betaFeedback

WorksPhysics and Astronomy

Theory of shift heat current and its application to electron-phonon coupled systems

Abstract

We propose a heat current analog of the shift current, "shift heat current". We study nonlinear heat current responses to an applied ac electric field by a diagrammatic method and derive a microscopic expression for the second order dc heat current response. As a result, we find that the shift heat current is related to the shift vector, a geometric quantity that also appears in the expression for the shift current. The shift heat current directly depends on and can be controlled through the chemical potential. In addition, we apply the diagrammatic method to electron-phonon coupled systems, and we find that even if only the phonons are excited by an external field, the amplitude of the shift heat current is determined by the energy scale of electrons, not of phonons.

Introduction

The nonlinear optical response is a subject of recent intensive research. For example, second order optical responses such as second harmonic generation have been attracting much attention and studied both theoretically and experimentally [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22]. Recently, it is also pointed out that even in nonlinear regime quantum geometry and topological properties of materials are important to understand the electric/optical responses [9, 20]. Among them, the bulk photovoltaic effect is an important issue for both applications and fundamental physics. The bulk photovoltaic effect, or sometimes called photogalvanic effect, is the generation of photocurrents that can occur in noncentrosymmetric materials, and a mechanism called shift current is proposed and well established [3, 7, 8, 9, 13, 20]. The shift current is characterized by a quantity called shift vector RR, and for two bands systems with the time reversal symmetry (TRS), the shift current Je(shift)J_e^{(\mathrm{shift})} induced by an electric field EE can be written as [3, 9]

Je(shift)∝E2∫dk δ(εv(k)−εc(k)+ℏΩ)∣vvc(k)∣2R(k),(1)J_e^{(\mathrm{shift})} \propto E^2 \int dk\,\delta\left(\varepsilon_v(k)-\varepsilon_c(k)+\hbar\Omega\right)\lvert v_{vc}(k)\rvert^2R(k), \tag*{(1)}

where εc(k)\varepsilon_c(k) and εv(k)\varepsilon_v(k) are the dispersion of the conduction band and the valence band, ℏΩ\hbar\Omega is the energy of the input photon, vvc(k)v_{vc}(k) is the matrix element of the velocity operator. Physically, the shift vector RR can be interpreted as the spatial shift of an electron wave packet during the interband transition due to an excitation by light (see Figure 1(b,Figure 1c)), and contributes to the dc electric current. The shift of wave packet can contribute other types of transport. Kim et al. pointed out that the shift can contribute to spin current and proposed a phenomenon named shift spin current [14]. In the present work, we propose another current induced by the shift: shift heat current.

Image not extracted

Figure 1. Schematic pictures of shift current and shift heat current. (a) Illuminating light of frequency ω\omega induces dc electric current (shift current) and dc heat current (shift heat current). (b) In the momentum space, an incident photon excites an electron from the valence band (with dispersion εv(k)\varepsilon_{v}(\boldsymbol{k})) to the conduction band (with dispersion εc(k)\varepsilon_{c}(\boldsymbol{k})). (c) In the real space, the center of a wave packet is shifted during the interband transition. This shift of the electron wave packet (indicated by the black arrow) induces an electric current, the shift current, and simultaneously induces the shift of energy of (εc+εv)/2−μ(\varepsilon_{c}+\varepsilon_{v})/2-\mu (indicated by the red arrow), which results in the shift heat current.

Electrons carry not only charge and spin, but also heat. Therefore, it is natural to expect that there exists a heat current analog of shift current. In fact, heat transport phenomena and thermoelectric responses are closely related to electric/optical responses. It is well known that several universal relations between thermal responses and electric responses, such as the Wiedemann-Franz law and the Mott relation, hold [23]. These universal relations can be derived theoretically by the semi-

classical Boltzmann theory [23], or quantum linear response theory [24, 25, 26, 27]. Within the linear response regime, the Mott relation or the Sommerfeld-Bethe relation [27] indicates that the amplitude of the electric current JeJ_e and the heat current JQJ_Q when an electric field is applied can be roughly related to each other as

JQ∼teJe,(2)J_Q \sim\frac{t}{e}J_e, \tag*{(2)}

where ee is the charge of an electron and tt is the characteristic energy scale an electron carries as heat. If we assume that the relation between JeJ_e and JQJ_Q in Eq. (2) holds even for nonlinear responses, there should be nonlinear heat responses corresponding to the shift current, namely, shift heat current. One can also expect that the shift heat current are related to the geometric property of the material as the shift current does.

Although the electric current responses and heat current responses are closely related, the theoretical investigation on heat transport and thermoelectric effects has not progressed compared with that of the electric conductivity or the optical responses. In particular, thermoelectric responses of macroscopic systems in nonlinear regime are rarely discussed in contrast to aforementioned optical responses. In previous studies, nonlinear thermoelectric effects were studied mainly for mesoscopic systems [28, 29, 30, 31, 32, 33], where large temperature bias can be relatively easily applied. Although there are several studies on nonlinear thermoelectric responses to dc external fields in macroscopic systems [34, 35, 36, 37], heat current responses to ac external fields including the shift current-like responses have not been studied so far.

In this work, we study the dc heat current responses to an ac electric field. We extend the diagrammatic method to calculate nonlinear responses proposed in Ref. [17] to heat current responses, and find that the relation in Eq. (2) also holds for the shift current. In particular, for two-band systems with TRS, the shift heat current response JQ(shift)J_Q^{(\mathrm{shift})} can be written as

JQ(shift)∝E2∫dk δ(εv(k)−εc(k)+ℏΩ)vvc(k)2R(k)(εc(k)+εv(k)2−μ).(3)J_Q^{(\mathrm{shift})} \propto E^2 \int dk\,\delta\left(\varepsilon_v(k)-\varepsilon_c(k)+\hbar\Omega\right)v_{vc}(k)^2R(k) \left(\frac{\varepsilon_c(k)+\varepsilon_v(k)}{2}-\mu\right). \tag*{(3)}

We also find that the shift heat current directly depends on the chemical potential μ\mu and thus we can control the shift heat current through the chemical potential.

Furthermore, we also discuss the phonon-induced shift heat current. Recently, the phonon-induced shift current is experimentally observed in BaTiO3_3 [38]. There, although the energy of the input photon (∼1\sim1 meV) is much smaller than the band gap of BaTiO3_3 (∼1\sim1 eV), phonons are excited by photons and then induce a large electric current of the order of 10 μ\muA. Although the mechanism is different from the usual shift current, we can also expect the similar relation as Eq. (2). We theoretically derive the expression for such electron-phonon coupled systems, and find again that the relation (2) holds. It should be noted that, although photons excite phonons only, the energy scale tt in Eq. (2) is still that of electrons, not of phonons.

This paper is organized as follows. In § 2, we describe the theoretical formalism used to calculate the shift heat current. In § 3, we present the calculation results for the shift heat current. In addition to the most general formula of shift heat current, formulas for systems with TRS and TR symmetric two band systems are also given. In § 3 D, we apply our theory to Rice-Mele model as an example. In § 4, we study the phonon-induced shift heat current. § 5 is devoted to discussions.

Calculation method

In this paper, we calculate the nonlinear responses of the heat current to an electric field. To this end, we follow the formalism adopted in Ref. [17]. We first summarize our notations regarding the basics of the band theory. We introduce Bloch Hamiltonian and covariant derivative of kk-dependent operators, which is convenient to expand the Bloch Hamiltonian with respect to the applied external field. Then we introduce the electric current operator, the energy current operator, and the heat current operator.

Band theory and covariant derivative

Let us consider a tight binding model of noninteracting electron systems in a periodic potential, written as

H^0=∑R,R′ψ^R†HR−R′ψ^R′.(4)\hat{H}_0 = \sum_{R,R'} \hat{\psi}^{\dagger}_{R} H_{R-R'} \hat{\psi}_{R'}. \tag*{(4)}
ψ^R=(ψ^R,1,ψ^R,2,…,ψ^R,s)T.(5)\hat{\psi}_{R} = \left(\hat{\psi}_{R,1},\hat{\psi}_{R,2},\ldots,\hat{\psi}_{R,s}\right)^{T}. \tag*{(5)}
ψ^R†=(ψ^R,1†,ψ^R,2†,…,ψ^R,s†).(6)\hat{\psi}^{\dagger}_{R} = \left(\hat{\psi}^{\dagger}_{R,1},\hat{\psi}^{\dagger}_{R,2},\ldots,\hat{\psi}^{\dagger}_{R,s}\right). \tag*{(6)}

Here, RR specifies the position of a unit cell. ψ^R,i\hat{\psi}_{R,i} (ψ^R,i†\hat{\psi}^{\dagger}_{R,i}) is an annihilation (creation) operator, which annihilates (creates) an electron in the ii-th orbital of the unit cell at RR. HR−R′H_{R-R'} is an s×ss \times s matrix. Note that we denote the operator in the Fock space a symbol with a hat ( ^\hat{\ }) while its matrix representation is denoted by a symbol without  ^\hat{\ }.

The Hamiltonian in the momentum space representation is

H^0=∑kψ^k†H0(k)ψ^k,(7)\hat{H}_0 = \sum_k \hat{\psi}^{\dagger}_k H_0(k)\hat{\psi}_k, \tag*{(7)}
(H0(k))ij=∑Re−ik⋅(R+ri−rj)(HR)ij,(8)\left(H_0(k)\right)_{ij} = \sum_R e^{-ik\cdot(R+r_i-r_j)}(H_R)_{ij}, \tag*{(8)}

where ψ^k,i=1N∑Re−ik⋅(R+ri)ψ^R,i\hat{\psi}_{k,i}=\frac{1}{\sqrt{N}}\sum_R e^{-ik\cdot(R+r_i)}\hat{\psi}_{R,i}, NN is the total number of unit cells and R+riR+r_i is the position of the ii-th orbital in the unit cell at RR.

By diagonalizing the Hamiltonian, one obtain the representation in the energy eigenstate basis:

Ek=Uk†H0(k)Uk,(9)\mathcal{E}_k = U_k^{\dagger}H_0(k)U_k, \tag*{(9)}

where UkU_k is a unitary matrix and Ek\mathcal{E}_k is a diagonal matrix and its element (Ek)ab=δabεk,a(\mathcal{E}_k)_{ab}=\delta_{ab}\varepsilon_{k,a} gives the dispersion of the aa-th band. Then the Hamiltonian can be written as

H^0=∑kc^k†Ekc^k,(10)\hat{H}_0 = \sum_k \hat{c}^{\dagger}_k\mathcal{E}_k\hat{c}_k, \tag*{(10)}

where ψ^k=Ukc^k\hat{\psi}_k=U_k\hat{c}_k.

For convenience, we introduce a covariant derivative. Consider an operator of the following form:

O^=∑kψ^k†O(k)ψ^k=∑kc^k†Okc^k,(11)\hat{O}=\sum_k\hat{\psi}^{\dagger}_k O(k)\hat{\psi}_k=\sum_k\hat{c}^{\dagger}_k O_k\hat{c}_k, \tag*{(11)}
Ok=Uk†O(k)Uk.(12)O_k=U_k^{\dagger}O(k)U_k. \tag*{(12)}

The covariant derivative appears when one considers the kk-derivative of the operator, namely,

D[O^]:=∑kψ^k†∇kO(k)ψ^k.(13)\mathcal{D}[\hat{O}]:=\sum_k\hat{\psi}^{\dagger}_k\nabla_k O(k)\hat{\psi}_k. \tag*{(13)}

Substituting ψ^k=Ukc^k\hat{\psi}_k=U_k\hat{c}_k leads to the expression in the energy eigen basis,

D[O^]=∑kc^k†D[Ok]c^k,(14)\mathcal{D}[\hat{O}]=\sum_k\hat{c}^{\dagger}_k\mathcal{D}[O_k]\hat{c}_k, \tag*{(14)}
D[Ok]:=∇kOk−i[Ak,Ok],(15)\mathcal{D}[O_k]:=\nabla_k O_k-i[\mathcal{A}_k,O_k], \tag*{(15)}

where Ak\mathcal{A}_k is the interband Berry connection defined as Ak=iUk†∇kUk\mathcal{A}_k=iU_k^{\dagger}\nabla_kU_k. Here, D\mathcal{D} is the covariant derivative. We note that the Leibniz rule holds for the covariant derivative:

D[α][O1O2]=D[α][O1]O2+O1D[α][O2].(16)\mathcal{D}^{[\alpha]}[O_1O_2]=\mathcal{D}^{[\alpha]}[O_1]O_2+O_1\mathcal{D}^{[\alpha]}[O_2]. \tag*{(16)}

Electromagnetic interaction

Next, let us consider the Hamiltonian describing the electromagnetic interaction, and expand it with respect to the applied electric field.

To express a spatially uniform external electric field E(t)E(t), we use the velocity gauge. In this gauge, the electromagnetic interaction with a spatially uniform electric field does not violate the translational symmetry and is incorporated by the minimal substitution as

H^A(t)=∑kψ^k†HA(k)ψ^k,(17)\hat{H}_{A}(t)=\sum_{\boldsymbol{k}}\hat{\psi}_{\boldsymbol{k}}^{\dagger}H_{A}(\boldsymbol{k})\hat{\psi}_{\boldsymbol{k}}, \tag*{(17)}
HA(k)=H0(k−eℏA(t)).(18)H_{A}(\boldsymbol{k})=H_{0}\left(\boldsymbol{k}-\frac{e}{\hbar}\boldsymbol{A}(t)\right). \tag*{(18)}

where e<0e<0 is the charge of an electron, and A(t)\boldsymbol{A}(t) is a vector potential satisfying ∂A(t)/∂t=−E(t)\partial\boldsymbol{A}(t)/\partial t=-\boldsymbol{E}(t).

To treat the external field perturbatively, we expand the Hamiltonian with respect to A(t)\boldsymbol{A}(t) up to second order,

HA(k)=H0(k)+(−eℏ)Aα(t)∂αkH0(k)+12(−eℏ)2Aα(t)Aβ(t)∂αk∂βkH0(k).(19)H_{A}(\boldsymbol{k})=H_{0}(\boldsymbol{k})+\left(-\frac{e}{\hbar}\right)A^{\alpha}(t)\partial_{\alpha_{k}}H_{0}(\boldsymbol{k})+\frac{1}{2}\left(-\frac{e}{\hbar}\right)^{2}A^{\alpha}(t)A^{\beta}(t)\partial_{\alpha_{k}}\partial_{\beta_{k}}H_{0}(\boldsymbol{k}). \tag*{(19)}

Here ∂α=∂/∂kα\partial_{\alpha}=\partial/\partial k_{\alpha} and αk\alpha_{k} is a spatial index. (In this paper, greek indices μ,α,β,…\mu,\alpha,\beta,\ldots always represent the spatial indices with an implicit summation henceforth). This expression is in the basis ψ^k\hat{\psi}_{\boldsymbol{k}}. For analytic calculation, it is convenient to move to the energy eigenstate basis, c^k\hat{c}_{\boldsymbol{k}}. As already discussed, the kk-derivative in the basis ψ^k\hat{\psi}_{\boldsymbol{k}} becomes the covariant derivative D\mathcal{D} in the basis c^k\hat{c}_{\boldsymbol{k}}. Therefore, up to O(A2)O(A^{2}), the Hamiltonian reads

H^A(t)=H^0+V^,(20)\hat{H}_{A}(t)=\hat{H}_{0}+\hat{V}, \tag*{(20)}
V^=∑kc^k†(V1,k+V2,k)c^k,(21)\hat{V}=\sum_{\boldsymbol{k}}\hat{c}_{\boldsymbol{k}}^{\dagger}(V_{1,\boldsymbol{k}}+V_{2,\boldsymbol{k}})\hat{c}_{\boldsymbol{k}}, \tag*{(21)}

with

V1,k=(−eℏ)Aα(t)hkα=∫dω2πieℏωEα(ω)e−iωthkα(22)V_{1,\boldsymbol{k}}=\left(-\frac{e}{\hbar}\right)A^{\alpha}(t)h_{\boldsymbol{k}}^{\alpha}=\int\frac{d\omega}{2\pi}\frac{ie}{\hbar\omega}E^{\alpha}(\omega)e^{-i\omega t}h_{\boldsymbol{k}}^{\alpha} \tag*{(22)}
V2,k=12(−eℏ)2Aα(t)Aβ(t)hkαβ=12∫dω12πieℏω1Eα(ω1)e−iω1t∫dω22πieℏω2Eβ(ω2)e−iω2thkαβ(23)\begin{aligned} V_{2,\boldsymbol{k}}&=\frac{1}{2}\left(-\frac{e}{\hbar}\right)^{2}A^{\alpha}(t)A^{\beta}(t)h_{\boldsymbol{k}}^{\alpha\beta}\\ &=\frac{1}{2}\int\frac{d\omega_{1}}{2\pi}\frac{ie}{\hbar\omega_{1}}E^{\alpha}(\omega_{1})e^{-i\omega_{1}t}\int\frac{d\omega_{2}}{2\pi}\frac{ie}{\hbar\omega_{2}}E^{\beta}(\omega_{2})e^{-i\omega_{2}t}h_{\boldsymbol{k}}^{\alpha\beta} \tag*{(23)} \end{aligned}

where hα1⋯αnh^{\alpha_{1}\cdots\alpha_{n}} is defined as

hα1⋯αn=Dα1⋯Dαn[Ek].(24)h^{\alpha_{1}\cdots\alpha_{n}}=\mathcal{D}^{\alpha_{1}}\cdots\mathcal{D}^{\alpha_{n}}[\mathcal{E}_{\boldsymbol{k}}]. \tag*{(24)}

Although hα1⋯αnh^{\alpha_{1}\cdots\alpha_{n}} depends on kk, we omit kk from its notation. (We often omit kk-dependencies in other quantities as well in the following.) Eα(ω)E^{\alpha}(\omega) is the Fourier transform of the electric field and is related to the Fourier transform of the vector potential Aα(ω)A^{\alpha}(\omega) as

Eα(ω)=∫dt eiωtEα(t)=iωAα(ω).(25)E^{\alpha}(\omega)=\int dt\,e^{i\omega t}E^{\alpha}(t)=i\omega A^{\alpha}(\omega). \tag*{(25)}

Particle current operator, energy current operator, and heat current operator

Here we will introduce current operators used in this work. The current operators defined below change their form in the presence of an external electric field, and thus we expand them with respect to the external field, as we did for the Hamiltonian in the previous section.

It is well known that the total particle current operator j^\hat{j} is given by the kk-derivative of the Hamiltonian, i.e.,

J^=1V∑kψ^k†1ℏ∇kHA(k)ψ^k,(26)\hat{J}=\frac{1}{V}\sum_{\boldsymbol{k}}\hat{\psi}_{\boldsymbol{k}}^{\dagger}\frac{1}{\hbar}\nabla_{\boldsymbol{k}}H_{A}(\boldsymbol{k})\hat{\psi}_{\boldsymbol{k}}, \tag*{(26)}

where VV is the volume of the system. Note that the application of an electric field changes the form of the electric current operator in this gauge. This can be rewritten in terms of the covariant derivative and c^k\hat{c}_{\boldsymbol{k}} as

J^μ=∑kc^k†Jμ(k)c^k=∑kc^k†(J0μ(k)+J1μ(k)+J2μ(k))c^k,(27)\hat{J}^{\mu}=\sum_{\boldsymbol{k}}\hat{c}_{\boldsymbol{k}}^{\dagger}\mathcal{J}^{\mu}(\boldsymbol{k})\hat{c}_{\boldsymbol{k}}=\sum_{\boldsymbol{k}}\hat{c}_{\boldsymbol{k}}^{\dagger}\left(\mathcal{J}_{0}^{\mu}(\boldsymbol{k})+\mathcal{J}_{1}^{\mu}(\boldsymbol{k})+\mathcal{J}_{2}^{\mu}(\boldsymbol{k})\right)\hat{c}_{\boldsymbol{k}}, \tag*{(27)}

with

J0μ(k)=1ℏhμ,(28)\mathcal{J}^{\mu}_{0}(\mathbf{k})=\frac{1}{\hbar}h^{\mu}, \tag*{(28)}
J1μ(k)=1ℏ∫dω2πieℏωEα(ω)e−iωthμα,(29)\mathcal{J}^{\mu}_{1}(\mathbf{k})=\frac{1}{\hbar}\int\frac{d\omega}{2\pi}\frac{ie}{\hbar\omega}E^{\alpha}(\omega)e^{-i\omega t}h^{\mu\alpha}, \tag*{(29)}
J2μ(k)=12ℏ∫dω12πieℏω1Eα(ω1)e−iω1t×∫dω22πieℏω2Eβ(ω2)e−iω2thμαβ.(30)\mathcal{J}^{\mu}_{2}(\mathbf{k})=\frac{1}{2\hbar}\int\frac{d\omega_{1}}{2\pi}\frac{ie}{\hbar\omega_{1}}E^{\alpha}(\omega_{1})e^{-i\omega_{1}t} \times\int\frac{d\omega_{2}}{2\pi}\frac{ie}{\hbar\omega_{2}}E^{\beta}(\omega_{2})e^{-i\omega_{2}t}h^{\mu\alpha\beta}. \tag*{(30)}

(Note again that we omit kk from the notation of hα1…αnh^{\alpha_{1}\ldots\alpha_{n}}.)

When we ignore the interaction between electrons, the total energy current operator in longitudinal responses to an external electric field is given by

J^E=1V∑kψ^k†12(HA(k)J(k)+J(k)HA(k))ψ^k=1V∑kψ^k†12ℏ∇k(HA(k)2)ψ^k.(31)\begin{aligned} \hat{J}_{E}=\frac{1}{V}\sum_{\mathbf{k}}\hat{\psi}^{\dagger}_{\mathbf{k}}\frac{1}{2}\left(H_{A}(\mathbf{k})\mathcal{J}(\mathbf{k})+\mathcal{J}(\mathbf{k})H_{A}(\mathbf{k})\right)\hat{\psi}_{\mathbf{k}} \\ &=\frac{1}{V}\sum_{\mathbf{k}}\hat{\psi}^{\dagger}_{\mathbf{k}}\frac{1}{2\hbar}\nabla_{\mathbf{k}}\left(H_{A}(\mathbf{k})^{2}\right)\hat{\psi}_{\mathbf{k}}. \tag*{(31)} \end{aligned}

Equation (31) shows that the application of an electric field changes the form of the energy current operator in the velocity gauge. This is similar to the electric current operator, which also changes its form in the presence of a vector potential.

The energy current operator can be rewritten in terms of the covariant derivative and c^k\hat{c}_{\mathbf{k}} as

J^Eμ=1V∑kc^k†JEμc^k=1V∑kc^k†(JE0μ+JE1μ+JE2μ)c^k,(32)\hat{J}_{E}^{\mu}=\frac{1}{V}\sum_{\mathbf{k}}\hat{c}^{\dagger}_{\mathbf{k}}\mathcal{J}_{E}^{\mu}\hat{c}_{\mathbf{k}}=\frac{1}{V}\sum_{\mathbf{k}}\hat{c}^{\dagger}_{\mathbf{k}}\left(\mathcal{J}_{E0}^{\mu}+\mathcal{J}_{E1}^{\mu}+\mathcal{J}_{E2}^{\mu}\right)\hat{c}_{\mathbf{k}}, \tag*{(32)}

with

JE0μ(k)=1ℏgμ,(33)\mathcal{J}_{E0}^{\mu}(\mathbf{k})=\frac{1}{\hbar}g^{\mu}, \tag*{(33)}
JE1μ(k)=1ℏ∫dω2πieℏωEα(ω)e−iωtgμα,(34)\mathcal{J}_{E1}^{\mu}(\mathbf{k})=\frac{1}{\hbar}\int\frac{d\omega}{2\pi}\frac{ie}{\hbar\omega}E^{\alpha}(\omega)e^{-i\omega t}g^{\mu\alpha}, \tag*{(34)}
JE2μ(k)=12ℏ∫dω12πieℏω1Eα(ω1)e−iω1t×∫dω22πieℏω2Eβ(ω2)e−iω2tgμαβ,(35)\mathcal{J}_{E2}^{\mu}(\mathbf{k})=\frac{1}{2\hbar}\int\frac{d\omega_{1}}{2\pi}\frac{ie}{\hbar\omega_{1}}E^{\alpha}(\omega_{1})e^{-i\omega_{1}t} \times\int\frac{d\omega_{2}}{2\pi}\frac{ie}{\hbar\omega_{2}}E^{\beta}(\omega_{2})e^{-i\omega_{2}t}g^{\mu\alpha\beta}, \tag*{(35)}

where gμα1…αng^{\mu\alpha_{1}\ldots\alpha_{n}} is defined as

gα1…αn=12Dα1…Dαn[Ek2].(36)g^{\alpha_{1}\ldots\alpha_{n}}=\frac{1}{2}D^{\alpha_{1}}\ldots D^{\alpha_{n}}[\mathcal{E}_{\mathbf{k}}^{2}]. \tag*{(36)}

The heat current operator J^Q\hat{J}_{Q} is defined as

J^Q=J^E−μJ^=1V∑kψ^k†1ℏ∇k(12HA(k)2−μHA(k))ψ^k=1V∑kψ^k†12ℏ∇k(HA(k)−μ)2ψ^k,(37)\begin{aligned} \hat{J}_{Q}=\hat{J}_{E}-\mu\hat{J} \\ &=\frac{1}{V}\sum_{\mathbf{k}}\hat{\psi}^{\dagger}_{\mathbf{k}}\frac{1}{\hbar}\nabla_{\mathbf{k}}\left(\frac{1}{2}H_{A}(\mathbf{k})^{2}-\mu H_{A}(\mathbf{k})\right)\hat{\psi}_{\mathbf{k}} \\ &=\frac{1}{V}\sum_{\mathbf{k}}\hat{\psi}^{\dagger}_{\mathbf{k}}\frac{1}{2\hbar}\nabla_{\mathbf{k}}\left(H_{A}(\mathbf{k})-\mu\right)^{2}\hat{\psi}_{\mathbf{k}}, \tag*{(37)} \end{aligned}

where μ\mu is the chemical potential. Therefore, one obtains the heat current operator by replacing Ek\mathcal{E}_{\mathbf{k}} in Eq. (32) with Ek−μ\mathcal{E}_{\mathbf{k}}-\mu, which yields

J^Qμ=1V∑kc^k†JQμc^k=1V∑kc^k†(JQ0μ+JQ1μ+JQ2μ)c^k,(38)\hat{J}_{Q}^{\mu}=\frac{1}{V}\sum_{\mathbf{k}}\hat{c}^{\dagger}_{\mathbf{k}}\mathcal{J}_{Q}^{\mu}\hat{c}_{\mathbf{k}}=\frac{1}{V}\sum_{\mathbf{k}}\hat{c}^{\dagger}_{\mathbf{k}}\left(\mathcal{J}_{Q0}^{\mu}+\mathcal{J}_{Q1}^{\mu}+\mathcal{J}_{Q2}^{\mu}\right)\hat{c}_{\mathbf{k}}, \tag*{(38)}

with

JQ0μ(k)=1ℏg~μ,(39)\mathcal{J}_{Q0}^{\mu}(\mathbf{k})=\frac{1}{\hbar}\tilde{g}^{\mu}, \tag*{(39)}
JQ1μ(k)=1ℏ∫dω2πieℏωEα(ω)e−iωtg~μα,(40)\mathcal{J}_{Q1}^{\mu}(\mathbf{k})=\frac{1}{\hbar}\int\frac{d\omega}{2\pi}\frac{ie}{\hbar\omega}E^{\alpha}(\omega)e^{-i\omega t}\tilde{g}^{\mu\alpha}, \tag*{(40)}
JQ2μ(k)=12ℏ∫dω12πieℏω1Eα(ω1)e−iω1t×∫dω22πieℏω2Eβ(ω2)e−iω2tg~μαβ,(41)\mathcal{J}_{Q2}^{\mu}(\mathbf{k})=\frac{1}{2\hbar}\int\frac{d\omega_{1}}{2\pi}\frac{ie}{\hbar\omega_{1}}E^{\alpha}(\omega_{1})e^{-i\omega_{1}t} \times\int\frac{d\omega_{2}}{2\pi}\frac{ie}{\hbar\omega_{2}}E^{\beta}(\omega_{2})e^{-i\omega_{2}t}\tilde{g}^{\mu\alpha\beta}, \tag*{(41)}

and

g~α1…αn=12Dα1…Dαn[(Ek−μ)2].(42)\tilde{g}^{\alpha_{1}\ldots\alpha_{n}}=\frac{1}{2}D^{\alpha_{1}}\ldots D^{\alpha_{n}}[(\mathcal{E}_{\mathbf{k}}-\mu)^{2}]. \tag*{(42)}

Our goal is to calculate the expectation value ⟨J^Qμ(t)⟩\langle\hat{J}_{Q}^{\mu}(t)\rangle. The heat current responses are characterized by the tensors αμα1α2…αn(t;t1,t2,…tn)\alpha^{\mu\alpha_{1}\alpha_{2}\ldots\alpha_{n}}(t;t_{1},t_{2},\ldots t_{n}) as

⟨J^Qμ(t)⟩=∑n=0∞1n!∫[∏k=1ndtkEαk(tk)]αμα1α2…αn(t;t1,t2,…tn).(43)\left\langle\hat{J}_{Q}^{\mu}(t)\right\rangle=\sum_{n=0}^{\infty}\frac{1}{n!}\int\left[\prod_{k=1}^{n}dt_{k}E^{\alpha_{k}}(t_{k})\right]\alpha^{\mu\alpha_{1}\alpha_{2}\ldots\alpha_{n}}(t;t_{1},t_{2},\ldots t_{n}). \tag*{(43)}

By Fourier transformation, we can write

⟨J^Qμ(ω)⟩=∫dt eiωt⟨J^Qμ(t)⟩=∑n=0∞1n!∫[∏k=1ndωk2πEαk(ωk)]αμα1α2…αn(ω;ω1,ω2,…ωn),(44)\begin{aligned} \left\langle\hat{J}_{Q}^{\mu}(\omega)\right\rangle=\int dt\,e^{i\omega t}\left\langle\hat{J}_{Q}^{\mu}(t)\right\rangle\\ &=\sum_{n=0}^{\infty}\frac{1}{n!}\int\left[\prod_{k=1}^{n}\frac{d\omega_{k}}{2\pi}E^{\alpha_{k}}(\omega_{k})\right]\alpha^{\mu\alpha_{1}\alpha_{2}\ldots\alpha_{n}}(\omega;\omega_{1},\omega_{2},\ldots\omega_{n}), \tag*{(44)} \end{aligned}
αμα1α2…αn(ω;ω1,ω2,…ωn)=∫dt2πeiωt∫[∏k=1ndtk2πe−iωktk]αμα1α2…αn(t;t1,t2,…tn).(45)\alpha^{\mu\alpha_{1}\alpha_{2}\ldots\alpha_{n}}(\omega;\omega_{1},\omega_{2},\ldots\omega_{n}) =\int\frac{dt}{2\pi}e^{i\omega t}\int\left[\prod_{k=1}^{n}\frac{dt_{k}}{2\pi}e^{-i\omega_{k}t_{k}}\right]\alpha^{\mu\alpha_{1}\alpha_{2}\ldots\alpha_{n}}(t;t_{1},t_{2},\ldots t_{n}). \tag*{(45)}

In particular, the second order response coefficient αμαβ(ω;ω1,ω2)\alpha^{\mu\alpha\beta}(\omega;\omega_{1},\omega_{2}) can be written with functional derivatives as

αμαβ(ω;ω1,ω2)=∫dt2π∫dt12π∫dt22πei(ωt−ω1t1−ω2t2)δδEα(t1)δδEβ(t2)⟨J^Qμ(t)⟩∣E=0.(46)\alpha^{\mu\alpha\beta}(\omega;\omega_{1},\omega_{2}) =\int\frac{dt}{2\pi}\int\frac{dt_{1}}{2\pi}\int\frac{dt_{2}}{2\pi}e^{i(\omega t-\omega_{1}t_{1}-\omega_{2}t_{2})} \frac{\delta}{\delta E^{\alpha}(t_{1})}\frac{\delta}{\delta E^{\beta}(t_{2})}\left\langle\hat{J}_{Q}^{\mu}(t)\right\rangle\bigg|_{E=0}. \tag*{(46)}

Note that we need to take functional derivatives in the time domain because we are considering nonlinear regime as in the case of the electric current responses [17].

We notice a formal similarity between the current operator and the energy/heat current operator. The current operator (Eq. (26)) is given by the kk-derivative of the Hamiltonian, while the energy current operator is the kk-derivative of the squared Hamiltonian. By replacing the Hamiltonian HH in the energy current operator with H−μH-\mu, we can also obtain the heat current operator. Because of this similarity, we can calculate the heat current response in the same manner as the one used for the shift current. Because the rest of the formulation is almost the same as Ref. [17], we only present the results in the next section.

Shift heat current

In this section, we present results for the second order dc responses of heat current with respect to an electric field with frequency Ω\Omega, which we call “shift heat current”. We derive a general expression for the shift heat current, and then reduce the expression to cases for time reversal invariant systems, and especially, time reversal symmetric two band systems.

General expression for shift heat current

We focus on the second order dc responses to an electric field with the frequency Ω\Omega. We also restrict ourselves to longitudinal responses, namely, αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) where η\eta is an infinitesimal positive quantity. First we calculate αxxx\alpha^{xxx} within the imaginary time formalism, and then continue the result to the real time expressions. From now on, we set ℏ=1\hbar=1 for simplicity.

The Feynman diagrams which contribute to the second order heat current responses are shown in Figure 2. After performing the Matsubara frequency summation, we obtain

Feynman diagrams which contribute to shift heat current

Figure 2. The Feynman diagrams which contribute to shift heat current. Solid lines with an arrow and wavy lines represent propagators of electrons and photons, respectively, and black dots with nn photons denote vertices for hα1…αnh^{\alpha_1\ldots\alpha_n} and cross dots with nn photons are the vertex for g~α1…αn\tilde{g}^{\alpha_1\ldots\alpha_n}

αxxx(2iη;Ω+iη,−Ω+iη)=1V∑ke2Ω2[∑ag~aaxxxfa+∑a,bg~abxxhbax(fabεab+Ω+iη+fabεab−Ω+iη)+∑a,bg~abxhbaxxfabεab+2iη+∑a,b,cg~acxhcbxhbaxεac+2iη(fabεab+Ω+iη+fcbεbc−Ω+iη+(Ω→−Ω))](47)\begin{aligned} \alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) &=\frac{1}{V}\sum_{\boldsymbol{k}}\frac{e^2}{\Omega^2}\left[\sum_a \tilde{g}^{xxx}_{aa}f_a+\sum_{a,b}\tilde{g}^{xx}_{ab}h^x_{ba}\left(\frac{f_{ab}}{\varepsilon_{ab}+\Omega+i\eta}+\frac{f_{ab}}{\varepsilon_{ab}-\Omega+i\eta}\right)+\sum_{a,b}\tilde{g}^x_{ab}h^{xx}_{ba}\frac{f_{ab}}{\varepsilon_{ab}+2i\eta}\right.\\ &\qquad\left.+\sum_{a,b,c}\frac{\tilde{g}^x_{ac}h^x_{cb}h^x_{ba}}{\varepsilon_{ac}+2i\eta}\left(\frac{f_{ab}}{\varepsilon_{ab}+\Omega+i\eta}+\frac{f_{cb}}{\varepsilon_{bc}-\Omega+i\eta}+(\Omega\to-\Omega)\right)\right] \tag*{(47)} \end{aligned}
=1V∑ke2Ω2[∑ag~aaxxxfa+∑a,bg~abxxhbax(fabεab+Ω+fabεab−Ω−iπfab(δ(εab−Ω)+δ(εab+Ω)))+∑a,bg~abxhbaxxεabfab+∑a,b,cg~acxhcbxhbaxεac(fabεab+Ω+fcbεbc−Ω)−iπ∑a,b,cg~acxhcbxhbaxεac+2iη(fabδ(εab+Ω)+fcbδ(εbc−Ω)+(Ω→−Ω))].(48)\begin{aligned} &=\frac{1}{V}\sum_{\boldsymbol{k}}\frac{e^2}{\Omega^2}\left[\sum_a\tilde{g}^{xxx}_{aa}f_a+\sum_{a,b}\tilde{g}^{xx}_{ab}h^x_{ba}\left(\frac{f_{ab}}{\varepsilon_{ab}+\Omega}+\frac{f_{ab}}{\varepsilon_{ab}-\Omega}-i\pi f_{ab}\bigl(\delta(\varepsilon_{ab}-\Omega)+\delta(\varepsilon_{ab}+\Omega)\bigr)\right)+\sum_{a,b}\frac{\tilde{g}^x_{ab}h^{xx}_{ba}}{\varepsilon_{ab}}f_{ab}\right.\\ &\qquad\left.+\sum_{a,b,c}\frac{\tilde{g}^x_{ac}h^x_{cb}h^x_{ba}}{\varepsilon_{ac}}\left(\frac{f_{ab}}{\varepsilon_{ab}+\Omega}+\frac{f_{cb}}{\varepsilon_{bc}-\Omega}\right)-i\pi\sum_{a,b,c}\frac{\tilde{g}^x_{ac}h^x_{cb}h^x_{ba}}{\varepsilon_{ac}+2i\eta}\left(f_{ab}\delta(\varepsilon_{ab}+\Omega)+f_{cb}\delta(\varepsilon_{bc}-\Omega)+(\Omega\to-\Omega)\right)\right]. \tag*{(48)} \end{aligned}

where εa=εk,a\varepsilon_a=\varepsilon_{\boldsymbol{k},a} is the band dispersion, fa=f(εa)f_a=f(\varepsilon_a), f(ε)=(exp⁡(β(ε−μ))+1)−1f(\varepsilon)=(\exp(\beta(\varepsilon-\mu))+1)^{-1}, fab=fa−fbf_{ab}=f_a-f_b, εab=εa−εb\varepsilon_{ab}=\varepsilon_a-\varepsilon_b and T,μT,\mu are the temperature and the chemical potential. In the last line, divergences due to terms such as 1/εab1/\varepsilon_{ab} with εa=εb\varepsilon_a=\varepsilon_b should be interpreted as zero. This is a general expression for shift heat current in noninteracing electronic systems. By replacing g~\tilde{g} with hh, we can see that the expression for αxxx\alpha^{xxx} reduces to the expression for the electric current response derived in Ref. [17]. The last term in Eq. (48) diverges as ∝1/η\propto1/\eta if TRS is broken. This term can be interpreted as the heat current analog of the injection current [6, 17, 20], and should explicitly depend on the scattering rate. We also note that the heat

current observed experimentally should include other contributions. For example, impurities and disorder can affect the momentum distribution of electrons resulting in finite ballistic contribution to the dc heat current as in the case of the electric current [4, 21, 39, 40]. Some of these contribution does not necessarily depend on the scattering rate explicitly [21, 39] as in the case of the side-jump contribution to the anomalous Hall effect [41]. Therefore, it would be difficult to fully distinguish the shift heat current contribution and other contribution due to the scattering in experiments.

Shift heat current under TRS

If the system has TRS, i.e., the Hamiltonian satisfies

H0(k)T=H0(−k),(49)H_0(\mathbf{k})^{T}=H_0(-\mathbf{k}), \tag*{(49)}

then the matrix element g~\tilde{g} and hh have the following symmetry:

(g~k(2n+1))T=−g~−k(2n+1),(50)\left(\tilde{g}_{\mathbf{k}}^{(2n+1)}\right)^{T}=-\tilde{g}_{-\mathbf{k}}^{(2n+1)}, \tag*{(50)}
(g~k(2n))T=g~−k(2n),(51)\left(\tilde{g}_{\mathbf{k}}^{(2n)}\right)^{T}=\tilde{g}_{-\mathbf{k}}^{(2n)}, \tag*{(51)}
(hk(2n+1))T=−h−k(2n+1),(52)\left(h_{\mathbf{k}}^{(2n+1)}\right)^{T}=-h_{-\mathbf{k}}^{(2n+1)}, \tag*{(52)}
(hk(2n))T=h−k(2n).(53)\left(h_{\mathbf{k}}^{(2n)}\right)^{T}=h_{-\mathbf{k}}^{(2n)}. \tag*{(53)}

With these symmetry properties, most of the terms in Eq. Eq. (48) cancel and only the terms with a delta-function remain. Specifically, αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) will be

αxxx(2iη;Ω+iη,−Ω+iη)=1V∑k−iπe2Ω2[∑a,bg~abxxhbaxfab(δ(εab−Ω)+δ(εab+Ω))+∑a,b,cg~acxhcbxhbax1εac+2iη(fabδ(εab+Ω)+fcbδ(εbc−Ω))+(Ω→−Ω)].(54)\begin{aligned} \alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) ={}&\frac{1}{V}\sum_{\mathbf{k}}\frac{-i\pi e^2}{\Omega^2}\left[\sum_{a,b}\tilde{g}_{ab}^{xx}h_{ba}^{x}f_{ab}\left(\delta(\varepsilon_{ab}-\Omega)+\delta(\varepsilon_{ab}+\Omega)\right)\right.\\ &\left.+\sum_{a,b,c}\tilde{g}_{ac}^{x}h_{cb}^{x}h_{ba}^{x}\frac{1}{\varepsilon_{ac}+2i\eta}\left(f_{ab}\delta(\varepsilon_{ab}+\Omega)+f_{cb}\delta(\varepsilon_{bc}-\Omega)\right)\right.\\ &\left.+(\Omega\rightarrow-\Omega)\right]. \tag*{(54)} \end{aligned}

Noting that the last term vanishes due to TRS when a=ca=c, one can remove the 2iη2i\eta in the denominator of the last term. By using TRS, this can be rewritten as

αxxx(2iη;Ω+iη,−Ω+iη)=1V∑kπe2Ω2[∑a,bIm⁡[g~abxxhbax]fab(δ(εab−Ω)+δ(εab+Ω))+∑a,b,cIm⁡[g~acxhcbxhbax]εac(fabδ(εab+Ω)+fcbδ(εbc−Ω)+(Ω→−Ω))]=1V∑k2πe2Ω2[∑a,bIm⁡[g~abxxhbax]fabδ(εab−Ω)+∑a,b,cIm⁡[g~acxhcbxhbax]fabεac(δ(εab+Ω)+δ(εab−Ω))].(55)\begin{aligned} \alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) ={}&\frac{1}{V}\sum_{\mathbf{k}}\frac{\pi e^2}{\Omega^2}\left[\sum_{a,b}\operatorname{Im}[\tilde{g}_{ab}^{xx}h_{ba}^{x}]f_{ab}\left(\delta(\varepsilon_{ab}-\Omega)+\delta(\varepsilon_{ab}+\Omega)\right)\right.\\ &\left.+\sum_{a,b,c}\frac{\operatorname{Im}[\tilde{g}_{ac}^{x}h_{cb}^{x}h_{ba}^{x}]}{\varepsilon_{ac}}\left(f_{ab}\delta(\varepsilon_{ab}+\Omega)+f_{cb}\delta(\varepsilon_{bc}-\Omega)+(\Omega\rightarrow-\Omega)\right)\right] \\ &=\frac{1}{V}\sum_{\mathbf{k}}\frac{2\pi e^2}{\Omega^2}\left[\sum_{a,b}\operatorname{Im}[\tilde{g}_{ab}^{xx}h_{ba}^{x}]f_{ab}\delta(\varepsilon_{ab}-\Omega)+\sum_{a,b,c}\operatorname{Im}[\tilde{g}_{ac}^{x}h_{cb}^{x}h_{ba}^{x}]\frac{f_{ab}}{\varepsilon_{ac}}\left(\delta(\varepsilon_{ab}+\Omega)+\delta(\varepsilon_{ab}-\Omega)\right)\right]. \tag*{(55)} \end{aligned}

One can also verify that the last term for a=ba=b or b=cb=c vanishes in the presence of TRS. It is clear that for a=ba=b the term vanishes due to fabf_{ab}. For b=cb=c, using the relation between g~x\tilde{g}^{x} and hxh^{x},

g~abx=εa+εb2habx,(56)\tilde{g}_{ab}^{x}=\frac{\varepsilon_a+\varepsilon_b}{2}h_{ab}^{x}, \tag*{(56)}

one can conclude Im⁡[g~abxhbbxhbax]∝Im⁡[habx∣hbbx∣2]=0\operatorname{Im}[\tilde{g}_{ab}^{x}h_{bb}^{x}h_{ba}^{x}]\propto\operatorname{Im}[h_{ab}^{x}|h_{bb}^{x}|^2]=0, and thus the b=cb=c terms vanish. Therefore, the last term in Eq. Eq. (55) represents three band contributions. In contrast, the first term corresponds to two band contributions.

Two band systems with TRS

Let us consider a two band model with TRS. Such description is justified when a TR symmetric system is effectively described by only two bands near the Fermi level. In this case, the three band contribution can be neglected, and we obtain

αxxx(2iη;Ω+iη,−Ω+iη)=1V∑k∑a,b2πe2Ω2Im⁡[g~abxhbax]fabδ(εab−Ω).(57)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta)=\frac{1}{V}\sum_{\mathbf{k}}\sum_{a,b}\frac{2\pi e^2}{\Omega^2}\operatorname{Im}[\tilde{g}_{ab}^{x}h_{ba}^{x}]f_{ab}\delta(\varepsilon_{ab}-\Omega). \tag*{(57)}

By using the Leibniz rule for the covariant derivative (Eq. Eq. (16)), one can verify

g~abxx=12(DxDx[(εk−μ)2])ab=(εa+εb2−μ)habxx+(hxhx)ab.(58)\tilde{g}_{ab}^{xx}=\frac{1}{2}\left(\mathcal{D}^{x}\mathcal{D}^{x}[(\varepsilon_k-\mu)^2]\right)_{ab} =\left(\frac{\varepsilon_a+\varepsilon_b}{2}-\mu\right)h_{ab}^{xx}+(h^{x}h^{x})_{ab}. \tag*{(58)}

Because we are considering a two band case, for a≠ba\ne b Im⁡[g~abxxhbax]\operatorname{Im}[\tilde{g}_{ab}^{xx}h_{ba}^{x}] can be rewritten as

Im⁡[g~abxxhbax]=(εa+εb2−μ)Im⁡[habxxhbax]+∑cIm⁡[hacxhcbxhbax]=(εa+εb2−μ)Im⁡[habxxhbax]+Im⁡[haaxhabxhbax]+Im⁡[habxhbbxhbax]=(εa+εb2−μ)Im⁡[habxxhbax]+Im⁡[(haax+hbbx)∣habx∣2]=(εa+εb2−μ)Im⁡[habxxhbax].(59)\begin{aligned} \operatorname{Im}[\tilde{g}_{ab}^{xx}h_{ba}^{x}] ={}&\left(\frac{\varepsilon_a+\varepsilon_b}{2}-\mu\right)\operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}]+\sum_c\operatorname{Im}[h_{ac}^{x}h_{cb}^{x}h_{ba}^{x}]\\ ={}&\left(\frac{\varepsilon_a+\varepsilon_b}{2}-\mu\right)\operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}]+\operatorname{Im}[h_{aa}^{x}h_{ab}^{x}h_{ba}^{x}]+\operatorname{Im}[h_{ab}^{x}h_{bb}^{x}h_{ba}^{x}]\\ ={}&\left(\frac{\varepsilon_a+\varepsilon_b}{2}-\mu\right)\operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}]+\operatorname{Im}[(h_{aa}^{x}+h_{bb}^{x})|h_{ab}^{x}|^2]\\ ={}&\left(\frac{\varepsilon_a+\varepsilon_b}{2}-\mu\right)\operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}]. \tag*{(59)} \end{aligned}

Therefore, αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) is given by

αxxx(2iη;Ω+iη,−Ω+iη)=1V∑k∑a,b2πe2Ω2(εa+εb2−μ)Im⁡[habxxhbax]fabδ(εab−Ω).(60)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta)=\frac{1}{V}\sum_{\mathbf{k}}\sum_{a,b}\frac{2\pi e^2}{\Omega^2}\left(\frac{\varepsilon_a+\varepsilon_b}{2}-\mu\right)\operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}]f_{ab}\delta(\varepsilon_{ab}-\Omega). \tag*{(60)}

Furthermore, the matrix element Im⁡[habxxhbax]\operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}] in a two band system can be transformed into [9, 17]

Im⁡[habxxhbax]=Im⁡[(∂kxhabx−i[Ax,hx]ab)hbax]=Im⁡[∣habx∣2(∂kxlog⁡habx−i(Aaax−Abbx)+iAabxhbax(haax−hbbx))]=Im⁡[∣habx∣2(∂kxlog⁡habx−i(Aaax−Abbx)+∣habx∣2ϵab(haax−hbbx))]=∣habx∣2(∂kxϕabx−(Aaax−Abbx))=∣habx∣2Rabx.(61)\begin{aligned} \operatorname{Im}[h_{ab}^{xx}h_{ba}^{x}] &= \operatorname{Im}\left[(\partial_{k_x}h_{ab}^{x}-i[\mathcal{A}^{x},h^{x}]_{ab})h_{ba}^{x}\right] \\ &= \operatorname{Im}\left[|h_{ab}^{x}|^{2}\left(\partial_{k_x}\log h_{ab}^{x}-i(\mathcal{A}_{aa}^{x}-\mathcal{A}_{bb}^{x})+i\mathcal{A}_{ab}^{x}h_{ba}^{x}(h_{aa}^{x}-h_{bb}^{x})\right)\right] \\ &= \operatorname{Im}\left[|h_{ab}^{x}|^{2}\left(\partial_{k_x}\log h_{ab}^{x}-i(\mathcal{A}_{aa}^{x}-\mathcal{A}_{bb}^{x})+\frac{|h_{ab}^{x}|^{2}}{\epsilon_{ab}}(h_{aa}^{x}-h_{bb}^{x})\right)\right] \\ &= |h_{ab}^{x}|^{2}\left(\partial_{k_x}\phi_{ab}^{x}-(\mathcal{A}_{aa}^{x}-\mathcal{A}_{bb}^{x})\right) \\ &= |h_{ab}^{x}|^{2}R_{ab}^{x}. \tag*{(61)} \end{aligned}

where ϕabx=Im⁡log⁡habx\phi_{ab}^{x}=\operatorname{Im}\log h_{ab}^{x}, and Rabx=∂kxϕabx−(Aaax−Abbx)R_{ab}^{x}=\partial_{k_x}\phi_{ab}^{x}-(\mathcal{A}_{aa}^{x}-\mathcal{A}_{bb}^{x}) is the quantity called shift vector, which also appears in the expression of the shift current [6]. Therefore, the expression for the response coefficient is simplified to the following form:

αxxx(2iη;Ω+iη,−Ω+iη)=1V∑k∑a,b2πe2Ω2(ϵa+ϵb2−μ)∣habx∣2Rabxfabδ(ϵab−Ω).(62)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta)=\frac{1}{V}\sum_{k}\sum_{a,b}\frac{2\pi e^{2}}{\Omega^{2}}\left(\frac{\epsilon_{a}+\epsilon_{b}}{2}-\mu\right)|h_{ab}^{x}|^{2}R_{ab}^{x}f_{ab}\delta(\epsilon_{ab}-\Omega). \tag*{(62)}

This is almost the same form as the one for the shift current, σxxx(2iη;Ω+iη,−Ω+iη)\sigma^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta):

σxxx(2iη;Ω+iη,−Ω+iη)=1V∑k∑a,b2πe3Ω2∣habx∣2Rabxfabδ(ϵab−Ω).(63)\sigma^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta)=\frac{1}{V}\sum_{k}\sum_{a,b}\frac{2\pi e^{3}}{\Omega^{2}}|h_{ab}^{x}|^{2}R_{ab}^{x}f_{ab}\delta(\epsilon_{ab}-\Omega). \tag*{(63)}

They only differ by the factor (ϵa+ϵb)/2−μ(\epsilon_{a}+\epsilon_{b})/2-\mu in α\alpha. From these expressions, one can interpret αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) as the dc heat current that originates from the difference in the intracell coordinate between conduction and valence bands. Since this coordinate difference is represented by the shift vector RabxR_{ab}^{x}, we call this nonlinear thermal response as "shift heat current". The shift heat current does not explicitly depend on the scattering rate as in the case of the expression for the shift current Eq. (63). The independence of the scattering rate of the shift heat current might be useful to distinguish from other contributions that explicitly depend on the scattering rate. However, we again note that some contributions due to impurities and/or disorder do not necessarily depend on the scattering rate [21, 39] and thus it would be difficult to fully distinguish the shift heat current contribution from those due to the scattering in experiments.

Shift heat current in Rice-Mele model

In this section, to exemplify our theory, we present a numerical calculation of the shift heat current for Rice-Mele model [42] which is a representative 1D model with broken inversion symmetry. The Rice-Mele model can be used to describe, for example, one dimensional dimerized systems [42, 43], single-layer monochalcogenides [44], and Perovskite materials [45]. The Hamiltonian for the Rice-Mele model is given by

H^=∑n(tABc^nB†c^nA†+tBAc^n+1,A†c^nB+h.c.)+∑n(tAAc^n+1,A†c^nA+tBBc^n+1,B†c^nB+h.c.)+∑n[(ϵ0+Δ2)c^nA†c^nA+(ϵ0−Δ2)c^nB†c^nB].(64)\hat{H}=\sum_{n}\left(t_{AB}\hat{c}_{nB}^{\dagger}\hat{c}_{nA}^{\dagger}+t_{BA}\hat{c}_{n+1,A}^{\dagger}\hat{c}_{nB}+ \mathrm{h.c.}\right) +\sum_{n}\left(t_{AA}\hat{c}_{n+1,A}^{\dagger}\hat{c}_{nA}+t_{BB}\hat{c}_{n+1,B}^{\dagger}\hat{c}_{nB}+\mathrm{h.c.}\right) +\sum_{n}\left[\left(\epsilon_{0}+\frac{\Delta}{2}\right)\hat{c}_{nA}^{\dagger}\hat{c}_{nA}+\left(\epsilon_{0}-\frac{\Delta}{2}\right)\hat{c}_{nB}^{\dagger}\hat{c}_{nB}\right]. \tag*{(64)}

Here nn is the index for unit cells, AA and BB represent the two sites in a unit cell, and their positions in the unit cell are given by rA,rBr_{A},r_{B}. The system breaks the inversion symmetry when, for example, Δ≠0\Delta\ne0 and tAB≠tBAt_{AB}\ne t_{BA} and thus nonvanishing shift current and heat current appear. The second-nearest-neighbor hopping is necessary to break particle-hole symmetry; otherwise heat current becomes zero (see Eq. (62)). We show a schematic picture of Rice-Mele model in Figure 3.

Image not extracted

Figure 3. An illustration of Rice-Mele model (Eq. (64)). In order to break the particle-hole symmetry, next-nearest-neighbor hoppings tAA,tBBt_{AA},t_{BB} are included in the model.

We perform the numerical calculation of αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) by directly applying the expression Eq. (47). The energy broadening η\eta needs to be large enough compared to the spacing between adjacent energy levels Δϵ\Delta\epsilon but small enough compared to the other energy scale. Δϵ\Delta\epsilon can be estimated as Δϵ∼t/N\Delta\epsilon\sim t/N with the energy scale of the system tt and the number of unit cell NN, and thus the condition for η\eta is t/N≪η≪tt/N\ll\eta\ll t. We also perform the calculation for σxxx(2iη;Ω+iη,−Ω+iη)\sigma^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) in the same manner.

In Figure 4, we show the band structure, the response coefficients αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) and σxxx(2iη;Ω+iη,−Ω+iη)\sigma^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) for a parameter set and several chemical potentials μ\mu shown in the caption of the Figure 4. We set the second-nearest neighbor hopping tAA=tBB=0.1tt_{AA}=t_{BB}=0.1t, and the other parameters are set following [45] where ferroelectric Perovskite BaTiO3_3 is discussed with the Rice-Mele model (without second-nearest-neighbor hopping) and the Hubbard interaction. One can see from Figure 4 that both the shift heat current and the shift heat current becomes large at the band gap around ℏΩ∼t\hbar\Omega\sim t as expected from Eq. (62) and Eq. (63). When μ\mu is set to zero, i.e., the middle of the conduction band and the valence band, the shift heat current is small reflecting the factor of (ϵa+ϵb)/2−μ(\epsilon_{a}+\epsilon_{b})/2-\mu in Eq. (55). By varying μ\mu, the shift heat current changes even its sign. We discuss on the sign and the chemical potential dependence of the shift heat current in Sec. V B and Sec. V C.

Image not extracted

Figure 4. The shift heat current and the shift current for the Rice-Mele model. We used the parameters: tAB=0.9tt_{AB}=0.9t, tBA=1.1tt_{BA}=1.1t, tAA=0.1tt_{AA}=0.1t, tBB=0.1tt_{BB}=0.1t, rA=0r_A=0, rB=0.5ar_B=0.5a, Δ=2t\Delta=2t, ε0=−0.20\varepsilon_0=-0.20 with lattice constant aa and an energy scale tt. ε0\varepsilon_0 is determined so that the middle of the minimum of the conduction band and the maximum of the valence band is zero. The temperature TT is set to be zero (T=0T=0). The total number of the unit cells NN is set to be 10001, and we set η=0.01t\eta=0.01t so that t/N≪η≪tt/N\ll\eta\ll t with the energy scale tt. (a) The band structure of the Rice-Mele model for the given parameters. The black, red and blue dashed lines correspond to the chemical potential μ=0\mu=0, −0.1t-0.1t, 0.1t0.1t respectively. (b) The response coefficients for shift heat current αxxx(2iη;Ω+iη,−Ω+iη)\alpha^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) and shift current σxxx(2iη;Ω+iη,−Ω+iη)\sigma^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta) as a function of the photon energy ℏΩ\hbar\Omega. αxxx\alpha^{xxx} is calculated for μ=0,−0.1t,0.1t\mu=0,-0.1t,0.1t and σxxx\sigma^{xxx} is calculated for μ=0\mu=0.

Phonon-induced shift heat current

Based on the diagrammatic method that we established for calculation of general heat current responses, in this section, we apply the formalism to electron-phonon coupled systems. Recently, it is experimentally shown that the optical excitation of phonons in a semiconductor induces large shift current responses in THz regions [38]. Specifically, a semiconductor BaTiO3_3 shows large second order responses to an electric field with frequency much smaller than its band gap energy. This can be understood as a phonon-induced shift current [38].

The physical picture for the phonon-induced shift current is as follows. In general, excitation of phonons in noncentrosymmetric systems is accompanied by finite electronic polarization PP due to the electron-phonon coupling. When phonons are excited by illumination of light, the number of phonons will increase in time, and thus the polarization PP also increases accordingly even at steady state in noncentrosymmetric systems. Since the polarization PP is related to an electric current JJ through J=dP/dtJ=dP/dt, the excitation of phonons results in a dc electric current. A more detailed theoretical description is given in Ref. [38]. It should be noted that the electrons are excited only virtually in this mechanism. This is in sharp contrast to, for example, a proposal in Ref. [46], where phonons create real excitations of electrons in a narrow gap quantum well. If the band gap of the system is sufficiently large compared to the energy of phonons, the phonon cannot create real excitations of electrons, in which case the contributions studied in Ref. [38] and the present paper are dominant.

From the analogy between shift current and shift heat current, we can also expect that the shift heat current can be induced by phonon excitations. Namely, we expect "electronic heat polarization" PQP_Q is induced through electron-phonon coupling along with PP when phonons are excited, and it also increases in time, resulting in finite dc heat current JQ=dPQ/dtJ_Q=dP_Q/dt. Following Ref. [38], let us calculate the phonon-induced shift heat current.

We consider only one mode of phonon with wavevector q=0\boldsymbol{q}=0 for simplicity. The Hamiltonian for the phonon is

H^ph=εpha^†a^,(65)\hat{H}_{\mathrm{ph}}=\varepsilon_{\mathrm{ph}}\hat{a}^{\dagger}\hat{a}, \tag*{(65)}

where a^\hat{a} (a^†\hat{a}^{\dagger}) is the annihilation (creation) operator of the phonon, and εph\varepsilon_{\mathrm{ph}} is the energy of the phonon. The electronphonon interaction is described by the following Hamiltonian:

H^el−ph=∑kc^k†λkVc^k(a^†+a^),(66)\hat{H}_{\mathrm{el-ph}}=\sum_k \hat{c}_k^\dagger\frac{\lambda_k}{\sqrt{V}}\hat{c}_k(\hat{a}^\dagger+\hat{a}), \tag*{(66)}

where λk\lambda_k is an s×ss \times s hermitian matrix (kk in λk\lambda_k is often suppressed below). The diagrams corresponding to phonon-induced shift heat current are shown in Figure 5. By calculating these diagrams, αphxxx(iω;iω1,iω2)\alpha_{\mathrm{ph}}^{xxx}(i\omega;i\omega_1,i\omega_2) is obtained as

Image not extracted

Figure 5. The Feynman diagrams which describe the phonon-induced shift heat current. Here, curly lines are phonon propagators and open dots are the electron-phonon interaction. Solid lines represent propagators of electrons. Black dots and crossdots represent vertices for hh and g~\tilde{g} as in Figure 2.

αphxxx(iω;iω1,iω2)=A(iω1)D(iω1)B1(iω1,iω2)+A(iω1)D(iω1)B2(iω1,iω2)+(iω1↔iω2),(67)\begin{aligned} \alpha_{\mathrm{ph}}^{xxx}(i\omega;i\omega_1,i\omega_2)={}&A(i\omega_1)D(i\omega_1)B_1(i\omega_1,i\omega_2)\\ &+A(i\omega_1)D(i\omega_1)B_2(i\omega_1,i\omega_2)\\ &+(i\omega_1\leftrightarrow i\omega_2), \tag*{(67)} \end{aligned}

with

A(iω1)=1V∑k∑a,b,εnieiω1habxGa(iεn+iω1)Gb(iεn)λba=1V∑k∑a,bieiω1λabhabxfabba(iω1),(68)\begin{aligned} A(i\omega_1)={}&\frac{1}{V}\sum_k\sum_{a,b,\varepsilon_n}\frac{ie}{i\omega_1}h_{ab}^xG_a(i\varepsilon_n+i\omega_1)G_b(i\varepsilon_n)\lambda_{ba}\\ ={}&\frac{1}{V}\sum_k\sum_{a,b}\frac{ie}{i\omega_1}\lambda_{ab}h_{ab}^x f_{ab}^{ba}(i\omega_1), \tag*{(68)} \end{aligned}
B1(iω1,iω2)=1V∑k∑a,b,εnλabGa(iεn+iω1)ieiω2g~baxxGb(iεn)=1V∑k∑a,bieiω2λabg~baxxfbaba(iω1),(69)\begin{aligned} B_1(i\omega_1,i\omega_2)={}&\frac{1}{V}\sum_k\sum_{a,b,\varepsilon_n}\lambda_{ab}G_a(i\varepsilon_n+i\omega_1)\frac{ie}{i\omega_2}\widetilde{g}_{ba}^{xx}G_b(i\varepsilon_n)\\ ={}&\frac{1}{V}\sum_k\sum_{a,b}\frac{ie}{i\omega_2}\lambda_{ab}\widetilde{g}_{ba}^{xx}f_{ba}^{ba}(i\omega_1), \tag*{(69)} \end{aligned}
B2(iω1,iω2)=1V∑k∑a,b,c,εnieiω2Ga(iεn)g~acxGc(iεn+iω)λcbGb(iεn+iω2)hbax+1V∑k∑a,b,c,εnieiω2Ga(iεn)g~caxGc(iεn−iω)λbcGb(iεn−iω2)habx−1V∑k∑a,b,cieiω2[g~acxλcbhbaxfababc(iω2,iω1)+g~caxλbchabxfababc(−iω2,−iω1)],(70)\begin{aligned} B_2(i\omega_1,i\omega_2)={}&\frac{1}{V}\sum_k\sum_{a,b,c,\varepsilon_n}\frac{ie}{i\omega_2}G_a(i\varepsilon_n)\widetilde{g}_{ac}^xG_c(i\varepsilon_n+i\omega)\lambda_{cb}G_b(i\varepsilon_n+i\omega_2)h_{ba}^x\\ &+\frac{1}{V}\sum_k\sum_{a,b,c,\varepsilon_n}\frac{ie}{i\omega_2}G_a(i\varepsilon_n)\widetilde{g}_{ca}^xG_c(i\varepsilon_n-i\omega)\lambda_{bc}G_b(i\varepsilon_n-i\omega_2)h_{ab}^x\\ &-\frac{1}{V}\sum_k\sum_{a,b,c}\frac{ie}{i\omega_2}\left[\widetilde{g}_{ac}^x\lambda_{cb}h_{ba}^x f_{ab}^{abc}(i\omega_2,i\omega_1)\right.\\ &\left.\qquad\qquad+\widetilde{g}_{ca}^x\lambda_{bc}h_{ab}^x f_{ab}^{abc}(-i\omega_2,-i\omega_1)\right], \tag*{(70)} \end{aligned}

and D(iω)D(i\omega) is the propagator of the phonon,

D(iω)=1iω−εph−1iω+εph.(71)D(i\omega)=\frac{1}{i\omega-\varepsilon_{\mathrm{ph}}}-\frac{1}{i\omega+\varepsilon_{\mathrm{ph}}}. \tag*{(71)}

By analytic continuation iω1→Ω+iηi\omega_1\to\Omega+i\eta, iω2→−Ω+iηi\omega_2\to-\Omega+i\eta, iω→2iηi\omega\to2i\eta, we obtain the phonon-induced shift heat current. In the following, we assume that Ω(>0)\Omega(>0) is much smaller than the band gap and of the order of the phonon energy, εph\varepsilon_{\mathrm{ph}}.

In the presence of TRS, λk\lambda_k satisfies

λkT=λ−k.(72)\lambda_k^T=\lambda_{-k}. \tag*{(72)}

After the analytic continuation, the following relations hold:

A(Ω)=A(Ω+iη)=A(−Ω+iη)=A(−Ω),(73)A(\Omega)=A(\Omega+i\eta)=A(-\Omega+i\eta)=A(-\Omega), \tag*{(73)}
Bi(Ω,−Ω)=Bi(Ω+iη,−Ω+iη)=−Bi(−Ω+iη,Ω+iη)=−Bi(−Ω,Ω)(i=1,2),(74)\begin{aligned} B_i(\Omega,-\Omega)&=B_i(\Omega+i\eta,-\Omega+i\eta)\\ &=-B_i(-\Omega+i\eta,\Omega+i\eta)\\ &=-B_i(-\Omega,\Omega)\qquad(i=1,2), \tag*{(74)} \end{aligned}

for infinitesimal η\eta. Therefore, the expression for αphxxx\alpha_{\mathrm{ph}}^{xxx} after the analytic continuation reduces to

αphxxx(2iη;Ω+iη,−Ω+iη)=−2πiδ(Ω−εph)A(Ω)(B1(Ω,−Ω)+B2(Ω,−Ω)).(75)\alpha_{\mathrm{ph}}^{xxx}(2i\eta;\Omega+i\eta,-\Omega+i\eta)=-2\pi i\delta(\Omega-\varepsilon_{\mathrm{ph}})A(\Omega)(B_1(\Omega,-\Omega)+B_2(\Omega,-\Omega)). \tag*{(75)}

For simplicity, we consider a two-band case with TRS below. By straightforward calculation, we obtain

A(Ω)=eV∑k∑a,bIm⁡[λabhbax]fbaεbaεba2−Ω2,(76)A(\Omega)=\frac{e}{V}\sum_k\sum_{a,b}\operatorname{Im}[\lambda_{ab}h_{ba}^x]\frac{f_{ba}\varepsilon_{ba}}{\varepsilon_{ba}^2-\Omega^2}, \tag*{(76)}
B1(Ω)=−ieVΩ∑k∑a,bRe⁡[λabg~baxx]fbaεbaεba2−Ω2,(77)B_1(\Omega)=-\frac{ie}{V\Omega}\sum_k\sum_{a,b}\operatorname{Re}[\lambda_{ab}\widetilde{g}_{ba}^{xx}]\frac{f_{ba}\varepsilon_{ba}}{\varepsilon_{ba}^2-\Omega^2}, \tag*{(77)}
B2(Ω)=2ieVΩ∑k∑a<bfab[1εab2−Ω2Re⁡[g~baxλab](haax−hbbx)+1εab2−Ω2Re⁡[g~baxhabx](λaa−λbb)+εab2+Ω2(εab2−Ω2)2Re⁡[λabhbax](g~aax−g~bbx)].(78)\begin{aligned} B_2(\Omega)=\frac{2ie}{V\Omega}\sum_k\sum_{a<b}f_{ab}\left[ \frac{1}{\varepsilon_{ab}^2-\Omega^2}\operatorname{Re}[\widetilde{g}_{ba}^x\lambda_{ab}](h_{aa}^x-h_{bb}^x)\right.\\ \left.+\frac{1}{\varepsilon_{ab}^2-\Omega^2}\operatorname{Re}[\widetilde{g}_{ba}^xh_{ab}^x](\lambda_{aa}-\lambda_{bb})\right.\\ \left.+\frac{\varepsilon_{ab}^2+\Omega^2}{(\varepsilon_{ab}^2-\Omega^2)^2}\operatorname{Re}[\lambda_{ab}h_{ba}^x](\widetilde{g}_{aa}^x-\widetilde{g}_{bb}^x)\right]. \tag*{(78)} \end{aligned}

This is almost of the same form as the phonon-induced shift current. One can obtain the expression for the phonon-induced shift current by replacing g~\widetilde{g} by hh.

Discussions

Symmetry condition to observe shift heat current

In order to observe shift heat current, the system needs to break the spatial inversion symmetry. This is obvious from symmetry consideration of the nonlinear response tensor, but we can also confirm explicitly that the response coefficient αxxx\alpha^{xxx} vanishes in centrosymmetric systems as follows. If the system preserves the spatial inversion symmetry, the Hamiltonian satisfies

H0(−k)=P†H0(k)P,(79)H_0(-k)=P^\dagger H_0(k)P, \tag*{(79)}

where PP is the unitary matrix which expresses the spatial inversion. For example, in the case of Rice-Mele model in § 3 C, P=σxP=\sigma_x where σx\sigma_x is the Pauli matrix. Under the inversion symmetry, the matrix element g~\widetilde{g} and hh have the following symmetry:

g~k(2n+1)=−g~−k(2n+1),(80)\widetilde{g}_k^{(2n+1)}=-\widetilde{g}_{-k}^{(2n+1)}, \tag*{(80)}
g~k(2n)=g~−k(2n),(81)\widetilde{g}_k^{(2n)}=\widetilde{g}_{-k}^{(2n)}, \tag*{(81)}
hk(2n+1)=−h−k(2n+1),(82)h_k^{(2n+1)}=-h_{-k}^{(2n+1)}, \tag*{(82)}
hk(2n)=h−k(2n).(83)h_k^{(2n)}=h_{-k}^{(2n)}. \tag*{(83)}

where g~k(m)\widetilde{g}_k^{(m)} denotes (1/2)(Dx)m[Ek2](1/2)(\mathcal{D}^x)^m[\mathcal{E}_k^2] and similar for hk(m)h_k^{(m)}. We also explicitly show the kk-dependence of g~\widetilde{g} and hh. Therefore, all the terms in Eq. (48) are odd in kk and vanish after the kk-summation. This means that the spatial inversion symmetry breaking is necessary to induce the second order response of the heat current, as in the case of the electric current.

One can also verify that the phonon-induced shift heat current vanishes under the inversion symmetry by using λk=λ−k\lambda_k = \lambda_{-k}.

Sign of shift heat current

The sign of shift heat current is determined by two factors. One is the shift vector or equivalently the polarity of the system, and the other is the factor of (εa+εb)/2−μ(\varepsilon_a + \varepsilon_b)/2 - \mu appearing in Eq. (62). The shift vector represents the inversion symmetry breaking of the system, and it also determines the direction of shift current. If the polarity of the system is reversed, the shift vector will be also reversed, hence shift heat current changes its sign. In noncentrosymmetric systems, the inversion symmetry breaking often results in finite polarization. In that case, the polarity of the system and the direction of shift heat current can be reversed by applying an electric field. This property is the same as shift current.

The other factor, (εa+εb)/2−μ(\varepsilon_a + \varepsilon_b)/2 - \mu, is characteristic to heat transport and its sign is determined in a similar way to the Seebeck coefficient. Let us consider the case of insulators with finite band gap. If (εa+εb)/2−μ<0(\varepsilon_a + \varepsilon_b)/2 - \mu< 0 where εa\varepsilon_a represents a conduction band while εb\varepsilon_b represents a valence band, the chemical potential is closer to the conduction band. Therefore, the situation is quite similar to a Seebeck effect where the dominant carrier is electron and the Seebeck coefficient is negative. However, we note that we are considering insulators here and thus there is almost no carriers nor the concept of the dominant carrier itself. A more appropriate interpretation may be the following: in the shift heat current, a pair of an electron and a hole excited by an incident photon carries the heat of (εa+εb)/2−μ(\varepsilon_a + \varepsilon_b)/2 - \mu in total. In this case, the number of excited electrons and holes are the same, but the net heat carried by them is finite.

Chemical potential and temperature dependence of shift heat current

As seen from the definition of the heat current Eq. (37), the heat current explicitly depends on the chemical potential μ\mu. In the case of linear dc responses in metals or doped semiconductors, only the carriers in levels near μ\mu contributes to the transport and the change of μ\mu results in change of both the electric current and the heat current. In contrast, shift current and shift heat current occur in insulators. In this case, as long as the change of Fermi distribution function due to the change of chemical potential is negligible, the shift current is almost independent of μ\mu while the shift heat current does depend on μ\mu, as seen from Eq. (63) and Eq. (62). Therefore, if the chemical potential is shifted by Δμ\Delta\mu because of, for example, impurities, the shift heat current changes by ∼−(Δμ/e)Je(shift)\sim-(\Delta\mu/e)J_e^{(\mathrm{shift})} with shift current Je(shift)J_e^{(\mathrm{shift})}. (Here we neglect the change of the shift current.) In other words, one can control the shift heat current by varying the chemical potential. Since the chemical potential in insulators strongly depends on the existence of impurities, we can also expect that a small amount of impurities induces a large shift of the chemical potential and thus drastically changes the magnitude of the shift heat current.

The temperature dependence of the shift heat current in insulators is determined almost only by the temperature dependence of the chemical potential. If the change of temperature ΔT\Delta T satisfies kBΔT≪EGk_{\mathrm{B}}\Delta T \ll E_{\mathrm{G}}, then the shift heat current Eq. (62) changes by ∼(Δμ/e)Je(shift)\sim(\Delta\mu/e)J_e^{(\mathrm{shift})}, where Δμ\Delta\mu is the change due to the temperature variation.

For example, if the density of states of conduction band and that of valence band are respectively given by Dc(ε+εc0)=AcεsD_c(\varepsilon+ \varepsilon_{c0}) = A_c\varepsilon^s and Dv(εv0−ε)=AvεsvD_v(\varepsilon_{v0} - \varepsilon) = A_v\varepsilon^{s_v} where εc0(v0)\varepsilon_{c0(v0)} is the bottom (top) of the conduction (valence) band and AcA_c, AvA_v, scs_c, svs_v are constants. If one further assumes that εc0−μ≫kBT\varepsilon_{c0} - \mu\gg k_{\mathrm{B}}T, μ−εv0≫kBT\mu- \varepsilon_{v0} \gg k_{\mathrm{B}}T and the charge neutrality condition

∫εc0∞f(ε)Dc(ε) dε=∫−∞εv0(1−f(ε))Dv(ε) dε,(84)\int_{\varepsilon_{c0}}^{\infty} f(\varepsilon)D_c(\varepsilon)\,\mathrm{d}\varepsilon= \int_{-\infty}^{\varepsilon_{v0}} (1-f(\varepsilon))D_v(\varepsilon)\,\mathrm{d}\varepsilon, \tag*{(84)}

then the temperature dependence of chemical potential is

μ≃εv0+εc02+kBT2[log⁡(AvΓ(sv+1)AcΓ(sc+1))+(sv−sc)log⁡kBT].(85)\mu\simeq\frac{\varepsilon_{v0}+\varepsilon_{c0}}{2}+\frac{k_{\mathrm{B}}T}{2}\left[\log\left(\frac{A_v\Gamma(s_v+1)}{A_c\Gamma(s_c+1)}\right)+(s_v-s_c)\log k_{\mathrm{B}}T\right]. \tag*{(85)}

Here, Γ(x)=∫0∞e−ttx−1 dt\Gamma(x) = \int_0^\infty e^{-t}t^{x-1}\,\mathrm{d}t is the Gamma function. In particular, if both the conduction and valence band are parabolic with effective mass mcm_c, mvm_v, then sc=sv=(d−2)/2s_c = s_v = (d-2)/2 with the spatial dimension dd and

μ≃εv0+εc02+d4kBTlog⁡mvmc.(86)\mu\simeq\frac{\varepsilon_{v0}+\varepsilon_{c0}}{2}+\frac{d}{4}k_{\mathrm{B}}T\log\frac{m_v}{m_c}. \tag*{(86)}

Therefore, Δμ∝ΔT\Delta\mu\propto\Delta T and thus the shift heat current is linearly dependent on the temperature in this case.

Candidate materials and order estimation

Materials that break inversion symmetry can support nonzero shift heat current responses. Because of the similarity between shift heat current (Eq. (62)) and shift current (Eq. (63)), one can expect that materials which exhibits large shift current responses also shows large shift heat current responses. Namely,

JQ(shift)∼tJe(shift),(87)J_Q^{(\mathrm{shift})} \sim t J_e^{(\mathrm{shift})}, \tag*{(87)}

where JQ(shift)J_Q^{(\mathrm{shift})} and Je(shift)J_e^{(\mathrm{shift})} are the shift heat current and the shift current, and tt is the characteristic energy scale corresponding to the factor (εa+εb)/2−μ(\varepsilon_a + \varepsilon_b)/2 - \mu in Eq. (62). As mentioned at the end of Sec. V B, the factor tt can be interpreted as a net heat carried by the excited electron and hole, and Eq. (87) is a generalization of Eq. (2) to the second order response. We can expect that the relation Eq. (87) holds also for general multiband systems with an appropriate modification of an expression for tt, because the general expression for αxxx\alpha^{xxx} (Eq. (48)) differs from σxxx\sigma^{xxx} only by g~\tilde{g} in place of hh.

One representative material which shows the shift current response is SbSI [16, 19]. SbSI exhibits shift current of the order of 0.1 nA when illuminated by a cw laser with ℏω=1.95\hbar\omega= 1.95 eV, and 10 µA at peak when irradiated by a pulsed laser with ℏω=2.05\hbar\omega= 2.05 eV and the power 0.6 µJ [16]. In this case, as the energy scale tt is estimated as t∼1t \sim1 eV, the expected shift heat current is JQ(shift)∼0.1J_Q^{(\mathrm{shift})} \sim0.1 nW for the cw excitation, while JQ(shift)∼10J_Q^{(\mathrm{shift})} \sim10 µW at peak is expected when the laser pulse is applied.

Another promising candidate is TaAs. TaAs is a Weyl semimetal with broken inversion symmetry that shows large shift current responses [18]. It is experimentally observed that a photocurrent ∼1\sim1 µA can be induced by a laser with wavelength 10.6 µm (117 meV) and power ∼100\sim100 mW. If we assume that the Weyl semimetal TaAs has Hamiltonian of the form H=v0k+vμνkμσνH = v_0 k + v_{\mu\nu} k_\mu\sigma_\nu where μ,ν\mu, \nu run over x,y,zx, y, z, then tt for photon with energy ℏω\hbar\omega can be estimated as t∼(v0/2v)ℏω−μt \sim(v_0/2v)\hbar\omega- \mu. Furthermore we estimate v02v∼0.2\frac{v_0}{2v} \sim0.2 and μ∼10\mu\sim10 meV [47], then t∼10t \sim10 meV and one obtains from Eq. (87),

JQ(shift)∼10 nW.(88)J_Q^{(\mathrm{shift})} \sim10\ \mathrm{nW}. \tag*{(88)}

As for the phonon-induced shift heat current, a material which exhibits a large shift current is promising as well. Since the expressions for the phonon-induced shift heat current is different from that of the electric current only by the factor g~\tilde{g} in the place of hh, Eq. (87) also holds for the phonon-induced currents. We emphasize that tt is determined by the energy scale of electrons even for the phonon-induced currents. In an experiment, it is observed that BaTiO3_3 shows a phonon-induced shift current as large as ∼10\sim10 µA [38]. Assuming t∼1t \sim1 eV for BaTiO3_3, the phonon-induced shift heat current is estimated as

JQ(shift,ph)∼10 μW,(89)J_Q^{(\mathrm{shift,ph})} \sim10\ \mu\mathrm{W}, \tag*{(89)}

which is quite large compared to the response of TaAs. We also note that by varying chemical potential, the shift heat current in the above materials can be changed by ∼(Δμ/e)Je(shift)\sim(\Delta\mu/e)J_e^{(\mathrm{shift})} as discussed in § 5 C.

In conclusion, we have established a diagrammatic formulation of the nonlinear heat current response to an ac electric field, and calculated the second order response, which we call shift heat current. We have derived the microscopic expression for the shift heat current and confirmed that the shift heat current is determined by the shift vector, as in the case of the shift current. The amplitude of shift heat current JQ(shift)J_Q^{(\mathrm{shift})} is roughly estimated as JQ(shift)∼(t/e)Je(shift)J_Q^{(\mathrm{shift})} \sim(t/e)J_e^{(\mathrm{shift})} where tt is the characteristic energy scale of electrons, and can be controlled by changing the chemical potential. We have also calculated the phonon-induced shift heat current and found that even for phonon-induced cases, JQ(shift,ph)∼(t/e)Je(shift,ph)J_Q^{(\mathrm{shift,ph})} \sim(t/e)J_e^{(\mathrm{shift,ph})} still holds and the amplitude of JQ(shift)J_Q^{(\mathrm{shift})} is determined by the energy scale of electrons, not that of phonons.

Acknowledgments

The authors are grateful to T. Sagawa, M. Hirschberger, and Y. Xu for useful comments and discussions. N.N. was supported by JST CREST Grant Number JPMJCR1874 and JPMJCR16F1, Japan, and JSPS KAKENHI Grant Number 18H03676. T.M. was supported by JST PRESTO (JPMJPR19L9) and JST CREST (JPMJCR19T3). Y.O. thanks the World-leading Innovative Graduate Study Program for Materials Research, Information, and Technology (MERIT-WINGS).

References

References

  1. [1]V. Belinicher, Physics Letters A 66, 213 (1978).
  2. [2]E. Ivchenko and G. Pikus, Pis’ma Zh. Eksp. Teor. Fiz. 27 (1978).
  3. [3]R. von Baltz and W. Kraut, Physical Review B 23, 5590 (1981).
  4. [4]V. Belinicher, E. Ivchenko, and B. Sturman, Zh. Eksp. Teor. Fiz. 83, 649 (1982).
  5. [5]D. F. Eaton, Science 253, 281 (1991).
  6. [6]J. E. Sipe and A. I. Shkrebtii, Physical Review B 61, 5337 (2000).
  7. [7]S. M. Young and A. M. Rappe, Physical Review Letters 109, 116601 (2012), arXiv:1202.3168.DOI
  8. [8]S. M. Young, F. Zheng, and A. M. Rappe, Physical Review Letters 109, 236601 (2012).
  9. [9]T. Morimoto and N. Nagaosa, Science Advances 2, e1501524 (2016), arXiv:1510.08112.DOI
  10. [10]L. Z. Tan, F. Zheng, S. M. Young, F. Wang, S. Liu, and A. M. Rappe, npj Computational Materials 2, 16026 (2016).
  11. [11]L. Wu, S. Patankar, T. Morimoto, N. L. Nair, E. Thewalt, A. Little, J. G. Analytis, J. E. Moore, and J. Orenstein, Nature Physics 13, 350 (2017), arXiv:1609.04894.DOI
  12. [12]A. M. Cook, B. M. Fregoso, F. de Juan, S. Coh, and J. E. Moore, Nature Communications 8, 14176 (2017), arXiv:1507.08677.DOI
  13. [13]N. Nagaosa and T. Morimoto, Advanced Materials 29, 1603345 (2017).
  14. [14]K. W. Kim, T. Morimoto, and N. Nagaosa, Physical Review B 95, 035134 (2017), arXiv:1607.03888.DOI
  15. [15]M. Nakamura, S. Horiuchi, F. Kagawa, N. Ogawa, T. Kurumaji, Y. Tokura, and M. Kawasaki, Nature Communications 8, 1 (2017).
  16. [16]N. Ogawa, M. Sotome, Y. Kaneko, M. Ogino, and Y. Tokura, Physical Review B 96, 241203(R) (2017).
  17. [17]D. E. Parker, T. Morimoto, J. Orenstein, and J. E. Moore, Physical Review B 99, 045121 (2019), arXiv:1807.09285.DOI
  18. [18]G. B. Osterhoudt, L. K. Diebel, M. J. Gray, X. Yang, J. Stanco, X. Huang, B. Shen, N. Ni, P. J. W. Moll, Y. Ran, and K. S. Burch, Nature Materials 18, 471 (2019), arXiv:1712.04951.DOI
  19. [19]M. Sotome, M. Nakamura, T. Fujioka, M. Ogino, Y. Kaneko, T. Morimoto, Y. Zhang, M. Kawasaki, N. Nagaosa, Y. Tokura, and N. Ogawa, Proceedings of the National Academy of Sciences 116, 1929 (2019).
  20. [20]I. Ahn, G.-Y. Guo, and N. Nagaosa, Physical Review X 10, 041041 (2020), arXiv:2006.06709.DOI
  21. [21]L. E. Golub, E. L. Ivchenko, and B. Spivak, Physical Review B 102, 085202 (2020), arXiv:2006.11323.DOI
  22. [22]A. Avdoshkin, V. Kozii, and J. E. Moore, Physical Review Letters 124, 196603 (2020), arXiv:1912.02193.DOI
  23. [23]K. Behnia, Fundamentals of thermoelectricity (Oxford University Press, Oxford, 2015).DOI
  24. [24]R. Kubo, M. Yokota, and S. Nakajima, Journal of the Physical Society of Japan 12, 1203 (1957).
  25. [25]R. Kubo, Journal of the Physical Society of Japan 12, 570 (1957).
  26. [26]J. M. Luttinger, Physical Review 135, A1505 (1964).
  27. [27]M. Ogata and H. Fukuyama, Journal of the Physical Society of Japan 88, 074703 (2019).
  28. [28]D. Sánchez and R. López, Comptes Rendus Physique 17, 1060 (2016), arXiv:1600.00855.
  29. [29]D. Sánchez and R. López, Physical Review Letters 110, 026804 (2013).
  30. [30]M. Leijnse, M. R. Wegewijs, and K. Flensberg, Physical Review B 82, 04512 (2010).
  31. [31]H. Karbaschi, J. Lovén, K. Courteaut, A. Wacker, and M. Leijnse, Physical Review B 94, 115414 (2016).
  32. [32]H. Thierschmann, R. Sánchez, B. Sothmann, F. Arnold, C. Heyn, W. Hansen, H. Buhmann, and L. W. Molenkamp, Nature Nanotechnology 10, 854 (2015).
  33. [33]R. López and D. Sánchez, Physical Review B 88, 045129 (2013), arXiv:1302.5557.DOI
  34. [34]P. Bhalla, Physical Review B 103, 115304 (2021).
  35. [35]T. Osada and A. Kiswandhi, Journal of the Physical Society of Japan 90, 053704 (2021).
  36. [36]Y. Gao and D. Xiao, Physical Review B 98, 060402(R) (2018).
  37. [37]S.-Y. Hwang, R. López, M. Lee, and D. Sánchez, Physical Review B 90, 115301 (2014), arXiv:1406.3187.DOI
  38. [38]Y. Okamura, T. Morimoto, N. Ogawa, Y. Kaneko, G.-Y. Guo, M. Nakamura, M. Kawasaki, N. Nagaosa, Y. Tokura, and Y. Takahashi, Proceedings of the National Academy of Sciences 119, 10 (2022), arXiv:2201.00133.DOI
  39. [39]V. I. Belinicher and B. I. Sturman, Soviet Physics Uspekhi 23, 199 (1980).
  40. [40]B. I. Sturman, Uspekhi Fizicheskikh Nauk 190, 441 (2020).
  41. [41]N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Reviews of Modern Physics 82, 1539 (2010).
  42. [42]M. J. Rice and E. J. Mele, Physical Review Letters 49, 1455 (1982).
  43. [43]S. Onoda, S. Murakami, and N. Nagaosa, Physical Review Letters 93, 167602 (2004), arXiv:0403532 [cond-mat].
  44. [44]T. Rangel, B. M. Fregoso, B. S. Mendoza, T. Morimoto, J. E. Moore, and J. B. Neaton, Physical Review Letters 119, 067402 (2017), arXiv:1610.06589.DOI
  45. [45]T. Egami, S. Ishihara, and M. Tachiki, Science 261, 1307 (1993).
  46. [46]G. V. Budkin and S. A. Tarasenko, New Journal of Physics 22, 013005 (2020), arXiv:1908.07252.DOI
  47. [47]C.-C. Lee, S.-Y. Xu, S.-M. Huang, D. S. Sanchez, I. Belopolski, G. Chang, G. Bian, N. Alidoust, H. Zheng, M. Neupane, B. Wang, A. Bansil, M. Z. Hasan, and H. Lin, Physical Review B 92, 235104 (2015).

Paper details

vlnc:W_39kasjg6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents