betaFeedback

WorksPhysics and Astronomy

Exciton valley depolarization in monolayer MoS2: non-Markovian quantum dynamics, intervalley scattering, and the breakdown of the Dyakonov-Perel mechanism

Abstract

In monolayer transition metal dichalcogenides, exciton valley relaxation is typically attributed to the Dyakonov-Perel (DP) mechanism, where frequent intravalley scattering with phonons or defects suppresses the intervalley exchange-induced precession and the role of intervalley scattering is considered secondary. Employing a first-principles nonequilibrium exciton Green's function (NEGF) approach with the generalized Kadanoff-Baym ansatz (GKBA), which treats exchange-driven precession and exciton-phonon scattering on an equal footing across the full Brillouin zone, we demonstrate that even in the small momentum regime most favorable to DP physics, realistic exciton-phonon scattering is too weak to induce the motional narrowing that defines the DP regime. Upon accounting for excitons across the entire Brillouin zone, large momentum intervalley scattering becomes the dominant pathway for valley relaxation, shortening the depolarization by a factor of 3-4 relative to intravalley-only models. We find that valley depolarization and decoherence time are approximately 50 fs at 300 K and lengthen to 130 fs at 10 K. By comparing our results with a Lindblad-type collision framework, we explicitly demonstrate the role of non-Markovian effects and reveal that the Markovian Lindblad framework is highly basis dependent. In the valley-pseudospin basis underlying prior analyses, the Lindblad approach artificially amplifies scattering-induced equilibration, biasing the dynamics toward a DP interpretation. Our study provides a comprehensive picture of valley relaxation, and establishes the exciton density matrix approach derived from NEGF+GKBA as a powerful tool for investigating ultrafast exciton dynamics.

Introduction

Valley physics in low-dimensional transition metal dichalcogenides (TMDs) is of particular interest owing to their unique valley-spin locking[1, 2] in KK and K′K' valleys connected by time-reversal symmetry and their experimental accessibility facilitated by large exciton binding energies[3, 4, 5, 6, 7]. The capability of selectively addressing a single valley with circularly polarized light[8, 9, 10] opens prospects for valleytronic and quantum science applications[11, 12, 13, 14, 15].

A critical metric for evaluating a material’s viability in valleytronics is the valley relaxation time. Interactions known for driving valley relaxation in TMDs include phonon-assisted intervalley electron scattering[16] [17, 18, 19, 20], Coulomb-interaction based intervalley electron-hole exchange and Dexter-like interaction21–24, and many-body correlations23,25–28. By drawing an analogy to spin relaxation physics, with the KK and K′K' valleys in TMDs treated as pseudospins, the Dyakonov-Perel mechanism (DP)29 (or Maialle-Silva-Sham (MSS) mechanism30) has been widely adopted to describe ultrafast valley relaxation. In this picture, the long-range exchange interaction acts as an effective momentum-dependent magnetic field that drives valley pseudospin precession. A hallmark of the DP regime is that the valley relaxation rate scales inversely with the momentum scattering rate29–31. This counterintuitive behavior originates from the motional narrowing of valley precession: frequent intravalley scattering from phonons or defects continuously resets the precession axis, thus averaging out the precession and slowing depolarization.

Despite its widespread use, the microscopic validity of the DP picture in realistic monolayer TMDs remains unsettled. Ultrafast optical spectroscopies, such as transient Kerr and Faraday rotation spectroscopy22,23, time-resolved photoluminescence32–35, helicity-resolved pump-probe experiments20,23,26,36, and four-wave mixing24,37,38 have been widely applied to study valley relaxation dynamics. Although the DP mechanism has been invoked to explain these experiments, these previous studies rely on simplified models that typically neglect large- ends_mid=1 momentum intervalley scattering processes22,24,33. In contrast, recent broadband femtosecond transient circular dichroism (CD) spectroscopy20 combined with model simulations identifies intervalley electron-phonon scattering as the primary relaxation channel during the subpicosecond regime. These results point to an important physical question: is valley relaxation in monolayer TMDs governed primarily by coherent exchange-driven precession near the light cone, or by exciton-phonon scattering across the full exciton Brillouin zone?

Answering this question requires a theory that goes beyond exciton population dynamics: it must retain valley coherences in the exciton density matrix while treating exchange-driven precession and exciton-phonon scattering on the same footing. Existing first-principles approaches have provided important insights, but each leaves out part of the relevant physics[16] [19, 39]. Specifically, the semiclassical Boltzmann framework neglects off-diagonal exciton coherences, which are fundamental to valley precession dynamics24,29,30,40 and thus cannot fully capture the phase-dependent evolution of the valley pseudospin. Conversely, previous electron density matrix approaches naturally retain off-diagonal coherences but omit, by construction, the long-range intervalley exchange interaction. This interaction vanishes at the zone center and is only finite for excitons with non-zero center-of-mass momentum (COM)41, making an exciton-basis treatment with full COM momentum resolution essential.

In this paper, we apply a first-principles nonequilibrium exciton Green’s function (NEGF) approach42 with the generalized Kadanoff-Baym ansatz (GKBA)43 to study exciton valley dynamics in monolayer MoS2_2. By including exciton-phonon scattering and Coulomb-based exchange interaction on an equal footing, we resolve their distinct contributions to valley relaxation dynamics across the entire Brillouin zone (BZ). We first use simulations restricted to small-momentum excitons to establish the dynamical signatures of the DP mechanism. These calculations reveal that, near the zone center, the intervalley exchange interaction drives the relaxation, but physically realistic exciton-phonon coupling is insufficient to place the system in the DP regime. Notably, DP motional narrowing only appears when the exciton-phonon coupling strength is artificially enhanced, regardless of the temperature. When excitons across the full BZ are included, large-momentum intervalley exciton-phonon scattering becomes the dominant valley relaxation pathway, shortening the depolarization time by a factor of 3–4 relative to zone-center-restricted models.

Our work also clarifies the theoretical requirements for simulating ultrafast exciton valley dynamics. We compare the GKBA dynamics with a Lindblad approach and show that the latter is basis dependent: formulated in the valley-pseudospin basis underlying prior analyses, it overestimates scattering-induced equilibration and produces a spurious motional-narrowing signature at realistic coupling strengths. In contrast, the GKBA formulation collision integral is covariant under basis transformations, ensuring representation-independent dynamics, while additionally incorporating non-Markovian scattering effects. These non-Markovian effects manifest as rapid transient oscillations in the population of high-energy exciton states during the initial stage of dynamics, which may be accessible to attosecond spectroscopies44,45. At longer times, our calculated valley depolarization and decoherence times are approximately 130 fs at 10 K, in good agreement with experimental measurements20,24.

Method

Equation of motion

We start by writing down the effective exciton Hamiltonian including exciton-phonon coupling46,47,

H=∑nQEnQanQ†anQ+∑νqℏωνqbνq†bνq+1Nq∑nmν,QqGnmν(Q,q)am,Q+q†anQ(bνq+bν,−q†),(1)H = \sum_{n\mathbf{Q}} E_{n\mathbf{Q}} a^\dagger_{n\mathbf{Q}} a_{n\mathbf{Q}} + \sum_{\nu\mathbf{q}} \hbar\omega_{\nu\mathbf{q}} b^\dagger_{\nu\mathbf{q}} b_{\nu\mathbf{q}} + \frac{1}{\sqrt{N_{\mathbf{q}}}} \sum_{nm\nu,\mathbf{Q}\mathbf{q}} \mathcal{G}_{nm\nu}(\mathbf{Q},\mathbf{q}) a^\dagger_{m,\mathbf{Q}+\mathbf{q}} a_{n\mathbf{Q}} \left(b_{\nu\mathbf{q}} + b^\dagger_{\nu,-\mathbf{q}}\right), \tag*{(1)}

where NqN_{\mathbf{q}} is the number of q\mathbf{q} points sampled in the BZ, anQ†a^\dagger_{n\mathbf{Q}} is a creation operator for an exciton with COM Q\mathbf{Q} and state nn, EnQE_{n\mathbf{Q}} is the corresponding excitation energy, bνq†b^\dagger_{\nu\mathbf{q}} is a creation operator for a phonon with momentum q\mathbf{q} and mode ν\nu, ωνq\omega_{\nu\mathbf{q}} is the phonon frequency, and Gnmν(Q,q)\mathcal{G}_{nm\nu}(\mathbf{Q},\mathbf{q}) is the exciton-phonon coupling matrix element46,47, which describes the scattering from an exciton in state (n,Q)(n,\mathbf{Q}) to state (m,Q+q)(m,\mathbf{Q}+\mathbf{q}) via a phonon (ν,q)(\nu,\mathbf{q}). The general equation of motion for the exciton density matrix reads,

dρnmQdt+i[H0,ρ]nmQ=dρnmQdt∣col,(2)\frac{d\rho_{nm\mathbf{Q}}}{dt} + i[H_0,\rho]_{nm\mathbf{Q}} = \left.\frac{d\rho_{nm\mathbf{Q}}}{dt}\right|_{\mathrm{col}}, \tag*{(2)}

where ρnmQ=⟨anQ†amQ⟩\rho_{nm\mathbf{Q}} = \langle a^\dagger_{n\mathbf{Q}}a_{m\mathbf{Q}}\rangle is the exciton density matrix diagonal in the exciton COM Q\mathbf{Q}, H0H_0 is the quadratic part of the Hamiltonian and the term on the right-hand side is the collision integral—here it describes the impact of phonon scattering on the evolution of the exciton density matrix.

In this work, we approximate the collision integral using two theoretically well established, but rarely implemented at the ab initioab\ initio level, formulations. The first is derived from a Lindblad superoperator based on the many-body Hamiltonian48,49, which has been utilized to investigate ultrafast spin dynamics within the electron density matrix formalism50. Hereafter, we refer to this as the “Lindblad” approach. For excitons, we can write,

dρijQdt∣col=2πℏNqRe⁡{(I+ρ)isQG~stν∗(Q,q)ρtmQ+qG~jmν(Q,q)−G~siν(Q−q,q)(I+ρ)stQ−qG~tmν∗(Q−q,q)ρmjQ}(nq+1)+2πℏNqRe⁡{(I+ρ)isQG~tsν(Q−q,q)ρtmQ−qG~mjν∗(Q−q,q)−G~isν∗(Q,q)(I+ρ)stQ+qG~mtν(Q,q)ρmjQ}nq,(3)\left.\frac{d\rho_{ij\mathbf{Q}}}{dt}\right|_{\mathrm{col}} = \frac{2\pi}{\hbar N_{\mathbf{q}}}\operatorname{Re}\left\{(I+\rho)_{i s\mathbf{Q}}\widetilde{G}^{*}_{s t\nu}(\mathbf{Q},\mathbf{q})\rho_{t m\mathbf{Q}+\mathbf{q}}\widetilde{G}_{j m\nu}(\mathbf{Q},\mathbf{q})-\widetilde{G}_{s i\nu}(\mathbf{Q}-\mathbf{q},\mathbf{q})(I+\rho)_{s t\mathbf{Q}-\mathbf{q}}\widetilde{G}^{*}_{t m\nu}(\mathbf{Q}-\mathbf{q},\mathbf{q})\rho_{m j\mathbf{Q}}\right\}(n_{\mathbf{q}}+1) +\frac{2\pi}{\hbar N_{\mathbf{q}}}\operatorname{Re}\left\{(I+\rho)_{i s\mathbf{Q}}\widetilde{G}_{t s\nu}(\mathbf{Q}-\mathbf{q},\mathbf{q})\rho_{t m\mathbf{Q}-\mathbf{q}}\widetilde{G}^{*}_{m j\nu}(\mathbf{Q}-\mathbf{q},\mathbf{q})-\widetilde{G}^{*}_{i s\nu}(\mathbf{Q},\mathbf{q})(I+\rho)_{s t\mathbf{Q}+\mathbf{q}}\widetilde{G}_{m t\nu}(\mathbf{Q},\mathbf{q})\rho_{m j\mathbf{Q}}\right\}n_{\mathbf{q}}, \tag*{(3)}

where nqn_{\mathbf{q}} is the phonon occupation and we define

G~jmν(Q,q)≡Gjmν(Q,q)δ(EmQ+q−EjQ−ℏωqν)1/2,\widetilde{G}_{j m\nu}(\mathbf{Q},\mathbf{q}) \equiv G_{j m\nu}(\mathbf{Q},\mathbf{q})\delta\left(E_{m\mathbf{Q}+\mathbf{q}}-E_{j\mathbf{Q}}-\hbar\omega_{\mathbf{q}\nu}\right)^{1/2},

and

G~imν∗(Q,q)≡Gimν∗(Q,q)δ(EmQ+q−EiQ+ℏωqν)1/2.\widetilde{G}^{*}_{i m\nu}(\mathbf{Q},\mathbf{q}) \equiv G^{*}_{i m\nu}(\mathbf{Q},\mathbf{q})\delta\left(E_{m\mathbf{Q}+\mathbf{q}}-E_{i\mathbf{Q}}+\hbar\omega_{\mathbf{q}\nu}\right)^{1/2}.

The plus sign in (I+ρ)(I+\rho) reflects the Bose statistics and one can show that the Boltzmann equation is recovered by taking only the state diagonal part of the density matrix. In deriving Eq. (3), a Markovian approximation is adopted, so memory effects are absent in this framework. The populations of phonons are assumed to follow a Bose distribution at a fixed temperature, which treats phonons as a bath, ignoring their dynamics. Equation (3) in the semiclassical Boltzmann limit51, where only the diagonal terms of the collision integral are included, describes the phonon-driven population transfers and relaxations. In contrast, the off-diagonal terms are vital to the decoherence processes. We do not include additional relaxation/dephasing channels in our dynamics.

Although the DP mechanism is traditionally derived using a collision term that is a linear functional of the density matrix, recent work on electron density matrices has shown that the DP regime persists even when nonlinearities are introduced as in the collision integral above50,52. We demonstrate in Sec. III A that the signature of the DP regime is clearly preserved within our excitonic framework.

The second formulation we investigate derives the collision integral from the NEGF approach, which goes beyond the Markov approximation53,54. In general, the derived equation of motion of the density matrix depends on several components of the two-time Green’s function. To close the equation and avoid solving the full Kadanoff-Baym equation, we make the generalized Kadanoff-Baym ansatz43. Hereafter, we refer to this as the “GKBA” approach. Taking into account the Fan diagram of exciton-phonon self-energy46,47, the collision integral reads, dρdt∣col=−[IL<+H.c.]\left.\frac{d\rho}{dt}\right|_{\mathrm{col}}=-[I^{<}_{L}+H.c.], where

IL,ipQ<(t)=i∑njsmνq∫0tdt′Gνin(Q−q,q)Gνjl∗(Q−q,q)[GnsQ−qR(t,t′)ρˉslQ−q(t′)Dνq<(t′,t)ρjmQ(t′)GmpQA(t′,t)−GnsQ−qR(t,t′)ρslQ−q(t′)Dνq>(t′,t)ρˉjmQ(t′)GmpQA(t′,t)].(4)I^{<}_{L,ip\mathbf{Q}}(t)=i\sum_{n j s m\nu\mathbf{q}}\int_{0}^{t}dt'G_{\nu in}(\mathbf{Q}-\mathbf{q},\mathbf{q})G^{*}_{\nu j l}(\mathbf{Q}-\mathbf{q},\mathbf{q})\left[G^{R}_{n s\mathbf{Q}-\mathbf{q}}(t,t')\bar{\rho}_{s l\mathbf{Q}-\mathbf{q}}(t')D^{<}_{\nu\mathbf{q}}(t',t)\rho_{j m\mathbf{Q}}(t')G^{A}_{m p\mathbf{Q}}(t',t)-G^{R}_{n s\mathbf{Q}-\mathbf{q}}(t,t')\rho_{s l\mathbf{Q}-\mathbf{q}}(t')D^{>}_{\nu\mathbf{q}}(t',t)\bar{\rho}_{j m\mathbf{Q}}(t')G^{A}_{m p\mathbf{Q}}(t',t)\right]. \tag*{(4)}

In Eq. (4), GRG^{R} and GAG^{A} are the retarded and advanced components of the exciton Green’s function, respectively; ρˉ=I+ρ\bar{\rho}=I+\rho; and D<D^{<} and D>D^{>} are the lesser and greater components of phonon propagators, respectively55. In our implementation, we approximate the exciton propagator with the Wigner-Weisskopf form53,55,

GnsQR(t,t′)=−iθ(t−t′)δnse−i(EnQ−iη)(t−t′)(5)G^{R}_{ns\mathbf{Q}}(t,t')=-i\theta(t-t')\delta_{ns}e^{-i(E_{n\mathbf{Q}}-i\eta)(t-t')} \tag*{(5)}

where θ(t)\theta(t) is the Heaviside function and η\eta is the numerical broadening parameter. The computational cost of Eq. (4) scales as O(t2)O(t^2), posing a significant bottleneck for long-time propagation. To overcome this, we employ the recently developed G1-G2 scheme, which reformulates the integro-differential equation as a system of coupled first-order ordinary differential equations56. Specifically, the memory kernel which involves a time-integral of the density matrix is handled by introducing an auxiliary density matrix, the formal derivation of which is provided in the Appendix I A V A.

A comparison of these two scattering formulations is instructive. Both frameworks describe the evolution of band- and momentum-resolved exciton coherences and populations driven by microscopic exciton-phonon scattering events. One can show that both collision terms strictly conserve the total exciton population, which we also verify numerically. In the Lindblad approach (Eq. (3)), the delta functions enforce energy conservation during exciton-phonon scattering. In contrast, the GKBA collision integral (Eq. (4)) allows for off-shell scattering that does not strictly conserve energy at short time scales, consistent with the energy-time uncertainty principle. Furthermore, the time-integral formulation preserves the memory effects inherent to the dynamics; notably, this framework has been successfully applied to capture the polaron features in transient absorption spectra55.

Helical versus spin basis

A subtle but fundamentally important issue in modeling dissipative dynamics is basis invariance. While both Eq. (3) and Eq. (4) account for exciton scattering, the Lindblad approach is not basis independent. This stems from the fact that the energy-conserving delta functions are typically defined with respect to a specific set of eigenstates; because these energy arguments do not transform co-variantly under basis changes, the resulting dynamics become sensitive to the representation. As shown in Ref.53,55, this basis dependence is a direct consequence of the infinite-time integration implicit in the Markov approximation, which ties the collision operator to the representation in which energy conservation is enforced.

This distinction is particularly relevant in the study of valley or spin dynamics, where two different bases are frequently employed. In the standard derivation of the DP mechanism22,24,29,30,40, excitons with small COM Q\mathbf{Q} are expanded in the basis of degenerate Q=0\mathbf{Q}=0 A excitons, which are localized in the KK or K′K' valley. In this representation, the exciton pseudospins are defined by their valley localization. The effective Hamiltonian in this basis includes the long-range intervalley exchange interaction—which vanishes at Q=0\mathbf{Q}=0—alongside momentum-dependent intravalley exchange and direct Coulomb terms. Because the intervalley exchange interaction flips the pseudospin, it acts as an effective transverse magnetic field, driving the precession dynamics. We denote the corresponding formulation of the collisions as the “Lindblad spin” approach.

Alternatively, in Eq. (1), exciton states are obtained by solving the first-principles Bethe-Salpeter equation (BSE), which incorporates both direct and exchange Coulomb kernels. We refer to the resulting eigenstates as the helical basis and denote the corresponding treatment as the “Lindblad helical” approach. In the following sections, we demonstrate that numerical results obtained from these two bases are indeed different. In stark contrast, Eq. (4) and Eq. (2) are co-variant under basis transformations, ensuring that physical observables remain strictly invariant.

Computational details

We have implemented a numerical solver for both Lindblad and GKBA approaches. All of the required ingredients, such as phonon modes, exciton energies, and exciton-phonon coupling matrix elements, are computed from first-principles. The time-propagation is integrated with a fourth order Runge-Kutta algorithm with a step size of 0.50.5 fs. Four exciton bands are considered in the calculations. Numerically, delta functions are implemented as Lorentzian functions with a width of 2020 meV, which is equivalent to η=10\eta=10 meV in the Wigner-Weisskopf exciton propagator.

Computational details of the first-principles calculations are given as follows. Electronic structure and density functional perturbation theory57 calculations are performed with the Quantum Espresso package58. Electron-phonon coupling matrix elements are computed puted with the EPW package59. We compute the GW quasi-particle energy and solve for excitons with the BerkeleyGW package60–62. Finite COM excitons41 and phonons are solved on a 48×4848 \times48 QQ-grid and qq-grid, respectively, on which exciton-phonon coupling matrix elements are computed. All other parameters can be found in our previous work63.

Intravalley Exciton Dynamics

The DP mechanism for small momentum excitons has previously been adopted to interpret experiments22,24. Indeed, a minimum four-state model31 clearly demonstrated the crossover of valley dynamics between the DP and other regimes depending on the magnitude of the momentum scattering rate, an approach we follow here to verify its applicability to our system. However, the role of large-momentum transfer scatterings, which leads to ill-defined valley quantum numbers, remains poorly understood. To disentangle the influence of various relaxation channels, we distinguish between scattering processes associated with small and large transfer of exciton momentum QQ. Small-QQ scattering processes include intravalley exciton-phonon scattering, wherein both the electron or the hole acquire a small momentum change, or exchange intervalley interaction, wherein both electrons and holes flip the valley, but the overall exciton momentum remains conserved [Figure 1 (a)]. In contrast, only exciton-phonon interactions lead to scattering processes that contribute to a large change in the exciton momentum QQ [Figure 1 (b)]. In this section, by focusing first on the small-momentum (Q≈0Q \approx0) exciton dynamics, we investigate the competition between these exchange-driven mechanisms and phonon-induced scattering, thereby identifying the hallmarks of the DP mechanism.

Figure 1. (a) Schematic representation of scattering processes associated with small transfer of exciton center-of-mass (COM) momentum QQ, and visualized both in the exciton and quasiparticle band structures. Small-QQ exciton scattering processes originate from both intervalley electron-hole exchange interactions, which strictly conserve QQ, and intravalley exciton-phonon scattering processes, which can be broken down into electron-phonon and hole-phonon scattering contributions. (b) Corresponding representation of processes associated with a large transfer of the exciton COM momentum QQ. For large-QQ exciton scattering, the primary mechanism is intervalley exciton-phonon scattering processes, which is composed of electron- and hole-phonon scattering contributions.

Four-state model

We start with a four-state model, where only A excitons and two excitons of Q=(1/48,0)Q = (1/48, 0) (The smallest QQ commensurate with our uniform grid) are included as shown in Figure 2 (a). The quadratic part of the Hamiltonian in the spin basis is

Figure 2. (a) Four exciton states selected from the exciton bands define the four-state model. ∣0⟩\lvert0\rangle and ∣1⟩\lvert1\rangle states are degenerate A excitons. ∣+⟩\lvert+\rangle and ∣−⟩\lvert-\rangle are exchange-split states at Q=(1/48,0)\mathbf{Q} = (1/48, 0). (b) Time evolutions of valley polarizations ⟨Sz⟩\langle S_z\rangle from the four approaches. (c)–(f) Time evolutions of populations of the four states, respectively. The results shown are simulated by Boltzmann (red dot-dashed line), Lindblad spin (blue solid line), Lindblad helical (green dashed line), and GKBA (purple dots) approach.

Hspin=ϵ22(σz+I)⊗I+Δ4(σz+I)⊗τx,(6)H_{\mathrm{spin}} = \frac{\epsilon_2}{2}(\sigma_z + I) \otimes I + \frac{\Delta}{4}(\sigma_z + I) \otimes\tau_x, \tag*{(6)}

where ϵ2\epsilon_2 is the energy of exciton at Q=(1/48,0)Q = (1/48, 0) without exchange interaction offset by the Q=0Q = 0 exciton energy, σz\sigma_z is the Pauli zz-matrix in the space of COM Q=0Q = 0 and Q=(1/48,0)Q = (1/48, 0) and τx\tau_x is the Pauli xx-matrix in the space of degenerate KK and K′K' valleys. The two states with finite momentum in the helical basis, obtained by diagonalizing Eq. (6), are denoted as ∣+⟩|+\rangle and ∣−⟩|-\rangle, respectively. Their energy splitting Δ\Delta is due to intervalley exchange interactions. Exciton-phonon interactions couple the A exciton states (∣0⟩|0\rangle and ∣1⟩|1\rangle) to the ∣+⟩|+\rangle and ∣−⟩|-\rangle states but couplings between orthogonal spin states are not possible since exciton-phonon interactions do not flip pseudospin in this energy range. The exciton energy and coupling strength are taken from our first-principles calculations (Sec. II. C).

To model photo-excitation by circularly polarized light, we initialize our simulation in state ∣0⟩|0\rangle, corresponding to a valley polarized configuration with ⟨Sz⟩=1\langle S_z\rangle= 1 and monitor the subsequent evolution of populations and valley polarizations. The temperature of the phonon bath is set at 300 K. In the following, the ⟨Sz⟩\langle S_z\rangle is computed by summing the contributions of the zone center exciton and its six nearest neighbors.

Figure 2(c)–(Figure 2f) illustrate the population dynamics for the four-state system calculated with our four numerical approaches. In the Boltzmann limit, Eq. (2) and 3 are reduced to their diagonal components, thereby neglecting exciton coherences. Qualitatively, all four methodologies yield consistent dynamical trends. Quantitatively, the results from the Boltzmann, Lindblad-helical, and GKBA approaches are in close agreement. In each case, an initial transition from the ∣0⟩\lvert0\rangle state to the higher-energy ∣+⟩\lvert+\rangle and ∣−⟩\lvert-\rangle states, manifesting as a rapid population surge in the latter, is followed by the decay of these high-energy states into the ∣1⟩\lvert1\rangle state. The convergence of the three approaches suggests that the helical basis effectively diagonalizes the dominant interactions.

Figure 2 (b) shows the time evolution of the valley polarization. While the four approaches yield qualitatively consistent trends with the Boltzmann, Lindblad-helical, and GKBA providing quantitatively close results, the Boltzmann approach fundamentally fails to capture the coherent oscillations dynamics in the valley polarization. Instead, it shows a mono-exponential decay, confirming that a population-based description is insufficient to resolve the quantum coherence of valley dynamics. The coherent oscillation can be understood as the precession of pseudospin around an effective transverse field, with the angular frequency determined by the energy splitting Δ\Delta. This precession is most intuitively described in the spin basis, where superposition of up and down valley states is encoded in the off-diagonal terms of the density matrix. Because these coherences are completely neglected in the Boltzmann framework, the precession can not be captured, resulting in the loss of the oscillating polarizations. Notably, the populations calculated using the Lindblad spin approach equilibrate more rapidly compared to the other three methods, all of which remain in close agreement. These results provide direct numerical evidence of the basis dependence inherent to the Lindblad formulation as discussed in Section II B.

We emphasize that the polarization relaxation in this model is a combined result of exchange interactions and exciton-phonon scatterings. Without exchange interactions, valley spin SzS_z is a conserved quantity while turning off the exciton-phonon scatterings leads to either frozen dynamics or persistent precession depending on the initial state.

As demonstrated in 31, the four-state model exhibits a dynamical crossover spanning three distinct regimes-Elliot-Yaffet (EY), DP, and strong precession-determined by the relative strength between the momentum scattering rate, the exchange splitting and the spin-flip scattering rate. Since exciton-phonon interactions do not introduce valley-pseudospin flips in this model, the EY mechanism is inactive, and we focus on the latter two regimes. Unlike the DP regime discussed in § 1, the strong precession regime is characterized by fast precession dynamics such that the time-averaged spin aligns with the precession axis; consequently, polarization relaxes through the realignment of this axis. This leads to a fundamental change in the scaling behavior. While the relaxation rate is inversely proportional to the scattering rate in the DP regime (motional narrowing), it becomes directly proportional to the scattering rate in the strong precession limit. In the following, we resolve this crossover by systematically scaling the exciton-phonon coupling strength by a factor ss.

Figure 3 (b)-(Figure 3e) display the polarization dynamics from four different approaches with s=1s = 1, 5, and 7, with the s≠1s \ne1 curves vertically offset for clarity. At s=1s = 1, the relaxation rates extracted from the Boltzmann, Lindblad helical, and GKBA simulations are in quantitative agreement. However, as ss increases, the results diverge significantly. The Boltzmann approach yields a relaxation time that monotonically decreases with ss. In contrast, the other three frameworks exhibit a transition to under-damped oscillatory behavior (Δτ/ℏ>1\Delta\tau/\hbar> 1, where τ\tau is the depolarization time). By fitting the relaxation time from the decay envelopes of these oscillations, we find that it exhibits a non-monotonic dependence on ss; it initially decreases before increasing in tandem with the oscillation amplitude. This ’upturn’ in depolarization time is the definitive hallmark of the DP mechanism, which underscores the fundamental importance of exciton coherence in describing valley dynamics.

Depolarization time and valley polarization dynamics for four approaches

Figure 3. (a) Depolarization time as a function of scattering rate from the simulations of the four approaches. The shaded region indicates the DP regime, which is characterized by the upturn of the curves from the three simulations other than the Boltzmann approach. (b)-(e) Valley polarizations dynamics simulated by the Boltzmann, Lindblad spin, Lindblad helical, and GKBA approach, respectively. Curves obtained from simulations with s≠1s \ne1 are vertically offset for clarity. In each panel, the dots are simulated data and dashed lines are the best fitting exponential functions. Curves with amplitude oscillations are fitted by their local maximums.

While the precession rates extracted from the two Lindblad approaches are consistent with each other, the GKBA yields a smaller rate, which might be attributed to the higher order vertex corrections embedded in the GKBA collision integral. A comparison between the two Lindblad implementations reveals a stronger precession in the spin basis. This can be understood by noting that the ∣+⟩|+\rangle and ∣−⟩|-\rangle states in the helical representation have an energy offset due to exchange, resulting in an imbalanced population across the basis states that effectively suppresses the coherent precession amplitude.

The transition between these regimes is summarized in Figure 3 (a), where the depolarization time is plotted against the momentum scattering rate extracted from early-time exciton populations. The clear upturn signature at high scattering rates confirms the crossover into the DP regime—a feature entirely absent in the Boltzmann results. These findings demonstrate that, while the predicted relaxation times vary, the exciton density matrix approach (both Lindblad and GKBA) is capable of capturing the complex interplay between coherence and dissipation that defines the DP mechanism.

Ab initio calculation restricted to small QQ

Next, we extend our analysis to the dynamics of excitons with COM Q\mathbf{Q} in a 0.150.15 Å−1^{-1} radius patch centered at Q=0\mathbf{Q} = 0 as illustrated in Figure 4 (a). The exciton energy, exchange couplings, and exciton-phonon coupling strength are taken from our ab initio calculations (Sec. II.C). Having established that the Boltzmann formulation is not capable of capturing the DP mechanism, we focus on the simulations using Lindblad and GKBA approaches. Figure 4 (b)-(Figure 4d) display the population dynamics for the four characteristic states within this patch. Qualitatively, all three simulations exhibit consistent trends of the population evolution. However, the Lindblad spin approach predicts a faster population decay rate, with the A exciton population reaching equilibrium at approximately 500500 fs, nearly twice as fast as the 11 ps scale predicted by the other two approaches. Quantitatively, Lindblad helical and GKBA results converge, suggesting that the helical representation, as opposed to spin representation, may be a more robust basis for the dissipative dynamics in this regime.

Exciton energy dispersion, population dynamics, depolarization times, and valley polarization dynamics

Figure 4. (a) Two dimensional exciton energy dispersion near center-of-mass momentum, Q=0\mathbf{Q} = 0. Red dots are exciton states included in the calculation restricted to small Q\mathbf{Q}. (b)-(d) are exciton populations dynamics of the four states defined in Figure 1, from Lindblad spin, Lindblad helical, and GKBA approach, respectively. (e) Depolarization time as a function of scattering rate from the simulations of the three approaches. Shaded region indicates the DP regime for the Lindblad spin and GKBA simulations. (f)-(h) Valley polarization dynamics with scaled exciton-phonon couplings from the three approaches. See the caption of Figure 2 for definitions of the curves and symbols.

Figure 4 (f)-(Figure 4h) display valley polarization dynamics for several scaled exciton-phonon coupling strengths, revealing divergent behaviors across the three simulations. The Lindblad spin approach predicts underdamped oscillations with a precession time of 200200 fs. In stark contrast, the other two methods exhibit heavily damped dynamics, with the GKBA specifically yielding a vanishing precession amplitude. As we increase the scaling factor ss, the Lindblad spin results show enhanced precession amplitudes in tandem with the depolarization time, which indicates that the system resides in the DP regime even at s=1s = 1. Conversely, the Lindblad helical simulation shown in Figure 4 (g) displays a monotonically decreasing depolarization time with increased coupling, suggesting the absence of motional narrowing. Notably, we no longer observe the coherent precession in the strongest coupling case, which can be rationalized by the fact that the scattering phase space in the calculation restricted to small Q\mathbf{Q} is much larger and the previously discussed population imbalances suppress the coherent oscillation.

In the GKBA simulation, we observe a non-monotonic trend of the depolarization time: it initially decreases but increases subsequently at higher coupling strength as shown in Figure 4 (h). While simulations at the high coupling values suffer from known numerical instabilities which can lead to negative exciton populations64 preventing us from scaling to s>3s > 3, the results at s=2.5s = 2.5 clearly show a longer relaxation time than at s=2s = 2. This indicates a crossover into the DP regime at a modest coupling strength.

The transition is summarized in Figure 4 (e). While the crossover between the DP and the strong precession regime is captured by both the Lindblad spin and the GKBA simulations, their predicted relaxation times differ. The Lindblad spin results align with previous model-based studies22,24 that attribute the depolarization time to the DP mechanism. However, our simulation with the advanced GKBA treatment suggests that for a single exciton valley, the intravalley exciton-phonon coupling is too weak to drive DP physics so that the system remains in the strong precession regime.

AB INITIO CALCULATIONS IN THE FULL BZ

Having analyzed the competition between exciton-phonon scattering and exchange interactions within a single exciton valley, we extend our analysis to the entire BZ, including all low-energy dark exciton branches illustrated in Figure 5 (a).

Figure 5. (a) Exciton energy dispersion along the high symmetry path. States at Γ\Gamma and KK are labeled by their energy orders at that valley. (b) and (c) show exciton population dynamics of six low energy exciton states from the Lindblad helical and GKBA simulations. Dashed lines indicate states which are optically dark. Red box in (c) highlights the transient oscillations. (d) Depolarization time as a function of exciton-phonon couplings scale factors. (e) and (f) show valley polarization dynamics from the two approaches, respectively. Symbols are simulated data while dashed lines are the best fitting function in the same time span.

Figures 5 (b) and (c) display the population dynamics of characteristic exciton states, integrated over 0.150.15 Å−1^{-1} patches centered at Q\mathbf{Q} points labeled in (a), as calculated using the Lindblad helical and GKBA frameworks at 300 K, respectively. Consistent with our previous results, the two approaches yield qualitatively similar trends of population dynamics. The population of Γ3\Gamma_3 (A) exciton exhibits a rapid decay within the first 250 fs before asymptotically approaching equilibrium. While the populations of Γ4\Gamma_4 (the other A) and K2K_2 exciton surge, overshooting their equilibrium values before slow subsequent decay. Crucially, the rapid rise of the K2K_2 population within the first 100 fs highlights the critical role of large-momentum intervalley exciton-phonon scattering as the dominant valley depolarization channel—a feature inherently absent in our previous phase space-restricted small QQ patch. The Lindblad approach predicts a slightly accelerated population dynamics, with the Γ4\Gamma_{4} and Γ3\Gamma_{3} exciton populations equalizing around 200 fs compared to 250 fs in the GKBA. More striking, the GKBA results exhibit non-Markovian transients, not seen in the Lindblad simulation. Specifically, in Figure 5 (c), we observe a sharp drop in the A exciton population accompanied by high-frequency oscillations within the first 50 fs. As this frequency does not correspond to characteristic phonon energies or exchange splittings, and are mirrored in the dynamics of the auxiliary density matrices, they signal a non-Markovian effect, mediated by virtual transition to the high-energy exciton states captured by the GKBA memory kernel, to be discussed in greater detail below.

To determine the accessibility of the DP regime for realistic simulations that include all states in the full BZ, we again perform simulations with scaled exciton-phonon couplings ss. Figure 5 (e) and (Figure 5f) depict the resulting valley polarization dynamics. We find that the short-time valley polarization dynamics is well-described by a mono-exponential decay. In the GKBA simulations, the high-frequency oscillations of the polarization within the first 50 fs are directly correlated with the Γ3\Gamma_{3} exciton population transients shown in Figure 5 (c). The amplitude of these features increases with the scale factor ss until the dynamics becomes numerically unstable at s=1.5s = 1.5, where the positivity of the population is no longer preserved. Notably, the depolarization time in both frameworks decreases monotonically with increasing coupling strength, showing no signatures of the motional narrowing hallmark of the DP mechanism. Moreover, exciton coherence, which is essential for precession dynamics, remains negligible across all tested coupling strengths. Figure 5 summarizes the dependence of the depolarization time on the scaling factor ss. We therefore conclude that the system remains in the strong precession regime, with intervalley scattering effectively suppressing the coherent DP pathway. This is further evidenced by a significant reduction in the valley depolarization time, which is 3-4 times smaller than the values obtained in the calculation restricted to small QQ, across the different approach employed. A similar reduction in the valley depolarization time when including the intervalley scattering is also found in our calculations at 10 K, even though the direct scattering rate to COM KK exciton is suppressed at low temperature.

The non-Markovian effect captured by the GKBA simulation is elucidated by further analyzing the evolution of exciton energy distribution functions during the initial stage of relaxation. Figure 6 (a) and (Figure 6f) depict the initial state for both frameworks, characterized by a single A exciton population. Although the low-energy exciton population dynamics appear qualitatively consistent trend (cf. Figure 5 (b) and (Figure 5c)), a striking difference emerges in the high-energy exciton distribution within 10 fs as shown Figure 6 (b) and (Figure 6g). In the GKBA simulation, we observe a significant transient population of high-energy excitons near 1.95 eV. In contrast, most exciton populations remain near the A exciton, with a small tail extending toward high energy in the Lindblad calculation. This distinction arises from the fundamental treatment of scattering: while the Lindblad framework imposes strict energy conservation constraint on the scattering phase space, the GKBA naturally incorporates off-shell processes over short timescales-governed by the energy-time uncertainty principle-allowing for transitions that violate energy conservation at t→0t \to0.

Figure 6. (a)–(e) and (f)–(j) display snapshots of exciton distribution functions obtained from GKBA and Lindblad dynamics, respectively. Red arrows in panel (b)–(d) highlight the oscillations of high energy populations. Symbols are data from the simulation. Dashed black lines are Bose distribution at 850 K.

By 40 fs, cf. Figure 6 (h), high-energy states begin to populate in the Lindblad calculations. Conversely, the GKBA population exhibits a rapid redistribution back toward low-energy states as shown in Figure 6 (c)

At 50 fs, while the overall distribution remains steady in the Lindblad simulation, a dramatic oscillatory resurgence at 1.95 eV is observed in the GKBA distribution. The rapid oscillations in the high-energy exciton population match those observed in the A exciton population in Figure 5 (c).

These high-frequency oscillations in the high-energy population precisely match the transients observed in the A-exciton and polarization dynamics (Figure 5(c) and (Figure 5f)), confirming their origin as non-Markovian memory effects unique to the GKBA.

In the long-time limit (t≈1t \approx1 ps), both distributions nearly converge to an identical high-temperature Bose-Einstein distribution at 850 K. While a Bose-Einstein distribution is a formal steady state solution for both frameworks, the elevated final temperature is a direct consequence of broadening in the numerical approximation of the energy-conserving delta function (or equivalently, the imaginary part of the self-energy used in the Wigner-Weisskopf propagator). We find that the effective final temperature can be systematically reduced by employing a smaller broadening parameter. The ultrafast non-Markovian transient predicted by the GKBA approach may be identified by attosecond spectroscopy44,45.

Valley depolarization and decoherence time of monolayer transition metal dichalcogenides have been extensively characterized through various experimental techniques. To facilitate a direct comparison with experiments, we present the temperature dependence of our calculated depolarization and decoherence times in Figure 7 (a) and (b), respectively. For decoherence simulations, we initialize the system in a pure state which is an equal superposition of the two degenerate A excitons, which corresponds to excitations by linearly polarized light, and monitor the subsequent evolution of the pseudospin expectation value, ⟨Sx(t)⟩\langle S_x(t) \rangle. Detailed simulation data and the corresponding fittings are provided in the Appendix II. Overall, the GKBA framework yields an approximately 20–30 fs longer valley depolarization and decoherence time than those obtained via the Lindblad approach. At 300 K, we calculate a valley depolarization and decoherence time of 50 fs, which increase monotonically to 130 fs at 10 K. These results are in agreement with an earlier four-wave mixing experiment that reported a valley decoherence time of 100 fs at 10 K24 in monolayer WSe2_2 as well as recent helicity-resolved transient absorption measurements showing a valley decoherence time of 100 fs in the initial stage of the dynamics20. While our simulations align with these two ultrafast measurements, we note that the latter experiment exhibits a weaker temperature dependence than our predictions and both measurements are done in monolayer WSe2_2 rather than MoS2_2. Furthermore, we acknowledge that several earlier studies reported depolarization times on the order of picoseconds. These discrepancies may arise from differences in sample quality, exciton densities, or the specific temporal resolution by different spectroscopic techniques.

Figure 7. (a) Depolarization and (b) decoherence time as a function of temperature fitted from Lindblad (red dots) and GKBA (blue squares) simulations.

Conclusion

In conclusion, we have employed a first-principles exciton density matrix approach, incorporating a non-Markovian collision integral within the GKBA framework, to investigate valley dynamics in monolayer MoS2_2. Our findings demonstrate that the DP mechanism is not responsible for valley relaxation in this material owing to insufficient small-momentum exciton-phonon scattering. Instead, large-momentum intervalley scattering serves as the primary channel for relaxation. This provides parameter-free microscopic support for the intervalley-scattering channel identified in recent transient circular dichroism experiments20, and explains why model analyses restricted to small-momentum excitons have instead inferred DP behavior.

By rigorously accounting for Coulomb-mediated interactions, including both direct and exchange terms, alongside exciton-phonon scatterings, our method allows for an unambiguous evaluation of these couplings on the dynamics. Equally importantly, we have shown that the widely used Lindblad collision integral is basisdependent—a direct consequence of its infinite-time Markov limit—and that, in the valley-pseudospin basis underlying prior model analyses, it produces a spurious motional-narrowing signature. Basis-covariant treatments such as the GKBA are therefore essential for reliably identifying relaxation regimes. Furthermore, we identify signatures of non-Markovian effects unique to the GKBA framework, which manifest as rapid transient oscillations during the initial stage of the dynamics. The calculated valley decoherence times show good agreement with existing four-wave mixing and transient absorption experiments. Our work clarifies the fundamental relaxation channel of the valley degree of freedom and establishes the GKBA-based density matrix approach as a predictive tool for studying ultrafast exciton dynamics in quantum materials.

Acknowledgments

This work was supported by the Center for Computational Study of Excited State Phenomena in Energy Materials (C2SEPEM), which is funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division under Contract No. DE-AC02-05CH11231, as part of the Computational Materials Sciences Program. F.H.J. was supported by the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences (BES), under award DE-SC0021984. We acknowledge the use of computational resources at the National Energy Research Scientific Computing Center (NERSC), a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. The authors acknowledge the Texas Advanced Computing Center (TACC) at The University of Texas at Austin for providing HPC resources that have contributed to the research results reported within this paper. YHC acknowledges the support of Academia Sinica under project No. AS-CDA-114-M04 and thanks the National Center for High-Performance Computing (NCHC) for providing computational and storage resources.

Derivation of the GKBA Collision Integral

In this appendix, we give the derivation of the collision term with GKBA and the details of the G1-G2 scheme. Our derivation closely follows those in Ref.53 and Ref.42. We start with the equation of motion of the density matrix, which is derived by taking the time-diagonal part of the lesser Green’s function in the Kadanoff-Baym equation (KBE),

dρijQdt+i[H(t),ρ(t)]ijQ=−[IL,ijQ<+H.c.].\frac{d\rho_{ij\mathbf{Q}}}{dt}+i[H(t),\rho(t)]_{ij\mathbf{Q}}=-\left[I^{<}_{L,ij\mathbf{Q}}+\mathrm{H.c.}\right].

where H(t)H(t) and IL<I^{<}_{L} are defined in the main text. The collision integral reads,

IL<(t)=∫0tdt′[Σ>(t,t′)G<(t′,t)−Σ<(t,t′)G>(t′,t)].I^{<}_{L}(t)=\int_{0}^{t}dt'\left[\Sigma^{>}(t,t')G^{<}(t',t)-\Sigma^{<}(t,t')G^{>}(t',t)\right].

where Σ>,<\Sigma^{>,<} and G>,<G^{>,<} are the greater and lesser components of self-energy and Green’s function, respectively. We consider a Fan term-like self energy for excitonphonon interactions46,47,

ΣijQ>(t,t′)=i∑ν,n,l,qGνin(Q−q,q)Gνjl∗(Q−q,q)GnlQ−q>(t,t′)Dνq<(t′,t).(7)\Sigma^{>}_{ij\mathbf{Q}}(t,t')=i\sum_{\nu,n,l,\mathbf{q}}\mathcal{G}_{\nu in}(\mathbf{Q}-\mathbf{q},\mathbf{q})\mathcal{G}^{*}_{\nu jl}(\mathbf{Q}-\mathbf{q},\mathbf{q})G^{>}_{nl\mathbf{Q}-\mathbf{q}}(t,t')D^{<}_{\nu\mathbf{q}}(t',t). \tag*{(7)}

The collision integral is a functional of the two-time Green’s function G>,<(t,t′)G^{>,<}(t,t'). Therefore, without further approximation, solving the density matrix dynamics would be equivalent to solve the full KBE. To make progress, we employ the GKBA, which is a standard approximation of the two-time Green’s function. Assuming t>t′t>t', we have

G<(t,t′)=−GR(t,t′)ρ(t′)G>(t,t′)=GR(t,t′)ρˉ(t′)G<(t′,t)=ρ(t′)GA(t′,t)G>(t′,t)=−ρˉ(t′)GA(t′,t)(8)\begin{aligned} G^{<}(t,t')&=-G^{R}(t,t')\rho(t')\\ G^{>}(t,t')&=G^{R}(t,t')\bar{\rho}(t')\\ G^{<}(t',t)&=\rho(t')G^{A}(t',t)\\ G^{>}(t',t)&=-\bar{\rho}(t')G^{A}(t',t) \tag*{(8)} \end{aligned}

, where ρˉ(t)=I+ρ(t)\bar{\rho}(t)=I+\rho(t) with the identity matrix II. Substituting Green’s function with the ansatz (8) in the collision integral, we obtain (4) in the main text. We proceed by using the Wigner-Weisskopf form for the propagators and further approximating the phonon propagator with their equilibrium form, 55

Dνq>(t,t′)=−i∑α=±Nνqαe−iαωνq(t−t′).D^{>}_{\nu\mathbf{q}}(t,t')=-i\sum_{\alpha=\pm}N^{\alpha}_{\nu\mathbf{q}}e^{-i\alpha\omega_{\nu\mathbf{q}}(t-t')}.

and

Dνq<(t,t′)=−i∑α=±Nνqαe−iαωνq(t−t′).D^{<}_{\nu\mathbf{q}}(t,t')=-i\sum_{\alpha=\pm}N^{\alpha}_{\nu\mathbf{q}}e^{-i\alpha\omega_{\nu\mathbf{q}}(t-t')}.

where Nνq±=nνq+12±12N^{\pm}_{\nu\mathbf{q}}=n_{\nu\mathbf{q}}+\frac{1}{2}\pm\frac{1}{2}. (4) can be reduced to a form analogous to the Lindblad framework by invoking the Markov approximation and extending the time-integration limit to t→∞t\to\infty53. We emphasize that the basis dependence inherent in the Lindblad approach is a direct mathematical consequence of this infinite-time integration; the resulting collision operator becomes tied to the specific representation in which the energy-conserving delta function is enforced, thereby breaking the form-invariance of the dynamics.

G1-G2 scheme

The GKBA facilitates a numerically efficient scheme that reduces the computational complexity of evaluating (4) from a scheme that scales as O(t2)O(t^{2})54,56 down to O(t)O(t). This is achieved by introducing a set of the auxiliary density matrices, defined as

FnQ−qpQν±<,>(t)=Gνjl,Q−q,qν∗∫0tdt′GnsQ−qR(t,t′)ρˉslQ−q(t′)Dνq±<,>(t′,t)ρjmQ(t′)GmpQA(t′,t).F^{\nu\pm<,>}_{n\mathbf{Q}-\mathbf{q}p\mathbf{Q}}(t)=\mathcal{G}^{\nu*}_{\nu jl,\mathbf{Q}-\mathbf{q},\mathbf{q}}\int_{0}^{t}dt'G^{R}_{ns\mathbf{Q}-\mathbf{q}}(t,t')\bar{\rho}_{sl\mathbf{Q}-\mathbf{q}}(t')D^{\pm<,>}_{\nu\mathbf{q}}(t',t)\rho_{jm\mathbf{Q}}(t')G^{A}_{mp\mathbf{Q}}(t',t).

where D>,<±D^{>,<\pm} represent the α=±\alpha=\pm branch of the phonon propagator. Within this framework, the collision integral (4) can be recast in terms of these auxiliary matrices,

IipQL(t)=i∑νq,α=±Gνin(Q−q,q)[FnQ−qpQνα,<(t)−FnQ−qpQνα,>(t)].I^{L}_{ip\mathbf{Q}}(t)=i\sum_{\nu\mathbf{q},\alpha=\pm}\mathcal{G}_{\nu in}(\mathbf{Q}-\mathbf{q},\mathbf{q})\left[F^{\nu\alpha,<}_{n\mathbf{Q}-\mathbf{q}p\mathbf{Q}}(t)-F^{\nu\alpha,>}_{n\mathbf{Q}-\mathbf{q}p\mathbf{Q}}(t)\right].

By taking the time derivative of the auxiliary density matrix, we obtain a closed equation of motion

iddtFnQ−qpQν+<(t)=hnm(t)FmQ−qpQν+<(t)−FnQ−qbQν+<(t)hbp(t)+ωνqFnQ−qpQν+<(t)+Nνq+Gνjl(Q−q,q)∗ρˉnlQ−q(t)ρjpQ(t).\begin{aligned} i\frac{d}{dt}F^{\nu+<}_{n\mathbf{Q}-\mathbf{q}p\mathbf{Q}}(t) ={}&h_{nm}(t)F^{\nu+<}_{m\mathbf{Q}-\mathbf{q}p\mathbf{Q}}(t)-F^{\nu+<}_{n\mathbf{Q}-\mathbf{q}b\mathbf{Q}}(t)h_{bp}(t)\\ &+\omega_{\nu\mathbf{q}}F^{\nu+<}_{n\mathbf{Q}-\mathbf{q}p\mathbf{Q}}(t)\\ &+N^{+}_{\nu\mathbf{q}}\mathcal{G}_{\nu jl}(\mathbf{Q}-\mathbf{q},\mathbf{q})^{*}\bar{\rho}_{nl\mathbf{Q}-\mathbf{q}}(t)\rho_{jp\mathbf{Q}}(t). \end{aligned}

The equation of motion for the remaining three auxiliary density matrices are derived similarly. This transformation converts the integro-differential equation of the density matrix into a coupled system of first-order ordinary differential equations, effectively ’mapping’ the memory kernel onto the dynamics of the auxiliary variables. In the numerical implementation, storage of the auxiliary density matrix becomes a bottleneck. We take advantages of large scale parallelization to overcome this issue.

While this approach bypasses the unfavorable O(t2)O(t^{2}) scaling, the high dimensionality of the auxiliary matrices—which must be stored for each Q\mathbf{Q}-point and exciton and phonon indices—presents a significant memory bottleneck. We address this through large-scale distributedmemory parallelization, enabling the simulation of full Brillouin zone dynamics.

Temperature Dependent Valley Dynamics

Figure 8 illustrates the temperature dependence of the valley dynamics calculated as calculated within the Lindblad and GKBA frameworks. Both valley depolarization (Figure 8 (a)) and decoherence (Figure 8 (c)) dynamics exhibit mono-exponential decay in the Lindblad simulations across all temperatures. In contrast, the GKBA results feature high-frequency oscillations during the initial 50 fs, which is subsequently transition into a mono-exponential decay regime. The amplitude of these non-Markovian transients increases with temperature, suggesting that their magnitude scales with the total scattering rate-a consequence of the increased phonon population at elevated temperature.

Temperature-dependent valley dynamics across Lindblad and GKBA frameworks

Figure 8. Temperature-dependent valley dynamics. Computed valley depolarization (panels a, b) and decoherence (panels c, d) dynamics across a temperature range of 10–500 K, comparing the Lindblad and GKBA frameworks. Symbols represent the calculated data points, while dashed lines indicate the corresponding mono-exponential fits used to extract the characteristic decay times τ\tau. In the GKBA results, note the presence of non-Markovian transients during the initial 50 fs that precede the onset of the exponential relaxation regime.

References

  1. [1]Ting Cao, Gang Wang, Wenpeng Han, Huiqi Ye, Chuanrui Zhu, Junren Shi, Qian Niu, Pingheng Tan, Enge Wang, Baoli Liu, and Ji Feng, “Valley-selective circular dichroism of monolayer molybdenum disulphide,” Nature Communications 3, 887 (2012).
  2. [2]Di Xiao, Gui-Bin Liu, Wanxiang Feng, Xiaodong Xu, and Wang Yao, “Coupled spin and valley physics in monolayers of mos2 and other group-vi dichalcogenides,” Phys. Rev. Lett. 108, 196802 (2012).DOI
  3. [3]Diana Y. Qiu, Felipe H. da Jornada, and Steven G. Louie, “Optical spectrum of mos2: Many-body effects and diversity of exciton states,” Phys. Rev. Lett. 111, 216805 (2013).pubmed.ncbi.nlm.nih.gov/24313514
  4. [4]Miguel M. Ugeda, Aaron J. Bradley, Su-Fei Shi, Felipe H. da Jornada, Yi Zhang, Diana Y. Qiu, Wei Ruan, Sung-Kwan Mo, Zahid Hussain, Zhi-Xun Shen, Feng Wang, Steven G. Louie, and Michael F. Crommie, “Giant bandgap renormalization and excitonic effects in a monolayer transition metal dichalcogenide semiconductor,” Nature Materials 13, 1091–1095 (2014).
  5. [5]Alexey Chernikov, Timothy C. Berkelbach, Heather M. Hill, Albert Rigosi, Yilei Li, Ozgur Burak Aslan, David R. Reichman, Mark S. Hybertsen, and Tony F. Heinz, “Exciton binding energy and nonhydrogenic rydberg series in monolayer ws2,” Phys. Rev. Lett. 113, 076802 (2014).DOI
  6. [6]Diana Y. Qiu, Felipe H. da Jornada, and Steven G. Louie, “Screening and many-body effects in two-dimensional crystals: Monolayer mos2,” Phys. Rev. B 93, 235435 (2016).DOI
  7. [7]Gang Wang, Alexey Chernikov, Mikhail M. Glazov, Tony F. Heinz, Xavier Marie, Thierry Amand, and Bernhard Urbaszek, “Colloquium: Excitons in atomically thin transition metal dichalcogenides,” Rev. Mod. Phys. 90, 021001 (2018).DOI
  8. [8]Hualing Zeng, Junfeng Dai, Wang Yao, Di Xiao, and Xiaodong Cui, “Valley polarization in mos2 monolayers by optical pumping,” Nature Nanotechnology 7, 490–493 (2012).DOI
  9. [9]Kin Fai Mak, Keliang He, Jie Shan, and Tony F. Heinz, “Control of valley polarization in monolayer mos2 by optical helicity,” Nature Nanotechnology 7, 494–498 (2012).DOI
  10. [10]G. Sallen, L. Bouet, X. Marie, G. Wang, C. R. Zhu, W. P. Han, Y. Lu, P. H. Tan, T. Amand, B. L. Liu, and B. Urbaszek, “Robust optical emission polarization in mos2 monolayers through selective valley excitation,” Phys. Rev. B 86, 081301(R) (2012).
  11. [11]Xiaodong Xu, Wang Yao, Di Xiao, and Tony F. Heinz, “Spin and pseudospins in layered transition metal dichalcogenides,” Nature Physics 10, 343–350 (2014).DOI
  12. [12]John R. Schaibley, Hongyi Yu, Genevieve Clark, Pasqual Rivera, Jason S. Ross, Kyle L. Seyler, Wang Yao, and Xiaodong Xu, “Valleytronics in 2d materials,” Nature Reviews Materials 1, 16055 (2016).DOI
  13. [13]Ziliang Ye, Dezheng Sun, and Tony F. Heinz, “Optical manipulation of valley pseudospin,” Nature Physics 13, 26–29 (2017).DOI
  14. [14]Thomas Mueller and Ermin Malic, “Exciton physics and device application of two-dimensional transition metal dichalcogenide semiconductors,” npj 2D Materials and Applications 2, 29 (2018).DOI
  15. [15]Siwen Zhao, Xiaoxi Li, Baojuan Dong, Huide Wang, Hanwen Wang, Yupeng Zhang, Zheng Han, and Han Zhang, “Valley manipulation in monolayer transition metal dichalcogenides and their hybrid systems: status and challenges,” Reports on Progress in Physics 84, 026401 (2021).DOI
  16. [16]Alejandro Molina-Sánchez, Davide Sangalli, Ludger Wirtz, and Andrea Marini, “Ab initio calculations of ultrashort carrier dynamics in two-dimensional materials: Valley depolarization in single-layer wse2,” Nano Lett. 17, 4549–4555 (2017).

Paper details

vlnc:W_gx38uc4s
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