WorksPhysics and Astronomy
Nonlinear optical response with relaxation in the velocity gauge
Abstract
We develop a framework for the evaluation of the nonlinear optical conductivity of electrons in crystals, driven by homogeneous optical fields, taking into account dissipation effects within the relaxation time approximation (RTA). The present formalism is based on Hamiltonians with minimal light-matter coupling within the velocity gauge, where the electric field is expressed in terms of a vector potential. The nonlinear conductivities are obtained by perturbatively solving the equation of motion for the single-particle reduced density matrix, which includes dissipative effects via the relaxation time approximation. We present a perturbative method to evaluate the reference equilibrium density matrix, towards which dissipation relaxes the system, which in the velocity gauge adiabatically follows the applied vector potential, and is also broadened by dissipative effects. Comparing with previous formulations based on the length gauge, the present approach has the benefit of avoiding the cumbersome evaluation of derivatives of Bloch states with respect to momentum. We test the presented method against length-gauge results, and calculate second- and third-order responses in a graphene-like tight-binding model with a staggered sublattice potential and a periodic modulation. We find that the relaxation time approximation gives results that deviate both quantitatively and qualitatively from the simple inclusion of an imaginary damping term in the frequencies, referred to as adiabatic switching. At equal relaxation parameter, the leading above-gap injection contribution differs by a factor of two; below the gap, the numerical results indicate a finite imaginary dc response in the small-relaxation limit of the relaxation-time calculation. These results show that adiabatic switching does not generally reproduce relaxation through the equation of motion, even in the small-relaxation limit.
Introduction
The study of nonlinear optical responses has become fundamental in understanding and characterizing the electronic properties of crystalline materials. In fact, ever since second harmonic generation was first observed [1], frequency conversion and wave mixing have become important methods to probe solids and interfaces [2, 3]. In graphene, coherent four-wave mixing [4], third-harmonic generation [5], and electrical tuning of nonlinear resonances [6, 7, 8] have been observed. Additionally, there have been developments in high-harmonic generation in solids [9], anisotropic second-harmonic generation [10], and spatially dispersive photogalvanic currents in Weyl semimetals [11].
Semiclassical band dynamics provides useful intuition for intraband motion, while interband transitions and coherences require a quantum description, naturally formulated within a single-particle reduced density-matrix, [9]. Light-matter coupling can be described with different gauge choices, and for a homogeneous electric field it is common to employ either the length gauge (coupling via a scalar potential, with the position operator and the electric field) or the velocity gauge (coupling via the minimal substitution, with the momentum operator, the vector potential and the particle charge) [12, 13, 14, 15]. Formal approaches to obtain second-order responses have been developed [16, 17], as well as numerical length-gauge density-matrix methods [18]. For graphene, there have been studies addressing intraband and interband processes and how they lead to non-linear behavior [19]. Moreover, graphene has been extensively studied in the context of doping and relaxation [20, 21, 22, 23], photocurrents [24], high harmonics in extended and confined geometries [25, 26, 27], and plasmon-assisted second-harmonic generation [28].
The length and velocity gauges provide equivalent descriptions of optical response when the light–matter coupling is treated consistently, and their respective computational properties are well established [14, 15, 29]. Here we adopt the velocity gauge, in which a spatially uniform electromagnetic field is described by a time-dependent vector potential, , with . Unlike the length gauge, where the position operator introduces momentum derivatives acting on the density matrix, the velocity gauge preserves translational symmetry and allows the coherent dynamics to be expressed in terms of matrices that are local in momentum space. This considerably simplifies the perturbative expansion of the density matrix. In the presence of relaxation, however, the field dependence of the reference equilibrium state can reintroduce momentum derivatives of the density matrix into the calculation [15]. As we show in this work, these derivatives can be avoided within the relaxation-time approximation, independently of the microscopic details of the single-particle Hamiltonian.
When simulating optical response of solids, it is important to include dissipative effects, which relax the system towards thermodynamic equilibrium. The simplest approach in which this is taken into account, within perturbation theory, is via adiabatic switching (AS), where the time dependent perturbation is assumed to have the form , where describes the slow switching-on of the external field and effectively acts as a broadening term, mocking dissipation. Another, more justifiable and commonly used, approach is the relaxation-time approximation (RTA), in which a term of the form , is included in the equation of motion for the density matrix , with being an equilibrium density matrix, towards which the system tends to relax. It has been pointed out that the AS and RTA do not, in general, coincide [15, 22, 30]. In the latter approach, and in the velocity gauge, the equilibrium state towards which the system relaxes must follow the applied field and therefore acquires an explicit dependence on the vector potential in the velocity gauge. It is worth pointing out that the RTA has been justified for systems where dissipation is mocked via hybridization with fermionic baths, Butikker probes [30]. Coupling to these baths not only modifies the dynamics of the density matrix, but also changes the equilibrium state. Hence, the choice of relaxation mechanism employed in the dynamics of the system can play a significant role for stationary injection currents and resonant nonlinear responses.
The treatment of relaxation matters when nonlinear optics probes quantum geometry and topology [31, 32, 33]. The predicted quantization of the circular photogalvanic injection-rate trace under suitable conditions [34], and proposals to detect topology through high-harmonic generation [35], connect nonlinear signals to Bloch-state structure. A consistent treatment of relaxation is therefore essential for disentangling intrinsic contributions rooted in the quantum geometry of Bloch states from extrinsic, lifetime-dependent contributions arising from the nonequilibrium carrier distribution.
Here we incorporate relaxation in the equation of motion, for the density matrix, via the relaxation-time approximation (RTA), within the velocity gauge , using a reference equilibrium state that adiabatically follows the applied field. Compared with previous formulations based on the length gauge, the present formulation avoids cumbersome derivatives of the density matrix with respect to momentum. To illustrate the formalism, we consider models for which minimal coupling is implemented through a shift of crystal momentum, , allowing the perturbative coefficients of the Hamiltonian to be obtained directly from momentum derivatives of the unperturbed Bloch Hamiltonian. This choice simplifies the applications presented here, but is not a requirement of the general framework. We obtain recursive nonlinear-conductivity expressions to arbitrary order and a closed formula for the field expansion of the equilibrium density matrix in terms of divided differences of the occupation function. The latter accommodates broadened occupations, induced by hybridization with fermionic baths.
We apply the framework to a gapped honeycomb model, subject to a periodic potential, computing: second-harmonic generation, dc shift and injection currents, and third-harmonic generation. At finite damping, the results can differ markedly from adiabatic switching: at equal , the injection-current response has a contribution with an approximately twofold larger prefactor in the relaxation-time scheme, while discrepancies persist near multiphoton resonances in the third-order response even for small damping.
Section 2 introduces the formalism, Sec. 3 derives the equilibrium density matrix expansion, Sec. 4 presents the numerical results, and Sec. 5 summarizes our conclusions.
General Formalism
Free Hamiltonian and conventions
We consider a non-interacting electronic system with translational symmetry, which may describe either a lattice model or electrons moving in a continuum periodic potential. In either case, the single-particle Hamiltonian admits a Bloch representation. In a basis of localized orbitals, the Hamiltonian can be written as
where label the orbital types, and denote the positions of the unit cells, and and are the corresponding creation and annihilation operators. The coefficients are matrix elements of the single-particle Hamiltonian and are not restricted to nearest-neighbor hopping. For continuum systems, such a representation can be obtained by expanding the Hamiltonian in a Wannier basis, which may be taken to be complete if no band truncation is performed.
Denoting by the position of orbital relative to the origin of its unit cell, we introduce the reciprocal-space operators through the Fourier convention
where is the number of unit cells. With this convention, the Hamiltonian becomes
where
The inclusion of the intracell orbital positions in the Fourier phases is particularly convenient when describing the coupling to a spatially uniform electromagnetic field. For models in which the position operator is diagonal in the localized-orbital basis, this convention allows the light–matter coupling to be implemented through a simple shift of crystal momentum. However, in a general Wannier representation, the position operator may also have off-diagonal matrix elements, which must be accounted for when constructing the electromagnetic coupling. This distinction does not depend on whether the underlying system is described by a lattice or continuum Hamiltonian.
Perturbative series of the Hamiltonian and velocity
In the velocity gauge, a spatially uniform electric field is represented by a time-dependent vector potential . We denote by the corresponding single-particle Hamiltonian, expressed in a fixed Bloch basis, and define
where . Assuming an expansion in powers of the vector potential, the perturbation can be written as
where is the electron charge, repeated Cartesian indices are implicitly summed, and
defines the th-order light-matter coupling tensor. The field-dependent velocity operator is obtained from the derivative of the Hamiltonian with respect to the vector potential,
It therefore admits the expansion
In particular, is the th Cartesian component of the unperturbed velocity operator. For the class of models considered in the applications below, the coupling to a uniform vector potential is implemented through the momentum shift
This prescription is exact for continuum Hamiltonians with conventional minimal coupling in a fixed cell-periodic Bloch representation. It also follows from the Peierls substitution for lattice models with a diagonal position operator in the localized-orbital basis, where the hopping amplitudes transform as . When (11) holds, the coupling tensors reduce to ordinary momentum derivatives,
and the field-dependent velocity can equivalently be expressed as
More generally, Eq. (11) need not hold in an arbitrary Wannier or band representation. For truncated band or Wannier Hamiltonians, additional care is required to construct the electromagnetic coupling consistently with the projection. The corresponding velocity matrix elements are obtained through covariant momentum derivatives, which include the Berry connection associated with the chosen basis [15, 29].
Importantly, the perturbative formalism developed in this work does not rely on Eq. (12). Instead, the tensors defined in Eq. (8) enter the density-matrix recursion as the coefficients of the field-dependent Hamiltonian. Thus, while their explicit construction may involve Berry connections or covariant derivatives of Bloch states for truncated Wannier models, the resulting perturbative equations do not require momentum derivatives acting on the density matrix, even in the presence of relaxation.
Current density expansion
Both the velocity operator and the reduced density matrix must be expanded to obtain the nonlinear current. The expectation value of the current density is
where is the spatial dimension and , label Bloch bands. The single-particle reduced density matrix is defined by
with and the creation and annihilation operators for a Bloch state labelled , and the many-body density operator.
With the Fourier convention , the current is expanded in powers of the electric field as
For compactness, we denote the ordered tuple of input frequencies by , with and . The th-order conductivity tensor is defined by
Equation of motion
We account for dissipation using a relaxation-time approximation in the equation of motion for the reduced density matrix,
where is the relaxation energy ( is the corresponding relaxation time) and is the field-dependent equilibrium density matrix. This reference state is a function of the instantaneous Hamiltonian and therefore carries the same vector-potential dependence. An equation of this form can be obtained from an open-system description with Büttiker probes in the slow-driving, wide-band limit [30]. Coupling to these Büttiker probes also leads to a broadening of the single-particle energy levels in the equilibrium state. This can be taken into account via the Lorentzian-broadened Fermi–Dirac function
where is the Fermi–Dirac distribution,
with being the inverse temperature and the Fermi energy. Here we use the same parameter for the Lorentzian half-width and the relaxation energy in Eq. (18).
Perturbation Theory
To solve the equation of motion, we introduce the deviation from the instantaneous equilibrium state,
Because is a function of , it commutes with the instantaneous Hamiltonian. Consequently,
We solve Eq. (22) recursively in frequency space. Writing , with of order in the vector potential, we consider a perturbative treatment, so that . We define the expansion coefficients by
where the coefficients of the expansion obey the recursive relation
with being band-energy difference in the eigenbasis of the unperturbed Hamiltonian . Note that in the equation above we also introduced the notation . The coefficient is defined by expanding in the same way as in Eq. (23). Its time derivative produces the final term in Eq. (24).
Conductivity tensor
The instantaneous-equilibrium contribution to the current vanishes upon integration over the full Brillouin zone. Indeed, with , the identity
makes this contribution a total momentum derivative, whose integral vanishes by Brillouin-zone periodicity. Thus only the deviation contributes to the net current. With this in mind, inserting Equation (23) into (14) yields the nonlinear conductivity tensor
The trace is over band indices. For reference we write here the first order expression,
and the second order expression,
where all the sums are over band indices. To obtain the corresponding adiabatic-switching expressions, we omit the relaxation term from the equation of motion, and take the initial equilibrium occupations to be given by the unbroadened Fermi–Dirac distribution. In practice, one only has to first take in the expressions above, use in the equilibrium-density coefficients, and subsequently apply the frequency shifts to each input frequency. In particular, at second order, the denominator associated with the frequency becomes , instead of in the relaxation-time approximation. Consequently, the two prescriptions generally yield different nonlinear conductivities at finite .
Because the electric-field factors in Eq. (17) commute, the physical conductivity tensor must be symmetric under simultaneous permutations of the field indices and their associated frequencies. Since this symmetry is not manifest in Eq. (26), we introduce the permutation-symmetric conductivity tensor,
where we sum over all permutations of the field index-frequency pairs. This is the quantity that appears in the figures throughout this work.
The remaining task is therefore the computation of the equilibrium-density coefficients entering Eq. (24), which are independent of the input frequencies. Their systematic construction is developed in the next section.
Expansion of the Equilibrium Density Matrix
Expansion in
Equation (24) depends on the expansion coefficients of the adiabatically evolved equilibrium density matrix. We now derive a systematic procedure for obtaining these coefficients to arbitrary order. The equilibrium density matrix is
Here, is the Lorentzian-broadened occupation of Eq. (19), and . Following Appendix A, we express this matrix as , and the terms , containing insertions of the potential , have the following expansion:
where the dependence on the equilibrium distribution is contained in the permutation-symmetric coefficient below:
This coefficient constitutes the basic building block of the equilibrium-density expansion and is valid whenever the energies are distinct; the cases for coincident energies are handled by taking the corresponding continuous limit , which involves derivatives of .
Equation (32) is one of the central results of this work. It achieves two main objectives: it shows that the coefficient is written solely in terms of the unperturbed eigenenergies and the occupation function — eliminating the need to explicitly evaluate the covariant derivative acting on the density matrix, as was initially believed necessary [15] — and it provides a systematic way to incorporate bath broadening into the initial equilibrium state through the occupation function defined in Eq. (19), without derivatives of Bloch eigenvectors. Finally, although derived for a Bloch Hamiltonian, this matrix-function identity also holds for a general finite-dimensional Hermitian Hamiltonian.
First-order
To illustrate the result, let . The first-order equilibrium-density coefficient is
For , the ratio becomes , including the cases of degenerate bands.
Second-order
At second order, one contribution arises directly from , while the other involves two first-order couplings and a sum over intermediate bands:
Here is computed by plugging and into:
obtained by developing the summations and products in Eq. (32); and then taking the proper energy limits for coinciding energies, as explained above. At arbitrary order, these equilibrium-density coefficients enter the recursion in Eq. (24), yielding the conductivity in Eq. (26).
Numerical results
We evaluate the response for a nearest-neighbor tight-binding graphene-like model with a staggered sublattice potential and a periodic on-site modulation,
Here is the nearest-neighbor hopping amplitude, () on sublattice A (B), and sets the staggered sublattice energy difference. The periodic on-site modulation is
where controls its strength, and are the reciprocal-lattice vectors of the supercell, and is chosen so that the average on-site potential vanishes over the supercell. Periodic potentials are experimentally realized in graphene–hexagonal-boron-nitride superlattices and can reconstruct the electronic bands [36]. Such structures also exhibit photocurrent responses sensitive to the symmetry and geometry of the minibands [37]. We use the modulation as a minimal model of symmetry breaking and band reconstruction, without attempting a quantitative description of a moiré device. Figure 1 illustrates the modulated lattice.
Figure 1. Top: Graphene-like honeycomb lattice with a supercell and a periodic on-site potential. The general model also allows a sublattice energy difference . The pristine-graphene limit is obtained by setting . Bottom left: Energy bands along the ––– path. Bottom right: Density of states for the supercell model. The density of states is normalized to states per supercell. The values were obtained with and .
Unless stated otherwise, we use , , and . The length used in the conductivity units denotes the carbon–carbon bond length. For , the model reduces to pristine graphene, for which the primitive two-atom unit cell suffices.
The electronic band structure and corresponding density of states for the default parameters are shown in the bottom part of Figure 1. The supercell gives rise to eight bands, with an energy gap of approximately separating the valence and conduction bands. This spectrum provides a reference for the optical transitions discussed below. In particular, lies within the gap, whereas , considered in the third-order response, lies near the conduction-band edge.
Length- and velocity-gauge agreement with relaxation
For the first benchmark, we set and compare our velocity-gauge calculation with the length-gauge relaxation results of Passos et al. [15]. Both calculations include relaxation in the density-matrix equation of motion, Eq. (18). We take and without equilibrium spectral broadening, while retaining . Figure 2 shows close agreement between the real and imaginary parts of the two THG spectra, including the three-photon resonance at . The benchmark extends the adiabatic-switching gauge comparison of Ref. [15] to relaxation in the equation of motion: the two gauges give the same response when the reference equilibrium state is transformed consistently.
Figure 2. Third-harmonic conductivity of pristine graphene. Circles: length-gauge relaxation data of Passos et al. [15]; solid curves: present velocity-gauge result. Blue and red indicate the real and imaginary parts, respectively, for (left) and (right), with and . Vertical dashed lines mark the three-photon () and two-photon () reference frequencies. The occupations are unbroadened, and . The length-gauge relaxation data was rescaled to our units and symmetrization convention.
Second-order optical response
We now turn to the second-order response of gapped graphene with periodic modulation. The same model introduced above describes both second-harmonic generation, at output frequency , and the dc photogalvanic response, at zero output frequency. We compare the relaxation-time prescription of Eq. (18) with adiabatic switching [15, 29]. In the latter, the relaxation term is omitted from the equation of motion, the equilibrium occupations are unbroadened, and each input frequency is shifted as .
Second-harmonic generation
We first consider the SHG conductivity . For this calculation we take , , and . The spectrum displays features near the two-photon () and one-photon () interband resonances. Because the periodic modulation modifies the band structure, these frequencies serve as approximate reference scales. Figure 3 compares the real and imaginary parts of the SHG conductivity
Figure 3. Real (blue) and imaginary (red) parts of the SHG conductivity , calculated with relaxation (dotted lines with circular markers), relaxation with broadening (solid lines) and adiabatic switching (dashed lines). The left and right panels show and 0.001, respectively, for , and . The spectral features occur near the two- and one-photon resonances, signaled by vertical lines at and , respectively. Conductivities are given in units of [38].
obtained using adiabatic switching and the relaxation-time approximation with either ordinary Fermi–Dirac or Lorentzian-broadened occupations. At , the three calculations yield appreciably different peak amplitudes and line shapes, especially near the resonances. Comparing the two relaxation-time results isolates the effect of equilibrium-occupation broadening while keeping the dynamical relaxation prescription fixed. In this case, the effect of broadening is the reduction of the magnitudes of the extrema in both the real and imaginary parts, while also modifying the shapes of the curves. Relaxation rates of this order have been reported in both monolayer and bilayer graphene [26, 27, 39, 40], indicating that relaxation-induced corrections to the response can be significant under experimentally relevant conditions. As , agreement between the three calculations improves. At , the spectra nearly coincide over most of the frequency range, with residual differences concentrated near the one- and two-photon resonances. This comparison therefore highlights the sensitivity of the finite- SHG response to both the dynamical relaxation prescription and the broadening of equilibrium occupations.
DC photogalvanic response
We next consider the dc conductivity , returning to the default parameters and , with . In noncentrosymmetric systems, the dc photogalvanic response can contain a shift-current contribution, associated with charge displacement during optical transitions, and an injection-current contribution, associated with asymmetric populations of photoexcited carriers [17, 31, 34]. The supercell modulation lowers the symmetry of uniformly gapped graphene, allowing additional second-order tensor components. For this tensor component, the real part describes the shift-type response, while the imaginary part describes the helicity-dependent channel containing the injection contribution.
Figure 4 summarizes the dependence of the dc response on frequency and relaxation strength. The upper left panel shows the real part of the conductivity, associated with the shift-current contribution. A pronounced negative feature develops near the approximate optical threshold . At finite , the relaxation-time and adiabatic-switching prescriptions broaden this structure differently. As , both approach the same finite response away from the threshold, as shown by comparison with the clean-limit reference curve.
Figure 4. Second-order dc conductivity for relaxation through the density-matrix equation of motion (solid lines) and adiabatic switching (dashed lines). Upper left panel: real part for and , including a clean-limit reference. Upper right panel: for , , and ; the rescaled curves above the gap are consistent with leading and scalings for relaxation and adiabatic switching, respectively. Bottom panel: imaginary part below the gap for , , , and , shown on a symmetric logarithmic scale. The subgap curves indicate a finite, nonzero small- limit for relaxation and a vanishing response for adiabatic switching. The vertical line marks the approximate optical threshold ; here , , and . Conductivities are given in units of .
The upper right panel shows , which isolates the leading inverse-relaxation contribution. Above the optical threshold, the rescaled quantity approaches a finite limit as is decreased, consistent with an injection-current contribution. In this frequency region, away from the immediate vicinity of the threshold, the leading terms obey
where is the limiting injection coefficient in this comparison. Thus, at the same numerical value of , the leading relaxation-time contribution is approximately twice the adiabatic-switching one. The factor of two follows from the different dc denominators: the second-order density-matrix response contains in the relaxation-time prescription, whereas adiabatic switching gives after shifting both input frequencies [15]. The factor of two therefore refers to the leading above-gap term at equal .
The lower panel resolves the imaginary part below the optical gap, where the leading injection divergence is absent. The numerical results indicate a finite, nonzero small- limit for the relaxation-time response and a vanishing limit for adiabatic switching. For the subgap frequencies shown, this behavior is summarized by
with generally different from zero. This finite subgap term is distinct from the divergent above-gap injection contribution. The imaginary parts thus indicate different pointwise small- limits, even where the real parts converge to the same clean-limit response.
Third-order optical response
We finally turn to third-harmonic generation (THG), described by . Unlike the preceding second-order calculations, we place the chemical potential in the conduction band at to investigate the effects of relaxation and equilibrium broadening for a partially occupied band. With , this is slightly above the unmodulated conduction-band minimum .
Figure 5 compares the THG spectra for and . At the larger value of , the two prescriptions give appreciably different amplitudes and line shapes. Reducing improves agreement over most of the spectrum, but a substantial difference remains near the three-photon feature at . The inset resolves this narrow region, where the relaxation-time result has a sharp structure in both parts of the conductivity. Agreement is closer near the two-photon feature at . These frequencies provide reference scales for the modulated system. The comparison shows that small damping alone does not ensure agreement near multiphoton resonances.
Figure 5. Real (blue) and imaginary (red) parts of the third-harmonic generation conductivity, , calculated with relaxation and broadened occupations (solid lines) and adiabatic switching (dashed lines), for (left panel) and (right panel). The staggered potential and the chemical potential are fixed at and , respectively. Vertical dashed lines mark the three-photon () and two-photon () reference frequencies. The inset resolves the three-photon feature, where a marked difference remains at . Conductivities are expressed in units of [41].
Conclusion
We have derived a velocity-gauge formulation of nonlinear optical response with relaxation towards a field-dependent equilibrium state. The conductivity follows recursively from the density-matrix equation of motion, while the equilibrium-density -expansion coefficients are given in terms of the unperturbed spectra and the occupation function. This provides a calculation at arbitrary order, includes bath-broadened occupations, and requires no derivatives of the density matrix. Additionally, for tight-binding or continuum Hamiltonians, derivatives of Bloch eigenvectors are likewise avoided.
The graphene calculations establish two distinct results. First, length- and velocity-gauge conductivities agree when relaxation is treated consistently. Second, this gauge equivalence does not imply equivalence between relaxation and adiabatic switching. The latter two prescriptions give different harmonic-generation spectra at finite , with discrepancies still resolved near the three-photon resonance at the smallest considered.
The dc response makes the distinction particularly clear. Away from the optical threshold, the shift-current component approaches the same finite limit in both prescriptions. The leading above-gap injection contribution instead scales as for relaxation and for adiabatic switching. This factor of two refers to the same numerical value of and follows from the respective dc denominators. Below the gap, the numerical results indicate a finite, nonzero imaginary response in the small- limit of the relaxation-time calculation, whereas the adiabatic-switching result vanishes. Thus the difference extends beyond resonance broadening to the limiting dc response itself.
A relaxation prescription therefore forms part of the physical model of nonlinear response. Adiabatic switching cannot in general replace a relaxation term, and a small broadening parameter alone does not ensure that the two calculations agree. The present formulation makes the field dependence of the relaxation state explicit and provides a direct way to retain it in multiband velocity-gauge calculations.
Funding
The authors acknowledge support from the Portuguese Foundation for Science and Technology (FCT-Portugal) in the framework of the Strategic Funding UID/04650/2025, Grant No. UID/PRR/04650/2025 (DOI: https://sciproj.ptcris.pt/177318UID/), and Grant No. UID/PRR2/04650/2025 (DOI: https://sciproj.ptcris.pt/178170UID/). J.P., B.A. and J.M.V.P.L also acknowledge support from FCT-Portugal through the project QPI-MoireS (DOI: https://sciproj.ptcris.pt/177781PRJ/). Finally, there has been support from the Erasmus+: Erasmus Mundus programme of the European Union.
Derivation of the Equilibrium Density Matrix Expansion
We first assume a power series convergent on the spectrum. Thus the equilibrium density matrix is
As and need not commute, each power of can be organized according to the number of insertions of the perturbation . Suppressing the time and dependence in the Hamiltonian for compactness:
Here, denotes the sum of all operator products containing exactly factors of and factors of . Accordingly, the equilibrium density matrix admits the perturbative expansion
where likewise collects all terms containing insertions of , not field order, since contains all powers of . In the eigenbasis , the insertion terms obey
with the symmetric polynomial given by
This leads us to the form used in Equation (31), repeated below:
with the coefficient defined as:
As it stands, the constrained nature of the summation in Equation (46) is not very practical for our purposes. Instead, one can prove that (see Appendix § 8.§ 1)
Substituting Eq. (48) into Eq. (47) and using the definition allows us to write
The second contribution vanishes identically due to the Lagrange interpolation identity (proved in Appendix § 8.§ 2),
leaving the remarkably compact result shown in Equation (32):
Algebraic identities
Derivation of the coefficients
The polynomial in Eq. (46) is the coefficient of in the generating function
Expanding each factor around gives
For distinct energies, the partial-fraction decomposition is
with
These coefficients follow by multiplying Eq. (53) by and evaluating at ; zero energies are included by continuity. Hence
Expanding once more,
which establishes Eq. (48).
Lagrange interpolation identity
We now prove the identity used in Appendix § 7,
for distinct energies . Consider a polynomial of degree at most . The Lagrange interpolation formula allows it to be written as
where
are the Lagrange basis polynomials. The coefficient of in each basis polynomial is
where denotes the coefficient of .
Choosing
and comparing the coefficients of on both sides of the interpolation formula gives
The coefficient on the left-hand side vanishes for and is equal to unity for . Therefore,
which establishes the identity used in the derivation of the density matrix expansion coefficients.
References
- [1]Franken P A, Hill A E, Peters C W and Weinreich G 1961 Physical Review Letters 7 118–119 ISSN 0031-9007 URL https://link.aps.org/doi/10.1103/PhysRevLett.7.118DOI
- [2]Bloembergen N 1982 Reviews of Modern Physics 54 685–695 ISSN 0034-6861 URL https://link.aps.org/doi/10.1103/RevModPhys.54.685DOI
- [3]Shen Y R 1989 Nature 337 519–525 ISSN 0028-0836, 1476-4687 URL https://www.nature.com/articles/337519a0
- [4]Hendry E, Hale P J, Moger J, Savchenko A K and Mikhailov S A 2010 Physical Review Letters 105 097401 ISSN 0031-9007, 1079-7114 URL https://link.aps.org/doi/10.1103/PhysRevLett.105.097401DOI
- [5]Kumar N, Kumar J, Gerstenkorn C, Wang R, Chiu H Y, Smirl A L and Zhao H 2013 Physical Review B 87 121406 ISSN 1098-0121, 1550-235X URL https://link.aps.org/doi/10.1103/PhysRevB.87.121406DOI
- [6]Soavi G, Wang G, Rostami H, Purdie D G, De Fazio D, Ma T, Luo B, Wang J, Ott A K, Yoon D, Bourelle S A, Muench J E, Goykhman I, Dal Conte S, Celebrano M, Tomadin A, Polini M, Cerullo G and Ferrari A C 2018 Nature Nanotechnology 13 583–588 ISSN 1748-3387, 1748-3395 URL https://www.nature.com/articles/s41565-018-0145-8
- [7]Jiang T, Huang D, Cheng J, Fan X, Zhang Z, Shan Y, Yi Y, Dai Y, Shi L, Liu K, Zeng C, Zi J, Sipe J E, Shen Y R, Liu W T and Wu S 2018 Nature Photonics 12 430–436 ISSN 1749-4885, 1749-4893 URL https://www.nature.com/articles/s41566-018-0175-7
- [8]Jiang T, Huang D, Cheng J, Fan X, Zhang Z, Shan Y, Yi Y, Dai Y, Shi L, Liu K, Zeng C, Zi J, Sipe J E, Shen Y R, Liu W T and Wu S 2018 Nature Photonics 12 634–634 ISSN 1749-4885, 1749-4893 URL https://www.nature.com/articles/s41566-018-0258-5
- [9]Ghimire S and Reis D A 2019 Nature Physics 15 10–16 ISSN 1745-2473, 1745-2481 URL https://www.nature.com/articles/s41567-018-0315-5
- [10]Wu L, Patankar S, Morimoto T, Nair N L, Thewalt E, Little A, Analytis J G, Moore J E and Orenstein J 2017 Nature Physics 13 350–355 ISSN 1745-2473, 1745-2481 URL https://www.nature.com/articles/nphys3969
- [11]Ji Z, Liu G, Addison Z, Liu W, Yu P, Gao H, Liu Z, Rappe A M, Kane C L, Mele E J and Agarwal R 2019 Nature Materials 18 955–962 ISSN 1476-1122, 1476-4660 URL https://www.nature.com/articles/s41563-019-0421-5
- [12]Sipe J E and Ghahramani E 1993 Physical Review B 48 11705–11722 ISSN 0163-1829, 1095-3795 URL https://link.aps.org/doi/10.1103/PhysRevB.48.11705DOI
- [13]Aversa C, Sipe J E, Sheik-Bahae M and Van Stryland E W 1994 Physical Review B 50 18073–18082 ISSN 0163-1829, 1095-3795 URL https://link.aps.org/doi/10.1103/PhysRevB.50.18073DOI
- [14]Aversa C and Sipe J E 1995 Physical Review B 52 14636–14645 ISSN 0163-1829, 1095-3795 URL https://link.aps.org/doi/10.1103/PhysRevB.52.14636DOI
- [15]Passos D J, Ventura G B, Lopes J M V P, Santos J M B L D and Peres N M R 2018 Physical Review B 97 235446 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/PhysRevB.97.235446DOI
- [16]Hughes J L P and Sipe J E 1996 Physical Review B 53 10751–10763 ISSN 0163-1829, 1095-3795 URL https://link.aps.org/doi/10.1103/PhysRevB.53.10751
- [17]Sipe J E and Shkrebtii A I 2000 Physical Review B 61 5337–5352 ISSN 0163-1829, 1095-3795 URL https://link.aps.org/doi/10.1103/PhysRevB.61.5337DOI
- [18]Virk K S and Sipe J E 2007 Physical Review B 76 035213 ISSN 1098-0121, 1550-235X URL https://link.aps.org/doi/10.1103/PhysRevB.76.035213DOI
- [19]Mikhailov S A 2007 Europhysics Letters (EPL) 79 27002 ISSN 0295-5075, 1286-4854 URL https://iopscience.iop.org/article/10.1209/0295-5075/79/27002
- [20]Glazov M and Ganichev S 2014 Physics Reports 535 101–138 ISSN 03701573 URL https://linkinghub.elsevier.com/retrieve/pii/S0370157313003785
- [21]Cheng J L, Vermeulen N and Sipe J E 2014 New Journal of Physics 16 053014 ISSN 1367-2630 URL https://iopscience.iop.org/article/10.1088/1367-2630/16/5/053014
- [22]Cheng J L, Vermeulen N and Sipe J E 2015 Physical Review B 91 235320 ISSN 1098-0121, 1550-235X URL https://link.aps.org/doi/10.1103/PhysRevB.91.235320DOI
- [23]Mikhailov S A 2016 Physical Review B 93 085403 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/PhysRevB.93.085403DOI
- [24]Hipolito F, Pedersen T G and Pereira V M 2016 Physical Review B 94 045434 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/PhysRevB.94.045434DOI
- [25]Al-Naib I, Sipe J E and Dignam M M 2014 Physical Review B 90 245423 ISSN 1098-0121, 1550-235X URL https://link.aps.org/doi/10.1103/PhysRevB.90.245423DOI
- [26]Mrudul M S and Dixit G 2021 Physical Review B 103 094308 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/PhysRevB.103.094308DOI
- [27]Gnawali S and Apalkov V 2024 Physical Review B 109 165121 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/PhysRevB.109.165121DOI
- [28]Pinho J M A, Cardoso S S, Bludov Y V, Lopes J M V P, Konotop V V, Cox J D and Peres N M R 2025 Physical Review B 112 195432 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/jxtp-18txDOI
- [29]Ventura G B, Passos D J, Lopes Dos Santos J M B, Viana Parente Lopes J M and Peres N M R 2017 Physical Review B 96 035431 ISSN 2469-9950, 2469-9969 URL http://link.aps.org/doi/10.1103/PhysRevB.96.035431DOI
- [30]Pinho J M A, Amorim B, Bludov Y V and Lopes J M V P 2026 Physical Review B 113 224309 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/3q7s-vmx6DOI
- [31]Morimoto T and Nagaosa N 2016 Science Advances 2 e1501524 ISSN 2375-2548 URL https://www.science.org/doi/10.1126/sciadv.1501524DOI
- [32]Orenstein J, Moore J, Morimoto T, Torchinsky D, Harter J and Hsieh D 2021 Annual Review of Condensed Matter Physics 12 247–272 ISSN 1947-5454, 1947-5462 URL https://www.annualreviews.org/doi/10.1146/annurev-conmatphys-031218-013712DOI
- [33]Jiang Y, Holder T and Yan B 2025 Reports on Progress in Physics 88 076502 ISSN 0034-4885, 1361-6633 URL https://iopscience.iop.org/article/10.1088/1361-6633/ade454
- [34]De Juan F, Grushin A G, Morimoto T and Moore J E 2017 Nature Communications 8 15995 ISSN 2041-1723 URL https://www.nature.com/articles/ncomms15995
- [35]Silva R E F, Jiménez-Galán Á, Amorim B, Smirnova O and Ivanov M 2019 Nature Photonics 13 849–854 ISSN 1749-4885, 1749-4893 URL https://www.nature.com/articles/s41566-019-0516-1
- [36]Yankowitz M, Xue J, Cormode D, Sanchez-Yamagishi J D, Watanabe K, Taniguchi T, Jarillo-Herrero P, Jacquod P and LeRoy B J 2012 Nature Physics 8 382–386
- [37]Delgado-Notario J A, Power S R, Knap W, Pino M, Cheng J, Vaquero D, Taniguchi T, Watanabe K, Velázquez-Pérez J E, Meziani Y M, Alonso-González P and Caridad J M 2025 ACS Nano 19 27338–27350
- [38]Taghizadeh A, Hipolito F and Pedersen T G 2017 Physical Review B 96 195413 ISSN 2469-9950, 2469-9969 URL https://link.aps.org/doi/10.1103/PhysRevB.96.195413DOI
- [39]Heide C, Eckstein T, Boolakee T, Gerner C, Weber H B, Franco I and Hommelhoff P 2021 Nano Letters 21 9403–9409 ISSN 1530-6984, 1530-6992 URL https://pubs.acs.org/doi/10.1021/acs.nanolett.1c02538DOI
- [40]Merboldt M, Schüler M, Schmitt D, Bange J P, Bennecke W, Gadge K, Pierz K, Schumacher H W, Momeni D, Steil D, Manmana S R, Sentef M A, Reutzel M and Matthias S 2025 Nature Physics 21 1093–1099 ISSN 1745-2473, 1745-2481 URL https://www.nature.com/articles/s41567-025-02889-7
- [41]Ventura G B, Passos D J, Viana Parente Lopes J M and Lopes Dos Santos J M B 2020 Journal of Physics: Condensed Matter 32 185701 ISSN 0953-8984, 1361-648X URL https://iopscience.iop.org/article/10.1088/1361-648X/ab62bc




