x{FFFF}8\x{FFFF}500\x{FFFF}250250500\x{FFFF}80\x{FFFF}70\x{FFFF}60\x{FFFF}50\x{FFFF}\x{FFFF}GHz\x{FFFF}P\x{FFFF}W\x{FFFF}\x{FFFF}\x{FFFF}ΝdBm49.85050.202.GHzP\x{FFFF}pW\x{FFFF}Ν\x{FFFF}100MHzCircularjunction\x{FFFF}50.\x{FFFF}25.25.50.10\x{FFFF}11 10\x{FFFF}10 10\x{FFFF}9 10\x{FFFF}8\x{FFFF}500\x{FFFF}250250500\x{FFFF}80\x{FFFF}70\x{FFFF}60\x{FFFF}50\x{FFFF}\x{FFFF}GHz\x{FFFF}P\x{FFFF}W\x{FFFF}\x{FFFF}\x{FFFF}ΝdBm49.85050.200.1GHzP\x{FFFF}nW\x{FFFF}Ν\x{FFFF}100MHzAnnularjunctionn\x{FFFF}1\x{FFFF}50.\x{FFFF}25.25.50. 10\x{FFFF}10 10\x{FFFF}910\x{FFFF}8\x{FFFF}500\x{FFFF}250250500\x{FFFF}70\x{FFFF}60\x{FFFF}50\x{FFFF}\x{FFFF}GHz\x{FFFF}P\x{FFFF}W\x{FFFF}\x{FFFF}\x{FFFF}ΝdBm11.81212.200.4GHzP\x{FFFF}pW\x{FFFF}AnnularjunctionΝ\x{FFFF}100MHzn\x{FFFF}2\x{FFFF}12.512.510\x{FFFF}1310\x{FFFF}11 10\x{FFFF}9\x{FFFF}125125\x{FFFF}100\x{FFFF}80\x{FFFF}50\x{FFFF}\x{FFFF}GHz\x{FFFF}P\x{FFFF}W\x{FFFF}\x{FFFF}\x{FFFF}ΝdBm!"#$!%#$!&#$!'#$50 GHz). By varying the number of fluxons, the junction diffraction pattern changes [see Fig. 2(c)]: Correspondingly, the dynam- ics of the junction is different, generating different emitted radiation power spectra. Figure 4(d) shows the spectrum gen- erated by an annular junction chain when two fluxons are trapped in each junction. The overall power emitted is smaller and the signal is accessible up to ∼ 10 GHz. The spectral fea- tures are very different with respect to the single fluxon ones [Fig. 4(c)]. In particular, the lower harmonics (a few multiple of ν) are now suppressed while the output maximum arises around a few GHz. The insets in Fig. 4 represent the blow-up of the cyan re- gions in the main panels: Notice that since the voltage pulse signals are almost rectified (Fig. 3), the spectra contain pre- dominantly the even harmonics, the odd ones being orders of magnitude smaller. The main sources of error that can limit the device perfor- mance are the imprecisions in the fabrication process. Small differences in the geometry of the junctions, i.e., length L for the rectangular junctions and radii R and r for the circular and annular junctions, will produce off-sets in the fluxes and de- lays in the phase jumps. The voltages will still sum up but the total voltage pulse shape will be broadened by these effects. In the frequency domain, this corresponds to an additional cut- off at high frequency. Another potential detrimental factor is the correction to the dynamics due to the intrinsic junction capacitance. However, this effect can be accounted for, mini- mized or corrected by a proper device design. Furthermore, notice that our whole description is done by considering the effect of the time-dependent magnetic field only, assuming there is no induced electric field on the junc- tions. It is well known that oscillating magnetic field can gen- erate in turn electric fields which, especially at high frequency (∼GHz), may significantly alter the effect we discuss. How- ever, this problem can be overcome by embedding the junction chain inside a suitably designed cavity where the electric (TE) and magnetic (TM) modes are spatially separated34,35. The junction array configurations discussed above are suit- able for the use as radiation emitters up to 50 GHz. The most straightforward way to detect the power in this frequency range is to couple the device to a transmission line and to feed 5 the signal to a commercial spectrum analyzer. To have access to higher frequency we must use different materials (for ex- ample, YBCO as discussed in Ref. 7) or adopt specific chain design. This change must be accompanied with a new de- tection schemes, for example, by using antennas coupled to the device electrodes7. Finally, in light of possible implemen- tations, besides the Nb/AlOx/Nb junctions considered in the present work, we signal that other materials could be promis- ing candidates. For instance Nb/HfTi/Nb junctions36,37, be- ing SNS-like (superconductor-normal metal-superconductor), would have the advantage of having almost negligible capaci- tance, despite having a slightly lower I+R product. III. CONCLUSIONS In summary, we have discussed the possibility to realize a Josephson radiation comb generator with extended Josephson junctions driven by a time-dependent magnetic field. With a linear array of N = 103 Nb/AlOx/Nb junctions and a driving frequency of 100 MHz, we estimate that substantial power [up to ∼100 pW] can be generated at 50 GHz (500-th harmon- ics), opening the way to a number of applications. The de- vice has room for optimization by modeling the geometry of the single junctions, the fabrication materials (see, for exam- ple, Ref. 7), the driving signal and the array design. The dis- cussed implementation would have the advantage to be built on-chip and integrated in low-temperature superconducting microwave electronics. IV. ACKNOWLEDGMENTS Stimulating discussions with C. Altimiras, S. Gasparinetti and D. Golubev are gratefully acknowledged. P.S. has re- ceived funding from the European Union FP7/2007-2013 un- der REA Grant agreement No. 630925 – COHEAT and from MIUR-FIRB2013 – Project Coca (Grant No. RBFR1379UX). The work of R.B. has been supported by MIUR-FIRB2013 – Project Coca (Grant No. RBFR1379UX). F.G. acknowl- edges the European Research Council under the European Union's Seventh Framework Program (FP7/2007-2013)/ERC Grant agreement No. 615187-COMANCHE for partial finan- cial support. 2 P. Del'Haye, A. Schliesser, O. Arcizet, T. Wilken, R. Holzwarth 88, 094506 (2013). ∗ paolo.solinas@spin.cnr.it † riccardo.bosisio@nano.cnr.it ‡ giazotto@sns.it 1 T. Udem, R. Holzwarth, and T. W. Hansch, Nature 416, 233 (2002). and T. J. Kippenberg, Nature 450, 1214 (2007). 3 T. Hansch and H. Walther, Rev. Mod. Phys. 71, S242 (1999). 4 N. Bloembergen, Nonlinear spectroscopy, vol. 64 (North Hol- land, 1977). 5 T. W. Hansch and M. Inguscio, Frontiers in Laser Spectroscopy: Varenna on Lake Como, Villa Monastero, 23 June-3 July 1992, vol. 120 (North Holland, 1994). 6 S. M. Foreman, K. W. Holman, D. D. Hudson, D. J. Jones, and J. Ye, Rev. Sci. Instrum. 78, 021101 (2007). 7 P. Solinas, S. Gasparinetti, D. Golubev, and F. Giazotto, Sci. Rep. 5, 12260 (2015). 8 F. Giazotto, M. J. Mart´ınez-P´erez, and P. Solinas, Phys. Rev. B 9 M. J. Mart´ınez-P´erez and F. Giazotto, Appl. Phys. Lett. 102, 10 M. J. Mart´ınez-P´erez and F. Giazotto, Appl. Phys. Lett. 102, 11 D. Bolmatov and C.-Y. Mou, Physica B: Condensed Matter 405, 182602 (2013). 092602 (2013). 2896 (2010). 12 G. J. Carty and D. P. Hampshire, Supercond. Sci. Technol. 26, 25 A. K. Jain, K. K. Likharev, J. E. Lukens, and J. E. Sauvageau, 13 M. Tinkham, Introduction to superconductivity (Courier Dover 26 P. Barbara, A. B. Cawthorne, S. V. Shitov, and C. J. Lobb, Phys. 065007 (2013). Publications, 2012). 14 R. Gross and A. Marx, Lecture on "Applied Superconductivity", http://www.wmi.badw.de/teaching/Lecturenotes/. 15 F. Giazotto and M. J. Mart´ınez-P´erez, Nature 492, 401(2012). 16 M. J. Mart´ınez-P´erez and F. Giazotto, Nat. Commun. 5 (2014). 17 V. Patel and J. Lukens, IEEE Trans. Appl. Supercond. 9, 3247 18 K. K. Likharev, Dynamics of Josephson Junctions and Circuits (Gordon and Breach, New York, 1986). 19 J. Dubois, T. Jullien, F. Portier, P. Roche, A. Cavanna, Y. Jin, W. Wegscheider, P. Roulleau, and D. C. Glattli, Nature 502, 659 (2013). 20 Y. Nakamura, Yu. A.. Pashkin, and J. S. Tsai, Nature 398, 786 21 Y. Makhlin, G. Schon, and A. Shnirman, Rev. Mod. Phys. 73, 357 22 S. Shapiro, Phys. Rev. Lett. 11, 80 (1963). 23 R. Kautz and F. L. Lloyd, Appl. Phys. Lett. 51, 2043 (1987). 24 J.-S. Tsai, A. K. Jain, and J. E. Lukens, Phys. Rev. Lett. 51, 316 (1999). (1999). (2001). (1983). 6 Physics Reports 109, 309 (1984). Rev. Lett. 82, 1963 (1999). 27 L. Ozyuzer, A. E. Koshelev, C. Kurter, N. Gopalsami, Q. Li, M. Tachiki, K. Kadowaki, T. Yamamoto H. Minami, H. Yamaguchi, T. Tachiki, K. E. Gray, W.-K. Kwok, and U. Welp Science 318, 1291 (2007). 28 J. Vanacken, L. Trappeniers, K. Rosseel, I. N. Goncharov, V. V. Moshchalkov, and Y. Bruynseraede Physica C 332, 411 (2000). 29 R. Bosisio, F. Giazotto, and P. Solinas, arXiv:1505.06333 (2015). 30 F. L. Lloyd, C. A. Hamilton, J. Beall, D. Go, R. Ono, and R. E. Harris (1987). 31 R. Popel, J. Niemeyer, R. Fromknecht, W. Meier, and L. Grimm, J. Appl. Phys. 68, 4294 (1990). 32 N. Martucciello and R. Monaco, Phys. Rev. B 53, 3471 (1996). 33 C. Nappi, Phys. Rev. B 55, 82 (1997). 34 G. R. Eaton, S. S. Eaton, D. P. Barr, and R. T. Weber, Quantitative EPR (Springer, 2010). 35 M. Goryachev and M. E. Tobar, New J. Phys. 17, 023003 (2015). 36 D. Hagedorn, R. Dolata, F.-Im. Buchholz, and J. Niemeyer, Phys- ica C 372-376, 7 (2002). 37 R. Wolbing, J. Nagel, T. Schwarz, O. Kieler, T. Weimann, J. Kohlmann, A. B. Zorin, M. Kemmler, R. Kleiner, and D. Koelle, Appl. Phys. Lett. 102, 192601 (2013). 

1203.6382

1

1203

Surface States of Topological Insulators

We develop an effective bulk model with a topological boundary condition to study the surface states of topological insulators. We find that the Dirac point energy, the band curvature and the spin texture of surface states are crystal face-dependent. For a given face on a sphere, the Dirac point energy is determined by the bulk physics that breaks p-h symmetry in the surface normal direction and is tunable by surface potentials that preserve T symmetry. Constant energy contours near the Dirac point are ellipses with spin textures that are helical on the S/N pole, collapsed to one dimension on any side face, and tilted out-of-plane otherwise. Our findings identify a route to engineering the Dirac point physics on the surfaces of real materials.

cond-mat.mes-hall

cond-mat

Surface States of Topological Insulators Fan Zhang,∗ C. L. Kane, and E. J. Mele Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA (Dated: December 22, 2013) We develop an effective bulk model with a topological boundary condition to study the surface states of topological insulators. We find that the Dirac point energy, the band curvature and the spin texture of surface states are crystal face-dependent. For a given face on a sphere, the Dirac point energy is determined by the bulk physics that breaks p-h symmetry in the surface normal direction and is tunable by surface potentials that preserve T symmetry. Constant energy contours near the Dirac point are ellipses with spin textures that are helical on the S/N pole, collapsed to one dimension on any side face, and tilted out-of-plane otherwise. Our findings identify a route to engineering the Dirac point physics on the surfaces of real materials. PACS numbers: 71.70.Ej, 73.20.-r, 73.22.Dj Introduction.—The discovery[1–6] of topological insu- lators (TI’s) and the synthesis[7–10] of three dimensional materials that realize their physics has opened up a new field in solid state physics. Particular interest has focused on the TI surface states with point degeneracies that are topologically protected by time reversal symmetry (T ) when a trivial insulator with topological index[1, 2] Z2 = 1 is joined to a TI with Z2 = −1. On the cleav- age surface of Bi2Se3 , a TI with a single Dirac cone has been identified[9–15] by angle resolved photoemission (ARPES) and scanning tunneling microscope (STM) ex- periments. The surface states form a spin-momentum locked helical metal with conduction and valence bands exhibiting opposite helicities near the Dirac point, and develop hexagonal warping[10, 15–17] on top of the cir- cular constant energy contour away from the Dirac point. Interest in the topological surface bands has led to the development of various bulk continuum models with the surface termination described by ad-hoc fixed node [8, 17–20] or by “natural” boundary conditions [21], where the existence of a topological surface state and its Dirac point energy are treated as inputs to the theories. The choice of boundary condition has profound conse- quences for the spatial form of the surface state wave- functions and their interactions with external fields and absorbed species. In this work we derive the appropriate topological boundary condition (TBC) at the surface from the matching of bulk waves to evanescent vacuum states. As expected a topologically protected Dirac point occurs in this model but its energy position and band curvature are both crystal face-dependent owing to bulk terms that break particle-hole (p-h) symmetry. Furthermore, the in- teraction with surface-localized potentials that preserve T symmetry shifts the energy position of the Dirac de- generacy and provides a robust degree of freedom for tun- ing the surface state spectrum into a convenient energy range to exploit their topological properties. The associ- ated spin texture is determined by the bulk symmetries and depends on the crystal face angle. Unlike the helical metal on a cleavage surface, the spin texture near Dirac point is compressed to a single direction on a side face, and is tilted out-of-plane otherwise. We start from a description of the low-energy model of Bi2Se3 , which applies generally to other TI’s with the same crystal structure with space group R3 ¯m. At the origin of Bi2Se3 Brillouin zone Γ, the effective Hilbert space near the bulk gap is spanned by states with angular 2 and parity P = ±1. Because of the momentum mj = ± 1 spin-orbit coupling (SOC), pz ↑(cid:105) and pz ↓(cid:105) states are mixed with p+ ↓(cid:105) and p− ↑(cid:105) states, respectively. Since the crystal-field splitting is much larger than the SOC, pz orbitals dominate and mj pseudospin is proportional to the electron spin. The P = ±1 hybridized states can be labelled approximately by +(cid:105) state from Bi atoms while −(cid:105) from Se, due to the large energy difference between 4p (Se) and 6p (Bi) principal quantum levels. Besides T and the parity inversion (P ) symmetries, Bi2Se3 crystal structure has threefold rotational (C3 ) symmetry along the z perpendicular to the quintuple layers, and twofold rotational (C2 ) symmetry along ΓM direction. By convention we choose the parity opera- tor P = τz and the time reversal operator T = iK σy where K is the complex conjugate operation. Therefore, to quadratic order in k the k · p bulk Hamiltonian that preserves the above four symmetries has a unique form H = (c0 + czk2 z + c(cid:113)k2(cid:113) ) + (−m0 + mzk2 z + m(cid:113)k2(cid:113) )τz + vzkz τy + v(cid:113) (ky σx − kxσy )τx , (1) where we assume mz , m(cid:113) , vz , v(cid:113) > 0 and  = 1 hereafter. The first parentheses is a scalar term that preserves T and P symmetries but breaks the p-h symmetry. The quadratic scalar terms intrinsically give rise to the cur- vature of Dirac surface bands and the rigid shift of Dirac point from the middle of the bulk gap, as we will demon- strate in the following. Eq.(1) has both T and P sym- metries with a topological index Z2 = sgn(−m0 ). Topological boundary condition.— To investigate the surface states, we use a TBC in which the mass term changes sign across the surface. On the TI side, m0 = m is positive and half of the bulk gap at k = 0, and c0 = 0 to define the middle of the k = 0 gap as energy zero. On the vacuum side, i.e., a trivial insulator with an infinite 2 (5) boundary problem can be generalized to any surface with τ replaced by S1 and σ by S2 . For an arbitrary crystal face Σ(θ) defined in Fig.1, the algebraic structure is S1 = {ατx + βσy τy , ατy − βσy τx , τz } , where v3 = (cid:112)(vz cos θ)2 + (v(cid:113) sin θ)2 , α = vz cos θ/v3 S2 = {ασx − βσz τz , σy , ασz + βσx τz} , (4) and β = v(cid:113) sin θ/v3 . These new pseudospins satisfy a , S j [S i b ] = 2iδab
ijkS k a . Rewritten in this new pseudospin basis, Eq.(1) reads H = − m0 S z 1 + (v3k3 + v0k1 ) S y 1 2 − v1k1 S y + (v(cid:113)ky S x 2 ) S x 1 , where v0 = (v2(cid:113) − v2 z ) sin θ cos θ/v3 and v1 = vz v(cid:113)/v3 . By matching the eigensystems of TI and vacuum sides, we obtain the Σ(θ) surface states similar to Eq.(3) where τ and σ are replaced by S1 and S2 , respectively. Note that κ = m/v3 + iv0k1/v3 in general. k1 coupling to S y 1 neither influences the energy spectrum nor the spin texture, only making the evanescent states oscillate and giving a phase accumulation along k1 away from Dirac point. Therefore, ignoring the quadratic corrections, we first derive the effective surface state Hamiltonian for an arbitrary face Σ(θ) to the linear order, 2 − v1k1 S y H(1) (θ) = v(cid:113)ky S x 2 , from which we can further explicitly demonstrate the sur- face state spin texture on Σ(θ): (cid:113) (cid:104)σx (cid:105)θ = ± vz v(cid:113)ky cos θ v2 1 + v2(cid:113) k2 1 k2 v3 (cid:113) y −vz v(cid:113)k1 1 + v2(cid:113) k2 v2 1 k2 y v3 (cid:104)σz (cid:105)θ = 0 , (7) where + (−) denotes the conduction (valence) band. The electron real spin (cid:104)s(cid:105) is proportional to but always smaller than (cid:104)σ(cid:105) due to the SOC[23]. In the local coordi- nates, (cid:104)σ1 (cid:105)θ = (cid:104)σx (cid:105)θ cos θ and (cid:104)σ3 (cid:105)θ = (cid:104)σx (cid:105)θ sin θ. Eq.(7) indicates that the surface state spin texture is rather dif- ferent from face to face while its pseudospin (S2 ) has a universal structure on the elliptic constant energy con- tour near the Dirac point. Clearly, there is no spherical symmetry to guarantee that the spin and orbital struc- tures are the same anywhere on a TI sphere. The crystal face-dependent spin texture is helical on the south and north poles, is compressed to a single dimension along the equator, and is tilted out-of-plane otherwise. The surface state anisotropy and spin textures for different faces are compared in Fig.2. The surface state spin textures on the poles and the equator of a TI sphere can be understood by symme- tries. For (¯100) face, the C2 symmetry along ΓM ( x) di- rection forbids σx coupling to any in-plane momentum. (cid:104)σy (cid:105)θ = ± (6) , , FIG. 1: (Color online) The definition of an arbitrary face of a B i2S e3 -type topological insulator. The south (S) pole de- notes the (00¯1) surface parallel to the quintuple layers and the equator represents the side faces perpendicular to the quintu- ple layers. For any face Σ(θ), k3 ⊥ Σ and k2 = ky . gap m0 = −M where M → +∞, both cz and c(cid:113) vanish since the vacuum has p-h symmetry around c0 . We first focus on the one dimensional quantum me- chanics of (00¯1) surface turning off the quadratic terms in Eq.(1). Notice that the two spin flavors are decoupled for this eigenvalue problem. The spin ↑ states in {τz , σ} representation are determined by the following coupled (cid:18) −m(z ) − E −vz∂z (cid:19) (cid:18) ψ1 (cid:19) Dirac equations m(z ) − E vz∂z ψ2 The wave function ψ↑ = (ψ1 , ψ2 )(cid:48) is continuous across the surface, integrating Eq.(2) over the vicinity of the surface. This continuity condition leads to a nontrivial solution which is isolated in the middle of the bulk gap and is localized on the surface. This midgap (E = 0) (cid:18) 1 (cid:19) (cid:18) 1 (cid:19) state has a simple and elegant exact solution, (cid:40) 1 0 0 A τx e−κz , eκ0 z , ψk(cid:113) ,↑ (x, y , z ) = ei(kx x+ky y)φ(z ) (TI) (Vac) z > 0 z < 0 φ(z ) = = 0 . σ , ⊗ (2) , (3) where κ = m/vz , κ0 = M /vz and A is a normalization factor. φ(z ) is evanescent on both sides of the surface where the mass m0 changes sign, analogous to Jackiw and Rebbi solution[22] of a two-band Dirac model. The spin ↓ solution can be obtained by ψ↓ = T ψ↑ . This isolated midgap state at k(cid:113) = 0 is identified as the Dirac point and at finite k(cid:113) spreads into a perfect Dirac cone- the ideal topologically protected surface bands. Surface state spin texture.— On the (00¯1) surface, the midgap solution of the boundary problem at k(cid:113) = 0 is determined by the operators τ and is free under any ro- tation of the operators σ . This τ ⊗ σ algebra of the kxk3kzk1(cid:2016)N (001)E (100)E (100)S (001)Σ(cid:4666)(cid:2016)(cid:4667)OIn linear order C3 symmetry upgrades to continuous ro- tational symmetry, consequently, the surface normal spin is zero along the TI equator. On the two poles, the mir- ror symmetry and the C3 symmetry insure that the spin- momentum locking into the form of ky σx − kxσy . σz decouples to any momentum in the linear order, taking into account all the four symmetries. It’s interesting to point out that the surface band is the positive eigenstate of S x 1 and the chiral counterpart is separated and localized on the opposite face. Thus the surface state Hilbert space is reduced by half and this pseudospin polarity (or chirality) blocks the back scattering on the surface as a result of interplay between T symmetry and band inversion physics with TBC. Dirac point energy and surface potentials.— The ab- sence of spherical symmetry in the bulk requires that the Dirac point has different energies on different crystal faces. When the quadratic mass terms and p-h symmetry breaking terms in Eq.(1) are turned on, the surface state wave function and spectrum are changed. In the coupled Schrodinger-type Eq.(1), the components of wave func- tion and their slopes are all continuous across the bound- ary. We find that the effective Hamiltonian for face Σ(θ) have two important[24] corrections: H(2) (θ) = c(cid:113)k2 y + (cz sin2 θ + c(cid:113) cos2 θ)k2 1 , cz cos2 θ + c(cid:113) sin2 θ · m . HDP (θ) = mz cos2 θ + m(cid:113) sin2 θ (8) (9) The breaking p-h symmetry terms give the surface Dirac cone a parabolic curvature described by Eq.(8) and shifts the Dirac point from the midgap to a nonzero energy given by Eq.(9). These two effects help to explain the origin of nonzero Dirac point energies and nonlinear Dirac cones of cleavage surface states observed in ARPES experiments[9, 10, 25–27]. Although the surface state solution obtained by Eq.(2) with TBC remains topologically stable, it is essential to understand how localized surface potentials influence the surface spectra and the associated wave functions as well. We focus on potentials that preserve T symmetry. Among these six types: I , τz , τx and (cid:126)στy list in Table I, τx and (cid:126)στy break P symmetry. It turns out that the same potential could play different roles on different crystal faces, as shown in Table I. On the south pole of a TI sphere, a potential ∆0 δ(z ) · 2vz /m·I changes the wave function continuity conditions, integrating Eq.(2) including surface potentials over the vicinity of z = 0. The amplitudes of ψ1 and ψ2 are either weakened or enhanced across the surface with changes are always opposite to each other. Consequently, to match the evanescent solutions on the vacuum and the mat- ter sides, the surface band energies are rigidly shifted by δEDP . τx potentials have similar effects to the I -type although τx breaks P symmetry. For a τz potential, it simply modifies the mass term on the surface and it is not surprising that it does not affect the surface spectra. 3 FIG. 2: (Color online) Dirac cones and spin textures of sur- face states on faces with (a) θ = 0, (b) θ = π/4 and (c) θ = π/2. The equal-energy contours are from −80 meV to 160 meV with 10 meV increment relative to the Dirac point. Surface band curvatures are taken into account. All three panels are in the same scale and the parameters are adopted from DFT results. The lower panels only show the spin tex- tures at 160 meV in the k1 − ky planes where ky is the vertical axis. A τz potential tunes the amplitudes of ψ1 and ψ2 in the same manner on the matter side but this gives no observ- able effect since ψ1 (z ) = ψ2 (z ) and ψ1,2 (z < 0) → 0 are still valid. Similarly, a potential like σn τy (∆nσn ≡ (cid:126)∆ · (cid:126)σ) does nothing to the surface spectra and ψ1/ψ2 ; however, TABLE I: Summary of the influence of T symmetry allowed momentum-independent surface potentials ∆δ(r3 ) · 2v3 /m on the inversion symmetry, and the wave function continuity and the Dirac point energy of surface states. This ∆ represents different surface potentials and their corresponding energy 2 S y 1 and ∆nS n scales: ∆0 I , ∆m τz , ∆ES x 1 . For the σn τy col- umn, only the results for S n 2 = 1 state is shown and their 2 = −1 state. complex conjugates represent the results for S n Σ(0) Σ(θ) P δEDP ψ1 (0+ ) ψ2 (0+ ) ψ1 (0+ ) ψ1 (0− ) ψ2 (0+ ) ψ2 (0− ) I I + 4m2∆0 (m2−∆2 0 ) (m2−∆2 0 )2+4m2∆2 0 m2−2m∆0−∆2 0 m2+2m∆0−∆2 0 m2−2m∆0−∆2 0 m2+∆2 0 m2+2m∆0−∆2 0 m2+∆2 0 τz τz + 0 1 m+∆m m−∆m m+∆m m−∆m τx S x 1 − (cid:0) m−∆E (cid:1)2 4m2∆E (m2+∆2 E ) E )2+4m2∆2 (m2+∆2 E m+∆E m−∆E m+∆E m+∆E m−∆E σn τy 2 S y S n 1 − 0 1 m−i∆n m+i∆n m−i∆n m+i∆n
a
Θ(cid:61)0Face
b
Θ(cid:61)Π4Face
c
Θ(cid:61)Π2Faceit couples the two spin flavors and shifts their phase in an opposite way. Note that ψ1 = ψ2 is always true on the vacuum side since M → ∞. The above results are summarized in Table I, providing sufficient information to construct the surface state wave function and to engi- neer the Dirac point energy position. On an arbitrary face Σ(θ), the types of surface poten- tials are the same but their combinations and the corre- sponding roles are rearranged. This can be fully under- stood by the fact that the spin-orbital structure τ ⊗ σ on the south pole is replaced by a pseudospin-pseudospin structure S1 (θ) ⊗ S2 (θ) on Σ(θ). Note that the combi- nation only occurs for potentials with the same parity. Although the surface states solutions are stable in the presence of localized surface potentials that preserve T symmetry, two terms play an essential role in determining the energy position of Dirac point 4m2 (∆0 + ∆E )(m2 − ∆2 0 + ∆2 EDP = HDP + E ) 4m2 (∆0 + ∆E )2 + (m2 − ∆2 E )2 , (10) 0 + ∆2 which implies that I and τx potentials (Table I) are able to tune the Dirac point from the midgap to the band edges ±m independently. Discussions.— The interactions of the topologically protected bands with surface potentials provides a ro- bust route to engineer and manipulate the topological surface states. In particular, an external surface poten- tial can raise or lower the Dirac point to the middle of the bulk gap and providing experimental access to the topo- logically protected band. This goal could be achieved by surface oxidation[28], or by other possible chemical processes[29–31] and interactions[32] on the surface. We also point out that the electrostatic gating[25, 31, 33– 35], Eeff (cid:104)r3 (cid:105)sf can act to Stark shift the TI surface state into the gap, with field strength determined by the pen- etration length of surface states. Typically, the screened field ∼ 100 meV · nm−1 are required for the surface states evanescent in a couple of quintuple layers. Our present results provide a framework to study the self-consistent band bending physics of real TI materials[9, 10, 25–27] and mean-field models of surface state many-body inter- actions. Our model is quite different from the fixed bound- ary condition (FBC) which arbitrarily clamps the sur- face state wavefunction to zero at the boundary. The FBC solution is insensitive to the mass inversion at the surface which topologically protects the surface bands, and consequently it provides no information about the energy of the symmetry protected degeneracy relative to the bulk bands or their interaction with surface-localized potentials. Furthermore FBC admits an infinite num- ber of (physically spurious) solutions with nonzero ener- gies for k1 = k2 = 0 that satisfy an (incorrect) surface boundary condition. By contrast TBC guarantees that there is only one isolated solution, i.e., the Dirac point of surface bands protected by the change of bulk topology. Moreover, TBC demonstrates that the energy position of 4 Dirac point is tunable in the bulk gap via the symmetry allowed scalar terms and surface potentials. On a non cleavage surface, dangling bonds and their reconstruction may add complexity to the surface state spectrum which can be modeled as surface potentials en- coded in the parameters ∆0 and ∆E . The fingerprint of the novel spin texture near the Dirac point on non cleavage surfaces are determined by the bulk symmetries along with the topological stability of the surface spec- trum and may be accessible in ARPES and STM ex- periments. On an equatorial face, Zeeman coupling or magnetic disorder coupled to the spin degree of freedom do not generally open a gap, and the diamagnetic suscep- tibility is anticipated to be unusually anisotropic. More spectacularly, there is intrinsic charge redistribution in the surface bands near the corners of a TI that connect different crystal faces with intrinsically different Dirac point energies. Acknow ledgements.—This work has been supported by DARPA under grant SPAWAR N66001-11-1-4110. ∗ Electronic address: zhf@sas.upenn.edu [1] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [2] L. Fu, C. L. Kane and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). [3] J. E. Moore and L. Balents, Phys. Rev. B 75, 121306 (2007). [4] R. Roy, Phys. Rev. B 79, 195322 (2009). [5] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [6] X. Qi and S. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [7] D. Hsieh et al., Nature (London) 452, 970 (2008); Science 323, 919 (2009). [8] H. Zhang et al., Nature Phys. 5, 438 (2009). [9] Y. Xia et al., Nature Phys. 5, 398 (2009). [10] Y. L. Chen et al., Science 325, 178 (2009). [11] P. Roushan et al., Nature 460, 1106 (2009). [12] J. Seo et al., Nature 466,343 (2010). [13] P. Cheng et al. Phys. Rev. Lett. 105, 076801 (2010). [14] H. Beidenkopf et al, Nature Phys. 7,939 (2011). [15] Z. Alpichshev et al, Phys. Rev. Lett. 104, 016401 (2010). [16] L. Fu, Phys. Rev. Lett. 103, 266801 (2009). [17] C. Liu, Phys. Rev. B 82, 045122 (2010). [18] J. Linder, T. Yokoyama and A. Sudbo, Phys. Rev. B 80, 205401 (2009). [19] H. Lu, W. Shan, W. Yao, Q. Niu and S. Shen, Phys. Rev. B 81, 115407 (2010). [20] B. Zhou, H. Lu, R. Chu, S. Shen and Q. Niu, Phys. Rev. Lett. 101, 246807 (2008). [21] A. Medhi, V. B. Shenoy, arXiv:1202.3863 (2012). [22] R. Jackiw and C. Rebbi, Phys. Rev. D 13, 3398 (1976). [23] O. V. Yazyev, J. E. Moore and S. G. Louie, Phys. Rev. Lett. 105, 266806 (2010). [24] The crossing terms ∝ k1 k3 sin 2θ have negligible effects and thus are ignored for simplicity. Eq.(9) is an exact result at Γ point while m gets reduced by the m1 k2 1 and m(cid:113) k2 y terms away from k = 0. [25] D. Kong et al., Nature Nano. 6, 705 (2011). [26] J. Zhang et al., Nature Comms. 2, 574 (2011). [27] T. Arakane et al., Nature Comms. 3, 636 (2012). [28] X. Wang, G. Bian, T. Miller and T. Chiang, Phys. Rev. Lett. (2012) [29] D. Hsieh et al., Nature 460, 1101 (2009). [30] M. Bianchi et al., Phys. Rev. Lett. 107, 086802 (2011). [31] P. King et al., Phys. Rev. Lett. 107, 096802 (2011). [32] O. V. Yazyev, E. Kioupakis, J. E. Moore and S. G. Louie, arXiv:1108.2088 (2011). [33] H. Steinberg et al., Nano Lett. 10, 5032 (2010). [34] J. G. Checkelsky et al, Phys. Rev. Lett. 106, 196801 (2011). [35] H. Yuan et al., Nano Lett. 11, 2601 (2011). 5 

1906.06395

1

1906

Scattering theory of the screened Casimir interaction in electrolytes

We apply the scattering approach to the Casimir interaction between two dielectric half-spaces separated by an electrolyte solution. We take the nonlocal electromagnetic response of the intervening medium into account, which results from the presence of movable ions in solution. In addition to the usual transverse modes, we consider longitudinal channels and their coupling by reflection at the surface of the local dielectric. The Casimir interaction energy is calculated from the matrix describing a round-trip of coupled transverse and longitudinal waves between the interacting surfaces. The nonzero-frequency contributions are approximately unaffected by the presence of ions. We find, at zero frequency, a contribution from longitudinal channels, which is screened over a distance of the order of the Debye length, alongside an unscreened term arising from transverse-magnetic modes. The latter defines the long-distance asymptotic limit for the interaction.

cond-mat.mes-hall

cond-mat

Scattering theory of the screened Casimir interaction in electrolytes P. A. Maia Neto,1 F. S. S. Rosa,1 L. B. Pires,1 A. B. Marim,1 A. Canaguier-Durand,2 R. Gu´erout,2 A. Lambrecht,2 and S. Reynaud2 1Instituto de F´ısica, Universidade Federal do Rio de Janeiro, CP 68528, Rio de Janeiro RJ 21941-909, Brazil 2Laboratoire Kastler Brossel, Sorbonne Universit´e, CNRS, ENS-PSL Universit´e, Coll`ege de France, Campus Pierre et Marie Curie, F-75252 Paris, France (Dated: June 18, 2019) We apply the scattering approach to the Casimir interaction between two dielectric half-spaces separated by an electrolyte solution. We take the nonlocal electromagnetic response of the interven- ing medium into account, which results from the presence of movable ions in solution. In addition to the usual transverse modes, we consider longitudinal channels and their coupling by reflection at the surface of the local dielectric. The Casimir interaction energy is calculated from the matrix de- scribing a round-trip of coupled transverse and longitudinal waves between the interacting surfaces. The nonzero-frequency contributions are approximately unaffected by the presence of ions. We find, at zero frequency, a contribution from longitudinal channels, which is screened over a distance of the order of the Debye length, alongside an unscreened term arising from transverse-magnetic modes. The latter defines the long-distance asymptotic limit for the interaction. I. INTRODUCTION context of metals [28] and electrolytes [21]. Over the last decade, the scattering approach [1, 2] to the Casimir effect [3] has allowed for the derivation of exact results for a number of non-trivial geometries, including the ones most often investigated experimen- tally: the plane-sphere [4 -- 11] and sphere-sphere [12 -- 14]. Within the scattering approach, the Casimir effect arises from the recurrent multiple scattering of electromagnetic fluctuations between the interacting surfaces [15]. In the particular case of two homogenous half-spaces sep- arated by a layer of empty space or of a third homo- geneous medium, one recovers the standard Lifshitz [16] and Dzyaloshinskii-Lifshitz-Pitaevskii (DLP) [17] results, respectively. The unretarded van der Waals interaction is obtained in the limit of short distances as a particular case [18]. Applications of the scattering approach in colloid sci- ences and biophysics requires the inclusion of the screen- ing caused by ions dissolved in a polar liquid (water in most cases). As in the double-layer interaction between prescribed charged surfaces [19, 20], movable ions could indeed be expected to screen slowly fluctuating charges. Alternatively, in the language of the scattering approach, the electrolyte solution displays a nonlocal electric re- sponse (spatial dispersion) allowing for the existence of longitudinal modes [21] in addition to the standard trans- verse ones. The complementary case in which the intervening medium is local, whereas the interacting half-spaces are nonlocal, has been extensively analyzed in connection with the anomalous skin depth effect in metals. Indeed, free electrons exhibit a nonlocal response that modifies the Casimir interaction between metallic plates [22 -- 26]. The nonlocal response of metals has been recently con- sidered in connection with quantum friction [27]. The unretarded van der Waals interaction between nonlocal half-spaces across a local medium has been derived in the In the case of electrolytes, the common view is that only the Matsubara zero-frequency contribution as given by DLP result [17] is modified by screening (see for in- stance [29] for a recent review), since the plasma fre- quency associated to the presence of ions is always much smaller than kBT /¯h (where T is the temperature). In other words, fluctuations at all nonzero Matsubara fre- quencies are too fast to be screened by ions in solu- tion. The zero-frequency contribution is then usually considered apart from the nonzero Matsubara terms, and results are derived from the linear Poisson-Boltzmann equation, either by considering its Green function [30], or by analyzing the zero-point energy of surface modes [31, 32] as in Ref. [33]. For the interaction between metallic plates, the zero frequency contribution is relevant only at very long dis- tances, in the micrometer range and beyond. Screening is hence negligible in the experiments probing the Casimir force between metallic surfaces across ethanol [34, 35] at distances up to ∼ 102 nm. On the other hand, the zero frequency contribution provides a sizable fraction of the total interaction energy even at distances in the nanome- ter range if the electric permittivities of the interacting and intervening media are approximately matched in the infrared spectral range. An important example, given its applications in cell biology, is that of lipid layers interact- ing across an aqueous medium [36, 37], particularly on account of the very large dielectric constant of water at zero frequency. The screening of the van der Waals force was inferred from measurements of the distance between lipid membranes (in the nanometer range) as a function of salt concentration [38]. Given the complexity of such systems, comparison with an ab-initio theoretical model for the van der Waals interaction appears as a daunting task. Simpler configurations, more amenable to theoretical descriptions, could provide a testing ground for inves- 9 1 0 2 n u J 4 1 ] l l a h - s e m . t a m - d n o c [ 1 v 5 9 3 6 0 . 6 0 9 1 : v i X r a tigating the salt screening effect. Indeed, the zero fre- quency contribution dominates the Casimir interaction between polystyrene surfaces across a layer of water even at distances as low as ∼ 102 nm [39 -- 41]. Unfortunately, in this range the overall attractive signal is weak and a comparison with theory is difficult to implement [42], also in part because of surface roughness effects [43, 44]. Re- cent force measurements with polystyrene microspheres for distances up to ∼ 20 nm are not sensitive to screening [45] (see also [46] for a review). Very weak double-layer forces between polystyrene mi- crospheres, of the order of 10 fN, were recently measured with the help of optical tweezers [41]. Optical tweezers are ideally suited to probe the zero-frequency contribu- tion to the Casimir interaction, and hence its reduction by salt screening, since trapping with a single laser beam requires a condition of nearly index matching at the laser wavelength (typically in the near infrared). Such experi- ment should allow for a comparison with theoretical mod- els built on scattering theory. In this paper, we develop the scattering theory of the Casimir interaction between two parallel planar surfaces separated by a layer of an electrolyte solution. We take the non-local response of the electrolyte into account and analyze the propagation of longitudinal and transverse modes and their coupling by reflection at the surface of the (local) dielectric medium. The Casimir interac- tion energy is then derived from the matrix describing a round-trip of the electromagnetic waves propagating in between the two interacting surfaces. When taking typical values for the salt concentration, we find that the presence of ions in solution does not change the contribution of nonzero Matsubara frequen- cies. For the zero frequency case, we recover the result of Refs. [30 -- 32], which is now reinterpreted as the screened contribution of longitudinal modes written in terms of the corresponding reflection coefficient. We also find an additional term, accounting for the contribution of trans- verse magnetic (TM) modes at zero frequency. Our model is based on macroscopic Maxwell equations and constitutive equations for the different materials in- volved. We take the constitutive equation of a bulk elec- trolyte to describe its nonlocal response. Note, however, that the surfaces bounding the electrolyte modify the constitutive equations and the derived reflection coeffi- cients [47 -- 50], which should modify our results at dis- tances smaller than the characteristic Debye screening length. Results for the van der Waals interaction beyond the bulk approximation were derived in Ref. [51]. A more microscopic theory, built on the analysis of charge fluctu- ations as in Refs. [52 -- 54], would be required to take into account ion specific effects and density correlations [55], which could play a role at very short distances. The paper is organized in the following way. In Sec. II we derive the reflection matrix describing the coupling between longitudinal and transverse modes propagating in the electrolyte solution. Such matrix is the build- ing block for developing the scattering formalism of the 2 Casimir interaction in Sec. III. Numerical results for the case of two polystyrene surface interacting across an aqueous medium are presented in Sec. IV. Sec. V con- tains concluding remarks. II. REFLECTION MATRIX FOR THE ELECTROLYTE-DIELECTRIC INTERFACE The key ingredient for the scattering theory to be de- veloped in the next section is the reflection matrix for the interface between the electrolyte solution and the lo- cal medium, describing the coupling between longitudi- nal and transverse magnetic waves. We start by review- ing the hydrodynamical model for 1:1 electrolytes in the bulk approximation [21]. In Fourier space, the constitu- tive equation for the ionic current is J(K, ω) = σ(cid:96)(K, ω)E(cid:96)(K, ω) + σt(ω)Et(K, ω), (1) where E(cid:96)(K) and Et(K) are the longitudinal and trans- verse components of the electric field, respectively. The transverse conductivity is local and given by the usual Drude-like model, whereas the longitudinal conductivity is nonlocal: σt(ω) = σ(cid:96)(K, ω) = ω2 P γ − iω , ω2 P γ − iω + iv2 th K2 ω , (2) (3) We have taken
0 = µ0 = 1. The plasma frequency is ωP = (cid:112)N e2/m where N is the number of free charge carriers per volume, e is the electric charge of cations and m is the mass of both cations and anions (assumed to be equal). The nonlocal behavior in real space trans- lates into the K− dependence (spatial dispersion) in (3), representing the thermal average velocity of the ions in solution (kB = Boltzmann constant). which is controlled by the parameter vth = (cid:112)kBT /m The electrolyte dielectric functions for transverse and longitudinal waves follow from the conductivities dis- cussed above:
1(ω) =
b(ω) − ω(ω + iγ) ω2 P
(cid:96)(K, ω) =
b(ω) −(cid:18) ω(ω + iγ) ω2 P (4) (5) λ2 D
b0 − K 2(cid:19)−1 where
b is the dielectric function of pure water at zero ionic concentration. We have introduced the Debye screening length in terms of the electrostatic permittivity of the medium
b0 : λD = √
b0 =(cid:114)
b0kBT N e2 , vth ωP (6) which can be tuned by changing the salt concentration N. The spatial dispersion explicit in Eq. (5) allows for the propagation of longitudinal waves satisfying the disper- sion relation
(cid:96)(K(cid:96), ω) = 0. (7) We write the wave-vector K(cid:96) = k(cid:96) z + k in terms of its projection k on the xy plane. Transverse waves satisfy the standard dispersion relation
1(ω)ω2/c2 = k2 1 + k2, with Kt = k1 z + k. We now consider the reflection of longitudinal and transverse waves propagating in the electrolyte by a pla- nar interface perpendicular to the z-axis. For a gen- eral oblique incidence, TM and longitudinal waves be- come coupled by reflection, while transverse electric (TE) waves are reflected following the standard Fresnel for- mula. The frequency ω and wavevector projection paral- lel to the surface k are conserved by reflection. In Appendix A, we derive the reflected fields for a gen- eral incident wave propagating from the electrolyte. In addition to the usual boundary conditions for the tangen- tial electric and magnetic fields, we take the condition for the ionic current Jz = 0 at the interface at z = 0 [21]. We use the indices s, p to represent TE and TM polar- izations, respectively. We cast the results in terms of the block-diagonal reflection matrix R giving the reflected fields as a linear combination of incident tranverse and longitudinal waves:  = R s p E(r) E(r) E(r)  R =(cid:32) rss 0 0 (cid:96) 0 rpp r(cid:96)p Ein s Ein p Ein (cid:96)  r(cid:96)(cid:96)(cid:33) 0 rp(cid:96) (8) rss is the standard Fresnel coefficient for TE polarization, which is not modified by the presence of ions: rss = k1 − k2 k1 + k2 . (9) The Fresnel coefficient for TM polarization rpp is modi- fied by the coupling with longitudinal waves: (
1 −
b) (
1 −
b)
2k1 −
1k2 + k2
2k1 +
1k2 − k2 rpp =
2
b
2
b (10) k(cid:96) k(cid:96) and the diagonal element for longitudinal waves in Eq. (8) is given by 3 At normal incidence (k = 0), the reflection matrix is diagonal, and rpp coincides with the standard Fresnel co- efficient rTM for TM polarization as expected. We also recover the standard TM Fresnel coefficient at frequen- cies ω (cid:29) ωP , since
1 ≈
b according to Eq. (4) in this case. The ions are too slow to couple transverse and lon- gitudinal waves at such large field frequencies, and then the reflection matrix is approximately diagonal. Such property entails that the contribution of nonzero Mat- subara frequencies are nearly unaffected by the presence of ions in solution, as discussed in the next section. III. ROUND-TRIP MATRIX AND THE CASIMIR INTERACTION ENERGY In this section, we derive the Casimir interaction en- ergy between two local dielectric half-spaces separated by a layer (thickness L) of a (non-local) electrolyte so- lution, as depicted in Fig. 1. For simplicity, we assume that the local media on both sides have the same electro- magnetic properties. Then the reflection matrices at the two interfaces are identical. FIG. 1: Casimir interaction across a layer of thickness L con- taining a non-local electrolyte solution, with dielectric func- tions
1 and
(cid:96) for transverse and longitudinal waves, respec- tively. For simplicity, we assume that the interacting (local) half-spaces share the same dielectric function
2. The Casimir energy is computed from the matrix R describing the cou- pling between longitudinal and transverse waves by reflection at the interface. After a Wick rotation, the Casimir free energy is written as a sum over the Matsubara frequencies ξn = 2π n kBT /¯h, with n integer. The (nonlocal) longitudinal dielectric constant (5) is then written as r(cid:96)(cid:96) =
2k1 +
1k2 +
2
b
2k1 +
1k2 −
2
b k2 k(cid:96) k2 k(cid:96) (
1 −
b) (
1 −
b) The nondiagonal matrix elements describe the conversion between TM-polarized and longitudinal waves: r(cid:96)p = rp(cid:96) = 2 k k(cid:96) k1
2
b (
1 −
b)
2k1 +
1k2 − k2 k(cid:96) 2
2k
2k1 +
1k2 − k2 k(cid:96)
2
b
2
b (
1 −
b) (cid:112)k2 (cid:112)k2 (
1 −
b) (cid:96) + k2 √
1 ω/c √
1 ω/c (cid:96) + k2 (11)
(cid:96)(K,ξn) =
b(iξn) + ω2 P ξn(ξn + γ) + v2 thK 2 (14) Eq. (14) shows that
(cid:96)(K,ξn) is a Lorentzian of width ≈ ξn/vth for any nonzero Matsubara frequency. In real space, the displacement field D is then given in terms of the electric field E by a convolution integral with an exponential kernel, corresponding to the characteristic length scale λn ≡ vth/ξn = λdB/[(2π)3/2n], (15) (12) (13) zelectrolyteL3casttheresultsintermsoftheblock-diagonalreflectionmatrixRgivingthereflectedfieldsasalinearcombina-tionofincidenttranverseandlongitudinalwaves:0@E(r)TEE(r)TME(r)`1A=R EinTEEinTMEin`!withR= RTE000RttRt`0R`tR``!(7)RTEisthestandardFresnelcoecientforTEpolariza-tion.Theadditionaldiagonalelementsin(21)aregivenbyRtt=✏2kt✏tk2+k2k`✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)(8)R``=✏2kt+✏tk2+✏2✏bk2k`(✏t✏b)✏2kt+✏tk2✏2✏bk2k`(✏t✏b)(9)Thenondiagonalmatrixelementsdescribethecou-plingbetweenTM-polarizedandlongitudinalwaves:R`t=2kk`kt✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)pk2`+k2p✏t!/c(10)Rt`=2✏2k✏2kt+✏tk2k2k`✏2✏b(✏t✏b)p✏t!/cpk2`+k2(11)Fornormalincidence(k=0),thereflectionmatrixisdiagonal,andRttcoincideswiththestandardFresnelco-ecientRTMforTMpolarizationasexpected.WealsorecoverthestandardTMFresnelcoecientatfrequencies!!P,since✏t⇡✏baccordingtoEq.(3)inthiscase.Theionsaretooslowtocoupletransverseandlongitu-dinalwavesatsuchlargefieldfrequencies,andthenthereflectionmatrixisapproximatelydiagonal.Suchprop-ertyentailsthatthecontributionofpositiveMatsubarafrequenciesarenearlyun-changedbythepresenceofionsinsolution,asdiscussedinthenextsection.III.ROUND-TRIPMATRIXANDTHECASIMIRINTERACTIONENERGYInthissection,wederivetheCasimirinteractionen-ergybetweentwolocaldielectrichalf-spacesseparatedbyalayer(thicknessL)ofa(non-local)electrolyteso-lution,asdepictedinFig.1.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectro-magneticproperties.Thenthereflectionmatricesatthetwointerfacesareidentical.AfterWickrotation,theCasimirfreeenergy(temper-atureTandinterfaceareaA)iswrittenasasumoverϵ2ϵlzϵ2(free charges)LϵtFIG.1:Casimirinteractionacrossanon-localmediumcon-tainingfreecharges.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectromagneticproper-ties.theMatsubarafrequencies⇠n=2⇡nkBT/¯h,withnanon-negativeinteger:F=kBTA1Xn=00Zd2k(2⇡)2lndet(1Mn)(12)withtheprimeindicatingthatthen=0termismulti-pliedby1/2.Theround-tripmatrixMnisgivenbyMn=ReKLReKL(13)ThereflectionmatrixRfortheelectrolyte-dielectricin-terfacewasderivedintheprevioussection.ThepropagationmatrixeKLisdiagonal:eKL= eL000eL000e`L!(14)WeemploytheresultsofSec.IIIfortheTM-logitudinalreflectionmatrixRaftertakingthereplace-ments!!i⇠n(15)✏t!✏(16)kt!i=ipk2+✏⇠2n/c2(17)k2!i2=ipk2+✏2⇠2n/c2(18)k`!i`=isk2+⇠n(⇠n+)+!2P2D!2P(19)The(nonlocal)longitudinaldielectricconstant(4)iswrittenas✏`(K,⇠n)=✏b(i⇠n)+!2P⇠n(⇠n+)+v2thK2(20)Eq.(17)showsthat✏`(K,⇠n)isaLorentzianofwidth⇡⇠n/vthforanypositiveMatsubarafrequency.Inrealspace,thedisplacementfieldDisthengivenintermsoftheelectricfieldEbyaconvolutionintegralwithanexponentialkernel,correspondingtothecharacteristiclengthscalesn⌘vth/⇠n=dB/[(2⇡)3/2n],(21)3casttheresultsintermsoftheblock-diagonalreflectionmatrixRgivingthereflectedfieldsasalinearcombina-tionofincidenttranverseandlongitudinalwaves:0@E(r)TEE(r)TME(r)`1A=R EinTEEinTMEin`!withR= RTE000RttRt`0R`tR``!(7)RTEisthestandardFresnelcoecientforTEpolariza-tion.Theadditionaldiagonalelementsin(21)aregivenbyRtt=✏2kt✏tk2+k2k`✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)(8)R``=✏2kt+✏tk2+✏2✏bk2k`(✏t✏b)✏2kt+✏tk2✏2✏bk2k`(✏t✏b)(9)Thenondiagonalmatrixelementsdescribethecou-plingbetweenTM-polarizedandlongitudinalwaves:R`t=2kk`kt✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)pk2`+k2p✏t!/c(10)Rt`=2✏2k✏2kt+✏tk2k2k`✏2✏b(✏t✏b)p✏t!/cpk2`+k2(11)Fornormalincidence(k=0),thereflectionmatrixisdiagonal,andRttcoincideswiththestandardFresnelco-ecientRTMforTMpolarizationasexpected.WealsorecoverthestandardTMFresnelcoecientatfrequencies!!P,since✏t⇡✏baccordingtoEq.(3)inthiscase.Theionsaretooslowtocoupletransverseandlongitu-dinalwavesatsuchlargefieldfrequencies,andthenthereflectionmatrixisapproximatelydiagonal.Suchprop-ertyentailsthatthecontributionofpositiveMatsubarafrequenciesarenearlyun-changedbythepresenceofionsinsolution,asdiscussedinthenextsection.III.ROUND-TRIPMATRIXANDTHECASIMIRINTERACTIONENERGYInthissection,wederivetheCasimirinteractionen-ergybetweentwolocaldielectrichalf-spacesseparatedbyalayer(thicknessL)ofa(non-local)electrolyteso-lution,asdepictedinFig.1.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectro-magneticproperties.Thenthereflectionmatricesatthetwointerfacesareidentical.AfterWickrotation,theCasimirfreeenergy(temper-atureTandinterfaceareaA)iswrittenasasumoverϵ2ϵlzϵ2(free charges)LϵtFIG.1:Casimirinteractionacrossanon-localmediumcon-tainingfreecharges.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectromagneticproper-ties.theMatsubarafrequencies⇠n=2⇡nkBT/¯h,withnanon-negativeinteger:F=kBTA1Xn=00Zd2k(2⇡)2lndet(1Mn)(12)withtheprimeindicatingthatthen=0termismulti-pliedby1/2.Theround-tripmatrixMnisgivenbyMn=ReKLReKL(13)ThereflectionmatrixRfortheelectrolyte-dielectricin-terfacewasderivedintheprevioussection.ThepropagationmatrixeKLisdiagonal:eKL= eL000eL000e`L!(14)WeemploytheresultsofSec.IIIfortheTM-logitudinalreflectionmatrixRaftertakingthereplace-ments!!i⇠n(15)✏t!✏(16)kt!i=ipk2+✏⇠2n/c2(17)k2!i2=ipk2+✏2⇠2n/c2(18)k`!i`=isk2+⇠n(⇠n+)+!2P2D!2P(19)The(nonlocal)longitudinaldielectricconstant(4)iswrittenas✏`(K,⇠n)=✏b(i⇠n)+!2P⇠n(⇠n+)+v2thK2(20)Eq.(17)showsthat✏`(K,⇠n)isaLorentzianofwidth⇡⇠n/vthforanypositiveMatsubarafrequency.Inrealspace,thedisplacementfieldDisthengivenintermsoftheelectricfieldEbyaconvolutionintegralwithanexponentialkernel,correspondingtothecharacteristiclengthscalesn⌘vth/⇠n=dB/[(2⇡)3/2n],(21)3casttheresultsintermsoftheblock-diagonalreflectionmatrixRgivingthereflectedfieldsasalinearcombina-tionofincidenttranverseandlongitudinalwaves:0@E(r)TEE(r)TME(r)`1A=R EinTEEinTMEin`!withR= RTE000RttRt`0R`tR``!(7)RTEisthestandardFresnelcoecientforTEpolariza-tion.Theadditionaldiagonalelementsin(21)aregivenbyRtt=✏2kt✏tk2+k2k`✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)(8)R``=✏2kt+✏tk2+✏2✏bk2k`(✏t✏b)✏2kt+✏tk2✏2✏bk2k`(✏t✏b)(9)Thenondiagonalmatrixelementsdescribethecou-plingbetweenTM-polarizedandlongitudinalwaves:R`t=2kk`kt✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)pk2`+k2p✏t!/c(10)Rt`=2✏2k✏2kt+✏tk2k2k`✏2✏b(✏t✏b)p✏t!/cpk2`+k2(11)Fornormalincidence(k=0),thereflectionmatrixisdiagonal,andRttcoincideswiththestandardFresnelco-ecientRTMforTMpolarizationasexpected.WealsorecoverthestandardTMFresnelcoecientatfrequencies!!P,since✏t⇡✏baccordingtoEq.(3)inthiscase.Theionsaretooslowtocoupletransverseandlongitu-dinalwavesatsuchlargefieldfrequencies,andthenthereflectionmatrixisapproximatelydiagonal.Suchprop-ertyentailsthatthecontributionofpositiveMatsubarafrequenciesarenearlyun-changedbythepresenceofionsinsolution,asdiscussedinthenextsection.III.ROUND-TRIPMATRIXANDTHECASIMIRINTERACTIONENERGYInthissection,wederivetheCasimirinteractionen-ergybetweentwolocaldielectrichalf-spacesseparatedbyalayer(thicknessL)ofa(non-local)electrolyteso-lution,asdepictedinFig.1.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectro-magneticproperties.Thenthereflectionmatricesatthetwointerfacesareidentical.AfterWickrotation,theCasimirfreeenergy(temper-atureTandinterfaceareaA)iswrittenasasumoverϵ2ϵlzϵ2(free charges)LϵtFIG.1:Casimirinteractionacrossanon-localmediumcon-tainingfreecharges.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectromagneticproper-ties.theMatsubarafrequencies⇠n=2⇡nkBT/¯h,withnanon-negativeinteger:F=kBTA1Xn=00Zd2k(2⇡)2lndet(1Mn)(12)withtheprimeindicatingthatthen=0termismulti-pliedby1/2.Theround-tripmatrixMnisgivenbyMn=ReKLReKL(13)ThereflectionmatrixRfortheelectrolyte-dielectricin-terfacewasderivedintheprevioussection.ThepropagationmatrixeKLisdiagonal:eKL= eL000eL000e`L!(14)WeemploytheresultsofSec.IIIfortheTM-logitudinalreflectionmatrixRaftertakingthereplace-ments!!i⇠n(15)✏t!✏(16)kt!i=ipk2+✏⇠2n/c2(17)k2!i2=ipk2+✏2⇠2n/c2(18)k`!i`=isk2+⇠n(⇠n+)+!2P2D!2P(19)The(nonlocal)longitudinaldielectricconstant(4)iswrittenas✏`(K,⇠n)=✏b(i⇠n)+!2P⇠n(⇠n+)+v2thK2(20)Eq.(17)showsthat✏`(K,⇠n)isaLorentzianofwidth⇡⇠n/vthforanypositiveMatsubarafrequency.Inrealspace,thedisplacementfieldDisthengivenintermsoftheelectricfieldEbyaconvolutionintegralwithanexponentialkernel,correspondingtothecharacteristiclengthscalesn⌘vth/⇠n=dB/[(2⇡)3/2n],(21)3casttheresultsintermsoftheblock-diagonalreflectionmatrixRgivingthereflectedfieldsasalinearcombina-tionofincidenttranverseandlongitudinalwaves:0@E(r)TEE(r)TME(r)`1A=R EinTEEinTMEin`!withR= RTE000RttRt`0R`tR``!(7)RTEisthestandardFresnelcoecientforTEpolariza-tion.Theadditionaldiagonalelementsin(21)aregivenbyRtt=✏2kt✏tk2+k2k`✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)(8)R``=✏2kt+✏tk2+✏2✏bk2k`(✏t✏b)✏2kt+✏tk2✏2✏bk2k`(✏t✏b)(9)Thenondiagonalmatrixelementsdescribethecou-plingbetweenTM-polarizedandlongitudinalwaves:R`t=2kk`kt✏2✏b(✏t✏b)✏2kt+✏tk2k2k`✏2✏b(✏t✏b)pk2`+k2p✏t!/c(10)Rt`=2✏2k✏2kt+✏tk2k2k`✏2✏b(✏t✏b)p✏t!/cpk2`+k2(11)Fornormalincidence(k=0),thereflectionmatrixisdiagonal,andRttcoincideswiththestandardFresnelco-ecientRTMforTMpolarizationasexpected.WealsorecoverthestandardTMFresnelcoecientatfrequencies!!P,since✏t⇡✏baccordingtoEq.(3)inthiscase.Theionsaretooslowtocoupletransverseandlongitu-dinalwavesatsuchlargefieldfrequencies,andthenthereflectionmatrixisapproximatelydiagonal.Suchprop-ertyentailsthatthecontributionofpositiveMatsubarafrequenciesarenearlyun-changedbythepresenceofionsinsolution,asdiscussedinthenextsection.III.ROUND-TRIPMATRIXANDTHECASIMIRINTERACTIONENERGYInthissection,wederivetheCasimirinteractionen-ergybetweentwolocaldielectrichalf-spacesseparatedbyalayer(thicknessL)ofa(non-local)electrolyteso-lution,asdepictedinFig.1.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectro-magneticproperties.Thenthereflectionmatricesatthetwointerfacesareidentical.AfterWickrotation,theCasimirfreeenergy(temper-atureTandinterfaceareaA)iswrittenasasumoverϵ2ϵlzϵ2(free charges)LϵtFIG.1:Casimirinteractionacrossanon-localmediumcon-tainingfreecharges.Forsimplicity,weassumethatthelocalmediaonbothsideshavethesameelectromagneticproper-ties.theMatsubarafrequencies⇠n=2⇡nkBT/¯h,withnanon-negativeinteger:F=kBTA1Xn=00Zd2k(2⇡)2lndet(1Mn)(12)withtheprimeindicatingthatthen=0termismulti-pliedby1/2.Theround-tripmatrixMnisgivenbyMn=ReKLReKL(13)ThereflectionmatrixRfortheelectrolyte-dielectricin-terfacewasderivedintheprevioussection.ThepropagationmatrixeKLisdiagonal:eKL= eL000eL000e`L!(14)WeemploytheresultsofSec.IIIfortheTM-logitudinalreflectionmatrixRaftertakingthereplace-ments!!i⇠n(15)✏t!✏(16)kt!i=ipk2+✏⇠2n/c2(17)k2!i2=ipk2+✏2⇠2n/c2(18)k`!i`=isk2+⇠n(⇠n+)+!2P2D!2P(19)The(nonlocal)longitudinaldielectricconstant(4)iswrittenas✏`(K,⇠n)=✏b(i⇠n)+!2P⇠n(⇠n+)+v2thK2(20)Eq.(17)showsthat✏`(K,⇠n)isaLorentzianofwidth⇡⇠n/vthforanypositiveMatsubarafrequency.Inrealspace,thedisplacementfieldDisthengivenintermsoftheelectricfieldEbyaconvolutionintegralwithanexponentialkernel,correspondingtothecharacteristiclengthscalesn⌘vth/⇠n=dB/[(2⇡)3/2n],(21)✏1,✏`!1!2 M=4.84mVP=2⇡c!Pr1r2=0.9r1r2=0.6k=1k=2 (ra,t)= 1(ra,t)+ 2(ra,t) k(ra,t),k=1,212 where λdB =(cid:16) 2π¯h2 mkB T(cid:17) 1 is the thermal de Broglie wave- length of the ions at room temperature. Since this length is extremely small, we conclude that the electrolyte be- haves approximately as a local medium for all nonzero Matsubara frequencies, as discussed in further detail be- low. On the other hand, for the zero frequency (14) yields
(cid:96)(K, 0) =
b0(1 + 1/(λDK)2), so that the scale of vari- ation with K is now controlled by the Debye screening length λD instead of the de Broglie wavelength. Thus, we expect strong nonlocal effects and the contribution of longitudinal modes as far as the zero-frequency contribu- tion is concerned. The Casimir free energy is written in terms of the round-trip operator M(ξn) describing the scattering of longitudinal and transverse waves between the interact- ing surfaces of area A: F = Fn A = Fn ∞(cid:88)n=−∞ (cid:90) kBT 2 d2k (2π)2 ln det[1 − M(ξn)] The round-trip matrix M(ξn) is given by M(ξn) = R e−KL R e−KL (16) (17) (18) The reflection matrix R for the electrolyte-dielectric in- terface was derived in the previous section. We replace ω by iξn and the axial wave-vector components for the transverse waves in medium m = 1, 2 by iκm = n/c2. The longitudinal waves in the elec- i(cid:112)k2 +
mξ2 trolyte correspond to iκ(cid:96) = i(cid:115)k2 + 1 v2 th(cid:18)ξn(ξn + γ) + ω2 P
b(iξn)(cid:19). The propagation matrix e−KL is diagonal: e−KL =(cid:32) e−κ1L 0 0 0 e−κ1L 0 0 0 e−κ(cid:96)L(cid:33) (19) When writing the scattering formula (17), we have as- sumed a condition of full thermodynamical equilibrium for all scattering channels, including the longitudinal modes associated to the presence of movable ions. In addition, our derivation is based on the bulk model for the non-local response reviewed in the previous section. Given such assumptions, we are not allowed to take the limit L (cid:28) λD, which would eventually suppress the longi- tudinal channel and introduce the effect of the interfaces already at the level of the constitutive equations. The determinant in Eq. (17) is evaluated explicitly: det[1 − M(ξn)] =(cid:0)1 − r2 sse−2κ1L(cid:1) a0 + a1 ∆ + a2 ∆2 (1 + ∆)2 (20) 4 with rss defined by Eq. (9). The parameter ∆ quantifies the strength of the coupling between TM and longitudi- nal waves: ∆ =
2
b k2(
1 −
b) κ(cid:96)(
2κ1 +
1κ2) (21) and also a0 = (1 − e−2κ(cid:96)L)(cid:34)1 −(cid:18)
2κ1 −
1κ2
2κ1 +
1κ2(cid:19)2 a1 = 2(1 + e−2κ(cid:96)L)(cid:18)1 +
2κ1 −
1κ2
2κ1 +
1κ2 e−2κ1L(cid:35) e−2κ1L(cid:19) 8
2κ1 +
2κ1 +
1κ2 e−(κ1+κ(cid:96))L a2 = (1 − e−2κ(cid:96)L)(1 − e−2κ1L) Since kBT (cid:29) ¯hωP for electrolytes, for nonzero Mat- subara frequencies we have
1 −
b (cid:28) 1 and then ∆ (cid:28) 1. Thus, the round-trip matrix is approximately diagonal leading to independent contributions from the three po- larizations: (22) Fn A ≈ kBT 2 d2k (2π)2 (cid:34)(cid:88)σ=s,p (cid:90) + ln(1 − e−2κ(cid:96)L)(cid:35) ln(1 − r2 σσe−2κ1L) (23) (n (cid:54)= 0) As
1 ≈
b, we can ignore the presence of ions when com- puting the TM Fresnel reflection coefficients rpp in (23) and use instead the standard result for local dielectric media rpp ≈
2κ1 −
1κ2
2κ1 +
1κ2 (n (cid:54)= 0) when computing the contribution of longitudinal modes, cor- responding to the second term in the rhs of (23): In addition, we can approximate κ(cid:96) ≈ (cid:113)k2 + ξ2 λn(cid:19) , (n (cid:54)= 0) exp(cid:18)− F long A ≈ − 8πλn L kBT n v2 th 2L n where λn, defined by Eq. (15), is the characteristic non- local length at nonzero Matsubara frequencies. Since λn is smaller than the thermal de Broglie wavelength of the ions and is below the Angstrom range, F long is exponen- tially suppressed. Thus, we recover the standard DLP result for local materials [17] for all nonzero Matsubara frequencies. n For the zero frequency contribution, we find a0 = a2 = (1 − e−2√k2+1/λ2 a1 = 2(1 + e−2√k2+1/λ2 D L)(1 − e−2kL) D L)(1 − e−2kL). (24) The expression given by (20) then simplifies, and Eq. (17) leads to F0 A = kB T 2 ζ(3) 8πL2 (cid:34)− +(cid:90) (25) d2k (2π)2 ln(cid:16)1 − r2 DL(cid:17)(cid:35) (cid:96)(cid:96)e−2√k2+1/λ2 with the longitudinal reflection coefficient obtained from (11) by taking ξ = 0 : r(cid:96)(cid:96) = (n = 0) (26)
b(cid:112)k2 + 1/λ2
b(cid:112)k2 + 1/λ2 D −
2k D +
2k The first term in the rhs of (25) accounts for the contribu- tion of TM modes and is written in terms of the value of the Riemann zeta function ζ(3) ≈ 1.202. The result (25) can also be obtained directly from the reflection matrix R by noting that rp(cid:96)r(cid:96)p → 0 and rpp → −1 when ξ → 0. In conclusion, we find that the modification of the nonzero Matsubara frequency contributions on account of the movable ions is very small. For the zero-frequency case, on the other hand, we find a (screened) contribu- tion from longitudinal waves, the second term in the r.- h.-s. of (25), which coincides with the result of Refs. [30 -- 32]. Within the scattering approach, such contribution is written in terms of the coefficient r(cid:96)(cid:96) describing the reflection of longitudinal waves at the limit of zero fre- quency. According to Eq. (25), the screened contribution of longitudinal waves is accompanied by an unscreened contribution from the TM-polarized modes, which is not suppressed even in the limit of strong screening. IV. A NUMERICAL EXAMPLE: POLYSTYRENE SURFACES INTERACTING ACROSS AN AQUEOUS SOLUTION In this section, we apply the formal expressions derived previously to the important example of polystyrene half- spaces interacting across a layer of an aqueous solution. We take T = 293 K and the Lorentz model with the pa- rameters given by Ref. [56] to describe the required di- electric functions. Similar results are obtained by taking the models proposed in Ref. [40]. It is convenient to define the Hamaker coefficient [40] H(L) = −12π L2 F(L) A . (27) In Fig. 2, we plot H (in units of kBT ) as a function of L/λD. We consider two different values for the monova- lent salt concentration: 90 mM yielding λD = 1 nm and 0.9 mM corresponding to λD = 10 nm. They are repre- sented by the black and blue (dark grey) lines in Fig. 2, respectively, which are calculated by combining (16)-(17) with the full exact expression (20). As discussed in con- nection with Eq. (23), the contribution from nonzero fre- quencies is well approximated by the DLP standard result 5 [17] neglecting the presence of ions in solution. For the examples shown in Fig. 2, the relative difference between the exact and DLP results for the nonzero-frequency con- tribution is of the order of 10−5. The red (light grey) line in Fig. 2 corresponds to the separate zero-frequency contribution as computed from Eq. (25). The resulting contribution to the Hamaker co- efficient is an universal function of L/λD exhibiting two well defined plateaus, with a crossover at L/λD ∼ 1. At short distances, L (cid:28) λD, we add the contribution of longitudinal channels, whose magnitude is controlled by the reflection coefficient r(cid:96)(cid:96) given by (26), to the univer- sal constant value H TM 4 ζ(3)kBT ≈ 0.9 kBT arising from TM-polarized modes. As the distance increases, the former is suppressed by screening, while the latter defines the asymptotic limit of the total Hamaker coefficient at long distances. 0 = 3 0 Indeed, as the distance approaches the thermal wave- length ¯hc/(kBT ), the nonzero-frequency contribution is also exponentially suppressed, and then the Hamaker coefficient goes to the zero-frequency asymptotic value H TM . Such behaviour is indicated in Fig. 2, particu- larly for the blue (dark grey) curve corresponding to λD = 10 nm, since larger distances are shown in this case. The contribution of nonzero frequencies is maximized at short distances. When added to the zero-frequency value, it defines the unretarded Hamaker 'constant' H(0) cor- responding to the short-distance plateau for the black and blue (dark grey) lines. The former, corresponding to λD = 1 nm, develops a second plateau at intermediate distances such that λD (cid:28) L (cid:28) λ0, with λ0 representing the typical scale for the resonance wavelengths of wa- ter and poystyrene. In this range, the longitudinal zero- frequency term is suppressed by ionic screening, but the nonzero-frequency contribution is still approximately un- affected by electrodynamical retardation. On the other hand, when considering λD = 10 nm, both screening and retardation take place at approximately the same dis- tance range, leading to the more steady decay of the Hamaker coefficient shown by the blue (dark grey) line in Fig. 2. The results obtained in this section can be directly ap- plied to the interaction between two polystyrene micro- spheres across an aqueous solution, provided that their radii R1 and R2 are much larger than the distance. In this case, we can take the proximity force approximation (PFA), also known as Derjaguin approximation [57], in order to derive the attractive Casimir force FSS between the two spheres from the free energy for parallel planar surfaces taken at the distance of closest approach L : FSS = 2πReff F(L) A (28) with Reff = R1R2/(R1 + R2). In Fig. 3, we plot FSS/Reff versus distance taking λD = 10 nm. The solid line corresponds to the scattering theory and is obtained from the results for the Hamaker coefficient displayed in Fig. 2 combined with Eqs. (27) 6 FIG. 2: Variation of the Hamaker coefficient (in units of kBT ) with distance (in units of the Debye screening length λD). We consider two polystyrene surfaces interacting across an aqueous solution. Black: λD = 1 nm; blue (dark grey): λD = 10 nm. The red (light grey) line represents the zero- frequency contribution, which is an universal function of L/λD. and (28). The dashed line shows the results calculated from the linear Poisson-Boltzmann equation for the zero- frequency contribution [30 -- 32], combined with the DLP formalism [17] for the nonzero frequencies. The numer- ical difference between the two curves arises from the contribution of TM modes at the limit ξ → 0, which is taken into account within the scattering theory but not by the approach of Refs. [30 -- 32]. At short distances, zero-frequency TM contribution is a relatively small frac- tion of the total interaction energy. Thus, the two curves shown in Fig. 3 are close to each other for L <∼ 10 nm, which is the typical range probed with polystyrene col- loids [45, 46]. Specifically, the relative discrepancy in- creases from 35% at L = 1 nm to 48% at L = 10 nm. On the other hand, Fig. 3 shows that the two models deviate strongly from each other as L increases past the length scales λD and λ0, due to the suppression of the longitudinal and nonzero-frequency terms as discussed in connection with Fig. 2. An order-of-magnitude discrep- ancy is found at L = 100 nm. Such range of distances might be reachable by employing optical tweezers [41], which allow for the measurement of very weak interac- tions. V. CONCLUSION We have developed the scattering approach to the Casimir interaction across an electrolyte solution. The key ingredient in our formalism is the matrix describing the reflection of transverse and longitudinal waves at the interfaces between the electrolyte and the interacting di- electric materials. Our derivation considers arbitrary fre- quencies, and the zero-frequency contribution is obtained by taking the limit ξ → 0 at the very end. As expected, we find that the ions in solution do not modify, to a very FIG. 3: Variation of the Casimir force FSS between two polystyrene microspheres with distance L. The microspheres of radii R1 and R2 interact across an aqueous solution with λD = 10 nm. We assume that R1, R2 (cid:29) L and calculate the force within the proximity force approximation, which is pro- portional to the effective radius Reff = R1R2/(R1 + R2). The dashed line is calculated by considering longitudinal chan- nels at zero frequency alongside the nonzero frequency modes, whereas the solid line also adds TM channels at the limit of zero frequency in accordance with the scattering approach. good approximation, the contributions at nonzero Mat- subara frequencies. At zero frequency, we find a screened contribution of longitudinal scattering channels which agrees with the result of previous derivations based on the linear Poisson-Boltzmann equation [30 -- 32]. Within the scattering approach, such screened contribution is cast in terms of the reflection amplitude r(cid:96)(cid:96) for longitu- dinal waves. Our derivation provides a new insight into the nature of the screened contribution to the Casimir force, and paves the way for the generalization to more general setups and geometries. In addition to the contribution of longitudinal chan- nels, we find a second contribution at zero frequency, associated to TM-polarized modes, which is not screened by the presence of ions and as a consequence defines the asymptotic behavior of the interaction at long distances. Our results are based on the bulk model for the electro- magnetic response of the electrolyte. Hence they should be valid for distances L >∼ λD. This condition overlaps with the distance range that allows for the suppression of the double-layer interaction between charged dielec- tric surfaces. Thus, our derivation is well adapted to experimental conditions aiming at isolating the Casimir interaction from electrostatic force signals. Acknowledgements. This work has been supported by Centre National de la Recherche Scientifique (CNRS) and Sorbonne Universit´e through their collaboration programs Projet International de Coop´eration Scien- tifique (PICS) and Convergence International, respec- tively. We also acknowledge partial financial support by the Brazilian agencies Conselho Nacional de De- senvolvimento Cient´ıfico e Tecnol´ogico (CNPq), Coor- dena¸cao de Aperfei¸coamento de Pessoal de N´ıvel Superior (CAPES), Funda¸cao de Amparo `a Pesquisa do Estado aqueous solution LpolystyrenepolystyreneL/λDH/(kBT)�����������������������������������������-���������L(nm)FSS/Reff(pN/μm)LR1R2de Minas Gerais (FAPEMIG), Funda¸cao Carlos Chagas Filho de Amparo `a Pesquisa do Estado do Rio de Janeiro (FAPERJ) and Funda¸cao de Amparo `a Pesquisa do Es- tado de Sao Paulo (FAPESP). Author contribution statement All authors contributed to the development of the the- oretical model and the interpretation of the results. Appendix A: Derivation of the reflection matrix elements In this appendix, we derive the matrix R describing re- flection at the interface between the electrolyte and the local medium. We consider longitudinal and transverse waves propagating in the electrolyte occupying the half- space z < 0. They are reflected at the interface at z = 0 and refracted into the local dielectric medium located at the haf-space z > 0. A similar derivation was presented for describing the optical excitation of plasmons in met- als [58]. However, we take Jz = 0 at the interface with the electrolyte, whereas Ref. [58] proposes the continuity of Ez at the interface between the metallic medium and vacuum as the additional boundary condition. Thus, the reflection matrix derived here diverges from the results of [58] for the metal-vacuum interface. We first consider the case of an incident TM wave. Without loss of generality, we assume that the incidence plane coincides with the xz plane. The total field in the p + E(r) nonlocal medium 1 is written as E = E(i) where p + E(r) (cid:96) E(i) p = E0 exp (iKp · r) y × Kp E(r) p = rpp E0 exp (iK ¯p · r) y × K ¯p E(r) (cid:96) = r(cid:96)p E0 exp (iK(cid:96) · r) K(cid:96) (A1) (A2) (A3) represent the incident, TM reflected, and longitudinal reflected waves, respectively. In the local medium 2, the refracted field is TM-polarized and given by 7 E2 = tpp E0 exp (iK2 · r) y × K2. The corresponding wave-vectors are written as Kp = (k, 0, k1) K ¯p = (k, 0,−k1) K(cid:96) = (k, 0,−k(cid:96)) K2 = (k, 0, k2) (A4) (A5) (A6) (A7) (A8) We now consider the boundary conditions at the inter- face z = 0 between the two media. The continuity of the tangential electric and magnetic fields lead to (1 − rpp) k1 + r(cid:96)p k = tpp k2
1 (1 + rpp) =
2 tpp . (A9) (A10) For the third boundary condition, Jz = 0 at z = 0, we take the dispersion relation (7) into account and then write (
1 −
b) (1 + rpp) k =
b r(cid:96)p k(cid:96) (A11) We now solve (A9), (A10) and (A11) for the coefficients rpp, r(cid:96)p and tpp. The case of a longitudinal incident wave can be solved in a similar way. Since the dimensionless vectors defin- ing the polarizations in (A1)-(A3) are non-unitary, the non-diagonal matrix elements describing the coupling be- tween transverse and longitudinal waves should be multi- plied by the ratio between the vectors moduli. The final results are given by Eqs. (10)-(13). [1] A. Lambrecht, P. A. Maia Neto, and S. Reynaud, The Casimir effect within scattering theory, New J. Phys. 8, 243 (2006). [2] S. J. Rahi, T. Emig, N. Graham, R. L. Jaffe, and M. Kardar, Scattering theory approach to electrodynamic Casimir forces, Phys. Rev. D 80, 085021 (2009). [3] H. B. G. Casimir, On the attraction between two per- fectly conducting plates, Proc. K. Ned. Akad. Wet. 51, 793 (1948). [4] T. Emig, Fluctuation-induced quantum interactions be- tween compact objects and a plane mirror, J. Stat. Mech. (2008) P04007. [5] P. A. Maia Neto, A. Lambrecht, and S. Reynaud, Casimir energy between a plane and a sphere in electromagnetic vacuum, Phys. Rev. A 78, 012115 (2008). [6] A. Canaguier-Durand, P. A. Maia Neto, I. Cavero-Pelaez, A. Lambrecht, and S. Reynaud, Casimir Interaction be- tween Plane and Spherical Metallic Surfaces, Phys. Rev. Lett. 102, 230404 (2009). [7] A. Canaguier-Durand, P. A. Maia Neto, A. Lambrecht, and S. Reynaud, Thermal Casimir Effect in the Plane- Sphere Geometry, Phys. Rev. Lett. 104, 040403 (2010). [8] A. Canaguier-Durand, P. A. Maia Neto, A. Lambrecht, and S. Reynaud, Thermal Casimir effect for Drude metals in the plane-sphere geometry, Phys. Rev. A 82, 012511 (2010). [9] R. Zandi, T. Emig, and U. Mohideen, Quantum and ther- mal Casimir interaction between a sphere and a plate: Comparison of Drude and plasma models, Phys. Rev. B 81, 195423 (2010). [10] M. Hartmann, G.-L. Ingold, and P. A. Maia Neto, Plasma versus Drude Modeling of the Casimir Force: Beyond the Proximity Force Approximation, Phys. Rev. Lett. 119, 043901 (2017). [11] M. Hartmann, G.-L. Ingold, and P. A. Maia Neto, Ad- vancing numerics for the Casimir effect to experimentally relevant aspect ratios, Phys. Scr. 93, 114003 (2018). [12] T. Emig, N. Graham, R. L. Jaffe, and M. Kardar, Casimir Forces between Arbitrary Compact Objects, Phys. Rev. Lett. 99, 170403 (2007). 8 [13] P. Rodriguez-Lopez, Casimir energy and entropy in the sphere-sphere geometry, Phys. Rev. B 84, 075431 (2011). [14] S. Umrath, M. Hartmann, G.-L. Ingold, and P. A. Maia Neto, Disentangling geometric and dissipative origins of negative Casimir entropies, Phys. Rev. E 92, 042125 (2015). [15] M.-T. Jaekel and S. Reynaud, Casimir force between par- tially transmitting mirrors, J. Phys. I (France) 1, 1395 (1991). [16] E.M. Lifshitz, The theory of molecular attractive forces between solids Soviet Phys. JETP 2, 73 (1956). [17] I. E. Dzyaloshinskii, E.M. Lifshitz, and L.P. Pitaevskii, General theory of van der Waals forces, Soviet Phys. Usp. 4, 153 (1961). [18] C. Genet, F. Intravaia, A. Lambrecht, and S. Reynaud. Electromagnetic vacuum fluctuations, Casimir and Van der Waals forces, Ann. Fondation Louis de Broglie 29, 311 (2004). [19] J. N. Israelachvili, Intermolecular and Surface Forces (Academic Press, Waltham, 2011). [20] H. J. Butt and M. Kappl, Surface and Interfacial Forces (Wiley-VCH, Weinheim, 2010). [21] B. Davies and B. W. Ninham, Van der Waals forces in electrolytes, J. Chem. Phys. 56, 5797 (1972). [22] E. I. Kats, Influence of nonlocality effects on van der Bezrukov, Measurements of the Casimir-Lifshitz force in fluids: The effect of electrostatic forces and Debye screen- ing, Phys. Rev. A 78, 032109 (2008). [35] A. Le Cunnuder, A. Petrosyan, G. Palasantzas, V. Sve- tovoy, and S. Ciliberto, Measurement of the Casimir force in a gas and in a liquid, Phys. Rev. B 98, 201408R (2018). [36] V. A. Parsegian and B. W. Ninham, Temperature- dependent van der Waals forces, Biophy. J. 10, 664 (1970). [37] V. A. Parsegian and B. W. Ninham, Toward the Cor- rect Calculation of van der Waals Interactions Between Lyophobic Colloids in an Aqueous Medium, J. Colloid Interface Sci. 37, 332 (1971). [38] H. I. Petrache, Th. Zemb, L. Belloni, and V. A. Parsegian, Salt screening and specific ion adsorption de- termine neutral-lipid membrane interactions, PNAS 103, 7982 (2006). [39] V. A. Parsegian, Long range van der Waals forces, in Physical Chemistry: enriching topics in colloid and sur- face science, edited by H. van Olphen and K. J. Mysels (Theorex, 1975). [40] W. B. Russel, D. A. Saville and W. R. Schowalter, Col- loidal Dispersions (Cambridge University Press, Cam- bridge, 1989). [41] D. S. Ether jr. et al., Probing the Casimir force with Waals interaction, Sov. Phys. JETP 46, 109 (1977). optical tweezers, EPL 112, 44001 (2015). [23] R. Esquivel, C. Villareal and W. L. Moch´an, Exact sur- face impedance formulation of the Casimir force: Ap- plication to spatially dispersive metals, Phys. Rev. A 68, 052103 (2003). Erratum: Phys. Rev. A 71, 029904E (2005). [24] R. Esquivel and V. B. Svetovoy, Correction to the Casimir force due to the anomalous skin effect, Phys. Rev. A 69, 062102 (2004). [25] A. M. Contreras-Reyes and W. L. M´ochan, Surface screening in the Casimir force, Phys. Rev. A 72, 034102 (2005). [26] R. Esquivel-Sirvent, C. Villarreal, W. L. Moch´an, A. M. Contreras-Reyes, and V. B. Svetovoy, Spatial dispersion in Casimir forces: a brief review, J. Phys. A: Math. Gen. 39, 6323 (2006). [27] D. Reiche, M. Oelschlager, K. Busch, and F. Intravaia, Extended hydrodynamic description for nonequilibrium atom-surface interactions, J. Opt. Soc. Am. B 36, C53 (2019). [28] G. Barton, Some surface effects in the hydrodynamic model of metals, Rep. Prog. Phys. 42, 65 (1979). [29] L. M. Woods, D. A. R. Dalvit, A. Tkatchenko, P. Rodriguez-Lopez, A. W. Rodriguez, and R. Podgornik, Materials perspective on Casimir and van der Waals in- teractions, Rev. Mod. Phys. 88, 045003 (2016). [30] D. J. Mitchell and P. Richmond, A General Formalism for the Calculation of Free Energies of Inhomogeneous Dielectric and Electrolyte System, J. Colloid Interface Sci. 46, 118 (1974). [31] J. Mahanty and B. W. Ninham, Dispersion Forces (Aca- demic, London) 1976. [32] V. A. Parsegian, Van der Waals Forces: A Handbook for Biologists, Chemists, Engineers, and Physicists (Cam- bridge, New York) 2006. [33] N. G. Van Kampen, B. R. A. Nijboer and K. Schram, On the Macroscopic Theory of Van der Waals Forces, Phys. Lett. 26A, 307 (1968). [34] J. Munday, F. Capasso, V. A. Parsegian and S. M. [42] M. A. Bevan and D. C. Prieve, Direct Measurement of Retarded van der Waals Attraction, Langmuir 15, 7925 (1999). [43] P. A. Maia Neto, A. Lambrecht, and S. Reynaud, Casimir effect with rough metallic mirrors, Phys. Rev. A 72, 012115 (2005). [44] P. J. van Zwol, G. Palasantzas, and J. Th. M. De Hosson, Influence of random roughness on the Casimir force at small separations Phys. Rev. B 77, 075412 (2008). [45] M. Elzbieciak-Wodka, M. N. Popescu, F. J. M. Ruiz- Cabello, G. Trefalt, P. Maroni, and M. Borkovec, Mea- surements of dispersion forces between colloidal latex particles with the atomic force microscope and compar- ison with Lifshitz theory, J. Chem. Phys. 140, 104906 (2014). [46] G. Trefalt, Th. Palberg, and M. Borkovec, Forces be- tween colloidal particles in aqueous solutions containing monovalent and multivalent ions, Curr. Opinion Colloid Interf. Sci. 27, 9 (2017). [47] G. S. Agarwal, D. N. Pattanayak, and E. Wolf, Struc- ture of the electromagnetic field in spatially dispersive medium, Phys. Rev. Lett. 27, 1022 (1971). [48] G. S. Agarwal, D. N. Pattanayak, and E. Wolf, Refrac- tion and reflection on a spatially dispersive medium, Opt. Commun. 4, 255 (1971). [49] G. S. Agarwal, D. N. Pattanayak, and E. Wolf, Bound- ary conditions on exciton polarization and mode coupling in a spatially dispersive medium, Opt. Commun. 4, 260 (1971). [50] A. A. Maradudin and D. L. Mills, Effect of spatial disper- sion on the properties of a semi-infinite dielectric, Phys. Rev. B 7, 2787 (1973). [51] V. N. Gorelkin and V. P. Smilga, Calculation of Inter- molecular Interaction Forces between Bodies Separated by a Film of a Strong Electrolyte Solution, Soviet Phys. JETP 36, 761 (1974). [52] J. Sarabadani, A. Naji, D. S. Dean, R. R. Horganm and R. Podgornik, Nonmonotoic fluctuation-induced interac- tions between dielectric slabs carrying charge disorder, J. Chem. Phys. 133, 174702 (2010). [53] D. S. Dean, V. A. Parsegian, and R. Podgornik, Fluc- tuation of thermal van der Waals forces due to dipole fluctuations, Phys. Rev. A 87, 032111 (2013). [54] D. S. Dean and R. Podgornik, Relaxation of the ther- mal Casimir force between net neutral plates containing Brownian charges, Phys. Rev. E 89, 032117 (2014). [55] Eletrostatics of soft and disordered matter, edited by D. S. Dean, J. Dobnikar, A. Naji and R. Podgornik (Pan Stanford, Singapore, 2014). [56] P. J van Zwol and G. Palasantzas, Repulsive Casimir forces between solid materials with high-refractive-index intervening liquids, Phys. Rev. A 81, 062502 (2010). [57] B. Derjaguin, Untersuchungen uber die Reibung und Adhasion, IV - Theorie des Anhaftens kleiner Teilchen, Kolloid-Zs. 69, 155 (1934). [58] A. R. Melnyk and M. J. Harrison, Theory of optical exci- tation of plasmons in metals, Phys. Rev. B 2, 835 (1970). 9 

1907.08092

1

1907

Theory of valley-resolved spectroscopy of a Si triple quantum dot coupled to a microwave resonator

We theoretically study a silicon triple quantum dot (TQD) system coupled to a superconducting microwave resonator. The response signal of an injected probe signal can be used to extract information about the level structure by measuring the transmission and phase shift of the output field. This information can further be used to gain knowledge about the valley splittings and valley phases in the individual dots. Since relevant valley states are typically split by several $\mu\text{eV}$, a finite temperature or an applied external bias voltage is required to populate energetically excited states. The theoretical methods in this paper include a capacitor model to fit experimental charging energies, an extended Hubbard model to describe the tunneling dynamics, a rate equation model to find the occupation probabilities, and an input-output model to determine the response signal of the resonator.

cond-mat.mes-hall

cond-mat

Theory of valley-resolved spectroscopy of a Si triple quantum dot coupled to a microwave resonator Maximilian Russ Department of Physics, University of Konstanz, D-78464 Konstanz, Germany Csaba G. P´eterfalvi Department of Physics, University of Konstanz, D-78464 Konstanz, Germany ‡ ‡ Guido Burkard Department of Physics, University of Konstanz, D-78464 Konstanz, Germany E-mail: Guido.Burkard@uni-konstanz.de Abstract. We theoretically study a silicon triple quantum dot (TQD) system coupled to a superconducting microwave resonator. The response signal of an injected probe signal can be used to extract information about the level structure by measuring the transmission and phase shift of the output field. This information can further be used to gain knowledge about the valley splittings and valley phases in the individual dots. Since relevant valley states are typically split by several µeV, a finite temperature or an applied external bias voltage is required to populate energetically excited states. The theoretical methods in this paper include a capacitor model to fit experimental charging energies, an extended Hubbard model to describe the tunneling dynamics, a rate equation model to find the occupation probabilities, and an input-output model to determine the response signal of the resonator. Submitted to: J. Phys.: Condens. Matter 9 1 0 2 l u J 8 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 9 0 8 0 . 7 0 9 1 : v i X r a ‡ Both authors contributed equally to this work. Valley-resolved spectroscopy in Si TQDs 2 Semiconductors with abundant nuclear spin-free isotopes are increasingly being silicon [1], germanium[2], and investigated as host material for spin qubits, e.g. It turns out that most of these materials comprise an electron graphene [3, 4]. valley degree of freedom [5] in the conduction band of the bulk material. In many nanostructures based on these semiconductor materials, the resulting valley splitting is still not fully understood and therefore represents in practice an unpredictable system parameter. It is known that the valley degree of freedom can be described as a pseudo- spin in a two dimensional electron gas (2DEG) whose attributes, i.e., valley-splitting and valley-phase, drastically depend on the interface of the heterostructure [6, 7, 8, 9, 10, 11, 12, 13]. A single atomic step can change the quantization axis of the pseudo-spin and the complex phase of the valley-orbit coupling of an electron can be modified by as much as π [11, 12, 13]. This has a large impact on silicon quantum computation [1] for most qubit implementations, which use the spin degree of freedom to encode quantum information [14]. For multi-qubit quantum processors [15, 16, 17, 18] and multi-spin qubit implementations [19, 20, 21, 22], the presence of the valley leads to several non-computational states into which the information can "leak". Since the number of leakage states exponentially increases with the number of electrons, the resulting complex energy diagram with a high density of states makes it difficult to find the optimal parameter regimes for encoding and operating such qubits. Therefore, a precise knowledge of the valley structure is required for high-fidelity qubit implementations and operations. A lower bound to the valley splitting can be obtained using ground-state magnetospectroscopy [23, 24, 25, 26, 27]. Recent advances in the coupling of electrons to superconducting microwave resonators [28, 29, 30, 31, 32, 33, 34, 35] allow for precise read-out of the valley splittings in double quantum dots [36, 32]. In this theoretical paper, the technique is extended and adapted to extract the valley splitting and valley phases in a silicon triple quantum dot (TQD) system using such superconducting microwave resonator. This paper is organized as follows. Firstly, we introduce a general theoretical model of the TQD system in Section 1. For this we use a classical capacitor model to find the electrostatic energies of the electrons and substitute them into an extended Hubbard model to account for hopping of the electrons between the dots (see subsection 1.1). Subsequently in subsection 1.2, we use a master equation to find the steady state solution of the electron dynamics in the presence of dissipative processes. This allows us to find the corresponding occupation probabilities. Finite temperature effects and an externally applied bias are included in our model. In subsection 1.3, we consider the response of a superconducting microwave cavity dispersively coupled to the TQD system and use input-output theory to derive analytical expressions for the response signal. Subsequently, in Section 2, we show how one can extract relevant system parameters from the cavity response signal. We explicitly demonstrate the case of a single electron in a triple quantum dot in subsection 2.1 and the case of three electrons in subsection 2.2. Valley-resolved spectroscopy in Si TQDs 3 1. Theoretical description We consider a triple quantum dot (TQD) connected to two leads and a superconducting transmission line resonator via the center gate (see Fig. 1). In order to describe the TQD theoretically, we first construct the bare electron Hamiltonian H of the system and introduce the interaction to the leads and the microwave resonator later. We consider a basis with 0, 1, 2 or 3 electrons with spin and two-fold valley degeneracy in each dot. For a fixed number of electrons ne, there are(cid:0)dd ds dv degeneracy ds, and the valley degeneracy dv. Therefore, we have N =(cid:80)3 with dd ds dv = 3 × 2 × 2 = 12 being the product of the number of dots dd, the spin (cid:1) possible basis states (cid:0)12 (cid:1) = 299 ne i=0 i basis states in total. Here, we restrict our analysis to the two energetically lowest laying orbital levels. Silicon quantum dots typically have relatively large orbital energies Eorb = 3 − 5 meV [37, 38], thus, the impact of higher orbital levels can be neglected for temperatures kBT (cid:28) Eorb and applied voltage biases Vl − Vr (cid:28) Eorb/eα where e is the electron charge and α is the norm of the lever arm matrix [39]. 1.1. Hamiltonian In order to obtain a good agreement of our theoretical studies with experiments we rely on a description with the extended Hubbard model. The electrostatic energies are given by a capacitor model of the TQD, schematically shown in Fig. 1. The free energy of the triple dot system reads [40] where T denotes the transposition and F = effC−1 QT dotQeff, 1 2  − Cgate nL nC nR  − Clead V1 V2 V3 (cid:33) (cid:32)VlVr (1) (2) Qeff = e quantifies the total effective charge on the quantum dots composed of the electron occupation number ni and the applied gate voltages V = (V1, V2, V3)T . Here e < 0 denotes the electron charge and ∆V = Vl −Vr the applied bias voltages between the left and right leads. The dot capacitance matrix reads which contains the self capacitances Ci and the mutual capacitances C12, C23 and C13. The capacitances between the gates and the dots reads C2 −C23 C3 −C12 −C13 −C23  C1 −C12 −C13  C1L C2L C3L C1C C2C C3C C1R C2R C3R,   Cdot = Cgate = − (3) (4) Valley-resolved spectroscopy in Si TQDs 4 Illustration of the setup: Figure 1. (a) a superconducting microwave resonator capacitively coupled to a linearly arranged silicon triple quantum dot (TQD) via the center plunger gate V2. The cavity is probed with the input signal ain. Measurement of the transmitted signal aout can be used to reconstruct the energy landscape of the TQD. (b) Capacitor model of the TQD where each dot is filled with ni electrons (i = L, C, R) and capacitively coupled to the electrostatic gates, V1, V2, V3. The Cvi denote the capacitance between gate v = 1, 2, 3 and dot i = L, C, R. The coefficients Ci,j describe the capacitances between the electrons in dot i and lead j = l, r. Applied voltages to lead reservoirs are denoted by Vj. The mutual capacitance Cik describes the electrostatic interaction between the electrons in the QDs i and k. Black lines capacitively couple neighboring dots, leads or gates, and gray lines denote next- neighbor coupling. Cross-coupling between the electrostatic gates is neglected since it leads only to an overall shift in energy [40]. and Clead =  −CL,l −CC,l −CC,r −CR,r 0  (5) 0 contains the capacitances between the dots and the left and right lead. For later convenience we also define the chemical potential in each dot µL(V ) =F(cid:0)(nL + 1, nC, nR), V(cid:1) − F(cid:0)(nL, nC, nR), V(cid:1), µC(V ) =F(cid:0)(nL, nC + 1, nR), V(cid:1) − F(cid:0)(nL, nC, nR), V(cid:1), µR(V ) =F(cid:0)(nL, nC, nR + 1), V(cid:1) − F(cid:0)(nL, nC, nR), V(cid:1). The total Hamiltonian of the hybrid TQD-resonator system is given by H = Hcharge + HZeeman + Hvalley + Htunnel, (6) (7) (8) (9) Valley-resolved spectroscopy in Si TQDs where the individual contributions are introduced below. The electrostatic interaction is described by the Hamiltonian Hcharge =F(cid:0)(0, 0, 0), V(cid:1) + + i ∂F(cid:0)(nL, nC, nR), V(cid:1) (cid:12)(cid:12)(cid:12)(cid:12)nL,C,R=0 (cid:88) ∂2F(cid:0)(nL, nC, nR), V(cid:1) (cid:12)(cid:12)(cid:12)(cid:12)nL,C,R=0 (cid:88) (cid:88) (cid:88) (cid:1) (cid:0)C−1 (cid:0)C−1 ∂ni∂nj (cid:1) ∂ni 1 2 ni i,j dot ij ni + e2 2 i,j =F0 + e Vj i,j ninj dot ij ninj with ni = (cid:80) † i,s,vci,s,v and the free energy F defined in Eq. (1). Here c † i,s,v) creates (annihilates) an electron in dot i = L, C, R, with spin s =↑,↓, and occupying the v = ± valley state. † i,s,v (c s,v c An externally applied magnetic field B breaks the spin-degeneracy. Considering a homogeneous magnetic field B = (0, 0, B), the Zeeman splitting is described by (n↑ − n↓) with ns =(cid:80) Hzeeman = (12) i,s,vci,s,v. The Zeeman energy is EZ = gµBB where g ≈ 2 is the electron † g-factor in silicon. To be precise, the electron g-factor depends on the valley and orbital level and is slightly anisotropic giving rise to small effective magnetic field gradients [41]. Here, this small anisotropy is neglected. i,v c EZ 2 For a silicon heterostructure the two minima in the conduction band [1] give rise to the valley degree of freedom. The valley splitting can be expressed locally in QD i as [6] 5 (10) (11) (13) (cid:32) (cid:33) Hv,j = 1 2 0 ∆j ∆∗ 0 j with the complex quantity ∆j = Ej V and valley phase φj in dot j = L, C, R. Because of atomistic defects at the silicon interface the valley pseudo-vector can have a different phase in each dot [36, 11, 42, 22]. The valley Hamiltonian in the valley eigenbasis of each dot can be written as V eiφj consisting of the valley splitting Ej with ni,v =(cid:80) Hvalley = (ni,+ − ni,−) Ei V 2 (14) † i,s,vci,s,v. In this particular choice of representation the valley phase is transferred to the coupling matrix elements between the quantum dots. The single-qubit inter-dot matrix elements in the valley eigenbasis can be expressed as [36] s c i,s,vcj,s,u → cos(θij) c † c † i,s,vcj,s,u + i sin(θij) c † i,s,vcj,s,¯u (15) 3(cid:88) i=1 Valley-resolved spectroscopy in Si TQDs 6 with i, j = L, C, R, v = ±, u = ± and ¯u = −u. The real-valued quantities θij = (φi − φj)/2 can be visually interpreted as the angle between the direction of the valley pseudo-spin in dot i and dot j. Off-diagonal elements of H allow for coherent hopping of electrons between neighboring quantum dots. In our model hopping is only allowed between basis states with the same total electron number, the same total spin, and conserves the valley. Because of Eq. (15), the tunneling Hamiltonian reads as (cid:88) (cid:104) Htunnel = cos(θij) c † i,s,vcj,s,v tij with tij = tji. We define i,j,s,v + i sin(θij) c † i,s,vcj,s,¯v tL = cos(θLC) t12, t(cid:48) L =i sin(θLC) t12, tR = cos(θRC) t23, R = − i sin(θRC) t23, t(cid:48) t13 = cos(θLR) t13, 13 = − i sin(θLR) t13. t(cid:48) (cid:105) , (16) (17) (18) (19) (20) (21) (22) The tunnel barriers are assumed to be adjusted such that the hopping matrix elements between the ground states are equal in strength, i.e., tL = tR. Note that, to warrant a unique stationary solution (see below), we chose the valley phases θ12 (cid:54)= n1 π 2 and θ23 (cid:54)= n2 π 2 with integer n1, n2. Because of the linear alinement of the TQD direct hopping between the left and the right dot becomes negligible, thus, we set t13 = 0. As a consequence the valley phase θLR becomes undetectable. 1.2. Occupation probabilities In order to calculate the occupation probabilities of the dots in the stationary state, we assume that incoherent transitions can occur between the eigenstates of H, both internally and via electron hopping between the TQD and the leads. These incoherent interactions with the environment can be taken into account with the Lindblad master equation (23) where  is the reduced Planck constant and ρ is the density matrix. The dissipative part D(ρ) consists of the following terms ρ = − i [H, ρ] + D(ρ), (cid:17) (cid:3)(ρ) (cid:3)(ρ) + D(cid:2)ci,s,v † i,s,v (cid:88) λ,λ(cid:48) + λλ(cid:48)D(cid:2)bλλ(cid:48)(cid:3)(ρ), τ−1 (24) D(ρ) = (cid:88) (cid:16) D(cid:2)c Γi v=±,s=↑↓ i=L,R 7 Valley-resolved spectroscopy in Si TQDs Here D(cid:2)O(cid:3)(ρ) = O†ρO − (ρOO† + OO†ρ)/2 is the usual Lindbald super operator, Γi † is the transition rate from the lead i = L, R to the dot i, and the operators c i,s,v, and ci,s,v create and annihilate an electron in dot i and valley v with spin s, respectively. The first and second terms of (24) correspond to the flow of electrons from lead i = l, r into dot i = L, R and in the opposite direction, out of the dot to the lead. The third term in (24) describes excitations within the TQD, i.e. incoherent interactions with a bosonic bath, such as phonons, that can induce transitions from one eigenstate of H to the other with the same total number of electrons in the dots with the same total spin. The operator bλλ(cid:48) = λ(cid:105)(cid:104)λ(cid:48) takes the system from an initial state λ(cid:48)(cid:105) to a final state λ(cid:105) with a transition rate τ−1 λλ(cid:48). We assume that the level broadenings, caused by the interaction with the leads and the bosonic bath, are smaller than the level splittings between the eigenstates of H which we ensure by an external magnetic field B. This is the so-called secular approximation [43], which results in a steady-state density matrix ρ diagonal in the eigenbasis of H. This significantly simplifies the Lindblad equation (23), where the commutator vanishes and after taking the matrix representation of the remaining dissipative term in the eigenbasis of H, we obtain Redfield equations for the diagonal elements of the steady-state solution Γj (ρmcmjn − ρncnjm + ρmcnjm − ρncmjn) + nmρm − τ−1 mnρn (cid:88) m j=L,R 0 = ρn = (cid:0)τ−1 (cid:88) m(cid:54)=n (cid:1) , of the density matrix ρ, and cmjn =(cid:80) where m, n runs over all diagonal elements of ρ. The terms in (25) are approximations of their respective counterparts in (24). Here ρn ≡ (cid:104)n ρn(cid:105) is the n-th diagonal element v,s (cid:104)m cj,s,v n(cid:105)2, which can be finite only if there is one more electron in state n(cid:105) than in m(cid:105). We can extend this description toward finite temperatures in the leads with the following replacement rules in Eq. (25) (25) ρmcmjn → ρmcmjnnnjm ρncmjn → ρncmjnnmjn, (cid:0)Em − En + (νm − νn)eVj, T(cid:1), Em and En are the eigenenergies, νm (26b) (26a) where nmjn = nFD and νn the number of electrons in the given eigenstates of H, and nFD(δEj, T ) = 1 exp(δEj/(kBT )) + 1 (27) is the Fermi-Dirac distribution of the electrons in the lead j, with kB and T being the Boltzmann constant and the electron temperature. To include finite temperature effects in Eq. (25) the τ−1 redefined as τ−1 mn = γmn nBE(Em − En, T ) with the temperature dependent prefactor mn transition rates in (25) are nBE(δE, T ) = 1 exp(δE /(kBT )) − 1 + Θ(−δE) (28) Valley-resolved spectroscopy in Si TQDs 8 which accounts for the Bose-Einstein statistics of the environmental thermal bath, that is assumed to be in thermal equilibrium with the electronic system and having an approximately constant density of states in the relevant energy window of the transitions. Moreover, Θ(·) denotes the Heaviside step function. We use the following phenomenological model to describe incoherent decay from m(cid:105) → n(cid:105) with rate γmn = (cid:104)m· Γeff n(cid:105)· (29) where n(cid:105) denotes the eigenstate of the unperturbed system given in Eq. (9) with eigenenergy En. In our model m(cid:105)· denotes the absolute-valued eigenvector obtained by taking the absolute value of each entry in m(cid:105) in the eigenbasis of H and the matrix elements of the effective decay rate read , Γeff,pq = γc (cid:104)p (cid:88) +γv (cid:104)p (cid:88) +γs (cid:104)p (cid:88) s,v s,v i−j=1 i−j≤1 s,v,v(cid:48) i−j≤1 i,s,vcj,s,v q(cid:105) † c i,s,vcj,s,v q(cid:105) † c i,s,vcj,s,v(cid:48) q(cid:105) , † c (30a) (30b) (30c) with i, j ∈ L, C, R, v = ±, s =↑,↓, and v (s) being the flipped valley (spin). Here, γc denotes the pure charge relaxation rate and γv describes the relaxation rate involving a valley flip. We neglect any spin-related decays, γs = 0 due to the long spin-flip time on the order of milliseconds. Because of γc (cid:29) γv a decay channel, where both the charge and the valley changes, is always limited by the smaller decay rate and the valley decay γv serves as a bottleneck of the process. The matrix elements are between eigenstates p(cid:105) ,q(cid:105) of H0 = Hcharge + HZeeman + Hvalley. Eq. (25) can cast into a more concise form, which also reflects the temperature dependence 0 = ρn = + (cid:88) (cid:88) m(cid:54)=n m(cid:54)=n (cid:0)τ−1 (Γnmρm − Γmnρn) nmρm − τ−1 mnρn (cid:1) , (31) where the total decay rate Γmn of the state n(cid:105) to state m(cid:105) with one electron hopping on or off the TQD is given by Γmn = Γj (cmjn + cnjm) nmjn. (32) j=L,R Note that depending on the direction of the hopping, either cmjn or cnjm will be zero. This set of classical rate equations can also be formulated as a matrix equation (cid:88) Valley-resolved spectroscopy in Si TQDs 9 M ρ = 0, where ρ is a vector of the diagonal elements of the density matrix ρ. The steady-state solution in the secular approximation is thus provided by the nullspace of the matrix M , as a normalized vector of the probabilities ρk for finding the system in its kth eigenstate. If the calculation of the nullspace of M does not return the expected, physically meaningful result because of numerical inaccuracies, then the direct integration of Eq. (31) with the initial condition of a thermal distribution can deliver the correct solution. 1.3. Input-ouput theory For read-out of the energies in the system, one can directly connect the oscillating voltage generated inside the microwave resonator to one of the gate potentials [see Fig. 1 (a)]. The response of the system to a microwave probe field due to this electric dipole coupling can be determined using cavity input-output theory [44]. We assume that the microwave field can induce transitions between all energy levels of the TQD, whereby transitions between neighboring energy levels are more likely for low temperatures and bias voltages. Following the calculation given in Refs. [45, 36, 46, 47] the transmission coefficient A of the output signal for the TQD is √ −i ωres − ωP − iκ/2 + gc (cid:80)N κ1κ2 (cid:80)N A = m=1 n=1 dm,nχm,n . (33) The electric susceptibility of the TQD is given by (34) χn,m = . mn/2) mn = γdep mn + τ−1 Em − En − ωP − i(γ∗ −gcdm,n(ρn − ρm) (cid:80) Here dn,m is the dipole matrix element pertaining to the n → m transition, τ−1 mn is i(∂(Em − En)/∂Vi)/α describes the relaxation rate [see Eq. (29)], and γ∗ pure dephasing with rate γdep due to charge noise [20]. The total cavity decay rate is κ = κ1 + κ2 + κi, where κ1 and κ2 are the photon decay rates through the input and output ports and κi is the intrinsic photon decay rate. The probe frequency and the cavity resonance frequency are denoted by ωP and ωres, and gc (also commonly known as gc = 2g0) is the electric dipole coupling strength. The charge noise is coupled through dotCgate. The summation the electrodes to the electrons via the lever arm matrix α = eC−1 in Eq. (33) runs over all the possible transitions within the N -electron states, and the eigenstates are indexed with increasing eigenenergies En. The dipole matrix can be calculated easily in the basis of H, by taking the derivative D = ∂H(V1, V2, V3) ∂V2 , (35) with gate V2 connected to the resonator [see Fig. 1 (a)]. The dipole matrix elements are then accordingly defined as dm,n = (cid:104)mD n(cid:105) . (36) Valley-resolved spectroscopy in Si TQDs 10 2. Results Our goal is to extract information about the energetic structure, in particular the valley splitting and valley phase, of the triple quantum dot system from a measurement of the output signal of the microwave resonator. We expect, in analogy to Ref. [36], that the finite dipole moment at avoided crossings in the triple quantum dot system yields measurable features in the output signal. Ideally, the location of these features as a function of detuning parameters allows us to reconstruct the energy spectrum of the triple dot. In order to limit the number of anti-crossings we first analyze the case of a single electron in the triple dot. Afterwards, we use the collected information to interpret the case of three electrons which has broad interest due to the realization as an exchange-only qubit [20]. We further consider a homogeneous magnetic field with Zeeman spin splitting Z = EZ = 0.3 meV (corresponding to ≈ 2.6 T in silicon) larger than typical valley Ei splittings EZ > Ei V for i = L, C, R in SiGe quantum dots. The presence of a magnetic field allows us to ignore the spin degree of freedom and focus solely on valley physics. The remaining simulation parameters are listed in appendix Appendix C. 2.1. One electron in a triple quantum dot Considering a single electron in the TQD the total Hamiltonian reads H1e = Hcharge,1e + Hvalley,1e + HZeeman,1e + Htunnel,1e (37) which can be obtained from Hamiltonian (9) using N = 1. The charge Hamiltonian (here in the basis L(cid:105) ,C(cid:105) ,R(cid:105)) ε − εM 3 0 0 Hcharge,1e =  + 0 2εM 0 0 3 0 −ε − εM 3 εg 2 13 (38) contains the electrostatic interactions from the capacitor model. The two detuning parameters are defined as ε =(µL − µR)/2 εM =µC − (µL + µR)/2 (39) (40) where µi is the chemical potentials of quantum dot i = L, C, R given in Eqs. (6)-(8) with N = (0, 0, 0). The average energy in the TQD is then given by εg = (µC + µL + µR)/3. (41) Through a variation of εg the total amount of electrons inside the TQD can be adjusted. Furthermore, we introduce two additional detuning parameters εL =(µL − µC)/2, εR =(µR − µC)/2. (42) (43) Valley-resolved spectroscopy in Si TQDs 11 Figure 2. (a) Calculated phase response ∆Φ = arg(A) of the probe signal of the TQD filled with a single electron coupled to the microwave cavity in thermal equilibrium at T = 1 K and without applied voltage bias ∆V = 0 as a function of the TQD detuning parameters ε and εM . Here (nL, nC, nR) denotes the occupation of dot i by ni with i = L, C, R electrons. The left and right white lines are cuts along the double quantum dot (DQD) detuning parameters εL and εR while keeping εg and the respective orthogonal detuning parameter fixed. This allows for the investigation of the (1,0,0)-(0,1,0) and (0,0,1)-(0,1,0) charge transition. (b) Phase response of the probe signal of the TQD coupled to the microwave cavity for T = 30 mK and applied bias ∆V = 0.3 mV. Similar features are visible as for the high-temperature phase response with zero bias. (c) Cut along εL (white line in (a)) and the energy of the four lowest eigenstates, E1, E2, E3, E4, is plotted as function of εL. The peaks
1 L, and L in the phase response correspond to an anti-crossing between states E2(cid:105) ↔ E3(cid:105),
4 E1(cid:105) ↔ E2(cid:105), E3(cid:105) ↔ E4(cid:105), and E2(cid:105) ↔ E3(cid:105). (d) Cut along εR [white line in (a)]. For identification of the avoided crossings also the energy of the four lowest eigenstates, E1, E2, E3, E4, is plotted as function of εR. For convenience the E2 + ωres (black- dashed) is also shown. The peaks
3 R in the phase response correspond to an anti-crossing between states E1(cid:105) ↔ E2(cid:105) and E3(cid:105) ↔ E4(cid:105). The very sharp peaks R correspond to the condition E3 − E2 = ωres (crossings of
1 R and
2 black-dashed line). The void area between the (0,1,0) and (0,0,1) originated from a steady state with a completely depleted triple quantum dot. R and
6 R and
5 L,
2 L,
3 R and
4 Valley-resolved spectroscopy in Si TQDs 12 These two detuning parameters have two implications. Firstly, εL and εR allow for a simple investigation of the (1,0,0)-(0,1,0) and (0,0,1)-(0,1,0) charge transitions. At these transitions the TQD behaves like a DQD with one charge state highly separated in energy. This effectively reduces the system to a conventional charge qubit. Secondly, unlike the set, ε, εM , and εg, the set εL, εR, and εg does not form an orthogonal set. Therefore, it is impossible to sweep through the left charge qubit along εL while keeping the average energy εg and the right-center detuning εR constant. Respective cuts along εL and εR seem to be non-orthogonal to the respective charge transition in (ε, εM ) space [see Fig. 2 (a)]. V − EC V V − EC L =(cid:0)EL (cid:1) /4, and
4 L =(cid:0)EL R = (cid:0)ER A cut along the right-center detuning
R shows a very similar phase response 2.1.1. Zero bias The valley degeneracy effectively creates two copies of the charge states which are coupled by the valley non-conserving tunnel amplitudes. Therefore, instead of a single anti-crossing between charge states we expect to see four anti- crossings [36]. In Fig. 2 (a) the phase shift of the cavity signal for the single electron is shown as a function of the two detuning parameters ε, εM . At the (1,0,0)-(0,1,0) and the (0,0,1)-(0,1,0) charge transitions we see the splitting of a single line into three and five distinct lines. A cut along the left-center detuning εL shows in comparison to the level diagram that the phase responses directly match the corresponding valley splittings [see Fig. 2 (c)]. We observe a phase response (peak) at
1
2 L = −(cid:0)EL (cid:1) /4. L =(cid:0)EL (cid:1) /4,
3 R = −(cid:0)ER (cid:1) /4. However, there is no phase response at εR = ∓(cid:0)ER [see Fig. 2 (d)]. We observe a phase response (peak) at
3
4 but instead two phase responses (each a sharp dip followed by a sharp peak) at R = −58.8 µeV,
2
1 R = 58.3 µeV (simulation parameters are listed in appendix Appendix C). This splitting into two signals appears if the energy splitting at the avoided crossing is smaller than the resonator frequency, 2t(cid:48) R < ωres. The microwave resonator becomes resonant with the ground-state excited- state transition ωres = E3 − E2 at exactly two points [see crossing between dashed and solid lines in Fig. 2 (d)]. For small tunnel couplings t(cid:48) V /4 the left anti- crossing between the first and second excited state in Fig. 2 (d) can be approximated by an isolated two-level system with energy splitting (cid:1) /4, (cid:1) /4 and (cid:1) /4 R = −50.1 µeV and
5 R = 49.6 µeV,
6 R (cid:28) EL V V + EC V V − EC V − EC V V V + EC V + EC V V + EC V (cid:113) ∆ER,1 = 2 (εR −
1 R)2 + t(cid:48) R2, (44) V + EC V R is the position in εR detuning space. From the equation above it follows that where
1 (
1 R +
2 between the first and second excited state at (
5 (cid:1) /4. Similarly, we find the position of the right anti-crossing R)/2 = −(cid:0)ER (cid:1) /4. V = 118.6 µeV which are roughly 2% smaller than the input settings (cid:101)EL V = 100 µeV, and (cid:101)ER (cid:101)EC V = 98.2 µeV, and V = 80 µeV, V = 120 µeV. We attribute this small systematic error to a deformation of the energy levels due to the mixing of the different levels via tunneling. R +
6 V = 77.7 µeV, EC R)/2 =(cid:0)ER In total we extract the valley splittings EL V + EC V ER Valley-resolved spectroscopy in Si TQDs 13 To mitigate these kind of errors the cuts along εL and εR can be performed further away from the triple point, ε = εM = 0 where all three charge configurations intersect. Note that we assumed an electron temperature T = 1 K to occupy the excited states and see features of the excited valley states in Fig. 2 (a). The phase response of the cold simulation with T = 30 mK but applied bias ∆V = 0.3 meV between the two leads shows similar features [see Fig. 2 (b)] in the vicinity of the triple point. However, there is only a small energy window in which a finite charge current is possible [40] and higher lying valley states have a finite occupation probability. At the relevant (1,0,0)-(0,1,0) and the (0,0,1)-(0,1,0) charge transitions the charge current is blocked suppressing any probe signal from higher states (see appendix Appendix B). An alternative measurement scheme for small temperature is discussed in the next subsection. L,
2 L,
3 L, and
4 The extraction of the valley phase is a more challenging task. Following the procedure in Ref. [32] the valley phase can be estimated by fitting the amplitudes of the phase response for the avoided crossings at
1 L in Fig. 2 (c) to the tunnel couplings tL and t(cid:48) L. Unfortunately, the fitting includes two more unknown parameters (taking into account charge noise) making the fits hard and unstable. Furthermore, this j > ωres with j = L, R to gain a single method requires large tunnel couplings 2tj, t(cid:48) response signal which for our parameter setting is not fulfilled for the (0,1,0)-(0,0,1) charge transition. Then the tunnel coupling strength tR and t(cid:48) R can be extracted by fitting to the energy gap. For small tunnel amplitudes Eq. (47) provides a sufficient approximation. Alternatively, for a frequency tunable resonator [35] the tunnel couplings can be extracted using spectroscopy by observing the splitting of the signal into two signals mentioned above. Using Eqs. (17)-(20) the two valley phases are given by θLC = tan−1 t(cid:48) L/tL = 0.23 π and θRC = tan−1 t(cid:48) R/tR = −0.2 π. The methods introduced here do not provide a way to measure the valley angle between the left and right valley pseudo-spin θLR. In our simplified picture for the tunneling between the dots, a direct tunnel matrix element t13 between the left and right dot is set to zero which is close to the real scenario for a linear array. For a triangular geometry of the triple dot all tunnel matrix elements are finite and the remaining valley phase difference θLR can be directly measured by performing the same type of measurement to extract θLC and θRC at the (1,0,0)-(0,0,1) charge transition. This requires the comparison of the tunnel couplings t13 and t(cid:48) 13. Furthermore, we note that in the presence of a triangular geometry a closed path can give rise to a non- vanishing geometric phase. This can in principle also be used to probe the valley in a complementary way. Instead of the two-step measurement to extract 2.1.2. Finite bias at low temperature the energy spectrum from cuts through the (1,0,0)-(0,1,0) and (0,1,0)-(0,0,1) charge configurations discussed above, an investigation of the charge intersection point (1,0,0)- (0,1,0)-(0,0,1) yields the same information about the valley splitting. This is especially interesting for measurements at low temperatures since the fine-structure of cuts through (1,0,0)-(0,1,0) and (0,1,0)-(0,0,1) charge transitions is invisible in the spectroscopic data Valley-resolved spectroscopy in Si TQDs 14 ε EL V
1 Q Q − EL
2
3 Q EL V V EL V
4 Q Q − EL
5
6 Q ER V V V 4 V 4 V 4 V 4 4 − ER 4 − ER 4 − ER 4 + ER 4 − ER 4 − EL 4 + ER 4 − EL V 4 V 4 V 4 V
7 Q
8 Q EL V ER V V V V V V 4 V 4 V 4 EC V EC V EC V εM 2 − EL 4 − ER 4 − ER 2 + EL 2 − EL 4 − ER − EC 2 − EL 4 + ER 4 − ER 2 + EL 4 + ER 2 + EL − EC 2 − EL 4 + ER − EC 4 + ER 2 + EL EC V V 4 V 4 V 4 V 4 V 4 V V V V V V V 4 − EC V Eex 0 EL V EC V ER V max(EL V , EC V ) max(EL V , ER V ) max(EC V , ER V ) max(EL V , EC V , ER V ) Table 1. Coordinates of the triple intersection points
n Q of (1,0,0)-(0,1,0)-(0,0,1) V , EC charge states in detuning space (ε, εM ) as a function of the valley splittings EL V , V . (Forth column) Estimated excitation energy Eex = E − EGS to populate the and ER state with respect to the ground state energy at the triple points in the absence of tunneling. The triple points are only visible in the output signal if there is a finite population of the corresponding states. due to the blocked charge current while current flow near the triple intersection points populates the necessary excited valley states (see appendix Appendix B). Considering the same setup as above there are n = 23 = 8 copies of the triple intersection point (1,0,0)-(0,1,0)-(0,0,1) due to the presence of the valley degree of freedom. The location of the intersection points
n Q in detuning space (ε, εM ) are shown in Table 1 together with an approximate energy necessary to populate the corresponding states. Each triple intersection point can be approximated for (ε, εM ) =
n Q by the three-level system with eigenenergies Q,1 = (cid:113)tn Q,3 = −(cid:113)tn Q,2 = 0, En En x2 + tn y2, x2 + tn y2, (45) (46) L} and tn x ∈ {tL, t(cid:48) Q,1(cid:105) and En En y ∈ {tR, t(cid:48) (47) R} depending on the intersection point. Close to where tn these points the three-level system forms a coupled two-level system between the states Q,i(cid:105) denotes En Q,3(cid:105) with the third state En the eigenstate with eigenenergy En Q,i. The three-level system posses a large electric Q,3(cid:105) [48, 49]; all dipole moments are quadrupole between the eigenstates En suppressed due to symmetry. Therefore, a symmetric architecture of the TQD resonator system, i.e, connecting the resonator via the center gate, is advantageous for probing Q,3 ≈ these triple points. The probe frequency is ideally set to (cid:101)ωP = (cid:101)ωres = En Q,2(cid:105) lying in the middle, where En Q,1(cid:105) and En Q,1 − En Valley-resolved spectroscopy in Si TQDs 15 detuning parameters ε and εM . The resonator and the probe frequency (cid:101)ωP = and (cid:101)ωres = Figure 3. Calculated phase response ∆Φ = arg(A) of the probe signal of the TQD filled with a single electron coupled to the microwave cavity for T = 30 mK and applied voltage bias (a) ∆V = 0.3 mV and (b) ∆V = −0.3 mV as a function of the TQD 2ωP 2ωres are adjusted to probe the charge quadrupole transitions [48, 49] Q (see V , and at the triple intersection points
i Table 1) are sufficient to extract the values of all three valley splittings EL ER V . Q (yellow). The positions of
1 Q, and
7 V , EC √ √ Q,
2 Q,
4 √ 2ωP . Fig. 3 shows the phase response of the probe signal in the vicinity of triple points for (a) ∆V = 0.3 mV and (b) ∆V = −0.3 mV. For an extraction of all three valley splittings a minimum of three triple points are necessary. If not enough features are visible in the phase response reversing the direction of the charge current helps to locate the position of missing triple intersection points. We see clearly a response in the phase Q = (49,−39) µeV. at
1 In total we extract with this method the valley splittings EL V = 100 µeV, and ER V = 80 µeV, V = 120 µeV. We again attribute this error to a deformation of the energy levels due to the mixing of the different levels via tunneling and the broadening of the response signal. Q = (−12, 0) µeV,
2 V = 122.2 µeV which are roughly 3% larger than the input settings (cid:101)EL (cid:101)EC V = 100 µeV, and (cid:101)ER Q = (−51, 41) µeV,
4 Q = (49, 61) µeV, and
7 V = 82.4 µeV, EC 2.2. Three electrons in a triple quantum dot In practice studying the three-electron case is more interesting since it allows one to measure the valley splitting and valley phase in the same charge configuration regime spin qubits can be implemented, i.e., three spin- 1 2 qubits or a exchange-only qubit are implemented in the (1,1,1) charge regime. The total Hamiltonian of the three-electron case is given by H3e = Hcharge,3e + Hvalley,3e + HZeeman,3e + Htunnel,3e (48) Valley-resolved spectroscopy in Si TQDs 16 which can be obtained from the Hamiltonian (9) with N = 3. Focusing only on the (2,0,1), (1,1,1), and (1,0,2) charge configuration regime where the resonant exchange (RX) qubit is typically realized, the charge Hamiltonian can be simplified to  ε − εM 0 0 0 0 0 0 0 −ε − εM  + 3εg13 Hcharge,3e = (49) containing the electrostatic interactions from the capacitor model. The detuning parameters ε and εM are (up to a constant energy shift) identical to Eqs. (39) and (40). We choose the average detuning εg such that the TQD is occupied by three electrons. The left-center and right-center detuning parameters, εL and εR, then allow us to investigate the (2,0,1)-(1,1,1) and (1,0,2)-(1,1,1) charge transitions. Analogously to the single-electron case, the dynamics is effectively reduced to an DQD filled with two electrons. 2.2.1. Extracting the valley splitting and phase The valley degeneracy effectively creates eight copies of the charge states, two from the valley DOF in each dot for the (1,1,1) configuration and two copies for the (2,0,1) and (1,0,2) configuration neglecting the spin. These states are coupled by the valley non-conserving tunnel matrix elements. Therefore, instead of a single anti-crossing between charge states we expect to see (in the ideal case) 16 anti-crossings between the (2,0,1)-(1,1,1) and (1,0,2)-(1,1,1) charge states. Of course, to observe all crossings requires a temperature or bias such that the excited states are populated. In Fig. 4 (a) and (b) the phase shift of the cavity signal for three electrons is shown as a function of the two detuning parameters ε, εM . At the (2,0,1)-(1,1,1) and the (1,0,2)-(1,1,1) charge transitions we could potentially see the splitting of a single line into multiple lines. The asymmetry in brightness between the (2,0,1)-(1,1,1) and the (1,0,2)-(1,1,1) charge transitions is due to different energy detunings ∆ = (E2 − E1) − ωres for the left and right charge transitions. A cut along the left-center detuning εL provides information about the level splittings [see Fig. 4 (c)]. The ground state in the (1,1,1) regime is a polarized valley state, where all electrons occupy the lower valley state, and in the (2,0,1) charge regime the two electrons form a valley-singlet state and the remaining electron in the right dot occupies the ground state. The respective energy level crossing occurs at εL = −(EL L ≈ 0, and V )/4. From Fig. 4 (c) we find
1
1 L = (EL V )/4 which are all consistent with the extracted valley splitting in the single electron case. V + EC V + EC L = −(EL L ≈
3 V + EC V )/4,
2 A cut along the right-center detuning εR between (1,0,2) and (1,1,1) charge states shows similar features [see Fig. 4 (d)]. The respective energy crossing occurs at εR = −(ER R. From Fig. 4 (d) V )/4 and we find again two surrounding peaks at
1 R and
2 V + EC Valley-resolved spectroscopy in Si TQDs 17 Figure 4. (a) Calculated phase response ∆Φ = arg(A) of the probe signal of the TQD coupled to the microwave cavity for T = 1 K and no applied voltage bias ∆V = 0. Here (nL, nC, nR) denotes the occupation of dot i = L, C, R with ni with electrons. The left and right white lines are cuts along the double quantum dot (DQD) detuning parameters εL and εR while keeping εg and the respective orthogonal detuning parameter fixed. (b) Calculated phase response for T = 30 mK and applied bias ∆V = 0.3 mV. This allows for the investigation of the (2,0,1)-(1,1,1) and (1,0,2)- (1,1,1) charge transition. (c) Cut along εL [white line in (a)] and the energy of the ten lowest eigenstates, Ei, is plotted as function of εL. (d) Cut along εR [white line in (a)]. For interpretation also the energy of the ten lowest eigenstates are plotted as function of εR. The sharp peaks
1 R correspond to the condition E2 − E1 = ωres. R and
2 we find −(ER This matches with the results in the single electron case. R ≈ 0, and (ER R ≈
4 V + EC V )/4 = (
1 R +
2 R)/2,
3 V + EC V )/4 = (
5 R +
6 R)/2. Unfortunately, further energy crossings are hardly visible for our choice of simulation parameters in the case of three electrons, thus, we refrain from a further analysis. 3. Conclusion In this paper we have theoretically investigated the response signal of a probed microwave resonator coupled to a linearly arranged triple quantum dot via the center Valley-resolved spectroscopy in Si TQDs 18 dot gate electrode. A realistic model of the TQD is used in our analysis which includes electrostatic cross-talk between the dots and gates via a capacitor model, valley and spin effects, and the solution of a Redfield master equation to find the occupation probabilities. We show that a setup consisting of a TQD filled with a single electron can be used to extract important information from the TQD system such as the valley splitting and the valley phase. The accuracy of the extracted valley splitting and phase becomes higher and the interpretation simpler if the TQD is detuned such that one chemical potential is significantly increased which reduces the triple dot system to an effective double dot. A setup consisting of three electrons in a triple quantum dot is in principle capable to deliver the same information but the larger number of energy levels makes the population of the relevant excited valley states and the corresponding interpretation of the signal more difficult. Acknowledgments We acknowledge funding from ARO through Grant No. W911NF-15-1-0149 and the DFG through SFB 767. We thank M. Benito and F. Ginzel for helpful discussions. We thank J. Petta for providing us with experimental datasets. Appendix A. Secular approximation In order to compute the occupation of the energy levels we relied on the secular approximation. However, since we have no all-to-all coupling there are energy levels which do not form an anti-crossing. At these points the energy splitting goes to zero, Ei − Ej → 0, thus, violating the secular approximation. The validity of our calculation, however, is unaffected since the ratio between the number of valid points NG and detected violations NF is small for large sample sizes NS, NG/NF (cid:29) 1. In all simulation we have used NS = 6002 sample points. Appendix B. Charge current As discussed in Ref. [36], excited energy states required for read-out of all relevant system parameters can be populated either by increasing the temperature in the system or by applying a dc bias voltage. While precise control over the temperature is experimentally challenging, biasing the left and right leads is not. The charge current from left to right can be given in two equivalent forms due to continuity (cmLn − cnLm) nnLmρm (cnLm − cmLn) nnLmρm, (B.1a) (B.1b) (cid:88) (cid:88) m(cid:54)=n I = eΓL = eΓR m(cid:54)=n where summations run for all m and n. The expressions for a charge current from right to left is similar. Fig. B1 shows the charge current for ∆V = ±0.3 mV. A finite Valley-resolved spectroscopy in Si TQDs 19 Figure B1. Calculated charge current I of the TQD coupled to the microwave cavity for T = 30 mK and applied voltage bias (a) ∆V = 0.3 mV and (b) ∆V = −0.3 mV. Here (nL, nC, nR) denotes the occupation of dot i = L, C, R with ni with electrons. The left and right white lines are cuts along the double quantum dot (DQD) detuning parameters εL and εR while keeping εg and the respective orthogonal detuning parameter fixed. The black dots mark the triple intersection points
i Q (black). Note, that in (b) the charge direction is reversed. current is only possible at charge quadruple points [40] where four charge configurations intersect which in our case is in the vicinity of the triple intersection points
n Q. Appendix C. Simulation parameters For the simulation in the main text we use the following parameters from experiments in undoped Si/SiGe performed in a triple quantum dot using the gate layout described in Ref. [50]. The extracted capacitance matrix consisting of the electrostatic capacitances between the dots reads and the extracted capacitance matrix consisting of the electrostatic capacitances between the dots and the gates reads     . Cdot = 56.2 −5.5 −0.5 −5.5 50.5 −11.7 −0.5 −11.7 59.4 Cgate = −6.9 −2.4 −0.3 −0.15 −5.9 −0.03 −0.4 −3.6 −6.9 (C.1) (C.2) -���-������-�������-���-������-�������������������Valley-resolved spectroscopy in Si TQDs 20 The capacitance matrix consisting of the electrostatic capacitances between the dots and the leads is set to (C.3)   . Clead = 40.6 0 13.6 13.6 0 36.4 All capacitances are given in units of (aF) attofarad. The remaining parameters for the simulation are the tunneling couplings, t12 and t23, between the dots, the valley-orbit parameters ∆j = Ej in each dot j = L, C, R, the incoherent decay rates γc and γv, and the charge dephasing rate γdep. The tunneling parameters used in all simulations in the main text are chosen to be t12 = 12.5 µeV and t23 = 11.5 µeV. For the valleys splittings we use EL V = 80 µeV, V = 120 µeV. The relative valley phases θLC = φL − φC = 0.23 π, V = 100 µeV, and ER EC θRC = φR− φC = −0.2 π, and θLR = φL− φR = 0 π, where the last phase is undetectable in a linear aligned triple quantum dot. The decay and dephasing rates are set to γc = 0.12 µeV, γc = 0.12 µeV, and γdep = 1.2 µeV. V eiφj Valley-resolved spectroscopy in Si TQDs 21 [1] Zwanenburg F A, Dzurak A S, Morello A, Simmons M Y, Hollenberg L C L, Klimeck G, Rogge S, Coppersmith S N and Eriksson M A 2013 Rev. Mod. Phys. 85(3) 961 [2] Watzinger H, Kukucka J, Vukusi´c L, Gao F, Wang T, Schaffler F, Zhang J J and Katsaros G 2018 Nature Communications 9 3902 [3] Eich M, Pisoni R, Pally A, Overweg H, Kurzmann A, Lee Y, Rickhaus P, Watanabe K, Taniguchi T, Ensslin K and Ihn T 2018 Nano Letters 18 5042 -- 5048 URL https://doi.org/10.1021/ acs.nanolett.8b01859 [4] Overweg H, Knothe A, Fabian T, Linhart L, Rickhaus P, Wernli L, Watanabe K, Taniguchi T, S´anchez D, Burgdorfer J, Libisch F, Fal'ko V I, Ensslin K and Ihn T 2018 Phys. Rev. Lett. 121(25) 257702 URL https://link.aps.org/doi/10.1103/PhysRevLett.121.257702 [5] Joyce B A 1993 Handbook on semiconductors, volume 1: Basic properties of semiconductors, 2nd ed. vol 5 URL https://onlinelibrary.wiley.com/doi/abs/10.1002/adma.19930051122 [6] Friesen M, Chutia S, Tahan C and Coppersmith S N 2007 Phys. Rev. B 75(11) 115318 [7] Culcer D, Cywi´nski L, Li Q, Hu X and Das Sarma S 2009 Phys. Rev. B 80(20) 205302 [8] Culcer D, Cywi´nski L, Li Q, Hu X and Das Sarma S 2010 Phys. Rev. B 82(15) 155312 [9] Veldhorst M, Ruskov R, Yang C H, Hwang J C C, Hudson F E, Flatt´e M E, Tahan C, Itoh K M, Morello A and Dzurak A S 2015 Phys. Rev. B 92(20) 201401 [10] Boross P, Sz´echenyi G, Culcer D and P´alyi A 2016 Phys. Rev. B 94(3) 035438 [11] Zimmerman N M, Huang P and Culcer D 2017 Nano Lett. 17 4461 -- 4465 [12] Gamble J K, Harvey-Collard P, Jacobson N T, Baczewski A D, Nielsen E, Maurer L, Montano I, Rudolph M, Carroll M S, Yang C H, Rossi A, Dzurak A S and Muller R P 2016 Applied Physics Letters 109 253101 [13] Tariq B and Hu X arXiv:1904.11944 URL https://arxiv.org/abs/1904.11944 [14] Loss D and DiVincenzo D P 1998 Phys. Rev. A 57(1) 120 [15] Veldhorst M, Yang C H, Hwang J C C, Huang W, Dehollain J P, Muhonen J T, Simmons S, Laucht A, Hudson F E, Itoh K M, Morello A and Dzurak A S 2015 Nature (London) 526 410 -- 414 [16] Zajac D M, Sigillito A J, Russ M, Borjans F, Taylor J M, Burkard G and Petta J R 2018 Science 359 439 -- 442 ISSN 0036-8075 URL http://science.sciencemag.org/content/359/6374/439 [17] Watson T F, Philips S G J, Kawakami E, Ward D R, Scarlino P, Veldhorst M, Savage D E, Lagally M G, Friesen M, Coppersmith S N, Eriksson M A and Vandersypen L M K 2018 Nature (London) 555 633 [18] Yang C H, Leon R C C, Hwang J C C, Saraiva A, Tanttu T, Huang W, Camirand Lemyre J, Chan K W, Tan K Y, Hudson F E, Itoh K M, Morello A, Pioro-Ladri`ere M, Laucht A and Dzurak A S arXiv:1902.09126 URL https://ui.adsabs.harvard.edu/#abs/2019arXiv190209126Y [19] Taylor J M, Srinivasa V and Medford J 2013 Phys. Rev. Lett. 111(5) 050502 [20] Russ M and Burkard G 2017 J. Phys. Condens. Matter 29 393001 [21] Sala A and Danon J 2018 Phys. Rev. B 98(24) 245409 URL https://link.aps.org/doi/10. 1103/PhysRevB.98.245409 [22] Russ M, Petta J R and Burkard G 2018 Phys. Rev. Lett. 121 177701 URL https://link.aps. org/doi/10.1103/PhysRevLett.121.177701 [23] Hada Y and Eto M 2003 Phys. Rev. B 68(15) 155322 URL https://link.aps.org/doi/10. 1103/PhysRevB.68.155322 [24] Lim W H, Zwanenburg F A, Huebl H, Mttnen M, Chan K W, Morello A and Dzurak A S 2009 Applied Physics Letters 95 242102 URL https://doi.org/10.1063/1.3272858 [25] Xiao M, House M G and Jiang H W 2010 Applied Physics Letters 97 032103 URL https: //doi.org/10.1063/1.3464324 [26] Lim W H, Yang C H, Zwanenburg F A and Dzurak A S 2011 Nanotechnology 22 335704 URL http://stacks.iop.org/0957-4484/22/i=33/a=335704 [27] Borselli M G, Ross R S, Kiselev A A, Croke E T, Holabird K S, Deelman P W, Warren L D, Alvarado-Rodriguez I, Milosavljevic I, Ku F C, Wong W S, Schmitz A E, Sokolich M, Gyure M F and Hunter A T 2011 Applied Physics Letters 98 123118 URL http://scitation.aip. Valley-resolved spectroscopy in Si TQDs 22 org/content/aip/journal/apl/98/12/10.1063/1.3569717 [28] Mi X, Cady J V, Zajac D M, Stehlik J, Edge L F and Petta J R 2017 Applied Physics Letters 110 043502 [29] Bruhat L E, Cubaynes T, Viennot J J, Dartiailh M C, Desjardins M M, Cottet A and Kontos T 2018 Phys. Rev. B 98(15) 155313 URL https://link.aps.org/doi/10.1103/PhysRevB.98.155313 [30] Samkharadze N, Bruno A, Scarlino P, Zheng G, DiVincenzo D P, DiCarlo L and Vandersypen L M K 2016 Phys. Rev. Applied 5(4) 044004 [31] Stockklauser A, Scarlino P, Koski J V, Gasparinetti S, Andersen C K, Reichl C, Wegscheider W, Ihn T, Ensslin K and Wallraff A 2017 Phys. Rev. X 7(1) 011030 [32] Mi X, P´eterfalvi C G, Burkard G and Petta J R 2017 Phys. Rev. Lett. 119(17) 176803 [33] Mi X, Benito M, Putz S, Zajac D M, Taylor J M, Burkard G and Petta J R 2018 Nature (London) 555 599 URL http://dx.doi.org/10.1038/nature25769 [34] Samkharadze N, Zheng G, Kalhor N, Brousse D, Sammak A, Mendes U C, Blais A, Scappucci G and Vandersypen L M K 2018 Science 359 1123 URL http://science.sciencemag.org/ content/359/6380/1123.abstract [35] Landig A J, Koski J V, Scarlino P, Mendes U C, Blais A, Reichl C, Wegscheider W, Wallraff A, Ensslin K and Ihn T 2018 Nature 560 179 URL https://doi.org/10.1038/s41586-018- 0365-y [36] Burkard G and Petta J R 2016 Phys. Rev. B 94(19) 195305 [37] Yang C H, Lim W H, Lai N S, Rossi A, Morello A and Dzurak A S 2012 Phys. Rev. B 86 115319 URL https://doi.org/10.1103/PhysRevB.86.115319 [38] Zajac D M, Hazard T M, Mi X, Nielsen E and Petta J R 2016 Phys. Rev. Applied 6(5) 054013 [39] van der Wiel W G, De Franceschi S, Elzerman J M, Fujisawa T, Tarucha S and Kouwenhoven L P 2002 Rev. Mod. Phys. 75(1) 1 -- 22 [40] Schroer D, Greentree A D, Gaudreau L, Eberl K, Hollenberg L C L, Kotthaus J P and Ludwig S 2007 Phys. Rev. B 76(7) 075306 URL http://link.aps.org/doi/10.1103/PhysRevB.76. 075306 [41] Ruskov R, Veldhorst M, Dzurak A S and Tahan C 2018 Phys. Rev. B 98(24) 245424 URL https://link.aps.org/doi/10.1103/PhysRevB.98.245424 [42] Tagliaferri M L V, Bavdaz P L, Huang W, Dzurak A S, Culcer D and Veldhorst M 2018 Phys. Rev. B 97(24) 245412 URL https://link.aps.org/doi/10.1103/PhysRevB.97.245412 [43] Breuer H P and Petruccione F 2007 The Theory of Open Quantum Systems (OUP Oxford) ISBN 978-0-19-921390-0 [44] Collett M J and Gardiner C W 1984 Phys. Rev. A 30 1386 -- 1391 [45] Kulkarni M, Cotlet O and Tureci H E 2014 Phys. Rev. B 90(12) 125402 URL https://link. aps.org/doi/10.1103/PhysRevB.90.125402 [46] Benito M, Mi X, Taylor J M, Petta J R and Burkard G 2017 Phys. Rev. B 96(23) 235434 [47] Kohler S 2018 Phys. Rev. A 98(2) 023849 URL https://link.aps.org/doi/10.1103/PhysRevA. 98.023849 [48] Friesen M, Ghosh J, Eriksson M A and Coppersmith S N 2017 Nat. Commun. 8 15923 [49] Koski J V, Landig A J, Russ M, Abadillo-Uriel J C, Scarlino P, Kratochwil B, Reichl C, Wegscheider W, Burkard G, Friesen M, Coppersmith S N, Wallraff A, Ensslin K and Ihn T arXiv:1905.00846 URL https://ui.adsabs.harvard.edu/abs/2019arXiv190500846K [50] Zajac D M, Hazard T M, Mi X, Wang K and Petta J R 2015 Applied Physics Letters 106 223507 

1208.1464

1

1208

Size Quantization in Planar Graphene-Based Heterostructures: Pseudospin Splitting, Interface States, and Excitons

A planar quantum-well device made of a gapless graphene nanoribbon with edges in contact with gapped graphene sheets is examined. The size-quantization spectrum of charge carriers in an asymmetric quantum well is shown to exhibit a pseudospin splitting. Interface states of a new type arise from the crossing of dispersion curves of gapless and gapped graphene materials. The exciton spectrum is calculated for a planar graphene quantum well. The effect of an external electric field on the exciton spectrum is analyzed.

cond-mat.mes-hall

cond-mat

ISSN 1063-7761, Journal of Experimental and Theoretical Physics, 2012, Vol. 114, No. 3, pp. 511 -- 527. c(cid:13) Pleiades Publishing, Inc., 2012. Original Russian Text c(cid:13) P.V. Ratnikov, A.P. Silin, 2012, published in Zhurnal ´Eksperimental'noı i Teoreticheskoı Fiziki, 2012, Vol. 141, No. 3, pp. 582 -- 601. ELECTRONIC PROPERTIES OF SOLID Size Quantization in Planar Graphene-Based Heterostructures: Pseudospin Splitting, Interface States, and Excitons P. V. Ratnikov∗ and A. P. Silin Lebedev Physical Institute, Russian Academy of Sciences, Leninskiı pr. 53, Moscow, 119991 Russia ∗e-mail: ratnikov@ lpi. ru Received April 21, 2011; in final form, August 1, 2011 Abstract -- A planar quantum-well device made of a gapless graphene nanoribbon with edges in contact with gapped graphene sheets is examined. The size-quantization spectrum of charge carriers in an asym- metric quantum well is shown to exhibit a pseudospin splitting. Interface states of a new type arise from the crossing of dispersion curves of gapless and gapped graphene materials. The exciton spectrum is calcu- lated for a planar graphene quantum well. The effect of an external electric field on the exciton spectrum is analyzed. DOI: 10.1134/S1063776112020094 1. INTRODUCTION The creation of graphene, a monolayer of carbon atoms forming a regular hexagonal lattice [1 -- 3], has stimulated extensive experimental and theoretical stud- ies along various lines of research. Graphene's unique properties make it a promising material for a new gen- eration of carbon-based nanoelectronic devices. In par- ticular, carrier mobility in graphene amounts to 2×105 cm2/V·s, and ballistic transport is possible on a sub- micrometer scale [4, 5]. Over the past seven years, numerous theoretical and experimental results have been reported on elec- tronic properties of nanometer-wide ribbons of graphene (nanoribbons). Among the first were studies of elec- tronic states of graphene nanoribbons using the Dirac equation under appropriate boundary conditions [6, 7]. The electronic properties of a graphene nanoribbon strongly depend on its size and edge geometry [8]. In terms of transport properties, graphene nanoribbons are highly reminiscent of carbon nanotubes [9, 10] since free carrier motion inside them is also one-dimensional. A field-effect transistor (FET) based on a 2 nm wide and 236 nm long graphene nanoribbon was fabri- cated in a recent study [11] (nanoribbons of widths be- tween 10 and 60 nm were also studied). The graphene nanoribbon was made narrow enough to open a gap of width required for room-temperature transistor oper- ation. However, it is less compact than the graphene quantum-dot transistor 30 nm in diameter discussed in [12]. In this study, we examine a planar quantum-well device made of a graphene nanoribbon whose edges are in contact with gapped graphene sheets. 1 A bandgap opening in graphene can be induced by several methods. First, graphene can be deposited on a hexagonal boron nitride (h -- BN) substrate instead of a silicon-oxide one. This makes its two triangu- lar sublattices nonequivalent, inducing in a bandgap of 53 meV [13]. Second, epitaxial graphene grown on a silicon-carbide substrate also has a nonzero bandgap [14]. According to angle-resolved photoemission data, a bandgap of 0.26 eV is produced by this method [15]. Third, a hydrogenated derivative of graphene synthe- sized recently, graphane [16], has been predicted to have a direct bandgap of 5.4 eV at the Γ point [17]. Fourth, ab initio calculations have shown that CrO3 adsorption on graphene induces a gap of 0.12 eV [18]. In the first two methods, a heterogeneous substrate can be used, such as an h -- BN -- SiO2 nanoribbon -- h -- BN or SiC -- SiO2 nanoribbon -- SiC one (Fig. 1a depicts a substrate with h -- BN). The last two meth- ods produce a graphene sheet containing a nanoribbon without hydrogenation (as the nonhydrogenated one in Fig. 1b) or a graphene strip without adsorbed CrO3 molecules, respectively. Furthermore, the bandgap can be varied by using partially hydrogenated graphene (where some carbon atoms are not bonded to hydrogen atoms). Combinations of these methods can also be employed. Extensive experimental studies of graphene on substrates made of various materials, including rare- earth metals, have been reported recently [19 -- 21]. It may be possible to open a bandgap via adsorption of other molecules on graphene or by using other materi- als as substrates. The use of gapped graphene to create potential barriers opens up additional possibilities for bandgap engineering in carbon-based materials [22]. We assume that both heterojunctions in the system 2 1 0 2 g u A 7 ] l l a h - s e m . t a m - d n o c [ 1 v 4 6 4 1 . 8 0 2 1 : v i X r a Fig. 1. Two configurations of the system under study: (a) graphene sheet on a substrate consisting of a SiO2 nanoribbon of width d inserted between h -- BN nanoribbons; (b) a graphene sheet on a SiO2 sub- strate containing a nonhydrogenated nanoribbon of width d, where open and closed circles are hydrogen atoms bonded to carbon atoms in different sublattices on opposite sides of the sheet, respectively. combining a nanoribbon with gapped graphene sheets are type I junctions (e.g., see [23] for classification of junctions); i.e., the Dirac points in gapless graphene fall within the bandgaps of the adjacent gapped graphene sheets. This prevents spontaneous electron -- hole pair creation, which would otherwise shunt graphene-based nanoelectronic devices such as FET. We believe that planar heterostructures made of gapless and gapped graphene are as prospective build- ing blocks in future carbon-based nanoelectronics. The use of only gapless graphene reduces the diverse op- portunities offered by bandgap engineering in gapped graphene. The paper is organized as follows. The preliminary remarks in Section 2 recall some aspects of the the- ory of quasi-relativistic fermions in graphene. Section 3 introduces a generalized Dirac model for graphene. Section 4 describes the size quantization of originally massless carrier states in a planar graphene quantum well. Interface states are examined in Section 5. An ef- fective bandgap opening in the size-quantization spec- trum of the heterostructure in question makes it pos- sible to generate excitons by optical pumping. The ex- citon spectrum is calculated in Section 6. In Section 7, the effect of an applied uniform electric field on an ex- citon is analyzed. Experimental manifestations of the effects examined in this paper are discussed in Section 8. The Conclusions section summarizes and discusses the main results obtained in this paper (Section 9). 2. PRELIMINARY REMARKS In our previous study [24], we analyzed a planar heterostructure combining two narrow-gap semiconduc- tors with a graphene ribbon. We considered states with √ lattice vectors b1 = (2π/3a, 2π/ Fig. 2. First Brillouin zone of graphene, with linear energy spectrum at the corners (Dirac points). The √ 3a) reciprocal and b2 = (2π/3a, −2π/ 3a), where a = 1.42 A is the lattice spacing, combined with the dashed lines equivalently represent the Brillouin zone as a rhombus. a definite parity λ, which is an eigenvalue of the parity operator [25] (cid:98)P = iγ4(cid:98)Λn. (1) Here, iγ4 is the inversion operator and 2 Σn = −iΣn is the operator of rotation by π about an n axis per- pendicular to the graphene plane. In standard repre- sentation, (cid:98)Λn = e−i π Σ =(cid:0) σ 0 (cid:1), γ4 ≡ β =(cid:0) I 0 (1) is analogous to the parity operator iγ5(cid:98)n in quantum electrodynamics, where(cid:98)n = γn, where I is the 2×2 unit matrix. It is clear that operator where σ denotes Pauli matrices, and γ =(cid:0) 0 σ−σ 0 (cid:1), γ5 = γ1γ2γ3γ4 = i(cid:0) 0 I (cid:1). I 0 0 σ (cid:1) 0 −I The eigenfunctions of this operator describe electron polarization states [26]. Charge carrier states in graphene can be described in terms of helicity define d as the eigenvalue of the operator (cid:98)h = σ · p/(2p). The projection of pseu- dospin on the direction of quasimomentum p indicates the valley in the Brillouin zone where electrons or holes belong (K or K(cid:48) point in Fig. 2). Positive helicity cor- responds to electrons and holes with wavevectors near the K and K(cid:48) points, respectively; negative helicity corresponds to electrons and holes with wavevectors near the K(cid:48) and K points, respectively [27]. Massless states with opposite helicities are decou- pled [28]. In addition, charge carriers have chiral sym- 2 metry (helicity is conserved), and parity can be de- fined for both massless and massive carriers1. In other words, a higher symmetry of massless charge carriers implies the existence of an additional quantum num- ber: helicity. Whereas parity distinguishes only be- tween the valleys where carrier states belong (λ = 1 and -- 1 for states close to the K and K(cid:48) points, re- spectively), helicity differs between a particle (electron) and an antiparticle (hole). However, chiral symmetry is broken for massive charge carriers (i.e., helicity is not a good quantum number any longer). Carrier states in a planar heterostructure combining gapless and gapped graphene should be characterized by parity. Recall that the Dirac equation describing massless carriers in graphene in terms of 4×4 matrices is de- rived by assuming that they are spinless and have two valley degrees of freedom [10]. When analysis is re- stricted to charge carriers in one valley, the Dirac equa- tion can be reduced to a 2×2 matrix representation by Weyl's equation for a massless fermion analogous to neutrino in two Euclidean dimensions. The car- rier energy spectrum with a pseudospin splitting in a planar heterostructure combining gapless and gapped graphene cannot be correctly analyzed in the 2×2 rep- resentation. For similar reasons, the representation of the Dirac algebra in terms of 2×2 matrices is not suf- ficient for describing the chiral symmetry breaking in quantum electrodynamics in two Euclidean dimensions [31]. Using the two-dimensional 4×4 Dirac equation to describe charge carriers in a graphene-based nanostruc- ture, we can study pseudospin effects following an ap- proach to narrow-gap semiconductor heterostructures based on the Dirac model [32]. This makes methods developed for solving problems in the spintronics of narrow-gap semiconductor heterostructures applicable to graphene-based ones [33 -- 40]. As applied to a planar graphene-based heterostruc- ture, these methods can be used to manipulate valley occupation in graphene via control of pseudospin. A heterostructure of this kind can be used in valleytron- ics [41 -- 47]. As a manifestation of pseudospin effect on charge carriers, we examine the pseudospin splitting of size-quantization spectrum (see Section 4 below). 3. MODEL To describe size quantization in graphene-based het- erostructures, an equation containing a mass term should be written for the envelope wavefunction. A bandgap opening in the energy spectrum of graphene results from the lack of symmetry between the two triangular sublattices of its hexagonal lattice. The corresponding tight-binding Hamiltonian taking into account nearest- neighbor hopping has the form [48] σ(B + di)bσ(B) + b† σ(B)aσ(B + di)(cid:3) (cid:98)H = −t (cid:2)a† (cid:88) B,i,σ 1Massless states can be characterized by two quantum num- bers: helicity and sign of energy or helicity and eigenvalue of the operator iγ5 [29]. Parity is analogous to the eigenvalue of iγ5. 3 Fig. 3. Part of hexagonal lattice, with highlighted vectors di from a B sublattice atom to the three nearest-neighbor A sublattice atoms. (cid:2)a† (cid:88) B,σ σ(B)bσ(B)(cid:3) , (2) + ∆ σ(B + d1)aσ(B + d1) − b† where t ≈ 2.8 eV is the nearest-neighbor hopping en- ergy; the sum runs over the position vectors B of all B sublattice atoms; the vectors di (i = 1, 2, 3) pointing from a B sublattice atom to the three nearest-neighbor A sublattice atoms are expressed in terms of the lattice spacing a as (cid:33) (cid:32) √ 3 2 1 2 a, d1 = a , d2 = (cid:32) (cid:33) √ a, − 1 2 3 2 a , d3 = (−a, 0) (see Fig. 3); σ =↑, ↓ is the (pseudo)spin index; aσ (a† σ) and bσ (b† σ) are the annihilation (creation) operators of A and B sublattice electrons, respectively; and the parameter ∆ quantifies the on-site energy difference between the two sublattices (setting ∆ = 0 restores the symmetry between sublattices so that graphene be- comes gapless, whereas nonzero ∆ equals the half-gap width in gapped graphene as shown below). Performing a Fourier transform, we change to the momentum representation (cid:90) (cid:90) ΩB aσ(A) = bσ(B) = d2k (2π)2 aσ(k)eik·A, d2k (2π)2 bσ(k)eik·B, ΩB where ΩB means integration over the first Brillouin zone. 0 vF σ∗ · k + ∆σz Fig. 4. The quantum well under analysis. , (4) (cid:17) (cid:88) σ 3a √ 2π 3 (cid:17) (cid:98)H = : (cid:90) 0 b† × σ ΩB σ(k) (cid:0)a† d2k (2π)2 (cid:98)H = −t(cid:80) σ(k)(cid:1) Hamiltonian (2) is rewritten as (cid:18)aσ(k) (cid:90) (cid:88)  −t(cid:80) Conduction and valence band extrema lie at the cor- (cid:16) 2π ners of the Brillouin zone. We use Hamiltonian (3) (cid:16) 2π expanded around the K point with quasimomentum or around the K(cid:48) point with q2 = 3a , q1 = 3a , − 2π √ 3 e−ik·di −∆ ∆ eik·di (cid:19) bσ(k) (3) 3a . i i (2π)2(cid:98)Ψ† d2k σ(k)(cid:98)H(cid:98)Ψσ(k). where integration is performed over small neighbor- hoods of the K and K(cid:48) points. Near the corners, the Hamiltonian reduces to (cid:18)vF σ · k + ∆σz (cid:98)H = (cid:19) where vF = 3 2 at is the carrier Fermi velocity, σ = (σx, σy) and σ∗ = (σx, −σy) are Pauli matrices in the sublattice space, and (cid:98)Ψσ(k) is the bispinor defined as (cid:33) σ (k) σ (k) , (cid:32)(cid:98)Ψ(1) (cid:98)Ψ(2) (cid:19) (cid:98)Ψσ(k) = (cid:18) 5πi (cid:18)aσ(q1,2 + k) (cid:19) bσ(q1,2 + k) . in terms of (cid:98)Ψ(1,2) σ (k) = exp σz σz 12 We consider the heterostructure combining gapped gra- phene regions 1 (x < 0) and 3 (x > d) with a graphene nanoribbon 2 (0 < x < d). The z axis is perpendicular to the graphene plane, the x axis is perpendicular to the heterojunction interfaces, and the y axis is parallel to the interfaces (see Fig. 1). Using Hamiltonian (4), we write an equation for the envelope wavefunction in a planar heterostructure: [vF j (τ0 ⊗ σx(cid:98)px + τz ⊗ σy(cid:98)py) + τ0 ⊗ σz∆j + τ0 ⊗ σ0 (Vj − E)] Ψ(x, y) = 0. (5) Here, ∆j = Egj/2 (j = 1, 2, 3) denotes half-width of bandgap (∆1 (cid:54)= 0 and ∆3 (cid:54)= 0 in regions 1 and 3, whereas ∆2 = 0 region 2); the respective work func- tions V1 and V3 of regions 1 and 3 depend on the mid-gap energies relative to the Dirac points for the corresponding materials (we set V2 = 0 to be specific, see Fig. 4); the 2×2 unit matrix σ0 acts in the sub- lattice space; the 2×2 unit matrix τ0 and the matrix τz defined similar to the Pauli matrix σz act in the valley space; ⊗ is the Kronecker product symbol; and ∂y are the momentum oper- ator components ( = 1). Assuming that the carrier ∂x and (cid:98)py = −i ∂ (cid:98)px = −i ∂ 4 Fermi velocities may differ between the three regions, we denote those for gapped regions 1 and 3 by vF 1 and vF 3 and use vF 2 ≈ 108 cm/s for gapless graphene. Charge carriers move freely along the y axis: Ψ(x, y) = Ψ(x)eikyy. The wavefunction Ψ(x) is a bispinor: (cid:18) ψK(x) (cid:19) ψK(cid:48)(x) , Ψ(x) = (cid:18)ψKA(x) (cid:19) ψKB(x) where the spinors ψK(x) and ψK(cid:48)(x) represent charge carriers in the K and K(cid:48) valleys, respectively: (cid:18)ψK(cid:48)A(x) (cid:19) ψK(cid:48)B(x) . ψK(x) = , ψK(cid:48)(x) = In the present context, the parity operator is expressed as follows: Equation (5) is solved here in the parity basis. The eigenfunctions Ψλ(x) of parity operator (6) are defined as follows: (cid:98)P = τz ⊗ σ0. (cid:98)P Ψλ(x) = λΨλ(x), (cid:19) (cid:18)ψ+1,K(x) (cid:18) (cid:19) 0 0 ψ−1,K(cid:48)(x) , . Ψ+1(x) = Ψ−1(x) = Rewriting Eq. (5) as the 2×2 matrix equations (cid:19) + vF jkyσy + λ∆jσz + Vj ψλK(x) (cid:18) −ivF jσx d dx = EλψλK(x), (6) (7) (8) (cid:18) −ivF jσx d dx − vF jkyσy − λ∆jσz + Vj = EλψλK(cid:48)(x). (cid:19) ψλK(cid:48)(x) (9) the K or K(cid:48) point, where the operator(cid:98)h can be defined. We see that setting ∆j = 0 and Vj = 0 brings us back to the spinor wavefunctions describing chiral states near However, chiral symmetry is broken when ∆ (cid:54)= 0 (see previous section). Defining parity λ as the eigenvalue of operator (6), we find that it indicates the valley where charge carriers belong: by virtue of (7), λ = +1 for the states near the K point described by Eq. (8) and λ = -- 1 for the states near the K(cid:48) point described by Eq. (9). In both gapped and gapless graphene, the valleys transform into each other under time reversal. This is indicated by the opposite signs of the terms propor- tional to ky in Eqs. (8) and (9), since ky → −ky under time reversal. It can be shown directly by using the time reversal operator T in explicit form that λ → −λ under T . Indeed, if ky is parallel to the line K−M−K(cid:48) (see Fig. 2) and its origin is set at M , then K → K(cid:48) and K(cid:48) → K under T . the 2×2 matrix equation Equations (8) and (9) are equivalently rewritten as (cid:19) + λvF jkyσy + ∆jσz + Vj ψλ(x) (10) (cid:18) −ivF jσx d dx = Eλψλ(x). Hereinafter, valley indices K and K(cid:48) are omitted as unnecessary since λ specifies the valley where charge carriers belong. We now discuss the boundary conditions at the in- terfaces between different graphene materials. At the outset, we note that they are easier to formulate than those used at the graphene -- free-space interface in mod- els of edge states [6, 49]. To derive boundary conditions in the present model, we must find a relation between ψλ(l) and ψλ(−l) as l → 0 in the neighbor-hood of x = 0, where l goes down to an atomic scale (condition at x = d is derived similarly). Multiplying Eq. (10) by ψ†(x) on the left, we integrate it over [−l, l]. Since a is small, we neglect all terms except those containing a derivative with respect to x to obtain2 λ (−l)v(−) λ (−l) = ψ(+)† F ψ(−) and ψ(+) ψ(−)† where ψ(−) are defined on the left- and right- hand sides of the boundary (at x < 0 and x > 0, re- spectively). Representing these functions as F ψ(+) (l)v(+) λ (l), λ λ λ λ (cid:12)(cid:12)(cid:12)ψ(±) (cid:12)(cid:12)(cid:12) exp (cid:113) (cid:12)(cid:12)(cid:12) = (−l) (cid:16) , iϕ(±)(cid:17) (cid:12)(cid:12)(cid:12)ψ(+)† λ v(+) F (cid:12)(cid:12)(cid:12) . (l) ψ(±) λ = (cid:12)(cid:12)(cid:12)ψ(−)† λ (cid:113) v(−) F we rewrite the equality above as 2From the given equality, we have the continuity of the cur- rent component normal to the interface in the heterostructure plane. It is necessary condition. To formulate the boundary condition in final form, we assume that the difference between phases of ψ(−) and ψ(+) λ near the interface is a multiple of 2π: ϕ(+) = ϕ(−) + 2πn, n ∈ Z. λ As l goes to zero, we obtain the following wavefunction- matching condition [34, 35] (cid:113) v(−) F ψ(−) λ = F ψ(+) v(+) λ , where minus and plus signs refer to the materials on the left- and right-hand sides of the boundary, respectively. The solution to Eq. (10) is expressed as follows. 1. At x < 0, ψλ(x) = C ek1x, (11) (cid:113) (cid:18) 1 (cid:19) q1 where q1 = −i (cid:113) vF 1(k1 − λky) Eλ − V1 + ∆1 1 − (Eλ − V1)2 + v2 ∆2 , F 1k2 y. vF 1k1 = 2. At 0 < x < d, ψλ(x) = C eik2x e−ik2x, q2κ∗ (cid:19) (cid:18) κ∗ (cid:18) κ−q2κ (cid:19) (cid:18) λky + C + k2 vF 1(k1 − λky)Eλ vF 2k2(Eλ − V1 + ∆1) (cid:113) (12) (cid:19)(cid:21) , (cid:114) vF 1 vF 2 where κ = 1 2 (cid:20) 1 + i q2 = vF 2(k2 + iλky) Eλ Eλ = ±vF 2 , k2 2 + k2 y, with plus and minus corresponding to electrons and holes, respectively. 3. At x > d, ψλ(x) = C (cid:18) ζ (cid:19) q3ζ e−k3(x−d), (13) (cid:114) vF 1 (cid:20) vF 3 where ζ = × cos(k2d) + (cid:18) λky k2 vF 1(k1 − λky)Eλ vF 2k2(Eλ − V1 + ∆1) + (cid:19) (cid:21) sin(k2d) , q3 = i (cid:113) vF 3(k3 + λky) Eλ − V3 + ∆3 3 − (Eλ − V3)2 + v2 ∆2 , F 2k2 y. vF 3k3 = The constant C is found by using the normalization condition for wavefunctions (11) -- (13), ∞(cid:90) † λ(x)Ψλ(x)dx = 1. Ψ −∞ 5 The carrier energy spectrum is determined by the dispersion relation tan(k2d) = vF 2k2f (λky; k1, k3, Eλ), (14) where f (λky;k1, k3, Eλ) = [vF 1(k1 − λky)(Eλ − V3 + ∆3) +vF 3(k3 + λky)(Eλ − V1 + ∆1)] × [Eλ(Eλ − V1 + ∆1)(Eλ − V3 + ∆3) − vF 2vF 3λky(k3 + λky)(Eλ − V1 + ∆1) + vF 1vF 2ky(k1 − λky)(Eλ − V3 + λ∆3) −vF 1vF 3(k1 − λky)(k3 + λky)Eλ] −1 is a function of k2 as well. Equation (14) must be solved for k2, and then the energy Eλ is found. In the case of an asymmetric quantum well, the de- pendence of (14) on λ gives rise to pseudospin splitting as the extrema of the dispersion curves shift away from Brillouin-zone corners. The dispersion relation predicts that Eλ(ky) (cid:54)= E−λ(ky), and an energy splitting ap- pears near the conduction-band bottom at ky = k∗ ye: δEe s = Ee−1(k∗ ye) − Ee +1(k∗ ye). A similar energy splitting appears near the valence- band top at ky = k∗ ye: δEh s = Eh−1(k∗ yh) − Eh +1(k∗ yh) Thus, a graphene nanoribbon becomes an indirect band- gap semiconductor analogous to silicon and germanium, where an electron -- hole plasma can exist [50]. In the case of a symmetric quantum well (∆1 = ∆3, V1 = V3, vF 1 = vF 3) band structure is invariant under parity and there is no pseudospin splitting [24]. 4. SIZE QUANTIZATION Fig. 5. Energy spectra: (a) symmetric quantum well (no pseudospin splitting), with matching branches for λ = +1 and λ = -- 1 (Eλ(ky) = E−λ(ky) = Eλ(−ky)); (b) asymmetric quantum well, with pseudospin split- ting manifested by the "spread-out" in quasimomen- tum between the extrema at k∗ ye for electrons, shown for b− = 1, and at k∗ yh for holes, shown for b+ = 1 (Eλ(ky) (cid:54)= E−λ(ky)). Solving Eq. (14), we determine the size-quantized energies Eλb∓ (ky) = ±vF 2 (cid:113) k2 2b∓ (λky) + k2 y, where b∓ = 1, 2, . . . labels electron ( -- ) and hole (+) branches, respectively. The size-quantized energy spec- tra for symmetric and asymmetric quantum wells are shown schematically in Fig. 5. We now determine the carrier effective masses aris- ing because of size quantization in the graphene nanorib- bon in a planar heterostructure. Note that the effective masses are invariant under parity regardless of pseu- dospin splitting. Hereinafter, we omit indices b∓, re- stricting ourselves to a particular branch of the electron spectrum and a particular branch of the hole spectrum. We write the dispersion law for electrons near an extremum at λk∗ (cid:0)ky − λk∗ ye (cid:1)2 ye as λ ≈ Ee Ee 0 + 1 2m∗ (cid:113) e 20e + k∗2 k2 1 + k(cid:48)2 20e + k20ek(cid:48)(cid:48) ye 20e m∗ e = 1 vF 2 , , (15) 6 where the respective values k20e, k(cid:48) and its first and second derivatives at ky = λk∗ independent of λ; Ee at the extremum. 20e of k2e(ky) ye are ye is the energy 20e + k∗2 k2 20e, k(cid:48)(cid:48) 0 = vF 2 (cid:113) Analogous expressions are obtained for hole ener- gies: λ ≈ Eh Eh 0 + (cid:0)ky − λk∗ yh (cid:1)2 , 1 2m∗ (cid:113) h 20h + k∗2 k2 1 + k(cid:48)2 20h + k20ek(cid:48)(cid:48) ye 20h m∗ h = 1 vF 2 (16) , (cid:113) 20h, k(cid:48)(cid:48) 20h + k∗2 k2 yh. where the respective values k20h, k(cid:48) 20h of k2h(ky) and its first and second derivatives at ky = −λk∗ yh; 0 = −vF 2 Eh To estimate characteristic values, we consider the planar heterostructure combining a gapless nanorib- bon with gapped graphene sheets with ∆1 = 0.75 eV, vF 1 = 1.1vF 2, ∆3 = 1 eV, and vF 3 = 1.2vF 2. The nanoribbon width is d = 2.46 nm (ten hexagonal cells). Since the unknown values of V1 and V3 can be found by Fig. 6. Electron and hole effective masses in the graphene nanoribbon (in units of free-electron mass m0) as functions of V1 for V3 = 0 (a) and as functions of V3 for V1 = 0 (b). comparing our results with experimental data, we seek the dependence of energy spectrum parameters on V1 and V3. Note that V1 ≤ ∆1 and V3 ≤ ∆3 to ensure that the heterostructure is type I. Figures 6 -- 9 show the results of numerical calcula- tions of electron and hole effective masses in the graphe- ne nanoribbon, extremum energies, k∗ xh val- ues, and pseudospin splitting δEe,h s plotted versus work function for one of the gapped graphene sheets given that the work function for the other is zero. xe and k∗ It is clear from Fig. 9 that the pseudospin split- ting energy may amount to approximately 10 meV. To obtain a larger pseudospin splitting, the quantum well must be more asymmetric. Both V1 and V3 can be var- ied by shifting the valley energies in gapped graphene under applied stress, with potential barriers playing the role of bandgaps in the gapped graphene sheets. An analogous effect is achieved by applying an electric field on the order of 106 V/cm perpendicular to the interfaces in the graphene plane [40]. As expected, the energy spectrum is symmetric un- der the change E → -- E when V1 and V3 = 0; i.e., the electron and hole spectra have equal effective masses, extremum energies, extremum positions, and pseudo- spin splitting energies. The electron and hole effec- tive masses in graphene are smaller than those in the gapped graphene sheets adjoining the gapless graphene nanoribbon (m∗ 3 = ∆3/v2 ≈ 0.15m0). F 3 F 1 ≈ 0.11m0 and m∗ 1 = ∆1/v2 7 Fig. 7. Electron and hole extremum energies and in the size-quantization spectra as functions of V1 for V3 = 0 (a) and as functions of V3 for V1 = 0 (b). The effective bandgap Eef f insignificantly. (cid:12)(cid:12) ≈ 629 meV varies 0 +(cid:12)(cid:12)Eh g = Ee 0 5. INTERFACE STATES The existence of surface states was predicted by Tamm in [52]. Extensive studies have been conducted of Tamm states in various systems including semicon- ductor superlattices [53, 54]. We consider interface states of a new type that arise in a narrow quasimo- mentum interval from the crossing of dispersion curves and are analogous to those in narrow-gap semiconduc- tor heterostructures [55, 56]. In the planar graphene- based heterostructure examined here, these states are localized near the heterojunction interfaces between the nanoribbon and the gapped graphene sheets. Inter- face states can exist not only in quantum wells but also in quantum barriers [10]. Note that interface states arise as well from the crossing of dispersion curves in a single heterojunction between different graphene ma- terials [57]. The wave function describing an interface electronic state is expressed as follows. (cid:19) eκ1x, (cid:18) 1(cid:101)q1 1. At x < 0,(cid:101)ψλ(x) = (cid:101)C (cid:101)q1 = −i (cid:113) vF 1κ1 = where u1(κ1 − λky) Eλ − V1 + ∆1 1 − (Eλ − V1)2 + v2 ∆2 , F 1k2 y. y = k∗ ye) and holes (k∗ Fig. 8. Extremum points of size-quantization bran- ches for electrons (k∗ yh) as functions V1 for V3 = 0 (a) and as functions of V3 for V1 = 0 (b). Inserts show the relative positions of dispersion curves for V1 = V3 = 0 (a) and at k∗ ye = k∗ yh (b); K and K(cid:48) points are set at the same position for simplicity. y = k∗ s and δEh Fig. 9. Pseudospin splitting in electron and hole spec- tra, δEe s , as functions V1 for V3 = 0 (a) and as functions of V3 for V1 = 0 (b). Vanishing δEe,h corresponds to vanishing k∗ ye,h in Fig. 8, as shown in inserts to (a). Inserts to (b) show positions of disper- ye and k∗ sion curves when k∗ yh coincide. s (cid:113) vF 3k3 = 3 − (Eλ − V3)2 + v2 ∆2 F 3k2 y. The expression for energy in (17) implies that an interface state exists only if3 κ2 < ky. We obtain the dispersion relation tanh(κ2d) = vF 2κ2f (λky; κ1, κ3, Eλ). (18) which is similar to (14) up to the substitutions k1 → κ1, k2 → iκ2, and k3 → κ3. We calculated numerically the energies of interface states as functions of quasimomentum component ky for heterostructures with ∆1 = 0.75 eV, ∆3 = 1 eV, V3 = 0, vF 1 = vF 3 = 0.7vF 2, and V1 = 0 and 100 meV. The results shown in Figs. 10 and 11 demonstrate that only electron interface states with λ = +1 and hole interface states with λ = -- 1 occur. When V1 = 0, the allowed quasimomenta for hole interface states (λ = -- 1) are similar to those for elec- tron states, but the hole and electron energies have opposite signs; i.e., the spectrum is symmetric under the change Eλ → -- Eλ. When V1 = 100 meV, the sym- metry is broken and hole interface states exist only at negative quasimomenta. 3The zero mode corresponding to κ2 = ky (with Eλ = 0) is irrelevant here because (cid:101)ψλ(x) ≡ 0. 2. At 0 < x < d, (cid:101)ψλ(x) = (cid:101)C (cid:114) vF 1 (cid:18) (cid:101)κ−(cid:101)q2(cid:101)κ− (cid:20) 1 ± λky κ2 1 2 vF 2(κ2 + λky) vF 2 where(cid:101)κ± = (cid:101)q2 = i (cid:19) e−κ2x + (cid:101)C (cid:19) (cid:18) (cid:101)κ+(cid:101)q(cid:48) 2(cid:101)κ+ eκ2x, ± vF 1(κ1 − λky)Eλ vF 2κ2(Eλ − V1 + ∆1) , (cid:101)q(cid:48) vF 2(κ2 − λky) 2 = −i (cid:113) Eλ , y − κ2 k2 2, Eλ Eλ = ±vF 2 (cid:21) , (17) with plus and minus corresponding to electrons and holes, respectively. 3. At x > d(cid:101)ψλ(x) = (cid:101)C (cid:114) vF 1 (cid:101)ζ = (cid:20) where vF 3 × ch(κ2d) + (cid:18) λky (cid:101)q3 = i κ2 + (cid:33) (cid:32) (cid:101)ζ(cid:101)q3(cid:101)ζ e−κ3(x−d), vF 1(κ1 − λky)Eλ vF 2κ2(Eλ − V1 + ∆1) vF 3(κ3 + λky) Eλ − V3 + ∆3 , (cid:19) (cid:21) sh(κ2d) 8 Fig. 10. Energy of electron interface states with λ = +1 in the heterostructure with ∆1 = 0.75 eV, ∆3 = 1 eV, vF 1 = vF 3 = 0.7vF 2, and V1 = V3 = 0 (electron interface states with λ = -- 1 do not exist). Allowed quasimomenta lie in two intervals: (a) negative ky and (b) positive ky. 6. EXCITON IN A PLANAR GRAPHENE QUANTUM WELL In gapless graphene, the carrier effective mass is zero and excitons do not exist. The existence of ex- citons in gapless graphene would lead to excitonic in- stability and excitonic insulator transition to a gapped state [58, 59]. The energy gap arising in a graphene nanoribbon as described in Section 4 makes it possible to generate ex- citons by optical excitation or electron-hole injection. Quantum-well excitons strongly affect optical proper- ties of the system considered here. Excitons in similar quasi-one-dimensional carbon- based systems (semiconducting single- and multi-walled nanotubes) have been studied theoretically in [60]. The exciton spectrum is calculated here for a planar graphene quantum well by using the model applied to quantum wires in [61]. This model yields simple analytical ex- pressions for exciton binding energy. Since formulas (15) and (16) are obtained in the nonrelativistic limit, the two-particle exciton wave func- tion depending on the electron and hole coordinates y− and y+ in a sufficiently narrow nanoribbon must obey the 1D Schrodinger equation with Coulomb potential: (cid:18) ∂2 ∂y2− − 1 2m∗ = E(cid:48)φ(y−, y+), e − 1 2m∗ h (cid:19) (cid:101)e2 − ∂2 ∂y2 + y− − y+ φ(y−, y+) (19) 9 g Fig. 11. Same as in Fig. 10, but with V1 = 100 meV; (c) energy of hole states with λ = +1 for allowed quasimomenta in a single interval of negative ky. and(cid:101)e2 ≡ e2/κef f . The effective where E(cid:48) = E − Eef f dielectric constant of graphene, κef f = (ε + ε(cid:48))/2, may vary widely with the dielectric constants ε and ε(cid:48) of the media in contact with graphene, such as free-space permittivity and substrate dielectric constant [62, 63]. The electron -- hole Coulomb interaction in a 1D gra- phene nanoribbon is three-dimensional, but the prob- lem can be reduced to one dimension (electron and hole y positions) for sufficiently narrow nanoribbons. ration y = y− − y+ and center-of-mass coordinate Rewriting Eq. (19) in terms of electron -- hole sepa- Y = ey− + m∗ m∗ e + m∗ m∗ h hy+ and introducing the function φ(y−, y+) = ψn(y)eiKY , where K is the total exciton momentum, we obtain (cid:18) − 1 2µ∗ (cid:19) ∂y2 − (cid:101)e2 y ∂2 ψn(y) = Enψn(y), (20) em∗ h/(m∗ e + m∗ where µ∗ = m∗ h) is the reduced mass and En is the energy of the nth exciton level (n = 0, 1, 2, . . . is the principal quantum number). The total exciton energy E(cid:48) is obtained by adding the total kinetic energy of the electron -- hole pair to En: E(cid:48) = En + K 2 e + m∗ h) 2(m∗ . To find the solution at y > 0, we substitute ψn(y) represented as ψn(y) = Bn exp (−y/an) Fn (cid:18) 2y (cid:19) . an into Eq. (20) and obtain the confluent hypergeometric differential equation ξF (cid:48)(cid:48) n − ξF (cid:48) (21) n + ηFn = 0, and η = µ∗(cid:101)e2an. We also have where ξ = 2y an En = − 1 2µ∗a2 n . (22) Equation (21) with η = n is solved by the associated Laguerre polynomial Fn(ξ) = 1 n! ξeξ dn dξn and the wavefunction is expressed as ψn(y) = Bn exp (−y/an) L−1 n (cid:0)ξn−1e−ξ(cid:1) ≡ L−1 (cid:19) (cid:18) 2y n (ξ). Here, an = n µ∗(cid:101)e2 (23) (n = 1, 2 . . .) is the Bohr radius of an exciton in the nth excited state. Combining (22) with (23), we find the exciton energy spectrum: En = − µ∗(cid:101)e4 2n2 . (24) The 1D ground-state (n = 0) Coulomb energy ex- hibits a logarithmic divergence at short distances [64]. Therefore, the lateral spread of the exciton wave func- tion (along the x axis) due to the three-dimensional nature of Coulomb interaction should be taken into account by introducing a cutoff parameter d0 (cid:46) d. Av- eraging the kinetic energy operator and the potential ∂2 ∂y2 2µ∗ (cid:98)T = − 1 V (y) = −(cid:101)e2 y θ (y − d0) over ground-state wave functions ψ0(y) = e−y/a0, 1√ a0 (25) where the ground-state Bohr radius a0 plays the role of a variational parameter, we express the ground-state exciton energy as Analogously, we find the solution to Eq. y < 0: ψn(y) = ±Bn exp (y/an) L−1 n (20) at E0 = 2µ∗a2 . (26) Minimizing (26) with respect to a0, we obtain an equa- tion for a0: , a0 = . (cid:19) an (cid:18) − 2y an a0 d 1 0 ln a0 − 2(cid:101)e2 d − 1(cid:1) . 2(cid:0)ln a0 a1 where "+" and " -- " are taken for n = 0 and n (cid:54)= 0, respectively, and the continuity of ψn(y) and its first derivative ψ(cid:48) n(y) are used as boundary conditions. Since ψn(0) = 0 and ψ(cid:48)(0) (cid:54)= 0 for n (cid:54)= 0, the excited-state wavefunction ψn(y) is odd (otherwise, it would be dis- continuous at the origin), whereas the ground-state wavefunction is even. To logarithmic accuracy, when (cid:29) 1, ln a1 d we find the relations E0 = 4E1 ln2 a1 d , (27) (28) (29) (30) The normalization condition ψn(y)2dy = 1 ∞(cid:90) −∞ an ∞(cid:90) is used to determine the coefficient Bn in the expression for ψn(y): Bn = (L−1 n (ξ))2e−ξdξ √ 0 √ 2an for n = 1, 2, . . . and B0 = 1/ a0 and Bn = 1/ for n = 0. −1/2 a0 = a1 2 ln a1 d . (cid:16) (cid:17) . a1 d Using (27), we easily obtain the next-order correction to E0: δE(1) 0 = −8E1 ln a1 d ln 2 ln We now examine the applicability of the formulas de- rived here. The semiconducting state induced in a graphene nanoribbon is stable with respect to spon- taneous electron -- hole pair creation (excitonic insula- tor transition) only if the exciton binding energy E0 is smaller than the effective bandgap in the graphene nanoribbon, E0 < Eef f g . 10 Furthermore, the quantum well width d must be much smaller than the exciton Bohr radius a1, d (cid:28) a1. Logarithmically accurate formula (29) is correct only if condition (28) holds. However, the asymmetric quan- tum well analyzed here to examine pseudospin effects may not admit even a single size-quantization level if the graphene nanoribbon width d is too narrow. As d decreases, the effective bandgap increases, approach- ing ∆+ + ∆−, where ∆± = min{∆1 ± V1, ∆3 ± V3} (with plus and minus corresponding to electrons and holes, respectively). When a certain dc is reached, the size-quantization levels are pushed into the continuum. This imposes a lower limit on d: d > dc, where dc can be estimated as dc (cid:39) πvF 2 ∆+ + ∆− . As d increases, condition (28) is violated. In this case, a more accurate variational calculation should be per- formed using the modified three-dimensional Coulomb potential (cid:101)V (y) = − , (cid:101)e2(cid:112)y2 + d2 − 2(cid:101)e2 0 1 tonian with potential (cid:101)V (y) over trial functions (25) to where d0 is a cutoff parameter. We average the Hamil- obtain 2µ∗a2 E0 = 2 [H0(ρ) − Y0(ρ)], Hν(ρ) is a Struve func- where I(ρ) = π tion, Yν(ρ) is a Bessel function of the second kind, and ρ = 2d0/a0 (ν is 0 here and 1 below). I(ρ), (31) a0 0 Minimizing (31) with respect to a0, we obtain an equation for a0: 2a0 a1 where J(ρ) = 1 − π I(ρ) + 4d0 a1 J(ρ) = 1, (32) 2 [H1(ρ) − Y1(ρ)]. Figure 12 shows the numerical results obtained by using both methods to calculate E0(d) for the het- erostructure described in Section 4, with d0 ∝ d ad- justed to match the curves at small d. Discrepancy at large d increases as ln(a1/d) approaches unity. 7. ELECTRIC FIELD EFFECT ON EXCITON LEVELS Interaction between an exciton and an external elec- trostatic field E is described by the operator (cid:98)Hi = −dE = (cid:101)e(Exx + Eyy) where x = x− − x+ and y = y− − y+ are the elec- tron -- hole relative position vector components and d is dipole moment. The electric field is supposed to be weak enough to ensure that the energy level shift is not Fig. 12. Exciton ground-state energy calculated by formula (29) (dashed line) and by formula (31) after Eq. (32) is solved numerically for a0 (solid line); d0 = 0.22d. only smaller than the spacing between size-quantization levels but also smaller than the spacing between exci- ton levels. These conditions can be written as where aE = (µ∗(cid:101)eE)−1/3 is the electric length. d (cid:28) a1 (cid:28) aE , We consider two cases: (1) the electric field is ap- plied parallel to the x axis and perpendicular to the nanoribbon edges in the graphene plane; (2) the elec- tric field is applied along the y axis, parallel to the nanoribbon edges. In the former case, the energy shift varies linearly with the difference between the average x components of the electron and hole position vectors: E(1)⊥λλ(cid:48) = (cid:101)eE ((cid:104)x−(cid:105)λ − (cid:104)x+(cid:105)λ(cid:48)) , (33) where average x components are calculated by using electron and hole single-particle wave functions, gen- erally depending on the electron and hole eigenvalues λ and λ(cid:48) of the operator (cid:98)P , respectively. Exciton en- ergy shift (33) is independent of the principal quan- tum number n. It may vary with λ and λ(cid:48), resulting in different exciton binding energies (more precisely, the binding energy of an electron -- hole pair with λ = ±1 and λ(cid:48) = ±1 may have four different values). In the latter case, the first-order electric field-induced correction is zero4, E(1)(cid:107)n = (cid:101)eE(cid:104)y(cid:105)n ≡ 0, (34) because the integral of yψn(y)2 with respect to y vanishes. To evaluate the second-order electric field- induced correction, we make use of the Dalgarno -- Lewis perturbation theory [65]. Defining a Hermitian opera- tor such that (35) [(cid:98)F , (cid:98)H0]n(cid:105) = (cid:98)Hin(cid:105), ∂y2 − (cid:101)e2 (cid:98)H0 = − 1 2µ∗ ∂2 y where 11 4Note that E(1)⊥n ≡ 0 in the former case if the electron and hole spectra transform into each other under field inversion. is the zeroth-order Hamiltonian, n(cid:105) = ψn(y) is the zeroth-order wave function of the nth exciton level, and (cid:98)Hi = (cid:101)eEy, we obtain E(2)(cid:107)n = (cid:104)n(cid:98)Hi(cid:98)Fn(cid:105) − (cid:104)n(cid:98)Hin(cid:105)(cid:104)n(cid:98)Fn(cid:105). (36) In the case in question, the second term in this formula vanishes by virtue of (34). Rewriting Eq. (35) as ∂2(cid:98)F ∂y2 + 2 ∂ψn ∂y (cid:98)F (y) = 2µ∗ y(cid:48)(cid:90) × dy(cid:48)(cid:48)ψ∗ ∂y = 2µ∗(cid:98)Hiψn, ∂(cid:98)F y(cid:90) n(y(cid:48)(cid:48))(cid:98)Hiψn(y(cid:48)(cid:48)). ψn(y(cid:48))2 dy(cid:48) −∞ (37) (38) ψn we find Fig. 13. Possible double resonant Raman processes involving electron scattering between valleys. To sim- plify presentation, analogous processes involving hole scattering between valleys are not shown. √ scattering involving phonons with wavenumbers q > K, 3a ≈ 1.7 × 108 cm−1 is the spacing where K = 4π/3 between adjacent K and K(cid:48) points. One process of this kind is indicated as A → B → C → D → A in Fig. 13. Pseudospin splitting enables intervalley scattering involving phonons with q(cid:48) ≈ q ∓ ∆k (with plus for electrons and minus for holes), where ∆k = 2k∗ ye and ∆k = 2k∗ yh in electron and hole scattering, respectively. These processes contribute to a peak blueshifted from R and a peak redshifted from D(cid:48) by ∆ω(−) D(cid:48) by ∆ω(+) R , giving rise to a doublet structure of the D(cid:48) peak. An estimate for ∆ωR can be obtained by using op- tical phonon dispersion ωph(q). The Raman shift is twice the optical phonon frequency: δωR(q) = 2ωph(q). The change in the Raman shift caused by pseudospin splitting is ∆ω(±) R ≈ δωR(∆K ∓ ∆k) − δωR(∆K). which amounts to ∆ωR ≈ 24 cm−1 for characteristic values of the heterostructure parameters. This value essentially exceeds the Raman spectral resolution of 1 cm -- 1 [68] and compares to the D(cid:48) peak width for gapless graphene, Γ0 = 30 cm−1 [69, 70]. Note that a blue shift of the D(cid:48) peak has also been observed in the Raman spectrum of epitaxial graphene on a SiC substrate [68]. This effect is attributed to the strain induced by the substrate in quasi-free graphene since the SiC lattice constant exceeds substantially that of graphene. Raman scattering contributions from gapped gra- phene sheets can be avoided either by using a laser beam whose width is smaller than that of the gapless graphene nanoribbon (d (cid:46) 10 nm) or by pumping at a frequency ω such that the beam cannot be absorbed by gapped graphene materials, g + 2ωph < ω < min{2∆1, 2∆3} Eef f The positions of the luminescence lines correspond- ing to exciton levels can be determined from optical −∞ Combining (36) with (38), we have the exciton gro- und-state energy shift E(2)(cid:107)0 = − 5 128 E 2, a3 1 ln4 a1 d (39) which is very small compared to E0 given by (29) be- cause of the fourth power of a logarithm in the denom- inator and a small numerical factor. For comparison, we write out the energy correction to the first excited exciton state: E(2)(cid:107)1 = − 3 8 (31 − 6γ)a3 1E 2, is Euler's constant. where γ = 0.577. . . By analogy with layered heterostructures [66], the ionizing (exciton-breaking) field strength Ec is estimated as (40) where (cid:104)y(cid:105)0 = a0/2 is the average electron -- hole sep- aration for the ground-state exciton. To logarithmic accuracy, it follows that , Ec = E0 8(cid:101)e(cid:104)y(cid:105)0 Ec = µ∗2(cid:101)e5 ln3 a1 . d (41) To get the order of magnitude of Ec, consider the quantum well discussed in Section 4. Setting m∗ e = h ≈ 0.0056m0, the SiO2 substrate dielectric constant m∗ κef f ≈ 5, d = 2.46 nm, and a1 ≈ 81 nm, we use for- mula (41) to obtain Ec = 9 kV/cm. 8. POSSIBLE EXPERIMENTS ON THE HETEROSTRUCTURE Pseudospin splitting can be observed by means of Raman spectroscopy. The D(cid:48) peak of interest for the present study (alternatively called 2D peak to empha- size that it is due to a two-phonon-assisted process) is located at 2700 cm -- 1 [67]. It arises from intervalley 12 experiments and compared to theoretical predictions. The splitting of exciton lines in an electric field is eval- uated by using formulas (33) and (39), respectively. Let us now discuss the Landau levels in a planar graphene heterostructure induced by a magnetic field perpendicular to the graphene plane. The magnetic field is supposed to be relatively weak, aH = (cid:112)c/eH (cid:29) d, so that size quantization plays a dominant role. As in analyses of layered narrow-gap heterostruc- tures [38, 71], the quasimomenta of free-moving carri- ers in size-quantized spectra (15) and (16) are replaced by the quantized magnetic momentum √ 2n aH , kHn = (cid:0)√ (cid:0)√ where n = 0, 1, 2 . . . labels Landau levels. (While each size-quantization level splits into a respective set of Landau levels, the present analysis is restricted to the lowest size-quantization level.) c , Ee Eh n − λ n + λ 0 + ωe c 0 − ωh Then, we have the carrier energies √ nHe √ nλ = Ee nλ = Eh eH e,hc is the cyclotron frequency and m∗ ye,h/2. where ωe,h nHe,h = a2 The minimum energy of electrons (near K point, with λ = +1) or holes (near K(cid:48) point, with λ = -- 1) may match a Landau level with n (cid:54)= 0. The zeroth Landau level is lowest only if nHe,h < 1/4. When c = H k∗2 nHh , (cid:1)2 (cid:1)2 where Nn =(cid:0)√ n(cid:1)2 √ Nn−1 < nHe,h < Nn, n + 1 + /4, the minimum energy (ground state) matches the nth Landau level. Such an unusual ground state can be observed only at temperatures sufficiently low that the excitation en- ergy from the lowest Landau level to the next one satisfies the condition ∆Ee,h It can easily be shown that ∆Ee,h H as a function of nHe,h has min- ima at nHmin = Nn (n = 0, 1, 2 . . .) and maxima at H > T . n − 1(cid:1)2 √ n + 1 + /4 (n = 1, 2 . . .): nHmax =(cid:0)√ (cid:104)(cid:0)√ (cid:104)(cid:0)√ ∆Ee,h H =  c n − 1 − √ n + 1 − √ ωe,h nHe,h Nn−1 < nHe,h < nHmax, ωe,h nHe,h c nHmax < nHe,h < Nn. (cid:1)2 −(cid:0)√ (cid:1)2 −(cid:0)√ n − √ n − √ nHe,h nHe,h (cid:1)2(cid:105) (cid:1)2(cid:105) , , A degenerate ground state corresponds to zero excita- tion energy. The excitation energy ∆Ee,h H is plotted as a function of nHe,h in Fig. 14. Finally, interface states can manifest themselves in the I -- V curve of the planar heterostructure carrying a current parallel to the gapless graphene nanoribbon. 14. Excitation energy ∆Ee,h Fig. nHe,h. The insert shows positions of Landau levels with N1 < nHe,h < N2. H vs. An increase in applied electric field may cause charge carriers to "drop" into interface states (preferable ener- gy-wise), giving rise to a region of negative differential conductivity in the I -- V curve. 8. CONCLUSIONS We have analyzed the characteristics of planar gra- phene nanostructures (quantum wells). On the one hand, they retain the unique properties of infinite gra- phene sheets. On the other hand, bandgap opening makes them important building blocks in carbon-based nanoelectronics, which can be used to control electron motion. Parameters of graphene quantum wells can easily be manipulated by varying the gapless nanoribbon width or the potential barriers in the adjoining gapped gra- phene sheets. We predict pseudospin splitting to occur in asymmetric graphene quantum wells and interface states to arise from the crossing of dispersion curves of gapless and gapped graphene materials. We have performed calculations of optical proper- ties of planar graphene nanostructures and suggested possible experiments to study the effects in question. Analysis of pseudospin (valley) characteristics in the heterostructure is simplified by using an effective Hamiltonian having a pseudospin-split energy spectrum. Note that an analogous spectrum was discussed in [72, 73]. Therefore, the effective Hamiltonian must contain a Bychkov -- Rashba spin-orbit coupling (cid:98)HR = αR[σ(cid:98)p]ν responsible for pseudospin splitting, where αR is a Rash- ba parameter and ν is the unit vector normal to the heterojunction interface in the graphene plane. In the coordinate system used in this paper, effective electron and hole Hamiltonians can be written as ef f = (cid:98)p2 (cid:98)H e y 2m∗ e − λαReσz(cid:98)py + ∆e 13 and ef f = (cid:98)p2 (cid:98)H h y + λαRhσz(cid:98)py + ∆h, 2m∗ ye/m∗ e and αRh = k∗ h where αRe = k∗ h re the re- spective electron and hole Rashba parameters (which are different when the symmetry under field inversion is broken) and the parameters ∆e,h are expressed in terms of extremum energies and Rashba parameters as yh/m∗ 1 2 ∆e = Ee 0 + ∆h = −Eh 0 + Re, m∗ 1 2 eα2 m∗ hα2 Rh. For characteristic values of the heterostructure param- eters, our estimates predict αRe,h (cid:39) 5 · 10−8 eV·cm, which is higher by two orders of magnitude than for GaAs [72] (mainly because of small effective mass). ACKNOWLEDGMENTS This work was supported by the Dynasty Founda- tion, the Scientific and Educational Complex of the Lebedev Physical Institute, and by the Presidium of the Russian Academy of Sciences under the Program for the Support of Young Scientists. References [1] K.S. Novoselov, A.K. Geim, S.V. Morozov et al., Science 306, 666 (2004). [2] K.S. Novoselov, A.K. Geim, S.V. Morozov et al., Nature 438, 197 (2005). [13] G. Giovannetti, P.A. Khomyakov, G. Brocks et al., Phys. Rev. B 76, 073103 (2007). [14] A. Mattausch and O. Pankratov, Phys. Rev. Lett. 99, 076802 (2007). [15] S.Y. Zhou, G.-H. Gweon, A.V. Fedorov et al., Na- ture Mater. 6, 770 (2007). [16] D.C. Elias, R.R. Nair, T.M.G. Mohiuddin et al., Science 323, 610 (2009). [17] S. Leb`egue, M. Klintenberg, O. Eriksson, and M.I. Katsnelson, Phys. Rev. B 79, 245117 (2009). [18] I. Zanella, S. Guerini, S.B. Fagan et al., Phys. Rev. B 77, 073404 (2008). [19] S. Marchini, S. Gunther, and J. Witterlin, Phys. Rev. B 76, 075429 (2007). [20] D. Martoccia, P.R. Willmon, T. Brugger et al., Phys. Rev. Lett. 101, 126102 (2008). [21] I. Pletikosi´c, M. Kralj, P. Pervan et al., Phys. Rev. Lett. 102, 056808 (2009). [22] M.Y. Han, B. Ozyilmaz, Y. Zhang, and P. Kim, Phys. Rev. Lett. 98, 206805 (2007). [23] L.E. Vorob'ev, E.L. Ivchenko, D.A. Firsov, and V.A . Shalygin , Optical Properties of Nanostruc- tures (Nauka, St. Petersburg, 2001) [in Russian]. [24] P.V. Ratnikov and A.P. Silin, Kratk. Soobshch. Fiz., No. 2, 11 (2009); arXiv:0808.3388. [25] B.G. Idlis and M.Sh. Usmanov, Sov. Phys. Semi- [3] Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, cond. 26 (2), 186 (1992). Nature 438, 201 (2005). [4] S.V. Morozov, K.S. Novoselov, M.I. Katsnelson et al., Phys. Rev. Lett. 100, 016602 (2008). [26] A. I. Akhiezer and V. B. Berestetskii, Quantum Electrodynamics (Wiley, New York, 1965; Nauka, Moscow, 1969). [5] X. Du, I. Skachko, A. Barker, and E.Y. Andrei, Nat. Nanotech. 3, 491 (2008). [27] A.H. Castro Neto, F. Guinea, N.M.R. Peres et al., Rev. Mod. Phys. 81, 109 (2009). [6] L. Brey and H.A. Fertig, Phys. Rev. B 73, 235411 (2006). [7] L. Brey and H.A. Fertig, Phys. Rev. B 75, 125434 (2007). [8] Y.-W. Son, M.L. Cohen, and S.G. Louie, Phys. Rev. Lett. 97, 216803 (2006). [9] R. Saito, G. Dresselhaus, and M.S. Dresselhaus, Physical Properties of Carbon Nanotubes, (Impe- rial College Press, London, 1998). [10] T. Ando, J. Phys. Soc. Jpn. 74, 777 (2005). [28] Yu.E. Lozovik, S.P. Merkulova, and A.A. Sokolik, Phys. -- Usp. 51 (7), 727 (2008). [29] S.S. Schweber, Introduction to Relativistic Quan- tum Field Theory (Halper and Row, New York, 1961). [30] A.M. Tsvelik, Quantum Field Theory in Con- densed Matter Physics (Cambridge University Press, 1998). [31] T.W. Appelquist, M. Bowick, D. Karabali, and L.C.R. Wijewardhana, Phys. Rev. D 33, 3704 (1986). [11] X. Wang, Y. Ouyang, X. Li et al., Phys. Rev. Lett. [32] B.A. Volkov, B.G. Idlis, and M.Sh. Usmanov, 100, 206803 (2008). Phys. -- Usp. 38 (7), 761 (1995). [12] L.A. Ponomarenko, F. Schedin, M.I. Katsnelson [33] A.V. Kolesnikov and A.P. Silin, JETP 82 (6), et al., Science 320, 356 (2008). 1145 (1996). 14 [34] A.V. Kolesnikov and A.P. Silin, J. Phys.: Con- [57] P.V. Ratnikov and A.P. Silin, Phys. Solid State dens. Matter 9, 10929 (1997). 52 (8), 1763 (2010). [35] A.P. Silin and S.V. Shubenkov, Phys. Solid State [58] A.A. Abrikosov, J. Low Temp. Phys. 2, 37 (1970). 40 (7), 1223 (1998). [36] E.A. Andryushin, Sh.U. Nutsalov, and A.P. Silin, Phys. Low-Dim. Struct. 7/8, 85 (1999). [59] S.A. Brazovskii, Sov. Phys. JETP 35 , 433 (1972). [60] T. Ando, J. Phys. Soc. Jpn. 66, 1066 (1997). [37] E.A. Andryushin, S.A. Vereshchagin, and A.P. Silin, Kratk. Soobshch. Fiz., No. 6, 21 (1999). [61] V.S. Babichenko, L.V. Keldysh, and A. P. Silin, Sov. Phys. Solid State 22 (4), 723 (1980). [38] E.A. Andryushin, A.P. Silin, and S.A. Vereshcha- gin, Phys. Low-Dim. Struct. 3/4, 85 (2000). [62] Yu.E. Lozovik and V.I. Yudson, Phys. Lett. 56A, 393 (1976). [39] E.A. Andryushin, Sh.U. Nutsalov, and A.P. Silin, Kratk. Soobshch. Fiz., No. 3, 3 (2001). [40] P.V. Ratnikov and A.P. Silin, Kratk. Soobshch. Fiz., No. 11, 22 (2005). [63] L.V. Keldysh, JETP Lett. 29 (11), 658 (1979). [64] R. Loudon, Am. J. Phys. 27, 649 (1959). [65] A. Dalgarno and J.T. Lewis, Proc. R. Soc. A 233, 70 (1955). [41] A. Rycerz, J. Tworzyd(cid:32)lo, and C.W.J. Beenakker, Nature Phys. 3, 172 (2007). [66] D.A.B. Miller, D.C. Chemla, T.C. Damen et al., Phys. Rev. Lett. 53, 2173 (1984). [67] D. Graf, F. Molitor, K. Ensslin et al., Nano Lett. 7, 238 (2007). [68] Z.H. Ni, W. Chen, X.F. Fan et al., Phys. Rev. B 77, 115416 (2008). [69] A.C. Ferrari, J.C. Meyer, V. Scardaci et al., Phys. Rev. Lett. 97, 187401 (2006). [70] I. Calizo, A.A. Balandin, W. Bao et al., Nano Lett. 7, 2645 (2007). [71] A.P. Silin and S.A. Vereshchagin, Phys. Low-Dim. Struct. 9/10, 115 (2001). [72] Yu.A. Bychkov and E.I. Rashba, JETP Lett. 39 (2), 78 (1984). [73] Yu.A. Bychkov and E.I. Rashba, J. Phys. C: Solid State Phys. 17, 6039 (1984). [42] J. Tworzyd(cid:32)lo, I. Snyman, A.R. Akhmerov, and C.W.J. Beenakker, Phys. Rev. B 76, 035411 (2007). [43] A.R. Akhmerov and C.W.J. Beenakker, Phys. Rev. Lett. 98, 157003 (2007). [44] D. Xiao, W. Yao, and Q. Niu, Phys. Rev. Lett. 93, 236809 (2007). [45] Z.Z. Zhang, K. Chang, and K.S. Chan, Applied Phys. Lett. 93, 062106 (2008). [46] J.L. Carcia-Pomar, A. Cortijo, and M. Nieto- Vesperinas, Phys. Rev. Lett. 100, 236801 (2008). [47] J.M. Pereira Jr., F.M. Peeters, R.N. Costa Filho, and G.A. Farias, J. Phys.: Condens. Matter 21, 045301 (2009). [48] G.W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984). [49] D.A. Abanin, P.A. Lee, and L.S. Levitov, Phys. Rev. Lett. 96, 176803 (2006). [50] T.M. Rice, J.C. Hensel, T. . Phillips, and G.A. Thomas, The Electron -- Hole Liquid in Semicon- ductors: Theoretical Aspects (Academic, New York, 1977; Mir, Moscow, 1980). [51] H. Suzuura and T. Ando, Phys. Rev. Lett. 89, 266603 (2002). [52] I.E. Tamm, Phys. Z. Sowjetunion 1, 733 (1932). [53] S.G. Tikhodeev, JETP Lett. 53 (3), 171 (1991). [54] S.G. Tikhodeev, Sol. St. Com. 78, 339 (1991). [55] A.V. Kolesnikov, R. Lipperheide, A.P. Silin, U. Wille, Europhys. Lett. 43, 331 (1998). [56] E.A. Andryushin, A.P. Silin, S.A. Vereshchagin, Phys. Low-Dim. Struct. 3/4, 79 (2000). 15 

1509.00250

1

1509

Pinning-induced pn junction formation in low-bandgap two-dimensional semiconducting systems

A model is presented for $pn$ junction formation near metal-semiconductor contacts in two-dimensional semiconducting systems such as graphene. Carrier type switching occurs in a region near the metal-semiconductor junction when energy band bending leads to a crossing between the junction Fermi level and the Dirac energy. A bias-dependent depletion region occurs due to the minimization of carrier density, which is shown to act as an additional parasitic resistance in devices. The $pn$ junction resistance is demonstrated by its implementation in a transfer length structure.

cond-mat.mes-hall

cond-mat

Pinning-induced pn junction formation in low-bandgap two-dimensional semiconducting systems David A. Deen1, a) independent submission A model is presented for pn junction formation near metal-semiconductor contacts in two-dimensional semiconducting systems such as graphene. Carrier type switching occurs in a region near the metal-semiconductor junction when energy band bending leads to a crossing between the junction Fermi level and the Dirac energy. A bias- dependent depletion region occurs due to the minimization of carrier density, which is shown to act as an additional parasitic resistance in devices. The pn junction resistance is demonstrated by its implementation in a transfer length structure. 5 1 0 2 p e S 1 ] l l a h - s e m . t a m - d n o c [ 1 v 0 5 2 0 0 . 9 0 5 1 : v i X r a a)Electronic mail: david.deen@alumni.nd.edu 1 The advent of two-dimensional (2D) semiconducting systems, such as graphene and the transition-metal dichalcogenides, has driven significant efforts to utilize these systems for electronic devices.1,2 In the vast majority of embodiments, electronic devices like the field effect transistor and various sensor designs utilize a modulated channel.3 In those cases, metallic electrodes are used to make ohmic contact to the semiconducting channel. Most studies address the minimization of parasitic access resistance via the contact resistance (Rc) of the contacts to graphene.4 -- 6 However, little attention has been given to the manifestation of spuriously high-resistance pn junctions formed in close proximity to the contact. The formation of the pn junction is due to the contrast between the Fermi-energy pinning at the metal-semiconductor interface and opposite carrier typing outside the contact region that results from ambient electrostatic bias. Ultimately, the pn junction resistance results as an additive parasitic resistance to the other parasitic resistances (access, Racc, and Rc) present, which may subsequently serve to impair device performance. This work presents the development of an electrostatic model that describes the formation of such a pn junction with low carrier density in proximity to a metal-2D semiconductor contact. The energy intersection between the pinned Fermi level under the metal-semiconductor contact and the 2D semiconductor Dirac energy (ED) is central to the development and results in the low carrier density bias-dependent pn junction. The model includes calculations for the width of minimum conductivity, its dependence on voltage bias, and how it is manifested in a conventional test structure. More complex effects such as Klein tunneling and the role of band chirality on carrier generation-recombination are not incorporated in the electrostatic model but likely play a role in the transport picture.7 Graphene and some transition-metal dichalcogenide semiconductors exhibit a unique lin- ear energy dispersion relation due to hybridized sp2 covalent bonding within their bravais lattice. In this work zero-bandgap graphene is used as an example from the broad 2D semiconducting families. Figure 1 shows an illustration of a graphene device cross-section near one ohmic contact (a) with the linearized energy band diagram (b) and resultant total charge distribution (c) for several voltage biases. The ohmic contact was assumed to pin the graphene Fermi level at an energy of ηp only under the contact.4,5,8 The linearization of the energy band diagram is an approximation made only in a small energy range about the Fermi level-Dirac energy (F-D) intersection where total charge density is minimized and resistance has a maximum. Realistically, the Dirac energy (or conduction and valence band 2 intersection) bends throughout the length of the device in the x-direction in coherence with a change in charge density (Poisson). However, for the small energy range about the F-D crossing the linearized energy is sufficient to describe the carrier typing and therefore, the electrostatics of the system in the vicinity of the ohmic contact. An approximation has been made for the resistance associated with the region beyond the linear energy region where (ED − Ef ) (cid:29) ηp. In this region, the total charge density exponentially increases according to Fermi-Dirac statistics and yields a resistance much less than in the region where the F-D crossing occurs. Therefore, this low-resistance extension has been treated as the sheet resistance outside the proximity of the F-D intersection (regions beyond Wd in Fig. 1 (c)). The pinning energy polarity at the metal-graphene junction and the ambient electrostatic bias away from the contact are responsible for setting up the conditions leading to the F-D intersection and therefore, the attributes of the pn junction. Those conditions determined the polarity of the ED slope, number of F-D intersection within a given region, and length of total charge minimum referred to as the 'depletion width' (Wd) in this work. If two F-D intersections occur, the total resistance becomes approximately twice the resistance of a single pn junction. The carrier distribution in graphene results from the linear dispersion relation and there- fore follows from the full Fermi-Dirac integral, n, p(η) = qNGFi(±η) where Fi(±η) = 0 ui(1 + e∓ηeu)−1du.3,9 The longitudinally-dependent total charge is given (Γ(i + 1))−1(cid:82) ∞ udu 1 + eueφF (x) − udu 1 + eue−φF (x) (1) 0 0 where NG = (gsgv/2π)(kBT /vF )2 is the 2D density of states with gs and gv are the spin and valley degeneracies, respectively, vF is the Fermi velocity, q the elemental charge, Γ(. . . ) is the gamma function,  is the reduced Planck's constant, φF (x) = kBT ηF (x) is the energy along the channel normalized by the thermal energy kBT , kB is Boltzmann's constant, and T the temperature. The linearized energy band was approximated as, E − Ef = Va(x− xo)/L, where xo is at the ohmic contact edge as illustrated in Fig. 1. The normalized longitudinal energy term used in Eq. 1 follows from the boundary conditions of the linear energy band, ηF (x) = x − ηp qVa L (2) where L is the length of the graphene sheet between the two biased contacts as defined in 3 as Qtot(x) = qNG Γ(2)  ∞(cid:90) ∞(cid:90)  FIG. 1. (a) Illustration of the 2D structure, (b) lateral band diagram in graphene along the source- drain direction at several different voltage biases, and (c) the resultant carrier density profiles along the graphene channel. 4 ohmic contactsubstratexQtotQoQminWd1Wd2Wd3ED,oEf,oηpED,1ED,2ED,3(a)(b)(c)L2D semiconductorqVaLEf = EDDirac coneEf > EDEf < EDenergyxoxzyxFig. 1, Va is the applied voltage, and ηp is the contact pinning energy. The depletion width is defined as twice the distance from contact pinning to the Fermi level crossing and follows Wd = 2ηpL qVa (3) In the conventional case of a three-dimensional semiconductor pn junction, the depletion width is defined by the balance of the internal field and the diffusion of free carriers. The linear energy approximation accounts for the complete electrostatic picture for the pinning- induced pn junction in graphene. The total resistance of the depletion width accounting for the single Fermi level crossing, is given by the integral relationship, Wd(cid:90) 0 Rd = 1 Wg dx σ(x) = 1 qWg Wd(cid:90) dx µnn(x) + µpp(x) 0 (4) where σ(x) = q(µnn(x) + µpp(x)) is the graphene conductivity and is dependent on the areal charge densities, n and p, and the respective carrier mobilities, µn and µp, and Wg is the cross-sectional width of the graphene sheet. Parametric calculations are shown in Fig. 2 wherein Figs. 2 (a) and (c) show total charge density and correspondent resistivity (ρ(x) = 1/σ(x)) in the graphene sheet, respectively, as a function of distance away from the ohmic contact for several longitudinal voltages. A pinning energy of 4 eV has been used, which is commensurate with values reported.4 -- 6 The charge density and resistivity minima progressively decreased with an increase in applied voltage as the F-D intersection migrated toward the ohmic contact. The rapidly diminished value of ρ away from the F-D crossing (total charge minima) validates the previously stated approximation. Figure 2 (b) and (d) show depletion width and depletion region resistance in the graphene sheet, respectively, as a function of applied voltage for several values of pinning energy. As a greater bias voltage is applied, energy-band bending is enhanced and the slope of the linear region becomes greater leading to the trends shown in Fig. 2. The bias-dependence of Rd is of note within the context that parasitic resistances are typically fixed (e.g. Rc or Racc). A dynamic parasitic resistance like Rd may serve to complicate the extraction of other intrinsic device metrics (i.e. transistor transconductance, frequency performance, etc.). One method that may prove to disable the formation of the pn junction is doping graphene. By doping, the Fermi level in graphene can be significantly shifted such that no F-D intersection may occur. Both Wd and Rd linearly decreased with increasing voltage following Eq. 3 and 4, respectively. 5 FIG. 2. Parametric characteristics for the pn junctions formed in graphene in proximity to an ohmic contact; (a) Charge density versus longitudinal distance, (b) resistivity versus distance for various applied voltages, (c) depletion width versus applied longitudinal voltage, and (d) depletion region resistance versus applied voltage for various pinning values. The effect of the pn junction on a device is illustrated through implementation into a transfer length structure (TLM). The TLM structure is a conventional test structure to extract information on Rc and Rsh. The total resistance of a TLM structure is conventionally given by R(L) = 2Rc + L W Rsh, where Rsh is the sheet resistance. For the scenario considered here, we assume there is a single pn junction between TLM contacts and the depletion width is small compared to the separation between contacts, Wd (cid:28) L. The latter assumption is verified by the calculation of Wd as shown in Fig. 2(b). These assumptions result in the TLM resistance relationship for graphene, RT LM = 2Rc + Rd + Rsh (cid:18) L − a (cid:19) with Rd being the depletion region resistance as given by Eq. 4 and a is the distance between 6 Wg (5) pinnedFIG. 3. Resistance of TLM structure comprised by graphene with and without the inclusion of a single pn junction showing a corresponding resistance increase. The inset shows the graphene TLM configuration and 1D linearized energy band diagram between two contacts. the contacts and the depletion width as illustrated in the inset in Fig. 3. Written in terms of the longitudinal graphene conductivity and invoking the constraint, L (cid:29) a, the total resistance of a graphene TLM structure with a single pn junction follows, R(L) = 2Rc + L Wg Rsh + 1 Wg (cid:124) dx σ(x) (cid:125) (6) Wd(cid:90) (cid:123)(cid:122) 0 Rd Equation 6 shows that the presence of a single pn junction in the graphene sheet adds a resistance proportional to the total bias-dependent resistivity of the junction to the TLM resistance. The pn junction resistance is independent of separation, L, and therefore is additive. The manifestation of additional pn junction resistance is demonstrated in Fig. 3 where a single pn junction is included in a TLM resistance curve with a Fermi pinning of 4 eV compared to a curve that does not include a pinning-induced pn junction. Other pertinent 7 contactsubstrateED,lEf,lseparation (L)energyxcontactWdaWg = 20μmηp = 4eVVa = 0.1mVno junctionηpηpEf,rED,rResistance (Ω)separation (μm)qVaV+graphene parameters used in the calculation are Wg = 20 µm, Rsh = 300 Ω/(cid:3), µn = µp = 1000 cm2/V s, T = 300K, and vF = 1.1× 106 cm/s. The linear energy approximation should hold for other 2D low bandgap semiconductors. An electrostatic model has been presented for pn junction formation in graphene in close proximity to ohmic contacts where Fermi level pinning occurs. It has been shown that electrostatic depletion of charge within the pn junction introduces a bias-dependent high- resistance region. The effect of pinning-induced pn junction formation in graphene has been illustrated for the case of a TLM structure, wherein the junction resistance was shown to be an additive term to the total TLM resistance. The broader impact of pn junction formation in low-bandgap two-dimensional systems may be spurious resistance that impairs device performance. We would like to thank J. G. Champlain at the Naval Research Laboratory for helpful technical discussion. REFERENCES 1Novoselov, K. S., Geim, A. K., Morozov, S. V., Jiang, D., Katsnelson, M. I., Grigorieva, I. V., Dubonos, S. V., Firsov, A. A.: 'Two-dimensional gas of massless Dirac fermions in graphene', Nature, 2005, 438, p. 197 2Radisavljevic, B., Radenovic, A., Brivio, J., Giacometti, V., Kis, A., 'Single-layer MoS2 transistors', Nat. Nanotech., 2011, 6, p. 147 3Champlain, J.G.: 'A first principles theoretical examination of graphene-based field effect transistors', J. Appl. Phys., 2011, 109, p. 084515 4Chaves, F. A., Jimenez, D., Cummings, A. W., Roch, S.: 'Physical model of the contact resistivity of metal-graphene junctions', J. Appl. Phys., 2014, 115, p. 164513 5Robinson, J. A., LaBella, M., Zhu, M., Hollander M., Kasarda R., Hughes, Z., Trumbull K., Cavalero R., Snyder D., 'Contacting graphene', Appl. Phys. Lett., 2011, 98, p. 053103 6Moon, J. S., Antcliffe M., Seo H. C., Curtis D., Lin S., Schmitz A., Milosavljevic I., Kiselev A. A., Ross R. S., Gaskill D. K., Campbell P. M., Fitch R. C., Lee K.-M., Asbeck P., 'Ultra-low resistance ohmic contacts in graphene field effect transistors', Appl. Phys. Lett., 2012, 100, p. 203512 7Katsnelson, M. I., Novoselov, K. S., Geim, A. K.,'Chiral tunneling and the Klein paradox 8 in graphene', Nat. Phys., 2006, 2, p. 620 8Giovannetti G., Khomyakov P. A., Brocks G., Karpan V. M., van den Brink J., Kelly P. J., 'Doping Graphene with Metal Contacts', Phys. Rev. Lett., 2008, 101, p. 026803 9Fang, T., Konar, A., Xing, H., Jena, D., 'Graphene statistics and quantum capacitance of graphene sheets and ribbons', Appl. Phys. Lett., 2007, 91, p. 092109 9 

1603.01578

2

1603

Role of Dzyaloshinskii-Moriya interaction for magnetism in transition-metal chains at Pt step-edges

We explore the emergence of chiral magnetism in one-dimensional monatomic Mn, Fe, and Co chains deposited at the Pt(664) step-edge carrying out an ab-initio study based on density functional theory (DFT). The results are analyzed employing several models: (i) a micromagnetic model, which takes into account the Dzyaloshinskii-Moriya interaction (DMI) besides the spin stiffness and the magnetic anisotropy energy, and (ii) the Fert-Levy model of the DMI for diluted magnetic impurities in metals. Due to the step-edge geometry, the direction of the Dzyaloshinskii vector (D-vector) is not predetermined by symmetry and points in an off-symmetry direction. For the Mn chain we predict a long-period cycloidal spin-spiral ground state of unique rotational sense on top of an otherwise atomic-scale antiferromagnetic phase. The spins rotate in a plane that is tilted relative to the Pt surface by $62^\circ$ towards the upper step of the surface. The Fe and Co chains show a ferromagnetic ground state since the DMI is too weak to overcome their respective magnetic anisotropy barriers. Beyond the discussion of the monatomic chains we provide general expressions relating ab-initio results to realistic model parameters that occur in a spin-lattice or in a micromagnetic model. We prove that a planar homogeneous spiral of classical spins with a given wave vector rotating in a plane whose normal is parallel to the D-vector is an exact stationary state solution of a spin-lattice model for a periodic solid that includes Heisenberg exchange and DMI. The validity of the Fert-Levy model for the evaluation of micromagnetic DMI parameters and for the analysis of ab-initio calculations is explored for chains. The results suggest that some care has to be taken when applying the model to infinite periodic one-dimensional systems.

cond-mat.mes-hall

cond-mat

Role of Dzyaloshinskii-Moriya interaction for magnetism in transition-metal chains at Pt step-edges B. Schweflinghaus, B. Zimmermann, M. Heide, G. Bihlmayer, and S. Blugel Peter Grunberg Institut and Institute for Advanced Simulation, Forschungszentrum Julich and JARA, 52425 Julich, Germany (Dated: June 16, 2016) We explore the emergence of chiral magnetism in one-dimensional monatomic Mn, Fe, and Co chains deposited at the Pt(664) step-edge carrying out an ab initio study based on density functional theory (DFT). The results are analyzed employing several models: (i) a micromagnetic model, which takes into account the Dzyaloshinskii-Moriya interaction (DMI) besides the spin stiffness and the magnetic anisotropy energy, and (ii) the Fert-Levy model of the DMI for diluted magnetic impurities in metals. Due to the step-edge geometry, the direction of the Dzyaloshinskii vector (D-vector) is not predetermined by symmetry and points in an off-symmetry direction. For the Mn chain we predict a long-period cycloidal spin-spiral ground state of unique rotational sense on top of an otherwise atomic-scale antiferromagnetic phase. The spins rotate in a plane that is tilted relative to the Pt surface by 62◦ towards the upper step of the surface. The Fe and Co chains show a ferromagnetic ground state since the DMI is too weak to overcome their respective magnetic anisotropy barriers. An analysis of domain walls within the latter two systems reveals a preference for a Bloch wall for the Fe chain and a N´eel wall of unique rotational sense for the Co chain in a plane tilted by 29◦ towards the lower step. Although the atomic structure is the same for all three systems, not only the size but also the direction of their effective D-vectors differ from system to system. The latter is in contradiction to the Fert-Levy model. Due to the considered step-edge structure, this work provides also insight into the effect of roughness on DMI at surfaces and interfaces of magnets. Beyond the discussion of the monatomic chains we provide general expressions relating ab initio results to realistic model parameters that occur in a spin-lattice or in a micromagnetic model. We prove that a planar homogeneous spiral of classical spins with a given wave vector rotating in a plane whose normal is parallel to the D-vector is an exact stationary state solution of a spin-lattice model for a periodic solid that includes Heisenberg exchange and DMI. In the vicinity of a collinear magnetic state, assuming that the DMI is much smaller than the exchange interaction, the curvature and slope of the stationary energy curve of the spiral as function of the wave vector provide directly the values of the spin stiffness and the spiralization required in micromagnetic models. The validity of the Fert-Levy model for the evaluation of micromagnetic DMI parameters and for the analysis of ab initio calculations is explored for chains. The results suggest that some care has to be taken when applying the model to infinite periodic one-dimensional systems. PACS numbers: 75.70.Tj, 71.15.Mb, 71.70.Gm, 73.90.+f I. INTRODUCTION In a seminal work, Gambardella et al.1,2 showed for the first time the presence of a truly one-dimensional (1D) metallic magnet. They succeeded in growing high- density arrays of monatomic Co chains on vicinal Pt(997) surfaces,3–5 denoted as Co/Pt(997), and investigated the magnetic properties by X-ray magnetic circular dichro- ism (XMCD). They found that, below a blocking temper- ature of about TB = 15 K, a long-range ordered collinear spin state is observed with magnetic moments aligned in the easy axis direction. The authors explained this fer- romagnetic order with a large magnetic anisotropy en- ergy (MAE) of ∆E = (2.0 ± 0.2) meV / Co atom that counteracts the magnetic fluctuations due to the finite temperature. The success in growing and measuring 1D magnetic monatomic chain structures as well as a detec- tion of an unusual easy axis direction pointing perpen- dicular to the chain direction and tilted by an angle of 43◦ towards the upper terrace triggered theoretical inves- tigations based on density functional theory (DFT),6–11 that affirmed the presence of an unusual direction of the easy axis. The strong MAE could be traced back to the large spin-orbit coupling (SOC) contribution of the Pt substrate. Succeeding these pioneering experiments al- ternative 1D systems had been investigated, among those FePt alloys12 and submonolayer Fe stripes,13,14 both on Pt(997), as well as Fe stripes15 and Co zigzag chains,16 both on an Ir(001) (5×1) surface. In this paper we address the question in how far these results and their interpretation remain unchanged in the light of the recently discovered interface in- duced Dzyaloshinskii-Moriya interaction (DMI).17 The DMI18,19 appears in magnetic systems that lack inversion symmetry and exhibit strong SOC. Only recently it was found to be an indispensable ingredient to understand non-collinear magnetic structures of unique rotational sense observed in thin films, for the first time demon- strated by Bode et al.17 who measured and analyzed a Mn monolayer on a W(110) substrate. Up to now a num- ber of similar systems are known in which the DMI leads to magnetic ground states that are described as cycloidal spin spirals20–22 or to the formation of a two-dimensional 6 1 0 2 n u J 5 1 ] l l a h - s e m . t a m - d n o c [ 2 v 8 7 5 1 0 . 3 0 6 1 : v i X r a generalization of spirals with one-dimensional propaga- tion vectors, the topological magnetic skyrmions.23,24 Also for biatomic Fe chains deposited on an Ir(001) (5×1) surface such a DMI-induced non-collinear mag- netic ground state has been predicted25,26 and experi- mentally verified shortly after.27 In the light of these analyses we turn to the Pt step- edge structure and investigate the leading magnetic in- teractions for different monatomic TM chains deposited along the step-edges. Due to the reduced symmetry oc- curring at step-edges, a complex interplay of DMI, MAE, and exchange interaction is found to determine the mag- netic ground state or the rotation type within a domain wall. The magnetic structures are explored in the context of a micromagnetic model that is introduced in Sec. II. There, we discuss consequences of the symmetry of the investigated structure on the magnetic anisotropy and DMI and derive two micromagnetic criteria that deter- mine the appearance of homogeneous and inhomogeneous spin spirals as magnetic ground states. In Sec. III we give details on the unit cell and the performed DFT calcula- tions. We proceed in Sec. IV with presenting the results of the performed calculations for the three investigated systems, monatomic chains of Mn, Fe, and Co at Pt(664) step-edges, and extract parameters for the previously dis- cussed micromagnetic model. Based on these parameters we predict the magnetic ground state for each system and characterize possible domain wall structures. We con- clude this paper with four appendices: In Appendix A we relate the micromagnetic parameters to the param- eters of a lattice-spin model. In Appendix B we show that the spin spiral as calculated from first principles is a stationary state of the lattice-periodic spin model con- taining Heisenberg interaction and DMI. In Appendix C we relate the micromagnetic parameters with the spin- spiral energetics as calculated from first principles. In Appendix D we analyze the relation between the micro- scopic DM vectors as obtained from the Fert-Levy model and the micromagnetic DM vectors. DM vectors are eval- uated and compared to the ab initio results from the main text. II. MICROMAGNETIC ANALYSIS OF THE STEP-EDGE STRUCTURE A. Symmetry considerations 2 structure discussed in this paper (see Fig. 1), however, only one mirror plane perpendicular to the chain direc- tion remains. A consequence of this reduction of sym- metry with respect to film structures is the previously mentioned easy axis direction for the Co chains, tilted by 43◦ towards the upper terrace. Similarly, the rules of Moriya19 only allow to reduce the possible orientation of the D-vector to the plane perpendicular to the chain axis, which is why ab initio calculations become neces- sary to determine not only the strength of the DMI but also the direction of the D-vector. Due to this particular symmetry at hand, the search for the magnetic ground state takes place in a higher- dimensional space. Besides the strength of the D-vector and the differences among easy, medium, and hard axes, one has to include in the final analysis the relative an- gle between D and the principal axes of the anisotropy tensor. B. The micromagnetic model To systematically study the magnetic phases in a solid from first principles one usually employs a multiscale ap- proach. DFT calculations are performed that allow to ex- tract system-specific parameters, which characterize the behavior of the system in terms of a suitable model, e.g., a (generalized) Heisenberg or spin-lattice model29 with spins placed on a discrete lattice. When the magnetic structure varies slowly across the crystal, meaning that the magnetic moments rotate on a length scale that is much larger than the interatomic distance, a micromag- netic model becomes favorable. Instead of a classical spin vector on each atomic site, such a model uses a continu- ous magnetization vector field m(r) (with m = 1) with effective parameters in which atom-specific contributions are implicitly contained. In case one deals with an an- tiferromagnetic spin-alignment, the classical spin vector is replaced by a staggered spin vector where the differ- ence of up and down spins on neighboring atoms form a new order parameter, that is treated then as a continu- ous field. Regarding the atomic structure we deal with in this paper, a linear chain of magnetic atoms along the y direction as depicted in Fig. 1, the magnetic energy for such spin textures can be expressed by the micromagnetic energy functional (cid:21) D 2π · (m × m) + mTKm , (cid:90) (cid:20) A E[m] = dy 4π2 ( m)2 + Many of the systems, in which the DMI is known to lead to a non-collinear magnetic ground state, consist of one or more layers of 3d transition-metal (TM) elements placed on top of a heavy element substrate17,21,22 and exhibit two mirror planes. This restricts the direction of easy, medium, and hard axis as well as the direction of the effective Dzyaloshinskii-vector28 (D-vector) to high- symmetry directions. Thus, the D-vector always points along either easy, medium, or hard axis. In the step-edge (1) with m = m(y) and m = d dy m. The first term in Eq. (1) contains the spin stiffness, A, and favors collinear spins ( m ≡ 0). In contrast, the second term is linear in m and thus shows a preference for a certain rota- tional sense of m with a strength and direction deter- mined by the Dzyaloshinskii-vector, D. Finally, the mag- netic anisotropy is accounted for by the last term, that features the anisotropy tensor, K, whose principal axes point along hard, medium, and easy axes.59 Note that in this work, without loss of generality, the energy-zero is given with respect to a magnetic configuration in which all spins are aligned along the easy axis. Furthermore, we point out that the local character of the integrand in Eq. (1) is reasonable as long as the range of the magnetic interactions is shorter than the characteristic length scale of the magnetic structure that is described. 3 In the following it is assumed to have knowledge of the model parameters A, D, and K. Considering the symme- try of the step-edge structure (cf. Fig. 1 and discussion in previous Sec. II A) the latter two are of the form Dx 0 Dz  , K = Kxx Kxz 0 Kyy 0 Kxz 0 0 Kzz  . D = (2) The direction of the Dzyaloshinskii vector, eDM = (sin ϑD, 0, cos ϑD)T, is described with respect to the z- axis by the angle60 ϑD = atan2 (Dx, Dz) ∈ (−180◦, 180◦] . (3) The eigenvalues of the anisotropy ellipsoid K, K1, K2, and K3, are the magnetic anisotropies along the prin- cipal axes. The principal axis corresponding to K2 is parallel to the y-axis. The axes e1 = Ry(ϑK)ex and e3 = Ry(ϑK)ez associated with K1 and K3 are obtained by a clockwise rotation, Ry(ϑK), of the magnetization m → Ry(ϑK)m around the y-axis by an angle ϑK = 1 2 atan2 (−2Kxz, Kxx − Kzz) ∈ (−90◦, 90◦] , (4) which results to K1 = Kxx cos2 ϑK − Kxz sin 2ϑK + Kzz sin2 ϑK , (5) (6) K2 = Kyy , K3 = Kxx sin2 ϑK + Kxz sin 2ϑK + Kzz cos2 ϑK . (7) C. Homogeneous versus inhomogeneous flat spin spirals It was first shown by Dzyaloshinskii30 that the magne- tization m(y) that minimizes the energy in functional (1) may correspond to spins that are periodically modulated rather than collinearly aligned along the easy axis. Ac- cording to the analysis of Heide et al.28 such a non- collinear spin structure can be either a three-dimensional (3D) spin spiral or a flat spin spiral with a propagation vector q along the step-edge (y direction) with magnetic moments rotating around the rotation axis of the spiral erot = (sin ϑr, 0, cos ϑr)T with an angle ϑr that encloses erot and the z-axis (the surface normal) and that is re- stricted to −90◦ < ϑr ≤ 90◦. In the following we restrict our analysis to flat spin spirals, i.e., the magnetization direction is always perpendicular to the rotation axis and Fig. 1: (color online) Step-edge structure, unit vectors and parameters used in the text with respect to the Cartesian coordinate system (x, y, z): The D-vector, being orthogonal to the y-axis, points along eDM and encloses an angle ϑD with the z-axis. The pairwise orthogonal principal axes of the anisotropy tensor K, e1, e2, and e3, are associated with K1, K2, and K3, respectively, where e3 encloses an angle ϑK with the z-axis. The rotation axis erot is perpendicular to the y-axis and encloses an angle ϑr with the z-axis. Dr denotes the projection of the D-vector onto erot. e(cid:107) and e⊥ are par- allel and perpendicular to the y-axis and are associated with K(cid:107) and K⊥, respectively, the anisotropy components within the rotation plane perpendicular to erot. The magnetization density m(y) varies as function of distance along the step- edge (y-axis) within the rotation plane (see semitransparent orange area) and encloses the spin-spiral rotation angle ϕ with e⊥. Note, that the angles ϑK , ϑD, and ϑr are positive (nega- tive) when pointing towards the lower (upper) terrace of the step-edge. erot is independent of y. For one part, this allows an an- alytical treatment of the problem, and for the other, we will show in Sec. IV that for all investigated systems the regime of truly 3D spin spirals can be excluded. Thus, the magnetization direction along the chain is given by  cos ϕ(y) sin ϕ(y) 0 m(y) = Ry(ϑr) (8) (cid:35) (9) and depends on a 1D parameter, the spin-spiral rotation angle ϕ(y). The matrix Ry describes a rotation around the y-axis. Inserting this into the energy functional from Eq. (1), normalized to one period length one arrives at an expression for the average energy density, A 4π2 ϕ + K⊥ cos2 ϕ + K(cid:107) sin2 ϕ ϕ (cid:34) (cid:90) 2π dϕ 0 Dr λ , Eλ [ϕ, ϑr] = 1 λ + with ϕ = dϕ vector onto the rotation axis and reads dy . Dr is the projection of the Dzyaloshinskii- Dr = D · erot = Dx sin ϑr + Dz cos ϑr . (10) K⊥ and K(cid:107) denote the anisotropy components in the rotation plane of the magnetization perpendicular and ye2e3e1ϑKeDMϑDxzϕm(y)erotϑreke⊥DDrparallel to the chain axis and are given by K⊥ = Kxx cos2 ϑr − Kxz sin 2ϑr + Kzz sin2 ϑr ,(11) (12) K(cid:107) = Kyy = K2 . average K = (cid:0)K⊥ + K(cid:107) (cid:1) /2. Note, that λ can become Note, that Dr and K⊥ depend explicitly on the an- gle of the rotation axis, ϑr, such that the functional of the average energy density in Eq. (9) depends on ϑr as well. For later purposes we additionally define Kmax = max{K⊥, K(cid:107)}, Kmin = min{K⊥, K(cid:107)}, and the negative since the present formalism also accounts for the rotational sense of the spiral. We distinguish a right- rotating spiral for ϕ > 0 and λ > 0 (energetically pre- ferred when Dr < 0, see Eq. (9)) and a left-rotating spi- ral for ϕ < 0 and λ < 0 (energetically preferred when Dr > 0) following the convention that a left-rotating spi- ral rotates clockwise when projecting the magnetic mo- ments onto the xy plane and reading the spiral rotation along the positive y direction (see Fig. 1). For a homogeneous spin spiral ϕ(y) changes linearly with distance within the chain and one finds ϕ(y) = y q· ϕ = ey, where q is the spin-spiral wave vector. Thus, q · ey = 2π/λ = const., and Eq. (9) simplifies to Ehom λ = A λ2 + Dr λ + K , (13) i.e., the energy density shows a parabolic behavior with respect to the inverse of the spiral length. Only when the minimum of this expression, Ehom λmin = − D2 r 4A + K , with λmin = −2 A Dr , (14) is below zero (corresponding to the energy of collinear spins aligned along the easy axis direction), a spiraling magnetic ground state can be established. This leads to the criterion for the appearance of a homogeneous spin spiral, f hom crit (ϑr) = 1 4 D2 r AK ! > 1 . (15) In the case of an inhomogeneous spin spiral ( ϕ (cid:54)= const.) the energy-density functional in Eq. (9) can be minimized by means of the Euler-Lagrange formalism30 resulting in Einh λ = −2K⊥ − K(cid:107) λinh = − sign(Dr) 2 π (cid:113) E(
)
2K(
) − c + Dr λinh , A/K⊥ − K(cid:107)
K(
) , (16) (17) (cid:112) with the Lagrange multiplier c > − min{K⊥, K(cid:107)}. K(
) and E(
) are the complete elliptic functions of first and second kind,61 respectively, with the ellipticity
=
(c) = K⊥ − K(cid:107)/(Kmax + c). It can be shown that the av- erage energy density in Eq. (16) gets minimal when Einh λ = −c. Together with Eq. (17) this leads to a condi- tional equation for c = c(
), Dr = 4 π AK⊥ − K(cid:107) E(
)
. (18) (cid:113) An inhomogeneous spiral appears for −c < 0, leading to the criterion 4 1 4 D2 r AK · α(K⊥, K(cid:107)) ! > 1 , (19) f inh crit(ϑr) = where α(K⊥, K(cid:107)) = K Kmax  2 π (cid:115) E −2 (20) K⊥ − K(cid:107) Kmax is a factor that depends on the ellipticity of the anisotropy energy within the plane of rotation of the magnetic moments spiral rotation axis: If the ellipticity
within the rotation plane is zero (K⊥ = K(cid:107)), then α = 1, which means that both criteria, Eqs. (15) and (19), be- come identical. One can show that elsewise α > 1, mean- ing that the criterion for the appearance of an inhomo- geneous spin spiral is always easier to be fulfilled than the criterion for the appearance of a homogeneous spiral, Eq. (15). For the case that the easy axis lies along the rotation axis (Kmin = 0) we have α = π2/8 and Eq. (19) simplifies to D2 r AKmax ! > 16 π2 , (21) which has been already discussed in literature.30,31 Eq. (9) can also be written as (cid:0)K⊥ − K(cid:107) (cid:1) cos2 ϕ + K(cid:107), As a final remark we state that the anisotropy term in which leads to the same expressions as derived above. D. Micromagnetic Parameters The three micromagnetic parameters A, D, and K are related to the site-dependent microscopic parameters of a spin-lattice model via (cid:88) (cid:88) j>0 1 2∆ j>0 A 4π2 = − 1 D ∆ 2π = R2 0jJ0j , R0jD0j , and K = 1 ∆K0 , (22) where ∆ defines the distance between two neighboring atoms within the chain, and R0j = j ∆ is the distance between atoms at sites 0 and j. J0j, D0j, and K0 are the exchange interaction, the Dzyaloshinskii vector between a pair of atoms at sites 0 and j, and the on-site anisotropy at the representative atom labeled 0, respectively (see Appendix A and Ref. 22 for details). The integrand of Eq. (1) is an energy density. For the quasi one-dimensional magnets studied in this work, it has the unit energy per length. Accordingly, the param- eters A, D, and K take the units energy times length, energy, and energy per length, respectively. However, it is often convenient to use another normalization, and rep- resent energy densities in units of energy per TM atom. The conversion from the first normalization to the sec- ond one is done by multiplication with ∆. In analogy, the units for the micromagnetic parameters A, D, and K change to energy times area per TM atom, energy times length per TM atom, and energy per TM atom, respec- tively. For the rest of this paper we use the same sym- bols for the two different normalizations, and the used normalization can be inferred from the unit. Notice, it is customary that both communities, the mi- cromagnetic and the spin-lattice model community, refer to D or Dij, respectively, as the Dzyaloshinskii-vector, although they are obviously different. We follow this tra- dition, but refer in addition to the D-vector in the spin- lattice model as microscopic D-vector, and in the micro- magnetic model either as the micromagnetic or effective D-vector or as the spiralization,32 whenever necessary. The spin stiffness and spiralization can be obtained di- rectly from first-principles calculations invoking the ho- mogeneous spin-spiral state. In Appendix B we prove that for each wave vector q there are two flat homoge- neous spin spirals of opposite handedness with a rota- tion axis parallel and antiparallel to the D-vector of that given mode. The lowest energy is found for a wave vector Q with a spin-chirality opposite to the D-vector of that mode. In Appendix C we show that if Q is in the vicin- ity of a high-symmetry point in the Brillouin zone, e.g., Q = 0 in case of the ferromagnetic state, typically this implies that the DMI is small compared to the exchange interaction. For the step-edge structure q becomes one- dimensional and we obtain the spin stiffness from the curvature A ∝ d2E(q)/dq2. The spiralization projected onto the direction of the DMI-vector is obtained from the slope [eDM · D] ∝ dE(q, eDM)/dq of the energy cal- culated for wave vectors q in the vicinity of the high- symmetry point. In the following Section we calculate A and D from E(q) for spin-spiral waves with q-vectors of different length from first principles in two separate steps: At first, E(q) is calculated without spin-orbit in- teraction employing the generalized Bloch theorem, from which the spin stiffness is determined and for which the spiralization is zero by definition, and then the spiral- ization is determined by calculating the change of the total energy ∆E(q) adding the spin-orbit interaction in first order perturbation calculated from electronic states related to the spin-spiral solution. III. FIRST-PRINCIPLES THEORY A. Structural Model and Computational Details The ab initio calculations based on DFT are car- ried out in film geometry of the full-potential linearized augmented plane-wave (FLAPW) method33,34 as imple- mented in the fleur code.35 The chains at stepped sur- faces are modeled like in earlier studies,10,11 where the chosen unit cell, a (664) step-edge structure, turned out to be a suitable structural model. The setup of the unit 5 Fig. 2: (color online) Sketch of the unit cell, a slab of a (664) vicinal surface decorated with a monatomic chain along the edge, and its repetition within the x(cid:48)y plane. The dark blue spheres correspond to the transition metals (Co, Fe, Mn) and the bright gray spheres represent the substrate atoms (Pt). The chain-to-chain distance is 13.24 A and the nearest- neighbor distance within the chain is 2.82 A. The inset in the lower left illustrates the use of the coordinates within the text. Note that the structure is periodic with respect to the x(cid:48)y plane. Although in the actual calculation all quantities are referenced with respect to (x(cid:48)yz(cid:48)), throughout this pa- per they are given with respect to (xyz), in accordance with Fig. 1. cell is inversion symmetric and consists of a tilted 8-layer Pt slab with two monatomic TMs deposited on both sides of the slab onto the step-edge. Throughout this investi- gation no relaxation of the structure is considered. This is motivated by the finding that relaxations can lead to an unphysically strong quenching of the orbital moment and, thus, to less accurate results for the MAE.10,11,36 In Fig. 2 we sketch the structural model and indicate the unit cell by the darker spheres in the foreground. The chain axis and, thus, the propagation direction of the investigated spiral structure is chosen as y-axis. To ensure a periodic repetition of the structure along the x(cid:48)-direction, as required by a solid-state code, the steps of the surfaces with normal [111] direction are tilted by an angle of about 10◦, so that the z(cid:48)-direction of the unit cell is [664] and the x(cid:48)-direction is [113] (see inset in the lower left of Fig. 2), resulting in a (6×1) surface unit cell. The used lattice constant is aPt = 3.99 A as calcu- lated by Baud et al.10 Along the x direction the unit cell has the length of the distance of two vicinal TM chains, ax(cid:48) = √11 · aPt ≈ 13.24 A. The width corresponds to one chain, ay = aPt/√2 ≈ 2.82 A. For all types of atoms the distance between two neighboring TM atoms within within the 2D unit cell the muffin-tin (MT) radius is chosen to be RMT = 1.16 A. If not stated otherwise, all energies obtained from first-principles calculations refer to energies per computational unit cell. Depending then TMPt13.24A2.82A[111]z[664]z0yxx0on the micromagnetic quantity under consideration, this energy can be related to energy per magnetic TM atom or per chain atom, respectively. For the exchange and correlation functional we chose the local density approximation (LDA) as proposed by Moruzzi, Janak, and Williams.37 The computational cut- off values for the expansion of the Kohn-Sham potential are Gmax = 12.0 a.u.−1 for the potential and Gxc max = 9.5 a.u.−1 for the exchange-correlation potential. The Hamiltonian matrix elements for all atoms in the unit cell due to the non-spherical part of the potential are expanded up to (cid:96)n−sph max = 6. The spherical harmonics expansion of the LAPW basis includes functions up to (cid:96)max = 8 within each MT sphere and all basis functions satisfying k(cid:107) + G(cid:107) < Kmax are included. If not stated otherwise, Kmax = 3.5 a.u.−1 is used. All self-consistent calculations have been carried out with 128 k-points in the full 2D Brillouin zone, whereas for one-shot calcu- lations employing the force theorem of Andersen38 512 k-points have been used. B. Spin stiffness The parameters A, D, and K are calculated as outlined in Refs. 22 and 39. The spin stiffness, A, is obtained by determining the total energy ESS(q) of the system as function of flat homogeneous spin spirals with wave vec- tors q of different lengths, all in the vicinity of the fer- romagnetic or antiferromagnetic state and all along the chain direction. Since the lengths of the q-vectors are small we applied the force theorem of Andersen38 to ob- tain these energies as deviations from the collinear state whose densities are calculated self-consistently employing the scalar relativistic approximation and which served as the initial state from which the force theorem is applied. To avoid numerical errors the magnetization in the in- terstitial region was set to zero before applying the force theorem. A detailed description can be found in Ref. 29. When calculating the spin stiffness we omitted the en- ergy correction due to SOC, because it proved small in tests and thus we can restrict ourselves to the use of the generalized Bloch theorem.40 C. Dzyaloshinskii-Moriya interaction The effective D-vector is determined treating SOC in first-order perturbation theory on top of flat homoge- neous spin-spiral solutions used to determine A. The DM energy is given by22,39 (cid:88) EDM(q, erot) = f (
0 kν, T ) δ
kν(q, erot) . (23) kν The occupation numbers are given by the Fermi function f (
, T ), which introduces a broadening of the occupation around the Fermi energy by the temperature T . They 6 depend on the wave vector q through the unperturbed (i.e., without SOC) eigenvalue spectrum
0 kν(q). The change of the eigenvalue spectrum δ
kν(q, erot) = (cid:104)U(erot)ψkν(q)HsoU(erot)ψkν(q)(cid:105) (24) due to SOC described by the Hamiltonian Hso, depends additionally on the rotation axis erot. The unitary trans- formation U(erot) directs the flat spin spiral of the unper- turbed state rotating around the z-axis to the global spin- rotation axis and ψkν(q) denotes the spin-spiral eigen- states of the unperturbed Hamiltonian. The summation in Eq. (23) runs over all states characterized by the Bloch vector k and band index ν. Due to the finite number of k-points (512 k-points in the whole 2D unit cell) the ef- fect of the broadening temperature will be a subject of study in Sec. IV B, which allows an estimation for the qualitative reliability of our results. We analyzed EDM, the change of the DM energy for a set of q-vectors that point along the chain direction (i.e., the y-axis) but vary in length, as well as two different ro- tation axes oriented along x- and z-direction (erot = ex and erot = ez, see Fig. 1 and Eq. (10)) to determine independently the two non-vanishing components of the D-vector (the third component vanishes due to symme- try, as already discussed in Sec. II A). In the micromag- netic limit, i.e., in the limit of long-period spirals, Eq. (9) is applicable. Therefore, if the spin-orbit interaction is included, the DM energy is expected to change linearly with the length of the wave vector q in the vicinity of the collinear spin alignment. Consequently we evaluate the effective D-vector as the slope of the energy change with respect to q in the limit q → 0. As outlined in Ref. 22, the spin-orbit coupling opera- tor Hso can be safely approximated by an atom-by-atom superposition of SOC operators limited to the muffin-tin spheres of the atoms, i.e., Hso = ξ(rµ) σ · Lµ , (25) (cid:88) µ where ξ is the spin-orbit strength related to the spherical muffin-tin potential V (rµ), ξ ∼ r−1 dV /dr, rµ = r − Rµ, MT. Rµ references the center and Rµ and rµ < Rµ MT is the radius of the µth muffin-tin sphere, with µ run- ning over all atoms in the unit cell. The atom-by-atom analysis is supported by the observation that ξ ∼ r−3 for small r. We observed for example in case of the Rashba effect that 90% of the Rashba strength is pro- duced by the wave function occupying a volume in the vicinity of the nucleus given by a radius of only about 10% (0.25 a.u.) of the muffin-tin radius.41 We expect an analogous behavior for the DMI. Thus, according to Eqs. (24) and (25) also δ
µ kν(q, erot) is atom dependent and the DM energy is a result of atom-by-atom con- DMI(q, erot), at least in first-order perturbation theory that we discuss here throughout the paper. The linear fit of Dµ(erot) q to Eµ DMI(q, erot) at the vicinity of a high symmetry point tributions EDMI(q, erot) = (cid:80) µ Eµ 7 in the Brillouin zone of propagation vectors gives then the decomposition of the D-vector into contributions Dµ, which satisfy (cid:88) kν (Dµ · erot) q (cid:39) f (
0 kν, T ) δ
µ kν(q, erot) . (26) For the interpretation of the atom dependent spiraliza- tion we refer to the discussion of the Fert-Levy model in Sec. IV B and in Appendix D. Since the structure of the unit cell setup in our ab ini- tio calculation is inversion symmetric, the contributions of the DMI to the total energy cancel when all atoms are taken into account. Thus, we manually break the inver- sion symmetry by considering only the energy differences due to SOC from the atoms that are placed in the upper half of the unit cell. D. Magnetic anisotropy For the magnetic anisotropy energy the force theorem of Andersen38 is applied, now in order to extract en- ergy differences between collinear systems with magneti- zations pointing in different directions. Starting point for the force theorem are self-consistent calculations includ- ing SOC, for which the magnetic moments point along the y direction. For each system we evaluate the total energy for several directions of the magnetic moments collinearly aligned within the xz plane and the yz plane. Out of the obtained energy landscape one is able to ex- tract K1, K2, and K3, the principal axes of the anisotropy tensor, K (cf. Eq. (2)). IV. RESULTS AND DISCUSSION A. Spin stiffness The results of the spin-spiral energy ESS(q) for all three investigated systems as function of the wave vector q along the one-dimensional Brillouin zone are summa- rized in Fig. 3. For Co and Fe chains the minimal energy is found for the ferromagnetic state, i.e., the state with wave vector q = 0, whereas the Mn chains align in the an- tiferromagnetic order. According to the micromagnetic model in Sec. II C (see Eq. (9)) and the discussions in Sec. II D we expect in the long-wavelength limit a lin- ear relationship between the exchange energy and the squared inverse wavelength, which is realized by these systems for a large fraction of the Brillouin zone (40%) and shown in the right panel of Fig. 3 with the result- ing fit. The slope gives the spin stiffness A. For the Mn system it is smallest (0.030 aJnm) and rises when going to Fe (0.041 aJnm) and Co (0.055 aJnm). The results are also collected in Table I. A small spin stiffness is fa- vorable for the stabilization of a chiral spin spiral and in this respect the Mn chain is the most favorable system. Fig. 3: (color online) Determination of the spin stiffness: In the left panel the total energies relative to their respective low- est energy, ESS, are shown as functions of the length of the wave vector, q, for Mn, Fe, and Co chains (magenta circles, red diamonds, and blue triangles, respectively). All systems show a collinear ground state, i.e., a ferromagnetic (q = 0) ground state for the Co and Fe chains and an antiferromag- netic (q = 0.5 in units of 2π/ay) ground state for the Mn chains. The right panel shows the energy as function of λ−2 in the linear regime with the corresponding linear fits. The slope represents the spin stiffness, A. Note that for the Mn chain we consider the antiferromagnetic ordering vector, meaning that λ → ∞ leads to the AFM spin alignment. The relative error due to the linear regression is in the order of 5% to 8% for the shown data range. B. The Dzyaloshinskii-vector Fig. 4 displays the DM energies per chain atom EDM and the x and z components of the spiralization vector D for the Mn chains. In the upper panel we present EDM(1/λ, ex/z) as function of the inverse wavelength, 1/λ, for clockwise rotating (negative values of λ) homo- geneous spin spirals of two rotational directions ex and ez. Analogously to the discussion of the spin stiffness, we utilize the micromagnetic model and expect a linear be- havior of EDM(1/λ, ex/z) ∝ Dx/z · 1/λ for corresponding wave vectors in the vicinity of high-symmetry points in the one-dimensional Brillouin zone, q = 0 and q = π/ay. Indeed we find a linear behavior for wave vectors covering 10% of the Brillouin zone measured from the antiferro- magnetic state at π/ay for EDM(1/λ, ex). However, for EDM(1/λ, ez) we notice a periodic modulation on top of the linear behavior. Such oscillations can occur due to fi- nite numerical resolutions, e.g., due to finite sampling of the Brillouin zone.22 In the lower panel of Fig. 4 we ana- lyze the effect of the electronic Fermi surface broadening temperature, T , on the obtained slopes that correspond to the D-vector components and the loss of linear behav- ior reflected in the error bars. When T is decreased, the values of the slopes and thus those of the D-vector com- ponents converge while at the same time the error bars 0.00.20.4q (2π / ay)050100150ESS (meV / TM atom)CoFeMn0.00.20.40.6λ−2 (nm−2)0204060CoFeMn8 tions come from atoms that are located next to the chain, albeit some contributions from some farther atoms can play a role as it is the case for Fe. A dominant contri- bution comes from the nearest-neighbor Pt atom at the upper terrace. For all systems they are of similar size, but for Pt next to Co, Dµ points in a direction differ- ent to the Mn or Fe case (cf. discussion at the end of this section). For the Co and Fe systems, each Pt atom µ with a dominant contribution Dµ has a vicinal atom with a Dµ vector of opposite sign and similar size. The dominant term for the Mn system, the nearest-neighbor atom at the upper terrace, has no counteracting contribu- tion and consequently leads overall to a larger size of the Dzyaloshinskii-vector D. This analysis shows that, al- though the atom resolved contributions might have large values themselves, the sum of all contributions can still lead to a rather moderate D-vector due to mutual com- pensation. In Fig. 5(d) we show an attempt to describe the re- garded structure in terms of the model proposed by Fert and Levy,42,43 where the DM energy Eµ DMI is given as sum over two distinct magnetic atoms within the chain interacting with the substrate atom µ (see Appendix D for details). Within this model, the direction of the atom- resolved Dµ-vectors is predefined to be perpendicular to the connection of the center of atom µ and the chain axis and perpendicular to the chain direction. This is in good agreement with the directions for the Dµ-vectors for the Mn and the Fe chains (cf. Figs. 5(a) and 5(b) with Fig. 5(d)), while it cannot be used to explain the direc- tions for the Dµ-vectors for the Co chain (cf. Fig. 5(c) with Fig. 5(d)). We next discuss the strength of the Dµ-vectors. Ap- plying the Fert-Levy model to a periodic infinite chain, one finds that Eµ DMI vanishes in the limit q → 0, while its derivative and thus the Dµ-vector, diverges. There- fore, this model is not applicable in this limit and the introduction of corrections attenuating or truncating the interaction between atoms in the infinitely long periodic chain after a certain interaction range, e.g., due to the lack of phase coherence or the presence of disorder will resolve this problem. Here, however, we avoid this sin- gularity by evaluating the strength of Eµ DMI(q0)/q0 for a finite wave vector q0 = 0.05 2π , which corresponds to ay λ−1 = 1.77 nm−1 and thus matches in length with a q- vector used in the presented ab initio calculations, see leftmost data points in upper panel of Fig. 4. The result- ing strengths of the Dµ-vectors decrease with distance to the chain (see Fig. 5(d)). The same behavior is also found for the three investigated chains, albeit the length of the vectors cannot be explained by the distance to the chain only. In Appendix D we furthermore show that the strength decays with distance much faster when the mag- netic moments of the atoms within the chain show a AFM short-range order, as compared to a FM short range order in the same chain. This observation, however, cannot be extracted from the ab initio results, e.g., when compar- ing the Mn chain (see Fig. 5(a)) to the Fe or the Co chain (color online) The upper panel displays Fig. 4: for Mn/Pt(664) the Dzyaloshinskii-Moriya energy EDM (see Eq. (23)) in the vicinity of the AFM state as function of the inverse wave length for flat homogeneous spirals rotating in the yz plane (magenta squares and solid line) and in the xy plane (magenta spheres and dashed line) for a temperature broadening of kBT = 27.2 meV. The slopes give the values for the components of the D-vector (see Eq. (2)) which are shown in the lower panel as function of a broadening tem- perature. Since for kBT = 27.2 meV (dotted vertical line) the values are converged and the error bars are still reason- ably small (for this system as well as for the other two the relative error due to the linear regression is in the order of 5% to 10%.), the corresponding values are considered in the following. are increasing. In the following, the values corresponding to kBT = 27.2 meV (see dotted vertical black line in the lower panel of Fig. 4) are used and can be found in Ta- ble I. Among the three systems the resulting D-vectors show remarkable differences in direction and strength. Therefore, we investigate its origin in more detail in the next paragraph. For the three investigated systems, a more detailed study of the atom-resolved contributions to the D-vectors is given in Figs. 5(a)-5(c). These atom-resolved contri- butions, i.e., Dµ for the atom with label µ, are obtained by switching on the SOC contribution for atom µ only. For each system they are plotted as vector with x and z components twice, (i) with respect to atom µ in the step-edge structure in the left part of each panel and (ii) with respect to the same origin in the right part of each panel, where in addition their sum, the D-vector, is shown as bold arrow. At first, we realize that for the Mn and Fe chain both D-vectors point into very similar directions. Although all three D-vectors point towards the upper step-edge, the direction of the D-vector of the Co chain is quite different from those of the Mn and of the Fe chain. The lengths of the D-vectors for Fe and Co chains are quite similar, but about only half as large as for the Mn chain. In general, the contribution Dµ of the 3d atom itself is nearly negligible. The largest contribu- 0.20.10.0λ−1 (nm−1)101EDM ‡meVMnatom·kBT=27.2 meV10-1100101102103kBT (meV)84048Dx, Dz ‡meVnmMnatom·DzDx9 Fig. 5: (color online) The atom resolved contributions {Dµ} to the D-vectors are shown (a) for the Mn chain, (b) for the Fe chain, and (c) for the Co chain, extracted from the performed ab initio calculations, as well as (d) for a TM chain when applying the Fert-Levy model (see Appendix D). For each of the four cases, these contributions are depicted twice. In the left image, a cross section of the step-edge structure is shown with these vectors {Dµ}, for convenience, located at the corresponding atom µ (in fact, these vectors act only on the TM atoms within the chain, represented by the light blue circles). In the right image, µ Dµ, they are given with respect to the same origin. In addition, the resulting D-vector, i.e., the sum over atoms µ, D =(cid:80) is printed in boldface. (see Figs. 5(b) and 5(c)). In conclusion, with regard to the structure of an infinite chain of magnetic atoms, we find that the model of Fert and Levy does not capture the diverse behavior of the three considered chains and we advise the application of this model to chains with some precaution. A more thorough investigation of the predictive power of the Fert-Levy model with respect to films and heterostructures would be interesting. Finally, we provide some arguments why the directions of Dµ from Pt atoms next to Co contributing to the total DMI vector are so different as compared to those next to Mn or Fe. From a simple tight-binding model that we developed in Ref. 44, we identified spin-flip transitions between occupied and unoccupied states as the relevant process for a non-vanishing DMI. For the Mn chain, the spin-up (spin-down) channels are entirely occupied (un- occupied) and all transitions yield a contribution to the DMI. Going now to Co, some spin-down states become occupied and transitions into these states do not con- tribute anymore to the DMI. Since the remaining empty states exhibit particular orbital characters, the Dµ vector may well be rotated as compared to Mn. The situation for Fe is similar to Mn: most of the spin-down states are still unoccupied. Of course, a quantitative analyze requires many more details, such as bandwidths, the na- ture of the chemical bond etc., but this goes beyond the scope of this paper. Fig. 6: (color online) Magnetic anisotropy energy for Mn/Pt(664). The energy is plotted for magnetic moments pointing along directions discretized by the angles ϕ and ϑ relative to the orientation in z direction. The symbols rep- resent ab initio calculated energy differences, whereas the fit functions correspond to Eq. (27). In the case of ϕ = 0◦ (solid red line) the xz plane and in the case of ϕ = 90◦ (dashed blue line) the yz plane is sampled. (a)Dµ=1MnMnDMnDµ=1Mn(b)FeDFe(c)CoDCo(d)DFL9060300306090polar angle ϑ (degrees)0.20.00.20.4Eϕ(ϑ)−Ez (meV / Mn atom)ϕ=0◦ϕ=90◦m(r)ϑϕC. The anisotropy tensor Following the findings of Sec. IV A we investigate the magnetic anisotropy tensor for the ferromagnetic order for the Co and Fe chains, and for the antiferromagnetic order for the Mn chains. The two required data sets for the latter system are shown in Fig. 6. The fit functions represent the leading order term and have the form Eϕ(ϑ) = Aϕ · cos2 (ϑ + Bϕ) − Ez (27) with the energy offset Ez = Eϕ=0(ϑ = 0), the polar angle ϑ as argument, the azimuth angle ϕ ∈ {0◦, 90◦} as parameter (see the inset in Fig. 6), and fit parameters Aϕ and Bϕ. The mirror plane perpendicular to the chain direction is reflected by the fact that Bϕ=90◦ = 0, leading to a symmetric function with respect to ϑ. The resulting hard, medium, and easy axes for all three systems are summarized in Table I. In the following we use the easy axis as energy offset. With respect to the resulting principal components the Mn system appears to be the most promising candidate for a non-collinear ground state. The anisotropy ener- gies for the medium and the hard axis are the smallest compared to those of the other two systems. In addition, the easy axis points along the chain direction which is of relevance for the following reason: The only spin-orbit driven spin spiral that the D-vector (perpendicular to the chain axis) can stabilize are of cycloidal character, meaning that the spiral rotation plane always contains the direction along the chain. Thus, the rotation over the easy axis is achieved automatically, regardless of the rotation axis. For the Co chains the easy axis is at about 61◦ tilted towards the upper terrace, which is in satisfy- ing agreement with other experimental1 and theoretical findings.6–8 The easy axis of the system containing the Fe chains is directed in approximately the same direction as the one for the Co chains. A remarkable finding for the Fe system is the strength as well as the orientation of the hard axis. It not only exhibits the largest value among all three systems, but also is oriented along the chain di- rection. Therefore, it shows the most unfavorable setup for a cycloidal spiral to appear since a rotation over the hard axis would be unavoidable. D. Magnetic ground states In the previous Secs. IV A, IV B, and IV C we extracted the parameters that now can be used to evaluate the cri- teria for the appearance of homogeneous and inhomoge- neous spin spirals (see Eqs. (15) and (19), respectively) and their respective properties.62 Those criteria depend on the spin stiffness A as well as Dr, the projection of the Dzyaloshinskii-vector onto the rotation axis erot, and K(cid:107), and K⊥, the two principal axes of the anisotropy tensor that describe spins rotating in the plane perpendicular to the rotation axis (see Fig. 1). Evaluating Eqs. (15) and 10 crit (ϑr) and f inh crit (ϑr) and f inh (19) the resulting magnetic ground state is determined by the functions f hom crit(ϑr) and whether their value exceeds the critical threshold of 1 for at least one rotation direction, described by the rotation angle ϑr. As we can see in Fig. 7 the Mn chains indeed fulfill both cri- teria when the direction of the spin-rotation axis, around which the magnetic moments of the spiral rotate, is in the regime between about −90◦ ≤ ϑr ≤ −20◦. The max- imum values of f hom crit(ϑr) are obtained for ϑr = −62◦, which at the same time represents the mini- mum of the total energy, i.e., the magnetic ground state. This rotation angle can be understood as a compromise between the optimal DMI contribution (ϑr = −38◦, ro- tation axis parallel to D-vector) and the minimal MAE barriers (ϑr = −72◦, rotation axis along K3, the hard axis). Since Dr is positive for ϑr = −62◦, the obtained magnetic structure is a left-rotating spiral, which modu- lates the otherwise antiferromagnetic order. As the mag- netic anisotropy within the yz plane (see Fig. 6, dashed blue curve) is small, the findings for homogeneous and inhomogeneous spirals are quite similar. For the same set of parameters, Eqs. (14) and (17) lead to a spiral length of λ = 15.7 nm for a homogeneous spin spiral and λ = 16.3 nm for an inhomogeneous spiral. Both values correspond to a period length of about 60 atoms along the chain, which is equivalent to an average rotation angle of ϕ ≈ 174◦ between neighboring Mn atoms and a q-vector of q = − 1 ey, measured from the AFM alignment. These large period lengths justify in retrospect our ansatz of a micromagnetic model. Employing Eqs. (14) and (16) we find an averaged energy gain of ∆E = −0.106 meV per chain atom (∆E = −0.376 meV nm−1) for the ho- mogeneous spin spiral and ∆E = −0.113 meV per chain atom (∆E = −0.399 meV nm−1) in the case of an inho- mogeneous spin spiral. In contrast to the analysis of the Mn chains, the ob- tained parameters for the Fe and Co chains confirm a ferromagnetic ground state, which is in line with previ- ous studies.10–12 For one part the resulting DMI is not large enough to change the collinear order favored by the spin stiffness, which we trace back to oppositely directed atom-resolved contributions to the D-vector of the Pt atoms nearby the chain (cf. Figs. 5(b) and 5(c)). On the other hand, the magnetic anisotropy causes energy barri- ers that prevent the system from forming a non-collinear ground state. Especially for the case of Fe chains the formation of a spin spiral turns out to be energetically unfavorable, as the spin moments would have to rotate over the hard axis, as mentioned in Sec. IV C. 2π ay 60 We conclude the investigation on the magnetic ground state with a brief discussion on the possibility of finding non-planar spin spirals. For systems with orthorhom- bic anisotropy, phase diagrams are known28 that take such three-dimensional non-collinear spin structures into account. However, to make our parameters match the Ansatz made in Ref. 28 one has to assume that the D- vector is oriented along one of the two principal axes of the anisotropy vector, K1 or K3. This is to some extent 11 Fig. 7: (color online) In the upper panels we show the values of the functions f hom crit for the appearance of homogeneous and inhomogeneous spin spirals (cf. Eqs. (15) and (19), respectively) in (a) Mn, (b) Fe, and (c) Co chains as functions of ϑr, the direction of the rotation axis, see Fig. 1. For each system the inset shows the relative orientation of the D-vector with respect to the principal axes of the anisotropy tensor. In the lower panels the corresponding parameters Dr, K(cid:107), and K⊥ are plotted. In the last two systems the critical threshold of 1 is missed by more than one order of magnitude. For the Mn chains, however, both criteria are fulfilled, and their respective curves reach their maxima for ϑr = −62◦. This can be seen as a compromise between finding the largest DMI contribution (dashed green line) and having the smallest anisotropy energy K⊥ (dashed brown line). crit and f inh A(cid:16) meV nm2 (cid:17) TM atom (cid:0) meV nm D (cid:1) TM atom 52 72 97 7.2 3.5 3.4 TM Mn Fe Co ϑD (degrees) −38 −49 −102 (cid:0) meV K1 (cid:1) (cid:0) meV K2 (cid:1) (cid:0) meV K3 (cid:1) TM atom TM atom TM atom 0.69 0.84 1.65 0.0 2.61 0.97 0.19 0.0 0.0 ϑK ϑmax r (degrees) −72 25 29 (degrees) −62 (−42) (+46) µmag (µB) 4.00 3.27 2.20 µorb (µB) 0.04 0.14 0.19 TABLE I: Collected results for the three investigated TM chains on Pt(664) step-edges: Spin stiffness A, absolute value of the Dzyaloshinskii-vector D and its orientation ϑD, principal components of the anisotropy tensor K (cf. Eq. (2)), K1, K2, K3, and ϑK , the orientation of the principal axis corresponding to K3, the rotation angle ϑmax , as well as the spin magnetic moment µmag and orbital magnetic moment µorb of the TM atom for the case that the spin-quantization axis points along the easy axis. All angles are measured with respect to the z-axis, see Fig. 1 and the insets of Figs. 7(a), 7(b), and 7(c). ϑmax indicates where the function f inh crit gets maximal, representing the planar inhomogeneous spin-spiral of lowest energy among all spirals. Only for the Mn chains this energy is lower than the one for the collinear state (i.e., criterion (19) is satisfied) and a spin-spiral state is formed as ground state. For the Fe and Co chains the collinear state always remains lower in energy. Note that K2, the anisotropy along the chain, is the easy axis for the Mn chains but the hard axis for the Fe system. A, D, and Ki can be expressed in units directly compatible to the micromagnetic equation (1) dividing the parameter values by ∆ = 0.282 nm / TM atom. r r only reasonable for the Fe chains where the angle be- tween easy axis direction and D-vector is 16◦ (see insets in Fig. 7). In addition this system is the best candidate for a three-dimensional spiral since a rotation over the unfavorable hard axis is avoided, so that we restrict our analysis onto the Fe chains only. Following the notation of Ref. 28 we arrive at DI = 0.32 and KI = −0.48, when the D-vector is assumed to point along the easy axis di- rection. Thus, we miss the critical regime of DI > 1 by a factor of 3, and this pair of parameters distinctly lies in the collinear region (cf. Fig. 3(b) in Ref. 28). E. Formation of domain walls Although for Fe and Co chains the DMI is not strong enough to introduce a chiral magnetic ground state its presence can influence the formation of domain walls.45 0123fcritfhomcritfinhcrit404Dr ‡meVnmTMatom·9060300306090ϑr (degrees)012K ‡meVTMatom·KK(a)MnchainsϑKϑD0.000.020.040.06fhomcritfinhcrit4049060300306090ϑr (degrees)012KK(b)FechainsϑKϑD0.000.020.040.06fhomcritfinhcrit4049060300306090ϑr (degrees)012KK(c)CochainsϑKϑDWe follow the analysis of chiral domain walls put for- ward by Dzyaloshinskii,30 but apply this analysis to the ferromagnetic Fe and Co chains. Once again the starting point is the energy functional as given in Eq. (1), now with the boundary condition TM Fe Co EB (meV) 17.49 28.26 wB (nm) 2.94 2.42 EN (EnoDMI N ) (meV) 29.06 (30.82) 17.73 (21.68) 12 wN (nm) 1.67 3.15 m(y) y→±∞−−−−−→ ±measy . (28) By this constraint a rotation by 180◦ is forced to take place spreading within the infinite chain. A distinction is made between a Bloch wall (helical rotation) and a N´eel wall (cycloidal rotation). For both cases a characteristic width of the planar domain wall can be defined by45 (cid:114) w = 1 π · A K , (29) TABLE II: The domain wall energies for a Bloch wall and a N´eel wall, EB and EN respectively, as well as the correspond- ing wall widths, wB and wN, are listed. Due to the DMI, the N´eel wall always exhibits a certain rotational sense that lowers the energy with respect to its value without taking the DMI into account (EnoDMI ). For the Fe chains the Bloch wall is energetically always more favorable, for the Co chains the N´eel wall is preferred. N where K represents the anisotropy energy for magnetic moments that point perpendicular to the easy axis direc- tion within the spin rotation plane. The expression for the minimal energy reads45 E = 2 π · √AK − erot · D 2 , (30) where for the N´eel wall erot·D is equal to the expression in Eq. (10), but vanishes for a Bloch wall (erot ⊥ D). Thus, only for a N´eel wall a preference in the rotation direction is expected and N´eel walls can be realized even if the MAE favors a Bloch wall. Note, that the energetically favored rotational sense of the domain wall is accounted for by the minus sign and the absolute value of the second term in Eq. (30). The resulting domain wall energies as well as the pre- dicted wall widths for Fe and Co chains are listed in Ta- ble II. Since for both systems the easy-axis direction is perpendicular to the chain direction, the rotation axis is fixed by Eq. (28) and the chain direction. If the easy axis points along the chain direction, erot is a compromise be- tween magnetic anisotropy and DMI as it was the case for the ground-state analysis. In such a case no Bloch wall can be established. For the Fe chains, a Bloch wall is energetically more favorable than the N´eel wall even when the DMI contri- bution is taken into account, so that we do not expect a preference in the rotational sense for the domain walls for this system. One reason is that a N´eel wall forces a rota- tion of the spins over the chain direction that is the hard axis of this system. Furthermore, the rotation plane is predefined by the easy axis direction. Since the D-vector is oriented nearly within this plane, the projection to the rotation axis erot is relatively small. For the Co chains, the energy of the Bloch wall, EB, is already by more than 6 meV higher in energy than the corresponding N´eel wall, EnoDMI , where the DMI contri- bution is neglected. When in Eq. (30) the DMI contri- bution is taken into account the preference of a N´eel wall is even higher. This gain in energy is achieved only for a right-rotating domain wall. This is because the rotation N Fig. 8: (color online) Schematic visualization of the energeti- cally preferred rotational sense for (a) the ground-state of the Mn chain (left-rotating spin-spiral) and (b) the domain wall for the Co chain (right-rotating N´eel wall). angle ϑr = 29◦ that describes a rotation plane perpendic- ular to the easy axis leads to a negative Dr, see Fig. 7(c). Such a spin-orbit driven preference of a particular rota- tional sense of the N´eel wall should be observable in an experiment. V. SUMMARY AND OUTLOOK Density-functional theory (DFT) calculations were performed to study the magnetic interactions in Mn, Fe, and Co chains at Pt step-edges. These calculations al- low to extract parameters for a micromagnetic model that takes into account the spin-stiffness constant, A, the magnetic anisotropy tensor, K, and the D-vector, which arises from the Dzyaloshinskii-Moriya interaction (DMI). y−62◦erotxz(a)y29◦xzerot(b)Using this model, the magnetic ground state for the three investigated systems is determined employing two differ- ent instability criteria for the appearance of spin-orbit driven non-collinear structures, one for the homogeneous and one for the inhomogeneous spin spiral. The main results are listed in Table III. Our results predict a spiral magnetic ground state for the Mn chains, that modulates the antiferromagnetic or- der with a period length of about 16 nm or 50-60 atoms along the chain. These findings establish Mn as a promis- ing candidate for experimental research groups to inves- tigate the DMI in 1D systems. A new aspect of this system, different to the systems studied in the literature, is the non-trivial direction of the D-vector, that is not fully determined by symmetry. As a result the spiral ro- tates in a plane that is tilted by about 62◦ towards the upper terrace (see Fig. 8(a)). For the Fe and Co chains we conclude that the formation of a non-collinear spi- ral magnetic structure is unlikely. For one part, this is due to magnetic anisotropies that are larger compared to the Mn chains. For the other part, their D-vectors are too small to overcome these anisotropy barriers. A detailed atom-resolved analysis of this quantity showed that their moderate strengths are due to compensation of the atomic contributions. For Co, the results are con- sistent with recent findings for the Co zigzag chain on Ir(001) (5×1),16 for which also no spiraling solution was observed. On the other hand, the Fe/Pt step-edge be- haves different to the biatomic Fe chain Ir(001),27 for which a spiral with a short period pitch was observed. The calculated directions and strengths of the D- vectors for the different chains were compared to those that result from the model of Fert and Levy. It appears that this model reproduces to some extent the directions of the D-vectors of the Mn and the Fe chains. For the Co chain, however, it fails to describe the direction of D cor- rectly. We noticed that the model of Fert and Levy may be used with some precaution at least for one-dimensional chains as the micromagnetic DM vector diverges for infi- nite chains in the limit of long wavelengths. A more thor- ough investigation of the predictive power of the Fert- Levy model with respect to films and heterostructures would be interesting. Furthermore, an analysis of planar domain wall struc- tures for the Fe and the Co chains was presented. It appears that in the Fe system a Bloch wall is energeti- cally more favorable. Since this type of domain wall is by symmetry not affected by the DMI, a preferred sense in the rotation direction is not expected. In contrast, the Co chains form a N´eel wall, which shows a homochiral preference in the wall rotation that is caused by the DMI (see Fig. 8(b)). We encourage experimental groups to verify our find- ings for the Mn chains in terms of the magnetic ground state. Furthermore, a statistically preferred rotational sense of domain walls in the Co chains should be observ- able in experiment. This could add a substantial aspect to the understanding of magnetism in low-dimensional 13 systems and could provide some insight into the conse- quences of surface and interface roughness on the DMI. Previous investigations revealed a strong dependence of the MAE on the number of transition-metal strands in the chain11,13,46. For example, a strong softening of the MAE was observed for Fe double-chains13, i.e., magnetic parameters may be tuned as functions of the number of strands to meet the criterion for a chiral ground state. In this paper we focused exclusively on infinite peri- odic chains. Here we comment briefly on the magnetism for chains of finite lengths. We may discuss the finite- ness in terms of a boundary effect, which are strongest where the chain terminates and whose effects decay away from the boundary into the chain. This affects shorter chains stronger than larger chains. Thus, in the center of larger chains we expect the same behavior as for periodic chains. In general, due to the finiteness of the chain three additional factors may play a role: (i) Atoms in a finite chain lose the mirror symmetry in a plane normal to the chain direction. (ii) Thus, edge effects of finite chains re- sult in non-vanishing components of D-vectors along the chain direction. Although the remaining non-vanishing component is small when averaged across the finite chain, locally we expect an additional tilt of the magnetic mo- ments and subsequently an additional energy gain. This supports the formation of a chiral magnetic ground state in a finite chain over an infinite one, similar to the surface twist in films of B20 alloys that stabilize the skyrmions phase in films over B20 bulk alloys.47 Even if we assume that the electronic structure at the boundary of the fi- nite chain does not change and all microscopic magnetic parameters remain unchanged, the micromagnetic DMI experience a change due to symmetry and this additional tilt of the magnetization has been investigated by S. Ro- hart and A. Thiaville48 but not for chains but for nanos- tructures. (iii) The change of the electronic structure at the boundary of the chain is an additional factor. Actu- ally we investigated this for finite clusters49 and it might be an important effect. Then all micromagnetic param- eters change, but most affected are the DMI and MAE. This can modify the threshold for the occurrence of a chi- ral magnetism in the chain. If the chain length becomes below two times the length scale, where the electronic structure is modified due to the presence of the finiteness of the chain, nothing can be said about the magnetic property of the short chain. Additional ab initio studies are required. On the methodological side we showed that for a spin- lattice model of classical spins on a Bravais lattice in- cluding Heisenberg and Dzyaloshinskii-Moriya interac- tion the homogeneous spin-spiral is an exact solution if the rotation vector of the spin-spiral points either par- allel or antiparallel to the D-vector, representing a so- lution of two different chiralities. This has important consequences since the spin-spiral state is a state that is frequently employed in the first-principles context using density functional theory. One consequence is that the slope and the curvature of the spiral energy as function of TM GS (no SOC) GS (with SOC) Mn Fe Co AFM FM FM (cid:96)-SS FM FM nian easy axis DW type (cid:107) chain ⊥ chain ⊥ chain - BW r-NW 14 TABLE III: Summary of the outcome of the paper: Whereas in the absence of SOC only collinear ground-states (GS) occur (FM and AFM), the Mn chain forms a left-rotating homochi- ral spin spiral ((cid:96)-SS) when SOC is taken into account. The easy axis can point along the chain (Mn) or perpendicular to it (Fe and Co). For the definition of ϑK see Fig. 7 or Eq. (4). The analysis of the domain wall (DM) type reveals that the Fe chain prefers a Bloch wall (BW) whereas for the Co chain a right-rotating N´eel wall (r-NW) is energetically favored. the wave vector as calculated in density functional theory provides directly the spin stiffness and the spiralization that enter a material specific micromagnetic model. VI. ACKNOWLEDGMENT We would like to thank Albert Fert, Miriam Hinzen, Daniel A. Kluppelberg and Christoph Melcher for fruit- ful suggestions and stimulating discussions during the course of this work. We gratefully acknowledge comput- ing time on the JUROPA supercomputer provided by the Julich Supercomputing Centre (JSC). B.S. acknowl- edges funding by the HGF-YIG Programme VH-NG-717 (Functional Nanoscale Structure and Probe Simulation Laboratory, Funsilab). B.Z. and S.B. acknowledge fund- ing from the European Union's Horizon 2020 research and innovation programme under grant agreement No 665095 (MAGicSky). Appendix A: Relation between micromagnetic and spin-lattice model A natural starting point for a multiscale analysis of a complex magnetic structure is the spin-lattice Hamilto- (cid:88) i<j (cid:88) i E{S} = [Jij Si · Sj + Dij · (Si × Sj)] + ST i Ki Si , (A1) where Jij is the exchange integral between atoms at sites i and j, Dij is the Dzyaloshinskii vector and Ki is the microscopic on-site anisotropy tensor. If these parame- ters are determined from first principles, one refers to a realistic spin-lattice model. Assuming lattice periodicity it follows that Jij = Jj−i, Dij = Dj−i = −Di−j, and Ki = K0 for all sites i, and considering that the exchange interaction and the DMI are even and odd functions, re- spectively, with respect to inversion symmetry. In this appendix we relate these parameters to the micromag- netic parameters of model (1). If we assume that the magnetic structure is slowly varying along the chain, meaning that the magnetic mo- ments rotate on a length scale that is much larger than the interatomic distance, then it is certainly possible to choose for the magnetization direction a continuous normalized function m(y) with m(y) = 1, such that m(j ∆) = Sj, where ∆ denotes the spacing between the lattice points along the y-axis. If we assume further that m does not vary much on a length scale at which the interactions J and D are relevant, then the interactions can be considered local over that length scale which is consistent with the formulation of the interactions in the micromagnetic energy functional (1). Under these condi- tions one can Taylor expand Sj = m(j∆) around m(i ∆). The energy expression (A1) treated within the lowest rel- evant order reads then (cid:88) j>i (cid:20) − 1 2 (cid:88) i ∆ E = (j − i)2∆Jj−i m2(i∆) + (j − i)Dj−i · (m(i∆) × m(i∆)) (cid:21) + 1 ∆  , m(i∆)T K0 m(i∆) (A2) For the exchange term we make explicitly use of the nor- malization as m2(y) = 1, d m(y) = 0 and d2 dy m2(y) = 2 m(y) · dy2 m2(y) = 2 m(y) · m(y) + 2 m2(y) = 0. Reminding that the distance Rij between atoms at site i and j is given by (j − i)∆ = Rij, replacing ∆ by dy in (A2) in the limit of small changes, the energy functional of spin-model (A1) approaches the energy functional of micromagnetic model (1), E{S} → E[m], with parame- ters A, D and K as summarized in Eq. (22). 15 replaced by a much simpler problem of minimizing the energy (B1) subject to the weak constraint Si · Si = N S2 , (B4) (cid:88) i where N is the number of lattice sites. This is a neces- sary condition and becomes sufficient if the solution also fulfills the strong constraint, as given in Eq. (B3). To take advantage of the translational symmetry of the crystalline solid, we transform the spin at lattice site i with the lattice vector Ri into momentum space Si = 1 √N Sq eiqRi and Sq = 1 √N (cid:88) i Si e−iqRi . (B5) but the derivations hold correct also for one- and two- Without loss of generality we assume here Ri ∈ R3, dimensional lattices. Since Si ∈ R3 is a three-tuple of real numbers, it holds that S∗q = S−q. With this definition, the quadratic form (B1) and the weak constraint (B4) can be expressed in momentum space as (cid:88) q 1 2 S†q Jq Sq (B6) (B7) (cid:17)∗ (cid:48) (cid:16) (cid:17)∗ = (cid:16) α J αα −q (cid:48) J α q (cid:88) q 1 2 E{Sq} = and ST−q Jq Sq = (cid:88) respectively, with (cid:48) J αα q = = e−iq(0−Rj ) = (cid:17)∗ (cid:17)∗ (cid:48) (cid:48) (cid:48) 0j (cid:16) J αα (cid:16) q J αα J αα q (cid:88) q q (cid:88) (cid:88)  j − Appendix B: Spin-spiral solution of spin-lattice model with Dzyaloshinskii-Moriya interaction From the view-point of first-principles calculations of a magnetic crystalline solid, the planar helical or cycloidal spin-spiral represents an interesting magnetic state, be- cause in the absence of spin-orbit coupling the spin-spiral state can be calculated by partitioning a solid into the same (chemical) unit cell that is used for non-magnetic or ferromagnetic calculations. This becomes possible by em- ploying the generalized Bloch theorem40 and holds true for any arbitrary wave vector q ∈ BZ taken from the Bril- louin zone (BZ) of wave vectors. It is a major concept to make such first-principles calculations feasible. In this Appendix B we show that the planar homoge- neous spin-spiral state of wave vector q, whose rotation axis points parallel or antiparallel to the Dzyaloshinskii- Moriya vector, is a stationary solution, and for a par- ticular wave vector Q, the spin-spiral state is also the energy minimizer of the spin-lattice model (A1) for a pe- riodic solid, when the magnetic anisotropy term is ig- nored. It is known that the spin-spiral state is the sta- tionary solution of a classical Heisenberg model on the Bravais lattice.50–53 Here we show that the solution holds true also for the Heisenberg exchange plus DMI. In differ- ence to the Heisenberg exchange only, where the energy is isotropic with respect to the rotation directions of the spirals, the DMI lowers the rotational symmetry, and se- lects spirals whose rotation directions are parallel and antiparallel to Dzyaloshinskii-Moriya vector Dq of mode q. In the following we assume a crystalline solid with lat- tice periodicity and restrict ourselves for simplicity to one atom per unit cell. We neglect the single-site anisotropy tensor in (A1). The spin-model (A1) on the Bravais lat- tice is then replaced by the quadratic form E{S} = 1 2 ST i Jij Sj , (B1) (cid:88) i,j with prefactor 1/2 preventing a double counting of terms and the exchange tensor  Jij Dz −Dz ij Dy ij −Dx ij −Dy ij Jij Dx ij Jij ij  ∈ R3×3 , Jij = and Jij = J T ji . The lattice periodicity implies Jij = J0,j−i = J T 0,i−j. The aim is to find the set of spins {Si}, with Si : Z → S2 ⊂ R3, that minimizes E{S} subject to the constraints that the length of spins are on sphere S2 of radius S and remain unchanged at all sites i, Si · Si = S2 ∀ i ∈ Z . (B3) Luttinger and Tisza54,55 realized that the minimization of a quadratic form under N strong constraints can be S−q · Sq = q S†q Sq = N S2 , Jq = (B2) for α = α(cid:48) for α (cid:54)= α(cid:48) , (B8) (cid:48) and α ∈ {x, y, z}. Since the off-diagonal elements of J αα are purely imaginary, we replace Dα q . The ex- change tensor in momentum space is then related to the tensor in real space (B2) as q by iDα q  ∈ C3×3 ,  Jq −i Dz q i Dy q ∈ R. Analogously, we find for the ex- (cid:88) i Dz q −i Dy q i Dx Jq q q −i Dx Jq (B9) q with Jq and Dα pression of the DM energy of the spin-model (A1) N EDM{S} = D0j · C0j with C0j = S0 × Sj , (B10) in terms of the vector chirality C0j, or sum over modes in momentum space, respectively, j 1 2 (cid:88) q EDM{Sq} = 1 2 Dq · Cq with C(Sq) = iS∗q × Sq , (B11) 16 Λ−q,min has at least a twofold degeneracy for q ∈ BZ, but the eigenvectors corresponding to the lowest and highest value exchange their roles, i.e., w−q,1 = (wq,1)∗ = wq,3 and vice versa. A lower bound to E{Sq} can be esti- mated considering that S†qJqSq ≥ S†qΛq,minSq is limited by the lowest eigenvalue, and thus E{Sq} ≥ 1 2 Λ S†q Sq = 1 2 ΛN S2 , (B16) (cid:88) q with Λ = Λ±Q = minqΛq the lowest eigenvalue of all q. In the state that minimizes the 3N -dimensional ellip- soid transformed to the principal axis (see Eq. (B15)) with respect to S†q subject to the constraint (B7) of a 3N -dimensional sphere of radius N S2, it is easy to show by the method of Lagrange multipliers that the Sq must satisfy (λq,γ − 2ξ) w†q,γRqSq = 0 ∀q, γ , (B17) where ξ is the Lagrange multiplier independent of q. Us- ing Eqs. (B17) and (B7), the energy (B15) becomes E{Sq} = ξN S2 . (B18) Hence the minimum E{SQ} is obtained for the minimum ξ for which solutions of Eq. (B17) exist, which is realized for ξ = 1/2Λ, proving that the ground state satisfies the equal sign in Eq. (B16). Now we have a closer look at Eq. (B17). If Sq (cid:54)= 0 for all q and since Wq spans the whole three-dimensional spin-space there exists for each vector q at least one eigenvector wq,γ for which w†q,γRqSq (cid:54)= 0. As a conse- quence for each vector q, there is at least one eigenvalue λq,γ for which λq,γ = 2ξ. Since ξ is independent of q, all eigenvalues and thus all Jq, and Dq should be inde- pendent of q, and this is unphysical. On the contrary a single-q state, (cid:114) (B13)  , (B14) where C(Sq) is the vector chirality of mode Sq. Obvi- ously energy is gained if the vector chirality is antiparallel to the Dzyaloshinskii-Moriya vector, Cq ∝ −Dq. To simplify the minimization problem it is convenient to transform the 3N dimensional quadratic form (B6) into the principal axes by diagonalizing the matrix Jq. Since Jq = J †q is Hermitian, Jq has 3 real eigenvalues λq,γ with γ ∈ {1, 2, 3}: λq,1(3) = Jq (∓) Dq and λq,2 = Jq (B12) with orthonormal eigenvectors Vq = {vq,1, vq,2, vq,3} ∈ C3×3. The eigenvector vq,2 points for each wave vector q into the direction of the Dzyaloshinskii vector vq,2 = eq,DM = Dq/Dq (see also Eq. (3)). Obviously eigenvec- tors vq,1 and vq,3 live in the orthogonal subspaces. With- out loss of generality we choose for each mode q the coor- dinate system of the spin space such that eq,DM coincides with the z-axis, eq,DM = ez. In this new frame of refer- ence RqDq = (0, 0, Dq)T with Dq = Dq ≥ 0, the eigen- vectors transform to Wq = RqVq = {wq,1, ez, wq,3} ∀q, where Rq ∈ R3×3 is the respective rotation matrix, which conserves handedness, i.e., detR = 1. With those defini- tions it is clear that R−q = −Rq, because of the symme- try D−q = −Dq (see Eq. (B8)) and our definition that the local z-axis points always parallel (not anti-parallel) to Dq. We can always choose a transformation such that the exchange tensor becomes block diagonal the eigenvectors simplify to Rq Jq RT q = 0 −iDq Jq 0 iDq 0 0 Jq  ,  Jq  0  1  = (wq,3)∗ , wq,2 = 3(cid:88) (cid:88) 0 1 i 0 and the energy in momentum space reads E{Sq} = 1 2 S†qRT q wq,γλq,γw†q,γRqSq . q γ=1 wq,1 = 1 √2 (B15) Sq¯q,γ = N S2 2 RT ¯q (w¯q,γδq,¯q + w−¯q,γδq,−¯q) , (B19) The eigenvalues and eigenvectors are invariant with re- spect to space inversion symmetry I transforming Iq = −q, and complex valued functions as IJ (q) = J ∗(−q), and Iwq,γ = w∗ −q,γ. The matrix of vector chirality for the three eigenvectors C(Wq) = {−ez, 0, ez} ∀q are mo- mentum independent and the chirality vector of the state with the lowest (highest) eigenvalue point antiparallel (parallel) to the direction of the DMI. Now we turn to our primary goal to find the state SQ that minimizes the energy expression E{Sq} of Eqs. (B6) or (B15), respectively. Irrespective of the sign of Jq, eigenvalue λq,1 is always the lowest eigenvalue for any wave vector q, Λq,min = λq,1. Considering the symme- try relation J (q) = J (−q)∗ (see Eq. B8)), both matri- ces have the same eigenvalues and subsequently Λq,min = i.e., a state for which Sq = 0, ∀ q \ {¯q,−¯q}, made of a superposition of two arbitrary modes with wave vectors ¯q and −¯q, for which eigenvalue λ¯q,γ = λ−¯q,γ is two- fold degenerate, with polarization directions determined by the principal axes of the exchange tensor J¯q satisfies Eq. (B17) for the Lagrange parameter ξ = 1/2λ¯q,γ and the respective energy E{Sq¯q,γ} = 1 2 λ¯q,γN S2 , (B20) and it is a stationary state of the energy functional, with λ¯q,γ from Eq. (B12). The term ∝ RT ¯q w¯q,γδq,¯q satisfies also condition (B17), but not the condition S∗q = S−q, and thus Si /∈ R. Therefore, this case is not further discussed. The three eigenmodes (B19) exhibit the chirality N S2 2 RT ¯q ez (δq,¯q − δq,−¯q) (B21) (B22) C(Sq¯q,1) = − C(Sq¯q,2) = 0 C(Sq¯q,3) = N S2 2 RT ¯q ez (δq,¯q − δq,−¯q) . (B23) Obviously, the modes Sq¯q,1 and Sq¯q,3 are of opposite chirality. A Fourier back-transformation of the eigenmodes shows that the modes with γ = 1 and γ = 3 correspond to flat spin spirals,  cos(¯q · Ri) (∓) sin(¯q · Ri)  , 0 Si¯q,1(3) = S RT ¯q (B24)  (cid:73) (cid:73) where the upper (lower) sign corresponds to mode 1 (3) and the rotation corresponds to a clockwise (counter- clockwise) rotation around e¯q,DM = RT ¯q ez. The assign- ment of sign and handedness is consistent (i) with the common definition of the rotation matrix Rz(ϕ) = cos ϕ (±) sin ϕ 0 cos ϕ 0 0 1 (∓) sin ϕ 0 (B25)  rotating vector Si¯q,1(3) in a right-handed coordinate sys- tem clockwise (counter-clockwise) around ez by an angle ϕ = ¯q · Ri with 0 ≤ ϕ ≤ π, (ii) as well as with the definition of the winding number w1(3) = = 1 2π 1 2π dϕ S1(3) × (cid:20) dϕ Sx1(3)(ϕ) dS1(3) dϕ dSy1(3)(ϕ) dϕ −Sy1(3)(ϕ) dSx1(3)(ϕ) dϕ = (∓)1 (cid:21) (B26) counting the total number of turns of the spin-spiral as a curve parametrized by ϕ with 0 ≤ ϕ ≤ 2π, where counter-clockwise motion counts as positive and clock- wise motion counts as negative integers, and (iii) with the vector spin-chirality between atom i and i + 1 de- fined in Ci,i+1(Si¯q,1(3)) = (∓)S2 sin (¯q (Ri+1 − Ri)) e¯q,DM . (B27) Alternatively we could also say, that mode 1 (3) rotates counter-clockwise (clockwise) around (∓)e¯q,DM, but this is not the definition we follow here. These two modes γ = 1, 3 are separated by an energy N S2D¯q. sity wave in the direction of D¯q, On the contrary, the mode Sq¯q,2 represents a spin den- Si¯q,2 = S e¯q,DM cos(¯q · Ri) . (B28) 17 All three modes satisfy per construction the weak con- straint (B4), but this mode does not fulfill the strong constraint (B3) and thus must be excluded from the set of solutions. ¯q takes the physical meaning of the propagation vector of the spin spiral. If the propagation vector is parallel to the DMI-vector, ¯q(cid:107)D¯q, then we call the spiral a helical or Bloch-type spin-spiral for which holds that curl S¯q,1(3) = ∇R × S¯q1(3) = (±)(¯q · e¯q,DM)S¯q1(3). If the propagation vector is perpendicular to the DMI-vector, ¯q ⊥ D¯q, then we name the spiral a cycloidal or N´eel-type spin-spiral for which curl S¯q,1(3) = (∓)(¯q · S¯q,1(3))e¯q,DM. Here the spin- spiral S¯q = S¯q(R) : R3 → S2 is a smooth function whose values coincide at the positions of the lattice vectors Ri with Si¯q. The details of the vector relation between ¯q and D¯q depend on the symmetry of the crystal lattice. So far we focused on physical realizations where the eigenvalues of the stationary state Λ¯q are exactly two- fold degenerate, namely for ¯q and −¯q for ¯q,−¯q ∈ BZ. For systems with a non-trivial point group, the ¯q-vector is equivalent to a star of p ¯q-vectors, {¯q}, formed by consideration of ¯qα = Pα ¯q for all symmetry operations α denoted by Pα of the symmetry group of the lattice. Accordingly, Λ{¯q} is p-fold degenerate, p different S¯qα can satisfy Eq. (B17) simultaneously, and the single-q state Sq¯q,γ in Eq. (B19) may be replaced by an alter- native ansatz describing a multi-q state by a superposi- tion of properly normalized Sq¯qα,γ for any choice of α taken from the symmetry group as long as the strong constraint (B3) is fulfilled. For ¯q = Q we may expect a multi-q ground state. The competition of the various possible multi-q states with the single-q state as possible ground state is typically determined by energy contribu- tions beyond the model discussed here.56 Appendix C: Extracting micromagnetic parameters from first-principles energetics of spin-spiral state An important aspect in undertaking multi-scale simu- lations of magnetic structures is the development of real- istic micromagnetic models with material-specific param- eters. Here, we show that the spin stiffness A and the spi- ralization D, which enter the micromagnetic model (see Eq. (1)) can be extracted directly from first-principles calculations of the total energy Etot(q, erot) given per magnetic atom for a planar homogeneous spin-spiral with wave vector q and fixed rotation axis erot related to the planar spiral as erot = mi × d mi/dRi. In the following we give a rather general derivation not restricted to one-dimensional chains and thus we than with the spin stiffness A and the spiralization ten- work with the spin-stiffness tensor A ∈ Rd×d rather sor D ∈ R3×d, also called matrix of Dzyaloshinskii- Moriya constants, rather than the Dzyaloshinskii-vector D. d ∈ {1, 2, 3} refers to the dimensionality of the micro- magnetic problem with d = 1 relevant for chains, domain- walls or magnetic spirals, d = 2 for films, interfaces or the treatment of skyrmions and d = 3 for bulk or bubbles for example. In the general case, the expression of the DM en- ergy density of the micromagnetic energy functional (1) translates from the one-dimensional case D· (m × m) to D : (m × ∇m) in case of higher dimensions. The expres- with matrix elements Lαβ = (cid:80) sion in the parenthesis is also called the Lifshitz matrix L(m) ∈ R3×d ⊆ R3×3, a matrix of differential one-forms α(cid:48)α(cid:48)(cid:48) , with α, β ∈ {x, y, z}. L(β) αα(cid:48) = −(mα∂βmα(cid:48) − mα(cid:48)∂βmα) are (cid:3) =(cid:80) trices: D : L(m) = tr(cid:2) the Lifshitz invariants and
is the Levi-Civita symbol. The operator ":" refers to the inner product of two ma- αβ DαβLαβ. Which of the Lifshitz invariants or which of the maximal 9 compo- nents of the D-matrix, respectively, are relevant depends on the point-group of the crystal, an aspect which is not considered here any further. α(cid:48)α(cid:48)(cid:48)
αα(cid:48)α(cid:48)(cid:48)L(β) DTL 1 Starting point of the derivation is the observation made in Appendix B that flat spirals with rotation axis ∓eDM(q) are the stationary states of the spin- lattice model with Heisenberg and DM interaction with an energy per atom of E(q, (∓)eDM(q)) = 1 2 λq,1(3) = 2 (J(q) (∓) D(q)) and two opposite rotation senses.63 The extraordinary nature of these states lies in the effi- cient realization in first-principles theory of the electronic structure. We recall from the discussion in Sec. III that by neglecting the magnetic anisotropy energy, the total energy of a spin-spiral state, Etot(q, erot) (cid:39) ESS(q) + EDM(q, erot), can be explicitly approximated by calcula- tions in two steps: self-consistent calculations of the spin- spiral energy without SOC, ESS(q), and the energy due to SOC, EDM(q, erot), in first order perturbation theory for a given rotation axis erot. We work here with a nor- malized length of magnetic spins, S2 = 1, as typical for spin-lattice and micromagnetic models. The parameters J(q) and D(q) of particular systems are then obtained from first principles by a direct comparison of the ener- gies E(q, (∓)eDM(q)) = 1 2 λq,1(3), δJ(q) = J(q) − J(0) = 2 (ESS(q) − ESS(0)) D(q) = 2EDM (q, eDM) , and (C1) (C2) ∀q ∈ BZ. Note, the equalities hold only if ESS and EDM are given as energy per magnetic atom. The back Fourier transformation of J(q) and D(q) according to Eq. (B8) provides then the Heisenberg exchange param- eter J0j and the microscopic DM-vectors D0j on the real space lattice. There might be cases where the direction eDM(q) is not known a priori. In this case we can deter- mine the three components of D(q) applying Eq. (C2) for each wavevector q and for three independent axes of spiral rotations erot, i.e., erot · D(q) = 2EDM (q, erot). In micromagnetic models one typically assumes that the ground state of the system is close to the collinear state qc, e.g., the ferromagnetic state at qc = 0 or an antiferromagnetic state at a high-symmetry point at the boundary of the Brillouin zone. At such a point qc in the Brillouin zone, J(qc) typically takes a local minimum. 18 If we assume that D(q) (cid:28) J(q) for q in the vicinity of qc and measured from qc, i.e., the long wavelength limit where q is small, the relevant energy landscape may be explored by Taylor expanding Eqs. (C1) and (C2), i.e., the exchange parameters δJ(q) = 2 J0j (cos(qRj) − 1) ≈ − J0j(Rjq)2, (C3) D(q) eDM(q) = D(q) (eDM(q) · e(k) rot) e(k) rot (cid:88) (cid:88) j≥1 j≥1 (cid:88) (cid:88) j≥1 j≥1 3(cid:88) k=1 = 2 ≈ 2 D0j sin(qRj) D0j (Rjq) (C4) and the total energy 1 VΩ Etot(q, erot) = (cid:21)T (cid:20) erot · D 2π q + qT A 4π2 q + O(q3) (C5) up to second order in q measured relative to qc, for a fixed rotation axis erot. k labels the maximally three linear independent rotation axes e(k) rot. In Eqs. (C3) and (C4) we made explicitly use of the symmetry relations J(q) = J(−q) and D(q) = −D(−q) and chose J(qc) as origin of energy. In Eq. (C5), VΩ represents the volume of the unit cell in a 3D system, the surface area in 2D, or, respectively, the spacing between magnetic atoms in 1D, so that the left-hand side represents an energy density. The numerical prefactors appear in order to keep consis- tency with the definition of the energy functional in the main text, Eq. (1). We arrive at the energy expression with the spin stiffness A obtained from the curvature of the total energy at wave vector qc, ∂2 ∂q2 ESS(q)qc A 4π2 = = 1 2VΩ 1 2VΩ (cid:88) j≥1 J0jRj ⊗ Rj . (C6) The projection of the spiralization matrix D onto the direction of the rotation vector erot, [erot · D] ∈ Rd, is obtained from the slope of the total energy calculated for the rotation axis erot, D 2π EDM(q, erot)qc = erot ⊗ = 1 VΩ ∂ ∂q 1 VΩ D0j ⊗ Rj . (cid:88) j≥1 (C7) The product denoted by "⊗" indicates the tensorial prop- erty of the spin-stiffness and the DM-matrix. These equa- tions provide the link between the micromagnetic param- eters, the ab initio results of spin-spiral calculations and a spin-lattice model. On the basis of the first expres- sion on the right-hand side of Eq. (C7) we can interpret the D-matrix as a tensor in the space spanned by the magnetization direction and real space. Appendix D: Atom-dependent micromagnetic D-vector in the Fert-Levy model obtained for a spin-spiral state Applying the Fert-Levy model42,43 for determining the microscopic DM vector in metals is very appealing due to its simplicity and clarity. It is frequently applied to determine the direction of the DM vector and interest- ing because it could provide a basis for interpreting our first-principles results. The assumptions under which the model is derived motivates this appendix, where we ex- plore the validity and applicability of the model to ex- tended periodic systems, primarily to chains. Fert and Levy investigated CuMnT ternary alloys, with a small concentration (1% or 2%) of Mn impurities and of heavy non-magnetic atoms T=Au or Pt. They found that experimental anisotropy data are explained by Mn atoms carrying a magnetic moment and interact- ing not only by the typical Rudermann-Kittel-Kasuya- 19 Yosida (RKKY) interaction, but also by a DM-type in- teraction due to spin-orbit scattering of the conduction electrons by the non-magnetic impurities. They derived the leading order expression for the DMI-energy that is first order in the spin-orbit coupling and second order in the exchange interaction, which results from an ex- pression for the third-order perturbation of the ground- state energy of the gas of conduction electrons due to the presence of the Mn spins and the non-magnetic impurity. Evaluating this expression under the assumption that (i) the magnetic moments are located at the Mn atoms and (ii) the spin-orbit interaction at the non-magnetic impu- rity atoms only, (iii) that both atom types are located in the host as impurities in the low, but not very low (> 1000 ppm), concentration limit, (iv) that Cu pro- vides the gas of homogeneous electrons described by the Fermi energy, EF, and wavevector, kF, that (v) scatter at the non-magnetic impurity with the scattering phase shift δ2(EF) and the spin-orbit strength λ, and hybridize with the 3d states of the Mn atoms described by Γ, the exchange interaction strength between the host electrons and local spins, they arrived at a trilinear expression for the DMI-energy, Eijµ DM = − 135π 32 λ Γ2 F k3 E2 F sin δ2(EF) sin[kF(Riµ + Rjµ + Rij) + δ2(EF)] · Riµ · Rjµ RiµRjµRij ( Riµ × Rjµ) · (Si × Sj) , (D1) relating three atoms: one non-magnetic impurity denoted by µ and two magnetic impurities denoted by i, j placed at the position Ri(j)µ measured from the position of the spin-orbit impurity. Rij measures the distance between the two atoms i and j. Now we apply this model to a single transition-metal chain on the Pt substrate. We are aware of the fact that neither the spin-orbit atoms Pt nor the magnetic chain atoms are in the low concentration limit. Also the Fermi surface of Pt is more complex than Cu and the isotropic approximation of the Fermi wavevector under- lying this model is a further approximation. Further, we assume that all Pt atoms, irrespective of their distance and position from the chain are electronically identical, i.e., δµ 2 = δ2, and λµ = λ. In difference to the ab ini- tio calculations we consider here a truly single magnetic chain and no periodic repetition of the surface unit cell. This is a good approximation for Pt atoms µ close to the chains, but differences are expected for atoms in the center of the terrace as they experience competitive in- teractions to chains at the upper and lower terrace. The DM-energy contribution of a magnetic texture which arises solely through the presence of a certain spin- orbit atom µ is given by (cid:88) Eµ DM = Eijµ DM . (D2) (cid:104)i,j(cid:105) The brackets denote a summation over all pairs of mag- netic chain-atoms i and j. For a fixed atom µ, the direc- tion of Riµ× Rjµ in Eq. (D2) is always the same, irrespec- tive of i and j, and we denote it by nµ = ey × dµ. Here, dµ is the unit vector pointing from the atom µ into the direction of shortest distance to the chain. Furthermore, we define ϕiµ as the angle between Riµ and dµ. In the spirit of the first-principles calculations, we next consider a homogeneous spin spiral for which Si × Sj = sin(q ay (j − i)) erot, where erot is the rotation axis, and we obtain Eµ DM(q) = −C εµ(q) nµ · erot with C = 135π sin δ2(EF) and λ Γ2 F k3 E2 F 32 cos(ϕjµ − ϕiµ) RiµRjµRij (cid:88) (cid:104)i,j(cid:105) εµ(q) = sin[kF(Riµ + Rjµ + Rij) + δ2(EF)] · sin(ϕjµ − ϕiµ) sin(q ay (j − i)) . 20 (D3) (D4) Clearly, Eµ nµ, depending on the sign of the prefactor. DM is lowest if erot is parallel (anti-parallel) to The contribution of this atom to the spiralization is defined as Dµ 2π = 1 ay ∂Eµ ∂q DM (cid:12)(cid:12)(cid:12)(cid:12)q=0 · nµ . (D5) Unfortunately, Dµ diverges in the limit q → 0 for this periodic model for one dimension, due to an effective 1/R-dependence for each of the two sums over i and j contained in Eq. (D5). In a realistic solid we expect a fi- nite phase coherence length of the wave function or some structural or chemical disorder, which truncates the in- teraction range of the atoms in the finite chain and thus the summations in the sums in Eq. (D4). This would prevent the divergence of Eq. (D5). In contrast, the energies are well behaved and we have εµ(q) → 0 for q → 0 (see Eq. (D4)). To compare to our ab initio results in Sec. IV (Fig. 5), we evaluate for the rest of this appendix the energy at a fixed wave vec- tor q0 = 0.05 2π (corresponding to a pitch of 5.6 nm) as ay magnetic state and calculate Eµ DM numerically. For this q0, the evaluation of the sum in Eq. (D4) in a supercell containing 4000 unit cells in the ±y-direction yields well converged results. We notice, however, that the conver- gence depends on the value of q0, i.e., that a lower value of q0 would require a larger number of unit cells in order to reach convergence. The atom-resolved contributions to the D-vector are then approximated by the finite dif- ference, Dµ ≈ const · εµ(q0) q0 · nµ . (D6) These values compare to the atom-resolved spiralization from our first-principles calculations presented in Fig. 5 and discussed in Sec. IV of the main text. Here, we only discuss the results predicted by the model: The type of the 3d atom (i.e., Mn, Fe, or Co) only enters the prefactor in Eq. (D6) through the parameter Γ. Up to a sign, the directions nµ (visualized by arrows in Fig. 5(d) in the main text) are independent of the type of magnetic chain 3d atoms. The dependence of the DMI strength on the substrate atom is governed by εµ for a fixed q0. For the results in this paper that utilize the Fert-Levy model (see Eq. (D4)) we use the following parameters: 2πk−1 F = 2 nm as given by Ref. 57 and δ2 = π 10 Zd, where Zd = 9.4 gives the number of d-electrons.42 Fig. 9: (color online) Dependence of the DMI energy origi- nating form a Pt atom, as function of the distance d of the Pt atom normal to the chain. The nearest-neighbor distance is denoted by ay. For the magnetic chain, we distinguish spin spirals of ferromagnetic (FM-SS) or antiferromagnetic (AFM- SS) short-range order, evaluated at a fixed q0 = 0.05 2π . ay Points and squares highlight the actual positions of atoms in the Pt(664) unit cell. Two different geometries need to be considered. For a description of the geometries see text. In Fig. 9, we analyze εµ as function of the distance d of Pt atom µ to the chain for two different cases: that the spin-spiral is of (i) ferromagnetic short-range order (FM- SS) as the case for Fe and Co, or of (ii) anti-ferromagnetic type (AFM-SS) as in the case of Mn (see circles and squares, respectively). More precisely, ε(dµ) is a func- tion of the distance d of atom µ to the chain if we dis- tinguish two geometrical cases: (a) that the projection of the position of atom µ coincides with the position of a 3d atom, or (b) that this projection is in the middle of two 3d atoms (see Fig. 9). This distance dependence is indi- cated by solid and broken lines, respectively. We observe typical RKKY-like oscillations that decay approximately as 1/d2. In total, we find that for the AFM case the DMI strength is smaller and decays faster with distance than for the FM case. As a result, in AFM-SS the first maxi- mum determines the overall DMI strength nearly alone. Moreover, the period length of the oscillations is nearly 012345distance d normal to the chain (ay)−0.2−0.10.00.10.20.30.40.5DMI strength ε(d) (arb. units)FM-SS, geom. aFM-SS, geom. bAFM-SS, geom. aAFM-SS, geom. bby a factor 2 larger in FM-SS (4.5 ay and 2.5 ay for FM- SS and AFM-SS, respectively). The disagreement between the Fert-Levy model and the ab initio results for Co (see Figs. 5(c) and 5(d)) can- not be resolved by adjusting the parameters for Co or the different Pt atoms (e.g., the phase shifts or Fermi wavevector) since for the most important Pt atoms, those next to Co, the direction of D does not coincide at all with the model of Fert and Levy, where the direction is exclusively determined by geometry. Maybe in the case of Co the interaction between the upper and lower Co chains, which is included in our ab initio calculations, but neglected in the Fert-Levy model contributes to this difference. This and the extension of the model to films and heterostructures will be a matter of future investiga- tions. 21 1 P. Gambardella, A. Dallmeyer, K. Maiti, M. C. Malagoli, W. Eberhardt, K. Kern, and C. Carbone, Nature 416, 301 (2002). 2 P. Gambardella, J. Phys.: Condens. Matter 15, S2533 (2003). 3 A. Dallmeyer, C. Carbone, W. Eberhardt, C. Pampuch, O. Rader, W. Gudat, P. Gambardella, and K. Kern, Phys. Rev. B 61, R5133 (2000). 4 P. Gambardella, M. Blanc, H. Brune, K. Kuhnke, and K. Kern, Phys. Rev. B 61, 2254 (2000). 5 P. Gambardella, M. Blanc, L. Burgi, K. Kuhnke, and K. Kern, Surf. Sci. 449, 93 (2000). 6 M. Komelj, C. Ederer, J. W. Davenport, and M. Fahnle, Phys. Rev. B 66, 140407 (2002). 7 A. B. Shick, F. M´aca, and P. M. Oppeneer, Phys. Rev. B 69, 212410 (2004). 8 B. ´Ujfalussy, B. Lazarovits, L. Szunyogh, G. M. Stocks, and P. Weinberger, Phys. Rev. B 70, 100404 (2004). 9 M. Komelj, D. Steiauf, and M. Fahnle, Phys. Rev. B 73, 134428 (2006). 10 S. Baud, C. Ramseyer, G. Bihlmayer, and S. Blugel, Phys. Rev. B 73, 104427 (2006). 11 S. Baud, G. Bihlmayer, S. Blugel, and C. Ramseyer, Surf. Sci. 600, 4301 (2006). 12 J. Honolka, T. Y. Lee, K. Kuhnke, A. Enders, R. Skom- ski, S. Bornemann, S. Mankovsky, J. Min´ar, J. Staunton, H. Ebert, et al., Phys. Rev. Lett. 102, 067207 (2009). 13 J. Honolka, T. Y. Lee, K. Kuhnke, D. Repetto, V. Sessi, P. Wahl, A. Buchsbaum, P. Varga, S. Gardonio, C. Car- bone, et al., Phys. Rev. B 79, 104430 (2009). 14 C. Carbone, S. Gardonio, P. Moras, S. Lounis, M. Heide, G. Bihlmayer, N. Atodiresei, P. H. Dederichs, S. Blugel, S. Vlaic, et al., Adv. Funct. Mater. 21, 1212 (2011). 15 L. Hammer, W. Meier, A. Schmidt, and K. Heinz, Phys. Rev. B 67, 125422 (2003). 16 B. Dup´e, J. E. Bickel, Y. Mokrousov, F. Otte, K. von Bergmann, A. Kubetzka, S. Heinze, and R. Wiesendanger, New J. Phys. 17, 023014 (2015). 17 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). 18 I. E. Dzyaloshinskii, Sov. Phys. JETP 5, 1259 (1957). 19 T. Moriya, Phys. Rev. 120, 91 (1960). 20 B. Santos, J. M. Puerta, J. I. Cerda, R. Stumpf, K. von Bergmann, R. Wiesendanger, M. Bode, K. F. McCarty, and J. de la Figuera, New J. Phys. 10, 013005 (2008). 21 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). 22 B. Zimmermann, M. Heide, G. Bihlmayer, and S. Blugel, Phys. Rev. B 90, 115427 (2014). 23 S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Ku- betzka, R. Wiesendanger, G. Bihlmayer, and S. Blugel, Nature Physics 7, 713 (2011). 24 N. Romming, C. Hanneken, M. Menzel, J. E. Bickel, B. Wolter, K. von Bergmann, A. Kubetzka, and R. Wiesen- danger, Science 341, 636 (2013). 25 Y. Mokrousov, A. Thiess, and S. Heinze, Phys. Rev. B 80, 195420 (2009). 26 R. Mazzarello and E. Tosatti, Phys. Rev. B 79, 134402 (2009). 27 M. Menzel, Y. Mokrousov, R. Wieser, J. E. Bickel, E. Vedmedenko, S. Blugel, S. Heinze, K. von Bergmann, A. Kubetzka, and R. Wiesendanger, Phys. Rev. Lett. 108, 197204 (2012). 28 M. Heide, G. Bihlmayer, and S. Blugel, J. Nanosci. Nan- otechnol. 11, 3005 (2011). 29 M. Lezai´c, P. Mavropoulos, G. Bihlmayer, and S. Blugel, Phys. Rev. B 88, 134403 (2013). 30 I. E. Dzyaloshinskii, Sov. Phys. JETP 20, 665 (1965). 31 Y. A. Izyumov, Sov. Phys. Usp. 27, 845 (1984). 32 F. Freimuth, S. Blugel, and Y. Mokrousov, J. Phys.: Con- dens. Matter 26, 104202 (2014). 33 H. Krakauer, M. Posternak, and A. J. Freeman, Phys. Rev. B 19, 1706 (1979). 34 E. Wimmer, H. Krakauer, M. Weinert, and A. J. Freeman, Phys. Rev. B 24, 864 (1981). 35 For program description see http://www.flapw.de. 36 A. Mosca Conte, S. Fabris, and S. Baroni, Phys. Rev. B 78, 014416 (2008). 37 V. L. Moruzzi, J. F. Janak, and A. R. Williams, Calculated electronic properties of metals (Pergamon Press, 1978). 38 A. R. Mackintosh and O. K. Andersen, Cambridge Univer- sity Press 77, 149 (1980). 39 M. Heide, G. Bihlmayer, and S. Blugel, Phys. B: Condens. Matter 404, 2678 (2009). 40 L. M. Sandratskii, J. Phys.: Condens. Matter 3, 8565 (1991). 41 G. Bihlmayer, Y. Koroteev, P. Echenique, E. Chulkov, and S. Blugel, Surf. Sci. 600, 3888 (2006). 42 A. Fert and P. M. Levy, Phys. Rev. Lett. 44, 1538 (1980). 43 P. M. Levy and A. Fert, Phys. Rev. B 23, 4667 (1981). 44 V. Kashid, T. Schena, B. Zimmermann, Y. Mokrousov, S. Blugel, V. Shah, and H. G. Salunke, Phys. Rev. B 90, 054412 (2014). 45 M. Heide, G. Bihlmayer, and S. Blugel, Phys. Rev. B 78, 140403 (2008). 46 P. Gambardella, A. Dallmeyer, K. Maiti, M. C. Malagoli, S. Rusponi, P. Ohresser, W. Eberhardt, C. Carbone, and K. Kern, Phys. Rev. Lett. 93, 077203 (2004). 47 F. N. Rybakov, A. B. Borisov, and A. N. Bogdanov, Phys. 22 Rev. B 87, 094424 (2013). 48 S. Rohart and A. Thiaville, Phys. Rev. B 88, 184422 (2013). 49 D. S. G. Bauer, P. Mavropoulos, R. Zeller, and S. Blugel, unpublished. 50 A. Yoshimori, J. Phys. Soc. Jpn 14, 807 (1959). 51 J. Villain, J. Phys. Chem. Solids 11, 303 (1959). 52 T. A. Kaplan, Phys. Rev. 116, 888 (1959). 53 T. A. Kaplan, Phys. Rev. B 79, 229901 (2009). 54 J. M. Luttinger and L. Tisza, Phys. Rev. 70, 954 (1946). 55 J. M. Luttinger, Phys. Rev. 81, 1015 (1951). 56 P. Kurz, G. Bihlmayer, K. Hirai, and S. Blugel, Phys. Rev. Lett. 86, 1106 (2001). 57 L. Zhou, J. Wiebe, S. Lounis, E. Vedmedenko, F. Meier, S. Blugel, P. H. Dederichs, and R. Wiesendanger, Nat. Phys. 6, 187 (2010). 58 G. Bihlmayer, Magnetism goes Nano (36th Spring School of the Institute of Solid State Research (IFF), 2005), chap. C2 Reduced Dimensions II: Magnetic Anisotropy. 59 In general, one would need to include the non-local dipole- dipole interaction as well. Following the estimate in Ref. 58, however, its contribution turns out to be negligible. 60 In Eqs. (3) and (4), we use atan2 (y, x) = Arg(x + iy) ∈ (−180◦, 180◦], the generalized form of the arcus-tangent x ∈ (−90◦, 90◦), allowing to properly ac- function arctan y count for all sign combinations of x and y. 61 The complete elliptic functions of first and second kind dφ (cid:0)1 −
2 sin2(φ)(cid:1)−1/2 and are defined by K(
) = (cid:82) π/2 E(
) =(cid:82) π/2 dφ (cid:0)1 −
2 sin2(φ)(cid:1)1/2, respectively. 0 0 62 In Sec. II C it was pointed out that for the same set of parameters the criterion for the appearance of an inhomo- geneous spin spiral is always more likely to be fulfilled than the one for the homogeneous spiral. Nevertheless we will still present our findings regarding both spiral types, which enables the reader to compare. 63 That λq,γ corresponds to an energy per magnetic atom can be inferred from Eq. (B20). 

1008.3209

1

1008

Tunneling Spin Injection into Single Layer Graphene (Supplementary Information)

We achieve tunneling spin injection from Co into single layer graphene (SLG) using TiO2 seeded MgO barriers. A non-local magnetoresistance ({\Delta}RNL) of 130 {\Omega} is observed at room temperature, which is the largest value observed in any material. Investigating {\Delta}RNL vs. SLG conductivity from the transparent to the tunneling contact regimes demonstrates the contrasting behaviors predicted by the drift-diffusion theory of spin transport. Furthermore, tunnel barriers reduce the contact-induced spin relaxation and are therefore important for future investigations of spin relaxation in graphene.

cond-mat.mes-hall

cond-mat

Tunneling Spin Injection into Single Layer Graphene Wei Han, K. Pi, K. M. McCreary, Yan Li, Jared J. I. Wong, A. G. Swartz, and R. K. Kawakami† Department of Physics and Astronomy, University of California, Riverside, CA 92521 † e-mail: roland.kawakami@ucr.edu Online Supplementary Information Content: 1. Dependence of non-local MR on contact resistance and gate voltage, based on the drift-diffusion model. 2. Local MR measurement. 1 1. Dependence of non-local MR on contact resistance and gate voltage, based on the drift-diffusion model. Following Takahashi and Maekawa [1], the non-local MR is given by: RNL = 4 RGe(cid:1) L / (cid:3)G RJ (cid:4) PJ (cid:8) RG (cid:8) 2 + 1 (cid:1) PJ (cid:8) (cid:8) (cid:5) PF RF RG 2 1 (cid:1) PF 2 (cid:6) (cid:9) (cid:9) (cid:9) (cid:9) (cid:7) (cid:4) (cid:4) (cid:8) (cid:8) (cid:8) (cid:2) 1 + (cid:8) (cid:8) (cid:8) (cid:8) (cid:8) (cid:5) (cid:5) 2 RJ RG 2 + 1 (cid:1) PJ 2 RF RG 2 1 (cid:1) PF 2 (cid:6) (cid:9) (cid:9) (cid:9) (cid:9) (cid:7) (cid:1) e(cid:1)2 L / (cid:3)G (cid:1)1 (cid:6) (cid:9) (cid:9) (cid:9) (cid:9) (cid:7) (S1) where RG = (cid:1)G (cid:2)GW( ) and RF = (cid:2)F(cid:1)F / AJ are the spin resistances of the SLG and FM electrodes, respectively, W is the width of the SLG, AJ is the junction area between the FM and SLG, (cid:2)G ((cid:2)F) is the spin diffusion length in the SLG (FM), (cid:2)G is the conductivity of SLG, (cid:3)F is the resistivity of the FM, PF is the spin polarization of the FM, PJ is the polarization of the interfacial current, RJ is the contact resistance between FM and SLG, and L is the spacing between the injector and detector electrodes. This equation shows that increasing the contact resistance produces a strong enhancement of (cid:1)RNL that saturates as RJ becomes significantly larger than RG. Figure S1 shows the non-local MR as a function of contact resistance based on equation S1 with typical values of W = 2 μm, L = 2 μm, (cid:1)G = 2 μm, (cid:2)G = 0.5 mS, PF = 0.4 [2], (cid:2)F = 6(cid:1)10-8 (cid:1) m [3], and (cid:1)F = 0.06 μm [4]. We plot two curves corresponding to junction polarizations of PJ = 0.4 and 0.2 because this parameter can vary depending on the microscopic properties of the interface. Equation S1 also shows that the relationship of (cid:1)RNL vs. (cid:2)G is strongly dependent on the contact resistance. Although the (cid:1)RNL vs. (cid:2)G relation is applicable to any material system, it has never been verified experimentally across different contact regimes. The gate tunable conductivity of SLG provides a unique opportunity to investigate this behavior. 2 For transparent contacts (RJ << RG), equation S1 reduces to: (cid:1)RNL = 2 4 PF 2 ) 2 RG (1 (cid:2) PF (cid:5) RF (cid:9) RG (cid:6) 2 (cid:7) (cid:10) (cid:8) e(cid:2) L / (cid:3)G 1 (cid:2) e(cid:2)2 L / (cid:3)G ~ (cid:4)G (S2) The top curve of Figure 2 shows the calculated gate dependence of the non-local MR (normalized by its value at zero gate voltage) for transparent contacts. The conductivity is assumed to vary linearly with Vg away from the Dirac point according to (cid:2)G = (cid:2)0 + μ(cid:1)Vg 2/h [5]), μ [5], where (cid:4)0 is the minimum conductivity (assumed to be 4e is the mobility (taken to be 2000 cm2/Vs), e is the electron charge, and (cid:2) is the capacitance per area (taken to be 1.15(cid:1)10-8 F/cm2 for 300 nm of SiO2). Due to the proportionality of (cid:1)RNL and (cid:4)G, the gate dependence of the non-local MR has a minimum at the Dirac point (Figure 2). For simplicity, we have assumed that (cid:3)G is independent of Vg. We note, however, that modest enhancements of (cid:3)G with increased carrier concentration have been reported [6], which should lead to slightly enhanced non-local MR away from the Dirac point. The low value of non-local MR (at RJ = 0 in Figure S1) in the calculation is due to the conductance mismatch term (RF/RG)2 << 1 in equation S2. Intuitively, the increase of non-local MR with increasing conductivity occurs because the conductance mismatch between the Co and SLG is reduced [7]. For intermediate contact resistance (RJ ~ RG, RF << RG, RF << RJ), equation S1 reduces to: (cid:1)RNL = 4 RG RJ (cid:4) PJ (cid:8) RG (cid:8) 2 1 (cid:2) PJ (cid:8) (cid:8) (cid:5) 2 (cid:6) (cid:9) (cid:9) (cid:9) (cid:9) (cid:7) (cid:4) (cid:4) (cid:8) (cid:8) (cid:8) 1 + (cid:8) (cid:8) (cid:8) (cid:8) (cid:8) (cid:5) (cid:5) 2 RJ RG 2 1 (cid:2) PJ 2 (cid:6) (cid:9) (cid:9) (cid:9) (cid:9) (cid:7) (cid:2) e(cid:2)2 L / (cid:3)G (cid:2)1 (cid:6) (cid:9) (cid:9) (cid:9) (cid:9) (cid:7) e(cid:2) L / (cid:3)G (S3) The middle curve of Figure 2 shows the gate dependence of non-local MR calculated 3 from equation S3 using RJ = 1 k(cid:1), PJ = 0.2, and (cid:2)G defined above. The non-local MR exhibits a relatively weak dependence on gate voltage with a shallow minimum at the Dirac point. For tunneling contacts (RJ >> RG), equation S1 reduces to: (cid:1)RNL = 1 (cid:4)G 2 PJ (cid:3)G W e(cid:2) L / (cid:3)G ~ 1 (cid:4)G (S4) The bottom curve of Figure 2 is a plot of the normalized non-local MR as a function of gate voltage based on equation S4. The presence of the tunnel barrier alleviates the conductance mismatch between the Co and SLG, as seen by the absence of the term (RF/RG)2 in equation S4. This leads to the increase in the non-local MR observed experimentally in Figure 1c and theoretically in Figure S1. Notably, the non-local MR has a maximum at the Dirac point and is inversely proportional to the SLG conductivity. 2. Local MR measurement. We measure a local MR signal of 0.4% at Vg = 0 V (Figure S2a), obtained by measuring the resistance across E2 and E3 as the magnetic field is ramped. Comparing the local and non-local MR scans (Figures S2a and S2b), the magnetization switching fields of the injector and detector electrodes match. It is also observed that the resistance change for the local MR is ~200 (cid:1), which is roughly twice the non-local MR (~100 (cid:1)). This relationship between local MR and the non-local MR is expected theoretically [8, 9]. References: [1] S. Takahashi, and S. Maekawa, Phys. Rev. B 67, 052409 (2003). [2] R. J. Soulen Jr. et al., Science 282, 85 (1998). 4 [3] D. R. Lide, CRC Handbook of Chem. and Phys., CRC Press, Boca Raton, Florida, USA (1998). [4] L. Piraux et al., Eur. Phys. J. B. 4, 413 (1998). [5] A. K. Geim, and K. S. Novoselov, Nature Mater. 6, 183 (2007). [6] M. Popinciuc et al., Phys. Rev. B 80, 214427 (2009). [7] G. Schmidt et al., Phys. Rev. B 62, 4790(R) (2000). [8] A. Fert, and S.-F. Lee, Phys. Rev. B 53, 6554 (1996). [9] F. J. Jedema et al., Phys. Rev. B 67, 085319 (2003). Figure Captions: Figure S1: The non-local MR as a function of contact resistance (RJ) with PJ = 0.4 and PJ = 0.2, based on equation S1 and parameters given in the text. Figure S2: (a, b) Local MR and non-local MR measurements of a graphene spin valve, respectively. Insets: Measurement geometries. 5 Figure S1 120 80 40 ) (cid:2) ( L N R (cid:1) 0 0 PJ = 0.4 PJ = 0.2 20 RJ (k(cid:2)) 40 Figure S2 50.4 50.2 50.0 49.8 0.2 0.0 -0.2 -0.4 ) (cid:1) k ( R ) (cid:1) k ( L N R (a) Local T = 4 K R (b) Non-local T = 4 K I VNL -100 -50 0 H (mT) 50 100 

1609.08982

1

1609

Light propagation in tunable exciton-polariton one-dimensional photonic crystals

Simulations of propagation of light beams in specially designed multilayer semiconductor structures (one-dimensional photonic crystals) with embedded quantum wells reveal characteristic optical properties of resonant hyperbolic metamaterials. A strong dependence of the refraction angle and the optical beam spread on the exciton radiative lifetime is revealed. We demonstrate the strong negative refraction of light and the control of the group velocity of light by an external bias through its effect upon the exciton radiative properties.

cond-mat.mes-hall

cond-mat

a Light propagation in tunable exciton-polariton one-dimensional photonic crystals E. S. Sedov,1, 2, ∗ E. D. Cherotchenko,1 S. M. Arakelian,3 and A. V. Kavokin1, 4, 5, † 1School of Physics and Astronomy, University of Southampton, SO17 1NJ Southampton, United Kingdom 2Department of Physics and Applied Mathematics, Vladimir State University named after A. G. and N. G. Stoletovs, Gorky str. 87, 600000, Vladimir, Russia 3Department of Physics and Applied Mathematics, Vladimir State University named after A. G. and N. G. Stoletovs, Gorky street 87, 600000, Vladimir, Russia 4CNR-SPIN, Viale del Politecnico 1, I-00133, Rome, Italy 5Spin Optics Laboratory, St. Petersburg State University, Ul'anovskaya 1, Peterhof, St. Petersburg 198504, Russia Simulations of propagation of light beams in specially designed multilayer semiconductor struc- tures (one-dimensional photonic crystals) with embedded quantum wells reveal characteristic optical properties of resonant hyperbolic metamaterials. A strong dependence of the refraction angle and the optical beam spread on the exciton radiative lifetime is revealed. We demonstrate the strong negative refraction of light and the control of the group velocity of light by an external bias through its effect upon the exciton radiative properties. I. INTRODUCTION Metamaterials are artificial composite structures that demonstrate unusual optical properties not achievable in natural materials1,2. Hyperbolic metamaterials (HMMs) represent an important group of metamaterials charac- terised by specific relations between components of di- electric permittivity (ε) and magnetic susceptibility (µ) tensors2. Namely, the diagonal components of either ε or µ tensors in HMMs have opposite signs. The presently studied HMMs are mostly based on metal-dielectric composites, whereas the presence of spe- cially designed metallic elements, e. g. metallic spheres, disks, rods, etc., provides the required properties of ε and µ tensors3 -- 6. The successful application of this approach to the fabrication of HMMs has been demonstrated in the microwave spectral range with use of metal wires and split ring resonators7,8. The metal-dielectric HMMs have been developed in the near-infrared and visible fre- quency bands, see e. g.9 -- 12. One of the intriguing effects observed in HMMs is negative refraction4,8,13,14. This effect is very promising for the development of hyper- lenses9,15 characterized by the spatial resolution beyond the diffraction limit as well as in optical cloaking16,17. As noticed in13, negative refraction materials can be based not only on HMMs but on photonic crystal (PC) structures as well. Although both HMMs and PCs are complex structured materials, unlike the former, which can be considered as a quasi-homogeneous medium with the effective constitutive parameters, the latter possesses the elementary building blocks being of the same size or- der as the impinging wavelength. Various optical effects in periodically stratified media that are, in fact, one- dimensional PC's caused by anomalous refraction have been observed in works of Russell (see e.g. Refs. 18 -- 20). It has been shown in Ref. 21 that the hyperbolic disper- sion of light modes can be achieved in such structures with positive ε and µ tensors. Recent publications 22 -- 24 provide further details on the design of pure dielectric PC-based materials with hyperbolic dispersion. Both in the metal-dielectric metamaterials and in the PC-based materials the propagation of light is governed by the structure parameters and composition, so that the external control of light trajectories is hardly achiev- able. Meanwhile, it is technologically important to pro- vide a significant tunability of the optical properties of the materials, e. g. using ultrashort optical pulse pump- ing of free carriers25, introducing small perturbations in the near field of the photon modes or temperature tun- ing of the refractive index26. To answer this technological challenge one may try to embed in PCs objects that will- ingly respond to the external impact and help modifying the optical properties of the whole system27. The role of such tunable elements can be played e. g. by ultracold two-level atoms28,29, quantum dots30, diamond nitrogen- vacancy centers31, or Cooper-pair boxes. Recently, specially designed planar multilayer semicon- ductor structures demonstrating properties of resonant HMMs (RHMMs) have been theoretically proposed. The model structure considered in32 represents a modified semiconductor Bragg mirror, containing periodically ar- ranged quantum wells (QWs). It was shown by modeling that such a structure should demonstrate properties of HMMs in a spectral range where the dispersion of its op- tical eigen-modes acquires a hyperbolic character. Based on this similarity of eigen-mode dispersion properties, we refer to such a structure as RHMM. This structure may be also qualified as a PC because its period is on the or- der of a half wavelength of light. We shall use the term HMM because it was introduced in the previous pub- lication32 and it correctly describes the phenomenology of light propagation in the considered structures. The structure proposed in Ref. 32 has a significant advan- tage over traditional HMMs since it contains no metallic elements that would absorb and scatter light leading to unavoidable losses. In addition, it offers the possibility of tuning its optical properties by applying external electric and magnetic fields that modify the exciton radiative life- time and, consequently, affect the exciton-light coupling strength. given by sin(kzjdj ) Tdj = cos(kzjdj) TQW = 1 1! , 2 kz1rQW k0tQW ikzj k0 0 ik0 kzj sin(kzjdj ) cos(kzj dj) ! , 2 (2a) (2b) FIG. 1. (Color online) Schematic picture of RHMM, a spatially-periodic array of alternating dielectric layers of dif- ferent widths and refractive indices, with the layers of one type contain single QWs placed in their centers. In this paper, we demonstrate how the proposed RHMM allows tailoring of the wave packets and control of the effective refractive index and the group velocity of light. We model propagation of light beams in such RHMMs and show the negative refraction and the control of the trajectory of the beams by external bias. 1,2k2 0 − k2 where k0 = ω/c, and kz1,2 = qn2 ρ, and kρ = (kx, ky) is the in-plane wavevector component; n1,2 are the refractive indices of the first and second sublayers respectively. In the calculations we assume QWs to be infinitely thin. The coefficients rQW and tQW = 1 + rQW are the amplitude reflection and transmission coefficients of the QW. According to34,35, the former is given by rQW = in1k0Γ0/ kz1 ωX − ω − i (Γ + n1k0Γ0/ kz1) . (3) where Γ0 and Γ determine the radiative and nonradiative decay rates, respectively, and ωX is the QW exciton res- onance frequency. The relative QW permittivity in the growth direction can be estimated with respect to the following expression: εQW = ε1(cid:18)1 + 2n1Γ0/kz1D ωX − ω − iΓ(cid:19) . (4) II. NEGATIVE EFFECTIVE MASS IN RHMM: THE LOW ANGLE OF INCIDENCE LIMIT The dispersion equation for the eigen-modes of an in- finite periodic structure is given by We consider the structure schematically shown in Fig. 1 which was first proposed by some of the authors of this paper in32. The structure represents a spatially- periodic array of alternating dielectric layers of different widths and refractive indices, with the layers of one type containing single QWs placed in their centers. We as- sume cylindric symmetry of the system and introduce the in-QW-plane radial coordinate ρ. To describe the optical dispersion properties of the structure, we use the transfer matrix technique following33,34. Namely, we introduce a vector Φ = (E(r, t), cB(r, t))T, where E(r, t) and B(r, t) are the am- plitudes of electric and magnetic fields, a superscript (T) denotes transposition. Considering propagation in the z direction that coincides with the structure growth axis, it is possible to link the field amplitude of a light wave entering the layer of width D with one of a light wave leaving this layer by the following equation: T Φz=zi = Φz=zi+D , (1) where T is the transfer matrix through the period of the structure. Since each period D is formed by four suc- cessive layers, namely, the first type half layer, QW, the first type half layer again, and the second type layer, the transfer matrix T is found as a product of the transfer matrices through the corresponding layers in the reverse TQW Td1/2. The submatrices are order, i.e., T = Td2 Td1/2 cos(KD) = Tr[ T ], 1 2 (5) where K is a normal to QW-plane pseudo-wave vector component. Figures 2 (a) and (b) demonstrate disper- sions of the light modes in a modified Bragg mirror struc- ture without [Fig. 2(a)] and with [Fig. 2(b)] embedded periodically arranged narrow QWs characterized by the radiative decay rate Γ0 = 2 meV. As a model struc- ture we consider a GaN/Al0.3Ga0.7N Bragg mirror with embedded thin In0.12Ga0.88N quantum wells. The thick- nesses of the layers and their refractive indices are taken as d1 = 64.8 nm, n1 = 2.55 and d2 = 115.3 nm, n2 = 2.15, the period of the lattice D = d1 + d2 is 180.1 nm. For the given parameters the structure exhibits a second pho- tonic band gap centered to ωB ≃ 3 eV in a QW-free case, see. Fig. 2(a). The QW exciton resonance energy ωX is tuned close to the lower boundary of the second photonic band gap, ωX ≃ 2.95 eV. The QW nonradia- tive decay rate is taken as Γ = 0.1 meV. The radiative decay rate Γ0 is a tunable parameter that strongly de- pends on the applied electric field. The tuning of Γ0 by the external bias is addressed in the Appendix. The presence of QWs leads to two principal changes in the dispersion of the eigen-modes. The first one is the appearance of four dispersion branches instead of the two characteristic of a QW-free structure due to the vac- uum field Rabi-splitting originated from the QW exciton- photon coupling. The resulting eigen-states of the system are exciton-polaritons. 3 The exciton radiative decay rate is controlled by the ap- plied electric field (see the Appendix). Figure 2(c) demonstrates equi-frequency contours (EFCs) in the (Kkρ) plane for a number of LB eigen energies ω for different values of Γ0. It is clearly seen, TABLE I. Relative QW permittivity εQW in the growth di- rection. Γ0 = 2 meV Γ0 = 10 meV ω εQW εQW 2.84 eV 6.59372 + i0.00008 6.95861 + i0.00041 2.89 eV 6.66685 + i0.00027 7.32424 + i0.00137 2.94 eV 7.47172 + i0.00969 11.34860 + i0.04846 that with the increase of Γ0 opposite branches of EFCs approach each other until the gap in the K direction closes and the gap in the kρ direction opens. Table I be- low gives values of the relative QW permittivity εQW in the growth direction estimated with respect to Eq. (4) for Γ0 and ω used in Fig. 2(c). The case Γ0 = 0 cor- responds to the absence of QWs in the structure, hence the permittivity of GaN layers remains unmodulated and equal to ε1. Hereafter, we consider only the LB and neglect three upper polariton branches that are split in energy and do not affect the propagation of light in the spectral range of our interest. Our structure behaves like a HMM in the specific frequency range in the vicinity of the LB saddle point (K, kρ = 0). A. A QW-free modified Bragg mirror Let us first consider the QW-free structure assuming Γ0 = 0. Following Ref. 22, we can easily obtain the saddle-point frequency ω0. To do this, we restrict our- selves to the limit K, kρ ≪ 1/D, i.e., we consider the system in the vicinity of the center of the first Brillouin zone (BZ). We make the Taylor expansion of Eq. (5) and obtain in the zeroth order 1 = cos (θ1) cos (θ2) − n2 1 + n2 2 2n1n2 sin (θ1) sin (θ2) , (6) where θ1,2 = ω0jn1,2d1,2/c. The subscript j numerates dispersion branches. For the QW-free structure j = 1, 2. In the general form the effective photonic mass tensor components are given by m∗ with kα = K, kρ. Taking the second derivative of both right and left parts of Eq. (5) over the wave vector components in the saddle point it is easy to obtain analytical expres- sions for the effective photonic mass tensor components: α(cid:1)−1 j,α = (cid:0) ∂2ωj(cid:14) ∂k2 FIG. 2. (Color online) Dispersion of the photonic eigen-modes for the structure (a) without and (b) with embedded QWs (b). (c) -- Equi-frequency contours in the reciprocal space showing the structure eigen modes belonging to the lowest dispersion band corresponding to the different values of Γ0: Γ0 = 0 for the red (solid) curves, Γ0 = 2 meV for the green (dashed) curves, and Γ0 = 10 meV for the blue (dash- dotted) curves. Different thicknesses of the curves correspond to different energies ω (from the thickest to the thinnest): 2.94 meV, 2.89 meV and 2.84 meV, respectively. (Table I con- tains the corresponding relative QW permittivities εQW.) (d) Inverse exciton-polariton effective mass tensor components in the structure. The red surface corresponds to the effective mass in the z-direction, m∗ z (z being the growth axis of the structure), and the green surface corresponds to the in-plane effective mass, m∗ ρ. The parameters used in the calculation are given in the text. The QW radiative decay rate for (d) is taken as Γ0 = 2 meV. The other change is the formation of the three- dimensional polaritonic band gap [see Fig. 2(b)]. It is necessary to mention, that the presence of QWs pushes the lowest dispersion branch (LB) to the lower energies and the greater the shift, the larger the value of Γ0. If a QW exciton is resonant with an eigenmode of the pho- tonic cavity structure, the exciton-photon results in the appearance of new exciton-polariton eigen modes, in the strong coupling regime. In the limit of weak coupling, exciton and photon modes would cross each other. In contrast, the strong coupling manifests itself in the anti- crossing (avoided crossing) of the modes. The dispersion of the structure eigenmodes is no more photonic or exci- tonic, but polaritonic. The increase of the exciton-photon coupling due to the increase of the exciton radiative de- cay rate results in the growing energy level repulsion. =  D2c(cid:20)sin(θ2) cos(θ1)(cid:18)d2n2 + d1(n2 1 + n2 2) 2n2 (cid:19) + sin(θ1) cos(θ2)(cid:18)d1n1 + d2(n2 1 + n2 2) 2n1 (cid:19)(cid:21) , (7a) 4 = D2ω0j c mph j,z(cid:20)sin(θ2) cos(θ1)(cid:18) d2 + n2 1 + n2 2) d1(n2 2n2 1n2 (cid:19) + sin(θ1) cos(θ2)(cid:18) d1 n1 + d2(n2 1 + n2 2) 2n1n2 2 (cid:19) − sin θ1 sin θ2 c(n2 2n3 2ω0j (cid:21)−1 1 − n2 2)2 1n3 , (7b) mph j,z(cid:12)(cid:12)(cid:12)K,kρ=0 j,ρ(cid:12)(cid:12)(cid:12)K,kρ=0 mph where we take θ1,2 = θ1,2ωj =ω0j . At kρ ≪ 1/D a photonic band gap appears at the K direction. Let us find a half-width of the band gap in the structure without embedded QWs following the method described in the Supplemental Material to Ref. 32. We introduce the parameter ζ = n1d1 n2d2 − 1 that characterizes the relative optical path lengths in the structure layers. Generally speaking, for a modified Bragg structure this parameter is close to zero. The center of the photonic band gap is characterized by a Bragg frequency that ac- cording to36 is given by ωB = 2πc/ (n1d1 + n2d2). We introduce the parameter δ = ω01/ωB−1 that is also small in comparison with 1. We expand Eq. (6) up to the sec- ond order in ζ and δ. As a result, we obtain the following expression for δ: δ = ± (n1−n2)ζ 2(n1+n2) . Now it is easy to find a half-width of the band gap as ΩB = ωBδ. In accordance with the foregoing, the eigenfrequency of the lower photonic branch at K, kρ = 0 is given by ω01 ≃ ωB − ΩB. (8) Θ1 = r(0) QW cD2t(0) QW (i cr(0) QW Γ0t(0) QW (cid:20)cos(θ2) sin(θ1) + sin(θ2)(cid:18) n1 n2 +(cid:20)d1n1 cos(θ1) cos(θ2) − sin(θ1) sin(θ2)(cid:18)d2n2 + d1(n2 1 + n2 2) 2n2 B. Modified Bragg mirror with embedded QWs When adding the QWs to the structure, a new term describing the exciton impact appears in the right-hand side of the Eq. (6): irQW tQW (cid:26)cos (θ2) sin (θ1) 2 (cid:19) − +(cid:20) n1 cos2(cid:18) θ1 n2 n2 n1 sin2(cid:18) θ1 2 (cid:19)(cid:21) sin (θ2)(cid:27) . (9) The expressions for the effective exciton-polariton masses in the vicinity of the saddle point can be obtained by the same technique as described above in the form = mph j,z + iΘ1, = D2ω c j,z (Θ2 − iΘ3)−1 . m∗ (10a) (10b) m∗ m∗ j,z(cid:12)(cid:12)K,kρ=0 j,ρ(cid:12)(cid:12)K,kρ=0 The parameters Θ1,2,3(ω0, Γ0) that characterize the QW exciton impact on the optical properties of the con- sidered RHMM are found in the form cos2(cid:18) θ1 n2 n1 2 (cid:19) − (cid:19) +d2n2(cid:18) n1 2 (cid:19)(cid:19)(cid:21) sin2(cid:18) θ1 cos2(cid:18) θ1 2 (cid:19) − n2 n2 n1 sin2(cid:18) θ1 2 (cid:19)(cid:19) cos(θ2)(cid:21)(cid:27) , (11a) Θ3 = c2r(0) QW t(0) c QW (cid:20)sin(θ1) sin(θ2)(cid:18) d2 n2 + d1(n2 1 + n2 2) 2n2n2 1 (cid:19) − d1 n1 cos(θ1) cos(θ2) + c 1ω0j n2 cos(θ2) sin(θ1) + n1n2ω0j (cid:18) n2 1 n2 2 cos2(cid:18) θ1 2 (cid:19) + sin2(cid:18) θ1 2 (cid:19)(cid:18)1 − 2n2 2 n2 1 (cid:19)(cid:19) sin(θ2) − d2 n1 (cid:18) n2 1 n2 2 cos2(cid:18) θ1 2 (cid:19) − sin2(cid:18) θ1 2 (cid:19)(cid:19) cos(θ2)(cid:21) , (11b) the parameter Θ2 is the rectangular bracket in the right part of Eq. (7b); r(0) QW ≡ rQWK,kρ=0, t(0) QW ≡ tQWK,kρ=0. ρ, z ≡ m∗ In the vicinity of the saddle point of LB, the effective mass tensor components m∗ 1,ρ, z have opposite signs. This is clearly seen in Fig. 2(d) where the depen- dencies of the inverse in-plane (green surface), m∗ ρ, and transverse (red surface), m∗ z, effective masses on the po- sition in the 1st BZ are shown. One can see that m∗ ρ > 0 while m∗ z < 0. It also should be mentioned that for the considered model structure the absolute value of m∗ z is at least one order of magnitude smaller than m∗ ρ. For example, according to Eqs. (10), the ratio m∗ z at the saddle point is about 20.1 for QW-free structure, 21.6 for the structure with embedded QWs with Γ0 = 2 meV and grows to 30.7 for the structure with Γ0 = 10 meV. Such a big difference introduces a strong anisotropy to the optical properties of the considered structure. ρ(cid:14) m∗ III. LIGHT SPEED MANIPULATION IN RHMM First, let us discuss the propagation of a femtosecond laser pulse in the growth direction of the structure (z- axis). We consider the Gaussian pulse in the form E(z, t) = E0 exp(cid:20)− (cid:21) exp[−iωct] exp[−ikzz], (t − t0)2 2t2 w (12) centered on the frequency ωc; E0 determines the pulse amplitude, tw is the half-width duration of the pulse. Here we consider the wave packet whose spatial width ρw exceeds the in-plane structure size, and we assume the intensity of light to be uniformly distributed in the QW plane in each layer. We consider the normal incidence geometry. In the numerical calculations we take tw = 50 fs and t0 = 0.1 ps, ωc = 0.95× ωX ≃ 2.8 eV; kρ = 0. Figures 3 (a) and (b) demonstrate light pulse propaga- tion in the structure calculated for different values of Γ0. In Fig. 3(a) we take Γ0 = 0, which corresponds to the QW-free Bragg mirror. In Fig. 3(b) we consider the case of a Bragg mirror with embedded QWs characterized by a high radiative decay rate, Γ0 = 10 meV. Propagation of light has been modelled in the system starting with a vacuum layer of width 25 × D on the left-hand side of the structure. In the middle part of the system we have placed an RHMM of 200 layers of width D each. The right-most part represents a 25 × D thick vacuum layer again. It is clearly seen that vg, z significantly decreases with the increase of Γ0. It is also confirmed by Figs. 3 (c) and 3(d) demonstrating the dependence of vg, z on Γ0 (c) for the fixed values of ωc and its dependence on ωc (d) for several fixed values of Γ0. Such a tendency can be qualitatively explained as follows. Once the pa- rameter Γ0 increases, the lowest branch moves down in energy, see Figs. 2 (a) and 2(b). Since in the vicinity of K, kρ ≃ 0 the dependence ω(K) for the lowest branch is convex, to conserve the energy the wave packet should reduce its wavevector and group velocity vg, z, see EFCs in Fig. 2(c). It is important to mention that this conclu- sion is only correct in a specific frequency range, namely, 5 FIG. 3. (Color online) Femtosecond laser pulse propagation in the multilayer structure schematically shown in Fig. 1. The parameter Γ0 is taken as (a) 0 meV (that is equivalent to the absence of QWs in the structure) and (b) 10 meV. Graphs (c) and (d) demonstrate the parametric dependencies of the group velocity of light in the z direction, vg, z, (c) on Γ0 for a number of fixed values of the wave-packet central frequency component ωc and (d) on ωc for different values of Γ0 (d) with kρ = 0. Values of vg, z are given in units of the speed of light in vacuum c. The vertical dashed lines correspond to Γ0 in (c) and ωc in (d) from (a) and (b). Horizontal dashed lines indicate the group velocities of the wave packet, with the considered values of Γ0. for ωc < ω0. The regular optical patterns in Figs. 3 (a) and 3(b) describe the interference of the propagating pulse and the pulses reflected from vacuum-RHMM and RHMM- vacuum interfaces. 6 IV. NEGATIVE REFRACTIVE RESPONSE OF RHMM Let us now consider a different geometry of the exper- iment, where a monochromatic spatially focalized light beam enters the structure from its side and propagates in the ρz plane; see the inset in Fig. 4. We consider the transmission of light through the interface between vac- uum and RHMM. The medium on the left is a vacuum characterized by a familiar linear dispersion ω = ckvac. The spatially modulated structure of a Bragg mirror we now consider in the continuous approximation as a ho- mogeneous effective dielectric medium characterized by the dispersion given by Eq. (5) and by the tensorial ef- fective mass of photons. This approximation is valid if the typical spatial size of the light beam (beam width zw) is much larger than period of the structure, zw ≫ D, and at sufficiently low incidence angles. Now we consider the propagation of a monochromatic Gaussian light beam of frequency ωc E(z, ρ, t) = E0 exp(cid:20)− (z − z0)2 2z2 w (cid:21) e−i(K0z+kρρ)e−iωct, (13) where zw is the beam spatial width. The beam propa- gates at an oblique angle to the structure, that is set by K0. In our simulations we take z = 0, zw = 12 × D, and ωc = 0.95ωX. The wave vector component K0 is taken as K0D = 0.6. To model the propagation of light in the ρz plane of the structure, we use the transfer matrix technique adapted for the new geometry. We check the accuracy of this numerical procedure in limiting cases by analytical calculations realised in the effective photonic mass approximation using the expressions (10) for the effective mass tensor components in the vicinity of the saddle point. Figure 4 demonstrates the light beam propagation in the ρz plane of the structure without [Fig. 4(a)] and with [Fig. 4(b)] embedded QWs in the regime of a high radia- tive decay rate Γ0 = 15 meV. We considered the model RHMM of 30×D width limited by vacuum on both sides. One can see from Figs. 4 (a) and 4(b) that the light beam undergoes the negative refraction in the consid- ered geometry. Moreover, since the absolute value of the effective mass in the z-direction is one order of magni- tude larger than the in-plane mass, the negative refrac- tion appears to be very strong. Clearly, the advantage of polaritonic RHMM over dielectric PC structures is in the suitability for the external control of the refraction angle in a range of several degrees that is achieved just by tuning the value of Γ0. This tuning can be done, e. g. by application of the external bias (see the Appendix). Figures 4 (c) and 4(d) demonstrate the dependencies of the refraction angle on Γ0 (c) for the fixed values of ωc and on ωc (d) for the fixed values of Γ0. One can see that the absolute value of the refraction angle decreases with the increase of Γ0. This tendency is maintained as long as the frequency of the beam ωc is less than ω0. If this FIG. 4. (Color online) Focalized light beam propagation in the ρz plane simulated for different values of Γ0. The param- eter Γ0 is taken as 0 eV for (a) and 15 meV for (b). Panels (c) and (d) show the dependencies of the refraction angle on Γ0 for a number of fixed values of ωc and on ωc for different values of Γ0 with K0 = 0.6D−1. The vertical dashed line on (d) corresponds to the value of ωc taken from (a) and (b). Horizontal lines indicate the refraction angles of the beams for the considered values of Γ0. condition is violated, some values of kρ in the vicinity of the first BZ center become forbidden, which affects the shape and angular dispersion of the beam. We also note that the parameter Γ0 significantly affects the spread of the optical beam. Comparing Figs. 4 (a) and 4(b), it can be concluded that the increase of Γ0 up to a certain limit reduces the light beam blurring. This is also correct only in the limit of ωc < ω0. Finally, we want to mention that Figs. 4 (a) and 4(b) also illustrate qualitatively the influence of Γ0 on the transmission prop- erties of RHMMs. The variation of transmittivity affects interference patterns in the vicinity of the surfaces of the considered structure. V. CONCLUSIONS We have considered a planar RHMM based on a mod- ified Bragg mirror with embedded periodically arranged QWs. The optical properties of this RHMM are tun- able by changing the radiative decay rate of embedded quantum wells. The latter can be done by application of external electric and magnetic fields due to their strong influence on the exciton oscillator strength (see the Ap- pendix). This enables one to control the group velocity and propagation direction of light as well as its spatial distribution. This work was supported by the Russian Foundation for Basic Research Grants No. 16-32-60104, No. 15- 59-30406, and No. 15-52-52001, by grant of President of Russian Federation for state support of young Rus- sian scientists No. MK-8031.2016.2, by the Russian Min- istry of Education and Science state tasks No. 2014/13, 16.440.2014/K and by the EPSRC Hybrid Polaritonics Programme grant. Appendix: Impact of the external electric field on the exciton radiative decay rate Here we discuss the mechanism of tuning of Γ0 by the external bias. It is known that the radiative exciton life- time τrad in a quantum well z direction is governed by the overlap of the electron and hole wave functions ψe(z) and ψh(z) and the exciton in-plane Bohr radius aB (Refs. 34 and 35). In general, τrad is larger, the smaller the over- . The following expression lap integral(cid:2)R ψe(z)ψh(z)dz(cid:3)2 links these parameters35,37: Γ0 = 1 2τrad = Γ0(cid:20)Z ψe(z)ψh(z)dz(cid:21)2 , (A.1) B(cid:14) (a2D where Γ0 = √εBωLT k0a3 B )2 is the quantity of the appropriate dimension, εB is a background dielectric con- stant, ωLT is the longitudinal-transverce splitting fre- quency, aB and a2D B are the exciton Bohr radii in the bulk and in QW, respectively. The electron-hole over- lap integral is strongly sensitive to the applied electric field normal to the QW plane due to the quantum con- fined Stark effect, see e. g. Ref. 38. Generally speaking, both a2D B and the overlap integral in (A.1) depend on the applied electric field, however this dependence for the for- mer is negligibly weaker than for the latter and can be ignored. To estimate the overlap integral, one should solve Schrodinger equations for both electron ψe(z) and hole ψh(z) envelope functions associated with the eigenener- gies Ee,h: 2 2m∗ i (cid:20)− ∂2 ∂z2 + Vi(z) + qiF z(cid:21) ψi(z) = Eiψi(z), (A.2) where i = e, h; m∗ is the effective mass; Vi(z) is the i potential for the i-th carrier that in the simplest case 7 FIG. 5. (Color online) Squared electron-hole overlap integral in dependence on the applied external electric field. The latter is given in logarithm scale. The inserts demonstrate schemat- ically the wave functions of an electron (red solid curves) and a hole (green dashed curves) in QW at specified values of F that are 10 kV/cm (for the lower insert) and 500 kV/cm (for the upper insert). we take equal to zero inside QW and to infinity outside. F is the stationary external field applied in the growth direction, and qi determines the charge of the i-th carrier. Next, we use the variational approach to retrieve elec- tron and hole wave functions. We take trial functions in the form39 −αe dQW(cid:19) e ψe(z) = Ae sin(cid:18) πz ψh(z) = Ah sin(cid:18) π(dQW − z) dQW z dQW , −αh dQW−z dQW , (A.3b) (cid:19) e (A.3a) where Ae,h = Ae,h(αe,h) are normalization param- eters determined from the orthonormality condition R ∞ −∞ ψe,h2dz = 1. αe,h are the only variational parame- ters that are found by minimization of the carrier energy Ei = hψiHiψii =Z ∞ −∞ ψ∗ i (z)Hiψi(z)dz, (A.4) e = 0.2m0 and m∗ where Hi represents the Hamiltonian corresponding to Eq. (A.2). For the estimations we take the carriers' ef- fective masses m∗ h = 0.8m0 with m0 being the mass of a free electron, and the QW width dQW = 10 nm. Figure 5 shows the squared electron- hole overlap integral as a function of the applied electric field. It is clearly seen that for the strong fields exceeding 1000 kV/cm the overlap integral is sufficiently small. At the same time, for the fields less than 1 kV/cm it remains unchanged. These estimates define the tunability range of the applied field that allows one to manipulate Γ0. It is necessary to mention that the theoretical curve presented in Fig. 5 does not reflect the true dependence of the estimated parameters on the external field. A num- ber of additional effects that have not been taken into account in the simulations can modify the dependence quite considerably. For example, the external field ap- plied in the QW plane direction leads to spatial separa- tion in electrons and holes and it also leads to change of 8 the overlap integral, see Ref. 40. Another possible effect is associated with the screening of the electric field by the counteracting field generated by free carriers. The partial cancellation of the internal field impact with the increase the free carrier's density have been discussed in41,42. In contrast with thin (on the order of 1 nm width) QWs, in the thicker (10 nm) GaN-based QWs, the carriers are spatially separated due to internal electric fields. In the case of small free carrier's density a large electric field in- deed induces a large spatial separation between electrons and holes, leading to a long recombination lifetime. The change of the radiative lifetime as a result of the inter- play between a built-in electric field inside the quantum well and a small external electric field is discussed in43. On the contrary, when the density of carriers increases, this leads to the enhancement of the induced electric field. The screening effects of the electric field due to carriers become important. They lead to the increase of the overlap integral and the decrease of the recombi- nation lifetime as a result. In this case, the maximum on the dependence in Fig. 5 appears for large enough values of F . To take into account the screening effect, an additional term qiΦi(z)zψi(z) should be included in Eq. (A.2), where Φi(z) depends on the carrier densities ψi2, see Refs. 38 and 39. Another important factor affecting the dependence of Γ0 on the external field is temperature. The temperature impact on the recombination lifetime was considered in, e. g.44,45. The authors concluded from the photolumines- cence intensity measurements that the radiative lifetime linearly increases with temperature, which also modifies the specified dependence. Last but not least, although we did our calculations for GaN-based QWs of 10 nm width, the initial choice of the QW width allows us to pick up the reference value of Γ0 in a wide range as well. To illustrate the remarkable dependence of the radiative lifetime on a QW width we refer to45,46 where this problem is the focus of attention. ∗ Electronic address: evgeny sedov@mail.ru † Electronic address: a.kavokin@soton.ac.uk 1 W. Cai and V. Shalaev, Optical Metamaterials: Fun- damentals and Applications (Springer-Verlag, New York, 2010). 2 A. Poddubny, I. Iorsh, P. Belov, and Y. Kivshar, Nature 18 P. S. J. Russell, Appl. Phys. B 39, 231 (1986). 19 P. S. J. Russell, Phys. Rev. A 33, 3232 (1986). 20 P. S. J. Russell and T. A. Birks, in Photonic Band Gap Materials, Vol. 315 of NATO Advanced Studies Institute, Series E: Applied Sciences, edited by C. M. Soukoulis (Kluwer, Dordrecht, 1996). 21 M. Notomi, Phys. Rev. B 62, 10696 (2000). 22 A. Kavokin, G. Malpuech, and Physics Letters A 339, 387 (2005). I. Shelykh, 23 A. Berrier, M. Mulot, M. Swillo, M. Qiu, L. Thyl´en, A. Tal- neau, and S. Anand, Phys. Rev. Lett. 93, 073902 (2004). 24 A. J. Hoffman, L. Alekseyev, S. S. Howard, K. J. Franz, V. A. Wasserman, D. Podolskiy, E. E. Narimanov, S. D. L., and C. Gmachl, Nature Mater. 6, 946 (2007). 25 P. M. Johnson, A. F. Koenderink, and W. L. Vos, Phys. Rev. B 66, 081102 (2002). 26 B. Wild, R. Ferrini, R. Houdre, M. Mulot, S. Anand, and Photon. 7, 948 (2013). 3 V. P. Drachev, W. Cai, U. Chettiar, H.-K. Yuan, A. K. Sarychev, A. V. Kildishev, G. Klimeck, and V. M. Shalaev, Laser Phys. Lett. 3, 49 (2006). 4 V. M. Shalaev, Nat. Photon. 1, 41 (2007). 5 D. D. Smith R. and Schurig, Phys. Rev. Lett. 90, 077405 (2003). 6 D. R. Smith, D. Schurig, J. J. Mock, P. Kolinko, and P. Rye, Applied Physics Letters 84, 2244 (2004). 7 J. B. Pendry, A. J. Holden, W. J. Stewart, and I. Youngs, Phys. Rev. Lett. 76, 4773 (1996). 8 R. A. Shelby, D. R. Smith, Science 292, 77 (2001). and S. Schultz, S. C. J. M., Appl. Phys. Lett. 84, 846 (2004). 27 A. Tomadin and R. Fazio, 9 Z. Liu, H. Lee, Y. Xiong, C. Sun, and X. Zhang, Science J. Opt. Soc. Am. B 27, A130 (2010). 315, 1686 (2007). 10 T. Tumkur, G. Zhu, P. Black, Y. A. Barnakov, C. E. Bon- ner, and M. A. Noginov, Appl. Phys. Lett. 99, 151115 (2011). 11 E. Schonbrun, M. Tinker, W. Park, and J.-B. Lee, IEEE Photon. Technol. Lett. 17, 1196 (2005). 12 C. L. Cortes, W. Newman, S. Molesky, and Z. Jacob, J. Optics 14, 063001 (2012). 28 E. S. Sedov, A. P. Alodjants, S. M. Arakelian, Y.- and R.-K. Lee, L. Chuang, Y. Y. Lin, W.-X. Yang, Phys. Rev. A 89, 033828 (2014). 29 D. Jaksch and P. Zoller, Annals of Physic 315, 52 (2005), special Issue. 30 K. Hennessy, A. Badolato, M. Winger, D. Gerace, M. Atature, E. L. H. S. Gulde, S. Falt, and A. Imamoglu, Nature 445, 896 (2007). 13 Y. Liu, G. Bartal, and X. Zhang, Opt. Express 16, 15439 31 C.-H. Su, A. D. Greentree, and L. C. L. Hollenberg, (2008). 14 V. M. Shalaev, W. Cai, U. K. Chettiar, H.-K. Yuan, and A. V. Kildishev, A. K. Sarychev, V. P. Drachev, Opt. Lett. 30, 3356 (2005). 15 Z. Jacob, L. V. Alekseyev, and E. Narimanov, Opt. Ex- press 14, 8247 (2006). 16 D. Schurig, J. Mock, B. Justice, S. A. Cummer, J. Pendry, A. Starr, and D. Smith, Science 314, 977 (2006). 17 W. Cai, U. K. Chettiar, A. V. Kildishev, and V. M. Sha- laev, Nat. Photonics 1, 224 (2007). Opt. Express 16, 6240 (2008). 32 E. S. Sedov, I. V. Iorsh, S. M. Arakelian, A. P. Alodjants, and A. Kavokin, Phys. Rev. Lett. 114, 237402 (2015). 33 F. Abeles, Ann. Phys. 3, 504 (1948). 34 A. Kavokin, J. Baumberg, G. Malpuech, and F. Laussy, Microcavities, Series on Semiconductor Science and Tech- nology (OUP Oxford, 2007). 35 L. E. Vorob'ev, E. L. Ivchenko, D. A. Firsov, and V. A. Shalygin, Optical properties of nanostructures (Nauka, St. Petersburg, 2001). 36 A. Y. Sivachenko, M. E. Raikh, and Z. V. Vardeny, T. Saitoh, Appl. Phys. Lett. 83, 4791 (2003). 9 Phys. Rev. A 64, 013809 (2001). 37 V. V. Solov'ev, I. V. Kukushkin, J. H. Smet, K. von Klitz- ing, and W. Dietsche, JETP Letters 84, 222 (2006). 38 P. Bigenwald, A. Kavokin, B. Gil, and P. Lefebvre, Phys. Rev. B 63, 035315 (2001). 39 F. G. Pikus, Fiz. Tekh. Poluprovodn. 26, 45 (1992), [Sov. Phys. Semicond. 26, 26 (1992)]. 40 D. A. B. Miller, D. S. Chemla, T. C. Damen, A. C. Gos- sard, W. Wiegmann, T. H. Wood, and C. A. Burrus, Phys. Rev. B 32, 1043 (1985). 41 H. Gotoh, T. Tawara, Y. Kobayashi, N. Kobayashi, and 42 E. Kioupakis, Q. Yan, and C. G. Van de Walle, Appl. Phys. Lett. 101, 231107 (2012). 43 E. Sari, S. Nizamoglu, I.-H. Lee, J.-H. Baek, and H. V. Demir, Appl. Phys. Lett. 94, 211107 (2009). 44 C. K. Sun, S. Keller, G. Wang, M. S. Minsky, J. E. Bowers, and S. P. DenBaas, Appl. Phys. Lett. 69, 1936 (1996). 45 E. Bekowicz, D. Gershoni, G. Bahir, and A. C. Abare, Phys. Status Solidi B 216, 291 (1999). 46 M. Zamfirescu, B. Gil, N. Grandjean, G. Malpuech, J. Massies, and A. Kavokin, P. Bigenwald, Phys. Rev. B 64, 121304 (2001). 

1208.4135

1

1208

Selective Molecular Sieving through Porous Graphene

Membranes act as selective barriers and play an important role in processes such as cellular compartmentalization and industrial-scale chemical and gas purification. The ideal membrane should be as thin as possible to maximize flux, mechanically robust to prevent fracture, and have well-defined pore sizes to increase selectivity. Graphene is an excellent starting point for developing size selective membranes because of its atomic thickness, high mechanical strength, relative inertness, and impermeability to all standard gases. However, pores that can exclude larger molecules, but allow smaller molecules to pass through have to be introduced into the material. Here we show UV-induced oxidative etching can create pores in micrometre-sized graphene membranes and the resulting membranes used as molecular sieves. A pressurized blister test and mechanical resonance is used to measure the transport of a variety of gases (H2, CO2, Ar, N2, CH4, and SF6) through the pores. The experimentally measured leak rates, separation factors, and Raman spectrum agree well with models based on effusion through a small number of angstrom-sized pores.

cond-mat.mes-hall

cond-mat

1 Selective Molecular Sieving through Porous Graphene Steven P. Koenig, Luda Wang, John Pellegrino, and J. Scott Bunch* Department of Mechanical Engineering, University of Colorado, Boulder, CO 80309 USA *email: jbunch@colorado.edu Membranes act as selective barriers and play an important role in processes such as cellular compartmentalization and industrial-scale chemical and gas purification. The ideal membrane should be as thin as possible to maximize flux, mechanically robust to prevent fracture, and have well-defined pore sizes to increase selectivity. Graphene is an excellent starting point for developing size- selective membranes1–8 because of its atomic thickness9, high mechanical strength10, relative inertness, and impermeability to all standard gases11–14. However, pores that can exclude larger molecules, but allow smaller molecules to pass through have to be introduced into the material. Here we show UV-induced oxidative etching15,16 can create pores in micrometre-sized graphene membranes and the resulting membranes used as molecular sieves. A pressurized blister test and mechanical resonance is used to measure the transport of a variety of gases (H2, CO2, Ar, N2, CH4, and SF6) through the pores. The experimentally measured leak rates, separation factors, and Raman spectrum agree well with models based on effusion through a small number of angstrom-sized pores. 2 Suspended graphene membranes were fabricated by mechanical exfoliation of graphene over predefined 5 µm diameter wells etched into silicon oxide17,18. After exfoliation, the pristine graphene flakes that span the microcavity form suspended membranes that are impermeable to all standard gas molecules11 and clamped to the silicon oxide substrate by surface forces18. Gas species can enter and exit the microcavity through the substrate by slow diffusion. To fill the microcavity with a desired gas species, the sample is put in a chamber pressurized to 200 kPa above ambient pressure with a “charging” gas (Fig. 1a). Prior to this pressurization, the chamber is flushed with the “charging” gas to exclude any other species. The samples are left in the pressure chamber for 4-12 d (depending on the gas species used) to allow for the internal, pint, and external pressure, pext, of the microcavity to equilibrate to the “charging” pressure, p0. Upon removing the sample from the pressure chamber the higher pressure inside the microcavity compared with ambient atmospheric pressure causes the membrane to bulge upward (Fig. 1b). This technique allows preparation of a graphene-sealed microcavity with an arbitrary gas composition at a prescribed pressure. To measure the leak rate of gas species we used both a pressurized blister test and mechanical resonance test11. The pressurized blister test was used for leak rates on the order of minutes to hours while the mechanical resonance was used to measure leak rates on the order of seconds to minutes. For the pressurized blister test, an atomic force microscope (AFM) is used to measure the shape of the bulged graphene membrane, which is parameterized by its maximum deflection, δ (Fig. 1e). The maximum deflection, δ, vs. time, t, for a pristine graphene membrane pressurized to 200 kPa, above atmospheric pressure, of H2 gas is shown in Fig. 1f (black). The deflection decreases 3 slowly with time consistent with a leak of H2 gas through the underlying silicon oxide11,18. UV-induced oxidative etching was used to introduce pores in the pristine graphene membranes15,16,19,20 (see supporting online text). The H2 gas pressurized graphene membranes were exposed to UV light (λ1 = 185 nm, λ2 = 254 nm; Jelight Model 42 UV ozone cleaner) at ambient conditions for several minutes. A number of other etching techniques have been proposed and demonstrated on graphene19,21–27 but the UV oxidative etching used here is simple and slow enough to allow for the creation of these subnanometer-sized selective pores as demonstrated later in this paper. Other etching techniques, including oxygen plasma etching, were tried but UV oxidative etching proved to be the only successful method for controllably introducing subnanometer pores. After the oxidative etch, δ is again measured versus t (Fig. 1e and 1f, red) (see supplementary info). The maximum deflection decreases rapidly (several minutes as opposed to hours for the unetched case) and eventually leads to a downward deflection of the membrane (Fig. 1c-1f). Figure 1e shows a series of cross sections through the centre of the membrane taken by AFM as time elapses from 0 to 8 min and Fig 1g shows a three dimensional rendering of the AFM image for t=0 in Fig. 1e. Here 0 min is defined to be the time at which the first AFM image was captured after removing the sample from the pressure chamber. The change in deflection, as depicted in Fig. 1c & d, results from increasing the H2 leak rate, through etching, while preventing significant changes in the N2 leak rate into the microcavity from the ambient atmosphere. The molecular selectivity of the fabricated porous graphene membrane is demonstrated by measuring the time rate of change of , -d/dt, for the same membrane 4 pressurized with a number of different gases. Figure 2a shows δ vs. t for H2, CO2, Ar, and CH4 before and after etching and N2 after etching. We did not measure the N2 leak rate for this particular device before etching, but measurements for 12 other ones located on the same flake are shown in Fig. 4 and labelled “Pristine Avg” for comparison with the after-etch leak rate. At short times, -d/dt is approximately linear (Fig. 2a). This rate, -d/dt, versus kinetic diameter28 is plotted for all the gases measured for the same membrane/microcavity in Fig. 1 before and after etching (Fig. 2b). After etching, there is an increase in -d/dt of two orders of magnitude for the leak rate of H2 and CO2, while Ar and CH4 remain relatively unchanged. This suggests that the etched pores change the transport mechanism for H2 and CO2, while leaving the transport of Ar and CH4 nearly unchanged. Since the kinetic diameter cut off in this bi-layer graphene membrane is nominally that of Ar, 3.4 Å28, this membrane will heretofore be referred to as “Bi- 3.4 Å”. The leak rate of various gases across the porous graphene membranes can also be measured by using a mechanical resonance test. This is accomplished by measuring changes in the mechanical resonant frequency, f, of the membrane vs. t using an optical drive and detection system previously used to measure mechanical resonance in suspended graphene resonators11,29. A pressure difference applied across the membrane leads to a pressure-induced tensioning of the membrane, which increases f of the stretched membrane. If the gas molecules introduced external to an initially evacuated microcavity can leak through the membrane, the gas will pass through and reduce the tension in the membrane, thus decreasing f. If the gas molecules cannot leak through the membrane, f stays constant. An example of this is shown in Fig. 3 where an etched 5 porous graphene membrane was put in a vacuum of 0.1 torr for a several days to ensure the microcavity has equilibrated to the pressure of the vacuum chamber. Next, a pure gas species is introduced into the vacuum chamber at a given pressure (~100 torr for the case in Fig. 3 and ~80 torr for the inlay of Fig. 3) and the resonant frequency is measured. The resonant frequency decreases with time, and from the rate of decrease, we determine the leak rate through porous graphene membrane. We could not observe the frequency return back to its original value due to significant gas damping when Δp ~ 0 (see supporting online text). As can be seen from Fig. 3, the leak rate of H2, CO2, N2, and CH4 is several seconds while SF6 shows no significant change in resonant frequency for the several minutes measured. This membrane will be referred to as “Bi- 4.9 Å” since it is a bilayer membrane with a nominal sieving kinetic diameter of SF6, 4.9 Å28. We derived the following expression for the molecular flux out of the pressurized "blister" microcavity, dn/dt, using the ideal gas law and Hencky’s solution for a clamped circular membrane30 (see supplementary info for derivation): where a is the radius of the membrane, E is the Young’s modulus, w is the thickness of the membrane, R is the molar gas constant, T is temperature, V(δ) is the total volume of the microcavity in the bulged state, and C(ν) and K(ν) are geometric coefficients which depend on the Poisson’s ratio, ν, of the membrane. For the case of graphene, the Young’s modulus and Poisson’s ratio are E = 1 TPa and ν = 0.16, respectively, and the thickness per layer is 0.34 nm10,11,18,31. Using ν = 0.16 gives coefficients of K(ν=0.16) = 3.09 and C(ν=0.16) = 0.52417. Figure 4 shows the normalized dn/dt (normalized to the partial pressure difference across the membrane) for the “Bi- 3.4 Å” membrane before UV 6 etching (black squares) and after UV etching (red squares). Also included is the average normalized dn/dt for 24 different unetched (12 for the case of N2) membranes on the same graphene flake shown in the Fig 1f inlay that contains “Bi- 3.4 Å” (black circles). Similarly, dn/dt, can be calculated from the linear approximation of the rate of frequency decay, df/dt (see supplementary info for details). The leak rate versus molecular size for the “Bi- 4.9 Å” membrane is shown in Fig. 4a (red diamonds). The changes in leak rates associated with UV etching are consistent with the introduction of a pore(s) which allow size selective permeation of gas molecules. For the “Bi- 3.4 Å” membrane in Fig. 2, the selectivity between CO2 and Ar suggests that the pore(s) size(s) introduced into the graphene membrane are comparable to the kinetic diameter of Ar (3.4 Å)28 and that the porous graphene is sieving molecules above and below this size. Similarly for the “Bi- 4.9 Å” membrane in Fig. 3, there are likely pore(s) larger in size than that of the “Bi- 3.4 Å” membrane, since effective molecular sieving is seen for molecules smaller than SF6 (4.9 Å compared to 3.8 Å for CH4)28. Due to the fact that there is likely only a small density of pores in the 5 μm diameter membranes, imaging of the pore is not possible (see supporting online text). However, the small density of pores is supported by Raman spectroscopy on the etched membranes (see supporting online text). The gas leak rates measured can be compared to results of computational modelling by Jiang et al. and Blankenburg et al.1,5. Following the work of Jiang et al., we estimate a H2 leak rate on the order of ~10-20 mol s-1 Pa-1 for a H-passivated pore in graphene consisting of 2 missing benzene rings at room temperature (see supporting online text)1. For the work of Blankenburg et al., the H2 leak rate was calculated to be on the order of ~10-23 mol s-1 Pa-1 through a smaller H-terminated pore consisting of a single 7 missing benzene ring5. Our measured H2 leak rate on “Bi- 3.4 Å” was ~4.5 x 10-23 mol s-1 Pa-1. This value is several orders of magnitude lower than Jiang et al., suggesting our pores have an overall higher energy barrier for H2 (and other species) than in their calculations. The similarity between our H2 leak rate with that modelled by Blankenburg et al. suggests a similar H2 energy barrier in our pore. Nonetheless, we do not match their calculated H2/CO2 selectivity (2 versus ~1017). This suggests that having a bilayer graphene membrane with different chemical pore termination from the oxidative etching can be quite important. We can also compare the H2 and CO2 measured leak rates between the “Bi- 3.4 Å” and “Bi- 4.9 Å” membranes (Fig. 4). The one with the smaller pore size, “Bi- 3.4 Å”, (red squares) had a H2 and CO2 leak rates (in units of 10-23 mol s-1 Pa-1) of 4.5 and 2.7, respectively, compared to H2 and CO2 leak rates (same units) of 75 and 25, respectively, for the larger pore membrane (red diamonds). The closeness between the magnitude of these 2 values, and the magnitudes calculated in the cited modelling, suggests that in both cases a low density of size-selective pores are participating in the transport across the graphene membrane and the faster leak rate for the “Bi- 4.9 Å” membrane is consistent with larger pores (and/or lower diffusional energy barriers) than the “Bi- 3.4 Å”. This is also consistent with the rapid effusion of gas expected from the ~µm3 confined volume of gas in the porous graphene sealed microchamber11. Both graphene membranes presented here were bilayer graphene membranes due to the more controlled etching and stability of the pores fabricated on bilayer versus monolayer graphene membranes. This is consistent with previous results showing slower etching for bilayer graphene compared with single layer graphene19. However, similar results were observed on monolayer graphene membranes (see supporting online text). In conclusion, we have demonstrated selective molecular sieving using porous , μm-sized, atomically-thin graphene membranes. Pores were introduced in graphene by 8 UV-induced oxidative etching and the molecular transport through them was measured using both a pressurized blister test and mechanical resonance. Our results are consistent with theoretical models in the literature based on effusion through angstrom-sized pores1,5. The results presented here are an experimental realization of graphene gas separation membranes by molecular sieving and represent an important step towards the realization of macroscopic, size-selective porous graphene membranes. The approach used here can also be used to probe the fundamental limits of gas transport by effusion through angstrom-sized pores with atomic-sized channel lengths. Methods Suspended graphene membranes are fabricated by a combination of standard photolithography and mechanical exfoliation of graphene. First, an array of circles with diameters of 5 µm and 7 µm are defined by photolithography on an oxidized silicon wafer with a silicon oxide thickness of 285 nm. Reactive ion etching is then used to etch the circles into cylindrical cavities with a depth of 250-500 nm leaving a series of wells on the wafer. Mechanical exfoliation of Kish graphite using Scotch® tape is then used to deposit suspended graphene sheets over the wells. The volume of the bulged graphene is on the order of the initial volume of the microcavity17, and we deduce the initial ∆p = pint – pext across the membrane, using the ideal gas law and isothermal expansion of the trapped gas with a constant number of molecules, N. Doing so leads to poVo = pint(Vo+Vb), where is the initial volume of the well and is the volume of the pressurized blister after the device is brought to atmospheric pressure and bulges upward. The constant, = 0.524 is determined from Hencky’s solution. AFM scans are then continuously taken in order to deduce the leak rate of molecules out of the membrane, dn/dt. For the resonance measurements, samples are placed in a vacuum chamber at 0.1 torr for several days to ensure the microcavity reaches equilibrium with the vacuum 9 chamber. A given pressure (ranging from 80-100 torr) of gas is then introduced into the vacuum chamber and the frequency is measured over time. After the introduction of a gas, the chamber is evacuated until the frequency returns to its original value when no pressure difference was present (or the signal was no longer detectable due to gas damping) and the next gas is then measured. An intensity modulated blue laser (405nm) was used to drive the graphene membranes and a red laser (633nm) was used to detect the motion of the graphene. References 1. 2. 3. 4. 5. 6. 7. 8. Jiang, D., Cooper, V.R. & Dai, S. Porous graphene as the ultimate membrane for gas separation. Nano Lett. 9, 4019–4024 (2009). Du, H. et al. Separation of Hydrogen and Nitrogen Gases with Porous Graphene Membrane. J. Phys. Chem. C 115, 23261–23266 (2011). Schrier, J. Helium Separation Using Porous Graphene Membranes. J. Phys. Chem. Lett. 1, 2284–2287 (2010). Hauser, A.W. & Schwerdtfeger, P. Nanoporous Graphene Membranes for Efficient 3 He/ 4 He Separation. J. Phys. Chem. Lett. 3, 209–213 (2012). Blankenburg, S. et al. Porous graphene as an atmospheric nanofilter. Small 6, 2266–2271 (2010). Suk, M.E. & Aluru, N.R. Water Transport through Ultrathin Graphene. J. Phys. Chem. Lett. 1, 1590–1594 (2010). Schrier, J. & McClain, J. Thermally-driven isotope separation across nanoporous graphene. Chem. Phys. Lett. 521, 118–124 (2012). Li, Y., Zhou, Z., Shen, P. & Chen, Z. Two-dimensional polyphenylene: experimentally available porous graphene as a hydrogen purification membrane. Chem. Commun. 46, 3672–3674 (2010). 9. Meyer, J.C. et al. The structure of suspended graphene sheets. Nature 446, 60–63 (2007). 10 10. Lee, C., Wei, X., Kysar, J.W. & Hone, J. Measurement of the Elastic Properties and Intrinsic Strength of Monolayer Graphene. Science 321, 385–388 (2008). 11. Bunch, J.S. et al. Impermeable atomic membranes from graphene sheets. Nano Lett. 8, 2458–2462 (2008). 12. Nair, R.R., Wu, H.A., Jayaram, P.N., Grigorieva, I.V. & Geim, A.K. Unimpeded Permeation of Water Through Helium-Leak-Tight Graphene-Based Membranes. Science 335, 442–444 (2012). 13. Leenaerts, O., Partoens, B. & Peeters, F.M. Graphene: A perfect nanoballoon. Appl. Phy. Lett. 93, 193107 (2008). 14. Chen, S. et al. Oxidation Resistance of Graphene-Coated Cu and Cu/Ni Alloy. ACS Nano 5, 1321–1327 (2011). 15. Ozeki, S., Ito, T., Uozumi, K. & Nishio, I. Scanning Tunneling Microscopy of UV-Induced Gasification Reaction on Highly Oriented Pyrolytic Graphite. Jpn. J. Appl. Phys. 35, 3772–3774 (1996). 16. Huh, S. et al. UV/Ozone-Oxidized Large-Scale Graphene Platform with Large Chemical Enhancement in Surface-Enhanced Raman Scattering. ACS Nano 5, 9799–9806 (2011). 17. Novoselov, K.S. et al. Two-dimensional atomic crystals. Proc. Natl Acad. Sci. USA 102, 10451–10453 (2005). 18. Koenig, S.P., Boddeti, N.G., Dunn, M.L. & Bunch, J.S. Ultrastrong adhesion of graphene membranes. Nature Nanotech. 6, 543–546 (2011). 19. Liu, L. et al. Graphene oxidation: thickness-dependent etching and strong chemical doping. Nano Lett. 8, 1965–1970 (2008). 20. Chang, H. & Bard, A.J. Scanning tunneling microscopy studies of carbon-oxygen reactions on highly oriented pyrolytic graphite. J. Am. Chem. Soc. 113, 5588–5596 (1991). 21. Bieri, M. et al. Porous graphenes: two-dimensional polymer synthesis with atomic precision. Chem. Commun. 6919–6921 (2009). 22. Girit, C.O. et al. Graphene at the edge: stability and dynamics. Science 323, 1705– 1708 (2009). 23. Schrier, J. Fluorinated and Nanoporous Graphene Materials As Sorbents for Gas Separations. ACS Appl. Mater. Interfaces 3, 4451–4458 (2011). 11 24. Bai, J., Zhong, X., Jiang, S., Huang, Y. & Duan, X. Graphene nanomesh. Nature Nanotech. 5, 190–194 (2010). 25. 26. 27. Sint, K., Wang, B. & Král, P. Selective ion passage through functionalized graphene nanopores. J. Am. Chem. Soc. 130, 16448–16449 (2008). Fan, Z. et al. Easy synthesis of porous graphene nanosheets and their use in supercapacitors. Carbon 50, 1699–1703 (2012). Fox, D. et al. Nitrogen assisted etching of graphene layers in a scanning electron microscope. Appl. Phys. Lett. 98, 243117 (2011). 28. Breck, D.W. Zeolites Molecular Sieves: Structure, Chemistry, and Use. 593–724 (Wiley: New York, NY, 1973). 29. Bunch, J.S. et al. Electromechanical Resonators from Graphene Sheets. Science 315, 490–493 (2007). 30. Hencky, H. Uber den spannungzustand in kreisrunden platten mit verschwindender biegungssteiflgeit. Z. fur Mathematik und Physik 63, 311–317 (1915). 31. Blakslee, O.L., Proctor, D.G., Seldin, E.J., Spence, G.B. & Weng, T. Elastic Constants of Compression-Annealed Pyrolytic Graphite. J.Appl. Phys. 41, 3373– 3382 (1970). Acknowledgements We thank Darren McSweeney and Michael Tanksalvala for help with the resonance measurements and Rishi Raj for use of the Raman microscope. This work was supported by NSF Grants #0900832(CMMI: Graphene Nanomechanics: The Role of van der Waals Forces), #1054406 (CMMI: CAREER: Atomic Scale Defect Engineering in Graphene Membranes), the DARPA Center on Nanoscale Science and Technology for Integrated Micro/Nano-Electromechanical Transducers (iMINT), the National Science Foundation (NSF) Industry/University Cooperative Research Center for Membrane Science, Engineering and Technology (MAST), and by the NNIN and the National Science Foundation under Grant No. ECS - 0335765. Author Contributions S.P.K. and L.W. performed the experiments. S.P.K. and J.S.B. conceived and designed the experiments. All authors interpreted the results and co-wrote the manuscript. 12 Additional Information The authors declare no competing financial interest. Supplementary information accompanies this paper at www.nature.com/naturenanotechnology. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/. Correspondence and requests for materials should be addressed to J.S.B. Figure Captions Figure 1: Measuring Leak Rates in Porous Graphene Membranes (a) Schematic of a microscopic graphene membrane on a silicon oxide substrate. We start with pristine graphene fabricated by exfoliation and fill the microchamber with 200 kPa of H2 (represented as red circles here) in a pressure chamber. Equilibrium is reached (pint = pext) by diffusion through the silicon oxide. (b) After removing the graphene membrane from the pressure chamber the membrane is bulged upward. We calculate pint using the ideal gas law and assuming isothermal expansion. The hydrogen molecules slowly leak out of the microchamber through the silicon oxide substrate. (c) Upon etching of the graphene membrane pore(s) bigger than that of H2 are introduced allowing the H2 to leak rapidly out of the microchamber through the graphene membrane. If the pore(s) are smaller than that of air molecules (mostly N2 and O2, denoted as green circles), air will be blocked from entering the microchamber causing the deflection of the graphene membrane to continue to 13 decrease until all of the H2 molecules exited the microchamber. (d) After all the H2 molecules have leaked out of the microchamber the membrane will be bulged downward. (e) Deflection versus position, 0 min (black) through 8 min (dashed blue) after etching, corresponding to some of the red points in (f). (f) Maximum deflection, δ, vs. t for one membrane that separates H2 from air as measured by AFM. The black points represent the leak rate of H2 before etching and the red points show the leak rate of H2 after introducing selective pores in the graphene. Inlay: Optical image of the bilayer graphene flake used in this study covering many cavities in the silicon oxide substrate. (g) Three dimensional rendering of an AFM image corresponding to the line cut at t = 0 in (e). Figure 2: Comparing Leak Rates between Pristine and Porous Graphene Membranes (a) Maximum deflection, δ, versus t before (black) and after etching (red). (b) Average -dδ/dt versus molecular size found from the slopes of membrane deflection versus t in (a) for before (black) and after (red) introducing pores in the same graphene membrane. The connecting lines show the measurements before (black) and after (red) etching. Figure 3: Measuring Leak Rates in a Porous Graphene Membrane Using Mechanical Resonance 14 Frequency, f, versus t for H2 (black), CO2 (red), N2 (green), CH4 (blue), and SF6 (cyan). With a pressure of 100 torr (~13.3 kPa) introduced into the vacuum chamber. Inlay is data from the same device with an 80 torr (~10.7 kPa) pressure introduced. Figure 4: Compilation of Measured Leak Rates Leak rate out of the microcavity for: “Bi- 3.4 Å” membrane before etching (black squares) and after etching (red squares), “Bi- 4.9 Å” membrane after etching (red diamonds), and the average before etching of 24 membranes (12 for N2) on the same graphene flake as “Bi- 3.4 Å” membrane (black circles with dot). (Note: the latter are hidden by black squares for several gases.) a pext= p0; Δp=pint - pext=0 b V0, N, pint=p0 V0+Vb, N, pint<p0 c e g d f 135 nm 0 nm Figure 1 012345-100-50050100150 0 min 5 min 1 min 6 min 2 min 7 min 3 min 8 min Z (nm)Y (m)0255075100125-100-50050100150 Maximum deflection,  (nm)Time, t (min)a b Figure 2 0.280.300.320.340.360.380.401E-41E-30.010.1110100 -d/dt (nm/min)Molecular size (nm) Before Etch After EtchH2CO2ArN2CH40255075100125150175-100-50050100150Before Etch After Etch H2 H2 CO2 CO2 Ar Ar N2 CH4 CH4 Maximum deflection,  (nm)Time, t (min)Figure 3 02004006008001000303540455055 Frequency, f (MHz)Time, t (s) H2 N2 SF6 CO2 CH405101520253028303234363840424446 Frequency, f (MHz)Time, t (s)Figure 4 0.280.300.320.340.360.380.401E-271E-261E-251E-241E-231E-221E-21 H2Normalized dn/dt (mol s-1 Pa-1)Molecular Size (nm) Pristine Avg Bi- 3.4Å Before Etching Bi- 3.4Å After Etching Bi- 4.9Å After EtchingCO2ArN2CH41 Selective Molecular Sieving Through Porous Graphene Steven P. Koenig, Luda Wang, John Pellegrino, and J. Scott Bunch* *email: jbunch@colorado.edu Supplementary Information: Raman Spectrum of Graphene Flakes Raman spectroscopy was used to support our conclusion that a small number of pores exist in the graphene flakes. The D-peak (1360 cm-1 wavenumber) is associated with defects in the graphene lattice. Figure S1 shows the Raman spectrum of the graphene flakes used in this study. Figure S1a shows the Raman spectrum of membrane “Bi- 3.4 Å” presented in the main text. This spectrum was taken “before-etching” but is identical to the “after-etching” Raman spectrum. Both spectrums show no D-peak. Figure S1b shows the ratio of the graphene G-peak to the silicon peak areas for the flake presented in this study (the red closed square) and a nearby flake which contained mono- and bi-layer portions (the black open circle for mono-layer and red open square for bi-layer). This follows the work by Koh et al to confirm we had bilayer graphene1. Figure S1c shows the Raman spectrum for membrane “Bi- 4.9 Å” presented in the main text “before- etching” showing the characteristic 2D peak shape of bilayer graphene. Figure S1d shows the Raman spectrum of the two monolayer membranes presented in the supplementary information, “Mono-3.4 Å”, from Figure S5, (upper red curve) and “Mono-5 Å”, presented in Figure S6 (lower black curve). Both spectrums presented in Figure S1d were taken “after-etching”. “Mono- 3.4 Å” showed similar H2 leaking behavior as membrane “Bi- 3.4 Å” presented in the main text. No spatial variation was seen in the Raman spectrum of these flakes after-etching. There has been no D-peak observed in etched monolayer samples that showed gas selectivity and we have observed a D-peak in bilayer samples showing selectivity but future work will be needed to correlate the D-peak with pore density since the top layer of bilayer graphene will likely etch before the bottom 2 layer does to open up pores. Etching Pores in Graphene Membranes In order to etch the graphene membranes, we first pressurized them with pure H2 up to 200 kPa (gauge pressure) above ambient pressure. After the microcavity reached equilibrium we removed it from the pressure chamber and measured the deflection using atomic force microscopy (AFM). We then did a series of short UV etches (30 s) followed by AFM scans between each etching step to see if the leak rate increased significantly. When pore(s) were created that were selective to allow the H2 to pass through, but not allow the molecules in the air to pass, the deflection would rapidly decrease and become negative, consistent with a vacuum inside the microcavity. For the case of the “Bi- 3.4 Å” membrane in the main text, this etching took 75 min (150, 30 s etching steps). Each etch step took about 5 min to complete. Once the sample was out of the pressure chamber for over an hour during the etching process, and the deflection had decreased 20 nm, we then returned the sample to the pressure chamber overnight to allow the pressure inside the microchamber to once again reach 200 kPa. The etching process was then continued the next day. For membrane “Bi- 4.9 Å” in the main text, the total etching time was 15 min using 1 min etching steps. From the etching experiments it was noted that longer etch steps required significantly less total etching time. 3 Since we conclude that there are only a small number of sub-nanometer pores in the 5 μm membranes, direct imaging of these pores is not possible. For classical effusion of gas out of the microcavity, the number of molecules in the microcavity is given by: √ where n0 is the initial number of molecules, A is the area of the hole, V is the volume of the container, kb is Boltzman’s constant, T is temperature, t is time, and m is the molecular mass of the gas undergoing effusion2,3. For a 3 Å diameter circular pore, and 100 kPa H2 pressure, the leak rate is ~10-20 mol·s-1·Pa-1 which should be fast enough to experimentally measure by our technique and on the order of the leak rates presented here. In order to visualize pores created by the UV induced oxidative etching reported in the main text, one membrane was over-etched to create much larger pores so we could image the pore formation and distribution with AFM. Figure S2 shows a monolayer membrane that was over-etched (22 min total with 1 min etching steps) in order to visualize the pore growth. Fig S2a shows the 500 nm x 500 nm AFM scan over the suspended region of the over-etched graphene membrane. This membrane was not selective to any of the gas species tested and the leak rates were too fast to measure. The results of the pore size distribution seen in Fig. S2b and Fig. S2c are comparable to previous oxidative etching of graphene and graphite (see references 14,15, and 19 from them main text). Calculating the Pressure Normalized Leak Rate from Deflection versus t Data The deformation of the membrane can be described using Hencky’s (1915) 4 solution for a pressurized clamped circular elastic membrane with a pressure difference of Δp across it: where E is the Young’s modulus, ν is the Poisson’s ratio, w is the membrane thickness, and K(ν) is a coefficient that depends only on ν 4. For the case of graphene, we take E=1TPa and ν=0.16, therefore K(ν=0.16)=3.095. In order to derive dn/dt, the leak rate of the microcavity, we start with the ideal gas law: where P is the absolute pressure inside the microcavity, V(δ) is the volume of the microcavity when the membrane is bulged with deflection δ, V(δ)=Vo+Vb(δ), Vb(δ)=C(ν)πa2δ, for graphene C(ν = 0.16 ) = 0.52, n is the number of moles of gas molecules contained in the microcavity, R is the gas constant, and T is temperature5. Substituting (Δp+patm) for P and dividing both sides by V(δ), and inserting Henckey’s solution for Δp we get: Now we can take the time derivative of both sides and solve for dn/dt to get the flux of gas molecules out of the membrane: [ ] To get the dn/dt (mol/s), we use the measured dδ/dt, the rate of the bulge decay from the linear fit of the membrane deflection versus time data. We then normalize the leak rate by dividing the calculated dn/dt by the pressure driving force for each of the gases measured to get the leak rate out of the microcavity. 5 Calculating the Pressure Normalized Leak Rate from Frequency versus t Data A schematic of the resonance measurement is presented in Fig S3. Figure S3a shows a membrane that is exposed to a gas smaller than the pore(s) in the graphene thus able to pass through after the membrane has been initially placed in vacuum. Over time, the molecules will leak into the microcavity causing the deflection, and thus the tension, to decrease which leads to a decreasing resonant frequency. Figure S3b shows the membrane in a gas species that is larger than the pore(s) in the graphene. Since the gas is larger than the pore(s) it is blocked and the resonant frequency does not change over time. For the case of the gas being able to pass through the graphene membrane, once the pressure begins to equilibrate on both sides, the signal is lost due to significant gas damping, and it is not possible to accurately experimentally resolve the resonant frequency. This can be seen in the CH4 data presented Fig S4. This data is resonant frequency curves from the CH4 leak rate found in Fig 3 inlay of the main text with 80 torr initially introduced across the membrane. The black curve is the original frequency, t = 0 s. The red curve is the frequency right after the pressure is introduced to the membrane, t = 1 s. From the red curve you can see there is already a significant gas damping which is evident because of the lower quality factor (i.e. broader peak). The green, blue, cyan, and magenta curves correspond to t = 3 s, t = 5 s, t = 7 s, and t = 11 s, respectively. At t = 13 s (orange curve) the damping is too large to discern a peak and we cannot determine what the resonant frequency is at or after this time. The frequency of a circular membrane under tension caused by a pressure difference Δp can be described using the following 3 equations: 6 √ where f is the resonant frequency of the membrane, a is the radius of the membrane, S is the tension in the membrane due to the applied pressure and S0 is the initial tension in the membrane, and ρA is the mass density6. K(ν) and E are elastic constants from Hencky’s solution and w is the thickness of the membrane and δ is the deflection in the membrane4. We do not take S0 to be zero in this case since the pressure difference and thus the deflection of the membrane are small compared with the case of the blister test. In order to derive the dn/dt, the leak rate of the microcavity we first need to solve for S by combining (S7) and (S8) to get: Since S is larger than S0 we can neglect the cubic order term of S0. Now we can insert the expression for S into equation (S6) and solve for Δp, and then insert this expression for Δp into the ideal gas law in a similar fashion as the bulge test equation. Since the deflection of the membrane is small in this case we take V to be constant. After doing this and taking time derivative and solving for dn/dt we arrive at the expression: [( ) ( ] ) where c1, c2, and c3 are constants equal to 8.74x103, 2.39x104, and 8.16x104, respectively. 7 To get dn/dt (mol/s) we can use df/dt, the rate of the frequency decay from the linear fit of the membrane frequency versus time data. We then normalize the leak rate by dividing the calculated dn/dt by the pressure driving force for each of the gases measured to get the leak rate (normalized dn/dt) into the graphene-sealed microcavity. Additional Membranes Measured Three additional membranes where measured, two monolayer and two bilayer samples. The monolayer sample in Fig. S5 (“Mono- 3.4 Å”) shows similar behaviour as seen in “Bi- 3.4 Å” of the main text. This monolayer sample was filled with 150 kPa above ambient pressure with pure H2. The pore was not stable and additional measurements could not be taken. The second monolayer sample shown in Fig. S6 was measured using the mechanical resonance scheme presented in the main text. This membrane showed a similar pore instability as the previous sample. The order of the leak rate measurements taken on this membrane were N2 (black), H2 (red), CO2 (green), and CH4 (blue). Next, N2 was measured a second time (cyan) showing a drastic increase in the N2 leak rate. After the repeat of the N2 data, we then introduced SF6, and the results show that the membrane is slowly allowing SF6 to permeate indicating that this pore is larger but similar in size to SF6 (4.9Ǻ)7. We attribute this increase in N2 leak rate to etching of the pore during the resonance measurement. Two additional bilayer membranes from the same graphene flake found in Fig. 1 (containing membrane “Bi- 3.4 Å”) of the main text are shown in Fig. S7. Figure S7a is a membrane that has larger pores than that of the sample presented in the main text. The membrane in Fig S7a was damaged before CH4 leak rate data could be taken. Fig S7b is the sample presented in the main text, and Fig S7c shows the leak rate of a membrane that showed molecular sieving of H2 versus CO2 and larger molecules (Ar, N2, and CH4). This suggests that the pore size for the membrane in Fig S7c is between 2.89 Ǻ and 3.3 8 Ǻ7. Comparison to Modeling Results and Effusion Jiang et al. simulated transport for two types of pores, a N-terminated one with a ~3 Å size and an H-terminated one with a ~2.5 Å size8. Their nominal H2 permeance of 1 mol m-2 s-1 Pa-1 was based on the N-terminated pore at 600 K with a pass through frequency of 1011 s-1 where a 1 bar pressure drop was estimated from their simulation . When discussing the H-terminated (2.5 Å) pore at room temperature Jiang et al states that for H2 the “passing-through frequency” is 109 s-1. This “passing-through frequency” is lower than the N-terminated by approximately two orders of magnitude for room temperature operation. Thus, we start with 1 mol m-2 s-1 Pa-1 at 600 K and lower it to 10-2 mol m-2 s-1 Pa-1 to accommodate the fact that our measurements were at room temperature. Then we multiply by the area that Jiang et al. used, which was 187 Å2 (1.87 x 10-18 m2), to arrive at ~10-20 mol s-1 Pa-1. To compare to selectivities predicted by the classical effusion model we plotted the leak rate for H2 and CO2 for membrane “Bi- 3.4 Å” and H2, CO2, N2, and CH4 for membrane “Bi- 4.9 Å” and included this in Figure S8 which is a plot of the normalized leak rate versus the inverse square root of the molecular mass of each gas species. Classical effusion predicts that the flow rate through a pore would scale with the inverse square root of the molecular mass and therefore would monotonically increase with 9 increasing inverse square root of the molecular mass. We can also compare the selectivity of H2 to CO2 for both membranes. For classical effusion the selectivity is the ratio of the square root of the molecular masses which is 4.7 for the case of H 2 to CO2. For membranes “Bi- 3.4 Å” and “Bi- 4.9 Å” presented in Fig S8 the H2 to CO2 selectivities are 1.7 and 3 respectively. Tables S1, S2, and S3 show the ideal selectivity for “Bi- 3.4 Å”, “Bi- 4.9 Å”, and “Mono- 5 Å”, respectively. This suggests that we are not in the classical effusion regime. Classical effusion requires the pore size to be smaller than the mean free path of the molecule which is ~60 nm at room temperature and ambient pressures. However, we are in a regime where the pore size is much smaller and on the order of the molecule size, therefore it is necessary to consider the molecular size and chemistry. Air leaking back into Microcavity Figure S9 shows air leaking back into a microcavity after all the H2 had rapidly escaped after etching. This is a bilayer sample that was etched in the same manner as the membranes presented in the main text. After etching the sample was filled with 200 kPa of H2 before being imaged. Hydrogen quickly leaks out leaving a near vacuum in the microcavity under the graphene. After 3000 min the deflections changed from -90 nm to -50 nm. This leak rate is consistent with previously measured leak rates for air leaking into an initially evacuated microcavity of similar geometry3. This result further suggests that we are measuring the transport thorough the porous graphene for H2 while N2 is diffusing through the silicon oxide substrate. 10 Supplementary References: Koh, Y.K., Bae, M.-H., Cahill, D.G. & Pop, E. Reliably counting atomic planes of few-layer graphene (n > 4). ACS Nano 5, 269-274 (2011). Reif, F. Fundamentals of Statistical and Thermal Physics. 651 (McGraw-Hill Book Company: New York, NY, 1965). Bunch, J.S. et al. Impermeable atomic membranes from graphene sheets. Nano Lett. 8, 2458-2462 (2008). Hencky, H. Uber den spannungzustand in kreisrunden platten mit verschwindender biegungssteiflgeit. Z. fur Mathematik und Physik 63, 311-317 (1915). Koenig, S.P., Boddeti, N.G., Dunn, M.L. & Bunch, J.S. Ultrastrong adhesion of graphene membranes. Nature Nanotech. 6, 543-546 (2011). Timoshenko, S., Young, D.H. & Weaver, W. Vibration Problems in Engineering. 481-484 (John Wiley and Sons, Inc.: New York, 1974). Breck, D.W. Zeolites Molecular Sieves: Structure, Chemistry, and Use. 593-724 (Wiley: New York, NY, 1973). Jiang, D.-en, Cooper, V.R. & Dai, S. Porous graphene as the ultimate membrane for gas separation. Nano Lett. 9, 4019-4024 (2009). 1. 2. 3. 4. 5. 6. 7. 8. Supplementary Figures: 11 Figure S1: Raman Spectrum of Graphene Samples (a) Raman spectrum of graphene flake containing membrane “Bi- 3.4 Å” from the main text taken before etching. (b) I(G)/I(Si) for flake in (a). The open circle and square were taken from a nearby flake containing both mono and bilayer sections. (c) Raman spectrum of membrane “Bi- 4.9 Å” from the main text before (black) and after (red) etching. (d) Raman spectrum for the monolayer samples presented in the supplementary information. Upper red curve is from the flake containing “Mono- 3.4 Å” from figure S5 of the supplementary information after etching. The lower black curve is for the monolayer membrane presented in Figure S6. Both were taken after etching. 12 Figure S2: Visualization of UV etching on suspended graphene (a) AFM scan of a membrane etched for a longer time to visualize the pore growth. The red areas are pits created by the UV etching. (b) Histogram of the number of pores versus the approximate pore area. (c) Histogram of the number of pores versus the equivalent radius of the pore. (b) and (c) indicate a nucleation and growth mechanism for pore evolution. 13 Figure S3: Schematic of Resonant Frequency Leak Rate Measurements (a) Schematic of the gas permeation through porous graphene membranes as measured by optical resonance. First the membrane is put in vacuum and the membrane is flat with a frequency of fo corresponding to zero tension in the membrane. After a pressure of a given gas species is introduced to the vacuum chamber the pressure difference across the membrane will induce tension causing the vibrational frequency to increase. If the gas species kinetic diameter is smaller than that of the pore size (red) it will pass through the pore(s) and the pressure difference will equalize and, therefore, the tension and resonant frequency will decrease with time. (b) If the gas species is larger than the pore size (green), the gas will not pass through the graphene membrane and the tension and resonant frequency will stay constant with time. 14 Figure S4: Sample Resonant frequency curves for CH4 Amplitude vs drive frequency for 80 torr of CH4. The data corresponds to the frequencies shown in Fig 3 inlay of main text taken at t = 0 s (black), t = 1 s (red), t = 3 s (green), t = 5 s (blue), t = 7 s (cyan), t = 11 s (magenta), and t = 13 s (orange). 15 Figure S5: Monolayer graphene showing selectivity H2/N2 selectivity (a) Maximum deflection, δ, vs, t for a monolayer membrane. The rapid decrease in deflection that becomes negative is consistent with the results seen in Fig 1 of the main text. Inlay: optical image of the monolayer graphene membrane covering one well in the substrate. (b) AFM line scans of the membrane in (a) as time passes. 16 Figure S6: Monolayer graphene showing SF6 permeation and pore instability (a) Frequency vs time for N2, H2, CO2, N2, CH4, and SF6, taken in that order. (b) A zoom in of (a). The change in N2 leak rate indicates that the pore(s) in monolayer graphene are not stable and the pore size can change. After the pore was enlarged, the membrane was able to allow SF6 to leak through the membrane. 17 Figure S7: Additional bilayer membranes measured (a) Normalized dn/dt vs. Molecular size showing permeation of all gas species larger than CH4 before and after etching. This membrane was damaged before the CH4 data could be taken. (b) Normalized dn/dt vs. Molecular size for the membrane “Bi-3.4 Å” before and after etching. (c) Normalized dn/dt vs. Molecular size for a membrane showing an increase in the leak rate of H2, and no significant increase in the leak rate for CO2, Ar, N2, and CH4. (a), (b), and (c) where all from the same graphene flake that can be found in the inlay of Fig 1f. 18 Figure S8: Comparing Flow Rates to Classical Effusion (a) Normalized dn/dt for membrane “Bi- 3.4 Å” plotted versus the inverse square root of the molecular mass of H2 and CO2. (b) Normalized dn/dt for membrane “Bi- 4.9 Å” plotted versus the inverse square root of the molecular mass of H2, CO2, N2 and CH4. 19 Figure S9: Air leaking back into Microcavity Maximum deflection, δ, vs, t showing the air leaking back into the microcavity after all the H2 has rapidly leaked out through pores created in the graphene. The microcavity was initially filled with 200 kPa of H2. Inlay show the optical image of this sample. Supplementary Tables: Table S1: Ideal gas separation factors for membrane “Bi- 3.4 Å” 20 Table S2: Ideal gas separation factors for membrane “Bi- 4.9 Å” Table S3. Ideal gas separation factors from membrane “Mono- 5 Å” 

1105.2511

1

1105

Directed motion of C60 on a graphene sheet subjected to a temperature gradient

Nonequilibrium molecular dynamics simulations is used to study the motion of a C60 molecule on a graphene sheet subjected to a temperature gradient. The C60 molecule is actuated and moves along the system while it just randomly dances along the perpendicular direction. Increasing the temperature gradient increases the directed velocity of C60. It is found that the free energy decreases as the C60 molecule moves toward the cold end. The driving mechanism based on the temperature gradient suggests the construction of nanoscale graphene-based motors.

cond-mat.mes-hall

cond-mat

Directed motion of C60 on a graphene sheet subjected to a temperature gradient A. Lohrasebi1, M. Neek-Amal2, and M. R. Ejtehadi3 1Department of Physics, University of Isfehan, Isfehan, Iran. 2Department of Physics, Shahid Rajaee Teacher Training University, Lavizan, Tehran 16788, Iran. 3 Department of Physics, Sharif University of Technology, Tehran P.O.Box 1155-9161, Iran. April 11, 2021 Abstract Nonequilibrium molecular dynamics simulations is used to study the motion of a C60 molecule on a graphene sheet subjected to a temperature gradient. The C60 molecule is actuated and moves along the system while it just randomly dances along the perpendicular direction. Increasing the temperature gradient increases the directed velocity of C60. It is found that the free energy decreases as the C60 molecule moves toward the cold end. The driving mechanism based on the temperature gradient suggests the construction of nanoscale graphene-based motors. 1 Introduction Since graphene has been discovered [1], many properties of this two-dimensional material have been studied both experimentally [2] and theoretically [3, 4]. In a recent experimental research the dynamic of light atoms deposited on a single- layer garphene has been studied by the mean of transmission electron microscopy (TEM) technic [5]. The two-dimensional structure of graphene suggests the possibility of motion with just two degrees of freedom. The free energy surface for a particle moving above a graphene sheet explains different motion-related phenomena at nanoscale as well as the various directed motions on the carbon nanotube-based motors [6, 7]. A net motion can be obtained from a nanoscale system subjected to a ther- mal gradient [8, 9]. Recently the motion of an experimentally designed nanoscale motor consisting of a capsule-like carbon nanotube inside a host carbon nan- otube has been explained successfully with molecular dynamics (MD) simula- tions [10]. The capsule travels back and forth between both ends of the host carbon nanotube along the axial direction. Barreiro et al. have designed an 1 artificial nanofabricated motor in which one short carbon nanotube travels rel- ative to another coaxial carbon nanotube [7]. This motion is actuated by a thermal gradient as high as 1 K nm−1 applied to the ends of the coaxial carbon nanotubes. Since graphene has a very high thermal conductivity (3000-5000 W K−1 m−1 [11, 12]), as high as diamond and carbon nanotubes [13, 14, 15, 16], it is a good can- didate for heat transferring designs in nano-electromechanical systems. Because of strong covalent bonds in graphene, thermal lattice conduction dominates the electrons contribution [11]. Recently Yang et al. [17] have studied the thermal conductivity and thermal rectification of trapezoidal and rectangular graphene nanoribbons and found a significant thermal rectification effect in asymmetric graphene ribbons [17]. Here we study the motion of a nanoscale object, e.g. C60, on a graphene sheet, in the presence of a temperature gradient. We show that the graphene is a good two-dimensional substrate for thermal actuation due to its high thermal conductivity, however as it is expected, in the absence of a thermal gradient, the C60 molecule randomly diffuses on the graphene sheet [6]. The average velocity along the temperature gradient direction and the free energy change throughout the system are calculated. This paper is organized as follows. In Sec. 2 we will introduce the atomistic model and the simulation method. Sec. 3 contains the main results including those for the produced temperature gradient, the trajectory of the C60 molecule over the graphene sheet and the free energy change. A brief summary and conclusions are included in Sec. 4. 2 The model and method The system was composed of a graphene sheet as a substrate, with dimensions Lx × Ly = 70×5 nm2 and a C60 molecule above the sheet. The graphene sheet with N = 14 400 carbon atoms was divided into 12 equal rectangular segments. Each segment with Nl = 1200 carbon atoms was arranged in 40 atomic rows (along armchair or x-direction). Each row has 30 atoms which were arranged along zigzag direction. The system was equilibrated for 300 ps at T = 300 K before the temperature gradient was applied. Once the system was equilibrated, the first (hot spot) and last (cold end) segments of the graphene sheet were kept at Th and Tc, respectively. A temperature gradient between the two ends was then produced, i.e. (Th − Tc)/Lx (top panel of Fig. 1). To make the model more efficient and prevent crumpling (see Fig. 2(a)) of the ends, we fixed the z-components of the first atomic row in the first segment and the last row of the last segment (see Fig. 2(b)). We carried out MD simulations employing two types of the interatomic po- tentials: 1) the covalent bonds between the carbon atoms in the graphene sheet and in the C60 molecule are described by Brenner potential [18] and 2) the non- bonded Van der Waals interactions between the graphene atoms and those of the C60 molecule. Brenner potential has been parameterized to model sp2 covalent 2 Lx = 70 nm Tc ∆T = 100K ∆T = 80K ∆T = 60K ∆T = 40K m n 5 = y L Th 340 320 K / T 300 280 260 2 4 6 l 8 10 12 Figure 1: (Color online) Top: The model which shows applied temperature gradient along x-direction. Bottom: Produced four temperature gradients after 450 ps of a nonequilibrium molecular dynamics simulation. Delta symbols are related to ∆T = 100 K, and for square symbols ∆T = 80 K, circle symbols ∆T = 60 K and for gradient symbols ∆T = 40 K, respectively. 3 (a) (b) Figure 2: (Color online) (a) A snapshot of a not-constrained system shows the effect of the ends crumpling. (b) Fixing z-components of the atoms at the first and last rows (see the text) prevents the crumpling. The black dots show a typical trajectory of the motion of a C60 molecule over the graphene sheet from the hot spot toward the cold spot. 4 bonds in the graphene, carbon nanotube and C60 structures. For the non- bonded potential a Lennard-Jones (LJ) potential gives reasonable results [19]. Here we choose the LJ parameters as ǫ = 2.413 meV and σ = 3.4 A [20] which represent the depth and range of the LJ potential energy, respectively. Note that the LJ potential is a simple and commonly used potential for modeling the interaction between carbon nanostructures [21, 22]. The equations of motion were integrated using a velocity-Verlet algorithm with a time step ∆t = 0.5 fs. The temperature of the hot end (Th) and the cold end (Tc) were held constant by a Nos´e-Hoover thermostat. The temperature of the inner segments were not controlled by the thermostat. Periodic boundary condition was applied only in the y-direction. Due to the applied temperature gradient, the system can no longer be described by equilibrium methods and we shall thus employ nonequi- librium molecular dynamics simulations. A temperature gradient were produced across the system during 450 ps and a stationary state was established. The C60 molecule was put above the second segment at zcm = 8 A (here index cm refer to the center of mass). During the production runs of 750 ps both the x and y positions of the center of mass of the C60 molecule were recorded. Moreover the temperature of each segment was calculated by measuring the total kinetic en- ergy of that segment. In our simulations a typical value for the relative standard deviation of the total energy of the extended system is about 3.5 × 10−5. A full simulation run takes about 50 h CPU time on a 3.2 GHz Pentium IV processor with 4 GB RAM. In order to compute the change in the free energy, one can employ the commonly used thermodynamic integration and perturbation methods [23]. A good estimation for the absolute value of the free energy requires sampling the whole phase space which is not feasible. Jarzynski's method removes this difficulty for nonequilibrium simulations [24]. There is an equality between the change of the free energy F and the work W applied on the system (here the C60 molecule) [24] ∆F = −β−1 ln exp(−βW ), (1) where β = 1/kB T (T is the temperature in each segment), and the average is taken over different configurations with different initial conditions. In fact, Eq. (1) connects the change of the free energy (between two equilibrium state) and the applied work on the system in a nonequilibrium process. 3 Results and discussion 3.1 Producing temperature gradient The temperature profiles for different temperature gradients with ∆T = 40, 60, 80, 100 K are shown in Fig. 1 (bottom panel). In this figure, the local temperatures of each segment which were obtained by averaging over 500 data are indicated by symbols. Corresponding error bars indicate the statistical errors and are in the range 4-6 K. Notice that the temperature profiles are nonlinear which is a commonly observed behavior in nonequilibrium molecular dynamics simulation 5 200 100 1 - s m / x v 62 60 58 56 54 52 50 Å / x 0 -100 -200 0 40 80 60 ∆T / K 100 (a) 200 400 t / ps 600 20 Å / y 0 -20 0 10 5 0 Å / > y < -5 (b) 600 200 400 t / ps -10 -200 0 <x> / Å (c) 200 Figure 3: (Color online) (a, b) Time series of x and y components of the center of mass of a C60 molecule over a monolayer graphene subjected to a temperature gradient. Thick curves show the average curves. The inset shows the variation of the velocity versus temperature gradient. (c) The trajectory of the motion of a C60 molecule (in x − y plane) moving over the monolayer graphene, averaged over six simulations with different initial conditions. 10000 0.5 0 F C A V 2 Å / > 2 ) ) 0 ( y - ) t ( y ( < 8000 -0.5 0 τ / ps 500 1000 6000 4000 2000 8000 t / ps 16000 Figure 4: (Color online) Mean square displacement for y-component of the center of mass of C60 molecule over the graphene. The inset shows the velocity autocorrelation function for the y-component of the motion as a function of time. 6 0 -2 -4 V e -6 / W 200 400 t / ps 600 -8 -10 -12 (a) -2 -4 V e -6 / F ∆ -8 -10 -12 (b) 200 400 t / ps 600 Figure 5: (Color online) (a) Total work performed on C60 molecules and (b) the change of free energy during C60 motion from the hot end to the cold end. The thick curve is the average over six simulations with different initial conditions. of thermal conductivity [14, 15, 16]. It is a consequence of the strong phonon scattering caused by the heat source or heat sink and can be explained by partly diffusive and partly ballistic energy transport along the system [14, 25]. Also, ripples in the graphene (Fig. 2(b)) are the other important source for phonon scattering [26, 27]. Therefore, these mechanisms cause a non-linear temperature profile in the middle segments. 3.2 Trajectory in the x − y plane The graphene ribbon is a two-dimensional way for the motion of the C60 molecule along it. C60 attempts to find its equilibrium state and looks for the local min- imum of the free energy. In other words, in the presence of a temperature gradient, the phonon waves created in the hot spot travel through the system, 7 interact and transfer momentum to the C60 molecule which results a net mo- tion [7]. The time series for the x and y coordinates, separately, are shown in Figs. 3(a,b). The trajectories of the six C60 molecules (from the six different simulations) in x−y plane over a time interval of 750 ps are depicted in Fig. 3(c). In all three cases the thick curve is the average of the six others. The motion along the x-direction is almost a linear uniform motion with velocities vx = 53, 56, 59 and 63 m/s for ∆T = 40, 60, 80, 100 K respectively (see inset of Fig. 3(a)). There are two computational methods that can be used to show that the motion along the y-direction is diffusive (Fig. 3(c)). First, one can use the Einstein relation to find the diffusion constant Dy by measuring the slope of the mean square displacement (MSD) of the y-components of the position, i.e. h(y(t) − y(0))2i. To show that the motion along the y-direction is driftless, we look at the MSD of y-component of the position of the C60 molecule. For a driftless motion the MSD should grows linearly with time, i.e. < (y(t)−y(0))2 > = 2Dyt. We depict the MSD in Fig. 4 for a long time simulation. As we see from this figure the slope of the MSD is almost one (after equilibration) which is a signature of diffusive regime. Alternatively one can use the Green- Kubo relation which makes use of the velocity autocovariance function (VACF). More specifically, one can take the integral over the autocovariance of the y- components of the position of the velocity of C60 as Dy = R ∞ 0 hvy(0)vy(τ )idτ , which gives the diffusion constant of random walk motion along the y-axis. The inset of Fig. 4 shows the VACF versus time. We have tested both methods in order to compute Dy. The result f or the diffusion constant is about 4 × 10−9 m2/s. The directed motion resulted from the temperature gradient along the graphene sheet, provides a nanoscale motor for the material transferring. Notice that in the absence of temperature gradient, when the system is equilibrated at T = 300 K, we found a diffusive motion on the graphene sheet [6]. 3.3 Free energy reduction Figures 5(a,b) show the variation of the total work and the changes in the free energy with time, respectively. The C60 loses on average almost 12 eV of free energy after 750 ps during the motion towards the cooler region. Note that here, the condition W > ∆F is always satisfied which is a well known criterion for a nonequilibrium (irreversible) thermodynamics evolution [24]. This method for calculating the change of the free energy is a well known method in computa- tional soft condensed matter [28] but it turns out to be useful also for studying various physical properties of graphene, particularly investigating the stability of new designs for nanoscale molecular devices as studied here. 4 Conclusions In summary, C60 moves directly toward the cold end of the graphene sheet when a temperature gradient is applied along armchair direction. It is found that the 8 lateral motion (along zigzag direction) is a diffusive motion. The reduction of the free energy of the system along the molecule motion is an indicative of the drifted motion. Comparing the free energy difference with the work performed on C60 shows that the process is indeed thermodynamically irreversible. The proposed mechanism for driving nanoparticles on a graphene sheet may be used in the design of novel nanoscale motors. 5 Acknowledgment We gratefully acknowledge valuable comments from Hamid Reza Sepangi and Ali Naji. References [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] A. K. Geim, and K. S. Novoselov, Nature Mater. 6, 183 (2007). [3] A. K. Geim and A. H. MacDonald, Phys. Today 60, 35 (2007). [4] A. H. Castro Neto, F. Guinea, N. M. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [5] J. C. Meyer, C. O. Girit, M. F. Crommie and A. Zettl, Nature 454, 319- 322 (2008). [6] M. Neek-Amal, N. Abedpour, S. N. Rasuli, A. Naji, and M. R. Ejtehadi, Phys. Rev. E, 82, 051605 (2010). [7] A. Barreiro, R. Rurali, E.R. Hernandez, J. Moser, T. Pichler, L. Forro, and A. Bachtold, Science 320, 775 (2008). [8] P. A. E. Schoen, J. H. Walther, S. Arcidiacono, D. Poulikakos, P. Koumout- sakos, Nano Lett. 6, 1910 (2006). [9] P. A. E. Schoen, J. H. Walther, D. Poulikakos, P. Koumoutsakos, Appl. Phys. Lett. 90, 253116 (2007). [10] H. Somada, K. Hirahara, S. Akita, and Y. Nakayama, Nano Lett. 9, 62 (2009). [11] A. A. Balandin, S. Ghosh, W. Bao, I. Calizo, D. Teweldebrhan, F. Miao, C. N. Lau, Nano Lett. 8, 902 (2008). [12] S. Ghosh, I. Calizo, D. Teweldebrhan, E. P. Pokatilov, D. L. Nika, A. A. Balandin, W. Bao, F. Miao, C. N. Lau, Appl. Phys. Lett. 92, 151911 (2008). [13] S. Berber, Y-K Kwon, and D. Tomanek, Phys. Rev. Lett. 84, 4613 (2000). 9 [14] P. K. Schelling, S. R. Phillpot, and P. Keblinski, Phys. Rev. B. 65, 144306 (2002). [15] E. Gonzalez Noya, D. Srivastava, Leonid A. Chernozatonskii, M. Menon, Phys. Rev. B. 70, 115416 (2004). [16] M. A. Osman and D. Srivastava, Nanotechnology, 12 21 (2001). [17] N. Yang, G. Zhang, and B. Li, Appl. Phys. Lett. 95, 033107 (2009). [18] D. W. Brenner, Phys. Rev. B. 42, 9458 (1990). [19] H. Ulbricht, G. Moos, and T. Hertel, Phys Rev Lett. 7, 095501 (2003). [20] G. Stan, M. J. Bojan, S. Curtarolo, S. M. Gatica, and M. W. Cole, Phys. Rev. B 62, 2173 (2000). [21] P. A. Gravil, M. Devel, Ph. Lambin, X. Bouju, Ch. Girard, and A. A. Lucas, Phys. Rev. B 53, 1622 (1996). [22] A. N. Kolmogorov, and V. H. Crespi, Phys. Rev. Lett 85, 4727 (2000). [23] F. Colonna, J. H. Los, A. Fasolino, and E. J. Meijer, Phys. Rev. B 80, 134103 (2009). [24] C. Jarzynski, Phys. Rev. Lett 78, 2690 (1997). [25] C. Oligschleger and J. C. Schon, Phys. Rev. B 59, 4125 (1999). [26] K. V. Zakharchenko, M. I. Katsnelson, and A. Fasolino, Phys. Rev. Lett 102, 046808 (2009). [27] N. Abedpour, M. Neek-Amal, R. Asgari, F. Shahbazi, N. Nafari, and M. R. Tabar, Phys. Rev. B 76, 195407 (2007). [28] H. Xiong, A. Crespo1, M. Marti, D. Estrin, and A. E. Roitberg, Theor. Chem. Acc. 116, 338 (2002). 10 

1812.10210

1

1812

Spin-valve Effect in Nanoscale Si-based Devices

The silicon (Si) based spin-MOSFET (metal-oxide semiconductor field-effect transistor) is considered to be the building block of low-power-consumption electronics, utilizing spin-degrees of freedom in semiconductor devices. In this paper, we review the latest results on the spin transport in nanoscale Si-based spin-valve devices, which is important to realize the nanoscale spin-MOSFET. Our results demonstrate the importance of ballistic transport in obtaining high spin-dependent output voltage in nanoscale Si spin-valve devices.

cond-mat.mes-hall

cond-mat

June 2018 vol. 28 no. 3 feature articles DOI: 10.22661/aaPPSBl.2018.28.3.07 Spin-valve effect in nanoscale Si-based Devices DUOnG DInH HIEP1, maSaaKI tanaKa2,3 anD PHam nam HaI1,3 1 DEPartmEnt OF ElECtrICal anD ElECtrOnIC EnGInEErInG, tOKyO InStItUtE OF tECHnOlOGy 2 DEPartmEnt OF ElECtrICal EnGInEErInG anD InFOrmatIOn SyStEmS, UnIvErSIty OF tOKyO 3 CEntEr FOr SPIntrOnICS rESEarCH nEtWOrK, GraDUatE SCHOOl OF EnGInEErInG, UnIvErSIty OF tOKyO ABSTRACT The silicon (Si) based spin-MOSFET (metal-oxide- semiconductor field-effect transistor) is considered to be the building block of low-power-consumption electron- ics, utilizing spin-degrees of freedom in semiconductor devices. In this paper, we review the latest results on the spin transport in nanoscale Si-based spin-valve devices, which is important to realize the nanoscale spin-MOS- FET. Our results demonstrate the importance of ballistic transport in obtaining high spin-dependent output volt- age in nanoscale Si spin-valve devices. INTRODUCTION Present computing technology is based on the manipu- lation of electron charge currents. However, this elec- tronic charge-based computing technology faces serious problems of high idling power consumption and heat generation due to the leakage off-current when a device' s size is miniaturized, which restricts a device's operating speed. Hence, there is demand for alternative low-power solutions to overcome these problems in the beyond- CMOS (complementary metal-oxide-semiconductor) era. The silicon (Si) based spin-MOSFET can be a promising solution because of its high compatibility with the well- established CMOS technology and long spin lifetime in Si [1,2,3,4,5]. For this reason, there has been a great interest in demonstrating spin injection and detection of spin transport in Si by ferromagnetic electrodes. In fact, spin injection into microns of Si channels by using the three terminal Hanle effect [6,7,8] or the four terminal spin-valve effect [9,10,11] has been demonstrated. How- ever, previous studies reported the spin transport only in micron-scale Si channels and the typical spin-dependent output voltages were less than 1 mV, which is not enough for realistic spin-MOSFET applications. To improve the spin-dependent output voltage, we have proposed and demonstrated the spin-valve effect utiliz- ing ballistic electron transport in nanoscale Si channels. It is expected that the ballistic transport of electrons in such nanoscale channels may overcome the conductivity mismatch problem that arises at the interface between a ferromagnetic (FM) electrode and a diffusive semicon- ductor (SC) channel [12,13], resulting in higher spin- dependent output voltage. In our studies, we fabricated Si-based spin-valve devices with a ferromagnetic spin in- jector / detector and a Si channel as short as 20 nm and sys-tematically investigated their spin-dependent trans- port characteristics [14,15]. We have observed a clear spin-valve effect up to 3% and a spin output voltage up to ~ 20 mV, which is the highest value reported so far in lateral spin-valve devices [15]. We also found that the sign of the spin-valve effect is reversed at low tempera- tures, suggesting the possibility of the spin-blockade effect of defect states in the MgO/Ge tunneling barrier [15]. DEvICE FABRICATION In this study, we used a highly doped n-type Si (100) sub- strate with an electron density n = 1 × 1018 cm -- 3. The Si substrates were first cleaned by an H2SO4/H2O2 solu- 7 feature articles Bulletin fig. 1: (a) Schematic spin-valve device structure with an Fe/Mgo/Ge spin injector / detector and a 20 nm-long Si channel. (b) Scanning electron microscopy image (top view) of a device. A silicon channel with a length of 20 nm was formed between the Fe electrodes. (c) Current -- voltage characteristics (i-v curves) of a MBe-device at various temperatures. (d)-(e) temperature dependence of the resistance of a eB-device measured at 20 mv and the MBe-device measured at 150 mv, respectively. Reprinted from Appl. Phys. lett. 109, 232402 (2016) and J. Appl. Phys. 122, 223904 (2017), with the permission of AiP Publishing. tion, then etched by diluted hydrofluoric acid solution to remove the native oxide layer, and rinsed in de-ionized water. To form a tunneling barrier and ferromagnetic electrodes for our devices (see Fig. 1(a)), we used two dif- ferent deposition methods; electron-beam (EB) evapora- tion and molecular-beam epitaxy (MBE). A. Thin film deposition by EB evaporation The first series of samples were fabricated by EB evapo- ration. After the cleaning process, the Si substrates were introduced into an ultra-high-vacuum EB evaporation chamber with a base pressure of 8 × 10 -- 6 Pa to deposit a 10 nm Fe layer and finally capped with a 3 nm Au thin film. In order to enhance the spin injection efficiency from Fe to Si, we inserted an MgO/Ge double layer be- tween the Fe electrodes and Si substrates. According to the literature, MgO is a promising spin-dependent tunnel barrier for efficient spin injection from a ferro- magnetic electrode into semiconductors (SCs) [16,17]. However, it is difficult to grow a very thin MgO layer with a thickness of 1~2 nm on Si at room temperature to be used as a tunnel barrier. On the other hand, epitaxial growth of MgO on Ge has been reported [18,19]. Moreo- ver, deposition of a smooth thin film of Ge on Si has been demonstrated at low deposition temperature [20]. Therefore, we deposited an ultrathin (1 nm) Ge film as a buffer layer between MgO and Si, to improve the quality of the MgO layer deposited at room temperature. Here, we used 2 nm-thick MgO / 1 nm-thick Ge double layers as a tunnel barrier between the Si channel and 10 nm- thick Fe electrodes. The Fe layer was capped with a 3 nm- thick Au layer to prevent oxidation. All of the deposition processes were conducted in the same EB evaporation chamber without breaking the vacuum. B. Thin film growth by MBE In the second series of samples, in order to improve the crystal quality of the tunnel barrier and the Fe electrodes, we grew the Fe/MgO/Ge stack by MBE. After the clean- ing process, the Si substrates were introduced into a MBE chamber with a base pressure of 1 × 10 -- 8 Pa for growing the stack, which consisting of (from top to bottom) a 10 nm Fe layer, 2 nm MgO tunnel barrier, 1 nm Ge, and 1.1 nm MgO cap layer. Knudsen cells were used for thermal evaporation of Ge and Fe, while a low-power electron- beam evaporator was used to deposit MgO with a slow rate of 0.03 Å/s. In-situ reflection high energy electron diffraction confirmed that the MgO layer grown on the Ge buffer layer became crystalline when the MgO thick- ness exceeded 1 nm. Furthermore, X-ray dif-fraction has been employed to investigate the crystal quality of the tunnel barrier, and has confirmed that the epitaxial 8 June 2018 vol. 28 no. 3 feature articles relationship is Fe(001) MgO(001) Ge(001). Based on these results, we expect that the crystal quality of the MBE-grown MgO/Ge double layer is much better than that grown by EB evaporation in the first series [14]. C. Nanofabrication of spin-valve devices After the thin film deposition, we used electron-beam lithography (EBL) and ion-milling techniques to fabri- cate nanoscale Si spin-valve devices. Fig. 1(a) shows the schematic structure of our devices examined here. First, we used the EBL and lift-off technique to pattern a 30 nm-thick Au hard mask with a 20 nm-long and 100 nm- wide gap in between, followed by Ar ion milling to etch the exposed Fe area and define a 20 nm-long Si channel. Then, the Au (40 nm) / Cr (5 nm) pad electrodes were deposited by EB evaporation. Fig. 1(b) shows a top-view scanning electron microscopy (SEM) image of a fabricat- ed device. A nanoscale Si channel with a length of 20 nm was formed between the Fe electrodes. In the following sections, the spin-valve devices with Fe/MgO/Ge layers deposited by EB (MBE) method are denoted as EB (MBE) spin-valve devices CURRENT -- vOLTAGE ChARACTERISTICS We investigated the conductance behavior of our de- vices by measuring current -- voltage characteristics (I-V curves). We found non-linear I-V in all the devices, sug- gesting that tunneling transport through the tunnel bar- rier contact is dominant. Fig. 1(c) shows representative I-V curves of a MBE spin-valve device at various tempera- tures. A strong dependence of I-V curves on temperature has been observed as well. At low temperatures, thermi- onic emission of electrons from the Fe electrodes to the Si over the tunnel barrier is suppressed. In addition, free carriers in the Si channel are partly quenched. There- fore, the current becomes smaller at low temperatures. This suggests that the resistance of the device is domi- nated by the transport through the spin-valve structure consisting of Fe(interface) / (MgO/Ge) / Si-channel / (Ge/MgO) / Fe(interface). Fig. 1(d) and Fig. 1(e) show the resistance of an EB-device and an MBE-device as a function of temperature, respectively. In both devices, we observed rapid increases in resistance as temperature decreased, from the order of ~ 101 Ω at 300 K to ~ 103 Ω and 108 Ω at 15 K, respectively, for the EB-device and the MBE-device. From these data, we can estimate the contribution of the parasitic anisotropic magnetoresis- tance (AMR) of the Fe electrodes to the total spin-valve effect MR as follows. The total device resistance R can be decomposed into two components: R = RFe + Rsv, where RFe is the parasitic resistance of the Fe electrodes and Rsv is the intrinsic resistance of the Fe(interface)/ (MgO/Ge)/ Si/(Ge/MgO)/ Fe(interface) spin-valve structure. Since the parasitic RFe(T) decreases as temperature decreases, RFe(T) ≤ RFe(300K) < R(300K) ≈ 30 Ω. The AMR ef- fect ∆RFe/RFe of Fe is in the order of 0.1%, thus ∆RFe is in the order of 0.3 Ω. From this consideration, we can estimate the contribution of the parasitic AMR effect of ∆RFe/R(T) at various temperatures. For example, in the MBE-device, ∆RFe/R(T) at 200 K should be in the order of 10 -- 4, while that at 15 K should be in the order of 10 -- 8. Therefore, we conclude that the contribution of the para- sitic AMR effect of the Fe electrodes is negligible at low temperatures. SPIN-DEPENDENT TRANSPORT ChARACTERISTICS A. Local spin-valve effect In this work, we use the 2-terminal local spin-valve ef- fect to characterize the spin transport. We show that it is possible to distinguish the intrinsic spin-valve effect from parasitic local effects by systematic measurements of the bias voltage dependence, temperature depen- dence, and magnetic-field direction dependence of the magneto-resistance (MR). Fig. 2 (a) and (b) show the MR characteristics of our spin-valve devices ((a) EB-device, (b) MBE-device) measured at low temperature with a magnetic field applied along the Si channel (x-direction in Fig. 1(a)). Blue dots are the resistance data taken when the magnetic field was swept from +3 kG to -3 kG, while red dots are taken when the magnetic field was swept from -3 kG to +3 kG. Fig. 2(a) shows a clear jump of re- sistance ∆R up to 12 Ω (corresponding to ∆R/R = 0.8%) of the EB-device. For the MBE-device, as shown in Fig. 2(b), we observed a huge drop of resistance ∆R ~ 57 kΩ, corresponding to ∆R/R = 3%, which is more than tri- ple that of the EB-device. This result shows the impor- tance of crystal quality for improving the spin-valve effect signal. Note that we observed negative MR at 15 K for the MBE-device, and the sign of its MR will be discussed later. The ∆R and ∆R/R values in the MBE-device are five orders of magnitude larger than that of the parasitic AMR effect of the Fe electrodes at 15 K (∆RFe ~ 0.3 Ω and ∆RFe/R=10 -- 8). Furthermore, we measured ∆R at various bias voltage V, and found that ∆R strongly de- pends on V, as shown in Fig. 2(c). The ∆R-V relation- ship closely follows R-V. These cannot be explained by the AMR effect, because ∆RFe does not depend on the 9 feature articles Bulletin fig. 2: Magnetoresistance characteristics of (a) the eB-device measured at 4.3 K with a bias voltage of 100 mv, and (b) the MBe-device measured at 15 K with a bias voltage of 300 mv. Blue dots are the resistance data taken when the magnetic field was swept from +3 kG to -3 kG, while red dots are taken when the magnetic field was swept from -3 kG to +3 kG. (c) Magnetoresistance ∆R and device resistance R of the MBe-device as a function of bias voltage at 15 K. Reprinted from Appl. Phys. lett. 109, 232402 (2016) and J. Appl. Phys. 122, 223904 (2017), with the permission of AiP Publishing. bias voltage. From the above quantitative and qualitative considerations, we conclude that the contribution of the parasitic AMR effect is negligible, and that the observed MR effect originates from the spin-dependent tunneling process of electrons between the Fe electrodes and the Si channel through the MgO/Ge barrier. In the following sec-tions, we will concentrate on spin-valve effect of the MBE-device. B. Magnetic-field direction dependence Besides the AMR effect, there is the TAMR effect at the Fe/MgO/Ge/Si interface, which possibly affects the two- terminal voltage as well. This TAMR effect is due to the dependence of the tunneling density of states (DOS) in the FM electrodes on the magnetization direction. In order to distinguish the intrinsic spin-valve effect from the TAMR effect at the Fe/MgO/Ge/Si interface, we have measured the dependence of the MR curve on the mag- neticfield direction ϕ with respect to the x-direction, as shown in Fig. 3(a). Fig.s 3(b)-3(f) show the MR curves of the MBE-device measured at 100 K when the magnetic field was applied in the film plane along ϕ = 0o, 30o, 45o, 60o, and 90o, respectively. In the case of TAMR, the resistance state of the FM electrodes depends on the magnetization direction, then the MR curve would be reversed or changed in shape when ϕ changes from 0o to 90o [23,24,25]. However, we have observed the same shape and polarity of the MR curves for all the ϕ values, indicating that the observed MR depends only on the relative angle between the magnetic moments of the two Fe electrodes. Thus, we conclude that the observed MR is caused by the spin-valve effect. C. Temperature dependence Fig. 4(a) shows the temperature dependence of the spin- 10 fig. 3: (a) Measurement configuration of the magnetic field direction dependence of the magnetoresistance of the MBe-device. Here, the magnetic field is applied in the x-y plane along the angle ϕ with respect to the Si channel (the x-direction). (b)-(f) Magnetoresistance curves measured at 100 K with a bias voltage of 100 mv, when ϕ = 0o, 30o, 45o, 60o, and 90o, respectively. Reprinted from J. Appl. Phys. 122, 223904 (2017), with the permission of AiP Publishing. valve MR ratio (∆R/R) of the MBE-device. In this Fig., the negative (positive) spin-valve MR ratio corresponds to the resistance drop (jump) in the MR curves. We found that the local spin-valve MR ratio is negative at tempera- tures lower than 200 K but turns to positive as usual at temperatures higher than 200 K. The inset of Fig. 4(a) shows a representative positive MR curve at 250 K. This phenomenon is unusual and has not been observed be- June 2018 vol. 28 no. 3 feature articles fore in Si-based spin-valve devices. For understanding the mechanism of this inverse spin spin-valve effect, we focus on the electron tunneling process between the Fe electrodes and Si channel in our devices. Fig. 1(d) shows that the resistance of the MBE- device increases exponentially as temperature decreases, from the order of 101 Ω at 300 K to 108 Ω at 4 K. This strong temperature dependence of the device resistance R suggests that electrons tunnel through the barrier by fig. 4: (a) temperature dependence of the spin-valve MR ratio (defined as ∆R/R) of the MBe-device measured at a bias voltage of 300 mv. inset shows the MR curve of this device at 250 K. (b)-(c) Proposed spin-dependent transport model through defect-induced gap states in the Mgo barriers for parallel and anti- parallel magnetization configuration, respectively. At low temperatures, defect- induced gap states in the Mgo barrier at the spin injector side are occupied with majority spins, thus only minority spins can tunnel through the Mgo barrier because of the Pauli exclusion principle (spin-blockade effect). this leads to the inverse spin-valve effect, where the resistance is high at (b) parallel and low at (c) antiparallel magnetization configuration. Reprinted from J. Appl. Phys. 122, 223904 (2017), with the permission of AiP Publishing. transport via some defect states in the barrier, whose en- ergy level is much lower than the intrinsic barrier height between Fe an MgO. Such defect states in MgO barri- ers have been widely studied. For instance, in Fe/MgO/ Fe magnetic tunnel junctions, oxygen vacancy defects in the MgO barrier form gap states at about 1.2 eV below the conduction band bottom of MgO, which reduces the MgO barrier height to 0.39 eV [26]. Hence, we assume the existence of such defect-induced gap states inside the MgO barrier of our spin-valve devices. We propose a model using these defect-induced gap states to explain the inverse spin-valve effect, as shown in Fig. 4(b) and (c). In our model, the two Fe electrodes work as a spin injector and detector. Spin-polarized electrons tunnel from the Fe spin injector to the Si channel through such gap states inside the MgO barrier. These electrons then transport in the Si channel without or with little scatter- ing (ballistic or quasi ballistic transport), and reach the opposite MgO barrier with higher kinetic energy. Finally, the electrons tunnel through gap states inside this MgO barrier to the Fe spin detector. In this picture, the rel- evant gap states are those with energy levels higher than the Fermi level of the Fe spin injector. At high tempera- tures, many electrons with high enough thermal energy in the Fe spin injector can tunnel to the gap states (ther- mally activated tunneling), and the device resistance is low (~ 101 Ω at 300 K). However, at low temperatures, the number of electrons that have enough thermal ener- gy rapidly decreases, resulting in much higher resistance (~ 108 Ω at 4 K). Then, only a limited number of the gap states whose energy levels are very close to the Fermi level of the Fe spin injector can allow such tunneling. If these gap states are filled with majority spins whose residence time is long enough at low temperatures, only minority spins from the spin injector can pass through. This is a so-called "spin-blockade", a phenomenon origi- nating from the Pauli exclusion principle, and has been observed in several quantum dot systems as well as defect states in semiconductors [27,28]. On the contrary, the electrons that arrive at the opposite MgO barrier can have higher kinetic energy, thus they can tunnel through many available gap states whose energy levels are higher than the Fermi level of the Fe spin detector. Hence, spin- blockade strongly occurs mostly at the spin injector side, not at the spin detector side. Therefore, the device resistance becomes high (low) at parallel (antiparallel) magnetization configuration, as shown in Fig. 4(b) and 4(c). This explains the observed inverse spin-valve effect at low temperatures. As temperature increases, there are more available gap states with higher energy levels for 11 feature articles Bulletin thermally activated tunneling from the spin injec-tor. Furthermore, the residence time of electrons at the de- fect states can be much shorter at high temperatures. For this reason, the spin-blockade is not effective anymore, thus the conventional spin-valve effect is dominant at higher temperatures. Our model is also consistent with the impurity assisted tunneling model proposed for expla-nation of the three-terminal Hanle effect observed in some FM/SC systems [29]. D. Spin-dependent output voltage In this subsection, we describe the spin-dependent out- put performance of our devices. The spin-dependent output voltage ∆V = (∆R/R)×V of the order 100 mV is important for correct read-out in realistic applications. However, previous studies on spin injection into Si chan- nels reported a low read-out voltage of only a few μV in 4 terminal measurements, and about 1 mV in 3 terminal measurements. In this study, by employing the nano- scale Si channel, we have significantly improved ∆V. Fig. s 5(a) and 5(b) show ∆V of our EB-device and MBE- device, respectively. In the EB-device, we succeeded in increasing ∆V up to 13 mV at V = 1.7 V. In the MBE- device, we have significantly improved the crystal quality of the Fe/MgO/Ge junctions, as evidenced by the much higher device resistance at 4 K (~ 108 Ω, compared to ~ 103 Ω of our EB-devices). Therefore, we achieved ∆V = 20 mV at V = 0.9 V, which is the highest value re- ported so far in lateral Si-based spin-valve devices. fig. 5: Bias voltage dependence of the spin-dependent output voltage ∆v = (∆R/R)v of (a) the eB-device, and (b) the MBe-device, respectively. the highest output voltage ∆v of 20 mv was achieved in the MBe-device at the bias voltage of 0.9 v. Reprinted from Appl. Phys. lett. 109, 232402 (2016) and J. Appl. Phys. 122, 223904 (2017), with the permission of AiP Publishing. E. Role of ballistic transport in the nanoscale Si channel In this subsection, we describe the advantages of ballistic transport in improving the spin-valve signal in our nano- scale Si channel. Theoretically, in the diffusive transport regime [12,13], the spin polarization of the current at the , FM/SC interface is given by (SP)I = in which, β is the spin-polarization of the FM layer, rF and rN are characteristic resistances defined by rF = ρF and rN = ρN , where ρF, ρN are the resistivity of the FM and SC, and are the spin-diffusion length in FM and SC. This equation clearly shows the important role of the ratio rN/rF in determining the spin polarization at the interface. For example, using ρF ~ 1 × 10 -- 5 Ωcm and ~ 2 nm for Fe, ρN = 2 × 10 -- 2 Ωcm for n-Si (n =1018 ~10 μm, the ratio rN / rF is cm -- 3) and assuming that ~ 1 × 107, which is much larger than β. This means that there is almost no spin polarization at the interface be- tween Fe and n-Si. In case of the FM/SC/FM structure, , 12 , the . For the assuming that the length of the SC layer tN ≪ spin-valve ratio is given by Fe/n-Si interface, even when tN = 20 nm, the ratio is of the order of 10 -- 9, which means that there is almost no spin-valve effect. This is so-called the conductivity mismatch problem and can be overcome by introducing a spin-dependent tunnel barrier at the interface between FM and SC [30], which has been widely used to obtain spin injection into SC. However, there are some problems remaining in this method. Even though spin injection from FM into SC through a tunnel barrier has been defi- nitely demonstrated, typical values of the spin-valve ratio reported so far in lateral devices are as small as 0.01% ~ 0.1%. On the other hand, the conductivity mismatch problem can also be overcome by using ballistic (or quasi- ballistic) transport in nanoscale SC channels. In the ballis- June 2018 vol. 28 no. 3 feature articles fig. 6: Magnetoresistance of several Si MBe spin-valve devices with different channel length of l = 20 nm, 500 nm, 1 µm, and 6 µm, (a)-(d) measured at 300 K with a bias voltage of 80 mv, and (e)-(h) measured at 15 K with a bias voltage of 300 mv. Here, the magnetic field was applied along the current direction. Reprinted from J. Appl. Phys. 122, 223904 (2017), with the permission of AiP Publishing. tic transport regime, Ohm's law and diffusion equations used to derive the spin-polarized current are no longer valid. If the SC channel length is comparable or shorter than the electron mean free path, the transport regime changes from diffusive to (quasi) ballistic. In this case, the transport in the SC channel may be modeled using quantum mechanics instead of classical transport equa- tions of electrons. Then, the transfer matrix method may be used to calculate the spin-polarized transport in the SC channel [2]. Utilizing ballistic transport to overcome the conductivity mismatch problem has obvious advan- tages such as (i) high spin-valve ratios, similar to tunnel- ing magnetoresistance (TMR), can be achieved; and (ii) the spin-valve effect can be observed even if there is no tunnel barrier, allowing high current driving capability of spin-transistors. In this study, in order to investigate the role of ballistic transport in our nanoscale Si channel, we also prepared and measured Si spin-valve devices with long channel lengths of L = 500 nm, 1 μm, and 6 μm by MBE. These lengths are long enough for electrons to transport in diffusive transport regime, as compared with our nano- scale Si spin-valve devices. Fig.s 6 (a) - (d) present the MR curves measured at room temperature for the de- vices with different channel lengths L = 20 nm, 500 nm, 13 feature articles Bulletin 1 μm, and 6 μm, respectively. At room temperature, all the devices show low re-sistance values of the order of 10 Ω. This low resistance indicates the ineffectiveness of tunnel barriers of these devices because of various ther- mally activated transport processes through defect states in the barriers. Without the barrier resistance, no spin- valve effect can be expected for the long-channel devices. Indeed, the data in Fig. 6(b)-(d) show no spin-valve ef- fect for the devices with L = 500 nm, 1 μm, and 6 μm, as predicted from the diffusion theory. In contrast, Fig. 6(a) shows a clear spin-valve effect of about 0.042% for the nanoscale device, even though there is effectively no barrier for this device at room temperature. This demon- strates the important role of (quasi) ballistic transport in generating the spin-valve effect when there is no tunnel barrier. Fig.s 6(e)-(f) show the MR curves of these devices at 15 K, measured with a bias voltage of 300 mV. At this low tem- perature, all the devices showed high resistance values of MΩ, indicating the effectiveness of the barrier in suppressing the thermally activated transport processes through defect states. Such barriers can eliminate the con-ductivity mismatch problem in long-channel devic- es. As a result, all of the devices showed relatively large spin-valve ratios. However, the data clearly show that the spin-valve ratio systematically decreases with increasing channel length; the MR ratio decreases from ~ 3% for L = 20 nm to 2.2%, 2% and 1.8% for L = 500 nm, 1 μm, and 6 μm, respectively. The drop of the spin-valve ratio is slow, consistent with the long spin-diffusion length in Si. The results demonstrate two important roles of (quasi) ballistic transport in the nanoscale Si channel; (i) the generation of the spin-valve effect even without a tunnel barrier at room temperature, and (ii) the suppression of spin-flip scattering to achieve a higher spin-valve ratio at low temperature. We have obtained a large spin-valve ratio of 3% in the nanoscale Si channel, which is much larger than those observed in μm-long Si channel devices reported before. The electron transport in our nano- scale devices may be quasi ballistic rather than fully bal- listic, because the channel length (20 nm) is not much shorter than that of the mean free path of 20 ~ 40 nm in n-type Si with n = 1018 cm -- 3 [31]. By using lightly-doped Si substrates and further downsizing the Si channel length to sub-10 nm, we expect to achieve fully ballistic transport and higher spin-valve signals. CONCLUSIONS In this study, we have systematically investigated spin transport in nano-scale Si spin-valve devices. For the MBE-device, a huge spin-valve effect with ∆R up to 57 kΩ, corresponding to ∆R/R = 3%, has been clearly observed. Interestingly, we observed the inverse spin- valve effect at low temperatures, suggesting the possibil- ity of the spin-blockade effect of defect states in the MgO tunnel barrier. The highest spin-dependent output volt- age is 20 mV at the bias voltage of 0.9 V at 15 K, which is the highest value reported so far in lateral Si-based spin- valve devices. Our results demonstrate the important role of ballistic transport in improving the spin-valve effect in nanoscale Si spin-valve devices. This work is an important step towards the realization of nanoscale spin-MOSFETs. By using lightly-doped Si substrates and further downsizing the Si channel length to sub-10 nm, we expect to achieve fully ballistic transport and higher spin-valve signals. Acknowledgments: This work was partly supported by Grants-in-Aid for Scientific Research (Grant Nos. 23000010, 16H02095, and 18H01492), Nanotechnology Platform 12025014 (F-17-IT-0010) from MEXT (Ministry of Education, Culture, Sports, Science and Technology), CREST of JST, the Yazaki Foundation, and the Spintron- ics Research Network of Japan. References [1] ItrS, Int. technol. roadmap Semicond. (2014). [2] S. Sugahara and m. tanaka, appl. Phys. lett. 84, 2307 (2004). [3] m. tanaka and S. Sugahara, IEEE trans. Electron Devices 54, 961 (2007). [4] I. Žutić, J. Fabian, and S. Das Sarma, rev. mod. Phys. 76, 323 (2004). [5] v. Zarifis and t.G. Castner, Phys. rev. B 36, 6198 (1987). [6] S.P. Dash, S. Sharma, r.S. Patel, m.P. de Jong, and r. Jansen, nature 462, 491 (2009). [7] S.P. Dash, S. Sharma, J.C. le Breton, J. Peiro, H. Jaffrès, J.m. George, a. lemaître, and r. Jansen, Phys. rev. B. Phys. rev. B 84, 054410 (2011). [8] C.H. li, O.m.J. van't Erve, and B.t. Jonker, nat. Commun. 2, 245 (2011). [9] B. Huang, D.J. monsma, and I. appelbaum, Phys. rev. lett. 99, 177209 (2007). [10] t. Sasaki, t. Oikawa, t. Suzuki, m. Shiraishi, y. Suzuki, and K. tagami, appl. Phys. Express 2, 053303 (2009). [11] y. aoki, m. Kameno, y. ando, E. Shikoh, y. Suzuki, t. Shinjo, m. Shiraishi, t. Sasaki, t. Oikawa, and t. Suzuki, Phys. rev. B. 86, 081201 (2012). [12] G. Schmidt, D. Ferrand, l.W. molenkamp, a.t. Filip, and B.J. van Wees, Phys. rev. B 62, r4790 (2000). [13] a. Fert and H. Jaffrès, Phys. rev. B 64, 184420 (2001). [14] D.D. Hiep, m. tanaka, and P.n. Hai, appl. Phys. lett. 109, 232402 (2016). [15] D.D. Hiep, m. tanaka, and P.n. Hai, J. appl. Phys. 122, 223904 (2017). [16] X. Jiang, r. Wang, r.m. Shelby, r.m. macfarlane, S.r. Bank, J.S. Harris, and S.S.P. Parkin, Phys. rev. lett. 94, 056601 (2005). 14 June 2018 vol. 28 no. 3 feature articles [17] C. martínez Boubeta, E. navarro, a. Cebollada, F. Briones, F. Peiró, and a. Cornet, J. Cryst. Growth 226, 223 (2001). [18] K.-r. Jeon, C.-y. Park, and S.-C. Shin, Cryst. Growth Des. 10, 1346 (2010). [19] W. Han, y. Zhou, y. Wang, y. li, J.J.I. Wong, K. Pi, a.G. Swartz, K.m. mcCreary, F. Brunner, G. Schmidt, and l.W. molenkamp, Phys. rev. lett. 94, 27203 (2005). [25] J. moser, a. matos-abiague, D. Schuh, W. Wegscheider, J. Fabian, and D. Weiss, Phys. rev. lett. 99, 056601 (2007). Xiu, K.l. Wang, J. Zou, and r.K. Kawakami, J. Cryst. Growth 312, 44 (2009). [26] S. yuasa, t. nagahama, a. Fukushima, y. Suzuki, and K. ando, nat. mater. 3, [20] C. Schöllhorn, m. Oehme, m. Bauer, and E. Kasper, thin Solid Films 336, 868 (2004). 109 (1998). [21] X. lou, C. adelmann, S.a. Crooker, E.S. Garlid, J. Zhang, K.S.m. reddy, S.D. Flexner, C.J. Palmstrøm, and P.a. Crowell, nat. Phys. 3, 197 (2007). [22] r. nakane, S. Sato, S. Kokutani, and m. tanaka, IEEE magn. lett. 3, 3000404 (2012). [27] K. Ono, D. G. austing, y. tokura, and S. tarucha, Science 297, 1313 (2002). [28] B. Weber, y.H.m. tan, S. mahapatra, t.F. Watson, H. ryu, r. rahman, l.C.l. Hollenberg, G. Klimeck, and m.y. Simmons, nat. nanotechnol. 9, 430 (2014). [29] O. txoperena, y. Song, l. Qing, m. Gobbi, l.E. Hueso, H. Dery, and F. [23] C. Gould, C. rüster, t. Jungwirth, E. Girgis, G.m. Schott, r. Giraud, K. Casanova, Phys. rev. lett. 113, 146601 (2014). Brunner, G. Schmidt, and l.W. molenkamp, Phys. rev. lett. 93, 117203 (2004). [30] E.I. rashba, Phys. rev. B 62, r16267 (2000). [31] B. Qiu, Z. tian, a. vallabhaneni, B. liao, J.m. mendoza, O.D. restrepo, X. [24] C. rüster, C. Gould, t. Jungwirth, J. Sinova, G.m. Schott, r. Giraud, K. ruan, and G. Chen, Europhys. lett. 109, 57006 (2015). Duong Dinh Hiep received his B.e. in physics from Ho Chi Minh City university of Science in 2008, and his M.e. in material science from Japan institute of Advance Science and technology in 2015. He is currently a PhD candidate at the Department of electrical and electronic engineering, tokyo institute of technology. His research focuses on spin transport in silicon-based nano-scale spintronics devices. Masaaki tanaka is the director of the Center for Spintronics Research network (CSRn), and a professor at the Department of electrical engineering & information Systems, Graduate School of engineering, the university of tokyo. He received his B.e., M.e., and PhD degrees in electronic engineering from the university of tokyo, Japan, in 1984, 1986, and 1989, respectively. His research field is electronic materials and device applications, in particular, spintronics materials, spin-related phenomena, and devices including magnetic semiconductors, ferromagnet/semiconductor heterostructures and nanostructures, magnetic tunnel junctions, spin transistors and other devices. He has authored and coauthored over 250 scientific publications, and presented over 120 invited talks at international conferences and meetings. Pham Nam Hai is an associate professor at the Department of electrical and electronic engineering, tokyo institute of technology, and a visiting associate professor at the Center for Spintronics Research network (CSRn), the university of tokyo. He received B.e., M.e. and PhD degrees in electronic engineering from the university of tokyo, Japan, in 2004, 2006, and 2009, respectively. His research interests include ferromagnetic nanoclusters, ferromagnetic semiconductors, ferromagnet/semiconductor hybrid systems, topological insulators, and their applications to spintronic devices. 15 

1308.3953

2

1308

Density of States Scaling at the Semimetal to Metal Transition in Three Dimensional Topological Insulators

The quantum phase transition between the three dimensional Dirac semimetal and the diffusive metal can be induced by increasing disorder. Taking the system of disordered $\mathbb{Z}_2$ topological insulator as an important example, we compute the single particle density of states by the kernel polynomial method. We focus on three regions: the Dirac semimetal at the phase boundary between two topologically distinct phases, the tricritical point of the two topological insulator phases and the diffusive metal, and the diffusive metal lying at strong disorder. The density of states obeys a novel single parameter scaling, collapsing onto two branches of a universal scaling function, which correspond to the Dirac semimetal and the diffusive metal. The diverging length scale critical exponent $\nu$ and the dynamical critical exponent $z$ are estimated, and found to differ significantly from those for the conventional Anderson transition. Critical behavior of experimentally observable quantities near and at the tricritical point is also discussed.

cond-mat.mes-hall

cond-mat

Density of States Scaling at the Semimetal to Metal Transition in Three Dimensional Topological Insulators Koji Kobayashi1, Tomi Ohtsuki1, Ken-Ichiro Imura2,4, and Igor F. Herbut3,4 1Department of Physics, Sophia University, Chiyoda-ku, Tokyo, 102-8554, Japan 2Department of Quantum Matter, AdSM, Hiroshima University, Higashi-Hiroshima, 739-8530, Japan 3Department of Physics, Simon Fraser University, Burnaby, British Columbia, V5A 1S6, Canada and 4Max-Planck-Institut fur Physik komplexer Systeme, Nothnitzer Str. 38, 01187 Dresden, Germany (Dated: November 5, 2018) The quantum phase transition between the three dimensional Dirac semimetal and the diffusive metal can be induced by increasing disorder. Taking the system of disordered Z2 topological insulator as an important example, we compute the single particle density of states by the kernel polynomial method. We focus on three regions: the Dirac semimetal at the phase boundary between two topologically distinct phases, the tricritical point of the two topological insulator phases and the diffusive metal, and the diffusive metal lying at strong disorder. The density of states obeys a novel single parameter scaling, collapsing onto two branches of a universal scaling function, which correspond to the Dirac semimetal and the diffusive metal. The diverging length scale critical exponent ν and the dynamical critical exponent z are estimated, and found to differ significantly from those for the conventional Anderson transition. Critical behavior of experimentally observable quantities near and at the tricritical point is also discussed. PACS numbers: 71.30.+h, 05.70.Jk, 71.23.-k, 71.55.Ak Topological classification of different insulating phases [1, 2] is an emerging new paradigm in condensed matter physics. Unlike in the Landau theory of phase transi- tions that is rooted in the idea of spontaneous breaking of symmetry [3], it is less clear how to describe different universality classes of the transitions between topologi- cally different phases. This is because the usual notion of the local order parameter characterizing the differ- ent phases is often lacking. At the transition between topologically distinct phases, on the other hand, the gap closes, and the system becomes a semimetal. In three dimensions (3D) such a critical phase is stable in pres- ence of weak disorder [4], but as disorder is increased it gives way to a diffusive metallic state [5]. This transition belongs to a distinct universality class that exhibits non- trivial dynamical and diverging length scale exponents z and ν, for example [5, 6]. The 3D Dirac Hamiltonian in presence of disorder is ubiquitous: it applies to certain phases of superfluid 3He [7], degenerate semiconductors [5], and to the Weyl semimetals [8 -- 11]. Related theories of disordered critical points for two-dimensional interact- ing Dirac fermions and bosons were also advanced in the past [12, 13]. In this paper we discuss how this disorder-induced fermionic criticality is reflected in the scaling behavior of a readily available physical quantity, the single parti- cle density of states (DOS), which can be understood as a proper order parameter that characterizes such a tran- sition. We then express the critical behavior of Dirac electron velocity, diffusion coefficient, conductivity and anomalous diffusion exponent in terms of z and ν. Such a surprisingly simple description is contrasted with the conventional Anderson transition [14 -- 16], where the DOS remains smooth through the transition. ) 2 / m W ( r e d r o s i d 10 8 6 4 2 0 (c) M (b) P c TI1 (WTI) (a) DSM TI2 (STI) -2.5 -2 -1.5 -1 mass term (m m 0 / 2 ) FIG. 1: (Color online) Typical phase diagram of the system under consideration. TI1 and TI2 correspond, respectively, to weak and strong topological insulator (WTI and STI), and DSM to the critical Dirac semimetal phase. The dotted line in the diffusive metal (M) phase (c) is an extrapolation [17] of the DSM line (a). The tricritical point (b) is denoted as Pc. In order to produce and control the semimetallic phase, we focus on a 3D time-reversal symmetric topological in- sulator under disorder. The Z2 topological insulator is interesting in itself, and has lately been a subject of in- tense theoretical and experimental research, with a num- ber of real material realizations [18]. Consider the phase diagram of a system exhibiting both weak and strong topological insulators (WTI and STI) as some parameter is varied [19 -- 21] (see Fig. 1). In three spatial dimensions disorder is irrelevant in the renormalization group sense, so that at weak disorder a direct transition between two topologically distinct insulating phases [4], say, between TI1 and TI2, remains. (In the specific situation we con- sider below, TI1 = WTI and TI2 = STI.) Only above a finite strength of disorder W > 0, does the bulk en- ergy gap become completely filled with impurity levels, so that the insulating phases are replaced by a diffusive metallic (M) phase [22] (see Fig. 1). Since TI1 and TI2 are characterized by a different topological number pro- tected by the bulk energy gap, at the phase boundary the bulk spectrum is in general closed. In the present case the system is also protected by time-reversal symmetry, and such a gap closing appears as a (Kramers) degen- erate pair of point nodes, i.e., as the Dirac semimetal (DSM) [23] line in the phase diagram. As disorder is increased the DSM line also terminates at the intersec- tion with the insulator-metal phase boundary. In the following we focus on the evolution of the DOS as one moves along the DSM line, through the tricritical point Pc where the DSM line terminates, and finally reaches inside the metallic phase. We have previously established, by a detailed numeri- cal study of the conductance [22], that although disorder W shifts the position of the phase boundary [24 -- 29] (de- termined, e.g., by the position of the conductance peak), it is nevertheless irrelevant; the peak height of the con- ductance on the DSM line is not influenced by the disor- der strength. It was also found [22] that on the DSM line the DOS remains a quadratic function of low energies, exactly as in the clean limit [see the curves (a) in Fig. 2]. Whereas the quadratic behavior is left intact by disorder, the coefficient of the quadratic term, which is related to the velocity v of Dirac electrons, is renormalized [30], as in Eq. (21) below. In this Letter we further quantify the behavior of the DOS on the DSM line toward the diffusive metal phase, and demonstrate that the DOS obeys a single parame- ter scaling typical of second order phase transitions, with new values of critical exponents. Our analysis is based on the single parameter scaling hypothesis, which is sub- stantially supported by numerical results. The scaling behavior of the DOS is studied using the kernel polyno- mial method (KPM) [31]. The 3D disordered Z2 topological insulator is modeled as a Wilson-Dirac-type tight-binding Hamiltonian with an effective momentum-dependent mass term [32], m(k) = m0 + m2 Xµ=x,y,z (1 − cos kµ) , (1) implemented on a cubic lattice. The topological nature of the model is controlled by the ratio of two mass parame- ters m0 and m2 such that an STI phase with Z2 (one strong and three weak) indices [19 -- 21] (ν0, ν1ν2ν3) = (1, 000) appears when −2 < m0/m2 < 0, while the regime of parameters: −4 < m0/m2 < −2 falls on a WTI phase with (ν0, ν1ν2ν3) = (0, 111) (see Fig. 1). 2 0.06 ρ) ( S O D 0.04 0.02 0 -1 (b) P c (a) DSM (c) M-phase 0 energy (ε) 1 FIG. 2: (Color online) Density of states calculated at differ- ent points of the phase diagram (2 ≤ W ≤ 7.5); (a) on the WTI/STI boundary, (b) at the tricritical point, and (c) in the M-phase. Its energy dependence ρ(ǫ) is quadratic on the WTI/STI boundary (a), becoming almost linear at the tri- critical point (b), while it acquires a finite value ρ(0) at ǫ = 0 on the M-side (c). We emphasize that these DOSs are not of the surface, but of the bulk. In real space our tight-binding Hamiltonian reads H =Xr Xµ=x,y,z (cid:20)r + eµi(cid:18) it 2 γµ − m2 2 γ0(cid:19) hr + h.c.(cid:21) +Xr rih(m0 + 3m2)γ0 + Vr14ihr , (2) where eµ is a unit vector in the µ-direction, and 14 rep- resents the 4 × 4 identity matrix. γµ and γ0 form a set of γ-matrices in a 4 × 4 representation, γµ = (cid:18) 0 σµ σµ 0 (cid:19) , γ0 = (cid:18)12 0 0 −12(cid:19) , (3) where σµ are Pauli matrices and 12 is 2 × 2 identity ma- trix. m0, m2 and t are mass and hopping parameters, and Vr represents a potential disorder distributed uniformly and independently between −W/2 and W/2. For simplicity, we have assumed the Hamiltonian Eq. (2) to be isotropic. In the actual computation we set the mass and hopping parameters to m2 = 1, t = 2. The linear size of the system L is taken to be 200 times the lattice constant, which is enough to reach the thermo- dynamic limit of DOS per unit volume. We also take the average over two samples, although the statistical error is already sufficiently small for L = 200, because of the self-averaging nature of DOS. The order of the Cheby- shev expansion in KPM is typically a few thousand, so that the DOS becomes smooth. The periodic boundary conditions are imposed on each direction. The scaling form of the density of states per volume near the Dirac point may be derived as follows. Be- gin with a dimensionless quantity, the number of states N (ǫ, L) below the energy ǫ in the system of size L in d dimensions, and assume that it is a function of dimen- sionless parameters L/ξ and ǫ/ǫ0, (a) 0.04 N (ǫ, L) = F (L/ξ, ǫ/ǫ0) , (4) where ξ is the characteristic length scale and ǫ0 is the characteristic energy scale. They are related via the dy- namical exponent z, ) 0 ( ρ ǫ0 ∝ ξ−z . (5) 0 5 Since the number of states should be proportional to Ld, the above scaling form should be N (ǫ, L) = (L/ξ)df (ǫξz) . (6) From N (ǫ, L), the DOS per volume ρ(ǫ) is calculated as ρ(ǫ) = 1 Ld dN (ǫ, L) dǫ , so that we finally obtain its scaling form, ρ(ǫ) = ρ(−ǫ) = ξz−df ′(ǫξz) . (7) (8) The first equality comes from the symmetry of DOS about ǫ = 0. Upon introducing the distance from the tricritical point δ = W − Wc/Wc, we may assume that the length scale ξ diverges near the tricritical point Pc as, ξ ∼ δ−ν , (9) where ν is the critical exponent. Around Pc, the scaling law, Eq. (8), therefore reads, For ǫ → 0, i.e., when the argument of the scaling function is small, one expects qualitatively different behavior in the M-phase and on the DSM line. If the system has Dirac cones, the DOS is expected to be proportional to ǫd−1 for ǫ ≪ ǫ0, so ρ(ǫ) ∼ δ(d−z)ν(ǫδ−zν)d−1 = ǫd−1δ−(z−1)dν . (11) In the M-phase, on the other hand, the DOS is finite at ǫ = 0, and ρ(0) ∼ δ(d−z)ν(ǫδ−zν)0 = δ(d−z)ν . (12) Right at the tricritical point δ = 0, ξ dependences in the prefactor and the argument of Eq. (8) should cancel, and consequently, ρ(ǫ) ∼ δ(d−z)ν(ǫδ−zν)(d−z)/z = ǫ(d−z)/z . (13) Armed with the above observations, we next study the DOS numerically. First, the DOS at ǫ = 0 vanishes [Fig. 3(a)] around ρ(ǫ) ∼ δ(d−z)νf ′(ǫδ−zν ) . (10) ρ(ǫ) ∼ c(δ)ǫ2 , 3 (b) ( ε) ρ 0.04 0.03 0.02 0.01 0 6 7 8 W 0 0.2 ε 0.4 0.6 FIG. 3: (Color online) (a) The DOS at ǫ = 0. The point Wc where ρ(0) → 0 indicates the tricritical point Pc. (b) The DOSs around Wc (solid lines, W = 6.3, 6.4, 6.5 from bottom to top). They can be approximated by a linear function (dot- ted line). The deviations for small energy regions are coming from the finite size effect ρ(0) ∼ L−2. We note that the effect of long ranged disorder [33], which might survive due to the finite lattice spacing, is not identified in our numerics. We use this value to define δ. The DOSs around W = Wc, i.e., near Pc, are plotted in Fig. 3(b). From the observed energy dependence and Eq. (13), we estimate (3 − z)/z = 1.00 ± 0.15 , z = 1.5 ± 0.1 . (15) (16) The result is consistent with the value z = 3/2 ob- tained to the first order in the critical disorder strength in Ref. [6]. Next we derive the critical exponent ν from the DOS for small ǫ. On the DSM line, by fitting the data to (17) (18) (19) (20) and then by fitting the coefficient c(δ) to the form c(δ)−1 ∼ δ3(z−1)νDSM , we find [Fig. 4(a)] 3(z − 1)νDSM ≃ 1.16 ± 0.05 , ∴ νDSM ≃ 0.81 ± 0.21 . The result can be interpreted physically as vanishing ve- locity of the Dirac electron along the DSM line towards the tricritical point δ = 0, v ∼ δ(z−1)ν ≈ δ0.4 . (21) In the M-phase, on the other hand, by fitting the data to Eq. (12), we find [Fig. 4(b)] (3 − z)νM ≃ 1.36 ± 0.09 , ∴ νM ≃ 0.92 ± 0.13 . (22) (23) Wc = 6.4 ± 0.1 . (14) The values of νM and νDSM agree within the margin of er- ror, and one expects in fact the same value on both sides (a) 100 50 ) δ ( c / 1 (b) 0.03 ) 0 ( ρ 0.02 0.01 0 0 0 0.2 0.4 0.6 0 0.05 0.1 0.15 0.2 δ  FIG. 4: (Color online) Dependence on δ (a) for Eq. (18) on the DSM line and (b) for Eq. (12) in the M-phase. We set Wc = 6.4. 1.E+01 1.E-01 v ) z - d ( - ρ 1.E-03 M-phase DSM 1.E-05 0.03 0.3 30 300 3 
-zv FIG. 5: (Color online) Single parameter scaling of the DOS. The upper branch corresponds to the DOS in M-phase, and the lower branch to the DSM line. We set the parameters Wc = 6.4, z = 1.5, and ν = 0.86 = (νDSM + νM)/2. of the transition. The first order perturbation theory in the location of the critical point [6] yields the character- istic νDSM = νM = 1, which also falls within our intervals on both sides. Lastly, and most importantly, we show that the single parameter scaling law, Eq. (10), fits successfully all of our numerical data. Figure 5 is the plot of the scaling combination ρ(ǫ)δ−(d−z)ν vs. ǫδ−zν, with the above es- timates of Wc, z, and with using the average of the two exponents, ν = (νDSM + νM)/2 = 0.86. A similar value for ν would also follow had we solved Eqs. (19) and (22) under the assumption that νM = νDSM. After cutting off the relatively large energy region outside the Dirac cone and the very small energy region where the DOS becomes too small to estimate numerically, all the curves in Fig. 2 collapse onto two distinct branches, corresponding to the M-phase and to the DSM line, respectively. This is the central result of the present work. The general scaling arguments imply interesting trans- port properties as well. Consider, for example, the wave packet dynamics [34]. We assume the mean square dis- 4 placement hr2(t, ǫ)i of the state with energy ǫ at time t, where h· · ·i represents both quantal and ensemble aver- ages to be of the form hr2(t, ǫ)i ∼ ξ2g(tξ−z, ǫξz) . (24) In the M-phase, one expects hr2(t, ǫ)i = 2dD(ǫ)t for large t with D(ǫ) the diffusion coefficient at energy ǫ. We focus only on the state with ǫ = 0, hr2(t, 0)i ∼ ξ2−zt , (25) implying the diffusion coefficient D(0) to diverge while the conductivity σ(0) ∼ ρ(0)D(0) to vanish towards Pc as D(0) ∼ δ−(2−z)ν , σ(0) ∼ δ(d−2)ν , (26) the latter coinciding with the Wegner's relation [35], and predicts σ(0) ∼ δ0.9. At Pc, the ξ dependence should vanish, leading to hr2(t, 0)i ∼ ξ2(tξ−z)2/z = t2/z ≈ t1.3 , (27) which implies superdiffusion: when z ≃ 1.5 < 2, the system at Pc is more diffusive than in the M-phase. The numerical verification of such a superdiffusive behavior is, however, difficult, since we need to focus on the wave packet dynamics of ǫ = 0 state, the DOS of which is vanishing. Study is in progress to improve the situation. Another interesting quantity is the conductance distri- bution along the DSM line. Away from Pc, the conduc- tance will be narrowly distributed about the value ex- pected in the absence of randomness as demonstrated in Ref. [22]. At Pc, we expect the scale independent broad conductance distribution as in the case of the Anderson transition [36, 37]. In summary, we have proposed the scaling of the den- sity of states as a characteristic of the semimetal to metal transition in general, or, of the tricritical point among the two topologically different insulating phases and the metallic phase, in particular. In contrast to the conven- tional Anderson transitions, the density of states plays the role of the order parameter and shows the universal single-parameter scaling. This idea of using DOS to char- acterize DSM is also relevant in different systems such as the ones reported recently in Refs. [33] and [38]. Fur- thermore, we have estimated numerically the dynamical exponent z ≃ 1.5, which is clearly different from the con- ventional value z = 3 [35] for the Anderson transition in 3D. The critical exponent of divergence of the length scale ν ≃ 0.9 is less accurate, but it also seems rather far from the conventional value ν ≃ 1.35 [39] for the Anderson transition in 3D symplectic class. The poor inaccuracy of ν originates from the uncertainty of Wc and z. High precision estimate of Wc by different methods such as the transfer matrix [22] would improve the estimate. In this paper, we have focused on the phase boundary of the strong and weak topological insulators. The rea- son is practical; the DSM line and the phase boundary of metal to topological insulator phases intersect with a large angle, allowing us to pinpoint Pc easily. For the phase boundary of the strong topological and ordinary in- sulators (STI/OI) [22], it is rather challenging to locate Pc, because the DSM line and the phase boundary of metal to insulator seem to intersect with a shallow angle. Because of the universal nature of critical phenomena, we expect similar scaling behavior with the same criti- cal exponents for the semimetal to metal transition for STI/OI. On the other hand, different critical behavior is expected for the case of Z topological superconductor de- scribed by a Bogoliubov-de Gennes Hamiltonian, which shows similar phase diagram but belongs to a different universality class (DIII). This work was supported by Grants-in-Aid for Scien- tific Research (C) (Grants No. 23540376) and Grants-in- Aid 24000013. I. F. H. is supported by the NSERC of Canada. K. K. and T. O. would like to thank Zhejiang In- stitute of Modern Physics, where fruitful discussion with V. E. Sacksteder and K. Slevin has been made. [1] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Lud- wig, Phys. Rev. B 78, 195125 (2008). [2] A. Kitaev, AIP Conference Proceedings 1134, 22 (2009). [3] I. Herbut, A Modern Approach to Critical Phenomena (Cambridge University Press, Cambridge, 2007). [4] R. Shindou and S. Murakami, Phys. Rev. B 79, 045321 (2009). [5] E. Fradkin, Phys. Rev. B 33, 3263 (1986). [6] P. Goswami and S. Chakravarty, Phys. Rev. Lett. 107, 196803 (2011). [7] G. E. Volovik, The Universe in a Helium Droplet (Oxford University Press, 2003). [8] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011). [9] A. A. Burkov, M. D. Hook, and L. Balents, Phys. Rev. B 84, 235126 (2011). [10] M. Neupane, S.-Y. Xu, R. Sankar, N. Alidoust, G. Bian, C. Liu, I. Belopolski, T.-R. Chang, H.-T. Jeng, H. Lin, et al., arXiv:1309.7892 (2013). [11] S. Borisenko, Q. Gibson, D. Evtushinsky, V. Zabolotnyy, B. Buchner, and R. J. Cava, arXiv:1309.7978 (2013). [12] I. F. Herbut, Phys. Rev. Lett. 87, 137004 (2001). 5 [13] I. F. Herbut, V. Jurici´c, and O. Vafek, Phys. Rev. Lett. 100, 046403 (2008). [14] P. W. Anderson, Phys. Rev. 109, 1492 (1958). [15] B. Kramer and A. MacKinnon, Rep. Prog. Phys. 56, 1469 (1993). [16] F. Evers and A. D. Mirlin, Rev. Mod. Phys. 80, 1355 (2008). [17] We employed a polynomial of the second order for the extrapolation. The result is not sensitive to the extrapo- lation method near Pc. [18] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [19] L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). [20] J. E. Moore and L. Balents, Phys. Rev. B 75, 121306 (2007). [21] R. Roy, Phys. Rev. B 79, 195322 (2009). [22] K. Kobayashi, T. Ohtsuki, and K.-I. Imura, Phys. Rev. Lett. 110, 236803 (2013). [23] S. M. Young, S. Zaheer, J. C. Y. Teo, C. L. Kane, E. J. Mele, and A. M. Rappe, Phys. Rev. Lett. 108, 140405 (2012). [24] J. Li, R.-L. Chu, J. K. Jain, and S.-Q. Shen, Phys. Rev. Lett. 102, 136806 (2009). [25] C. W. Groth, M. Wimmer, A. R. Akhmerov, J. Tworzyd lo, and C. W. J. Beenakker, Phys. Rev. Lett. 103, 196805 (2009). [26] H.-M. Guo, G. Rosenberg, G. Refael, and M. Franz, Phys. Rev. Lett. 105, 216601 (2010). [27] A. Yamakage, K. Nomura, K.-I. Imura, and Y. Ku- ramoto, J. Phys. Soc. of Jpn. 80, 053703 (2011). [28] E. Prodan, Phys. Rev. B 83, 195119 (2011). [29] A. Yamakage, K. Nomura, K.-I. Imura, and Y. Ku- ramoto, Phys. Rev. B 87, 205141 (2013). [30] Q. Wu, L. Du, and V. E. Sacksteder, Phys. Rev. B 88, 045429 (2013). [31] A. Weisse, G. Wellein, A. Alvermann, and H. Fehske, Rev. Mod. Phys. 78, 275 (2006). [32] C.-X. Liu, X.-L. Qi, H.-J. Zhang, X. Dai, Z. Fang, and S.-C. Zhang, Phys. Rev. B 82, 045122 (2010). [33] R. Nandkishore, D. A. Huse, and S. L. Sondhi, arXiv:1307.3252 (2013). [34] T. Ohtsuki and T. Kawarabayashi, J. Phys. Soc. Jpn. 66, 314 (1997). [35] F. Wegner, Z. Phys. B 25, 327 (1976). [36] B. Shapiro, Phys. Rev. Lett. 65, 1510 (1990). [37] K. Slevin and T. Ohtsuki, Phys. Rev. Lett. 78, 4083 (1997). [38] Y. Ominato and M. Koshino, arXiv:1309.4206 (2013). [39] Y. Asada, K. Slevin, and T. Ohtsuki, J. Phys. Soc. Jpn., Supplement 74, 258 (2005). 

1511.02389

1

1511

Programmable Extreme Pseudomagnetic Fields in Graphene by a Uniaxial Stretch

Many of the properties of graphene are tied to its lattice structure, allowing for tuning of charge carrier dynamics through mechanical strain. The graphene electro-mechanical coupling yields very large pseudomagnetic fields for small strain fields, up to hundreds of Tesla, which offer new scientific opportunities unattainable with ordinary laboratory magnets. Significant challenges exist in investigation of pseudomagnetic fields, limited by the non-planar graphene geometries in existing demonstrations and the lack of a viable approach to controlling the distribution and intensity of the pseudomagnetic field. Here we reveal a facile and effective mechanism to achieve programmable extreme pseudomagnetic fields with uniform distributions in a planar graphene sheet over a large area by a simple uniaxial stretch. We achieve this by patterning the planar graphene geometry and graphene-based hetero-structures with a shape function to engineer a desired strain gradient. Our method is geometrical, opening up new fertile opportunities of strain engineering of electronic properties of 2D materials in general.

cond-mat.mes-hall

cond-mat

Accepted by Physical Review Letters on 6 November 2015 Programmable Extreme Pseudomagnetic Fields in Graphene by a Uniaxial Stretch Shuze Zhu1, Joseph A. Stroscio2, Teng Li1,* 1Department of Mechanical Engineering, University of Maryland, College Park, MD 20742, USA 2Center for Nanoscale Science and Technology, NIST, Gaithersburg, MD 20899, USA Many of the properties of graphene are tied to its lattice structure, allowing for tuning of charge carrier dynamics through mechanical strain. The graphene electro-mechanical coupling yields very large pseudomagnetic fields for small strain fields, up to hundreds of Tesla, which offer new scientific opportunities unattainable with ordinary laboratory magnets. Significant challenges exist in investigation of pseudomagnetic fields, limited by the non-planar graphene geometries in existing demonstrations and the lack of a viable approach to controlling the distribution and intensity of the pseudomagnetic field. Here we reveal a facile and effective mechanism to achieve programmable extreme pseudomagnetic fields with uniform distributions in a planar graphene sheet over a large area by a simple uniaxial stretch. We achieve this by patterning the planar graphene geometry and graphene-based hetero-structures with a shape function to engineer a desired strain gradient. Our method is geometrical, opening up new fertile opportunities of strain engineering of electronic properties of 2D materials in general. * To whom correspondence should be addressed. Email: lit@umd.edu Being able to influence the motion of charge carriers, strain-induced pseudomagnetic fields in graphene have been explored as a potential approach to engineering the electronic states of graphene. There has been experimental evidence of enormous pseudomagnetic fields (up to 300 T) in locally strained graphene nanobubbles [1] and graphene drumheads [2], which inspires enthusiasm in exploring the abundant potential of strain engineering of graphene, as well as charge carrier behavior under extreme magnetic fields that otherwise do not exist in normal laboratory environments [3-9]. Enthusiasm aside, there exist significant challenges that hinder further explorations of these fertile opportunities to full potential. For example, existing experiments demonstrate pseudomagnetic fields in highly localized regions of graphene with a non-planar morphology [1,2], which poses tremendous challenge for experimental control and characterization of the resulting fields. Further challenge originates from the dependence of the symmetry of the strain-induced pseudomagnetic field on the strain gradient in graphene. As a result, an axisymmetric strain field in graphene leads to a pseudomagnetic field of rotational threefold symmetry [2,4-7]. By contrast, a uniform pseudomagnetic field in a planar graphene with tunable intensity is highly desirable for systematic investigations [10]. In principle, such a uniform pseudomagnetic field can be achieved by introducing a strain field of threefold symmetry in graphene [4,8], which requires equal-triaxial loading of atomically thin graphene, a technical challenge already prohibitive in bulk materials. So far, a viable solution to generate a pseudomagnetic field in graphene with controllable distribution and amplitude over a large planar area under a feasible loading scheme still remains highly desirable but elusive. 2 The ever-maturing programmable patterning [11-22] and functionalization [23-30] of graphene has enabled a class of graphene-based unconventional nanostructures with exceptional functionalities, such as nanoribbon [31], nanomesh [16] and hybrid superlattices [23]. Significant progress has also been made on fabricating high quality in-plane heteroepitaxial nanostructures that consist of different monolayer two-dimensional (2D) crystals, such as graphene, hydrogenated graphene (graphane) and hexagonal boron nitride (h-BN) [32-35]. Furthermore, controllable and nondestructive generation of uniaxial strains (up to more than 10 %) in graphene has been successfully demonstrated recently [36]. Motivated by these advances, here we reveal a feasible and effective mechanism to achieve programmable pseudomagnetic fields in a planar graphene by a simple uniaxial stretch. We demonstrate two new possible approaches: 1) by tailoring the planar edge geometry of a graphene strip, and 2) by patterning in-plane graphene-based hetero-structures. These feasible-to-implement approaches yield rich features necessary for systematic studies of pseudomagnetic fields in strain engineered graphene geometries, as demonstrated below. When the graphene lattice is strained, the main effect is to modify the hopping energy between the two graphene sublattices. The modified energies add a term to the momentum operators in the low energy Dirac Hamiltonian, in the same way a vector potential is added for electromagnetic fields. This gives a very useful way to relate the mechanical deformation in graphene with a gauge field that acts on the graphene electronic structure [2-9]. The pseudomagnetic field, Bps, is given by the 2D curl of the mechanically derived gauge field. For 3 elastic deformations, the pseudomagnetic field in graphene is related to the strain field in the plane of the graphene as [2-9] 𝐵ps = 𝑡𝛽 𝑒𝑣F [−2 𝜕𝜖𝑥𝑦 𝜕𝑥 − 𝜕 𝜕𝑦 (𝜖𝑥𝑥 − 𝜖𝑦𝑦)], (1) where 𝛽 = 2.5 is a dimensionless coupling constant, 𝑡 = 2.8 eV is the hopping energy, 𝑣F = 1 × 106 m s−1 is the Fermi velocity, and 𝜖𝑥𝑥, 𝜖𝑦𝑦, and 𝜖𝑥𝑦 are the components of the strain tensor of the graphene. The x-axis is along the zigzag direction of graphene lattice. The field in Eq. (1) is for one graphene valley, with opposite sign for the other valley. We consider the pseudomagnetic field under the special case of uniaxial stretch (see Section I in Supplemental Material [47]) given by, 𝐵ps = 3𝑡𝛽 𝑒𝑣F (1 + 𝜈) 𝜕𝜖𝑦𝑦 𝜕𝑦 . (2) The above formulation reveals that a programmable pseudomagnetic field is achieved if the strain gradient 𝜕𝜖𝑦𝑦 𝜕𝑦 in graphene can be engineered under a simple uniaxial stretch. For example, a constant strain gradient 𝜕𝜖𝑦𝑦 𝜕𝑦 in graphene (i.e., a linear distribution of tensile strain in the armchair direction) can result in a uniform pseudomagnetic field over a large area of graphene; a highly desirable feature to enable direct experimental characterization of the resulting field. To demonstrate the feasibility to engineer the strain field in graphene under a simple uniaxial stretch, we first consider a graphene nanoribbon of length L that is patterned into a 4 shape with a varying width 𝑊(𝑦) and subject to an applied uniaxial tensile strain 𝜖𝑎𝑝𝑝 along its length in the y direction [Fig. 1(a)]. The geometry of the two long edges of the graphene nanoribbon is defined by a shape function 𝑓(𝑦) = 𝑊(𝑦)/𝑊0, where 𝑊0 = 𝑊(0) denotes the basal width of the graphene nanoribbon. When 𝐿 ≫ 𝑊(𝑦), it is reasonable to assume that 𝜖𝑦𝑦 is constant along any cross-section cut in x direction and only varies along y direction. This assumption is justified in the majority part of the graphene nanoribbon except in the vicinity of its four corners, as verified by both finite element modeling and atomistic simulations [47]. Considering the force balance along any cross-section cut in x direction, it is shown that [47] 𝜕𝜖𝑦𝑦 𝜕𝑦 = − 𝐹 𝐸𝑔𝑊0ℎ 1 𝑓2 𝑑𝑓 𝑑𝑦 , (3) where 𝐹 is the applied force at the ends of graphene nanoribbon necessary to generate the uniaxial tensile strain 𝜖app, 𝐸g and ℎ are the Young’s Modulus and thickness of graphene, respectively. Thus from Eq. (2), the resulting pseudomagnetic field in such a patterned graphene nanoribbon is given by, 𝐵ps = − 3𝑡𝛽𝐹 𝑒𝑣F𝐸g𝑊0ℎ (1 + 𝜈) 1 𝑓2 𝑑𝑓 𝑑𝑦 . (4) Equation (4) reveals that a tunable pseudomagnetic field is achieved under a uniaxial stretch by engineering the shape of the graphene nanoribbon. For example, to achieve a uniform pseudomagnetic field, the corresponding shape function is shown to be 5  𝑓(𝑦) = 𝑓r 𝐿 𝑓r (𝐿 − 𝑦)+ 𝑦 , (5) where 𝑓r = 𝑓(𝐿) denotes the ratio between the widths of the top and bottom ends of the graphene nanoribbon. The intensity of the resulting uniform pseudomagnetic field (see Section III in [47] for details) is given by, 𝐵ps = 6𝑡𝛽 𝑒𝑣F 𝜖app 𝐿 (1−𝑓r) (1 + 𝑓r) (1 + 𝜈). (6) To verify the above elasticity-based theoretical prediction, we performed numerical simulations using both finite element method and atomistic simulations (Sections V and VII in [47] for simulation details) to calculate the strains and pseudomagnetic field using Eq. (1), which lead to results well in agreement with the above theory, Eqs. (2-6), as elaborated below. Figure 1(a) shows the schematic of a graphene nanoribbon of 𝐿 = 25 nm, 𝑊0 = 10 nm, 𝑓r = 0.5, with two long edges prescribed by the shape function in Eq. (5). The ribbon is subject to an applied unidirectional stretch of 5 % in its length direction. Figure 1, (b) to (d), plots the components of the resulting strain in the graphene, 𝜖𝑥𝑥, 𝜖𝑦𝑦 and 𝜖𝑥𝑦, respectively, from finite element simulations. In the majority portion of the graphene except its four corners, 𝜖𝑥𝑥 and 𝜖𝑦𝑦 show a linear distribution along y direction [also see Fig. S1(a)], while 𝜖𝑥𝑦 shows a linear distribution along x direction. From Eq. (2), such a strain distribution will result in a rather uniform pseudomagnetic field in the graphene nanoribbon. Figure 1(e) plots the resulting pseudomagnetic fields in the graphene nanoribbon under 6 an applied uniaxial stretch of 5 %, 10 % and 15 %, respectively, all of which clearly show a uniform distribution in nearly the entire graphene ribbon except at its four corners. The intensity of the pseudomagnetic field as the function of location along the centerline of the graphene ribbon is shown in Fig. 1(f), for various applied uniaxial stretches (See Section V in [47] for detailed discussions). For each case, the plateau in a large portion of the curve shows a rather uniform and strong pseudomagnetic field along the centerline of the graphene nanoribbon (e.g., ≈150 T under a 15 % stretch). Further parametric studies [Fig. 1(g)] reveal that the intensity of resulting pseudomagnetic field is linearly proportional to the applied uniaxial stretch 𝜖𝑎𝑝𝑝 and inversely proportional to the length of the graphene ribbon L, in excellent agreement with the dependence from the theoretic prediction in Eq. (6) (See Section V in [47] for details). Our atomistic simulation results [Fig. S4] further verify both the uniform distribution of the resulting pseudomagnetic field in the graphene nanoribbon and the agreement on the field intensity with the results from finite element simulations. As additional verification, our density functional theory calculation produces pseudo-Landau levels, corresponding to cyclotron motion in a magnetic field [Fig. 1(h)], attesting to the presence of a strain-induced pseudomagnetic field for a graphene under a strain field of constant 𝜕𝜖𝑦𝑦 𝜕𝑦 (See Section II in [47] for details). Equation (6) also suggests another geometric dimension to tailor the intensity of pseudomagnetic field: tuning the top/bottom width ratio 𝑓r of the graphene nanoribbon. For nanoribbons with the same length, a smaller 𝑓r leads to more strain localization (i.e., a higher strain gradient) in the graphene nanoribbon, and thus a higher intensity of the pseudomagnetic 7 field. Figure S3 shows the geometry of 25 nm long graphene nanoribbons with three top/bottom width ratios, 𝑓r = 0.35, 0.5, and 0.7, with the two long edges of each nanoribbon prescribed by Eq. (5). The corresponding intensities of the resulting pseudomagnetic field from finite element simulations, as shown in Fig. S3(b), are in excellent agreement with the prediction from Eq. (6). The programmable pseudomagnetic field in planar graphene demonstrated above essentially originates from determining a shape function that yields a tunable effective stiffness in various locations of the graphene, which in turn leads to non-uniform distribution of strain under a uniaxial stretch. From a different point of view, the graphene nanoribbon in Fig. 1(a) can be regarded as a lateral 2D hetero-structure, consisting of a pristine graphene nanoribbon and two patches on its side made of 2D material (vacuum) with zero stiffness [e.g., Fig. 2(a)]. As a result, the effective stiffness of the graphene nanoribbon at different cross-section decreases from the wider end to the narrower end. The above mechanistic understanding indeed opens up more versatile approaches to achieving a programmable pseudomagnetic field in planar graphene hetero-structures, which we further explore as follows. Recent experiments demonstrate facile fabrication of high quality in-plane hetero-epitaxial nanostructures such as graphene/graphane and graphene/h-BN hetero-structures in a single 2D atomic layer [32-35]. The more corrugated lattice structures of graphane and h-BN lead to an in-plane stiffness smaller than that of pristine graphene. It is expected that such in-plane hetero-structures with proper geometry (shape function) can be tuned to have a suitable variation of effective stiffness, and thus allow for a desirable strain distribution to enable 8 programmable pseudomagnetic fields in the graphene portion under a uniaxial stretch. Consider a rectangular 2D hetero-structure with a graphene nanoribbon and two patches of another 2D crystal of effective stiffness 𝐸ℎ [e.g., graphane or h-BN, Fig. 2(a)]. Following a similar theoretical formulation as for the graphene nanoribbon shown above, it is shown that a programmable pseudomagnetic field in the graphene domain can be achieved by tailoring its geometry in the 2D hetero-structure. For example, a suitable shape function 𝑓(𝑦) of the two long edges of the graphene domain can be solved so that a uniaxial stretch in y direction can generate a uniform pseudomagnetic field in the graphene domain (see Section IV in [47] for details). To verify the above theoretical prediction, we carried out both finite element modeling and atomistic simulations of two types of 2D hetero-structures, graphene/graphane and graphene/h-BN, respectively, under uniaxial stretch, as shown in Fig. 2(a). The intensity of the resulting pseudomagnetic field in the graphene domain of a graphene/graphane and a graphene/h-BN hetero-structure, are shown in Fig. 2(b), respectively. Here the top/bottom width ratio of the graphene domain 𝑓r = 0.5, and the applied stretch is 15 %. A rather uniform distribution of the pseudomagnetic field is clearly evident, with an intensity of ≈33 T (graphene/graphane) and ≈22 T (graphene/h-BN), respectively, in good agreement with theoretical predictions. There exists a unique advantage of using a 2D hetero-structure over a pure graphene nanoribbon. It is shown that a stronger pseudomagnetic field can be generated in a graphene nanoribbon (or domain in 2D hetero-structure) with a smaller top/bottom width ratio 𝑓r, 9 with all other parameters kept the same (Eq. (S20) in [47]). To maximize such a tunability on field intensity, a graphene nanoribbon with 𝑓r = 0 (the narrower end shrinks to a point) is desirable, but applying uniaxial stretch to such a nanostructure becomes prohibitive given its sharp tip. By contrast, a tipped graphene domain in a 2D hetero-structure is feasible to fabricate and a uniaxial stretch can be readily applied to the rectangular 2D hetero-structure. Figure 2C demonstrates the resulting pseudomagnetic field in two types of such a hetero-structure, with an elevated average intensity of ≈70 T (graphene/graphane) and ≈45 T (graphene/h-BN), respectively, in comparison with those in Fig. 2(b) (𝑓r = 0.5, all other parameter being the same). Further atomistic simulations [Fig. S5] show good agreement with the above finite element modeling results in terms of both distribution and intensity of the resulting pseudomagnetic field. In conclusion, we offer a long-sought solution to achieving a programmable pseudomagnetic field in planar graphene over a large area via a feasible and effective strain-engineering mechanism. Our method utilizes a shape function applied to a planar graphene sheet to achieve a constant strain gradient when applying a simple uniaxial stretch to a graphene ribbon [Fig. 1(a)]. We demonstrate such a mechanism in both graphene nanoribbons and graphene-based 2D hetero-structures with resulting pseudomagnetic fields possessing a uniform distribution and a tunable intensity over a wide range of 0 T to 200 T. Such a programmable pseudomagnetic field under a uniaxial stretch results from the tunable effective stiffness of graphene by tailoring its geometry, so that the challenge of generating controllable strain gradient in graphene can be resolved by patterning the shape of a graphene nanoribbon or 10 the graphene domain in a 2D hetero-structure, a viable approach with the ever advancing 2D nanofabrication technologies. These feasible-to-implement approaches can yield rich rewards from systematic studies of pseudomagnetic fields in graphene, which are extreme fields compared to normal laboratory field strengths, and can be arbitrarily patterned in 2D. For example, a repeating programmable pseudomagnetic field can be generated in a wide range of structures over large areas by repeating the suitable geometrical patterns, e.g., a long graphene ribbon [Fig. 3(a)], a graphene nanomesh [Fig. 3(b)], and a graphene-based 2D superlattice structure [Fig. 3(c)]. The geometrical nature of the concept demonstrated in the present study is applicable to other 2D materials, and thus sheds light on fertile opportunities of strain engineering of a wide range of 2D materials for future investigations. Acknowledgements We would like to thank N. Zhitenev for valuable discussions. T.L. and Z.S. acknowledge the support by the National Science Foundation (Grant Numbers: 1069076 and 1129826). ZS thanks the support of the Clark School Future Faculty Program and a Graduate Dean’s Dissertation Fellowship at the University of Maryland. 11 [1] N. Levy, S. Burke, K. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. Neto, and M. Crommie, Science 329, 544 (2010). [2] N. N. Klimov, S. Jung, S. Zhu, T. Li, C. A. Wright, S. D. Solares, D. B. Newell, N. B. Zhitenev, and J. A. Stroscio, Science 336, 1557 (2012). [3] D. Guo, T. Kondo, T. Machida, K. Iwatake, S. Okada, and J. Nakamura, Nat. Commun. 3, 1068 (2012). [4] F. Guinea, M. Katsnelson, and A. Geim, Nat. Phys. 6, 30 (2010). [5] S. Zhu, Y. Huang, N. Klimov, D. Newell, N. Zhitenev, J. Stroscio, S. Solares, and T. Li, Phys. Rev. B 90, 075426 (2014). [6] K.-J. Kim, Y. M. Blanter, and K.-H. Ahn, Phys. Rev. B 84, 081401(R) (2011). [7] Z. Qi, A. Kitt, H. Park, V. Pereira, D. Campbell, and A. Neto, Phys. Rev. B 90, 125419 (2014). [8] M. Neek-Amal, L. Covaci, K. Shakouri, and F. Peeters, Phys. Rev. B 88, 115428 (2013). [9] M. Vozmediano, M. Katsnelson, and F. Guinea, Phys. Rep. 496, 109 (2010). [10] G. Verbiest, S. Brinker, and C. Stampfer, Phys. Rev. B 92, 075417 (2015). [11] J. Feng, W. Li, X. Qian, J. Qi, L. Qi, and J. Li, Nanoscale 4, 4883 (2012). [12] D. Bell, M. Lemme, L. Stern, J. RWilliams, and C. Marcus, Nanotechnology 20, 455301 (2009). [13] L. Ci, Z. Xu, L. Wang, W. Gao, F. Ding, K. Kelly, B. Yakobson, and P. Ajayan, Nano Res. 1, 116 (2008). [14] L. Ci, L. Song, D. Jariwala, A. Elias, W. Gao, M. Terrones, and P. Ajayan, Adv. Mater. 21, 4487 (2009). [15] M. Fischbein and M. Drndic, Appl. Phys. Lett. 93, 113107 (2008). [16] J. Bai, X. Zhong, S. Jiang, Y. Huang, and X. Duan, Nat. Nanotechnol. 5, 190 (2010). [17] A. Sinitskii and J. Tour, J. Am. Chem. Soc. 132, 14730 (2010). [18] Z. Zeng, X. Huang, Z. Yin, H. Li, Y. Chen, Q. Zhang, J. Ma, F. Boey, and H. Zhang, Adv. Mater. 24, 4138 (2012). [19] X. Liang, Y. Jung, S. Wu, A. Ismach, D. Olynick, S. Cabrini, and J. Bokor, Nano Lett. 10, 2454 (2010). [20] M. Wang, L. Fu, L. Gan, C. Zhang, M. Rummeli, A. Bachmatiuk, K. Huang, Y. Fang, and Z. Liu, Sci. Rep. 3, 1238 (2013). [21] S. Zhu, Y. Huang, and T. Li, Appl. Phys. Lett. 104, 173103 (2014). [22] L. Tapaszto, G. Dobrik, P. Lambin, and L. Biro, Nat. Nanotechnol. 3, 397 (2008). [23] Z. Sun et al., Nat. Commun. 2, 559 (2011). [24] D. Elias et al., Science 323, 610 (2009). [25] R. Balog et al., Nat. Mater. 9, 315 (2010). [26] P. Sessi, J. Guest, M. Bode, and N. Guisinger, Nano Lett. 9, 4343 (2009). [27] C. Reddy, Y. Zhang, and V. Shenoy, Nanotechnology 23, 165303 (2012). [28] S. Zhu and T. Li, J. Phys. D: Appl. Phys. 46, 075301 (2013). [29] S. Zhu and T. Li, ACS Nano 8, 2864 (2014). [30] L. Zhang and X. Wang, Phys. Chem. Chem. Phys. 16, 2981 (2014). 12 [31] J. Baringhaus et al., Nature 506, 349 (2014). [32] L. Liu et al., Science 343, 163 (2014). [33] Y. Gao et al., Nano Lett. 13, 3439 (2013). [34] G. Han et al., ACS Nano 7, 10129 (2013). [35] P. Sutter, R. Cortes, J. Lahiri, and E. Sutter, Nano Lett. 12, 4869 (2012). [36] H. Garza, E. Kievit, G. Schneider, and U. Staufer, Nano Lett. 14, 4107 (2014). [37] J. Soler, E. Artacho, J. Gale, A. Garcia, J. Junquera, P. Ordejon, and D. Sanchez-Portal, J. Phys.: Condens. Matter 14, 2745 (2002). [38] Y. Zhang, Y. Tan, H. Stormer, and P. Kim, Nature 438, 201 (2005). [39] Q. Peng, W. Ji, and S. De, Comput. Mater. Sci. 56, 11 (2012). [40] E. Cadelano, P. Palla, S. Giordano, and L. Colombo, Phys. Rev. B 82, 235414 (2010). [41] E. Munoz, A. Singh, M. Ribas, E. Penev, and B. Yakobson, Diamond Relat. Mater. 19, 368 (2010). [42] Q. Pei, Y. Zhang, and V. Shenoy, Carbon 48, 898 (2010). [43] S. Plimpton, J. Comput. Phys. 117, 1 (1995). [44] S. Stuart, A. Tutein, and J. Harrison, J. Chem. Phys. 112, 6472 (2000). [45] J. Tersoff, Phys. Rev. B 37, 6991 (1988). [46] A. Kinaci, J. Haskins, C. Sevik, and T. Cagin, Phys. Rev. B 86, 115410 (2012). [47] See Supplemental Material at [URL will be inserted by publisher] for a detailed discussion on computational method and theoretical calculation. Figure captions: FIG. 1. (color online). Producing uniform pseudomagnetic fields in a planar shaped graphene strip under a uniaxial stretch. (a) Schematic showing a graphene nanoribbon of varying width under a uniaxial stretch producing a pseudomagnetic field, Bps. The red circle denotes cyclotron orbits in the field giving rise to pseudo-Landau levels in (h). (b to d) Contour plots of the resulting strain components in the graphene, 𝜖𝑥𝑥, 𝜖𝑦𝑦 and 𝜖𝑥𝑦, respectively, under a 5 % uniaxial stretch. (e) Resulting pseudomagnetic fields in the graphene nanoribbon shown in (a) under a uniaxial stretch of 5 %, 10 % and 15 %, respectively. (f) Intensity of the pseudomagnetic field as the function of location along the centerline of the graphene ribbon for various applied uniaxial stretches. (g) Intensity of the pseudomagnetic field is shown to be linearly proportional to the applied uniaxial stretch and inversely proportional to the length of the graphene ribbon L. (h) Local density of states of unstrained graphene and graphene with a constant strain 13 gradient determined by density functional theory calculations. N=0 and N=±1, ±2, ±3 Landau levels, corresponding to cyclotron motion in a magnetic field are seen to emerge in the strained graphene, demonstrating a uniform pseudomagnetic field. The wiggles in the results for the unstrained case result from finite size effects in the calculations. See Supplemental Material for further discussion [47]. FIG. 2. (color online). Producing uniform pseudomagnetic fields in planar graphene-based hetero-structures under a uniaxial stretch. (a) Schematic showing a 2D hetero-structure consisting of graphene and graphane (or h-BN) bonded to a center piece of graphene under a uniaxial stretch. (b) Left: Intensity of the resulting pseudomagnetic field in the graphene domain of a graphene/graphane and a graphene/h-BN hetero-structure, respectively, under a 15 % uniaxial stretch. Here the top/bottom width ratio of the graphene domain 𝑓r = 0.5; Right: Contour plot of the resulting pseudomagnetic field in the graphene/graphane hetero-structure. (c) Left: Intensity of the resulting pseudomagnetic field in the graphene domain of a graphene/graphane and a graphene/h-BN hetero-structure, respectively, under a 15 % uniaxial stretch. Here the top/bottom width ratio of the graphene domain 𝑓r = 0; Right: Contour plot of the resulting pseudomagnetic field in the graphene/graphane hetero-structure. FIG. 3. (color online). Pseudomagnetic fields in patterned graphene hetero-structures supperlattices. (a) Schematic of a suitably patterned long graphene nanoribbon (left) and the contour plot of the resulting pseudomagnetic field under a 15 % uniaxial stretch (right). (b) Schematic of a suitably patterned graphene nanomesh (left) and the contour plot of the resulting pseudomagnetic field under a 15 % uniaxial stretch (right). (c) Schematic of a suitably patterned graphene-based 2D superlattice structure (left) and the contour plot of the resulting pseudomagnetic field under a 15 % uniaxial stretch (right). The scale for Bps is from – 200 T to 200 T. 14 (a) (b) (c) (d) Bps (cid:94)(cid:346)(cid:258)(cid:393)(cid:286)(cid:3)(cid:296)(cid:437)(cid:374)(cid:272)(cid:415)(cid:381)(cid:374) W(y) (e) Stretch 10% 5% 15% (f) (g) (h) 15% 10% 5% -3 -2 -1 N=0 2 3 1 15% 10% 5% FIG. 1. Producing uniform pseudomagnetic fields in a planar shaped graphene strip under a uniaxial stretch. (a) Schematic showing a graphene nanoribbon of varying width under a uniaxial stretch producing a pseudomagnetic field, Bps. The red circle denotes cyclotron orbits in the field giving rise to pseudo-Landau levels in (h). (b to d) Contour plots of the resulting strain components in the graphene, , respectively, under a 5 % uniaxial stretch. (e) Resulting pseudomagnetic fields in the graphene nanoribbon shown in (a) under a uniaxial stretch of 5 %, 10 % and 15 %, respectively. (f) Intensity of the pseudomagnetic field as the function of location along the centerline of the graphene ribbon for various applied uniaxial stretches. (g) Intensity of the pseudomagnetic field is shown to be linearly proportional to the applied uniaxial stretch and inversely proportional to the length of the graphene ribbon L. (h) Local density of states of unstrained graphene and graphene with a constant strain gradient determined by density functional theory calculations. N=0 and N=(cid:102)1, (cid:102)2, (cid:102)3 Landau levels, corresponding to cyclotron motion in a magnetic field are seen to emerge in the strained graphene, demonstrating a uniform pseudomagnetic field. The wiggles in the results for the unstrained case result from finite size effects in the calculations. See supplemental materials for further discussion [47]. and , (b) (c) Graphane/Graphene Graphane/Graphene (a) Hybrid sheet Graphane/Graphene OR Boron Nitride/Graphene FIG. 2. Producing uniform pseudomagnetic fields in planar graphene-based hetero-structures under a uniaxial stretch. (a) Schematic showing a 2D hetero-structure consisting of graphene and graphane (or h-BN) bonded to a center piece of graphene under a uniaxial stretch. (b) Left: Intensity of the resulting pseudomagnetic field in the graphene domain of a graphene/graphane and a graphene/h-BN hetero-structure, respectively, under a 15 % uniaxial stretch. Here the top/bottom width ratio of the graphene domain ; Right: Contour plot of the resulting pseudomagnetic field in the graphene/graphane hetero-structure. (c) Left: Intensity of the resulting pseudomagnetic field in the graphene domain of a graphene/graphane and a graphene/h-BN hetero-structure, respectively, under a 15 % uniaxial stretch. ; Right: Contour plot of the resulting pseudomagnetic Here the top/bottom width ratio of the graphene domain field in the graphene/graphane hetero-structure. (a) (b) (c) FIG. 3. Pseudomagnetic fields in patterned graphene hetero-structures supperlattices. (a) Schematic of a suitably patterned long graphene nanoribbon (left) and the contour plot of the resulting pseudomagnetic field under a 15 % uniaxial stretch (right). (b) Schematic of a suitably patterned graphene nanomesh (left) and the contour plot of the resulting pseudomagnetic field under a 15 % uniaxial stretch (right). (c) Schematic of a suitably patterned graphene- based 2D superlattice structure (left) and the contour plot of the resulting pseudomagnetic field under a 15 % uniaxial stretch (right). The scale for Bps is from – 200 T to 200 T. Supplemental Material Programmable Extreme Pseudomagnetic Fields in Graphene by a Uniaxial Stretch Shuze Zhu1, Joseph A. Stroscio2, Teng Li1,* 1Department of Mechanical Engineering, University of Maryland, College Park, MD 20742, USA 2Center for Nanoscale Science and Technology, NIST, Gaithersburg, MD 20899, USA Contents: I. Pseudomagnetic field under a uniaxial stretch II. Pseudo-Landau levels when 𝝏𝝐𝒚𝒚 𝝏𝒚 = constant III. Solving optimal shape function for a uniform pseudomagnetic field in a graphene nanoribbon under a uniaxial stretch IV. Solving optimal shape function for a uniform pseudomagnetic field in a graphene-based 2D hetero-structure under a uniaxial stretch V. Finite element modeling and comparison with theoretical prediction VI. Effect of top/bottom width ratio on pseudomagnetic field in a graphene nanoribbon VII. Atomistic simulations * To whom correspondence should be addressed. Email: lit@umd.edu S1 I. Pseudomagnetic field under a uniaxial stretch The constitutive law of a 2D elastic material correlates the stress 𝜎𝑖𝑗 and the strain 𝜖𝑖𝑗 in the form of 𝜎𝑥𝑥 = 𝐸 1−𝜈2 (𝜖𝑥𝑥 + 𝜈𝜖𝑦𝑦), 𝜎𝑦𝑦 = 𝐸 1−𝜈2 (𝜖𝑦𝑦 + 𝜈𝜖𝑥𝑥), and 𝜎𝑥𝑦 = 2𝐺𝜖𝑥𝑦, (S1) where 𝐸 is the Young’s modulus, 𝜈 is Poisson’s ratio and 𝐺 = 𝐸 2(1+𝜈) is the shear modulus of the material. Stress equilibrium requires 𝜕𝜎𝑥𝑥 𝜕𝑥 + 𝜕𝜎𝑦𝑥 𝜕𝑦 = 0, 𝜕𝜎𝑥𝑦 𝜕𝑥 + 𝜕𝜎𝑦𝑦 𝜕𝑦 = 0. (S2) Combining Eqs. (S1-S2) leads to 𝐸 1−𝜈2 ( 𝜕𝜖𝑥𝑥 𝜕𝑥 + 𝜈𝜕𝜖𝑦𝑦 𝜕𝑥 ) + 2𝐺 𝜕𝜖𝑥𝑦 𝜕𝑦 = 0, 𝐸 1−𝜈2 ( 𝜕𝜖𝑦𝑦 𝜕𝑦 + 𝜈𝜕𝜖𝑥𝑥 𝜕𝑦 ) + 2𝐺 𝜕𝜖𝑥𝑦 𝜕𝑥 = 0. (S3) When the graphene is subject to a uniaxial stretch along the armchair direction in its plane, 𝜖𝑥𝑥 = −𝜈𝜖𝑦𝑦, and Eq. (S3) becomes 𝜕𝜖𝑥𝑦 𝜕𝑦 = 0, 𝜕𝜖𝑥𝑦 𝜕𝑥 = −(1 + 𝜈) 𝜕𝜖𝑦𝑦 𝜕𝑦 . (S4) Substituting Eq. (S4) into Eq. (1) of the main text leads to Eq. (2), which is a main result for determining the requirements for generating a programmable uniform pseudomagnetic field. II. Pseudo-Landau levels when 𝝏𝝐𝒚𝒚 𝝏𝒚 = constant We calculate the local density of states (LDOS) using density functional theory (DFT) applied to a scaled down graphene nanoribbon. Pseudo-Landau levels appear in the LDOS due to the strain generated pseudomagnetic field. To compare the pseudofields to the results from finite element calculations, we point out that the DFT calculations will underestimate the pseudomagnetic field. As to be explained in details below, the molecular model for DFT calculations is subject to a finite constant strain gradient 𝜕𝜖𝑦𝑦 𝜕𝑦 , but 𝜕𝜖𝑥𝑦 𝜕𝑥 = 0 and 𝜕𝜖𝑥𝑥 𝜕𝑦 = 0. By contrast, in a graphene nanoribbon under uniaxial stretch, 𝜕𝜖𝑥𝑦 𝜕𝑥 and 𝜕𝜖𝑥𝑥 𝜕𝑦 are related to 𝜕𝜖𝑦𝑦 𝜕𝑦 (e.g., Eq. (S4)) and thus generally non-zero. Following the analysis in section I and comparing with Eq. (2), the resulting DFT generated pseudomagnetic field will then be given by, DFT = 𝐵ps 𝛽 𝛼 𝜕𝜖𝑦𝑦 𝜕𝑦 = 𝐵ps . 3(1+𝜈) (S5) Figure S1(a) shows the distribution of 𝜖𝑦𝑦 in a graphene nanoribbon [as in Fig. 1(a)] subject to a uniaxial applied stretch of 15 %, obtained from finite element simulations. The S2 bottom panel in Fig. S1(a) clearly shows a constant gradient of 𝜖𝑦𝑦 in the nanoribbon. We first consider a molecular model representing a local region (indicated by the boxed area in Fig. S1(a)) in the nanoribbon. Fig. S1(b) shows the atomistic details of the molecular model labeled with characteristic bond lengths and bond-stretching strains. The positions of the carbon atoms in this molecular model are prescribed so that the corresponding strain gradient of 𝜖𝑦𝑦 is 0.0036 nm−1, in accordance with strain gradient obtained from finite element simulations. Two such molecular models are patched head-to-head along their short edges with higher 𝜖𝑦𝑦 to define the molecular model for DFT calculations [Fig. S1(c)]. Periodic boundary conditions are applied along both horizontal and vertical directions of the DFT molecular model while a vacuum region of 10 nm is set along out-of-plane direction. As a result, such a model indeed represents an infinitely large graphene subject to alternating strain gradients along its armchair direction, which can effectively eliminate the artificial edge effects in the LDOS [4]. As a control calculation, we also construct the unstrained molecular model by relaxing all carbon-carbon bond length in Fig. S1(c) to that in pristine graphene [1.42 Å, Fig. S1(d)]. We perform first- principle DFT calculations in a supercell configuration by utilizing the SIESTA code [37]. The generalized gradient approximation (GGA) in the framework of Perdew-Burke-Ernzerhof (PBE) is adopted for the exchange-correlation potential. Numerical atomic orbitals with double zeta plus polarization (DZP) are used for basis set, with a plane-wave energy cutoff of 4080 eV (300 Ry). Self-consistent Field (SCF) tolerance is set to 10-6. A 160 × 5 × 1 Monkhorst-Pack k-point mesh is used for Brillouin zone integration in the strained model [Fig. S1(c)], while a 140 × 5 × 1 Monkhorst-Pack k-point mesh is used in the unstrained model [Fig. S1(d)], in order to ensure comparable k points separation. For LDOS calculations, the mesh along x and y directions is increased to 50 times of its initial size while the number of k points in out-of-plane direction is kept as one. For example, an 8000 × 250 × 1 mesh is used for the LDOS calculation of the strained model. The peak broadening width for LDOS calculation is 0.02 eV. The electronic smearing temperature during the calculation is 300 K. The LDOS of all carbon atom in the supercell for both strained and unstrained cases, and simulated Landau Levels are compared in Fig. 1(h). The appearance of additional peaks in the LDOS for the strained case is clearly shown, which is comparable to the pseudo-Landau Levels generated by a real magnetic field of 30 T. Figure S2 further shows the linear scaling relation between the DFT pseudo peak energies and the square root of the orbital index, N. S3  FIG. S1. (color online). (a) Distribution of 𝜖𝑦𝑦 in a graphene nanoribbon [as in Fig. 1(a)] subject to a uniaxial applied stretch of 15 %. The bottom panel clearly shows the linear distribution (i.e., constant gradient) of 𝜖𝑦𝑦 in the nanoribbon. (b) A molecular model within a local region in the nanoribbon (indicated by the boxed area in (a)). The box in (b) denotes the molecular model containing 32 carbon atoms. The lengths of characteristic carbon-carbon bonds are labeled and the corresponding bond- stretching strains are shaded using the same color scale as in (a). (c) The DFT model is made of two molecular models in (b) that are patched head-to-head along their short edges with higher 𝜖𝑦𝑦. The box denotes the supercell containing 64 carbon atoms and periodic boundary conditions are applied to the edges of the supercell. (d) The DFT model for the unstrained case of the molecular model in (c). S4 FIG. S2. (color online). Linear scaling between the pseudo peak energies 𝐸𝑁 from DFT results as shown in Fig. 1(h) and √𝑁, where 𝑁 is the peak index. The intensity of the pseudomagnetic field can be estimated by the energy spacing between pseudo Landau level peak positions from DFT calculations [1, 38] DFT = 𝐵ps ( 𝐸𝑁−𝐸Dirac sgn(𝑁) 2𝑒ħ𝑣F 2𝑁 2 ) (S6) From Fig. S2, the energy spacing between pseudo peak N=0 (𝐸Dirac) and N=1 (𝐸1) is DFT ≅ 33.63 T. Using Eq. (S5) gives 𝐵ps ≅ 118 T, which roughly ~0.21 eV, which gives 𝐵ps agrees with the finite element modeling result (≈150 T), as shown in Fig. 1(f). Since the pseudomagnetic field intensity predicted by DFT is corrected by a scaling factor to the elasticity- based theoretic prediction via Eq. (S5), we believe such a difference in the estimated pseudomagnetic field intensity essentially originates from the fact that the elasticity-based theoretic prediction modestly underestimates the field intensity in comparison with the finite element modeling results (see Section V for detailed discussion). Nonetheless, the above DFT calculations offers solid evidence attesting to the presence of a strain-induced pseudomagnetic field, as suggested by Eq. (2). III. Solving optimal shape function for a uniform pseudomagnetic field in a graphene nanoribbon under a uniaxial stretch Force balance along any cross-section cut of the graphene nanoribbon in Fig. 1(a) in y direction gives S5  𝐹 = 𝐸g𝑓(𝑦)𝑊0ℎ𝜖𝑦𝑦, (S7) where 𝐹 is the applied force at the ends of graphene nanoribbon necessary to generate the uniaxial tensile strain 𝜖app . 𝐸g and ℎ are the Young’s Modulus and thickness of graphene, respectively. Re-arrange Eq. (S7) and take derivative with respect to y, we get 𝜕𝜖𝑦𝑦 𝜕𝑦 = − 𝐹 𝐸g𝑊0ℎ 1 𝑓2 𝑑𝑓 𝑑𝑦 . (S8) For a uniform strain gradient 𝜕𝜖𝑦𝑦 𝜕𝑦 , Eq. (3) gives the governing equation of the optimal shape function, 1 𝑓2 𝑑𝑓 𝑑𝑦 = 𝐶, (S9) where C is a constant. Using the boundary conditions of 𝑓(0) = 1 and 𝑓(𝐿) = 𝑓r(≠ 0), the solution of the shape function is given by and 𝑓(𝑦) = 𝑓r 𝐿 𝑓r (𝐿 − 𝑦)+ 𝑦 , 𝐶 = 𝑓r−1 𝑓r 𝐿 . (S10) (S11) The unknown quantity 𝐹 in Eq. (S7) can be related to the global deformation by the loading- deformation condition where 𝐿 ∫ 𝜖𝑦𝑦 0 𝑑𝑦 = Δ𝐿 𝜖𝑦𝑦 = 𝐹 1 𝐸g𝑊0ℎ 𝑓(𝑦) (S12) (S13) and Δ𝐿 is the change in length of the graphene nanoribbon under uniaxial stretch so that 𝜖𝑎𝑝𝑝 = Δ𝐿/𝐿. Equations (S12) and (S13) lead to 𝐹 = 2 Δ𝐿 𝐸g 𝑓r 𝑊0ℎ (1 + 𝑓r)𝐿 Substituting Eqs. (S10) and (S14) into Eq. (4) gives (S14) S6  𝐵ps = 6𝑡𝛽 𝑒𝑣F 𝜖app 𝐿 (1−𝑓r) (1 + 𝑓r) (1 + 𝜈), (S15) IV. Solving the optimal shape function for a uniform pseudomagnetic field in a graphene- based 2D hetero-structure under a uniaxial stretch Force balance along any cross-section of the 2D hetero-structure gives so that 𝐸g𝑊0ℎ𝑓(𝑦)𝜖𝑦𝑦 + 𝐸h𝑊0ℎ(1 − 𝑓(𝑦))𝜖𝑦𝑦 = 𝐹, 𝜕𝜖𝑦𝑦 𝜕𝑦 = − 𝐹 𝐸g𝑊0ℎ (1− 𝐸h 𝐸g ) 𝑑𝑓 𝑑𝑦 ((1− 𝐸h 𝐸g )𝑓+ 𝐸h 𝐸g ) 2. Solving 𝜕𝜖𝑦𝑦 𝜕𝑦 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 gives (S16) (S17) 𝑓(𝑦) = − 𝐸h 𝐸g −((1−(𝑓r(1− 𝐸h 𝐸g )+ −1 ) ) 𝑦 𝐿 𝐸h 𝐸g −1 −1) 1− 𝐸h 𝐸g (S18) Following a similar strategy as in Section III, one gets and 𝜕𝜖𝑦𝑦 𝜕𝑦 = 2 𝜖app (1− 𝑓r)(1− 𝐿(1 + 𝑓r (1− 𝐸h 𝐸g )+ 𝐸h 𝐸g 𝐸h 𝐸g ) ) 𝐵ps = 6𝑡𝛽 𝑒𝑣F 𝜖app (1− 𝑓r)(1− 𝐿(1 + 𝑓r (1− 𝐸h 𝐸g )+ ) 𝐸h 𝐸g 𝐸h 𝐸g (1 + 𝑣). ) (S19) (S20) Equation (S20) suggests that the larger the difference in stiffness of the constituent materials in the 2D hetero-structure (i.e., smaller 𝐸h 𝐸g ), the more intensive the resulting pseudomagnetic field. Equation (S20) reduces to Eq. (6) when 𝐸h 𝐸g = 0 (i.e., a graphene nanoribbon). V. Finite element modeling and comparison with elasticity-based theoretical prediction Graphene, as well as graphane and h-BN, are modeled as linear elastic materials with Young’s Moduli of 1 TPa, 0.73 TPa, 0.82 TPa and Poisson’s ratios of 0.17, 0.08, 0.22, S7 respectively [39-42]. Large-strain quadrilateral shell elements (S4R), which allow for finite membrane strain, are used for modeling all materials. All material intrinsic thickness is set to 0.34 nm. In finite element models, the bottom edge of the graphene nanoribbon (or the 2D hetero-structure) is fixed in y direction. A displacement u in y direction is applied at the top edge, so that 𝜖𝑎𝑝𝑝𝑙 = 𝑢/𝐿. Both top edge and bottom edge are allowed to deform along x direction. The two long edges are free. The modeling is carried out using finite element method. The strain distribution obtained from finite element modeling is then plugged into Eq. (1) to calculate the intensity of the resulting pseudomagnetic field [Fig. 1, (e) to (g)], which is then compared with the field intensity theoretically predicted using Eq. (6). We design the loading scheme in which the edges are allowed to move in x direction rather than completely clamped edge condition out of two concerns: (1) If the two ends of the ribbon are completely clamped, significant shear strain rises near four corners of the ribbon due to the mechanical constraint from the clamp and the lateral contraction of the ribbon due to the Poisson’s ratio effect. Such significant shear strains at the corners are undesirable because they could cause complicate strain distribution in the center portion of the ribbon and thus compromise the uniformity of the resulting pseudomagnetic field. The loading scheme adopted in this work can avoid such an undesirable edge effect and allow for achieving rather uniform pseudomagnetic field in a large portion of the graphene ribbon, as shown in Fig. 1(e); (2) The suitably shaped graphene ribbon based on the optimal shape function (e.g., Fig. 1(a) and Fig. 2) can be viewed as a unit cell representation for the patterned graphene hetero-structures supperlattices (e.g., Fig. 3) to achieve uniform pseudomagnetic fields in a much larger structure. In practice, when such a superlattice structure with a sufficiently large length is uniaxially stretched, even though its two ends are clamped, the edge effect decays rapidly along the length directly, and majority portion of the superlattice structure away from the two clamped ends deform under the condition rather close to the loading scheme adopted in our approach. The comparison between the results from finite element simulations [Fig. 1, (e) to (f)] and those from elasticity-based theoretic prediction (Eq. (6)) shows that the theory modestly underestimates the intensity of resulting pseudomagnetic field (defined by the plateau value). This can be attributed to the assumption of a linear distribution of 𝜖𝑦𝑦 (or 𝜖𝑥𝑥) over the entire length and linear distribution of 𝜖𝑥𝑦over the entire width of the graphene nanoribbon in the theory. As shown in Fig. 1 (b-d), such an assumption holds for the graphene nanoribbon except at its four corners, leading to an effective length of the ribbon shorter than the entire length and thus a higher intensity of pseudomagnetic field. It is noted that in Fig. 1(f), as the applied stretch increases, the resulting pseudomagnetic field in the middle section of the graphene ribbon shows a slight deviation from a perfectly uniform plateau, which becomes more pronounced as the applied stretch further increases. Such a slight deviation can be understood by the nature of the theoretical derivation of the optimal shape function (e.g., Eq. (S10)). The optimal shape function is derived by assuming the ribbon is subject to a uniaxial stretch and then considering the force balance of the ribbon to achieve a uniform strain gradient. The resulting optimal shape function is independent of the magnitude of S8 the applied stretch. Such a function is then used to define the original shape of the ribbon simulated in the finite element modeling. As the applied stretch increases, the ribbon elongates in the loading direction, and due to Poisson’s ratio effect, also contracts in the lateral direction. As a result, the shape of the ribbon is no longer the same as described by the optimal shape function. Specifically, the narrower end of the ribbon contracts more than the wider end, which results in a slightly smaller effective value of fr, and thus in turn leads to a slightly higher magnitude of pseudomagnetic field according to Eq. (6). The above effect becomes more pronounced as the applied stretch increases, which is accountable for the slightly deviating trend of pseudomagnetic field as shown in Fig. 1(f). VI. Effect of top/bottom width ratio on pseudomagnetic field in a graphene nanoribbon Figure S3(a) shows three 25-nm long graphene nanoribbons with different top-bottom ratios 𝑓r = 0.35, 0.5, and 0.7. The basal width is 10 nm. Their two long edges are prescribed by the optimized shape function given in Eq. (S10). Figure S3(b) shows the finite element modeling results on the effect of top-bottom width ratio 𝑓r on the intensity of the resulting pseudomagnetic field. The smaller the top-bottom width ratio, the higher the strain gradient in the graphene, and thus the stronger the resulting pseudomagnetic field, which agrees well with our theoretical prediction (Eq. (S15)). FIG. S3. (color online). (a) The geometry of 25 nm long graphene nanoribbons of three top/bottom width ratio 𝑓𝑟 = 0.35, 0.5, and 0.7, respectively, with two their long edges of each nanoribbon prescribed by Eq. (S10). (b) The corresponding intensities of the resulting pseudomagnetic field from finite element simulations under a 5 % uniaxial stretch. VII. Atomistic simulations The atomistic simulations are carried out using Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) [43]. Figures S4(a) and S5(a) show the atomistic simulation models for the graphene nanoribbon and graphene-based 2D hetero-structure, respectively. Each model contains two graphene nanoribbons, the same as shown in Fig. 1(b) [or two 2D hetero- S9 structures as same as shown in Fig. 2(a)], which are covalently bonded along their wider ends in a mirroring fashion. Periodic boundary condition is applied along the vertical (loading) direction, therefore the atomistic model indeed represents a long graphene nanoribbon [e.g., Fig. 3(a)] or a large graphene-based 2D hetero-superlattice structure [e.g., Fig. 3(c)] with a repeating unit defined by Figs. S4(a) and S5(a), respectively. For simulations of graphene nanoribbons and graphene/graphane hetero-structures, the carbon-carbon (C-C) and carbon-hydrogen (C-H) bonds in the graphene as well as the non- bonded C-C and C-H interactions are described by the Adaptive intermolecular Reactive Empirical Bond Order (AIREBO) potential [44]. For simulations of graphene/h-BN hetero- structures, the atomic interactions are described by the Tersoff potential [45, 46]. The molecular mechanics simulations are carried out at zero K temperature. The loading is applied by gradually elongating the simulation box along vertical direction. At each loading step, the energy of the system is first minimized by using conjugate gradient algorithm until either the total energy change between successive iterations divided by the energy magnitude is less than or equal to 10- 8 or the total force is less than 10-5 eV/nm. The strain components, determined by Lagrange strain tensor in the deformed state, are used to calculate the resulting pseudomagnetic field. FIG. S4. (color online). (a) Atomistic simulation models for the graphene nanoribbon. (b)-(d) Atomistic simulations results on pseudomagnetic fields in the graphene nanoribbon under a uniaxial stretch of 5 %, 10 % and 15 %, respectively. S10 Figure S4 (b-d) plots the resulting pseudomagnetic field in the graphene nanoribbon under a uniaxial stretch of 5 %, 10 % and 15 %, respectively. The corresponding averaged intensity of the pseudomagnetic field near the central region of the top or bottom part is approximately 55 T, 100 T, and 125 T, respectively, in excellent agreement with the prediction from finite element modeling [Fig. 1(e)]. The ripple-like feature in the contour of pseudomagnetic field along the long edges coincides with the non-smooth and discrete nature of the long edges of the graphene nanoribbon to fit the shape function. FIG. S5. (color online). (a) Atomistic simulation models for a graphene/graphane 2D hetero-structure with straight material domain boundary. (b) Atomistic simulation results on pseudomagnetic fields in the graphene domain in (a) under a uniaxial stretch of 15 %. (c) Atomistic simulation models for the graphene/h-BN 2D hetero-structure with straight material domain boundary. (d) Atomistic simulation results on pseudomagnetic fields in the graphene domain in (c) under a uniaxial stretch of 15 %. Figure S5 shows the quasi-uniform pseudomagnetic field in graphene/graphane and graphene/h-BN 2D hetero-structures with straight domain boundary, under a uniaxial stretch of 15 %. S11 

1906.10902

1

1906

Quantum ammeter

We present the theoretical model of the "quantum ammeter", a device that is able to measure the full counting statistics of an electron current at quantum time scales. It consists of an Ohmic contact, perfectly coupled to chiral quantum Hall channels, and of a quantum dot attached to one of the outgoing channels. At energies small compared to its charging energy, the Ohmic contact fractionalizes each incoming electron and redistributes it between outgoing channels. By monitoring the resonant tunneling current through the quantum dot, one gets an access to the moment generator of the current in one of the incoming channels at time scales comparable to its correlation time.

cond-mat.mes-hall

cond-mat

Quantum ammeter Edvin G. Idrisov,1 Ivan P. Levkivskyi,2 and Eugene V. Sukhorukov3 1Physics and Materials Science Research Unit, University of Luxembourg, L-1511 Luxembourg, Luxembourg 2 Dropbox Ireland, One Park Place, Hatch Street Upper, Dublin, Ireland 3D´epartement de Physique Th´eorique, Universit´e de Gen`eve, CH-1211 Gen`eve 4, Switzerland (Dated: June 27, 2019) We present the theoretical model of the "quantum ammeter", a device that is able to measure the full counting statistics of an electron current at quantum time scales. It consists of an Ohmic contact, perfectly coupled to chiral quantum Hall channels, and of a quantum dot attached to one of the outgoing channels. At energies small compared to its charging energy, the Ohmic contact fractionalizes each incoming electron and redistributes it between outgoing channels. By monitoring the resonant tunneling current through the quantum dot, one gets an access to the moment generator of the current in one of the incoming channels at time scales comparable to its correlation time. PACS numbers: 73.21.-b, 73.43.-f, 73.43.Lp, 73.63.-b, 85.35.-p Since their discovery, the integer [1] and fractional [2] quantum Hall (QH) effect, admittedly being a very complex phenomena [3], for many years remained pre- dominantly a playground for various theoretical mod- els. For instance, although Lauglin quasi-particles with fractional charge [4] were successfully observed in exper- iments based on shot noise measurements [5, 6], their coherence and anyonic exchange statistics are still illu- sive. Only recently, we have witnessed a number of thor- ough experimental studies exploring different aspects of the integer QH effect physics, such as the quantum co- herence of the edge states [7], the energy relaxation and heat transport at the QH edge [8], and many others. A further progress in experimental techniques at nanoscale provided one with the remarkable control of the electronic quantum states in QH systems, thereby giving birth to the new field in the QH effect physics, dubbed the quantum electron optics [9]. In this field, ex- perimentalists combine different nanoscale systems, such as quantum point contacts (QPC), quantum dots (QD), QH edge channels, to study specific quantum phenomena. Between these basic elements of the electron quantum optics, an Ohmic contact, a piece of metal perfectly cou- pled to QH edge channels, was always considered merely a reservoir of equilibrium electrons [10], i.e., an analog of a black body in quantum optics. However, in contrast to photons electrons strongly interact. As a consequence, if the capacitance C of an Ohmic contact is relatively small, so that its charging energy Ec = e2/2C is not negligible, new interaction effects arise. The notable examples are the suppression of charge quantization caused by quan- tum fluctuations [11, 12], heat Coulomb blockade (CB) effect [13, 14], interaction induced recovery of the phase coherence [15 -- 17], charge Kondo effect [18, 19], quanti- zation of the anyonic heat flow [20], and the observation of the half-integer thermal Hall conductance [21]. In this Letter, we consider an Ohmic contact as an ideal electron fractionalizer. By this we mean that at ener- gies smaller than its charging energy Ec, the Ohmic con- FIG. 1: The quantum ammeter, a system that can be used to measure an electron current fluctuations (injected from the source) at quantum time scales, is schematically shown. It consists of an Ohmic contact (a small piece of metal con- taining the charge Q) perfectly coupled to QH edge states at integer filling factor and loaded by the resistance R, and of a QD attached to the outgoing channel. The Ohmic contact splits every incoming electron and transmits Tin = R/(R+Rq) part of it to the outgoing channel, where Rq is the quantum of resistance. The electron distribution function in the out- going channel f (ε) is measured by monitoring the resonant tunneling current through a QD, as in Ref. [8]. The function f (ε) contains the contribution from the moment generation function of the current jin injected in the incoming channel. The constant λin = 2πTin plays the role of a counting vari- able in the moment generator, which can be controlled by the the resistance R. We assume, that the load resistor does not create partition noise, which is the case where, e.g., it fully transmits N non-chiral modes, so that R = Rq/N . Other notations are explained in the text of the paper. tact simply splits every electron injected in the incoming channel so that only the fraction Tin ≤ 1 of the incoming electron charge e is transmitted to one of the outgoing channel, as shown in Fig. 1. This process, caused by the strong Coulomb interaction at the Ohmic contact, is fully deterministic and has to be distinguished from electron beam splitting, where the outgoing state is a su- perposition of the full (not fractionalized) electron in the outgoing channels. We furthermore propose to use this system in conjunc- arXiv:1906.10902v1 [cond-mat.mes-hall] 26 Jun 2019 Ohmic contact jc f(ε) ine T jout e jin source jR j'out R QD tion with a QD as a "quantum ammeter", i.e., a device that is able to detect the full counting statistics (FCS) of the fluctuations of the injected current at quantum time scales [22]. Namely, by measuring the resonant tun- neling current through a QD, on can record the electron distribution function f (ε) in the outgoing channel (as ex- perimentally demonstrated in Ref. [8]), which is directly related to the electron correlation function in the chan- nel, see Eq. (2). Every time an injected electron crosses the tunneling point, it adds the Friedel's phase shift λin = 2πTin to the electron correlation function, which af- ter time t acquires the overall contribution in the form of the moment generating function of the charge (7) injected to the incoming channel [23]. The parameter λin plays the role of the counting variable, which can be controlled by a load resistor R, ideally without adding extra parti- tion noise. In the following, we present the theory of the quantum ammeter and discuss its limitations. Through- out the paper, we set e =  = kB = 1, which also implies that the resistance quantum Rq = 2π/e2 = 2π. Model of the quantum ammeter. We follow Refs. [12, 13, 15, 18] and use the effective theory [24] and the bosonization technique [25] to describe QH edge states perfectly coupled to the Ohmic contact. Accordingly, the collective fluctuations of the charge densities ρα(x, t) and currents jα(x, t) in the QH edge channels are expressed in terms of the bosonic fields φα(x, t), ρα = (1/2π)∂xφα, jα = −(1/2π)∂tφα, (1) where the index α enumerates channels, entering and leaving the Ohmic contact (see Fig. 1). The electrons in the channels are represented by the vertex operators, ψα(x, t) ∝ eiφα(x,t), so that the electron distribution function in the outgoing channel reads: ∞Z−∞ dt 2π f (ε) = e−iεtK(t), K(t) ∝ he−iφout(t)eiφout(0)i. (2) Here, the prefactor in the expression for K(t) can be fixed in the end of calculations by comparing to the equilibrium correlation function for free electrons. Next, we apply the Langevin equation method [13, 26] to evaluate electron correlation function K(t). This method has been successfully used [11, 12] to describe experiments on the decay of CB oscillations [11, 14, 19]. According to this method, an Ohmic contact serves as a resorvoir of neutral excitations, which are accounted by introducing Langevin currents jc and jR in the equation of motion for the charge Q of the contact (see Fig.1). The details of the evaluations are presented in the sup- plementary material, while the results are rather intu- itive. At frequencies small compared to the charging en- ergy, ω (cid:28) Ec, the outgoing current is simply expressed in terms of the incoming current jin, and the Langevin 2 sources, jc and jR: jout(ω) =Xα Tαjα(ω), j0out(ω) =Xα T 0αjα(ω), (3) where α = (in, c, R), and the frequency-independent transmission coefficients take the following form − TR = Tin = 1 − Tc = R/(Rq + R), − T 0c = T 0in = 1 − T 0R = Rq/(Rq + R). (4a) (4b) By solving Eqs. (1-4), one arrives at the following expres- sion for the electron correlation function heiλαQα(t)e−iλαQα(0)i, (5) K(t) ∝ Yα=in,c,R where Qα(t) = R t −∞ jα(t0)dt0 is the total charge in the channel α, and λα = 2πTα. quire correlation functions in the form At zero bath temperature the Langevin sources jα ac- hjα(ω)jα(ω0)i = 2πSα(ω)δ(ω + ω0), α = c, R, Sc(ω) = + , SR(ω) = ω/Rq ω/R 1 − e−ω/Tc 1 − e−ω/Tc (6a) ωθ(ω) R , (6b) where Tc is the temperature of the Ohmic contact. The current jc is chiral, while jR is not, therefore the sec- ond term in the expression for SR is the contribution of the incoming states. These relations follow from the fluctuation-dissipation theorem for currents. Assuming for the moment a cold Ohmic contact, Tc = 0 (the heat- ing effects are studied below), and using Eqs. (1-5), we ar- rive at the following expression for the distribution func- tion df (ε) =Z ∞ dt 2π dε − G(λin, t) = (t + i0)T 2 −∞ e−iεtG(λin, t), inheiλinQin(t)e−iλinQin(0)i, (7a) (7b) where the correlation function G(λin, t), normalized to its ground-state part 1/(t + i0)T 2 in , can be considered the quantum FCS generator [22] of the injected non- equilibrium current jin, with λin = 2πTin playing the role of the counting variable [23]. In particular, if no current is injected in the ammeter, then G = 1 and df (ε)/dε = −δ(ε), as expected. Applications. In what follows, we consider examples of various systems injecting currents with different statistics into the incoming channel, as shown in Fig. (1). We start with the simplest example of the tunneling current through a barrier with the transparency D (cid:28) 1. At long times, t∆µ (cid:29) 1, where ∆µ is the applied potential difference, the fluctuations of charge can be considered classical, and their statistics is known to be Poissonian log(G) = (t∆µ/2π)D(eiλin − 1) for t > 0. Taking [22]: into account the fact, that all the odd (even) cumulants are odd (even) functions of time, one can write 3 (8a) (8b) [1 − cos(λin)]. ∆µD 2π γ (ε − ε0)2 + γ2 , 1 π df (ε) = dε ∆µD − ε0 = sin(λin), γ = 2π FIG. 2: Minus derivative of the distribution function in the outgoing channel (blue line), normalized to the voltage bias ∆µ, is plotted as a function of normalized energy for the load resistance R = Rq/2, and consequently, for λin = 2πR/(R + Rq) = 2π/3. Broadening of the distribution function, caused by the Gaussian noise in the incoming channel, created by the QPC with the transparency D = 1/2, is characterized by the effective temperature Tout in Eq. (11). It is found by comparing the energy flux in the channel, Jout = (Rq/2)hj2 ini, to the one for the equilibrium distribution (shown by the red line). Both distribution are shifted in energy by the value ∆µout = DTin∆µ = ∆µ/6. Interestingly, the equilibrium distribution appears to be broader, which can be attributed to the fact that it saturates faster at high energies, than the none-equilibrium one. the time integral in Eq. (2) numerically and show the re- sult in Fig. 2 for the transparency of the QPC D = 1/2 and for the load resistance R = Rq/2 (blue line). In order to characterize the non-equilibrium distribu- tion function in the outgoing channel, we compare it to the equilibrium one for the same effective temperature Tout, which is determined as follows. The electronic en- ergy flux of the equilibrium chiral channel reads dεε[f (ε)−θ(∆µout−ε)] = πT 2 out/12, (10) ∞Z−∞ q 1 R This result is justified in the limit D (cid:28) 1, since the inte- gral in Eq. (2) comes from long (Markovian) time scales. For example, if the load resistor transmits only one chan- nel, R = Rq, one has λin = π, and the broadening ac- quires the maximum value of γ = ∆µD/π, while the energy shift vanishes, ε0 = 0. Let us compare this result to the Gaussian noise case, where the functions in Eq. (8b) have to be expanded to second order in λin. Then, the energy shift ε0 = ∆µD/2 is simply induced by the average bias in the outgoing QH edge channel, and the broadening γ = π∆µD/4 is larger compared to the one for the non-Gaussian noise. Finally, we note, that the contributions to G from short time scales t ∼ 1/∆µ lead to the asymmetry of the tails of the peak in −df (ε)/dε, studied in Refs. [27, 28]. Next, we consider the classical noise induced by se- quential tunneling through a resonant level at zero tem- perature and with the in and out rate Γ1 and Γ2, re- spectively. The FCS generator of this process can be obtained by solving the generalized master equation [29] with the result G = Pm=1,2 GmeΛmt for t > 0, where Λm(λin) are the two eigenvalues of the transition matrix, and Gm(λin) are the corresponding weights (see the sup- plementary material). In the long time limit, one of the exponential function dominates, and the noise becomes Markovian with the known generator [29]. In the sym- metric case, Γ1 = Γ2 = Γ, the weights are real functions, Gm = [1 ± cos(λin/2)]/2, while eigenvalues take the sim- ple form: Λm = −Γ ± Γeiλin/2. Therefore, the function −df (ε)/dε acquires the double-peak structure, (11) 2 D(1 − D) (Tin∆µ)2 . 3 π T 2 out = which is nothing but the heat flux quantum. For the non- equilibrium channel, one can directly use the expression (7) with the definition of the energy flux (10), compare the results, and thus find the effective temperature of the Gaussian noise. Alternatively, one can use the expression Jout = (Rq/2)hj2 outi in terms of bosons (see, e.g., Ref. [13]), and compare it to the energy flux quantum (see the supplementary material). With both methods one obtains One can now use this expression for the effective tem- perature Tout and the effective bias ∆µout = DTin∆µ of the channel to plot the equilibrium distribution function where εm = ±Γ sin(λin/2) and γm = Γ[1 ∓ cos(λin/2)]. In particular, for R = Rq one has λin = π, and thus the function −df /dε shows two equal peaks with Gm = 1/2, γm = Γ, and εm = ±Γ. Finally, we consider the noise of a QPC at interme- diate transparencies D as an example of the quantum process. In this case, one may observe the noise induced phase transition, studied in Ref. [30], which originates from the singularity in the Markovian cumulant gener- ator at D = 1/2 and λin = π, i.e., R = Rq. Here, we concentrate on the Gaussian noise case (for the de- tails of calculations, see the supplementary material) and present the full time dependence of the cumulant gener- ator log G = it∆µDTin − 2D(1 − D)T 2 inF (t∆µ), where 0 dx(1 − x)x−2[1 − cos(t∆µx)]. We evaluate F (t∆µ) =R 1 Gmγm (ε − εm)2 + γ2 m , (9) Jout = Xm=1,2 1 π = df (ε) dε − 5 4 3 2 1 -0.4 0 -0.2 0.0 0.2 0.4 0.6 (see Fig. 2, red line). Interestingly, the equilibrium dis- tribution functions appears to be broader than the non- equilibrium one, which, we think, is compensated by its faster decay at large energies. Effects of heating. So far, we have neglected the effect of heating of the Ohmic contact assuming that its tem- perature vanishes, Tc = 0, so that the Langevin sources jc and jR contain only ground-state fluctuations. This assumption greatly simplifies the operation of the quan- tum ammeter, however, it is justified only if coupling to phonons is sufficiently strong, so that they are able to evacuate extra heat. Now, we relax this requirement and assume full thermalization inside the Ohmic contact, so the temperature Tc is to be found from the energy bal- ance equations (for the recent closely related analysis, which accounts for coupling to phonons, see Ref. [31]). The incoming energy flux is given by Jin = (Rq/2)hj2 ini (see, e.g., Ref. [13]), while the outgoing energy flux reads Jout = (Rq/2)hj2 outi. The total energy flux is conserved, Jin = Jout. By using Eqs. (3) and (4), we obtain Jin = (Rq/2)hj2 Ri. By substituting the correlators (6) to this equation, integrating over ω, and subtracting the ground-state contribution, we arrive at the following result: outi + (R/2)hj02 ci+(R2/2Rq)hj2 T 2 c = (12) 6R2 q π(Rq + R)(cid:2)hj2 ini − hj2 ini0(cid:3) , where hj2 ini0 is the ground-state part [32]. Finally, one can use the temperature Tc from this equation in order to evaluate the correlators (6), and eventually, the distri- bution function (2) with the help of Eqs. (3-5). The difficulty that arises in the last step is that for the situations of interest, R/Rq ∼ 1, where the coupling to the charged mode is not small, Tin ∼ 1, the neutral modes in of the equi- librium single boson channel at the temperature Tc to the outgoing state, which is not small, either. Therefore, the FCS generator in Eq. (7b) is modified as follows: contribute an arbitrary fraction q =p1 − T 2 G(λin, t) = G(λin, t)Kq(t, Tc), (13) where Kq(t, Tc) = he−iqφc(t)eiqφc(0)i is the equilibrium correlator of the fractional quasiparticle with the charge q ≤ 1 normalized to its ground state value, so that Kq(t, 0) = 1. This correlator smears out the distribution function (7a) introducing the effect that is difficult to ac- count analytically. Therefore, we suggest to calibrate the quantum ammeter by measuring −df /dε for a noiseless biased incoming channel, jin = hjini, and evaluating the Fourier transform to find the function Kq(t, Tc). Then, for the case of the non-zero noise, one can bias all the incoming channels in order to cancel the average current ini = hjini2. In this contribution in Eq. (12) to obtain hδj2 case, the function Kq remains the same, and one can sim- ply cancel it in the results of the measurements in order to obtain the FCS generator G(λin, t). 4 Discussion. First, we note, that in the limit R → ∞ (floating Ohmic contact) λin = 2π, as follows from Eq. (4a). Heating is also negligible, according to Eq. (12). Therefore G is given by Eq. (7b) and is equal to the correlation function of electrons in the incoming channel. Thus, in this case the Ohmic contact not only conserves the phase of incoming electrons in the outgoing channel, as has been experimentally demonstrated in Ref. [17], but should also conserve the electron distribution function at all energies smaller than Ec. Second, throughout the paper we assumed that incom- ing and outgoing channels do not interact with other QH channels, if they are present. This assumption is justi- fied, if the distance to the injection and detection point from the Ohmic contact is shorter than v/∆µ, where v is the velocity of the slowest neutral mode. In the oppo- site limit, one should take into account the charge frac- tionalization caused by the interaction between channels. At large distances one can neglect accumulated phases, which leads to the relatively simple modification of the scattering coefficients, Eq. (4). For instance, at the fill- ing factor ν = 2 the strong Coulomb interaction splits the spectrum of the edge plasmons into the neutral and charged mode, and equally distributes each injected elec- tron between two channels [33]. Finally, we have focused on the QH systems at integer filling factors, where the fermionic description is allowed. At fractional fillings the physics is more intricate. First of all, the spectrum of collective excitations at the edge is more complex, e.g., due to the emergence of upstream neutral mode [34]. Second, the statistics of injected cur- rents might be less simple at large transparencies. And what is more important, the physics of the Ohmic con- tact at general filling fractions is not yet well understood. However, at ν = 1/m, where m is the odd integer num- ber, edge of the QH system is known to support solely one charged mode. In this case, our model may still be applied by replacing Rq → mRq, and after some modifi- cation of the correlation functions [35]. To summarize, we have investigated an Ohmic con- tact, a small piece of metal perfectly coupled to chiral QH edge states, and demonstrated that at energies small compared to its charging energy Ec, it fractionalizes elec- tron states, thus only a fraction Tin of an incoming elec- tron is transmitted to one of the outgoing channels. We have shown that this system can serve as a "quantum ammeter" device: By monitoring the resonant tunneling current through a QD attached to the outgoing channel one can measure the generator of the FCS of the incoming current, with λin = 2πTin playing the role of the count- ing variable. After presenting several applications of this device, we have discussed how to calibrate the quantum ammeter to account for the effects of heating. We are grateful to F. Pierre for fruitful discussion. E. I. acknowledges financial support from the Fonds Na- tional de la Recherche Luxembourg under the grants AT- TRACT 7556175 and INTER 11223315. E. S. acknowl- edges financial support from the Swiss National Science Foundation. [1] K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [2] D. C. Tsui, H. L. Stormer, and A. C. Gossard, Phys. Rev. Lett. 48, 1559 (1982). [3] Z. F. Ezawa, Quantum Hall Effects: Recent Theoretical and Experimental Developments, (3rd ed., World Scien- tific, 2013). [4] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983). [5] R. de-Picciotto, M. Reznikov, M. Heiblum, V. Umansky, G. Bunin, and D. Mahalu, Nature 389, 162 (1997). [6] L. Saminadayar, D. C. Glattli, Y. Jin, and B. Etienne, Phys. Rev. Lett. 79, 2526 (1997). [7] Y. Ji, Y. Chung, D. Sprinzak, M. Heiblum, D. Mahalu, and H. Shtrikman, Nature 422, 415 (2003); E. Bieri, M. Weiss, O. Goktas, M. Hauser, S. Csonka, S. Oberholzer, C. Schonenberger, Phys. Rev. B 79, 245324 (2009); L. V. Litvin, H.-P. Tranitz, W. Wegscheider, and C. Strunk, Phys. Rev. B 75, 033315 (2007); I. Neder, F. Marquardt, M. Heiblum, D. Mahalu, and V. Umansky, Nat Phys 3, 534 (2007); P. Roulleau, F. Portier, D. C. Glattli, P. Roche, A. Cavanna, G. Faini, U. Gennser, and D. Mailly, Phys. Rev. B 76, 161309 (2007); E. Bieri, M. Weiss, O. Goktas, M. Hauser, S. Csonka, S. Oberholzer, C. Schonenberger, Phys. Rev. B 79, 245324 (2009); H. Duprez, E. Sivre, A. Anthore, A. Aassime, A. Cavanna, A. Ouerghi, U. Gennser, and F. Pierre, Phys. Rev. X 9, 021030 (2019). [8] C. Altimiras, H. le Sueur, U. Gennser, A. Cavanna, D. Mailly, and F. Pierre, Nature Physics 6, 34 (2010); H. le Sueur, C. Altimiras, U. Gennser, A. Cavanna, D. Mailly, and F. Pierre, Phys. Rev. Lett. 105, 056803 (2010); C. Altimiras, H. le Sueur, U. Gennser, A. Cavanna, D. Mailly, and F. Pierre, Phys. Rev. Lett. 105, 226804 (2010); K. Itoh, R. Nakazawa, T. Ota, M. Hashisaka, K. Muraki, and T. Fujisawa, Phys. Rev. Lett. 120, 197701 (2018). [9] For a review, see C. Grenier, R. Herv´e, Gwendal F´eve, and P. Degiovanni, Mod. Phys. Lett. B 25, 1053 (2011). [10] M. Buttiker, IBM J. Res. Develop. 32, 63 (1988). [11] S. Jezouin, Z. Iftikhar, A. Anthore, F. D. Parmentier, U. Gennser, A. Cavanna, A. Ouerghi, I. P. Levkivskyi, E. Idrisov, E. V. Sukhorukov, L. I. Glazman, and F. Pierre, Nature 536, 58 (2016). [12] E. G. Idrisov, I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. B 96, 155408 (2017). 5 [13] A. O. Slobodeniuk, I. P. Levkivskyi, and E. V. Sukho- rukov, Phys. Rev. B 88, 165307 (2013). [14] E. Sivre, A. Anthore, F. D. Parmentier, A. Cavanna, U. Gennser, A. Ouerghi, Y. Jin, and F. Pierre, Nature Physics 14, 145 (2018). [15] E. G. Idrisov, I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. Lett., 121, 026802 (2018). [16] A. A. Clerk, P. W. Brouwer, and V. Ambegaokar, Phys. Rev. Lett. 87, 186801 (2001). [17] H. Duprez, E. Sivre, A. Anthore, A. Aassime, A. Ca- vanna, U. Gennser, F. Pierre, arXiv:1902.07569. [18] A. Furusaki and K. A. Matveev, Phys. Rev. B 52, 16676 (1995). [19] Z. Iftikhar, A. Anthore, A. K. Mitchell, F. D. Parmen- tier, U. Gennser, A. Ouerghi, A. Cavanna, C. Mora, P. Simon, and F. Pierre, Science (2018): eaan5592, DOI: 10.1126/science.aan5592. [20] M. Banerjee, M. Heiblum, A. Rosenblatt, Y. Oreg, D. E. Feldman, A. Stern, and V. Umansky, Nature 545, 75 (2017). [21] M. Banerjee, M. Heiblum, V. Umansky, D. E. Feldman, Y. Oreg, and A. Stern, Nature 559, 210 (2018). [22] L. S. Levitov, H. Lee, and G. B. Lesovik, J. Math. Phys 37, 4845 (1996). [23] In the long time limit Qin(0) and Qin(t) commute, and G becomes the classical moment generating function of the injected charge ∆Qin(t) = Qin(t) − Qin(0), i.e., ∂m iλG(iλ)λ=0 = h(∆Qin)mi. [24] X. G. Wen, Phys. Rev. B 41, 12838 (1990). [25] T. Giamarchi, Quantum Physics in One Dimension (Claverdon Press Oxford, 2004). [26] E. V. Sukhorukov, Physica E 77, 191 (2016). [27] I. Chernii, I. P. Levkivskyi, E. V. Sukhorukov, Phys. Rev. B 90, 245123 (2014). [28] A. Borin, E. Sukhorukov, Phys. Rev. B 99, 085430 (2019). [29] M. J. M. de Jong, Phys. Rev. B 54, 8144 (1996). [30] I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. Lett. 103, 036801 (2009). [31] E. Sivre, H. Duprez, A. Anthore, A. Aassime, F.D. Par- mentier, A. Cavanna, A. Ouerghi, U. Gennser, and F. Pierre, unpublished. [32] Note, that according to Eq. (12) T 2 c can be used as a measure of instant (coincedent in time) fluctuations of the current. [33] I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. B 78, 045322 (2008). [34] A. H. MacDonald, Phys. Rev. Lett. 64, 220 (1990); M. D. Johnson and A. H. MacDonald, Phys. Rev. Lett. 67, 2060 (1991); C. L. Kane and M. P. A. Fisher, Phys. Rev. B 55, 15832 (1997). [35] Ivan P. Levkivskyi, Phys. Rev. B 93, 165427 (2016) Quantum ammeter. Supplementary material. Edvin G. Idrisov,1 Ivan P. Levkivskyi,2 and Eugene V. Sukhorukov3 1Physics and Materials Science Research Unit, University of Luxembourg, L-1511 Luxembourg, Luxembourg 2Dropbox Ireland, One Park Place, Hatch Street Upper, Dublin, Ireland 3D´epartement de Physique Th´eorique, Universit´e de Gen`eve, CH-1211 Gen`eve 4, Switzerland (Dated: June 27, 2019) LANGEVIN EQUATIONS AND ELECTRON DISTRIBUTION FUNCTION Here, we follow the Refs. [1 -- 3] and write the equations of motion for the currents and the charge Q in the Ohmic contact in the form (see Fig. 1 of the main text): Q(t) = jin(t) − jout(t) − j0out(t), jout(t) = Q(t) RqC + jc(t), j0out(t) = Q(t) RC + jR(t), (1) where the first equation expresses the conservation of charge, while the second and the third one are the Langevin equations, which have the following physical meaning: On the right hand side the contributions Q(t)/RqC and Q(t)/RC are the currents induced by the time-dependent potential Q(t)/C of the Ohmic contact, while the terms jc(t) and jR(t) are the Langevin current sources. After straightforward calculations, we present the outgoing currents in the following form jout(ω) = Xα=in,c,R Tα(ω)jα(ω), j0out(ω) = Xα=in,c,R T 0α(ω)jα(ω) (2) where TR(ω) = −Tin(ω) = Tc(ω) − 1 = 1/[iωRqC − Rq/R − 1], and T 0c (ω) = −T 0in(ω) = T 0R(ω) − 1 = (Rq/R)TR(ω). At frequencies small compared to the charging energy, ω (cid:28) Ec ∼ 1/RqC, the outgoing currents are given by Eq. (3) of the main text. Next, we use the Eq. (5) of the main text to obtain the following expression for the electron correlation function at zero temperature: K(t) ∝ (it + 0)−2RqT 2 c Gc−2RqT 2 RGRheiλinQin(t)e−λinQin(0)i. (3) We note, that the coefficients Tα satisfy the summation rule: Pα Tα2Gα = 1/2Rq, where Gin = Gc = 1/2Rq, GR = 1/R. Taking this relation into account and using Eq. (2), we straightforwardly get Eqs. (7) of main text, namely df (ε) dε − =Z dt 2π e−iεtG(λin, t), where G(λin, t) ∝ (it + 0)T 2 inheiλinQin(t)e−λinQin(0)i. (4) FULL COUNTING STATISTICS OF SEQUENTIAL TUNNELING According to Ref. [4], the FCS generator of sequential electron tunneling through a single level in a QD can be obtained by solving the generalized master equation. Here, we briefly derive it by considering the evolution of the joint probability distribution function Pl(n, t), where l = 1, 2 enumerates states of the empty and occupied level, and n is the number of electrons in the right reservoir (see Fig. 1). At zero temperature, the probability distribution function satisfies the infinite set of master equations: ∂tP1(n, t) = −Γ1P1(n, t) + Γ2P2(n − 1, t), ∂tP2(n, t) = −Γ2P2(n, t) + Γ1P1(n, t). Then, the FCS generator can formally be written as: G(λin, t) = Xl=1,2 gl(λin, t), where gl(λin, t) = Xn,l0,n0 Pl(n, t)P (0) l0 (n0)eiλin(n−n0), (5) (6) with the initial condition Pl(n, 0) = δl,l0δn,n0, while P (0) l (n) is the stationary distribution of master equations (5). arXiv:1906.10902v1 [cond-mat.mes-hall] 26 Jun 2019 2 FIG. 1: Sequential tunneling via the resonant level ε0 in a double-barrier structure, e.g., in a QD, is schematically shown. At strong biases, EF + ∆µ > ε0 > EF and ∆µ (cid:29) Γ1,2, the transport of electrons arises due to tunneling from the left to the right reservoir through two barriers with the rates Γ1 and Γ2. We have intentionally introduced the moment generator gl(λin, t) in the mixed representation, because it satisfies simple generalized master equations ∂tg1(λin, t) = −Γ1g1(λin, t) + Γ2eiλin g2(λin, t), ∂tg2(λin, t) = −Γ2g2(λin, t) + Γ1g1(λin, t), as follows from its definition and from the master equations (5). The formal solution of these equations reads: gl(λin, t) = Xm=1,2 cmvmleiΛmt, t > 0, (7) (8) where vml and Λm are the eigenvectors and eigenvalues of the matrix of the coefficients in Eq. (7), respectively, and ck are arbitrary constants. These coefficients can be found from the initial condition for gl(λin, t), which follows from Eq. (6), that it is equal to the stationary occupations of the QD level: g1 = Γ2/(Γ1 + Γ2) and g2 = Γ1/(Γ1 + Γ2). Taking the final trace, G(λin, t) = Pl=1,2 gl(λin, t), after straightforward calculations we arrive at the following result G(λin, t) = Xm=1,2 Gm(λin)eΛm(λin)t, t > 0, (9) where the "weights" and eigenvalues are given by G1,2(λin) = [1 ± F (λin)]/2 and Λ1,2(λin) = −[Γ1 + Γ2 ∓ β(λin)]/2, respectively, with Finally, using Eqs. (7) of the main text the electron distribution function can be presented in the following form F (λin) = β(λin) Γ1 + Γ2 − 2Γ1Γ2(eiλin − 1) β(λin)[Γ1 + Γ2] , β(λin) =q(Γ1 + Γ2)2 + 4Γ1Γ2(eiλin − 1). Re(cid:20) 1 e−iεtG(λin, t)(cid:21) = Xm=1,2 dt π π = Re(cid:20)Z ∞ 0 df (ε) dε − (10) (11) γm + i(ε − εm)(cid:21) , Gm where εm = Im[Λm(λin)] and γm = −Re[Λm(λin)] > 0. For the case of the symmetric system, Γ1 = Γ2 = Γ, these expressions take the simple form: Λ1,2 = −Γ ± Γeiλin/2 and G1,2 = [1 ± cos(λin/2)]/2. BROADENING OF THE ELECTRON DISTRIBUTION FUNCTION BY A GAUSSIAN NOISE Typically, the effective noise temperature is defined by comparing the zero-frequency noise power to its equilibrium value given by the fluctuation-dissipation relation. However, in order to compare equilibrium and non-equilibrium distribution functions, it is more appropriate to define the effective noise temperature via the energy fluxes of two distributions (for the details, see Ref. [1]), which characterize their broadening. The total energy flux in the outgoing channel can be written as Jout = (Rq/2)[hj2 outi − hj2 outi0] = (1/2)Z dω[Sout(ω) − S0(ω)], (12) Γ1 Γ2 EF+Δµ ε0 EF 3 FIG. 2: Minus derivative of the distribution function (blue lines), normalized to the voltage bias ∆µ, is plotted as a function of normalized energy for different values of the load resistance R [consequently, for different coupling constants λin = 2πR/(R + Rq)]. Broadening of the distribution function, caused by the Gaussian noise of the QPC with the transparency D = 1/2, is characterized by the effective temperature in Eq. (18), evaluated by comparing the energy flux in the channel to the one for the equilibrium distribution (red lines). where Sout(ω) =Z dteiωthjout(t)jout(0)i (13) is the noise power of the outgoing current, and S0(ω) is its ground-state value. For the equilibrium chiral channel with the temperature Tout the fluctuation-dissipation relation reads: Sout(ω) = ω/Rq 1 − e−ω/Tout , and S0(ω) = ωθ(ω)/Rq, (14) (15) Xα=in,c,RZ dω(cid:2)T 2 α Sα(ω) − S0(ω)(cid:3) = (T 2 in/2)Z dω [Sin(ω) − S0(ω)] , 1 2 Jout = thus the equilibrium channel carries the energy flux Jout = πT 2 In the case of a Gaussian non-equilibrium noise, one can use Eq. (3) of the main text and the propertyPα T 2 1/2Rq, where Gin = Gc = 1/2Rq and GR = 1/R, to write out/12, which is nothing but the flux quantum. α Gα = where Sin(ω) is the noise power in the incoming channel. By applying the scattering theory [5], one obtains for the voltage biased QPC at zero temperature: Sin(ω) = S0(ω) + D(1 − D) ×"Xσ=± S0(ω + σ∆µ) − 2S0(ω)# . Substituting this expression into Eq. (15), one easily obtains Jout = 1 4π D(1 − D) (Tin∆µ)2 (16) (17) ∆µout = DTin∆µ. (19) (18) 2 D(1 − D) (Tin∆µ)2 . 3 π T 2 out = Interestingly, for Tin = 1 this expression agrees with the electron energy flux evaluated right after the QPC. Thus, the suppression of high-order cumulants conserves the energy flux. out/12 to defind the effective temperature of the Finally, we compare the result (17) to the flux quantum Jout = πT 2 channel Note, that only the fraction DTin of electrons from the reservoir reaches the outgoing channel. Thus, the effective bias in this channel reads: R=Rq 5 4 3 2 1 R=Rq/2 8 6 4 2 R=Rq/3 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 ε/Δµ -0.4 0 -0.2 0.0 0.2 ε/Δµ 0.4 0.6 -0.2 0 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 ε/Δµ 2.5 2.0 1.5 1.0 0.5 0.0 -Δµdf/dε In the next step, we use Eqs. (7) of the main text, the Gaussian character of the noise (expanding the correlator in Eqs. (7) to second order in λin, averaging, and re-exponentiating the result), and Eq. (16), to obtain 4 ∂f ∂ε − =Z dt 2π e−iεtGin(λin, t), log G = it∆µDTin − 2D(1 − D)T 2 inF (t∆µ), (20) where F (t∆µ) =R 1 0 dx(1 − x)x−2[1 − cos(t∆µx)] is the dimensionless function, and we set λin = 2πTin. We are now in the position to compare the equilibrium and non-equilibrium distribution function. For doing so, we evaluate the integral in Eq. (20) numerically and present the results in Fig. 2 together with the equilibrium distributions with the temperature Tout and the bias ∆µout from Eqs. (18) and (19) for different values of λin. The apparent difference in broadening of equilibrium and non-equilibrium distribution functions (with the same effective temperatures) can, most likely, be attributed to the fact, that the non-equilibrium distribution decays slower at high energies. This effect is more pronounced for smaller values of λin, because at λin (cid:28) 1 the function −df /dε acquires power-law tails at ε (cid:29) ∆µ. [1] A. O. Slobodeniuk, I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. B 88, 165307 (2013). [2] E. G. Idrisov, I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. B 96, 155408 (2017). [3] E. G. Idrisov, I. P. Levkivskyi, and E. V. Sukhorukov, Phys. Rev. Lett., 121, 026802 (2018). [4] M. J. M. de Jong, Phys. Rev. B 54, 8144 (1996). [5] Ya. M. Blanter, M. Buttiker, Phys. Rep. 336, 1 (2000). 

0907.0182

2

0907

Control of Spin Blockade by AC Magnetic Fields in Triple Quantum Dots

We analyze coherent spin phenomena in triple quantum dots in triangular configuration under crossed DC and AC magnetic fields. In particular, we discuss the interplay between Aharonov-Bohm current oscillations, coherent electron trapping and spin blockade under electron spin resonance conditions. We demonstrate that, for certain field frequencies, AC magnetic fields induce an antiresonant behavior in the current, allowing for both removal and restoration of entangled spin blockaded states by tuning the AC field frequency. Our theoretical predictions indicate how to manipulate spin qubits in a triangular quantum dot array.

cond-mat.mes-hall

cond-mat

Control of Spin Blockade by ac Magnetic Fields in Triple Quantum Dots 1Instituto de Ciencia de Materiales de Madrid, CSIC, Cantoblanco, 28049, Madrid, Spain Maria Busl1, Rafael S´anchez2 and Gloria Platero1 2 D´epartement de Physique Th´eorique, Universit´e de Gen`eve, CH-1211 Gen`eve 4, Switzerland We analyze coherent spin phenomena in triple quantum dots in triangular configuration under crossed dc and ac magnetic fields. In particular, we discuss the interplay between Aharonov-Bohm current oscillations, coherent electron trapping and spin blockade under two-electron spin resonance configurations. We demonstrate an unexpected antiresonant behaviour in the current, allowing for both removal and restoration of maximally entangled spin blockaded states by tuning the ac field frequency. Our theoretical predictions indicate how to manipulate spin qubits in a triangular quantum dot array. PACS numbers: Electronic transport through mesoscopic systems can become correlated not only by charge interaction but also by the spin degree of freedom. A dramatic combination of both can be found in systems where strong Coulomb interaction limits the population to a small number of electrons (Coulomb blockade) and where Pauli exclusion principle avoids certain internal transitions -- spin block- ade (SB). This was first observed as a rectification effect in the current through a double quantum dot (DQD)1. Recent experiments have taken advantage of SB to achieve qubit operations in a double dot by electric gate control2 or by electron spin resonance (ESR)3. It con- sists in inducing transitions between the electron's spin- up and spin-down states, which are splitted by the Zee- man energy coming from a dc magnetic field, Bdc. Differ- ent mechanisms have been considered: crossed dc and ac magnetic fields (Bac), where the ac frequency is resonant with the Zeeman splitting4,5, effective Bac induced by ac electric fields in the presence of spin-orbit interaction6, slanting Zeeman fields7 or hyperfine interaction8. Lately, a next step towards quantum dot arrays has been reached: tunnel spectroscopy measurements with triple quantum dots (TQD), both in series9 and in trian- gular configurations10 have been achieved. Theoretical works in these systems11,12 analyze their eigenstates and stability diagram, as well as the effect of a magnetic field penetrating the structure. TQDs with strongly corre- lated electrons have also been investigated in the Kondo regime13 and have been proposed as spin entanglers14. Additionally, these systems show a more peculiar prop- erty which is intrinsic to three-level systems, namely co- herent population trapping, which is a well-known ef- fect in quantum optics and which was observed in three- level atoms excited by two resonant laser fields15. There, the electronic wave function evolves towards an eigen- state superposition, a so-called dark state, which is de- coupled from the laser fields and therefore it manifests as an antiresonance in the emission spectrum. An anal- ogy in transport has been made when coherent superpo- sitions avoid transport by interference between tunnel- ing events. These dark states can be achieved by driv- FIG. 1: (Color online) Coherent processes in a triangular TQD, with one electron confined in the dot connected to the drain. ∆1 = ∆2 6= ∆3, where ∆i is the Zeeman splitting in dot i. Transport through the system depends on the mag- netic flux Φ penetrating the system and on the frequency ω of the time-dependent magnetic field Bac. The shaded regions indicate the existence of a dark state. ing three-level double dots with bichromatic ac electric fields16 or by the interference of tunneling processes in TQDs17,18. It was shown19 how coherent trapping can be lifted in closed-loop TQDs by means of the Aharonov- Bohm (A-B) effect20. Here, we will discuss the electron spin dynamics and transport for the case where a triangular TQD contains up to two extra electrons, as shown in Fig. 1. In contrast to the single electron case, spin correlations can influence transport due to SB21. We will show that at certain Bac frequencies and sample configurations, the magnetic field brings the electronic wave function into a superposition of parallel spins states, unexpectedly bringing the system back to SB. Model -- We consider a system consisting of three dots lkσ which are coupled through tunnel barriers, and dots 1 and 3 are also connected to source and drain contacts respectively. The Hamiltonian of the system is H(t) = HTQD+ Hτ + HT+ Hleads+ HB(t). HTQD = Piσ ǫic† iσ ciσ + Pi Ui ni ni + Pi,j,i6=j Vij ni nj describes the uncoupled TQD, Hτ = −Pijσ(τij c† iσ cjσ + h.c.), the coherent tun- neling between the dots, HT = Pl∈L,Rkσ(γl d† lkσ clσ +h.c.) describes the coupling of the dots to the leads, and Hleads = Plkσ ǫlk d† dlkσ the leads themselves. ǫi is the energy of an electron located in dot i, Ui is the intra- dot and Vij = V the inter-dot Coulomb repulsion. If not stated otherwise, we set τij = τ . The Hamiltonian for the magnetic field, HB(t), has two components: a time-independent dc component along the z-axis that breaks the spin-degeneracy by a Zeeman-splitting ∆i = giBzi, and a circularly polar- ized ac component in the xy-plane that rotates the z-component of the electron spin when its frequency fulfills the resonance condition ωi = ∆i. It reads HB(t) = P3 i=1[∆i Szi + Bac(cos(ωt) Sxi + sin(ωt) Syi)], be- 2 Pσσ′ c† ing Si = 1 iσσσσ′ ciσ′ the spin operator of the −ith dot. As shown in5, the ac magnetic field has no effect on SB unless the Zeeman splitting is inhomogeneous in the sample. Here we will consider the simplest config- uration that allows to analyze the relevant mechanisms: ∆1 = ∆2 6= ∆3. The experimental feasibility7,22 justifies the choice of the present configuration. The dynamics of the system is given by the time evo- lution of the reduced density matrix whose equations of motion read as, within the Born-Markov-approximation: ρln(t) = −i hl[ HTQD + Hτ + HB(t), ρ]ni (1) (Γnkρkk − Γknρnn)δln − Λlnρln(1 − δln) + X k6=n 2 Pk(Γkl + Γkn). The commutator accounts for the coherent dynamics in the TQD, Γln are the transition rates from state ni to state li induced by the coupling to the leads and deco- herence appears by Λln = 1 We consider a configuration where the dot coupled to the drain is permanently occupied by one electron (see Fig. 1), and only up to two electrons can be in the sys- tem. Double occupancy is only allowed in the drain dot. This is the case when the chemical potentials in the leads satisfy ǫ3 +V < µR < ǫ3 +U3 and µL < ǫ1 +2V . For reso- nant tunneling, ǫ1 = ǫ2 and ǫ1,2 +V = ǫ3 +U3. Out of the full TQD basis with up to two electrons, there are then eleven one- and two-electron states that dominate the dy- 0, 0, σi, χ1σσ′i = σ, 0, σ′i, χ2σσ′i = 0, σ, σ′i namics: and S3i = 0, 0,↑↓i, with σ, σ′ = {↑,↓}. Transport is biased from left to right and only state S3i contributes to tunneling through to the drain, acting as a bottleneck for the current: I(t) = Pn ΓnS3ρS3S3 (t). Though be- ing confined, the electron in dot 3 is essential to induce spin correlated transport. A Bdc perpendicular to the plane of the triangular dot structure (Fig. 1) encloses a magnetic flux Φ such that electron tunneling acquires an 2 additional phase φ = 2πΦ/Φ0, with Φ0 = h/e being the flux quantum. We accumulate the phase between dot 1 and dot 2, τ12 = τ e−iφ. Undriven case (Bac = 0): -- It is well known19 for a TQD with up to one extra electron, that due to interfer- ence, the current oscillates with Φ (A-B oscillations) and periodically drops to zero with a periodicity of Φ0/2. For the understanding of the two-electron spin dynamics, it is crucial to look at the eigenstates of this system, which change depending on the flux Φ. For Φ/Φ0 = n 2 ψ− σσ′i = ψ+ σσi = 1 √2 1 √2 (χ2σσ′i − χ1σσ′i) (χ2σσi + χ1σσi) σ, σ′ = {↑,↓} σ = {↑,↓} (2) (3) are eigenstates of the closed system. States (2) avoid σσ′ HτS3i = 0, which is why they tunneling to S3i: hψ− are also called dark states, see Fig. 1. Occupation of S3i thus decays by the coupling to the drain (Fig. 2b) and current is blocked. The states in (2) remind of the dark states found in the single electron case17. A significant difference is that for two electrons the spin degree of free- dom plays a role: Pauli exclusion principle introduces spin correlation such that dark states ψ− σσ′i with σ = σ′ are avoided. The electrons are rather being trapped in combinations of dark states ψ− σσ′i with σ 6= σ′ and spin- blockaded states ψ+ σσi. Thus, SB competes with coher- ent population trapping in the blocking of the current, and the relative occupation of ψ− σσ′i (σ 6= σ′), and ψ+ σσi depends on the initial condition. If however Φ = Φ0/4, the A-B phase removes the dark state and only eigen- states with parallel spins are decoupled from S3i: ξ± σσi = 1 √2 (χ2σσi ± ıχ1σσi) σ = {↑,↓} (4) Coherent trapping is hence lifted, however, transport is still cancelled by SB (Figs. 1, 2a). One can appreciate that without Bac, the system is always blocked for trans- port -- the stationary current is insensitive to A-B effect due to SB. Driven case (Bac 6= 0): -- In order to remove SB, we apply a time-dependent Bac. Fig. 2c shows the I-Φ char- acteristics of the TQD excited by Bac, where for every value of Φ, the magnetic field frequency fulfills the reso- nance condition ω = ∆1,2. For Φ/Φ0 6= n/2, dark states are avoided by A-B effect, and Bac enables transitions of the form χlσσi → χlσ′σi → S3i that produce a finite current. It can be shown that Bac does not affect the destructive tunneling interference of the superpositions (2). Then, if Φ/Φ0 = n/2 the system evolves towards a state which is only composed of dark states performing spin rota- tions: ψ− σσi, as shown schematically in Fig. 1. Since the dark states are decoupled from transport, the oscillations can only be affected by decoherence due to spin scattering processes, which are not considered here. Hence, a Bac induces current through the system only σσ′i ↔ ψ− a) Φ/Φ0 = 0.25 Bac = 0 SB S3〉 Time [2π/Ωτ] AC = ∆1,2 ω c) 0 8 Bac ≠ 0 Bac = 0 b) Φ/Φ0 = 0.5 Bac = 0 dark state S3〉 Time [2π/Ωτ] SB 8 0.06 a) Γ / I ∆3 0 0.06 c) ∗ Γ / I Bac = τ/3 Bac = τ/2 Bac = τ ∆1 ω0ω Γ = τ Γ = 1.5τ Γ∗ = 2τ 3 τ12 = 2τ; τ23 = 4τ; τ13 = τ τ13 = 0; τ12,23 = τ (linear TQD) τ12,23 = 0; τ13 = τ (DQD) ∆3 ω0ω ∆1 2ESR 0.07 b) Γ / I 0 0.03 d) Γ / I SB SB 0.25 0.5 Φ/Φ0 0.75 1 0 ∆3 ω0ω ∆1 0 0 0.25 Φ/Φ0 0.5 0.5 y t i l i b a b o r p . c c O 0 0 0.03 Γ / I 0 0 FIG. 2: (Color online) (a) ρii for Φ/Φ0 = 0.25, Bac = 0. I is blocked due to SB, once the parallel spin states (dashed and dotted red lines) are occupied. (b) ρii for Φ/Φ0 = 0.5, Bac = 0. Due to SB, there is a finite occupation of parallel spin states χlσσi (dashed and dotted red lines), while electrons with antiparallel spin form dark states as in (2) (solid and dashed-dotted blue lines), all of them contributing to quench the current. (c) I-Φ: for Bac = 0, I = 0 due to SB (dashed blue line); Bac 6= 0: for ωac = ∆1,2, SB is removed and the current shows A-B-like oscillations (solid purple line). Rabi frequency: Ωτ = 2τ . τ = 0.0026, Γ = 0.01 in meV, ∆3 = 0.77∆1, ∆1 = ∆2. FIG. 3: 2ESR in a TQD by tuning ω, for Φ/Φ0 = 0.25, is manifested in the current as an antiresonance at ω0 = (∆1 + ∆3)/2. (a) For different Bac and fixed τij = τ : The width of the antiresonance depends on the Rabi frequencies associated with τ and Bac. (b) For different τij and fixed Bac = τ /2. The antiresonance appears for any configuration of the τij . (c) For different Γ and τij = τ . (d) I-Φ for fixed ω while tuning Bdc, so both Φ and ∆i are modified. A-B oscillations are suppressed by SB except when the Zeeman splittings are close to resonance with ω. At Φ/Φ0 = 0.25, ω = ω0 and current vanishes. Parameters: τ = 0.005, Γ = Γ∗ = 0.01. when assisted by the A-B lifting of dark states -- i.e. for Φ/Φ0 6= n/2 (Fig. 2c). Remarkably, not imposing the resonance condition ω = ∆i, one can find a novel kind of SB induced by Bac, quenching the current even in the presence of A-B effect. This is the main result of our work. As can be seen in Fig. 3a, the current shows a resonant behaviour as the frequency of Bac approaches the ESR condition (i.e. ω ∼ ∆1, ∆3). Surprisingly though, an antiresonance ap- pears for ω0 = (∆1 + ∆3)/2, i.e. when the two electrons are equally far from the ESR condition. Note that the two peaks around the antiresonance are not Lorentzian- like and cannot be identified as two different resonance peaks centered at the conditions ω = ∆1 = ∆2 and ω = ∆3, but as a collective effect due to the simultane- ous rotation of the two electron spins (2ESR), cf. Fig 1. We want to stress that the appearance of the antireso- nance does not depend on the field intensity Bac or tun- nel couplings τij (see Fig. 3a,b): it occurs for different τij as well as for linear TQD configurations (setting τ13 = 0) and DQDs in series5 (setting τ12 = τ23 = 0), see Fig. 3b. The width of the antiresonance scales with the Rabi fre- quency of the coherent processes involved23: spin rota- tion (∝ Bac) and interdot tunneling (∝ τij ) (Figs. 3a and 3b, respectively); it also depends on the tunneling rate through the contact barriers, which induce decoherence, see Fig. 3c. The quenching of the current can be understood ana- lytically by transforming the Hamiltonian into the ro- tating frame. Applying the unitary operator U (t) = exp{−iωtP3 the magnetic field term reads: i=1[(∆i− ω) Szi +Bac Sxi]. One can easily verify H ′ B = P3 that, at ω0 = (∆1 + ∆3)/2, the coherent superpositions Szi}, i=1 Ψ±i = 1 √2 (cid:16)ξ± ↓↓i − ξ± ↑↑i(cid:17) (5) are eigenstates of the Hamiltonian H ′ = Hτ + H ′ B. Since the electrons in (5) have parallel spins, current is quenched due to SB. Note that the electron spins in (5) are maximally entangled. We want to emphasize that SB can be switched on and off by tuning the frequency of Bac, which is usually introduced to lift it, or by changing the flux Φ at a fixed frequency ω, see Fig. 3d. In TQDs, a necessary condition for (5) to be eigen- states of H ′ and thus for the current blocking to occur, is the equal coupling of dots 1 and 2 to Bdc, i. e. ∆1 = ∆2 (6= ∆3). If ∆1 6= ∆2 though, this symmetry is broken and H ′ B couples all parallel to antiparallel spin states and thus to the transport state S3i. However, numerical results show that even in the asymmetric case, a pronounced antiresonance due to SB still appears in the current. By means of a perturbative analysis for ∆1 − ∆2 ≪ τ it can be shown that the antiresonance occurs at a frequency ω1 ≈ 1 2 + ∆3), see Fig. 4. The electrons drop into 2 ( ∆1+∆2 0.02 Γ / I 0.8 1 ω/ω0 1.2 FIG. 4: I-ω in a TQD for ∆1 6= ∆2. For τ ≫ ∆1 − ∆2, I shows an antiresonance with a pronounced minimum at ω1 ≈ 1 + ∆3). Dashed-dotted line: ∆2 − ∆1 = 0.0013; dashed line: ∆1 − ∆2 = 0.0013; solid line: ∆1 = ∆2. Parameters: τ = Γ = 0.01. 2 ( ∆1+∆2 2 y t i l i b a b o r p . c c O 1 0 0 σ,0,σ 0,σ,σ . b o r p . c c O S3〉 Time [2π/Ωτ] 10 20 Time [2π/Ωτ] 30 FIG. 5: ρii in a TQD excited by a bichromatic Bac, where ω1 = ∆1,2 and ω2 = ∆3. In the stationary limit, parallel spin states decouple from Bac and form coherent superpositions as in (5) thereby blocking the current. an eigenstate Ψ⋆i which is similar to (5) but includes a small contribution of antiparallel spin states which pro- duces a small leakage current. This leakage current in- creases as ∆1 − ∆2 becomes of the order of τ . Bichromatic Bac -- Finally we will show that for ∆1 = ∆2 SB can also be induced by a bichromatic Bac, tuning its frequencies to ω1 = ∆1,2 and ω2 = ∆3, 4 so every electron is kept in resonance regardless of its location. Assuming that the inhomogeneity in the Zee- man splittings is high enough so one can neglect the off- resonance terms, the Hamiltonian HB(t) can be written as HB,2 = P3 i=1[∆i Szi + Bac(cos(∆it) Sxi + sin(∆it) Syi)]. Again, by means of the unitary transformation U (t) = exp{−iPi ∆i Szit}, the states (5) turn out to still be eigenstates of the transformed H ′ B,2, even if the two spins are now rotating in resonance. The bichromatic Bac washes out all the states with antiparallel spins, driving the system into SB in spite of the Zeeman inhomogeneity, see Fig. 5. Conclusions -- In summary, we have shown theoreti- cally that TQD systems in triangular configuration under dc and ac magnetic fields exhibit rich dynamics due to the interplay of different coherent phenomena induced by the magnetic fields. For two extra electrons in the system the interplay of Pauli exclusion principle and coherent trap- ping is discussed in terms of the magnetic flux piercing the TQD. We have shown that, in contrast to the one electron case, due to SB, electrons remain trapped even for Φ/Φ0 6= n/2. We demonstrate that a generic prop- erty of monochromatic and bichromatic magnetic fields is to induce SB at certain frequencies in both DQDs and TQDs. Furthermore, the coherent superposition induced by the Bac constitutes a novel SB state and is decoupled from the field. Its experimental realization will allow one to infer properties of the system such as Zeeman inho- mogeneities, and to manipulate spin qubits in DQDs and TQDs. It opens new perspectives for manipulating spin transport properties, thereby providing possibilities for designing spintronic devices. We thank S. Kohler for critical the manuscript and T. Brandes, C. Emary and C. Poltl for helpful discussions. This work has been supported by grant MAT2008-02626. M. B. acknowledges support from CSIC (JAE). R.S. acknowledges support from Swiss NSF and MaNEP. reading of 1 K. Ono et al., Science 297, 1313 (2002). 2 J.R. Petta et al., Science 309, 2180 (2005). 3 F.H.L. Koppens et al., Nature 442, 766 (2006). 4 M. S. Rudner et al., Phys. Rev. Lett. 99, 036602 (2007). 5 R. S´anchez et al., Phys. Rev. B, 77, 165312 (2008). 6 K.C. Nowack et al., Science 318, 1430 (2007). 7 M. Pioro-Ladri`ere et al., Nature Physics 4, 776 (2008). 8 E. A. Laird et al., Phys. Rev. Lett. 99, 246601 (2007). 9 D. Schroer et al., Phys. Rev. B 76, 075306 (2007). 10 M. C. Rogge et al., Phys. Rev. B 77, 193306 (2008); L. Gaudreau et al., Phys. Rev. Lett. 97, 036807 (2006). 11 M. Korkusinski et al., Phys. Rev. B 75, 115301 (2007); F. Delgado et al., Phys. Rev. B 76, 115332 (2007); F. Delgado et al., Phys. Rev. Lett. 101, 226810 (2008). 12 T. Kostyrko et al., Phys. Rev. B 79, 075310 (2009). 13 T. Kuzmenko et al., Phys. Rev. Lett. 96, 046601 (2006); E. Vernek et al., Phys. Rev. B 80, 035119 (2009); 14 D. S. Saraga et al., Phys. Rev. Lett. 90, 166803 (2003). 15 R.M. Whitley et al. Jr., Phys. Rev. A 14, 1498 (1976); H.R. Gray et al., Opt. Lett. 3, 218 (1978); E. Arimondo et al., Lett. Nuovo Cim. 17, 333 (1976). 16 T. Brandes et al., Phys. Rev. Lett. 85, 4148 (2000). 17 B. Michaelis et al., Europhys. Lett. 73, 677 (2006); C. Poltl et al., Phys. Rev. B 80, 115313 (2009). 18 M. Busl et al., Physica E 42, 830 (2010). 19 C. Emary, Phys. Rev. B 76, 245319 (2007). 20 Y. Aharonov et al., Phys. Rev. 115, 485 (1959); M. Buttiker, et al., Phys. Rev. A 30, 1982 (1984). 21 J. Inarrea et al., Phys. Rev. B 76, 085329 (2007); R. S´anchez et al., New J. Phys. 10, 115013 (2008); F. Dom´ınguez et al., Phys. Rev. B 80, 201301 (2009). 22 S.M. Huang et al., arXiv:0904.1046, Phys. Rev. Lett. (to be published). 23 T. Brandes, Phys. Rep. 408, 315 (2005). 5 

1006.5915

1

1006

Thresholdless Coherent Light Scattering from Subband-polaritons in a Strongly-Coupled Microcavity

We study a "strongly-coupled" (SC) polariton system formed between the atom-like intersubband transitions in a semiconductor nanostructure and the THz optical modes that are localised at the edges of a gold aperture. The polaritons can be excited optically, by incoherent excitation with bandgap radiation, and we find that they also coherently scatter the same input laser, to give strikingly sharp "sideband" (SB) spectral peaks in the backscattered spectrum. The SB intensity is a sensitive track of the polariton density and they can be detected down to a quantum noise floor that is more than 2500 times lower than the excitation thresholds of comparable quantum cascade laser diodes. Compared with other coherent scattering mechanisms, higher order SB scattering events are readily observable, and we speculate that the effect may find utility as a passive all-optical wavelength shifting mechanism in telecommunications systems.

cond-mat.mes-hall

cond-mat

Thresholdless Coherent Light Scattering from Subband-polaritons in a Strongly-Coupled Microcavity. Johannes Gambari1, Antonio I Fernandez-Dominguez1, Stefan A Maier1, Ben S Williams2,3, Sushil Kumar3, John L Reno4, Qing Hu3 and Chris C Phillips1 1 Physics Dept., Imperial College London, London, SW7 2AZ UK.  2 Dept of Electrical Engineering and California NanoSystems Institute, University of California, Los Angeles, California 90095, USA.  3 Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science and Research Laboratory of Electronics, Cambridge,  Massachusetts, 02139, USA.  4Sandia National Laboratories, Department 1123, MS 0601, Albuquerque, New Mexico, 87185‐0601, USA.5  (Received ???? :published ???) We study a “strongly-coupled” (SC) polariton system formed between the atom-like intersubband transitions in a semiconductor nanostructure and the THz optical modes that are localised at the edges of a gold aperture. The polaritons can be excited optically, by incoherent excitation with bandgap radiation, and we find that they also coherently scatter the same input laser, to give strikingly sharp “sideband” (SB) spectral peaks, in the backscattered spectrum. The SB intensity is a sensitive track of the polariton density and they can be detected down to a quantum noise floor that is more than 2500 times lower than the excitation thresholds of comparable quantum cascade laser diodes. Compared with other coherent scattering mechanisms, higher order SB scattering events are readily observable, and we speculate that the effect may find utility as a passive all-optical wavelength shifting mechanism in telecommunications systems. In a strongly-coupled (SC) system, the electron and photon sub-systems are so strongly coupled to each other that they form a new hybrid entity, a polariton, where the coupling is characterised by a vacuum-Rabi energy, ħΩVR, that exceeds the original natural linewidths of both the electron and photon modes. Viewed in the time domain, the -1 excitation energy cycles coherently, at a rate ħΩVR between electronic and photonic forms, before being lost, either by optical emission or by other non-radiative loss channels [1]. Here we study a SC system whose polaritons are formed from the THz photon modes of a tightly-confined metal- semiconductor microcavity, that are hybridised with the electronic “Intersubband transitions” (ISBT’s) in a semiconductor nanostructure. These polaritons carry an optical dipole, so when we create a population of them incoherently, (with an interband near-infrared pump laser, ωNIR (fig.1)), we find that they coherently scatter that same input beam, generating strikingly sharp, multiple-order sidebands (SB’s), at ωSB = ωNIR +nωTHz (n=-2,-1,0,+1+2) in the spectrum of backscattered light. The devices studied were fabricated from epilayers comprising many (~175) repeats [2,3] of a GaAs/ [Al,GaAs] semiconductor nanostructure module [fig 1]. The epilayers were fabricated into ~10μm thick gold-epilayer-gold sandwich waveguides ideal gold the near that used conductivity [4,5] to confine the THz fields into layers ~ 10 times thinner than the λ ~ 100 μm free-space wavelength. Three device structures were studied, identified here as “λ~80μm” [3], “λ~100 μm”[2] and “λ~120” μm[3] corresponding to the approximate THz free-space wavelengths they emitted when they were electrically driven, in separate experiments [2,3], as standard quantum cascade lasers (QCL’s). 100 ωNIR Energy (meV) B 5〉 1.52 eV A 50μm  4〉 2〉 3〉 1〉 80 60 40 20 ) W n (   y t i s n e t n I 100 50 0 THz Polariton field spectrometer resolution ωSB=  ωNIR + nωTHz  0 730 732 734 736 738 740 wavelength (nm) Fig 1. A single period of the nanostructure, embedded in a THz microcavity which is symbolised here by external mirrors. THz polaritons are created (A) by photo carriers, excited by a near IR laser, ωNIR, cascading down through the subband system until they reach the polariton state. The same near IR laser (B), is also coherently scattered from these ωTHz polaritons, and generates “sidebands” at ωSB = ωNIR +nωTHz.. Upper left inset:- Raw spectra of the light backscattered from the top of the device, as the ~50 μm diameter λ~744nm ωNIR laser spot is scanned along the centre of the ridge, across one of the slots, at distances from the centre of the slot of 0 μm (squares) , 20μm (triangles), 40μm (circles) and 100μm (inverted triangles). Coherent Sideband emitted at ωSB = ωNIR + ωTHz 20 30 Z (nm) 0 10 40 50 60 1  n=-1 Input Laser wave- length n=+1 n=+2 50 40 30 20 10 0 n=-2 m μ 0 8 = λ , ) W n ( r e w o P d n a b e d i S 400 350 300 250 200 150 100 50 0 -50 m μ 0 2 1 = λ ) W n ( r e w o P d n a b e d i S ) W p ( r e w o P 150 hωQCL Input Wave- length n=+1 Sideband Anti-Stokes Raman (x 100) ) W 50 n ( r 40 e w o 30 P d n 20 a b e 10 d i S 0 1 = n 800 810 820 830 Sideband Wavelength (nm) (a) 5 ) W 70 n 4 60 ( r e 50 w 3 o 40 P 2 d 30 n a 20 b 1 e hωLO d 10 i 0 S 785 790 795 800 805 810 100 50 0 2 = Wavelength (nm) Temperature (K) n (b) (c) Fig 2.(a) T~14K Background-subtracted spectra taken, from the “λ~120 μm” (solid line, right hand vertical axis), and “λ~80 μm” (dotted line, left hand vertical axis) devices. Both spectra show coherent sideband features at ωSB = ωNIR +nωTHz. (n=-2,-1,+1,+2), whose linewidths are set by the spectrometer resolution [Δλ~ 1.4nm in this case]. (b) Comparison between the T~ 14K Anti-Stokes GaAs Raman line, [enhanced by x100 on this plot] and a typical n=+1 sideband feature. (c) Temperature dependence of the SB intensity measured with a near IR wavelength of 810nm and ~1mW power entering the “λ~80 μm” device; squares, the n=-1 SB intensity ( right hand axis), filled circles, the n=-2 sideband intensity (left hand axis); open circles, output of a QCL made from the same heterostructure [3] (arb units). Each period of the the “λ~100μm” nanostructure in fig. 1 [2] comprises 4.9/7.9/2.5/6.6/4.1/15.6/3.3/9.0 nm thick layers of Al0.3Ga0.7As/GaAs and supports ~5 confined electron states. The 15.6 nm well is doped at 1.9 x 1016 cm-3, giving an areal electron density, ns = 3 x 1010 cm-2 and a Fermi energy of ~1meV, so only the lowest subband is occupied at equilibrium. The 1.52 eV photoluminescence peak implies an effective bandgap close to that of bulk GaAs and the 2> ⇒ 1> ISBT had a modelled energy of E12 =15.8 meV and a transition dipole of z12 = 2.3 nm. The waveguide is ~10μm x 50 μm x 834 μm long. It has a periodic array (fig. 1 inset) of 6 30 μm x 8 μm slots, spaced by Λ = 31 μm, etched into it’s top surface. These were originally designed to outcouple the THz fields when the structure was used as a QCL [4]. The devices were illuminated normally, with a tuneable continuous-wave titanium:sapphire near-infrared laser, ωNIR, with a 50μm spot size and a linewidth ΔωNIR < 0.1 meV. The backscattered light was polarisation filtered before being analysed with a 0.25m grating monochromator and a standard, background-subtracting, photon-counting setup whose cooled photomultiplier had a dark count < 10 cps. The sharpness of the SB’s meant careful attention to the system’s mechanical and laser wavelength stability was needed to see them. Backscattered spectra were typically dominated by an elastically scattered background peak and a λ =815 nm / 1.52 eV photoluminescence peak (not shown), but they also featured strikingly sharp “sideband” (SB) peaks. The data presented in figures 1 - 3 were taken with the spectrometer slits opened up to improve the signal-to-noise ratio, but separate trials (not shown) always found SB linewidths that were resolution-limited, down to the 0.3nm/0.5meV working limit of the spectrometer. For a given device, the SB features stayed at a fixed frequency interval from ωNIR, as the Ti:sapph laser was tuned over a wide wavelength range (fig.3). All the SB’s disappeared if ħωNIR was tuned below the ~1.52 eV/ 815 nm effective bandgap of the nanostructure. Both 1st and 2nd order peaks vanished [fig.2 (c)] by ~120K, similar to the maximum operation temperature of comparable QCL’s [3]. The optical polarisation of the SB’s matched the polarisation of the ωNIR input to better than 99%, whether it was linear or circular. When a small bias was applied across the waveguide it acted as a photoconductive detector which could be calibrated to measure the fraction of the ωNIR beam that made it through the illuminated slot in the top gold layer. When the laser spot was scanned along the centre of the waveguide, across a given slot, (Fig 1 inset) the SB intensity scaled with this photocurrent. The SB’s were reproducible over months of measurement and through numerous changes to the optical setup. They were never seen in control experiments, when the ωNIR spot was focussed onto (i) un-metallised epilayer samples, (ii) onto GaAs test wafers, (iii) onto highly scattering parts of the cryostat mount or (iv) onto the gold contacting layers adjacent to the slots in the structure of fig.1. The way the SB’s tune with ωNIR [figs. 3(b) and 3(c)] immediately implies that they are generated by a form of coherent scattering. We rule out normal phonon Raman scattering because the SB’s are ~ 100 x more intense, at least 12 times sharper, and appear at the wrong energy offsets [fig. 2(b)] compared with typical LO phonon Raman lines. Coherent scattering processes generate SB’s whose lineshape is a convolution of the linewidths of the input laser (<0.1meV) and that of the scattering excitation. Therefore, the sharpness of the SB lines [< 0.5meV] compared with an estimated bare ISBT linewidth of ~ meV [6] argues that they cannot be due to standard electronic Raman scattering from the ISBT. SB features which are superficially similar to the ones we see here have previously been generated by exploiting optical frequency mixing effects which arise due to non- linear components of the optical polarisability in the material of a QCL [7,8]. However, this mechanism requires the presence of a spectrally sharp THz optical field inside the structure, and this could only be happening here if we had somehow managed to produce an optically-pumped 2  0 2 15 10 5 1 300 200 100 ) W n ( r e w o P d n a b e d i S ) W n ( r e w o p d n a b e d i S ) m 820 n ( h t g 800 n e l e v 780 a w t u 760 p n i R I N 50 ) 5 - 0 1 40 ( y c 30 n e i c 20 i f f E 10 n o i 0 0 s r e v 0.3 0.0 0.2 0.1 n o NIR power entering QCL (mW) C (a) ) m 800 n ( h ) W t g n 775 n ( e r l e e w v a o 750 w P t u p n 0 725 i 740 800 780 760 R 820 800 780 760 I Sideband wavelength (nm) Sideband Wavelength (nm) N (b) (c) Fig.3 (a) Sideband power, (squares, left hand axis) and conversion efficiency (circles, right hand axis) from the “λ~100μm” device of fig.1. The Dashed (solid) lines are guides to the eye, evidencing the quadratic(linear) dependencies of the power (efficiency) at low NIR pump levels (b) tuning behaviour (right hand axis) and n=1 sideband spectra (left hand axis) of the “λ~100 μm” structure as the near IR input wavelength (1mW power) is scanned . (c) tuning behavior (right hand axis) of the n= +2 (squares), n=+1 (circles), n=-1 (diamonds) and n= -2 (triangles) SB peaks from the “λ~80 μm” structure. Spectra (left hand axis), of the n=-2 SB’s peaks from the same sample at the input wavelength denoted by the corresponding triangle data point on the tuning line. All data taken at T~14K. analogue of a standard QCL. This possibility, however, is strongly at odds with the fact that we see no excitation threshold for the onset of the SB generation (Fig.3). Defining the SB generation efficiency, ηSB , as the ratio between the detected sideband intensity and the ωNIR power entering the slot [Fig.3(a)], ηSB climbs linearly from the detector system’s photon noise floor as the ωNIR pump power is increased. When a comparable structure to the device of fig. 1 was configured as a QCL and electrically excited [2], it’s threshold current density, Jth ~ 435 A cm-2, corresponded to ~4.8 x 1027 sec-1 m-2 ISBT transitions throughout the epilayer’s 175 periods. In our optical experiments (fig.3) the SB first appears with ~0.1mW entering the 2.4 x 10-10 m2 slot, an areal excitation rate ~1.7 x 1024 sec-1 m-2 , i.e. ~2500 times smaller than the comparable QCL threshold. Finally, the ease with which the n=±2 higher order processes are seen [fig.2(a)] contrasts sharply with what is in Raman and non-linear frequency mixing observed experiments. To understand the origin of the sidebands we first need to estimate the coupling between the ISBTs and the photon modes in these devices. The ISBT energy is independent of electron in-plane wave vector, so strong electron correlation effects [9] concentrate all the oscillator strength into a single, dispersionless, atom-like Lorentzian line, with a large transition dipole, z12 , that has already been shown to generate SC with giant ħΩVR energies in planar structures [10,11], especially with the wider wells used in THz devices [12]. In fact, ħΩVR values have been achieved that not only exceed the linewidths, but are also are comparable with the transition energy itself [13,14], the so-called ultra-SC (USC) condition[15]. In our non-planar devices, the photon modes are confined in all 3 dimensions, so to estimate the degree to which they couple with the ISBT’s we must first compute their mode shapes and volumes. This is done with a standard finite-difference-time-domain (FDTD) calculation, on a 125 nm mesh, which treats the gold as a perfect conductor and the semiconductor as an insulator with a dielectric constant of 13.3 [16]. It models the ridge structure of fig 1 as an infinite array of slots so that periodic boundary conditions allow the modes to be plotted in terms of the superlattice wavevectors, 2π/Λ, where Λ = 31 μm is the slot repeat distance (fig.4) The model correctly reproduces the “radiating” modes (fig. 4(a) squares) that were originally intended to outcouple [4] the THz when biased as a QCL. However, at ) z H T ( f 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0.0 0.5 0.4 0.3 0.2 0.1 Normalised wave vector [2π/Λ] (a) 1 0.9 0.6 0.3 2E-4 15 20 25 30 x (μm) (b) (c)  12 10 8 6 ) m μ ( z 0 5 10 z x y   Fig. 4 (a) THz photon modes supported by the device of Fig.1, which is modelled as an superlattice array of slots on the top of an infinite (in x) ridge waveguide. Modes are enumerated in terms of the superlattice wavevector, 2π/Λ, where Λ = 31 μm is the slot spacing. Squares, the “radiating” mode designed to out-couple radiation when electrically driven as a QCL Red triangles:- the “localised” family of modes. Dotted (solid) lines are light lines in the vacuum (semiconductor). (b) energy distribution, across a single slot in the periodic structure, of the “localised” modes. (c) Schematic 3D distribution of the fields in the “localised” mode concentrated at the edges of the slots in the structure. almost the same energy (12.5 meV/ ~3 THz) there is another family of “localised” THz modes [fig. 4(a) triangles] whose field distributions are tightly localised at the slot edges, similar to the ultra-confined modes recently reported [17,18] in other sub-wavelength structures. These originate from a vertical ¼ wave “organ–pipe” resonance, with a node 3 produced in these experiments (i.e. to demonstrate a THz analogue of a so called “inversionless laser”) were frustrated by the poor sensitivity of current THz detectors, and the very weak coupling of the “localised” modes to the outside world. That said, we believe that this effect may prove more useful as a simple, passive coherent optical mixing devices than as a source of THz radiation. Although the conversion efficiencies and operating temperatures are low at the moment, this effect still has potential for frequency-shifting e.g. an optical bit stream by a fixed frequency interval that can be tightly specified at the design stage and would be full optical data-transparent and operate across the telecommunications bandwidth. Moving to the [In,Al,Ga],As materials system and optimising ħΩVR by judicious choice of doping levels, THz mode shapes and ISBT z12 values may move the operating wavelengths, temperatures and efficiencies towards technologically useful values. Helpful conversations with Paul Eastham are gratefully acknowledged. This work was supported by the Engineering and Physical Sciences Research Council (EPSRC), and by the US Air Force Office of Scientific Research AFOSR. [1] M S Skolnick, T A Fisher and D M Whittaker, Semicon. Sci . Tech, 13, 645 (1998). [2] B S Williams, S Kumar, Q Hu and J L Reno, Opt. Express 13, (9), 3331 (2005). [3] B S Williams, et al. , Appl. Phys. Lett., 88, 261101 (2006). [4] S Kumar, et al. Opt. Express, 15 , 113. (2006). [5] SA Maier, Opt. Express 14, 1957 (2006). [6] Electroluminescence (EL) spectra from test diodes had ~4meV linewidths, but these came from “diagonal” ISBT’s with strong interface broadening arising from the high electric fields needs for the measurements. ~2 meV ISBT linewidths were seen in reflectance measurements of Planar samples with similar (3.7THz) ISBT energies to the device of fig. 1. [12]. [7] S. S. Dhillon, et.al., Appl. Phys. Lett., 87, 071101 (2005). [8] C. Zervos et.al. Appl. Phys. Lett. 89, 183507 (2006) [9] D. E. Nikonov , A. Imamoglu, L. V. Butov, and H. Schmidt,. Phys. Rev. Lett. 79, 4633 (1997). [10] D. Dini, Phys.Rev.Lett. 90, 116401 (2003). [11] E. Dupont, H. C. Liu, S. Schmidt and A. Seilmeier, Appl.Phys.Lett. 79 4295 (2001). [12] Y Todorov et al. Phys Rev Lett., 102 186404 (2009). [13] A A Anappara, et al. Appl. Phys.Lett. 91, 231118, (2007). [14] Jonathan Plumridge, Edmund Clarke, Ray Murray and Chris Phillips, Solid State Comm., 146, 406, (2008). [15] C Ciuti, G Bastard, I Carusotto Phys. Rev .B 72, 115303 (2005). [16] From the point of view of interband absorption and dielectric response, the heterostructure slab region closely approximated n~ 5 x 10 15 cm-3 bulk GaAs, with absorption properties as described in Blakemore, JS, J. Appl. Phys. 53, R123 (1982) . [17] M A Seo et al. Nat. Photon., 3, 152 (2009). [18] Rupert F. Oulton et.al. Nature, 461, 629 (2009). [19] J McKeever et al. Nature, 425, 268. (2003). at the lower gold layer and an antinode at the slot opening, giving a resonance roughly corresponding to a free-space wavelength λ~4nh, where n is the semiconductor refractive index and h~10 μm the slab thickness. The field localisation means that photon modes on adjacent slots oscillate almost independently, so they are practically mono-energetic and dispersionless in the photonic superlattice plot (fig . 4(a)), and all the THz modes in the family can strongly couple to the electronic ISBT’s at the same time. This contrasts with the anti-crossing behaviour previously seen in dispersive 2D systems [11-15]. The field localisation around the slot edge also gives very weak outcoupling to the free-space THz modes, raising the Q factor to ~1100, compared with ~57 for the “radiating” modes. The computed volume of the “localised” mode, V~ 496 (μm)3 , is only ~λ3/2000 of the λ~100 μm free space wavelength and ~1/50 of λ3 in the semiconductor material. It’s frequency resonates closely with the modelled E12 ~ 15 meV (fig. 1) ISBT energy and its electric field is mainly vertically polarised, so it couples strongly to the vertically polarised z12 of the ISBT. Also, it’s half-height energy density (fig. 4b) is only ~0.96 μm below the semiconductor- air interface, so overlaps well with the ~1μm penetration depth of the ωNIR incoherent pump light from the the Ti:sapph laser[16]. We compute ħΩVR for the “localized” photon mode by equating the classical stored electromagnetic energy, Vεrε0E2 vac, with the quantum photon ground state energy, ħωTHz/2, to give a mean zero-point vacuum field of E ~ 142 V m-1. The interaction of a single electron ISBT oscillator with this field will give ħΩVR = 2Ee z12, which, with εr = 13.3 [16] , z12 ~ 2.3 nm, and ħωTHz =15 meV gives 6.3 x 10 - 7 eV. A factor f=0.92 of the mode energy lies inside the semiconductor, and N~2.4 x 106 ISBT electrons lies within this volume, giving a total coupling energy of ħΩVR = 2Ee z12 N ½ ~ 1.0 meV. Even with no photoexcitation this is some ~7% of the ISBT energy, and will increase further under the experimental conditions. Assuming an interband carrier recombination time of ~1 nsec, ~1mW of absorbed laser power would triple the local electron concentration and increase ħΩVR by ~√3. This ħΩVR value exceeds both the <0.5meV upper bound to the linewidth of the excitation responsible for the SB generation and the ~15 μeV modelled linewidth of the localised photon mode. This confirms the SC nature of the electron-photon system, i.e. the SB’s arise from coherent scattering mechanism from polaritons whose linewidths lie between [6] the ~meV ISBT linewidth and the ~15 μeV localised photon mode linewidth. At higher optical excitation levels (not shown), the λ ~ 100μm device n=1 SB conversion efficiency peaks at ~ 5 x 10 -5 [at ~ 1 mW input power], and then drops, because of sample heating (fig. 2(c)]. There are strong parallels between this SC system, and previous atom-cavity studies [19] , where coherent output radiation was seen without the system needing to be driven into population inversion. Unfortunately, our attempts to detect the emitted optical component of the THz polaritons 4 

1801.02153

1

1801

Theoretical investigation of charge transport in germanium doped phosphorene nanoribons using DFT + NEGF

New two diemensional structures nanoribbon including phosphorus and germanium atoms are introduced for the nanoelectronic applications. Under various bias voltages, the electronic transport in the systems have been studied within the noneqilibrium Green's function formalism. The $I-V$ characteristics have been extracted. DOS and $T(E,V_{bias})$ have been investigated and show that the charge transport occurs when the bias voltage reaches about 1 \textit{V}. The calculated MPSH shows that the spatial distribution of orbital levels has been affected by the electrodes. The studied structures have a bandgap of about 0.7 \textit{eV} which absorbs light in the visible range and thus could be an interesting contender for solar cells applications.

cond-mat.mes-hall

cond-mat

Theoretical investigation of charge transport in germanium doped phosphorene nanoribons using DFT + NEGF Maryam Azizi and Badie Ghavami School of Nanoscience, Institute for Research in fundamental Sciences, P.O.Box: 19395-5531, Tehran, Iran (Dated: July 11, 2021) New two diemensional structures nanoribbon including phosphorus and germanium atoms are introduced for the nanoelectronic applications. Under various bias voltages, the electronic transport in the systems have been studied within the noneqilibrium Green's function formalism. The I − V characteristics have been extracted. DOS and T (E, Vbias) have been investigated and show that the charge transport occurs when the bias voltage reaches about 1 V. The calculated MPSH shows that the spatial distribution of orbital levels has been affected by the electrodes. The studied structures have a bandgap of about 0.7 eV which absorbs light in the visible range and thus could be an interesting contender for solar cells applications. I. INTRODUCTION Two dimensional (2D) graphene-like materials, have recently attracted considerable attention due to their po- tential applications in nano- and optoelectronics1 -- 4. In particular, their unique size dependent properties allowed for the exploration of a large number of novel phenom- ena at nanoscale5 -- 8. Among 2D materials, the ones with sizable bandgap are used in field effect transistor (FET) devices. In the recent years, phosphorene9 -- 11 (monolayer of black phosphorus) has attracted great attention due to its reasonable mobility and bandgap which makes it an attractive material for electronic applications. The ef- fects of native defects12, vacancies, and adatoms13 have been investigated for this material. More over, tunability of electronic properties, for example due to strain14 -- 16, paves the way to interesting applications of the material. In addition to phosphorene sheets, phosphorene nanorib- bons (PNR) have also been studied theoretically8,16,17. Another promissing material with a tunable bandgap is germanene18,19 (monolayer of germanium atoms) Although both phoshorene and germanene nanorib- bons show certain advantages in comparison to graphene and 2D dichalcogenides such as their application in en- ergy conversion/storage and high performance field effect transistors (FET), there have been few transport stud- 20 -- 23. Moreover, to ies compared to graphene and M oS2 our knowledge, the only recent work on the combination of these two elements24 considers the n layer of GeP3 and focuses on the low indirect bandgaps and high car- rier mobilities. Since monolayers of germanium atoms show semimetallic properties, combining this element with phosphorus which is known as a wide bandgap ma- terial may result in a moderate bandgap material. Hence, in this letter, we for the first time introduce two nanorib- bon structures composed of both germanium and phos- phorus atomes. These new nanoribbons have a bandgap about 0.7 eV which absorbs light in the visible range and could be suitable for solar cells applications. Non- equilibrium Green's function method based on density functional theory (DFT) has been used to study charge and quantum transport properties. Applying bias volt- age, Density of States (DOS) and transmission spectrum as a function of energy have been investigated. The paper is organized as follows. In Sec.II, we first introduce the system and explain the method we use to derive current-voltage characteristics, density of states (DOS) and transmission spectrum. Then in Sec.III, we present our numerical results. In particular, the role of bias voltage on the transmission spectrum has been shown. Finally, we conclude and summarize the main achivement in Sec.IV. II. MODEL A. Structure In this work we study a nanoribbon structures com- posed of germanium and phosphorus atoms, with a bandgap between pure germanene and pure phosphorene nanoribbon, depicted in Fig.1. In this model phosphorus atoms in the both zigzag edges of the phoshorene nanoribbon, Fig.1a, were replaced by germanium atoms to construct Fig.1c. To go further with the investigation of the effect of germanium atoms, the second chain of phosphorus atomes was replaced by ger- manium ones Fig.1c. We assumed that each edge atom was passivated with enough H to remove dangling bonds. This replacement decreases the bandgap of the pure phosphorene nanoribbon and makes it more suitable for the certain electronic applications discussed in Sec.III. 8 1 0 2 n a J 7 ] l l a h - s e m . t a m - d n o c [ 1 v 3 5 1 2 0 . 1 0 8 1 : v i X r a 2 FIG. 1. phosphorus, germanium and hydrogen are in yellow, purple and blue spheres respectively. (Color online) The computational setup for a) zPNR, b) zGeNR, c) zGePNR-1chain and d) zGeNPR-2chain. B. Computational details The first principle calculations are performed within the Density Functional Theory(DFT) method, as imple- mented in SIESTA open-source package25. We assume the infinite length and priodic boundary conditions in the x-direction. To avoid interactions between the adjacent layer, more than 20A◦ vacuum space considered to separate the nano-ribbons in two other directions, . Spin-unrestricted DFT calculations are performed us- ing the Perdew Burke Ernzerhof (PBE) for generalized gradient approximation (GGA) exchange and correlation functional approach to obtain the band structures for the selected structures. In our calculations, Brillouin zone is sampled with 50× 1×1 k-point grid and the cutoff energy is fixed to be 150- Ry. The double-ζ polarized basis sets are used for the va- lence band electrons. The structures are relaxed until the interatomic forces are less than 0.002eV /A◦. In order to perform the calculations of charge trans- port and electrical properties of the systems, the Non- equilibrium Green's Functions (NEGF)26 equations are solved using the Kohn-Sham wave functions obtained from DFT,27 as implemented in TRANSIESTA open- source package28 at the room temperature. To investigate the transport properties, we specify three regions within the sample: two electrodes and the central (device) region (Fig.1). The transmission function of the system is calculated according to the following equation:29 T (E, Vb) = T r[ΓL(E, Vb)G(E, Vb)ΓR(E, Vb)G†(E, Vb)] where E, Vb and G are energy, bias voltage and green function, respectively. ΓL(R) is the spectral density de- scribing the coupling between the right (left) electrode and the scattering region. In these structures current vs bias voltage is extracted by the Landauer- Buttiker30,31 formula, (cid:90) ∞ −∞ I(Vb) = 2e h dET (E, Vb)[f (E − µL) − f (E − µR)] where µR(L) is the chemical potential of the right (left) electrode, eVb = µL − µR and f (E − µR(L)) is the Fermi function. We change the bias voltage from 0.0V to 2.0V with the step of 0.05V to achieve convergence of the density ma- trix. Left electrodeRight electrodeDevicea) zPNRb) zGePNRLeft electrodeRight electrodeDevicec) zGePNR-1chainLeft electrodeRight electrodeDeviced) zGePNR-2chainLeft electrodeRight electrodeDevice3 FIG. 2. (Color online) The band structures of a) zPNR, b) zGeNR, c) zGePNR-1chain and d) zGeNPR-2chain.. III. RESULTS AND DISCUSSION The band structures of the studied configurations are illustrated in Fig.2. In the case of phoshorene nanorib- bon, each phosphorus atom is covalently bonded to three other P atomes and forms a sp3 hybridization to con- struct a puckered honeycomb structure with a DFT bandgap about 1.3eV . The chemical bonds in germa- nium are actually of sp3 -like32 type for the partial hy- bridization of s and pz in Ge but it has a larger covalent radius and its standard atomic weight is two times big- ger than that for P. All this specifications along with its electronegativity which is almost of the same order as in the case of P, result in DFT bandgap of 1.67 eV which is not very far from P DFT bandgap. As shown in Fig.3, zPNR (zigzag edge phoshorene nanoribbon) has a higher Density of States (DOS) in the region close to Fermi energy in comparison to zGeNR (zigzag edge germanene nanoribbon). This explains why zGeNR has bigger DFT bandgap. This is true for the two other configurations studied in the present work. Replacing edge phosphorus atoms by germanium atoms, Fig.1c, leads to increase in zGePNR- 1chain's DOS around Fermi energy. As a result, the DFT bandgap is decreased in comparison to pure zPNR and zGeNR, Fig.2. Going further and replacing the next series of phosphorus atoms by Ge ones, Fig.1d, zGePNR-2chain shows less DOS around Fermi energy. This is the reason for zGePNR-2chain to have a rel- atively larger DFT bandgap rather than zGePNR-1chain. To investigate the charge transport in our systems, we determine the current-voltage bias characteristic of the configurations. As shown in Fig. 4, nonlinear I-V -traces characteristic for tunneling appear at elevated bias. One could notice that for bias voltages lower than 1V the current nearly vanishes for three systems while it in- creases with a relatively fast slope for higher bias volt- ages. FIG. 3. zGePNR-1chain and zGeNPR-2chain. (Color online) Density of states for zPNR, zGeNR, In fact, applying a bias voltage shifts the Fermi level of the left electrode with respect to the Fermi level of the right one. The current starts flowing once the top of the valence band of the left electrode matches in energy with the bottom of the conduction band of the right electrode. This occurs in both systems, but for zPNR is about three orders of magnitude smaller than for germanium doped zPNR and it rapidly increases at about 1.4V . The I − Vbias curve can be divided into three interest- ing regions. In the first one, up to 1.2V both configu- rations (zGePNR-1chain and zGePNR-2chain) show the same behavior with almost the same values of currents. With increasing the bias voltage, they start to split with small differences up to 1.75V . Then, again they behave in almost the same way. This is confirmed in Fig. 5, where transmission spec- trum of both configurations is shown at Vbias = 1.5 V . In the range of bias voltage, between 1.2 V and 1.75 V the overlap of DOS for both electrodes and scattering re- -3-2-1 0 1 2 3Γ Xa) zPNREnergy (eV)-3-2-1 0 1 2 3Γ Xb) zGeNR-3-2-1 0 1 2 3Γ Xc) zGePNR-1chain-3-2-1 0 1 2 3Γ Xd) zGePNR-2chain 0 1 2 3 4 5 6 7 8-2-1.5-1-0.5 0 0.5 1 1.5 2DOSEnergy (eV)zPNRzGeNRzGePNR-1chainzGePNR-2chain4 of the transmission spectrum for different bias voltages. According to Fig. 6, the bandgaps at zero bias volt- age are of about 1.3eV , 0.63eV and 0.74eV for zPNR, zGePNR-1chain and zGePNR-2chain, respectively which shows that the presence of germanium decrease the bandgap and causes corresponding increase in the trans- mission spectrum and charge transport, Fig. 6a. FIG. 4. zGePNR-1chain and zGeNPR-2chain. (Color online) I -V characteristics of zPNR, FIG. 5. 1chain and zGePNR-1chain at Vbias = 1.5 V (Color online) Transmission spectra of zGePNR- gion in zGePNR-2chain increases. This results in more eigenchannells open and thus larger number of electrons transfered. That is why the I − Vbias curve shows higher values of current for zGePNR-2chain in this bias voltage range. For the zPNR system, the current increases with the applied bias voltage and reaches the maximum value of 1.36 µA at Vbias = 1.55 V . However, when Vbias in- creases further, the current decreases dramatically, and consequently the negative differential resistance (NDR) phenomenon arises. An important aspect is the path through which the current flows in the channel material. We have traced the pathway for the current by studying the eigenvectors FIG. 6. sity of states for the three configurations at Vbias = 0 V (Color online) a) Transmission spectra and b) Den- In fact, at zero bias voltage, both electrodes have ex- actly the same DOS, and hence the transmission function is large wherever the electrodes have electronic states. However, since the bias window is zero, there is no cur- rent, as it is confirmed by Fig. 4. Fig. 6b, shows the DOS for zPNR, zGePNR-1chain and zGePNR-2chain and it can be seen that there exists no overlap between the electrodes in the absence of bias volt- age. By increasing Vbias, the chemical potential of elec- trodes changes and it reaches to the region where the overlap between the states in the electrodes and the de- vice is not zero anymore. In this case, the charge trans- port takes place as it is shown in Fig. 7, for germanium doped PNR. The upper panel of Fig. 7 shows the transmission spec- trum and the Density of States (DOS) at the Fermi en- 0.00.51.01.52.0Bias Voltage(V)05101520µ AnAzGePNR−1chainzGePNR−2chainzPNR0.00.51.01.52.0Bias Voltage(V)0.00.40.81.2nAP0.81.21.6Bias Voltage(V)0.00.40.81.2µ AnA2.01.51.00.50.00.51.01.52.0Energy(eV)02468101214TransmisionVbias=1.5VzGePNR−1chainzGePNR−2chain0.40.20.00.20.00.10.20.30.4 0 5 10 15 20-1-0.5 0 0.5 1(a) TransmissionzPNRzGePNR-1chainzGePNR-2chain 0 200 400 600 800-1-0.5 0 0.5 1(b) DOSEnergy (eV)zPNRzGePNR-1chainzGePNR-2chain5 FIG. 7. Vbias = 2 V respectively. (Color online) Transmission and Density of States for zGePNR-1chain and zGePNR-2chain at Vbias = 1 V and ergy of the device for a bias voltage of 1V . One can see that the amplitude of transmission spectrum is large only at the two edges of DOS. Thus it is clear that the cur- rent is carried by the edge states, and there is virtually no current in the central region. The DOS has large amplitude near the left electrode. The amplitude decreases as we move from the left elec- trode into the device and near the right electrode the amplitude nearly vanishes resulting in a very small cur- rent (Fig. 4(a) shows almost zero value for current at Vbias = 1 V ). Lower panel of Fig. 7 represents the transmission spec- trum and the Density of States (DOS) for a bias voltage of 2V . One can notice that the amplitude of transmis- sion spectra in the bias voltage interval increases, Fig. 7c. This results from the overlap between DOS in both elec- trods and the device, Fig. 7d. To better understand the effect of Vbias, we show the eigenstates of the molecule placed in two different probe environments. The eigenvalues were calculated accord- ing the molecular projected self-consistent Hamiltonian (MPSH), Fig. 8. The clearly nonzero transmission spec- tra emerges around the Fermi level and the isosurfaces appear inside the device. Fig. 8 shows the highest occu- pied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) for the first (Fig. 8a and b) and second (Fig. 8c and d) molecular orbital levels for zGePNR-1chain at Vbias = 2 V . The state overlap can be clearly seen in the right upper panels where a non- zero current flows from left electrode to the right one (Fig. 4(a) shows 4.4µA). The good compatibility can be seen between Fig. 8(a) and Fig. 8(c). This is what one would expect in a sizable bandgap semiconductor. The applied bias voltage changes the eigen-channels in the electrode-device interfaces and the bandgap narrow- ing occurs due to induced electrostatic potential across the nanoribbon. The calculated molecular projected self-consistent Hamiltonian (MPSH) (Fig. 8) shows the eigenstates of the molecule which placed in a two-probe environment. This confirms that the spatial distribution of orbital lev- els has been affected by the electrodes. The same effect occurs for zPNR, Fig. 9. The only difference is that for zPNR the maximum amplitude of current is about three orders of magnetide smaller. This allows to explain the DOS spectra shown in Fig. 6. 0 0.2 0.4 0.6 0.8 1-0.8-0.6-0.4-0.2 0 0.2 0.4 0.6 0.8 (c) Vbias=2 VTransmission Energy (eV)zGePNR-1chainzGePNR-2chain 0 200 400 600-1-0.5 0 0.5 1 (d) Vbias=2 VDOSEnergy (eV)zGePNR-1chainzGePNR-2chain 0 0.02 0.04 0.06 0.08-0.8-0.6-0.4-0.2 0 0.2 0.4 0.6 0.8 (a) Vbias=1 VTransmission Energy (eV)zGePNR-1chainzGePNR-2chain 0 200 400-1-0.5 0 0.5 1 (b) Vbias=1 VDOSEnergy (eV)zGePNR-1chainzGePNR-2chain6 FIG. 8. (Color online) The HOMO, LUMO, HOMO-1 and LUMO+1 MPSH orbital for zGeRNR-1chain at Vbias = 2 V . FIG. 9. LUMO+1 MPSH orbital for zRNR at Vbias = 2 V . (Color online) The HOMO, LUMO, HOMO-1 and If Ge is embeded in the structure, the spectra penetrate deeper into the conduction region (about 1 eV ) than those for zPNR. In order to investigate the calculated rectification be- havior, we studied the transmission spectrum T (E, Vbias) as a function of energy E and the bias voltage Vbias, Fig. 10a, which we compare the corresponding DOS and transmission spectra of zPNR and germanium doped PNR. As seen in Fig. 10, transmission happens to some extend in both structure, however in zPNR the amplitude is almost zero even in the highest Vbias. FIG. 10. (Color online) a) Transmission spectrum and b) Density of States for zPNR and zGePNR-1chain at Vbias = 2 V . The calculated DOS shows a peak at the Fermi level, which implies a strong correlation between transmission (c) HOMO-1 at V=2 V(d) LUMO+1 at V=2 V(a) HOMO at V=2 V(b) LUMO at V=2 V(c) HOMO-1 at V=2 V(d) LUMO+1 at V=2 V(a) HOMO at V=2 V(b) LUMO at V=2 V 0 5 10 15 20-4-3-2-1 0 1 2 3 4(a) Vbias = 2 VTransmissionEnergy (eV)zPNRzGePNR-1chain 0 500 1000 1500-4-3-2-1 0 1 2 3 4(b) Vbias = 2 VDOSEnergy (eV)zPNRzGePNR-1chain7 FIG. 11. charge carrier energy and bias voltage. (Color online) The transmission spectra of a) zPNR, b) zGeNPR-1chain and c) zGePNR-2chain as a function of and DOS. This occurs because the transport at the Fermi level is dominated by resonant tunneling through inter- face states (see Fig. 10), not barrier tunneling33. Under a positive bias voltage the chemical potential of the left (right) electrode shifts by eV 2 ). Therefore, the DOS of the left (right) electrode shifts toward higher (lower) energy by eV 2 (− eV 2 , Fig. 10. From Fig. 10, one expects to have higher transmission for the higher applied bias voltage. This is confirmed by Fig. 11, which proves increasing the bias voltage causes higher transmission values (lighter color) in both struc- tures. The two solid lines in Fig. 11 show the total current ob- tained from the integration of the transmission function in a given bias window. The positive (negative) trans- mission refer to electron and hole conductivities, respec- tively. One may notice that in positive (negative) re- gion, the contributions of electrons (holes) in conductiv- ity dominate. As can be seen in Fig. 11b and Fig. 11c replacing P edge atoms by Ge atoms, results in the in- creased transmission and thus also conductivity and cur- rent. IV. CONCLUSIONS We studied two new systems consisting of phosphorus, germanium and hydrogen atoms in a zigzag nanoribbon. The structures were relaxed using DFT method under exchange-correlation potential GGA. The electronic and transport properties of the systems are investigated within the noneqilibrium Green's func- tion formalism and density functional theory. DOS and T (E, Vbias) were investigated and show that the charge transport occurs when the bias voltage reaches to about 1 V. First, the electronic structures of pure zPNR and zGeNR were analyzed then we showed how replacing edge phosphorus atoms with germanium atoms results in de- creasing of the bandgap of zPNR. The transport channels are studied via the calculations of the current density and local electron transmission pathway. The calculated MPSH shows that the spatial distri- bution of orbital levels was affected by the electrodes. The visualized transmission pathways show how charge carriers propagate through the scattering region in all systems. The characteristics of the new structure were compared to those of zPNR. The results confirm that the bandgap of phosphorene nanoribbon in the presence of germanium decreases which results in the increase of charge trans- port. The studied structure has a bandgap of about 0.7 eV which absorbs light in the visible range and thus is an interesting contender for solar cells. The characteristics of the present systems make them suitable for practical applications in nanoelectronic and optoelectronics and devices. Finally, our calculations show that negative differential resistivity behavior in zPNR device vanishes when the edge phosphorus atoms are substituted by Ge ones. Acknowledgments This work was partially supported by Iran Science Elites Federation grant. 8 1 M. E. D´avila and G. Le Lay, Scientific reports 6, 20714 18 Z. Ni, Q. Liu, K. Tang, J. Zheng, J. Zhou, R. Qin, Z. Gao, (2016). D. Yu, and J. Lu, Nano letters 12, 113 (2011). 2 F. Bonaccorso, Z. Sun, T. Hasan, and A. Ferrari, Nature 19 N. Drummond, V. Zolyomi, and V. Fal'Ko, Physical Re- photonics 4, 611 (2010). 3 F. Schwierz, Nature nanotechnology 5, 487 (2010). 4 V. Nicolosi, M. Chhowalla, M. G. Kanatzidis, M. S. Strano, and J. N. Coleman, Science 340, 1226419 (2013). 5 Z. Wang, H. Jia, X. Zheng, R. Yang, Z. Wang, G. Ye, X. Chen, J. Shan, and P. X.-L. Feng, Nanoscale 7, 877 (2015). 6 V. Saroka, I. Lukyanchuk, M. Portnoi, and H. Abdelsalam, Physical Review B 96, 085436 (2017). 7 L. Hallıoglu, Ph.D. thesis, bilkent university, 2015. 8 A. Maity, A. Singh, P. Sen, A. Kibey, A. Kshirsagar, and D. G. Kanhere, Physical Review B 94, 075422 (2016). 9 C. Guo, T. Wang, C. Xia, and Y. Liu, Scientific reports 7, 12799 (2017). 10 Y. Lv, S. Chang, Q. Huang, H. Wang, and J. He, Scientific reports 6, 38009 (2016). view B 85, 075423 (2012). 20 H. Liu, Y. Du, Y. Deng, and D. Y. Peide, Chemical Society Reviews 44, 2732 (2015). 21 S. Das, M. Demarteau, and A. Roelofs, ACS nano 8, 11730 (2014). 22 B. Ghavami and A. Rastkar-Ebrahimzadeh, Molecular Physics 113, 3696 (2015). 23 M. V. Kamalakar, B. Madhushankar, A. Dankert, and S. P. Dash, Small 11, 2209 (2015). 24 Y. Jing, Y. Ma, Y. Li, and T. Heine, Nano Letters 17, 1833 (2017). 25 J. M. Soler, E. Artacho, J. D. Gale, A. Garc´ıa, J. Junquera, P. Ordej´on, and D. S´anchez-Portal, Journal of Physics: Condensed Matter 14, 2745 (2002). 26 J. Taylor, H. Guo, and J. Wang, Physical Review B 63, 11 A. Carvalho, A. Rodin, and A. C. Neto, EPL (Europhysics 245407 (2001). Letters) 108, 47005 (2014). 12 V. Wang, Y. Kawazoe, and W. Geng, Physical Review B 91, 045433 (2015). 13 P. Srivastava, K. Hembram, H. Mizuseki, K.-R. Lee, S. S. Han, and S. Kim, The Journal of Physical Chemistry C 119, 6530 (2015). 14 X. Peng, Q. Wei, and A. Copple, Physical Review B 90, 085402 (2014). 27 M. Brandbyge, J.-L. Mozos, P. Ordej´on, J. Taylor, and K. Stokbro, Physical Review B 65, 165401 (2002). 28 K. Stokbro, J. Taylor, M. Brandbyge, and P. Ordejon, Annals of the New York Academy of Sciences 1006, 212 (2003). 29 S. Datta, Electronic transport in mesoscopic systems (Cambridge university press, ADDRESS, 1997). 30 R. Landauer, IBM Journal of Research and Development 15 M. Elahi, K. Khaliji, S. M. Tabatabaei, M. Pourfath, and 1, 223 (1957). R. Asgari, Physical Review B 91, 115412 (2015). 16 J.-W. Jiang and H. S. Park, Physical Review B 91, 235118 31 M. Buttiker, Physical review letters 57, 1761 (1986). 32 S. Wang, Physical Chemistry Chemical Physics 13, 11929 (2015). (2011). 17 J. Zhang, H. Liu, L. Cheng, J. Wei, J. Liang, D. Fan, J. Shi, X. Tang, and Q. Zhang, Scientific reports 4, 6452 (2014). 33 Q. Wu, L. Shen, M. Yang, Y. Cai, Z. Huang, and Y. P. Feng, Phys. Rev. B 92, 035436 (2015). 

1701.00760

1

1701

Optical Spectra of p-Doped PEDOT Nano-Aggregates Provide Insight into the Material Disorder

Highly doped Poly(3,4-ethylenedioxythiophene) or PEDOT is a conductive polymer with a wide range of applications in energy conversion due to its ease of processing, optical properties and high conductivity. The latter is influenced by processing conditions, including formulation, annealing, and solvent treatment of the polymer, which also affects the polymer arrangement. Here we show that the analysis of the optical spectra of PEDOT domains reveals the nature and magnitude of the structural disorder in the material. In particular, the optical spectra of objects on individual domains can be used for the elucidation of the molecular disorder in oligomer arrangement which is a key factor affecting the conductivity.

cond-mat.mes-hall

cond-mat

Optical Spectra of p-Doped PEDOT Nano-Aggregates Provide Insight into the Material Disorder David Gelbwaser-Klimovsky,1 Semion K. Saikin,1 Randall H. Goldsmith,2 and Alán Aspuru-Guzik1 1Department of Chemistry and Chemical Biology, Harvard University, Cambridge, MA 02138∗ 2Department of Chemistry, University of Wisconsin-Madison, Madison, WI 53706 Highly doped Poly(3,4-ethylenedioxythiophene) or PEDOT is a conductive polymer with a wide range of applications in energy conversion due to its ease of processing, optical properties and high conductivity. The latter is influenced by processing conditions, including formulation, annealing, and solvent treatment of the polymer, which also affects the polymer arrangement. Here we show that the analysis of the optical spectra of PEDOT domains reveals the nature and magnitude of the structural disorder in the material. In particular, the optical spectra of objects on individual domains can be used for the elucidation of the molecular disorder in oligomer arrangement which is a key factor affecting the conductivity. their conventional inorganic counterparts in a number of energy related applications [1], including photovoltaics [2, 3], fuel cells [4, 5] and low-cost and environmentally-safe thermoelectrics [6–9]. A central feature of many of these devices is a transparent, flexible conductor layer. Among several materials employed for this layer, materials based on poly(3,4-ethylenedioxythiophene) (PEDOT) stand out due to their high hole conductivity, chemical stability, and transparency to visible light [10]. PEDOT-based structures have been incorporated in touch screens, light-emitting diodes, and photovoltaic elements, to mention a few examples [11]. Though multiple experimental investigations have addressed the molecular organization of PEDOT-based materials, the microscopic electronic states that lead to high conductance and the interplay between these states, optical properties, and the material structure has yet to be definitively characterized. The main obstacle to elucidating this relationship is the strong structural disorder that appears on multiple length scales and is highly sensitive to the preparation procedure [10]. In this work, we link the supramolecular packing of PEDOT with its optical absorption spectra. We argue that optical spec- troscopy of PEDOT aggregates on a single domain level can be used for characterization of the microscopic electronic structure in the material that is a function of molecular disorder. Specifically, we show how disorder in two packing models, widely accepted for PEDOT aggregates, leads to a transition between J- and H-aggregation [12, 13] reflected in optical spectra. Finally, we identify the degree of disorder potentially responsible for this transition and study the changes in localization of the bright and dark states, in order to better understand the character of the spectral peaks. As we show below, disorder tends to localized the states and bright states are more delocalized than the dark ones. As a model system, we focus on structures composed of multiply charged oligomers. This model matches the experimentally determined charge density in highly conducting PEDOT. A high-conductance regime is observed for p-doped PEDOT [16] when the injection of positive charges induces a struc- tural relaxation effectively forming polaron and bipolaron states [17]. In this case, PEDOT-based materials consist of short oligomers, see Figure 1(a), bound to counterions that balance the charges of oxidized PEDOT oligomers and affect the struc- tural arrangement of the PEDOT oligomers. The most common forms of high-conducting PEDOT are PEDOT:tosylate and PEDOT:polystyrene sulfonate (PEDOT:PSS) [18]. In the latter form the PSS polymer also plays the role of a structural back- bone. As compared to neutral oligomers, the ground state of p-doped oligomers has a quinoid form [19, 20], which results in a delocalization of the charge carriers over several monomers. During deposition, typically via spincoating, the PEDOT oligomers rearrange, resulting in formation of supramolecular aggregates with a characteristic π − π-stacking intermolecular distance of 0.34 nm.[21] Figure 1(b-c) shows two widely discussed packing motifs – parallel stacks [14] and head-to-tail chains [15]. Both motifs would exhibit similar π − π-stacking distances in x-ray scattering measurements. [21]. These general motifs have also been discussed in several variations including tilted stacking [20] and formation of 3D nanocrystals [21]. The physical reasoning behind the proposed packing begins with π − π interactions that induce a parallel alignment of PEDOT oligomers and result in the formation of oligomer stacks. The electrostatic attraction to a backbone polyelectrolyte chain balances the self-repulsions from positive charge on the oligomers. This interaction tends to minimize the volume-to-surface ratio of the stacks, therefore limiting the stack size. The details of the deposition procedure, including solvent choice and post-processing, affect the balance between these two interactions and thus, controls the morphology of the structure [10, 22]. In contrast with inorganic materials, the interactions mentioned above are relatively weak, frequently leading to high disorder in most conductive organic materials. This disorder influences both the optical spectra [23] and the charge transport properties [24]. For PEDOT-based materials, we can identify several specific types of disorder. First, the length and doping of each oligomer are not well controlled properties, resulting in a distribution of mesoscopic realizations and as a consequence, electronic states. Second, in a pair or stack of PEDOT oligomers, each individual oligomer can be displaced relative to its neighbor along the 7 1 0 2 n a J 3 ] l l a h - s e m . t a m - d n o c [ 1 v 0 6 7 0 0 . 1 0 7 1 : v i X r a ∗Electronic address: dgelbwaser@fas.harvard.edu 2 Figure 1: PEDOT oligomer geometries (a) Optimized structure of a single PEDOT oligomer (some hydrogens not visible) (b-c). Two packing models considered in the study: (b) a stack of parallel oligomers [14] and (c) a double-layer head-to-tail packing [15]. The blue ellipsoid and ribbon represent counterions equilibrating the PEDOT charges. length of the oligomer. Finally, domains with different packing structures, and with a different relative oligomer orientation, can be formed on a larger scale. These structural variations affect both the material conductance and optical spectra [22, 25]. Heavily doped PEDOT is mostly transparent to visible light. However, it has a very broad and structureless absorption spectrum below 1 eV [26]. This spectral feature is associated with the HOMO-1 to HOMO and also HOMO to LUMO transitions of cation oligomers [17]. As compared to isolated PEDOT oligomers, the absorption spectra of PEDOT aggregates are modified by the Coulomb coupling between molecular excitations (excitonic interactions) [27–29] as well as by charge tunneling [30]. Depending on the relative alignment of the molecules, the excitonic interactions can result in the shift of the absorption peak to longer (J-aggregation) or shorter (H-aggregation) wavelengths [12, 31]. Many examples of J-aggregates can be found in nature, where aggregates of pigments play an essential role in light-harvesting and energy transfer in phototrophic organisms [32–34]. To predict and characterize optical properties of PEDOT aggregates we use a theoretical model, where single oligomer prop- erties are computed using Time-Dependent Density Functional Theory (TDDFT), while the emergent aggregate spectra are calculated based on a Hubbard-type model [35]. Details of TDDFT calculations are provided in the Supplementary Information. 3 As compared to previous theoretical studies of PEDOT optical spectra [14, 15, 36], here we focus on the formation of small disordered clusters that include random displacements rather than a periodic crystalline orientation or arrangement of PEDOT oligomers on a polymer backbone. To our knowledge, the role of structural disorder in the optical properties of PEDOT-based materials has not yet been explored in detail. The formation of PEDOT aggregates is driven by inter-oligomer forces. π-π interactions between the aromatic oligomers orients them face-to-face and fixes the perpendicular distance to 0.34 nm as determined by X-ray scattering [21, 37, 38]. The energetic benefit of π-orbital overlap makes it energetically favorable for neighboring monomers to be cofacially aligned in a way that maximizes orbital overlap. Using this tendency as a simplifying assumption, in our models we constrain the relative parallel displacement between oligomers to multiples of the monomer length. Below, we analyze the packing motifs shown in figs. 1(b-c). These models are then used as building blocks for more complicated structures, e.g. 2D films [39] or 3D nanocrystals [21, 40]. The analyzed structures are composed of around 100 oligomers, which is enough to saturate the spectral shifts related to the molecular arrangement size. In the cases where disorder is allowed, we average over 10,000 realizations, which is sufficient to obtain a wide sampling of the possible noise realizations. Considering larger structures or averaging over more realizations introduces only minor changes in the results. Figure 2: Average of DOS (top) and absorption spectra (bottom) over 10,000 random realizations of single chain arrangements as function of the detuning. Zero detuning is the single oligomer transition energy. From left to right we raise the disorder level by increasing the maximum allowed parallel displacement (1,5,9 from left to right). By enlarging the maximum displacement, see insets, the number of possible realizations is escalated and with it the disorder level. This is reflected in the DOS width which grows with the level of disorder. On the other hand the spectra experience a red shifting due to the change in the structural arrangement (see the main text). Even in the case with the largest red shift, right plots, the lowest energy states of the DOS are not bright. Insets: Examples of small parts of realizations of the corresponding disordered single oligomer stack. Parallel Stack of Oligomers. As a first case, we studied a cofacially stacked column of 100 parallel oligomers as shown on 1b. Perfectly aligned oligomers form an H-aggregate, with a blue-shifted spectrum relative to the single isolated oligomer. The introduction of a constant parallel displacement between oligomers results in a transition from an H-aggregate behavior to J-aggregate behavior,[35] see also Supplementary Information for more details. This change is reflected in the initial blue- shift reduction, which eventually turns into a red shift relative to the isolated oligomer as the constant displacement between neighboring oligomers is further enlarged. Nevertheless, even at the maximum possible parallel displacement that maintains at least one monomer overlap, the lowest states do not contribute to the absorption spectra, in contrast to the case of a perfect J-aggregate. We also analyzed the effect of disordered or random parallel displacements that keep the monomers aligned and with at least a one monomer overlap. Each realization consists of a disordered chain composed of 100 parallel oligomers, with a random parallel displacement between them. Figure 2 shows the density of states (DOS) and absorption spectra averaged over 10,000 realizations. To mimic structural disorder, we allow random displacements along the long axis of the oligomer Detuning (eV)Detuning (eV)Detuning (eV)(eV-1)x102(µ2/Νµ02)x10-2Max. Displacement: 1Max. Displacement: 5Max. Displacement: 9Absorption spectrumDOS 4 Figure 3: Average spectra of 10,000 random realizations of the structural arrangement found in Ref. 21 as function of the detuning. This structure is composed of two parallel 13-oligomer chains. From left to right we rise the disorder level by increasing the maximum allowed parallel displacement (1,5,9 from left to right). As a simplifying assumption we assume that the two 13 oligomer chains have the same disorder realization, therefore they are identical. Insets: Examples of corresponding realizations of the structural arrangement proposed in Ref. 21, with disorder included in the form of random parallel displacement between oligomers. over a range of amplitudes in both directions (forward or backward) up to a maximum value. The maximum value for the lowest disorder level is a single monomer displacement. The low disorder level is reflected in the small widths of the DOS and the absorption spectra (see Figure 2) as wells as the spectral blue shift relative to the isolated oligomer. A low disorder chain is similar to the ordered case, with parallel arrangement resulting in a blue-shifted spectra. The disorder level grows by increasing the maximum parallel displacement (see Figure 2-Insets). Larger disorder results in broader spectra and DOS, Figure 2. The increase in the amplitude of random parallel displacements transforms the arrangement into a mixture of a parallel and a head-to-tail chain. In turn, this reduces the spectral blue-shift relative to the isolated oligomer, eventually becoming a red-shifted spectra as the maximum allowed displacement is further increased. In spite of this, these arrangements also fail to form a perfect J-aggregate, because the lowest states do not participate in the absorption spectra. The spectral shifting behavior also arises in more complex models. Delocalization analysis clarifies the nature of the intense bands of the absorption spectra. In particular, we calculate the inverse participation ratio [41] and find that the bright states are consistently more delocalized than the dark states, independent of the oligomer displacement (see Supplementary Information). This behavior also persists in more structurally complicated models. Nevertheless, increasing the degree of disorder increases the localization independently of whether the state is dark or bright, as expected. To test our approach, we used the structural arrangement proposed in Ref. 21 for the PEDOT:PSS micelles in a solid film, where the PEDOT oligomers form nanocrystals in the micelle core surrounded by PSS counterions. The PEDOT oligomers are arranged in two side-by-side stacks, where each stack is composed of a 13 cofacial oligomers. Figure 3 shows the average spectra for the PEDOT nanocrystal for different disorder levels. Similar to the previous model, augmenting the maximum parallel displacement, the spectrum is red shifted. The peak in the absorption spectrum of the most ordered case (maximum displacement of a single monomer) at around 0.6 eV is a consequence of the short length of the chains. For a completely ordered aggregate, this peak corresponds to the next odd delocalized excitation after the brightest state. Head to tail chains We next considered an arrangement consisting of two parallel chains [15] as shown in Figure 1c. Each chain is composed of 50 oligomers in a head-to-tail arrangement. The two chains are cofacially stacked upon each other. The minimal inter-oligomer distance within a chain is limited by the Van der Waals radius, and the interchain distance is 0.34 nm, the same as in the previous model. We first considered two identical and perfectly aligned chains with a minimum distance between oligomers. The absorption spectrum is blue shifted relative to the single monomer spectrum. As we displaced one chain relative to the other, the blue- shifting is reduced, eventually evolving into a red-shifted spectrum, as seen above. A more complex model allows structural disorder due to a random degree of overlap between oligomers in the two cofacially stacked chains that preserves the following constraints: i) Facing monomers are aligned; ii) the minimal head to tail distance is 2.5 A; iii) chain continuity is ensured by requiring that every oligomer overlaps with two oligomers from the second chain. The randomness of alignment allows several arrangements and the results shown are the average over 10,000 realizations (see Figure 4). In contrast to the single cofacial stack of oligomers, the red shifting for the head to tail pair of chains is larger and more narrow. Also here, the lowest states contribute to the absorption spectra, in resemblance of a perfect J-aggregate. As in the single cofacial stack of oligomers, the bright states are delocalized while the dark are more localized (see Supplementary Information.) The insights obtained from the two basic models introduced earlier can help in the examination of the electronic structure of more elaborate arrangements. In order to demonstrate these insights we analyzed a 2D layer, as the one proposed in Ref. 39 for the condensed state of a vapor-phase polymerized PEDOT, where a 2D layer of PEDOT is "sandwiched" by 2D layers of PSS counterions. The structure of the PEDOT layer is composed of pairs of head-to-tail chains. The arrangement of the effective (µ2/Νµ02)x10-2Detuning (eV)Detuning (eV)Detuning (eV)Max. Displacement: 1Max. Displacement: 5Max. Displacement: 9Absorption spectrum5 Figure 4: a) Part of a disordered double chain, shown for illustration purpose (the complete double chain is formed of 100 oligomers). b) Average DOS (top) and spectra (bottom) over 10,000 realizations. In contrast to the other structural arrangements, the lowest energy states of the DOS are bright. dipoles in the 2D layer is shown in Figure 5. Due to the disorder, nearest neighbors may be relatively far away from each other, therefore only extended dipole-dipole interactions are considered. Each 2D layer is composed of 95 oligomers. The DOS and absorption spectrum are shown in figure 5. This structure can be understood as a combination of the parallel oligomers chain and the head-to-tail chains models described above. In contrast to previous models, there is a low energy peak in the DOS. The origin of this red shifted peak is the π − π stacking of more head-to-tail chains that augments the total interaction, red shifting part of the DOS. In agreement with our results on parallel stacks of oligomers above, the lowest states do not contribute to the absorption spectra. This example shows how the intuitions obtained from simple models can be combined to analyze more intricate spectra. A common feature of the studied models is the significant changes on the DOS and absorption spectra caused by the intro- duction of structural disorder. Furthermore, we showed that structural disorder is needed for reproducing the red shifted or low energy absorption spectra seen experimentally. Nevertheless, in general the lowest states do not participate in the absorption spectra. The only exception is the disordered head to tail stack of oligomers, which becomes a perfect J-aggregate in the sense that the lowest states contribute to the energy absorption (see Figure 4). Thin films of PEDOT:PSS show a broad, featureless absorption spectrum below 1 eV [26]. This spectral behavior is in contrast to that of the radical cations and dications in phenyl-capped PEDOT oligomers, as demonstrated via solution-phase spectroelec- trochemistry [42]. These solution phase measurements can be viewed as representative of the absorption spectra of isolated, un-aggregated PEDOT oligomers, essentially a starting point for building a thin film from a molecular building block. Our calculations provide a guide to predict the thin film DOS and optical properties from the properties of the component molecules. Absorption features of the phenyl-capped oligomers showed the expected 1/N length dependence, and extrapolation of dication absorption features of shorter oligomers to the oligomers of length ten considered above suggest a bipolaron absorption feature at 1.2 eV. If disorder is not included, a net blue shift of 1-2 eV is predicted due to the Coulomb interaction, in stark contrast to the experimentally observed thin film spectra. In contrast, inclusion of disorder faithfully reproduces a red shift of bipolaron spectral features, suggesting broad absorption at 1 eV and below in the thin film, as observed. In conclusion, we have shown the magnitude of structural disorder in the overlap between neighboring PEDOT oligomers in pi-pi stacked geometries has a significant impact on electronic absorption spectra. Specifically, the experimentally observed red- shifted absorption spectra relative to isolated PEDOT oligomers could only be obtained if structural order was introduced above a critical displacement. Therefore we conclude that in PEDOT:PSS films, structural disorder is significant and has important contributions to both optical and electronic properties. A variety of structural models were considered and all showed the need for structural disorder in order to reproduce red-shifted spectra. Finally, we point out that though X-ray scattering has provided important clues as to the nature of molecular packing in PEDOT materials, scattering measurements alone cannot be used to (a)(b)Detuning (eV)(eV-1)x102(µ2/Νµ02)x10-2Absorption spectrumDOS6 Figure 5: 2D layer structure (a) and average DOS (top-b) and spectrum (bottom-b) over 10,000 realizations of the disordered 2D layer. There is a low energy peak in the DOS, due to the π-π stacking of head-to-tail chains. Here again the lowest energy states of the DOS are dark. differentiate between different models of structural disorder. Electronic spectroscopy on whole films can provide additional clues, but the significant heterogeneous broadening of spectral features limits the utility of optical spectra. However, single- particle spectroscopy [43, 44] on individual polymer domains [23, 45] can potentially add significant new information on the structural and electronic properties of the PEDOT family of materials. I. ACKNOWLEDGMENTS We acknowledge the support from the Center for Excitonics, an Energy Frontier Research Center funded by the U.S. Depart- ment of Energy under award de-sc0001088 (D. G.-K.,S.S and A. A.-G.) and from the National Science Foundation under award DMR-1610345 (R.H.G.). We appreciate useful discussion with Dr. Dmitrij Rappoport and Dr. Dmitry Zubarev. First principles computations were run on Odyssey cluster of the Harvard University, supported by the Research Computing Group of the FAS Division of Science. (µ2/Νµ02)x10-2(eV-1)x102Detuning (eV)(a)(b)DOSAbsorption spectrumTable I: Properties of neutral, and double-charged PEDOT oligomers computed with TDDFT (B3LYP functional): number of monomers,Nmon, oligomer charge, Q, ground state energy,Eg, energy of the lowest strong electronic excitation, Eex, and the contribution of HOMO-to-LUMO transition into these excitations. 7 Nmon Q [e] Eg [Hartree] Eex [eV] Osc. strength Orbitals 2 4 6 6 8 8 10 10 12 12 -1559.716 -3118.261 -4676.784 -4676.415 -6235.318 -6234.965 -7793.881 -7793.565 -9352.386 -9352.049 98% 99% 98% 99% 97% 99% 95% 98% 93% 98% 0 0 0 2 0 2 0 2 0 2 3.97 2.83 2.39 1.61 2.16 1.28 2.01 1.06 1.94 0.88 0.47 1.28 2.17 2.72 3.03 3.81 3.85 4.66 4.67 5.29 II. OPTICAL SPECTRA OF P-DOPED PEDOT NANO-AGGREGATES PROVIDE INSIGHT INTO THE MATERIAL DISORDER: SUPPLEMENTARY INFORMATION For the TDDFT calculations of single oligomer properties, we used the quantum chemistry package Turbomole, version 6.0. [46]. The geometry optimization was done using a triple-ζ valence-polarization basis set (def2-TZVP [47]) and B3LYP hybrid functional [48]. In order to account for the neutral, polaron and bi-polaron ground states of the oligomers, the structures were optimized with the charges Q = 0, +1, +2, respectively. We also computed several quadruple-charged n = 4 oligomer structures. The effects of the polarizable medium were accounted for using the conductor-like solvation model (COSMO) with effective dielectric constant
= 4.[63] The optimized structures are provided as edtN_n.xyz files, where N is the number of monomers and n is the charge. The five lowest electronic excitations were computed for each oligomer. The influence of the oligomer length on the spectra of electronic excitations was studied in a set of oligomers of size Nmon = 2, 4, 6, 8, 10 and 12 units. We observe a strong electronic transition for neutral and double charged oligomers longer than 4 units, and for quadruple charged oligomers longer than 6 monomer units. This transition occurs mostly between the HOMO and LUMO orbitals of the oligomers. The examined structures span the experimentally determined estimates of oligomer length (6-18 monomer units) [49] and charge density (4 monomers/polaron) [25]. For the studied oligomer lengths, the energy of the transition scales as 1/Nmon, see Figure 6(a), similar to the expected dependence of a particle-in-a-box model assuming that the electron state is delocalized over the conjugated chain [50, 51]. Similar 1/Nmon dependencies have been observed in a variety of experimental investigations of optical properties of solution-phase (isolated) polymers [52–54]. In Table I we collected the total energies, energies of the lowest electronic excitations, oscillator strengths associated with these excitations, and the leading contribution of the molecular orbitals for neutral and doubly charged oligomers. As the oligomers are allowed to increase in size the absorption peak saturates at about 1.5 eV for the neutral oligomers, and the double and quadruple charged molecules exhibit transition frequencies significantly below 1 eV. In Figure 7 we show computed spectra of six oligomers with net charges Q = 0, +2. As demonstrated in previous calculations, [17] the single-charged chains exhibit two transitions, a weak transition composed mostly of HOMO-1 - HOMO orbitals in the sub-eV range and a stronger HOMO - LUMO transition (data not shown). To verify that a B3LYP hybrid functional gives correct estimates for delocalized electronic excitations in the studied oligomers, [55, 56] we also computed the excitation energies and the transition dipoles with a range-separated functional wB97XD [57] and the basis set 6-31+G* as implemented in Gaussian 09 [58]. The computed results (open circles and open triangles in Figure 6), show linear trends similar to the ones obtained for B3LYP with a systematic shift. In particular, the transition energies of the bipolarons, the critical species in this work, show little difference between the two functionals. We expect that the deviations due to long-range corrections[55] will be more pronounced for longer oligomers, but this work only considered the short oligomers that are present in PEDOT:PSS. The electronic excitation spectra of molecular aggregates are significantly modified as compared to the isolated molecules that compose these aggregates. The main sources of the modifications are the Coulomb coupling between electronic excitations, including both the Förster [28] and the Dexter exchange [29] interactions, formation of charge-transfer excitations or delocal- ized hole bands [30], and changes in the dielectric properties of the local environment. The Coulomb interaction between the transition dipole moments, specifically the Förster coupling, can be strong in PEDOT aggregates. The transition dipole scales approximately linearly with the number of monomers, reflecting a delocalization length of the electronic excitations over the length of the oligomer, see Figure 6(b). For a double-charged molecule composed of ten monomers the value of the transition dipole is about 30 Debye. For a more generalized characterization of the Coulomb coupling between a pair of oligomers we 8 Figure 6: Computed transition energies (a) and the transition dipoles (b) of neutral and double-charged PEDOT oligomers (bi-polaron and two bi-polaron) as functions of the oligomer length. computed the Förster and Dexter couplings as VF = VD = d3x d3x (cid:90) (cid:90) (cid:90) (cid:90) d3y ψH1 (x)ψL2(y) d3y ψL2 (x)ψH1(y) e2 x − y ψL1(x)ψH2(y), e2 x − y ψL1(x)ψH2(y), (1) (2) where the indices Hi and Li correspond to the HOMO and LUMO orbitals of an i-th molecule [35]. Equation 1 describes the Förster interaction beyond the dipole approximation as a Coulomb transition between the initial ψL1(x)ψH2 (y) and the final ψH1(x)ψL2 (y) states of a two-oligomer system. In Equation 2, the transition involves tunneling of electrons between oligomers, and this process requires an overlap of electronic wavefunctions. For all intermolecular distances relevant to our system, the Dexter interaction between the molecules is less than 5% of the Förster coupling. Figure 8 shows a map of the Förster coupling as a function of the relative spatial positions of two 10-monomer oligomers. The negative sign of the Förster 9 Figure 7: Electronic excitation spectra of (a) neutral and (b) double-charged PEDOT oligomers (bi-polaron) composed of Nmon monomers. The transition frequencies within the 0 − 5 eV range, red lines, are computed using TDDFT, B3LYP functional. The empirically broadened spectra with a 100 meV linewidth are show in blue. Figure 8: A 2D map of the Förster interaction between two 10-monomer PEDOT chains parallel displaced relative to each other (see inset). The axis values refer to center-center distances. The color scheme shows the intermolecular coupling in meV. coupling corresponds to the J-aggregation of the molecules, while the positive sign of the coupling indicates H-aggregation, with the sign of the coupling directly linked to the expected ordering and symmetry of the new J- or H-aggregated hybrid orbitals. In general, if the oligomers are aligned in parallel the excitonic coupling between the oligomers leads to a blue-shift of the collective absorption peak. Only oligomers with a large relative displacement (more than 2/3 of the oligomer length) along the x-axis show J-aggregation in the absorption spectra, with a consequent red-shift of the absorption peak. It should be noted that due to delocalization of the electronic excitations the nearest neighbor Förster interaction between oligomers decreases with the length of the chains,[59, 60] which may result in reduced Förster coupling for sufficiently long chains. We observe this effect qualitatively, with a 10-monomer chain showing an interaction subtly reduced from the 8-monomer chain. However, at the short oligomer lengths determined in experimental characterizations of PEDOT:PSS, we expect this effect to have only a minor effect. In order to accelerate the computation of the electronic excitation spectra of aggregated oligomers, we treated the Förster 10 Figure 9: Average IPR as a function of the square dipole strength of 10,000 realizations of single-chain structures. The error bars indicate the mean absolute deviation from the average number. coupling beyond nearest neighbor interactions with an extended dipole model [35, 61] (cid:18) 1 (cid:19) V ed F = keq2 + 1 r−− − 1 r+− − 1 r−+ r++ , (3) where q is the effective dipole charge and rij is the distance between the i charge on one oligomer and the j charge on the another oligomer and ke is the Coulomb constant. The Förster interaction 2D map is used to calibrate the effective dipole charge as well as its length. This information together with a model of the structural arrangement of the oligomers is enough to construct the collective Hamiltonian and the electronic absorption spectrum [35]. Charge transfer (CT) excitations in conducting polymers have been observed previously [30] and analyzed theoretically [62]. Usually, the CT absorption line is red-shifted as compared to the intramolecular excitations. These transitions are allowed if there is a sufficient overlap between the electron wavefunctions of different oligomers. The lowest order term contributing to the CT transition in an oligomer dimer is due to the excitation from a HOMO orbital of one oligomer to a LUMO orbital of another oligomer. For two parallel-stacked 10-monomer units displaced by 0.34 nm the CT transition dipole, calculated as a matrix element of a dipole operator taken between HOMO and LUMO orbitals of adjacent oligomers, the interaction strength is µCT = 1.4 D. This value is about one order of magnitude smaller that the transition dipoles for intramolecular excitations. Previous theoretical studies of delocalized electronics states in P3HT also showed that the intensities of charge- transfer excitations are small compared to the intramolecular excitations [62]. While we expect that the far infrared part of the absorption spectra of PEDOT aggregates will have significant contribution from CT excitations, the near infrared and visible parts of the absorption spectrum can be largely attributed to the Coulomb interaction between the oligomers. In order to characterize the states delocalization we calculate the inverse participation ratio (IPR), which is a standard measure States delocalization of delocalization [41]. For an exciton state of the form ψ(cid:105) =(cid:80)N i λii(cid:105), the IPR is defined as N(cid:88) IP R = λ4 i . (4) i=1 The IPR ranges from 1/N for a totally delocalized state state to 1 for a totally localized state. Below we plot the IPR as function of the square dipole strength, which determines the oscillator strength for each spectral feature, for different structures composed of 100 oligomers. The dipole strength is normalized by the single oligomer dipole strength, therefore it can range from zero for a completely dark state to 100 for a maximally bright state. A dipole strength of one corresponds to a single oligomer. The IPR ranges from 1/100 to 1. Figure 9 shows the average IPR for 10,000 realizations of the single-chain arrangements with a maximum displacements of 1,5 and 9 (from left to right). The error bars indicate the mean absolute deviation from the average. Figure 10 shows the IPR for the double chain structure. A common feature between all structures is that the brightest states are more delocalized than the dark states. This trend includes the case of the single chain with a maximum monomer displacement of one. Even though its IPR is quite flat compared to the other structures, there is a small slope for the bright states. As expected, IPRMax. Displacement: 1Max. Displacement: 5Max. Displacement: 911 Figure 10: Average IPR as a function of the square dipole strength of 10,000 realizations of double-chain structures. The error bars indicate the mean absolute deviation from the average number. we see also that disorder localizes the states. While the most ordered structure, the single chain with a maximum displacement of 1, has a IPR around 0.02, the others demonstrate IPR values that are much higher. [1] K. Sun, S. Zhang, P. Li, Y. Xia, X. Zhang, D. Du, F. H. Isikgor, and J. Ouyang, J Mater Sci : Mater Electron 26, 4438 (2015). [2] A. Facchetti, Chem Mater 23, 733 (2011), URL http://dx.doi.org/10.1021/cm102419z. [3] D. J. Lipomi, B. C.-K. Tee, M. Vosgueritchian, and Z. Bao, Advanced Materials 23, 1771 (2011), ISSN 1521-4095, URL http: //dx.doi.org/10.1002/adma.201004426. [4] Z. Guo, Y. Qiao, H. Liu, C. Ding, Y. Zhu, M. Wan, and L. Jiang, J Mater Chem 22, 17153 (2012). [5] B. Winther-Jensen, O. Winther-Jensen, M. Forsyth, and D. R. MacFarlane, Science 321, 671 (2008). [6] N. Dubey and M. Leclerc, Journal of Polymer Science Part B: Polymer Physics 49, 467 (2011), URL http://dx.doi.org/10. [7] H. Wang, U. Ail, R. Gabrielsson, M. Berggren, and X. Crispin, Advanced Energy Materials 5, 1500044 (2015), URL http://dx. 1002/polb.22206. doi.org/10.1002/aenm.201500044. [8] Q. Wei, M. Mukaida, K. Kirihara, Y. Naitoh, and T. Ishida, Materials 8, 732 (2015). [9] O. Bubnova, Z. U. Khan, A. Malti, S. Braun, M. Fahlman, M. Berggren, and X. Crispin, Nat Mater 10, 429 (2011), URL http: //dx.doi.org/10.1038/nmat3012. [10] L. Groenendaal, F. Jonas, D. Freitag, H. Pielartzik, and J. R. Reynolds, Advanced Materials 12, 481 (2000), URL http://dx.doi. org/10.1002/(SICI)1521-4095(200004)12:7<481::AID-ADMA481>3.0.CO;2-C. [11] W. Lövenich, Polymer Science, Ser. C 56, 135 (2014). [12] F. C. Spano and C. Silva, Annu Rev Phys Chem 65, 477 (2014). [13] S. Siddiqui and F. Spano, Chem Phys Lett 308, 99 (1999). [14] A. Dkhissi, D. Beljonne, R. Lazzaroni, F. Louwet, and B. Groenendaal, Theor Chem Acc 119, 305 (2008). [15] A. Lenz, H. Kariis, A. Pohl, P. Persson, and L. Ojamäe, Chemical Physics 384, 44 (2011), ISSN 0301-0104, URL http://www. sciencedirect.com/science/article/pii/S0301010411001571. [16] L. Groenendaal, G. Zotti, P. H. Aubert, S. M. Waybright, and J. R. Reynolds, Advanced Materials 15, 855 (2003). [17] J. L. Brédas and G. Street, Acc. Chem. Res. 1305, 309 (1985). [18] A. Elschner, S. Kirchmeyer, W. Lovenich, U. Merker, and R. Knud, PEDOT: Principles and Applications of an Intrinsically Conductive Polymer (CRC Press, Taylor & Francis Group, 2011). [19] A. Dkhissi, F. Louwet, L. Groenendaal, D. Beljonne, R. Lazzaroni, and J. L. Brédas, Chem Phys Lett 359, 466 (2002). [20] E. G. Kim and J. L. Brédas, J Am Chem Soc 130, 16880 (2008). [21] T. Takano, H. Masunaga, A. Fujiwara, H. Okuzaki, and T. Sasaki, Macromolecules 45, 3859 (2012). [22] J. Gasiorowski, R. Menon, K. Hingerl, M. Dachev, and N. S. Sariciftci, Thin Solid Films 536, 211 (2013). [23] A. Thiessen, J. Vogelsang, T. Adachi, F. Steiner, D. V. Bout, and J. M. Lupton, Proceedings of the National Academy of Sciences 110, E3550 (2013). [24] J. T. Friedlein, S. E. Shaheen, G. G. Malliaras, and R. R. McLeod, Advanced Electronic Materials 1, 1500189 (2015), ISSN 2199-160X, 1500189, URL http://dx.doi.org/10.1002/aelm.201500189. [25] X. Crispin, S. Marciniak, W. Osikowicz, G. Zotti, A. van der Gon, F. Louwet, M. Fahlman, L. Groenendaal, F. De Schryver, and W. R. Salaneck, J. Polym. Sci. B Polym. Phys. 41, 2561 (2003). [26] N. Massonnet, A. Carella, O. Jaudouin, P. Rannou, G. Laval, C. Celle, and J.-P. Simonato, Journal of Materials Chemistry C 2, 1278 IPR12 (2014), URL http://pubs.rsc.org/en/content/articlehtml/2014/tc/c3tc31674b. [27] T. Förster, Annalen der physik 437, 55 (1948). [28] T. Förster, in Modern Quantum Chemistry, edited by O. Sinanoglu (Academic Press, 1965), chap. III B 1, pp. 93–137. [29] D. L. Dexter, J. Chem. Phys. 21, 836 (1953), URL http://scitation.aip.org/content/aip/journal/jcp/21/5/10. 1063/1.1699044. [30] R. Österbacka, C. P. An, X. M. Jiang, and Z. V. Vardeny, Science 287, 839 (2000). [31] S. K. Saikin, A. Eisfel, S. Valleau, and A. Aspuru-Guzik, Nanophotonics 2, 21 (2013). [32] B. . Green and W. W. Parson, eds., Light-Harvesting Antennas in Photosynthesis. (Kluwer, Dordrecht, 2003). [33] R. Croce and H. van Amerongen, Nat Chem Biol 10, 492 (2014), URL http://dx.doi.org/10.1038/nchembio.1555. [34] J. Huh, S. K. Saikin, J. C. Brookes, S. Valleau, T. Fujita, and A. Aspuru-Guzik, J Am Chem Soc 136, 2048 (2014). [35] S. Valleau, S. K. Saikin, M.-H. Yung, and A. A. Guzik, J. Chem. Phys. 137, 034109 (2012). [36] R. Gangopadhyay, B. Das, and M. R. Molla, RSC Advances 4, 43912 (2014), URL http://dx.doi.org/10.1039/C4RA08666J. [37] K. Aasmundtveit, E. Samuelsen, L. Pettersson, O. Inganäs, T. Johansson, and R. Feidenhans, Synth Met 101, 561 (1999). [38] S. E. Atanasov, M. D. Losego, B. Gong, E. Sachet, J.-P. Maria, P. S. Williams, and G. N. Parsons, Chem Mater 26, 3471 (2014). [39] D. C. Martin, J. Wu, C. M. Shaw, Z. King, S. A. Spanninga, S. Richardson-Burns, J. Hendricks, and J. Yang, Polymer Reviews 50, 340 [40] B. Cho, K. S. Park, J. Baek, H. S. Oh, Y.-E. Koo Lee, and M. M. Sung, Nano Lett 14, 3321 (2014). [41] C. Smyth, F. Fassioli, and G. D. Scholes, Phil. Trans. R. Soc. A 370, 3728 (2012). [42] R. Martin, U. Gubler, J. Cornil, M. Balakina, C. Boudon, C. Bosshard, J. Gisselbrecht, F. Diederich, P. GÃijnter, M. Gross, et al., Chem (2010). –Eur J 6, 3622 (2000). [43] K. D. Heylman, K. A. Knapper, and R. H. Goldsmith, J. Chem. Phys. 5, 1917 (2014). [44] K. A. Knapper, K. D. Heylman, E. H. Horak, and R. H. Goldsmith, Advanced Materials (2016). [45] P. F. Barbara, A. J. Gesquiere, S.-J. Park, and Y. J. Lee, Acc Chem Res 38, 602 (2005). [46] R. Ahlrichs, M. Bär, M. Häser, H. Horn, and C. Kölmel, Chem. Phys. Lett. 162, 165 (1989), URL http://www.sciencedirect. com/science/article/pii/0009261489851188. [47] F. Weigend and R. Ahlrichs, Phys. Chem. Chem. Phys. 7, 3297 (2005), URL http://dx.doi.org/10.1039/B508541A. [48] A. D. Becke, J. Chem. Phys. 98, 5648 (1993), URL http://scitation.aip.org/content/aip/journal/jcp/98/7/10. 1063/1.464913. [49] S. Kirchmeyer and K. Reuter, J. Mater. Chem. 15, 2077 (2005), URL http://dx.doi.org/10.1039/B417803N. [50] H. Kuhn, Ang. Chem. 71, 93 (1959), ISSN 1521-3757, URL http://dx.doi.org/10.1002/ange.19590710302. [51] F. Bär, W. Huber, G. Handschig, H. Martin, and H. Kuhn, J. Chem. Phys. 32, 470 (1960), URL http://scitation.aip.org/ content/aip/journal/jcp/32/2/10.1063/1.1730718. [52] J. L. Brédas, R. Silbey, D. S. Boudreaux, and R. R. Chance, J. Am. Chem. Soc. 105, 6555 (1983), http://dx.doi.org/10.1021/ja00360a004, URL http://dx.doi.org/10.1021/ja00360a004. [53] L. M. Tolbert, Acc. Chem. Res. 25, 561 (1992), http://dx.doi.org/10.1021/ar00024a003, URL http://dx.doi.org/10.1021/ ar00024a003. Chem. Eur. J. 8, 2384 (2002). [54] J. J. Apperloo, L. B. Groenendaal, H. Verheyen, M. Jayakannan, R. A. J. Janssen, A. Dkhissi, D. Beljonne, R. Lazzaroni, and J.-L. Brédas, [55] I. H. Nayyar, E. R. Batista, S. Tretiak, A. Saxena, D. L. Smith, and R. L. Martin, J. Phys. Chem. Lett. 2, 566 (2011). [56] U. Salzner and A. Aydin, J. Chem. Theory Comput. 7, 2568 (2011). [57] J.-D. Chai and M. Head-Gordon, Phys. Chem. Chem. Phys. 10, 6615 (2008). [58] M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, G. Scalmani, V. Barone, B. Mennucci, G. A. Petersson, et al., Gaussian 09 Revision D.01, gaussian Inc. Wallingford CT 2009. [59] E. S. Manas and F. C. Spano, J. Chem. Phys. 109, 8087 (1998), URL http://scitation.aip.org/content/aip/journal/ jcp/109/18/10.1063/1.477457;jsessionid=-RiU2J8FsUSK3U2ZTFbF9XbM.x-aip-live-03. [60] W. Barford, J. Chem. Phys. 126, 134905 (2007). [61] T. Kobayashi, ed., J-Aggregates (World Scientific, 1996). [62] D. Beljonne, J. Cornil, H. Sirringhaus, P. J. Brown, M. Shkunov, R. H. Friend, and J.-L. Brédas, Adv. Funct. Mater. 11, 229 (2001). [63] Variation of
between 1 and 4 results in a small shift in transition frequencies. 

1806.01548

1

1806

Electromagnetic effects induced by time-dependent axion field

We studied the dynamics of the so-called $\theta$-term, which exists in topological materials and is related to a hypothetical field predicted by Peccei-Quinn in particle physics, in a magnetic superlattice constructed using a topological insulator and two ferromagnetic insulators, where the ferromagnetic insulators had perpendicular magnetic anisotropies and different magnetic coercive fields. We examined a way to drive the dynamics of the $\theta$-term in the magnetic superlattice through changing the inversion symmetry (from an anti-parallel to a parallel magnetic configuration) using an external magnetic field. As a result, we found that unconventional electromagnetic fields, which are magnetic field-induced charge currents and vice versa, are generated by the nonzero dynamics of the $\theta$-term.

cond-mat.mes-hall

cond-mat

Electromagnetic effects induced by time-dependent axion field Katsuhisa Taguchi,1 Tatsushi Imaeda,2 Tetsuya Hajiri,3 Takuya Shiraishi,4 Yukio Tanaka,2 Naoya Kitajima,4 and Tatsuhiro Naka4, 5 1Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan 2Department of Applied Physics, Nagoya University, Nagoya 464-8603, Japan 3Department of Materials Physics, Nagoya University, Nagoya 464-8603, Japan 4Department of Physics, Nagoya University, Nagoya 464-8602, Japan 5Kobayashi Masukawa Institute for the Origin of Particles and the Universe, Nagoya 464-8602, Japan (Dated: June 6, 2018) We studied the dynamics of the so-called θ-term, which exists in topological materials and is re- lated to a hypothetical field predicted by Peccei-Quinn in particle physics, in a magnetic superlattice constructed using a topological insulator and two ferromagnetic insulators, where the ferromagnetic insulators had perpendicular magnetic anisotropies and different magnetic coercive fields. We ex- amined a way to drive the dynamics of the θ-term in the magnetic superlattice through changing the inversion symmetry (from an anti-parallel to a parallel magnetic configuration) using an external magnetic field. As a result, we found that unconventional electromagnetic fields, which are magnetic field-induced charge currents and vice versa, are generated by the nonzero dynamics of the θ-term. Introduction.– In the context of high-energy physics, an axion is an additional scalar degree of freedom, which gives a natural solution to the charge conjugation parity problem in the standard model of particle physics[1]. In the cosmological context, it can also take the role of dark matter in the present universe, and the axion detection experiment is currently one of the most exciting fields of study. The axion couples with electromagnetic fields E and B as follows: La = α 2π g a(t) fa E · B, (1) where α is a fine structure constant, g is a coefficient of O(1) [2–5], a(t) is the (time-dependent) axion field, and fa is the so-called Peccei-Quinn scale. Because of an extremely small value of a(t)/fa . 10−19 in the present universe, it is difficult to detect the signature of a dark- matter axion. Intriguingly, a Lagrangian similar to Eq. (1) can be realized in topological materials: Lθ = e2 2πh θE · B, (2) where e is an elementary charge, and the term pro- portional to θ is the so-called θ-term. This is analo- gous to a(t)/fa in Eq. (1), but θ is basically a static constant in the time-reversal symmetry. The finite θ can be realized in the family of topological insulators (TIs), which includes magnetic-doped TIs[6, 7], multilay- ers of magnetic TIs[8, 9], Weyl semimetals (WSs)[10–13], and superlattices[14–16]. In the presence of θ, charac- teristic electromagnetic effects, [16–23], anomalous Hall effects[24–28], chiral magnetic effects[29–37], and Kerr effects[38, 39], have been intensively studied. One of the most interesting physical phenomena driven by the θ-term is the electromagnetic effect via the dynam- ics of θ (i.e., ∂tθ), which could be analogous to the axion. So far, the unconventional optical effect [40] and electric field-induced magnetic field [41] have been discussed un- der a nonzero ∂tθ, whose dynamics is caused by magnetic fluctuations, in materials with breaking time-reversal and inversion symmetries. Then, the time-average of ∂tθ be- comes zero, and its manipulation by an external field might be difficult. In this letter, we discuss a way to drive the dynamics of the θ-term using an external magnetic field and con- sider the electromagnetic effects via ∂tθ in a magnetic superlattice while breaking both the time- and inversion- reversal symmetries. The magnetic superlattice we con- sider is constructed using a TI and two ferromagnetic insulators (FIs) [FI1/TI/FI2/spacer]n [Fig. 1(a)], where FI1 and FI2 have perpendicular magnetic anisotropies and different magnetic coercive fields. Here, to clearly define the θ-term, we consider the axion insulator phase realized in the FI/TI superlattice [Fig. 1(b)]. Then, θ can correspond to magnetic configurations of the FIs, where the inversion-reversal symmetry is preserved or broken corresponding to parallel (P) or anti-parallel (AP) magnetic configurations. These configurations could be controlled by an external magnetic field because of the different magnetic coercive fields. Through this control, the nonzero ∂tθ is induced during the AP→P process. Furthermore, we found unconventional electromagnetic effects under a nonzero ∂tθ. Unlike the conventional (static) electromagnetic effects, a dynamical magnetic field-induced charge current is generated and vice versa. Model.– We start with a model in the superlattice constructed by [FI1/TI/FI2/spacer]n as shown in Fig. 1(a). Its effective Hamiltonian model can be described 2 and sx,y,z (τx,y,z) are a 2 × 2 identical matrix and Pauli matrices acting on the spin (the surface) degrees of freedom, respectively. M indicates the exchange interaction of the FIs, where M1 and M2 correspond to the z-components of the magnetizations of FI1 and FI2, respectively. ℓ = L/N is the superlattice period, and L is the length of the whole layer. In the limit N → ∞, the effective Hamiltonian is also described by a Fourier transformed form as follows [14, 42]: H =Z dkψ† k" 5 Xn=1 dn(k)Γn + M+τ0sz# ψk (9) with dn=1,2,3,4,5 = [−vkx, vky, −∆M sin kzℓ, ∆T + ∆M cos kzℓ, M−], , k = (kx, ky, kz), M± ≡ (M1 ± M2)/2, and Γn=1,2,3,4,5 = (τzsy, τzsx, τys0, τxs0, τzsz), which is defined by {Γn, Γm} = 2δnm. (Color online) (a) Superlattice constructed by FIG. 1. [FI1/TI/FI2/spacer]n . (b) Phase diagram of superlattice, which is categorized by an order parameter O = M1M2 with several tunneling parameter ∆± = ∆T ± ∆M. In the AI phase −), the θ-term is given by θ = tan−1 (d5/d4) within (O < ∆2 the lowest order of M+. (c) Illustration of the magnetic hys- teresis loops and magnetic configurations of the superlattice with different magnetic coercive fields (b1 and b2) under an external magnetic field Bex along the layered direction. M1 and M2 correspond to the magnetizations of the FIs. The an- tiparallel (AP) and parallel (P) magnetic configurations can be generated by Bex though 1 → 2 → 3 (red arrows) and 3 → 4, respectively. (d) Time-dependence of M2(t) by the external magnetic field Bex (inset) during the 3 → 4 process. (e) Time-dependence of the dynamical θ-term with several M1/M2(t = 0) at M1 = 0.01 eV. by [14, 42] = ∆2 M + ∆2 − + ∆2 kz ]2 with ∆2 kz +, respectively, where O ≡ M 2 k + [M+ ±qM 2 The dispersion of the Hamiltonian is given by ǫ2(k) = 2v2k2 T + 2∆M∆T cos kzℓ. Based on this, the gap-full state, i.e., ax- ion insulator (AI) and quantum anomalous Hall (QAH) systems are characterized by the conditions O > ∆2 − and O > ∆2 − = M1M2 is an order parameter in the superlattice [Fig. 1(b)]. The gapless phase (WS phase) is realized for ∆2 − < O < ∆2 +. The θ-term was studied in the AI phase and WS phase. The former is clearly described by δθ = tan−1(d5/d4) within the lowest order of M+ [Fig. 1(b)], and the lat- ter is given by the distance of each point-node of the dispersion[14]. In the following, to discuss the dynamical θ-term clearly, we focus on the AI phase. + − M 2 Electromagnetic effect via δθ.– The electromagnetic ef- fects in the AI phase can be given by[17–20, 43, 44] L= LMaxwell + Lθ + Le H = N Xn,mZ dkkψ† kk,nHnm(kk)ψkk,m, Hnm(kk) = h0(kk)δn,m + h1δn,m+1 + h2δn,m−1, (4) h0(kk) = vτz(σ × kk)z + ∆Tτxs0 + M, (5) h1 = h2 = 1 2 1 2 ∆M(τx + iτy)s0, ∆M(τx − iτy)s0, M = − 1 2 M1(τ0 + τz)sz − 1 2 M2(τ0 − τz)sz, (6) (7) (8) ψ† kk,n operator −,↑,kk,n ψ† the +,↑,kk,n ψ† creation +,↓,kk,n ψ† where = (ψ† −,↓,kk,n) of an elec- tron, where ↑ and ↓ are the spin indices and + and – indicate the top and bottom of the same TI layer, respectively. The labels n and m indicate the TI layers, and kk and v are the momentum and velocity of the surface states of a TI layer, respectively. ∆T (∆M) denotes the tunneling between the top and bottom of the same TI layer (of neighboring TI layers). s0 (τ0) (3) = 1 2(cid:2)ǫE2 − 1 µ B2(cid:3) + e2 2πh (θ0 + δθ)E · B + je · A.(10) Here, Le denotes the gauge coupling between charge cur- rent je and vector potential A. ǫ and µ are the electrical permittivity and magnetic permeability of the medium, respectively. θ0 (= 0 or π) is a static value and δθ denotes the deviation from θ0 [40–42, 45, 46]. Below, we simply consider δθ in the linear response of d5 and d4 ∼ ∆+. Then, δθ is given by [34, 42] ∆+(cid:21). δθ = tan−1(cid:20) M− The Maxwell equations are given by ∇ × E = −∂tB ∇ × B/µ = ǫ∂tE + je + jθ, (11) (12) where jθ = −[(∇δθ) × E + (∂tδθ)B]cǫα/π is a charge current due to δθ, and c is the velocity of light. Then, the wave equation becomes (∂2 t − c2∇2)B = 1 ǫ ∇ × (je + jθ). (13) The right-hand side of the above equation indicates the source term for the wave equation. The first term is conventional but the second term, which de- is unconventional and is pends on ∇δθ and ∂tδθ, given by ∇ × jθ/ǫ = − cα π [(∇ · E)∇δθ − (∇δθ) · ∇E] − cαµ π ∂tδθ (ǫ∂tE + je + jθ) . In particular, for ∇δθ = 0 and ∇ × je = 0, the wave equation becomes (∂2 t − c2∇2)B = − cαµ π ∂tδθ (ǫ∂tE + je + jθ) . (14) Dynamics of δθ.– To drive the nonzero dynamics of θ (∂tδθ), we focus on magnetic configurations in the su- perlattice. Here, we assume that the magnetizations of FI1 and FI2 have perpendicular magnetic anisotropies and different magnetic coercive fields b1 and b2, as illus- trated in Fig. 1(c). Then, an external magnetic field Bex = Bex z along the layered direction can trigger a change in the magnetic configurations ( z is a unit vec- tor). For example, first we set the AP magnetic configu- ration, which can be generated through the 1 → 2 → 3 process, with b1 < Bex < b2, as shown in Fig. 1(c). Then, the AP magnetic configuration becomes stable even when Bex = 0. After Bex 6= 0 is applied, the M2 values of the FIs are changed by Bex, and the magnetic configuration can be dynamically changed from the AP configuration to the P magnetic configuration via the 3 → 4 process. The changes in the magnetic configurations are caused by the change in the magnetization of FI2, which is switched from the −z to the +z direction during 3 → 4. The above process could drive the nonzero ∂tδθ. The magnetization switch can be phenomenologically understood using the Bloch equation for the magneti- zation motion as ∂tM2(t) = [M2(t = 0) − M2(t)]/τf, where τf is the characteristic relaxation time of the mag- netization switching, which depends on the material. To solve this equation, we assume the initial condition M2(t = 0) = −M2 with M2 > 0, as illustrated in Fig. 1(d). Then, we have M2(t) = M2[1 − 2 exp (−t/τf)] for M−(t) = [M1 − M2(t)]/2 [Fig. 1(d)]. As a result, ∂tδθ is given by Eq. (11) as follows: ∂tδθ(t) = ∆+M2 exp (−t/τf) ∆2 + + M 2 −(t) 1 τf . (15) Using the ∂tδθ [Fig. 1(e)], we consider the following three cases. Below, we simply focus on the time-dependence of B and ∂tδθ, and we ignore c2∇2B and set ∇δθ = 0. ∂tδθ-induced magnetic field and charge current.– We consider the responses by the dynamical θ term ∂tδθ(t) during the 3 → 4 process under a static external magnetic field Bex = Bex z, which is applied at t = 0. Here, we focus on the unconventional magnetic field [Bθ(t)] via a nonzero ∂tδθ, which can be defined and decomposed as B(t) ≡ Bex + M+(t) + Bθ(t). Then, the time-revolution of Bθ(t) is given from ∂2 B(t): π ∂tδθ(cid:3)2 α2 π2 (∂tδθ)2(Bθ + Bex + M+) − ∂2 t B(t) =(cid:2) α ∂2 t Bθ = t M+. (16) 3 To solve this, we use the following final and initial con- ditions: Bθ(t → +∞) = 0 ∂tBθ(t = 0) = −∂tM+(t)t→0, (17) where the former indicates the removal of a divergent solution, and the latter is determined from ∂tB(t) = −∇ × E at t = 0, while assuming ∇ × E → 0 at t = 0. Figure 2 illustrates the dynamical θ-term that induces magnetic field Bθ. The nonzero Bθ is only generated dur- ing the 3 → 4 process. Furthermore, from Eq. (12), the ∂tδθ couples with the external magnetic field and drives the charge current jθ = −[∂t(δθ)]Bcǫα/π only during 3 → 4, as described in Fig. 2. This magnetic-field in- duced charge current can be regarded as a kind of chiral magnetic effect[29–37]. Under charge current along Bex.– Next, we discuss the magnetic field via ∂tδθ in the presence of both the mag- netic field Bex = Bex z and the charge current je = je z, which is spatially uniform. Then, the time-revolution of Bθ = Bθ z becomes α2 π2 (∂tδθ)2(Bθ + Bex + M+) − ∂2 cαµ0 t M+ ∂tδθje. π (18) !! t ∼ τf … !! t ≫ τf … ∂2 t Bθ = − (a) t ≤ 0 … FI1 TI FI2 spacer … z B = M + j θ B θ(t) … … B = B θ + Bex + M + B = Bex + M + 0.4 !"& 0.3 !"% 0.2 !"$ !"# 0.1 x e B / ) t ( θ B 0 !"! 0 ! (b) M1/M2(t = 0) M1/M2(t = 0) 0.5 1.0 2.0 (c) M1/M2(t = 0) M1/M2(t = 0) 0.5 1.0 2.0 ) ) 2 2 m m / / A A 2 2 0 0 1 1 × × ( ( θ θ j j 12 &"!! 10 &!!! 8 %!! 6 $!! 4 #!! 2 "!! 0 ! !! ! !! 4 ! !! ! 8 ! !! !" ! t/τf 0 ! !! ! !! 4 ! !! ! t/τf !! 8 ! !" ! (Color online) Dynamical θ term induces magnetic FIG. 2. field Bθ and charge current jθ. (a) Illustration of Bθ (red arrows) and jθ (orange arrows) in the presence of the dynam- ical θ term during the 3(t ≤ 0) → 4(t ≫ tf) process. Bex is an external magnetic field and M+ = [M1 +M2(t)] z corresponds to the total magnetization of the FIs. (b) Time-dependence of Bθ and (c) time-dependence of jθ. We use the parameters ∆+ = 0.1 eV, M1 = 10 meV, τf = 1 ns, ǫ = 8.8 × 10−12F/m, and Bex = 0.1 T. 0.2 !"%! (a) (b) x e B / ) t ( θ B !"$# 0.1 !"$! !"!# 0.0 !"!! je = 0 je = 107A/m2 je = 108A/m2 je V 0 !"!! "!"!$ x e B / ) t ( θ B "!"#! −0.1 je = 0 −= 107A/m2 je = 10 = 108A/m2 je = 10 − je V Bθ "!"#$ Bθ 0 ! ! 4 ! ! ! ! t/τf 8 ! ! ! ! 0 !! ! !! 4 ! !! ! t/τf 8 ! !! !" ! FIG. 3. (Color online) Time-dependence of magnetic field Bθ ≡ Bθ(je = 0) + Bθ(je 6= 0) due to ∂tδθ in presence of external magnetic field Bex and charge current je = je z along layered direction in (a) je > 0 and (b) je < 0, with several je at M1/M2(t = 0) = 1. The parameters are the same as those used in Fig. 2. Bθ ≡ Bθ(je = 0) + Bθ(je 6= 0) is generated not only by the coupling between ∂tδθ and Bex but also by the coupling between ∂tδθ and je. The former Bθ(je = 0) is given in Fig. 2. The latter Bθ(je 6= 0) will be dominant when the magnitude of je is sufficiently large (Fig. 3). It should be noted that during the 3 → 4 process, the charge current via the chiral magnetic effect jθ is also generated, and the charge current je includes jθ as je → je + jθ. However, the magnitude of jθ is smaller than the sufficient value of current (e.g., 107A/m2) [see Fig. 2(c)]. Hence, the small jθ hardly affects Bθ. Charge current perpendicular to Bex.– We consider Bθ in the presence of the external magnetic field Bex = Bex z and the charge current je = je z. Then, we have ∂2 t Bθ,z = ∂2 t Bθ,x = − α2 π2 (∂tδθ)2(Bθ,z + Bex + M+) − ∂2 cαµ0 t M+, ∂tδθje. π (19) (20) (Color online) Magnetic field Bθ = Bθ,z z + Bθ,x x FIG. 4. induced by dynamics of θ under charge current along x di- rection (je = je x). (a) Time-dependence of Bθ,z with several M1/M2 [see also Fig.2(b)]. (b) Time-dependence of Bθ,x at je = ±107A/m2 with M1/M2(t = 0) = 1, 2. The parameters are the same as those used in Fig. 2. (inset) Illustration of the magnetic fields due to the dynamics of θ. 4 The induced magnetic field Bθ = Bθ,z z + Bθ,x x is solved under the initial condition [Eq. (17)], and it can be de- composed into two terms. Bθ,z is the same magnetic field Bθ(je = 0) parallel to Bex [Fig. 4(a)], and Bθ,x corre- sponds to the same magnetic field Bθ(je 6= 0) parallel to je [Fig. 4(b)]. It should be noted that Bθ,x is the current-induced magnetic field. The direction of Bθ,x is parallel to the applied charge current direction, which is different from that of the charge currents via Amp´ere's law. In addi- tion, Bθ,x is regarded as a kind of electromagnetic ef- fect (current-induced magnetic field) or Edelstein effect. The conventional current-induced magnetic field[47–49] is represented by Bm = χmnje,n, where Bm is an in- duced magnetic field, je,n is a static charge current, and χmn(m, n = x, y, z) is a coefficient that is usu- ally determined by the characteristic parameters in an equilibrium state in materials. Hence, χmn is static. On the other hand, Bθ,x(t) is described by the form Bθ,x = χθ,xn(t)je,n, where the coefficient χθ,xn is rep- resented by χθ,xn(t) = − cαµ0 π δxn(cid:20)Z t 0 dt1δθ(t1) + c1t + c2(cid:21). (21) Here, c1 and c2 are arbitrary constants determined by the initial condition. Because the theta-term δθ is dynamical, χθ,xn is dynamical, unlike the conventional χmn. Finally, we discuss an experimental instrument for the obtained results in the response to the dynamics of θ. Although we consider the model in the superlattice [FI1/TI/FI2/spacer]n to study a simple model of ∂tθ, the number of ferromagnetic metals is practically larger than that of the FIs. Recently, for example, an axion-insulator multilayer constructed of a TI film and Cr-doped TI film with different magnetic coercive fields was experimentally studied[8]. A superlattice of tailored axion-insulators could be used for the model we considered. In such mate- rials, it is expected that the manipulation of the magnetic configurations of FIs (AP and P) is driven by an external magnetic field along the layered direction, as illustrated in Fig. 2(a). As a result, the dynamics of θ can be driven via the AP → P process. Then, three electromagnetic ef- fects occur during the process. First, the chiral magnetic effect, which is a magnetic field-induced charge current, could be expected. In other words, in the absence of an applied charge current, the chiral magnetic effect, which is caused by the dynamical θ-term and applied magnetic field, drives a charge current of ≈ 102 A/m2 [Fig. 2(c)], which could be detectable. Second, another electromag- netic effect is Bθ,z, which is driven by the dynamics of the θ-term and the external magnetic field Bex. The magni- tude of Bθ,z, for example, becomes ≈ 20 mT during a few tens of τf [Fig. 4(a)]. We expect that it could be difficult to distinguish Bθ,z from the total magnetic field B [Fig. 2(a)] because the magnetic fields are in the same direc- tion. Third, the obtained electromagnetic effect is the current-induced magnetic field Bθ,x, which occurs when we apply the charge current. Under the in-plane charge current je, Bθ,x, which is along the applied current direc- tion, reaches ≈ ±10mT when M1 = M2 = 10 meV and d4 = 0.1 eV with je = ±107 A/m2. This current-induced magnetic field could be measurable by manipulations of the current direction. It should be noted that under the in-plane charge current, a spin-transfer-torque-induced magnetic field[50–52] could also be generated by the cou- pling between the applied current and magnetizations on the surfaces of the TIs. However, this magnetic field could be much smaller than that of Bθ,x, which would make it experimentally negligible. Conclusion.– We have theoretically studied a way to drive the dynamics of θ in a magnetic superlattice with an axion insulator phase. As a result, we have shown the unconventional electromagnetic effects via ∂tθ: The dy- namical magnetic field-induced charge current, which is a kind of chiral magnetic effect, and the current-induced magnetic effect, which is a kind of Edelstein effect, have been discussed. The obtained results, including the artifi- cial control of ∂tθ by an external field, could be analogous to the time-dependent axion field a(t)/fa in dark-matter axion physics. This work was supported by the institute for advanced research (IAR) in Nagoya University through a Grant-in- Aid. In addition, this work was supported by a Grant-in- Aid for Scientific Research on Innovative Areas, Topolog- ical Material Science (No. JP15H05853), and a Grant- in-Aid for Challenging Exploratory Research (Grant No. JP15K13498 and 16K13803) from the Ministry of Edu- cation, Culture, Sports, Science, and Technology, Japan (MEXT). K.T. also gives thanks for the support of the Core Research for Evolutional Science and Technology (CREST) of the Japan Science and Technology Corpora- tion (JST) [JPMJCR14F1]. N.K. acknowledges the sup- port by Grant-in-Aid for JSPS Fellows. 5 T. Yokoyama, S.-i. Kimura, S. Hasegawa, Chulkov, Nano Letters 0 (2017). and E. V. [8] M. Mogi, M. Kawamura, R. Yoshimi, A. Tsukazaki, Y. Kozuka, N. Shirakawa, K. S. Takahashi, M. Kawasaki, and Y. Tokura, Nat Mater , 4855 (2017). [9] Q. L. He, X. Kou, A. J. Grutter, G. Yin, L. Pan, X. Che, Y. Liu, T. Nie, B. Zhang, S. M. Disseler, B. J. Kirby, W. Ratcliff II, Q. Shao, K. Murata, X. Zhu, G. Yu, Y. Fan, M. Montazeri, X. Han, J. A. Borchers, and K. L. Wang, Nat Mater 16, 94 (2017). [10] S.-M. Huang, S.-Y. Xu, I. Belopolski, C.-C. Lee, G. Chang, B. Wang, N. Alidoust, G. Bian, M. Neupane, C. Zhang, S. Jia, A. Bansil, H. Lin, and M. Z. Hasan, Nat. Commun. 6, 7373 (2015). [11] 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). [12] 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). [13] H. Weng, C. Fang, Z. Fang, B. A. Bernevig, and X. Dai, Phys. Rev. X 5, 011029 (2015). [14] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011). [16] J. Tominaga, [15] J. Tominaga, a. V. Kolobov, P. Fons, T. Nakano, and S. Murakami, Adv. Mater. Interfaces 1, 1300027 (2014). J. Fons, Saito, T. Nakano, M. Hase, and Y. Takagaki, X. Wang, Y. S. Murakami, Sci. Technol. Adv. Mater. 16, 14402 (2015). A. V. Kolobov, P. J. Herfort, [17] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. Lett. 102, 146805 (2009). [18] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). J. Zang, [19] X.-L. Qi, R. Li, Science 323, 1184 (2009). and S.-C. Zhang, [20] A. Karch, Phys. Rev. Lett. 103, 171601 (2009). [21] G. Rosenberg and Phys. Rev. B 82, 035105 (2010). M. Franz, [22] M. M. Vazifeh and M. Franz, Phys. Rev. B 82, 233103 (2010). [23] Y. Lan, S. Wan, and S.-C. Zhang, Phys. Rev. B 83, 205109 (2011). [24] R. Yu, W. Zhang, H.-J. Zhang, S.-C. Zhang, X. Dai, and [1] R. D. Peccei and H. R. Quinn, Z. Fang, Science 329, 61 (2010). Phys. Rev. Lett. 38, 1440 (1977). [25] K. Nomura and N. Nagaosa, [2] M. Dine, W. Fischler, and M. Srednicki, Physics Letters Phys. Rev. Lett. 106, 166802 (2011). B 104, 199 (1981). [3] A. R. Zhitnitsky, Sov. J. Nucl. Phys. 31, 260 (1980). [4] J. E. Kim, Phys. Rev. Lett. 43, 103 (1979). [5] M. Shifman, A. Vainshtein, and V. Zakharov, Nuclear Physics B 166, 493 (1980). [6] 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). [7] T. Hirahara, S. V. Eremeev, T. Shirasawa, Y. Okuyama, T. Kubo, R. Nakanishi, R. Akiyama, A. Takayama, T. Hajiri, S.-i. Ideta, M. Matsunami, K. Sumida, K. Miyamoto, Y. Takagi, K. Tanaka, T. Okuda, [26] J. Wang, B. Lian, X.-L. Qi, Phys. Rev. B 92, 081107 (2015). and S.-C. Zhang, [27] A. A. Zyuzin and A. A. Burkov, Phys. Rev. B 86, 1 (2012). [28] 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). [29] A. Vilenkin, Phys. Rev. D 22, 3080 (1980). [30] D. E. Kharzeev, L. D. McLerran, and H. J. Warringa, Nuclear Physics A 803, 227 (2008). [31] K. Fukushima, D. E. Kharzeev, Phys. Rev. D 78, 074033 (2008). and H. J. Warringa, [32] M. M. Vazifeh and M. Franz, Phys. Rev. Lett. 111, 027201 (2013). [33] H. Sumiyoshi and S. Fujimoto, Phys. Rev. Lett. 116, 166601 (2016). Phys. Rev. Lett. 108, 161803 (2012). and [42] J. Wang, Lian, B. [34] A. Sekine and K. Nomura, Phys. Rev. B 93, 45115 (2016). 6 S.-C. Zhang, Phys. Rev. Lett. 116, 96401 (2016). [35] K. Taguchi, T. Imaeda, M. Sato, Phys. Rev. B 93, 201202 (2016). and Y. Tanaka, [36] Q. Li, D. E. Kharzeev, C. Zhang, Y. Huang, I. Pletikosic, A. V. Fedorov, R. D. Zhong, J. A. Schneeloch, G. D. I. Pletikosi´c, A. V. Fedorov, R. D. Gu, T. Valla, Zhong, J. A. Schneeloch, G. D. Gu, and T. Valla, Nat. Phys. 12, 550 (2016). [37] T. Hayata, arXiv:1705.09926. [38] W.-K. and Tse A. H. MacDonald, [43] P. Sikivie, Phys. Rev. Lett. 51, 1415 (1983). [44] P. Sikivie, Phys. Rev. D 32, 2988 (1985). [45] J. Wang, R. Li, S.-C. Zhang, and X.-L. Qi, Phys. Rev. Lett. 106, 126403 (2011). [46] Y.-L. Lee, H. C. Park, J. Ihm, and Y.-W. Son, Proceedings of the National Academy of Sciences 112, 11514 (2015). [47] E. Dzyaloshinskii, I, Sov. Phys. JETP 10, 628 (1960). [48] D. N. Astrov, Sov. Phys. JETP 11, 708 (1960). [49] M. Fiebig, Journal of Physics D: Applied Physics 38, R123 (2005). [50] I. Garate Franz, and M. Phys. Rev. Lett. 105, 057401 (2010). Phys. Rev. Lett. 104, 146802 (2010). [39] J. Maciejko, X.-L. Qi, H. D. Drew, and S.-C. Zhang, [51] T. Yokoyama, J. Zang, and N. Nagaosa, Phys. Rev. Lett. 105, 166803 (2010). Phys. Rev. B 81, 241410 (2010). [40] R. Li, J. Wang, X.-L. Qi, and S.-C. Zhang, [52] A. Sakai and H. Kohno, Phys. Rev. B 89, 165307 (2014). Nat Phys 6, 284 (2010). [41] H. Ooguri and M. Oshikawa, 

1707.04051

3

1707

Transmission Line Model for Materials with Spin-Momentum Locking

We provide a transmission line representation for channels exhibiting spin-momentum locking (SML) which can be used for both time-dependent and steady-state transport analysis on a wide variety of materials with spin-orbit coupling such as topological insulators, heavy metals, oxide interfaces, and narrow bandgap semiconductors. This model is based on a time-dependent four-component diffusion equation obtained from the Boltzmann transport equation assuming linear response and elastic scattering in the channel. We classify all electronic states in the channel into four groups ($U^+$, $D^+$, $U^-$, and $D^-$) depending on the spin index (up ($U$), down ($D$)) and the sign of the $x$-component of the group velocity ($+,-$) and assign an average electrochemical potential to each of the four groups to obtain the four-component diffusion equation. For normal metal channels, the model decouples into the well-known transmission line model for charge and a time-dependent version of Valet-Fert equation for spin. We first show that in the steady-state limit our model leads to simple expressions for charge-spin interconversion in SML channels in good agreement with existing experimental data on diverse materials. We then use the full time-dependent model to study spin-charge separation in the presence of SML, a subject that has been controversial in the past. Our model shows that the charge and spin signals travel with two distinct velocities resulting in well-known spin-charge separation which is expected to persist even in the presence of SML. However, our model predicts that the lower velocity signal is purely spin while the higher velocity signal is largely charge with an additional spin component which has not been noted before. Finally, we note that our model can be used within standard circuit simulators like SPICE to obtain numerical results for complex geometries.

cond-mat.mes-hall

cond-mat

Transmission Line Model for Materials with Spin-Momentum Locking Shehrin Sayed,1, ∗ Seokmin Hong,2 and Supriyo Datta1, † 1School of Electrical and Computer Engineering, 2Center for Spintronics, Korea Institute of Science and Technology, Seoul 02792, Republic of Korea.‡ Purdue University, West Lafayette, IN 47907, USA We provide a transmission line representation for channels exhibiting spin-momentum locking (SML) which can be used for both time-dependent and steady-state transport analysis on a wide variety of materials with spin-orbit coupling such as topological insulators, heavy metals, oxide interfaces, and narrow bandgap semiconductors. This model is based on a time-dependent four- component diffusion equation obtained from the Boltzmann transport equation assuming linear response and elastic scattering in the channel. We classify all electronic states in the channel into four groups (U +, D+, U−, and D−) depending on the spin index (up (U ), down (D)) and the sign of the x-component of the group velocity (+,−) and assign an average electrochemical potential to each of the four groups to obtain the four-component diffusion equation. For normal metal channels, the model decouples into the well-known transmission line model for charge and a time-dependent version of Valet-Fert equation for spin. We first show that in the steady-state limit our model leads to simple expressions for charge-spin interconversion in SML channels in good agreement with existing experimental data on diverse materials. We then use the full time-dependent model to study spin-charge separation in the presence of SML (p0 (cid:54)= 0), a subject that has been controversial in the past. Our model shows that the charge and spin signals travel with two distinct velocities resulting in well-known spin-charge separation which is expected to persist even in the presence of SML. However, our model predicts that the lower velocity signal is purely spin while the higher velocity signal is largely charge with an additional spin component proportional to p0, which has not been noted before. Finally, we note that our model can be used within standard circuit simulators like SPICE to obtain numerical results for complex geometries. I. INTRODUCTION Background : Transport properties in materials with spin-orbit coupling (SOC) are of great interest for po- tential spintronic applications, especially because of the unique spin-momentum locking (SML) observed in di- verse classes of materials such as topological insulator (TI) [1 -- 3], heavy metals [4 -- 8], oxide interfaces [9, 10], and narrow bandgap semiconductors [11 -- 13]. There has been an immense effort to model the interplay between spin and charge in such materials using time-dependent classical [14] or quantum Boltzmann equation [15, 16], nonequilibrium Green's function [17 -- 21], phenomenologi- cal equations coupled to magnet dynamics [22], and time- independent diffusion equation used to explain bulk spin Hall effect [23]. Four-Component Diffusion Equation: In this paper, we propose a time-dependent four-component diffusion equation that can be used for transport analysis on multi- contact based structures implemented with materials ex- hibiting SML. The model is obtained from the Boltzmann transport equation assuming linear response and elastic scattering processes in the channel. The basic approach is to assign one electrochemical potential µ((cid:126)p, s) to each of the eigenstates ((cid:126)p, s) where (cid:126)p is the momentum con- fined in the z-x plane and s = ±1 is the spin index with ∗ ssayed@purdue.edu † datta@purdue.edu ‡ shong@kist.re.kr with respect to the spin quantization axis y ×(cid:16) +1 and −1 denoting the up (U ) and the down (D) spins (cid:126)p − q (cid:126)A (cid:17) ( (cid:126)A is the vector magnetic potential). We then classify the eigenstates into four groups (U +, D+, U−, and D−) based on the spin index (U , D) and the sign of the x-component of the group velocity (+,−) and define an average electrochemical potential correspond- ing to the each of the four groups resulting in a four- component diffusion equation. This can be viewed as an extension of the Valet-Fert equation which uses two elec- trochemical potentials for U and D states [24]. The four- component diffusion equation in steady-state reduces to our prior model [25, 26] that we used to predict a unique three resistance state on SML materials with two FM contacts in a multi-terminal spin valve structure [25]. The prediction has been observed recently on Pt [27, 28] and InAs [29] up to room temperature. We expect the prediction to be observed on any channel exhibiting SML. In our generalized view, U + (and U−) states have same number of modes M (and N ) as D− (and D+) states due to the time reversal symmetry. The degree of SML in our model is given by [25, 26] M − N M + N , p0 = (1) where M and N are evaluated at Fermi energy for zero temperature and in general require thermal averaging. For normal metal (NM) channels p0 = 0 i.e. M = N . For a perfect topological insulator (TI) N = 0 leading to p0 = 1, however, p0 gets effectively lowered by the presence of parallel channels. p0 has been quantified for 8 1 0 2 t c O 4 ] l l a h - s e m . t a m - d n o c [ 3 v 1 5 0 4 0 . 7 0 7 1 : v i X r a different TIs by a number of groups [30 -- 36] by measuring the charge current induced spin voltage using a ferromag- netic (FM) contact, motivated by a theoretical proposal [21]. For a Rashba channel with a coupling coefficient αR, p0 ≈ αRkF /(2EF ) (cid:28) 1 [13, 21] which can be quan- tified with similar spin voltage measurements [11, 12]. kF and EF are the Fermi wave vector and Fermi energy respectively. Recently, spin voltage measurements have been reported on heavy metals like platinum [27, 28] and gold [37]. These experiments can also be quantified by p0, though the underlying mechanism is subject to active debate [38 -- 40] and could involve a bulk spin Hall effect [7, 23, 41] or interface Rashba-like channel [8, 42 -- 44]. Transmission Line Model : We translate our four- component (U +, D+, U−, and D−) semiclassical model into a transmission line model with two component (charge and z-component of spin) voltages and currents where the coupling between charge and spin in a SML channel is characterized by p0 in Eq. (1). The model is compatible within a standard circuit simulator tool like Simulation Program with Integrated Circuit Emphasis (SPICE) which will enable straightforward analysis of complex geometries. The transmission line model is a new addition to our multi-physics spin-circuit framework [45] which has been previously used to explain experi- ments and evaluate spin-based device proposals [46, 47]. For NM channels (i.e. p0 = 0), the proposed transmis- sion line model decouples into the well-known model for charge that has been previously used to analyze transport in quantum wires [48 -- 50] and a time-dependent version of Valet-Fert equation [24] for spin. For SML channels (i.e. p0 (cid:54)= 0), our model lead to several prior results on charge-spin interconversion in the steady-state limit [21, 26] that have been previously used by a number of ex- perimental groups [29 -- 36] to quantify their spin voltage measurements using potentiometric ferromagnetic con- tacts. We further derive a simple expression and present SPICE simulation results for a parameter that has been widely used to quantify the inverse Rashba-Edelstein ef- fect (IREE) in 2D channels, which are in good agreement with existing experiments [9, 51 -- 53] on diverse materials. We then use the full time-dependent transmission line model to study the spin-charge separation in the pres- ence of SML in materials with SOC, a subject that has been controversial in the past (see, for example, [54 -- 57]). Our model suggests that depending on the channel cross- section, the charge signal can travel faster than the spin signal resulting in spin-charge separation which is well- known for materials without SOC [48, 58 -- 60] based on the Luttinger liquid theory. We argue using our model that the spin-charge separation persists even in the SOC materials exhibiting SML (i.e. p0 (cid:54)= 0). Similar argu- ments have been made in the past considering the pres- ence of spin-orbit coupling (SOC) [54 -- 56] although there exists counter arguments that the presence of SOC de- stroys the spin-charge separation [57]. However, we pre- dict that the high velocity charge signal in SML channels accompanies an additional spin component proportional 2 to p0 having the same velocity as the charge, which has not been discussed before. Note that the proposed model does not take into ac- count the effects such as spin precession involving the off- diagonal elements of the density matrix which we assume to be negligible. An extension of this model to include x and y components of spin could possibly address such is- sues, as done earlier for materials without SOC (see [46], and references therein). The assumptions made to derive the model have been discussed in detail in Section V. Several predictions from our model for steady-state [25] have already received support from experiments [27 -- 29] suggesting that the assumptions are within the reason- able limits. The assumptions can be revisited as the field evolves leading to revised model parameters, but the ba- sic model should remain valid. Outline: The paper is organized as follows. In Sec- tion II, we describe the transmission line model for SML channels and show that special cases lead to prior well- known models. In Section III, we derive several re- sults on charge-spin interconversion from our transmis- sion line model in steady-state and present comparison with SPICE simulations using the full model. We obtain a simple expression for a parameter that has been widely used to quantify IREE and show that it is in good agree- ment with available experiments on diverse materials. In Section IV, we study the spin-charge separation in terms of spin and charge signal velocities obtained from our time-dependent transmission line equations. We show that the separation persists even in SML channels, how- ever, in SML channels there exists an additional spin component accompanied by the high velocity charge sig- nal. In Section V, we derive the transmission line model starting from the Boltzmann transport equation with all the assumptions clearly stated. We discuss different scat- tering mechanisms in the channels and their effects on charge and spin transport. Finally, in Section VI, we end with a brief summary. II. TRANSMISSION LINE MODEL II.1. Model Description We consider the structure and axes in Fig. 1(a) to derive the transmission line model. The model has two components: charge and z-component of spin with cou- pling between them characterized by p0 in Eq. (1). The charge model is given by (cid:18) 1 (cid:19)−1 ∂ + 1 CQ CE (LK + LM ) ∂ ∂t ∂t Ic, Vc = − ∂ ∂x Ic + Rc Ic = − ∂ ∂x (2) Vc + p0ηcVs, where Ic and Vs are charge current and voltage along x- direction, CE and CQ are the electrostatic and quantum capacitances per unit length, LM and LK are the mag- 3 (4b) (4c) (5a) (5b) (5c) (5d) (5e) (5f) (5g) (5h) (5i) (5j) (5k) 2 RB (cid:104)vx(cid:105) , RB 2(cid:104)vx(cid:105) , RB λ α2RB , , λ0 4 CQ = LK = Rc = Rs = Gsh = gm = rm = ηs = ηc = γs = , , α2RBλs 2 , , αRB α (cid:104)vx(cid:105) CE 2α λ0 2 αλr 2 αλt h q2 1 , , Vn = ηsVc + rmIem, and, In = γsIc − gmVem. The parameters of Eqs. (2) and (3) are given by FIG. 1. (a) Structure and corresponding axes of the channel with spin-momentum locking (SML) under consideration, for which a transmission line model is derived. The model has two components: (b) charge (corrseponds to Eq. (2)) and (c) spin (corrseponds to Eq. (3)), with coupling between them described by degree of SML p0 (see Eq. (1)). Coupling be- tween charge and spin are modeled by dependent voltage and current sources with Vm = ηcVs, Vn = ηsVc + rmIem, and In = γsIc − gmVem. netic and kinetic inductances per unit length, and Rc is the charge resistance per unit length. The spin model is given by ∂ CQ α2 ∂t α2LK Vs + GshVs + p0gmVem = − ∂ ∂x Is + RsIs − p0rmIem = − ∂ ∂ ∂x ∂t Is + p0γsIc, Vs + p0ηsVc, (3) where Is and Vs are spin current and voltage along x- direction with spin polarization along the z-direction. In this discussion, y-direction is out-of-plane. Here, α = 2/π is an angular averaging factor, Rs is the spin resistance per unit length of the channel and Gsh is the shunt con- ductance per unit length that captures the spin lost in the channel due to the spin relaxation. Detailed derivation of Eqs. (2) and (3) from the Boltzmann transport equa- tion will be discussed in Section V with clearly stated assumptions. Distributed circuit models for charge and spin are shown in Fig. 1(b) and (c), which are based on Eqs. (2) and (3) respectively. The dependent sources propor- tional to p0 represent charge-spin inter-coupling between the two models. The dependent source parameters in Fig. 1 are given by Vm = ηcVs, (4a) and, RB = . M + N Here, λ, λ0, and λs are three distinct mean free paths (cid:104)vx(cid:105) is that determine Rc, Rs, and Gsh respectively. the the magnitude of the thermally averaged electron ve- locity (cid:104)vx(cid:105). ηc represents spin to charge conversion coeffi- cient and ηs, γs represent charge to spin conversion coef- ficients. These coefficients depend on different scattering mechanisms in the channel, which will be discussed later in Section V. rm and gm are transient charge-spin cou- pling coefficients which are in the units of resistance per unit length and conductance per unit length respectively. RB is the ballistic resistance of the channel, h is the Planck's constant, and q is electron charge. RB is inversely proportional to the total number of modes (M +N ) in the channel which represents a material prop- erty and does not imply ballistic transport. The models and related results discussed in this paper are valid all the way from ballistic to diffusive regime of operation. II.2. Presence of an External Contact In the presence of an external contact on the channel (see Fig. 2(a)), the charge model in Eq. (2) is modified under the external contact. They are given as (cid:27) (cid:27) (cid:26) ic (cid:26) ∆vc is ∆vs = G0 , and  1 pf α pf α 1 α2 (cid:27) vs − Vs (cid:26) vc − Vc (cid:20) 1 αpf (cid:27) (cid:21)(cid:26) Ic (cid:34) (cid:35)(cid:26) vc (cid:27) αpf α2 Is 1 pf α α pf B = − G0R2 4 p0G0RB + 2 4 (8) (9) , vs where vc = (vu + vd)/2 and vs = α(vu − vd)/2 are charge and spin voltages applied at the external contact, G0 is the contact conductance per unit length, and pf is the contact polarization with pf = 0 indicating a nor- mal metal contact and pf (cid:54)= 0 indicating a ferromagnetic contact. The derivation of the model starting from the Boltzmann transport equation is shown in Section V with clearly stated assumptions streamlined with subheadings. Modified distributed circuit models for charge and spin are shown in Fig. 2(b) and (c), which are based on Eqs. (6)-(7) respectively. The presence of a contact with con- ductance G0 adds series resistances Rc cont = α2G0R2 B 4 cont = Rs in the charge and spin models as shown in Fig. 2. This effect exists even if the channel is NM. The presence of the external contact also modulates the dependent sources in Eqs. (4a) and (4b) as G0R2 B 4 and (cid:16) vs (cid:17) V (cid:48) m = ηcVs + G0RB 2 α + pf vc , (10a) V (cid:48) n = ηsVc + rmIem + G0RB 2 (αvc + pf vs) , (10b) with Va and Vb representing the voltage drop across Rc cont and Rs cont in charge and spin models in Fig. 2 respec- tively. Note that the additional terms are proportional to G0 and negligible for potentiometric contacts where G0 is very low. Vc + p0ηcVs + ∆vc, (6) II.3. Special Case: Normal Metals (p0 = 0) and the spin model in Eq. (3) is modified as Is + p0γsIc + is, CQ ∂ α2 ∂t α2LK Vs + GshVs + p0gmVem = − ∂ ∂x Is + RsIs − p0rmIem = − ∂ ∂ ∂x ∂t Vs + p0ηsVc + ∆vs, (7) where ic and is represent charge and spin currents en- tering into the channel per unit length from the external contact, ∆vc and ∆vs represent the change in channel charge and spin voltages per unit length in the region We consider a special case of Eqs. (2) and (3) for a normal metal (NM) channel i.e. p0 = 0. For NM channel, charge and spin model decouples to well-known models as described below. II.3.1. Charge Model: Transport Model for Quantum Wires For NM channels (p0 = 0), the charge model in Eq. (2) reduces to the well-known transmission line model for FIG. 2. (a) Structure of the spin-momentum locked (SML) channel in the presence of an external contact. Both (b) charge and (c) spin transmission line models are modified as compared to Fig. 1, which now correspond to Eqs. (6) and (7) respectively. The dependent sources are V (cid:48) m = ηcVs + (αvc + pf vs), and In = γsIc − gmVem. Note that the model reduces to that shown in Fig. 1 in the limit G0 → 0. n = ηsVc + rmIem + G0RB (cid:1), V (cid:48) (cid:0) vs α + pf vc G0RB 2 2 as(cid:18) 1 CE + 1 CQ (cid:19)−1 ∂ (LK + LM ) ∂ ∂t Ic + ic, ∂t Vc = − ∂ ∂x Ic + Rc Ic = − ∂ ∂x charge that has been previously used to analyze transport in quantum wires, given by tact in Eqs. (6) and (7), given by 5 (16a) (16b) (16c) (cid:18) 1 (cid:19)−1 ∂ + 1 CQ CE (LK + LM ) ∂ ∂t ∂t Ic, Vc = − ∂ ∂x Ic + Rc Ic = − ∂ ∂x (11) Vc. The model was first derived from Luttinger liquid the- ory [48, 49] and then from Boltzmann transport equation with one electrochemical potential [50]. In the quantum wire limit LK (cid:29) LM and CQ (cid:28) CE while in the classical transmission line limit LK (cid:28) LM and CQ (cid:29) CE [48, 50]. In steady-state (∂/∂t → 0), we get the diffusion equa- tion for charge from Eq. (11), given by d2 dx2 Vc = 0. (12) II.3.2. Spin Model: Valet-Fert Equation For NM channels (p0 = 0), the spin model in Eq. (3) becomes a time-dependent spin-diffusion equation, given by CQ ∂ α2 ∂t α2LK Vs + GshVs = − ∂ ∂x Is + RsIs = − ∂ ∂ ∂x ∂t Is, (13) Vs, which in steady-state (∂/∂t → 0), becomes the well- known Valet-Fert equation [24], given by d2 dx2 Vs = Vs λ2 sf , with the spin diffusion length given by √ λsf = 1√ RsGsh = λ0λs 2 . (14) (15) Spin model similar to Eq. (13) has been discussed pre- viously based on Luttinger liquid theory [48], however, the model did not take into account the spin relaxation processes in the channel (the shunt conductance Gsh). III. STEADY-STATE TRANSPORT RESULTS In this section, we discuss several established steady- state results on charge-spin interconversion in the poten- tiometric limit. We start from the steady-state (∂/∂t → 0) form of the transmission line model with external con- d dx d dx d dx d dx Ic = ic, Vc = −Rc Ic + p0ηcVs + ∆vc, Is = −GshVs + p0γsIc + is, Vs = −RsIs + p0ηsVc + ∆vs. and (16d) Under the steady-state condition (∂/∂t → 0), the ca- pacitors and inductors in Fig. 2 become open and short circuits respectively. The steady-state form in Eq. (16) is equivalent to our prior time-independent semiclassical equations with four electrochemical potentials [25] and all our previous results can be reproduced using Eq. (16). We first derive a resistance matrix for a three terminal setup (two charge and one spin) with potentiometric con- tacts (see Fig. 3) and assuming reflection with spin-flip to be the dominant scattering mechanism in the channel. We present dc simulation results on charge-spin intercon- version in the SML channel using the full model (Eqs. (6) and (7)) in SPICE and compare with the results from the resistance matrix. We then derive a simple expression for a parameter that has been widely used to quantify inverse Rashba-Edelstein effect (IREE) in 2D channels. We compare the expression with available experiments on diverse materials as well as dc SPICE simulation results using the full model. III.1. Resistance Matrix for Potentiometric Setup We consider a structure with three NM contacts (pf = 0) on top of a SML channel, as shown in Fig. 3. Con- tacts 1 and 2 are the charge terminals and contact 3 is the spin terminal with no charge current flowing through it (i.e. ic = 0). We start from Eq. (16) and make two as- sumptions to drive the resistance matrix: (i) the contacts are potentiometric i.e. the contact conductance per unit length G0 is very low such that the following condition is satisfied 1 λ , 1 λ0 , 1 λr (cid:29) G0RB 4 , (17) and (ii) the reflection with spin-flip is the dominant scat- tering mechanism in the channel. The details of the derivation are given in Appendix A. The resistance matrix is given by (cid:26) ∆Vc (cid:27) vs =  RBL 2 G(cid:48)(cid:48) 0 + λ αp0RB 2 − αp0RB 2 α2 (cid:48) G0 (cid:26) Ic is tot (cid:27) , (18) where ∆Vc is the charge voltage difference between con- tacts 1 and 2, Ic is the charge current flowing in the 6 TABLE I. Estimated Material Parameters. Material λ [nm] p0 λIREE [nm] λIREE [nm] (measured) (Eq. (23)) AgBi CuBi 0.05†† 0.05†† LAOSTO 180.8† 0.0616†† 0.025‡ Bi2Se3 22.6† 0.88† 3.2† 0.36 0.014 3.55 0.026 0.3 [51] 0.009 [52] 6.4 [9] 0.035 [53] † Estimated from the measured sheet resistance of the samples. †† Estimated from the Rashba coupling coefficient and Fermi velocity of the materials. ‡Estimated from spin-pumping induced voltage and Eq. (20). The estimations are discussed in detail in Appendix C. Eq. (19) was originally proposed in Ref. [21] which was later confirmed by a number of experiments [29 -- 36] on different TI materials using potentiometric ferromagnetic contact. Note that the spin voltage measured using a contact with higher conductance will be lower than Eq. (19) due to the current shunting effect in the contact [25]. The full model in Eqs. (6) and (7) takes into account such effect related to contact conductances. We have simulated the structure in Fig. 3(a) by con- necting the full two-component circuit models given in Figs. 1 and 2 in a distributed manner using the stan- dard circuit rules. The details of the simulation are dis- cussed in Appendix B. The simulation results of vs/Ic as a function of p0 is shown in Fig. 3(b), which is in good agreement with Eq. (19). III.3. Inverse Effects: Spin Current Induced Charge Voltage The reciprocal effect of Eq. (19) is the spin current is induced open circuit charge voltage difference ∆Vc across the SML channel [26]. For Ic = 0 we have from Eq. (18) ∆Vc = − αp0 2GB is tot. (20) The reciprocal relation between Eqs. (19) and (20) in- cluding the negative sign has been observed experimen- tally [27, 29, 33]. Note that Eq. (20) is exact for a potentiometric contact. For a contact with higher con- ductance, ∆Vc will be lower than Eq. (20) due to current shunting by the same amount as that for the direct effect (in Eq. (19)) [25], which can be analyzed using the full model in Eqs. (6) and (7). We have simulated the struc- ture shown in Fig. 3(c) using the full two-component models in Figs. 1 and 2. The simulation results of ∆Vc/is tot as a function of p0 is shown in Fig. 3(d), which show good agreement with Eq. (20). The details of the simulation are discussed in Appendix B. FIG. 3. (a) Setup to observe charge current Ic induced spin voltage vs in the SML channel. Charge terminal of contact 3 is kept open and spin terminals of contacts 1 and 2 are grounded. (b) vs/Ic vs. p0 for is tot = 0 from SPICE simula- tion and comparison with Eq. (19). (c) Setup to observe spin current is tot induced charge voltage difference ∆Vc across the SML channel. Charge terminal of contact 3 is kept open and spin terminals of contacts 1 and 3 are grounded. (d) ∆Vc/is tot vs. p0 for Ic = 0 from SPICE simulation and comparison with Eq. (20). The SPICE setup is shown in Fig. 6 of Appendix B. Parameters: α = 2/π, G0 ≈ 0.05GB, M + N = 100, and scattering rate per unit mode = 0.04 per lattice points. We assumed reflection with spin-flip to be the dominant scatter- ing process in the channel. 0 , and G(cid:48) channel with length L, vs is the spin voltage at contact tot = isL is the spin current at contact 3, contacts 1 3, is and 2 have equal conductance G(cid:48)(cid:48) 0 = G0L is the contact conductance of contact 3. Eq. (18) is the similar to that previously reported in Ref. [26] with corrections in the diagonal components. The diagonal components (1, 1) and (2, 2) in the matrix represent the charge resis- tance between contacts 1 and 2 and spin resistance at contact 3 respectively. Note that Eq. (18) is derived under the assumption that reflection with spin-flip is the dominant scattering mechanism. In the presence of other scattering mecha- nisms, the basic structure of Eq. (18) remains the same, however, additional factors related to scattering rates multiply each of the components in the matrix (see Ap- pendix A). III.2. Direct Effects: Charge Current Induced Spin Voltage For a charge current Ic flowing through the SML chan- nel, the open circuit spin voltage vs at the contact 3 (i.e. is tot = 0) can be derived from Eq. (18) as vs = αp0 2GB Ic. (19) III.4. Inverse Rashba-Edelstein Effect (IREE) Length The phenomena described by Eq. (20) is often known as the inverse Rashba Edelstein effect (IREE) for 2D channels. IREE is often quantified with a parameter called IREE length defined as λIREE = Jc Js , (21) where Jc is the longitudinal short circuit charge current density in A/m induced by the injected transverse spin current density Js in A/m2. We derive a simple expression for IREE length starting from the first row of Eq. (18) for short circuit condition between contacts 1 and 2 (i.e. vc1 = vc2) and assuming large channel resistance compared to contact resistance (i.e. L/(GBλ) (cid:29) 2/G(cid:48)(cid:48) 0 ). The expression is given by λIREE = αp0λ 2 . (22) The derivation is given in Appendix A. Note that both p0 and λ are two completely independent parameters. Here, α is an angular averaging factor that can vary between 0 and 1 depending on the angular variation of the spin polarization of the eigenstates and details of the scattering mechanism. We assume that the distribution is such that the angle between z-axis and the spin po- larization of the eigenstates with particular group veloc- ity (+ or −) vary between −π/2 to +π/2, which yields α = 2/π. Thus, from Eq. (22) we have the following expression λIREE = p0λ π , (23) which has been previously reported in Ref. [25]. We have simulated the setup in Fig. 4(a) in order to esti- mate λIREE of diverse ranges of p0 and λ. The setup in Fig. 4(a) is same as that in Fig. 3(c), except we observe the short circuit charge current Ic between the charge terminals of contacts 1 and 2 induced by the injected spin current is through the spin terminal of contact 3. [51], CuBi We have compared the simulation results with Eq. (23) as well as available measurements from spin-pumping and lateral spin valve experiments on different Rashba interfaces: AgBi [52], LaAlO3SrTiO3 (LAOSTO) [9], and Bi2Se3 [53]. The comparison is shown in Fig. 4(b). We have estimated λ from the re- ported sheet resistance or resistivity of the samples. p0 for AgBi, CuBi, and LAOSTO are estimated using the Rashba coupling coefficient (v0) and the Fermi velocity (vF ) quoted in the literature, using the following expres- sion: v0(cid:112)v2 0 + v2 F p0 = . (24) For Bi2Se3, we have estimated p0 from spin-pumping in- duced inverse voltage and Eq. (20). Note that p0 esti- mated for the sample in Ref. [53] is much lower than 7 FIG. 4. (a) Setup similar to that in Fig. 3(c) but short circuit charge current between contacts 1 and 2 is being ob- served. (b) Inverse Rashba Edelstein effect (IREE) length (λIREE) vs. p0λ from SPICE simulation and comparison to Eq. (23) and experiments on different interfaces with Rashba SOC: AgBi [51], CuBi [52], LaAlO3SrTiO3 (LAOSTO) [9], and Bi2Se3 [53]. The back scattering length λ is estimated from measured sheet resistance or resistivity. The degree of spin-momentum locking p0 is estimated from the Rashba cou- pling coefficient and the Fermi velocity using Eq. (24). The details of estimations are given in Appendix B. SPICE simu- lation parameters: α = 2/π, G0 ≈ 0.05GB, and M +N = 100. We assumed reflection with spin-flip to be the dominant scat- tering process in the channel. previous reports [30 -- 36], which may be due to the pres- ence of large number of parallel channels. We expect higher λIREE for Bi2Se3 samples with higher p0. The derivation of Eq. (24) from the Rashba Hamilto- nian is shown in Appendix C. The estimations are sum- marized in Table I and the details are given in Appendix C. These two independent estimations of p0 and λ when applied to Eq. (23), agrees very well with experimentally reported λIREE, as shown in Fig. 4 and Table I. IV. TIME-DEPENDENT TRANSPORT RESULTS In this section, we use the full time-dependent model in Eqs. (2) and (3) to discuss a well-known phenomenon called the spin-charge separation. The spin-charge sepa- ration in the presence of spin-orbit coupling is a subject that has been controversial in the past. Our model shows that the charge and spin propagates with two distinct ve- locities which persist even in the materials with spin-orbit coupling exhibiting spin-momentum locking (p0 (cid:54)= 0). However, we show that the lower velocity signal is purely spin while the higher velocity signal is largely charge with an additional spin component proportional to p0. IV.1. Velocities of Charge and Spin Signals The velocities for charge and spin signals can be de- rived by finding the eigenvalues of Eqs. (2) and (3) as- suming the low loss limit and constant coefficients. In addition, we find the corresponding eigenvectors as well to analyze the coupling between spin and charge in the channel due to SML. The details of the derivation are given in Appendix D. The lower velocity eigenvalue is given by = ±(cid:104)vx(cid:105), vg,s = ± 1(cid:112)LKCQ (25) which is determined by quantum capacitance CQ and kinetic inductance LK, resulting in thermally averaged electron velocity (cid:104)vx(cid:105). The corresponding eigenvector is given by(cid:40) (cid:41) Vc Ic (cid:40) (cid:41) 0 0 = , and = (cid:40) (cid:41) Vs Is  ±α2 (cid:115) 1 LK CQ  , (26) which shows that the lower velocity signal is purely spin and no charge accompanies the signal even in channels with SML i.e. p0 (cid:54)= 0. The higher velocity eigenvalue is given by vg,c = ± (27) where Cef f is a series combination of CE and CQ and Lef f is a series combination of LM and LK. The corre- sponding eigenvector is given by , 1(cid:112)Lef f Cef f  ±  , and (cid:18) Lef f Cef f 1 −α2gm (cid:115) LM CQ ±vg,cv2 g,sCQ =   , (cid:19) v2 g,c + rmv2 g,s −gm + LM CQ rm α2 (cid:40) (cid:41) Vc Ic (cid:41) = (cid:40) Vs Is p0 g,c − v2 v2 g,s (28) which shows that the higher velocity signal is largely charge which will be accompanying an additional spin signal proportional to p0, which has not been discussed before. This additional spin component vanishes in a NM channel where there is no SML (i.e. p0 = 0) and the signal is purely charge. Further evaluation of this high velocity spin component we leave as future work. 8 The quantum capacitance CQ is proportional to the total number of modes (M + N ) in the channel (see Eq. (5a)) while the kinetic inductance LK is inversely propor- tional to M + N (see Eq. (5b)). M + N is proportional to the channel width (for 2D) or cross-sectional area (for 3D) [61]. For a conductor with very large cross-section, we may have CE (cid:28) CQ and LM (cid:29) LK which is the standard transmission line limit. In this limit, the velocity in Eq. (27) becomes c = 1√ LM CE , which is the velocity predicted by standard transmission line theory and can be as high as the speed of light. For a conductor with very small cross-section like quantum wires, we may have CE (cid:29) CQ and LM (cid:28) LK. In this limit, velocity in Eq. (27) becomes the thermally averaged electron velocity, given by (cid:104)vx(cid:105) = 1(cid:112)LKCQ , which is same as Eq. (25). Note that the two velocity eigenvalues are equal at this limit. IV.2. Spin-Charge Separation From Eqs. (25) and (27) we have (cid:118)(cid:117)(cid:117)(cid:117)(cid:117)(cid:117)(cid:116) 1 + δ 1 + CQ CE CQ CE vg,s vg,c = , (29) where δ = (LM CE) / (LKCQ) = ((cid:104)vx(cid:105)/c)2 (see Eq. 6 of Ref. [50]) which is usually much less than one, making the spin signal slower than the charge signal, given by vg,s < vg,c. (30) This results in spin-charge separation which is well- established for channels without SML (i.e. p0 = 0) from Luttinger liquid theory (see for example, Refs. [48, 58 -- 60], and references therein). Electrons' charge excites CE and LM hence the charge signal velocity is determined by Cef f and Lef f given by Eq. (27). However, pure spin signal do not excite CE and LM , hence its velocity is de- termined by CQ and LK only given by Eq. (25). Similar arguments have been made in the past [48, 49] in the context of carbon nanotubes without SOC. Note that the argument in Eq. (30) is independent of p0 (see Appendix D), which indicates that the spin-charge separation persists even in channels with SOC exhibiting SML (i.e. p0 (cid:54)= 0). Similar arguments have been dis- cussed previously by considering SOC [54 -- 56] although there exists argument that the presence of SOC may de- stroy the spin-charge separation [57]. In SML channels, an additional spin signal propor- tional to p0 accompanies the charge signal at the same ve- locity as the charge (vg,c) as seen from Eq. (28). This ad- ditional spin component is induced by the instantaneous voltage drop across LM and the instantaneous current through CE of the channel. However, the low velocity signal (see Eq. (25)) remains purely spin since the spin signal do not excite LM and CE [48, 50] to induce a sim- ilar accompanying charge component. V. TRANSMISSION LINE MODEL FROM BOLTZMANN FORMALISM In this section, we derive the transmission line model in Eqs. (6) and (7) starting from the time-dependent Boltz- mann transport equation under several clearly stated as- sumptions, which allow us to obtain the simple expres- sions for the model parameters stated in Eq. (5). Several of our predictions for steady-state [25] have already re- ceived experimental support [27 -- 29] suggesting that the assumptions are reasonable, but they could be revisited as the field evolves. V.1. Boltzmann Transport Equation (cid:88) (cid:126)p(cid:48),s(cid:48) We assume a structure where the spatial variations and the applied fields are along x-direction. The time- dependent Boltzmann transport equation is given by ∂f ∂t + vx ∂f ∂x + Fx ∂f ∂px = S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) (f − f(cid:48)) , (31) where we have assumed elastic scattering so that the scat- tering rates are same in both directions i.e. S((cid:126)p, s → (cid:126)p(cid:48), s(cid:48)) = S((cid:126)p(cid:48), s(cid:48) → (cid:126)p, s) ≡ S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)). Here, f ≡ f (x, t, (cid:126)p, s) is the occupation factor of a state for a particular position x, time t, momentum (cid:126)p and spin index s = ±1, f(cid:48) ≡ f (x, t, (cid:126)p(cid:48), s(cid:48)) with momentum (cid:126)p(cid:48) and spin index s(cid:48) = ±1, vx = ∂E/∂px is the x-component of the group velocity, Fx = −∂E/∂x is the force on electrons along x-direction, and E is the total energy. Note that the spin index +1 and −1 correspond to up (U ) and down (D) spin polarized states for a particular (cid:126)p. V.2. Occupation Factor 9 We write the occupation factor f in terms of an elec- trochemical potential µ ≡ µ(x, t, (cid:126)p, s), in the form f (x, t, (cid:126)p, s) = 1 + exp (cid:18) E(x, t, (cid:126)p, s) − µ(x, t, (cid:126)p, s) 1 (cid:19) , kBT (32) where kB is the Boltzmann constant, T is the tempera- ture. Note that V.3. Linearization We apply a variable transform ξ = E − µ on the left hand side of Eq. (31). On the right hand side of Eq. (31), we expand both f and f(cid:48) into Taylor series around f0 = 1 + exp (cid:18) (E(x, (cid:126)p, s) − µ0) 1 (cid:19) , kBT (33) with constant electrochemical potential µ0 and apply lin- ear response approximation. Thus Eq. (31) can be writ- ten as (see Appendix E for details of the derivation) (cid:19) (cid:18) (cid:19)(cid:18) ∂µ = −(cid:88) − ∂f0 ∂E (cid:126)p(cid:48),s(cid:48) − ∂E ∂µ + vx ∂t ∂t ∂x S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) (cid:18) ∂µ ∂px (cid:19) + Fx − ∂f0 ∂E (µ − µ(cid:48)). (34) We assume that there are no internal fields in the present discussion. Hence, Fx comes from the applied voltage and the term Fx(∂µ/∂px) depends on the higher order of the applied voltage, which can be neglected in the linear response regime [61]. Thus Eq. (34) is given by(cid:18) (cid:19)(cid:18) ∂µ = −(cid:88) − ∂f0 ∂E (cid:126)p(cid:48),s(cid:48) − ∂E ∂µ + vx ∂t ∂t ∂x S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) (cid:19) (µ − µ(cid:48)). (35) (cid:19) (cid:18) − ∂f0 ∂E The term ∂E/∂t can be evaluated from the dispersion relation of a given Hamiltonian in the semiclassical ap- proximation as discussed below. V.4. Dispersion Relation (cid:16) We start from the following Rashba Hamiltonian H = (cid:126)p − q (cid:126)A2 2m I2×2 − v0 (cid:126)σ × ((cid:126)p − q (cid:126)A) (36) Here, I2×2 is a 2× 2 identity matrix, (cid:126)p and (cid:126)A are the mo- mentum and vector magnetic potential respectively in the (cid:17) · y + UEI2×2. z-x plane, (cid:126)σ is the Pauli's matrices, v0 is the Rashba coef- ficient, UE is the electrostatic potential, m is the electron mass, and q is the electron charge. Eigenstates of Eq. (36) are given by E((cid:126)p, s) = (cid:126)p − q (cid:126)A2 2m − s v0(cid:126)p − q (cid:126)A + UE, (37) with (cid:126)p is confined to the z-x plane. We assume that UE and (cid:126)A varies slowly with x and t, so that in the semiclassical approximation we have E(x, t, (cid:126)p, s) = (cid:126)p − q (cid:126)A(x, t)2 2m −s v0(cid:126)p − q (cid:126)A(x, t)+UE(x, t). (38) Differentiating Eq. (38) with respect to t yields (cid:32) (cid:33) = (cid:126)v · ∂E ∂t −q ∂ (cid:126)A ∂t + ∂UE ∂t , (39) where (cid:126)v = ∇(cid:126)pE (see Appendix F for the derivation). The electrostatic potential UE and the vector magnetic potential (cid:126)A on the structure of interest can be evaluated from the theory of electromagnetism. V.5. From Potentials to Charge and Current The electrostatic potential UE is related to the total charge Q in the channel by the electrostatic capacitance CE of the structure under consideration, given by UE q = Q CE . (40) We assume that the charge current Ic flows along x- direction which is uniform in the channel. The vector magnetic potential (cid:126)A is related to Ic by the magnetic inductance LM of the channel, given by (cid:126)A ≡ xAx = xLM Ic. (41) Thus Eq. (39) can be written as = −qvxLM ∂E ∂t ∂Ic ∂t + q CE ∂Q ∂t . (42) We combine Eq. (35) with Eq. (42) to get (cid:18) ∂f0 ∂E (cid:19)(cid:18) ∂µ = −(cid:88) ∂t + vx (cid:126)p(cid:48),s(cid:48) (cid:19) + qvxLM ∂µ ∂x S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) ∂Ic ∂t (cid:18) ∂f0 (cid:19) ∂Q ∂t − q CE (µ − µ(cid:48)). ∂E (43) where the factor of 2 appeared since we are summing over all s states. Note that D0(U +) = D0(D−) and D0(U−) = D0(D+) due to time-reversal symmetry. V.6. Classification 10 We classify all (cid:126)p, s states into four groups based on the sign of vx (+ or −) and the spin index s = ±1, given by (cid:60) :  by y ×(cid:16) U + ∈ {(cid:126)p, s vx > 0, s = +1}, D− ∈ {(cid:126)p, s vx < 0, s = −1}, U− ∈ {(cid:126)p, s vx < 0, s = +1}, and D+ ∈ {(cid:126)p, s vx > 0, s = −1}. (44) (cid:17) (cid:126)p − q (cid:126)A where s = +1 and −1 denote up (U ) and down (D) spins with respect to the spin quantization axis defined , which is different for each direction of (cid:126)p. Such classification can be mapped onto the two Fermi circles of a Rashba channel (see Fig. 5(a)). The large circle corresponds to U + and D− groups which share the same number of modes nm(U +) = nm(D−) = M sat- isfying the time-reversal symmetry. Similarly, the small circle corresponds to U− and D+ groups sharing the same number of modes nm(U−) = nm(D+) = N . Note that the eigenstates belonging to each of the four half Fermi circles in Fig. 5(a) has an average spin polarization along z-direction with an averaging factor of α = 2/π, which we will use later when writing spin currents and voltages. V.7. Averaging We define the thermal average of a variable ψ ≡ ψ((cid:126)p, s) within each of the group (cid:126)p, s ∈ (cid:60) as (cid:88) (cid:18) (cid:88) (cid:126)p,s∈(cid:60) (cid:19) ψ((cid:126)p, s) (cid:19) (cid:18) − ∂f0 ∂E − ∂f0 ∂E (cid:104)ψ(cid:105)(cid:126)p,s∈(cid:60) = (cid:126)p,s∈(cid:60) . (45) We sum both sides of Eq. (43) over all (cid:126)p states within range (cid:126)p, s ∈ (cid:60) in the z-x plane as D0((cid:60)) ∂ (cid:104)µ(cid:105) 2 ∂t qD0((cid:60)) + = − (cid:88) 2 D0((cid:60)) + 2 (cid:104)vx(cid:105)LM ∂Ic ∂t ∂(cid:104)vxµ(cid:105) ∂x − q CE S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) (cid:88) (cid:126)p,s∈(cid:60) (cid:126)p(cid:48),s(cid:48) D0((cid:60)) (cid:18) (cid:19) ∂Q 2 ∂t − ∂f0 ∂E (46) (µ − µ(cid:48)). Here, D0((cid:60)) is the thermally averaged density of states within (cid:126)p, s ∈ (cid:60) given by D0((cid:60)) (cid:88) (cid:18) (cid:19) (47) = 2 (cid:126)p,s∈(cid:60) − ∂f0 ∂E , We make the following assumption in Eq. (46) (cid:104)vxµ(cid:105) ≈ (cid:104)vx(cid:105)(cid:104)µ(cid:105), (48) 11 which yields ∂ (cid:104)µ(cid:105) D0((cid:60)) ∂t 2 qD0((cid:60)) + = − (cid:88) 2 (cid:126)p,s∈(cid:60) (cid:126)p(cid:48),s(cid:48) D0((cid:60)) + 2 (cid:104)vx(cid:105)LM ∂Ic ∂t (cid:104)vx(cid:105) ∂(cid:104)µ(cid:105) (cid:18) − q CE S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) (cid:88) ∂x D0((cid:60)) (cid:19) ∂Q 2 ∂t − ∂f0 ∂E (µ − µ(cid:48)) + iext q . (49) Note that the term iext/q on the right hand side has been added to take into account the total current entering into (cid:126)p, s ∈ (cid:60) states of the channel an external contact. The number of modes within (cid:126)p, s ∈ (cid:60) in the channel is given by [61] nm((cid:60)) = hD0(cid:104)vx(cid:105) , 2L (50) where (cid:104)vx(cid:105) is the magnitude of (cid:104)vx(cid:105) and L is the channel length. Note that (cid:104)vx(cid:105) is same in all four groups for the Rashba Hamiltonian considered here (see Appendix F). Thus Eq. (49) can be written as nm ((cid:60)) (cid:104)vx(cid:105) + sgn ((cid:104)vx(cid:105)) q nm ((cid:60))LM + sgn ((cid:104)vx(cid:105)) nm ((cid:60)) nm ((cid:60)) (cid:104)vx(cid:105) = S((cid:60)) + ∂ (cid:104)µ(cid:105) ∂t ∂(cid:104)µ(cid:105) ∂x ∂Q ∂t , − q CE iext L h q ∂Ic ∂t where S((cid:60)) = − h L (cid:88) (cid:88) (cid:126)p,s∈(cid:60) (cid:126)p(cid:48),s(cid:48) (cid:19) (cid:18) − ∂f0 ∂E (51) (µ − µ(cid:48)) . S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)) (51) applies to each group in Eq. (52) Eq. (44) with an average electrochemical potential given by (cid:104)µ(cid:105)(cid:126)p,s∈U + ≡ µ (U +), (cid:104)µ(cid:105)(cid:126)p,s∈D− ≡ µ (D−), (cid:104)µ(cid:105)(cid:126)p,s∈U− ≡ µ (U−), and (cid:104)µ(cid:105)(cid:126)p,s∈D+ ≡ µ (D+) respectively. FIG. 5. (a) Two Fermi circles of a Rashba channel with spin- momentum locking having opposite spin polarizations at a given energy EF . Left (right) half of each circle represents negative (positive) group velocity. The large circle corre- sponds to the large number of modes M in the channel and have a net spin polarization along +z (−z) on right (left) side. The small circle corresponds to smaller number of modes N and have net spin polarization along −z (+z) on right (left) side. We classify all electronics states into four groups based on the z-component of spin polarization (up (U ), down (D)) and the sign of x-component of the group velocity (+, −). (b) Assuming high scattering among states in each individual groups, we assign four electrochemical potentials for these four groups: µ(U +), µ(U−), µ(D+), and µ(D−). External contact is modeled as up (vu) and down (vd) spin voltages applied to up and down states of the channel through up (gu) and down (gd) spin conductances per mode, respectively. Four − − different currents (i+ D) enter the four differ- U , and i ent groups in the channel. The contact can be either normal metal gu = gd or ferromagnet gu (cid:54)= gd. U , i+ D, i V.8. Scattering Matrix We make the following assumption (cid:104)Sµ(cid:105) ≈ (cid:104)S(cid:105)(cid:104)µ(cid:105), with (53) which when applied to the term related to the scattering rate S((cid:60)) in Eq. (52) becomes S(cid:60)↔(cid:60)(cid:48) = h L S((cid:60)) = S(cid:60)↔U + (cid:0)(cid:10)µ(cid:0)U +(cid:1)(cid:11) − (cid:104)µ ((cid:60))(cid:105)(cid:1) + S(cid:60)↔D−(cid:0)(cid:10)µ(cid:0)D−(cid:1)(cid:11) − (cid:104)µ ((cid:60))(cid:105)(cid:1) + S(cid:60)↔U−(cid:0)(cid:10)µ(cid:0)U−(cid:1)(cid:11) − (cid:104)µ ((cid:60))(cid:105)(cid:1) (cid:0)(cid:10)µ(cid:0)D+(cid:1)(cid:11) − (cid:104)µ ((cid:60))(cid:105)(cid:1) , + S(cid:60)↔D+ (cid:88) (cid:88) (cid:18) (cid:126)p,s∈(cid:60) (cid:126)p(cid:48),s(cid:48)∈(cid:60)(cid:48) − ∂f0 ∂E (cid:19) S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)). (55) (54) We can evaluate Eq. (54) for each of the group in Eq. (44) i.e. (cid:60) ≡ U +, D−, U−, and D+, which together in the {µ(U +), µ(D−), µ(U−), µ(D+)}T basis becomes the following matrix (see Appendix G for details) Eq. (51) for each group in Eq. (44) is given as 12 1 (cid:104)vx(cid:105) ∂ ∂t  =  M µ (U +) M µ (D−) N µ (U−) N µ (D+)   ∂ ∂x   +  + −u1 rs1 r ts rs1 −u1 ts r ts −u2 r rs2 rs2 −u2 r ts M −M −N N  ∂Ic ∂t   + M µ (U +) −M µ (D−) −N µ (U−) N µ (D+) µ (U +) µ (D−) µ (U−) µ (D+)   q ∂Q ∂t M M N N − qLM (cid:104)vx(cid:105) CE i+ U i− i− U i+ D (59) Note that the electrochemical potentials are referenced with respect to the constant µ0 i.e. µ = µ − µ0. i+ U , i− D, i− U , and i+ D are the currents per unit length entering into the four groups from an external contact (see Fig. 5(b)), which are given as [25] h q D   . q (cid:19) (cid:18) (cid:19) (cid:18) vu − µ(U +) (cid:19) (cid:18) vd − µ(D+) vu − µ(U−) (cid:19) (cid:18) vd − µ(D−) q q , , , . q i+ U = i+ D = i− U = and i− D = q2 h q2 h q2 h q2 h M gu N gd N gu M gd (60) Here, vu and vd are up and down spin voltages at the external contact respectively. gu and gd are up and down spin conductances per unit mode per unit length of the contact. The contact can be either NM (gu = gd) or FM (gu (cid:54)= gd). In steady-state, Eq. (59) reduces to our prior model [25]. V.9. Conversion to Charge-Spin Basis The charge and spin voltages and currents in the chan- nel are defined in terms of the four average electrochem- ical potentials as  IcRB 2Vs IsRB 2Vc  =  q h RB 1 −1 −1 1 α −α α −α − 1 1 α α 1 1 − 1 α 1 1 α 1    , M µ(U +) M µ(D−) N µ(U−) N µ(D+) (61) where RB is the ballistic resistance of the channel given in Eq. (5k) and α is an angular averaging factor. The derivation of Eq. (61) is given in Appendix H. −u1 −u(cid:48) SU +↔D− SU +↔U− SU +↔D+ SD−↔U− SD−↔D+ SU−↔D+ SU +↔D− SU +↔U− SD−↔U− SU +↔D+ SD−↔D+ SU−↔D+ −u2 −u(cid:48) 1 2  , (56)  [S] = where u1 = SU +↔D− + SU +↔U− + SU +↔D+, u(cid:48) 1 = SU +↔D− + SD−↔U− + SD−↔D+, u2 = SU +↔U− + SD−↔U− + SU−↔D+, 2 = SU +↔D+ + SD−↔D+ + SU−↔D+. and u(cid:48) The scattering matrix is such that the sum of each col- umn is zero satisfying the charge conservation and the sum of each row is zero satisfying the zero current re- quirement under equal potential. In addition, the time-reversal symmetry requires that SU +↔U− = SD−↔D+ and SU +↔D+ = SD−↔U− . (57) There are three types of scattering processes consid- ered in the channel: (a) reflection with spin flip rs1 = SU +↔D− and rs2 = SU−↔D+, (b) reflection without spin flip r = SU +↔U− = SD−↔D+, and (c) transmission with spin flip ts = SU +↔D+ = SD−↔U− . They are given by (cid:19) (cid:88) (cid:18) (cid:88) (cid:126)p,s∈U + (cid:126)p(cid:48),s(cid:48)∈D− (cid:88) (cid:18) (cid:88) (cid:126)p,s∈U− (cid:126)p(cid:48),s(cid:48)∈D+ rs1 = rs1 = h L h L − ∂f0 ∂E S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)), (58a) (cid:19) − ∂f0 ∂E S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)), (58b) (cid:88) (cid:88) (cid:126)p,s∈U + (cid:88) (cid:88) (cid:126)p(cid:48),s(cid:48)∈U− (cid:126)p,s∈D− (cid:126)p(cid:48),s(cid:48)∈D+ (cid:88) (cid:88) (cid:126)p,s∈U + (cid:88) (cid:88) (cid:126)p(cid:48),s(cid:48)∈D+ (cid:126)p,s∈D− (cid:126)p(cid:48),s(cid:48)∈U− r = = h L h L and ts = = h L h L (cid:18) (cid:18) − ∂f0 ∂E − ∂f0 ∂E (cid:19) (cid:19) S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)), S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)), (58c) (cid:18) (cid:18) − ∂f0 ∂E − ∂f0 ∂E (cid:19) (cid:19) S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)), S((cid:126)p, s ↔ (cid:126)p(cid:48), s(cid:48)). (58d) We have multiplied the second row of Eq. (61) with a factor 0 ≤ α ≤ 1 to take into account the angular distri- bution of the spin polarization of the eigenstates on the half Fermi circles indicated by U +, U−, D+, and D− in Fig. 5(a). The net z-spin polarization (or z-spin voltage) is expected to be lower by the factor α [21, 25, 26] which depends on the distribution of the spin polarization of the eigenstates and the details of the scattering mechanisms. The α factor introduced in Eq. (61) appears in Eq. (19) to indicate a lowering of the charge current induced spin voltage in the channel from the ideal value due to such angular distribution. Onsager reciprocity requires that the spin current induced charge voltage in the channel will be lowered by the same factor α as shown in Eq. (20), which has been taken into account by multiplying 1/α to the third row of Eq. (61). In the simplest ap- proximation, the angle between the z-axis and the spin polarization of the eigenstates of a particular half Fermi circle in Fig. 5(a) varies from −π/2 to +π/2 which yields α = 2/π. Combining Eq. (59) with Eq. (61) yields 1 (cid:104)vx(cid:105) ∂ ∂t  0 0 2 αλ(cid:48) − 2 λ s = −2LM ∂Ic ∂t   IcRB 2Vs IsRB 2Vc  0 0 0 − 2α2 λ0 0 0 2αp0 λ0 0 0 0 1 (cid:104)vx(cid:105) 2 CE ∂Q ∂t ∂ ∂x  +   IcRB 2Vs IsRB 2Vc 1 αp0 0 0    +  0 0 0 1 0 0 1 0 α2 0 1 α2 0 0 0 0 0 − 2 α2λs 2 αλ(cid:48) 0 0 p0 α 1   + (cid:0)i+ with the external contact terms given by U + i− ic = i+ 1 α is = (cid:1) D + i− U − i− RB 2 αRB (cid:0)i+ (cid:0)i+ U + i+ D, U − i+ D + i− D − i− U − i− D − i− U + i− D 2 D U + i+ U − i+ D (cid:1) , and (cid:1) . ∆vc = ∆vs =  IcRB 2Vs IsRB 2Vc   , RBic 2∆vs RBis 2∆vc (62) (63) 13 (64) (64) equation given by ∂Q ∂t + ∂Ic ∂x = ic. The first row of Eq. (62) combined with Eq. becomes ∂ ∂t Vc = (cid:18)(cid:104)vx(cid:105) RB 2 (cid:19)(cid:18) (cid:19) + 1 CE ic − ∂ ∂x Ic , (65) which is the first equation in Eq. (6). The second row of Eq. (62) combined with Eq. (64) becomes α2 RB 2(cid:104)vx(cid:105) ∂ ∂t Is + ∂ ∂x (cid:18) Vs = − α2RB λ0 1 (cid:104)vx(cid:105) 1 CE (cid:19) Is + 2αp0 λ0 ic − ∂Ic ∂x Vc + ∆vs. + αp0 We replace the expression for ic − ∂ Ic from Eq. (65) to ∂x get the second equation in Eq. (7). For contact conduc- tance G0 → 0, Eqs. (6) and (7) reduces to Eqs. (2) and (3) respectively. (66) V.11. Mean Free Paths given by We have three distinct mean free paths in Eq. (62), (cid:19) + 1 M 1 N 2 r + ts (cid:16) rs2 (cid:16) rs2 2 1 2 N 1 λ 1 λ0 1 λs = = = (cid:17) rs1 M + (cid:18) 1 + N + rs1 M + 1 M (cid:17) + r 2 (cid:18) 1 (cid:19) (cid:18) 1 N , and ts 2 + 1 M N , (cid:19) , (67) where λ, λ0, and λs determine the series charge resis- tance Rc (see Eq. (5c)), the series spin resistance Rs (see Eq. (5d)), and the shunt spin conductance Gsh (see (5e)) respectively. Note that λs depends on the spin-flip processes in the channel and determines the shunt con- ductance Gsh that takes into account the spin relaxation process. V.12. Charge-Spin Coupling Coefficients The other terms of Eq. (62) are given by (cid:16) rs2 (cid:16) rs2 N N (cid:17) (cid:17) − rs1 M − rs1 M + + r 2 ts 2 (cid:18) 1 (cid:18) 1 N N (cid:19) (cid:19) − 1 M − 1 M , and , (68) Eq. (63) combined with Eq. (60) yields Eqs. (8) and (9). V.10. Continuity Equation 1 λ(cid:48) = 1 = λ(cid:48) s 1 2 1 2 The term ∂Q/∂t on the right hand side of Eq. (62) is related to the charge currents according to the continuity representing coupling coefficients between charge and spin. λ(cid:48) s and λ(cid:48) cause a charge induced spin signal and 14 from a four-component diffusion equation obtained from the Boltzmann transport equation assuming linear re- sponse and elastic scattering in the channel. The four- component diffusion equation uses four average electro- chemical potentials based on a classification depending on the sign of z-component of spin (up (U ) or down (D)) and the sign of x-component of group velocity (+ or −). Such classification can be viewed as an extension of the Valet-Fert equation [24] which uses two electrochem- ical potentials for U and D states. For a normal metal channel, the time-dependent model presented here de- couples into (i) the well-known transmission line model for charge transport in quantum wires [48 -- 50] and (ii) a time-dependent version of Valet-Fert equation [24] for spin transport. We first derive several results on charge- spin interconversion starting from our model in steady- state. The steady-state results show good agreement with existing experiments on diverse materials. We then study the phenomenon of spin-charge separation using our full time-dependent model, especially in the mate- rials with SOC exhibiting SML. Our model shows the expected spin-charge separation with two distinct veloc- ities for charge and spin, which persist even in channels exhibiting SML. However, we show that the lower veloc- ity signal is purely spin while the higher velocity signal is largely charge with an additional spin component pro- portional to the degree of SML. spin induced charge signal respectively. Note that the first terms of Eq. (68) indicate a purely scattering in- duced spin-charge coupling even if M = N (i.e. p0 = 0) since rs1 and rs2 are two independent parameters and there could be situations where rs1 (cid:54)= rs2. In this paper, we restrict ourselves to SML caused by difference between M and N (i.e. p0 (cid:54)= 0). We can eliminate the first terms in Eq. (68) by assuming either of the followings: rs1 = rs2 = rs, or, rs1 M = rs2 N . (69a) (69b) The first assumption Eq. (69a) when applied in Eq. (cid:18) 1 (cid:18) 1 N N 1 λ(cid:48) = (rs + r) 1 = (rs + ts) λ(cid:48) s (cid:19) (cid:19) = , and − 1 M − 1 M p0 λ p0 λs = (70) , (68) yields which in turn gives λr = λ and λt = λs in Eqs. (5i) and (5h) respectively. The second assumption Eq. (69b) when applied in Eq. (68) yields (cid:18) 1 (cid:18) 1 N N 1 λ(cid:48) = r 1 λ(cid:48) s = ts (cid:19) (cid:19) − 1 M − 1 M = = p0 λr p0 λt , and . V.13. Comments on the Assumptions The assumptions in Eqs. (48), (53), and (69) result in an effective change in the transmission line model param- eters in Eqs. (5), but does not change the models in Eqs. (2), (3) (Fig. 1) and Eqs. (6), (7) (Fig. 2) themselves. The assumptions made to derive the model can be re- visited as the field evolves. However, several predictions from our model for steady-state [25] have already received support from experiments [27 -- 29] suggesting that the as- sumptions are within the reasonable limits. (71) ACKNOWLEDGMENTS This work was in part supported by FAME, one of six centers of STARnet, a Semiconductor Research Corpora- tion (SRC) program sponsored by MARCO and DARPA and in part by ASCENT, one of six centers in JUMP, a SRC program sponsored by DARPA. Appendix A: Steady-State Results This appendix provides the details of the derivation of Eq. (18) and Eq. (23). 1. Derivation of the Resistance Matrix VI. SUMMARY Potentiometric NM Contacts: Eqs. (8) and (9) for pf = 0 becomes We have proposed a two component (charge and z- component of spin) transmission line model for channels with spin-momentum locking (SML) which is a new ad- dition to our SPICE compatible multi-physics model li- brary [45 -- 47]. The model enables easy analysis of com- plex geometries involving materials with spin-orbit cou- pling (SOC) observed in diverse classes of materials e.g. topological insulators, heavy metals, oxide interfaces, and narrow bandgap semiconductors. The model is derived G0 ic = G0 (vc − Vc) , (cid:18) α2 (vs − Vs) , is = (cid:18) ∆vc = G0RB 2α αG0RB and ∆vs = 2 p0vs − αIcRB 2 p0vc − αIsRB 2 (cid:19) (cid:19) , (A1a) (A1b) (A1c) . (A1d) We apply Eqs. (A1a) and (A1b) in Eqs. (A1c) and (A1d) respectively, which yields ∆vc = ∆vs = αp0 2GB αp0 2GB is + p0 ic + p0 G0RB 2α αG0RB Vs − Rc Vc − Rs 2 contIc, contIs. (A2) We assume that the contact conductance per unit length G0 is very low such that the potentiometric con- dition in Eq. (17) is satisfied. Thus, we have Rc (cid:29) Rc ηc (cid:29) G0RB 2α cont, Rs (cid:29) Rs cont, , and ηs (cid:29) αG0RB 2 . Under this condition, we combine Eq. (16) with Eq. (A2) for ic = 0, which yields d dx d dx d dx d dx Ic = 0, Vc = −Rc Ic + p0ηcVs + αp0 2GB Is = −GshVs + p0γsIc + is, Vs = −RsIs + p0ηsVc. and (A3a) is, (A3b) (A3c) (A3d) Uniform Spin Voltage: We assume that the spin volt- age Vs is uniform in the channel region of interest (from x = 0 to x = L where L is the channel length) d dx Vs = 0. Thus from Eq. (A3d) we can write Is = 2p0GB α Vc, (A4) (A5) where we have used the definitions in Eqs. (5d) and (5h). Differentiating both sides of Eq. (A5) with respect to x and combining with Eqs. (A3b) and (A3c) yields (cid:18) From Eq. (A1b) we have Vs = vs − α2is G0 , −GshVs+p0γsIc+is = 2p0GB α −Rc Ic + p0ηcVs + αp0 2GB is which in conjunction with the definitions in Eq. (5) yields (cid:18) 1 (cid:18) 1 λt λs + + 1 λ p2 0 λr (cid:19) (cid:19) Ic + α2(cid:0)1 − p2 (cid:1) 0 4GB Vs = αp0 2GB (cid:19) is. (cid:18) 1 λs 1 + p2 0 λr (A6) (A7) which when combined with Eq. (A6), becomes + λt αp0 2GB (cid:18) 1 (cid:18) 1  α2(cid:0)1 − p2 4GB λs 0 + (cid:1) 1 λ p2 0 λr (cid:19) (cid:19) Ic (cid:18) 1 1 + λs vs = +  is. (cid:19) + α2 G0 p2 0 λr 15 (A8) Charge Voltage in the Channel : Let us assume that the contact 1 is located at x = 0 and contact 2 is located at x = L as shown in Fig. 3, where L is the channel length between the two contacts. The charge voltage in the channel at x = 0 and x = L are Vc1 and Vc2 respectively. According to (A3a), Ic is constant along the channel. Since Vs and is is also uniform along the channel, we can integrate Eq. (A3b) from x = 0 to x = L as Vs − αp0 2GB RBIc − 2p0L αλr Vc1 − Vc2 = (A9) isL. L λ From Eq. (A1a), we can write equation for Vc1 in terms of the terminal voltage vc1 of contact 1 at x = 0 as Vc1 = vc1 − Ic G(cid:48)(cid:48) 0 . (A10) noting that ic = Ic. Here, G(cid:48)(cid:48) 0 is the contact conductance of contact 1. Similarly, we can write an equation for contact 2 noting that ic = −Ic given by Vc2 = vc2 + Ic G(cid:48)(cid:48) 0 , (A11) where we assume that contact 1 and 2 has same conduc- tance. We combine Eq. (A9) with Eqs. (A6), (A10), and (A11) to have vc1 − vc2 = (cid:19) ,  L λ − p2 0L λr (cid:16) 1 (cid:16) 1 λt λs (cid:17) (cid:17) + 1 λ + p2 0 λr  RB + (cid:18) 1 (cid:18) 1 λr + + λs (cid:48)(cid:48)  Ic (cid:19) (cid:19) isL. 2 G0 1 λs p2 0 λr − αp0 2GB (A12) Scattering Condition: We assume that the reflection with spin-flip scattering mechanism is dominant in the channel rs1,2 (cid:29) r, ts. (A13) This condition in Eqs. (67), (70), and (68) yields 1 λ ≈ 1 λs , and 1 λ , 1 λs (cid:29) 1 λ0 , 1 λr , 1 λt . (A14) Applying Eq. (A14) in Eqs. (A12) and (A8) we have (18) respectively by the first and second rows of Eq. setting ∆Vc = vc1 − vc2, G(cid:48) 0 = G0L and is tot = isL. 2. Derivation of the IREE Length We start from the first row of Eq. (18) given by (cid:18) L (cid:19) ∆Vc = + 2 G(cid:48)(cid:48) 0 Ic − αp0 2GB is tot. GBλ We assume that the channel resistance is much larger than the contact resistances, i.e. L/(GBλ) (cid:29) 2/G(cid:48)(cid:48) 0 . This yields ∆Vc = L GBλ Ic − αp0 2GB is tot. (A15) We apply the short circuit condition vc1 = vc2 at con- tacts 1 and 2 i.e. ∆Vc = 0. Thus we get Ic = αp0λ 2 is tot L . The IREE length is given by Eq. (21) as λIREE = Jc Js = Ic/w Is/(wL) = αp0λ 2 , (A16) (A17) which is the expression in Eq. (22). Applying α = 2/π yields Eq. (23). Appendix B: Simulation Setup This appendix provides the details of the simulation setup in SPICE that was used to analyze steady-state results of charge-spin interconversion in Section III. We have discretized the structure in Figs. 3(a), 3(c), and 4(a) into 100 small sections and represented each of the small sections with the corresponding circuit model. For example, Block 1 and block 2 indicated in Fig. 6 are represented with the models in Figs. 1 and 2 respectively. Note that each of the nodes in Fig. 6 are two component: charge (c) and z-component of spin (s). We have con- nected the charge and spin terminals of the models for all the small sections in a modular fashion using stan- dard circuit rules as shown in Fig. 6. We perform a dc simulation in SPICE. Note that during dc simulation in SPICE, capacitors and inductors automatically become open and short circuit respectively and correspond to the stead-state (∂/∂t → 0) form in Eq. (16). The contacts (1, 2, and 3) in this discussion are point contacts. The contact polarizations pf = 0 and conduc- tances are in the potentiometric limit with G0 ≈ 0.05GB. We set the total number of modes M + N in the chan- nel to be 100. We have assumed that the reflection with spin-flip scattering mechanism is dominant in the chan- nel i.e. rs1,2 (cid:29) r, ts. The scattering rate per unit mode was set to 0.04 per lattice point. We apply the charge open and spin ground boundary condition at the two boundaries given by (cid:40) (cid:41) ic vs = L (cid:40) (cid:41) 0 0 (cid:40) (cid:41) ic vs = R (cid:40) (cid:41) 0 0 , and . (B1) Here, ic and vs indicates boundary charge current and boundary spin voltage. Indices L and R indicate left and right boundaries respectively. 16 Setup in Fig. 3(a): For setup in Fig. 3(a), we apply a current Ic at the charge terminals of contacts 1 and 2, (cid:41) given by(cid:40) (cid:40) (cid:41) ic vs = 1 Ic 0 (cid:40) (cid:41) ic vs = 2 (cid:40) −Ic (cid:41) 0 and . (B2) where indices 1 and 2 represent contacts 1 and 2 re- spectively. The spin terminals of contacts 1 and 2 are grounded to take into account the spin relaxation within the contact. Both charge and spin terminals of contact 3 are open and we observe open circuit spin voltage at the spin terminal. The boundary condition at contact 3: , (B3) (cid:40) (cid:41) ic is = 3 (cid:40) (cid:41) 0 0 (cid:41) (cid:40) (cid:41) (cid:40) ic is = 3 0 is tot where index 3 indicate the contact 3. Setup in Fig. 3(c): For the setup in Fig. 3(c), the charge terminal of contact 3 is open and we apply a cur- rent is tot at the spin terminal, given by . (B4) The spin terminals of contacts 1 and 2 are grounded to take into account the spin relaxation within the contact and charge terminals are kept open. Here we observe the open circuit voltage difference between the charge terminals of contacts 1 and 2. The boundary conditions are: (cid:40) (cid:41) (cid:40) (cid:41) ic vs = 1 (cid:40) (cid:41) ic vs = 2 (cid:40) (cid:41) 0 0 and . (B5) Setup in Fig. 4(a): For the setup in Fig. 4(a), charge terminal of contact 3 is open and we inject a current is tot through the spin terminal. The boundary condition at contact 3: 0 0 (cid:40) (cid:41) (cid:40) (cid:41) ic is = 3 0 is tot . (B6) We short circuit the charge terminals of contacts 1 and 2 and observe the short circuit charge current Ic flowing in the channel induced by is tot. The spin terminals of con- tacts 1 and 2 are grounded to take into account the spin relaxation within the contact. The boundary conditions at contacts 1 and 2 are: (cid:40) (cid:41) (cid:40) (cid:41) (cid:40) (cid:41) (cid:40) (cid:41) vc vs = 1 vc 0 and 2 vc vs = 2 vc 0 (B7) . 1 17 TABLE III. Estimation of GB. Material w [µm] t [nm] M + N GB AgBi† CuBi† Bi2Se3 5 [51] 400 [51] 4.6 × 107 (Eq. (C2)) 0.15 [52] 20 [52] 8.8 × 104 (Eq. (C2)) 3.3 × 105 (Eq. (C3)) 1000 [53] 9†† [53] 3.6 × 106 (Eq. (C2)) LAOSTO 400 [9] - (Eq. (C4)) 1.77 kS 3.4 S 12.6 S 140 S † We have used the thicknesses of the most conductive layer (Ag and Cu respectively) to estimate GB considering bulk conduction. †† We considered 6 quintuple layer (QL) sample in Ref. [53]. 1 QL ≈ 1.5 nm. where w is the width and t is the thickness. For a 2D channel the expression is M + N = 2kF w π . (C3) We estimate the ballistic conductance of the channel using the following expression GB = q2 h (M + N ). (C4) Estimations are summarized in Table III. c. Estimation of Mean Free Path (λ) FIG. 6. SPICE setup for dc simulation. The setup connects the transmission line circuit models in a distributed manner. Block 1 corresponds to model in Fig. 1 and Block 2 corre- sponds to model in Fig. 2. All the external contact have pf = 0. All terminals are two component: charge (c) and spin (s). All indicated spin terminals are grounded except the spin terminal of contact 2. TABLE II. Estimation of kF . kF [nm−1] Electron Density Material 5.86 × 1028 m−3 [62] AgBi† 12 (Eq. (C1a)) 8.49 × 1028 m−3 [62] 13.6 (Eq. (C1a)) CuBi† LAOSTO 2.6 × 1017 m−2 [9] 1.278 (Eq. (C1b)) 1.59 (Eq. (C1b)) Bi2Se3 4 × 1017 m−2 [53] † We have used the electron densities of Ag and Cu respectively as they are the most conductive layer in the corresponding bi-layer. Appendix C: Parameter Estimations This appendix provides the details of the estimations made for the IREE lengths on diverse materials using Eq. (23). 1. Estimations a. Estimation of Fermi Wave Vector (kF ) The mean free path (λ) is estimated from the measured sheet resistance RS of the sample using the following ex- pression: We estimate the Fermi wave vector kF of the chan- nel from the electron density n3D (units of m−3) or n2D (units of m−2), using the following expressions kF = 3(cid:112) 3π2n3D, √ 2πn2D. kF = RS w = 1 GBλ , (C5) or from the resistivity ρ of the sample using the following expression: (C1a) ρ wt = 1 GBλ . (C6) (C1b) Estimations are summarized in Table IV. Estimations are summarized in Table II. b. Estimation of Ballistic Conductance (GB) We estimate the total number of modes in the 3D chan- nel using the following expression M + N = k2 F wt 2π , (C2) d. Estimation of Degree of Spin-Momentum Locking (p0) For AgBi, CuBi, and LAOSTO, the degree of SML p0 is estimated using the Rashba coupling coefficient (v0 = αR/) and the Fermi velocity (vF ) of the mate- rials, using Eq. (24). For Bi2Se3, p0 is estimated using spin current density reported from spin pumping (Js), measured inverse effect voltage ∆Vc, and Eq. (20). The estimations are summarized in Table V. λ [nm] the expression in Eq. (24). Assuming UE = 0 and applying EF = 18 1 2 mv2 F we have TABLE IV. Estimation of λ. Material RS [Ω/(cid:3)] ρ [µΩ-cm] AgBi CuBi 10† - LAOSTO 176†† Bi2Se3 - - 100‡ - 2000‡‡ 22.6 (Eq. (C5)) 0.88 (Eq. (C6)) 180.8 (Eq. (C5)) 3.2 (Eq. (C6)) † Taken from Fig. 3(a) of Ref. [51] for AgBi sample with 5 nm Ag. †† Taken from Fig. 1(d) of Ref. [9] for LAOSTO at 7K. ‡ ρ of Bi layer was used which was taken from Ref. [52]. ‡‡ Taken from Fig. 2(b) of Ref. [53] at ∼300K. TABLE V. Estimation of p0. Material αR [eV·A] vF [×106 m·s−1] AgBi 1.39 (Ag) [62] 1.87 (Bi) [62] 1.57 (Cu) [62] 1.87 (Bi) [62] CuBi 0.56 [51] 0.56 [52] LAOSTO 0.03 [9] 0.074† Material ∆Vc [µV] 40‡ [53] Bi2Se3 Js [A-m−2] 1.39×105 ‡‡ p0 0.05 0.05 0.0616 p0 0.025 (Eq. (20)). † kF m∗ . m∗ ≈ 2 × We estimate the Fermi velocity using vF = 9.1 × 10−31 kg as reported in Ref. [9]. ‡Inverse effect induced voltage due to spin-pumping at 3 GHz. ‡‡ Taken from table 1 in supplementary information of Ref. [53] for 6 QL. The sample dimension is 1 mm × 5 mm [53], which yields is tot ≈ 0.695 A. 2. Derivation of Eq. (24) We start from the eigenstates in Eq. (37) of the Rashba Hamiltonian in Eq. (36), given by E = p(cid:48)2 2m − s v0p(cid:48) + UE. with p(cid:48) = (cid:126)p − q (cid:126)A. Solutions for p(cid:48) are given by 0 + 2m(E − UE), 0 + 2m(E − UE), 1(s) = s mv0 +(cid:112)m2v2 2(s) = s mv0 −(cid:112)m2v2 p(cid:48) p(cid:48) 1(s = +1) and p(cid:48) 1(s = −1) 2(s = 2(s = +1) correspond to M and N respec- noting that s2 = 1. Here, p(cid:48) correspond to M and N respectively. Similarly p(cid:48) −1) and p(cid:48) tively. Thus the degree of SML p0 is given by 1(s = −1) 1(s = −1) p0(EF ) = p(cid:48) p(cid:48) 1(s = +1) − p(cid:48) (cid:114) 1(s = +1) + p(cid:48) v0 2(EF − UE) v2 0 + m = . (C7) Appendix D: Charge and Spin Velocities This appendix provides the derivation of eigenvalues and eigenvectors of Eqs. (2) and (3) to find the charge and spin velocities and their coupling while propagation. The matrix form of Eqs. (2) and (3) is given by  Vc Vs Ic Is  ∂t  ∂   +  = − −p0rmCef f Cef f 0 0 0 0 0 0 0 0 0 CQ α2 p0gmLM 0 0 Lef f 0 0  α2LK Vc Vs Ic Is   .  ej(kx−ωt), 0 0 0 Gsh −p0γs 0 −p0ηc Rc 0 Rs 0 0 −p0ηs 0 0 0 1 1 0 0 0 0 1 0 0  0 0 1 0  Vc Vs Ic Is ∂x  ∂  ≡   Vc Vs Ic Is Vc Vs Vc Vs (D1) In the low-loss limit, we assume a solution of the form (D2)    Vc Vs Vc Vs which results in  −jω  + = −jk Cef f 0 0 0 CQ α2 p0gmLM 0 0 0 Lef f 0 0 0 0  α2LK Vc Vs Vc Vs   0 −p0rmCef f 0 0 0  0 0 1 0 0 0 0 1 1 0 0 0 0 1 0 0 0 0 Gsh −p0γs 0 −p0ηc Rc 0 0 Rs 0   Vc Vs Vc Vs  . −p0ηs (D3) We assume that the co-efficients of the transmission line model (Eqs. (2) and (3)) are constant with frequency ω and the propagation vector k. We differentiate both sides of Eq. (D3) with respect to k and the following  , matrix equation  Vc Vs Vc Vs  = vg  0 − α2p0gmLM CQLef f 1 Lef f 0 0 0 0 1 α2LK 1 Cef f 0 0 p0rm α2LK 0 α2 CQ 0 0 where vg is the group velocity given by vg = ∂ω ∂k .   Vc Vs Vc Vs (D4) (D5) 1 1 = Lef f Vc Ic (cid:41) (cid:40) (cid:40) Note that the eigenvalues of Eq. (D4) give the veloc- ities in Eqs. (25) and (27). For a particular eigenvalue vg, we can write the following equation from Eq. (D4) (cid:40) (cid:41)  α2gmLM  −vg  .  −vg For the eigenvalue vg = vg,s = 1/(cid:112)LKCQ in Eq. (D6) (26). Again, for the eigenvalue vg = vg,c = 1/(cid:112)Lef f Cef f we have Vc = Ic = 0 and assuming Is = 1 we get Eq. Cef f −vg 1 CQ −vg CQLef f − rm α2LK 1 LK (cid:41) Vs Is = p0 (D6) 0 0 Vc Ic , 19 E(x, t, (cid:126)p(cid:48), s(cid:48)) − µ(x, t, (cid:126)p(cid:48), s(cid:48)) and ξ0(x, (cid:126)p, s) ≡ E(x, (cid:126)p, s) − µ0 and assume that ξ and ξ(cid:48) are close to ξ0 which gives (ξ − ξ0) , and (ξ(cid:48) − ξ0) . ξ=ξ0 ξ(cid:48)=ξ0 (E2) (cid:18) (cid:18) (cid:18) ∂f − ∂f ∂ξ − ∂f ∂ξ(cid:48) (cid:19) (cid:19) (cid:19) = (cid:19) f ≈ f0 + f(cid:48) ≈ f0 + (cid:18) ∂f ∂ξ (cid:19) (cid:18) = ∂f0 ∂E (cid:18) ∂f0 (cid:19) ∂E We set ξ=ξ0 ∂ξ(cid:48) ξ(cid:48)=ξ0 hand side of Eq. (31) becomes Thus the right f − f(cid:48) = − ∂f0 ∂E (ξ − ξ(cid:48)) = (µ − µ(cid:48)), (E3) noting that E(x, t, (cid:126)p(cid:48), s(cid:48)) = E(x, t, (cid:126)p, s) in the elastic scat- tering limit. Appendix F: E-p relation (cid:32) This appendix provides the derivation of Eq. (42), start- ing from the E-p relation in Eq. (38). Differentiating Eq. (38) with respect to time t yields (cid:126)p − q (cid:126)A = − s v0 ∂E ∂t (F1) The velocity (cid:126)v(E) = ∇(cid:126)pE is derived from Eq. (38) as ∂UE ∂t −q m + · . ∂ (cid:126)A ∂t (cid:126)p − q (cid:126)A (cid:126)p − q (cid:126)A (cid:33) (cid:32) (cid:33) we get Eq. (28) assuming Ic = 1. Appendix E: Derivation of semiclassical model (cid:126)v(E) = (cid:126)p − q (cid:126)A m − s v0 (cid:126)p − q (cid:126)A (cid:126)p − q (cid:126)A . (F2) This appendix provides the derivation of Eq. (35), start- ing from the Boltzmann transport equation in Eq. (31). We apply a variable transformation ξ(x, t, (cid:126)p, (cid:126)s) ≡ E(x, t, (cid:126)p, (cid:126)s)−µ(x, t, (cid:126)p, (cid:126)s) on the left hand side of Eq. (31), which yields that Eq. (F2) can be written as (cid:32)(cid:126)p − q (cid:126)A (cid:33) (cid:126)v(E) = + s v0 m (cid:126)p − q (cid:126)A (cid:126)p − q (cid:126)A . (F3) Combining Eqs. (F1) and (F2) yields Eq. (39). Note , (cid:19) (cid:18) ∂E (cid:18) ∂E (cid:18) ∂E ∂t ∂f ∂ξ − ∂µ ∂t − ∂µ ∂x − ∂µ ∂px (cid:19) ∂f ∂ξ ∂px ∂x ∂f ∂t = ∂f ∂ξ vx Fx ∂f ∂x ∂f ∂px = vx = Fx , and (cid:19) where (cid:126)p − q (cid:126)A (cid:126)p − q (cid:126)A is an unit vector along (cid:126)p − q (cid:126)A. From Eq. (E1) (38) we get (cid:114) . ∂E ∂px . Finally, We first substitute Fx = − ∂E ∂x and vx = we set ∂f ∂ξ = ∂f0 ∂E to get the left hand side of Eq. (35). On the right hand side of Eq. 1(cid:14) (1 + exp ((E(x, (cid:126)p, s) − µ0)/kBT )) with constant elec- (31), we ex- pand both f and f(cid:48) into Taylor series around f0 = trochemical potential µ0. We set ξ(cid:48)(x, t, (cid:126)p(cid:48), s(cid:48)) ≡ (cid:126)p − q (cid:126)A = −s m v0 ± m 2 (E − UE) m v2 0 + , (F4) noting that s2 = 1. Thus from Eq. (F3) we get (cid:126)p − q (cid:126)A (cid:126)p − q (cid:126)A . 2 (E − UE) (cid:126)v(E) = ± v2 0 + m (cid:114) (F5) Note that the magnitude of the velocity for a particular energy of interest is same for all four groups in Eq. (44) for the Rashba Hamiltonian in Eq. (36) considered here. Appendix G: Scattering Rates We assume the following This appendix provides the derivation of the scattering matrix in Eq. (56). For the group (cid:60) ≡ U +, D−, U−, and D+, we have from Eq. (54) + SU +↔D+ S(D−) = SD−↔U + S(U +) = SU +↔D−(cid:0)(cid:10)µ(cid:0)D−(cid:1)(cid:11) −(cid:10)µ(cid:0)U +(cid:1)(cid:11)(cid:1) + SU +↔U−(cid:0)(cid:10)µ(cid:0)U−(cid:1)(cid:11) −(cid:10)µ(cid:0)U +(cid:1)(cid:11)(cid:1) (cid:0)(cid:10)µ(cid:0)D+(cid:1)(cid:11) −(cid:10)µ(cid:0)U +(cid:1)(cid:11)(cid:1) , (cid:0)(cid:10)µ(cid:0)U +(cid:1)(cid:11) −(cid:10)µ(cid:0)D−(cid:1)(cid:11)(cid:1) + SD−↔U−(cid:0)(cid:10)µ(cid:0)U−(cid:1)(cid:11) −(cid:10)µ(cid:0)D−(cid:1)(cid:11)(cid:1) (cid:0)(cid:10)µ(cid:0)D+(cid:1)(cid:11) −(cid:10)µ(cid:0)D−(cid:1)(cid:11)(cid:1) , (cid:0)(cid:10)µ(cid:0)U +(cid:1)(cid:11) −(cid:10)µ(cid:0)U−(cid:1)(cid:11)(cid:1) + SU−↔D−(cid:0)(cid:10)µ(cid:0)D−(cid:1)(cid:11) −(cid:10)µ(cid:0)U−(cid:1)(cid:11)(cid:1) (cid:0)(cid:10)µ(cid:0)D+(cid:1)(cid:11) −(cid:10)µ(cid:0)U−(cid:1)(cid:11)(cid:1) , (cid:0)(cid:10)µ(cid:0)U +(cid:1)(cid:11) −(cid:10)µ(cid:0)D+(cid:1)(cid:11)(cid:1) + SD+↔D−(cid:0)(cid:10)µ(cid:0)D−(cid:1)(cid:11) −(cid:10)µ(cid:0)D+(cid:1)(cid:11)(cid:1) + SD+↔U−(cid:0)(cid:10)µ(cid:0)U−(cid:1)(cid:11) −(cid:10)µ(cid:0)D+(cid:1)(cid:11)(cid:1) . S(U−) = SU−↔U + S(D+) = SD+↔U + + SD−↔D+ + SU−↔D+ and (G1) (G2) (G3) (G4) (G1), Eqs. yields the scattering matrix in Eq. {µ(U +), µ(D−), µ(U−), µ(D+)}T basis. (G3), (G2), and (G4) together (56) in the Appendix H: Charge and Spin Currents and Voltages This appendix provides the derivation of the Eq. (61). The current in any group is given by (cid:88) (cid:126)p,s∈(cid:60) I ((cid:60)) = q L where f is given by Eq. (32). Under the linear response approximation, we can write f (x, t, (cid:126)p, s) ≈ f0 + − ∂f0 ∂E (µ (x, t, (cid:126)p, s) − µ0) (H2) (cid:18) (cid:19) where, f0 is given by Eq. (33) with constant electrochem- ical potential µ0. Thus from Eq. (H1), we can write f0 (cid:88) (cid:126)p,s∈(cid:60) I ((cid:60)) = q L D0 ((cid:60)) 2 vx + (cid:16)(cid:104)vxµ(cid:105)(cid:126)p,s∈(cid:60) − µ0 (cid:104)vx(cid:105)(cid:17) , (H3) where D0((cid:60)) is given by Eq. (47) and the averaging is defined by Eq. (45). vx f (x, t, (cid:126)p, s) . (H1) (cid:126)p,s∈U + (cid:126)p,s∈D− (cid:104)vxµ(cid:105)(cid:126)p,s∈(cid:60) ≈ (cid:104)vx(cid:105)(cid:104)µ(cid:105)(cid:126)p,s∈(cid:60) , (cid:17) (cid:16)(cid:104)µ(cid:105)(cid:126)p,s∈(cid:60) − µ0 (cid:88) vx, f0 (cid:126)p,s∈(cid:60) nm ((cid:60)) q h +sgn (vx) which results in I ((cid:60)) = sgn ((cid:104)vx(cid:105)) q L where nm((cid:60)) is given by Eq. (50). 1. Charge Current which in conjunction to Eq. (H4) yields Ic = The charge current in the channel is given by L f0 q h Ic = I(cid:0)U +(cid:1) + I(cid:0)D+(cid:1) + I(cid:0)U−(cid:1) + I(cid:0)D−(cid:1) , (cid:1) vx + M(cid:0)µ(cid:0)U +(cid:1) − µ0 (cid:1) vx + N(cid:0)µ(cid:0)D+(cid:1) − µ0 (cid:1) vx + N(cid:0)µ(cid:0)U−(cid:1) − µ0 (cid:1) , vx + M(cid:0)µ(cid:0)D−(cid:1) − µ0  h  h  h  h (cid:88) (cid:88) (cid:88) (cid:88) − q h (cid:126)p,s∈D+ (cid:126)p,s∈U− (cid:126)p,s∈U + q h f0 f0 L f0 L + − q h L (cid:126)p,s∈D− where nm(U +) = nm(D−) = M and nm(U−) = nm(D+) = N . Note that (U +, D−) and (U−, D+) are time-reversal symmetric pairs, hence (cid:88) (cid:88) vx = vx = (cid:88) (cid:88) (cid:126)p,s∈U− (cid:126)p,s∈D+ vx, and vx, (H7) Thus the expression for charge current is given by (cid:0)M µ(cid:0)U +(cid:1) − N µ(cid:0)U−(cid:1) + N µ(cid:0)D+(cid:1) − M µ(cid:0)D−(cid:1)(cid:1) , q h Ic = where we defined µ = µ − µ0. 2. Spin Current The spin current in the channel is given by (cid:0)I(cid:0)U +(cid:1) + I(cid:0)U−(cid:1) − I(cid:0)D+(cid:1) − I(cid:0)D−(cid:1)(cid:1) , Is = 1 α 20 (H4) (H5) (H6) (H8) (H9) 21 We subtract the equilibrium part from our definition of spin current given by (cid:0)M µ(cid:0)U +(cid:1) − N µ(cid:0)U−(cid:1) − N µ(cid:0)D+(cid:1) + M µ(cid:0)D−(cid:1)(cid:1) . Is = 1 α q h (H12) 3. Spin Voltage where α = 2/π is an angular averaging factor to take into account the average z · (cid:126)s on a half Fermi circle (see Fig. 5(a)). Eq. (H9) in conjunction to Eq. (H4) yields q h q h q h L  h  h  h  h L L L f0 f0 f0 f0 (cid:126)p,s∈U + (cid:88) (cid:88) (cid:88) (cid:88) (cid:126)p,s∈D+ (cid:126)p,s∈U− (cid:1) vx + M(cid:0)µ(cid:0)U +(cid:1) − µ0 (cid:1) vx + N(cid:0)µ(cid:0)U−(cid:1) − µ0 (cid:1) vx + N(cid:0)µ(cid:0)D+(cid:1) − µ0 (cid:1) . vx + M(cid:0)µ(cid:0)D−(cid:1) − µ0 (cid:126)p,s∈D− Is = 1 α − 1 α − 1 α + 1 α q h Is = 1 α q h Applying the condition in Eq. (H7), we have the ex- pression for channel spin current as (cid:0)M µ(cid:0)U +(cid:1) − N µ(cid:0)U−(cid:1) − N µ(cid:0)D+(cid:1) + M µ(cid:0)D−(cid:1)(cid:1)  vx − (cid:88)  (cid:88) vx f0 + 2 α q L (cid:126)p,s∈U + (cid:126)p,s∈U− (H11) Note first term is zero at equilibrium condition µ(U +) = µ(U−) = µ(D+) = µ(D−) = µ0. However, the second term is non-zero even at equilib- rium since (cid:88) (cid:88) (H10) The spin voltage in the channel is given by q Vs = α M µ (U +) + N µ (U−) − N µ (D+) − M µ (D−) 2 (M + N ) . (H13) Subtracting each electrochemical potential by µ0 which gives q Vs = α M µ(U +) + N µ(U−) − N µ(D+) − M µ(D−) 2 (M + N ) 4. Charge Voltage . (H14) The charge voltage in the channel is given by q V c = M µ(U +) + N µ(U−) + N µ(D+) + M µ(D−) 2 (M + N ) . (H15) Subtracting each electrochemical potential by µ0 ac- cording to q V c = M µ(U +) + N µ(U−) + N µ(D+) + M µ(D−) 2 (M + N ) . vx (cid:54)= (cid:126)p,s∈U + (cid:126)p,s∈U− vx where Vc = Vc − µ0 q . (H16) and represent the equilibrium spin current in the channel. written in a matrix form given by Eq. (61). Eqs. (H8), (H12), (H14), and (H16) together can be [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [2] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [3] Y. Fan and K. L. Wang, SPIN 06, 1640001 (2016). [4] I. M. Miron, K. Garello, G. Gaudin, P.-J. Zermatten, M. V. Costache, S. Auffret, S. Bandiera, B. Rodmacq, A. Schuhl, and P. Gambardella, Nature 476, 189 (2011). [5] T. Suzuki, S. Fukami, N. Ishiwata, M. Yamanouchi, S. Ikeda, N. Kasai, and H. Ohno, Appl. Phys. Lett. 98, 142505 (2011). [6] C.-F. Pai, L. Liu, Y. Li, H. W. Tseng, D. C. Ralph, and R. A. Buhrman, Appl. Phys. Lett. 101, 122404 (2012). Vaz, H. Naganuma, G. Sicoli, J.-P. Attan´e, M. Jamet, E. Jacquet, J.-M. George, A. Barth´el´emy, H. Jaffr´es, A. Fert, M. Bibes, and L. Vila, Nature Mater. 15, 1261 (2016). [10] Q. Song, H. Zhang, T. Su, W. Yuan, Y. Chen, and W. Han, Sci- 10.1126/sciadv.1602312, W. Xing, J. Shi, J. Sun, ence Advances 3 (2017), http://advances.sciencemag.org/content/3/3/e1602312.full.pdf. [11] P. R. Hammar and M. Johnson, Phys. Rev. B 61, 7207 (2000). [12] Y. H. Park, H. Cheol Jang, H. C. Koo, H.-j. Kim, J. Chang, S. H. Han, and H.-J. Choi, Journal of Ap- plied Physics 111, 07 (2012). [7] L. Liu, C.-F. Pai, Y. Li, H. W. Tseng, D. C. Ralph, and [13] R. H. Silsbee, Journal of Physics: Condensed Matter 16, R. A. Buhrman, Science 336, 555 (2012). R179 (2004). [8] B. Yan, B. Stadtmller, N. Haag, S. Jakobs, J. Seidel, D. Jungkenn, S. Mathias, M. Cinchetti, M. Aeschlimann, and C. Felser, Nat. Commun. 6, 10167 (2015). [9] E. Lesne, Y. Fu, S. Oyarzun, J. C. Rojas-S´anchez, D. C. [14] S. Tolle, U. Eckern, and C. Gorini, Phys. Rev. B 95, 115404 (2017). [15] X. Liu and J. Sinova, Phys. Rev. B 86, 174301 (2012). [16] X. Liu and J. Sinova, Phys. Rev. Lett. 111, 166801 22 (2013). (2013). [17] E. G. Mishchenko, A. V. Shytov, and B. I. Halperin, [41] H. L. Wang, C. H. Du, Y. Pu, R. Adur, P. C. Hammel, Phys. Rev. Lett. 93, 226602 (2004). [18] A. A. Burkov, A. S. N´unez, and A. H. MacDonald, Phys. Rev. B 70, 155308 (2004). [19] A. A. Burkov and D. G. Hawthorn, Phys. Rev. Lett. 105, 066802 (2010). [20] A. N. M. Zainuddin, S. Hong, L. Siddiqui, S. Srinivasan, and S. Datta, Phys. Rev. B 84, 165306 (2011). [21] S. Hong, V. Diep, S. Datta, and Y. P. Chen, Phys. Rev. B 86, 085131 (2012). [22] Y. Tserkovnyak and S. A. Bender, Phys. Rev. B 90, 014428 (2014). [23] 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). [24] T. Valet and A. Fert, Phys. Rev. B 48, 7099 (1993). [25] S. Sayed, S. Hong, and S. Datta, Sci. Rep. 6, 35658 (2016). and F. Y. Yang, Phys. Rev. Lett. 112, 197201 (2014). [42] H. J. Zhang, S. Yamamoto, Y. Fukaya, M. Maekawa, and H. Li, A. Kawasuso, T. Seki, E. Saitoh, K. Takanashi, Sci. Rep. 4, 4844 (2014). [43] M. Hoesch, M. Muntwiler, V. N. Petrov, M. Hengsberger, L. Patthey, M. Shi, M. Falub, T. Greber, and J. Oster- walder, Phys. Rev. B 69, 241401 (2004). [44] A. Tamai, W. Meevasana, P. D. C. King, C. W. Nichol- son, A. de la Torre, E. Rozbicki, and F. Baumberger, Phys. Rev. B 87, 075113 (2013). [45] K. Y. Camsari, S. Ganguly, S. Sayed, and S. Datta, "Modular approach to spintronics," (2013), https:// nanohub.org/groups/spintronics. [46] K. Y. Camsari, S. Ganguly, and S. Datta, Sci. Rep. 5, 10571 (2015). [47] S. Sayed, V. Q. Diep, K. Y. Camsari, and S. Datta, Sci. Rep. 6, 28868 (2016). [26] S. Hong, S. Sayed, and S. Datta, Sci. Rep. 6, 20325 [48] P. J. Burke, IEEE Transactions on Nanotechnology 1, (2016). 129 (2002). [27] V. T. Pham, L. Vila, G. Zahnd, A. Marty, W. Savero- Torres, M. Jamet, and J.-P. Attan, Nano Lett. 16, 6755 (2016). [28] V. T. Pham, G. Zahnd, A. Marty, W. Savero Torres, M. Jamet, P. Nol, L. Vila, and J. P. Attan, Appl. Phys. Lett. 109, 192401 (2016). [29] J.-H. Lee, H.-J. Kim, J. Chang, S. H. Han, H.-C. Koo, S. Sayed, S. Hong, and S. Datta, Scientific Reports (Ac- cepted) (2018). [30] C. H. Li, O. M. van 't Erve, J. T. Robinson, Y. Liu, L. Li, and J. B. T., Nature Nanotechnol. 9, 20325 (2014). [31] J. Tang, L.-T. Chang, X. Kou, K. Murata, E. S. Choi, M. Lang, Y. Fan, Y. Jiang, M. Montazeri, W. Jiang, Y. Wang, L. He, and K. L. Wang, Nano Letters 14, 5423 (2014). [49] P. J. Burke, IEEE Transactions on Nanotechnology 2, 55 (2003). [50] S. Salahuddin, M. Lundstrom, and S. Datta, IEEE Transactions on Electron Devices 52, 1734 (2005). [51] J. C. R. S´anchez, L. Vila, G. Desfonds, S. Gambarelli, J. P. Attan´e, J. M. De Teresa, C. Mag´en, and A. Fert, Nat. Commun. 4, 2944 (2016). [52] M. Isasa, M. C. Mart´ınez-Velarte, E. Villamor, C. Mag´en, L. Morell´on, J. M. De Teresa, M. R. Ibarra, G. Vig- nale, E. V. Chulkov, E. E. Krasovskii, L. E. Hueso, and F. Casanova, Phys. Rev. B 93, 014420 (2016). [53] H. Wang, J. Kally, J. S. Lee, T. Liu, H. Chang, D. R. Hickey, K. A. Mkhoyan, M. Wu, A. Richardella, and N. Samarth, Phys. Rev. Lett. 117, 076601 (2016). [54] A. De Martino, R. Egger, K. Hallberg, and C. A. Bal- [32] A. Dankert, J. Geurs, M. V. Kamalakar, S. Charpentier, seiro, Phys. Rev. Lett. 88, 206402 (2002). and S. P. Dash, Nano Letters 15, 7976 (2015). [55] A. Calzona, M. Carrega, G. Dolcetto, and M. Sassetti, [33] L. Liu, A. Richardella, I. Garate, Y. Zhu, N. Samarth, Phys. Rev. B 92, 195414 (2015). and C.-T. Chen, Phys. Rev. B 91, 235437 (2015). [56] T. Stauber, G. G´omez-Santos, and L. Brey, Phys. Rev. [34] J. Tian, I. Miotkowski, S. Hong, and Y. P. Chen, Sci. B 88, 205427 (2013). Rep. 5, 14293 (2015). [57] A. V. Moroz, K. V. Samokhin, and C. H. W. Barnes, [35] J. S. Lee, A. Richardella, D. R. Hickey, K. A. Mkhoyan, Phys. Rev. Lett. 84, 4164 (2000). and N. Samarth, Phys. Rev. B 92, 155312 (2015). [58] B. I. Halperin, Journal of Applied Physics 101, 081601 [36] F. Yang, S. Ghatak, A. A. Taskin, K. Segawa, Y. Ando, M. Shiraishi, Y. Kanai, K. Matsumoto, A. Rosch, and Y. Ando, Phys. Rev. B 94, 075304 (2016). (2007). [59] M. Polini and G. Vignale, Phys. Rev. Lett. 98, 266403 (2007). [37] P. Li and I. Appelbaum, Phys. Rev. B 93, 220404 (2016). [38] P. M. Haney, H.-W. Lee, K.-J. Lee, A. Manchon, and [60] A. Schroer, B. Braunecker, A. Levy Yeyati, P. Recher, Phys. Rev. Lett. 113, 266401 (2014). and M. D. Stiles, Phys. Rev. B 87, 174411 (2013). [39] J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, Rev. Mod. Phys. 87, 1213 (2015). [61] S. Datta, Lessons from Nanoelectronics: A New Perspec- tive on Transport (World Scientific Publishing Co. Pte. Ltd., Singapore, 2012). [40] A. Hoffmann, IEEE Transactions on Magnetics 49, 5172 [62] N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saunders, 1976). 

1108.5529

1

1108

Simulation of energy barrier distributions using real particle parameters and comparison with experimental obtained results

In this work we compare previously measured energy barriers over the course of temperature with the results of simulations of the behaviour of the energy barriers. For the measurements the temperature dependent magnetorelaxation method (TMRX) was used. For the simulations of the energy barrier distribution we have used the real particles properties such as anisotropy and core size volume of the fractions of two magnetically fractionated ferrofluids. There is a good agreement between the simulated behaviour and the experimental obtained results. The influence of the particle volume concentration and agglomeration on the energy barrier distribution has been investigated. Finally the simulations confirm a previously published explanation for an experimentally obtained relaxation effect.

cond-mat.mes-hall

cond-mat

Simulation of energy barrier distributions using real particle parameters and comparison with experimental obtained results M Büttner1, M Schiffler2, P Weber1 and P Seidel1 1 Institut für Festkörperphysik, Friedrich-Schiller-Universität Jena, Helmholtzweg 5, 07743 Jena, Germany 2 Institut für Geowissenschaften, Friedrich-Schiller-Universität Jena, Burgweg 11, 07749 Jena, Germany Short Title: Simulation of energy barrier distributions Corresponding author: Markus Büttner, Friedrich-Schiller-Universität Jena, Institut für Festkörperphysik, Helmholtzweg 5, 07743 Jena, Germany, Email: Markus.Buettner@uni-jena.de Tel.: ++49-3641-947387 Fax: ++49-3641-947412 Abstract In this work we compare previously measured energy barriers over the course of temperature with the results of simulations of the behaviour of the energy barriers. For the measurements the temperature dependent magnetorelaxation method (TMRX) was used. For the simulations of the energy barrier distribution we have used the real particles properties such as anisotropy and core size volume of the fractions of two magnetically fractionated ferrofluids. There is a good agreement between the simulated behaviour and the experimental obtained results. The influence of the particle volume concentration and agglomeration on the energy barrier distribution has been investigated. Finally the simulations confirm a previously published explanation for an experimentally obtained relaxation effect. Keywords Temperature dependent magnetorelaxation measurements, Néel relaxation, Magnetic nanoparticles, Energy barrier distribution, Simulation, Reduced anisotropy, Agglomeration, Interaction 1. Introduction The examination of the magnetic properties of many-particle systems of magnetic nanoparticles (MNP) is still a field of intensive research [1, 2]. One special method for the characterisation of MNP is based on the analysis of the temperature-dependent Néel-relaxation signal (TMRX). The theoretical mediations in this field of research are done by R.W. Chantrell et al. [3] and D. V. Berkov et al. [4-6]. Based on this theoretical work a system for the measurement of the temperature dependence of the Néel-relaxation signal were developed. In a first step measurements between room temperature and liquid nitrogen temperature were realised [7] and in a second step the temperature range were expanded down to liquid helium temperature [8]. With this measurement set-up a lot of interesting measurements were done [9, 10]. The missing connective link was the comparison between the results of a concrete simulation of the energy barriers using the real particle parameters and the experimental obtained results. In this work we compare previously published experimentally measured energy barrier distributions over the course of temperature [8, 10] with the calculated energy barrier distributions. For this comparison we use two varying ferrofluids that were magnetically fractionated to obtain smaller distribution of energy barriers and magnetically active core sizes, respectively. 2. Experimental details A detailed description of the measurement system and the preparation procedure has been published elsewhere [8]. The first ferrofluid we use is a water-based ferrofluid (DDM 128N, Meito Sangyo, Japan) consisting of MNP with a core of iron oxide and a shell of carboxydextran. The second one also is a water-based ferrofluid (V190 [11]) consisting of MNP with a core of iron oxide but with a shell of carboxymethyldextran. The original ferrofluid with a broad particle size distribution was fractionated in an inhomogeneous magnetic field using an electromagnet with variable magnetic field strength (Bruker B-E 10v) and a magnetic separation column (MACS XS, Miltenyi Biotech) generating a strong field gradient. The particles are retained in the column in dependence on this field gradient as well as the particle magnetic moment. In what follows, the samples will be named with the used fractionation coil current. To obtain more reliable data we only use the fractionated samples but not the inititial solutions for our investigations. The magnetic data that is necessary for the simulation and the experimental obtained energy barrier distributions are taken from [10] for sample V190 und from [8] for sample DDM 128N. 3. Theoretical details 3.1. Derivation of the important energy terms Our aim is to simulate the full shape of the energy barrier distribution found in TMRX measurements. The energy barrier distribution is the probability distribution to find an energy barrier to be overcome at one transition from an initial metastable state of magnetic nanoparticles to a final one. The energy is barrier therefore =∆ EE (1) E x ) − opt , A (max (max E where optx ) is the maximum of the optimal path through the energy landscape which minimises the action as function of the magnetisation directions of the MNP and metastable state. According to [5, 12] we now have to find this optimal path via numerical minimization of the action. S∇ ≡ is fulfilled, this optimal path should lead from the first metastable 0 If the minimum condition state over the saddle point to the second metastable state. There are two possible methods to find the minimum: The first one is the derivation the N Euler-Lagrange-equations from the action and solving the corresponding boundary value problem. For systems with e.g. N=128 MNP the inversion of the resulting equation system is time-consuming. For faster calculation the evaluation of the integral is better performed by numerical computation of the discretised variant of the action AE the initial energy of one S disc ( Ω = ∆ ) t − 1 N K ∑ ∑ = = 0 1 i k  θ , i k   − + 1 ∆ t θ , i k + 1 2     ∂ Ω { E θ ∂ , i k } k + 1 + + 1 } k ∂ Ω { E θ ∂ , i k 2         2 (2) fixed representing the two (ΩE ) includes the units (3) + + sin θ , i k      2 sin   for the set Ω of all magnetisation orientation angles θ φ , , i k i k + 1 ∆ t φ , i k sin − + 1 2 + 1 1 θ , i k + 1 ∂ Ω { E ∂ φ , i k k + 1 } +        . Via minimisation of this functional of ∂ Ω { E ∂ φ , i k 1 θ , i k sin } + 1 k all particle angles for the time slices ki φθ , ( , , ki Kk = 0=k = − ,1 k … with metastable states we obtain the optimal trajectory. Thereby the system energy anisotropy energy in reduced β − ∑ (cid:2) ⋅ (cid:2) ( m n i i 2 i ε = an and K 1 ) ) , 2 where ß is proportional to the anisotropy constant and in(cid:2) is achieved via dividing the magnetisation by its absolute is the unit vector of the easy axis of each j j ) 3 + 3( 3 r ij π 2 = − ε dip ⋅ (cid:2) (cid:2) m m i 〈 〉 η (cid:2) m m i particle. The magnetisation unit vector im(cid:2) value. Below the Curie temperature it can be identified with the saturation magnetisation of the field The stray system. energy is calculated with   ⋅ ⋅ − (cid:2) (cid:2) (cid:2) (cid:2) )( e m e m ∑ ∑ ij ij i       < r r i ij rest with the distance units rescaled by R=r/a where a denotes the average particle radius. To reduce calculation time significantly the Lorentz cavity method is used. The direct calculation of the stray field via the dipolar dependence acting on the i-th particle only is done inside a sphere around it with a restr twice the average interparticle distance. The remnant interactions are considered restriction radius by the system magnetisation. The average system magnetisation is denoted by m〈 〉 in equation 4. 3.2. Necessary simulation parameters Summarising the initialisation procedure for the simulation following parameters are needed for the run: (4) 2 K The reduced anisotropy constant β = is required for calculation of equation 3 and can be 2 µ M 0 S obtained via magnetometric measurements. The anisotropy constant K is the product of the coercivity C MHK = and the saturation magnetisation: . S The particle concentration has to be calculated from the sample geometry and the individual average volume of the magnetic nanoparticles. If we denote the iron concentration with MNPn , the molar mass MNPM and the density MNPρ of the material the magnetic nanoparticles consist in and the sample V we obtain: volume with sample η . = MNP n M MNP ρ V MNP sample The third important parameter is the number of runs of the minimisation procedure. Because the transition from one metastable state to another can be reverted we obtain two energy barriers in one run. With as many runs as possible we achieve statistical behavior of the system and therefore a more realistic energy barrier distribution. The latter expenses on calculation time: for getting 256 energy barriers we need one and a half weeks to perform full computation. Because of the complexity of the simulation program we are prevented to use powerful parallel computing techniques. Nevertheless multiprocessing can be done by running the program in several processes separately. By the way this workaround improves the stability of the whole simulation. (5) , 2 3.3. Simulation procedure To find a certain energy barrier the simulation program has to do the following: Firstly we need the initial configuration of the system of the MNP. It consists in the initial and final set of magnetisation directions describing a system of N single-domain particles. In our model we assume uniaxial single-particle anisotropy. To calculate the stray field between these particles we have to consider the spatial distribution of the MNP. The initialisation of the positions of these is done in an external script: The spatial matrix with 3N values representing N particles with three dimensions is obtained by using a random number generator and scaling by 1/ 3   N   η   whereηdenotes the particle concentration. Factor 2 takes the scale of the spatial information to the average particle radius into account. The more particles are in the system is or the less the concentration is, so further the particles are divided. The spatial matrix also can be edited manually for achieving the possibility to let the simulation be influenced by the knowledge about the particles as much as possible. To obtain the initial configuration in magnetisation the search for the two metastable states has to be proceeded. There we have to find two energy minima via minimisation similar to the action minimiser. The main difference between the minimisation of the system consists in having only the first stage and calculating the values and gradients only of the energy. The break condition here is fulfilled, if the energy has undergone a target precision. The action minimisation consists in three stages. In the first step several relaxation steps are made. pΩ into the to . According This means the movement of the system coordinates from the initial configuration 1+Ω p p S Ω∇ ) ( the new position local antigradient the to (6) direction of −Ω=Ω + 1 p p α rel ∇ S , and this shift is scaled by the step length (7) relα . This parameter is chosen arbitrarily, so that the action decreases. If after some step the action value increases we return to the previous configuration and rel αα = 2/ . After ca. 10 relaxation steps the action should decrease after each step and therefore let rel we can turn towards step two. In this stage the full function minimisation along the local antigradient optα is performed. With equation 7 we move the system coordinates with a step length of an optimised into the direction of the function minimum. In the third stage the gradient of the action has to be relα at the new minimised. This step is to be used for determination of the relaxation step length relaxation steps. Afterwards the program returns to step one and the relaxation steps are restarted. The break condition of the minimiser loop is achieved if for the components of the action gradient p p ∂Ω∂ ∀< ∂Ω∂ ∀< δφ) δθ) is fulfilled. The target precisionδ is an arbitrary ( ( S i S i i i 01.0=δ constant, e.g. leads to good results. The described method for action minimisation seeks only for the shortest path over minimal energy values. Therefore it could happen that energy maxima are crossed by the “optimal” trajectory. To prevent this, a check procedure is enforced. So we move the system coordinates in the point with the maximum energy value into an arbitrary direction. The step length is chosen so that the new maximum gradient component at the new site is around ten times larger than the old one. The arbitrary moves of all other points are scaled with the average coordinate distance from one time step to another. With this configuration the minimisation procedure is resumed. If the second minimisation results in an energy value close to the first one we can surely assume that we have found an optimal trajectory leading over the saddle points of the energy landscape and the output is performed. The result of the simulation program is a set of found energy barriers, the energy course in time, the corresponding action and the coordinates of the optimal trajectory. To obtain the energy barrier distribution we have to perform a simple last step manually. The distribution is nothing other than the histogram of the given energy barriers. To calculate the energy barrier distribution over temperature we have to rescale the system from the reduced values to the normal ones. According to ε = E 2 VM S , (8) (9) 0µ there is to consider the scaling with the average particle volume V . We can identify it with the maximum volume of the energy barrier density curves obtained from the TMRX measurements. The next step is to scale the information to one individual particle. Initially we observe the energy summarised for all N nanoparticles. Because the anisotropy energy determines the average height of the energy barrier this can be obtained by dividing the system energy by the expectation value of n n ∑ ∑ i i where iψ denotes the angle between the easy axis and the magnetisation direction of each particle. The 2N . So the conversion factor of the energy between the reduced questioned value is nothing else than and units by described the is units measurement the 2 µ VM 0 S 2 N ⋅ (cid:2) (cid:2) m n i i (10) ε . sin 2 ψ i E = 2 ) = , ( If we would like to get information about the temperature course in order to compare it with our Bk20 where the factor 20 TMRX measurements we have to divide the energy axis of the histogram by is derived from the equation for the blocking volume [13] and Bk is the Boltzmann constant. 4. Simulation results and discussion The simulation program and the scaling described above were firstly run for two ferrofluids samples measured previously by TMRX. There are four interesting fractions of each sample prepared which are provided with information about the anisotropy constant obtained from magnetometric measurements. For the sample DDM 128N we obtain a particle volume concentration of η = 0.000724 using equation (6). Due to the mixture ratio between maghemite and magnetite in the η = 0.000814 ferrofluids system V190 this value is . For each fraction 256 simulation runs are performed for finding a pair of energy barriers. The contribution of the anisotropy energy to the energy barrier is narrowly distributed around a center where the distribution of the concentration dependent stray field energy is broader. The simulations with a fit of the lognormal distribution to the histogram data are shown in figure 1. Table 1 shows the simulation fit parameters for the investigated fractions of sample V190 and table 2 for the fractions of DDM 128N. The values for the anisotropy constant K, the volume of the particles Vmax and the experimental obtained temperature maximum of the energy barrier distribution TmaxExp were taken from [8] for sample DDM 128N and from [10] for sample V190. The fit was done supposing a lognormal distribution of the particles. The factor ß in both tables denotes the calculated reduced anisotropy, Tm the median and σ the standard deviation of the log-normal distribution. We firstly observe that the simulations reflect the lognormal distribution of the MNP found in real experiments. The differences between the peaks determined by simulation and fit on the one hand and the temperature detection deviation of the TMRX measurements on the other are caused by model-like character of the simulations and the error tolerance of the lognormal fit. To improve the amount of data two other simulations with the same systems of magnetic nanoparticles are performed. The first one should demonstrate the influence of the particle volume concentration on the shape of the energy barrier distribution (see fig. 2). This step is done for the 400 mA fraction of sample V190 and for the 0 A fraction of sample DDM 128N. It could be clearly shown that for very low concentrations =η 0.0001 the energy barrier closely is distributed around the expectation value of the anisotropy =η 0.001 energy. Switching to higher concentrations ( ) for the sample DDM 128N we observe a broader distribution around anε . The broadening is less for the sample V190. A noticeable broadening 01.0=η . For sample DDM 128N we observe a slightly shift of the maximum for this sample starts at 01.0=η of the energy barrier distribution towards lower temperatures. For the sample V190 and =η 0.05 is not applicable. For the highest simulated energy barrier distributions ( ) we observe in this scaled by x lx That means, for a spatial part of for the agglomerated particles their spatial information is accordance to [4] a shift to lower temperatures and a broadening in the energy barrier distribution. Besides this, we also obtain energy barriers corresponding to higher temperatures. Due to the higher concentration the interparticle distance decreases. These increasing values of the stray field energy lead to higher energy values. After that a check with a system of partially agglomerated particles is done (see fig. 3). The agglomeration is done by compressing the spatial extension in all three dimensions for one fourth of the nanoparticles to the half of its value around the center of the system. 3 8=l 1 . The investigations are performed for the same fractions described above. → − + 1 ) 2 (1 l The results are shown in fig. 3. Therefore the energy barrier distribution broadens in comparison to the non-agglomerated investigations. For the sample V190 we only observe a slight shift to lower temperatures. For the fractions 250 mA and 0 mA of the sample DDM 128N the broadening is more significant. This could be lead back to the lower anisotropy constant. Figure 4 shows the comparison of the simulated energy barrier distribution over the course of temperature (bars) and the experimental obtained distribution (curves) for the fractions of sample V190 and DDM 128N. The experimental obtained energy barrier distributions are taken from [10] for sample V190 und from [8] for sample DDM 128N. As we see we are able to reconstruct the complete energy barrier distribution by only using average values of the (reduced) anisotropy constant and the average particle volume. The simulated energy barrier distributions over the course of temperature are in agreement with the measured ones. The simulations also show that the experimental obtained peak at 18 K for all fractions of the V190 sample is not a result of strongly interacting particles. Relaxation signals in an equal temperature range were also measured on magnetite nanoparticles produced by magnetotactic bacteria (so called magnetosomes) [9]. The origin of that signals was explained as memory effects [14]. This explanation can be confirmed by using the results of the simulations shown in figures 2 and 3. Conclusions The results of the simulations of the energy barrier distributions under use of the real particle parameters are in good agreement with the experimental obtained results. For very low particle concentrations the energy barrier closely is distributed around the expectation value of the anisotropy energy. In case of agglomeration the energy barrier distribution broadens in comparison to the non- agglomerated particle systems. The simulations reflect the lognormal distribution of the MNP found in real experiments and confirm a previously published explanation for an experimentally obtained relaxation effect which is not a result of the Néel relaxation process and therefore not an effect of the energy barrier. Acknowledgements The authors would like to thank D. V. Berkov for the fruitful discussions and the helpful suggestions. References [1] V. Schaller, G. Wahnstrom, A. Sanz-Velasco, P. Enoksson, C. Johansson, Monte Carlo simulation of magnetic multi-core nanoparticles, Journal of Magnetism and Magnetic Materials, 321 (2009) 1400-1403. [2] R. Muller, S. Dutz, T. Habisreuther, M. Zeisberger, Investigations on magnetic particles prepared by cyclic growth, Journal of Magnetism and Magnetic Materials, 323 (2011) 1223- 1227. [3] R.W. Chantrell, E.P. Wohlfarth, Dynamic and Static Properties of Interacting Fine Ferromagnetic Particles, Journal of Magnetism and Magnetic Materials, 40 (1983) 1-11. [4] D.V. Berkov, Calculation of the energy barriers in strongly interacting many-particle systems, Journal of Applied Physics, 83 (1998) 7390-7392. [5] D.V. Berkov, Evaluation of the energy barrier distribution in many-particle systems using the path integral approach, Journal of Physics-Condensed Matter, 10 (1998) L89-L95. [6] D.V. Berkov, R. Kotitz, Irreversible relaxation behaviour of a general class of magnetic systems, Journal of Physics-Condensed Matter, 8 (1996) 1257-1266. [7] E. Romanus, D.V. Berkov, S. Prass, C. Gross, W. Weitschies, P. Weber, Determination of energy barrier distributions of magnetic nanoparticles by temperature dependent magnetorelaxometry, Nanotechnology, 14 (2003) 1251-1254. [8] E. Romanus, T. Koettig, G. Glockl, S. Prass, F. Schmidl, J. Heinrich, M. Gopinadhan, D.V. Berkov, C.A. Helm, W. Weitschies, P. Weber, P. Seidel, Energy barrier distributions of maghemite nanoparticles, Nanotechnology, 18 (2007) 115709. [9] M. Buttner, P. Weber, C. Lang, M. Roder, D. Schuler, P. Gornert, P. Seidel, Examination of magnetite nanoparticles utilising the temperature dependent magnetorelaxometry, Journal of Magnetism and Magnetic Materials, 323 (2011) 1179-1184. [10] M. Buttner, P. Weber, F. Schmidl, P. Seidel, M. Roder, M. Schnabelrauch, K. Wagner, P. Gornert, G. Glockl, W. Weitschies, Investigation of magnetic active core sizes and hydrodynamic diameters of a magnetically fractionated ferrofluid, Journal of Nanoparticle Research, 13 (2011) 165-173. [11] K. Wagner, A. Kautz, M. Roder, M. Schwalbe, K. Pachmann, J.H. Clement, M. Schnabelrauch, Synthesis of oligonucleotide-functionalized magnetic nanoparticles and study on their in vitro cell uptake, Applied Organometallic Chemistry, 18 (2004) 514-519. [12] D.V. Berkov, Numerical calculation of the energy barrier distribution in disordered many-particle systems: the path integral method, Journal of Magnetism and Magnetic Materials, 186 (1998) 199-213. [13] K.H.J. Buschow, F.R.d. Boer, Physics of magnetism and magnetic materials, Kluwer Academic/Plenum Publishers, New York, 2003. [14] M. Sasaki, P.E. Jonsson, H. Takayama, H. Mamiya, Aging and memory effects in superparamagnets and superspin glasses, Physical Review B, 71 (2005) 104405. Table 1. Parameters for the fractions of sample V190 used in simulation and for the fit. fraction [mA] 1000 400 200 100 K [kJ/m³] 13.7 12 13 12.3 β σ Vmax Tmax Sim [10-25m³] Tm [K] [K] 104.24 0.397 89.10 0.18551 16.9 167.95 0.437 138.72 0.16303 28.2 0.17632 39.2 222.23 0.443 182.71 272.28 0.410 230.06 0.16713 46.0 Tmax Exp [K] 92 150 190 225 difference -3.2% -7.5% -3.8% 2.2% Table 2. Parameters for the fractions of sample DDM 128N used in simulation and for the fit. fraction [mA] 6000 1000 250 0 K [kJ/m³] 22.6 19.1 11.4 11.8 β σ Vmax Tmax Sim [10-25m³] Tm [K] [K] 0.242 7.15 0.31120 0.61 7.58 16.83 0.287 15.51 0.23101 1.99 181.39 0.463 146.44 0.15482 35.4 0.16225 56.4 322.71 0.414 271.93 Tmax Exp [K] 5 13 137 210 difference 43.0% 19.3% 6.9% 29.5% Figure captions Figure 1. Simulation with a fit of the lognormal distribution (curves) to the histogram data (bars) for the fractions of V190 (left) and DDM 128N (right). Figure 2. Influence of the MNP concentration on the simulation result for the 400 mA fraction of sample V190 (left) and for the 0A fraction of sample DDM 128N (right). At low concentration there is only anisotropy energy. Figure 3. Agglomeration of 25 percent of the particles for the fractions of sample V190 (left) and DDM 128N (right). Figure 4. Comparison of the simulated energy barrier distribution over the course of temperature (bars) and the experimental obtained distribution (curves) for the fractions of sample V190 (left) and DDM 128N (right). Figure 1. Simulation with a fit of the lognormal distribution (curves) to the histogram data (bars) for the fractions of V190 (left) and DDM 128N (right) V190 100 0 200 300 0 DDM 128 N 100 200 1.2 1.0 0.8 0.6 0.4 0.2 1.20.0 1.0 0.8 0.6 0.4 0.2 1.20.0 1.0 0.8 0.6 0.4 0.2 1.50.0 1.0 0.5 0.0 0 1000 mA 400 mA 200 mA 100 mA 100 200 300 Temperature T [K] ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 1.0 0.8 0.6 0.4 0.2 1.20.0 1.0 0.8 0.6 0.4 0.2 1.20.0 1.0 0.8 0.6 0.4 0.2 1.20.0 1.0 0.8 0.6 0.4 0.2 0.0 300 6000 mA 1000 mA 250 mA 0 mA 0 100 200 300 Temperature T [K] ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ Figure 2. Influence of the MNP concentration on the simulation result for the 400 mA fraction of sample V190 (left) and for the 0A fraction of sample DDM 128N (right). At low concentration there is only anisotropy energy. V190 2 Energy ε [a.u.] 4 6 8 0 DDM 128 N Energy ε [a.u.] 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 η = 0.05 η = 0.01 η = 0.001 η = 0.0001 50 100 150 200 250 Temperature T [K] ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 100 50 0 15 10 5 80 60 40 20 0 60 40 20 0 0 η = 0.05 η = 0.01 η = 0.001 η = 0.0001 100 200 300 Temperature T [K] ] . u . a [ ) ε ( ρ 100 80 60 40 20 80 ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 60 40 20 80 60 40 20 0 80 60 40 20 0 0 Figure 3. Agglomeration of 25 percent of the particles for the fractions of sample V190 (left) and DDM 128N (right). V190 100 200 300 400 500 600 700 800 0 50 DDM 128 N 100 150 Temperature T [K] Temperature T [K] 0 50 100 150 200 250 1000 mA 400 mA 200 mA 100 mA 100 200 300 400 500 600 700 800 ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 100 80 60 40 20 0 150 100 50 0 80 60 40 20 0 100 80 60 40 20 0 200 250 6000 mA 1000 mA 250 mA 0 mA ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 0 150 100 50 0 80 60 40 20 0 150 100 50 0 150 100 50 0 0  Figure 4. Comparison of the simulated energy barrier distribution over the course of temperature (bars) and the experimental obtained distribution (curves) for the fractions of sample V190 (left) and DDM 128N (right). V190 100 0 200 300 0 DDM 128 N 100 200 Temperature T [K] Temperature T [K] 0 100 200 300 ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 1.2 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 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1000 mA 400 mA 200 mA 100 mA 0 100 200 300 ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ ] . u . a [ ) ε ( ρ 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 0 mA 300 6000 mA 1000 mA 250 mA 

1508.02583

1

1508

Analysis of radiation effect on the threshold voltage of flash memory device

Flash memory experiences adverse effects due to radiation. These effects can be raised in terms of doping, feature size, supply voltages, layout, shielding. The the operating point shift of the device forced to enter the logically-undefined region and cause upset and data errors under radiation exposure. In this letter, the threshold voltage shift of the floating gate transistor (FGT) is analyzed by a mathematical model.

cond-mat.mes-hall

cond-mat

For Reference: Nahid M. Hossain, Jitendra Koppu, Masud H Chowdhury, “Analysis of radiation effect on the threshold voltage of flash memory device”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2896-2899, Lisbon, Portugal, 2015. Analysis of Radiation Effect on the Threshold Voltage of Flash Memory Device Nahid M. Hossain1, Jitendra Koppu1, Masud H Chowdhury1 1Computer Science and Electrical Engineering, University of Missouri – Kansas City, Kansas City, MO 64110, USA Email: mnhtyd@mail.umkc.edu, jkvxf@mail.umkc.edu and masud@ieee.org Abstract- Flash memory experiences adverse effects due to radiation. These effects can be raised in terms of doping, feature size, supply voltages, layout, shielding. The the operating point shift of the device forced to enter the logically-undefined region and cause upset and data errors under radiation exposure. In this letter, the threshold voltage shift of the floating gate transistor (FGT) is analyzed by a mathematical model. I. INTRODUCTION As the device size is shrinking down to nanometer range, the impacts of radiations on circuit and device performance and reliability would become more prominent. In the recent time study of radiation hardness of micro & nano-electronic devices for extreme conditions are gaining wide spread attention. The radiation effect on the floating gate transistor (FGT) used in flash memory leads to charge loss from the programmed floating gate (FG). Due to the imposture to certain type of radiation, an extra electron/hole pair can be generated in the device. For example, a minimum of 10-Kev X-rays exposure would initiate this process. In the FGT, the radiation effect can be neglected if the oxide thickness is 40-47.5 nm [1]. But it is no longer negligible because the oxide thickness is going down to 10 nm in scaled FGTs. Radiation induced charge in the oxide depends on several physical mechanisms i.e. electron-hole pair generation and election-hole recombination. The electron-hole recombination depends on the applied electrical field and linear energy transfer. Even the effect changes over time, i.e. the change of threshold voltage (∆VTH) varies over the 10-6~108 Sec time scale. When VG>0V, holes are trapped into the oxide due to the radiation effect. These trapped holes creates conduction, which leads to “ON” state even when VGS=0V [3]. Figure 1a shows the schematics of a MOSFET floating gate transistor (FGT). The only difference with the standard MOSFET is the addition of a new gate, called the floating gate, between the original gate and the channel. The original gate (topmost) is now called the control gate. A floating gate is basically a polysilicon gate surrounded by insulator and it has no electrical connection with other layers. The working principle of a FGT is almost same as the conventional MOSFET, where the source-drain current is monitored and controlled by the control gate voltage. The floating gate voltage or in other words the stored charge on the floating gate can control the channel between the drain and the source. Thinner tunnel oxide is required to facilitate tunneling between the channel and the floating gate. To program or write a FGT (Figure 1b), a positive gate voltage is applied. This positive voltage attracted electrons from the channel through the tunnel oxide insulating layer into the floating gate. Charge accumulated in the floating gate is protected by the insulating layer. So, the stored data is retained for years. To erase the data, a negative voltage is applied at the control gate (Figure 1c). This negative voltage pushes the electrons out of the floating gate [7]. (a) (b) (c) For Reference: Nahid M. Hossain, Jitendra Koppu, Masud H Chowdhury, “Analysis of radiation effect on the threshold voltage of flash memory device”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2896-2899, Lisbon, Portugal, 2015. Figure 1: (a) A floating gate transistor, (b) programming of floating gate transistor, and (c) erasing of floating gate transistor [7]. In this paper, we present how threshold voltage of the FGT is affected by the radiation. To the best of our knowledge, this is the first work to present the analysis of radiation hardness of a flash memory device. An analytical model has been developed, which shows direct relation between radiation level and threshold voltage (Vth). This equation directly shows how dynamic characteristics are changed due to the radiation exposure. The rest of the paper is organized as follows. Section II presents the effects of radiation on the FGT. Section III provides the results and analysis. Finally, Section IV concludes the paper with a brief introduction to future work. II. A. Mechanism of Radiation Effect EFFECTS OF RADIATION ON FGT The changes in characteristics of a FGT due to radiation can be explained by four major steps as shown in Figure 2. The FGT is a modified MOS structure. Therefore, according to device point of view, the radiation effects on the FGT can be explained by the simplified MOS structure. Step-1: Schematic energy band diagram for MOS structure is illustrated to understand each step of VTH variation. When a positive voltage is applied to the control gate, the electrons flow towards the floating gate and holes to the substrate. Due to irradiation, electron hole pair is generated in SiO2, which is considered as the most sensitive part in the MOS structure. As electrons are more mobile than hole, they swept out in picoseconds or less. In the first picosecond, recombination of electrons and holes takes place; and holes that escape from recombination are relatively immobile and remain near the point of generation, and these holes cause the negative threshold shift in the MOS transistor. This initial stage leads to the maximum drift in the VTH [3]. Figure 2: The change of band energy inside the FGT under irradiation [3]. Step-2: Holes tends to shift towards Si/SiO2 interface that causes short-term recovery in the VTH, which depends on mainly applied electrical field, temperature and oxide thickness. Generally, it takes about 1sec at room temperature but it may need extended time at low temperature [3]. Step-3: Holes reach to silicon interface, fraction of holes transported to deep long lived trap states. These trapped holes cause the threshold voltage to make a negative shift. This effect continues for hours to years. Gradual annealing can recover the memory from this damage [3]. Step-4: The radiation induced traps at the Si/SiO2 interface are determined by the Fermi level. Generally interface traps are highly dependent on oxide processing [3]. Thus, it can be summarized that radiation induces charge in the oxide, which is dependent on several physical mechanisms like electron-hole pair generation and election-hole recombination. The electron-hole recombination depends on the applied electrical field and linear energy transfer. B. VTH Variation Figure 3 shows the distributions of the threshold voltage (VTH) for the memory device in both programmed (“0”) and erased states (“1”). Here the impacts are shown before and after the exposure to radiation. The VTH of the FGT in the programming (“0”) state is high because a large amount of electrons are stored in the floating gate. As a consequence, higher control gate voltage is needed to form the channel in the “0” state. In the programming (“0”) state, an electron-hole pair needs 17eV energy around the floating gate (FG) and oxide region by photoemission, which is defined by a process where electrons are emitted from solids under For Reference: Nahid M. Hossain, Jitendra Koppu, Masud H Chowdhury, “Analysis of radiation effect on the threshold voltage of flash memory device”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2896-2899, Lisbon, Portugal, 2015. irradiation with photons of sufficiently low wavelength and high energy. Under irradiation, the threshold voltage of programming (“0”) state reduces uniformly. Therefore, a comparatively lower control gate voltage creates the conducting channel when the FGT is affected by radiation. On the other hand, the reduction of due to radiation VTH in the erased state (“1”) is less prominent. Portion of the generated electron-hole pairs are recombined, which depends on the electric field around the oxide. Higher electric field leads to lower recombination as more electrons can escape from the recombination. The photoemission is responsible for injection of holes, which escape from the recombination into the FG. These positive holes are recombined partially with the negative electron in the FG. The electrons, which are stored in the FG, get enough energy to jump over the oxide layer barrier when exposed to radiation and by photoemission [1]. Figure 3: Probability distribution of the threshold voltage for the FGT device for both programming states before and after irradiation [1]. Figure 4 : Threshold voltage data as a function of radiation dose [2] . To observe the VTH variation due to radiation effect, in [2] a memory cell is exposed to cobalt-60 radiation with a dose rate greater than 100rad.s-1. The VTH variations are observed over 100krad. The VTH in “0” state is seen to go down significantly even to negative values (see Figure 4). While for the “1” state, the VTH increases slightly. Figure 4 shows that the VTH in programming state (“0”) goes down whereas the VTH in the erasing (“1”) state increases slightly due to the radiation effect. Therefore, under high radiation doses the logic “0” can be read as logic “1” incorrectly [2]. C. Time Dependent Effect Even the radiation effect changes over time, i.e. the change of threshold voltage (∆VTH) varies over the 10-6~108 Sec time scale. Figure 5 shows that ∆VTH is not fixed after radiation exposure. When VGS>0V, holes are trapped into the oxide due to the radiation effect. These trapped holes shifts the operation of the FGT “OFF” to “ON” state even when VGS=0V [3]. Therefore, it gives a wrong reading. For Reference: Nahid M. Hossain, Jitendra Koppu, Masud H Chowdhury, “Analysis of radiation effect on the threshold voltage of flash memory device”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2896-2899, Lisbon, Portugal, 2015. Figure 5: Time dependent post irradiation threshold voltage change of the FGT [3]. III. RESULT AND ANALYSIS The change of VTH depends on the charge loss of the FG, which is caused by the photoemission and electron/hole pair generation in the tunnel oxide and control oxide. The change of VTH can be expressed by (1). ∆VTH= ∆Q CFG = ∆QTO+∆QCO+∆QPH CFG (1) Here, ∆VTH is the change of the threshold voltage, ∆Q is the total charge loss, CFG is the capacitance between the floating gate and the control gate, ∆QTO and ∆QCO are the charge losses in the tunnel oxide and control oxide respectively, and ∆QPH is the charge loss due to photo emission. According to the conventional FGT geometry (Figure 1a), the horizontal area of the FG is parallel to the substrate and the lateral area is perpendicular to the substrate. In the existing semiconductor industry, horizontal and vertical FG areas are equal [1]. Radiation causes both electron and hole generation in the surrounding oxides. Therefore, ∆QTO and ∆QCO linearly depend on both AFGH and AFGV. ∆VTH depends on charge density per area, rather than on the absolute number of stored electrons [9]. In principle, photoemission can happen wherever the electric field is nonzero, i.e., it can depend on both the planar and lateral dimensions of FG. Equation (1) can be rewritten as in (2). ∆VTH= (∆QTO).(AFGH)+(∆QCO).(AFGH)+(∆QPH) CFG (2) Here, AFGH is the horizontal area and AFGV is the vertical area of the floating gate. Here, CFG is the function of thickness and process of the control oxide. It should be noted that ∆VTH equation of FGT is completely different from charge trap memory, which explained in [1], because of the structural and charge trapping mechanism differences. Figure 6: ∆VTH variation as a function of the floating gate area. For Reference: Nahid M. Hossain, Jitendra Koppu, Masud H Chowdhury, “Analysis of radiation effect on the threshold voltage of flash memory device”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2896-2899, Lisbon, Portugal, 2015. Figure 6 shows ∆VTH variation with respect to floating gate area for a fixed radiation exposure and oxide thicknesses. It is observed that ∆VTH is inversely proportional to the floating gate area. 20 nm thick SiO2 control oxide is considered for the computation. The floating gate area, AFG is varied from 0.007~0.01µm2. The above-mentioned values are industry standard, which leads to better ∆VTH estimation. For convenience, the radiation exposure is assumed fixed and the fringing effects are neglected. Figure 7: ∆VTH shift as a function of dielectric constant. Figure 7 illustrates how threshold voltage changes for different oxide materials. It clearly shows that high-K oxides exhibits low VTH shift which leads high radiation hardness. Therefore, high-k oxide is recommended as the control oxide to make the circuit radiation hard. A comparative study is done for the most popular oxides [6] (SiO2, Si3N4, Al2O3 and HfO2) in the current semiconductor industry. This study suggests that if high k-dielectric oxide is used as the control oxide, the VTH variation tends to be less and at a certain higher value of dielectric constant (k) the variation tends become zero, which leads to better radiation hardness. According to the analysis, HfO2 is the best control oxide choice for flash memory when radiation hardness is the major concern. Many researches have shown that how Vth changes after the device is kept under radiation. For CAD tool and IC designer community it is required to translate the radiation effect quantitatively. Keeping that in mind, we have considered a black box where a FGT/MOSFET is kept as shown in Figure 8. The role of the model is to compute VTH values for increasing radiation levels for given device which is already fabricated or designed i.e. other parameters will not change. We are stable to the condition because the experimental results which are available followed the approach. Figure 8: Black box of a flash memory under radiation exposure. The variation of the VTH as a function of radiation data are collected from the experimental results [2],[4]. Then the data is statistically analyzed. These steps and data are not provided in the paper because of space limitation. The statistical analysis is concluded by the result shown Figure 9, where x-axis presents TID (dose of radiation in the Krad(SiO2) unit) and y-axis represents VTH. For Reference: Nahid M. Hossain, Jitendra Koppu, Masud H Chowdhury, “Analysis of radiation effect on the threshold voltage of flash memory device”, IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2896-2899, Lisbon, Portugal, 2015. Figure 9: Threshold shift as a function of total ionization dose (TID). Figure 9 shows that VTH decreases as a function of the radiation level. As the radiation exposure increases, VTH tends to fall rapidly. The black dots are the experimental value collected from [4], while the continuous red curve represents the simulated result. It should be noted that TID(Krad(SiO2)) is a well-defined universal radiation measurement unit, which is very popular in experimental and commercial radiation measurement. In order to validate the model, the result is verified with experimental research works. The simulated result of the VTH variation with radiation shows good agreement with the experimental data of [2],[4]. IV. CONCLUSION AND FUTURE WORK A mathematical model of FGT is proposed where the threshold voltage (VTH) is considered as the key parameter. The VTH of the FGT drops when radiation exposure rises. From our analysis, we observed that the variation of VTH in the FGT is (i) inversely proportional to the floating gate area, (ii) directly proportional to the control oxide thickness, and (iii) drops exponentially at the higher value of dielectric constant. Therefore, the mathematical model will be useful to analyze the radiation hardness of flash memory design and allow trade-off between important parameters. Our future work involves the radiation hardness test at every single design step of a device which will allow designers more flexibility in the radiation hardened memory design in future. REFERENCES Cellere et al., “Ionizing radiation effects on floating gates”, Appl. Phys. Lett., Volume 85, Issue 3, pp. 485-487, 2004. Snyder et al., “Radiation response of floating gate EEPROM memory cells”, IEEE Transactions On Nuclear Science, Vol. 36, No. 6, pp.2131-2139, 1989. [1] [2] [3] Oldham et al., “Total Ionizing Does Effect In Mos Oxide And Device”, IEEE Transactions On Nuclear Science, Vol. 50, No. 3, 2003. [4] Giorgio et al., “A model for TID effects on floating gate memory cells”, IEEE Transactions On Nuclear Science, Vol. 51, No. 6, pp.3753-3758, 2004. “Total Ionizing Dose Test Report”, Microsemi www.microsemi.com/document-portal/doc_view/133046rtsx72su-cq256-d1wwa1-tid-report [5] [6] John Robertson, “High dielectric constant gate oxides for metal oxide Si transistors”, Reports on Progress in Physics, Vol.69, No.2, pp. 327–396, 2006. [7] Nahid M. Hossain, Masud H Chowdhury, “Multilayer Graphene Nanoribbon Floating Gate Transistor for Flash Memory”, Proc. Of ISCAS, pp.806-809, 2014. 

1111.4458

1

1111

Carbon Nanotube Initiated Formation of Carbon Nanoscrolls

The unique topology and exceptional properties of carbon nanoscrolls (CNSs) have inspired unconventional nano-device concepts, yet the fabrication of CNSs remains rather challenging. Using molecular dynamics simulations, we demonstrate the spontaneous formation of a CNS from graphene on a substrate, initiated by a carbon nanotube (CNT). The rolling of graphene into a CNS is modulated by the CNT size, the carbon-carbon interlayer adhesion, and the graphene-substrate interaction. A phase diagram emerging from the simulations can offer quantitative guideline toward a feasible and robust physical approach to fabricating CNSs.

cond-mat.mes-hall

cond-mat

Carbon Nanotube Initiated Formation of Carbon Nanoscrolls Zhao Zhanga, Teng Lia,b,* aDepartment of Mechanical Engineering,University of Maryland, College Park, MD 20742 bMaryland NanoCenter, University of Maryland, College Park, MD 20742 Abstract The unique topology and exceptional properties of carbon nanoscrolls (CNSs) have inspired unconventional nano-device concepts, yet the fabrication of CNSs remains rather challenging. Using molecular dynamics simulations, we demonstrate the spontaneous formation of a CNS from graphene on a substrate, initiated by a carbon nanotube (CNT). The rolling of graphene into a CNS is modulated by the CNT size, the carbon-carbon interlayer adhesion, and the graphene-substrate interaction. A phase diagram emerging from the simulations can offer quantitative guideline toward a feasible and robust physical approach to fabricating CNSs. * Author to whom correspondence should be addressed. Electronic mail: LiT@umd.edu 1 A carbon nanoscroll (CNS) is formed by rolling up a graphene sheet into a spiral multilayer structure.1-3 CNSs are topologically open. For example, the core size of a CNS can vary significantly by relative sliding between adjacent layers.4,5 By contrast, a multiwall carbon nanotube (MWCNT) consists of several coaxial carbon cylinders and is topologically closed. The open and highly tunable structure of CNSs, combining with the exceptional mechanical and electronic properties inherited from the basal graphene,6,7 has inspired potential applications of CNSs, such as hydrogen storage medium,8,9 water and ion channels,10 nano-oscillators,11 and nano-actuators.12 Enthusiasm aside, the realization of these promising applications hinges upon feasible and reliable fabrication of high quality CNSs, which remains as a significant challenge. Here we use molecular dynamics (MD) simulations to demonstrate a simple physical approach to fabricating CNSs via CNT-initiated rolling of graphene on a substrate. The experimental discovery of CNSs were achieved via a chemical approach, in which graphite is first intercalated using alkali metals, and the resulting exfoliated graphite sheets can curl into scrolls upon sonication.2 The chemical approach results in scrolls of graphite sheets with undetermined number of layers, which also suffer from the contamination of residual solvent. The surge of interests in graphene in the past several years has enabled the fabrication of graphene monolayers via mechanical exfoliation.13 Recent experiments show that a SiO2- supported graphene monolayer immersed in isopropyl alcohol (IPA) solution can roll up to form a CNS.14 The formation of CNSs is highly sensitive to the concentration of IPA solution and the shape of the graphene. In general, the existing chemical approaches to fabricating CNSs suffer from the possible contamination of chemical residue, and also the difficulty in controlling the rolling initiation and rolling direction. So far, no physical approach to fabricating CNSs has been experimentally demonstrated. 2 Theoretical analysis and MD simulations have been conducted to investigate mechanisms governing the formation of CNSs from graphene monolayer.4,15,16 Simulations show that a sufficiently large overlap between two edges of a freestanding graphene monolayer can lead to further relative sliding of the overlapped area and eventually forming a CNS.4 It is also shown that a long and narrow freestanding carbon nanoribbon can spontaneously form a short CNS driven by low temperature (<100K) thermal fluctuation.16 Recent simulations demonstrate that water nanodroplets could activate the folding of freestanding graphene to form different carbon nanostructures, including CNSs, depending on the size of both graphene and nanodroplets.15 These simulation demonstrations, however, are rather challenging to be realized in experiment, given the challenge to manipulate freestanding graphene and the difficulty to control water nanodroplets. Inspired by recent experiments and simulations,12,14,15 in this letter we use MD simulations to demonstrate an all-dry physical approach to fabricating CNSs, in which the rolling of a substrate- supported graphene monolayer is initiated by a CNT. Figure 1 depicts the simulation model, in which a CNT is placed along the left edge of a flat rectangular graphene monolayer supported by a SiO2 substrate. The CNT-initiated formation of a CNS from the substrate-supported graphene is governed by the interplay among the following energies: the CNT-graphene interaction energy Etg, the graphene-graphene interlayer interaction energy Egg (once graphene starts to roll into a CNS), the graphene strain energy Eg, and the graphene-substrate interaction energy Egs. The non- bonded CNT-graphene interaction and graphene-graphene interlayer interaction can be characterized by vdW force. The weak interaction between a mechanically exfoliated graphene and its substrate can also be characterized by vdW force. Due to the nature of vdW interaction, Etg and Egg minimize when the carbon-carbon (C-C) interlayer distance reaches an equilibrium 3 value, so does Egs when the distance between the graphene and the substrate surface reaches its equilibrium. When the graphene separates from the substrate, curls up to wrap the CNT, and later starts to roll into a CNS, Etg and Egg decrease; on the other hand, Egs increases, so does Eg due to the mechanical deformation of the graphene associated with the wrapping and rolling. Above said, Etg and Egg serve as the driving force, while Egs and Eg represent the resistant force in the CNT-initiated formation of a CNS from a substrate-supported graphene. In the simulations, the C-C bonds in the CNT and graphene are described by the second generation Brenner potential.17 The non-bonded CNT-graphene interaction and graphene- graphene interlayer interaction are described by a Lennard-Jones pair potential )( rV CC = 4 12 σελ ( CC CC CC 12 / r − 6 σ CC / r 6 ) , where ε CC = 0.00284 eV , σ CC = 0.34 nm and λ is a CC tuning factor that is used to vary the C-C interaction energy to study its effect on the CNS formation. It has been shown that the effective C-C interaction energy in a CNS can be tuned by an applied dc/ac electric field 12,18. The non-bonded graphene-SiO2 substrate interaction is described by a Si-C pair potential and an O-C pair potential, both of which take the same form of )(rVCC but with different parameters, that is, ε SiC = 0.00213 eV , σ SiC = 0.15 nm , ε OC = 0.00499 eV and σ OC = 0.23 nm , respectively. The tuning factor for the graphene-substrate λ is taken to be the same for both Si-C and O-C pair potentials. To reduce the interaction CS computation size, all atoms in the SiO2 substrate are fixed during simulation. This assumption is justified by the weak graphene-substrate interaction and the rigidity of bulk SiO2. The graphene used in the simulations is 50 nm long and 4 nm wide, with the right edge constrained on the substrate by a linear spring. CNTs with length of 4 nm but various diameters are used to study 4 the effect of CNT size on the CNS formation. The MD simulations are carried out using LAMMPS19 with NVT ensemble at temperature 300K and with time step 1 fs. Depending on the CNT size, C-C interaction strength and C-SiO2 interaction strength, the CNT-graphene-substrate system shown in Fig. 1 evolves in three different modes. Figure 2 illustrates the time sequential snapshots of each mode of evolution and the corresponding variation in the total potential energy of the system. In the three cases shown in Fig. 2, λ CC 1= and λ CS 1= . CNTs of various diameters (i.e., (10,10), (12,12) and (18,18)) are used, respectively. The graphene strain energy Eg due to wrapping a CNT is roughly inversely proportional to the square of the CNT diameter. If the CNT diameter is too small, the significant increase of Eg and the corresponding increase of Egs due to graphene-substrate separation can overbalance the decrease of Etg due to graphene wrapping the CNT. As a result, instead of wrapping the CNT, the graphene remains flat on the substrate, while the CNT glides on the graphene driven by the thermal fluctuation, as shown in Fig. 2b-e. The translational motion of the CNT on the graphene is expected to cause negligible variation to the total potential energy of the system except the thermal fluctuation, which is evident in Fig. 2a. If a CNT of intermediate diameter is used, the increase of Eg due to graphene bending and Egs due to graphene-substrate separation can be outweighed by the corresponding decrease of Etg. Consequently, graphene can separate from the substrate under thermal fluctuation and start to wrap around the CNT (Fig. 2g). The total potential energy continues decreasing until the whole surface of the CNT is nearly wrapped by the graphene (Fig. 2h). Further rolling of the graphene is hindered by a local energy barrier due to the step formed by the left edge of the graphene adhering to the CNT. If the CNT radius is not sufficiently large, the energy barrier due to the 5 graphene edge step can be too high and thus prevent the further rolling of graphene. This is analogous to a roller moving toward a speed bump of a fixed height. If the roller is too thin, instead of passing the bump, it can be bounced up in the air. As shown in Fig. 2h-i, the graphene- wrapped CNT rolls toward the graphene edge step, and is then bounced upward. The kinetic energy of the graphene-wrapped CNT leads to further separation of a short segment of graphene from the substrate, and then the graphene-substrate interaction pulls the separated graphene re- adhered back to the substrate. Such two processes repeat several times and the graphene segment eventually re-adheres back to the substrate after the excess translational kinetic energy is dissipated. The CNT remains wrapped by the graphene (Fig. 2j). If the diameter of the CNT is sufficiently large, the translational kinetic energy of the graphene-wrapped CNT can overcome the fixed energy barrier due to the graphene edge step. As a result, CNT-initiated rolling of the graphene continues and then an overlap between the left edge and the flat portion of the graphene forms. Such an overlap leads to the decrease of Egg, which drives further rolling of graphene into a CNS (Fig. 2l-o, and also the supplemental materials). As shown in Fig. 2k, the continuous rolling of the graphene results in further decrease of the overall potential energy, which is more substantial than that due to the graphene wrapping the CNT. The resulting CNS is energetically stable against thermal perturbations at 300 K. When the CNT size and C-SiO2 interaction strength are fixed, the evolution of the CNT- graphene-substrate system can be modulated by the C-C interaction strength. Figure 3a defines a phase diagram of the evolution of the CNT-graphene-substrate system in the space of C-C interaction strength and CNT size, for a given C-SiO2 interaction strength (i.e., λ CS 1= ). The same three modes of evolution as described above are observed. For a given CNT size, the mode 6 of evolution changes from CNT gliding to graphene wrapping and then to graphene forming a CNS, as the C-C interaction becomes stronger. For a given C-C interaction strength, the similar change of the mode of evolution is shown as the CNT size increases. Emerging from the simulations are a boundary between mode I and mode II and that between mode II and mode III, the latter of which can serve as a guidance for controlling CNS formation by varying C-C interaction and selecting CNT size. Figure 3b plots the case of λ CS 4= . When the C-SiO2 interaction strength increases, a stronger C-C interaction is needed for the CNS formation, for a given CNT size; similarly, a CNT with larger diameter is needed to initiate the CNS formation, for a given C-C interaction strength. For the case of λ CS 4= λ , , there exists a critical value of CC below which graphene rolling into a CNS cannot be initiated by a CNT of any given size. In summary, we demonstrate the CNT-initiated formation of a CNS from substrate-supported graphene, using MD simulations. The CNT is shown to help overcome the energy barrier to form an overlap in graphene. Once the overlap is formed, the graphene can spontaneously roll up into a CNS. The successful formation of a CNS depends on the CNT diameter, the C-C interaction strength and the graphene-substrate interaction strength. The phase diagram obtained from this study elucidates critical parameters governing the formation of CNSs from graphene. With the ever maturing fabrication of high quality CNTs and large area graphene on substrates, and the nanopatterning technique to position these building blocks at high precision, the CNT-initiated formation of CNSs holds great potential leading to a feasible, all-dry, physical fabrication technique of high quality CNSs. The resulting CNS nanostructures hold potential to enable unconventional nanoscale electromechanical devices.12 7 Acknowledgement: This work is supported by a UMD GRB summer research award and NSF Grant No. 0928278. Z.Z. also thanks the support of A. J. Clark Fellowship and UMD Clark School Future Faculty Program. 8 References 1 R. Bacon, J. Appl. Phys. 31, 283 (1960). 2 L. M. Viculis, J. J. Mack, and R. B. Kaner, Science 299, 1361 (2003). 3 H. Shioyama and T. Akita, Carbon 41, 179 (2003). 4 S. F. Braga, V. R. Coluci, S. B. Legoas, R. Giro, D. S. Galvao, and R. H. Baughman, Nano Lett. 4, 881 (2004). 5 X. H. Shi, N. M. Pugno, and H. J. Gao, J. Comp. Theo. Nanosci. 7, 517 (2010). 6 C. Lee, X. Wei, J. W. Kysar, and J. Hone, Science 321, 385 (2008). 7 A.H.C. Neto, F. Guinea, N.M.R. Peres, K.S. Novoselov, and A.K. Geim, Rev. Mod. Phys. 81, 109 (2009). 8 V. R. Coluci, S. F. Braga, R. H. Baughman, and D. S. Galvao, Phys. Rev. B 75 (2007). 9 G. Mpourmpakis, E. Tylianakis, and G. E. Froudakis, Nano Lett. 7, 1893 (2007). 10 X. Shi, Y. Cheng, N. M. Pugno, and H. Gao, Small 6, 739 (2010). 11 X. Shi, N. M. Pugno, Y. Cheng, and H. Gao, Appl. Phys. Lett. 95, 163113 (2009). 12 X. Shi, Y. Cheng, N. M. Pugno, and H. Gao, Appl. Phys. Lett. 96, 517 (2010). 13 A. K. Geim and K. S. Novoselov, Nat Mater 6, 183 (2007). 14 X. Xie, L. Ju, X. Feng, Y. Sun, R. Zhou, K. Liu, S. Fan, Q. Li, and K. Jiang, Nano Lett. 9, 2565 (2009). 15 N. Patra, B. Wang, and P. Král, Nano Lett. 9, 3766 (2009). 16 B. V. C. Martins and D. S. Galvao, Nanotech. 21 (2010). 17 D. W. Brenner, O. A. Shenderova, J. A. Harrison, S. J. Stuart, B. Ni, and S. B. Sinnott, J. Phys.: Cond. Matt. 14, 783 (2002). 18 R. Langlet, M. Devel, and P. Lambin, Carbon 44, 2883 (2006). 19 S. Plimpton, J. .Comp. Phys. 117, 1 (1995). (http://lammps.sandia.gov) 9 CNT Graph en e S iO2 FIG. 1. (Color online) The MD simulation model. A graphene is supported by a SiO2 substrate, with a CNT placed along the left edge of the graphene. 10 Mode I CNT gliding on graphene (e) (c)(d) (b) Mode II Graphene wrapping CNT (j) (i) (h) (g) (f) 50 100 150 200 Time (ps) (g) (h) (i) (j) Mode III Graphene rolling into CNS 400 200 0 -200 -400 0 (l)(m) (n) (o) (k) 50 100 Time (ps) 150 (l) (m) (n) (o) 40 20 0 -20 -40 0 (a) 50 100 Time (ps) 150 200 (b) (c) (d) (e) ) eV 40 20 0 -20 -40 0 Po te nti al E n erg y V ar iati on ( FIG. 2. (Color online) Three modes of evolution of the CNT-graphene-substrate system. (a, f, k) plot the variation in the total potential energy of the system as a function of simulation time for each mode, respectively; (b-e): Snapshots of a (10, 10) CNT gliding on the substrate-supported graphene at 0 ps, 15 ps, 40 ps and 100 ps, respectively; (g-j): Snapshots of the graphene wrapping a (12, 12) CNT at 20 ps, 65 ps, 120 ps and 200 ps, respectively; (l-o): Snapshots of the graphene rolling into a CNS, initiated by a (18,18) CNT, at 20 ps, 45 ps, 60 ps and 160 ps, respectively. The two dotted fitting curves in (k) show that the graphene further rolling into a CNS (from (n) to (o)) leads to more substantial decrease of potential energy than that due to graphene wrapping CNT (from (l) to (m)). In all three cases shown here, λ CC 1= and λ CS 1= . (enhanced online). 11 CNT Chirality (10,10) (12,12) (14,14) (16,16) (18,18) (20,20) Mode I Mode II Mode III (a) 2.0 1.5 λCC 1.0 λCS=1 0.5 12 16 24 20 CNT Diameter (Å) 2.0 (b) 1.5 λCC 1.0 0.5 12 Mode I Mode II Mode III λCS=4 16 24 20 CNT Diameter (Å) 28 28 FIG. 3. (Color online) Phase diagrams of the evolution of the CNT-graphene-substrate system in the space of C-C interaction strength and CNT size, for a given C-SiO2 interaction strength. Here, (a) λ CS 1= , (b) λ CS 4= . 12 

1504.04550

1

1504

Prediction of inelastic light scattering spectra from electronic collective excitations in GaAs/AlGaAs core-multishell nanowires

We predict inelastic light scattering spectra from electron collective excitations in a coaxial quantum well embedded in a core-multishell GaAs/AlGaAs nanowire. The complex composition, the hexagonal cross section and the remote doping of typical samples are explicitly included, and the free electron gas is obtained by a DFT approach. Inelastic light scattering cross sections due to charge and spin collective excitations belonging to quasi-1D and quasi-2D states, which coexist in such radial heterostructures, are predicted in the non-resonant approximation from a fully three-dimensional multi-subband TDDFT formalism. We show that collective excitations can be classified in azimuthal, radial and longitudinal excitations, according to the associated density fluctuations, and we suggest that their character can be exposed by specific spectral dispersion of inelastic light scattering along different planes of the heterostructure.

cond-mat.mes-hall

cond-mat

Prediction of inelastic light scattering spectra from electronic collective excitations in GaAs/AlGaAs core-multishell nanowires Miquel Royo1,2,∗ Andrea Bertoni2, and Guido Goldoni2,3 1Departament de Qu´ımica F´ısica i Anal´ıtica, Universitat Jaume I, E-12080, Castell´o, Spain 2CNR-NANO S3, Institute for Nanoscience, Via Campi 213/a, 41125 Modena, Italy and 3Department of Physics, Informatics and Mathematics, University of Modena and Reggio Emilia, Via Campi 213/a, 41125 Modena, Italy (Dated: May 21, 2021) Abstract We predict inelastic light scattering spectra from electron collective excitations in a coaxial quantum well embedded in a core-multishell GaAs/AlGaAs nanowire. The complex composition, the hexagonal cross section and the remote doping of typical samples are explicitly included, and the free electron gas is obtained by a DFT approach. Inelastic light scattering cross sections due to charge and spin collective excitations belonging to quasi-1D and quasi-2D states, which coexist in such radial heterostructures, are predicted in the non-resonant approximation from a fully three-dimensional multi-subband TDDFT formalism. We show that collective excitations can be classified in azimuthal, radial and longitudinal excitations, according to the associated density fluctuations, and we suggest that their character can be exposed by specific spectral dispersion of inelastic light scattering along different planes of the heterostructure. 5 1 0 2 r p A 7 1 ] l l a h - s e m . t a m - d n o c [ 1 v 0 5 5 4 0 . 4 0 5 1 : v i X r a 1 I. INTRODUCTION Inelastic light scattering (ILS) spectroscopy is a fundamental optical technique to study electronic properties of excess carriers in semiconductors.1 -- 4 First, it enables to probe elec- tronic elementary excitations, which in semiconductors occur in the range of few tens of meV, with near-infrared and visible light for which high-quality tunable lasers and detectors exist. Second, ILS enables to study separately different types of elementary excitations thanks to the polarization selection rules.3 Indeed, the ILS cross section arises from collective charge density excitations (CDEs), or plasmons, when the polarizations of the incident and scat- tered light are parallel (polarized configuration), and from collective spin density excitations (SDEs) when the two polarizations are orthogonal (depolarized configuration). In addition, single-particle excitations (SPEs) can be observed in both polarization configurations under strong inter-band resonance conditions.5 Finally, by adjusting the relative direction of the incident and scattered photons, it is possible to tune the momentum transferred to the elec- tronic system in a given direction, which, due to the momentum dependent selection rules,3 enables to map the dispersion of the elementary excitations. The wavevector dispersion of elementary excitations is strongly dependent on the di- mensionality of the electronic system.4,6,7 This is particularly important for intra-subband plasmons, as well as the Landau damping in SPEs continua. Such features have been demon- strated in ILS experiments in semiconductor heterostructures of reduced dimensionality such as quantum wells (QWs),8 -- 12 quantum wires6,13,14 and quantum dots.15,16 The comparison of the experimental results with theoretical models yielded information about the carriers energy structure, density, mobility and many-body contributions. Recently, ILS experiments and theoretical predictions have been used to confirm the accu- mulation of excess electrons in the conduction band of radial heterostructures in nanowires, namely, core-multishell nanowires (CSNW) hosting a coaxial quantum well (coQW), similar to the one shown in Fig. 1(a).17 The extra signals observed in the ILS spectra of remotely- doped coQWs, in comparison with those of undoped reference samples, were shown to origi- nate from the collective excitations of a high-mobility electron gas confined inside the coQW. Furthermore, ILS resonances could be selectively assigned to quasi-one-dimensional (q1D) and quasi-two-dimensional (q2D) states which coexist in the structure, due to the hexagonal nanowire cross section, being localized at the corners and facets of the coQW, respectively. 2 Such a spatial-dependent dimensionality and the tubular shape of the electron channel, which bridges between q2D and a q1D electron gas, guarantee a particular dispersion of the collective excitations in the confined in-plane directions and a diameter-dependent momen- tum dispersion along the invariant longitudinal direction. However, results in Ref. 17 were restricted to a backscattering geometry with incidence normal to the facets of the coQW and did not probe the energy dispersion of such excitations. It is important to note that ILS in coQWs substantially differs from ILS in planar struc- tures. On the one hand, in coQWs there is only one translational invariant direction (along the nanowire axis), the other direction being wrapped around the CSNW core, removing the continuous energy dispersion. The discrete symmetry also induces a inhomogeneous carrier distribution, with coexisting q1D and q2D states (see Fig. 1(b)). On the other hand, the photon field always impinges on more than one coQW planes, and with different angles, at the same time (see Fig. 1(a)). Since selection rules depend on the photon field direction, ILS spectra in coQWs are expected to be substantially more complex than in traditional QWs. The dispersion of plasmon modes has been theoretically investigated in cylindrical tubu- lar geometries.18 -- 22 These studies, mostly RPA calculations, illustrated the dimensional crossover from q2D to q1D electron gas for very small radius of the tubule, a regime appro- priate, e.g., to carbon nanotubes but not to CSNWs. Furthermore, these works neglected any discrete symmetry characteristic of the nanowire cross section. The lowest collective ex- citations of an electron gas in CSNWs, including the hexagonal symmetry of the electronic system, have been studied recently by our group,23 which enabled to rationalize the complex excitation spectra in terms of symmetry allowed excitations and symmetry selective Landau damping. However, the high electron density regime, relevant for ILS experiments, and the ensuing self-consistent field was not included. Likewise, ILS dispersion was not discussed. In this paper, we study theoretically the ILS spectra of a prototypical GaAs/Al0.3Ga0.7As CSNW hosting a remotely doped coQW at large electron densities.17,24 Assuming non- resonant conditions, the scattering cross section is obtained from the dynamic response functions of the excess electrons. Calculations of CDE and SDE spectra are performed in a 3D real-space multi-subband DFT and TDDFT formalism which accounts for the complex geometry and composition of the structure, the self-consistent field of excess electrons, and the plasmon-phonon coupling. Three different classes of collective excitations arise in the spectra, which involve density fluctuations occurring in the azimuthal, radial or longitudinal 3 directions of the coQW. We show that these collective modes can be singled out by properly designed ILS experiments. II. THEORETICAL MODEL A. Self-consistent ground-state calculation To obtain the electronic states of a CSNW we employ a standard envelope function approach in a single parabolic band approximation. Electron-electron interactions have been treated at the mean-field level through a typical Kohn-Sham (KS) LDA procedure. We consider a CSNW which is spatially invariant along the NW axis direction z (see Fig. 1(a) for axis definitions). Therefore, we factorize the envelope functions as Ψn,kz (r, z) = ϕn(r) eikzz, . In the r ≡ (x, y) plane, the NW cross with parabolic energy dispersion εn(kz) =
n + section is hexagonal and the material and doping modulations are mapped on a hexagonal 2 k2 2 m∗ z e domain using a symmetry-preserving triangular grid. The self-consistent potential vKS(r) = v(r) + vH(r) + vXC(r) includes the effect of the spatial confinement v(r) determined by the materials conduction band offset, the Hartree potential vH(r) generated by the free electrons and the static doping, and an approximate exchange-correlation potential vXC(r).25 Further details can be found in Ref. 26. B. ILS spectra calculation ILS spectra have been calculated within a multi-subband TDLDA non-resonant formal- ism.27 The scattering cross section due to CDEs is obtained from the imaginary part of the momentum-dependent density-density response function (DDRF),12,28 ICDE(Q, ω) ∝ −(cid:61)(cid:104) Π(Q, ω) (cid:105) = −(cid:61) (cid:20)(cid:90)(cid:90) (cid:21) dr dr(cid:48)e−i q(r−r(cid:48)) Π(r, r(cid:48), qz, ω) , (1) where, q is the in-plane and qz the in-wire, longitudinal components of the total momentum Q ≡ (q, qz) exchanged in the scattering process, i.e., Q = Qi − Qs, Qi and Qs being the momenta of the incident and scattered photons, respectively, and ω = ωi − ωs is the ILS energy shift. To obtain the dynamic DDRF, we expand Π(r, r(cid:48), qz, ω) in terms of the in-plane KS envelope functions, 4 Π(r, r(cid:48), qz, ω) = (cid:88) ijlm Πijlm(qz, ω) ϕ∗ i (r) ϕj(r) ϕl(r(cid:48)) ϕ∗ m(r(cid:48)), (2) where the elements Πijlm(qz, ω) are defined by the Dyson equation (cid:88) ijlm Πijlm(qz, ω) = (cid:88) ijlm Π0 ij(qz, ω) δil δjm + (cid:88) ij Π0 ij(qz, ω) (cid:2)vijkn(qz) + uXC ijkn (cid:3) Πknlm(qz, ω). (cid:88) knlm Here, Π0 ij(qz, ω) are the elements of the DDRF for the KS system, (cid:90) dkz 2π Π0 ij(qz, ω) = g fi(kz) − fj(kz + qz) ω − (εj(kz + qz) − εi(kz)) + iη , (3) (4) where g = 2 accounts for electron spin degeneracy, fn(kz) is the Fermi occupation function and η is the electron damping parameter. vijkn(qz) and uXC ijkn are the direct Coulomb and exchange-correlation matrix elements, respectively, which describe the dynamic interaction of two electrons, one of which gets scattered from state i to j and the other from k to n, with an exchange of momentum qz. The direct Coulomb matrix elements read (cid:90) (cid:90) dr vijkn(qz) = (5) where V (r − r(cid:48), qz) is the Fourier transform of the Coulomb operator in the z direction. We have used a frequency dependent dielectric constant ε(ω), entering the denominator of the dr(cid:48) φi(r) φ∗ j (r) V (r − r(cid:48), qz) φ∗ k(r(cid:48)) φn(r(cid:48)), Coulomb operator instead of the semiconductor high-frequency dielectric constant ε∞, that phenomenologically accounts for the coupling of CDEs with the phonons of the underlying crystal lattice, ε(ω) = ε∞ ω2 − ω2 ω2 − ω2 LO + i Γph ω T O + i Γph ω . (6) Here, ωLO and ωT O are the resonant frequencies of the longitudinal and transverse phonons, respectively, and Γph is a damping parameter. The exchange-correlation matrix elements read (cid:90) (cid:90) ijkn = − uXC dr dr(cid:48) φi(r) φ∗ j (r) fXC(r, r(cid:48)) φ∗ k(r(cid:48)) φn(r(cid:48)), (7) 5 with fXC(r, r(cid:48)) being the dynamic exchange-correlation kernel defined in the adiabatic LDA as the derivative of the static exchange-correlation potential with respect to the ground state density, fXC(r, r(cid:48)) = δvXC [n0](r) δn0(r(cid:48)) δ(r − r(cid:48)). The TDLDA formalism also provides the ILS intensity due to SDEs. These are only cou- pled through indirect-exchange interactions. Consequently, they always appear redshifted with respect to their plasmonic counterparts. The ILS cross section due to SDEs is therefore obtained from the imaginary part of the so-called momentum-dependent irreducible response function (IRF), Π(Q, ω), which is calculated as above for the DDRF, but with the direct Coulomb integrals set to zero in Eq. (3). From the DDRF and the IRF we can obtain the density fluctuation induced by the electro- magnetic field at a given energy and momentum, the so-called induced density distribution (IDD), from Kubo's correlation formula.29 Thus, for CDEs the IDD is given by (cid:90) δnCDE(r, q, qz, ω) = dr(cid:48) Π(r, r(cid:48), qz, ω) ei qr(cid:48) , (8) whereas for SDEs, δnSDE(r, q, qz, ω) is equivalently calculated substituting Π(r, r(cid:48), qz, ω) by Π(r, r(cid:48), qz, ω). III. NUMERICAL RESULTS A. Electron gas ground-state distribution Calculations are performed for the CSNW shown in Fig. 1(a). A GaAs core is surrounded by Al0.3Ga0.7As/GaAs/Al0.3Ga0.7As/GaAs layers with thickness indicated in the figure. A n-type δ-doping layer is included in the middle of the outer Al0.3Ga0.7As layer. System pa- rameters are similar to those of the sample recently studied in Ref. 17, although the thickness of the GaAs coQW is slightly narrower here (26 nm in the experiment) to spectrally separate different classes of transitions, which are partially overlapping in Ref. 17, as discussed in Sec. III B. Material parameters for GaAs (Al0.3Ga0.7As) are the isotropic electron effective mass e = 0.067 (0.092) and the static dielectric constant ε = 13.18 (12.24). The GaAs/Al0.3Ga0.7As m∗ conduction band offset is taken as 284 meV. The Fermi level is pinned at the middle of the GaAs band gap (taken as 1.43 eV), the temperature is 4 K, and a common homogeneous 6 FIG. 1. (a) Schematics of a prototypical CSNW investigated in the paper. Dark, middle and light green colors are used for Al0.3Ga0.7As, GaAs, and n-doping regions, respectively. (b) Calculated electron gas distribution over the CSNW cross section. Only the GaAs core and coQW regions are shown. triangular grid with ∼ 3.21 nodes/nm2 is used to numerically integrate the KS and Poisson equations. Unless otherwise stated, we have used a constant density of static donors over the δ-doping layer of 2 × 1018 cm−3, which yields an accumulated electron gas with a total density ∼ 1.13 × 107 cm−1. As shown in Fig. 1(b), electrons are almost exclusively localized in the GaAs coQW with only a minority of them distributed over the GaAs core. In the coQW, in turn, electrons are preferentially localized in the hexagonal corners in order to minimize electron-electron interactions and single-particle confinement energies. This is an electron distribution previously observed in different hexagonal CSNW heterostructures at 7 (b)(a)sufficiently large electron density,17,26,30 -- 34 which points to coexisting channels of different dimensionalities, namely, q1D states in the corners and q2D states in the facets of the coQW. B. ILS at normal incidence We now study the ILS spectra due to the elementary excitations of the conduction free electrons accumulated in the CSNW in a back-scattering geometry (i.e., with antiparallel incident and scattered photons) and with the photon field at normal incidence with respect to the top facet of the coQW. In this configuration no momentum is transferred to the electronic system along the invariant CSNW axis direction. In the calculation, we have assumed a typical excitation energy of 1.92 eV. We approximate Qi = Qs, yielding a transferred momentum Q = 1.38×106 cm−1, which in our sample corresponds to ∼ 1.03 kF , kF being the Fermi wave-vector. We have used a basis set formed by the 100 lowest-lying KS subbands which assures convergence within the considered energy range. Finally, in order to account for the plasmon-phonon coupling effect we have used the GaAs high-frequency dielectric constant ε∞ = 10.86, the GaAs phonon resonances ωT O(LO) = 33.72 (36.69) meV, and a damping parameter Γph = 1 meV. In Fig. 2 we show the calculated CDE (a) and SDE (b) ILS spectra with an electron damping η = 2 meV, which leads to peak bandwidths comparable to those measured in recent ILS experiments in CSNWs.17 We also show a breakdown of the spectra in contribu- tions due to excitations in the GaAs coQW (blue) and core (red). As can be seen, the cross section is almost entirely originated in coQW excitations, with some weak core excitations at low energies that hardly influence the total spectrum. Likewise, we have found no appre- ciable coupling between core and coQW excitations by comparing the spectra in Fig. 2 with equivalent ones obtained by deactivating the dynamic Coulomb and exchange-correlation integrals between core and coQW states (not shown). Such a result contrast with the strong inter-tubule coupling observed in coaxial tubules18,21 in which, however, the inter-tubule separation was of the order of few A. Both CDE and SDE spectra are dominated by a low-energy broad band, labeled c and f in Fig. 2(a) and (b), peaking at ∼ 18.9 and ∼ 10.8 meV, respectively. From the energy of the corresponding peak in the SPE spectrum (not shown), we estimate a depolarization 8 FIG. 2. CDE (a) and SDE (b) spectra in a back-scattering configuration with the photons at normal incidence on the top CSNW facet. Blue and red lines show the contribution to the total spectra from excitations occurring in the GaAs coQW and core, respectively. The dashed line in (a) shows the CDE spectrum calculated neglecting the plasmon-phonon coupling. Vertical gray lines mark the position of the GaAs LO and TO phonon resonances. (c)-(g) Real (left) and imaginary (right) parts of the IDDs calculated at the resonances indicated by arrows in the spectra (a) and (b). shift ∼ 5.8 meV and an excitonic correction ∼ 2.8 meV. As reported in Ref. 17, this broad signal is due to excitations between the discrete subbands which arise from the discretization of the 2D continuum of an equivalent flat QW due to azimuthal periodicity of the coQW. Note that in a planar QW these collective modes would correspond to genuine intra-subband excitations and would not be excited at normal incidence. In our CSNW these modes are 9 0123456ILS Intensity (arb. units)01234567801020304050ILS Intensity (arb. units)ILS shift (meV)(a)(b)cdefg(c)(d)(e)(f)(g)TOLOactivated through the lateral facets which offer oblique orientation to the incident photons. By this reason, and in order to avoid confusion with the genuine 1D intra-subband excitations along the CSNW axis, we will refer to these excitations as 2D intra-subband excitations. This assignment is confirmed by the calculated IDDs shown in panels (c) and (f) of Fig. 2, respectively. These clearly correspond to density fluctuations running along the lateral facets of the coQW with nodal surfaces normal to the coQW planes (azimuthal nodes). No nodal surfaces parallel to the QW plane (radial nodes) are present, since these are associated to inter-subband excitations lying at larger energies, as we discuss next. The CDE spectrum in Fig. 2(a) also shows two less intense peaks below and above the GaAs phonon resonances (gray vertical lines), labeled d and e. The spectrum calculated neglecting the plasmon-phonon coupling (dashed line in Fig. 2(a)) demonstrates that these two peaks are the split pair of a single CDE resonance due to coupling with the phonons. The corresponding SDE peak, labeled g in Fig. 2(b), is at ∼ 31 meV. This collective mode corresponds to the first inter-subband excitation in the coQW, as clearly demonstrated by the calculated IDDs shown in Figs. 2(d),(e) and (g) featuring one radial node in the top and bottom facets. The depolarization shift for this mode (calculated using the energy of the CDE without phonon coupling) is ∼ 3.9 meV and the excitonic correction is ∼ 2 meV. Therefore, many-electron corrections are smaller for the inter-subband than for the 2D intra-subband mode, which is reasonable due to the higher confinement regime of the former. Note that the position of the inter-subband peaks is very sensitive to the coQW thickness, and for thicker coQWs (as in Ref. 17) these modes are likely to merge to the broad 2D intra-subband peak. In Fig. 3 we show the effect of reducing the value of the electron damping, which might be achieved in samples with higher electron mobilities, on the calculated ILS spectra. In contrast to a flat QW, where the damping only modifies the bandwidth of single intra- or inter-subband excitations, in a CSNW a small damping exposes a complex fine-structure. Indeed, at the energy of the broad 2D intra-subband band, numerous close by collective excitations show up. As discussed above, these excitations are associated to transitions between states with different number of azimuthal nodes, arising from the azimuthal dis- cretization of the 2D continuum. Also the inter-subband peak, which is mainly associated to transitions with a change in one radial node, shows several resonances at small damping, which correspond to different changes in the number of azimuthal nodes, as shown by the 10 FIG. 3. (a) CDE and (b) SDE spectra calculated using an electron damping of 2, 1 and 0.5 meV from top to bottom lines. For clarity, the curves have been offset and scaled, as indicated by labels. corresponding IDDs (Figs. 2(d),(e) and (g)). Note indeed that the two types of excitations cannot be excited individually due to the different impinging angle on different facets. Since in Fig. 3 the fine structure emerges already at a damping of ∼ 1 meV, the fine structure is likely to be observed in future ILS experiments. C. Longitudinal dispersion We now study the dispersion of the ILS spectra as the direction of the photons is tilted (always assuming a back-scattering geometry) in the longitudinal direction, along the CSNW axis. We indicate with θ the angle between the direction of the photon beam and the y-axis, and with ϕ the angle between the projection of the beam on the (x, z) plane with the z-axis (see inset in Fig. 4(b)). We consider the dispersion over the (y, z) plane (ϕ = 0◦) as the 11 x1x2x30481216ILS Intensity (arb. units)0481216202401020304050ILS Intensity (arb. units)ILS shift (meV)x1x2x3(a)(b)beam is tilted from normal to parallel to the CSNW top facet (i.e., from θ = 0◦ to θ = 90◦). In this configuration the momentum transferred along the invariant direction excites genuine intra-subband transitions within the 1D parabolic subbands. At large angles, moreover, SPE continua gain spectral weight and induce Landau damping, which in a hexagonal NW is a symmetry restrictive process.23 FIG. 4. CDE (a) and SDE (b) spectra dispersion over the (y, z) plane. Calculated spectra with different photon orientations θ in steps of 10◦ are shown. Each curve is shifted proportionally to the in-wire transferred momentum qz. The top curves correspond to qz = 0, the bottom curves correspond to q =∼ 1.03 kF. The inset in (b) illustrates the scattering geometry. The (y, z) plane dispersion of the CDE and SDE spectra are respectively shown in Figs. 4 (a) and (b). Clearly, the spectra are strongly affected by the photon orientation. A main feature is the disappearance of the inter-subband resonances as the photons are tilted toward the longitudinal direction. For SDE, e.g, the peak at ∼ 31 meV disappears at θ >∼ 30◦. The 12 024681012141618ILS Intensity (arb. units)051015202501020304050ILS Intensity (arb. units)ILS shift (meV)(a)(b)=0º=90º=0º=90ºFIG. 5. (a) CDE spectra dispersion over the (y, z) plane calculated using an electron damping of 0.1 meV. Curves corresponding to different photon orientations are shown for selected values of θ and have been shifted proportionally to the in-wire transferred momentum qz. (b) IDD calculated at the resonant position of the long-lived CDE marked with an arrow in Fig. (a) for θ = 30◦. inter-subband phonon-coupled peaks in the CDE spectrum also disappear, but a narrow peak appears at ∼ 39 meV as a consequence of the coupling between the phonons and the persisting lower energy broad band which approaches the phonon resonances as θ increases. Vanishing of the inter-subband resonances with increasing transferred momentum in the invariant direction has been reported for planar systems both in simulations35 and experi- ments.36,37 This is due, on the one hand, to Landau damping of some collective modes inside SPE continua and, on the other hand, to the reduction of the momentum transferred normal to the QW plane. In our calculations this results from the ILS selection rules arising from the integrals in Eq. (1), which vanish for inter-subband transitions when q = 0 (θ = 90◦). Correspondingly, the 2D intra-subband excitations associated with azimuthal density fluc- tuations are also expected to weaken as θ is increased since the momentum transferred along the azimuthal direction is reduced. In fact, the the 2D intra-subband CDE resonance at ∼ 19 meV for θ = 0◦ (top curve in Fig. 4(a)) reduces in intensity as θ increases. However, 13 θ=0ºθ=20ºθ=40ºθ=60º(a)01234567891001020304050ILS Intensity (arb. units)ILS shift (meV)(b)θ=30ºbθ=10ºθ=30ºan additional peak starts to be observed at θ = 30◦ as a low-energy shoulder of the 2D intra-subband resonance. This new peak originates in 1D intra-subband excitations and it shows the characteristic strong blueshift with qz of an intra-subband plasmon.23 In order to better analyze this resonance, in Fig. 5(a) we examine its dispersion in detail using a smaller electron damping (0.1 meV). The figure shows that the resonance is origi- nated in a single collective mode which becomes the only one optically active for θ ≥ 20◦. Therefore, contrary to the numerous 2D intra-subband and inter-subband modes showed in Fig. 3, only a single intra-subband plasmon is observed in spite of the high number of occupied subbands. A similar result was reported in Ref. 18 for a zero-thickness cylindrical tubule. Specifically, when multiple 1D subbands are occupied, one among all possible intra- subband plasmons gets an increasingly larger spectral weight. This is the intra-subband plasmon corresponding to a coherent oscillation of all electrons along the longitudinal di- rection of system. The IDDs calculated for the long-lived peak at θ = 30◦ (Figs. 5(b)) confirms this assignment, since almost no density fluctuation is found over the CSNW cross section because the 1D intra-subband character of the mode induces density fluctuations in the longitudinal direction. Finally, note that in the longitudinal limit (θ ≥ 60◦ in Fig. 4(a)) the ILS intensity is exclusively due to the intra-subband SPE continuum, and the long-lived intra-subband plasmon is Landau damped. This is a consequence of the effective dimensionality of the system induced by the large density of states along the facets, as pure 1D plasmons would not be Landau damped.6 D. In-plane dispersion We next discuss the dispersion when the photon beam is tilted in the plane of the cross- section of the CSNW, that is the (x, y) plane (see Fig. 6(b) inset). To estimate the expected anisotropy, we compare the spectra calculated at normal incidence (θ = 0◦) and with photons propagating parallel to a maximal diameter (θ = 30◦). ILS selection rules were discussed in Ref. 23 for these scattering configurations in a model coQW system with few occupied subbands.23 While some of the excitations are excited in both configurations, some other excitations are selectively totally or strongly suppressed in either configurations. Similar arguments apply to the present study. However, in a realistic experimental situation, which 14 FIG. 6. (a) CDE and (b) SDE spectra for a sample with doping density 2× 1018 cm−3. Black-solid (red-dashed) lines correspond to ϕ = 90◦, θ = 0◦ (θ = 30◦). The inset illustrates the scattering geometry. (c) CDE and (d) SDE spectra with doping density 2.7× 1018 cm−3. (e) and (f) are IDDs calculated at the CDE energies marked with arrows in panel (c). we simulate here using a large electron gas density and electron damping in the meV range, it would not be possible to single out specific excitations, due to the large density of states of the coQW. Nevertheless, these selection rules are at the origin of the results discussed below. 15 0123456ILS Intensity (arb. units)0123456789012345678901020304050ILS Intensity (arb. units)ILS shift (meV)02468101201020304050ILS shift (meV)(a)(b)(c)(d)(e)(f)efThe CDE and SDE spectra for the sample studied in the previous sections are shown in Figs. 6 (a) and (b). Tilting the photon beam over the (x, y) plane slightly reshapes the peaks but otherwise it has a small effect on the spectra. In general, at θ = 30◦ the intensity of the 2D intra-subband peaks is increased, while inter-subband peaks are weakened, in comparison with the peaks at θ = 0◦. This is originated in the fact that for θ = 0◦ more momentum is transfered in the radial than in the azimuthal direction, thus favoring inter- subband excitations, whereas the opposite holds for θ = 30◦. In Figs. 6(c) and (d) we show the spectra for a larger doping density, 2.7 × 1018 cm−3, with a total free electron density ∼ 3.11 × 107 cm−1.38 Compared with the spectra for the lower density, all peaks appear blueshifted due to the larger subband splitting. For CDEs, however, this is also due to the dynamic many-electron Coulomb contributions typical of larger electron densities.17 Moreover, the 2D intra-subband and the inter-subband peaks partially overlap due to the larger blueshift of the former. In this larger carrier density regime, the spectra are strongly anisotropic in the (x, y) plane. For instance, in the SDE spectrum for θ = 0◦ the inter-subband peak at ∼ 31 is stronger than the 2D intra-subband resonance at ∼ 16 meV, while the opposite is true at θ = 30◦ Similar, but less pronounced, reshaping is also predicted for the CDE spectrum. Indeed, the IDD calculated at the energy of the most intense CDE peak for θ = 0◦ (Fig. 6(e)) shows mainly a radial density fluctuation in the top and bottom facets, while the most intense peak for θ = 30◦ (Fig. 6(f)) is rather originated in azimuthal density fluctuations. IV. SUMMARY AND CONCLUSIONS CSNWs are quite complex systems from the electronic point of view. On the one hand, due to the tubular shape, the electron gas bridges between a q2D and a q1D system. On the other hand, the prismatic cross-section induces an inhomogeneous distribution of carriers, forming q1D and q2D channels at the corners and facets of the heterostructure, respec- tively. Furthermore, in ILS experiments the photon field inevitably impinges with different directions on the different facets of the embedded coQW at the same time, exciting differ- ent types of excitations simultaneously. Therefore, ILS spectra in CSNWs are considerably more complex than in planar heterostructures. At the same time, ILS cross-section is an extremely informative probe of the nature of the electronic states, particularly if dispersion 16 is measured in appropriate directions, as we have discussed in this paper. To show the potential of ILS experiments in CSNWs, we have predicted the spectra by CDEs and SDEs at experimentally relevant regimes. The real-space 3D multi-subband DFT and TDDFT formalism used here proved of predictive quality in a previous theoretical- experimental study which has been conducted at normal incidence.17 Here, we have gener- alized this approach to predict the dispersion of the ILS resonances with the photon field rotated in different directions. We have shown that the ILS spectra hold information on three different type of collective excitations which form in CSNWs, namely, 2D-like intra- subband, inter-subband and intra-subband excitations. These are associated with density fluctuations in the azimuthal, radial and longitudinal directions, respectively, and can be singled out in ILS from their spectral position and bandwidth, and from their specific dis- persion in properly designed experiments. In particular, i) the 2D intra-subband and inter-subband peaks vanish as the photon field is rotated toward the CSNW axis; ii) in this regime a single long-lived intra-subband plasmon can be observed; iii) the ILS spectra are anisotropic with respect to the azimuthal angle of the photon field, the degree of anisotropy being larger for higher electron densities. ACKNOWLEDGMENTS We acknowledge partial financial support from Universitat Jaume I - Project P1.1B2014- 24, from Generalitat Valenciana Vali+d Grant (MR) APOSTD/2013/052, from European Unions 7th framework programme - Marie Curie ITN INDEX under grant agreement n 289968, and from University of Modena and Reggio emilia, through Grant "Nano- and emerging materials and systems for sustainable technologies". ∗ mroyo@qfa.uji.es 1 M. Cardona and G. Abstreiter, Light scattering in solids V: superlattices and other microstruc- tures, edited by M. Cardona and G. Guntherodt, Topics in applied physics (Springer-Verlag, 1989). 2 A. Pinczuk and G. Abstreiter, Light Scattering in Solids V , edited by M. Cardona and 17 G. Guntherodt, Topics in Applied Physics, Vol. 66/1 (Springer Berlin Heidelberg, 1989) pp. 153 -- 211. 3 C. Schuller, Inelastic light scattering of semiconductor nanostructures: fundamentals and recent advances, 219 (Springer-Verlag, 2006). 4 M. S. Kushwaha, Surface Science Reports 41, 1 (2001). 5 C. Schuller, G. Biese, K. Keller, C. Steinebach, D. Heitmann, P. Grambow, and K. Eberl, Phys. Rev. B 54, R17304 (1996). 6 S. Das Sarma and E. H. Hwang, Phys. Rev. B 54, 1936 (1996). 7 K. Linghua, Y. Baorong, and H. Xiwei, Plasma Science and Technology 9, 519 (2007). 8 F. A. Blum, Phys. Rev. B 1, 1125 (1970). 9 S. Katayama and T. Ando, Journal of the Physical Society of Japan 54, 1615 (1985). 10 P. Hawrylak, J.-W. Wu, and J. J. Quinn, Phys. Rev. B 32, 5169 (1985). 11 S. Das Sarma and D.-W. Wang, Phys. Rev. Lett. 83, 816 (1999). 12 M. S. Kushwaha, AIP Advances 2, 032104 (2012). 13 C. Steinebach, R. Krahne, G. Biese, C. Schuller, D. Heitmann, and K. Eberl, Phys. Rev. B 54, R14281 (1996). 14 M. R. S. Tavares, Phys. Rev. B 71, 155332 (2005). 15 C. P. Garc´ıa, V. Pellegrini, A. Pinczuk, M. Rontani, G. Goldoni, E. Molinari, B. S. Dennis, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 95, 266806 (2005). 16 S. Kalliakos, M. Rontani, V. Pellegrini, C. P. Garc´ıa, A. Pinczuk, G. Goldoni, E. Molinari, L. N. Pfeiffer, and K. W. West, Nature Physics 4, 467 (2008). 17 S. Funk, M. Royo, I. Zardo, D. Rudolph, S. Morkotter, B. Mayer, J. Becker, A. Bechtold, S. Matich, M. Doblinger, M. Bichler, G. Koblmuller, J. J. Finley, A. Bertoni, G. Goldoni, and G. Abstreiter, Nano Letters 13, 6189 (2013). 18 M. F. Lin and K. W. K. Shung, Phys. Rev. B 47, 6617 (1993). 19 O. Sato, Y. Tanaka, M. Kobayashi, and A. Hasegawa, Phys. Rev. B 48, 1947 (1993). 20 L. Wendler and V. G. Grigoryan, Phys. Rev. B 49, 14531 (1994). 21 B. Tanatar, Phys. Rev. B 55, 1361 (1997). 22 Y.-N. Wang and Z. L. Miskovi´c, Phys. Rev. A 66, 042904 (2002). 23 M. Royo, A. Bertoni, and G. Goldoni, Phys. Rev. B 89, 155416 (2014). 24 B. Ketterer, J. Arbiol, and A. Fontcuberta i Morral, Phys. Rev. B 83, 245327 (2011). 18 25 O. Gunnarsson and B. I. Lundqvist, Phys. Rev. B 13, 4274 (1976). 26 A. Bertoni, M. Royo, F. Mahawish, and G. Goldoni, Phys. Rev. B 84, 205323 (2011). 27 A thorough description of the computational approach can be found in Ref. 23. 28 M. S. Kushwaha, AIP Advances 3, 042103 (2013). 29 R. Kubo, Journal of the Physical Society of Japan 12, 570 (1957). 30 G. Ferrari, G. Goldoni, A. Bertoni, G. Cuoghi, and E. Molinari, Nano Letters 9, 1631 (2009). 31 B. M. Wong, F. L´eonard, Q. Li, and G. T. Wang, Nano Letters 11, 3074 (2011). 32 M. Royo, A. Bertoni, and G. Goldoni, Phys. Rev. B 87, 115316 (2013). 33 J. Jadczak, P. Plochocka, A. Mitioglu, I. Breslavetz, M. Royo, A. Bertoni, G. Goldoni, T. Smolenski, P. Kossacki, A. Kretinin, H. Shtrikman, and D. K. Maude, Nano Letters 14, 2807 (2014), pMID: 24745828. 34 M. Royo, C. Segarra, A. Bertoni, G. Goldoni, and J. Planelles, Phys. Rev. B 91, 115440 (2015). 35 G.-Q. Hai, N. Studart, and G. E. Marques, Phys. Rev. B 57, 2276 (1998). 36 R. Sooryakumar, A. Pinczuk, A. Gossard, and W. Wiegmann, Phys. Rev. B 31, 2578 (1985). 37 A. Schmeller, A. R. Goni, A. Pinczuk, J. S. Weiner, J. M. Calleja, B. S. Dennis, L. N. Pfeiffer, and K. W. West, Phys. Rev. B 49, 14778 (1994). 38 We have used a basis set of 130 subbands to obtain converged spectra at this doping density. 19 

1102.4207

1

1102

A Josephson Quantum Electron Pump

A macroscopic fluid pump works according to the law of Newtonian mechanics and transfers a large number of molecules per cycle (of the order of 10^23). By contrast, a nano-scale charge pump can be thought as the ultimate miniaturization of a pump, with its operation being subject to quantum mechanics and with only few electrons or even fractions of electrons transfered per cycle. It generates a direct current in the absence of an applied voltage exploiting the time-dependence of some properties of a nano-scale conductor. The idea of pumping in nanostructures was discussed theoretically a few decades ago [1-4]. So far, nano-scale pumps have been realised only in system exhibiting strong Coulombic effects [5-12], whereas evidence for pumping in the absence of Coulomb-blockade has been elusive. A pioneering experiment by Switkes et al. [13] evidenced the difficulty of modulating in time the properties of an open mesoscopic conductor at cryogenic temperatures without generating undesired bias voltages due to stray capacitances [14,15]. One possible solution to this problem is to use the ac Josephson effect to induce periodically time-dependent Andreev-reflection amplitudes in a hybrid normal-superconducting system [16]. Here we report the experimental detection of charge flow in an unbiased InAs nanowire (NW) embedded in a superconducting quantum interference device (SQUID). In this system, pumping may occur via the cyclic modulation of the phase of the order parameter of different superconducting electrodes. The symmetry of the current with respect to the enclosed magnetic flux [17,18] and bias SQUID current is a discriminating signature of pumping. Currents exceeding 20 pA are measured at 250 mK, and exhibit symmetries compatible with a pumping mechanism in this setup which realizes a Josephson quantum electron pump (JQEP).

cond-mat.mes-hall

cond-mat

A Josephson Quantum Electron Pump F. Giazotto,1, ∗ P. Spathis,1 S. Roddaro,1 S. Biswas,1 F. Taddei,1 M. Governale,2 and L. Sorba1 1NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, Piazza S. Silvestro 12, I-56127 Pisa, Italy 2School of Chemical and Physical Sciences and MacDiarmid Institute for Advanced Materials and Nanotechnology, Victoria University of Wellington, P.O. Box 600, Wellington 6140, New Zealand A macroscopic fluid pump works according to the law of Newtonian mechanics and transfers a large number of molecules per cycle (of the order of 1023). By contrast, a nano-scale charge pump can be thought as the ultimate miniaturization of a pump, with its operation being subject to quantum mechanics and with only few electrons or even fractions of electrons transfered per cycle. It generates a direct current in the absence of an applied voltage exploiting the time-dependence of some properties of a nano-scale conductor. The idea of pumping in nanostructures was discussed theoretically a few decades ago [1–4]. So far, nano-scale pumps have been realised only in sys- tem exhibiting strong Coulombic effects [5–12], whereas evidence for pumping in the absence of Coulomb-blockade has been elusive. A pioneer- ing experiment by Switkes et al. [13] evidenced the difficulty of modulating in time the proper- ties of an open mesoscopic conductor at cryogenic temperatures without generating undesired bias voltages due to stray capacitances [14, 15]. One possible solution to this problem is to use the ac Josephson effect to induce periodically time- dependent Andreev-reflection amplitudes in a hy- brid normal-superconducting system [16]. Here we report the experimental detection of charge flow in an unbiased InAs nanowire (NW) embed- ded in a superconducting quantum interference device (SQUID). In this system, pumping may occur via the cyclic modulation of the phase of the order parameter of different superconducting electrodes. The symmetry of the current with respect to the enclosed magnetic flux [17, 18] and bias SQUID current is a discriminating sig- nature of pumping. Currents exceeding 20 pA are measured at 250 mK, and exhibit symme- tries compatible with a pumping mechanism in this setup which realizes a Josephson quantum electron pump (JQEP). The microscopic mechanism that enables the transport properties of the NW to be affected by the phases of the superconducting order parameter is Andreev reflection [19]. This is the quantum process for which an electron impinging from the normal side onto the interface be- tween a normal metal and a superconductor, is retrore- flected as a hole (i.e., a time-reversed electron) which picks up the phase of the superconducting order param- 1 1 0 2 b e F 1 2 ] l l a h - s e m . t a m - d n o c [ 1 v 7 0 2 4 . 2 0 1 1 : v i X r a FIG. 1. InAs Josephson quantum electron pump. (a) Pseudo-color scanning electron micrograph of a JQEP. An InAs nanowire (NW) is connected to three (cid:39) 250-nm-wide V/Ti superconducting contacts forming two (cid:39) 50-nm-long Josephson weak-links and realizing a superconducting quan- tum interference device (SQUID). Two Au/Ti leads, placed at relative distance of (cid:39) 1.5 µm, are contacted to the ends of the NW to allow current detection. The structure was fab- ricated with electron-beam lithography and evaporation of metals. The normal-state resistance of the SQUID is ∼ 250 Ω whereas that of the Au/NW/Au line is ∼ 3.5 kΩ. (b) Blow- up of the device core showing the two V/InAs/V Josephson junctions as well as the two Au electrodes. (c) Inset: SQUID voltage (VSQU ID) versus current (ISQU ID) characteristics at Φ = 0 and Φ = Φ0/2 (Φ is the applied magnetic flux whereas Φ0 is the flux quantum) showing a maximum critical current of ∼ 235 nA. Φ0 corresponds to a magnetic field of (cid:39) 1.4 Oe applied through an effective loop area of ∼ 14.6 µm2. Main panel: Φ-dependent modulation of the SQUID critical current Ic. Dashed line is the theoretical behavior of a tunnel and resistively-shunted junction SQUID assuming an asymmetry of ∼ 4% between the critical currents of the two weak-links. Data in (c) are taken at T = 250 mK. eter. When two or more superconductors are connected to the NW, multiple Andreev scattering processes can occur between them so that transport through the NW will depend on the differences between the phases of the order parameters [20]. The physical realization of this scheme is shown in Fig. 1a and consists of a heavily-doped InAs semiconducting NW on top of which three fingers of superconducting (S) vanadium (V) are deposited thus implementing a SQUID [21]. Two Au normal-metal electrodes (N) are coupled to the ends of the NW to allow detection of the current Iwire flowing through the wire. A close-up of the device core is shown in Fig. 1b. Time-dependence, and possi- bly pumping, arises from biasing the loop with a current ISQU ID larger than the critical current Ic of the SQUID so that the phase differences ϕ1(t) and ϕ2(t) across the two Josephson junctions cycle in time at the Josephson frequency νJ = VSQU ID/Φ0, where VSQU ID is the volt- age developed across the SQUID and Φ0 (cid:39) 2× 10−15 Wb In addition, ϕ1(t) and ϕ2(t) can is the flux quantum. be shifted by a constant term δϕ = 2πΦ/Φ0 originating from an applied magnetic flux Φ threading the loop. This scheme has the advantage that no high-frequency signal needs to be brought to the sample thus simplifying the setup and minimizing the impact of stray capacitancies: the time-dependent signal is self-generated thanks to the ac Josephson effect. Below the critical temperature of the superconductors (Tc (cid:39) 4.65 K) a Josephson current flows through the SQUID across the NW. The SQUID voltage-current char- acteristics at 250 mK is shown in the inset of Fig. 1c for two representative values of Φ. Whereas for Φ = 0 the characteristic shows a clear dissipationless regime with a critical current Ic (cid:39) 235 nA, for Φ = Φ0/2 it behaves almost linearly with Ic largely suppressed. The full Ic(Φ) dependence (main panel of Fig. 1c) shows the character- istic pattern of a superconducting interferometer. The theoretical curve of a conventional (i.e., described by the RSJ model) SQUID [22] is shown for a comparison (dashed line, see Supplementary Information). Figure 2a shows a sketch of the pumping measurement setup. A dc current ISQU ID is fed through the SQUID terminals while the voltage drop VSQU ID is measured against Φ. The N electrodes are grounded and Iwire is sensed with an amperometer. The N and S parts of the circuit have no common ground therefore preventing any direct net charge transfer from the SQUID to the NW. In the following we will concentrate our attention on the symmetries in Φ and ISQU ID displayed by the measured signal, as these are of crucial impor- tance for the interpretation of the experiment. The low-temperature SQUID flux-to-voltage transfer function VSQU ID = ∂VSQU ID/∂Φ versus Φ and ISQU ID is dis- played in Fig. 2b. In particular, VSQU ID is a Φ0-periodic function of Φ and is antisymmetric in Φ and ISQU ID. By contrast, the flux-to-current transfer function of the NW, Iwire = ∂Iwire/∂Φ (Fig. 2c), besides exhibiting the same Φ0-periodicity shows a drastically different behav- ior, being almost symmetric either in Φ or in ISQU ID. A similar behavior with the same symmetries of Iwire is displayed by the NW flux-to-voltage transfer function, 2 Experiment setup and transfer functions FIG. 2. characteristics. (a) Schematic drawing of the JQEP setup. A dc current ISQU ID is fed into the SQUID termi- nals through a floating source while the voltage drop VSQU ID is recorded against the applied magnetic flux Φ threading the ring. When ISQU ID exceeds the SQUID critical super- current the ac Josephson effect sets up inducing a current Iwire which flows in the NW. Iwire is sensed through an amperometer. S and N denote superconductors and normal metals, respectively. (b) Color plot of the SQUID flux-to- voltage transfer function VSQU ID = ∂VSQU ID/∂Φ versus Φ and ISQU ID. VSQU ID is antisymmetric in Φ and ISQU ID. (c) Color plot of the NW flux-to-current transfer function Iwire = ∂Iwire/∂Φ versus Φ and ISQU ID. (d) Color plot of the NW flux-to-voltage transfer function Vwire = ∂Vwire/∂Φ versus Φ and ISQU ID. Data are taken with a voltmeter in an open-circuit configuration, i.e., without allowing Iwire to flow. Note the markedly different behavior displayed by Iwire and Vwire which are almost symmetric in Φ as well as in ISQU ID. All measurements are taken at T = 250 mK with low-frequency phase-sensitive technique to get higher sensi- tivity and reduced noise. Vwire = ∂Vwire/∂Φ (Fig. 2d), where Vwire is measured with open NW contacts. Iwire and Vwire result from different but complementary measurements, and the ev- idence of such a similarity suggests that both reflect the same physical mechanism (see Supplementary Informa- tion). As we shall argue, the nature of the symmetries displayed by Iwire and Vwire is compatible with a quan- tum pumping mechanisms. In general, the pumped current is not expected to show definite parity with Φ [17, 18], therefore Iwire can have a flux-symmetric component as well. This, however, could be ascribed also to other mechanisms than pumping. In 3 addition, Iwire is even not expected to possess any def- inite parity with ISQU ID. In order to extract a pure pumped current contribution from the whole measured signal we focus on the component of Iwire which is an- tisymmetric in Φ, I A wire, as it is predicted to be a fin- gerprint of quantum pumping in the JQEP [16]. Af- ter Φ-integration of Iwire, I A wire is therefore obtained as wire = [Iwire(Φ, ISQU ID) − Iwire(−Φ, ISQU ID)]/2. The I A result of this procedure is shown in Fig. 3a which dis- plays I A wire versus Φ and ISQU ID at 250 mK. The Φ0 pe- riodicity joined with the antisymmetry imply that I A wire vanishes at Φ = Φ0/2, while its sign and magnitude can be changed by varying Φ. Notably, I A wire is almost symmetric in ISQU ID. The theoretical I A wire calculated for the JQEP geometry through a dynamical scattering approach [23–25] assuming for the NW multiple inde- pendent modes is shown in Fig. 3b (see Supplementary Information). Although rather idealized, the model is an essential tool to predict the pumped current symme- tries of the JQEP. Remarkably, summing over many NW modes yields I A wire which is almost symmetric in ISQU ID, in agreement with the experiment. Figure 3c shows I A wire versus Φ and ISQU ID over a wider range of SQUID currents. Specifically, I A wire turns out to be a non-monotonic function of ISQU ID, initially increasing then being suppressed for large ISQU ID. This is emphasized in Fig. 3d where I A wire(Φ) is plotted for selected values of ISQU ID. I A wire is a sinusoidal-like func- tion of Φ whose amplitude depends on ISQU ID, and is maximized at Φ ∼ (1/4)Φ0 and Φ ∼ (3/4)Φ0. The full I A wire(VSQU ID) dependence for a few values of flux is displayed in Fig. 3e and highlights both the monotonic linear increase for low VSQU ID and suppres- sion at large VSQU ID. The symmetry in VSQU ID (i.e., in ISQU ID) is emphasized as well. Furthermore, I A wire is maximized at V max SQU ID ≈ 0.4 mV independently of Φ, where it reaches values exceeding 20 pA. By convert- ing V max SQU ID in terms of the Josephson frequency we get νJ (cid:39) 190 GHz whose corresponding time, ν−1 J ∼ 5 ps, is comparable to τD = W 2/D (cid:39) 4 ps, i.e., the time re- quired by electrons to diffuse in the NW between the Josephson junctions. In the above expression W (cid:39) 250 nm is the width of the SQUID central electrode (Fig. 1b) which we assume to coincide with the separation between the weak-links, whereas D (cid:39) 0.015 m2/s is the diffusion coefficient of the NW [26]. The transition between the regime of I A wire enhancement as a function of VSQU ID to the one of I A wire suppression can be explained in terms of the ability of the electrons to follow adiabatically the time-dependent parameters up to a maximum frequency set by τ−1 D . Another possible contribution to the suppres- sion observed at larger VSQU ID might stem from weaken- ing of the ac Josephson coupling at high applied current [27]. The I A bias range is displayed in Fig. 3f. wire(VSQU ID) dependence plotted over a reduced wire In particular, I A c FIG. 3. Flux and ISQU ID dependence of the antisym- metric part of current flowing in the NW. (a) Color plot of I A wire versus ISQU ID and Φ. (b) Color plot of the the- oretical zero-temperature I A wire versus ISQU ID and Φ. The calculation was performed for the JQEP geometry assuming the same asymmetry between the Josephson junctions as in the experiment. I max is the sum of the critical currents of the two Josephson junctions, R is the total shunting SQUID resistance, and RK (cid:39) 25.8 kΩ is the Klitzing resistance (see Supplementary Information for further details). (c) Color plot of I A wire versus ISQU ID and Φ shown over a wider range of ISQU ID. (d) I A wire versus Φ for a few representative values of ISQU ID. The latter are indicated as dashed lines of the same color in panel (c). (e) I A wire versus VSQU ID for a few selected values of Φ. (f) I A wire versus VSQU ID plotted over a smaller range of VSQU ID for the same Φ values as in panel (e). The slope in the linear regime, expressed in pA/GHz, is de- noted with η. In (e) and (f) the error bars represent the stan- dard deviation of the current values calculated over several measurements, and the upper horizontal scale is expressed in terms of the Josephson frequency νJ . All measurements are taken at T = 250 mK. shows a linear behavior with slope η which depends on the applied flux, and obtains values as high as several 10−1 pA/GHz. In the so-called 'adiabatic regime', i.e., where pumped current is expected to vary linearly with frequency, η would therefore correspond to some 10−3 electrons per pump cycle. 4 always dominated by a component symmetric in Φ and antisymmetric in ISQU ID; (2) any thermocurrent gener- ated by a different power dissipated in the two junctions, which is expected to be predominantly symmetric in both Φ and ISQU ID. We finally note that other normal conductors than InAs NWs could be used for the implementation of the JQEP. This might pave the way to the investigation of the interplay between superconductivity-induced quan- tum pumping and exotic electronic states existing, for instance, in graphene [28] or in carbon nanotubes [29]. We gratefully acknowledge L. Faoro, R. Fazio, L. B. Ioffe, J. Konig, J. P. Pekola, V. Piazza, H. Pothier, and S. Russo for fruitful discussions, and D. Ercolani for providing the InAs nanowires. The work was partially supported by the NanoSciERA project "NanoFridge". F.T acknowledges financial support from EU through the projects "SOLID" and "GEOMDISS". METHODS SUMMARY Selenium doped InAs NWs were grown by chemical beam epitaxy on an InAs 111B substrate. Gold catalyst particles were formed by thermal dewetting (at 520◦ C for 20 min) of a 0.5-nm-thick Au film under TBA flux. NWs were grown for two hours at 420◦ C using TBA, TMI and DTBSe metallorganic precursors with line pressures of 2.0 Torr, 0.3 Torr, and 0.4 Torr, respectively. NWs have diameters of 90±10 nm and are around 2.5 µm long. Transport parameters were estimated over an ensemble of nominally identical 1 µm-long NW field effect transis- tors using a charge control model [30] and a numerical evaluation of the gate capacitance. Carrier density was estimated to be n = 1.8 ± 0.8 × 1019 cm−3 and elec- tron mobility µ = 300 ± 100cm2/Vs. The devices were fabricated using a technique of dry cleavage of the NWs onto Si/SiO2 substrates (500 nm oxide on intrinsic Si). Contacts were obtained by a two-step aligned process: thermal evaporation of Ti/Au (10/80 nm) was performed first and followed by electron-beam deposition of Ti/V (15/120 nm) in an UHV chamber [21]. InAs NWs were treated with a NH4Sx solution before each evaporation step to get transparent metal-NW contacts [26]. The magneto-electric characterization of the devices was performed in a filtered 3He refrigerator (two-stage RC- and π-filters) down to ∼ 250 mK using a standard 4-wire technique. Current injection at the SQUID ter- minals was obtained by using a battery-powered float- ing source, whereas voltage and current were measured by room-temperature preamplifiers. Derivative measure- ments (flux-to-voltage as well as flux-to-current transfer functions) were performed with standard low-frequency lock-in technique by superimposing a small modulation to the applied magnetic field. FIG. 4. Temperature dependence of the antisymmet- ric part of the current flowing in the NW. (a) I A wire versus ISQU ID measured at several bath temperatures T . (b) I A wire versus T at selected bias currents ISQU ID. Note the sat- uration of I A wire at low temperature as well as its suppression at high T . The error bars represent the standard deviation of the current values calculated over several measurements. Dashed lines in both panels (a) and (b) are guides to the eye, and all measurements are taken for Φ = (3/4)Φ0. The role of temperature (T ) is shown in Fig. 4a which displays I A wire versus VSQU ID at Φ = (3/4)Φ0 for several increasing temperatures. I A wire monotonically decreases upon increasing T , which can be ascribed to the influence of thermal smearing as well as thermal-induced dephas- ing, and is suppressed for T (cid:38) 3.5 K. We stress that the aforementioned temperature is substantially smaller than Tc, the latter setting the disappearance of both Josephson effect and superconductivity in the JQEP. The I A wire(T ) dependence at the same flux is shown in Fig. 4(b) for a few ISQU ID values. Specifically, I A wire begins to round off at lower temperatures indicating a saturation, whereas it is damped at higher T . Low-temperature behavior sug- gests that current tends to saturate upon reducing tem- perature when the "effective" separation between Joseph- son junctions becomes of the same order of the electron coherence length in the NW, LT =(cid:112)D/(2πkBT ) ∼ 270 nm at 250 mK, where  is the reduced Planck's constant while kB is the Boltzmann's constant. By contrast, the decay of LT at higher temperatures may be considered as one of the predominant decoherence mechanisms leading to I A wire suppression. Further study is needed to clarify this point. It is worthwhile to emphasize that other effects which might manifest in the JQEP would yield currents charac- terized by symmetries markedly different from the ones predicted for quantum pumping (see Supplementary In- formation). Among these we recall (1) any spurious cur- rent due to asymmetry between the junctions which is 0123401020I Awire (pA)VSQUID (mV) 0.25 1.26 1.5 2.1 2.8 3.4T (K)0.00.61.21.8νJ (THz)0123401020 I Awire (pA) 0.25 1.5 2.5 5 10 T (K)ISQUID (µA)ab(cid:27) . 5 (2) − ∂[S(cid:63) he]i,j ∂x1 ∂[She]i,j ∂x2 In Eq. (2), See and She are, respectively, the normal and the Andreev scattering matrices between the two N leads evaluated at the Fermi energy. Assuming that all leads support a single propagating channel, [See]i,i ([She]i,i) is the amplitude for an electron entering from lead i = 1, 2 to be reflected back as an electron (hole), while [See]j,i ([She]j,i), with j (cid:54)= i, is the transmission amplitude for an electron entering from lead i and exiting through lead j as an electron (hole). See and She can be determined by the scheme proposed in Ref. [25], which requires the calculation of the scattering matrix S of the system de- picted in Fig. 5 when all contacts are in the normal state. S, in turns, is computed as a composition of three (three- legged) beam splitters, indicated by dashed circles in Fig. 5 and labelled by the index λ = {t, m, b}, connected to each other through a pair of ballistic N wires of dif- ferent lengths. The scattering matrix of beam splitter λ can be written as 1 − 2γλ eiψλ √ √ γλ γλ γλ √ (cid:113) 1−γλ (cid:113) 1−γλ 2 2 eiαλ eiβλ γλ √ (cid:113) 1−γλ (cid:113) 1−γλ 2 2 eiβλ eiαλ  ,  −√ Sλ = (cid:113) (3) where γλ takes values between 0 and 1/2, αλ = −ψλ + qλ arccos[−γλ/(1 − γλ)] and βλ = −ψλ − qλ arccos[−γλ/(1− γλ)] with qλ = ±1. The three S leads, described by constant pair potentials ∆i = ∆exp(iφi) (with i = 3, 5), are assumed to be ideally coupled to the structure so that perfect Andreev reflection occurs at the S interfaces. When the bias current ISQU ID is larger than the critical current of the SQUID, a voltage VSQU ID de- velops across the latter. For the SQUID we assume the RSJ voltage-current relation [22] VSQU ID(δϕ) = sign(ISQU ID)R SQU ID − Ic(δϕ)2, I 2 (4) where R is the total shunting SQUID resistance and Ic(δϕ) is the flux-dependent SQUID critical current. The latter, used to fit the data in Fig. 1c, can be written as Ic(δϕ) = (Ic1 + Ic2)(cid:112)r2 + (1 − r2) cos2(δϕ/2), (5) where Ic1 and Ic2 are the critical currents of the indi- vidual Josephson junctions composing the SQUID, and r = (Ic1 − Ic2)/(Ic1 + Ic2) is the degree of asymmetry of the SQUID. From a practical point of view, we first calculate Q1 and Q2 through Eq. (1) assuming that the N leads 1 and 2 and lead 5 are grounded, while S leads 3 and 4 are kept at the potential VSQU ID. This choice sets the phases of the superconductors as follows: φ3 = sign(ISQU ID)ωJ t (6) FIG. 5. Sketch of the Josephson quantum electron pump. The system is modeled as three, three-legged beam splitters (denoted by dashed circles) labelled t, m and b, and connected by two ballistic normal-metal (N) wires. Electrodes 1 and 2 are normal-metallic and grounded, while electrodes 3, 4 and 5 are superconducting (S) and arranged to form a SQUID thread by a magnetic flux Φ. A dc current ISQU ID is fed into the SQUID terminals through a floating source determining a voltage drop VSQU ID. Qp represents the charge pumped per cycle in the N electrodes. SUPPLEMENTARY INFORMATION Theoretical model Let us consider the system de- picted in Fig. 5, where a NW is connected to two normal- metal (N) leads, labelled by 1 and 2, and to three super- conducting (S) leads, which implement a SQUID, labelled by 3, 4, and 5. If the SQUID is polarized by a current ISQU ID larger than its critical current, the ac Josephson effect sets in introducing a time-dependence in the scat- tering amplitudes through the NW and enabling pump- ing to occur. A magnetic flux Φ threads the SQUID in- troducing a phase shift δϕ = 2πΦ/Φ0, where Φ0 = π/e is the flux quantum, e is the electron charge and  the reduced Planck's constant. In the adiabatic regime, the charge pumped per cycle in one of the N leads Qi (with i = 1, 2) can be calculated in the scattering approach through a generalization of the Brouwer's formula [3] to hybrid systems [23–25]. If x1(t) and x2(t) are the two pumping parameters varying along a closed path in the (x1, x2)-space, at zero temperature one finds that dx1dx2 Πi,j(x1, x2), (1) Ω j=1 where Ω is the area enclosed by the path in parameter space, and Πi,j(x1, x2) = (cid:61) ee]i,j ∂x1 ∂[See]i,j ∂x2 − (cid:90) Qi = e π 2(cid:88) (cid:26) ∂[S(cid:63) (cid:19)(cid:21) (7) (cid:18) δϕ 2 (cid:20) 1 φ4 = sign(ISQU ID)ωJ t − δϕ φ5 = − δϕ 2 π 2 r ) − π 2 − arctan δϕ 2 cot + sign(sin sign(ISQU ID), (8) where ωJ = 2πνJ = 2πVSQU ID/Φ0 is the Josephson angular frequency and the function arctan takes val- ues between −π/2 and π/2. The value of φ5 is cho- sen to ensure that the supercurrent is maximized in the limit of ωJ → 0. As a consequence of this, all observ- able quantities exhibit the standard Φ0 periodicity. The pumping parameters are defined as x1(t) = cos(ωJ t) and x2(t) = sin(ωJ t) so that exp(iφ3) = x1 + ix2, exp(iφ4) = (x1 +ix2)exp(−iδϕ). From this choice is clear that the two parameters are maximally out of phase, in- dependently of δϕ, and that the path is a circle of radius one centered around the origin. Since the N and S parts of the circuit have no common ground, the actual chem- ical potentials of the S electrodes with respect to the potential of the N electrodes have to arrange themselves so that no net current flows between the two parts of the circuit. As a consequence, the charge pumped per cycle can be written as Qp(δϕ) = Q1(δϕ)G2(δϕ) − Q2(δϕ)G1(δϕ) G1(δϕ) + G2(δϕ) , (9) where Gi = [She]i,12 + [She]i,22 is the conductance (in units of e2/π) relative to lead i. Note that, in general, Qp(δϕ) has no definite parity in δϕ, in agree- ment with the results of Refs. [17, 18], and no def- inite parity in ISQU ID. The δϕ-antisymmetric com- ponent of the pumped current is obtained as I A wire = (ωJ /2π) [Qp(δϕ)− Qp(−δϕ)]/2. In Fig. 3b I A wire is plot- ted in units of I max = Ic1 + Ic2 and RK = 2π/e2 is the Klitzing resistance. The current has been computed assuming that the NW carries 50 inde- pendent channels, each of which described by a scattering matrix obtained taking ψλ, qλ and the phases accumu- lated along the two N wires as random parameters, while setting γt = 1/10, γm = 1/11 and γb = 1/13. c R/RK, where I max c In the configuration where lead 1 is a voltage probe (rather than connected to ground) one can calculate the voltage Vp which develops at lead 1 as a consequence of the charge pumped. Vp, determined by setting to zero the current flowing in the NW, can be written as Vp(δϕ) = VSQU ID(δϕ) G1(δϕ) + G2(δϕ) G1(δϕ)G2(δϕ) Qp(δϕ). (10) in general, no definite parity in δϕ and The Gi has, ISQU ID. Furthermore, G1(δϕ)+G2(δϕ) is approximately G1(δϕ)G2(δϕ) even in both quantities also in the presence of a small asymmetry between the two Josephson junctions (which is typically the case of any realistic situation), so that Vp and Qp show the same parity both in δϕ and ISQU ID. 6 The flux-antisymmetric component of Vp is defined as wire(δϕ) = [Vp(δϕ) − Vp(−δϕ)]/2. V A We shall further discuss the spurious effects which can occur in the presence of a shunting dissipative current across the Josephson weak-links. If the two Josephson junctions are not equal, a spurious voltage Vs (contain- ing a constant and a time-oscillating component) arises in the NW between the beam splitters t and b in Fig. 5. This produces a current Is in the NW that is not origi- nated by quantum pumping. On the one hand, the cur- rent Is,const related to the constant component of Vs re- verses by changing the sign of ISQU ID, in contrast to I A wire, and it is an even function of δϕ. On the other hand, it turns out that the quantum rectified current Is,rect associated to the oscillating component of Vs has no definite parity both in δϕ and ISQU ID, similarly to Qp of Eq. (9). However, Is,rect exists only in the pres- ence of a finite Is,const, since they have the same physical origin. Yet, Is,rect is smaller than Is,const because the amplitude of the oscillating components of Vs is set by I max , whereas the constant component of Vs is propor- c tional to VSQU ID. Therefore, the total spurious current is dominated by the component that is even in flux and odd in ISQU ID which would be detected, if present, in the transfer function Iwire. Since the measured deriva- tive signal Iwire is almost flux-symmetric [see Fig. 2(c)], we can rule out the presence of Is,const and therefore of quantum rectification. We stress that even in the pres- ence of a sizable Is,const, our calculations predict I A wire to be typically several orders of magnitude larger than the flux-antisymmetric component of Is,rect (which is even in ISQU ID) thus fully dominating the measured signal. In analogy, the current ISQU ID might produce a differ- ent power dissipated between points t and b in Fig. 5 lead- ing to a thermocurrent flowing through the NW. Since Vs is dominated by its constant component, this thermocur- rent would be almost symmetric both in δϕ and ISQU ID, in contrast to I A wire. In addition, there could be a small contribution to the thermocurrent due to the oscillating component of Vs which would have no definite parity both in ISQU ID and δϕ. Since the power dissipated is propor- tional to V 2 s such contribution to the thermocurrent is a fortiori negligible. In conclusions, all the mechanisms envisioned above to produce a spurious dc current can be distinguished from quantum pumping by their parity with respect to magnetic flux Φ or bias current ISQU ID. Supplementary data Here we present additional data for another JQEP device with nominally-identical geometry. Its essential parameters are the SQUID normal-state resistance of ∼ 187 Ω and the resistance of the Au/NW/Au line of ∼ 2.1 kΩ. The general behavior of this device is similar to that discussed in the main text although it is characterized by less symmetry between the two Josephson junctions. Figure 6 (a) displays the 7 wire is maximized at V max full I A wire(VSQU ID) dependence for a few selected values of Φ at 250 mK is displayed in Fig. 6d, and emphasizes the overall symmetry in VSQU ID. For the present de- vice I A SQU ID ≈ 0.25 mV where it obtains values exceeding ∼ 27 pA. V max SQU ID corre- sponds to a Josephson frequency νJ (cid:39) 120 GHz (and J ∼ 8 ps). This difference from the de- related time ν−1 vice presented in the main text could originate from a slightly larger width W of the SQUID central electrode combined with a reduced NW diffusion constant which lead to an increased diffusion time τD. The I A wire(T ) de- pendence at Φ = (3/4)Φ0 is shown in Fig. 6e for a few selected ISQU ID currents. Specifically, I A wire is rounded off at low temperature, whereas it is strongly damped and suppressed for T (cid:38) 3 K. The general behavior of I A wire and the arguments of the previous section there- fore suggest that in this sample I A wire is fully dominated by quantum pumping, although a small component of quantum rectification might perhaps be present as well. ∗ f.giazotto@sns.it [1] D. J. Thouless, Phys. Rev. B 27, 6083 (1983). [2] M. Buttiker, H. Thomas, and A. Pretre, Z. Phys. B 94, 133 (1994). [3] P. W. Brouwer, Phys. Rev. B 58, R10135 (1998). [4] F. Zhou, B. Spivak, and B. Altshuler, Phys. Rev. Lett. 82, 608 (1999). [5] H. Pothier, P. Lafarge, C. Urbina, D. Esteve, and M. H. Devoret, Europhys. Lett. 17, 249 (1992). [6] J. M. Martinis, M. Nahum, and H. D. Jensen, Phys. Rev. Lett. 72, 904 (1994). [7] N. E. Fletcher, J. Ebbecke, T. J. B. M. Janssen, F. J. Ahlers, M. Pepper, H. E. Beere, and D. A. Ritchie, Phys. Rev. B 68, 245310 (2003). [8] J. Ebbecke N. E. Fletcher, T. J. B. M. Janssen, F. J. Ahlers, M. Pepper, H. E. Beere, and D. A. Ritchie, Appl. Phys. Lett. 84, 4319 (2004). [9] J. P. Pekola, J. J. Vartiainen, M. Mottonen, O.-P. Saira, M. Meschke, and D. V. Averin, Nature Phys. 4, 120 (2008). [10] A. Fuhrer, C. Fasth, and L. Samuelson, Appl. Phys. Lett. 91, 052109 (2007). [11] M. R. Buitelaar, V. Kashcheyevs, P. J. Leek, V. I. Talyan- skii, C. G. Smith, D. Anderson, G. A. C. Jones, J. Wei, and D. H. Cobden, Phys. Rev. Lett. 101, 126803 (2008). [12] B. Kaestner, V. Kashcheyevs, G. Hein, K. Pierz, U. Sieg- ner, and H. W. Schumacher, Appl. Phys. Lett. 92, 192106 (2008). [13] M. Switkes, C. M. Marcus, K. Campman, and A. C. Gos- sard, Science 283, 1905 (1999). [14] P. W. Brouwer, Phys. Rev. B 63, 121303(R) (2001). [15] L. DiCarlo, C. M. Marcus, and J. S. Harris, Jr., Phys. Rev. Lett. 91, 246804 (2003). [16] S. Russo, J. Tobiska, T. M. Klapwijk, and A. F. Mor- purgo, Phys. Rev. Lett. 99, 086601 (2007). [17] T. A. Shutenko, B. L. Altshuler, and I. L. Aleiner, Phys. Rev. B 61, 10 366 (2000). FIG. 6. Experimental data for a different JQEP. (a) Φ-dependent modulation of the SQUID critical current Ic. Dashed line is the theoretical behavior of a tunnel and resistively-shunted junction SQUID assuming an asymmetry r ∼ 9% between the critical currents of the two weak-links. (b) Color plot of the NW flux-to-current transfer function Iwire = ∂Iwire/∂Φ versus Φ and ISQU ID. (c) Color plot of I A wire versus ISQU ID and Φ. (d) I A wire versus VSQU ID for a few selected values of Φ. Data in (a)-(d) are taken at T = 250 mK. (e) I A wire versus temperature T at selected bias currents ISQU ID for Φ = (3/4)Φ0. The error bars represent the stan- dard deviation of the current values calculated over several measurements, and dashed lines are guides to the eye. full Ic(Φ) dependence of the SQUID measured at 250 mK which shows a maximum critical current of ∼ 330 nA. Su- perimposed for a comparison (dashed line) is the model for a tunnel and resistively-shunted junction SQUID [22] assuming an asymmetry r ∼ 9% between the critical cur- rents of the two weak-links. The low-temperature flux- to-current transfer function Iwire = ∂Iwire/∂Φ versus Φ and ISQU ID is shown in Fig. 6b. Iwire shows no defi- nite parity both in Φ and ISQU ID which stems from the presence of a spurious current Is in the NW, and might be attributed to the reduced symmetry of the SQUID junctions. Figure 6c shows the extracted I A wire versus Φ and ISQU ID at 250 mK which highlights both the non- monotonic dependence and symmetry in ISQU ID. The [18] M. Moskalets and M. Buttiker, Phys. Rev. B 72, 035324 [25] F. Taddei, M. Governale, and R. Fazio, Phys. Rev. B 70, (2005). 052510 (2004). [19] A. F. Andreev, Sov. Phys. JETP 19, 1228 (1964). [20] F. Giazotto, J. T. Peltonen, M. Meschke, and J. P. [26] S. Roddaro, A. Pescaglini, D. Ercolani, L. Sorba, F. Gi- azotto, and F. Beltram, Nano Res. (2011) (in print). Pekola, Nature Phys. 6, 254 (2010). [21] P. Spathis, S. Biswas, S. Roddaro, L. Sorba, F. Giazotto, and F. Beltram, Nanotechnology 22, 105201 (2011). [22] M. Tinkham, Introduction to Superconductivity, 2nd Edi- tion (McGraw-Hill, Inc., New York, 1996). [23] J. Wang, Y. Wei, B. Wang, and H. Guo, Appl. Phys. [27] R. E. Harris, Phys. Rev. B 10, 84 (1974). [28] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [29] Y. Wei and J. Wang, Phys. Rev. B 66, 195419 (2002). [30] X. Jiang, Q. Xiong, S. Nam, F. Qian, Y. Li, and C. M. Lett. 79, 3977 (2001). [24] M. Blaauboer, Phys. Rev. B 65, 235318 (2002). Lieber, Nano Lett. 7, 3214 (2007). 8 

1306.3135

1

1306

Electrical control of phonon mediated spin relaxation rate in semiconductor quantum dots: the Rashba vs the Dresselhaus spin-orbit couplings

In symmetric quantum dots (QDs), it is well known that the spin-hot spot (i.e., the cusp-like structure due to the presence of degeneracy near the level or anticrossing point) is present for the pure Rashba case but is absent for the pure Dresselhaus case [Phys. Rev. Lett. 95, 076805 (2005)]. Since the Dresselhaus spin-orbit coupling dominates over the Rashba spin-orbit coupling in GaAs and GaSb QDs, it is important to find the exact location of the spin-hot spot or the cusp-like structure even for the pure Dresselhaus case. In this paper, for the first time, we present analytical and numerical results that show that the spin-hot spot can also be seen for the pure Dresselhaus spin-orbit coupling case by inducing large anisotropy through external gates. At or nearby the spin-hot spot, the spin transition rate enhances and the decoherence time reduces by several orders of magnitude compared to the case with no spin-hot spot. Thus one should avoid such locations when designing QD spin based transistors for the possible implementation in quantum logic gates, solid state quantum computing and quantum information processing. It is also possible to extract the exact experimental data (Phys. Rev. Lett. 100, 046803 (2008)) for the phonon mediated spin-flip rates from our developed theoretical model.

cond-mat.mes-hall

cond-mat

Electrical control of phonon mediated spin relaxation rate in semiconductor quantum dots: the Rashba vs the Dresselhaus spin-orbit couplings Sanjay Prabhakar,1 Roderick Melnik1,2 and Luis L. Bonilla2 1M 2NeT Laboratory, Wilfrid Laurier University, Waterloo, ON, N2L 3C5 Canada 2Gregorio Millan Institute, Universidad Carlos III de Madrid, 28911, Leganes, Spain (Dated: May 15, 2013) In symmetric quantum dots (QDs), it is well known that the spin-hot spot (i.e., the cusp-like structure due to the presence of degeneracy near the level or anticrossing point) is present for the pure Rashba case but is absent for the pure Dresselhaus case [Phys. Rev. Lett. 95, 076805 (2005)]. Since the Dresselhaus spin-orbit coupling dominates over the Rashba spin-orbit coupling in GaAs and GaSb QDs, it is important to find the exact location of the spin-hot spot or the cusp-like structure even for the pure Dresselhaus case. In this paper, for the first time, we present analytical and numerical results that show that the spin-hot spot can also be seen for the pure Dresselhaus spin-orbit coupling case by inducing large anisotropy through external gates. At or nearby the spin- hot spot, the spin transition rate enhances and the decoherence time reduces by several orders of magnitude compared to the case with no spin-hot spot. Thus one should avoid such locations when designing QD spin based transistors for the possible implementation in quantum logic gates, solid state quantum computing and quantum information processing. It is also possible to extract the exact experimental data (Phys. Rev. Lett. 100, 046803 (2008)) for the phonon mediated spin-flip rates from our developed theoretical model. I. INTRODUCTION Manipulation of a single electron spin with the ap- plication of gate controlled electric fields in a confined semiconductor QDs is a promising way for developing spin based quantum logic gates, spin memory devices for various quantum information processing applications.1 -- 12 Sufficiently short gate operation time combined with long decoherence time is one of the requirements for quan- tum computing.1,13,14 When a qubit is operated on by a classical bit, then its decay time is given by a spin re- laxation time which is also supposed to be longer than the minimum time required to execute one quantum gate operation.2,13,15 -- 17 Long spin relaxations have been mea- sured experimentally in both symmetric and asymmet- ric QDs.4,5,15 Balocchi et. al.18 have recently measured larger spin relaxation times (30 ns) in GaAs QDs. More specifically, both isotropic and anisotropic spin relax- ations can be tuned with spin orbit coupling by choosing the growth direction parallel to the crystallographic axis [001], [110] and [111] of III-V zinc blend semiconductor QDs.6,18 -- 20 In addition to the lengthening spin coherence time, the electric field tuning of spin relaxation forms the basis for turning the spin current on and off in some spin transistor proposals that can help to initialize elec- tron spin based quantum computers.16,21 These experi- mental studies confirm that the manipulation of spin-flip rates mediated by phonons due to spin-orbit coupling is an important ingredient for the design of robust spin- tronics logic devices. The spin-orbit coupling is mainly dominated by the Rashba22 and the linear Dresselhaus23 terms in III-V semiconductor QDs.17,24 -- 33 The Rashba spin-orbit coupling arises from structural inversion asym- metry along the growth direction while the Dresselhaus spin-orbit coupling arises from the bulk inversion asym- metry of the crystal lattice.22,23 In Ref. 34 and 35, the authors report that the cusp-like structure in the phonon mediated spin transition rate can be seen for the pure Rashba case. For the pure Dressel- haus case, the spin transition rate is a monotonous func- tion of the magnetic fields and QDs radii. Since the Dres- selhaus spin-orbit coupling dominates over the Rashba spin-orbit coupling in some materials such as GaAs and GaSb QDs,25 it is important to find the exact location of spin-hot spot or the cusp-like structure even for the pure Dresselhaus case. The cusp like structure implies shorter spin relaxation and decoherence time which is hazardous for spin based applications such as quantum logic gates, solid state quantum computing and quantum informa- tion processing. For these applications, the spin-hot spot in the phonon mediated spin relaxation rate is something to avoid during the design of QD spin based transistors. Very recently, the authors in Ref. 36 measured the spin- hot spot in the phonon mediated spin relaxation rate in Silicon QDs with the application of tuning very weak spin orbit coupling when Zeeman energy and valley splittings induce degeneracy. At the spin-hot spot in Silicon QDs, the dramatic rate enhancement decreases the decoher- ence time which is not supposed to be the ideal location for the qubit operation.37 -- 40 In this paper, we obtain new analytical and numerical results for the behavior of the spin relaxation rate in anisotropic III-V semiconductor QDs. For the first time, we show the spin-hot spot in the phonon mediated spin transition rate can be seen for the pure Dresselhaus case by creating large anisotropy through external gates. Note that such location (spin-hot spot) is hazardous for quantum computing and quantum information processing, and must therefore be avoided during the design of spin based transistors. The paper is organized as follows. In section II, we de- velop a theoretical model for anisotropic spin relaxation mediated by piezo-phonons that will allow us to inves-  GaAs GaSb InAs InSb R/ D=1 30 40 50 20 10 Electric Field, E [104 V/cm] 60 70 80 90 100 10 D R 1 0.1 FIG. 1. (Color online) Interplay between Rashba-Dresselhaus spin-orbit coupling vs the applied electric field along the z- direction. The Rashba spin-orbit coupling is seen to dominate in InAs and InSb QDs whereas the Dresselhaus spin-orbit coupling is seen to dominate in GaAs and GaSb QDs. / ] s 1 [ e t a r p i l f - n p S i Asymmetric QDs 103 102 Symmetric QDs Theory (obtained from Eq.18) Experiment (see Ref.5, Fig3(b)) 1.2 1.6 2.0 2.4 2.8 Ey (meV) II. THEORETICAL MODEL 2 We consider 2D anisotropic semiconductor QDs in the presence of a magnetic field along the growth direction. The total Hamiltonian of an electron in anisotropic QDs including spin-orbit interactions can be written as26,34,42 H = Hxy + Hso, where Hso = HR + HD is the Hamilto- nian associated with the Rashba-Dresselhaus spin-orbit couplings and Hxy is the Hamiltonian of the electron in anisotropic QDs. Hxy can be written as Hxy = ~P 2 2m + 1 2 mω2 o(ax2 + by2) + 1 2 goµBσzB, (1) where ~P = ~p+e ~A is the kinetic momentum operator, ~p = −i¯h(∂x, ∂y, 0) is the canonical momentum operator, ~A = B(cid:16)−y√b, x√a, 0(cid:17) /(cid:16)√a + √b(cid:17) is the vector potential in the asymmetric gauge, m is the effective mass, µB is the Bohr magneton, ~σ = (σx, σy, σz) are the Pauli spin matrices, g0 is the bulk g-factor, ω0 = ¯h/(mℓ2 0) is the parabolic confining potential and ℓ0 is the radius of the QDs. The energy spectrum of Hxy can be written as26,43 ε0 n+,n−,± = (n+ + n− + 1) ¯hω+ + (n+ − n−) ¯hω− ± , ∆ 2 (2) where ω± = 1 c + ω2 , ∆ = g0µBB 2(cid:20)ω2 0(cid:16)√a ± √b(cid:17)2(cid:21)1/2 ±a±. Here, a± and a† and n± are the eigenvalues of the Fock-Darwin number operators a† ± are usual annihilation and creation operators. Also, we label the Fock-Darwin states as n+, n−,±i with ± being the eigenvalues of the Pauli spin matrix along z-direction.24,26 FIG. 2. (Color online) Relaxation rate vs anisotropy in QDs. We choose B = 3 T , ℓ0 = 10 nm, λR = λD = 1.7 µm and a = 5. Here we define λR = ¯h2/mαR, λD = ¯h2/mαD, Ex = ¯hω0√a and Ey = ¯hω0√b. The choice of these parameters mimics the experimentally reported values in Ref. 15. It can be seen that the theoretically obtained spin relaxation rate is in excellent agreement with the experimentally reported values in Ref. 15. For symmetric QDs, (lower panel, inset plot), we chose a = b = 5. tigate the interplay between the Rashba and the linear Dresselhaus spin-orbit couplings in QDs. In section III, we provide details of the diagonalization technique used for finding the energy spectrum and the matrix elements of the phonon mediated spin transition rate in QDs. In section IV, we plot both isotropic and anisotropic spin relaxation rates vs. magnetic fields and QDs radii for the pure Rashba and the pure Dresselhaus case in III-V semiconductor materials of zinc blend structure such as GaAs, GaSb, InAs and InSb. Finally, in section V, we summarize our results. Finally, Hso can be written as26 Hso = αR ¯h (σxPy − σyPx) + αD ¯h (−σxPx + σyPy) , (3) where αR = γReE, αD = 0.78γD(cid:18) 2meE ¯h2 (cid:19)2/3 . (4) Here γR and γD are the Rashba and Dresselhaus spin- orbit coefficients. In Fig. 1, we have plotted the con- tribution of the Rashba-Dresselhaus spin-orbit coupling (αR/αD) with the variation of applied electric fields (E) along the z-direction. It can be seen that the Rashba spin-orbit coupling dominates in InAs and InSb QDs whereas the Dresselhaus spin-orbit coupling dominates in GaAs and GaSb QDs. In section IV we will focus our investigation on the phonon mediated spin-relaxation in both symmetric and asymmetric QDs. The Hamiltonian (3) can be written in terms of raising and lowering operators as  GaAs QDs 3 (i) Dresselhaus case Isotropic QDs (a=b=1) (ii) Rashba case Isotropic QDs (a=b=1) (iii) Dresselhaus case Anisotropic QDs (a=0.5,b=2) 6 2 4 0 (iv) Rashba case Anisotropic QDs (a=0.5,b=2) 6 2 4 Magnetic Field, B (T) / ] s 1 [ 1 T 1 / , e t a r n o i t a x a e R l 1012 108 104 100 10-4 1012 108 104 100 10-4 0 FIG. 3. (Color online) Contributions of the Rashba and the Dresselhaus spin-orbit couplings to the phonon induced spin- flip rate as a function of magnetic fields. Material constants are chosen the same as in Fig. 1, but ¯hω0 = 1.1 meV and λR = λD = 8µm. Solid lines (blue) are obtained from Eq. 17. Open circles and squares are obtained numerically from Eq. 16 by an exact diagonalization scheme implemented via Finite Element Method.41 Notice that a cusp-like structure can be seen for the pure Dresselhaus case in asymmetric QDs (Fig. 2 (iii), a 6= b), but not for symmetric QDs (Fig. 2 (i), a = b). Also, the spin-flip rate vanishes like B 5 (see Eq. 25). Fig. 2(i) is Loss et. al. proposal for symmetric QDs (see Ref. 34). Fig. 2(iii) is our proposal for asymmetric QDs. We also expect a similar cusp-like structure for the pure Dresselhaus case with heavy holes in asymmetric QDs which is different from Ref. 34. Hso = αR (1 + i)hb1/4κ+ (s+ − i) a+ + b1/4κ+ (s− + i) a− + a1/4η− (i − s−) a+ + a1/4η− (i + s+) a−i +αD (1 + i)ha1/4κ− (i − s−) a+ + a1/4κ− (i + s+) a− + b1/4η+ (−i + s+) a+ + b1/4η+ (i + s−) a−i + H.c., (5) where s± = cq b ω2 + a 1 ω2 ω+ 4  a − 1 ± r b   ωc(cid:0) b a(cid:1) 2 (s+ − s−)(cid:26) 1 σx ± i 2 (s+ − s−)(cid:26) 1 σy ± i 1 1 ℓ ℓ η± = κ± = 1 ,(6) a!2  2 + 1 −r b  ¯h (cid:18) √a + √b(cid:19) σy(cid:27) , (7) ¯h (cid:18) √a + √b(cid:19) σx(cid:27) , (8) 1 1 eBℓ eBℓ H.c. represents the Hermitian conjugate, ℓ = length and Ω = the hybrid orbital p¯h/mΩ is rω2 c /(cid:16)√a + √b(cid:17)2 0 + ω2 At low electric fields and small QDs radii, we treat the Hamiltonian associated with the Rashba and linear . Dresselhaus spin-orbit couplings as a perturbation. Us- ing second order non degenerate perturbation theory, the energy spectrum of the two lowest electron spin states in QDs (for details, see Ref. 43) is given by ε0,0,+ = ¯h+ − ε0,0,− = ¯h− − Rξ+ + α2 α2 Dς+ ¯hωx − ∆ − α2 Rς+ + α2 Dξ+ ¯hωx + ∆ − Rς− + α2 α2 ¯hωy − ∆ Rξ− + α2 α2 ¯hωy + ∆ Dξ− Dς− , (9) ,(10) where ± = ω+ ± ωz/2, ωz = ∆/¯h is the Zeeman fre- quency, ωx = ω+ + ω−, and ωy = ω+ − ω−. Also, ±(cid:27) ,(11) β2 ∓(cid:27) ,(12) 2(s+ − s−)(cid:26)± 2(s+ − s−)(cid:26)± α2 ∓ − 2α∓β∓ ∓ α2 ± + 2α±β± ∓ 1 s± 1 s± 1 s∓ 1 s∓ ξ± = ς± = β2 1 1  GaAs QDs 4 (i) Dresselhaus case Isotropic QDs (a=b=1) (ii) Rashba case Isotropic QDs (a=b=1) (iii) Dresselhaus case Anisotropic QDs (a=0.5,b=2) 15 30 45 60 75 (iv) Rashba case Anisotropic QDs (a=0.5,b=2) 15 30 45 60 75 QDs radii (nm) / ] s 1 [ 1 T 1 / , e t a r n o i t a x a e R l 1013 108 103 10-2 10-7 1013 108 103 10-2 10-7 FIG. 4. (Color online) Same as Fig. 2 but 1/T1 vs ℓ0. Here we chose, B = 1T . Again, notice the cusp-like structure can only be seen for the pure Dresselhaus case in asymmetric QDs (Fig. 3 (iii), a 6= b) but not for symmetric QDs (Fig. 3 (i), a = b). Also, the spin-flip rate vanishes like ℓ8 0 (see Eq. 25). 1 ℓ ± eBℓ ¯h 1 ℓ ± eBℓ ¯h α± = a1/4  β± = b1/4  1  (cid:16)√a + √b(cid:17)   (cid:16)√a + √b(cid:17)  1 We now turn to the calculation of the phonon in- duced spin relaxation rate at absolute zero temperature between two lowest energy states in QDs. Following Ref. 26, 27, 42, and 44, the interaction between electron and piezo-phonon can be written45 uqα ph (r, t) =s ¯h 2ρV ωqα ei(q·r−ωqαt)eAqαb† qα + H.c., (15) where ρ is the crystal mass density and V is the vol- ume of the QD. b† qα creates an acoustic phonon with wave vector q and polarization eα, where α = l, t1, t2 are chosen as one longitudinal and two transverse modes of the induced phonon in the dots. Aqα = qi qkeβijkej qα is the amplitude of the electric field created by phonon strain, where q = q/q and eβijk = eh14 for i 6= k, i 6= j, j 6= k. The polarization directions of the induced phonon are el = (sin θ cos φ, sin θ sin φ, cos θ), et1 = (cos θ cos φ, cos θ sin φ,− sin θ) and et2 = (− sin φ, cos φ, 0). Based on the Fermi Golden Rule, the ,(13) phonon induced spin transition rate in the QDs is given by26,44 1 T1 = .(14) 2π q ¯h Z d3 (2π)3 Xα=l,t M (qα) 2δ (¯hsαq − εf + εi) , (16) where sl,st are the longitudinal and transverse acoustic phonon velocities in QDs. The matrix element M (qα) = hψiuqα ph (r, t) ψfi with the emission of one phonon qα has been calculated perturbatively and numerically.41,44,46 Here ψii and ψfi correspond to the initial and finial states of the Hamiltonian H. Based on second order non degenerate perturbation theory, after long algebraic tran- formations, we have: 1 T1 = c(cid:0)Mx2 + My2(cid:1) , (17) where Mx = c = 2 (eh14)2 (gµBB)3 t(cid:19) , (18) (is− + 1) Ξ1 (¯hωx + ∆) + (−is− + 1) Ξ3 (¯hωx − ∆) 1 s5 s5 l 35π¯h4ρ (cid:18) 1 a1/4h(¯hωx)2 − ∆2i 4 3 + (−is+ + 1) Ξ2 (¯hωy + ∆) + (is+ + 1) Ξ4 (¯hωy − ∆) + , (19) a1/4h(¯hωy)2 − ∆2i InAs QDs Perturbation results Symmetric QDs (a=b=4) Asymmetric QDs (a=1, b=16) Numerical results Symmetric QDs (a=b=4) Asymmetric QDs (a=1, b=16) 5 (i) QDs radius=15nm Dresselhaus case (ii) QDs radius=15nm Rashba case 0 3 6 9 12 3 Magnetic Field, B (T) 15 0 6 9 12 15 1016 1012 108 104 100 (iii) B=1T Dresselhaus case 30 15 45 QDs radii (nm) 60 (iv) B=1T Rashba case 30 45 60 / ] s 1 [ 1 T 1 / , e t a r n o i t a x a e R l 1016 1012 108 104 100 15 FIG. 5. (Color online) Same as Figs. 2 and 3 but for InAs QDs. We chose E = 105 V /cm. Again, notice the cusp-like structure can also be seen for the pure Dresselhaus case in asymmetric QDs (a 6= b), but not for symmetric QDs (a = b). My = (is+ + 1) Ξ1 (¯hωx + ∆) + (−is+ + 1) Ξ3 (¯hωx − ∆) b1/4h(¯hωx)2 − ∆2i + (is− − 1) Ξ2 (¯hωy + ∆) + (−is− − 1) Ξ4 (¯hωy − ∆) b1/4h(¯hωy)2 − ∆2i , (20) Ξ1 = ℓ Ξ3 = Ξ4 = Ξ2 = ℓ 2 (s+ − s−)2 [αR {(s+ + i) β+ + (1 − is−) α+} + αD {(−s− − i) α− + (−1 + is+) β−}] , 2 (s+ − s−)2 [αR {(s− − i) β+ + (1 + is+) α+} + αD {(s+ − i) α− + (1 + is−) β−}] , 2 (s+ − s−)2 [αR {(s+ − i) β− + (−1 − is−) α−} + αD {(−s− + i) α+ + (1 + is+) β+}] , 2 (s+ − s−)2 [αR {(s− + i) β− + (−1 + is+) α−} + αD {(s+ + i) α+ + (−1 + is−) β+}] . ℓ ℓ (21) (22) (23) (24) In the above expression, we use c = clIxl + 2ctIxt, where cα = and q = gµB B ¯hsα g-factor. is the Land´e longitudinal phonon modes,13,44 . Also, g = ε0,0,−−ε0,0,+ (2π)2¯h2sαεqα2, εqα2 = q2¯h 2ρωqα For q2e2 µB B Aq,l2 = 36h2 14 cos2 θ sin4 θ sin2 φ cos2 φ we have = 16πh2 For and thus we find Ixl transverse phonon modes, we have = 2h2 and thus we find Ixt = 32πh2 Aq,t2 14(cid:2)cos2 θ sin2 θ + sin4 θ(cid:0)1 − 9 cos2 θ(cid:1) sin2 φ cos2 φ(cid:3) 14/105. 14/35. GaSb QDs Perturbation results Symmetric QDs (a=b=4) Asymmetric QDs (a=1, b=16) Numerical results Symmetric QDs (a=b=4) Asymmetric QDs (a=1, b=16) 6 (i) QDs radius=25nm Dresselhaus case 0 2 4 6 Magnetic Field (T) 0 (ii) QDs radius=25nm Rashba case 4 2 6 (iii) B=1T Dresselhaus case 45 30 15 (iv) B=1T Rashba case 60 15 QDs radii (nm) 30 45 60 1016 1012 108 104 100 / ] s 1 [ 1 T 1 / , e t a r n o i t a x a e R l 1017 1013 109 105 101 FIG. 6. (Color online) Same as Fig. 2 and 3, but for GaSb QDs. We chose E = 105 V /cm. Again, notice the cusp-like structure can also be seen for the pure Dresselhaus case in asymmetric QDs (a 6= b), but not for symmetric QDs (a = b). For isotropic QDs (a = b = 1, s+ = 1 and s− = −1), the spin relaxation rate is given by as 1 T1 = 2 (eh14)2 (gµBB)3 35π¯h4ρ (cid:18) 1 s5 l + 4 3 1 s5 t(cid:19)(cid:0)MR2 + MD2(cid:1) , (25) where MR and MD are the coefficients of matrix elements associated with the Rashba and Dresselhaus spin-orbit couplings in QDs and are given by MR = MD = αR√2¯hΩ  1 − αD√2¯hΩ  1 + 1 ∆ ¯h(Ω+ ωc 2 ) − 1 + 1 ∆ ¯h(Ω− ωc 2 ) 1 ∆ ¯h(Ω+ ωc 2 ) − 1 − 1 ∆ ¯h(Ω− ωc 2 ) , (26) . (27)     Since ∆ = g0µBB is negative for GaAs and InAs QDs, we see the degeneracy only appears in the Rashba case (see the 2nd term of Eq. 26) and the degeneracy is absent in the Dresselhaus case. The degeneracy in the Rashba case induces the level crossing point and cusp-like structure in the spin-flip rate in QDs. The spin relaxation rate for isotropic QDs can be written in a more convenient form 1 T1 = 2 (eh14)2 (gµBB)3 35π¯h4ρ (cid:18) 1 s5 l + 4 3 1 s5 t(cid:19) 2∆2m4 ¯h8 R + α2 (cid:0)α2 D(cid:1) ℓ8 0h1 + O (ωc/ωo)2i . (28) From Eq. 28, it is clear that the spin-flip rate vanishes like B5 and ℓ8 0 (see Ref. 26). TABLE I. The material constants used in our calculations are taken from Refs. 26 and 47 0.0239 InAs GaSb −7.8 -15 0.0412 33 187 2 ] Parameters g0 m γR [A γD [eV A eh14 [10−5erg/cm] sl [105cm/s] st [105cm/s] ρ [g/cm3] ] 3 GaAs -0.44 0.067 4.4 26 2.34 5.14 3.03 5.3176 110 130 0.54 4.2 2.35 5.667 1.5 4.3 2.49 5.6137 InSb -50.6 0.0136 500 228 0.75 3.69 2.29 5.7747 InSb QDs Perturbation results Symmetric QDs (a=b=4) Asymmetric QDs (a=1, b=16) 7 Numerical results Symmetric QDs (a=b=4) Asymmetric QDs (a=1, b=16) / ] s 1 [ 1 T 1 / , e t a r n o i t a x a e R l 1018 1014 1010 106 102 10-2 1020 1016 1012 108 104 100 10-4 0 (i) QDs radius=15nm Dresselhaus case 0 2 4 6 8 (ii) QDs radius=25nm Rashba case 0 10 4 Magnetic Field (T) 2 6 8 10 (iii) B=1T Dresselhaus case (iv) B=1T Rashba case 10 20 30 40 50 0 10 20 30 40 50 QDs radii (nm) FIG. 7. (Color online) Same as Fig. 2 and 3 but for InSb QDs. We chose E = 104 V /cm. Again, notice the cusp-like structure can also be seen for the pure Dresselhaus case in asymmetric QDs (a 6= b), but not for symmetric QDs (a = b). III. COMPUTATIONAL METHOD We suppose that a QD is formed at the center of a 400 × 400 nm2 geometry. Then we diagonalize the to- tal Hamiltonian H numerically using the Finite Element Method.41 The geometry contains 24910 elements. Since the geometry is much larger compared to the actual lat- eral size of the QD, we impose Dirichlet boundary con- ditions, find the eigenvalues, eigenfunctions and the ma- trix elements M (qα) of the total Hamiltonian H. From Figs. 2 to 7, the analytically obtained spin-flip rates from Eq. 17 (solid and dashed-dotted lines) are seen to be in excellent agreement with the numerical values (open cir- cles and squares). The material constants are taken from table I. IV. RESULTS AND DISCUSSIONS In Fig. 2, we compare theoretically obtained spin-flip rates from Eq. 17 to the experimentally reported values in Ref. 15. Theoretical and experimental data are in ex- cellent agreement. Inset plots (from left to right) show realistic in-plane wavefunctions of QDs for the spin states 0, 0, +1/2i, 0, 0,−1/2i and 0, 1, +1/2i. It can be seen that anisotropy breaks the in-plane rotational symmetry. As a result, we find that the in-plane wavefunction of anisotropic QDs for the states 0, 1, +1/2i split into two which has a direct consequence on inducing accidental degeneracy even for the pure Dresselhaus spin-orbit cou- pling case in the phonon mediated spin flip rate. This will be separately discussed from Figs. 3 to 7. In Fig. 3 (i) we see that the cusp-like structure is ab- sent (i.e., the spin-flip rate is a monotonous function of the magnetic field) for the pure Dresselhaus case in sym- metric QDs. However in Fig. 3(iii) we see that the cusp- like structure is present for the pure Dresselhaus case in asymmetric QDs. In Fig. 4, again we see that the cusp- like structure is absent in isotropic QDs (a = b) but is present in anisotropic QDs (a 6= b) for the pure Dressel- haus case. The cusp-like structure in anisotropic QDs is thus due to the fact that the anisotropy induces the acci- dental degeneracy in the matrix elements (M (qα)) near the level crossing or anticrossing point. The accidental degeneracy point where the cusp-like structure appears is referred to as the spin-hot spot while tuning on the spin- orbit coupling removes the degeneracy.46 Thus, we apply degenerate perturbation theory and the energy spectrum of the unperturbed spin states 0, 0,−i and 0, 1, +i for anisotropic QDs are given by pure Dresselhaus spin-orbit coupling case in anisotropic QDs (a 6= b). 8 Rξ− + α2 Rξ− + α2 ¯hω− +(cid:2)α2 ¯hω− −(cid:2)α2 , (29) . (30) Dζ−(cid:3)1/2 Dζ−(cid:3)1/2 ε0 0,0,− = ε0 0,1,+ = 3 ¯hω+ − 2 3 ¯hω+ − 2 1 2 1 2 We have substituted Eqs. 29 and 30 into 17 and found the spin-flip rate at the level crossing point from Figs. 2 to 7. Lifting the degeneracy with the application of spin-orbit couplings mixes spin up and spin down states where the phonon mediated spin transition rate between states of opposite magnetic moment will involve spin flips with a much more enhanced probability compared to the normal states. For example, the spin-hot spot for the pure Dres- selhaus case in symmetric GaAs QDs (Figs. 3 (i) and 4 (i)) can not be observed while tuning the anisotropy (a 6= b), however can be observed at B = 5.1 T and ℓ0 = 69 nm as shown in Figs. 3 (iii) and 4 (iii), respec- tively. Notice that the spin-flip rates of the pure Dressel- haus case found near the spin-hot spot in Figs. 3 (iii) and 4 (iii) are 6 orders of magnitude larger than those values found in Figs. 3 (i) and 4(i). This result (i.e., the spin-hot spot in asymmetric QDs for the pure Dresselhaus case yet to be experimentally verified) provides small relaxation and decoherence time which should be avoided during the design of spin based transistors for the possible im- plementation in quantum logic gates, quantum comput- ing and quantum information processing. From Figs. 4 to 7, we investigated the spin relaxation rate in InAs, GaSb and InSb QDs. Analyzing all plots, the spin-hot spot and associated cusp-like structure can be seen in the V. CONCLUSIONS We have shown that the anisotropy breaks the in-plane rotational symmetry. As a result, we found that the cusp- like structure (i.e., where the spin-hot spot) is present in the phonon mediated spin transition rate in anisotropic QDs for the pure Dresselhaus spin-orbit coupling case. In contrast, for isotropic QDs, the spin transition rate is a monotonous function of magnetic fields and QDs radii (i.e., where the spin-hot spot is absent) for the pure Dres- selhaus spin-orbit coupling case. These results (yet to be experimentally verified) provide new information for finding the spin hot-spot in anisotropic spin relaxation for the pure Dresselhaus case during the design of QD spin transistors. At or nearby the spin-hot spot, the relax- ation and decoherence time are smaller by several orders of magnitude. One should avoid such locations during the design of QD spin based transistors for the possible implementation in quantum logic gates, quantum com- puting and quantum information processing. VI. ACKNOWLEDGEMENTS This work has been supported by Natural Science and Engineering Research Council (Canada) and Canada Re- search Chair programs. The authors acknowledge the Shared Hierarchical Academic Research Computing Net- work (SHARCNET) community and Dr. Philip James Douglas Roberts for the helpful and technical support. 10 M. P. Nowak and Phys. Rev. B 83, 035315 (2011). 11 M. P. Nowak and Phys. Rev. B 80, 195319 (2009). 12 M. P. Nowak and Phys. Rev. B 81, 235311 (2010). 13 V. N. Golovach, A. Khaetskii, Phys. Rev. Lett. 93, 016601 (2004). B. B. B. Szafran, Szafran, Szafran, and D. Loss, 14 X. Wang, L. S. Bishop, J. Kestner, E. Barnes, K. Sun, and S. Das Sarma, Nat Commun 3, 997 (2011). 15 S. Amasha, K. MacLean, I. P. Radu, D. M. Zumbuhl, M. A. Kastner, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 100, 046803 (2008). 16 S. Bandyopadhyay, Phys. Rev. B 61, 13813 (2000). 17 T. Fujisawa, Y. Tokura, and Y. Hirayama, Phys. Rev. B 63, 081304 (2001). 18 A. Balocchi, Q. H. Duong, P. Renucci, B. L. Liu, and X. Marie, C. Fontaine, T. Amand, D. Lagarde, Phys. Rev. Lett. 107, 136604 (2011). 19 M. Griesbeck, M. M. Glazov, E. Y. Sherman, D. Schuh, W. Wegscheider, C. Schuller, and T. Korn, Phys. Rev. B 85, 085313 (2012). 1 D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). 2 D. D. Awschalom, D. Loss, and N. Samarth, Semicon- ductor Spintronics and Quantum Computation (Springer, Berlin, 2002). 3 R. Hanson, L. H. W. van Beveren, I. T. Vink, J. M. Elzerman, W. J. M. Naber, F. H. L. Kop- pens, L. P. Kouwenhoven, and L. M. K. Vandersypen, Phys. Rev. Lett. 94, 196802 (2005). 4 M. Kroutvar, Y. Ducommun, D. Heiss, M. Bichler, D. Schuh, G. Abstreiter, and J. J. Finley, Nature 432, 81 (2004). 5 J. M. Elzerman, R. Hanson, L. H. Willems van Beveren, B. Witkamp, L. M. K. Vandersypen, and L. P. Kouwen- hoven, Nature 430, 431 (2004). Sherman, and V. Dugaev, 6 M. Glazov, E. 7 J. I. Climente, A. Bertoni, G. Goldoni, M. Rontani, and E. Molinari, Phys. Rev. B 75, 081303 (2007). 8 P. Pietilainen and T. Chakraborty, Phys. Rev. B 73, 155315 (2006). and Chakraborty 9 T. Phys. Rev. Lett. 95, 136603 (2005). P. Pietilainen, Physica E: Low-dimensional Systems and Nanostructures 42, 2157 (2010). 20 O. Olendski T. Phys. Rev. B 75, 041306 (2007). and 21 M. E. Flatt´e, Physics 4, 73 (2011). 22 Y. Bychkov and A. V. Shahbazyan, E. I. Rashba, J. Phys. C: Solid State Phys. 17, 6039 (1984). 23 G. Dresselhaus, Phys. Rev. 100, 580 (1955). 24 S. Prabhakar and E. J. Raynolds, Phys. Rev. B 79, 195307 (2009). 25 S. Prabhakar, J. Raynolds, A. Inomata, and R. Melnik, Phys. Rev. B 82, 195306 (2010). 26 R. de Sousa and S. Das Sarma, Phys. Rev. B 68, 155330 (2003). 27 S. Prabhakar, R. V. N. Melnik, and L. L. Bonilla, Applied Physics Letters 100, 023108 (2012). 28 J. A. Folk, C. M. Marcus, C. I. Duruoz, Phys. Rev. Lett. 86, 2102 (2001). S. R. Patel, K. M. Birnbaum, and J. S. Harris, 29 T. Fujisawa, D. G. Austing, Y. Tokura, Y. Hirayama, and S. Tarucha, Nature 419, 278 (2002). 30 R. Hanson, B. Witkamp, L. M. K. Vandersypen, L. H. W. van Beveren, J. M. Elzerman, and L. P. Kouwenhoven, Phys. Rev. Lett. 91, 196802 (2003). 9 36 C. Yang, A. Rossi, R. Ruskov, N. Lai, F. Mohiyaddin, S. Lee, C. Tahan, G. Klimeck, A. Morello, and A. Dzurak, arXiv:1302.0983. 37 W. Peng, Z. Aksamija, S. A. Scott, J. J. Endres, D. E. Savage, I. Knezevic, M. A. Eriksson, and M. G. Lagally, Nat Commun 4, 1339 (2013). 38 T. S. Koh, J. K. Gamble, M. Friesen, M. A. Eriksson, and S. N. Coppersmith, Phys. Rev. Lett. 109, 250503 (2012). 39 Z. Shi, C. B. Simmons, J. R. Prance, J. K. Gamble, T. S. Koh, Y.-P. Shim, X. Hu, D. E. Savage, M. G. Lagally, M. A. Eriksson, M. Friesen, and S. N. Coppersmith, Phys. Rev. Lett. 108, 140503 (2012). 40 J. R. Prance, Z. Shi, C. B. Simmons, D. E. Savage, M. G. Lagally, L. R. Schreiber, L. M. K. Vandersypen, M. Friesen, R. Joynt, S. N. Coppersmith, and M. A. Eriks- son, Phys. Rev. Lett. 108, 046808 (2012). 41 Comsol Multiphysics version 3.5a (www.comsol.com). 42 A. V. Khaetskii and Y. V. Nazarov, Phys. Rev. B 61, 12639 (2000). 43 S. Prabhakar, J. E. Raynolds, and R. Melnik, Phys. Rev. B 84, 155208 (2011). 44 A. V. Khaetskii and Y. V. Nazarov, Flatt´e, Phys. Rev. B 64, 125316 (2001). 45 V. F. Gantmakher and Y. B. Levinson, Carrier Scattering in Metals and Semiconductor (North-Holland, Amster- dam, 1987). 46 P. Stano and J. Fabian, Phys. Rev. B 74, 045320 (2006). 47 M. Cardona, N. E. Christensen, and G. Fasol, Phys. Rev. B 38, 1806 (1988). 31 C. E. Pryor and M. Phys. Rev. Lett. 96, 026804 (2006). 32 C. E. Pryor and M. Phys. Rev. Lett. 99, 179901 (2007). E. E. Flatt´e, 33 M. P. Nowak, B. Szafran, F. M. Peeters, B. Partoens, and W. J. Pasek, Phys. Rev. B 83, 245324 (2011). D. and Phys. Rev. Lett. 95, 076805 (2005). and Bulaev Bulaev D. V. V. 34 D. 35 D. Phys. Rev. B 71, 205324 (2005). Loss, Loss, 

1501.02092

1

1501

Cross-Kerr nonlinearity in optomechanical systems

We consider the response of a nanomechanical resonator interacting with an electromagnetic cavity via a radiation pressure coupling and a cross-Kerr coupling. Using a mean field approach we solve the dynamics of the system, and show the different corrections coming from the radiation pressure and the cross-Kerr effect to the usually considered linearized dynamics.

cond-mat.mes-hall

cond-mat

Cross-Kerr nonlinearity in optomechanical systems Raphael Khan,1, ∗ F. Massel,2, † and T. T. Heikkila2, ‡ 1Low Temperature Laboratory, Aalto University, P.O. Box 15100, FI-00076 AALTO, Finland 2Department of Physics and Nanoscience Center, University of Jyvaakyla, P.O. Box 35 (YFL), FI-40014 University of Jyvaskyla, Finland We consider the response of a nanomechanical resonator interacting with an electromagnetic cavity via a radiation pressure coupling and a cross-Kerr coupling. Using a mean field approach we solve the dynamics of the system, and show the different corrections coming from the radiation pressure and the cross-Kerr effect to the usually considered linearized dynamics. 5 1 0 2 n a J 9 ] l l a h - s e m . t a m - d n o c [ 1 v 2 9 0 2 0 . 1 0 5 1 : v i X r a I. INTRODUCTION Cavity optomechanics offers a framework to study the coupling between an electromagnetic field and the vibra- tions of a mechanical resonator. The interaction between these two systems is usually mediated by a radiation- pressure type coupling proportional -- through a coupling constant g -- to the number of photons nc in the cavity and the displacement of the mechanical resonator. The radi- ation pressure coupling offers the possibility of altering the resonant frequency of the mechanical resonator and its damping. The latter can be used for cooling [4 -- 6] or amplification [7]. Moreover, the nonlinearity of the in- teraction may allow for the observation of macroscopic quantum phenomena such as quantum superpositon of states [1, 2] or quantum squeezed states [3]. The re- quirements for observing these quantum phenomena are the necessity of being close to the ground state and be- ing in the strong coupling regime [8, 9] where g is larger than the cavity and the mechanical resonator decay rate. However, g is usually weak and to bypass this constraint a strong drive to the cavity is applied at the cost of losing the nonlinear property of the interaction. Our recent proposal [10], in which the cavity and the resonator are coupled to a Josephson junction, shows that the interaction between the cavity and the resonator can be enhanced via the non-linearity of the Josephson effect. Quadratic and higher-order interactions in the displace- ment have been investigated also in different setups such as membrane in the middle geometries [11 -- 13]. Analogously to the setup mentioned above, the non- linearity of the Josephson effect leads to an additional nonlinear interaction, namely a cross-Kerr coupling gck between the cavity and the resonator. The difference resides in the fact that in the Josephson junction setup the relative value of gck and g depends on the value of the gate charge to a superconducting island, whereas in [12, 13], it generally reflects the position of the resonator within the cavity. In the context of optomechanical systems, the cross- Kerr coupling between the resonator and the cavity in- ∗ raphael.khan@aalto.fi † francesco.p.massel@jyu.fi ‡ tero.t.heikkila@jyu.fi duces a change to the refractive index of the cavity de- pending on the number of phonons in the resonator, whereas the radiation pressure coupling gives rise to an analogous effect, but depending on the displacement of the mechanical resonator. In this paper we solve the dynamics of the cavity and the mechanical resonator in the presence of the cross- Kerr and the radiation pressure coupling. We determine the effects of the cross-Kerr coupling on the red and blue sidebands within a mean field approach. In particular, we demonstrate that the sideband peak is shifted due to the cross-Kerr coupling. In addition, the cross-Kerr coupling induces a nonmonotonuous response of the ef- fective mechanical damping as a function of the number of photons pumped into the cavity. FIG. 1. Schematic picture of the system. A cavity and a mechanical resonator coupled via a radiation type coupling g and a cross-Kerr coupling gck. The number of photons in the cavity nc is coupled to the oscillations of the mechanical resonator x and the number of phonons in the mechanical resonator nm. II. MEAN FIELD APPROACH We consider an electromagnetic cavity with frequency ωc and linewidth κ coupled to a mechanical resonator with frequency ωm and linewidth γ. The number of phonons in the cavity nc is coupled to the vibration am- CavityMechanicalresonatorncnmx^gckgωcωmκγ(cid:21) (cid:21) plitude of the mechanical resonator x via a radiation- pressure type coupling g. In addition the number of pho- tons nc is coupled to the number of phonons nm in the mechanical resonator through a cross-Kerr coupling gck (Fig. 1). The Hamiltonian of the system is ( = 1) H = ωca†a + ωmb†b − ga†a(b† + b) − gcka†ab†b, (1) where a and b are the annihilation operators of the cavity and the mechanical resonator, respectively. We treat the interactions with a mean-field (MF) approach. Within this frame, the radiation-pressure interaction becomes ga†a(b† + b) = g ((cid:104)a†(cid:105)a + (cid:104)a(cid:105)a† − (cid:104)a†a(cid:105))(b† + b) + (a†a − (cid:104)a(cid:105)a† − (cid:104)a†(cid:105)a)(cid:104)b† + b(cid:105) , (2) where (cid:104)A(cid:105) stands for the average of A over the static nonequilibrium state of the system (mean field). The negative terms in Eq. (2) are included to suppress dou- ble counting. The first line of Eq. (2) describes exchange processes between the resonator and the cavity while the second line gives a frequency shift of the cavity which is proportional to the average displacement of the res- onator. This decomposition allows us to find the usual results of the weak radiation pressure coupling [7, 16]. In MF, the cross-Kerr coupling becomes (cid:20) (cid:20) gcka†ab†b = gck (cid:104)a†a(cid:105)b†b + (cid:104)b†b(cid:105)a†a + (cid:104)b†a(cid:105)ba† + (cid:104)ba†(cid:105)b†a + (cid:104)ba(cid:105)b†a† + (cid:104)b†a†(cid:105)ba . (3) The term (cid:104)a†a(cid:105)b†b ((cid:104)b†b(cid:105)a†a) describes a Hartree-like in- teraction between the resonator and the cavity while the other terms describe exchange processes between the res- onator and the cavity. Thus we can rewrite the Hamilto- nian as H = [ωc − gck(cid:104)b†b(cid:105)]a†a + [ωm − gck(cid:104)a†a(cid:105)]b†b − G[(cid:104)a†(cid:105)ab† + (cid:104)a(cid:105)a†b†] − G∗[(cid:104)a†(cid:105)ab + (cid:104)a(cid:105)a†b] + g[(a†a − (cid:104)a(cid:105)a† − (cid:104)a†(cid:105)a)(cid:104)b† + b(cid:105) − (cid:104)a†a(cid:105)(b† + b)], (4) where the expectation values of the different operators have to be determined self-consistently within the MF picture and G = g + gck(cid:104)b(cid:105). We assume the usual experi- mental situation where ωc (cid:29) ωm and where the cavity is driven with a coherent field of strength fp oscillating at frequency ωp = ωc + ∆. Using the input-output formal- ism [16] the equations of motion are a = −i[−∆ − gck(cid:104)b†b(cid:105)]a − κ 2 + iG∗(cid:104)a(cid:105)b + iG(cid:104)a(cid:105)b† − ig(cid:104)b† + b(cid:105)[a − (cid:104)a(cid:105)] b = −i[ωm − gck(cid:104)a†a(cid:105)]b − γ 2 γbin κfp a + √ √ b + (5) +iG(cid:104)a†(cid:105)a + iG(cid:104)a(cid:105)a† − ig(cid:104)a†a(cid:105), (6) 2 √ κfp Here we have written the cavity operator a in a frame rotating with frequency ωp neglecting the fast oscillati- ing terms. We define bin to be the thermal input of † in(t)bin(t(cid:48))(cid:105) = the resonator satisfying (cid:104)bin(t)(cid:105) = 0 and (cid:104)b nthδ(t − t(cid:48)), where nth is the number of phonons in the thermal bath damping the resonator. We split the cav- ity and the mechanical operators into a sum of coherent and fluctuation parts, i.e., a ≡ δa + α and b ≡ δb + β with α = (cid:104)a(cid:105), β = (cid:104)b(cid:105) and (cid:104)δa(cid:105) = (cid:104)δb(cid:105) = 0. As usual, we assume that α and β oscillate at the same frequency as the coherent drive so that α = β = 0. With these approximations, the solutions of Eqs. (5, (6)) are α = β = κ 2 − i[∆ − gck(cid:104)b†b(cid:105) − (G∗β + Gβ∗)] i(2G − g)α2 − ig(cid:104)δa†δa(cid:105) γ/2 + i(ωm − gck(cid:104)a†a(cid:105))) . , (7) (8) 2 (9) √ (cid:105) (cid:105) (cid:104) κ (cid:104) γ In the derivation of Eqs. (7, 8), we have assumed, in agreement with what is usually done in the optomechan- [18]), ∆ + g(cid:104)b† + b(cid:105) ≈ ∆. The ical literature (see e.g. equations of motion for the fluctuations in the Fourier space are given by − i(ω + ∆) 2 − i(ω − ωm) δa = iGαδb† + iG∗αδb δb = iGα∗δa + iGαδa† + γbin,(10) where ∆ = ∆ + gck(cid:104)b†b(cid:105) and ωm = ωm − gck(cid:104)a†a(cid:105). The effect of the thermal drive bin on the response of the cavity is mediated by the coupling G. Through this cou- pling the oscillations of the mechanical resonator produce sideband peaks at ωd ± ωm in the cavity response. They allow for the exchange of energy between the cavity and the resonator when ∆ ≈ ±ωm [5, 17]. These processes are depicted in Fig. 2. For ∆ ≈ −ωm the system is in the red sideband regime and one can transfer energy from the resonator to the cavity, thus the mechanical resonator is damped and cooled. For ∆ ≈ ωm, the system is in the blue sideband regime and one can transfer energy from the cavity to the resonator, thus the mechanical resonator is excited and heated. In order to find the correction to the damping, we solve the response function of δa for the thermal input δbin. We find that it is a Lorentzian function peaked at ωm + ωshift with m + κ2 4 ) ωshift = −G2α2( ∆2 − ω2 − (cid:32) ωm 1 κ2 4 + (ωm + ∆)2 1 κ2 4 + (ωm − ∆)2 and whose linewidth is γ + Γopt with Γopt = G2α2κ (cid:32) 1 κ2 4 + (ωm + ∆)2 − 1 κ2 4 + (ωm − ∆)2 (cid:33) (cid:33) , (11) . (12) 3 FIG. 3. Schematic picture of the red sideband with and with- out cross-Kerr coupling for ∆ < 0. For gck > 0 the sideband peak is shifted to lower values while for gck < 0 the sideband peak is shifted to higher values. up only in the frequency shift which now depends on the number of coherent and thermal photons in the cavity Eq. (16). In Fig. 3 we show a schematic picture of what happens to the sideband in the presence of the cross-Kerr coupling. In the Doppler limit (ωm ≤ κ) the frequency shift and optical damping are given by ωshift = ∓4G2α2 Γopt = ± 4G2α2 κ ωm − gck(cid:104)a†a(cid:105) κ2 4 + 4(ωm − gck(cid:104)a†a(cid:105))2 4(ωm − gck(cid:104)a†a(cid:105))2 4 + 4(ωm − gck(cid:104)a†a(cid:105))2 κ2 (18) . (19) Now both the frequency shift and the optical damping depend on the cross-Kerr coupling. In Figs. 4-5 we plot the number of phonons and the optical damping as a function of ωm/κ for the red sideband in the Doppler limit. Since the cross-Kerr coupling shifts the mechanical frequency, the value of Γopt is shifted as well. The sign of the shift is given by the sign of gck. Otherwise we recover the cooling of the resonator for the red sideband (Fig. 4) and the parametric instability when Γopt = −γ for the blue sideband (Fig. 5). IV. CASE WITH ∆ = ωm. In experiments the parameter one can tune directly is the detuning ∆ and not ∆ as it can be difficult to set ∆ = ∓ωm for each value of α as the pump strength is varied. Therefore, another regime we consider is the case where ∆ = ∓ωm, i.e, setting ∆ = ∓ωm + gck(cid:104)b†b(cid:105). In this case the frequency shift and optical damping in the red (upper sign) and in the blue (lower sign) sideband FIG. 2. Cooling (heating) process. The cavity is driven with a frequency ωd = ωc − ωm (ωd = ωc + ωm). The drive does not allow a transition from nm, nc(cid:105) → nm, nc + 1(cid:105) but allow the transition from nm, nc(cid:105) → nm − 1, nc + 1(cid:105) (nm, nc(cid:105) → nm + 1, nc + 1(cid:105)). The cavity relaxes then to the state nm − 1, nc(cid:105) (nm + 1, nc(cid:105)) resulting into cooling (heating) of the mechanical resonator. Integrating the Lorentzian function obtained above, we obtain the number of phonons and photons coming from the thermal vibrations of the resonator. We get [6] γnth + Γoptnm0 (cid:104)δb†δb(cid:105) = γ + Γopt (cid:104)δa†δa(cid:105) = G2α2(cid:104)δb†δb(cid:105) , (cid:32) 1 κ2 4 + (ωm + ∆)2 + 1 κ2 4 + (ωm − ∆)2 (13) (14) (cid:33) with nm0 = − (ωm + ∆)2 + κ2 4 4 ∆ωm . (15) Eqs. (7), (8), (13) and (14) form a set of self-consistency equations and are solved in the next sections in order to find the number of phonons in the resonator and photons in the cavity. We now focus on the sidebands. III. OPTIMAL COOLING/HEATING In order to minimize/maximize the optical damping Γopt, we set ∆ = ∓ωm . The upper sign refers to the red sideband (Γopt > 0) and the lower sign to the blue sideband (Γopt < 0). In the resolved sideband limit, ωm (cid:29) κ (cid:29) γ, the frequency shift and the optical damp- ing become ωshift = ∓G2α2 Γopt = ± 4G2α2 ωm . κ = ∓ G2α2 ωm − gck(cid:104)a†a(cid:105) , (16) (17) The result for the optical damping Eq. (17), is identical to the one usually obtained in optomechanics in the ab- sence of the cross-Kerr coupling. The effect of gck shows nmncnc+1nmnc+1nmnc-1nm-1ωcωdωmnmnc+1nc+1nm+1ωmΔ=-ωshiftgck=0γ+Γoptgck<0gck>0ωpγωcωshiftωshift4 FIG. 4. Steady-state phonon number in the resonator and Γopt (inset) as a function of the ratio ωm/κ at the red sideband for the optimal case ∆ = −ωm with γ = 10−3κ, g = 10−2κ. The number of photons pumped into the cavity is fixed to α2 = 100 and the bath temperature corresponds to nth = 100. FIG. 6. Steady-state phonon number in the resonator and the optical damping Γopt (inset) as a function of the number of photons pumped into the cavity in the case where ∆ = −ωm. The values for the parameters are γ = 10−3ωm, κ = 10−1ωm, g = 10−2ωm and the bath temperature corresponds to nth = 100. become ωshift = ∓G2α2 Γopt = ± ωm [κ2/4 − 2gckωm((cid:104)b†b(cid:105) − (cid:104)a†a(cid:105))] ck((cid:104)b†b(cid:105) − (cid:104)a†a(cid:105))2 + κ2/4] [g2 ck((cid:104)b†b(cid:105) − (cid:104)a†a(cid:105))2 + κ2/4 . g2 G2α2κ , (20) (21) In Figs. 6 and 7 the steady-state phonon number and the optical damping are plotted as a function of the number of photons pumped into the cavity, for the red and blue sidebands respectively. For the red sideband (Fig. 6) the optical damping increases with increasing α. ckα2 (cid:29) κ2/4 the optical damping becomes in- When g2 versely proportional to the number of photons pumped into the cavity, consequently, the cooling deteriorates when pumping more phonons in the cavity. In the blue sideband (Fig. 7) the main effect of a small cross-Kerr coupling is to limit the instability to a finite number of phonons, (cid:104)b†b(cid:105) ≈ (cid:112)κ/γGα/gck + α2. FIG. 5. Steady-state phonon number in the resonator and Γopt (inset) as a function of the ratio ωm/κ at the blue sideband for the optimal case ∆ = ωm with γ = 10−2κ, g = 10−2κ. The number of photons pumped into the cavity is fixed to α2 = 100 and the bath temperature corresponds to nth = 10. The dashed lines indicate the onset of the para- metric instability for Γopt = −γ. This thus competes with the usual limitation coming from the intrinsic (Duffing) nonlinearity of the resonator. In Figs. 8-9 we plot the frequency shift as a function of the number of photons pumped into the cavity for the red and blue sidebands. For the red sideband when gck > 0 (gck < 0) the frequency shift increases (decreases) as α increases until α ≈ (cid:104)b†b(cid:105) after which it increases (de- creases). For the blue sideband when gck > 0 the fre- quency shift decreases while for for gck < 0 it increases. The difference at small α between the red and blue side- bands arises from the fact that in the red sideband when pumping more photons into the cavity the cooling im- proves, thus the number of phonons in the mechanical 0.20.40.60.81100101102gck=0gck=0.1ggck=−0.1g0.20.40.60.810.0050.010.0150.020.0250.030.0350.04ωm/κResonator phonon numberΓopt/κωm/κgck=0gck=0.02ggck=−0.02g0.050.10.15−15−10−505x10−30.050.10.150.2101102103Resonator phonon numberΓopt/κωm/κωm/κ02040608010010−110010110202040608010000.050.10.150.20.250.30.350.4α2Γopt/ωmgck=0gck=0.1gResonator phonon numberα25 depending on the number of photons in the cavity and phonons in the resonator. In addition, we have shown that when the detuning of the pump is equal to the fre- quency of the mechanical resonator the variation of the optical damping saturates instead of being linearly de- pendent on the number of phonons pumped into the cav- ity. This work was supported by the European Re- FIG. 8. Frequency shift as a function of the number of pho- tons pumped into the cavity when ∆ = −ωm with γ = 10−3ωm κ = 10−1ωm, g = 10−2ωm and the bath tempera- ture corresponds to nth = 100. FIG. 7. a) Steady-state phonon number in the resonator and b) the optical damping Γopt as a function of the number of photons pumped into the cavity in the case where ∆ = ωm. The values for the parameters are γ = 10−3ωm, κ = 10−1ωm, g = 10−2ωm and the bath temperature corresponds to nth = 10. resonator decreases, making it possible to have a num- ber of photons in the cavity of the same order and larger than the number of phonons in the resonator. V. CONCLUSION In conclusion, we have solved the dynamics of a me- chanical resonator coupled to an electromagnetic cavity via a radiation pressure coupling and a cross-Kerr cou- pling using a mean field approach. We have shown that the cross-Kerr coupling shifts the frequency of the me- chanical resonator and of the optical cavity, the shift FIG. 9. Frequency shift as a function of the number of pho- tons pumped into the cavity when ∆ = ωm with γ = 10−3ωm κ = 10−1ωm, g = 10−2ωm and the bath temperature corre- sponds to nth = 100. search Council (Grant No. 240362-Heattronics) and the Academy of Finland. [1] W. Marshall, C. Simon, R. Penrose, and D. Bouwmeester, Phys. Rev. Lett. 91, 130401 (2003). A. Guerreiro, V. Vedral, A. Zeilinger, and M. As- pelmeyer, Phys. Rev. Lett. 98, 030405 (2007). [2] D. Vitali, S. Gigan, A. Ferreira, H. R. Bohm, P. Tombesi, [3] A. A. Clerk, F. Marquardt, and K. Jacobs, New J. Phys. 01020304050X10-3-2-1.5-1-0.50gck=0gck=0.1ggck=0.01gΓopt/ωmα201020304050101102103104gck=0gck=0.1ggck=0.01gα2Resonator phonon numbera)b)05101520−6−5−4−3−2−1012x 10−3gck=0gck=0.01ggck=−0.01gωshiftωmα201020304050X10-3-1-0.500.51gck=0gck=0.01ggck=-0.01gωshiftωmα26 10, 095010 (2008). 073601 (2007). [4] J. D. Teufel, J. W. Harlow, C. A. Regal, and K. W. Lehnert, Phys. Rev. Lett. 101, 197203 (2008). [5] A. Schliesser, P. Del'Haye, N. Nooshi, K. J. Vahala, and [12] J. D. Thomson, B. M. Zwickl, A. M. Jayich, F- Mar- quardt, S. M. Girvin, and J. G. E. Harris, Nature 452, 72 (2007). T. J. Kippenberg, Phys. Rev. Lett. 97, 243905 (2006). [13] A. Xuereb, and M. Paternostro Phys. Rev. A 87, 023830 [6] F. Marquardt, J. P. Chen, A. A. Clerk, and S. M. Girvin, (2013). Phys. Rev. Lett. 99, 093902 (2007). [14] L. Mandel and E. Wolf. Optical coherence and quantum [7] F. Massel, T. T. Heikkila, J. Pirkkalainen, S. U. Cho, H. Saloniemi, P. J. Hakonen, and M. A. Sillanpaa, Nature 480, 351 (2011). optics (Cambridge university press, 1995). [15] M. Bartkowiak, L. A. Wu, and A. Miranowicz, J. Phys. B 47(14), 145501 (2014). [8] S. Groblacher, K. Hammerer, M. R. Vanner, and M. As- [16] C. W. Gardiner and M. J. Collett, Phys. Rev. A 31, 3761 pelmeyer, Nature 460, 724 (2009). (1985). [9] J. D. Teufel, Dale Li, M. S. Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, and R. W. Simmonds, Nature 460, 724 (2009). [10] T. T. Heikkila, F. Massel, J. Tuorila, R. Khan, and M. A. [17] 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 475, 359 (2011). [18] M. Aspelmeyer, T. J. Kippenberg, and F. Marquardt Sillanpaa, Phys. Rev. Lett. 112, 203603, (2014). arXiv:1303.0733. [11] M. Bhattacharya, and P. Meystre Phys. Rev. Lett. 99, 

1902.00403

1

1902

Collective magnetization dynamics in nano-arrays of thin FePd discs

We report on the magnetization dynamics of a square array of mesoscopic discs, fabricated from an iron palladium alloy film. The dynamics properties were explored using ferromagnetic resonance measurements and micromagnetic simulations. The obtained spectra exhibit features resulting from the interactions between the discs, with a clear dependence on both temperature and the direction of the externally applied field. We demonstrate a qualitative agreement between the measured and calculated spectra. Furthermore, we calculated the mode profiles of the standing spin waves excited during a time-dependent magnetic field excitations. The resulting maps confirm that the features appearing in the ferromagnetic resonance absorption spectra originate from the temperature and directional dependent inter-disc interactions.

cond-mat.mes-hall

cond-mat

Collective magnetization dynamics in nano-arrays of thin FePd discs Agne Ciuciulkaite,1, ∗ Erik Ostman,1 Rimantas Brucas,2, 3 Ankit Kumar,2 Marc A. Verschuuren,4 Peter Svedlindh,2 Bjorgvin Hjorvarsson,1 and Vassilios Kapaklis1, † 1Department of Physics and Astronomy, Uppsala University, Box 516, SE-75120 Uppsala, Sweden 2Department of Engineering Sciences, Uppsala University, Box 534, SE-751 21 Uppsala, Sweden 3Angstrom Microstructure Laboratory, Uppsala University, Box 534, SE-751 21 Uppsala, Sweden 4Philips Research Laboratories, High Tech Campus 4, Eindhoven, The Netherlands (Dated: August 14, 2019) We report on the magnetization dynamics of a square array of mesoscopic discs, fabricated from an iron palladium alloy film. The dynamics properties were explored using ferromagnetic resonance measurements and micromagnetic simulations. The obtained spectra exhibit features resulting from the interactions between the discs, with a clear dependence on both temperature and the direction of the externally applied field. We demonstrate a qualitative agreement between the measured and calculated spectra. Furthermore, we calculated the mode profiles of the standing spin waves excited during a time-dependent magnetic field excitations. The resulting maps confirm that the features appearing in the ferromagnetic resonance absorption spectra originate from the temperature and directional dependent inter-disc interactions. I. INTRODUCTION Arrays of closely packed mesoscopic magnets provide a rich playground for investigations of collective magne- tization dynamics. The (stray field induced) interaction between the discs forms a link between the internal mag- netization dynamics of the elements and the global re- sponse of the system1 -- 4. Magnetic discs are interesting in this context, due to the richness of internal magnetic textures and the absence of shape induced anisotropy in their plane. Discs of certain radius and height ratio exhibit a ground state referred to as a vortex 5,6 charac- terized by an in-plane magnetic flux closure. Since the magnetic moments are curling in-plane of the discs, the stray field from the discs is negligible when vortices are formed, in stark contrast to the collinear state 7 -- 10. The application of an external magnetic field drives the vortex core out of the center, towards the edge of the disc. At a given field, the vortex is annihilated and the magnetic moment is aligned parallel to the direction of applied field (collinear state). When the discs are in a collinear state, their stray fields result in inter-disc interactions. Pre- vious investigations of iron-palladium (Fe20Pd80) discs, arranged in a square array (See Figure 1) showed that a change of temperature is sufficient to alter the mag- netization dynamics10. Above a given temperature the energy barrier for switching from a vortex to the collinear state and vice versa was even found to be free from hys- teresis. As a consequence the magnetization state of the system under the certain applied field becomes bi-stable: the vortex and the collinear state have the same energy. Here we explore the effect of changes in the inter-disc interactions on the magnetization dynamics of soft mag- netic iron-palladium alloy discs, using ferromagnetic res- onance (FMR) and micro-magnetic simulations11. The changes in the inter-disc interactions are obtained by ro- tating the magnetisation of the discs as well as altering the stray field from the discs by changing the sample FIG. 1. Atomic force microscopy image of a square array of Fe20Pd80 alloy discs with the high symmetry directions indicated as [10] and [11] by arrows. temperature. II. METHODS A. Sample The investigated array consists of iron-palladium al- loy (Fe20Pd80) discs arranged in a square lattice as illus- trated in Figure 1. Each disc has a radius of 225 nm and a thickness of 10 nm, with a center-to-center distance of the discs equal to 513 nm in the [10] direction. This re- sults in an inter-disk distance of 53 nm and 275 nm along the [10] and along the [11] direction, respectively. A de- tailed description of the sample preparation is provided by Ostman et al. 10. 9 1 0 2 b e F 1 ] l l a h - s e m . t a m - d n o c [ 1 v 3 0 4 0 0 . 2 0 9 1 : v i X r a B. Ferromagnetic resonance measurements Magnetization dynamics of the Fe20Pd80 disc arrays were measured using X-band cavity FMR equipped with with variable-temperature sample holder. A static mag- netic field was applied in-plane, while a time-dependent spatially uniform (wave vector (cid:126)k=0) magnetic field ex- citation, with a frequency of 9.8 GHz was applied per- pendicular to the plane of the sample7,12. The static magnetic field was swept from 0 to 300 mT and the mea- surements were carried out at temperatures ranging from 80 K to 293 K. The strength of the applied static field ensured that discs were measured in the collinear state. A second set of FMR measurements were performed using a vector-network-analyzer (VNA) at a room tem- perature, by an in-plane field sweep, utilizing a copla- nar waveguide. The detailed description of this measure- ment setup is described by Wei et al. 13. The linewidth ∆Ht versus frequency f data obtained from these mea- surements were fitted to a linear function, extracting the Gilbert damping coefficient α. The extracted value for α was 1.8×10−2 and was later employed as a parameter in the micromagnetic simulations of the magnetization dynamics in the Fe20Pd80 disc array. It was adjusted in order to match the calculated FMR absorption peak amplitude to one measured experimentally. C. Micromagnetic simulations and standing spin wave map calculations Micromagnetic simulations were performed using MUMAX314. The exchange stiffness constant, Aex, defin- ing the inter-spin coupling in the magnetic material was adjusted to qualitatively reproduce the experimental ob- servations. It is known that Aex is temperature de- pendent and decreases with increasing temperature15. Therefore a higher temperature implies softer standing spin wave modes, excited within the discs. After the initial simulations, a value of Aex=3.36 pJ/m was used for the subsequent micromagnetic simulations, in order to qualitatively reproduce the experimentally observed FMR spectra features. A broader discussion and motiva- tion regarding this choice is provided by Ciuciulkaite 16. Finally, the Gilbert damping parameter, α, describing the losses in the system and proportional to the FMR absorption linewidth was chosen to be 1.9×10−2, after initially taking the value determined from the VNA-FMR measurements. In the micromagnetic simulations, four discs with 225 nm radius and 10 nm thickness were placed in 2-by-2 square lattice with a lattice parameter of 513 nm. The in-plane cell size was defined as 0.513(18)·lex, where lex is the exchange length, a material parameter determined by 14. The cell size along the z-direction, the Aex and Msat i.e. the thickness of the structure, was set to 5 nm for all simulations. Periodic boundary conditions (PBC)14 were applied in both lateral directions (PBCx=PBCy=3, 2 PBCz=0). The FMR simulations were performed ap- plying a static magnetic field in-plane of the lattice, re- laxing the system, and then applying a time-dependent field excitation out-of-plane, with an analytical expres- sion of A·sin(2πf t) or A·sinc(2πf t). The static mag- netic field is applied at an angular offset of 2◦from the principal in-plane directions, in order to lift any degen- eracy in the simulations, related to the high symme- try directions we will be investigating, being parallel to the [10] and [11] direction of the disc lattice. The am- plitude of the time-dependent excitation was A=5 mT, the frequency was f =9.8 GHz, the duration of the si- nusoidal time-dependent magnetic field excitation was 10 ns, and the sampling period was 1 ps. The fre- quency bandwidth for the sinc function excitation was 20 GHz. The temperature-dependent saturation magnetiza- tion, Msat(T ), was used to account for the temperature dependence of the FMR response of the soft magnetic discs. The Fe20Pd80 alloy has a Curie temperature, TC, of 463 K10. To a first approximation the temperature de- pendence of the magnetization can be described by the modified Bloch behaviour: (cid:20) 1 − (cid:18) T (cid:19)β(cid:21) TC Msat(T ) = Msat(0) , (1) where Msat(0) is the saturation magnetization at 0 K and is 5.9×105 A/m for the Fe20Pd80 alloy and β=1.6910. Here we would like to note that we do not account for any other temperature induced effects in our simulations than the temperature dependence of the saturation mag- netization. A complementary method for describing the observed features in the FMR spectra are standing spin wave (SSW) mode maps. The maps were calculated from the spatial micromagnetic evolution of the magnetization as excited by the time-dependent magnetic field. A fast Fourier transform (FFT) for each of the discrete spatial elements was performed, resulting in the spatial maps of the amplitude of the magnetization dynamics. III. RESULTS AND DISCUSSION A. FMR The FMR spectra measured at room temperature, with the static field applied along [10] and [11] in-plane directions of the array, as indicated in Figure 1, are shown in Figure 2. The ferromagnetic resonance shifts to slightly higher applied magnetic fields, when the field is applied along [10] direction as compared to the [11] direction. This shift reflects the effect of the inter-disc interaction strength, due to the change in the applied static magnetic field and hence the stray field direction. Furthermore, the measured FMR absorption spectra exhibits a split in the 3 FIG. 2. The measured FMR spectra of square arrays at room temperature with the field applied along [10] and [11] direc- tions. The shaded grey regions indicate characteristic FMR absorption features of the investigated structure: a shoulder feature, appearing before the main absorption peak, and a main FMR field, µ0Hres0. main absorption peak in the [10] direction. We will call this feature before the main absorption peak a shoulder feature (See Figure 2 in the following). The change in interaction strength is due to a modification of inter-disc coupling by changing the magnetisation direction of the discs. Effectively, along the [10] direction a disc has two nearest neighbours while along the [11] direction number of interacting nearest neighbours is doubled as described below. B. Micromagnetic simulations on a single disc Micromagnetic simulations of a single disc were car- ried out to identify the basic features of the elements, i .e. the response of the discs in absence of interactions. A FMR frequency versus applied static field map was calculated for a wide range of frequencies, using the ex- pression A·sinc(2πf t) for the time-dependent magnetic field excitation and is provided in Appendix A, Figure A.1. Since the FMR measurements were carried out at the frequency of 9.8 GHz, a line cut along this frequency was taken and is shown as an FMR absorption spectrum in Figure 3 (a). The shaded grey regions indicate the strength of the applied static magnetic field, for which the SSW modes excited in the disc were simulated and are presented as maps of normalized magnetization pre- cession amplitude and phase in Figure 3 (b) and in Ap- pendix B Figure B.1. Figure 3 (b) presents the SSW maps for an amplitude and phase for applied magnetic field corresponding to the region where the shoulder feature in Figure 2 is observed. It is indicated on the left-hand-side (LHS) of the main absorption peak as µ0HLHS in Figure 3 (a). This mode is mainly constrained near the middle of the disc and along a line parallel to the direction of the applied mag- FIG. 3. (a) FMR absorption spectrum for a single disc, taken as a linecut from the map in Figure A.1 (see Appendix A) at 9.8 GHz frequency. The shaded grey regions indicate external magnetic fields at which the SSW profiles (mz component, amplitude of precession) were calculated; (b) FFT amplitude and phase maps calculated at a shoulder feature at 197 mT applied magnetic field. netic field. In the remainder of this communication we will refer to the area in the middle of a disc as the central area. The largest precession takes place at two symmet- ric points outside of the disc center, along the direction of static magnetic field. In addition to these symmetry points, the magnetic moments also resonate at the edges of the disc along a line perpendicular to the magnetic field direction. In the following, we will refer to this mode as a perpendicular edge mode. To SSW modes appearing at the edges along a line parallel to the direction of the applied magnetic field, we will be referring to as parallel edge modes. The SSW spatial mode maps calculated at higher magnetic fields indicate the following spatial mode profiles: at 210 mT field - uniform precession, at 226 and 255 mT - edge modes (see Appendix B Figure B.1). C. Micromagnetic simulations of square arrays Subsequent simulations were carried out on square ar- rays containing interacting discs, but otherwise identical to the building block described in the previous section. In order to explain the shoulder feature in the FMR spec- tra for the [10] direction (Figure 2) and to further inves- tigate which SSW modes are excited in the discs, spatial amplitude and phase maps (see Figure 4 (b)-(c)) were 4 ified, due to effective increase of nearest neighbors along the [11] direction and the magnetic moment precession amplitude significantly increases in the center and per- pendicular edge areas. Simultaneously, the phase angles in the center and at the parallel edges of a disc decrease, while the overall phase angle distribution over each disc becomes more uniform as compared to the case when ex- ternal field is along the [10] direction. Furthermore, the amplitude of the perpendicular edge mode and the exten- sion over the edges is smaller for the [10] external field direction as compared to the [11] direction. This again is a hallmark of the effect the proximity to the nearest neighboring discs has on the magnetization dynamics. Comparison of the SSW mode amplitude and phase spa- tial maps, to those of an isolated single disc reveals that the amplitude of magnetic moments in the center of an isolated disc is smaller as compared to the same area in the arrays. These results clearly demonstrate the effect of the stray field coupling between discs in an array on the internal magnetization dynamics of the elements. Modifying the inter-disc interaction strength alters the magnetic mo- ment fluctuations at the center of a disc which results in appearance (or disappearance) of the shoulder feature in an FMR absorption spectrum. A further important observation arising from the simulations, is the absorp- tion modes appearing at higher fields after the main ab- sorption peak. These features, indicated as µ0Hres1 and µ0Hres2 at around 229 mT and 251 mT, respectively in Figure 4 (a)), are not observed in the experimental FMR spectra (Figure 2). They arise from the idealised circular shape of the discs in the simulations. In real samples the lithographically fabricated elements are imperfect, lead- ing to the absence of these modes17. The origin of spin wave maps for these simulated modes is discussed in more detail in the Appendix C. D. Ferromagnetic resonance response of iron-palladium arrays: comparison of experiments and simulations A comparison of the measured and calculated spectra is provided in Figure 5, where we display the change in the ferromagnetic resonance field with increasing tem- perature, for applied fields in the [10] and [11] directions. The strength of the interaction alters the resonance field as observed in the measured FMR spectra. Same qual- itative changes are observed in the micromagnetic sim- ulations. The resonance field increases as the temper- ature is increased (see Figure 5). Thermal fluctuations become more prominent with increasing temperature and a stronger magnetic field is needed to align the moments, shifting the resonance field to higher values. This trend is seen in both calculated and measured FMR spectra. At low temperatures the experimental and calculated reso- nance fields are comparable. At elevated temperatures the resonance fields calculated from micromagnetic sim- (a) Computed FMR absorption spectra with FIG. 4. Msat(T = 300K) and static magnetic field applied along the [10] and [11] directions. Grey shaded regions represent exter- nal magnetic fields at which SSW modes were calculated; (b) spatial amplitude and (c) phase maps of SSW modes calcu- lated for the two applied field directions. Color bars indicate normalized amplitude and phase angles in degrees; the bottom panels in (b) and (c) represent amplitude and phase linecuts in the respective maps along the [10] and [11] directions. calculated at external magnetic fields of 191 mT and 186 mT along the [10] and [11] directions, respectively. The shoulder feature on the left-hand-side of the FMR absorption peak, observed in the experimentally mea- sured spectra (Figure 2), is also present in the calculated results, when the external magnetic field is applied along the [10] direction (Figure 4 (a)). In this case, calculated SSW mode spatial maps show that the magnetic mo- ments are strongly out of phase at the parallel edge and center areas of the discs. When the direction is changed to [11], the interaction strength between the discs is mod- 5 tively becomes zero, while beyond that it seems to in- crease again. This behaviour potentially relates to the internal magnetization dynamics within the discs. As already reported previously by Ostman et al. 10, a tem- perature range exists where bi-stability is attainable for the collinear and vortex states, which should also be ac- companied by a strong modification in the magnetization dynamics of the individual discs. This range was found to be independent of applied field direction, for temper- atures above ≈ 220 K from measurement protocols in- volving very low frequencies (hysteresis curves, recorded employing the magneto-optical Kerr effect and field cy- cling of 0.4 Hz). In the present study, Hres[10] − Hres[11] becomes zero at ≈ 270 K, considerably higher than that in Ostman et al. 10. The time scale of dynamics, probed in this work, is much shorter (sub-ns, FMR measured at 9.8 GHz), hinting for this strong apparent tempera- ture shift. A more thorough survey of this effect would greatly benefit from more detailed micromagnetic simu- lations, incorporating also the magnetization dynamics of the material at all relevant length-scales (inter- and intra-disc). The shift in the resonance field is attributed to a dif- ference in the stray field induced inter-disc coupling, along the [10] and [11] directions. In Figure 6, we show the spatial maps of the demagnetizing field amplitude, computed for arrays with initial collinear magnetization states, along [10] and [11] directions. Having the mag- netisation of the elements along the [11] direction results in an increased stray field coupling between neighbor- ing discs. Clearly, a simple point-dipole-like approxima- tion is not sufficient to describe the effective magneto- static coupling. Even though the spacing between neigh- boring discs in the [10] direction is smaller - and thus one would expect a stronger coupling - the demagnetiz- ing field strength is distributed broadly along the disc perimeter for the [11] case (see the bottom panels in Fig- ure 6). This leads to an stronger coupling for the applied fields along the [11] direction. The root of these effects are clearly seen in Figure 7, where we provide spatial maps of static magnetization components perpendicular to the direction of applied magnetic field. A map of the my component with Hres[10], shows a dominant posi- tive direction for these, in accordance to the 2◦offset of Hres from the [10] direction in our simulations. When Hres[11], the magnetization components mxy perpen- dicular to applied fields, obtain non-zero values only at the discs' rim and with maxima at the positions where the gap to neighboring disc is minimum. In contrast to the case where Hres[10], now all of the four gaps re- lated to neighboring discs are active. The maps show that mxy ⊥ [11] is of the same sign at the disc edges along the [10] and [01] directions which further confirm- ing inter-disc coupling via stray fields. FIG. 5. Comparison of the resonance fields obtained via mi- cromagnetic simulations (red squares) and experimental val- ues (black squares) extracted from FMR measurements of the Fe20Pd80 alloy disc array. The external static magnetic field was applied along [10] and [11] directions (empty and filled symbols, respectively). Lines are guides to the eye. Error bars and resonance fields were obtained from fitting of FMR amplitude to a Lorentzian peak profile. ulations increase faster than the measured values. The micromagnetic simulations offer a qualitative agreement with the experimental data which is more clearly seen from a comparison of measured (Figure 2) and calcu- lated FMR absorption spectra (Figure 4(a)) at elevated temperatures. In Figure 5 we can further observe that the splitting between the simulated resonance field values along [10] and [11] directions, is always present in the investigated temperature range. The curves for the [10] direction lie higher in field, compared to those for the [11] direc- tion in the investigated temperature range. In related studies, where neighboring elongated micromagnetic ele- ments were antiferromagnetically coupled by stray fields, resonance field shifts depending on the strength of the interaction, were also reported18,19. However, the dif- ference in resonance field for [10] and [11] directions (Hres[10]− Hres[11]) calculated from micromagnetic FMR results (by fitting amplitude to a Lorentzian profile peak and extracting its position), follows a different trend as compared to the resonance field extracted from exper- imentally obtained data (Figure 5). This can be at- tributed to the fact that while exchange stiffness and Gilbert damping parameters are temperature dependent, in micromagnetic simulations we kept these parameters constant for all temperatures. Furthermore, no thermal fluctuations were accounted for in micromagnetic simu- lations. Finally, we would like to comment upon the scaling of the experimental resonance field difference Hres[10] − Hres[11] versus temperature. The difference decreases monotonically up to some temperature, where it effec- nance field as well as the fine structures of the resonance. The inter-island interaction is found to be larger when 6 FIG. 7. Spatial maps of magnetization components distribu- tion perpendicular to the direction of the external magnetic field. Maps were obtained from micromagnetic simulations of arrays with initial magnetization along [10] and [11] direc- tions relaxed in a static magnetic field corresponding to Hres in magnitude for each of the directions. the magnetisation of the islands are in the [11] as com- pared to [10] principal directions. The origin of this effect arises from the transverse component of the magnetisa- tion, which enhances the inter-island interactions in the transverse directions, illustrating the need of including inner magnetic structure of the magnetisation to explain both dynamic and static results. The results presented in this study, show that mi- cromagnetic simulations are suitable for investigations of spatial profiles of SSW modes, excited in arrays of stray field coupled nanomagnets. These can be uti- lized for analyzing and designing the microwave response of extended arrays of thermally active and interacting nanomagnets20 -- 22, having interesting topological and in- teraction schemes, such as artificial spin ices1,3,4,23 -- 26. The latter could enable the design and fabrication of re- configurable magnonic devices3,24. FIG. 6. Spatial distribution maps of demagnetizing field strengths, obtained from micromagnetic simulations of ar- rays with initial magnetization along [10] and [11] directions (first and second row, respectively) relaxed in a 0 mT static magnetic field and a field corresponding to the ferromag- netic resonance field (left and right columns, respectively). Bottom panels represent polar plots of the demagnetizing field, along the rim of a disc, as indicated by a circle in the "m[11]µ0H= 0 mT" map. ACKNOWLEDGMENTS IV. CONCLUSIONS We have investigated the magnetization dynamics in a square array of Fe20Pd80 alloy discs by means of FMR measurements and micromagnetic simulations. The fer- romagnetic resonance is found to increase in field with in- creasing temperature which we attribute to the decrease of the magnetic moment. The results were qualitatively reproduced and confirmed by micromagnetic simulations. The effects of stray field induced interaction of the is- lands is seen in the orientation dependence of the reso- The authors would like to acknowledge financial sup- port from the Swedish Research Council (VR), the Swedish Foundation for International Cooperation in Re- search and Higher Education (STINT) and the Knut and Alice Wallenberg Foundation project "Harness- ing light and spins through plasmons at the nanoscale" (2015.0060). This work is part of a project which has re- ceived funding from the European Union's Horizon 2020 research and innovation programme under grant agree- ment no. 737093. 7 Nat. Nanotechnol 9, 514 (2014). 22 M. S. Andersson, S. D. Pappas, H. Stopfel, E. Ostman, A. Stein, P. Nordblad, R. Mathieu, B. Hjorvarsson, and V. Kapaklis, Sci. Rep 6, 37097 (2016). 23 S. Gliga, A. K´akay, R. Hertel, and O. G. Heinonen, Phys. Rev. Lett 110, 117205 (2013). 24 E. Iacocca, S. Gliga, R. L. Stamps, and O. Heinonen, Phys. Rev. B 93, 134420 (2016). 25 X. Zhou, G.-L. Chua, N. Singh, and A. O. Adeyeye, Adv. Funct. Mater 26, 1437 (2016). 26 E. Ostman, H. Stopfel, I.-A. Chioar, U. B. Arnalds, A. Stein, V. Kapaklis, and B. Hjorvarsson, Nat. Phys 14, 375 (2018). 27 M. Kammerer, M. Weigand, M. Curcic, M. Noske, M. Sproll, A. Vansteenkiste, B. Van Waeyenberge, H. Stoll, G. Woltersdorf, C. H. Back, and G. Schutz, Nat. Commun 2 (2011). ∗ agne.ciuciulkaite@physics.uu.se † vassilios.kapaklis@physics.uu.se 1 M. Krawczyk and D. Grundler, J. Phys. Condens. Matter 26, 123202 (2014). 2 C. Nisoli, V. Kapaklis, and P. Schiffer, Nat. Phys 13, 200 (2017). 3 V. S. Bhat and D. Grundler, Phys. Rev. B 98, 174408 (2018). 4 M. B. Jungfleisch, W. Zhang, E. Iacocca, J. Sklenar, J. Ding, W. Jiang, S. Zhang, J. E. Pearson, V. Novosad, J. B. Ketterson, O. Heinonen, and A. Hoffmann, Phys. Rev. B 93, 100401 (2016). 5 T. Shinjo, T. Okuno, R. Hassdorf, K. Shigeto, and T. Ono, Science 289, 930 (2000). 6 R. P. Cowburn, D. K. Koltsov, A. O. Adeyeye, M. E. Welland, and D. M. Tricker, Phys. Rev. Lett. 83, 1042 (1999). 7 V. Castel, J. Ben Youssef, F. Boust, R. Weil, B. Pigeau, G. de Loubens, V. V. Naletov, O. Klein, and N. Vukadi- novic, Phys. Rev. B 85, 184419 (2012). 8 O. Heinonen, Phys. Rev. B 92, 054420 (2015). 9 J. Ding, P. Lapa, S. Jain, T. Khaire, S. Lendinez, W. Zhang, M. B. Jungfleisch, C. M. Posada, V. G. Yefre- menko, J. E. Pearson, A. Hoffman, and V. Novosad, Sci. Rep 6, 25196 (2016). 10 E. Ostman, U. B. Arnalds, E. Melander, V. Kapaklis, G. K. P´alsson, A. Y. Saw, M. A. Verschuuren, F. Kronast, E. T. Papaioannou, C. S. Fadley, and B. Hjorvarsson, New J. Phys 16, 053002 (2014). 11 R. Dutra, D. E. Gonzalez-Chavez, T. L. Marcondes, R. L. Sommer, S. O. Parreiras, and M. D. Martins, Phys. Rev. B 99, 014413 (2019). 12 M. Hanze, C. F. Adolff, B. Schulte, J. Moller, M. Weigand, and G. Meier, Sci. Rep 6, 22402 (2016). 13 Y. Wei, R. Brucas, K. Gunnarsson, I. Harward, Z. Celinski, and P. Svedlindh, J. Phys. D 46, 495002 (2013). 14 A. Vansteenkiste, J. Leliaert, M. Dvornik, M. Helsen, F. Garcia-Sanchez, and B. Van Waeyenberge, AIP Ad- vances 4, 107133 (2014). 15 M. Mulazzi, A. Chainani, Y. Takata, Y. Tanaka, Y. Nishino, K. Tamasaku, T. Ishikawa, T. Takeuchi, Y. Ishida, Y. Senba, H. Ohashi, and S. Shin, Phys. Rev. B 77, 224425 (2008). 16 A. Ciuciulkaite, Micromagnetic simulations of magneti- zation dynamics in iron-palladium nanostructure arrays, Master's thesis, Department of Physics and Astronomy, Uppsala University (2016), http://www.diva-portal. org/smash/record.jsf?pid=diva2%3A968632. 17 M. B. Jungfleisch, J. Sklenar, J. Ding, J. Park, J. E. Pear- son, V. Novosad, P. Schiffer, and A. Hoffmann, Phys. Rev. Appl. 8, 064026 (2017). 18 M. Demand, A. Encinas-Oropesa, S. Kenane, U. Ebels, I. Huynen, and L. Piraux, J. Magn. Magn. Mater 249, 228 (2002), international Workshop on Magnetic Wires. 19 N. Kuhlmann, A. Vogel, and G. Meier, Phys. Rev. B 85, 014410 (2012). 20 A. Farhan, P. M. Derlet, A. Kleibert, A. Balan, R. V. Chopdekar, M. Wyss, L. Anghinolfi, F. Nolting, and L. J. Heyderman, Nat. Phys 9, 375 (2013). 21 V. Kapaklis, U. B. Arnalds, A. Farhan, R. V. Chopdekar, A. Balan, A. Scholl, L. J. Heyderman, and B. Hjorvarsson, 8 Appendix A: Dispersion relation of a single FePd disc The complete map of available resonance modes for a single nanodisc resonator is shown in Figure A.1 as a fre- quency vs external magnetic field map. The color bar represents the log FFT amplitude of spin precession. Be- low the annihilation field, the spectrum is dominated by low frequency modes, corresponding mostly to the gy- rotropic motion of the vortex core27. The highest inten- sity peak corresponding to a vortex mode shows similar dispersion relation as that for isolated cylinders exhibit- ing single domain state18. At stronger external magnetic fields above the saturation field, when the disc is in a collinear magnetic state, a significant increase of the res- onance frequency is observed and highest intensity peak follows Kittel-like behaviour of resonance frequency for continuous ferromagnetic films. Below the main reso- nance peak are two smaller intensity features correspond- ing to the edge modes. FIG. A.1. The calculated map of the log and normal- ized FMR amplitude response to a 20 GHz bandwidth sinc function form magnetic field excitation, for different external static fields. Appendix B: Magnetization maps of single disc SSW modes SSW maps calculated at the resonance field, which is 210 mT for a single disc, reveal that the area of con- strained magnetic moments becomes more concentrated at the edges along the direction of an applied magnetic field. Meanwhile, magnetic moment fluctuations become more intense at the perpendicular edges. In contrast, the SSW modes calculated at 226 mT and 255 mT external magnetic fields appear at the parallel edges. The spa- tial log normalized FFT amplitude and phase maps cal- culated at fields just below the ferromagnetic resonance field and above it are shown in Figure B.1. The evolu- tion of parallel edge modes can be observed. Just before and after the ferromagnetic resonance (at 205 mT and FIG. B.1. The calculated spatial log and normalized FFT amplitude and phase maps of an isolated single disc at differ- ent fields indicated in Figure 3 (a). 216 mT fields, respectively), spin magnetic moment fluc- tuation amplitude is uniform throughout the whole area of a disc except for the parallel edges. At the additional absorption features observed after the FMR at 232 mT and 256 mT fields, respectively, fluctuation amplitude becomes larger at the edges while throughout the rest of a disc, it goes to zero, in other words, the rest of the spins "freeze in". Appendix C: Magnetization maps of edge modes in an array In this section we discuss the resonance and edge modes observed in the calculated FMR spectra in Fig- ure 4 (a). At the resonance field, indicated by an arrow 9 at 216 mT field in Figure 4 (a), the log FFT magneti- sation amplitude of spin magnetic moments is highest at the perpendicular edge area as in the single disc case but with the constrained magnetic moment area reduced (compare Figure B.1 amplitude maps of modes at 205 and 216 mT fields with a top panel in Figure C.1. When the external magnetic fields are approximately 229 mT and 250 mT, the parallel edge modes are excited. Mode maps reveal that in this case the magnetic moments fluc- tuate the most at the edges along disc coupling direction C.1. This hints that disc interaction occurs through the fluctuating magnetic moments. As a result, along [10] direction the system becomes less stiff and resonance oc- curs at lower fields when the field is applied along [10] than along [11] direction as can be seen in the absorp- tion spectra in Fig. 4 (a) at around 229 and 251 mT fields. Spatial maps of SSW modes calculated for the FIG. C.1. Fe20Pd80 alloy disc array at the field values indicated by shaded grey regions in Figure 4 (a). In the real arrays such modes are absent, due to shape imperfections and edge roughness of the discs arising from a lithographic fabrication method. Furthermore, due to computational limitations we calculated signifi- cantly smaller amount of nanodiscs than the real sam- ple actually contains17. This is also further supported by micromagnetic simulations performed accounting for shape and edge imperfections which do not reproduce these modes16. 

1912.09177

1

1912

Effects of electron-hole asymmetry on electronic structure of helical edge states in HgTe/HgCdTe quantum wells

We study the effects of electron-hole asymmetry on the electronic structure of helical edge states in HgTe/HgCdTe quantum wells. In the framework of the four-band kp-model, which takes into account the absence of a spatial inversion centre, we obtain analytical expressions for the energy spectrum and wave functions of edge states, as well as the effective g-factor tensor and matrix elements of electro-dipole optical transitions between the spin branches of the edge electrons. We show that when two conditions are simultaneously satisfied -- electron-hole asymmetry and the absence of an inversion centre -- the spectrum of edge electrons deviates from the linear one, in that case we obtain corrections to the linear spectrum.

cond-mat.mes-hall

cond-mat

Effects of electron-hole asymmetry on electronic structure of helical edge states in HgTe/HgCdTe quantum wells M. V. Durnev Ioffe Institute, 194021 St. Petersburg, Russia We study the effects of electron-hole asymmetry on the electronic structure of helical edge states in HgTe/HgCdTe quantum wells. In the framework of the four-band kp-model, which takes into account the absence of a spatial inversion centre, we obtain analytical expressions for the energy spectrum and wave functions of edge states, as well as the effective g-factor tensor and matrix elements of electro-dipole optical transitions between the spin branches of the edge electrons. We show that when two conditions are simultaneously satisfied (cid:22) electron-hole asymmetry and the absence of an inversion centre (cid:22) the spectrum of edge electrons deviates from the linear one, in that case we obtain corrections to the linear spectrum. I. INTRODUCTION The study of helical edge states emerging at the edge of two-dimensional topological insulators is an important area of physics of two-dimensional crystalline systems with nontrivial topological properties [1 -- 3]. The key ar- eas of research include the study of local and nonlocal electron transport through edge channels [4 -- 7], backscat- tering mechanisms [8 -- 14] and photogalvanic effect [15, 16]. Among various systems, where one-dimensional helical channels are experimentally observed [2, 3, 17], HgTe/HgCdTe quantum wells attract the most attention. In such wells, the transition between the trivial and topo- logical phases occurs upon variation of the quantum well width. Electronic states in HgTe/HgCdTe quantum wells of close-to-critical width are usually obtained in the frame- work of the four-band kp-model, which includes closely lying electron and hole subbands. The isotropic modi- fication of this model is called Bernevig-Hughes-Zhang (BHZ) model [1] and is widely used for calculations of the electronic structure of the bulk and edge states [18 -- 25]. However, as shown by the atomistic calcula- tions [26, 27], the absence of a center of spatial in- version in the zinc blende lattice and the low symme- try of the HgTe/HgCdTe quantum well heterointerfaces lead to strong mixing of the electron and hole subbands, which modifies the kp-model [19, 27, 28] and leads to a substantial rearrangement of bulk and edge electronic states [27, 29, 30]. The absence of a center of spatial in- version in the HgTe/HgCdTe quantum well is responsi- ble, for example, for the emergence of optical transitions between helical states with opposite spin in the frame- work of the strong electro-dipole mechanism [31]. Both in the isotropic BHZ model and in the noncen- trosymmetric kp-model, there are diagonal terms pro- portional to the square of the wave vector and related to the presence of remote energy bands. These con- tributions are different for electron and hole subbands, which leads to violation of electron-hole symmetry -- the Hamiltonian does not coincide with itself after replac- ing an electron with a hole and simultaneously changing the sign of the energy. Although the terms quadratic in wave vector do not directly affect the group velocity and localization width of the edge states (these quan- tities are controlled by the large inter-subband mixing, which is linear in wave vector), electron-hole asymme- try encoded in diagonal terms, leads to significant mod- ification of the energy spectrum and wave functions of edge states. Electron-hole asymmetry may also result in new effects. For instance, it is responsible for the ap- pearance of circular photogalvanic effect in helical edge channels [15, 24, 31, 32]. The influence of electron-hole asymmetry on the spectrum and structure of edge states has been studied earlier in the framework of the BHZ model (see, for example, works [24, 25]). However, such studies were not carried out for realistic quantum wells that do not have a center of spatial inversion. In this paper we study the effects of electron-hole asymmetry on helical edge states in HgTe/HgCdTe quan- tum wells. In particular, we analyze the dispersion of edge electrons and derive expressions for the wave func- tions of edge states in the framework of noncentrosym- metric kp- model, we calculate the components of the g-factor tensor of edge electrons and the matrix elements of optical transitions between the spin branches of edge channel. The paper is organised as follows: in Sec. II a general model is presented and expressions for the en- ergy and wave functions of the edge states are obtained for a zero wave vector along the edge; in Sec. III the g-factor tensor of edge electrons is calculated; in Sec. IV the spectrum and wave functions of edge states are found in a wide range of wave vectors; in Sec. V the matrix el- ements of optical transitions between the spin branches of edge states are calculated; and finally, in Sec. VI we discuss the effect of boundary conditions on the obtained results. II. EDGE STATES AT ZERO WAVE VECTOR In this section we formulate the model and obtain wave functions of helical states at zero wave vector of the mo- tion along the edge. Let us consider a HgTe/CdHgTe quantum well grown along the [001] crystallographic di- rection with a width, which is close to the critical width, 9 1 0 2 c e D 9 1 ] l l a h - s e m . t a m - d n o c [ 1 v 7 7 1 9 0 . 2 1 9 1 : v i X r a when the transition to a topological phase occurs. Such a structure possesses D2d point symmetry lacking the center of a spatial inversion. The states in the vicin- ity of the Fermi level are formed from the close-in- energy electron-like and hole-like subbands E1,±1/2(cid:105) 2 and H1,±3/2(cid:105), respectively [1]. In the E1, +1/2(cid:105), H1, +3/2(cid:105), E1,−1/2(cid:105), H1,−3/2(cid:105) basis the quantum well states are described by the following Hamilto- nian [29]:  δ0 − (B + D)k2 −iAk− −iγe2iθ 0 H0(kx, ky) = −δ0 + (B − D)k2 iAk+ −iγe2iθ 0 δ0 − (B + D)k2 0 iγe−2iθ iAk+  . (1) iγe−2iθ −iAk− 0 −δ0 + (B − D)k2 Here k = (kx, ky) is the electron in-plane wave vector, k = k, k± = kx ± iky, A, B, D, δ0 and γ are real band structure parameters. At B < 0 (B > 0) and δ0 < 0 (δ0 > 0) the quantum well is in the topological insulator phase, and its' edges support helical edge states. The γ parameter takes into account the absence of a spatial in- version center in the structure, and results mainly from the mixing of E1,±1/2(cid:105) and H1,∓3/2(cid:105) subbands at the quantum well interfaces [27]. The value of this parameter is not yet determined experimentally, and theoretical val- ues lie in the range 2 ÷ 5 meV for a HgTe/Hg0.3Cd0.7Te quantum well [19, 26 -- 28]. In what follows, we use a coordinate system with the z-axis that coincides with the growth axis of the quan- tum well, the edge of the sample is parallel to the y axis, and the sample occupies the half-space x > 0, see Fig. 1. The y axis makes the angle θ with the [010] crystallographic axis, which allows one to consider struc- tures with different crystallographic orientations of the edge. In particular, at θ = 0 the edge of the sam- ple is parallel to the [010] axis, while at θ = π/4 the edge is parallel to the [110] axis. While obtaining ana- lytical expressions and analysing the results, we assume that B < 0, δ0 < 0, D < B, A > 0 and γ > 0. For numerical estimates, we use the set of parameters A = 3.6 eV·A, B = −68 eV·A2, D = −51 eV·A2 [19], γ = 5 meV [27], and δ0 = −10 meV corresponding to a HgTe/Hg0.3Cd0.7Te quantum well with a 8 nm width. The edge of topological Figure 1. insulator based on HgTe/CdHgTe quantum well with a sketch of helical edge states. Diagonal terms in the Hamiltonian (1), proportional to k2, arise due to the mixing of the four considered sub- where bands with remote subbands that are not included in the Hamiltonian. This mixing is described by the parameters B and D. As seen, in the case of D (cid:54)= 0 the diagonal terms enter asymmetrically for the electron and hole subbands, and therefore break the symmetry of the Hamiltonian with respect to the replacement of an electron by a hole. Hence, the ratio D/B can be regarded as the electron- hole asymmetry parameter. As will be shown below, the spectrum and wave functions of the edge states will be largely determined by this parameter. For the mentioned parameters, D/B ≈ 0.75. The edge states are derived from the solution of the (cid:18) Schrodinger equation −i H0 ∂ ∂x (cid:19) , ky ψkys = εkysψkys (2) with the boundary conditions ψkys(x = 0) = 0 и ψkys(x → +∞) = 0. For each value of ky there ex- ist a pair of such states with different pseudospin s = ±1/2. We use here boundary conditions of the simplest form, other types of boundary conditions are discussed in Sec. VI. First, let us consider ky = 0. We will now show that in this case there exist an analytical solution of Eq. (2) with an energy For this purpose we substitute energy (3) in Eq. (2) and write the sought wave functions in the form ε0s = −δ0 D B .  ,  , a(x)−αa(x) −ib(x)e2iθ −iαb(x)e2iθ   −ib(x)e−2iθ (cid:114)B + D iαb(x)e−2iθ a(x) αa(x) B − D . ψ0+1/2 = ψ0−1/2 = eikyy√ 2L eikyy√ 2L α = (3) (4) (5) yx[100][010]After such an ansatz the equations for a(x) and b(x) are analogous to the equations at D = 0. Using the solution at D = 0 [29] we obtain e−x/l1 cos e−x/l1 sin a(x) = N(cid:104) b(x) = N(cid:104) − e−x/l2 cos + e−x/l2 sin (cid:17)(cid:105) (cid:17)(cid:105) (cid:16) (cid:16) , (6) k0x − ϕ 2 k0x − ϕ 2 ϕ 2 ϕ 2 , where N = 2(cid:112)l2(1 + α2) , (7) A , l2 = − A κδ0 l1 = −κB and , k0 = γ A , tan ϕ = − γ κδ0 , (8) √B2 − D2 κ = . B (9) To derive (6) we used the relation l1 (cid:28) l2, which is valid for realistic quantum wells. Indeed, it follows from (8), that l1 ≈ 10 A, l2 ≈ 540 A for above mentioned parameters. In what follows we assume that l1 (cid:28) l2 relation is fulfilled. Equations (3)-(9) generalise results of Ref. [32] (D (cid:54)= 0, γ = 0) and Ref. [29] (D = 0, γ (cid:54)= 0). It is seen from the derived relations, that despite the small values of B and D (which manifests itself in small l1), all the results depend significantly on D/B. The key consequence of electron-hole asymmetry is the shift of Dirac point (ε0s) from the middle of the gap (ε = 0). As seen from (3), the Dirac point shift is pro- portional to D/B: at D < 0 the Dirac point shifts to- wards the conduction band, whereas at D > 0 -- towards the valence band. The shift of the Dirac point leads to a redistribution of the relative contributions of the sub- bands E1,±1/2(cid:105) and H1,±3/2(cid:105) to the wave functions, increase of the decay length l2, and change of the edge In the limiting cases D = ±B, the electrons velocity. Dirac point falls on the boundary of the energy gap, and the edge states at ky = 0 hybridise with the bulk ones. As follows from (8), in this case l2 → ∞. By projecting the part of the Hamiltonian (1) with ky y terms onto the wave functions (4) we obtain the and k2 energy of the edge electrons up to linear in ky terms: εkys = −δ0 where the edge velocity A  v0 = D B + 2sv0ky , (cid:112)κ2δ2 δ0κ2 0 + γ2 . (10) (11) III. ZEEMAN EFFECT FOR EDGE STATES In the framework of kp-model the interaction of elec- tron with magnetic field B is described by the sum of 3  ,  g⊥ e Bz 0 (cid:107) g e B+ 0 the orbital contribution, given by the Hamiltonian (1) with the Peirels substitution H0[k − (e/c)A], where A is the vector potential of magnetic field, and the Zeeman contribution [29] HZ = µB 2 0 g⊥ h Bz 0 (cid:107) e B− g 0 −g⊥ e Bz 0 0 (cid:107) he−4iθB+ g 0 −g⊥ h Bz e , g (cid:107) e, g⊥ (cid:107) he4iθB− g (12) h are the g-factors of E1(cid:105) and where g H1(cid:105) subbands, that contain contributions from the mix- ing with remote electron and hole subbands, B± = Bx ± iBy, and µB is the Bohr magneton. (cid:107) h and g⊥ The interaction of edge electrons with magnetic field is described by the following effective Hamiltonian in the basis (ψ0+1/2, ψ0−1/2): H(B) edge = µB 2 α,β=x,y,z gαβσαBβ , (13) (cid:88) where gαβ are the components of the g-factor tensor of edge electrons, and σx, σy, σz are the Pauli matrices. By projecting the Hamiltonian (12) onto the wave func- tions (4), we obtain the components of the g-factor tensor for B lying in the quantum well plane gxx = g1 cos2 2θ + g2 sin2 2θ , gyy = g1 sin2 2θ + g2 cos2 2θ , (g1 − g2) sin 4θ , gxy = gyx = 1 2 where g1 = g2 = 1 2 1 2 (cid:18)B − D (cid:18)B − D B g(cid:107) B g(cid:107) e + e − B + D B g B + D B g (cid:107) h (cid:107) h (cid:19) (cid:19) , (cid:112)κ2δ2 δ0κ 0 + γ2 (14) (15) . Equation (15) generalises results of [29] for D (cid:54)= 0. (cid:107) Since the heavy-hole g-factor g h is close to zero in quan- tum wells made of materials with a zinc-blende lat- tice [33], the main contribution to g1 and g2 is given by the first terms in (15). In real structures, the (B − D)/B factor may strongly deviate from unity, for example, for the parameters listed in Sec. II, (B − D)/B ≈ 1/4, hence the g-factors for the in-plane magnetic field are decreased by about four times as compared with D = 0. Estimates give g1 ≈ 2.5, g2 ≈ 2, consistent with the results of nu- merical modeling of the edge electrons spectrum in [31]. Magnetic field directed normal to the quantum well mixes the E1(cid:105) and H1(cid:105) subbands leading to to large orbital contribution to the gαz components. The diagonal component gzz, similarly to the case of D = 0, depends on the vector potential gauge and can be set to be zero [29], while the off-diagonal components are gauge-independent and have the following form: gxz = −g3 sin 2θ, gyz = g3 cos 2θ , (16) where 2m0A2 2 g3 = γδ0κ2 κ2 + γ2)3/2 (δ2 0 Estimations give g3 ≈ 130, and thus, g3 (cid:29) g1, g2. Magnetic field mixes edge states and opens a gap in its spectrum. This energy gap at θ = 0 is (cid:113) εg = µB g2 1B2 x + (g2By + g3Bz)2 . (18) a(x, ky) = . (17) ψ(0) ky−1/2 = eikyy√ 2L 0 0 a(x,−ky) αa(x,−ky)   . e−x/l1 − e−x/l2(ky)(cid:105) (cid:104) 4 , (21) Here q = D/B, function a has the form [see Eq. (6) at γ = 0] (cid:112)l2(ky)(1 + α2) 2 As follows from (18), the gap opens for any direction of the field, except when the field lies in the yz-plane and is directed at an angle − arctan g2/g3 with respect to y- axis. We note, that for any edge orientation (any θ) there exists a direction in space, when applied magnetic field does not open a gap. Giant anisotropy of the g-factor is confirmed in the magneto-transport experiments. It was shown that per- pendicular magnetic field suppresses edge conductivity due to the opening of the Zeeman gap, whereas the influ- ence of the in-plane magnetic field on the conductivity is much weaker [7] . IV. EDGE STATES AT NONZERO WAVE VECTOR In order to study electron transport and optical tran- sitions involving edge electrons, it is necessary to know the wave functions and the spectrum of edge states for a nonzero wave vector of motion along the edge. In this section, we find the spectrum and wave functions of edge states for ky (cid:54)= 0. In the framework of the isotropic model corresponding to γ = 0 in the Hamiltonian (1), and also in the framework of the model without an in- version center, but possessing electron-hole symmetry (γ (cid:54)= 0, D = 0), analytical expressions are derived for the spectrum and wave functions in the entire range of wave vectors. In a more general case, approximate re- sults are obtained that are valid for a small value of D/B or γ/δ0. with the edge states localization width l2 that depends on the wave vector l2(ky)−1 = l−1 2 − qky − k2 yl1 , (22) and l1 and l2 are given by Eqs. (8). The derived expres- sion for a(x, ky) is valid, as before, at l1 (cid:28) l2(ky). The yl1 term is small in the range of the studied ky, and thus, k2 is neglected in the following. Important consequence of the electron-hole asymmetry (q (cid:54)= 0) is the dependence of the edge localization length on ky. As shown below, this dependence leads to nonzero matrix element of the electric dipole operator between the edge states. (cid:113) 0 + A2k2 δ2 Let us consider the case q > 0 in more detail. As follows from Eq. (22), the length l2(ky) diverges at ky = k∗ = 1/(ql2) for the edge state with s = +1/2 and at ky = −k∗ for the s = −1/2 state. One may show that at ky = ±k∗ dispersion curves of the edge states (19) touch the lower boundary of the conduction band εc = y. At these points the edge states "merge" with the bulk (2D) states. However, note that the elastic scattering from the edge states to the conduc- tion band comes to play significantly earlier, already at εkys = −δ0, i. e., ky = 1/l2(1 + q). While moving along the edge branches towards the valence band the l2(ky) length decreases. At εkys = δ0, when the energy of the edge states coincides with the maximum energy of the valence band one has l2(ky) = l2(1 − q). At subsequent increase of the wave vector l2(ky) may become compara- ble with l1, and Eqs. (21), (22) are not valid in this case. This situation is analyzed in detail in Ref. [25]. A. Isotropic model B. Model without inversion center First, let us consider the case γ = 0, which corre- sponds to the isotropic model. Detailed study of the edge states structure within isotropic approximation is presented in [25]. In this case the Hamiltonian (1) is composed of two independent 2×2 blocks. At γ = 0 the Schrodinger equation (2) has an analytical solution, valid for a wide range of ky [20, 25] kys = −qδ0 + 2sκAky , ε(0)  a(x, ky) −αa(x, ky)  , (19) (20) 0 0 ψ(0) ky+1/2 = eikyy√ 2L Let us now consider the model of realistic quantum wells lacking the center of spatial inversion. This model is described by the Hamiltonian (1) with γ (cid:54)= 0. We will derive analytical results in two limits, in the limit γ/δ0 (cid:28) 1, when anti-diagonal terms of the Hamilto- nian (1) can be considered as a small perturbation, and in the limit D/B (cid:28) 1, i.e., in the limit of weak electron- hole asymmetry. 1. The limit γ/δ0 (cid:28) 1 In this section we consider anti-diagonal terms of the Hamiltonian (1) as a small perturbation. Non-perturbed wave functions ψ(0) ky±1/2 of the edge states are found in Sec. IV A for arbitrary ky, see Eq. (20). Note that in the first order of perturbation theory over γ the pertur- bation does not couple or shift in energy the edge states ψ(0) ky±1/2. It means that the corrections to the edge states wave functions are due to the admixture of bulk states of conduction or valence band to the edge states. Hence, in the first order of perturbation theory over γ the energy of edge states is not changed and coincides with Eq. (19). We will seek the wave functions of the edge states in the following form (23) a1(x, ky) −αa2(x, ky) −ib1(x, ky)e2iθ −iαb2(x, ky)e2iθ  ,   ,  −ib1(x,−ky)e−2iθ iαb2(x,−ky)e−2iθ a1(x,−ky) αa2(x,−ky) ψky+1/2 = eikyy√ 2L ψky−1/2 = eikyy√ 2L where b1 and b2 are unknown functions ∝ γ. In the first order of perturbation theory a1 and a2 are equal and coincide with the non-perturbed wave function (21). The function ψky−1/2 is related to ψky+1/2 by the time inversion. Eq. (2) with non-perturbed energy (19) leads to the following set of equations on b1 and b2: 2 +(cid:2)l2(ky)−1 + ky 1 +(cid:2)l2(ky)−1 − ky (cid:3) b1 + kyb2 = −k0a , (24) (cid:3) b2 − kyb1 = −k0a . l1b(cid:48)(cid:48) l1b(cid:48)(cid:48) 1 + b(cid:48) 2 + b(cid:48) The solution of these equations that satisfies the bound- ary conditions has the form b1 = b(x) + B(x) , b2 = b(x) − B(x) , (25) where b(x) = and (cid:26) N (ky)k0l2(ky) 2 e−x/l1 + × × (cid:20) 2x l2(ky) (cid:21) − 1 (cid:27) , e−x/l2(ky) B(x) = −N (ky)k0kyl2(ky) xe−x/l2(ky) , (26) N (ky) = (cid:112)l2(ky)(1 + α2) 2 . At ky = 0 the functions b1 and b2 are equal and coincide with Eq. (6), taken in the limit γ/δ0 (cid:28) 1. The pertur- bation theory is valid when b1,2 are small compared to a, i.e. at k0l2(ky) (cid:28) 1. It limits the range of ky, over which the derived equations can be applied. Using corrections to the wave functions, obtained in the first order of perturbation theory over γ/δ0, we will 5 derive the second order corrections to the edge states energy: (cid:68) (cid:69) ε(2) kys = kys Vγ ψ(1) ψ(0) kys , (27) where ψ(0) kys are the wave functions of the zero order (20), ψ(1) kys is the linear-in-γ part of the wave functions (23), and Vγ is the anti-diagonal part of the Hamiltonian (1). Calculations show that kys = − γ2 ε(2) 2δ0 kyl2 (1 − qkyl2)2 . (28) Equation (28) contains quadratic in γ correction to the edge velocity, as well as, for q (cid:54)= 0, the terms of higher powers in ky. Hence, when two conditions are simul- taneously fulfilled, electron-hole asymmetry and the ab- sence of spatial inversion center (q (cid:54)= 0 and γ (cid:54)= 0), the spectrum of edge electrons deviates from the linear one. As before, Eq. (28) is valid when k0l2(ky) (cid:28) 1, i.e. at 1 − qkyl2 (cid:29) k0l2. Figure 2 shows the results of numerical calculation of edge and bulk states in the HgTe/CdHgTe well and com- parison with the obtained analytical dependences. It can be seen that the spectrum of edge electrons deviates from the linear one with a velocity (11), and this deviation is well described by the dependence εkys = ε(0) kys. Cor- rections to the linear dispersion of edge electrons were also studied in [25] in the framework of the isotropic model, but with boundary conditions of a more complex form. The nonlinearity of the spectrum leads, for exam- ple, to the generation of edge photocurrent in the edge channels due to indirect optical transitions [16]. kys +ε(2) Figure 2. Energy spectrum of electronic states in HgTe/HgCdTe quantum well for band parameters shown in the text and γ = 2 meV, so that γ/δ0 = 0.2. The edge dis- persion branches are shown by blue and red. Panel b) shows increased part of the edge dispersion. Solid lines are numeric calculations, dashed lines are linear dependences with veloc- ity calculated by (11), dashed-dotted lines are dependences ky s, which take into account deviation from the linear ε(0) ky s + ε(2) behaviour. Energy (meV)−10−5051015Wave vector ky (106 cm-1)−0.4−0.200.20.4Energy (meV)56789Wave vector ky (106 cm-1)−0.100.1a)b)2. The limit of weak electron-hole asymmetry For arbitrary γ, but q = D/B (cid:28) 1 it is also possible to obtain analytical expressions for edge states. As before, we will seek the wave functions in the form (23). Let us consider ψky+1/2 in more detail. The wave function ψky−1/2 is related to ψky+1/2 by Eq. (23), and εky−1/2 = ε−ky+1/2. The Schrodinger equation (2) for ψky+1/2 leads to the following set of equations (cid:18) (cid:18) (cid:18) (cid:18) 2 + 1 + 2 + 1 + l1a(cid:48)(cid:48) l1a(cid:48)(cid:48) 1 + a(cid:48) 2 + a(cid:48) 1 + b(cid:48) 2 + b(cid:48) l1b(cid:48)(cid:48) l1b(cid:48)(cid:48) l−1 E αA 2 + 2 − αE l−1 A l−1 E αA 2 + 2 − αE l−1 A (cid:19) (cid:19) (cid:19) (cid:19) a1 − kya2 − k0b2 = 0 , a2 + kya1 − k0b1 = 0 , b1 + kyb2 + k0a2 = 0 , b2 − kyb1 + k0a1 = 0 ,(29) where we decomposed the edge state energy as εky+1/2 = −qδ0 + E. The set of equations (29) has analytical solution at q = 0 (α = 1) with energy (cid:112)δ2 Akyδ0 0 + γ2 . (30) E(0) = Thus, the spectrum of edge states obtained in Sec. II using perturbation theory for small ky remains linear in the case q = 0 in the entire range of wave vectors. Details of the solution, as well as expressions for the functions a1,2 and b1,2 are given in the Appendix. At q (cid:28) 1 to the first order in q we have α ≈ 1 + q and α−1 ≈ 1 − q. Hence, Eqs. (29) have the same form as at q = 0 with l2 that depends on energy 2 − qE A . l−1 2 (E) = l−1 (31) In this equation up to terms linear in q we can set E = E(0) and therefore obtain 2 (ky) ≈ l−1 l−1 (cid:112)δ2 2 − qkyδ0 0 + γ2 . (32) Equation (32) describes the dependence of localization width of the s = +1/2 edge state on ky at q (cid:28) 1. To find corrections to the dispersion of the edge states (30) at q (cid:28) 1 we substitute the expression (32) for l2(ky) in the right side of Eq. (30) for energy. Up to the first order in q we find E ≈ E(0) − qA2k2 y γ2δ0 0 + γ2)2 . (δ2 (33) It follows from Eq. (33) that in the presence of electron- hole asymmetry the spectrum of edge states deviates from linear one, and corrections quadratic in the wave vector appear. This result is consistent with that ob- tained in the limit γ/δ0 (cid:28) 1, see Eq. (28). V. MATRIX ELEMENTS OF OPTICAL TRANSITIONS 6 Excitation of the edge of topological insulator by elec- tromagnetic wave causes optical transitions between the spin branches of edge states. Due to the absence of a spatial inversion center in HgTe/HgCdTe quantum wells, optical transitions occur due to electro-dipole mecha- nism [31]. The matrix element of the electron-photon interaction is proportional to the matrix elements of the velocity operator v = ∂H0/∂k between the states ψkys and ψky−s [31]: s−s =(cid:10)ψkys vx ψky−s s−s =(cid:10)ψkys vy ψky−s v(x) v(y) (cid:11) = u1e−4isθ , (cid:11) = 2isu2e−4isθ . (34) In this section we calculate u1 and u2 in the limit γ/δ0 (cid:28) 1. Wave functions ψkys for that case are found in Sec. IV B 1, see Eqs. (23), (25), (26). The calculation gives: v0k0kyl2 2 [l2(ky) − l2(−ky)] 2(cid:112)l2(ky)l2(−ky) (cid:20) (cid:112)l2(ky)l2(−ky) 1 − l2(ky) + l2(−ky) v0k0l2 2 2l2 , (35) (cid:21) , u1 = u2 = where v0 = κA/ is the edge velocity at γ/δ0 (cid:28) 1. After simplifications we obtain (cid:113) yl3 qk0k2 2 1 − q2k2 yl2 2 u1 = v0 , u2 = −qu1 . (36) Equations (35), (36) are valid for arbitrary value of q and wave vector ky, for which the relation 1 − qkyl2 (cid:29) k0l2 holds. Also in calculations we set l1 = 0 both in the wave functions and the velocity operator. As follows from (35), the matrix elements u1 and u2 are nonzero due to the dependence of the edge states localization width on ky, and therefore vanish if the system possesses electron-hole symmetry (q = D/B = 0). It can be shown, however, that, taking into account small contributions ∝ l1/l2, the matrix element u2 (cid:54)= 0 even at q = 0. Figure 3 shows the results of calculations of the matrix elements u1 and u2 using two methods -- numerical di- agonalization of the Hamiltonian (1) and analytical for- mulas (36). The plots show that at γ/δ0 = 0.1, the analytical calculations are in good agreement with the numerical ones, however at γ/δ0 = 0.2, a significant discrepancy is already observed, so that Eqs. (36) over- estimate the values of u1 and u2. One of the reasons is that with increasing γ the edge velocity v0, which enters expressions for u1 and u2, decreases. Optical transitions between edge states can also be characterized by the matrix elements of the dipole mo- ment operator, which at small ky have the form [31]: s−s = −2sie−4isθD1ky , d(y) d(x) s−s = e−4isθD2ky , (37) where D1,2 = eu1,2/kyωs−s and ωs−s = εkys − εky−s. With account for Eq. (36) VI. THE ROLE OF BOUNDARY CONDITIONS 7 D1 = e 2 qk0l3 2 = D2 = −qD1 = − γA2 δ3 0 γA2 δ3 0 (38) , . 2(B2 − D2)3/2 eDB2 eD2B 2(B2 − D2)3/2 Estimations by Eq. (38) at γ/δ0 = 0.1 give D1/e ≈ 1.7 × 10−12 cm2 and D2/e ≈ −1.2 × 10−12 cm2. This estimation agrees on the order of magnitude with the numeric estimations obtained in [31] for γ/δ0 = 0.5. Figure 3. Matrix elements of the velocity operator u1 and u2 for γ/δ0 = 0.1 (а) and γ/δ0 = 0.2 (b). Solid curves show numeric calculations, dashed curves show calculations by Eqs. (36). The inset sketches the optical transitions in the system. Upon absorption of the circularly polarized light by he- lical edge states, optical transitions occur asymmetrically in k space leading to the generation of spin polarization and edge photocurrent [31]. Relative difference in the rates of optical transitions from the states ψky−1/2 and ψ−ky+1/2, gky−1/2 and g−ky+1/2, respectively, is propor- tional to the degree of circular polarization of the incident light Pcirc and is equal to gky−1/2 − g−ky+1/2 gky−1/2 + g−ky+1/2 = KPcirc , (39) where K = − 2D1D2 1 + D2 2 D2 = 2DB B2 + D2 . (40) We note that the asymmetry parameter K for transitions between edge states is equal to the similar coefficient for transitions from edge to bulk states [24]. In both cases, K = 0 if the system possesses electron-hole symmetry (D = 0). This result is a consequence of a more general statement about the absence of spin polarization and the photogalvanic effect upon the absorption of circularly po- larized radiation by electron-hole symmetric system [31]. The above results are obtained for the simplest "open" boundary condition ψ(x = 0) = 0. Solutions corre- sponding to a more general boundary condition ψ(cid:48)(x = 0) + hψ(x = 0) = 0 differ from those considered in the work only by the pre-factors before the exponents e−x/l1 and e−x/l2 in the functions a(x) and b(x), see for example Eq. (6). As a result, a change of h parameter leads to a change of the wave functions exactly near the edge at the small l1 scale, and therefore all the obtained results are independent of h, since they are determined by the wave functions behaviour on a much larger scale l2. The most general form of the boundary conditions at the boundary of the topological insulator with vacuum was obtained from general physical considerations in [23]. Boundary conditions of the general form take into ac- count the admixture of remote subbands by the edge of the structure. In particular, it is possible that the bound- ary condition itself violates the electron-hole symmetry in the system. In this case, even at D = B = 0 in the quantum well Hamiltonian, the spectrum of edge states becomes asymmetric with respect to the center of the band gap and deviates from the linear [23, 25]. Another example of a system with helical states is the boundary of two HgTe/HgCdTe quantum wells of differ- ent widths in the phase of trivial and topological insu- lators, respectively. In the simplest case, such a contact can be modelled by the spatial dependence δ0(x) in the Hamiltonian (1), assuming that the remaining band pa- rameters are weakly dependent on the well width. In the case B = D = 0, such a model predicts the presence of edge states with a symmetric linear spectrum [34, 35]. However, if B (cid:54)= 0 and D (cid:54)= 0, the situation becomes more complicated. The Dirac point position and the edge ve- locity depend on the ratio of δ0 values at the right and left sides of the contact. Let us consider δ0(x > 0) = −δr, δ0(x < 0) = δl, and δl,r > 0. Then, if the relation A/δr, A/δl (cid:29) l1 holds, the electron-hole asymmetry does not lead to a noticeable change in the spectrum of edge states. In this case, the ∝ k2 diagonal terms in the Hamiltonian can be neglected without changing the physical results. The same situation is realized in the case of a smooth contact. However if A/δr (cid:29) l1 and A/δl (cid:46) l1, the electron-hole asymmetry begins to play a significant role in the spectrum of edge states. In the limit A/δl (cid:28) l1 the wave function does not penetrate into the region x < 0, which corresponds to the open boundary condition. VII. CONCLUSION In this work we have studied the effect of electron-hole asymmetry on the electronic structure of helical edge states in HgTe/HgCdTe quantum wells. We have ob- tained analytical expressions for the wave functions and the energy spectrum of helical states, the g-factor ten- u1u2Velocity matrix element (105 cm/s)−50510Wave vector ky (106 cm-1)−0.100.1Velocity matrix element (105 cm/s)−100102030Wave vector ky (106 cm-1)−0.100.1a)b)sor, and the matrix elements of optical transitions be- tween edge states with opposite spin in the framework of the electro-dipole mechanism. We have shown that in the presence of electron-hole asymmetry, the spectrum of edge states deviates from the linear one, and have found corrections of higher orders in the wave vector. It has been shown that electron-hole asymmetry has the great- est impact on the structure of helical states for a sharp boundary with vacuum, while in the case of a smooth boundary, for example, a contact between two insulators, its role is significantly reduced. Obtained results can be used in the analysis of magneto-transport phenom- ena and the edge photogalvanic effect in HgTe/HgCdTe quantum wells. ACKNOWLEDGMENTS (cid:110) (cid:115) 8 These equations show that for E < δ0 (the energy of the edge state lies in the bulk gap) only the oscillation period of the wave function varies, while the decay length does not change. For E > δ0, the decay length starts to increase, and finally, at E = ±δ0 0/γ2, it goes to infinity. As in the case γ = 0, this occurs at the point, where edge and bulk dispersion curves touch each other. (cid:112)1 + δ2 The final expressions for a1,2 and b1,2 read (cid:111) e−x/l1 cos ϕ1 − e−x/l2 cos[k0(E)x − ϕ1] a1(x) = v1 a2(x) = v1e−x/l1 cos ϕ1 − v2e−x/l2 cos[k0(E)x − ϕ1 − ϕ2] , b1(x) = e−x/l1 sin ϕ1 + v1v2e−x/l2 sin[k0(E)x − ϕ1 − ϕ2] , b2(x) = e−x/l1 sin ϕ1 + e−x/l2 sin[k0(E)x − ϕ1] . (A2) Here: , The author is grateful to S.A. Tarasenko for fruitful discussions. The author acknowledges financial support from the Russian Federation President Grant (project MK-2943.2019.2) and the "Basis" Foundation for the Ad- vancement of Theoretical Physics and Mathematics. k0(E) = k0 1 − E2 δ2 0 , tan 2ϕ1 = k0(E)l2 1 + kyl2 , (A3) Appendix A: Wave functions of edge states at D = 0 In this section, we obtain analytic expressions for the wave functions ψkys in the case D = 0 and in the limit D/B (cid:28) 1. The functions ψkys are sought in the form (23), where the set of equations for the functions a1,2 and b1,2 is given by Eq. (29). We will look for a solution in the form a1,2, b1,2 ∝ e−λx with positive λ. Substituting this solution into the set of equations, we find the roots of the characteristic equation λj and the corresponding vectors as functions of the energy E. It can be shown that the boundary condition ψkys(x = 0) = 0 2, see Eq. (30) can be satisfied only if E = Aky/(cid:112)1 + k2 (cid:115) In the l1 (cid:28) l2 limit we have: in the main text. 0l2 λ1 = λ2 = , λ3 = λ∗ 4 = 1 l1 − ik0 1 l2 1 − E2 δ2 0 . (A1) [1] B. A. Bernevig, T. L. Hughes, S.-C. Zhang. Quan- tum Spin Hall Effect and Topological Phase Transition in HgTe Quantum Wells. Science 314, 1757 (2006). [2] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buh- mann, L. W. Molenkamp, X.-L. Qi, S.-C. Zhang. Quan- tum Spin Hall Insulator State in HgTe Quantum Wells. Science 318, 766 (2007). [3] I. Knez, R.-R. Du, G. Sullivan. Evidence for Helical Edge Modes in Inverted InAs/GaSb Quantum Wells. Phys. Rev. Lett. 107, 136603 (2011). [4] A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X.-L. Qi, S.-C. Zhang. Nonlocal Transport (cid:115)δ0 − E δ0 + E v1 = , v2 = (cid:115)δ0(cid:112)1 + k2 δ0(cid:112)1 + k2 2 − E 0l2 0l2 2 + E , (cid:112)δ2 0 − E2(cid:112)1 + k2 Ek0l2 0l2 2 . tan ϕ2 = As shown in Sec. IV B 2, in the limit D/B (cid:28) 1 Eqs. (29) has the same form as at D = 0, but with l2 that depends on energy according to Eq. (31). Thus, to find the wave functions of the edges states in the limit D/B (cid:28) 1, one should substitute the l2 length in Eqs. (A2) and (A3) with its expression Eq. (31). in the Quantum Spin Hall State. Science 325, 294 (2009). [5] G. M. Gusev, Z. D. Kvon, O. A. Shegai, N. N. Mikhailov, S. A. Dvoretsky, J. C. Portal. Transport in disordered two-dimensional topological insulators. Phys. Rev. B 84, 121302 (2011). [6] E. Y. Ma, M. R. Calvo, J. Wang, B. Lian, M. Muhlbauer, C. Brune, Y.-T. Cui, K. Lai, W. Kundhikanjana, Y. Yang, M. Baenninger, M. Konig, C. Ames, H. Buh- mann, P. Leubner, L. W. Molenkamp, S.-C. Zhang, D. Goldhaber-Gordon, M. A. Kelly, Z.-X. Shen. Un- expected edge conduction in mercury telluride quantum wells under broken time-reversal symmetry. Nature Com- munications 6, 7252 (2015). [7] S. U. Piatrusha, E. S. Tikhonov, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoretsky, V. S. Khrapai. Topological Protection Brought to Light by the Time-Reversal Sym- metry Breaking. Phys. Rev. Lett. 123, 056801 (2019). [8] Y. Tanaka, A. Furusaki, K. A. Matveev. Conductance of a Helical Edge Liquid Coupled to a Magnetic Impurity. Phys. Rev. Lett. 106, 236402 (2011). [9] A. M. Lunde, G. Platero. Helical edge states coupled to a spin bath: Current-induced magnetization. Phys. Rev. B 86, 035112 (2012). [10] B. L. Altshuler, I. L. Aleiner, V. I. Yudson. Localiza- tion at the Edge of a 2D Topological Insulator by Kondo Impurities with Random Anisotropies. Phys. Rev. Lett. 111, 086401 (2013). [11] J. I. Vayrynen, M. Goldstein, Y. Gefen, L. I. Glazman. Resistance of helical edges formed in a semiconductor het- erostructure. Phys. Rev. B 90, 115309 (2014). [12] M. V. Entin, L. I. Magarill. Localization of edge electrons in a 2D topological insulator strip. JETP Letters 100, 566 (2015). [13] P. D. Kurilovich, V. D. Kurilovich, I. S. Burmistrov, M. Goldstein. Helical edge transport in the presence of a magnetic impurity. JETP Letters 106, 593 (2017). [14] K. E. Nagaev. AC Response of the Edge States in a Two- Dimensional Topological Insulator Coupled to a Conduct- ing Puddle. physica status solidi (RRL) -- Rapid Research Letters 12, 1700422 (2018). [15] K.-M. Dantscher, D. A. Kozlov, M. T. Scherr, S. Gebert, J. Barenfanger, M. V. Durnev, S. A. Tarasenko, V. V. Bel'kov, N. N. Mikhailov, S. A. Dvoretsky, Z. D. Kvon, J. Ziegler, D. Weiss, S. D. Ganichev. Photogalvanic probing of helical edge channels in two-dimensional HgTe topological insulators. Phys. Rev. B 95, 201103 (2017). [16] M. V. Durnev, S. A. Tarasenko. High-Frequency Nonlin- ear Transport and Photogalvanic Effects in 2D Topologi- cal Insulators. Annalen der Physik 1800418 (2019). [17] Z. Fei, T. Palomaki, S. Wu, W. Zhao, X. Cai, B. Sun, P. Nguyen, J. Finney, X. Xu, D. H. Cobden. Edge con- duction in monolayer WTe2. Nature Physics 13, 677 EP (2017). [18] X.-L. Qi, S.-C. Zhang. Topological insulators and super- conductors. Rev. Mod. Phys. 83, 1057 (2011). [19] M. Konig, H. Buhmann, L. W. Molenkamp, T. Hughes, C.-X. Liu, X.-L. Qi, S.-C. Zhang. The Quantum Spin Hall Effect: Theory and Experiment. Journal of the Physical Society of Japan 77, 031007 (2008). [20] B. Zhou, H.-Z. Lu, R.-L. Chu, S.-Q. Shen, Q. Niu. Finite Size Effects on Helical Edge States in a Quantum Spin- Hall System. Phys. Rev. Lett. 101, 246807 (2008). [21] E. B. Sonin. Edge accumulation and currents of moment in two-dimensional topological insulators. Phys. Rev. B 9 82, 113307 (2010). [22] P. C. Klipstein. Structure of the quantum spin Hall states in HgTe/CdTe and InAs/GaSb/AlSb quantum wells. Phys. Rev. B 91, 035310 (2015). [23] V. V. Enaldiev, I. V. Zagorodnev, V. A. Volkov. Bound- ary conditions and surface state spectra in topological in- sulators. Pis'ma Zh. Eksp. Teor. Fiz. 101, 94 (2015). [24] V. Kaladzhyan, P. P. Aseev, S. N. Artemenko. Photogal- vanic effect in the HgTe/CdTe topological insulator due to edge-bulk optical transitions. Phys. Rev. B 92, 155424 (2015). [25] M. V. Entin, M. M. Mahmoodian, L. I. Magarill. Lin- earity of the edge states energy spectrum in the 2D topo- logical insulator. EPL (Europhysics Letters) 118, 57002 (2017). [26] X. Dai, T. L. Hughes, X.-L. Qi, Z. Fang, S.-C. Zhang. Helical edge and surface states in HgTe quantum wells and bulk insulators. Phys. Rev. B 77, 125319 (2008). [27] S. A. Tarasenko, M. V. Durnev, M. O. Nestoklon, E. L. Ivchenko, J.-W. Luo, A. Zunger. Split Dirac cones in HgTe/CdTe quantum wells due to symmetry-enforced level anticrossing at interfaces. Phys. Rev. B 91, 081302 (2015). [28] R. Winkler, L. Wang, Y. Lin, C. Chu. Robust level co- incidences in the subband structure of quasi-2D systems. Solid State Communications 152, 2096 (2012). [29] M. V. Durnev, S. A. Tarasenko. Magnetic field effects on edge and bulk states in topological insulators based on HgTe/CdHgTe quantum wells with strong natural in- terface inversion asymmetry. Phys. Rev. B 93, 075434 (2016). [30] G. M. Minkov, A. V. Germanenko, O. E. Rut, A. A. Sherstobitov, M. O. Nestoklon, S. A. Dvoretski, N. N. Mikhailov. Spin-orbit splitting of valence and conduction bands in HgTe quantum wells near the Dirac point. Phys. Rev. B 93, 155304 (2016). [31] M. V. Durnev, S. A. Tarasenko. Optical properties of helical edge channels in zinc-blende-type topological in- sulators: selection rules, circular and linear dichroism, circular and linear photocurrents. Journal of Physics: Condensed Matter 31, 035301 (2019). [32] M. Entin, L. Magarill. Edge absorption and circular pho- togalvanic effect in 2D topological insulator edges. Pis'ma Zh. Eksp. Teor. Fiz. 103, 804 (2016). [33] X. Marie, T. Amand, P. Le Jeune, M. Paillard, P. Renucci, L. E. Golub, V. D. Dymnikov, E. L. Ivchenko. Hole spin quantum beats in quantum-well structures. Phys. Rev. B 60, 5811 (1999). [34] R. Jackiw, C. Rebbi. Solitons with fermion number 1/2. Phys. Rev. D 13, 3398 (1976). [35] B. A. Volkov, O. A. Pankratov. Two-dimensional mass- less electrons in an inverted contact. Pis'ma Zh. Eksp. Teor. Fiz. 42, 145 (1985). 

1108.5478

2

1108

Band gap in graphene induced by vacuum fluctuations

The electrons in undoped graphene behave as massless Dirac fermions. Therefore graphene can serve as an unique condensed-matter laboratory for the study of various relativistic effects, including quantum electrodynamics (QED) phenomena. Although theoretical models describing electronic properties of graphene have been elaborated in details, the QED effects were usually neglected. In this paper we demonstrate theoretically that QED can drastically modify electronic properties of graphene. We predict the following QED effect - the opening of the band gap in a graphene monolayer placed inside a planar microcavity filled with an optically active media. We show that this phenomenon occurs due to the vacuum fluctuations of the electromagnetic field and is similar to such a well-known phenomenon as a vacuum-induced splitting of atomic levels (the Lamb shift). We estimate the characteristic value of the band gap and find that it can sufficiently exceed the value of the Lamb shift.

cond-mat.mes-hall

cond-mat

Band gap in graphene induced by vacuum fluctuations O. V. Kibis,1, 2, ∗ O. Kyriienko,3 and I. A. Shelykh3, 2 1Department of Applied and Theoretical Physics, Novosibirsk State Technical University, Karl Marx Avenue 20, 630092 Novosibirsk, Russia 2International Institute of Physics, Av. Odilon Gomes de Lima, 1772, Capim Macio, 59078-400, Natal, Brazil 3Science Institute, University of Iceland, Dunhagi-3, IS-107, Reykjavik, Iceland The electrons in undoped graphene behave as massless Dirac fermions. Therefore graphene can serve as an unique condensed-matter laboratory for the study of various relativistic effects, including quantum electrodynamics (QED) phenomena. Although theoretical models describing electronic properties of graphene have been elaborated in details, the QED effects were usually neglected. In this paper we demonstrate theoretically that QED can drastically modify electronic properties of graphene. We predict the following QED effect -- the opening of the band gap in a graphene monolayer placed inside a planar microcavity filled with an optically active media. We show that this phenomenon occurs due to the vacuum fluctuations of the electromagnetic field and is similar to such a well-known phenomenon as a vacuum-induced splitting of atomic levels (the Lamb shift). We estimate the characteristic value of the band gap and find that it can sufficiently exceed the value of the Lamb shift. PACS numbers: 78.67.Wj, 31.30.jf I. INTRODUCTION Graphene -- a monolayer of carbon atoms -- pos- sesses unusual physical properties that make it attrac- tive for various applications.1 -- 3 Usually treated as a plat- form for the novel high-speed electronics,4,5 graphene is of great interest from the point of view of the fundamen- tal physics as well. Indeed, the low-energy electron exci- tations in graphene are massless Dirac fermions with the linear energy spectrum, ε(k) = vFk.6 -- 9 That makes graphene a condensed-matter playground for the study of various relativistic quantum phenomena, such as the Klein tunnelling10,11 and the Casimir effect.12 -- 14 Up to now, most of graphene-related studies were focused on its unusual transport properties, and quantum electrody- namics (QED) effects arising from interaction of electrons in graphene with a quantized electromagnetic field were neglected.7 This paper is aimed to fill partially this gap in the theory. We show that due to the giant Fermi ve- locity of electrons in graphene, vF ≈ c/300, QED effects are pronounced and can lead to qualitative modifications of the spectrum of elementary excitations. The linear energy spectrum of electrons in graphene comes from its specific honeycomb lattice structure which makes the band gap between the valence and conductiv- ity bands to be exactly zero.7 There is the long-standing problem of the opening of the band gap. The appear- ance of a controllable band gap is required for various electronical and optical applications of graphene.1,15,16 Aside from this, it is interesting from the fundamental viewpoint to analyze how massless Dirac fermions can acquire a mass. This question is relevant, particularly, in the context of the observation of Majorana fermions in condensed matter systems.17 Several mechanisms of the band gap opening in monolayer graphene have been proposed. Among them are breaking of the symme- try between two sublattices of the honeycomb lattice of graphene,18,19 the spin-orbit coupling20 and the many- body interactions leading to the excitonic instability.21,22 Recently, one of us put forward the proposal of open- ing the band gap by illuminating graphene with a circu- larly polarized light.23 In this case the gap in the spec- trum of elementary electron excitations appears due to the formation of composite electron-photon states which are similar to polaritons in ionic crystals and quantum microcavities.24 -- 27 It should be noted that, within the framework of QED, the electron-photon interaction can be observed even if "real" photons are absent and elec- trons interact only with vacuum fluctuations of electro- magnetic field due to emitting and reabsorbing virtual photons.28 Therefore, one can expect that the photon- induced splitting of valence and conductivity bands in graphene23 will take place due to the vacuum fluctuations even in the absence of an external field pumping. This QED effect is similar to the well-known Lamb shift in the atomic physics, i.e., the vacuum-induced splitting of the states 2s1/2 and 2p1/2 of a hydrogen atom with the char- acteristic splitting energy ∆ ≈ 4 µeV. The Lamb shift, discovered experimentally by Lamb and Retherford29 and theoretically explained by Bethe30 more than 60 years ago, is extremely important for understanding and verifi- cation of basic principles of QED. That is why it attracts the undivided attention of the physics community up to now.31 Since clockwise and counterclockwise polarized pho- tons shift electron levels in graphene in mutually oppo- site directions,23 the band gap opening needs breaking the symmetry between virtual photons with different cir- cular polarizations. This can be achieved by placing a graphene monolayer inside a planar cavity filled with an optically active material (see Fig. 1). As it will be shown below, in this case the vacuum fluctuations lead to the opening of the band gap in graphene even in the ab- sence of an external circularly polarized optical pumping. 2 of a planar microcavity. Generally, electron states in graphene near the Fermi energy are described by the eight-component wave function which accounts for two elementary sublattices of graphene, two electron valleys, and two orientations of electron spin.7 In what follows intervalley scattering processes and spin-flip effects will be beyond consideration, which reduces the number of necessary components of wavefunction to two. The single-particle Hamiltonian of electron in graphene coupled to the cavity mode reads (see the Appendix A for details of the derivation) H = Hfield + Hk + Hint , (cid:88) q,± Hfield = ωq,± a† q,± aq,± (1) (2) is the photonic part of the Hamiltonian written in the basis of circularly polarized states, Hk = vF σ · k (3) FIG. 1: (color online) Sketch of the system. A graphene sample placed inside a planar cavity filled with an optically active media. The arrows with signs + and − correspond to clockwise and counterclockwise circularly polarized virtual photons, respectively, which are emitted and reabsorbed by electrons in graphene. where It should be noted, that the QED mass renormalization in an optically active media has been considered,32 but, surprisingly, the most interesting case of massless Dirac fermions was not analyzed before. II. THE MODEL Let us consider the problem of interaction between a single electron in graphene and a single photon mode is the electron Hamiltonian near the point where the valence and conductivity bands of graphene touch each other (the Dirac point) and (cid:114) 2 (cid:20) (cid:88)
0LS q 1√ ω+,q Hint = −evF (cid:16) † σ−a+,qeiqr + σ+a +,qe−iqr(cid:17) (cid:16) −,qe−iqr + σ+a−,qeiqr(cid:17)(cid:21) † σ−a (4) + 1√ ω−,q is the Hamiltonian of electron-photon interaction in the cavity. For definiteness, we assume the graphene sheet to be placed in the center of the cavity. In Eqs. (2) -- (4) the subscript indices, ±, correspond to the photon modes with clockwise and counterclockwise circular polariza- tions, k = exkx + eyky and q = exqx + eyqy denote in- plane electron and photon wave vectors, respectively, ex,y are unit vectors directed along the x, y-axis, e is the elec- tron charge,
0 is the vacuum permittivity, L is the dis- tance between two mirrors of the planar cavity (the cavity length), S is the area of graphene sample, ω±,q are the eigenfrequencies of clockwise and counterclockwise circu- † larly polarized photons, a±,q and a ±,q are photonic anni- hilation and creation operators. The Pauli vector opera- tor, σ, acts in the space of two orthogonal electron states, ±(cid:105), corresponding to the two sublattices of graphene in accordance with the following rules: σz±(cid:105) = ±±(cid:105) and σ±∓(cid:105) = ±(cid:105), where σ± = (σx ± iσy)/2. Thus, it corre- sponds to the pseudospin of electron. Eigenstates of the electron Hamiltonian (3) are given by the expression7 (cid:16) e−iθk/2+(cid:105) ± eiθk/2−(cid:105)(cid:17) k,±(cid:105) = eikr√ 2S , (5) where θk = arctan(ky/kx) and the signs ± correspond to electron states in the conductivity and valence bands of graphene (the upper and lower Dirac cones, respectively). The corresponding eigenenergies are ε(0)±,k = ±vFk. The eigenstates of the photon Hamiltonian (2) can be written as N±,q(cid:105), where N±,q are photon occupations number for photons with different circular polarizations (±) and wave vectors q. Then eigenstates of the full electron-photon Hamiltonian (1) can be decomposed in the basis of the orthogonal electron-photon states k,±, N+, N−(cid:105) = k,±(cid:105) ⊗ N+,q(cid:105) ⊗ N−,q(cid:48)(cid:105) (6) with the energies k,±,N+,N− = ±vFk + ω+,qN+,q + ω−,q(cid:48)N−,q(cid:48) . (7) ε(0) In order to find eigenstates and eigenenergies of the full Hamiltonian (1), we will use the perturbation theory, considering the interaction Hamiltonian (4) as a pertur- values of the 2 × 2 matrix (cid:101)H(1) having matrix elements bation. To calculate the energy corrections in the lowest order of the perturbation, one needs to find the eigen- (cid:101)H(1) ss(cid:48) (k) = (cid:104)k, s, 0, 0 Hintk, s(cid:48), 0, 0(cid:105). However, it is easy to see that all matrix elements of this type are zero and one needs to use the second order of the perturbation theory. Physically, we need to account for the follow- ing processes: the electron with a wave vector k emits a virtual photon with a momentum q and then reabsorbs it. Note, that in such a process the momentum of the electron in the initial state, k, should be equal to its mo- mentum in the final state, but the value of the index s can be changed: the electron can remain in the same Dirac cone or move from one Dirac cone to another one. The last process becomes efficient around k = 0 point where the energies of the Dirac cones are close to each other, which can lead to the lifting of the degeneracy as we show below. Therefore, to calculate the spectrum of the Hamiltonian (1), we need to use the perturbation theory for degenerate states. in Ref. Let us briefly remind how the second-order correc- tions can be accounted for within the framework of per- turbation theory for degenerate states (the details can be found, e.g., Imagine that we have a set of states {m} which are close in energy to each m(cid:48) m − ε(0) other (this means that the energy distances ε(0) between them are comparable or smaller relative to a characteristic energy of the perturbation). The pertur- bation does not couple any states m and m(cid:48) directly [33]). 3 (cid:101)H = (cid:101)H(0) + (cid:101)H(2), (otherwise the standard first-order perturbation theory is applicable), but couples them to a set of the states {l} whose energies lie far from energies of the states {m} (this means that the energy distances ε(0) m − ε(0) are large as compared with the characteristic energy of In our case the set {m} consists the perturbation). of the two states {k, +, 0, 0(cid:105), k,−, 0, 0(cid:105)}, and the set {l} corresponds to the states k(cid:48),±, N+,n,q, N−,n,q(cid:48)(cid:105) with (cid:54)= 0. Then energies of the perturbed N+,n,q + N−,n,q(cid:48) {m} states can be obtained by diagonalization of the ma- trix Hamiltonian l l l 1 1 + 1 2 (cid:88) (cid:101)H(2) where (cid:101)H(0) is the matrix of unperturbed Hamiltonian (3) elements of the Hamiltonian (cid:101)H(2) can be found as (cid:19) written in the subspace of states {m}, and the matrix (cid:18) (8) mm(cid:48) = m − ε0 ε0 (9) where the summation goes over all set of the states {l}. k for a given electron wave vector k can be written as the 2× 2 matrix m(cid:48) − ε0 ε0 × (cid:104)m Hintl(cid:105)(cid:104)l Hintm(cid:48)(cid:105) , In the case we consider, the Hamiltonian (cid:101)H(2) (cid:19) (cid:101)H(2) −vF k + (cid:101)H−− where the vacuum-fluctuation corrections (cid:101)H++ k , (cid:101)H−− and (cid:101)H+− (cid:18) vF k + (cid:101)H++ (cid:101)H−+ (cid:17)∗ (cid:16)(cid:101)H−+ (cid:101)H+− are given by k = (10) k = k k , k l k k k (11) (12) (13) (cid:90) (cid:90) (cid:88) (cid:88) λ=± λ=± d2q d2q ωλ,q − vFk ωλ,q [(ωλ,q − vFk)2 − v2 Fk − q2] ωλ,q + vFk ωλ,q [(ωλ,q + vFk)2 − v2 Fk − q2] , , (cid:101)H++ k = − e2v2 F 4π
0L (cid:101)H−− k = − e2v2 F 4π
0L (cid:88) (cid:90) λ d2q λ=± (cid:101)H+− k = − e2v2 F 4π
0L F (k2 + k − q2) λ,q − v2 ω2 Fk − q2] · [(ωλ,q − vFk)2 − v2 [(ωλ,q + vFk)2 − v2 cone. The off-diagonal matrix elements (cid:101)H+− Fk − q2] , k = (cid:101)H−+ k correspond to the processes in which the electron after reabsorption of the photon changes the Dirac cone. Di- agrammatic representation of these terms is shown in Fig. 2. Diagonalization of the Hamiltonian (10) gives the renormalized energy spectrum of the elementary ex- and the symbol λ = ± corresponds to the two different circular polarizations of virtual photons. The physical meaning of the terms in the Hamiltonian (10) is the fol- corresponds to the pro- cess: an electron in the upper Dirac cone emits a virtual photon and then reabsorbs this photon while returning corresponds to the same process for the electron in the lower Dirac lowing. The matrix element (cid:101)H++ to the same cone. The matrix element (cid:101)H−− k k 4 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∞(cid:88) n=0 (cid:88) λ=± (cid:20) (cid:21)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) , In the discussion above we restricted our analysis to the single-mode approximation, accounting for the cou- pling of the electron in graphene with only one photon mode. Going beyond this approximation, one needs to perform the summation over all modes n in Eqs. (11) -- (13). Keeping in mind that photon modes with even numbers n correspond to the zero field intensity in the center of the cavity and, thus, do not interact with the graphene sheet, one gets the following expression for the band gap: εg = e2 2π
0L λβ2 λ 1 − β2 λ ln 1 + 0L2(1 − β2 q2 λ) π2(2n + 1)2 (18) where β± = vF /c±, q0 ∼ 1/a0 is the cut-off parameter of integration in Eq. (13) with a0 being the lattice constant of graphene. III. RESULTS AND DISCUSSION It is seen that Eq. (18) contains the summation over photon polarizations, λ = ±. As expected, the contri- butions of clockwise and counterclockwise polarized pho- tons in the band gap (18) have opposite signs and the band gap vanishes for β+ = β− and appears only in the presence of an optically active media with β+ (cid:54)= β−. For instance, the cavity can be filled with a magneto- gyrotropic media based on ferrite garnets which possess the giant difference between the velocities of light with different circular polarizations.34 The effect becomes even more pronounced if the cavity is filled with an active me- dia with the circular dichroism.35 In this case one of the two circularly polarized photon modes in the cavity is suppressed and its contribution to the band gap (18) can be neglected. As a result, the summation over λ in Eq. (18) can be omitted, which leads to the drastic increasing of the band gap. Figure 3(a) shows the dependence of the band gap, εg, on the cavity length, L, for the cavity filled by such a media with the circular dichroism. In the physically relevant region of the cavity lengths (the white area in Fig. 3(a)), the band gap calculated in the single-mode approximation (n = 1) is of several µeV, which is compa- rable with the Lamb shift.29 The summation over higher modes increases this value by 1 -- 2 orders of magnitude: for 100 modes the value of the band gap increases to 50 -- 100 µeV, while the summation over all modes gives value of about 200 µeV. However, the summation over infinite number of modes overestimates the band gap. Indeed, if the characteristic photon wavelength, 2π/qz = 2L/n, is comparable to the interatomic distance, the macro- scopic description of an optically active media becomes irrelevant. Therefore the band gap can be reasonably es- timated to be about tens of µeV, that is one order of magnitude bigger than the Lamb shift.29 The energy spectrum of massive Dirac fermions in graphene is plotted in Fig. 3(b). The renormalized dis- FIG. 2: Diagrammatic representation of the terms entering in the Hamiltonian (10). The solid lines correspond to the elec- trons and the dashed lines correspond to the virtual photons. Index λ = ± denotes the two different circular polarizations of the photons, and the index s = ± denoted the two different Dirac cones. Summation should be performed over both the indices λ and s. ± citations in graphene, k /2 + (cid:101)H−− (cid:114)(cid:16)(cid:101)H++ (cid:17)2 (cid:16)(cid:101)H+− k /2 − (cid:101)H−− k = (cid:101)H−− Taking into account that at k = 0 we have (cid:101)H++ ε±(k) = (cid:101)H++ k /2 + vFk(cid:17)2 k , the renormalized electron energy (14) at the Dirac point can be written as k /2 (14) + k . ε±(0) = (cid:101)H++ 0 ±(cid:12)(cid:12)(cid:12)(cid:101)H+− 0 (cid:12)(cid:12)(cid:12) . It follows from Eq. (15) that the vacuum fluctuations of electromagnetic field in the cavity can open the band gap between the conductivity and valence bands of graphene at the Dirac point, which is (cid:12)(cid:12)(cid:12)(cid:101)H+− 0 (cid:12)(cid:12)(cid:12) . εg = 2 (15) (16) the term (cid:101)H+− It should be stressed that in the absence of an opti- cally active media, the eigenfrequencies of clockwise and counterclockwise circularly polarized photons are equal, ω+,q = ω−,q. According to the equation (13), in this case is zero and the band gap (16) vanishes. Therefore for the band gap opening one needs to fill the cavity by an optically active media which splits modes of virtual photons with different circular polarizations. In this case the photonic dispersions read as 0 ω±,q = c±(cid:112)q2 + q2 z , (17) where qz = πn/L is the quantized z-component of photon wave vector in the cavity, n is the number of photon mode, c± = c/n± are the speeds of light with clockwise and counterclockwise circular polarizations, and n± are the refractive indices for clockwise and counterclockwise polarized light, which are different in the optically active media, n+ (cid:54)= n−. Appendix A: Derivation of interaction Hamiltonian 5 The introduction of the electron-photon interaction in graphene can be done by the conventional replacement k → k − e A, where A is the operator of the vector- potential of electromagnetic field. Then the full Hamil- tonian of the electron-photon system reads H = vF σ·(cid:16)k − e A (cid:17) (cid:90) + 1 2 (cid:16) †
0 E †
E + µ0 B µ−1 B , dV (cid:17) (A1) where E, B are the operators of electric and magnetic fields, and
, µ are the tensors of electric and magnetic permittivity of the media, respectively. The integration in the last term, giving the energy of free electromagnetic field, goes over all space where the field is present. In the current paper we consider a graphene sheet placed in a planar microcavity. In this case it is convenient to represent the operators of the fields in terms of the eigenmodes of the cavity as (cid:88) (cid:88) (cid:88) λ,n,q λ,n,q λ,n,q A(r) = E(r) = B(r) = Aλ,n,q(r) , Eλ,n,q(r) , Bλ,n,q(r) , (A2) (A3) (A4) where n = 1, 2, 3, ... is the number of field mode in the cavity. Using the Coulomb gauge, we can write the field operators (A2) -- (A4) as Aλ,n,q(r) = Eλ,n,q(r) = i (cid:115)  Bλ,n,q(r) = 2
0ωλ,n,q (cid:16) (cid:16) 2
0ωλ,n,q (cid:115)  (cid:114)ωλ,n,q (cid:16) 2
0 aλ,n,quλ,n,q(r) + (cid:17) † λ,n,qu∗ +a λ,n,q(r) (cid:17) † λ,n,q(r) − λ,n,qu∗ a −aλ,n,quλ,n,q(r) , (A5) , (A6) aλ,n,q∇ × uλ,n,q(r) + † λ,n,q∇ × u∗ +a λ,n,q(r) (cid:17) , (A7) where uλ,n,q are the cavity eigenmodes. If the cavity is filled with an optically active media, the eigenmodes are circularly polarized and can be found as (cid:114) 2 (cid:114) LS sin (cid:17) (cid:16) πnz (cid:17)2 (cid:16) πn L L eiq·r , (A8) (A9) ω±,n,q = c± q2 + u±,n,q(z, r) = e± where FIG. 3: (color online) (a) The band gap in graphene induced by vacuum fluctuations, calculated with accounting different numbers of cavity modes n; (b) Energy spectrum of free elec- trons in graphene (dashed lines) and electrons dressed by vir- tual photons (solid lines). The calculation is performed for the cavity length L = 300 nm and the number of accounted cavity modes is n = 100. (cid:113) persion relation can be approximated by the analytical expression ε±(k) = ± (vFk)2 + (m∗v2 F )2 , (19) where the effective mass of electron dressed by virtual photons is m∗ = εg/2v2 F . It should be noted that the considered single-electron problem can be easily generalized for the realistic situa- tion when the valence band is filled by the Fermi sea of electrons. In this case the Pauli principle forbids virtual transitions into the lower Dirac cone filled with electrons, which reduces both the matrix elements (11) -- (13) and the band gap (18) by the factor of 1/2. IV. CONCLUSIONS We predicted the quantum electrodynamical effect in graphene placed inside a planar cavity filled by an op- tically active media. Due to the vacuum fluctuations of electromagnetic field in the cavity, the spectrum of el- ementary excitations in graphene undergoes qualitative changes. Namely, the valence and conductivity bands of graphene are split at the Dirac points. The value of the vacuum-induced band gap can be one order of magnitude bigger then the famous Lamb shift in hydrogen atom. Acknowledgements. The work was partially supported by Rannis "Center of Excellence in Polaritonics", Eim- skip foundation, the RFBR projects 10-02-00077 and 10- 02-90001, the Russian Ministry of Education and Sci- ence, the 7th European Framework Programme (Grants No. FP7-230778 and FP7-246784), and ISTC Project No. B-1708. are the photon eigenfrequencies. Therefore the Hamilto- nian of the interaction between the graphene sheet and the electromagnetic field in the cavity can be written as Hint = −evF σ · A(r) = evF (e+ σ− + (cid:88) √ 2 λ=±,n,q + e− σ+) · Aλ,n,q(r). (A10) 6 Since the graphene sheet is placed in the center of the cav- ity (at z = L/2), it is coupled only with modes (A8) cor- responding to odd numbers n. This means that the sum- mation index, n, in Eq. (A10) is odd: n = 1, 3, 5, 7, . . . . Then, using the expression (A5) for the vector potential operator Aλ,n,q(r), the interaction Hamiltonian (A10) can be rewritten in the form (4). ∗ Electronic address: Oleg.Kibis@nstu.ru 1 A. Geim, Science 324, 1530 (2009). 2 A. K. Geim and K. S. Novoselov, Nature Mat. 6, 183 W. A. de Heer, D.-H. Lee, F. Guinea, A. H. Castro Neto, and A. Lanzara, Nat. Mater. 6, 770 (2007). 19 F. Guinea, M. I. Katsnelson, and A. K. Geim, Nature Phys. (2007). 6, 30 (2010). 3 K. S. Novoselov, Angew. Chem. Int. Ed. 50, 31, 6986 20 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2011). 4 S. V. Morozov, K. S. Novoselov, M. I. Katsnelson, F. Schedin, D. C. Elias, J. A. Jaszczak, and A. K. Geim, Phys. Rev. Lett. 100, 016602 (2008). 5 S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83, 407 (2011). 6 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, Nature (London) 438, 197 (2005). 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 I. A. Luk'yanchuk and Y. Kopelevich, Phys. Rev. Lett. 97, 256801 (2006). 9 G. W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984). 10 C. W. J. Beenakker, Rev. Mod. Phys. 80, 1337 (2008). 11 M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nature Phys. 2, 620 (2006). (2005). 21 O. V. Gamayun, E. V. Gorbar, and V. P. Gusynin, Ukr. J. Phys. 55, 95 (2010). 22 O. V. Gamayun, E. V. Gorbar, and V. P. Gusynin, Phys. Rev. B 81, 075429 (2010). 23 O. V. Kibis, Phys. Rev. B 81, 165433 (2010). 24 A. V. Kavokin, J. J. Baumberg, G. Malpuech, and F. P. Laussy, Microcavities (Oxford University Press, Oxford, 2007). 25 T. C. H. Liew, I. A. Shelykh, and G. Malpuech, Physica E 43, 1543 (2011). 26 I. A. Shelykh, A. V. Kavokin, Y. G. Rubo, T. C. H. Liew, and G. Malpuech, Sem. Sci. Technol. 25, 013001 (2010). 27 T. C. H. Liew, A. V. Kavokin, and I. A. Shelykh, Phys. Rev. B 75, 241301 (2007). 28 V. B. Berestetskii, E. M. Lifshitz, L. P. Pitaevskii, Quan- tum Electrodynamics (Pergamon Press, Oxford, 1982). 29 W. E. Lamb and R. C. Retherford, Phys. Rev. 72, 241 12 S. M. Dutra, Cavity Quantum Electrodynamics (Wiley, (1947). Hoboken, 2005). 13 I. V. Fialkovsky, V. N. Marachevsky, and D. V. Vassilevich, 30 H. A. Bethe, Phys. Rev. 72, 339 (1947). 31 M. O. Scully and A. A. Svidzinsky, Science 328, 1239 Phys. Rev. B 84, 035446 (2011). (2010). 14 B. E. Sernelius, EPL 95, 57003 (2011). 15 P. Avouris, Z. Chen, and V. Perebeinos, Nature Nanotech. 32 K.-P. Marzlin, Phys. Rev. A 53, 2074 (1996). 33 G. L. Bir and G. E. Pikus, Symmetry and Strain-induced 2, 605 (2007). 16 R. R. Hartmann, N. J. Robinson, and M. E. Portnoi, Phys. Rev. B 81, 245431 (2010). 17 F. Wilczek, Nature Phys. 5, 614 (2009). 18 S. Y. Zhou, G.-H. Gweon, A. V. Fedorov, P. N. First, Effects in Semiconductors (IPST, Jerusalem, 1975). 34 P. S. Pershan, J. Appl. Phys. 38, 1482 (1967). 35 P. J. Stephens, Ann. Rev. Phys. Chem. 25, 201 (1974). 

1911.11104

1

1911

Cavity Optomagnonics

In the recent years a series of experimental and theoretical efforts have centered around a new topic: the coherent, cavity-enhanced interaction between optical photons and solid state magnons. The resulting emerging field of Cavity Optomagnonics is of interest both at a fundamental level, providing a new platform to study light-matter interaction in confined structures, as well as for its possible relevance for hybrid quantum technologies. In this chapter I introduce the basic concepts of Cavity Optomagnonics and review some theoretical developments.

cond-mat.mes-hall

cond-mat

Cavity Optomagnonics Silvia Viola Kusminskiy Max Planck Institute for the Science of Light Staudtstrasse 2, 91058 Erlangen, Germany and Friedrich-Alexander University Erlangen-Nuremberg Staudtstrasse 7, 91058 Erlangen, Germany 9 1 0 2 v o N 5 2 ] l l a h - s e m . t a m - d n o c [ 1 v 4 0 1 1 1 . 1 1 9 1 : v i X r a Abstract In the recent years a series of experimental and theoretical efforts have centered around a new topic: the coherent, cavity-enhanced interaction between optical photons and solid state magnons. The resulting emerging field of Cavity Opto- magnonics is of interest both at a fundamental level, providing a new platform to study light-matter interaction in confined structures, as well as for its possible relevance for hybrid quantum technologies. In this chapter I introduce the basic concepts of Cavity Optomagnonics and review some theoretical developments. 1 Contents 1 Introduction 2 Optomagnonic Hamiltonian 2.2.1 Homogeneous magnon mode 2.2.2 Magnetic textures 2.1 Faraday Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Optomagnonic coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Total Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Free Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Driving term . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Total Hamiltonian for the Kittel mode . . . . . . . . . . . 2.3.4 Total Hamiltonian and linearization . . . . . . . . . . . . 3 Equations of Motion 3.1 Heisenberg equations of motion . . . . . . . . . . . . . . . . . . 3.2 Dissipative terms . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Landau-Lifschitz-Gilbert equation of motion . . . . . . . . 3.2.2 Coupling to an external bath for the photon field . . . . . 3.3 Light-induced dynamics of a classical macrospin . . . . . . . . . . 4 Optomagnonics with a magnetic vortex 4.0.1 Magnetic vortex . . . . . . . . . . . . . . . . . . . . . . . 4.0.2 Optomagnonic coupling for the gyrotropic mode . . . . . 5 A quantum protocol: all-optical magnon heralding 5.1 Hamiltonian and Langevin Equations of Motion . . . . . . . . . . 5.2 Write and read protocol . . . . . . . . . . . . . . . . . . . . . . . 5.3 Solution of the linear quantum Langevin equations . . . . . . . . 5.4 Probability of heralding a magnon . . . . . . . . . . . . . . . . . 5.5 Magnon cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Outlook 2 3 7 7 9 11 15 17 17 18 19 19 21 21 22 22 22 24 29 29 31 35 35 37 39 40 40 43 Chapter 1 Introduction The last two decades have seen enormous advances towards the realization of quantum technologies [1]. The ability to bring systems into the quantum regime, to design them, and to control them, can enable ultra-sensitive measurement and the manipulation of information at the quantum level, from quantum computers [2] to a quantum internet [3]. At the same time, it permits testing the predictions of quantum mechanics at unprecedented macroscopic scales [4]. Harnessing the power of quantum mechanics for applications implies being able to design a system for a certain desired quantum functionality. In gen- eral this means going beyond the single-atom limit, into the mesoscopic regime. Mesoscopic systems are comprised of millions of atoms and their behavior is de- scribed by collective excitations: e.g. the mechanical vibrations of a nanobeam. These systems, with characteristic length scales from tens of nanometers to hundreds of microns, are such that their collective excitations can be designed and brought into the quantum regime. This is however very challenging, since quantum states are fragile and require temperatures lower than the frequencies of the corresponding collective excitations. Breakthrough experiments in 2011 used active cooling to bring a macroscopic mode of mechanical vibration into its quantum ground state [5, 6], by using the backaction of electromagnetic ra- diation in a cavity. This is an example of cavity optomechanical systems [7]. Very generally, a cavity is a "box" which serves to confine the electromagnetic fields and can be used to enhance and even modify the interaction between electromagnetic radiation and matter. The field of cavity optomechanics has evolved rapidly, showing, for example, the possibility of entangling micrometer sized oscillators via the optomechanical interaction [8]. Cavity optomechanical systems form part of a broader class of systems de- nominated hybrid quantum systems [9]. These combine different degrees of free- dom, such as photonic, mechanical, electronic, or magnetic, with the aim of con- trolling and optimizing their quantum functionality. For example, while quan- tum information can be processed with superconducting qubits at microwave frequencies [10], transmitting the information through long distances and at room temperature can be done with optical photons, due to their much higher 3 CHAPTER 1. INTRODUCTION 4 frequencies. In turn, storing quantum information requires systems with long coherence times such that they can act as quantum memories. Promising results have been obtained in this regard using ensembles of spin impurities in a solid matrix [11]. In recent years, solid state magnetic systems have emerged as promising candidates for integrating them in hybrid quantum systems. Current research directions include spintronics [12], which aims at using the spin degree of free- dom as a carrier replacing the electron, with the advantage of no energy loss due to Joule heating. Magnetic and electronic degrees of freedom couple well, and the concept of current-induced spin torque, proposed theoretically in 1996 [13], is being nowadays used in the development of random access memories [14]. The field of spin mechanics, in turn, deals with the coupling of the spin and mechanical degrees of freedom [15]. A hybrid spin mechanical system in- corporating also an optomechanical cavity has been moreover demonstrated for ultra-sensitive magnetometry [16]. The interaction between electromagnetic radiation and magnetically ordered solid state systems in the context of hybrid quantum systems has however been an unexplored path until quite recently [17]. This changed with seminal ex- periments from 2013 to 2015, in which strong coherent coupling between mi- crowave photons and magnons in Yttrium Iron Garnet (YIG) was demonstrated [18, 19, 20, 21], following a theoretical proposal in 2010 [22]. Magnons are the collective elementary excitations of magnetic systems, the quanta of the corre- sponding spin waves in the material. In these experiments, a microwave cavity was used to enhance the spin-photon interaction. The collective nature of the magnons, involving all spins in the magnetic sample, also provides a factor of enhancement to the magnon-photon coupling. The microwave field in the cavity can serve moreover as an intermediary field to couple the YIG magnons coher- ently to a superconducting qubit, indicating the potential of these magnetic systems for quantum information platforms [23]. In 2016, the first experimental [24, 25, 26] and theoretical [27, 28] works on cavity optomagnonics appeared, in which an optical cavity enhances the interaction between the spins and optical photons. Since MW photons and the probed YIG magnons have similar energies in the GHz range, the magnon-phonon coupling can be tuned to be resonant, for example by applying an external magnetic field which controls the frequency of the magnonic excitations. The coupling term is of the form gMW S+a† + S−a , (1.0.1) (cid:16) (cid:17) written in terms of the spin ladder operators S± and the microwave photons operators a(†). The first term creates a photon in mode a by annihilating a magnon, and vice-versa for the second term. The coupling strength between magnons and photons gMW is enhanced with respect to the single-spin coupling g0 due to the collective character of the magnons by a factor √N, where N is the number of spins participating in the magnon mode [22, 17]. MW cavity systems with magnetic elements can be by now routinely brought into the strong CHAPTER 1. INTRODUCTION 5 coherent coupling regime, with coupling strengths in the order of hundreds of MHz [29, 30, 31, 32, 33, 34]. In this regime, photons and magnons hybridize, forming a quasiparticle denominated a magnon polariton. Strong coupling is a prerequisite for quantum information manipulation, since it indicates the rate at which information is transferred between the different degrees of freedom. New routes towards tunability and quantum control [35, 36, 37, 38, 30, 39], and on-chip [31, 40, 41] realizations of these systems are also starting to be explored. The frequency of optical photons is, on the other hand, in the range of hundred THz and the coupling to magnons is necessarily parametric, giving rise to inelastic Brillouin scattering [42]. In its simplest form, the coupling reads gOP Sia†a, (1.0.2) where Si is the i = x, y, or z component of the spin operator and in this case optical photons operators a(†). The spin-photon coupling gOP in this regime is inherently weak. The use of an optical cavity has been predicted to boost the interaction and, under certain conditions, to allow the system to enter in the strong coupling regime [27, 28, 43, 44, 45]. These conditions are how- ever challenging, and the current experimental implementations are far from the strong coupling regime. In particular, given the weakness of the intrin- sic interaction, optimal mode matching between magnon and photonic modes is required. Given the complexity of structured magnetic systems, this neces- sitates nontrivial theoretical design and challenging experimental implementa- tion. The challenge makes for exciting times in cavity optomagnonics research [46, 44, 47, 48, 49, 50, 51, 52, 45]. It is to be expected that the shortcomings in the coupling will be overcome, opening the door to applications in quan- tum platforms. For example, magnons could be used as a quantum transducer, converting information up or down between MW and telecom photons. Besides applications, cavity optomagnonics provides a unique setup in which concepts of cavity quantum electrodynamics (QED) can be applied to a mag- netic system and its excitations. Originating from studying the electromag- netic radiation emission and absorption properties of single atoms in a cavity [53, 54, 55], cavity QED is nowadays a well-established framework to study light-matter interaction with confined electromagnetic fields. The concepts have been extended with great success to electromagnetic circuits [56] (circuit QED) and, as mentioned above, optomechanics. The extension to magnetic systems promises rich physics to discover. The following sections cover the basics of cavity optomagnonics, restricted to the coupling of magnetic systems to photons in the optical domain. We derive the optomagnonic Hamiltonian starting from the Faraday rotation of light in magnetized solids and analyze in detail two solvable limits. In one case, we treat the interaction of light with the homogeneous magnon mode of the magnetic material (denominated the Kittel mode). For this mode, all spins precess in phase and they can be treated as a single degree of freedom, consisting of a macrospin. This allows for treating arbitrary dynamics of the macrospin, including nonlinear dynamics away for the equilibrium point. In the other case, CHAPTER 1. INTRODUCTION 6 we study the coupling to arbitrary magnon modes but restricted to the spin- wave limit, where only small deviations of the spins from their equilibrium can be treated. In this limit we can address the coherent interaction of light with magnetic textures. We will derive both the quantum Langevin and semiclassical equations of motion for the coupled open quantum system, and study the spin induced dynamics due to the light in the cavity. Finally, we will go over a proposal of a quantum protocol for creating non-classical macroscopic states of the magnetic system by using light. Chapter 2 Optomagnonic Hamiltonian Cavity optomagnonic systems rely on the interaction between magnetic insula- tors and electromagnetic fields at optical frequencies. At these high frequencies, the magnetic permeability of the material can be taken as that of the vacuum, µ0, and the magneto-optical interaction modeled solely through the dielectric permittivity tensor, εij [57]. One is therefore interested in the coupling between the electric field component of the electromagnetic wave, and the magnetiza- tion of the host material. This coupling is responsible for the classical Faraday effect, where the plane of polarization of light is rotated as the light propa- gates through the magnetized medium, see Fig. (2.1.1). In this configuration, a linearly polarized plane wave propagates along the magnetization direction, and the resulting Faraday rotation of the polarization's plane per unit length of propagation is given by θF, which is a characteristic of the material (sometimes the Faraday rotation is given in terms of the Verdet constant of the material, which is the angle of rotation per unit length, per unit magnetic field). 2.1 Faraday Rotation The rotation of the plane of polarization of the light can be understood at a phenomenological level by noting that, if one writes the linear polarization in a circularly polarized basis, right and left polarizations inside of the material are not equivalent since time reversal symmetry is broken due to the magnetization M. Effectively, this results on a different index of refraction for the two circular polarizations, which accumulates as a phase difference as light propagates, and results on the rotation of the polarization in the linear basis. This phenomenon is also known as magnetic circular birefringence and it is non-reciprocal, that is, the acquired phase adds up if the propagation direction is reversed. Using the zero energy loss condition, implying that no energy from the elec- tromagnetic wave is absorbed in the material, together with the Onsager reci- procity condition for response functions in the presence of a magnetic field, one can derive symmetry conditions on the permittivity tensor of the magnetic ma- 7 CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 8 Figure 2.1.1: Faraday rotation of linearly polarized light propagating through a magnetized material. The sketch shows the Faraday configuration, where the light propagates along the magnetization direction. After a length L, the plane of polarization has rotated an angle θFL in the plane perpendicular to the propagation direction. terial. The zero-loss condition is a good approximation for transparent media. One finds that the permittivity tensor (i) is Hermitian, (ii) its real part is sym- metric in the magnetization, and (iii) its imaginary part is antisymmetric in the magnetization [58, 57, 59]. It is easy to see that a matrix of the form εij(M) = ε0 (εrδij − if
ijkMk) (2.1.1) where
ijk is the Levi-Civita tensor and ijk are spatial indices, fulfills condi- tions (i) to (iii). Throughout this chapter we use the Einstein convention of summation over repeated indices. Eq. (2.1.1) assumes that the unmagnetized material is isotropic, this can, however, be readily generalized to non-isotropic materials by simply replacing εrδij by the corresponding (symmetric and real) permittivity tensor. The linear dependence of Eq. (2.1.1) on the magnetization is valid as long as the correction of the magnetization on the permittivity is small (f M (cid:28) εr), which is usually the case. By considering a linearly polarized plane wave propagating along the material, it is straightforward to show that the permittivity from Eq. (2.1.1) leads to the Faraday rotation of light (see e.g. Ref. [59]). Denoting the saturation magnetization by Ms, f is related to the Faraday rotation coefficient by the expression θF = ω 2c√εr f Ms (2.1.2) where c = 1/√ε0µ0, ω is the angular frequency of the light, and the condition f Ms (cid:28) εr has been used. ✓FLLMagnetization M<latexit sha1_base64="bHsELfbH1mMaVgzae30NOiz9bFA=">AAAB8XicbVDLSgMxFL3js9ZX1aWbYBFclZkq6LLoxo1QwT6wLSWT3mlDM5khyQhl6F+4caGIW//GnX9jpp2Fth4IHM65l5x7/FhwbVz321lZXVvf2CxsFbd3dvf2SweHTR0limGDRSJSbZ9qFFxiw3AjsB0rpKEvsOWPbzK/9YRK80g+mEmMvZAOJQ84o8ZKj92QmpEfpHfTfqnsVtwZyDLxclKGHPV+6as7iFgSojRMUK07nhubXkqV4UzgtNhNNMaUjekQO5ZKGqLupbPEU3JqlQEJImWfNGSm/t5Iaaj1JPTtZJZQL3qZ+J/XSUxw1Uu5jBODks0/ChJBTESy88mAK2RGTCyhTHGblbARVZQZW1LRluAtnrxMmtWKd16p3l+Ua9d5HQU4hhM4Aw8uoQa3UIcGMJDwDK/w5mjnxXl3PuajK06+cwR/4Hz+AL0wkPU=</latexit>Light's polarizationCHAPTER 2. OPTOMAGNONIC HAMILTONIAN 9 In optomagnonic systems, one is usually interested in coupling light to the excitations of the magnetically ordered ground state, and not to the ground state itself. From a classical point of view, these excitations are time-dependent deviations of the magnetization with respect to the ground state, in a collective, phase-locked way, and constitute spin waves. Their respective quanta are de- nominated magnons. Hence, in optomagnonics the usual configuration between the optical fields and the magnetization is not the one from Faraday's original experiment, where the light propagates along the magnetization direction, but perpendicular to it. This configuration is denominated the Voigt configuration. The Faraday configuration can however also be probed and leads to interesting effects concerning angular momentum conservation rules, as discussed in Ref. [52] and also in the previous chapter of this book. In cavity optomagnonics, the optical fields form standing waves in a cavity, and part of the task is to opti- mize the coupling between the so-called optical spin density and the magnetic excitations, as we show below. We are therefore interested in contributions to the permittivity tensor that are linear in the deviations of the magnetization M = M0 + δM, where M0 is the ground state magnetization and δM the deviation. It is easy to see that terms quadratic in M, not included in Eq. (2.1.1), also give linear contributions in δM [42]. The quadratic contribution in M to the permittivity gives rise to the Cotton-Mouton effect, also referred to as magnetic linear birefringence. In contrast to the Faraday effect, the Cotton-Mouton effect is reciprocal. In the simplest situation, the ground state magnetization is uniform and equal to the saturation magnetization, M0 =Ms. Defining the z-axis along Ms, the deviations of the magnetization can be written as δM =(Mx, My, 0). In this case, the only finite components of εij(δM) to first order in δM, including the Cotton-Mouton term, are given by εyz(δM) = ε∗zy(δM) = −iε0f Mx + 2ε0g44MyMs εxz(δM) = ε∗zx(δM) = iε0f My + 2ε0g44MxMs (2.1.3) where g44 is the Cotton-Mouton coefficient related to the corresponding rotation angle per unit length θCM = g44M 2 s , (2.1.4) ω 2c√εr which can be obtained via linear birefringence measurements [57]. In the fol- lowing we will discuss the optomagnonic coupling focusing only on the Faraday term, considering a permittivity tensor of the form given by Eq. (2.1.1). The ex- tension to include Cotton-Mouton terms is however straightforward. From Eq. (2.1.3) one can see that these terms introduce an asymmetry in the couplings [28]. 2.2 Optomagnonic coupling If we consider a medium in the presence of time-dependent electromagnetic fields, we can define an internal energy by taking the time average of the instan- CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 10 taneous energy. For non-dispersive media the permittivity tensor is independent of frequency, and it can be shown that the averaged internal energy can be writ- ten as [58] EEM = 1 4 dV (E∗i εijEj + H∗i µijHj) , (2.2.1) where E is the electric field, H the magnetic field, µij the permittivity ten- sor, and we have used the complex representation of the fields, i.e. such that Re{E} = 1/2 (E∗ + E). It is straightforward to show that the expression Eq. (2.2.2) is real. The magnetization-dependent part of the permittivity introduces a correction to the electromagnetic energy. Considering for simplicity only the Faraday rotation term in the permittivity (see Eq. (2.1.1)), from Eq. (2.2.1) it is straightforward to show that this correction is given by (cid:88) ij (cid:88) (cid:104) ξ UMO = − i 4 f ε0 drM(r) · [E∗(r) × E(r)] . The cross product term is proportional to the optical spin density Slight(r) = ε0 2iω [E∗(r) × E(r)] (2.2.2) (2.2.3) implying the need of a non-trivial polarization of the optical field to obtain a finite coupling term. One is usually interested in problems where the magnetization M has a dy- namical part, M(r, t) = M0(r)+δM(r, t), where δM(r, t) is the term due to the spin-wave excitations. The static ground state can be uniform (M0(r) ≡ Ms) if the sample is saturated by an external magnetic field, or also in the case of nanometer samples, where the exchange interaction is predominant and -- for a ferromagnetic interaction -- , sufficient to align all spins. In general, the in- terplay of exchange interactions and dipole-dipole interactions (or other type of interactions, such as Dzyalozinskii-Moriya [60]) gives rise either to the formation of domains in macroscopic samples, or to textured ground states for interme- diate sizes (typically microns) [61], where the surface to volume ratio is large enough as to make the boundaries relevant for the minimization of the magne- tostatic energy [62]. The formation of domains or textures have an exchange energy cost, but minimize stray fields. The relevant term for the optomagnonic coupling is the dynamical part of the magnetization δM(r, t). Therefore Eq. (2.2.2) implies that, besides a nontrivial polarization of the optical fields, the symmetry of the modes should be such that the integral is finite. In particular, the overlap between magnetic and optical modes should be maximized. Quantizing Eq. (2.2.2) leads to the optomagnonic coupling Hamiltonian for ferromagnets [27]. The electric fields E(r, t) can be readily quantized in terms of bosonic creation a†ξ and annihilation operators aξ, E(r, t) → E(r, t) = 1 2 Eξ(r)aξ(t) + E∗ξ(r)a†ξ(t) (2.2.4) (cid:105) CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 11 where ξ labels the corresponding optical mode including the polarization index. The mode functions Eξ(r) satisfy the Helmholtz equation (cid:0) ∇2 + n2k2 0 (cid:1) Eξ(r) = 0 , (2.2.5) (cid:104) a†ξ, a†ξ(cid:48) (cid:105) (cid:105) where n is the index of refraction of the medium and k0 the vacuum wave vec- tor. The mode functions are found from Eq. (2.2.5) together with appropriate boundary conditions for the geometry and material of the optical cavity [63]. The photon operators obey the usual bosonic commutation rules = δξξ(cid:48) (δξξ(cid:48) the Dirac delta) and [aξ, aξ(cid:48)] = aξ, a†ξ(cid:48) = 0. (cid:104) r, sj r algebra(cid:2)si (cid:3) = i
ijksk The magnetization, in turn, can be quantized in terms of local spin operators sr (r indicates the position of the spin) fulfilling locally the angular momentum r and commuting otherwise. The spin operators can be written exactly in terms of bosonic ones via a Holstein-Primakoff transformation. In order to preserve the algebra, however, this transformation is necessarily nonlinear and introduces extra interaction terms between magnons [59]. There are two cases that one can treat, up to a certain extent, analytically: (i) the uniform case, in which the ground state is uniform and one is interested in the homogeneous, k = 0 spin-wave mode, denominated the Kittel mode, and (ii) the general, spatial-dependent case in the spin-wave limit, valid for small deviations of the spins from their equilibrium configuration. We give the resulting form of the optomagnonic Hamiltonian for the two cases in the following. 2.2.1 Homogeneous magnon mode We start with the homogeneous case (i), where both the ground state magne- tization and the excitation are spatially independent: M(r, t) = Ms + δM(t), with Ms the uniform saturation magnetization. In this case, the magnetization can be quantized simply in terms of a macrospin S M Ms → S S , (2.2.6) where S is the total spin of the considered system. This quantization scheme allows to retain the spin algebra and to treat fully the nonlinearity of the problem (δM(t) does not need to be small), but it is restricted to the homogeneous case. From Eqs. (2.2.6), the optomagnonic coupling Hamiltonian in this case reduces to [27] (2.2.4) and (2.2.2), using the substitution rule Eq. HM O =  SjGj βγ a†β aγ (2.2.7) with coupling constants Gj βγ = −i ε0f Ms 4S jβγ ξ
jmn drE∗βm(r)Eγn(r) , (2.2.8) (cid:88) CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 12 Figure 2.2.1: Schematic representation of the geometry for the calculation of the optomagnonic coupling to the Kittel mode, Eqs. (2.2.12) and (2.2.13). Adapted from Ref. [59]. where the Greek indices label the optical modes, and the Roman indices label the spatial components (x, y, z). The factor ξ ≤ 1 is a measure of the overlap between the Kittel mode and the corresponding optical modes, with ξ = 1 corresponding to optimal mode-matching. The Gj are hermitian matrices which in general cannot be simultaneously diagonalized. One sees that there are two possible kinds of processes: intra-mode coupling, given by the diagonal elements of Gj, and inter-mode coupling, given by the off-diagonal elements. The coupling constants Gj βγ are uniquely determined once the normalization of the electromagnetic field is specified. We follow the normalization procedure common in optomechanical systems, where the electromagnetic field amplitude is normalized to one photon over the EM vacuum [64]. The normalization condition is given by ωα = ε0ε(cid:104)α d3r E(r)20(cid:105), where α(cid:105) is a state with a single photon in mode α, and 0(cid:105) is the cavity vacuum. One obtains (2.2.9) d3r E(r)2α(cid:105)−ε0ε(cid:104)0 ´ ´ ωα = 2ε0ε drEα(r)2 . With this normalization, Eq. (2.2.8) reads ´ (cid:113)´ √ωβωγξ
jmn (cid:113)´ Gj βγ = −i f Ms 8Sε drE∗βm(r)Eγn(r) drEβ(r)2 drEγ(r)2 . (2.2.10) For processes involving a single optical mode, as we already pointed out, some degree of circular polarization of the mode is necessary for a finite coupling. Assuming an optical mode circularly polarized in the yz plane, the optical spin density is along the x axis and couples to the x component of the spin operator, Kittel modeOptical modeE⇤⇥EexeyezezΩMs<latexit sha1_base64="NkfavkyOt2LOKRjxnLeHqaMp7Co=">AAAB/HicbVDLSsNAFJ3UV62vaJduBovgqiRV0GXRjRuhgn1AE8JkOmmHzkzCzEQoIf6KGxeKuPVD3Pk3TtostPXAwOGce7lnTpgwqrTjfFuVtfWNza3qdm1nd2//wD486qk4lZh0ccxiOQiRIowK0tVUMzJIJEE8ZKQfTm8Kv/9IpKKxeNCzhPgcjQWNKEbaSIFd9zjSkzDK7vIg8yTPVJ4HdsNpOnPAVeKWpAFKdAL7yxvFOOVEaMyQUkPXSbSfIakpZiSveakiCcJTNCZDQwXiRPnZPHwOT40yglEszRMaztXfGxniSs14aCaLqGrZK8T/vGGqoys/oyJJNRF4cShKGdQxLJqAIyoJ1mxmCMKSmqwQT5BEWJu+aqYEd/nLq6TXarrnzdb9RaN9XdZRBcfgBJwBF1yCNrgFHdAFGMzAM3gFb9aT9WK9Wx+L0YpV7tTBH1ifP8IllX0=</latexit>CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 13 (2.2.1)). The relevant coupling matrix is therefore Gx Sx (see Fig. αβ which, in the considered geometry, is diagonal in the circularly polarized basis for the optical fields eR/L = (ey ∓ iez) /√2, as one can easily obtain from (2.2.8). We quantize the optical field for simplicity in terms of plane waves E(r, t) → E+(r, t) = i aj(t)eikj·r (2.2.11) E∗(r, t) → E−(r, t) = −i a†j(t)e−ikj·r , (cid:114) ω (cid:114) ω 2εV ej 2εV (cid:88) (cid:88) j ej j where the ± superscripts follows the usual convention which indicates the pos- itive and negative frequency components of the optical field, ω = ωR = ωL, j = R, L and kj the corresponding wave vector. V is the volume of the optical cavity. Using these expressions, one can easily show that Eq. (2.2.10) reduces to G = Gx LL = −Gx RR = 1 S cθF 4√ε ξ (2.2.12) where we have used Eq. (2.1.2) to write f in terms of θF. The numerical factor ξ ≤ 1 takes into account the mode overlap of the electric field with the magnon mode and other geometric factors. In current optomagnonic experiments, in- volving YIG spheres, the optical modes in the cavity are actually whispering gallery modes (WGM), see Fig. (2.2.2). We will discuss a system with optical whispering gallery modes in Sec. (4), for now these details are hidden in ξ. Considering Eqs. (2.2.7), (2.2.11), and (2.2.12), the coupling Hamiltonian reads [27] HM O =  SxG a†LaL − a†RaR . (2.2.13) (cid:112) In the spin-wave approximation, for small oscillations of the macrospin around its equilibrium position, we can replace the spin operator Sx by a position oper- ator Sx → S/2( m + m†) using a Holstein-Primakoff transformation truncated to first order in the bosonic operators (see Eq. (2.2.15)). In this limit, (cid:16) (cid:16) HM O ≈  (cid:112) 1 √2S cθF 4√ε ξ a†LaL − a†RaR ( m + m†) (2.2.14) which is reminiscent of the Hamiltonian in the related field of cavity optome- chanics, where light couples to mechanical vibrations by pressure forces. The coupling g0 = G S/2 (given as an angular frequency) is a measure of the sin- gle photon-magnon coupling and equivalent to the vacuum coupling strength in optomechanics, where g0 is proportional to the zero-point motion of the me- chanical oscillator [7]. As we see from the dependence on 1/√S in Eq. (2.2.14), the single photon-magnon coupling is enhanced by small magnetic volumes. The material of choice for optomagnonic systems is the insulating ferrimag- net YIG, due to the small losses both for the optics (absorption coefficient α ∼ 0.069cm−1 at λ = 1, 2 µm) and for the magnon modes (Gilbert damp- ing coefficient ηG ≈ 10−4) and large Faraday rotation (θF = 240 deg/cm at (cid:17) (cid:17) CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 14 Figure 2.2.2: Sketch of a YIG sphere supporting optical WGM. Within the material, photons and magnons interact via the optomagnonic interaction. The frequency of the magnons can be controlled by an external magnetic field H and the optical modes can be driven and probed by an optical fiber, which couples evanescently to the WGM. λ = 1, 2 µm). Note that although YIG is technically a ferrimagnet, one mag- netic sublattice is dominant and mostly behaves like a ferromagnet. The optical cavity is formed by the magnetic material itself, due to total internal reflection of light inside of the dielectric material. At optical frequencies, the index of re- fraction of YIG is n = √ε ≈ 2.24, which combined with its low absorption makes it a reasonably good optical cavity if patterned appropriately. Experiments so far have used YIG spheres, since they are commercially available and relatively easy to polish into small sizes while preserving the quality of the optical cavity. Sizes nevertheless remain still too large, in the range of 100µm radius. The YIG sphere supports optical modes in the way of whispering gallery modes which can be accessed through a tapered fiber, see the scheme of Fig. (2.2.2). Assuming optimal mode-matching and a diffraction-limited volume of YIG of 1µm3, one obtains G S/2 ≈ 0.1MHz [27], which would be comparable to state of the art optomechanical systems [65]. Current experimental setups are still far from this limit, due to fabrication and design issues. Note for example that for the case of a sphere, the Kittel mode is a bulk mode, whereas the WGMs live near the surface, leading to a small overlap between the modes. Improving the current values of the coupling is however highly desirable for applications in the quantum regime. (cid:112) YIGinoptical fiberWhisperingGalleryModesoutHmagneticfieldCHAPTER 2. OPTOMAGNONIC HAMILTONIAN 15 2.2.2 Magnetic textures One alternative to improve the value of the optomagnonic coupling is to go beyond the Kittel mode, searching for modes that would be better suited for mode matching with the optics. This is starting to be explored both theoretically [66, 44, 45] and experimentally [50, 49]. This brings us to case (ii) from our two limiting cases, where one allows for non-uniform ground states (also called magnetic textures) and/or magnetic excitations with a spatial structure, and uses the Holstein-Primakoff transformations to represent the excitations in terms of bosonic operators mi, where i is the lattice site: m†i mi 2s mi (cid:115) i = √2s S+ (cid:16) 1 − s − m†i mi S−i = √2s m†i Sz i = (cid:115) 1 − (cid:17) m†i mi 2s . (2.2.15) In these, s is the total spin per lattice site i, so that the total spin is given by S = N s with N the number of lattice sites, and S±i = Sx i are the spin ladder operators. Eqs. (2.2.15) assume a quantization axis along the z direction. If the magnetic ground state is textured, the quantization axis is local, defined by ez(r) = Ms(r)/Ms. The bosonic operators mi fulfill the usual i ± i Sy bosonic commutation rules(cid:104) (cid:105) (cid:104) mi, m†j = δij [ mi, mj] = m†i , m†j (cid:105) = 0 . (2.2.16) The problem is simplified by cutting off the Holstein-Primakoff transformation to first order in the bosonic operators, √2S mi √2S m†i S+ i ≈ S−i ≈ Sz i ≈ S , (2.2.17) and is therefore a linear approximation for the local spin operators, which are treated as harmonic oscillators. The elementary magnetic excitations are col- lective, since given a spin Hamiltonian (e.g. the Heisenberg Hamiltonian), after performing the approximation Eq. (2.2.17) one still needs to bring the Hamilto- nian to a diagonal form, so that is is a sum of independent harmonic oscillators (e.g. in the bulk by going to Fourier space, mk). These collective excitations are denominated magnons: essentially, one magnon is a "flipped" spin which is shared by the whole system. Higher-order terms in the expansion can be included and represent magnon-magnon interactions. CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 16 The quantization of the coupling term Eq. (2.2.2) in this case follows by writing the excitation δM(r, t) in terms of the magnon modes (by magnon modes we mean the bosonic operators which diagonalize the magnetic Hamiltonian). It is convenient to work in terms of the normalized magnetization δm(r, t) = δM(r, t)/Ms. For small deviations δm (cid:28) 1 we can quantize the spin wave in analogy to Eq. (2.2.4) for the electric fields, by the substitution (cid:2)δmγ(r) mγ + δm∗γ(r) m†γ (cid:3) , (cid:88) γ δm(r, t) → 1 2 (2.2.18) where m(†) γ annihilates (creates) a magnon in mode γ. The information on its spatial structure is contained in the mode functions δmγ(r). Together with Eq. (2.2.4), from Eq. (2.2.2) and using Eq. (2.1.2) we obtain the optomagnonic coupling Hamiltonian linearized in the spin fluctuations [44] (cid:88) (cid:88) HM O =  Gαβγ a†αaβ mγ +  G∗αβγ a†β aα m†γ (2.2.19) αβγ αβγ where Gαβγ = −i θFλn 4π ε0ε 2 dr δmγ(r) · [E∗α (r) × Eβ (r)] (2.2.20) is the optomagnonic coupling in terms of the Faraday rotation per wavelength of the light in the material λn = λ0/n, with λ0 the vacuum wavelength. For YIG one obtains θFλn/2π ≈ 10−5. Within the linear regime for the spins, Eq. (2.2.21) is very general and allows to treat arbitrary geometries and modes, both optical and magnetic. In Eq. (2.2.20) one still needs to specify the normalization of the modes. For the optical fields the normalization was given in Eq. (2.2.9). For the magnon modes, we impose a total magnetization corresponding to one Bohr magneton (times the corresponding gyromagnetic factor g) in the excitation. This nor- malizes the coupling to one magnon. From the definition Eq. (2.2.18), this is equivalent to imposing [44] 1 4 drδmγ(r)2 = gµB Ms . (cid:114) gµB (2.2.21) (2.2.22) . The normalized coupling therefore reads Gαβγ = −i (cid:113)´ × θFλn 4π 1 2 ´ √ωβωα Ms (cid:113)´ dr δmγ(r) · [E∗α (r) × Eβ (r)] (cid:113)´ drEβ(r)2 drEα(r)2 drδmγ(r)2 ´ For the optical fields, it is common to define an effective mode volume V α E V α E = d3rEα(r)2 max{Eα(r)2} , (2.2.23) CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 17 which for a homogeneous electric field reduces simply to the volume occupied by the field. Analogously, we can define an effective magnetic volume , (2.2.24) ´ V γ M = d3rδmγ(r)2 max{δmγ(r)2} ´ according to which Gαβγ = −i θFλn 4π 1 2 (cid:114) gµB Ms (cid:113) √ωβωα V γ MV α E V β E dr δmγ(r) · [E∗α (r) × Eβ (r)] δmγ(rm)Eα(rm)Eβ(rm) , (2.2.25) where for simplicity of notation we have defined rm such that δmγ(rm) = max{δmγ(r)} and analogously for Eα. From Eq. (2.2.25) we see that the strength of the coupling is cut off by the smallest volume in the integral factor. Assuming similar effective mode volumes E the coupling is suppressed by for the optical fields, V α E instead, the E ≡ V α M, favoring small magnetic volumes. Recalling that E , favoring small optical volumes. For V γ a factor (cid:112)V γ coupling goes as 1/(cid:112)V γ M/V α E ≈ V β M ≥ V α E , if V γ M ≤ V α Ms = SgµB V γ M (2.2.26) where S = N s is the total spin in the volume V γ M (N number of spins, s spin value), we recover in this case the behavior ∝ 1/√S found above for the Kittel mode in the spin-wave approximation. From these scaling arguments, we see that small mode volumes and optimal mode matching are required for large coupling. In Sec. (4) we will use Eq. (2.2.25) to calculate the optomagnonic coupling in a cavity system consisting of a micromagnetic disk. 2.3 Total Hamiltonian In the previous section, we derived the Hamiltonian that governs the coupling between optical photons and magnons in a cavity. In order to study dynamical processes, we need the total Hamiltonian of the system. Besides the coupling Hamiltonian, we need to include the free Hamiltonian both for magnons and photons (the kinetic terms). The cavity is an open system, which can be driven and is also subject to dissipation, both for magnons and photons. We can include the driving term in the Hamiltonian, the dissipation terms we will include at the level of the equations of motion in Sec. (3). 2.3.1 Free Hamiltonian The total optomagnonic Hamiltonian H consists of the optomagnonic coupling term, given by either Eqs. (2.2.7) and (2.2.8) or Eqs.(2.2.19) and (2.2.20), plus the free photon Hamiltonian Hph =  ωαa†αaα (2.3.1) (cid:88) α CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 18 and the free magnetic term Hm. For the Kittel mode (case (i) from the previous section), this free term is simply the Larmor precession of the macrospin S, H K m = −Ω Sz , (2.3.2) where we have assumed an external magnetic field B0 applied along the ez axis and Ω is the free precession frequency of the macrospin, see Fig. (2.2.1). We assume also that the ground state magnetization is saturated and along ez. The frequency Ω is in general controlled by B0. In the case of a spherical magnet, due to the high symmetry of the system Ω is independent of the demagnetization fields [62], and given simply by Ωsphere = gµBB0 . (2.3.3) For YIG, g = 2 and the gyromagnetic factor equals that of the electron: γe = gµB/ = 1.76 × 1011rad/s · T. Applied magnetic fields in the range of tens of mT therefore lead to frequencies in the GHz range. For other geometries, e.g. ellipsoids or thin films, Ω depends also on the demagnetization fields which can be taken into account through demagnetization factors [67]. For general magnon modes in the spin-wave approximation (case (ii)), one writes the free magnetic term also as a sum of harmonic oscillators H SW m =  Ωβ m†β mβ (2.3.4) β where Ωβ is the dispersion of the β magnon mode. Part of the problem in confined geometries is finding the magnon modes and corresponding dispersion, and, except for very simple geometries, micromagnetic simulations must be employed. Note that by using the Holstein-Primakoff expression for Sz, c.f. Eq. (2.2.15), Eq. (2.3.2) reduces, asides from a constant, to an expression like Eq. (2.3.4). 2.3.2 Driving term The cavity system can be driven by an external laser. The magnon modes can in principle also be driven by an external MW field, but we will not consider this in the following. The driving term can be included in the Hamiltonian as (cid:88) where α indicates the mode that is being driven, ωL is the laser frequency and HD = i
α(aαeiωLt − a†αe−iωLt) , (cid:114) 2καPα ωL
α =  (2.3.5) (2.3.6) depends on the driving laser power Pα and on the cavity decay rate κα of the pumped mode due to the coupling to the the driving channel, e.g. an optical fiber or waveguide. It is common to work in a rotating frame at the CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 19 laser frequency ωL, so that the trivial time dependence eiωLt is removed. This † is achieved by the unitary transformation U = e−iωLta αaα under which the † Hamiltonian transforms as H → U H U† − i U ∂ U ∂t . In the rotating frame, for a single photon mode α one obtains Hph + HD → −∆αa†αaα + i
α(aα − a†α) (2.3.7) where ∆α = ωL−ωα is the detuning of the driving laser frequency with respect to the resonance frequency of the optical cavity for the α-mode. The generalization to multiple driven modes is straightforward. If ∆α > 0 (∆α < 0 ) the system is said to be blue (red) detuned. In the literature, it is usual to write the Hamiltonian in the rotating frame omitting the driving term (second term on the RHS of Eq. (2.3.7)). In that case, the driving term is added at the level of the equations of motion, together with the dissipation and fluctuation terms. 2.3.3 Total Hamiltonian for the Kittel mode The total cavity optomagnonic Hamiltonian, in the rotating frame and omitting the driving and dissipation terms, are given in the following both for the Kittel mode (case (i)) and in the spin-wave approximation (case (ii)). For the Kittel mode we choose a simplified model in which the light is circularly polarized in the yz plane giving rise to the coupling in Eq. (2.2.13). Hence (2.3.8) HK = −∆a†a − Ω Sz + G Sxa†a with G given by Eq. (2.2.12). Since in this simple case the Hamiltonian is diagonal in the circularly polarized basis, right and left handed modes are not coupled and we can restrict the Hamiltonian to a single photon mode, which we denote with the operator a (compare with Eq. (2.2.13)). Note also that, as long as we work with the Voigt geometry, we can always find a system of coordinates such that the Hamiltonian can be expressed as in Eq. (2.3.8). The Hamiltonian in Eq. (2.3.8) seems deceptively simple, since, as we will see below, it leads to rich nonlinear dynamics even in the classical limit. The parametric coupling in the photon operators (the coupling is a two-photon process) gives rise to nonlinearities even in the spin-wave approximation. These are equivalent to the nonlinear behavior present in optomechanical systems [7]. Retaining the full macrospin dynamics introduces new nonlinear behavior unique to optomagnonic systems. 2.3.4 Total Hamiltonian and linearization In the spin-wave approximation the total Hamiltonian reads HSW = − Ωβ m†β mβ +  ∆αa†αaα +  (cid:88) (cid:88) α (cid:88) αβγ β Gαβγ a†αaβ mγ + h.c. (2.3.9) with Gαβγ given in Eq. (2.2.25). This Hamiltonian is a three-particle interacting Hamiltonian. Diagonalization is possible by linearizing the optical fields around CHAPTER 2. OPTOMAGNONIC HAMILTONIAN 20 the steady state solutions, aα = (cid:104)aα(cid:105) + δaα (2.3.10) such that d(cid:104)aα(cid:105)/dt = 0 and all the dynamics is contained in the fluctuation fields δaα. The input laser power determines the average number of photons circulating in the cavity in mode α, nα = (cid:104)aα(cid:105)2. Considering terms up to linear order in the fluctuations δaα, Eq. (2.3.9) reduces to a quadratic Hamiltonian (cid:88) (cid:88) Hfl = − α +  (cid:88) β ∆αa†αaα +  Ωβ m†β mβ (cid:0)√nαδaβ mγ + √nβδa†α mγ Gαβγ (2.3.11) (cid:1) + h.c. . αβγ This kind of Hamiltonian is well know from quantum optics and related systems (see e.g. Refs. [68, 7]), and it can be turned into a parametric amplifier (δaβbγ and m†γδa†β terms) or a beam splitter Hamiltonian (δaβ mγ and m†γδa†β) by tuning the external laser driving frequency. The combination Geff = √nG (2.3.12) (with indices as appropriate) shows that the coupling G is enhanced by the square root of the number of photons trapped in the cavity, and can in this way be controlled. Chapter 3 Equations of Motion In this section we split again for simplicity the discussion into the two cases (i) and (ii) detailed above. We will obtain the equations of motion for the macrospin dynamics, only valid for the Kittel mode but retaining the spin algebra and the full non-linearity of the problem, and and the spin-wave approximation, where we restrict the problem to the dynamics of coupled harmonic oscillators. 3.1 Heisenberg equations of motion The Heisenberg equation of motion for an operator O evolving under a Hamil- tonian H is given by In the macrospin approximation using HK given in Eq. (2.3.8) and imposing the = i
ijk Sk one obtains the following Si, Sj coupled equations of motion for the macrospin and the light field (cid:104) commutation relations(cid:2)a, a†(cid:3) = 1, (cid:16) (3.1.1) (3.1.2) d O dt  = i[ H, O] . (cid:105) a (cid:1) a = −i G Sx − ∆ (cid:17) S =(cid:0)Ga†a ex − Ω ez (cid:88) (cid:88) βγ G∗αβγ a†β aα , × S . (cid:88) αβ 21 In the spin wave approximation, considering a general multimode system given by the Hamiltonian in Eq. (2.3.9) we obtain aα = i∆αaα − i mγ = −iΩγ mγ − i Gαβγ aβ mγ − i G∗βαγ aβ m†γ βγ (3.1.3) CHAPTER 3. EQUATIONS OF MOTION 22 and analogously for a†α and m†γ. Eqs. (3.1.2) and (3.1.3) are written in a rotating frame but do not contain either driving or dissipative terms, we will include these below. For now, we note that for G = 0 Eqs. (3.1.2) decouple into a simple harmonic oscillator for the optical field, and a Larmor precession equation for the spin operator. In particular, the equation of motion for the spin in this case is S = −Ω ez × S, which reduces to the well known Landau-Lifschitz equation of motion for the magnetization by taking the classical expectation values and proper rescaling. Note that the Landau-Lifschitz equation of motion is therefore semiclassical, since it is derived as the classical limit of the Heisenberg equation of motion. 3.2 Dissipative terms The dissipative rates in cavity systems are very important since they determine how fast information in the system is lost to the environment. A very important figure of merit in hybrid systems is the cooperativity C, which is defined as the ratio of the effective coupling strength (see Eq. (2.3.12)) to the decay channels in the system, in our case the photon decay rate κ and the magnon decay rate which we take as Γ = ΩηG, C = 4 (3.2.1) Quantum protocols require at least C > 1, so that information transfer can occur before the information is lost. Coherent state transfer between magnons and photons requires moreover that C/ntm > 1, where ntm is the number of thermal magnons (the photon environment can be safely assumed to be at zero temperature for optical photons, since kBT (cid:28) ωph). 3.2.1 Landau-Lifschitz-Gilbert equation of motion To recover the classical dynamics of the spin, it is still necessary to include dissipation. This is done phenomenologically by adding a Gilbert damping term to the Landau-Lifschitz equation G2 eff κΓ . S = −Ω ez × S + ηG S ( S × S) (3.2.2) where ηG is a characteristic of the magnetic material and denominated the Gilbert damping coefficient. For YIG, ηG ≈ 10−4, which is very low when com- pared to other magnetic materials. The Eq. (3.2.2) for the classical macrospin is denominated the Landau-Lifschitz-Gilbert equation of motion. Note that the dissipative term damps the precession of the spin, but does not alter the norm of the vector. 3.2.2 Coupling to an external bath for the photon field The optical cavity fields are also subject to dissipative processes due to the interaction with an environment. This can be modeled by a thermal bath of CHAPTER 3. EQUATIONS OF MOTION 23 harmonic oscillators bk of frequency ωk which couple linearly to the photonic field with strength gk (cid:88) (cid:88) (cid:16) (cid:17) Henv = ωaa†a + ωkb†k bk +  gka†bk + g∗k b†ka . (3.2.3) k k The environment is then integrated out and its effect is taken into account by a dissipation term in the equations of motion for the degree of freedom of interest, in this case, the optical field a [69]. In the Markov approximation, this procedure results in a quantum Langevin equation of motion a(t) = − κ 2 a(t) + F (t) , (3.2.4) where we have already transformed to the rotating frame with frequency ωa. The environment-induced dissipation is encoded in the cavity decay rate κ, (3.2.5) where D(ωa) is the density of states (DOS) of the bath, which in the Markov approximation can be evaluated at the cavity resonance frequency. F (t) is the noise operator κ = πD(ωa)g(ωa)2 F (t) = −i gkbk(0)ei(ωa−ωk)t , (3.2.6) (cid:88) k which represents the random "quantum kicks" of the environment on the cavity mode. The expectation value of the noise operator for a reservoir in thermal equilibrium is zero (3.2.7) and it is related to the cavity decay rate κ by the fluctuation-dissipation theorem (cid:104) F (t)(cid:105)R = 0 ∞ −∞ κ = 1 ¯n dτ(cid:104) F †(τ ) F (0)(cid:105)R (3.2.8) (3.2.9) , with ¯n = (cid:104)b†(Ω)b(Ω)(cid:105)R = e 1 βΩ − 1 where β = 1/kBT , kB the Boltzmann constant and T the temperature of the bath. The vacuum noise correlator is local in time (cid:104)0 F (t(cid:48)) F †(t(cid:48)(cid:48))0(cid:105)R = κδ (t(cid:48) − t(cid:48)(cid:48)) , (3.2.10) in accordance with the Markov approximation: the dynamics of the bath is fast compared to that of the cavity, and it has no memory. In input-output theory, the noise operator is normalized to an operator ain = F /√κ such that (cid:104)0ain(t(cid:48))a†in(t(cid:48)(cid:48))0(cid:105)R = δ (t(cid:48) − t(cid:48)(cid:48)) (3.2.11) CHAPTER 3. EQUATIONS OF MOTION represents the input noise, and Eq. (3.2.4) reads a(t) = − κ 2 a(t) + √κain(t) 24 (3.2.12) in Eq. If the system is driven, the expectation value < ain(t) > is taken over a coherent state and it is finite, whereas for the fluctuations < δain(t) >= 0. The decay rate is in general added to the equations of motion (e.g. (3.1.2)) as a phenomenological parameter, since the microscopic details of the bath are usually not known. There can be several decay channels for the photons in a cavity, due for example to scattering with phonons or leaky mirrors. However, there are also "wanted" decay channels, those which allow probing the cavity (for example, an external optical fiber coupled to the cavity). Sometimes is it useful to split the total decay rate into the unwanted losses κ0 and the losses associated with the input-output channel κin, which contain information about the state of the cavity system and can to a certain degree be tuned. The total decay rate in this case (for a high quality cavity) is simply the sum of the two contributions κ = κin + κ0, and the quantum Langevin equation reads a(t) = − κ 2 a(t) + √κinain(t) + +√κ0 d0(t) , (3.2.13) where d0 is a noise operator associated with the unwanted losses, with zero expectation value (cid:104) d0(cid:105). It is sometimes customary to define the dimensionless parameter η = κ0/κ as the ratio of useful loss to total loss. The procedure of "integrating out" in order to obtain an effective equation of motion for the degrees of freedom of interest is quite general. The Markov approximation is widely used for systems in contact with a thermal reservoir, since its assumptions are in this case well justified. The Langevin equation is an example of an equation of motion which includes backaction. In particular, the dissipative term is the first correction to the instantaneous coupled system-bath dynamics, due to the retardation in the response of the bath to a change in the system of interest. In general the instantaneous response also causes an energy shift in the frequency of the cavity field, which we have ignored in the previous discussion. 3.3 Light-induced dynamics of a classical macrospin We now turn to the classical dynamics of the coupled photon-spin system, follow- ing Ref. [27]. We therefore replace the operators by their classical expectation values a = (cid:104)a(cid:105) and S = (cid:104)S(cid:105), κ 2 (a − αmax) (3.3.1) (3.3.2) a = −i (GSx − ∆) a − S = (Ga∗a ex − Ω ez) × S + ηG S ( S × S) , and ignore the fluctuations. In Eq. (3.3.1) we included the driving laser ampli- tude αmax for the optical mode. This corresponds to the steady state value of a CHAPTER 3. EQUATIONS OF MOTION 25 for the uncoupled system at zero detuning. As we anticipated, this is a highly nonlinear system of equations and the full dynamics can be solved only numer- ically. Analytical progress is however possible in certain limits. In particular, in the fast cavity limit we can follow the previous example for a system in con- tact with a fast environment. In this case, however, we integrate out the cavity field, in order to obtain an effective equation of motion for the macrospin. The term "fast cavity" refers to the limit in which the photons in the cavity decay very rapidly compared to the dynamics of the macrospin, so that a photon in the cavity "sees" mostly a static spin. The fast cavity condition in this case is given by G Sx (cid:28) κ2. We will find that the light field is responsible for extra dissipation and a frequency shift for the spin precession. In order to find the effective equation of motion for the macrospin in this limit, we expand the photon field a(t) in powers of Sx, a(t) = a0(t) + a1(t) + . . . (3.3.3) where the subscript indicates the order in the expansion in Sx. a0(t) there- fore corresponds to the instantaneous equilibrium. From Eq. (3.3.1), imposing a(t) = 0 one finds a0(t) = κ 2 αmax 1 κ 2 − i (∆ − GSx(t)) . (3.3.4) Inserting Eqs. (3.3.3) and (3.3.4) into Eq. (3.3.1) and keeping terms to first order in Sx we obtain the correction a1 a1(t) = − 1 κ 2 − i (∆ − GSx) ∂a0 ∂Sx Sx , (3.3.5) where we have used that a1 includes, by definition, only terms up to first order in Sx. Finally, replacing a2 ≈ a02 + a∗1a0 + a∗0a1 in Eq. (3.3.2) and keeping again terms only up to first order in Sx, we obtain and effective equation of motion for the classical macrospin S S = Beff × S + ηopt S ( Sx ex × S) + ηG S ( S × S) , (3.3.6) with Beff = −Ωez + Bopt. Both Bopt and ηopt are light induced and depend implicitly on time through Sx(t). The quantity Bopt(Sx) = Ga0(Sx)2 ex (3.3.7) is the instantaneous response of the light field and acts as an optically induced magnetic field, consequently giving rise to a frequency shift for the precession of the spin. The second term in the RHS of Eq. (3.3.6) is due to retardation effects, and it is reminiscent of Gilbert damping, albeit with spin-velocity component only along ex due to the geometry we chose, where the spin of light lies along the ex axis. CHAPTER 3. EQUATIONS OF MOTION 26 (cid:17)2 αmax ex (cid:16) κ 2 The optically induced field Bopt and the dissipation coefficient ηopt are highly non-linear functions of the spin coordinate Sx(t), Bopt = G [( κ 2 )2 + (∆ − GSx)2] ηopt = −2GκS Bopt (∆ − GSx) 2 )2 + (∆ − GSx)2]2 , [( κ (3.3.8) (3.3.9) and can be tuned externally by the laser drive, both by the power and by the detuning. The strength of the induced field is controlled by a combination of the input power and the decay rate of the photons in the cavity. The detuning however has a qualitative effect. In particular when the condition ∆ > GSx is fulfilled, the optically induced dissipation is negative, leading to instabilities of the original stable equilibrium points when it dominates over the Gilbert damping ηG. The instabilities are a consequence of driving the system blue- detuned, which pumps energy into the spin system. This can be seen by studying the stability of the north pole (which is the stable solution without driving) once the driving laser is applied. From Eq. (3.3.6) assuming ηG (cid:28) ηopt (this for YIG), we can obtain an equation of condition can be easily achieved e.g. motion for Sx. Setting Sz = S, (3.3.10) we consider small deviations δSx of Sx from the equilibrium position that sat- x. To linear isfies S0 order we obtain x = −SBopt/Ω, where Bopt in Eq. (3.3.7) is evaluated at S0 Sx = −ΩSBopt − Ω2Sx − ηoptΩ Sx , (cid:18) (cid:19) δSx = −Ω Ω + S ∂Bopt ∂Sx δSx (cid:104) (cid:104) + 2GSκΩBopt (∆ + GSBopt/Ω) (κ/Ω)2 + (∆ + GSBopt/Ω)2 (3.3.11) (cid:105)2 δSx . Therefore the effective dissipation coefficient is in this case ηopt ≈ −2GSκBopt (∆ + GSBopt/Ω) (κ/Ω)2 + (∆ + GSBopt/Ω)2 (3.3.12) (cid:105)2 , which is always negative for blue detuning. Comparing with Eq. (3.3.11), we see that the solutions near the north pole are unstable for ∆ > 0. The runaway solutions can fall either into a new static equilibrium point at more or less the opposite pole on the Bloch sphere (aligned with the equilibrium value of Beff = −Ωez + Bopt, but Bopt (cid:28) Ω) resulting in an effective switching of the magnetization. The other possibility is to fall into a limit cycle attractor, where the solution is a periodic motion of the spin on the Bloch sphere. Which attractor is selected can be determined by analyzing instead the stability near CHAPTER 3. EQUATIONS OF MOTION the south pole, Sz = −S. In this case S0 x = SBopt/Ω and (cid:18) δSx − Ω Ω − S ∂Bopt ∂Sx (cid:19) δSx (cid:104) 27 (3.3.13) (cid:105)2 δSx . − 2GSκΩBopt (∆ − GSBopt/Ω) (κ/Ω)2 + (∆ − GSBopt/Ω)2 Therefore for ∆ > GSBopt/Ω, the effective ηopt in this case is positive and the solution is a stable fixed point resulting in magnetization switching, which can be seen as a population inversion driven by a blue detuned laser. In the opposite case (∆ < GSBopt/Ω), ηopt < 0 and there are runaway solutions. These two instabilities at the north and south pole are indicative of a limit cycle. In general, the limit cycles behavior is due to the change of sign of the dissipation function on the Bloch sphere (note that the condition ∆ > GSx can be fulfilled ∀Sx or for just a region on the Bloch sphere, depending on the magnitude of the detuning), analogous to the Van der Pol oscillator dynamics. These self-sustained oscillations can be the working principle for magnon lasing [27]. Beyond the fast cavity limit, Eqs. (3.3.1) and (3.3.2) can be solved numeri- cally. Besides magnetization switching and limit cycles, the full dynamics of the system can be driven into a chaotic regime, reached by period doubling as the power of the laser is increased. The chaotic regime requires sideband resolution (Ω > κ), and strong coupling G or equivalently a high density of circulating photons in the cavity, determined by the laser power. The estimated minimum values for attaining chaos seem to be outside of the current capabilities with YIG, see Ref. [27] for details. Whereas magnetization switching and self os- cillations can be attained at more moderate values of the parameters, other dissipation processes not considered here (such as three and four-magnon scat- tering processes) could hinder their realization in solid state systems [70]. These regimes can however be attained in cavity cold atom systems [71]. In the limit of small oscillations of the macrospin, we can fix Sz = S ( Sz = 0) and the remaining dynamical variables of the spin Sx and Sy behave as the conjugate coordinates of a harmonic oscillator. This is the classical limit of the Holstein-Primakoff approximation and it is valid as long as Sx, Sy (cid:28) S. In this limit (neglecting the Gilbert damping) we obtain and hence Sx = ΩSy Sy = −GSc2 − ΩSx Sx = −GSΩc2 − Ω2Sx (3.3.14) (3.3.15) with c(t) given by Eq. (3.3.1). The nonlinear dynamics in this case reduces to the one known for optomechanical systems in the classical limit [72]. Fig. (3.3.1) presents a qualitative, schematic phase diagram for the optomagnonic CHAPTER 3. EQUATIONS OF MOTION 28 Figure 3.3.1: Qualitative phase diagram for the classical nonlinear behavior of a blue detuned optomagnonic system as a function of the laser driving strength and inverse optomagnonic coupling. The axis have been rescaled into dimen- sionless quantities. The Bloch sphere in the switching region shows an example of the trajectory of the macrospin (in green) to the fixed point near the south pole, together with the sign of the optically induced dissipation coefficient ηopt. The white boxes in the limit cycle and chaos region show the trajectory of the macrospin (in red) once the corresponding attractor has been reached, show- ing a single and double-period limit cycle and an example of chaotic dynamics. Adapted from Ref. [27]. system in the blue detuned regime as a function of laser power and coupling strength, highlighting the different possible nonlinear regimes. 112xy-plane limit cycles"optomechanics"chaosoptomagnonic limit cyclesswitchingchaos yz plane limit cycles⌦GSxy-plane limit cyclesB↵max⌦chaosswitchingInverse coupling strengthLaser driving amplitudeG↵2max⌦<latexit sha1_base64="b6raiFTifBzvkXiwHefkkcfyH1s=">AAACDHicbVDLSsNAFJ34rPVVdelmsAiuSlIFXRZd6M4K9gFNLDfTSTt0JgkzE7GEfIAbf8WNC0Xc+gHu/BunbRbaemDgcM653LnHjzlT2ra/rYXFpeWV1cJacX1jc2u7tLPbVFEiCW2QiEey7YOinIW0oZnmtB1LCsLntOUPL8Z+655KxaLwVo9i6gnohyxgBLSRuqWyG0gg6aULPB5AN3WlSAU8ZNldNUvda0H7kJmUXbEnwPPEyUkZ5ah3S19uLyKJoKEmHJTqOHasvRSkZoTTrOgmisZAhtCnHUNDEFR56eSYDB8apYeDSJoXajxRf0+kIJQaCd8kBeiBmvXG4n9eJ9HBmZeyME40Dcl0UZBwrCM8bgb3mKRE85EhQCQzf8VkAKYdbformhKc2ZPnSbNacY4r1ZuTcu08r6OA9tEBOkIOOkU1dIXqqIEIekTP6BW9WU/Wi/VufUyjC1Y+s4f+wPr8ARGInEM=</latexit>Optomagnonic phase diagram for blue detuning⌘opt>0<latexit sha1_base64="Jx3J5bW00AjCRXorigl3qj0F3sE=">AAAB+nicbVBNS8NAEN34WetXqkcvwSJ4KkkV9CRFLx4r2A9oQthst+3S3WzYnSgl5qd48aCIV3+JN/+N2zYHbX0w8Hhvhpl5UcKZBtf9tlZW19Y3Nktb5e2d3b19u3LQ1jJVhLaI5FJ1I6wpZzFtAQNOu4miWEScdqLxzdTvPFClmYzvYZLQQOBhzAaMYDBSaFd8CjjMfCUymUCeX7mhXXVr7gzOMvEKUkUFmqH95fclSQWNgXCsdc9zEwgyrIARTvOyn2qaYDLGQ9ozNMaC6iCbnZ47J0bpOwOpTMXgzNTfExkWWk9EZDoFhpFe9Kbif14vhcFlkLE4SYHGZL5okHIHpDPNwekzRQnwiSGYKGZudcgIK0zApFU2IXiLLy+Tdr3mndXqd+fVxnURRwkdoWN0ijx0gRroFjVRCxH0iJ7RK3qznqwX6936mLeuWMXMIfoD6/MHnuaUOg==</latexit>⌘opt<0<latexit sha1_base64="xeolw0D7Ksb+nIHtsUgSfi6GRj8=">AAAB+nicbVBNS8NAEN34WetXqkcvwSJ4KkkV9OCh6MVjBfsBTQib7bZdupsNuxOlxPwULx4U8eov8ea/cdvmoK0PBh7vzTAzL0o40+C639bK6tr6xmZpq7y9s7u3b1cO2lqmitAWkVyqboQ15SymLWDAaTdRFIuI0040vpn6nQeqNJPxPUwSGgg8jNmAEQxGCu2KTwGHma9EJhPI8ys3tKtuzZ3BWSZeQaqoQDO0v/y+JKmgMRCOte55bgJBhhUwwmle9lNNE0zGeEh7hsZYUB1ks9Nz58QofWcglakYnJn6eyLDQuuJiEynwDDSi95U/M/rpTC4DDIWJynQmMwXDVLugHSmOTh9pigBPjEEE8XMrQ4ZYYUJmLTKJgRv8eVl0q7XvLNa/e682rgu4iihI3SMTpGHLlAD3aImaiGCHtEzekVv1pP1Yr1bH/PWFauYOUR/YH3+AJvclDg=</latexit>Chapter 4 Optomagnonics with a magnetic vortex In this section we come back to the issue of calculating the optomagnonic cou- pling in the presence of a smooth magnetic textures and structured optical modes, as given by Eq. (2.2.25). By smooth magnetic textures we refer to a magnetization profile that can be represented by a continuous vector field with constant length, given in our case by M(r) = Msm(r). As an example, we will consider a cavity system with cylindrical geometry. The optical cavity is a dielectric microdisk which can also host magnetic modes. Due to the size and geometry, the magnetic ground state is a vortex. We will calculate explicitly the optomagnonic coupling of whispering gallery modes in the disk, to a magnetic excitation localized at the vortex core. In this section we follow Ref. [44]. 4.0.1 Magnetic vortex A paradigmatic example of a magnetic texture is a magnetic vortex, which is the stable ground state configuration in magnetic disks with radial dimensions in the µm range. The vortex forms due to a competition between demagnetizing fields, which tend to align with the surfaces of the disk to avoid the formation of sources of stray fields, and the exchange interaction, which tends to align the spins among themselves. The demagnetizing fields are determined by the mag- netostatic Poisson equation and have their origin in the dipolar interactions, and are therefore weak but long ranged. The exchange interaction, whose physical origin is the Coulomb interaction together with the Pauli principle of exclusion, is instead strong but short ranged. The effects of these fields become compa- rable at the microscale, leading to ordered flux closure configurations that, in the case of a cylindrical geometry, take the form of a vortex. In a thin disk (for heights comparable to the exchange length lex) the spins are mostly in-plane and curl around the center of the disk. The core of this vortex is situated at the center of the disk, and it consists of spins pointing out of the plane [73]. Besides 29 CHAPTER 4. OPTOMAGNONICS WITH A MAGNETIC VORTEX 30 Figure 4.0.1: Magnetic vortex in a microdisk. The spins curl in the plane of the disk and at the core of the vortex point out of the plane. Adapted from Ref. [44]. being ubiquitous, vortices are interesting since they are topological objects with two independent degrees of freedom, the chirality C = ±1 (the spins can curl clockwise or anti-clockwise) and the parity P = ±1 , indicating if the spins at the vortex core point up or down, see Fig. (4.0.1). Manipulating these degrees of freedom can give rise to new forms of storing and processing information with magnetic systems [74]. The vortex in the thin disk can be parametrized as where b is an effective core radius (of the order of a few lex), we have assumed ,C = P = 1 and used cylindrical coordinates with origin at the center of the vortex core, eϕ = (cos ϕ, sin ϕ, 0). The lowest energy magnon mode in this system is a translational mode of the vortex core, which can be shown to be a circular motion. This is due to an effective gyrotropic force proportional to the topological charge of the vortex, which effectively acts on the vortex core as a magnetic field acts on a charged particle [75]. This mode is denominated gyrotropic and is usually in the range of hundreds of MHz. It gives rise to a time-dependent magnetization that can be approximated as mex(r, t) = m(r − rc(t)) ≈ m(r) − (rc(t) · ∇) m(r) , (4.0.3) m((cid:126)ρ) = eϕ = 1 ρ2 + b2  −2by (cid:0)b2 − ρ2(cid:1) 2bx  for ρ ≥ b for ρ ≤ b (4.0.1) (4.0.2) Vortex coreP=+1C=+1CHAPTER 4. OPTOMAGNONICS WITH A MAGNETIC VORTEX 31 where rc(t) = rc [cos(ωgt)ex + sin(ωgt)ey] parametrizes the time-dependent po- sition of the vortex core measured from its equilibrium position (we have ignored the damping of the mode). Therefore we can write (4.0.4) From Eqs. (4.0.1) and (4.0.4) one can obtain analytical expressions for the gyrotropic mode profile δm(r). Together with the normalization prescription Eq. (2.2.21) one obtains the profile of the mode normalized to one magnon [44]. δm(r, t) = − (rc(t) · ∇) m(r) . 4.0.2 Optomagnonic coupling for the gyrotropic mode We proceed now to calculate an analytical expression for the coupling of the gyrotropic mode to an optical WGM in the 2D limit. In order to obtain a finite coupling, the gyrotropic mode needs to have overlap with the WGM, which lives near the rim of the disk, see Fig. (4.0.2) (a). The core of the vortex can be displaced from the center of the magnetic disk by applying an external in-plane magnetic field, as the spins will try to align with the field. To a first approximation, the position of the disk s varies linearly with magnetic field, although this breaks down as the vortex approaches the rim. We will moreover use the "rigid vortex" approximation, which implies that the vortex moves but does not deform, so that Eqs. (4.0.1) are always valid as a parametrization of the vortex. It is known that this approximation also fails close to the rim of the disk [73]. Our model therefore is valid for thin disks and its accuracy will diminish as the vortex approaches the rim of the disk. The WGMs in cylindrical geometry, considering as an approximation an infinite cylinder along z, so that the results will be independent of z due to translation invariance and effectively two-dimensional, are given by the solution to the Helmholtz equation ∇2 + n2k2 (4.0.5) where ψ = Ez or ψ = Bz respectively for the TM (B ⊥ ez) and the TE (E ⊥ ez) mode, and n is the index of refraction of the confining dielectric (e.g. YIG). In cylindrical coordinates (r, θ) with origin at the center of the disk, the solutions for r < R (R the radius of the disk) are of the form (k = nk0) 0 (4.0.6) with Jm the Bessel function of the first kind together with the boundary condi- tion ψ(r, θ) = AmJm(kr)ei(±mθ) , K∂rJ m(nkR)/Jm(nkR) = ∂rH (1) (4.0.7) m the Hankel function of the first kind and K = n for a TM and K = 1/n m(kR)/H (1) m (kR) with H (1) for a TE mode. One obtains (cid:0) (cid:1) ψ = 0, ETM mp = ψmp(r, θ)ez , (cid:18) 1 ETE mp = i εωmp r ∂θψmper − ∂rψmpeθ , (cid:19) (4.0.8) (4.0.9) CHAPTER 4. OPTOMAGNONICS WITH A MAGNETIC VORTEX 32 Figure 4.0.2: Gyrotropic magnon mode and WGM mode and their respec- tive coupling. (a) WGM of a micromagnetic disk together with the gyrotropic magnon mode of a displaced magnetic vortex core. The vortex core has been statically displaced by an applied magnetic field Hx along the x direction. (b) Spatial dependence of the optomagnonic coupling, G(r, Hx) before integrating over the volume to obtain the to total coupling, G = V d3rG(r, Hx). The re- sults were obtained with finite element simulations for the optics and MuMax3 [76] micromagnetic simulations for the magnetics, for a thin YIG disk. In order to confine the optical WGM, a cylindrical heterostructure consisting of a thin YIG layer between SiN layers was considered, see Ref. [44] for details. ´ k = ω + iκ 2 , and where the tilde indicates that the solutions are complex: ω = c n the subscripts mp indicates that the expressions are evaluated for a particular solution k = kmp = kmp − ik(cid:48)(cid:48)mp of Eq. (4.0.7). In the following we consider only WGMs solutions, which correspond to p = 1 (one node in the radial direction), hence we will omit this index. Well defined WGMs are solutions with small imaginary part k(cid:48)(cid:48)m1, since this is related to the leaking of the optical mode out of the cavity. The imaginary part gives the decay rate of the mode due to coupling to external unbounded optical modes, and will enter in the total decay rate of the mode. The normalization of the WGM can found by imposing Eq. (2.2.9). We consider for simplicity the optomagnonic coupling to only one WGM. One can easily show that the only possibility for finite coupling in the 2D limit is to couple to the TE mode [44]. It is straightforward to obtain 2π dϕmz(ρ, ϕ)(cid:0)ETE∗m × ETE m (cid:1) Gm =−i θFλn 4π ε0ε 2 h b ρdρ 0 0 · ez (4.0.10) where (ρ, ϕ) are polar coordinates in the system with origin at the center of the vortex (note that the center of the vortex can be displaced by an external WGMmz(~r)[A/m]0[Hz]0abG(r,Hx)CHAPTER 4. OPTOMAGNONICS WITH A MAGNETIC VORTEX 33 Figure 4.0.3: Optomagnonic coupling (red) as a function of the vortex position according to Eq. (4.0.11). Also shown is the (normalized) magnitude of the optical spin density (see Eq. (2.2.3)) at the position of the vortex. From the plot it is evident that the coupling is proportional to the gradient of the optical spin density. The inset shows the magnetic vortex at zero field and at an arbitrary finite magnetic field along x. Radius of the YIG thin disk R = 1µm), WGM with m = 6, ωopt/2π ≈ 200THz. Adapted from Ref. [44]. G±m =± θFλn 2π ωm 2πNJ 0 0 (ρ2 + 1)2 ∂rJm(r)2 (s/b) ey + ρeρ , magnetic field, see Fig. (4.0.2)). From Eq.(4.0.9) we obtain rckmm 1 dρ 2π dϕeiϕ ρ2 (4.0.11) with r = kmr, r = sey + ρeρ, and ωm = c n km. rc and NJ are the corresponding normalization factors for the magnon and optical modes and depend on the geometrical parameters of the system. An example of the spatial structure of the coupling (4.0.11) is shown in Fig. (4.0.2) (b), where the results were obtained via micromagnetics [76] and finite element simulations for the optics. One can show that the first order contribution to Eq. (4.0.11) is proportional to the gradient of the optical spin density with respect to the vortex position. This can be also seen in the results of Fig. (4.0.3). The 2D approximation works quite well for a thin YIG disk, as we have shown in Ref. [44], where we compared the analytical results with results obtained by combining micromagnetic and finite element simulations. Whereas the thin disk is perfectly fine to host the magnon modes, it is a bad optical cavity since it is not able to confine the light. As a rule of thumb, the thickness of the dielectric has to be of the order of at least half of the wavelength of the light in the material 204060801001020304050Optomagnonic Coupling - Analytical Result Optomagnonic Coupling G [Hz]Vortex Position s [nm]20040060080010005040302010CouplingOptical Spin Density (a. u.)Hx=0mTHx=0.85mTHx=4.3mTVortexShifted vortexC-vortexsCHAPTER 4. OPTOMAGNONICS WITH A MAGNETIC VORTEX 34 in order to confine it effectively. For this reason, we proposed a "sandwich" heterostructure, where the YIG thin disk is sandwiched between thicker layers of SiN which serve to confine the light. The index of refraction of the transparent dielectric SiN is similar to that of YIG in the optical range, so it is a good choice of material. This however has the effect of decreasing the magnon-to- optical volume ratio, which is detrimental for the optomagnonic coupling as discussed previously. A solution to enhance the coupling was proposed in Ref. [44], where instead of a thin disk a thicker YIG structure was considered. In this case however, the 2D approximation breaks down and the results for the coupling necessarily must be obtained numerically. Promising high values for the coupling and cooperativity were obtained, indicating the value of studying and designing optomagnonic systems beyond the homogeneous, Kittel mode case. Chapter 5 A quantum protocol: all-optical magnon heralding To finish this chapter, we study a possible quantum protocol in a cavity opto- magnonic system, as proposed in Ref. [77]. In the protocol, a one-magnon Fock state is created in the magnetic material by the detection of a photon. This is referred to as heralding, since the detected photon announces the creation of the desired state. A Fock state, also called a number state since it has a well-defined occupation number, is a purely nonclassical state (as compared for example with a coherent state) characterized by a negative Wigner function. A magnon Fock state is, therefore, a macroscopic collective (involving millions of spins) nonclassical state of the magnetic system, and its realization can be the first step towards the generation, manipulation, and transfer of quantum states in optomagnonic systems [35, 39]. Our protocol proposal includes generating the magnon Fock state optically and reading the state, also optically, at some time later. Due to the optomagnonic interaction, photons and magnons are entangled during the evolution, and detecting a photon projects the magnon sate. The successful generation of the nonclassical state is tested by measuring the two-photon correlations of a "reading" laser. We consider a system with two optical modes and one relevant magnon mode, in line with current experimental setups involving optical whispering gallery modes in YIG spheres [24, 25, 26]. 5.1 Hamiltonian and Langevin Equations of Mo- tion In this section, we analyze the quantum Langevin equations of motion of the cavity optomagnonic system in the spin-wave regime for a system with two non-degenerate optical modes a1 and a2 interacting with one magnon mode m. We will find the analytical solutions for the evolution of the quantum fields by linearizing also in the optical fields, as in Eq. (2.2.19). Our analysis is valid 35 CHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING36 for any magnetization texture and magnon mode (including the homogeneous case), but it is restricted to small oscillations of the spins. In this case the Hamiltonian in Eq. (2.3.9) reduces to H = −∆1a†1a1 − ∆2a†2a2 + Ω m† m +  G12a†1a2 + G21a†2a1 (cid:17) m† + h.c. (5.1.1) (5.1.2) (cid:16)
j(a†j − aj) , (cid:88) j + i where we consider that modes j = 1, 2 are driven at frequency ωL and ampli- tude
j given by Eq. (2.3.6), and ∆j = ωL − ωj is the respective detuning. The particularity of the cavity optomagnonic system involving spherical WGMs is the asymmetry between the scattering rates G12 and G21, due to energy and an- gular momentum conservation rules [26]. In particular, one of the two processes is highly suppressed, which we reflect by setting G21 = 0 = G∗21 (5.1.3) in Eq. (5.1.1). Accordingly, we consider two optical modes satisfying approxi- mately the resonance condition ω2 − ω1 ≈ Ω. Applying Eq. (3.1.1) to each of the field operators, and including dissipative and noise terms both for photons and magnons following Eq. (3.2.12), we obtain da1 dt da2 dt d m dt a1 + √κ a1,in(t) +
1 , κ = i∆1a1 − iG12a2 m† − 2 a2 + √κ a2,in(t) +
2 , κ = i∆2a2 − iG∗12a1 m − 2 m + +√Γ min(t) , Γ = −iΩ m − iG12a†1a2 − 2 (5.1.4) where we have assumed for simplicity that both photon modes are subject to the same decay rate κ. Whereas the photon bath can be considered to be at zero temperature due to the high frequency of the optical photons, (cid:104)ai,in(t)a†j,in(t(cid:48))(cid:105) = δijδ(t − t(cid:48)) , (cid:104)a†i,in(t)aj,in(t)(cid:105) = 0 , (5.1.5) the magnons have usually GHz frequencies and the temperature of the thermal bath cannot be ignored, unless the system is cooled to mK temperatures. The general expressions for the magnon correlators are (cid:104) min(t) m†in(t(cid:48))(cid:105) = (ntm + 1)δ(t − t(cid:48)), (cid:104) m†in(t) min(t(cid:48))(cid:105) = ntmδ(t − t(cid:48)) , (5.1.6) (5.1.7) CHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING37 where the subscript tm indicates that ntm is the mean number of thermal magnons given by the Bose-Einstein distribution ntm(T ) = 1 exp ( Ω/kBT ) − 1 , (5.1.8) where T is the temperature of the magnon bath and kB the Boltzmann constant, and Ω is the magnon mode frequency. The steady state values (cid:104)ai(cid:105) = αi and (cid:104) m(cid:105) = β are found by setting Eqs. (5.1.4) to zero and ignoring the noise fluctuations. The expectation values of the operators are understood to be taken with respect to a coherent state. In the side-band resolved limit ( Ω (cid:29) κ, γ ) it is straightforward to see that if only one mode j (=1 or 2) is driven, β = 0 and the steady state circulating number of photons in the cavity is given by αj2 with αj = −
j i∆j − κj/2 . (5.1.9) By considering the fluctuations around the steady state ai → αi +ai, m → m (where for simplicity of notation we denote now the fluctuations by ai and m) one obtains the Hamiltonian valid in the linear regime, as in Eq. (2.3.11). In the interaction picture the resulting Hamiltonian reads HIP ≈ α∗1G12a2 m†ei(∆2+Ω)t + α2G12a†1 m†e−i(∆1−Ω)t + h.c. . (5.1.10) (cid:104) (cid:105) 5.2 Write and read protocol From Eq. (5.1.10) one can immediately observe that by pumping the optical mode 2 at resonance (ωL = ω2 ≈ ω1 + Ω) while mode 1 is not driven (α1 = 0), the condition ∆1 = Ω is satisfied and one obtains HW ≈  α∗2G∗12a1 m + α2G12a†1 m† , (5.2.1) in which a magnon and a photon in mode 1 are either created or annihilated in pairs. We denote this effective Hamiltonian as HW since it corresponds to the writing Hamiltonian is our protocol. Starting from the vacuum of magnons and photons, the evolution under this Hamiltonian creates entangled pairs of magnons and mode-1 photons. Detecting a photon in mode 1 collapses the entangled state and determines the magnon state, with a certain probability of collapsing into a one-magnon Fock state [77]. The Fock state created in this form is denominated heralded. The successful heralding of a magnon Fock state can be corroborated by a reading protocol, as long this is done within the lifetime of the state, that is , the lifetime of the magnon. The reading Hamiltonian is obtained from Eq. (5.1.10) CHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING38 Figure 5.2.1: Write and read protocol for heralding a magnon Fock state. The detection of a photon at time tm is a probabilistic process. Figure taken from Ref. [77]. by this time driving the optical mode 1 at resonance (ωL = ω1 ≈ ω2 − Ω) and not driving mode 2 (α2 = 0). One obtains (cid:105) (cid:104) HR ≈  α∗1G12a2 m† + α1G∗12a†2 m . (5.2.2) In the strong coupling regime, such that the effective coupling strength Geff 1,12 = α1G12 is larger than the decay channels α1G12 > κ, Γ, the magnon mode and the photon mode hybridize, giving rise to eigenmodes which are part magnon, part photon, in analogy with the cavity magnon polariton discussed previously in the MW regime. Moreover, if α1G12 > κ, ntmΓ (denominated the coher- ent coupling regime) the interaction is quantum-coherent, allowing to transfer the magnon state coherently to the photons. Therefore measuring the mode-2 photons give information on the magnon state, that is, we can "read" the state. The write-and-read protocol is described schematically in Fig. (5.2.1). As we pointed out in the introduction, the strong coupling regime in opto- magnonics is challenging to attain and experiments have not yet reached this point. Nevertheless, photons in mode 2 can be used to probe the heralded state even in the weak coupling regime. This can be done by measuring the two-photon correlation function [68] Read(t, t + τ ) = (cid:104)a†2(t)a†2(t + τ )a2(t + τ )a2(t)(cid:105) g(2) (cid:104)a†2(t)a2(t)(cid:105)(cid:104)a†2(t + τ )a2(t + τ )(cid:105) , (5.2.3) !1!2//⌦!L⇠!2=!1+⌦Driven resonance:⌦Write ModeRead Mode//!L⇠!1=!2⌦!2!1⌦⌦Time (arbitrary units)Write PhaseClick at = 
magnon Fock stateRead PhaseTo↵tmHW=~↵2G12a†1m†+h.c.Driven resonance:HR=~↵⇤1G12a2m†+h.c.CHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING39 which can be done interferometrically. In order to use this quantity as a herald- ing witness, the expectation values have to be taken after the measurement of a "write" photon (a mode-1 photon). Given the form of the read Hamiltonian from Eq. (5.2.2), which "swaps" magnons with photons, measuring g(2) Read for the photons is equivalent to measure the corresponding correlation function for the magnons. If the state is non-classical, g(2) Read(t, t) < 1, given an indication that the heralded magnon state is a Fock state. The condition g(2) Read(t, t) < 1 is denominated antibunching, since there is a reduced probability of two photons being detected simultaneously, and it is an example of a quantum state violat- ing a classical inequality (g(2) Read(t, t) > 1 necessarily for classical states). The reliability of g(2) Read as a true heralding witness depends however on the temper- ature of the magnon bath, worsening as the temperature and consequently the number of thermal magnons increases [77]. 5.3 Solution of the linear quantum Langevin equa- tions The probability of heralding a magnon, as well as the correlation function g(2) Read, can be calculated by the linear quantum Langevin equations dictated by the Hamiltonians of Eq. (5.2.1) and (5.2.2) with the steady state given by Eqs. (5.1.4) and using the noise correlators of Eqs. (5.1.5) and (5.1.6). The linear Langevin equations can be written in compact form as where A = d A dt  a†1 m a2  , = M X · A(t) + N (t),  , (5.3.1) (5.3.2)  √κ(ain 1 )† √γ min √κain 2 N = X = W, R indicates the write or read phases of the protocol, and M X is a matrix given by the Heisenberg equation of motion deriving from Hamiltonians (5.2.1) or (5.2.2) . The solution can be written formally as A(t) = U X(t) · A(0) + t 0 dτ U X(t − τ ) · N (τ ), (5.3.3) where U W(t) and U R(t) are the respective evolution matrices. These can be found analytically by going to a diagonal basis such that d A(cid:48)i dt = λP i A(cid:48)i(t) + N(cid:48)i (t), (5.3.4) which can be easily integrated and transformed back to find U W(t) and U R(t). The expressions for U W(t) and U R(t) are lengthy and can be found in Ref. [77]. CHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING40 5.4 Probability of heralding a magnon The probability of heralding a magnon is given by the probability of detecting a photon in mode 1 during the write phase. This is given by (see e.g. Ref. [68]) P1,W(t) = (cid:104): a†1a1 exp(−a†1a1) :(cid:105) ∼ (cid:104)a†1a1(cid:105) − (cid:104)a†1a†1a1a1(cid:105) , where the approximation is valid as long as (cid:104)a†1a†1a†1a1a1a1(cid:105) (cid:28) (cid:104)a†1a†1a1a1(cid:105). This probability can be computed by the method outlined previously, on the basis of the solution for the equations of motion (5.3.1) for the write-Hamiltonian. One finds that the heralding probability grows linearly with the square of the effective optomagnonic coupling during the write phase. A very large coupling is however detrimental, since it will tend to generate a larger number of magnons within the same period, compromising the Fock state. Therefore an optimal effective coupling strength Geff 1,12 must be found, see Fig. (5.4.1). The mean number of magnons nhm = (cid:104) m† m(cid:105) after the heralding event depends crucially on the temperature of the magnon bath, as shown in Fig. (5.4.1). 5.5 Magnon cooling As we have seen in the heralding example, the presence of thermal magnons is highly detrimental for quantum protocols. Due to their larger frequencies (usually in the GHz range in optomagnonic setups), magnons are generally more amenable to direct cooling than, for example, phonons in optomechanical systems. However light can be used for active cooling in cases in which direct cooling is not sufficient or inconvenient from and experimental point of view. Magnon cooling in optomagnonic systems has been studied in detail in Ref. [78]. The formalism described to solve the quantum Langevin equations of motion can be used to study the cooling protocol exactly, including all decay channels and quantum fluctuations. In particular, our "read" Hamiltonian can be used for magnon cooling, since it effectively annihilates magnons when mode 1 is driven. If we consider an initial thermal state state with a mean number of magnons nth, the initial density matrix of the optomagnonic system is given by where ρth, m = 1 1 + nth ρ(0) = 0(cid:105)(cid:104)01 ⊗ 0(cid:105)(cid:104)02 ⊗ ρth,m, (cid:88) (cid:20) nth (cid:21)n 1 + nth n(cid:105)(cid:104)n, n≥0 indicates that the initial population of magnons os a thermal state. The tem- poral evolution of the mean number of magnons is given by (5.5.1) (5.5.2) (5.5.3) (cid:104) m† m(cid:105)(t) =(cid:80) i,j + i2(t)(cid:1)∗ U R (cid:0)U R (cid:88) t dτ1dτ2 i,j 0 2j(t)(cid:104) A†i (0) Aj(0)(cid:105) (cid:0)U R i2(t − τ1)(cid:1)∗ U R 2j(t − τ2)(cid:104) N†i (τ1) Nj(τ2)(cid:105) , CHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING41 Figure 5.4.1: Probability of heralding a magnon and mean number of heralded 1,12/κΓ. The shaded magnons as a function of the write cooperativity, C = 4Geff area indicates values of the cooperativity which give an appreciable heralding probability while keeping the number of heralded magnons near one. The num- ber of thermal magnons ntm is dictated by the temperature of the magnon bath, larger temperatures are, as expected, detrimental for the heralding protocol. Figure taken from Ref. [77]. Cooperativity 10710610510410310210111.11.21.31.41.5106105104103102101nTh=0.01nTh=0.5nTh=1.5nTh=2.5Write mode: magnon heraldingMagnon mean number nhmHeralding probabilityP1,WCHAPTER 5. A QUANTUM PROTOCOL: ALL-OPTICAL MAGNON HERALDING42 where i, j indicate the components of A, N, and U R as defined in Eqs. (5.3.2) and (5.3.3), and the expectation values are taken over the initial state deter- mined by ρ(0). Imposing the noise correlators of Eqs. (5.1.5) and (5.1.6) one obtains (cid:104) m† m(cid:105)(t) = U R 22(t)2nth + U R 12(t)2 dτ(cid:2) t 0 + 12(t − τ )2 + U R U R 22(t − τ )2nth (cid:19) (cid:18) The steady state value n0 = (cid:104) m† m(cid:105)(t → ∞) can be shown to be n0 = Γnth (κ + Γ) 1 + κ Γ(1 + CR) , (5.5.4) (cid:3) . where CR = 4α∗1G122/κΓ is the read-phase cooperativity. This is a thermal state with n0 < nth, and therefore it is cooled. Whereas active cooling is nec- essary for most implementations of optomechanical systems, given that phonon frequencies are usually low, for optomagnonic systems this can be circumvented by cooling the system by usual dilution fridge refrigeration techniques. Chapter 6 Outlook Cavity optomagnonic systems are at the interface between condensed matter and quantum optics, and present new opportunities to study and control the in- teraction between light and magnetic systems, in particular at the single quanta level. Magnons are robust elementary excitations, highly tunable and can couple well to several other degrees of freedom beyond photons, such as phonons and electrons. The incorporation of magnetically ordered systems into hybrid plat- forms for quantum information is therefore very promising. From a fundamental point of view, cavity optomagnonic systems are very rich. Topics such as the interaction of structured light with magnetic textures and topological defects, the nonlinear dynamics of the coupled system, or collective quantum effects in cavity optomagnonics taking into account the strongly correlated nature of the magnetically ordered systems, are still largely unexplored. There are exciting times ahead for theorists and experimentalists alike. 43 Bibliography [1] A. G. J. MacFarlane, Jonathan P. Dowling, and Gerard J. Milburn. Quan- tum technology: The second quantum revolution. Philosophical Transac- tions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences, 361(1809):1655 -- 1674, August 2003. [2] Frank Arute, Kunal Arya, Ryan Babbush, Dave Bacon, Joseph C. Bardin, Rami Barends, Rupak Biswas, Sergio Boixo, Fernando G. S. L. Bran- dao, David A. Buell, Brian Burkett, Yu Chen, Zijun Chen, Ben Chiaro, Roberto Collins, William Courtney, Andrew Dunsworth, Edward Farhi, Brooks Foxen, Austin Fowler, Craig Gidney, Marissa Giustina, Rob Graff, Keith Guerin, Steve Habegger, Matthew P. Harrigan, Michael J. Hartmann, Alan Ho, Markus Hoffmann, Trent Huang, Travis S. Humble, Sergei V. Isakov, Evan Jeffrey, Zhang Jiang, Dvir Kafri, Kostyantyn Kechedzhi, Julian Kelly, Paul V. Klimov, Sergey Knysh, Alexander Korotkov, Fe- dor Kostritsa, David Landhuis, Mike Lindmark, Erik Lucero, Dmitry Lyakh, Salvatore Mandrà, Jarrod R. McClean, Matthew McEwen, Anthony Megrant, Xiao Mi, Kristel Michielsen, Masoud Mohseni, Josh Mutus, Ofer Naaman, Matthew Neeley, Charles Neill, Murphy Yuezhen Niu, Eric Ostby, Andre Petukhov, John C. Platt, Chris Quintana, Eleanor G. Rieffel, Pe- dram Roushan, Nicholas C. Rubin, Daniel Sank, Kevin J. Satzinger, Vadim Smelyanskiy, Kevin J. Sung, Matthew D. Trevithick, Amit Vainsencher, Benjamin Villalonga, Theodore White, Z. Jamie Yao, Ping Yeh, Adam Zal- cman, Hartmut Neven, and John M. Martinis. Quantum supremacy using a programmable superconducting processor. Nature, 574(7779):505 -- 510, October 2019. [3] H. J. Kimble. The quantum internet. Nature, 453:1023 -- 1030, June 2008. [4] A. D. O'Connell, M. Hofheinz, M. Ansmann, Radoslaw C. Bialczak, M. Lenander, Erik Lucero, M. Neeley, D. Sank, H. Wang, M. Weides, J. Wenner, John M. Martinis, and A. N. Cleland. Quantum ground state and single-phonon control of a mechanical resonator. Nature, 464(7289):697 -- 703, April 2010. [5] Jasper Chan, T. P. Mayer Alegre, Amir H. Safavi-Naeini, Jeff T. Hill, Alex Krause, Simon Gröblacher, Markus Aspelmeyer, and Oskar Painter. 44 BIBLIOGRAPHY 45 Laser cooling of a nanomechanical oscillator into its quantum ground state. Nature, 478(7367):89 -- 92, October 2011. [6] J. D. Teufel, T. Donner, Dale Li, J. W. Harlow, M. S. Allman, K. Ci- cak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and R. W. Simmonds. Sideband cooling of micromechanical motion to the quantum ground state. Nature, 475(7356):359 -- 363, July 2011. [7] Markus Aspelmeyer, Tobias J. Kippenberg, and Florian Marquardt. Cavity optomechanics. Rev. Mod. Phys., 86(4):1391 -- 1452, December 2014. [8] Ralf Riedinger, Andreas Wallucks, Igor Marinković, Clemens Löschnauer, Markus Aspelmeyer, Sungkun Hong, and Simon Gröblacher. Remote quantum entanglement between two micromechanical oscillators. Nature, 556(7702):473 -- 477, April 2018. [9] Gershon Kurizki, Patrice Bertet, Yuimaru Kubo, Klaus Mølmer, David Petrosyan, Peter Rabl, and Jörg Schmiedmayer. Quantum technologies with hybrid systems. PNAS, 112(13):3866 -- 3873, March 2015. [10] M. H. Devoret and R. J. Schoelkopf. Superconducting Circuits for Quantum Information: An Outlook. Science, 339(6124):1169 -- 1174, March 2013. [11] Mikael Afzelius, Nicolas Gisin, and Hugues de Riedmatten. Quantum mem- ory for photons. Physics Today, 68(12):42 -- 47, November 2015. [12] Fabio Pulizzi. Spintronics. Nature Mater, 11(5):367 -- 367, May 2012. [13] J. C. Slonczewski. Current-driven excitation of magnetic multilayers. Jour- nal of Magnetism and Magnetic Materials, 159(1):L1 -- L7, June 1996. [14] Sabpreet Bhatti, Rachid Sbiaa, Atsufumi Hirohata, Hideo Ohno, Shun- suke Fukami, and S. N. Piramanayagam. Spintronics based random access memory: A review. Materials Today, 20(9):530 -- 548, November 2017. [15] Joseph E. Losby and Mark R. Freeman. Spin Mechanics. arXiv:1601.00674 [cond-mat], January 2016. [16] Marcelo Wu, Nathanael L.-Y. Wu, Tayyaba Firdous, Fatemeh Fani Sani, Joseph E. Losby, Mark R. Freeman, and Paul E. Barclay. Nanocavity optomechanical torque magnetometry and radiofrequency susceptometry. Nature Nanotechnology, 12(2):127 -- 131, February 2017. [17] Dany Lachance-Quirion, Yutaka Tabuchi, Arnaud Gloppe, Koji Usami, and Yasunobu Nakamura. Hybrid quantum systems based on magnonics. Appl. Phys. Express, 12(7):070101, June 2019. [18] Hans Huebl, Christoph W. Zollitsch, Johannes Lotze, Fredrik Hocke, Moritz Greifenstein, Achim Marx, Rudolf Gross, and Sebastian T. B. Goen- nenwein. High Cooperativity in Coupled Microwave Resonator Ferrimag- netic Insulator Hybrids. Phys. Rev. Lett., 111(12):127003, September 2013. BIBLIOGRAPHY 46 [19] Yutaka Tabuchi, Seiichiro Ishino, Toyofumi Ishikawa, Rekishu Yamazaki, Koji Usami, and Yasunobu Nakamura. Hybridizing Ferromagnetic Magnons and Microwave Photons in the Quantum Limit. Phys. Rev. Lett., 113(8):083603, August 2014. [20] Xufeng Zhang, Chang-Ling Zou, Liang Jiang, and Hong X. Tang. Strongly Coupled Magnons and Cavity Microwave Photons. Phys. Rev. Lett., 113(15):156401, October 2014. [21] J. A. Haigh, N. J. Lambert, A. C. Doherty, and A. J. Ferguson. Disper- sive readout of ferromagnetic resonance for strongly coupled magnons and microwave photons. Phys. Rev. B, 91(10):104410, March 2015. [22] Ö. O. Soykal and M. E. Flatté. Strong Field Interactions between a Nano- magnet and a Photonic Cavity. Phys. Rev. Lett., 104(7):077202, February 2010. [23] Yutaka Tabuchi, Seiichiro Ishino, Atsushi Noguchi, Toyofumi Ishikawa, Rekishu Yamazaki, Koji Usami, and Yasunobu Nakamura. Coherent cou- pling between a ferromagnetic magnon and a superconducting qubit. Sci- ence, 349(6246):405 -- 408, July 2015. [24] J. A. Haigh, A. Nunnenkamp, A. J. Ramsay, and A. J. Ferguson. Triple- Resonant Brillouin Light Scattering in Magneto-Optical Cavities. Physical Review Letters, 117(13), September 2016. [25] Xufeng Zhang, Na Zhu, Chang-Ling Zou, and Hong X. Tang. Opto- magnonic Whispering Gallery Microresonators. Physical Review Letters, 117(12), September 2016. [26] A. Osada, R. Hisatomi, A. Noguchi, Y. Tabuchi, R. Yamazaki, K. Us- ami, M. Sadgrove, R. Yalla, M. Nomura, and Y. Nakamura. Cav- ity Optomagnonics with Spin-Orbit Coupled Photons. Phys. Rev. Lett., 116(22):223601, June 2016. [27] Silvia Viola Kusminskiy, Hong X. Tang, and Florian Marquardt. Coupled spin-light dynamics in cavity optomagnonics. Phys. Rev. A, 94:033821, 2016. [28] Tianyu Liu, Xufeng Zhang, Hong X. Tang, and Michael E. Flatté. Opto- magnonics in magnetic solids. Phys. Rev. B, 94(6):060405, August 2016. [29] Yi-Pu Wang, Guo-Qiang Zhang, Dengke Zhang, Tie-Fu Li, C.-M. Hu, and J. Q. You. Bistability of Cavity Magnon Polaritons. Phys. Rev. Lett., 120(5):057202, January 2018. [30] H. Maier-Flaig, M. Harder, S. Klingler, Z. Qiu, E. Saitoh, M. Weiler, S. Geprägs, R. Gross, S. T. B. Goennenwein, and H. Huebl. Tunable magnon-photon coupling in a compensating ferrimagnet -- from weak to strong coupling. Appl. Phys. Lett., 110(13):132401, March 2017. BIBLIOGRAPHY 47 [31] R. G. E. Morris, A. F. van Loo, S. Kosen, and A. D. Karenowska. Strong coupling of magnons in a YIG sphere to photons in a planar superconduct- ing resonator in the quantum limit. Sci Rep, 7(1):1 -- 6, September 2017. [32] Isabella Boventer, Christine Dörflinger, Tim Wolz, Rair Macêdo, Ro- main Lebrun, Mathias Kläui, and Martin Weides. Control of the Coupling Strength and the Linewidth of a Cavity-Magnon Polariton. arXiv:1904.00393 [cond-mat], March 2019. [33] Yi-Pu Wang, Guo-Qiang Zhang, Da Xu, Tie-Fu Li, Shi-Yao Zhu, J. S. Tsai, and J. Q. You. Quantum Simulation of the Fermion-Boson Composite Quasi-Particles with a Driven Qubit-Magnon Hybrid Quantum System. arXiv:1903.12498 [cond-mat, physics:quant-ph], March 2019. [34] Graeme Flower, Maxim Goryachev, Jeremy Bourhill, and Michael E. To- bar. Experimental implementations of cavity-magnon systems: From ultra strong coupling to applications in precision measurement. New J. Phys., 21(9):095004, September 2019. [35] Dany Lachance-Quirion, Yutaka Tabuchi, Seiichiro Ishino, Atsushi Noguchi, Toyofumi Ishikawa, Rekishu Yamazaki, and Yasunobu Nakamura. Resolving quanta of collective spin excitations in a millimeter-sized ferro- magnet. Science Advances, 3(7):e1603150, July 2017. [36] J. W. Rao, C. H. Yu, Y. T. Zhao, Y. S. Gui, X. L. Fan, D. S. Xue, and C.-M. Hu. Level attraction and level repulsion of magnon coupled with a cavity anti-resonance. New J. Phys., 21(6):065001, June 2019. [37] Isabella Boventer, Mathias Kläui, Rair Macêdo, and Martin Weides. Steer- ing between Level Repulsion and Attraction: Beyond Single-Tone Driven Cavity Magnon-Polaritons. arXiv:1908.05439 [cond-mat, physics:quant- ph], August 2019. [38] Yi-Pu Wang, J. W. Rao, Y. Yang, Peng-Chao Xu, Y. S. Gui, B. M. Yao, J. Q. You, and C.-M. Hu. Nonreciprocity and Unidirectional Invisibility in Cavity Magnonics. Phys. Rev. Lett., 123(12):127202, September 2019. [39] Dany Lachance-Quirion, Samuel Piotr Wolski, Yutaka Tabuchi, Shingo Kono, Koji Usami, and Yasunobu Nakamura. Entanglement-based single-shot detection of a single magnon with a superconducting qubit. arXiv:1910.09096 [cond-mat, physics:quant-ph], October 2019. [40] Justin T. Hou and Luqiao Liu. Strong Coupling between Microwave Pho- tons and Nanomagnet Magnons. Phys. Rev. Lett., 123(10):107702, Septem- ber 2019. [41] Yi Li, Tomas Polakovic, Yong-Lei Wang, Jing Xu, Sergi Lendinez, Zhizhi Zhang, Junjia Ding, Trupti Khaire, Hilal Saglam, Ralu Divan, John Pear- son, Wai-Kwong Kwok, Zhili Xiao, Valentine Novosad, Axel Hoffmann, BIBLIOGRAPHY 48 and Wei Zhang. Strong Coupling between Magnons and Microwave Pho- tons in On-Chip Ferromagnet-Superconductor Thin-Film Devices. Phys. Rev. Lett., 123(10):107701, September 2019. [42] Michael G. Cottam and David J. Lockwood. Light Scattering in Magnetic Solids. Wiley-Interscience, New York, 1 edition edition, August 1986. [43] P. A. Pantazopoulos, N. Stefanou, E. Almpanis, and N. Papanikolaou. Pho- tomagnonic nanocavities for strong light -- spin-wave interaction. Phys. Rev. B, 96(10):104425, September 2017. [44] Jasmin Graf, Hannes Pfeifer, Florian Marquardt, and Silvia Viola Kusmin- skiy. Cavity optomagnonics with magnetic textures: Coupling a magnetic vortex to light. Phys. Rev. B, 98(24):241406, December 2018. [45] Sanchar Sharma, Babak Zare Rameshti, Yaroslav M. Blanter, and Gerrit E. W. Bauer. Optimal mode matching in cavity optomagnonics. Phys. Rev. B, 99(21):214423, June 2019. [46] Petros-Andreas Pantazopoulos, Nikolaos Papanikolaou, and Nikolaos Ste- fanou. Tailoring coupling between light and spin waves with dual pho- tonic -- magnonic resonant layered structures. J. Opt., 21(1):015603, Decem- ber 2018. [47] Evangelos Almpanis. Dielectric magnetic microparticles as photomagnonic cavities: Enhancing the modulation of near-infrared light by spin waves. Phys. Rev. B, 97(18):184406, May 2018. [48] A. Osada, A. Gloppe, Y. Nakamura, and K. Usami. Orbital angular mo- mentum conservation in Brillouin light scattering within a ferromagnetic sphere. New J. Phys., 20(10):103018, October 2018. [49] A. Osada, A. Gloppe, R. Hisatomi, A. Noguchi, R. Yamazaki, M. No- mura, Y. Nakamura, and K. Usami. Brillouin Light Scattering by Magnetic Quasivortices in Cavity Optomagnonics. Phys. Rev. Lett., 120(13):133602, March 2018. [50] J. A. Haigh, N. J. Lambert, S. Sharma, Y. M. Blanter, G. E. W. Bauer, and A. J. Ramsay. Selection rules for cavity-enhanced Brillouin light scattering from magnetostatic modes. Phys. Rev. B, 97(21):214423, June 2018. [51] Petros Andreas Pantazopoulos, Kosmas L. Tsakmakidis, Evangelos Alm- panis, Grigorios P. Zouros, and Nikolaos Stefanou. High-efficiency triple- resonant inelastic light scattering in planar optomagnonic cavities. New J. Phys., 21(9):095001, September 2019. [52] R. Hisatomi, A. Noguchi, R. Yamazaki, Y. Nakata, A. Gloppe, Y. Naka- mura, and K. Usami. Helicity-changing brillouin light scattering by magnons in a ferromagnetic crystal. Phys. Rev. Lett., 123:207401, Nov 2019. BIBLIOGRAPHY 49 [53] Purcell, E. M. Proceedings of the American Physical Society. Physical Review, 69(11-12):674 -- 674, June 1946. [54] Serge Haroche and Daniel Kleppner. Cavity Quantum Electrodynamics. Physics Today, 42(1):24 -- 30, January 1989. [55] Herbert Walther, Benjamin T. H. Varcoe, Berthold-Georg Englert, and Thomas Becker. Cavity quantum electrodynamics. Rep. Prog. Phys., 69(5):1325 -- 1382, April 2006. [56] R. J. Schoelkopf and S. M. Girvin. Wiring up quantum systems. Nature, 451(7179):664 -- 669, February 2008. [57] Daniel D. Stancil and Anil Prabhakar. Spin Waves: Theory and Applica- tions. Springer US, 2009. [58] L. D. Landau, L. P. Pitaevskii, and E. M. Lifshitz. Electrodynamics of Con- tinuous Media: Volume 8 (Course of Theoretical Physics S). Butterworth- Heinemann, 2nd edition. [59] Silvia Viola Kusminiskiy. Quantum Magnetism, Spin Waves, and Opti- cal Cavities. SpringerBriefs in Physics. Springer International Publishing, 2019. [60] Markus Garst, Johannes Waizner, and Dirk Grundler. Collective spin ex- citations of helices and magnetic skyrmions: Review and perspectives of magnonics in non-centrosymmetric magnets. Journal of Physics D: Ap- plied Physics, 50(29):293002, July 2017. [61] C. L. Dennis, R. P. Borges, L. D. Buda, U. Ebels, J. F. Gregg, M. Hehn, E. Jouguelet, K. Ounadjela, I. Petej, I. L. Prejbeanu, and M. J. Thornton. The defining length scales of mesomagnetism: A review. J. Phys.: Condens. Matter, 14(49):R1175, 2002. [62] Charles Kittel. Physical Theory of Ferromagnetic Domains. Rev. Mod. Phys., 21(4):541 -- 583, October 1949. [63] John Heebner, Rohit Grover, and Tarek Ibrahim. Optical Microresonators: Theory, Fabrication, and Applications. Springer Series in Optical Sciences. Springer-Verlag, New York, 2008. [64] Markus Aspelmeyer, Tobias J. Kippenberg, and Florian Marquardt, edi- tors. Cavity Optomechanics: Nano- and Micromechanical Resonators In- teracting with Light. Quantum Science and Technology. Springer-Verlag, Berlin Heidelberg, 2014. [65] Gregory S. MacCabe, Hengjiang Ren, Jie Luo, Justin D. Cohen, Hengyun Zhou, Alp Sipahigil, Mohammad Mirhosseini, and Oskar Painter. Phononic bandgap nano-acoustic cavity with ultralong phonon lifetime. arXiv:1901.04129 [cond-mat, physics:quant-ph], January 2019. BIBLIOGRAPHY 50 [66] Sanchar Sharma, Yaroslav M. Blanter, and Gerrit E. W. Bauer. Light scattering by magnons in whispering gallery mode cavities. Phys. Rev. B, 96(9):094412, September 2017. [67] J. A. Osborn. Demagnetizing Factors of the General Ellipsoid. Phys. Rev., 67(11-12):351 -- 357, June 1945. [68] D. F. Walls and Gerard J. Milburn. Quantum Optics. Springer-Verlag, Berlin Heidelberg, 2 edition, 2008. [69] Pierre Meystre and Murray Sargent. Elements of Quantum Optics. Springer-Verlag, Berlin Heidelberg, 4th edition, 2007. [70] A. M. Clogston, H. Suhl, L. R. Walker, and P. W. Anderson. Ferromag- netic resonance line width in insulating materials. Journal of Physics and Chemistry of Solids, 1(3):129 -- 136, November 1956. [71] Jonathan Kohler, Nicolas Spethmann, Sydney Schreppler, and Dan M. Stamper-Kurn. Cavity-Assisted Measurement and Coherent Control of Col- lective Atomic Spin Oscillators. Phys. Rev. Lett., 118(6):063604, February 2017. [72] Florian Marquardt, J. G. E. Harris, and S. M. Girvin. Dynamical Multi- stability Induced by Radiation Pressure in High-Finesse Micromechanical Optical Cavities. Phys. Rev. Lett., 96(10):103901, March 2006. [73] K. Yu. Guslienko. Magnetic Vortex State Stability, Reversal and Dynamics in Restricted Geometries. Journal of Nanoscience and Nanotechnology, 8(6):2745 -- 2760, June 2008. [74] Benjamin Pigeau, Grégoire de Loubens, Olivier Klein, Andreas Riegler, Florian Lochner, Georg Schmidt, and Laurens W. Molenkamp. Optimal control of vortex-core polarity by resonant microwave pulses. Nat Phys, 7(1):26 -- 31, January 2011. [75] A. A. Thiele. Steady-State Motion of Magnetic Domains. Phys. Rev. Lett., 30(6):230 -- 233, February 1973. [76] Arne Vansteenkiste, Jonathan Leliaert, Mykola Dvornik, Mathias Helsen, Felipe Garcia-Sanchez, and Bartel Van Waeyenberge. The design and ver- ification of MuMax3. AIP Advances, 4(10):107133, October 2014. [77] Victor A. S. V. Bittencourt, Verena Feulner, and Silvia Viola Kusminskiy. Magnon heralding in cavity optomagnonics. Phys. Rev. A, 100(1):013810, July 2019. [78] Sanchar Sharma, Yaroslav M. Blanter, and Gerrit E. W. Bauer. Optical cooling of magnons. Phys. Rev. Lett., 121:087205, Aug 2018. 

1112.4091

1

1112

Capillary-Wave Description of Rapid Directional Solidification

A recently introduced capillary-wave description of binary-alloy solidification is generalized to include the procedure of directional solidification. For a class of model systems a universal dispersion relation of the unstable eigenmodes of a planar steady-state solidification front is derived, which readjusts previously known stability considerations. We, moreover, establish a differential equation for oscillatory motions of a planar interface that offers a limit-cycle scenario for the formation of solute bands, and, taking into account the Mullins-Sekerka instability, of banded structures.

cond-mat.mes-hall

cond-mat

Capillary-Wave Description of Rapid Directional Solidification Alexander L. Korzhenevskii Institute for Problems of Mechanical Engineering, RAS, Bol'shoi prosp. V. O., 61, St Petersburg, 199178, Russia Institut fur Theoretische Physik IV, Heinrich-Heine-Universitat Dusseldorf, Universitatsstrasse 1, D-40225 Dusseldorf, Germany Richard Bausch Institut fur Theoretische Physik A, RWTH Aachen University, Templergraben 55, D-52056 Aachen, Germany (Dated: March 8, 2021) Rudi Schmitz A recently introduced capillary-wave description of binary-alloy solidification is generalized to include the procedure of directional solidification. For a class of model systems a universal dispersion relation of the unstable eigenmodes of a planar steady-state solidification front is derived, which readjusts previously known stability considerations. We, moreover, establish a differential equation for oscillatory motions of a planar interface that offers a limit-cycle scenario for the formation of solute bands, and, taking into account the Mullins-Sekerka instability, of banded structures. PACS numbers: 68.35.Dv,81.10.Aj,05.70.Np INTRODUCTION The main feature of a recently introduced capillary- wave model [1] for the solidification of a dilute binary alloy is the use of the interface position as a basic field variable, in addition to the concentration of the solute component. In the present paper this approach will be generalized to cover also the description of directional solidification, especially with regard to the rapid-growth regime. As outlined in reviews by Langer [2] and by Muller-Krumbhaar et al. [3], in directional solidification the growth of a crystal is accomplished by pulling it in opposite direction of an externally applied temperature gradient. We will mainly consider the case of a constant temperature gradient, which enters via a driving force in the equation of motion for the interface position. This form of description arises in the limit of an infinite heat conductivity from a more general model, involving energy density as an additional field variable. In general, such a model would allow to include the effect of heat diffusion. As a first application of our approach we scrutinize the possibility of stationary motions of a planar solidification front. The stability of such a front has been investigated in the rapid-growth regime with increasing regard of non- equilibrium effects by Mullins and Sekerka [4], Coriell and Sekerka [5], and by Merchant and Davis [6]. In Refs. [5] and [6] new oscillatory interface instabilities have been discovered in addition to the previously-known Mullins- Sekerka instability. Our own approach demonstrates that these effects are closely related to an instability, found by Cahn [7] in grain-boundary motion. The threshold of this instability represents a border line between regimes of steady-state and of non-steady-state motions of the solidification front. Non-steady interface motions operate in generating the periodic growth of layers with alternating homogeneous and dendritic micro-structures in binary alloys. This so- called banded structure occurs in many metallic alloys, as described in the review [8] by Carrard et al. who also offer a phenomenological explanation of the effect, using a quasi-stationary approximation. In a more microscopic treatment, Karma and Sarkissian [9] pointed out that the banding phenomenon is due to relaxation oscillations of the solidification front. This behavior, described in more detail in Ref. [10], was derived from numerical solutions of the diffusion equations for the solute concentration in a dilute binary alloy and for the temperature, supported by non-equilibrium boundary conditions, as formulated by Aziz and Boettinger [11]. Starting from a phase-field model, Conti has performed one- and two-dimensional simulations, describing the generation of solute bands [12], and of banded structures [13], respectively. He also has confirmed in Ref. [14] the observation by Karma and Sarkissian that the inclusion of heat diffusion leads to an increasing suppression of the formation of bands with decreasing heat conductivity. In the application of our capillary-wave description we are going to analyze the simplest-possible model, which shows the banding effect. We, accordingly, consider the dynamics of a planar interface, neglecting heat diffusion, and assuming an overall constant diffusion coefficient for the solute component. For a class of model systems with arbitrary equilibrium profiles of the solute concentration these properties lead to an integro-differential equation for the interface position. In case of a sufficiently small temperature gradient, realized in many experiments, this equation can be reduced to the differential equation of a damped nonlinear oscillator [15]. Stationary solutions of this equation turn out to exist in some region, limited by the threshold of the Cahn instability. This instability is attended by an oscillation, blowing up in the unstable regime to a limit-cycle behavior of the solidification front and of the solute concentration at the interface. Close to the stability threshold the transition from uniform to periodic solutions can analytically be evaluated by the Bogoliubov-Mitropolsky method [16]. The nature of the periodic solutions can be tuned from almost-harmonic to distinctive relaxation oscillations by changing the pulling velocity from the stability threshold to values deep inside the unstable regime. A benefit of our approach is that, apart from solving the oscillator equation in the unstable regime, all steps of the procedure could be accomplished analytically, which deepens our understanding of the banding effect. The discussion of micro-segregation effects at an oscillating solidification front requires to consider the stability of such a front in the transverse direction. Since, however, our results reveal an almost stationary behavior of the interface motion in the so far barely understood low- velocity regime [10], we presently only complement the established limit-cycle scenario by the standard Mullins- Sekerka procedure. Then, in some window of the model parameters, the high- and low-velocity sections of a limit cycle are located inside the Mullins-Sekerka stable and unstable regimes. As a result, a dendritic microstructure will develop in the low-velocity bands, which we consider as a kind of noise on the more macroscopic scale of the periodic array of the widely flat bands. CAPILLARY-WAVE MODEL The effective Hamiltonian of our capillary-wave model is a functional of the interface position Z(x, t) and of the excess concentration C(r, t) of the solute relative to its value CS in the solid phase. In terms of these field variables the effective Hamiltonian has the form (cid:90) (cid:90) H = + σ 2 κ 2 d2x (∂Z)2 (cid:104) d3r C − U (z − Z) (cid:105)2 , (1) 2 FIG. 1: Temperature-concentration phase diagram, showing the liquidus and solidus lines TL(C), TS(C) which meet at TM . The values CL and CS refer to the temperature TS. involving the solute concentration CL in the liquid phase, the latent heat L per unit volume, the miscibility gap ∆C ≡ CL − CS , (3) visible in Fig. 1, and the solute-concentration profile in thermal equilibrium, U (z − Z) = CE(z − Z) . (4) Whereas the expression (2) has been derived in Ref. [1], Eq. (4) directly follows from the equilibrium condition δH/δC = 0. From Ref. [1] we also adopt the equations of motion (cid:18) (cid:19) ∂tZ = Λ ∂tC = D ∇2 1 κ F − δH δZ δH δC , (5) where the rate Λ measures the interface mobility, and D is the diffusion coefficient of the solute, here assumed to have an overall constant value. The externally applied temperature gradient S, and the pulling velocity VP enter via the driving force F = L TS − T TM (6) where, in terms of the temperature TP at the steady-state position Z(t) = VP t, established already in Ref. [1]. It determines all static properties of the system in thermal equilibrium at some fixed temperature TS < TM where TM denotes the melting temperature of the solvent, showing up in the temperature-concentration phase diagram, Fig. 1. The input quantities in the Hamiltonian (1) are the surface tension σ, the coupling parameter (cid:18) ∂CL (cid:19)−1 L ∂T TM κ = − 1 ∆C , (2) T = TP + S(Z − ZP ) , ZP = VP t . (7) In Appendix A we will consider a more general model, which includes energy density as an additional field. We SC LC C MT ST T liquidsolidalso will show that the reduced model equations (1) - (7) emerge in the limit of an infinite heat conductivity. A dimensionless form of the model equations (5) can be obtained by adopting from Ref. [1] the mappings r → r , 1 ξ D ξ2 t → t , ξ σ F → F , C → C , 2 ∆C U → U , 2 ∆C (8) where, in the present context, the length ξ is defined by ξ ≡ ∆C 2 1 U(cid:48)(0) . In terms of the dimensionless quantities , p ≡ Λσ D , (cid:18) ∆C (cid:19)2 γ ≡ ξκ σ m2 ≡ ξ2L σ 2 S TM , (cid:90) +∞ the resulting equations of motion read dz U(cid:48)(z − Z)[C − U (z − Z)] , ∂tZ = F (Z) − γ 1 p ∂tC = D ∇2[C − U (z − Z)]) , with the driving force F given by −∞ F = FP − m2(Z − ZP ) , FP ≡ ξL σ TS − TP TM . (12) As a first application of Eqs. (11) and (12) we now will consider the steady-state motion of a planar interface with velocity VP in z-direction. STATIONARY PLANAR GROWTH Measuring velocities in units of the diffusion velocity, v ≡ V VD , VD ≡ D ξ , (13) the stationary growth of a planar solidification front is described in the co-moving frame z = vP t + ζ by the equations vP = FP + GP (vP ) − GP (0) , (14) 1 p GP (vP ) ≡ − γ CP (ζ; vP ) = (cid:90) +∞ (cid:90) ζ −∞ dζ U(cid:48)(ζ) CP (ζ; vP ) , −∞ dζ(cid:48) U(cid:48)(ζ(cid:48)) exp [vP (ζ(cid:48) − ζ)] . 3 These equations are identical to those, derived in Ref. [1] for the case of solidification by under-cooling the liquid phase from TS to TP . In particular, the result (14) is independent of the parameter m2, which only enters in discussing the stability of the planar morphology. For perturbations of the form h(x, t) ≡ Z(x, t) − vP t , c(x, ζ, t) ≡ C(x, ζ, t) − CP (ζ; vP ) + C(cid:48) (15) P (ζ; vP )h(x, t) , the resulting equations of motion read (cid:90) +∞ ζ + ∂2) c + [C(cid:48) ∂th = (∂2 − m2)h − 1 p ∂tc = vP ∂ζc + (∂2 −∞ dζ U(cid:48)(ζ) c(x, ζ, t) , (16) P (∂t − ∂2) + U(cid:48)∂2] h ζ . These equations have eigensolutions where ∂2 ≡ ∇2−∂2 of the form h(x, t) = h(q, ω) exp (iq · x + ωt) , c(ζ, x, t) = c(ζ, q, ω) exp (iq · x + ωt) , (17) which, after elimination of the component c, lead to the eigenmode dispersion relation (9) (10) (11) ω p + q2 + m2 − vP [GP (vP + λ) − GP (vP )] = λ2 − q2 vP + 2λ [GP (vP + λ) + GP (λ)] (18) where, deriving from the equation of motion for c, λ ≡ −(vP /2) +(cid:112)(vP /2)2 + ω + q2 . (19) The result (18) is similar to that, established in Ref. [1] and only differs by the additional term m2. Inspection of the low - q, ω behavior of the dispersion relation (18) leads to identify an eigenfrequency ω1(q), which captures the Mullins-Sekerka instability. Since the parameter m acts as a long-wavelength cutoff, the wave- number threshold qc for this instability is shifted from qc = 0 at m = 0 to some finite value, determined by the relations ω1(qc) = ω(cid:48) 1(qc) = 0. Elimination of qc then generates the neutral stability curve of the instability in form of a function vP (γ), with a parametric dependence on m. An explicit form of this neutral line will later be derived for a specific expression of U (z − Z). A second branch ω2(q) comprises an instability, similar to that, discovered by Cahn [7] in the process of grain- boundary motion. This instability is characterized by a gap at q = 0, determined by the relation ω2(0) Ω(vP ) = − F (cid:48) P (vP )Ω(vP ) ± 2m2 [F (cid:48) P (vP )Ω(vP )]2 4m4 − 1 , (cid:114) (cid:26) d 2 dv 2 P Ω(vP ) ≡ m (cid:20) − GP (vP ) + GP (0) 2 vP (cid:21)(cid:27)−1/2 . (20) In the limit m → 0 a single nonzero value of the gap [1]. survives, which is identical to that, found in Ref. For m (cid:54)= 0 the neutral stability curve F (cid:48) P (vP ) = 0 of the Cahn instability is attended by an oscillation of period Ω(vP ). Similar oscillatory instabilities have previously been observed by Coriell and Sekerka [5], and later by Merchant and Davis [6]. However, the neutral line, found in Ref. [6], differs from ours, which, we conjecture, arises from an unsettled generalization of the Gibbs-Thomson relation. We next will demonstrate that the instability, described by Eq. (20), acts as a seed for the limit-cycle behavior in the unstable regime. NON-STATIONARY PLANAR GROWTH For general unsteady motions of a planar interface the equations of motion (11) reduce to the form Z(t) = FP − m2[Z(t) − ZP (t)] (cid:90) +∞ 1 p − γ dz U(cid:48)(z − Z(t))[C(z, t) − U (z − Z(t))] , −∞ (∂t − ∂2 z )C(z, t) = − U(cid:48)(cid:48)(z − Z(t)) . (21) We are mainly interested in the late-stage behavior of Z(t), and, therefore, are going to replace in the first of the equations (21) the solution C(z, t) of the second equation, complemented by the boundary condition C(z,−∞) = 0. This leads to the expression dz(cid:48) ∂z(cid:48)G(z − z(cid:48), t − t(cid:48)) · U(cid:48)(z(cid:48) − Z(t(cid:48))) , (22) (cid:90) t −∞ dt(cid:48)(cid:90) +∞ −∞ C(z, t) = involving the Green function (cid:90) +∞ −∞ G(z, t) = exp (−k2t + ik z) . (23) dk 2π After the variable substitutions ζ ≡ z − Z(t) , ζ(cid:48) ≡ z(cid:48) − Z(t(cid:48)) , (24) and expansion of Z(t(cid:48)) around Z(t), we obtain 4 (25) G(ζ − ζ(cid:48) + Z(t) − Z(t(cid:48)), t − t(cid:48)) = (cid:90) +∞ −ik −∞ · exp dk 2π (cid:88) n≥2 exp [−k2(t − t(cid:48)) + ik(ζ − ζ(cid:48)) + ik(t − t(cid:48))v(t)]  , (−1)n n! (t − t(cid:48))n∂ n−1 t v(t) using the notation v(t) ≡ Z(t) = vP + h(t) . (26) For m2 (cid:28) 1, the higher-order contributions in n are increasingly negligible, as seen from the scaling procedure h → m−2h , ∂t → m2∂t , (27) t which leaves v(t) invariant, and attaches a factor m2n−2 to the contributions ∝ ∂ n−1 v(t). In the so-called quasi- stationary approximation all terms of order n ≥ 2 are neglected. The scenario, developed by Carrard et al. [8], is based on this procedure, applied to a phenomenological model where a low-velocity dendritic branch is added to the curve F = FP (v) for a planar interface. Without this additional dendritic branch all trajectories in the F, v - plane would inevitably run to v = 0. The phase-field simulations by Conti [12] effectively include all n-order terms in Eq. (25), and for the planar interface lead to the appearance of limit cycles, which are well separated from the line v = 0. An almost identical behavior arises, if in Eq. (25) we just include the term n = 2 , thereby going one step beyond the quasi-stationary approximation. After expansion of the exponential in Eq. (25) up to n = 2, and collection of all terms in Eq. (22) depending on t(cid:48), integration over τ ≡ t − t(cid:48) yields (cid:20) (cid:90) ∞ 0 (cid:21) − ik dτ 1 − ik τ 2 2 v exp [(−k2 + ik v) τ ] = − i 1 k − iv − 1 (k + iε)(k − iv)3 v . (28) The shift + iε in the denominator of the last term arises from including a term − ε v in the preceding exponential, which regularizes the singular point k = 0 at the upper bound of the integral. Next, we take care of the, so far, ignored contribution ik(ζ − ζ(cid:48)) in Eq. (25), and perform the integrations over k separately for the two final contributions in Eq. (28). The resulting equations (cid:90) +∞ (cid:90) +∞ −∞ ∂2 ∂v2 ∂2 ∂v2 1 2 1 2 v v v dk 2π (cid:90) +∞ (cid:110) −∞ 1 v dk 2π exp [ik(ζ − ζ(cid:48))] −∞ Θ(ζ − ζ(cid:48)) exp [−v(ζ − ζ(cid:48))] , −i k − iv = (29) exp [ik(ζ − ζ(cid:48))] −1 (k + iε)(k − iv)3 = exp [ik(ζ − ζ(cid:48))] (k + iε)(k − iv) dk 2π = Θ(ζ − ζ(cid:48)) exp [−v(ζ − ζ(cid:48))] + Θ(ζ(cid:48) − ζ) (cid:111) have, finally, to be multiplied with U(cid:48)(ζ(cid:48)) and integrated over ζ(cid:48), in order to evaluate the expression (22) for the solute concentration in the assumed approximation. In terms of the stationary concentration profile CP , presented in Eqs. (14), the result for C(z, t) reads C(z, t) = CP (ζ; v)+ v 1 2 ∂2 ∂v2 1 v [CP (ζ; v)+CP (ζ; 0)] . (30) Insertion of this into the first equation in Eqs. (21) leads to the closed equation of motion for Z(t) in the form 1 p + GP (v) − GP (0) + v v = FP − m2(Z − ZP ) ∂2 ∂v2 1 2 (31) 1 v [GP (v) + GP (0)] where GP (v) has been defined in Eqs. (14). For v = vP the result (31) consistently reduces to the first line in Eqs. (14). Subtracting the latter from Eq. (31), we find for the displacement 5 We mention that in the limit v(t) = vP + h(t) → 0 the inertial term shows, after application of the scaling procedure (60), the behavior M ( h) h ∝ m2 v3 h . (35) The most singular terms in higher-order contributions turn out to carry a pre-factor (m2/v3)n−1, so that our oscillator equation is only valid for velocities above the cross-over line v3 ∝ m2 . (36) We, furthermore, observe that, for small h, the second definition in Eq. (34) implies the behavior (cid:20) 1 p (cid:21) R( h) ≡ − G(cid:48) P (vP ) h + O(h2) (37) for the friction term in Eq. (33), so that, due to the first line in Eqs. (14), M (0)h + F (cid:48) P (vP ) h + m2 h + O(h2) = 0 , (38) in agreement with our linear stability analysis. P (vP ) > 0. In the regime F (cid:48) The nonlinear differential equation (33) obviously has the trivial solution h(t) = 0, which, however, according to Eq. (20), is unaffected by the Cahn instability in the regime F (cid:48) P (vP ) < 0 we will find solutions h(t), showing an oscillatory behavior in the limit t → ∞. This behavior is shared by the solute concentration C(Z(t), t) at the oscillating interface, as can be seen from Eq. (30), taken at ζ = 0. h(t) ≡ Z(t) − ZP (t) (32) LIMIT-CYCLE SOLUTIONS the simpler differential equation M ( h(t)) h(t) + R( h(t)) + m2 h(t) = 0 (33) where we have introduced the mass and friction functions (cid:34) (cid:35) ∂2 ∂v2 P GP (vP + h) + GP (0) M ( h) ≡ − 1 2 h − GP (vP + h) + GP (vP ) . vP + h R( h) ≡ 1 p , (34) Together with these definitions, Eq. (33) represents one of the central results of the present paper. It describes a damped nonlinear oscillator, general properties of which have been discussed in Ref. [16]. The equations (30) and (33) are valid for a whole class of models with varying equilibrium-concentration profiles U (z − Z). In order to obtain explicit solutions h(t) and C(Z(t), t), we choose the model U (z − Z) = Θ(Z − z) exp (z − Z) + Θ(z − Z)[2 − exp (Z − z)] , (39) derived in Ref. [1] from a two-parabola phase-field model. As also explained in Ref. [1], the choice (39) leads to the expressions CP (0, v) = 1 v + 1 , GP (v) = − γ v + 2 (v + 1)2 . (40) 6 FIG. 2: The functions M ( h), R( h) for γ = 0.01, p = 100, and vP = 0.3, resulting from the model (39) of the solute concentration. FIG. 3: Solutions h(t) for γ = 0.01, p = 100, m = 0.003, and pulling velocities vP = 0.522, and v = 0.52. The latter result allows us to determine the quantities (34), which e. g. for γ = 0.01, p = 100, vP = 0.3 have the form, shown in Fig. 2. A conspicuous property of the function M ( h) is its monotonous growth with decreasing velocity, which decisively affects the solutions h(t) in this regime, and, therefore, supports the procedure to include the inertial term in the oscillator equation (33). Another implication of Fig. 2 is that, for the present choice of the model parameters, the function R( h) is negative in some finite region where the solution h(t) = 0 is unstable. The first result in Eqs. (40), finally, permits to calculate the solute concentration (30) at the interface, once the solution h(t) of Eq. (33) has been found. Numerically obtained solutions h(t) for the parameter values γ = 0.01, p = 100, m = 0.003, and for the pulling velocities vP = 0.522 and vP = 0.52 are shown in Fig. 3. The threshold condition F (cid:48) P (vC) = 0 generally defines a critical velocity, which in the present case has the value vC ≈ 0.521. Above vC the oscillating trajectories h(t) converge to the value h(∞) = 0 whereas below vC they approach a limit cycle. Fig. 4 shows the same behavior further away from vC, so that, comparing these figures, one observes a kind of critical slowing down. In the regime vP − vC/vC (cid:28) 1 the envelopes in Fig. 3 can be calculated analytically by the Bogoliubov- Mitropolsky procedure [16], the application of which to the present case is described in Appendix B. The result for the solution of Eq. (33) then is found to be h(t) = a(t) cos ψ(t) (41) where ψ(t) is a rapidly oscillating phase, and a(t) is an amplitude, obeying the differential equation da dt = −ρ1 a − ρ3 a3 . (42) Here, ρ1 ≡ r1(vP − vC), and the coefficients r1, ρ3 are determined by the values of the model parameters γ, p, m. Eq. (42) has the solution a(t) = a0 1 + ρ3 ρ1 a2 0 exp [2ρ1 t] − ρ3 ρ1 a2 0 , (43) which for ρ1 > 0 and ρ1 < 0 describes the envelopes in Fig. 3. The asymptotic value of the limit-cycle amplitude (cid:26)(cid:20) (cid:21) (cid:27)−1/2 0.00.20.40.60.801234Mih0.00.20.40.60.8-0.0010.0000.001Rih020000400006000080000-101th020000400006000080000-10-505h t7 FIG. 5: Orbit of the cycle for γ = 0.01, p = 100, m = 0.003, and vP = 0.5. in Ref. [12]. From the associated behavior of C(Z(t), t) we, moreover, see that the transitions between high- and low-concentration layers are joined by large-acceleration sections. As already pointed out by Carrad et al. [8], this explains the appearance of relatively sharp interfaces between these layers. Fig. 8, finally, presents our result for the orbit of the limit cycle belonging to Fig. 7. BANDED-STRUCTURE FORMATION The layer formation, induced by the above limit-cycle solutions is unaffected by the Mullins-Sekerka instability, deriving from Eq. (18). This follows from Fig. 9, which shows the neutral stability lines, enclosing the unstable regions of the Cahn and the Mullins-Sekerka instabilities, and the projection of the limit cycle in Fig. 8. We have to point out, however, that the form of the the Mullins- Sekerka neutral line is only an approximate one, since it is related to a steady-state reference motion with velocity vP . The approximation seems, however, to be acceptable due to the almost stationary behavior of h(t) in Fig. 7 at low velocities. We, accordingly, expect that Fig. 7 induces the formation of precipitation-free solute bands. The approximation for the Mullins-Sekerka neutral line is apparently more justified for the limit cycle, belonging to Fig. 10. In this case, the projection of the cycle enters the unstable region of the Mulins-Sekerka instability, as seen in Fig. 11. Accordingly, the interface will develop a dendritic microstructure at low velocities, which dissolves again in the high-velocity regime. This is just what one expects to happen in the creation of banded structures, and it is in agreement with the simulations in Ref. [13]. FIG. 4: Solutions h(t) for γ = 0.01, p = 100, m = 0.003, and pulling velocities vP = 0.53, and vP = 0.51. shows the critical behavior a(∞) = (cid:112)−ρ1/ρ3. In the marginal case ρ1 = 0 Eq. (43) implies the algebraic decay (cid:112)1 + 2ρ3(a(0))2t a0 a(t) = . (44) The rapid oscillations in a fully developed limit cycle, gleaming through in Fig. 4, are most suitably analyzed by numerical computations. A first result is the orbit of a limit cycle in the h, h - plane, shown in Fig. 5 for the parameter values γ = 0.01, p = 100, m = 0.003, and the pulling velocity vP = 0.5. The related oscillations of the trajectories h(t), h(t), and of C(Z(t), t) are displayed in Fig. 6. Since the term m2h(t) measures the temperature at the oscillating interface, this quantity is effectively also included in Fig. 6. By reducing the pulling velocity to the value vP = 0.3 at constant parameters γ, p, m, one obtains the shape of the trajectories h(t), h(t), C(Z(t), t) deeper inside the limit-cycle regime. The results for h(t), h(t), displayed in Fig. 7, are remarkably close to the findings by Conti 020000400006000080000-101h t020000400006000080000-200h t-40-200-0.10.00.10.2 hih8 FIG. 6: Solutions h(t), h(t), C(Z(t), t) for γ = 0.01, p = 100, m = 0.003, and vP = 0.5. FIG. 7: Solutions h(t), h(t), C(Z(t), t) for γ = 0.01, p = 100, m = 0.003, and vP = 0.3. DISCUSSION A crucial point of our analysis is the observation that the periodic motion of a planar solidification can only be explained, if we go one step beyond the quasi-stationary approximation in the expansion (25). For a quantitative evaluation higher-order terms can be neglected, because the definition of the effective expansion parameter m2 in Eqs. (10) implies m2 ≈ 5 × 10−5, if we adopt from Ref. [8] the value S = 2 × 105K/cm of the temperature gradient, and from Ref. [12] the material parameters TM = 1728 K, L = 2350 J/cm3, σ = 3.7 × 10−5J/cm2 for Nickel, and the interface thickness 2ξ = 1.68 × 10−7cm. The fact that the restoring force in Eq. (33) is given by m2h(t) may raise the suspicion that the presence of a temperature gradient is an essential ingredient of our 02000400060008000-40-200 th02000400060008000-0.10.00.10.2 tih020004000600080000.600.660.72 Ct200002500030000350004000045000-600-3000ht2000025000300003500040000450000.00.61.2tih2000025000300003500040000450000.30.60.9C t9 FIG. 8: Orbit of the cycle for γ = 0.01, p = 100, m = 0.003, and vP = 0.3. FIG. 9: Neutral lines, enclosing the regions of the Cahn (solid line), and of the Mullins-Sekerka (dashed line) instability. The vertical line is the projection of the limit cycles in Fig. 9. theory. This is only true, however, for a planar geometry of the solidification front. In case of a growing spherical nucleus the parameter m2 turns out to be proportional to the ratio ξ/Rc where Rc is the critical radius of the droplet. Generally, the parameter m determines the period Ω0 of a limit cycle. Close to the threshold at vC the friction term in Eq. (33) can be neglected, so that, due to scaling, Ω0 ∝ m. Deep inside the limit-cycle regime accelerations are negligible in most parts of the trajectories h(t) in Figs. 7 and 10, suggesting to neglect the inertial term in Eq. (33). Its scaling behavior then implies Ω0 ∝ m2, in accordance with the statement in Ref. [8] that the band width in a pronounced banded structure is inversely proportional to the temperature gradient. As a final point we note that most phenomenological approaches are based on the assumption of an N -shaped force-velocity relation. This suggests the formation of a hysteresis loop, which is considered to represent the limit cycle, describing the defect oscillations. In our model the FIG. 10: Solutions h(t), h(t), C(Z(t), t) for γ = 0.02, p = 100, m = 0.003, and vP = 0.5. driving force is a convex function of velocity, excluding the existence of a hysteresis loop. Instead the necessary turnaround of a trajectory at low velocities is provided by the inertial term in the oscillator equation, which proves the importance of including this term in the equation for the interface position. -600-30000.00.61.2hih0.0010.010.10.0010.010.1110Pvγ1000020000300004000050000-3000-2000-10000 ht1000020000300004000050000024 tih10000200003000040000500000.00.61.2C tthe parameters κ0 and D0. The only generally important constraint on the coupling constants is κ0(cid:101)κ − ν2 ≥ 0 , 10 (49) which ensures stability of the Hamiltonian (45). The equations of motion of the generalized model read , δH ∂tZ = − Λ δZ ∂tC = D0 ∇2 1 κ0 ∂t(cid:101)C = (cid:101)D ∇2 1(cid:101)κ δH δC δH δ(cid:101)C (50) , where (cid:101)D is the heat diffusion constant. The relation , (51) δH T (r, t) ≡ TS + defines a temperature field via a shifted local Legendre δ(cid:101)C transform of (cid:101)C(r, t), obeying the condition T (r, t) = TS in thermal equilibrium δH/δ(cid:101)C = 0. (cid:101)D → ∞, and the last of the Eqs. (50) is solved, for the In the limiting case of an infinite heat conductivity, boundary conditions T (ZP ) = TP , T (cid:48)(ZP ) = S, by the static temperature field T (z) = TP + S(z − ZP ) . (52) Insertion of the Hamiltonian (45) and the result (52) into Eq. (51) leads to the relation (cid:105) (cid:105) (cid:104)(cid:101)C(r, t) −(cid:101)U (z − Z) (cid:101)κ = −(cid:104) If the expression for (cid:101)C − (cid:101)U , extracted from Eq. (53), + ν TS − TP + S(z − ZP ) C(r, t) − U (z − Z) (cid:104) (cid:105) (53) . is inserted into the second of the Eqs. (50), one recovers the corresponding equation in Eqs. (5) with the reduced diffusion constant (cid:18) D ≡ 1 − ν2 κ0(cid:101)κ D0 . (54) In the calculation of the force − δH/δZ, entering the first of the Eqs. (50), the relations (48) and (53) can be used to eliminate the quantities (cid:101)U ,(cid:101)C, which leads to the result − δH (cid:90) +∞ −κ C − U (z − Z) dz U(cid:48)(z − Z) = σ ∂2Z (55) (cid:104) (cid:105) δZ dz U(cid:48)(z − Z) TS − TP + S(z − ZP ) (cid:105) (cid:90) +∞ −∞ L + TM ∆C −∞ (cid:19) (cid:104) FIG. 11: Neutral lines of the Cahn (solid line), and of the Mullins-Sekerka (dashed line) instability. The vertical lines are the projections of the limit cycles in Figs. 6 and 10. APPENDIX A Within our capillary-wave approach the most general effective Hamiltonian for the directional solidification of a dilute binary alloy reads H = (cid:26) κ0 (cid:90) (cid:90) (cid:104) (cid:105) (cid:105)(cid:104)(cid:101)C −(cid:101)U (z − Z) (cid:104) (cid:105)2(cid:27) (cid:104)(cid:101)C −(cid:101)U (z − Z) + (cid:101)κ d2x (∂Z)2 + C − U (z − Z) + ν d3r σ 2 2 . (cid:105)2 C − U (z − Z) (45) 2 Here, we have introduced a field (cid:101)C(r, t), which is related TM ∆C (cid:101)C(r, t) ≡ to the energy density E(r, t) by the equation (cid:18) 1 − ν(cid:101)κ (cid:19) E(r, t) From this and the equilibrium condition δH/δ(cid:101)C = 0 we conclude that (cid:101)U (z − Z) obeys the relation (cid:19) L (cid:101)U (+∞) −(cid:101)U (−∞) = TM ∆C (46) (47) TM L . , (cid:18) 1 − ν(cid:101)κ L TM which, remembering the relations (3) and (4), suggests to refine our model by assuming (cid:101)U (z − Z) = (cid:18) L TM ∆C (cid:19) − ν(cid:101)κ U (z − Z) . (48) For the derivation of the model (1) - (7) the physical meanings of the coupling constants ν,(cid:101)κ are irrelevant, because they will be absorbed into renormalizations of 0.0010.010.10.0010.010.1110Pv γwith the renormalized coupling constant Following Ref. [16], we look for solutions of Eq. (61) 11 κ ≡ κ0 − ν2(cid:101)κ . in form of the expansion (56) f (t) = α(t) cos ψ(t) + (cid:88) ν=1,2 ενuν(α(t), ψ(t)) , (63) The first two terms on the right-hand side of Eq. (55) are identical to the force − δH/δZ in the first equation in Eqs. (5). Assuming that U(cid:48)(ζ) is an even function, as in case of the model (39), the last integral in Eq. (55) reduces to the driving force (6). We mention that the coupling term ∝ ν in the Hamiltonian (1) only gives rise to a shift in Eqs. (46), (47), (48), and to parameter renormalizations in Eqs. (54), (56), which all disappear in the commonly considered case ν = 0. APPENDIX B In order to derive the differential equation (42) for the amplitude a(t), we start from the oscillator equation (33), rewritten in the form (cid:20) 1 (cid:21) h + Ω2h = − R( h) M ( h) + m2 − 1 M ( h) M (0) h . (57) where Ω, R( h) and M ( h) depend parametrically on vP . Close to the stability threshold of Eq. (38) is sufficient to evaluate all terms in Eq. (57) to leading order of an expansion in ε ≡ vP − vC , F (cid:48) P (vC) ≡ 0. (58) Since, according to Eq. (38), the linear part of R( h) in h is of order ε, the leading expression for Eq. (57) has the general structure complemented by the constraints (cid:90) 2π (cid:90) 2π 0 0 dψ uν(α, ψ) sin ψ = 0 (64) dψ uν(α, ψ) cos ψ = 0 , which ensure that the uν(α, ψ) for ν = 1, 2 only contain higher harmonics in ψ. The equations = ενAν(α) , (65) (cid:88) dα dt dψ dt ν=1,2 = Ω + (cid:88) ν=1,2 ενBν(α) , also assumed in Ref. [16], reflect the conditions that the amplitude α(t) and the difference ψ(t) − Ω t are slowly varying variables. We are mainly interested in the functions Aν(α) and Bν(α), which can be obtained by projecting Eq. (61) onto the first harmonics sin ψ and cos ψ. Then, due to Eqs. (64), the contributions uν(α, ψ) in Eq. (63) cancel, which allows us to look from the beginning for solutions of the simplified form f (t) = α(t) cos ψ(t) . (66) h + Ω2(vC)h = εX1(vC) h + Z(h, h; vC) . (59) Its first and second derivatives are given by If we, finally, apply the scaling transformation h(t) ≡ ε f (t) , (60) + f (t) = − Ω α(t) sin ψ(t) εν(Aν cos ψ − αBν sin ψ) , Q1(f, f ) = X1 Q2(f, f ) = X3 f + X2 2 f 3 + Y2f f f 2 + Y1f f , (62) . From this and the relations (62) we, finally, obtain we obtain, up to second order in ε, the representation f + Ω2f = ενQν(f, f ) , (61) (cid:88) ν=1,2 which is the starting point of the work by Bogoliubov and Mitropolsky [16] where in the present case + ε2 (67) (68) (cid:88) ν=1,2 (cid:88) (cid:20)(cid:18) (cid:18) A1 ν=1,2 − f (t) = − Ω2 α(t) cos ψ(t) − 2Ω εν(Aν sin ψ + αBν cos ψ) (cid:19) dA1 dα − αB2 1 2A1B1 + αA1 (cid:19) (cid:21) sin ψ . cos ψ dB1 dα (cid:90) 2π (cid:90) 2π (cid:90) 2π 0 0 0 B1(α) = − 1 4πΩ A1(α) = − 1 4πΩ A2(α) = − 1 4πΩ Q1(f, f )) cos ψ = 0 , (69) Q1(f, f )) sin ψ = Q2(f, f )) sin ψ = 1 4 X1α , 3 16 X3α3 . After scaling back to the variable h(t) via Eq. (60), one recovers the result (42), where 12 [2] J. S. Langer, Rev. Mod. Phys. 52, 1 (1980). [3] H. Muller-Krumbhaar, W. Kurz, E. Brener, Solidification in Phase Transformations in Materials, ed. G. Kostorz (Wiley-VCH, Weinheim, FRG, 2005). [4] W. W. Mullins, R. F. Sekerka, J. Appl. Phys. 35, 444 (1964). [5] S. R. Coriell, R. F. Sekerka, J. Cryst. Growth. 61, 499 (1983). [6] G. J. Merchant, S. H. Davis, Acta metall. mater. 38, 2683 (1990). [7] J. W. Cahn, Acta Metal. 10, 789 (1962). [8] M. Carrard, M. Gremaud, M. Zimmermann, W. Kurz, Acta metall. mater. 40, 983 (1992). [9] A. Karma, A. Sarkissian, Phys. Rev. Lett. 68, 2616 (1992). a(t) ≡ ε α(t) . (70) [10] A. Karma, A. Sarkissian, Phys. Rev. E 47, 513 (1993). [11] M. J. Aziz, W. J. Boettinger, Acta metall. mater. 42, The coefficients r1, ρ3, appearing in this equation, depend on the parameters γ, p, m, and on the critical velocity vC, and can be calculated from Eqs. (59), (34), and (62). For the choice γ = 0.01, p = 100, m = 0.003, and vC = 0.5214 one finds r1 = 1.5575 · 10−2 and ρ3 = 9.885 · 10−5. A. L. Korzhenevskii wants to express his gratitude to the University of Dusseldorf for its warm hospitality. This work has been supported by the DFG under BA 944/3-3, and by the RFBR under N10-02-91332. 527 (1994). [12] M. Conti, Phys. Rev. E 58, 2071 (1998). [13] M. Conti, Phys. Rev. E 58, 6101 (1998). [14] M. Conti, Phys. Rev. E 58, 6166 (1998). [15] A. L. Korzhenevskii, R. Bausch, R. Schmitz, to be pub- lished. [16] N. N. Bogoliubov, Y. A. Mitropolsky, Asymptotic Meth- ods in the Theory of Nonlinear Oscillations, (Gordon and Breach Science Publishers, New York, 1961). [1] A. L. Korzhenevskii, R. Bausch, R. Schmitz, Phys. Rev. E 83, 041609 (2011). 

1704.00782

3

1704

Robustness of symmetry-protected topological states against time-periodic perturbations

The existence of gapless boundary states is a key attribute of any topological insulator. Topological band theory predicts that these states are robust against static perturbations that preserve the relevant symmetries. In this article, using Floquet theory, we examine how chiral symmetry-protection extends also to states subject to time-periodic perturbations $-$ in one-dimensional Floquet topological insulators as well as in ordinary one-dimensional time-independent topological insulators. It is found that, in the case of the latter, the edge modes are resistant to a much larger class of time-periodic symmetry-preserving perturbations than in Floquet topological insulators. Notably, boundary states in chiral time-independent topological insulators also exhibit an unexpected resilience against a certain type of symmetry-breaking time-periodic perturbations. We argue that this is a generic property for topological phases protected by chiral symmetry. Implications for experiments are discussed.

cond-mat.mes-hall

cond-mat

a Robustness of symmetry-protected topological states against time-periodic perturbations Department of Physics, University of Gothenburg, SE 412 96 Gothenburg, Sweden Oleksandr Balabanov and Henrik Johannesson The existence of gapless boundary states is a key attribute of any topological insulator. Topo- logical band theory predicts that these states are robust against static perturbations that preserve the relevant symmetries. In this article, using Floquet theory, we examine how chiral symmetry- protection extends also to states subject to time-periodic perturbations − in one-dimensional Floquet topological insulators as well as in ordinary one-dimensional time-independent topological insula- tors. It is found that, in the case of the latter, the edge modes are resistant to a much larger class of time-periodic symmetry-preserving perturbations than in Floquet topological insulators. Notably, boundary states in chiral time-independent topological insulators also exhibit an unexpected re- silience against a certain type of symmetry-breaking time-periodic perturbations. We argue that this is a generic property for topological phases protected by chiral symmetry. Implications for experiments are discussed. PACS numbers: 71.10.Fd, 71.23.An, 73.20.At, 67.85.-d I. INTRODUCTION A hallmark of a symmetry-protected topological (SPT) phase of matter − topological insulators and topologi- cal superconductors being well-known examples 1,2 − is the presence of gapless boundary states3. The very ex- istence of these states is a consequence of the nontriv- ial topology of the bulk band structure ("bulk-boundary correspondence"), with the symmetry protection ensur- ing their robustness against static gap-preserving per- turbations as long as the relevant symmetries remain unbroken4–6. As is well known, this robustness, along with other unique properties of the boundary states, has raised the prospects for exploiting SPT phases for future technologies − from applications in spintronics7 to topo- logical quantum computation8. How does the symmetry protection play out when the boundary states are subject to time-dependent perturba- tions? While a comprehensive answer has to await fur- ther advances in the theory of SPT phases, the case of time-periodic perturbations can be addressed efficiently by using Floquet theory9,10. In this work we exploit this advantage to answer the question to what extent the midgap boundary states in a SPT phase are robust against a time-periodic disordering perturbation. While such perturbations may not occur naturally in physical systems, they have recently been realized in highly con- trolled experiments with cold atomic11 and optical12 se- tups. As such, their study could open a new inroad to explore the physics of SPT phases. We focus on two typical brands of one-dimensional (1D) SPT phases − Floquet topological insulators13,14 and ordinary time-independent insulators1,2 − with boundary states protected by chiral symmetry3. The first type − the Floquet insulator − is obtained by driving a system with a time-periodic field, resulting in symmetry- protected boundary states in nonequilibrium13–19. The protection of the states comes about as a feature of the time-evolution operator, implying that robustness against added time-periodic perturbations can be investi- gated within the same Floquet formalism which describes the driving of the bulk. As noted by Asb´oth et al.20, the robustness of boundary states in a chiral Floquet system driven by two time-periodic fields of the same frequency depends critically on the relative phase of the driving. By a systematic study we here extend this picture to the case where one field drives the bulk, with the other acting as a time-periodic disordering perturbation. As one would expect, we find that boundary states remain robust to time-periodic perturbations that preserve chiral symme- try. In its turn, the very existence of chiral symmetry in an unperturbed driven system depends crucially on the phase of the bulk driving. It follows that the states re- main protected only when the phase of the time-periodic disordering perturbation is properly tuned to the driv- ing in the bulk. This is explicitly shown analytically and confirmed numerically. Turning to the time-independent topological insula- tors, we can again use Floquet theory to study the ef- fect of time-periodic perturbations. This is so, since any time-independent Hamiltonian is trivially periodic in time. Our analysis and findings here can be summa- rized as follows: First, we make explicit how the free- dom in choosing starting point in the stroboscopic Flo- quet time evolution implies that chiral time-independent systems actually possess an infinite number of chiral symmetries. As a consequence, the edge states are ro- bust against a much larger class of symmetry-preserving time-periodic perturbations compared with those of a Floquet topological insulator. Second, we establish a class of symmetry-breaking time-periodic perturbations for which the boundary states display an unexpected re- silience. A detailed analysis reveals how this property is manifested in Floquet perturbation theory: The ef- fect coming from this class of perturbations gets sup- pressed by the very structure of the unperturbed chiral- symmetric spectrum, implying that its expected leading- order contribution vanishes identically. Such protection is a very interesting feature because it hints that, even when the chiral symmetry is broken, a residue of it can still have an effect on the system's behavior. For simplicity, our analysis proceeds by way of exam- ple, with the Su-Schrieffer-Heeger (SSH) model21 as a case study. To cover the two classes of topological insu- lators − Floquet and time-independent ones − we con- sider a periodically driven version of the SSH model as well as the original time-independent variety. Our choice is motivated by the fact that the SSH model serves as a prototype for band insulators exhibiting topological phases22–25. We should point out, however, that our analysis can be carried over to any 1D chirally symmet- ric topological phase (symmetry classes AIII, BDI, DIII, and CII in the Altland-Zirnbauer classification26). The paper is planned as follows: In the next section, af- ter a brief introduction to Floquet formalism, we present the harmonically driven SSH model and describe its topo- logical properties. We then discuss the topological pro- tection of the boundary states and explicitly identify the types of time-periodic disorder in the presence of which the states remain robust. In Sec. III we turn to the un- driven (time-independent) SSH model and describe first the symmetries that the system possesses within Floquet theory. It is then argued that the boundary states are robust to a much broader class of time-periodic pertur- bations than in the driven case and this is verified nu- merically. Next, we present our perturbative analysis re- vealing the enhanced resilience of the boundary states for a certain type of symmetry-breaking time-periodic per- turbations. This is followed by a qualitative argument why this property may hold also outside the perturbative regime, supplemented by supporting data from numeri- cal computations. In the same section we discuss time- periodic disorder in the chemical potential − an impor- tant type of perturbations for making contact with exper- iments. Finally, we comment on the feasibility to test our predictions in an experiment using optically trapped cold atoms. A brief summary and outlook is given in Sec. IV. II. HARMONICALLY DRIVEN SSH MODEL A. Floquet formalism To set the stage, let us recall that the time evolution of any quantum system driven by time-periodic fields can be described using Floquet theory9,10. Within this formal- ism an equivalent of energies, so-called quasienergies, can be defined and one may consider the band structure of the system in terms of its quasienergy spectrum. Specifically, the Schrodinger equation with a time-periodic Hamilto- nian H(t) = H(t + T ) has a complete set of solutions ψn(t)(cid:105) = exp(−iεnt)un(t)(cid:105), commonly called steady states, where εn denotes the quasienergies and un(t)(cid:105) = un(t+T )(cid:105) for all times t (with ≡ 1). The quasienergies, defined modulo 2π/T , appear in the dynamical phase ac- quired by the steady states, and in this sense they are similar to ordinary energies. The quasienergy spectrum can be found by using the fact that the states un(t)(cid:105) are 2 FIG. 1: A schematic illustration of the harmonically driven SSH chain, with hopping amplitudes γ1/2± v(t), where v(t) ∼ cos(Ωt). eigenstates of the evolution operator U (t, T + t) associ- ated with the eigenvalues exp(−iεnT ). To find the band structure it is thus sufficient to diagonalize U (t, T + t) for some conveniently chosen fixed time t. Alternatively, one can Fourier transform the Schrodinger equation and perform the calculations in the frequency domain9,10. B. Topological characteristics of the model The SSH model consists of spinless fermions hopping on a 1D staggered lattice21. Here we assume that the hopping amplitudes have both static and time-dependent harmonically modulated components as shown in Fig. 1. Within a tight-binding approximation the Hamiltonian with vanishing chemical potential can be written as27 (cid:17) (cid:17) , (1) H(t) = −(cid:88) (cid:88) j (cid:16) (cid:16) + j † † A,jcB,j + γ2c γ1c B,j−1cA,j + H.c. † A,jcB,j − v(t)c v(t)c † B,j−1cA,j + H.c. † where c σ,j and cσ,j (σ = A, B) are creation and an- nihilation operators, γ1 and γ2 are the static intra- cell and intercell hopping amplitudes respectively, and v(t) = 2Vac cos(Ωt) is the harmonically modulated com- ponent of the hopping. The undriven SSH Hamiltonian H0 [defined by setting v(t) = 0 in Eq. (1)] has chiral symmetry, implying the existence of a unitary operator Γ such that ΓH0Γ = −H0 where ΓcA,jΓ = cA,j and ΓcB,jΓ = −cB,j. The represen- tation [28] Γ ≡ eiπ(cid:80) † B,j cB,j j c (2) allows for an easy check of chiral symmetry in the second- quantized formalism, more convenient than if one were to directly use a representation of the second-quantized an- tiunitary chiral operator [26]. Clearly, the chiral symme- try here reflects a sublattice symmetry, i.e., the property that the SSH Hamiltonian H0 does not couple sites on the same sublattice. As it turns out, the harmonically driven SSH model has a chiral symmetry as well, but now given for the evolution operator. This can be established by first noticing that H(t) in Eq. (1) satisfies the relation ΓH(t)Γ = −H(−t). (3) A B A A B unit cell B A B 3 FIG. 2: The topological invariants ν0 and νπ calculated for Vac = 0.2 Ω and different values of γ1 and γ2. νπ/0 = 1(0) cor- responds to a topologically nontrivial (trivial) winding num- ber. As shown in Appendix A, this property is sufficient for proving that the evolution operators F ≡ U (0, T /2) and G ≡ U (T /2, T ) are related as F = ΓG†Γ. From this follows immediately that the periodically driven model has chiral symmetry, as expressed by ΓU (0, T )Γ = U−1(0, T )20. With chiral symmetry in hand, we refer to a general result20 to conclude that the topological phases of the harmonically driven SSH model can be characterized by two integer topological invariants, v0 and vπ. These invariants count the number of SPT boundary states at each end of the chain, corresponding to quasiener- gies 0 and Ω/2 respectively. It has been suggested that both of them can be extracted from the operator F de- fined above, after imposing spatially periodic boundary conditions20,29. For the present problem we have car- ried out the calculation numerically by discretizing the time-evolution into a large number of intervals and as- suming that the Hamiltonian H(t) is constant in each of them. The values of the topological invariants for differ- ent static hopping amplitudes are displayed in Fig. 2, in excellent agreement with that obtained from an analysis of the Zak phase27. C. Protection against symmetry-preserving boundary perturbations In exact analogy to the time-independent case, the symmetry protection of the topological invariants against time-periodic perturbations comes from the restriction that the chiral symmetry places on the quasienergy spec- trum. For details, see Appendix B. As a consequence, the boundary states in the thermodynamic limit of the SSH model, Eq. (1), are expected to be robust against gap- preserving time-periodic perturbations V (t) which do not violate the chiral relation ΓV (t)Γ =−V (−t). This condi- tion is satisfied for any site-dependent perturbation of the hopping amplitudes γ1 and γ2 that is even in time (which, trivially, includes static perturbations). In contrast to a disordering of the hopping amplitudes, a perturbation from an added time-periodic staggered chemical poten- FIG. 3: Quasienergy spectra of harmonically driven SSH chains subject to time-periodic boundary perturbations, here realized as a spatial disordering of the amplitudes for hopping (∼ γ1, γ2) or of an added staggered chemical potential (∼ ∆). The chains have 80 sites, with unperturbed γ1 = 0.15 Ω, Vac = 0.2 Ω, and with γ2 swapped from 0 to 1.2 Ω. The disorder is added over 20 sites from one of the boundaries, with the disorder amplitudes varying randomly within the in- terval [−0.2 Ω, 0.2 Ω]. Here we display levels for bulk states (black), perturbed (red), and unperturbed (green) edge states corresponding to single disorder realizations. For each type of perturbation 100 different disorder realizations were con- sidered, with the midgap quasienergies always to be found confined to the corresponding pink regions. † A,jcA,j − c tial ∆ (proportional to c in time in order to respect chiral symmetry20. † B,jcB,j) has to be odd In Fig. 3 we numerically examine the robustness of the boundary states against various types of time-periodic perturbations, here added to one of the boundary regions (taken to extend over several sites near the left edge of the chain) as a spatial disordering of the amplitudes of the hopping (∼ γ1, γ2) or the staggered chemical poten- tial (∼ ∆). By confining the perturbation to a boundary region, we are ensured that the quasienergy bulk gaps stay open for all disorder realizations. In contrast, bulk perturbations may close the gap for certain realizations of large-amplitude disorder, removing the protection. The quasienergies were obtained by truncating the Hamilto- nian in the frequency domain9 and then diagonalizing it numerically. All perturbations were chosen to be har- monic in time, however, our approach can be generalized to any time-periodic perturbation. The numerical re- sults validate our predictions above: The midgap levels γ20Ωγ10Ωγ20Ωγ10Ω000101100ν0νπ"[+]"[+]"[+] .2[+] .2[+] -12-1212-12"cos(+t).2cos(+t).1cos(+t).1sin(+t)"sin(+t)12.2sin(+t)121100at quasienergies 0 and Ω/2 (in red color in Fig. 3), cor- responding to the perturbed boundary states, are robust against perturbations of hopping amplitudes (staggered chemical potential) which are even (odd) in time, other- wise not. It is important to realize that the zero reference time gets fixed by us when we take the bulk driving to be proportional to cos(Ωt). Thus, the phases of the allowed boundary perturbations for which the midgap states re- main robust are determined by the phase of the bulk driving. We should also mention that the hopping disor- ders were taken to be complex in the computations, thus disabling particle-hole symmetry to protect the boundary states when the chiral symmetry is broken by a pertur- bation. It is interesting to note the appearance of additional edge states in the topologically trivial regime of the har- monically driven SSH chains, with values of γ2 near Ω (see Fig. 3). Similar states have been seen also in other 1D Floquet systems30. These boundary modes are not expected to be of a topological origin. Still, such states are robust against weak perturbations because they are separated from the bulk modes by a finite gap. Intrigu- ingly, the response of these states to various time-periodic perturbations seems to be correlated with the robust- ness of the topological edge states: In Fig. 3 we see that the corresponding quasienergy shifts are much more pro- found in the cases where chiral symmetry is broken by the perturbation. This feature warrants further study. III. TIME-INDEPENDENT SSH MODEL A. Symmetry preservation in the model Time-independent models can also be handled within the Floquet formalism because any static Hamiltonian is periodic in time for any frequency Ω. Thus, we may write the evolution operator of a time-independent model as U (t0, t0 + T ), where t0 is a fixed reference time and T is interpreted as a period in the Floquet formalism. By this simple change of perspective we can systematically explore the robustness of the boundary states against time-periodic perturbations. It should be stressed that once we enter Floquet theory the notion of energy is re- placed by quasienergy and this must be carefully taken into account. Having expressed the evolution operator for a time- independent chirally symmetric model as U (t0, t0 + T ), it is essential to note that within the Floquet formalism the model actually supports an infinite number of chi- ral symmetries. This is so because ΓU (t0, t0 + T )Γ = U−1(t0, t0 + T ) for any choice of reference time t0. Since the effect of a time-periodic perturbation is independent of the choice of t0, the perturbation has to break all these chiral symmetries in order to kick the quasienergies away from zero. Therefore, the symmetry-protected boundary modes in static chiral models are expected to be robust to a much broader class of time-periodic perturbations in 4 FIG. 4: Quasienergy spectra of time-independent SSH chains [v(t) = 0 in Eq. (1)] under influence of various time-periodic boundary perturbations. These perturbations were imple- mented as a spatial disordering of the hopping amplitudes (∼ γ1, γ2) or of an added staggered chemical potential (∼ ∆). The chains were taken with γ1 = 0.15 Ω and γ2 varying be- tween 0 and 1.2 Ω. Each chain consists of 80 sites and is disordered over the first 20 sites from one of the bound- aries, with the corresponding disorder amplitudes chosen ran- domly within the interval [−0.2 Ω, 0.2 Ω]. Here we illustrate quasienergies for bulk states in black and edge states in red, respectively (the quasienergies of the perturbed and unper- turbed edge modes perfectly match in this case). comparison with the Floquet topological insulators dis- cussed above. In Fig. 4 we show numerical data for undriven SSH chains [v(t) = 0 in Eq. (1)] subject to the same time- periodic disorders that we considered in Fig. 3 for the harmonically-driven case. The quasienergies correspond- ing to the symmetry-protected states are seen to be com- pletely unaffected by the harmonic single-parameter dis- orders, independently of the phase of the perturbative driving. This is in full agreement with the discussion above because, in each of these cases, the chiral symme- try is preserved for some t0. We briefly note that the band structure now supports only a single gap, with all dynamical gaps being closed because of the absence of bulk driving. Also, the edge states for γ2 near Ω are not present anymore. .2[+] "[+]"[+]"[+] .2[+] .2cos(+t).2sin(+t).1cos(+t).1sin(+t)"cos(+t)"sin(+t)-1212120101-1212-12B. Resilience of the boundary states against symmetry-breaking perturbations 5 (cid:88) In general, topological boundary states are not pro- tected against symmetry-breaking perturbations. Still, in what follows we show that boundary states of time- independent 1D chiral systems inherit a residual protec- tion also against a large class of symmetry-breaking time- periodic perturbations. To be more specific, we show that boundary states of these systems show a resilience against perturbations of the form V (t) = Vn cos(nΩt + φn), (4) n where ∀n ∈ N : ΓVnΓ = ±Vn (± can depend on n) and φn ∈ R. Importantly, this class of perturbations which in general break chiral symmetry for all choices of reference time t0, neither depends on the specifics of the model considered nor on any spatial fine-tuning. To analyti- cally uncover this surprising resilience of the boundary states we turn to Floquet perturbation theory and estab- lish that the expected leading-order quasienergy correc- tion vanishes identically. Analogous to conventional perturbation theory, Flo- quet perturbation theory allows us to estimate correc- tions to the eigenvalues (quasienergies) in powers of the strength of the time-periodic perturbation V (t). Within this formalism the first- and second-order quasienergy corrections to any nondegenerate level are given by (cid:104)ψ0(t)V (t)ψ0(t)(cid:105)dt, (cid:90) T (cid:12)(cid:12)(cid:12) 1 (cid:82) T 0 (cid:104)β0(t)V (t)ψ0(t)(cid:105)dt T 0 ψ − ε0 ε0 β ε1 ψ = 1 T (cid:88) β(cid:54)=ψ ε2 ψ = (cid:12)(cid:12)(cid:12)2 (5) , (6) ψ and ε0 where ψ0(t)(cid:105),β0(t)(cid:105) are unperturbed steady modes as- sociated with quasienergies ε0 β, V (t) is a time- periodic perturbation, and T is a driving period. The sum runs over all steady modes β0(t)(cid:105) differing from the mode under consideration ψ0(t)(cid:105). The corrections above are given by the same expressions as in stationary pertur- bation theory but modified in accordance with the Flo- quet formalism, with matrix elements of operators being replaced by their time averages over the period T 9. Let us now consider a 1D topologically nontriv- ial system described by an unperturbed Hamiltonian H0(t). While we are here primarily interested in time- independent nonperturbed systems where H0(t) = H0, for now we keep the time argument in the Hamiltonian which allows us to discuss both cases on the same foot- ing: driven and undriven nonperturbed systems. We assume that the system is chirally symmetric, implying that Eq. (3), ΓH0(t)Γ = −H0(−t), is satisfied. For sim- plicity we suppress the reference time t0 in all formulas FIG. 5: Scaling of maximum zero-quasienergy shifts with maximum disorder amplitude for time-independent (pan- els with subindex 1) and harmonically driven (panels with subindex 2) SSH chains with γ1 = 0.15 Ω and γ2 = 0.5 Ω. The chains consist of 80 sites and in the driven case the harmonic modulation was fixed to v(t) = 0.4 cos(Ωt). The added time- periodic boundary perturbation, extending over 20 sites from one of the edges, was implemented as a spatial disordering of two independent parameters: a) γ1,j cos(Ωt) + ∆j cos(2Ωt); b) γ2,j cos(Ωt) + γ1,j sin(2Ωt); c) ∆j sin(Ωt) + γ2,j sin(2Ωt) with j = 1, 2, ..., 10. In red color we plot the largest zero- quasienergy shift ε[Ω] maximized over 100 disorder realiza- tions versus the largest allowed disorder site-amplitude, de- noted by V [Ω]. The blue curve represents smoothed data obtained by replacing every 20 points by their average. but keep in mind its presence whenever relevant. Equa- tion (3) restricts the unperturbed steady modes to come β0(t)(cid:105) is a steady mode in symmetry-bounded pairs: β if and only if β0(−t)(cid:105) is also a steady with quasienergy ε0 mode but with quasienergy −ε0 β. Also, it is assumed that in the thermodynamic limit the zero-quasienergy level is nondegenerate at each of the boundaries (with the zero-quasienergy boundary mode at the left edge hav- ing vanishingly small overlap with the zero-quasienergy boundary mode at the right edge), and therefore we may apply nondegenerate perturbation theory separately for each of the boundary modes. Without loss of general- ity we focus on the symmetry-protected mode satisfy- ing ψ0(t)(cid:105) = Γψ0(−t)(cid:105). This is the time-dependent analog of the relation ψ0(0)(cid:105) = Γψ0(0)(cid:105) discussed in Appendix B. Here, we focus on 1D chiral systems with V[+] "[+]"[+]"[+]V[+] "1=3[+1=3]V[+] "1=3[+1=3]V[+] "1=3[+1=3]V[+] 0:71100:701a1)c1)b1)a2)c2)0101b2)000000:050:050:20:70:040:30:3only one localized state per edge but our approach can be straightforwardly generalized to the degenerate case, leading to the same result. We are interested in the leading-order correction to the zero-quasienergy level ε0 ψ = 0 associated with the state ψ0(t)(cid:105) = Γψ0(−t)(cid:105) under influence of the time-periodic perturbations V (t) introduced in Eq. (4). According to Eq. (5), the first-order correction is generally nonzero for a driven state ψ0(t)(cid:105); however, it does vanish in the case when ψ0(t)(cid:105) = ψ0(cid:105) is a stationary state, i.e., an eigen- state of a time-independent system [H0(t) = H0]. This is so because of the integration over the period T . This result is not surprising because the first-order correc- tion represents energy-conserving transitions disallowed in time-independent systems by requiring the perturba- tive driving to have zero time average. Given that the first-order correction vanishes identi- cally when the unperturbed system is time-independent, we now consider the second-order correction (6) for this case. The unperturbed modes β0(t)(cid:105) are given now by β0(t)(cid:105) = einΩtβ0(cid:105), with n ∈ N and β0(cid:105) being eigen- states of the static Hamiltonian. By this, we can split the sum in (6) into a sum over quasienergy phases einΩt and eigenstates β0(cid:105), with ε0 β + nΩ. By using that ε0 ψ = 0, we thus obtain (cid:82)(cid:82) (cid:104)β0V (t)ψ0(cid:105)(cid:104)ψ0V (t(cid:48))β0(cid:105)einΩ(t−t(cid:48)) β → ε0 −ε0 β − nΩ (cid:104)β0V (−n)ψ0(cid:105)(cid:104)ψ0V (n)β0(cid:105) −ε0 β − nΩ , (7) (cid:88) (cid:88) β(cid:54)=ψ, n β(cid:54)=ψ, n ε2 ψ = = T (cid:82) T 0 e−inΩtV (t)dt. where t and t(cid:48) are both integrated over one period T and then time-averaged (divided by T ). Also, in the second line we have introduced the Fourier components V (n) ≡ 1 It is easy to verify that these satisfy the relation V (±n) = e±iφn Vn/2 in accor- dance with the assumed form of perturbations, Eq. (4). Together with the property ΓVnΓ = ±Vn and the chi- ral symmetry of the unperturbed Hamiltonian, implying that the unperturbed eigenmodes β0(cid:105) with ε0 β always come paired to β0 β, this rela- tion allows us to derive Γ(cid:105) ≡ Γβ0(cid:105) with ε0 ≡ −ε0 βΓ ε2 ψ = β(cid:54)=ψ (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) β(cid:54)=ψ n n = = = 0. (cid:104)β0V (−n)ψ0(cid:105)(cid:104)ψ0V (n)β0(cid:105) −ε0 β − nΩ (cid:104)β0ΓΓV (−n)ΓΓψ0(cid:105)(cid:104)ψ0ΓΓV (n)ΓΓβ0(cid:105) −ε0 β − nΩ ΓV (−n)ψ0(cid:105)(cid:104)ψ0V (n)β0 (cid:104)β0 Γ(cid:105) (8) β(cid:54)=ψ n ε0 βΓ + nΩ 6 The correction ε2 ψ vanishes because the first and last expressions in Eq. (8) are the same but with opposite signs (only the terms in the summations are ordered dif- ferently). As follows from the derivation, the vanishing of the second-order correction for this class of pertur- bations crucially hinges on the chiral symmetry of the unperturbed Hamiltonian. In Fig. 5 we numerically verify the scaling of the quasienergy shifts when perturbed by various driven dis- orders, again employing the SSH model. To test our pre- diction we choose to disorder two independent parame- ters and drive the corresponding perturbations periodi- cally in time by using a superposition of a first and second harmonic. Such perturbations may be less accessible ex- perimentally but represent well the class of perturbations assumed in the calculation above. In agreement with our perturbative prediction, in the undriven case the leading- order scaling is indeed only cubic in the disorder strength (see Fig. 5). The deviations from c

1704.07361

2

1704

Strongly directional scattering from dielectric nanowires

It has been experimentally demonstrated only recently that a simultaneous excitation of interfering electric and magnetic resonances can lead to uni-directional scattering of visible light in zero-dimensional dielectric nanoparticles. We show both theoretically and experimentally, that strongly anisotropic scattering also occurs in individual dielectric nanowires. The effect occurs even under either pure transverse electric or pure transverse magnetic polarized normal illumination. This allows for instance to toggle the scattering direction by a simple rotation of the incident polarization. Finally, we demonstrate that directional scattering is not limited to cylindrical cross-sections, but can be further tailored by varying the shape of the nanowires.

cond-mat.mes-hall

cond-mat

Strongly directional scattering from dielectric nanowires Peter R. Wiecha,1, ∗ Aurélien Cuche,1 Arnaud Arbouet,1 Christian Girard,1 Gérard Colas des Francs,2 Aurélie Lecestre,3 Guilhem Larrieu,3 Frank Fournel,4 Vincent Larrey,4 Thierry Baron,5 and Vincent Paillard1, † 1CEMES-CNRS, Université de Toulouse, CNRS, UPS, Toulouse, France 2ICB, UMR 6303 CNRS - Université Bourgogne-Franche Comté, Dijon, France 3LAAS-CNRS, Université de Toulouse, CNRS, INP, Toulouse, France 4CEA-LETI/MINATEC, CEA, Grenoble, France 5CNRS, LTM, Université Grenoble Alpes, Grenoble, France 7 1 0 2 l u J 3 1 ] l l a h - s e m . t a m - d n o c [ 2 v 1 6 3 7 0 . 4 0 7 1 : v i X r a It has been experimentally demonstrated only recently that a simultaneous excitation of interfer- ing electric and magnetic resonances can lead to uni-directional scattering of visible light in zero- dimensional dielectric nanoparticles. We show both theoretically and experimentally, that strongly anisotropic scattering also occurs in individual dielectric nanowires. The effect occurs even under either pure transverse electric or pure transverse magnetic polarized normal illumination. This al- lows for instance to toggle the scattering direction by a simple rotation of the incident polarization. Finally, we demonstrate that directional scattering is not limited to cylindrical cross-sections, but can be further tailored by varying the shape of the nanowires. The search for ways to control light at subwavelength dimensions has increasingly attracted the interest of re- searchers for about the last two decades. Due to their strong polarizability and tunable plasmon resonances, metallic nanostructures are particularly suitable for the nanoscale manipulation of light – especially at visible frequencies.1 However, such plasmonic structures suffer from certain drawbacks like strong dissipation associated to the large imaginary part of the dielectric function in metals. Recently, dielectric nanostructures from high-index materials have proven to offer a promising alternative platform with far lower losses.2 Like in plasmonics, it is possible to tune optical resonances from the near ultra- violet to the near infrared, yet with almost no dissipative losses. At these resonances, which can be of both elec- tric or magnetic nature, strong local field enhancements3 and intense scattering4 occur, tunable via the mate- rial and the geometry of the nanostructure. Promi- nent dielectric materials are, among others, silicon, ger- manium or III-V compound semiconductors with indi- rect band-gap.5,6 Conventional geometries include spher- ical nanoparticles4 or nanowires (NWs),7,8 but also more complex dielectric nanostructures9. Dielectric optical an- tennas are promising candidates for applications in field- enhanced spectroscopy,10–13 imaging,14 to enhance and control nonlinear effects15–17 or to increase the efficiency in photovoltaics18. A peculiarity of dielectric particles is the possibil- ity to simultaneously obtain a strong electric and mag- netic response using very simple geometries.3,4,19,20 Re- cently, it has been independently shown by two research groups,21,22 that exclusive forward (FW) or backward (BW) scattering, predicted by Kerker et al. in 1983 for hypothetical magneto-dielectric particles,23 can be real- ∗ e-mail : peter.wiecha@cemes.fr † e-mail : vincent.paillard@cemes.fr ized in the visible spectral range using dielectric nanopar- ticles. Kerker et al. described two possible configura- tions, called the Kerker conditions. At the first Kerker condition zero backward scattering occurs for equal elec- tric permittivity and magnetic permeability (
r = µr). The second Kerker condition predicts zero forward scat- tering in small spherical particles when the first order magnetic and electric Mie coefficients are of equal abso- lute value and with opposite sign (a1 = −b1). In con- trast to particularly designed metamaterials,24 the mag- netic permeability µr is unitary in dielectric nanoparti- cles. Nevertheless, simultaneously occurring electric and magnetic resonances can de-facto fulfill the first Kerker condition.25,26 While the Kerker conditions were origi- nally derived for spherical particles, it has been shown that the conditions are a result of a cylindrical symme- try and therefore can be generalized accordingly.27 Recent publications have confirmed the possibility to obtain optimum FW scattering also from elon- gated structures such as nanopillars,21 spheroids28 or even from cuboidal dielectric particles8. Other dielec- tric particles on which directional scattering was in- vestigated include nanodiscs for FW/BW directional metasurfaces,29,30 patch antennas,31 V-shaped structures for multi-directional color routing,32 or asymmetric hol- low nanodiscs for bi-anisotropic scattering33, all of them based on the interplay between electric and magnetic modes. Control over the directional scattering can also be obtained using arrangements of nanoparticles like dimers34–36 or via hybrid metal/dielectric nano- structures37. Even directional shaping of nonlinear emis- sion has been demonstrated: The radiation pattern of the third harmonic generation from silicon dimers can be controlled via the geometry of the structure.38 In a plasmonic-dielectric hybrid structure, the interplay of electric and magnetic resonances in dielectric TiO2 spheres was used to impose an uni-directional radiation pattern on the second harmonic generation from a plas- monic driver element.39 Compared to zero-dimensional (0D) particles of deeply 2 leading to the occurrence of directional scattering. We then compare experimental results from individual, sin- gle crystal silicon nanowires (SiNWs) of cylindrical and rectangular cross-sections. In both cases spectral zones of strongly anisotropic FW/BW scattering ratios can be identified and we find that asymmetric wire geometries allow even further tailoring of directional scattering and offer the incident angle of the illumination as a supple- mentary free parameter. We confront our experimental results with simulations using the Green dyadic method (GDM), yielding a very good agreement. I. DIRECTIONAL SCATTERING FROM CYLINDRICAL NANOWIRES Mie theory can be applied to infinitely long cylinders by expanding the fields in vector cylindrical harmonics. The Mie scattering coefficients ai and bi of the expansion can be regarded as weights for corresponding electric and magnetic multipole moments, representing the response of the wire to an external illumination. Under normal incidence, the Mie S-matrix, connecting the incident (Ei) and the asymptotic, scattered field (Es) writes1 (cid:20)T1 0 (cid:21)(cid:20)Ei,TM (cid:21) 0 T2 Ei,TE (1) (cid:20)Es,TM (cid:21) Es,TE (cid:114) 2 = ei3π/4 eikR πkR with the wavenumber k = 2π/λ, the distance R to the cylinder axis and ∞(cid:88) ∞(cid:88) n=1 T1 = b0 + 2 T2 = a0 + 2 bn cos(nϕ) an cos(nϕ). (2) n=1 ϕ is the scattering angle with respect to the incident wave vector and ϕ = 0 corresponds to the forward scattering direction. As can be seen from Eq. (1), under normal incidence the transverse magnetic (TM) and transverse electric (TE) polarized components of the scattered fields are proportional to the S-matrix components T1, respectively T2. Therefore, according to Eqs. (2) scattering from a TE polarized normally incident plane wave (E ⊥ NW axis) is only due to the "electric" multipole contributions ai. On the other hand, a TM polarized illumination (E (cid:107) NW axis) induces scattering exclusively via the "magnetic" Mie terms bi. The expressions "electric" and "magnetic" refer to the fact that in the TE and TM case, the mag- netic, respectively electric field components are zero in the scattering plane. Let us now consider the case of sufficiently small nanowires, where only the first two orders of the Mie ex- pansion contribute significantly to scattering. For SiNWs in the visible spectral range, this assumption is a good approximation for diameters up to at least D = 100 nm FIG. 1. Average electric (red lines) and magnetic (blue lines) field intensity enhancement inside a silicon nanowire of diam- eter D = 100 nm as function of wavelength for a) a TE and b) a TM polarized incident plane wave, calculated using Mie theory. Normalized to the illumination field intensity. For comparison, the FW/BW scattering ratio in the far-field is shown as dashed green line (right axis ticks). c) and d) show the internal field intensity distributions at λ = 550 nm for TE and TM polarization, respectively (left subplot: electric, right: magnetic field). D = 100 nm, plane wave incident from the top. sub-wavelength size in all directions (like nano-spheres), one-dimensional (1D) nanowires have a strongly polar- ization dependent optical response, which offers an addi- tional degree of freedom, supplementary to the choice of material and modifications of the nano-structure's geometry. Despite this opportunity, research on di- rectional scattering from dielectric nanowires is still scarce. As for scattering, it has been shown theoretically that multi-layer dielectric40 or plasmonic-dielectric core- shell41 nanowires can be rendered invisible by tailoring destructive interference between different order modes or electric and toroidal dipole modes, respectively. Such entities might be useful for cloaking applications, for in- stance to create "invisible" electric contacts or circuits.42 Dense arrays of vertical dielectric NWs on the other hand, possess a strong light extinction due to multiple scatter- ing inside the nanowire assembly, which can be used as an anti-reflection coating for photovoltaics.43,44 At the level of an individual nanowire, angle-dependent absorp- tion can be tailored via the incident angle through the interplay between Mie-type and guided modes.45 Also, the directionality of photoluminescence from individual III-V nanowires can be tailored through the supported Mie and guided modes46–48. In contrast to these former works on complex geome- tries or effects, in this paper we study the directional scattering of light from individual normally illuminated dielectric nanowires. Using Mie theory we analyze in a first step the interplay of different order Mie modes, 400500600700800900wavelength(nm)010203040avg.fieldenhancementa)TEE2×10H24005006007008009001000wavelength(nm)b)TM10−1100101FW/BWscat.ratioFW/BWFW/BW=1c)0.00.51.01.5E2/E02TE,λ=550nm0255075H2/H02d)0.02.55.07.5E2/E02TM,λ=550nm0255075H2/H023 excitation, revealing zones of strongly directional scatter- ing. For the dispersion of silicon we use tabulated data from Ref. 6 throughout this paper. In the SI Figs S3-S4 we show the first two order Mie coefficients for SiNWs as function of the wavelength and the nanowire diame- ter and find that zones where the conditions in Eqs. (3) are approached do exist at several wavelength / diameter combinations for normally illuminated cylindrical silicon nanowires. We want to note that a similar derivation has been per- formed for core / shell metal / dielectric nanowires, where the plasmonic core was introduced in the NW to shift the electric and magnetic dipolar modes to spectrally over- lap, leading to directional scattering effects.51 In contrast to this former theoretical work, we study interference be- tween different order multipole contributions in homoge- neous dielectric nanowires, leading to strongly directional scattering even in such very simple systems. It is possible to expand the electromagnetic fields in- side the cylinder in the same way as the scattered fields of Eq. (1). For details, we refer to the textbook of Bohren and Huffmann (chapters 4 and 8.4).1 In figure 1 the av- erage internal field intensity enhancement is shown for a SiNW with diameter D = 100 nm with a TE (a) and TM (b) polarized incident plane wave, respectively. In- terestingly, at the resonant wavelengths we observe not only high electric field intensities (red lines) but also a very strong enhancement of the magnetic field (blue lines), regardless of the incident polarization orientation. For both polarizations, the magnetic field increases even significantly stronger compared to the electric field in- tensity. We remark that this observation is in agree- ment with recent results from dielectric cylinders in the GHz regime.52 We conclude that a simultaneous exci- tation of strong electric and magnetic fields occurs in dielectric, non-magnetic (i.e. µr = 1) nanowires, even under pure TE or TM polarized illumination and normal incidence. Hence, in analogy to the findings of Kerker for the case of spherical dielectric nanoparticles, et al. the observed directionality (dashed green lines in Fig. 1) can be interpreted as a result of the interference between "effective" electric and magnetic modes. In particular, no directional scattering is obtained at the non-degenerate, fundamental TM01 mode, where only the internal electric field shows a resonant enhancement while the magnetic field intensity follows a flat line beyond λ (cid:38) 700 nm (see Fig. 1b). For illustration the electric and magnetic field intensity patterns inside the NW cross section are shown for an incident wavelength λ = 550 nm in Fig. 1(c-d). FIG. 2. Illustration of the experimental setup for (a) forward scattering and (b) backward scattering measurements. (c) sketch of the incident polarization configurations. The electric field is orientated either perpendicular (TE) or parallel (TM) to the NW axis. (d) SEM image of a cylindrical SiNW, de- posited on a silicon substrate. The NWs for the spectroscopy experiments were deposited on a glass substrate, which could not be imaged in SEM due to charging. Scale bar is 100 nm. (e) SEM images of rectangular SiNWs. SEM images are ob- tained from a second sample on silicon on insulator (SOI) substrate. The spectroscopy sample is fabricated on trans- parent, but insulating SOQ substrate. Scale bars are 500 nm. Zoomed insets on the right are 200 × 200 nm2 (white dashed squares indicate zoom area). (see also supporting informations (SI), Figs. S1-S2). By setting ϕ = 0 or ϕ = π in equations (2) we obtain the con- ditions for zero scattering in forward (FW), respectively backward (BW) direction. For TM polarized illumina- tion we find: ≈ b0 + 2b1 = 0 ≈ b0 − 2b1 = 0 for pure BW scat. for pure FW scat. (3) (cid:12)(cid:12)(cid:12)ϕ=0 (cid:12)(cid:12)(cid:12)ϕ=π T1 T1 The same conditions hold for T2 and a0,1 in the TE con- figuration. Hence if the conditions in Eqs. (3) are met, it is possible to obtain uni-directional scattering from 1D dielectric nanowires, which is the result of interference between the simultaneously excited first two multipole contributions. In figure 1(a-b), the FW/BW scattering ratio is shown (dashed green lines) for a SiNW of di- ameter D = 100 nm for (a) TE and (b) TM polarized II. SPECTROSCOPY ON CYLINDRICAL SILICON NANOWIRES In order to compare the forward and backward scatter- ing of a normally incident plane wave on silicon nanowires (SiNWs) we perform standard darkfield microscopy ei- ther in reflection (backward scattering, "BW") or in trans- a) FW scatteringDF ObjectiveDF condenserb) BW scatteringc) incidentpolarizationsTE (E ∥ x)TM (E ∥ y)d) SEMcylindrical NWe) SEMrectangular NWsW=70nmW=220nmW=430nmGlass substrateindividualSiNWxyzzoom4 FIG. 3. Experimental (solid lines) and simulated (dashed lines) FW/BW scattering spectra (red/blue) from cylindrical NWs of different diameters. (a) D ≈ 50 nm, (b) D ≈ 80 nm and (c) D ≈ 100 nm. Diameters are estimates by comparison to simulations. Top row and bottom row show the case of TE and TM polarized incident plane waves, respectively. (d) calculated nearfield distributions (top row: TE, center row: TM incidence) and farfield patterns (bottom row; TE/TM: red/blue) for a nanowire of D = 100 nm diameter at selected wavelengths, indicated by dashed vertical lines in (c). Plane wave incident from the top. Scale bar is 50 nm. FIG. 4. Forward (top row) and backward (center row, multiplied by ×2 for better contrast) scattering intensity from rectangular silicon nanowires as function of wavelength and NW width. Data is normalized to the global maximum intensity in FW and BW spectra. FW/BW ratios are shown in the bottom row on a logarithmic color scale. (a-b) TE, (c-d) TM polarized illumination. (a) and (c) show measured dark field spectra, (b) and (d) corresponding GDM simulations. Fixed NW height and length (H = 90 nm, L = 7 µm). mission geometry (forward scattering, "FW"). The mea- surement geometry is schematically shown in Fig. 2(a) and (b) for FW and BW scattering, respectively. For details on the measurement technique, see Methods. We compare our experimental results to 2D simulations (in XZ-plane, assuming infinitely long structures along Y ) by the Green dyadic method. Likewise, the GDM simu- lation technique is explained in the methods section. In the SI Figs. S5-S10, the accuracy of the method is veri- fied by comparing GDM simulations to Mie theory and FDTD simulations. We measure the scattering from cylindrical SiNWs of 0123scatteringintensity(a.u.)a)NW1-D=50nmFWBWexp.sim.50060070080002468b)NW2-D=80nm500600700800wavelength(nm)c)TENW3-D=100nm500600700800TMd)TEλ=480.0nmTMfarfield×2.5λ=550.0nm180◦(BW)0◦(FW)λ=750.0nm×5TETME2/E02max.min.b)simulation×25006007008000.00.20.40.60.81.0scatteringintensity(a.u.)10−1100101FW/BWratio100150200250300350400450TEa)experimentforward100150200250300350400450width(nm)×2backward500600700800wavelength(nm)100150200250300350400450FW/BWratiod)simulation×25006007008000.00.20.40.60.81.0scatteringintensity(a.u.)10−1100101FW/BWratio100150200250300350400450TMc)experimentforward100150200250300350400450width(nm)×2backward500600700800wavelength(nm)100150200250300350400450FW/BWratio5 FIG. 5. (a-d) selected scattering spectra from rectangular silicon nanowires. Experimental FW (red) and BW (blue) scattered intensity (solid lines) is compared to GDM simulations (dashed lines) for TE (top row) and TM (bottom row) polarized plane wave illumination. NW height and length are fixed (H = 90 nm, L = 7 µm), widths W are (a) 60 nm, (b) 120 nm, (c) 180 nm and (d) 330 nm, referring to the simulation parameters. Intensities in (a-b) are multiplied by ×2 for better visibility. Same data as in figure 4. Panels (e-i) show selected electric field intensities in the cross section of rectangular SiNWs for (e) W = 60 nm, λ = 750 nm; (f) W = 120 nm, λ = 580 nm; (g) W = 180 nm, λ = 580 nm; (h) W = 330 nm, λ = 560 nm and (i) W = 330 nm, λ = 650 nm. Plane wave illumination from the top, polarized perpendicularly (TE, top row) or along the NW axis (TM, bottom row). Farfield radiation patterns are shown in the bottom row (TE: red, TM: blue), adjusted by a variable scaling factor for better visibility (factor indicated on the upper left of each plot). Scale bar in (e) is 50 nm, same scale for all field plots. Vertical dashed lines in (a-d) indicate the spectral positions of the field-plots in (e-i). diameters between D ≈ 50 nm and D ≈ 100 nm, epi- taxially grown by the vapor-liquid-solid (VLS) technique and drop-coated on a transparent glass substrate. The NW size-dispersion was obtained via scanning electron microscopy (SEM) from a second sample consisting of the same SiNWs, drop-coated on a silicon substrate (see Fig. 2d). For details, see the methods section. Results of the scattering experiments are shown in Fig. 3(a-c) for TE (top row) and TM (bottom row) polarized illumina- tion. The given NW diameters are estimates, obtained by comparison with simulations. The very weak (TE), respectively omnidirectional (TM) scattering in case of the the smallest nanowire (Fig. 3a, NW1: D ≈ 50 nm) is in perfect agreement with the GDM simulations and confirms the assumption of a purely dipolar response if only the non-degenerate TM01 mode is excited (see also field plot for λ = 750 nm in figure 3d on the right). As expected, with increas- ing NW diameter (figure 3b-c), we observe a simultane- ous red-shift of the FW and BW scattering peaks. For the largest SiNWs (Fig. 3c, D ≈ 100 nm), in the case of TE polarized illumination mainly FW scattering oc- curs in a limited spectral range (550 nm (cid:46) λ (cid:46) 700 nm), before BW scattering takes over at shorter wavelengths (λ (cid:46) 550 nm). Note that around 480 nm, we obtain the possibility to invert the main scattering direction by sim- ply flipping the polarization from TM to TE (see Fig. 3d, bottom left or also SI, Fig. S5). By changing the NW diameter this spectral zone can be also tuned to other wavelengths (see SI, Figs. S6-S7). At longer wavelengths (λ (cid:38) 700 nm), only the TM01 mode exists, leading to very weak overall scattering under TE incidence and to the above mentioned, omnidirectional radiation pattern in the TM geometry. Interestingly, while TE excitation can induce uni-directional BW scattering, under TM polar- ization BW scattering is generally very weak and mainly FW scattering occurs as soon as higher order contribu- tions are excited together with the TM01 mode. In summary, we note that although we do observe BW scattering (mainly under TE polarization) it mostly re- mains weak compared to the FW scattered light. This has been also observed in the case of 0D-particles22,36 and can be explained with the finding that the second Kerker's condition (describing BW scattering) is contra- dicting the optical theorem and thus cannot be exactly fulfilled.53 Furthermore it is difficult to even closely meet the second condition.27 In the case of TM polarized il- lumination, the FW/BW scattering ratio is even almost exclusively (cid:38) 1 (see also SI, Figs. S5-S7 for more simula- tions on cylindrical NWs). III. ASYMMETRIC, RECTANGULAR SILICON NANOWIRES In a second step we analyze what happens if the cylin- drical symmetry of the nanowire cross section is broken. We therefore fabricate SiNWs of rectangular section by electron beam lithography (EBL) and subsequent dry- etching on a silicon-on-quartz (SOQ) substrate. For de- tails on the fabrication process, see Methods. We want to point out a great advantage of our top-down approach on SOQ: The possibility to create silicon nanostruc- tures of arbitrary shape by EBL on a transparent sub- strate from single crystalline silicon. Using crystalline material rather than polycrystalline or amorphous sili- con guarantees that the high refractive index of silicon is obtained and facilitates furthermore the modeling, since accurate tabulated data for the dispersion is available in literature.6 This gives us the opportunity for a system- atic analysis of the scattering from rectangular SiNWs of variable width. The height of the structures however remains constant, defined by the thickness of the silicon layer on the SOQ substrate (H = 90 nm in our case). We choose SiNWs with length long compared to the fo- cal spot of the illuminating optics (all data acquired on NWs with L = 7 µm) in order to obtain a purely Mie-like response.54 SEM images of selected rectangular SiNWs are shown in figure 2e. The images are obtained from a second sample, fabricated with the exact same process parameters but on a non-insulating silicon-on-insulator (SOI) substrate. For the purpose of better illustration NWs of L = 2 µm length are shown. We point out that all NWs have excellent surface properties, low roughness and steep flanks. The results of our systematic FW/BW scattering mea- 6 surements are shown in figure 4. Spectra for TE po- larized illumination are shown in (a-b), TM spectra in (c-d). The comparison of experiments ( (a) and (c) ) with GDM simulations ( (b) and (d) ) show a very good qualitative and quantitative agreement. Having a look at the FW/BW ratios (Fig. 4, bottom row) we observe that uni-directional scattering occurs mostly towards the FW direction. Spectrally relatively narrow peaks of strong forward scattering can be identified for both, TE and TM polarized illumination. On the other hand, a zone of significant BW scattering is observed only in the case of TE polarization, similar to our observations on cylindri- cal SiNWs. Selected spectra from Fig. 4, as well as several near- field intensity plots are shown in figure 5. The spectra in (a-d) confirm the very good agreement between experi- ment (solid lines) and simulations (dashed lines). For the smallest nanowire widths as well as for long wavelengths, we find again an omnidirectional scattering, correspond- ing to a purely dipolar response (Fig. 5e). In analogy to cylindrical NWs, this first order resonance can only be ex- cited in the TM geometry, TE polarized illumination re- sults in a very weak scattering in the corresponding spec- tral range. While small nanowires have spectrally broad optical resonances (figure 5a-b), narrower higher order resonances appear with increasing NW width (figure 5c- d). In contrast to the omnidirectional TM scattering in the smallest wires, the first occurrences of strong forward scattering seem to be the result of interference between a sharp quadrupolar resonance and a broad dipolar contri- bution. We deduce this conclusion from the similarity in field intensity distributions in cylindrical and rectangular NWs (compare Figs. 3d and 5e,f,i). Under TM excitation, we also observe branches of a different kind of Fano-like resonances where the FW/BW ratio is almost unity (see Fig. 4c-d and Fig. 5c-d, bot- tom row). In a narrow spectral window the otherwise strong forward scattering is suddenly suppressed while BW scattering increases. Resonances with such field profiles are observed neither under TE polarization, nor in symmetric SiNWs (see SI Figs. S11-S15, where scat- tering from rectangular wires is compared to different symmetric geometries). We attribute these sharp fea- tures to horizontal guided modes along the SiNW width (X-direction), the wire side facets acting as the Fabry- Perot cavity mirrors. To verify this assumption we as- sess the effective index for the supported guided mode(s) assuming a 1D waveguide of infinite extensions in X- and Y -direction. For the case of TM polarized illumina- tion at λ = 580 nm, the corresponding first order guided mode along the width of the NW has an effective index of neff = 3.45 (nSi ≈ 4.0). The first and second order standing wave patters in figure 5g and h match perfectly with the nanowire widths of 180 nm, respectively 330 nm (at λ = 560 nm for the latter). For more details on the supported guided modes, see section V in the SI. It is possible to exploit these Fabry-Perot like modes in order to toggle between FW and omni-directional scattering 7 FIG. 6. (a) sketch illustrating the oblique incident illumination. (b-g) Electric field intensities (individually normalized) in the cross section of rectangular SiNWs of width W = 250 nm (height H = 90 nm) for increasing incident angles from normal incidence in (b) to ϑ = 72◦ in (g). The NW cross section is indicated by dashed white lines. Left: TE, center: TM polarized plane wave of wavelength λ0 = 600 nm. Right: In the farfield panels, the corresponding angle of incidence is depicted by a black arrow. (h-j) show corresponding farfield spectra for TE (left) and TM (right) polarization. (h) forward scattering (towards lower hemisphere), (i) backward scattering (to upper hemisphere, multiplied by 2) and (j) FW/BW scattering ratio for normal incidence (black lines), averaged incident angles from 0◦ to 72◦ (magenta lines) and 72◦ (green dashed lines). 72◦ corresponds to the maximum angle of the darkfield illumination in our experiments. farfield pattern for TE (red) and TM (blue). within a narrow spectral window – simply by switching the incident polarization from TE to TM (see e.g. around 560 nm in figure 5d). Figs. 5(g-h) (TM: center row) correspond to near-field intensity distributions inside the SiNW at the first two orders of these TM-excited Fabry-Perot resonances. The field patterns inside the NW cross section correspond in- deed to standing wave patterns of guided modes, which eventually lead to an almost equal FW and BW scatter- ing in a narrow spectral region (see Fig. 5(c-d) bottom, 550 nm (cid:46) λ (cid:46) 600 nm). We remark that the assump- tion of spectrally sharp, guided modes between the NW side facets is also in agreement with former studies on scattering from rectangular dielectric NWs.55,56 Finally, we want to assess the influence of non-normal illumination on FW and BW scattering from rectangular SiNWs. For spectroscopy on single nanowires, we need to use a high-NA microscope objective, which results in a large dispersion of incident angles, with angles up to ≈ 72◦ in our setups (NA 0.95, see also Methods). In fig- ure 6, we show simulated field intensity distributions in the scattering plane and scattering spectra for a rectan- gular SiNW of 250 nm width (height 90 nm) for differ- ent incident angles (depicted in figure 6a). To calculate the spectra, while the angle of incidence is varied, we fix the integration solid angle to the upper (BW) and lower (FW) hemisphere, corresponding to a fixed position of the collecting microscope objective. The wavelength in (b-g) is λ0 = 600 nm, the electric field is either polar- ized parallel to the scattering plane (TE) or along the NW axis (TM). (b-g) show the field intensity and an- gular farfield intensity patterns in the NW section for increasing incident angles. We observe, that while the original field distribution can still be recognized for an- gles up to ≈ 36◦, at high angles the nearfield distribution as well as the farfield radiation pattern are completely xyzkrectangularSiNWTETMdifferent. In (h-j) we show FW scattering (h), BW scat- tering (i) and FW/BW ratio (j) spectra for the same rectangular SiNW. The black lines correspond to normal incidence and the green dashed lines to an incident angle of 72◦. The magenta lines denote the average scatter- ing for illumination angles from 0◦ to 72◦, hence an ap- proximation to our experimental conditions. The spectra for an illumination angle corresponding to NA 0.95 show often even completely opposite trends compared to nor- mal incidence. In (j) for instance regions exist where the directionality flips from FW under normal incidence to BW for an angle of 72◦. On the other hand, the spectra calculated using the average illumination angles follow the trends of normal incidence. We conclude, that a normally incident plane wave gives a valid first-order approximation even for rectangular NWs illuminated by tightly focused beams. This is also supported by the good agreement between normal-incident simulations and the experimental data (see figure 4). In the SI Figs. S17- S19, we show nearfield intensity maps for different ex- citation wavelengths and nanowire geometries as well as further spectra for comparison with figure 5, confirming the trends discussed above. In the supporting informa- tions, we also discuss a selected example, where the nor- mal incidence approximation breaks down. IV. CONCLUSIONS In conclusion, we demonstrated strongly directional light scattering from symmetric an asymmetric dielectric nanowires at optical frequencies. We derived the con- ditions for exclusive forward or backward scattering in cylindrical nanowires from Mie theory and showed both theoretically and experimentally that anisotropic scat- tering can be obtained from individual silicon nanowires. Unlike 0D-structures, directionality is a result of inter- ference between different orders of either transverse elec- tric or -magnetic modes. Also, compared to 0D nano- particles the incident light polarization offers an addi- tional degree of freedom in nanowires – apart from mate- rial and geometry variations. Besides being able to adjust the scattering intensity, the directionality and the spec- tral width of the resonances by varying the NW width, it is for example possible to switch between FW and BW scattering simply by rotating the polarization of the in- cident light. Finally we showed that by breaking the symmetry of cylindrical nanowires and using rectangular cross sections instead, Fano-like resonances occur, lead- ing to spectrally very sharp features with unity FW/BW scattering ratio, due to guided modes along the NW width. Again, this isotropic scattering can be triggered by switching the incident polarization between TE and TM. We demonstrated that asymmetric nanowires also allow to tune the scattering response via the angle of incidence. Our results open perspectives for various ap- plications where spectrally tunable directional guiding of light is required at a nano scale, for instance in optical 8 nano-circuits. Furthermore, dielectric nanowires may be used for example for anisotropic cloaking purposes, hence observer-dependent invisibility, in wavelength-selective field-enhanced spectroscopies or for directional non-linear emission. Finally, due to the simplicity of the silicon nanowires and their compatibility with production-ready technology, our results render nanowires very interesting for light harvesting applications in photovoltaics. V. METHODS A. Crystal Growth of Cylindrical SiNWs The cylindrical silicon nanowires were grown in a hot- wall reduced pressure chemical vapor deposition system via the Vapor-Liquid-Solid (VLS) process. Au droplets were used as catalysts, their sizes can be easily controlled and define the diameter of the nanowires. The reac- tor pressure was 4.5 Torr, as precursor for the Si growth silane (SiH4) was used, diluted with hydrogen as carrier gas. More details can be found in reference 57. Subse- quent to the expitaxy, the grown SiNWs were removed from the substrate by careful sonication in isopropyl al- cohol solution. For the scattering experiments the NWs were finally drop-coated onto a transparent glass sub- strate with lithographic markers. A second sample was prepared on a silicon substrate for scanning electron mi- croscopy, on which we determined a dispersion in NW diameters between around 50 nm and 100 nm. B. Lithographic SiNWs by Electron Beam Lithography on Silicon on Quartz Substrates The so-called silicon-on-quartz (SOQ) substrates were fabricated starting from commercially available SOI sub- strates. A first thinning step is realized by thermal oxida- tion and oxide removal in order to adjust the top silicon thicknesses according to specifications. Direct bonding is then performed to join the SOI to fused silica substrates. As silicon and silica present a large CTE mismatch (CTE: Coefficient of thermal expansion), high thermal processes are prohibited to improve the bonding strength and the annealing temperature is therefore limited to 200◦C.58 This implies dedicated surface preparation to achieve high adherence energy at low temperature: SOI are pre- pared with a nitrogen plasma and the fused silica with a chemical mechanical process step. The bonded structure is then thinned by grinding and chemical etching in order to remove the base silicon substrate as well as the buried oxide layer (BOX). The rectangular NWs were fabricated by a top-down technique via electron beam lithography (EBL) and sub- sequent anisotropic plasma etching.59,60 A transparent SOQ wafer with a 90 nm thick Si overlayer served as sub- strate. A RAITH 150 writer at an energy of 30 keV was used for EBL, employing a thin (60nm) negative- tone resist layer (hydrogen silsesquioxane, "HSQ"). Af- ter the EBL step, HSQ was developed by immersion in 25 % tetramethylammonium hydroxide (TMAH) for 1 min. The patterns in the resist were finally written into the SOQ silicon top layer by reactive ion etching (RIE). For the RIE step, a SF6/C4F8 plasma was used. Etching was stopped when the silica layer was reached, follow- ing in-situ analysis. A residual layer of ≈ 20 nm of SiO2 (developed resist) remains on the structures, which has no major impact on the optical properties due to its low index and small thickness. A comparison with a sample where the residual layer was removed using fluorhydric acid showed no significant difference in the spectra, but smaller structures were destroyed due to under-etching. For SEM imaging, a second sample was prepared with identical process parameters on a silicon on insulator (SOI) substrate (Si-Layer: 90 nm, BOX layer: 145 nm) C. Forward/Backward Confocal Darkfield Spectroscopy Forward, "FW": Transmission confocal optical dark- field (DF) microscopy was performed on an inverted op- tical microscope. A broadband white lamp was focused through the sample substrate on the individual SiNWs using a DF condenser (Nikon, NA 0.8 − 0.95). The FW scattered light was collected using a Nikon NA 0.75 mi- croscope objective, spatially selected via a confocal hole and guided through a polarization filter ((cid:107) or ⊥ to the NW axis for TM and TE geometry, respectively). Fi- nally, the light was dispersed by a 150 grooves per mm grating on a highly sensitive CCD (Andor Newton). Backward, "BW": Reflection confocal optical DF ex- periments were carried out on a separate spectrometer (Horiba XploRA). A white lamp was focused on individ- ual SiNWs by a ×50 dark-field objective (NA 0.5, con- denser: NA 0.8 − 0.95). The backscattered light was filtered by a confocal hole and a polarization filter and dispersed by a 300 grooves per mm grating on a highly sensitive CCD (Andor iDus 401). In both setups (FW and BW), for practical reasons we put the polarization filters in the detection path (detec- tion direction normal to the NW axis). Because we limit our investigations to normal illumination (no depolariza- tion) and due to the linearity of the fields, we assume that the polarization filter after the sample is equivalent to a polarized, normal incidence. Normalization: In the FW as well as in the BW scat- tering experiments, the intensity distribution of the re- spective lamp as well as the spectral response of the optical components was accounted for by the following normalization scheme: First, the background noise was substracted (DF measurement on the bare SOQ / glass substrate). Subsequently the data was normalized using the spectra of the respective lamps, measured through the bare substrate ("FW" setup) and via a white refer- 9 ence sample ("BW" setup). Since we measure FW and BW scattering on two entirely different setups, we can- not compare the spectra quantitatively. Therefore we use the scattering from the smallest cylindrical nanowire ("NW1", D ≈ 50 nm, Fig. 3a bottom) for normalization of the data between the two measurement geometries: In sufficiently small NWs illuminated by TM polarized light, exclusively the fundamental dipolar TM01 mode is excited. This results in an omni-directional scattering, corresponding to a dipolar source along the SiNW axis. Thus we can assume that the FW and BW scattered in- tensities under these conditions are of equal strength and normalize all spectra according to this reference. Incident and detection angles: The upper limit for the incident angle from the dark-field condensers (FW and BW both NA = 0.95) is ϕin = sin−1 (NA) ≈ 72◦. (4) The high incident angles can modify the response of the Mie resonances. However, since the illumination is conic, only a fraction of the incident light has actually a large wavevector component parallel to the nanowire . Thus, assuming normally incident plane wave illumi- k(cid:107) nation in the simulations is a valid approximation to our experimental conditions, further justified by the excellent agreement between our simulations and measurements. The upper limits of the detection angles can be calcu- lated by the collecting objectives' numerical apertures: NAFW = 0.75 NABW = 0.5 ⇒ ϕFW ≈ 49◦ ⇒ ϕBW ≈ 30◦. (5) With the maximum incident angle ϕin ≈ 72◦ this eventu- ally leads to the following angular detection ranges (per polar quadrant): 23◦ (cid:46) ϕcol., FW (cid:46) 42◦ (cid:46) ϕcol., BW (cid:46) 120◦ 102◦. (6) The simulated range is 0◦ ≤ ϕcol., sim. ≤ 90◦. Given the excellent agreement between simulations and mea- surements, the deviations between the collection angles seem not to cause a serious problem. We attribute this to the fact that scattering is either omni-directional or else strongly forward/backward directional. In both cases, large angles with respect to the incident wave-vector di- rection contribute only weakly to the total scattering. See also SI, Fig. S16 for a detailed sketch, illustrating the collection angles. D. GDM Simulations Because of the high aspect ratio of the nanowires, we assume infinitely long structures (NW axis along Y ). In this manner we simplify the computation to a two di- mensional (2D) problem. The electric field is calculated using the Green dyadic method (GDM). Practically, we solve the vectorial Lippman Schwinger equation (cid:90) E2D(r(cid:107), ω, ky) = E2D , ω, ky) · χ(ω)E2D(r(cid:48) (cid:107) 0 (r(cid:107), ω, ky) + , ω, ky)dr(cid:48) (cid:107) G2D 0 (r(cid:107), r(cid:48) (cid:107) (7) 0 at the frequency ω, by discretizing the nanostructure us- ing square meshes. E2D and E2D are the incident field and resulting field inside the nanostructure, respectively. is the Green's dyad and χ the electric susceptibility G2D of the structure material (χ =
− 1). ky is the inci- 0 dent wavevector along the NW axis. For normal inci- dence, ky = 0. The integral in Eq. (7) runs over the sur- face of the 2D particle cross section. This self-consistent equation is solved (e.g. using standard LU decomposi- tion) and the field can be computed in the nanowire. Knowing the field inside the nanostructure, application of Eq. (7) permits to compute the field everywhere, both in the near-field and far-field zone. For details we refer the interested reader to Refs. 2 and 3. All simulations are carried out for SiNWs in vacuum (nenv. = 1) to allow direct comparison with Mie theory for the cylindrical wires (see SI, Figs. S5-S10). Given the large index contrast with silicon (nSi ≈ 4 in the vis- ible), this is a good approximation, which is supported also by the excellent agreement between simulations and measurements. 1 Mühlschlegel, P., Eisler, H.-J., Martin, O. J. F., Hecht, B. & Pohl, D. W. Resonant Optical Antennas. Science 308, 1607–1609 (2005). URL http://science.sciencemag.org/ content/308/5728/1607. 2 Kuznetsov, A. I., Miroshnichenko, A. E., Brongersma, M. L., Kivshar, Y. S. & Luk'yanchuk, B. Optically res- onant dielectric nanostructures. Science 354, aag2472 (2016). URL http://science.sciencemag.org/content/ 354/6314/aag2472. 3 Bakker, R. M. et al. Magnetic and Electric Hotspots with Silicon Nanodimers. Nano Letters 15, 2137–2142 (2015). URL http://dx.doi.org/10.1021/acs.nanolett. 5b00128. 4 Kuznetsov, A. I., Miroshnichenko, A. E., Fu, Y. H., Zhang, J. & Luk'yanchuk, B. Magnetic light. Scientific Reports 2, 492 (2012). URL http://www.nature.com/articles/ srep00492. 5 Kallel, H. et al. Tunable enhancement of light absorption and scattering in Si(1-x)Ge(x) nanowires. Physical Review B 86, 085318 (2012). URL http://link.aps.org/doi/10. 1103/PhysRevB.86.085318. 6 Albella, P., Alcaraz de la Osa, R., Moreno, F. & Maier, S. A. Electric and Magnetic Field Enhancement with Ul- tralow Heat Radiation Dielectric Nanoantennas: Consider- ations for Surface-Enhanced Spectroscopies. ACS Photon- ics 1, 524–529 (2014). URL http://dx.doi.org/10.1021/ ph500060s. 7 Cao, L., Fan, P., Barnard, E. S., Brown, A. M. & Brongersma, M. L. Tuning the Color of Silicon Nanos- 10 ACKNOWLEDGMENTS This work was supported by Programme Investisse- ments d'Avenir under the program ANR-11-IDEX-0002- 02, reference ANR-10-LABX-0037-NEXT, by LAAS- CNRS micro and nanotechnologies platform member of the French RENATECH network, by the Région Midi- Pyrénées and by the computing facility center CALMIP of the University of Toulouse under grant P12167. SUPPORTING INFORMATIONS In the supporting informations we show further details on FW/BW scattering from nanowires by Mie theory as well as additional simulations. NOTES We note that a similar, theoretical work has been published during the review of this manuscript, describ- ing scattering phenomena in nanowires with a radially anisotropic refractive index.63 The authors declare no competing financial interest. tructures. Nano Letters 10, 2649–2654 (2010). URL http://dx.doi.org/10.1021/nl1013794. 8 Ee, H.-S., Kang, J.-H., Brongersma, M. L. & Seo, M.-K. Shape-Dependent Light Scattering Properties of Subwave- length Silicon Nanoblocks. Nano Letters 15, 1759–1765 (2015). URL http://dx.doi.org/10.1021/nl504442v. 9 Wiecha, P. R. et al. Evolutionary multi-objective optimization of colour pixels based on dielectric nanoan- tennas. Nature Nanotechnology 12, 163–169 (2017). URL http://www.nature.com/nnano/journal/v12/n2/abs/ nnano.2016.224.html. 00002. 10 Gérard, D. et al. Strong electromagnetic confinement near dielectric microspheres to enhance single-molecule fluorescence. Optics Express 16, 15297 (2008). URL https://www.osapublishing.org/oe/abstract.cfm?uri= oe-16-19-15297. 11 Wells, S. M., Merkulov, I. A., Kravchenko, I. I., Lavrik, N. V. & Sepaniak, M. J. Silicon Nanopillars for Field- Enhanced Surface Spectroscopy. ACS Nano 6, 2948–2959 (2012). URL http://dx.doi.org/10.1021/nn204110z. 00052. 12 Regmi, R. et al. All-Dielectric Silicon Nanogap Antennas To Enhance the Fluorescence of Single Molecules. Nano Letters 16, 5143–5151 (2016). URL http://dx.doi.org/ 10.1021/acs.nanolett.6b02076. 13 Cambiasso, J. et al. Bridging the Gap between Dielec- tric Nanophotonics and the Visible Regime with Effectively Lossless Gallium Phosphide Antennas. Nano Letters 17, 1219–1225 (2017). URL http://dx.doi.org/10.1021/acs. nanolett.6b05026. 00004. 14 Kallel, H. et al. Photoluminescence enhancement of silicon nanocrystals placed in the near field of a silicon nanowire. Physical Review B 88, 081302 (2013). URL http://link. aps.org/doi/10.1103/PhysRevB.88.081302. 00011. 15 Shcherbakov, M. R. et al. Enhanced Third-Harmonic Gen- eration in Silicon Nanoparticles Driven by Magnetic Re- sponse. Nano Letters 14, 6488–6492 (2014). URL http: //dx.doi.org/10.1021/nl503029j. 16 Wiecha, P. R. et al. Enhanced nonlinear optical response from individual silicon nanowires. Physical Review B 91, 121416 (2015). URL http://link.aps.org/doi/10.1103/ PhysRevB.91.121416. 00000. 17 Wiecha, P. R., Arbouet, A., Girard, C., Baron, T. & Paillard, V. Origin of second-harmonic generation from individual silicon nanowires. Physical Review B 93, 125421 (2016). URL http://link.aps.org/doi/10.1103/ PhysRevB.93.125421. 00000. 18 Brongersma, M. L., Cui, Y. & Fan, S. Light management for photovoltaics using high-index nanostructures. Nature Materials 13, 451–460 (2014). URL http://www.nature. com/nmat/journal/v13/n5/full/nmat3921.html. 00279. 19 Mirzaei, A. & Miroshnichenko, A. E. Electric and magnetic hotspots in dielectric nanowire dimers. Nanoscale 7, 5963– 5968 (2015). URL http://pubs.rsc.org/en/content/ articlelanding/2015/nr/c5nr00882d. 20 Valuckas, V., Paniagua-DomÃnguez, R., Fu, Y. H., Luk'yanchuk, B. & Kuznetsov, A. I. Direct observation of resonance scattering patterns in single silicon nanoparticles. Applied Physics Letters 110, 091108 (2017). URL http:// aip.scitation.org/doi/abs/10.1063/1.4977570. 00000. 21 Person, S. et al. Demonstration of Zero Optical Backscat- tering from Single Nanoparticles. Nano Letters 13, 1806–1809 (2013). URL http://dx.doi.org/10.1021/ nl4005018. 22 Fu, Y. H., Kuznetsov, A. I., Miroshnichenko, A. E., Yu, Y. F. & Luk'yanchuk, B. Directional visible light scattering by silicon nanoparticles. Nature Communications 4, 1527 (2013). URL http://www.nature.com/ncomms/journal/ v4/n2/full/ncomms2538.html. 23 Kerker, M., Wang, D.-S. & Giles, C. L. Electromagnetic scattering by magnetic spheres. Journal of the Optical Society of America 73, 765 (1983). URL https://www. osapublishing.org/abstract.cfm?URI=josa-73-6-765. 24 Pendry, J. B., Schurig, D. & Smith, D. R. Control- Science 312, 1780–1782 ling Electromagnetic Fields. (2006). URL http://science.sciencemag.org/content/ 312/5781/1780. 25 Gómez-Medina, R. et al. Electric and magnetic dipo- lar response of germanium nanospheres: interference ef- fects, scattering anisotropy, and optical forces. Journal of Nanophotonics 5, 053512–053512–9 (2011). URL http: //dx.doi.org/10.1117/1.3603941. 00000. 26 Nieto-Vesperinas, M., Gomez-Medina, R. & Saenz, J. J. Angle-suppressed scattering and optical forces on sub- micrometer dielectric particles. JOSA A 28, 54–60 (2011). URL http://www.osapublishing.org/abstract. cfm?uri=josaa-28-1-54. 00107. 27 Zambrana-Puyalto, X., Fernandez-Corbaton, I., Juan, M. L., Vidal, X. & Molina-Terriza, G. Duality sym- metry and Kerker conditions. Optics Letters 38, 1857 (2013). URL https://www.osapublishing.org/abstract. cfm?URI=ol-38-11-1857. 28 Luk'yanchuk, B. S., Voshchinnikov, N. V., Paniagua- 11 Domínguez, R. & Kuznetsov, A. I. Optimum Forward Light Scattering by Spherical and Spheroidal Dielectric Nanoparticles with High Refractive Index. ACS Photon- ics 2, 993–999 (2015). URL http://dx.doi.org/10.1021/ acsphotonics.5b00261. 29 Staude, I. et al. Tailoring Directional Scattering through Magnetic and Electric Resonances in Subwavelength Silicon Nanodisks. ACS Nano 7, 7824–7832 (2013). URL http: //dx.doi.org/10.1021/nn402736f. 00295. 30 Decker, M. & Staude, I. Resonant dielectric nanostruc- tures: a low-loss platform for functional nanophotonics. Journal of Optics 18, 103001 (2016). URL http://stacks. iop.org/2040-8986/18/i=10/a=103001. 31 Yang, Y., Li, Q. & Qiu, M. Controlling the angular ra- diation of single emitters using dielectric patch nanoan- tennas. Applied Physics Letters 107, 031109 (2015). URL http://scitation.aip.org/content/aip/journal/ apl/107/3/10.1063/1.4927401. 32 Li, J. et al. All-Dielectric Antenna Wavelength Router with Bidirectional Scattering of Visible Light. Nano Letters 16, 4396–4403 (2016). URL http://dx.doi.org/10.1021/acs. nanolett.6b01519. 00010. 33 Alaee, R. et al. All-dielectric reciprocal bianisotropic nanoparticles. Physical Review B 92, 245130 (2015). URL http://link.aps.org/doi/10.1103/PhysRevB.92. 245130. 00010. 34 Albella, P., Shibanuma, T. & Maier, S. A. Switchable di- rectional scattering of electromagnetic radiation with sub- wavelength asymmetric silicon dimers. Scientific Reports 5, 18322 (2015). URL http://www.nature.com/articles/ srep18322. 35 Campione, S., Basilio, L. I., Warne, L. K. & Sinclair, M. B. Tailoring dielectric resonator geometries for direc- tional scattering and Huygens' metasurfaces. Optics Ex- press 23, 2293 (2015). URL https://www.osapublishing. org/abstract.cfm?URI=oe-23-3-2293. 36 Yan, J. et al. Directional Fano Resonance in a Silicon Nanosphere Dimer. ACS Nano 9, 2968–2980 (2015). URL http://dx.doi.org/10.1021/nn507148z. 00049. et al. 37 Guo, R. Multipolar Coupling in Hybrid Metal–Dielectric Metasurfaces. ACS Photonics 3, 349–353 (2016). URL http://dx.doi.org/10.1021/acsphotonics. 6b00012. 38 Wang, L. et al. Shaping the third-harmonic radi- ation from silicon nanodimers. Nanoscale 9, 2201– 2206 (2017). URL http://pubs.rsc.org/en/content/ articlelanding/2017/nr/c6nr09702b. 00000. 39 Xiong, X. Y. Z., Jiang, L. J., Sha, W. E. I., Lo, Y. H. & Chew, W. C. Compact Nonlinear Yagi-Uda Nanoan- tennas. Scientific Reports 6, 18872 (2016). URL http: //www.nature.com/articles/srep18872. 40 Mirzaei, A., Miroshnichenko, A. E., Shadrivov, I. V. & Kivshar, Y. S. All-Dielectric Multilayer Cylindrical Struc- tures for Invisibility Cloaking. Scientific Reports 5, 9574 (2015). URL http://www.nature.com/srep/2015/150330/ srep09574/full/srep09574.html. 00014. 41 Liu, W., Zhang, J., Lei, B., Hu, H. & Mirosh- nichenko, A. E. Invisible nanowires with interfering electric and toroidal dipoles. Optics Letters 40, 2293 (2015). URL https://www.osapublishing.org/abstract. cfm?URI=ol-40-10-2293. 42 Fan, P. et al. An invisible metal-semiconductor pho- todetector. Nature Photonics 6, 380–385 (2012). URL http://www.nature.com/nphoton/journal/v6/n6/abs/ 12 Express 24, 29760–29772 (2016). URL https://www. osapublishing.org/abstract.cfm?uri=oe-24-26-29760. 00000. 57 Dhalluin, F. et al. Silicon nanowires: Diameter dependence of growth rate and delay in growth. Applied Physics Letters 96, 133109 (2010). URL http://aip.scitation.org/doi/ full/10.1063/1.3373546. 00043. 58 Moriceau, H., Fournel, F. & Rieutord, F. Materials and manufacturing techniques for SOI wafer technology. In Kononchuk, O. & Nguyen, B.-Y. (eds.) Silicon-on-Insulator (SOI) Technology, 38–46 (Woodhead Publishing, 2014). 00000. 59 Han, X.-L., Larrieu, G., Fazzini, P.-F. & Dubois, E. Re- alization of ultra dense arrays of vertical silicon nanowires with defect free surface and perfect anisotropy using a top- down approach. Microelectronic Engineering 88, 2622–2624 (2011). URL http://www.sciencedirect.com/science/ article/pii/S0167931710005903. 60 Guerfi, Y., Carcenac, F. & Larrieu, G. High resolu- tion HSQ nanopillar arrays with low energy electron beam lithography. Microelectronic Engineering 110, 173–176 (2013). URL http://www.sciencedirect.com/science/ article/pii/S0167931713002724. 61 Martin, O. J. F., Girard, C. & Dereux, A. General- ized Field Propagator for Electromagnetic Scattering and Light Confinement. Physical Review Letters 74, 526– 529 (1995). URL http://link.aps.org/doi/10.1103/ PhysRevLett.74.526. 62 Paulus, M. & Martin, O. J. F. Green's tensor technique for scattering in two-dimensional stratified media. Physical Review E 63, 066615 (2001). URL http://link.aps.org/ doi/10.1103/PhysRevE.63.066615. 00072. shaping Superscattering 63 Liu, W. anisotropic ences. http://arxiv.org/abs/1704.07994. 1704.07994. for through multipolar (2017). 00001 arXiv:1704.07994 [physics] radially interfer- URL arXiv: nanowires 43 Grzela, G., Hourlier, D. & Gómez Rivas, J. Polarization- dependent light extinction in ensembles of polydisperse vertical semiconductor nanowires: A Mie scattering ef- fective medium. Physical Review B 86, 045305 (2012). URL https://link.aps.org/doi/10.1103/PhysRevB.86. 045305. 00017. 44 Muskens, O. L., Rivas, J. G., Algra, R. E., Bakkers, E. P. A. M. & Lagendijk, A. Design of Light Scattering in Nanowire Materials for Photovoltaic Applications. Nano Letters 8, 2638–2642 (2008). URL http://dx.doi.org/ 10.1021/nl0808076. 00405. 45 Abujetas, D. R., Paniagua-Domínguez, R. & Sánchez- Gil, J. A. Unraveling the Janus Role of Mie Resonances and Leaky/Guided Modes in Semiconductor Nanowire Ab- sorption for Enhanced Light Harvesting. ACS Photon- ics 2, 921–929 (2015). URL http://dx.doi.org/10.1021/ acsphotonics.5b00112. 00027. 46 Grzela, G. et al. Nanowire Antenna Emission. Nano Let- ters 12, 5481–5486 (2012). URL http://dx.doi.org/10. 1021/nl301907f. 00090. 47 Paniagua-Domínguez, R., Grzela, G., Rivas, J. G. & Sánchez-Gil, J. A. Enhanced and directional emission of semiconductor nanowires tailored through leaky/guided modes. URL http://pubs.rsc.org/en/content/articlelanding/ 2013/nr/c3nr03001f. 00000. Nanoscale 5, 10582–10590 (2013). 48 Brenny, B. J. M. et al. Directional Emission from Leaky and Guided Modes in GaAs Nanowires Measured by Cathodoluminescence. ACS Photonics (2016). URL http://dx.doi.org/10.1021/acsphotonics.6b00065. 49 Bohren, C. F. & Huffman, D. R. Absorption and scattering of light by small particles (Wiley, 1998). nphoton.2012.108.html. 50 Edwards, D. F. Silicon (Si)*. In Palik, E. D. (ed.) Handbook of Optical Constants of Solids, 547 – 569 (Academic Press, Burlington, 1997). URL http://www.sciencedirect.com/ science/article/pii/B9780125444156500273. 51 Liu, W. et al. Scattering of core-shell nanowires with the interference of electric and magnetic resonances. Optics Letters 38, 2621–2624 (2013). URL https://www. osapublishing.org/abstract.cfm?uri=ol-38-14-2621. 00037. 52 Kapitanova, P. et al. Giant field enhancement in high- index dielectric subwavelength particles. Scientific Reports 7, 731 (2017). URL http://www.nature.com/articles/ s41598-017-00724-5. 00001. 53 Alu, A. & Engheta, N. How does zero forward-scattering in magnetodielectric nanoparticles comply with the optical theorem? Journal of Nanophotonics 4, 041590–041590– 17 (2010). URL http://dx.doi.org/10.1117/1.3449103. 00057. 54 Traviss, D. J., Schmidt, M. K., Aizpurua, J. & Muskens, O. L. Antenna resonances in low aspect ra- tio semiconductor nanowires. Optics Express 23, 22771 (2015). URL https://www.osapublishing.org/abstract. cfm?URI=oe-23-17-22771. 55 Fan, P., Yu, Z., Fan, S. & Brongersma, M. L. Optical Fano resonance of an individual semiconduc- tor nanostructure. 471–475 (2014). URL http://www.nature.com/nmat/journal/v13/ n5/full/nmat3927.html. 00072. Nature Materials 13, 56 Landreman, P. E., Chalabi, H., Park, J. & Brongersma, M. L. Fabry-Perot description for Mie resonances of rectangular dielectric nanowire optical resonators. Optics (cid:20)Es,TM (cid:21) Es,TE ∞(cid:88) ∞(cid:88) SUPPORTING INFORMATIONS: STRONGLY DIRECTIONAL SCATTERING FROM DIELECTRIC NANOWIRES I. CONTRIBUTING MIE ORDERS TO FW/BW SCATTERING INTENSITY 1 Mie theory can be applied to infinitely long cylinders by expanding the fields in vector cylindrical harmonics (instead of the vector spherical harmonics as in "classical" Mie theory for spherical particles). The expansion coefficients for the scattered fields are the so-called scattering coefficients an for electric multipole contributions and bn magnetic multipoles. Details can be found e.g. in the book of Bohren and Huffman.S1 In the next step, the asymptotic scattered field far from the structure has to be calculated. In Mie theory, the S-matrix, relating the scattered field Es with the incident field Ei, can be used for this means. Under normal incidence on an infinite cylinder, the S-matrix, connecting the incident and the asymptotic scattered field, writes (cid:114) 2 (cid:20)T1 0 (cid:21)(cid:20)Ei,TM (cid:21) 0 T2 Ei,TE = ei3π/4 eikR πkR (S.1) with the wavenumber k = 2π/λ and R the distance to the cylinder origin. Under normal incidence the TM and TE polarized components of the scattered fields are proportional to the S-matrix components T1, respectively T2: T1 = b0 + 2 bn cos(nϕ) T2 = a0 + 2 n=1 n=1 an cos(nϕ) (S.2) where ϕ is the polar angle of scattering in the cross-sectional plane of the cylinder. ϕ = 0◦ corresponds to the forward scattering direction. If two-dimensional objects of arbitrary cross section need to be calculated, Mie theory cannot be used and numerical simulations become necessary. The Green dyadic methodS2 (GDM) can be applied on infinitely long, two dimensional problems ("2D-GDM"), by deriving appropriate Green's tensors in a similar way to Mie theory for infinitely long cylinders.S3 In order to determine the scattering to the far-field, each discretization cell can be considered a dipolar emitter, whose emission to the far-field can be obtained using the two-dimensional Green's tensor. The coherent superposition of the dipolar emission from all discretization cells gives the total scattered field at any arbitrary location outside the discretized object. The BW and FW scattered far-field intensities write for both, Mie theory and 2D-GDM simulations: 3π/2(cid:90) Es(ϕ)2dϕ Is,BW = π/2(cid:90) Is,FW = Es(ϕ)2dϕ. (S.3) π/2 −π/2 Directional scattering is a result of interference between multiple simultaneously excited modes (see Eqs. (S.1)-(S.2)), and therefore the FW/BW resolved scattered intensity cannot be plotted individually for the different contributing scattering coefficients an and bn (for TE and TM polarized normal incidence, respectively). In Figs. S.1 and S.2, the FW and BW scattered intensity from a normally illuminated SiNW is calculated successively for an increasing number of contributing terms. For a NW of D = 100 nm diameter (Fig. S.1), obviously only the first two orders contribute significantly, while for a larger NW (D = 200 nm, Fig. S.2), the interplay becomes more complex and several orders of the Mie scattering coefficients are necessary to describe all spectral features. We want to stress the fact that despite some missing spectral features, using only the first two Mie orders gives already a very good approximation also in the case of larger SiNWs (see Fig. S.2). We calculate furthermore the FW/BW scattering using the simple conditions derived in Eqs. (S.4) and (S.5), i.e. considering exclusively the FW (ϕ = 0◦) or BW (ϕ = 180◦) direction instead of performing the integration of Eq. (S.3). The results are shown in the very bottom row of Figs. S.1 and S.2 and closely match the spectra obtained via integration of a solid angle of ∆ϕ = π. 2 FIG. S.1. Mie development of the FW/BW scattering from a SiNW with diameter D = 100 nm for the first 4 Mie coefficients an (TM: bn). Nmax corresponds to the number of Mie terms: n = 0, n ∈ {0, 1}, n ∈ {0, 1, 2}, n ∈ {0, 1, 2, 3} (from top to bottom). Very bottom: FW/BW scattering intensity calculated using the Mie conditions derived for pure FW/BW scattering at ϕ = 0◦ / ϕ = 180◦ (no integration). Left: TE, right: TM polarized normal incident plane wave. For better visibility, TE/TM data is normalized separately. 0.00.20.40.60.81.0Nmax = 1TEFW scat.BW scat.FW/BW0.00.20.40.60.81.0TM0.00.20.40.60.81.0Nmax = 210-11001010.00.20.40.60.81.010-11001010.00.20.40.60.81.0Nmax = 310-11001010.00.20.40.60.81.010-1100101400500600700800900wavelength (nm)0.00.20.40.60.81.0scattering intensity (a.u.)Nmax = 410-1100101forward / backward ratio400500600700800900wavelength (nm)0.00.20.40.60.81.0scattering intensity (a.u.)10-1100101forward / backward ratiowavelength (nm)0.00.51.01.52.0a0±2a1210-1100101FW / BW ratiowavelength (nm)0123456789b0±2b1210-1100101FW / BW ratio3 FIG. S.2. Same as Fig. S.1 for SiNW with diameter D = 200 nm. 0.00.20.40.60.81.0Nmax = 1TEFW scat.BW scat.FW/BW0.00.20.40.60.81.0TM0.00.20.40.60.81.0Nmax = 210-11001010.00.20.40.60.81.010-11001010.00.20.40.60.81.0Nmax = 310-11001010.00.20.40.60.81.010-1100101400500600700800900wavelength (nm)0.00.20.40.60.81.0scattering intensity (a.u.)Nmax = 410-1100101forward / backward ratio400500600700800900wavelength (nm)0.00.20.40.60.81.0scattering intensity (a.u.)10-1100101forward / backward ratiowavelength (nm)012345a0±2a1210-1100101FW / BW ratiowavelength (nm)012345678b0±2b1210-1100101FW / BW ratioII. MIE SCATTERING COEFFICIENTS FOR AN INFINITELY LONG CYLINDER 4 The Mie scattering coefficients ai and bi (see e.g. Ref. S1) can be regarded as corresponding electric and magnetic multipole moments, representing the response of the nanowire to an external illumination. As can be seen from Eqs. (S.1) and (S.2), under normal incidence on an infinitely long cylinder a TE polarized plane wave scattering occurs only due to electric multipole contributions (ai) and a TM polarized excitation induces a purely magnetic scattering (via bi). Furthermore, if considering only contributions of first and second order, the scattered fields in direct forward direction (ϕ = 0◦) are for TM and TE polarization, respectively: Es,TM,FW ∝ b0 + 2b1 Es,TE,FW ∝ a0 + 2a1 (S.4) while for the exact backward direction (ϕ = 180◦) one finds: (S.5) The first two order Mie coefficient an and bn are shown in figures S.3 and S.4, respectively, separately for their real and imaginary parts. The absolute values of the sums/differences corresponding to pure FW/BW scattering (Eqs. (S.4) and (S.5)) are plotted as well (on the right). Es,TE,BW ∝ a0 − 2a1. Es,TM,BW ∝ b0 − 2b1, FIG. S.3. Mie electric scattering coefficients a0 (left) and a1 (second left) as function of wavelength and cylinder diameter. Contributions to FW scattering (a0 + 2a1, second right) and BW scattering (a0 − 2a1, right) are shown separately. Top: real parts, bottom: imaginary parts. FIG. S.4. Same as Fig. S.3 but for the magnetic scattering coefficients b0 and b1. 50100150200250300350NW diameter (nm)Re(a0)0.00.10.20.30.40.50.60.70.80.91.0Re(a1)0.00.10.20.30.40.50.60.70.80.91.0450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)Im(a0)1.00.80.60.40.20.00.20.40.60.81.0450500550600650700750800wavelength (nm)Im(a1)1.00.80.60.40.20.00.20.40.60.81.0450500550600650700750800wavelength (nm)50100150200250300350FW scattering:a0 + 2a10123450500550600650700750800wavelength (nm)BW scattering:a0 - 2a1012350100150200250300350NW diameter (nm)Re(b0)0.00.10.20.30.40.50.60.70.80.91.0Re(b1)0.00.10.20.30.40.50.60.70.80.91.0450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)Im(b0)1.00.80.60.40.20.00.20.40.60.81.0450500550600650700750800wavelength (nm)Im(b1)1.00.80.60.40.20.00.20.40.60.81.0450500550600650700750800wavelength (nm)50100150200250300350FW scattering:b0 + 2b10123450500550600650700750800wavelength (nm)BW scattering:b0 - 2b10123III. CYLINDRICAL NANOWIRES: COMPARISON MIE-THEORY VS. 2D-GDM SIMULATIONS VS. FDTD SIMULATIONS Figure S.5 shows FW and BW scattering spectra for a D = 100 nm large SiNW, calculated by Mie theory (solid lines) and 2D-GDM simulations (dashed lines). Both, the qualitative and quantitative agreement is excellent, apart from a small spectral shift in the TM case and small intensity variations towards small diameters. Figures S.6 and S.7 further demonstrate the excellent agreement between analytical theory and 2D-GDM simula- tions, which justifies the use of the 2D-GDM for the comparison with our experimental results. This is particularly important in the case of our rectangular nanowires, which cannot be treated analytically. 5 FIG. S.5. Comparison between Mie theory (solid lines) and 2D-GDM simulations (dashed lines) of FW/BW scattered inten- sities as function of the wavelength of the incident plane wave, polarized TE (top) or TM (bottom). Right: Comparison of Mie/2D-GDM calculated radiation patterns at selected wavelengths, all normalized to the overall maximum (TM, λ = 550 nm. Normalization factors are shown on the upper right of each polar plot). All data for a SiNW of diameter D = 100 nm. In a final step we double-check our methods by comparing them to finite difference time domain (FDTD) simulations. Therefore we calculate the electric and magnetic field intensity distributions inside infinitely long SiNWs using Mie theory, 2D-GDM simulation and via the FDTD simulation toolkit "Meep"S4. For the FDTD simulations we use the model from Ref. S5 for the dispersion of silicon, for Mie and 2D-GDM we use tabulated data from Ref. S6. The results of this comparison are shown for selected SiNW diameters (D ∈ [50 nm, 90 nm, 150 nm]) at fixed incident wavelength (λ = 500 nm) in figures S.8-S.10. Again, we obtain a very good qualitative and quantitative agreement between the three methods. We observe that 2D-GDM becomes more prone to numerical noise in the TE excitation configuration, which is probably due to the cubic mesh, not ideal in the description of a round structure such as a cylinder. In the TE configuration fields are not continuous along the air/NW interface, which is why in this case the numerical stability suffers more compared to a TM polarized incident plane wave. This might also explain the minor quantitative deviations between Mie and 2D-GDM, which are larger for TE than for TM polarization (see figure S.5). We note that FDTD can become quantitatively inaccurate for near field calculations when strong field confinement exists.S7 This might explain the deviations in field intensities between FDTD and Mie/2D-GDM. 0.00.51.01.52.02.5scattered intensity (a.u.)TE480nm550nm750nmFW - MieFW - 2D-GDMBW - MieBW - 2D-GDM400450500550600650700750800wavelength (nm)0246810scattered intensity (a.u.)TM×8λ=480nmTE×5λ=550nmTE×50λ=750nmTEMie2D-GDM×2.5TM×1TM×5TM6 FIG. S.6. FW/BW scattering intensities as function of NW diameter for an infinitely long silicon nanowire, calculated by Mie theory. FIG. S.7. Same as Fig. S.6, calculated using 2D-GDM. 50100150200250300350NW diameter (nm)TE FW scat0.000.080.160.240.320.400.480.560.64scat. intensity (a.u.)×2TE BW scat0.000.080.160.240.320.400.480.560.64scat. intensity (a.u.)450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)TM FW scat0.000.150.300.450.600.750.90scat. intensity (a.u.)450500550600650700750800wavelength (nm)×2TM BW scat0.000.150.300.450.600.750.90scat. intensity (a.u.)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-110010150100150200250300350NW diameter (nm)TE FW scat0.000.080.160.240.320.400.480.560.640.72scat. intensity (a.u.)×2TE BW scat0.000.080.160.240.320.400.480.560.640.72scat. intensity (a.u.)450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)TM FW scat0.000.150.300.450.600.750.90scat. intensity (a.u.)450500550600650700750800wavelength (nm)×2TM BW scat0.000.150.300.450.600.750.90scat. intensity (a.u.)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-11001017 FIG. S.8. Comparison of electric and magnetic field intensity distributions inside silicon nanowires, calculated with identical conditions using Mie theory (left), 2D-GDM simulations (center) and FDTD simulations (right). Incident plane wave with λ = 500 nm from the top, polarized normal to (TE, top row) or along (TM, bottom row) the NW axis. SiNW diameter is D = 50 nm. FIG. S.9. Same as Fig. S.8, with SiNW diameter of D = 90 nm. FIG. S.10. Same as Fig. S.8, with SiNW diameter of D = 150 nm. TEMie Theory0.0000.0150.0300.045E2/E020.00.81.62.4H2/H02TM0.00.30.60.9E2/E020246H2/H022D-GDM0.000.020.040.06E2/E020.00.81.62.4H2/H020.00.30.60.9E2/E020246H2/H02FDTD0.000.020.040.06E2/E020.00.81.62.4H2/H020.00.30.60.9E2/E020.02.55.07.5H2/H02TEMie Theory0.00.30.60.9E2/E020204060H2/H02TM0.01.53.04.5E2/E020204060H2/H022D-GDM0.00.51.01.5E2/E020255075H2/H020.01.53.04.5E2/E020204060H2/H02FDTD0.000.250.500.75E2/E02081624H2/H020.00.61.21.8E2/E02081624H2/H02TEMie Theory0.000.250.500.75E2/E02081624H2/H02TM0.00.81.62.4E2/E02081624H2/H022D-GDM0.00.40.81.2E2/E020102030H2/H020.00.81.62.4E2/E020102030H2/H02FDTD0.00.20.40.6E2/E0204812H2/H020.00.30.60.9E2/E0204812H2/H02IV. COMPARISON OF FW/BW SCATTERING FROM DIFFERENT NANOWIRE GEOMETRIES We show below, that the Fano-like horizontally guided modes occur only in non-symmetric nanowires (e.g. with rectangular cross sections, see case of TM polarization in Fig. S.11). Symmetric cross sections do not produce these kind of resonances (see Figs. S.12-S.15) 8 FIG. S.11. FW/BW scattering from infinitely long SiNW of rectangular section (fixed height H = 90 nm). FIG. S.12. FW/BW scattering from infinitely long SiNW of cylindrical section. rectangleD50100150200250300350400450NW diameter (nm)TE FW scat0150300450600750900105012001350scat. section (nm)×2TE BW scat0150300450600750900105012001350scat. section (nm)400450500550600650700750800wavelength (nm)50100150200250300350400450NW diameter (nm)TM FW scat0200400600800100012001400scat. section (nm)400450500550600650700750800wavelength (nm)×2TM BW scat0200400600800100012001400scat. section (nm)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-1100101circleD50100150200250300350NW diameter (nm)TE FW scat0.000.080.160.240.320.400.480.560.640.72scat. intensity (a.u.)×2TE BW scat0.000.080.160.240.320.400.480.560.640.72scat. intensity (a.u.)450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)TM FW scat0.000.150.300.450.600.750.90scat. intensity (a.u.)450500550600650700750800wavelength (nm)×2TM BW scat0.000.150.300.450.600.750.90scat. intensity (a.u.)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-11001019 FIG. S.13. FW/BW scattering from infinitely long SiNW of regular hexagonal section. The size parameter is the inscribed diameter of the hexagon. FIG. S.14. FW/BW scattering from infinitely long SiNW of square section. FIG. S.15. FW/BW scattering from infinitely long SiNW of equilateral triangular section. hexagonD50100150200250300350NW diameter (nm)TE FW scat0.00.10.20.30.40.50.60.70.80.9scat. intensity (a.u.)×2TE BW scat0.00.10.20.30.40.50.60.70.80.9scat. intensity (a.u.)450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)TM FW scat0.000.150.300.450.600.750.901.05scat. intensity (a.u.)450500550600650700750800wavelength (nm)×2TM BW scat0.000.150.300.450.600.750.901.05scat. intensity (a.u.)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-1100101square D 50100150200250300350NW diameter (nm)TE FW scat0.00.10.20.30.40.50.60.70.8scat. intensity (a.u.)×2TE BW scat0.00.10.20.30.40.50.60.70.8scat. intensity (a.u.)450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)TM FW scat0.000.150.300.450.600.750.901.05scat. intensity (a.u.)450500550600650700750800wavelength (nm)×2TM BW scat0.000.150.300.450.600.750.901.05scat. intensity (a.u.)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-1100101triangleD50100150200250300350NW diameter (nm)TE FW scat0.00.10.20.30.40.50.60.70.8scat. intensity (a.u.)×2TE BW scat0.00.10.20.30.40.50.60.70.8scat. intensity (a.u.)450500550600650700750800wavelength (nm)50100150200250300350NW diameter (nm)TM FW scat0.000.150.300.450.600.750.901.05scat. intensity (a.u.)450500550600650700750800wavelength (nm)×2TM BW scat0.000.150.300.450.600.750.901.05scat. intensity (a.u.)TE - FW/BW10-1100101450500550600650700750800wavelength (nm)TM - FW/BW10-110010110 V. LATERALLY GUIDED MODES IN RECTANGULAR NANOWIRES OF FINIT WIDTH It is possible to assess the effective index for the supported guided mode(s) assuming a 1D waveguide of infinite extensions in X- and Y -direction.S8 The thickness of this waveguide corresponds to the height of the rectangular nanowires (H = 90 nm). To stay with the examples shown in the main paper figure 5, we calculate the supported modes and corresponding effective indices for the wavelengths λ = 560 nm and λ = 580 nm (both: nSi ≈ 4.0). Under TM incidence, "TE" guided modes are excited and the following fundamental guided modes are found: • 580 nm → mode "TE0": neff = 3.45, λeff = 170 nm. → first order standing wave pattern in the width of a NW of W = 180 nm. • 560 nm → mode "TE0": neff = 3.45, λeff = 162 nm. → second order standing wave pattern in the width of a NW of W = 330 nm. Under TE polarized incidence "TM" guided modes are induced. compared to the "TE" guided modes • 580 nm → mode "TM0": neff = 2.58, λeff = 225 nm. In this case, the effective index is much lower In this case also the field confinement is significantly lower inside the slab, therefore the formation of standing wave patterns is less pronounced compared to the case of a TM polarized illumination of the nanowires. We note that theoretically also higher order guided modes are allowed in the slab, however with very low effective indices and weak field confinements. Therefore the lowest order guided mode dominates and we never observed standing wave patterns of higher order guided modes. VI. COLLECTION ANGLES FIG. S.16. Sketch illustrating the experimental collection angle range. (a) FW scattering setup, (b) BW scattering setup. Simulated angular range is 0◦ < β < 90◦, corresponding to the entire upper, respectively lower hemisphere. a)FW scatteringβmin=23°βmax=120°ObjectiveNA=0.75DF condenserNA=0.8-0.95b)BW scatteringαin=0°βmin=42°βmax=102°DF ObjectiveNA=0.5(condenser NA=0.8-0.95) αin=0°α: illumination angleβ: collection anglesGlass substrateGlass substrateindividualSiNWFW directioncollection:23°< β < 120°BW directioncollection:42°< β < 102°VII. OBLIQUE INCIDENCE The fields in figure S.17 are chosen for a spectral position, where for TM polarization the incident angle seems to have weak influence on the scattering (c.f. figure 6 (g-i) of main text). By comparing the field patterns, it seems that a similar kind of guided mode is excited at this particular configuration for incidence either along Z ("top") or along X ("side"). 11 FIG. S.17. Nearfield distributions in the sectional plane of a W = 250 nm, H = 90 nm rectangular NW under oblique incidence. Same as main paper, figure 6 (a-f), but for λ0 = 705 nm. Mostly, however, the angle of incidence does have a significant impact on the scattering from rectangular dielectric nanowires. To assess this effect, we show in figure S.18 additional spectra, analogously to main text figure 6, but for different NW widths. FIG. S.18. FW/BW scattering spectra from rectangular NWs under oblique incidence. Same as main paper, figure 6 (g-i), for: (a-c) W = 100 nm, (d-f) W = 150 nm and (g-i) W = 200 nm. For most of the spectra normal incidence gives a sufficient first order approximation compared to averaging over several incident angles. However, for larger asymmetries and at long wavelengths, a FW-directed resonance occurs under oblique angles, which does not exist under normal incidence. In order to explain the remarkable deviation from normal incidence around the fundamental TM mode, we plot the nearfield distribution in the NW section for a W = 180 nm, H = 90 nm NW at λ0 = 810 nm (see figure S.19). While under normal incidence, the fundamental mode along the "height" of the NW is excited (corresponding to a dipolar W=80nmW=120nmW=180nmresponse), under oblique incidence the "effective" height of the NW increases rapidly due to the large aspect ratio. Soon, a different resonance, corresponding to a significantly higher but narrower nanowire is excited, which contains a quadrupolar mode (see Fig. S.19d-f), leading to strongly directional scattering in that case. 12 FIG. S.19. Nearfield distributions in the sectional plane of a W = 180 nm, H = 90 nm rectangular NW under oblique incidence with λ0 = 810 nm. [S1] C. F. Bohren and D. R. Huffman, Absorption and scattering of light by small particles (Wiley, 1998). [S2] O. J. F. Martin, C. Girard, and A. Dereux, Physical Review Letters 74, 526 (1995). [S3] M. Paulus and O. J. F. Martin, Physical Review E 63, 066615 (2001), 00072. [S4] A. F. Oskooi, D. Roundy, M. Ibanescu, P. Bermel, J. D. Joannopoulos, and S. G. Johnson, Computer Physics Communi- [S5] A. Deinega, I. Valuev, B. Potapkin, and Y. Lozovik, Journal of the Optical Society of America A 28, 770 (2011), 00084. [S6] D. F. Edwards, in Handbook of Optical Constants of Solids, edited by E. D. Palik (Academic Press, Burlington, 1997) pp. cations 181, 687 (2010). 547 – 569. [S7] C. Forestiere, Y. He, R. Wang, R. M. Kirby, and L. Dal Negro, ACS Photonics 3, 68 (2016), 00008. [S8] M. Hammer, "1-D multilayer slab waveguide mode solver "OMS"," 00000.