Adiabatically Induced Orbital Magnetization

Cong Xiao, Yafei Ren, and Bangguo Xiong Department of Physics, The University of Texas at Austin, Austin, Texas 78712, USA

A semiclassical theory for the orbital magnetization due to adiabatic evolutions of Bloch electronic states is proposed. It renders a unified theory for the periodic-evolution pumped orbital magnetiza- tion and the orbital magnetoelectric response in insulators by revealing that these two phenomena are the only instances where the induced magnetization is gauge invariant. This theory also accounts for the electric-field induced intrinsic orbital magnetization in two-dimensional metals and Chern insulators. We illustrate the orbital magnetization pumped by microscopic local rotations of atoms, which correspond to phonon modes with angular momentum, in toy models based on honeycomb lattice, and the results are comparable to the pumped spin magnetization via strong Rashba spin orbit coupling. We also show the vital role of the orbital magnetoelectricity in validating the Mott relation between the intrinsic nonlinear anomalous Hall and Ettingshausen effects.

I. INTRODUCTION current in the second Chern form of Berry curvatures. Noticeably, the orbital magnetoelectric effect and the periodic-evolution pumped orbital magnetization emerge Understanding the orbital magnetization in crystalline as the only instances where the induced magnetization is solids is among the most important objectives of orbi- gauge invariant due to the elimination of its explicit time tronics [1–3]. Unlike its spin companion, the orbital mag- dependence. Our work thus renders a unified theory of netization of Bloch electrons in the absence of external both phenomena in insulators with vanishing Chern num- perturbations is already hard to access quantum mechan- bers. Besides, unlike the Chern-Simons contribution de- ically, as it does not correspond to a bounded operator. duced from the second Chern form current, the induced This nonlocality is finally accounted for by a Berry phase magnetization due to the perturbed Berry connection is formula [4–6] that gives significantly distinct orbital mag- well defined irrespective of Chern invariants and of insu- netization from the atom-centered approximation when lators or metals. As a result, the orbital magnetoelectric- combined with ab-initio calculations in various magnetic ity in two-dimensional (2D) metals and Chern insulators, materials [7–9]. In the presence of external driving elec- which had long been hard to approach, is also attained tric fields, the extrinsic responses of spin and orbital mag- in our theory. netization, namely the spin and orbital Edelstein effects, are given similarly by the magnetic moments averaged over current-carrying states [10–15]. In contrast, the in- We apply our theory to illustrate the orbital magne- trinsic responses of them, i.e., the spin and orbital mag- tization pumped by microscopic atomic rotations, which netoelectric effects, are completely different due to the correspond to phonon modes with angular momentum nonlocal nature of the magnetic dipole operator. Specifi- [25, 26], in toy models based on the honeycomb lattice. cally, the spin magnetoelectricity [16, 17] is dictated by a The results are comparable to the pumped spin mag- Berry curvature following the ubiquitous character of in- netization via a strong Rashba spin orbit coupling [24]. trinsic linear responses of a local operator [18], while the We also show the vital role played by the electric-field orbital one consists of a Chern-Simons three form and induced orbital magnetization in the nonlinear intrinsic a perturbative term of the reciprocal-space Berry con- anomalous Ettingshausen effect in 2D metallic systems. nection [19–22]. Besides, the magnetization pumped by In particular, the Mott relation is validated in intrin- periodic adiabatic processes in band insulators has been sic nonlinear transport by subtracting the magnetization studied recently by a density matrix approach evaluat- component of the thermal current in the second order of ing the time-averaged expectation value of the spin and the electric field. magnetic dipole operators [23, 24]. In this approach, the orbital magnetization can only be obtained in the Wan- nier basis [23], in contrast to the spin one that can be This paper is organized as follows. In Sec. II we lay arXiv:2012.08750v3 [cond-mat.mes-hall] 3 Jan 2021 evaluated in the Bloch representation. out the semiclassical theory of Bloch electrons, which is Up to date, the orbital magnetization induced by elec- employed to study the adiabatically induced orbital mag- tric fields and by periodic adiabatic processes are treated netization in metals and insulators, respectively, in Sec. by different theories. Whether both phenomena have a III and Sec. IV. A case study of the orbital magnetization deep connection and if they can emerge in a unified the- pumped by local rotations of atoms and the application oretical framework are still unknown. In this work, we of the orbital magnetoelectricity to the intrinsic nonlin- develop a semiclassical theory for the magnetization in- ear anomalous Ettingshausen effect are shown in Sec. V, duced by adiabatic evolutions of Bloch electronic states. followed by a summary in Sec. VI. Some technical details In general, the adiabatically induced orbital magnetiza- of the theory are presented in Appendices A and B for tion is gauge dependent due to the presence of the electric the convenience of interested readers. 2

0 II. SEMICLASSICAL THEORY of local Hamiltonian Hˆc but is modified by Hˆ up to the linear order of spatial gradients [22]. In the well es- In the semiclassical description, the Hamiltonian felt tablished second-order theory [29] the inhomogeneity ap- pears only in electromagnetic gauge potentials, whereas by a narrow wave packet centered around position rc is 0 the following results account for weak inhomogeneities Hˆ = Hˆc +Hˆ in the first order gradient expansion, where ˆ ˆ
P ˆ(α) (α) of mechanical fields conjugate to general bounded opera- Hc = H0 rˆ, pˆ; w (rc) ,R (t) + α θ · h (rc, t) is 0 1 ∂Hˆc tors. This generalization is necessary to obtain the adia- the local Hamiltonian and Hˆ = {(ˆr − rc) , } is 2 i ∂rci batically induced orbital magnetization carried by Bloch the gradient correction [27]. The Einstein summation electrons by calculating the magnetization current, which convention is implied for repeated Cartesian indices i, j is manifested only in nonuniform systems [30, 31]. For ˆ and s henceforth. H0 is the local approximation of the this purpose, it is sufficient to retain results up to the ˆ genuine Hamiltonian. The most general H0 considered order of the product spatial and time derivatives. here includes w (rc) that represents possible nonuniform In the second-order theory the wave packet takes the static mechanical fields varying slowly on the scale of form of [29] the wave packet as well as R (t) serving as a parameter X X whose time evolution is adiabatic. Besides, in order to |W (k, r , t)i = eip·rˆ[C |u i + C |u i], implement the variational approach to obtain the local n c np np n1p n1p p n16=n density of a bounded observable, we add the auxiliary (2) P ˆ(α) (α) term α θ · h (r, t) into the Hamiltonian and ex- where (derivations presented in Appendix A) (α) pand it around rc. Here θˆ (α = 1, 2, ..) is a set of bounded observable operators, each of which is assumed ∆n1nCn rc p Cn = − iA · (i∂p + A − rc)Cn (3) to be a vector for simple notations, without losing gen- 1 n1n n εnp − εn1p erality. h(α) (r, t) denotes the conjugate slowly varying (α) external fields, and h (rc, t) changes adiabatically in incorporates spatial-gradient induced corrections. Here the parameter space (t, rc). At the end of the calculation (α) i X p the auxiliary term is set to zero, i.e., h = 0. ∆n n ≡ (∂pˆ · ∂r Hˆc)n n + (∂r Hˆc)n n · A 1 2 c 1 c 1 2 n2n In the semiclassical theory that is accurate to the n26=n first order of spatial gradients and of time derivative r − A c · v (4) P n1n n [27, 28], the wave packet |Wn (k, rc, t)i = p Cnp|ψnpi is constructed by superposing the local Bloch states rc has the dimension of energy, with A = hun|i∂r |un i. ip·rˆ ˆ nn1 c 1 |ψnp (rc, t)i = e |unp (rc, t)i of Hc. Here n and ~p are The wave-packet center appearing in Eq. (3) is deter- the band index and crystal momentum, respectively, and k k mined by rc = hW |rˆ|W i = ∂kγ(k) + A + a , where the coefficient Cnp is sharply distributed around the wave −γ(p) is the phase of C . While the first two terms P 2 np vector k of the wave packet, obeying p Cnp = 1. To appear already in the first order theory [27], the third simplify notations, all the band quantities without ex- term ak ≡ 2 Re P P C∗ C Ap is the inhomo- n p n16=n np n1p nn1 plicit band index are considered for band n, unless oth- geneity induced positional shift of the wave-packet cen- erwise noted. Throughout this study we consider nonde- ter. Gathering the above results we get a gauge invariant generate bands to simplify the analysis and assume they expression are so separated that adiabatic evolutions are feasible. The wave packet Lagrangian reads (set = 1) Ak ∆ ~ k X nn1 n1n rck an = 2 Re − ∂k ·Gn , (5) d εn − εn1 L = hW |i −Hˆ |W i = r˙ ·k+k˙ ·Ak+r˙ ·Arc +At−ε,˜ (1) n16=n dt c c rck P rc k k/r with G = Re A A being the quantum c t n n16=n nn1 n1n where A = hun|i∂k/rc |uni and A = hun|i∂t|uni are the Berry connections derived from the periodic part metric tensor in (rc, k) space. One can then find that the wave-packet Lagrangian |un (k, rc, t)i of the Bloch wave [27]. The noncanonical form of the Lagrangian due to Berry connections implies takes the same form [Eq. (1)] as in the first order theory, but the involved Berry connections are modified by in- that (rc, k) are not canonical variables, thus the measure ˜k k k ˜t t t of the phase space spanned by (rc, k) should be modified, homogeneity, i.e., A = A + a and A = A + a . The krc krc corrected Berry connections in fact take similar struc- with the result [4] D = 1+Ωii . Here Ωii is the trace of krc rc k tures, e.g., the Berry curvature Ωij ≡ ∂ki Aj − ∂rcj Ai , and other Berry curvatures are formed similarly. The wave-packet At ∆ t X nn1 n1n rct energy is given byε ˜ = ε + δε up to first order gradi- an = 2 Re − ∂k · Gn , (6) εn − εn1 ents, where εn (k, rc, t) is the local Bloch energy, δεn = n16=n Re P Ak · (∂ Hˆ ) and Ak = hu |i∂ |u i is n16=n nn1 rc c n1n nn1 n k n1 where Grct = Re P Arc At , and are gauge in- the interband Berry connection. n n16=n nn1 n1n When going beyond the above first-order theory, the variant. Note that the correction to Arc is not needed, wave packet is no longer dictated only by the Bloch states since Arc is already in the first order of spatial gradients. 3

Meanwhile, the wave-packet energy does not receive fur- eˆs is the unit vector in the s direction, and ther corrections at the order of the product spatial and Z time derivatives. M 0 = (f 0m + eg0Ω). (10) Having identified the wave-packet Lagrangian and the concomitant action S, one gets directly the semiclassical Here m = e P Ak × v is the orbital moment dynamics of Bloch electrons following from the Euler- n 2 n16=n nn1 n1n Lagrange equation in (rc, k) space. Furthermore, one of a Bloch electron, Ω is the vector form of Berry curva- kk 0 R ∞ can consider the local density of a bounded observable ture Ωij , and g = − ε f(η)dη is the grand potential θˆ, of which the conjugate external field is marked by h, density contributed by a Bloch electron. With the sym- contributed by a Bloch-electron ensemble with the occu- metric gauge for the uniform magnetic field, Eq. (6) gives pation function fn (rc, k, t). The general recipe for this −At m e has been given recently as [18] t X nn1 n1n kt ∂Ba = 2 Re + ∂k × Gn , (11) Z εn − εn1 2 δS n16=n θ (r, t) = − [dk] dr Df | , (7) c δh (r, t) h→0 where Gkt = Re P Ak At is the quantum met- P d n n16=n nn1 n1n where [dk] ≡ n dk/(2π) with d as the spatial dimen- ric in (k, t) space, and mn1n = −∂B∆n1n, which will sionality. In what follows we suppress the notation h → 0 be elaborated later in combination with a more specific but the results for various adiabatic responses are calcu- physical context. lated in this limit. We take f(˜ε) as the Fermi distribution Equation (8) is the pivotal result of this paper. The in order to focus on adiabatic intrinsic contributions de- first line is of total spatial and time derivatives, hence is termined solely by band structures. certainly intimately related to the orbital magnetization The spin magnetoelectricity and spin pumping by peri- and electric polarization. Apparently, M 0 is the mag- odic adiabatic processes can be readily obtained from the netization that relies solely on instantaneous electronic above formula in the first order of time derivative, as de- states and corresponds to the equilibrium orbital mag- tailed in Appendix B, while the orbital counterparts can netization in the static case [6]. On the other hand, the only be acquired through calculating the electric current, magnetization and polarization may not be determined which is much more involved and is elaborated in the next by the first line of Eq. (8) alone, as the second line can two sections. be relevant as well. This line consists of first and second Chern forms of Berry curvatures, which underline various electronic topological responses of insulators [28, 33–35]. III. ORBITAL MAGNETIZATION IN METALS Besides, the last line signifies intrinsic Fermi-surface con- tributions to the charge current density in metals, which A. Nonlinear electric current are beyond the conventional Boltzmann transport pic- ture of conductors [36] and are distinct from intrinsic In order to address the orbital magnetization induced Fermi-sea contributions to linear response. by adiabatic evolutions, we need to formulate the local Now we are in a position to compare Eq. (8) with ex- charge current density up to the order of the product isting results at the same order. The second line of this spatial and time derivatives. To achieve this Eq. (7) is equation has been formulated in inhomogeneous insula- considered in the case of θˆ = evˆ being the charge current tors [33, 37] and metals [38]. The specific case where the operator, hence h = −A is the electromagnetic vector inhomogeneity enters only through the magnetic vector potential with a minus sign, which enters through the potential has been studied in insulators with degenerate minimal coupling, resulting in the chain rule ∂−A = e∂k. bands [20, 34] and in metals [39]. Meanwhile, these pio- We still use k to denote the gauge invariant crystal mo- neering studies disregarded the magnetization current in mentum. In the following the center label c is suppressed, the first line of Eq. (8), especially the orbital magnetiza- unless otherwise noted. After some manipulations, as tion induced by the Berry connection at due to adiabatic shown in Appendix B, we arrive at (hereafter R without time evolutions [Eq. (11)]. However, there is an impor- integral variable is shorthand for R [dk], f 0 = f (ε)) tant physical context: orbital magnetoelectricity in 2D Z Z metals, which is contributed entirely by this gauge in- 0 0 t 0 k j (r, t) = ∇ × (M + f ∂Ba ) + ∂t f ea variant term and hence is beyond the scope of the afore- mentioned theories. We discuss this subject shortly. Z 0 kt k[kr]t − e f (Ω + Ωs eˆs) Z Z 0 kt 0 t 0 k B. Orbital magnetoelectricity in 2D − e ∂εf δεΩ + e (∂kf a − ∂tf a ), (8) where the second Chern form of the Berry curvature [32, To address the orbital magnetoelectricity, we consider 33] is labeled as the case that the adiabatic time dependence stems en- tirely from the vector potential, i.e., E = −∂tA, with E k[kr]t kk rt kr tk kt kr Ωs ≡ Ωsi Ωi + Ωsi Ωi + Ωs Ωii , (9) being a weak constant electric field, then ∂t = eE · ∂k, 4 at = ak · eE and Ωk[kr]t = Ωk[kr]keE . Thus the local with Gkk = Re P Ak Ak as the k-space quan- s sj j n n16=n nn1 n1n charge current density [Eq. (8)] reduces to tum metric [41]. o Z It is interesting to compare αij with the spin magneto- 0 0 k j (r) = ∇ × [M + f ∂B(eE · a )] electric susceptibility contributed by each Bloch electron, which is given by the first term of Eq. (17) with m Z Z n1n − e2E × f 0Ω − e2E f 0Ωk[kr]keˆ replaced by the interband elements of spin magnetic mo- i si s e P k ment. Since mn1n = − n 6=n(vn1n2 +δn2n1 vn)×An n Z 2 2 2 2 0 k reduces to the familiar orbital moment mn when n1 = n, + e E × ∂εf (v × a − δεΩ). (12) it can be deemed as an interband orbital magnetic mo- ment. Despite this similarity between spin and orbital To understand this current we first inspect the case magnetoelectric susceptibility, the distinction is appar- when the spatial dependence originates only from the ent: the k-space dipole moment of the quantum metric vector potential, i.e., B = ∇ × A. Then it is apparent kk ∂kl Gsi does not have a counterpart in spin magnetoelec- that tricity. Noticeably, for two-band metallic systems with Z o 2 0 0 k B particle-hole symmetry, the first term of αij vanishes, j = −e E × {f Ω − ∂εf [v × (a ) + (m · B) Ω]}, 2 o e kk hence αij = 2 jls∂kl Gsi is given solely by the quantum (13) metric dipole, which is an intrinsic Fermi surface effect. k B where (a ) is proportional to the magnetic field. This Before closing this section, we note that in 3D insula- result recovers the intrinsic magneto-nonlinear Hall cur- tors with nonvanishing k-space Chern invariants or 3D rent of order EB that was obtained previously by a dif- metals, the second Chern form current in Eq. (12) obvi- ferent method [22]. ously poses a difficulty in pursuing a gauge invariant de- On the other hand, to identify the orbital magneti- composition in the form of Eq. (14). This difficulty raises zation one could introduce the spatial dependence from the question as to whether the electric-field induced or- other inhomogeneous external mechanical fields. By do- bital magnetization can be defined as a bulk quantity in ing so one may expect that the local current density in such systems. At the present stage this is still an open bulk can be decomposed into a transport and a magne- question [42–44] and is left for future efforts. tization component, namely [30] j = jtr + ∇ × M, (14) IV. ORBITAL MAGNETIZATION IN tr where the transport current j contributes to the net NON-CHERN INSULATORS flow through the sample. In 2D the second Chern form k[kr]k current is enforced to vanish due to Ωsi = 0, hence tr Now we turn to the nonlinear electric current in in- Eq. (14) is rescued with j taking the same form as the sulators in the general case of adiabatic time evolutions above magneto-nonlinear Hall current Eq. (13) and and spatial dependence, under the assumption of vanish- Z ing Chern numbers in all the pertinent parameter spaces. 0 0 k M = M + f ∂B(eE · a ). (15) Great simplifications of Eq. (8) occur in insulators. First, the Fermi-surface terms vanish and the Fermi-sea ones Recall that ak is physically a positional shift of a semi- are contributed by fully occupied bands. Then, accord- classical Bloch electron, the electric work upon this ing to the antisymmetric decomposition of the second shift implies immediately an orbital magnetization. This Chern form electric-field induced magnetization is in agreement with k[kr]t tkr tkr tkk kkr what is obtained recently by a different method [40], Ωs = ∂ks CSii − ∂ki CSsi − ∂ri CSis − ∂tCSsii , but the present derivation is much simpler, even though (18) starting from a more generic framework. In 2D, M is a where the involved Chern-Simons three forms read, e.g., tkr 1 t kr k rt r tk tkk 1 t kk pseudoscalar and is well defined irrespective of metals or CSsi = 2 (A Ωsi +As Ωi +Ai Ωs ), CSis = 2 (A Ωis + Chern insulators. k kt k tk kkr 1 k kr k rk Ai Ωs + As Ωi ) and CSsii = (As Ωii + Ai Ωis + k 2 Owing to the gauge invariance of a , it is legitimate ArΩkk), the current density takes the form of to define the orbital magnetoelectric susceptibility con- i si tributed by each Bloch electron in 2D αo via ij j (r, t) = ∇ × M (r, t) + ∂tP (r, t) . (19) Z 0 o ∂Mj/∂Ei = f αij. (16) Here we have taken the k-space periodic gauge for Bloch wave functions, and αo takes a gauge invariant form (j = z in 2D) ij Z 0 t 1 tkk k 2 M = M + (∂Ba − e lisCSis eˆl), (20) o X (Ai )nn1 (mj)n1n e kk 2 αij = −2e Re + jls∂kl (Gsi )n, εn − εn 2 Z n 6=n 1 k k kkr 1 P = e (A + a + CSsii eˆs) (21) (17) 5 can be deemed as the orbital magnetization and polar- in agreement with the previous theory [22], and verify ization induced, respectively, by the adiabatic time evo- ∂Mi/∂Ej = ∂Pj/∂Bi. lution and spatial inhomogeneity. One can tell from Eq. (8) that the perturbative con- t tribution ∂Ba to the orbital magnetization is well de- B. Periodic-evolution pumped orbital fined regardless of Chern numbers in (k, t) space and is magnetization invariant under a gauge transformation of Bloch wave functions (a phase transformation is compatible with the It is also possible to define, based on M and P , the k-space periodic gauge). In contrast, the Chern-Simons time averaged orbital magnetization in time periodic sys- orbital magnetization deduced from the second Chern tems and spatially averaged polarization in spatially pe- form current is only well defined in insulators with van- riodic systems. Here we concentrate on the magnetiza- ishing (k, t)-space Chern numbers. It changes under the tion, and the polarization can be discussed similarly. If gauge transformation. It can be readily shown that this the adiabatic time dependence of the electronic Hamilto- gauge dependence is permitted by the inherent degrees nian is periodic with period T and the Chern invariants of freedom of M (r, t) and P (r, t) determined by the in- in (k, t) space are zero, then the time averaged M is variance of the local current density Eq. (19) [45] (e.g., gauge invariant and can be perceived as the orbital mag- in the 2D case the inherent degrees of freedom of M and netization pumped by the periodic evolution, namely P are M zˆ → M zˆ−∂ χzˆ and P → P +∇×(χzˆ), with z z t Z T dt a scalar field χ(r, t)). M¯ (r) = M (r, t) . (23) This gauge dependence also implies, on the other hand, 0 T the necessity of removing the time dependence of the or- We use the notation M¯ to distinguish from the instan- bital magnetization and the spatial dependence of the taneous magnetization M (r, t) in Eq. (19). In 2D the electric polarization if one would like to pursue gauge in- ∇φ degree of freedom of the magnetization is irrelevant variant definitions of them. Therefore, there are generally and the so defined M¯ (r) gives the orbital magnetiza- two ways to have a gauge invariant orbital magnetization: tion unambiguously. In both 2D and 3D this M¯ coin- to either eliminate the explicit time dependence of M or cides with the abelian version of the so-called geometric pursue the definition upon an average over time. These orbital magnetization obtained by a density matrix ap- two approaches correspond to two important physical proach evaluating the time-averaged expectation value of contexts – orbital magnetoelectric response and orbital the magnetic dipole operator in the Wannier basis in ho- magnetization pumping – that are addressed separately mogeneous band insulators [23]. On the other hand, the in the following two subsections. present theory does not invoke the Wannier basis and accounts also for weak inhomogeneous systems.

A. Orbital magnetoelectric response V. APPLICATIONS When the time and spatial dependence concerns only the electromagnetic gauge potentials, the explicit time A. Model illustration of orbital magnetization dependence of M and spatial dependence of P are re- pumped by local rotations of atoms moved due to the minimal coupling. This is the case of the orbital magnetoelectric response in insulators, which To illustrate the above theory, we consider a minimal includes two dual effects: a constant electric (magnetic) model for the orbital magnetization due to the periodic field induces an orbital magnetization (electric polariza- adiabatic evolution of electronic states induced by mi- tion). Most previous derivations are designed for only croscopic local rotations of atoms. Such a model is not one of the two dual effects [19–22], while a theory ca- required to possess the spin-orbit coupling, in contrast pable of both simultaneously is rare [46]. Here we show to the spin magnetization induced by local circulations that they are readily derived from the present theory. of atoms that is only possible with the aid of spin-orbit On the one hand, when the time dependence appears coupling [24]. The minimal spatial dimensionality for ro- R tkk −θe solely as E = −∂tA, one has CSis = 4π2 isjEj, tational motions is two, and the model should have a gap. R d3k k Moreover, the second Chern form current can be nonzero where θ = − 4π A · Ω is the abelian version of the so called θ-term [28, 35]. Then M [Eq. (20)] becomes a only if the dimension of the Hamiltonian is larger than time-independent orbital magnetization two [33]. Therefore, we here consider a two-band model hence focus exclusively on the contribution Z 2 0 k e Z T Z dt dk t M = M + ∂B a · eE + 2 θE. (22) ¯ 4π M = 2 ∂Ba (24) 0 T (2π) On the other hand, when the spatial dependence ap- from the perturbed Berry connection. According to the R kkr θe t pears only as a magnetic field, CSsii = 4π2 Bs. Thus expression for ∂Ba (only the first term of Eq. (11) mat- one can identify P as a uniform polarization, which is ters in insulators), one can easily verify that it can be 6

A and B (see also Fig 1(a)), thus the nearest neighbor bond lengths change with time by the microscopic local rotations, while the next nearest neighbor ones do not. One can hence take these rotations as the modulation of the nearest neighbor hopping. By writing down the tight-binding Hamiltonian and converting it to a k-space one, the resultant adiabatic perturbation to Hˆ (k) reads √ δHˆ (k, t) = − δt0(σx sin y + σy cos y) 3 sin x cos ωt

+δt0[(cos x cos y − cos 2y)σx

− (cos x sin y + sin 2y)σy] sin ωt, (27)

where δt0 ∝ 2u0 arises from the change of the first near- est neighbor hopping energy due to the variation of the inter-atomic distance by the local rotations [24]. In our calculation, we consider the chemical potential inside the band gap and set t0 as the energy unit, t1 = 0.1t0, and δt0 = 0.1t0. With ∆ = 0.2t0, we plot energy bands in the absence FIG. 1. Model illustration of the orbital magnetization of of phonons in Fig. 1(a) where a band gap opens and the Bloch electrons induced by the microscopic local rotation of energy bands do not show particle-hole symmetry. In atoms. (a) Band structure of a spinless-graphene toy model the presence of phonons, the adiabatic evolution of the with first and second nearest neighbor hoppings. (b) Time electronic states due to the local rotations of A and B dependence of the adiabatically induced magnetic moment in 2 atoms leads to a time-dependent orbital magnetic mo- units of eωa per unit cell. Here ω is the angular frequency ment, which is plotted in Fig. 1(b) in units of et0 a2 ~ω of the atomic rotation, a is the lattice constant, and we take ~ t0 (magnetic moment upon an area of a2) in an evolution the parameters as ∆ = 0.2t0, t1 = 0.1t0 and δt0 = 0.1t0. The insert shows that the pumped orbital magnetization in period. It is apparent that the induced orbital magne- one period of the local rotations of atoms is proportional to tization is proportional to the phonon frequency ω. By the next nearest neighbor hopping parameter. (c) k-space using the parameters of√ graphene with t0 = 3 eV and t et0 2 distribution of ∂Ba of the valence band. (d) Gap dependence the lattice constant a = 3a0 = 2.46 A,˚ a is about ~ of the pumped orbital magnetization. 4.77µB with µB as the Bohr magneton. One can find a weak oscillation with a nonzero net contribution (about −3 ω −3.5 × 10 µB ~ ) upon one period of the local rota- t0 nonvanishing in a two-band model only if the particle- tions of atoms, which implies a nonzero pumping of or- hole symmetry is broken. bital magnetization. The k-resolved instantaneous or- Such a minimal model can thus be chosen as a spinless t bital magnetization ∂Ba is plotted in Fig. 1(c) where graphene-type one based on the honeycomb lattice taking one can find that the main contribution is from K and into account the next nearest neighbor hopping, which is K’ points with minimal interband spacings. As the band described by the Hamiltonian gap decreases, the magnitude of the induced magnetiza-

Hˆ (k) = t0(FRσx − FImσy) + ∆σz + t1FNNσ0, (25) tion increases rapidly (Fig. 1(d)). We verify that the induced orbital magnetization van- where FR = 2 cos x cos y +√ cos 2y, FIm =√ 2 cos x√sin y − ishes with the next nearest neighbor hopping parameter sin 2y, and F = 2 cos(2x/ 3)+4 cos(x/ 3) cos 3y+3 NN√ t1 in the insert in Fig. 1(b), which also shows that the with x = kxa0 3/2, y = kya0/2 and a0 being the inter- pumped magnetization is proportional to t1 at least up atomic distance. The first and second nearest neighbor to t1 = 0.1t0. The resultant orbital magnetization pump- hoppings are t0 and t1, respectively, and a nonzero t1 ing in the above toy model is of the same order as the breaks the particle-hole symmetry. The staggered sub- pumped spin magnetization in a honeycomb lattice with lattice potential strength is ∆. very strong Rashba spin-orbit coupling (the Rashba co- Next, we add an adiabatic perturbation term due to efficient equals to 0.4t0) [24]. We also mention that, in the microscopic local rotation of atoms, and mainly fol- this latter four-band model, breaking the particle-hole low the treatment introduced in Ref. [24], where a right symmetry is not required for supporting nonzero orbital handed circularly polarized optical phonon mode at Γ magnetization pumped by rotations of atoms, and the point is considered, with frequency ω and displacement pumped orbital and spin magnetization are also gener- vectors ally comparable (not shown here). Furthermore, taking u = u (cos ωt, sin ωt), u = −u (26) ~ω/t0 = 0.1 (the ratio of the typical energy scales of A 0 B A phonons and electrons), the magnetic-moment pumping of A and B atoms on the two sublattices. There is a due to the periodic adiabatic change of electronic states phase difference π between the circular rotations of atoms induced by microscopic rotations of atoms is of the order 7 of the nuclear magneton. potential, T is the temperature, and we have made use of the result for the intrinsic nonlinear Hall electric current [22] j = −e2E × R f 0Ω˜. B. Intrinsic nonlinear anomalous Ettingshausen By integration by parts, the entropy density takes the effect in 2D metals R form of s (ε) = dη (η − µ) ∂µf (η) θ (η − ε) /T , which renders the thermal transport current to be Not only the electric current but also a thermal cur- rent carried by Bloch electrons can be induced by an ap- jh,tr 1 Z η − µ ∂f (η) = − dη j (η) . (33) plied electric field. In the intrinsic linear thermal current T e T ∂η response to the electric field, the zero-field orbital mag- netization plays a vital role [30, 31, 47]. It is therefore Here j (η) = −e2E × R θ (η − ε) Ω˜ is the nonlinear Hall anticipated that the electric-field induced orbital mag- electric current at zero temperature with Fermi energy netization is indispensable in the second-order nonlinear η. This equation is completely parallel to the generalized intrinsic thermal current response to the electric field, Mott relation between the transport thermal current and i.e., the nonlinear intrinsic anomalous Ettingshausen ef- electric current in the linear order of electric fields. At fect. In this subsection we discuss in more detail the low temperatures much less than the distances between semiclassical picture of the electric-field induced orbital the chemical potential and band edges, the Sommerfeld magnetization in 2D metals, and point out its key role in expansion is legitimate [48], hence the entropy density re- the intrinsic nonlinear anomalous Ettingshausen effect. 1 2 2 duces to s (ε) = 3 π kBT δ (µ − ε), which is concentrated First, the electric-field modified orbital magnetization on the Fermi surface and decays dramatically away from can be recast into an instructive form it. Then the standard form of the Mott relation follows Z 0 0 ˜ h 2 2 M = (f m˜ + eg Ω) (28) j /T = (π kBT/3e)[∂j (ε) /∂ε]|ε=µ. (34) in analogy to the magnetization (10) in the absence of Therefore, we extend the regime of validity of the Mott relation to the second-order intrinsic thermoelectric cur- electric fields. Here Ω˜ = ∂ ×[Ak +(ak)E] is the electric- k rent responses to a constant electric field. field modified Berry curvature, and

k 0 k E m˜ = m + ∂B(eE · a ) + ev × (a ) (29) VI. SUMMARY e is the orbital moment m˜ = 2 hW | (rˆ − rc) × vˆ|W i up to the first order of the electric field. m is the zero-field We have formulated a semiclassical theory for the or- orbital moment, v0 is the band velocity, and (ak)E is bital magnetization induced by general adiabatic evolu- the positional shift linear in the electric field [22]. tions of Bloch electronic states. This theory starts from Second, a coarse graining process based on the wave- formulating the electric current density in bulk, from packet description of Bloch electrons [47] shows that, the which the magnetization can be extracted. The induced electric-field induced local energy current density up to orbital magnetization is gauge dependent in general case the second order is given by but is gauge invariant only when the adiabatic time de- Z Z pendence is implicit or averaged out. These two cases jE = −eE × f 0εΩ˜ − E × f 0m˜ . (30) correspond to the orbital magnetoelectric response and the periodic-evolution pumped orbital magnetization. In the orbital magnetoelectric effect the adiabatic evo- A magnetization current jE,mag should be discounted to lution is driven by a constant electric field, and the time obtain the transport energy current density [30] jE,tr = dependence is only implicit through the evolution of me- jE − jE,mag. In uniform crystals, the energy magnetiza- chanical crystal momentum. Thus the pertinent second tion current at the linear order of the electric field is given Chern form current vanishes in 2D, making the 2D or- by the the material-dependent part of the Poynting vec- bital magnetoelectricity governed completely by the per- tor describing the energy flow [47]: jE,mag = −E × M 0. turbative term of the reciprocal-space Berry connection In the present nonlinear response, one has (Eqs. (16) and (17)), irrespective of insulators or metals. jE,mag = −E × M. (31) The role of the orbital magnetoelectricity in the non- linear intrinsic anomalous Ettingshausen effect, which is Consequently, the transport thermal current is given by proposed here as a transverse thermal current response in the second order of the driving electric field, has also been µ Z revealed in 2D metals. On the other hand, the orbital jh,tr = jE,tr − j = −eE × T s (ε) Ω˜, (32) e magnetoelectricity and the nonlinear intrinsic anomalous Ettingshausen effect in 3D metals are beyond the scope where s (ε) = (ε − µ) f 0 − g0 /T is the entropy density of the present theory. They may not be determined solely contributed by a particular Bloch state, µ is the chemical by bulk considerations and are left for future efforts. 8

In the context of the orbital magnetization pumped After some manipulations we get Eq. (3) with by periodic adiabatic evolutions in non-Chern insulators, the Chern-Simons contribution deduced from the second 1 p p ∆n n = (i∂p + A − A ) · (∂r Hˆc)n n Chern form current can be present even in 2D. Mean- 1 2 n1 n c 1 while, as a second Chern form can be nonzero only if the 1 X p + A · (∂r Hˆc)n n system has more than two bands [33], in a two-band min- 2 n1n2 c 2 imal model the pumped magnetization is dictated solely n26=n1 1 by the perturbative term of the time component of the X ˆ p rc + (∂rc Hc)n1n2 An n − An n · vn. (A4) Berry connection [Eqs. (24) and (11)]. We illustrated the 2 2 1 n26=n orbital magnetization pumping due to the periodic adia- batic change of electronic states induced by microscopic By using rotations of atoms in toy models based on the honeycomb ˆ ˆ lattice. The induced magnetization is of the same order ∂p · (∂rc Hc)n1n = (∂pˆ · ∂rc Hc)n1n as the pumped spin magnetization via strong Rashba spin X p + i [A · (∂r Hˆc)n n orbit coupling. n1n2 c 2 The presented formulation is based on the assumption n2 − (∂ Hˆ ) · Ap ], (A5) of well separated nondegenerate Bloch bands, whereas to rc c n1n2 n2n explore the semiclassical theories in the case of degen- erate bands and of closely located bands with possible one gets Eq. (4). Here we also note that Eqs. (3) and (4) non-adiabatic effects [49, 50] need separate studies. has also been obtained by a different method in a recent ˆ preprint [51]. In addition, when [∂rc Hc, rˆ] = 0, Eq. (4) reduces to ∆ ≡ P (∂ Hˆ ) ·Ap −Arc ·v . n1n n26=n rc c n1n2 n2n n1n n ACKNOWLEDGMENTS

We thank Luka Trifunovic and Liang Dong for en- lightening discussions. This work was supported by NSF Appendix B: Derivation of Eq. (8) (EFMA-1641101) and Welch Foundation (F-1255). As shown in Ref. [18], one has

Appendix A: Derivation of gradient corrected Z δS Z ∂L d ∂L wave-packet state [dk] drcDf = Df[ − ] δh ∂h dt ∂(∂th) Z ∂L ∂L Given that the perturbative Hˆ 0 is the form of the gra- − ∂ri Df[ − r˙i], dient correction, the first-order wave-packet reads ∂(∂ih) ∂(∂th) (B1) X  |Wn(k)i = Cnp |ψnpi p then the field variation formula (7) yields  ˆ 0 Z Z ∂at X hψn1p1 |H |ψnpi ˜ hT θ + |ψ i θs = Df(∂hs ε˜− Ω ) − ∂ri f[d − ]. (B2) n1p1  s is ∂ (∂ h ) εnp − εn1p1 i s n1p16=np X ip·rˆ X ˜ hT hk ˙ hr ˜ ht ˙ = e [Cnp|unpi + Cn p|un pi], (A1) Here Ωs = Ωsi ki + Ωsi r˙i + Ωs , where ki = −∂iε + 1 1 rt kt p n16=n Ωi andr ˙i = ∂ki ε − Ωi up to the order of the product spatial and time derivatives according to the equations where of motion, and only the Berry curvature Ωht need be ˆ 0 modified by inhomogeneity: X hψn1p|H |ψnp1 i Cn p = Cnp 1 1 t εnp1 − εn1p h h ˜ p1 ht t ∂A h h Ω˜ = ∂hA˜ − ∂tA˜ , A˜ = = A + a . (B3) ˆ ∂(∂th) X (∂rc Hc (p))n1n2 · i(Dp)n2nδpp1 = Cnp1 εnp1 − εn1p n2p1 Then we arrive at ˆ t X (∂rc Hc (p1))n2n · i(Dp)n1n2 δpp1 Z Z ∂a + Cnp . 0 t θ 0 1 θs (r, t) = ∂hs (G − f a ) − ∂ri [Dis − f ] εnp1 − εn1p n2p1 ∂ (∂ihs) Z Z (A2) 0 ht h[kr]t 0 h − f (Ωs + Ωs ) + ∂t f as Here we introduced the notation Z Z 0 ht 0 t 0 h − ∂εf δεΩs + (∂hs f a − ∂tf as ), (B4) i(D ) ≡ i∂ δ + Ap − r δ . (A3) p n2n p n2n n2n c n2n 9 where each term is gauge invariant, and the second Chern The first term on the right hand side is simply the average form of the Berry curvature is labeled as value of θ in the electron system obtained by using the instantaneous Hamiltonian, whereas the second term is hk rt hr tk ht kr h[kr]t Ωsi Ωi + Ωsi Ωi + Ωs Ωii ≡ Ωs . (B5) a geometric term related to the Berry curvature in (t, h) space Besides, G = R Dg (˜ε) is the electronic grand poten- tial density and is evaluated to the first order of gra- dients, g (˜ε) = g0 + f 0δε, and ∂g (˜ε) /∂ε˜ = f (˜ε) = 0 0 0 R ∞ 0 t f + ∂εf δε, with g = − f(η)dη and f = f (ε). A θ ε ht th X nn1 n1n Dθ = R (f 0dθ + g0Ωkh) is the θ-dipole density of the Ω = −Ω = 2 Re . (B7) is is is εn − εn1 θ
n16=n electron system, with dij = ∂δεn/∂ ∂ihj being the θ- dipole moment of a semiclassical Bloch electron [18]. In the case of θˆ = evˆ and h = −A, Eq. (B4) reduces to Eq. (8). Equation (B6) gives a unified account of diverse adiabatic On the other hand, in the absence of inhomogeneity responses of bounded operators in band insulators, such R 0 ht Eq. (B4) reduces to θ (t) = f (∂hε − Ω ). In insula- as the spin magnetoelectric effect, where the spin magne- tors, hence zero temperature for electrons, one has tization is induced by a weak electric field, and the spin Z Z magnetization pumped by microscopic local rotation of θ (t) = hu|θˆ|ui − Ωht. (B6) atoms [24].

[1] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Phys. New J. Phys. 12, 053032 (2010). Rev. Lett. 95, 066601 (2005). [20] A. M. Essin, A. M. Turner, J. E. Moore, and D. Vander- [2] D. Go, J.-P. Hanke, P. M. Buhl, F. Freimuth, G. bilt, Phys. Rev. B 81, 205104 (2010). Bihlmayer, H.-W. Lee, Y. Mokrousov, and S. Blugel, Sci. [21] K.-T. Chen and P. A. Lee, Phys. Rev. B 84, 205137 Rep. 7, 46742 (2017). (2011). [3] S. Bhowal and S. Satpathy, Phys. Rev. B 101, 121112(R) [22] Y. Gao, S. A. Yang, and Q. Niu, Phys. Rev. Lett. 112, (2020). 166601 (2014). [4] D. Xiao, J. Shi, and Q. Niu, Phys. Rev. Lett. 95, 137204 [23] L. Trifunovic, S. Ono, and H. Watanabe, Phys. Rev. B (2005). 100, 054408 (2019). [5] T. Thonhauser, D. Ceresoli, D. Vanderbilt, and R. Resta, [24] M. Hamada and S. Murakami, Phys. Rev. Res. 2, 023275 Phys. Rev. Lett. 95, 137205 (2005). (2020). [6] J. Shi, G. Vignale, D. Xiao, and Q. Niu, Phys. Rev. Lett. [25] L. Zhang and Q. Niu, Phys. Rev. Lett. 112, 085503 99, 197202 (2007). (2014). [7] D. Ceresoli, U. Gerstmann, A. P. Seitsonen, and F. [26] D. A. Garanin and E. M. Chudnovsky, Phys. Rev. B 92, Mauri, Phys. Rev. B 81, 060409(R) (2010). 024421 (2015). [8] M. G. Lopez, D. Vanderbilt, T. Thonhauser, and I. [27] G. Sundaram and Q. Niu, Phys. Rev. B 59, 14915 (1999). Souza, Phys. Rev. B 85, 014435 (2012). [28] D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, [9] J.-P. Hanke, F. Freimuth, A. K. Nandy, H. Zhang, S. 1959 (2010). Blugel, and Y. Mokrousov, Phys. Rev. B 94, 121114(R) [29] Y. Gao, Frontiers of Physics 14, 33404 (2019). (2016). [30] N. R. Cooper, B. I. Halperin, and I. M. Ruzin, Phys. [10] V. M. Edelstein, Solid State Commun. 73, 233 (1990). Rev. B 55, 2344 (1997). [11] T. Yoda, T. Yokoyama, and S. Murakami, Sci. Rep. 5, [31] C. Xiao and Q. Niu, Phys. Rev. B 101, 235430 (2020). 12024 (2015). [32] J.-H. Zhou, J. Hua, Q. Niu, and J.-R. Shi, Chin. Phys. [12] J. Lee, Z. Wang, H. Xie, K. F. Mak, and J. Shan, Nat. Lett. 30, 027101 (2013). Mater. 16, 887 (2017). [33] D. Xiao, J. Shi, D. P. Clougherty, and Q. Niu, Phys. Rev. [13] C. S¸ahin, J. Rou, J. Ma, and D. A. Pesin, Phys. Rev. B Lett. 102, 087602 (2009). 97, 205206 (2018). [34] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B [14] L. Salemi, M. Berritta, A. K. Nandy, and P. M. Oppe- 78, 195424 (2008). neer, Nature Communications 10, 5381 (2019). [35] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. [15] A. Johansson, B. Gobel, J. Henk, M. Bibes, and I. Mer- Lett. 102, 146805 (2009). tig, arXiv: 2006.14958 [36] J. M. Ziman, Principles of the Theory of Solids (Cam- [16] I. Garate and A. H. MacDonald, Phys. Rev. B 80, 134403 bridge University Press, Cambridge, 1972). (2009). [37] D. Bulmash, P. Hosur, S.-C. Zhang, and X.-L. Qi, Phys. [17] I. Garate and M. Franz, Phys. Rev. Lett. 104, 146802 Rev. X 5, 021018 (2015). (2010). [38] T. Hayata and Y. Hidaka, Phys. Rev. B 95, 125137 [18] L. Dong, C. Xiao, B. Xiong, and Q. Niu, Phys. Rev. Lett. (2017). 124, 066601 (2020). [39] D. Varjas, A. G. Grushin, R. Ilan, and J. E. Moore, Phys. [19] A. Malashevich, I. Souza, S. Coh, and D. Vanderbilt, Rev. Lett. 117, 257601 (2016). 10

[40] C. Xiao, H. Liu, J. Zhao, S. A. Yang, and Q. Niu, arXiv: [46] P. T. Mahon and J. E. Sipe, Phys. Rev. Res. 2, 033126 2002.01637 (2020). [41] A. Gianfrate, O. Bleu, L. Dominici, V. Ardizzone, M. De [47] D. Xiao, Y. Yao, Z. Fang, and Q. Niu, Phys. Rev. Lett. Giorgi, D. Ballarini, G. Lerario, K. W. West, L. N. Pfeif- 97, 026603 (2006). fer, D. D. Solnyshkov, D. Sanvitto, and G. Malpuech, [48] C. Xiao, D. Li, and Z. Ma, Phys. Rev. B 93, 075150 Nature 578, 381 (2020). (2016). [42] K.-T. Chen and P. A. Lee, Phys. Rev. B 86, 195111 [49] Matisse Wei-Yuan Tu, C. Li, H. Yu, and W. Yao, Phys. (2012). Rev. B 102, 045423 (2020). [43] D. L. Bergman, Phys. Rev. Lett. 107, 176801 (2011). [50] T. Stedman and L. M. Woods, Phys. Rev. Res. 2, 033086 [44] M. Barkeshli and X.-L. Qi, Phys. Rev. Lett. 107, 206602 (2020). (2011). [51] Y. Zhao, Y. Gao, and D. Xiao, arXiv: 2009.09306 [45] L. L. Hirst, Rev. Mod. Phys. 69, 607 (1997).