betaFeedback

WorksMathematics

Incorporating Memory into Propagation of 1-Electron Reduced Density Matrices

Abstract

For any linear system with unreduced dynamics governed by invertible propagators, we derive a closed, time-delayed, linear system for a reduced-dimensional quantity of interest. This method does not target dimensionality reduction: rather, this method helps shed light on the memory-dependence of 11-electron reduced density matrices in time-dependent configuration interaction (TDCI), a scheme to solve for the correlated dynamics of electrons in molecules. Though time-dependent density functional theory has established that the 11-electron reduced density possesses memory-dependence, the precise nature of this memory-dependence has not been understood. We derive a symmetry/constraint-preserving method to propagate reduced TDCI electron density matrices. In numerical tests on two model systems (H2\text{H}_2 and HeH+\text{HeH}^+), we show that with sufficiently large time-delay (or memory-dependence), our method propagates reduced TDCI density matrices with high quantitative accuracy. We study the dependence of our results on time step and basis set. To implement our method, we derive the 44-index tensor that relates reduced and full TDCI density matrices. Our derivation applies to any TDCI system, regardless of basis set, number of electrons, or choice of Slater determinants in the wave function.

Introduction

For an NN-electron system, the time-dependent Schrödinger equation (TDSE) is a partial differential equation for the system’s electronic wave function Ψ\Psi. Ignoring spin and assuming fixed nuclear coordinates within the Born-Oppenheimer approximation, Ψ\Psi depends on 3N3N spatial variables and time. This electronic wave function governs the electron dynamics of the system, important in ultrafast charge transfer in the initial steps of photosynthesis and laser-induced processes, among others[1, 2, 3, 4]. Unless NN is very small, we cannot expect to solve for Ψ\Psi. Within a finite-dimensional basis set, a numerically exact solution of the time-independent problem is possible with configuration interaction (CI)[5, 6, 7, 8]. CI expands Ψ\Psi in a basis of Slater determinants; TDCI uses the same expansion but allows the coefficients to depend on time, enabling the solution of genuinely time-dependent problems—see Sections III and IV for details. When all possible determinants are used, as in the present paper, we refer to the resulting Ψ\Psi as a full CI wave function.

The full CI wave function was recently calculated for the C3H8\mathrm{C}_3\mathrm{H}_8 molecule (N=26N=26), which used distributed computing to determine the energy of the system with 1.31 trillion determinants, the largest known full CI calculation to date[9]. The full CI wave function can be approximated with various levels of truncation of the expansion of determinants, as is commonly done for molecular systems. In the limit where we use only a single Slater determinant to approximate Ψ\Psi, we have Hartree-Fock theory, which neglects electron correlation[10].

Note that Ψ=Ψ(r1,r2,…,rN,t)\Psi=\Psi(\mathbf{r}_1,\mathbf{r}_2,\ldots,\mathbf{r}_N,t) where each rj\mathbf{r}_j is three-dimensional. Let us now define the 1-electron reduced density ρ(r,t)\rho(\mathbf{r},t) as the integral of ∣Ψ(r,r2,…,rN,t)∣2|\Psi(\mathbf{r},\mathbf{r}_2,\ldots,\mathbf{r}_N,t)|^2 with respect to r2,…,rN\mathbf{r}_2,\ldots,\mathbf{r}_N. Clearly, ρ\rho is a much lower-dimensional object than Ψ\Psi. Many observables of practical interest can be computed easily and explicitly from ρ\rho; additionally, given any physical observable, there exists[11] a mapping expressing it as a function of ρ\rho.

Time-dependent density functional theory (TDDFT), which can be derived[11] from the TDSE, gives us an alternative route to compute 1-electron reduced densities ρ\rho without computing the full wave function Ψ\Psi. For the same NN-electron system, TDDFT yields computationally tractable evolution equations in three spatial variables plus time; solving these equations enables us to compute the 1-electron reduced density ρ(r,t)\rho(\mathbf{r},t). The Runge-Gross theorem establishes the existence of exchange-correlation potentials VxcV_{\mathrm{xc}} such that the 1-electron densities computed via TDDFT exactly match those computed from solutions of the TDSE12. Thus, in principle, TDDFT provides an accurate, scalable path to computing electron dynamics. Note that VxcV_{\mathrm{xc}} is shorthand for Vxc[ρ](r,t)V_{\mathrm{xc}}[\rho](\mathbf{r},t)—this potential depends functionally on the ρ\rho trajectory {ρ(w,s)∣w∈R3,s≤t}\{\rho(\mathbf{w},s) \mid\mathbf{w} \in\mathbb{R}^{3},s \le t\}. Because exact, time-dependent exchange-correlation functionals are unavailable, TDDFT practitioners use approximations13 in which Vxc[ρ](r,t)V_{\mathrm{xc}}[\rho](\mathbf{r},t) depends only on {ρ(w,t)∣w∈R3}\{\rho(\mathbf{w},t) \mid\mathbf{w} \in\mathbb{R}^{3}\}, ignoring memory-dependence on ρ(⋅,s)\rho(\cdot,s) for s<ts < t. This ubiquitous approximation, known as the adiabatic approximation within the TDDFT community, leads to errors that have been quantified in prior work[14, 15, 16, 17, 18, 19].

One way to devise more accurate VxcV_{\mathrm{xc}} models that incorporate memory-dependence is to use machine learning. For a spatially one-dimensional, two-electron model problem, the approach of learning memory-dependent VxcV_{\mathrm{xc}} models from data has been pursued recently[20, 21]. To scale these machine learning methods to larger, three-dimensional molecular systems, it will be essential to incorporate prior knowledge regarding the memory-dependence of VxcV_{\mathrm{xc}}. This will enable targeted searches over neural network architectures that match physically relevant dynamics. In TDDFT, memory-dependence of VxcV_{\mathrm{xc}} results in time-delayed equations of motion for 1-electron Kohn-Sham orbitals[22]. The sum of the squared magnitudes of all such 1-electron Kohn-Sham orbitals equals the 1-electron reduced density ρ\rho. Memory-dependence of VxcV_{\mathrm{xc}} gives rise to memory-dependence of ρ\rho.

Let P(t)P(t) and Q(t)Q(t) denote, respectively, the full density matrix and 1-electron reduced density matrix (1RDM) that can be computed from the TDCI solution Ψ\Psi. In this paper, we use all possible Slater determinants in our CI expansion for Ψ\Psi. Then the resulting Ψ\Psi, PP, and QQ all incorporate the effects of electron correlation and exchange. Our TDCI approach circumvents the approximations of methods such as time-dependent Hartree-Fock and TDDFT. Therefore, we can logically aim to improve a theory such as TDDFT by studying the dynamics of TDCI 1-electron reduced density matrices Q(t)Q(t). The missing piece has been an equation of motion for Q(t)Q(t).

In this paper, we derive and study a closed, time-delay system that enables propagation of TDCI 1-electron reduced density matrices Q(t)Q(t). As this system requires the full many-body propagator, it does not achieve computational savings. Instead, we use the system to generate insight into the explicit memory-dependence of Q(t)Q(t) on past values Q(s)Q(s) for s<ts < t. As we explain in Section IV, the TDCI 1-electron reduced density matrices Q(t)Q(t) are closely related to the 1-electron reduced densities ρ\rho in TDDFT. Therefore, insight into the memory-dependence of the TDCI reduced electron density matrices Q(t)Q(t), enabled by the methods of this paper, will assist us in improving future models of VxcV_{\mathrm{xc}} for TDDFT.

The idea of trading high-dimensionality of one dynamical system (2) for time-delayed dynamics in a reduced-dimensional space (4) is connected to notions from Mori-Zwanzig theory[23, 24, 25, 26, 27], Takens embedding theory[28, 29, 30], and Koopman operator theory[31, 32, 33]. We now review these connections.

Given a dynamical system in a high-dimensional space, Mori-Zwanzig theory produces closed systems for the time-evolution of low-dimensional observables[23, 24]. The resulting equations trade (i) non-time-delayed dynamics in the original, high-dimensional space for (ii) time-delayed dynamics in a low-dimensional space[27]. The time delays are expressed in terms of a memory kernel. Recent work has connected Mori-Zwanzig theory to Koopman operator methods, showing how to learn the memory kernel from data[34].

Our method shares with Mori-Zwanzig theory the trade-off between (i) and (ii). A key difference is that Mori-Zwanzig methods often assume that the unobserved or irrelevant variables can be treated as noise, leading to generalized Langevin equations[25, 26]. If we view the transformation from full electron density matrices P(t)P(t) to 1-electron reduced density matrices Q(t)Q(t) as projection onto a subspace SS, then the unobserved/irrelevant variables are the projection of P(t)P(t) onto the orthogonal complement S⊥S^{\perp}. We make no assumptions regarding the dynamics of these variables; they do not appear at all in our self-contained scheme for propagating Q(t)Q(t). This is a key difference between our work and Mori-Zwanzig approaches. In Section IIB, we explain this difference further through a concrete example.

Takens justified trading (i) for (ii) in the sense of establishing a sufficient criterion for how much memory is required (i.e., the number of time-delayed scalar observations) to accurately represent dynamics in the original phase space[28]. Takens’ theory establishes the existence of time-delayed embeddings, but does not give methods to find them. Time-delayed embeddings are linked to Koopman operator methods[35], which are themselves closely connected to the dynamic mode decomposition (DMD)[33, 36, 37]. In Section II A, we show concretely that our method results in the same type of model as time-delayed DMD, but without any need to learn from data.

Our method shares structural features with self-energy methods where one solves for a many-body system’s Green’s function G(t,t′)G(t,t') via Dyson’s equation[38, 39]

(i∂t−H0)G(t,t′)=δ(t−t′)+∫0tΣ(t−t′′)G(t′′,t′) dt′′.(\mathrm{i}\partial_t-H_0)G(t,t')=\delta(t-t')+\int_0^t\Sigma(t-t'')G(t'',t')\,\mathrm{d}t''.

As in Mori-Zwanzig theory, forward evolution of GG depends on past states via a memory integral, here represented by convolution with the self-energy Σ\Sigma, which itself depends on GG. Here H0H_0 is the non-interacting Hamiltonian; Σ\Sigma incorporates interactions that go beyond mean-field theories[40]. There are close analogies between the terms that comprise Σ\Sigma and the Coulomb and exchange-correlation (VxcV_{\mathrm{xc}}) potentials that one finds in TDDFT[41]. As with VxcV_{\mathrm{xc}}, in general, the analytical form of Σ\Sigma is unknown, necessitating approximations such as GW[42, 43, 44]. Recent work has sought to use machine learning to improve these approximations[45, 46]. The present paper’s methods may yield insight into how to model Σ\Sigma, especially its temporal decay within the memory integral.

A key feature of our system is that the propagator is unitary, implying the reversibility of the dynamics. The Liouville-von Neumann equation (1) that governs the dynamics of P(t)P(t) is not dissipative. Hence we cannot directly apply balanced truncation methods that have been developed for open quantum systems[47, 48]. Variants of these methods, when applied to the TDCI system, may yield substantial computational savings. This is outside the scope of the present work.

Liouville-von Neumann

In several contexts, electron density matrices P(t)P(t) satisfy an equation of motion known as the Liouville-von Neumann equation:

idPdt=[H,P].(1)\mathrm{i}\frac{\mathrm{d}P}{\mathrm{d}t}=[H,P]. \tag*{(1)}

Here [H,P]=HP−PH[H,P]=HP-PH denotes the commutator. In fact, we show in Section III that the TDCI full electron density matrices P(t)P(t) satisfy (1). The 1RDM P~(t)\widetilde{P}(t) from time-dependent Hartree-Fock theory satisfies (1) with a density-dependent Hamiltonian H(P~)H(\widetilde{P})[6, 49]. Thus it is natural to ask whether the TDCI 1-electron reduced density matrices Q(t)Q(t) evolve via (1). If they do, one can show that they must satisfy Q(t)=U(t)†Q(0)U(t)Q(t)=U(t)^{\dagger}Q(0)U(t), where U(t)U(t) is unitary and † denotes conjugate transpose; consequently, the eigenvalues of Q(t)Q(t) must equal those of Q(0)Q(0) for all t≥0t\geq0.

This gives us a way to test if it is possible that TDCI 1-electron reduced density matrices Q(t)Q(t) satisfy the Liouville-von Neumann equation. For four two-electron molecular systems (HeH+{}^{+} and H2_2 in both STO-3G and 6-31G atomic orbital basis sets, for which we can compute the full CI solution with all possible determinants), we have applied TDCI methods—see Sections III and § 0.5 A—to compute full density matrices P(t)P(t), from which we have computed 1-electron reduced density matrices Q(t)Q(t). Let us focus attention on HeH+{}^{+} in the STO-3G atomic orbital basis set; here Q(t)Q(t) is 2×22\times2 and thus has two eigenvalues, λ1(t)\lambda_1(t) and λ2(t)\lambda_2(t). In Figure 1, we plot the absolute deviation ∣λj(t)−λj(0)∣|\lambda_j(t)-\lambda_j(0)|, and we see that the eigenvalues of the Q(t)Q(t) matrices do not stay constant in time. We omit plots for the other three molecular systems, all of which would show substantial fluctuations in the eigenvalues of Q(t)Q(t).

Figure 1. For the molecule HeH+^{+} in the STO-3G atomic orbital basis set, we apply the methods described in § 0.5 A to compute 1-electron reduced density matrices Q(t)Q(t). From the left plot, we conclude that the eigenvalues of Q(t)Q(t) do not stay constant in time, implying that Q(t)Q(t) cannot satisfy the Liouville-von Neumann equation for any choice of Hamiltonian H(t)H(t). The right plot shows that trace⁡(Q(t))=2\operatorname{trace}(Q(t)) = 2. Since HeH+^{+} has N=2N = 2 electrons, this agrees with trace⁡(Q(t))=N\operatorname{trace}(Q(t)) = N, which we prove in Appendix B.

Because the eigenvalues of the TDCI 1-electron reduced density matrices Q(t)Q(t) do not stay constant in time, there does not exist a Hamiltonian HH such that these Q(t)Q(t) matrices satisfy the Liouville-von Neumann equation. Thus we cannot apply methods from our prior work[50, 51] to learn a Hamiltonian HH such that these Q(t)Q(t) satisfy (1).

Reduced Density Matrices (RDMs)

A subset of the many-body quantum physics and chemistry literature is concerned with deriving governing equations for RDMs, for both time-independent[52, 53, 54] and time-dependent[55, 56, 57, 58] problems. Consider the operator obtained by integrating ∣Ψ⟩⟨Ψ∣|\Psi\rangle\langle\Psi| with respect to all spins and all but kk of the electrons’ positions; the matrix representation of this operator is the kk-electron reduced density matrix or kkRDM. Here we have used bra-ket notation—see Definition 4. Starting from the Schrödinger equation, one can derive a governing equation for the 1RDM Q(t)Q(t), but it will necessarily involve the 2RDM[59]. While there have been multiple attempts to close the 1RDM equation via approximate reconstruction of the 2RDM from the 1RDM[53, 56, 60, 61, 62], we have not seen any prior attempts to use memory (present and past 1RDMs) to derive a closed system. More generally, if one seeks an equation of motion for the 2RDM, it will either involve reconstruction of higher-order matrices such as the 3RDM or 4RDM[53, 63, 64], or a reexpression of constraints on higher-order RDMs (the NN-representability conditions)[65, 66]. Even within this literature on higher-order RDM systems, the use of memory appears to be unexplored.

Summary of Contributions

In Section II, we give a general method to derive a self-contained linear delay equation for the propagation of a reduced quantity of interest. We provide this section to highlight the key steps of the derivation, which already constitutes one novel contribution of this paper, prior to introducing any notions from quantum mechanics. The general setting also allows us to contrast our method with both time-delayed DMD and a Mori-Zwanzig approach, which we do in Section II.

Section III reviews essential background material, and is intended to increase the accessibility of the paper to applied mathematicians. Here we define a subset of concepts from quantum/computational chemistry such as orbitals, Slater determinants, and basis functions. We review TDCI and its relation to the TDSE. We show that the full density matrix P(t)P(t) satisfies the Liouville-von Neumann equation. Theorem 1 in Appendix B shows that 1-electron reduced density matrices Q(t)Q(t) have constant trace equal to NN, as exemplified by the right plot in Figure 1.

In Section IV, we specialize the derivation from Section II to the TDCI setting. We begin by deriving several useful properties of 1-electron reduced density matrices that stem from TDCI. The derivations here are general with respect to the number of electrons (NN), the number of atomic/molecular orbitals (KK), and the number of configurations (or Slater determinants) included in the wave function (NCN_{\mathrm{C}}). Proposition 1 enables concrete evaluation of the 4-index tensor BB that relates full configuration interaction (CI) electron density matrices to 1-electron reduced density matrices. Theorem 6 establishes a self-contained delay equation for Q(t)Q(t) that preserves Hermitian symmetry, constant trace, and identically zero elements.

We present numerical results in § 0.5, applying the time-delayed scheme from Theorem 6 to propagate reduced density matrices Q(t)Q(t) for four different molecular systems. First, we establish the accuracy of the Q(t)Q(t) matrices computed using the scheme from Theorem 6. For each molecular system, accuracy is controlled by the total memory length, which is defined to be the physical time corresponding to the maximum time delay in the evolution equation for Q(t)Q(t). With sufficiently large total memory length, the resulting Q(t)Q(t) matrices computed using our method match those computed directly from full TDCI simulations. We also quantify how the accuracy of Q(t)Q(t) depends on our choice of time step, basis set, and stride. A stride of kk means that, in the equation to propagate Q(t)Q(t), we use every kk-th previous reduced density matrix, up to a preset limit.

Core Ideas

Here we present derivations that illustrate the core ideas of our approach without the complexity of the quantum system that we present and analyze in Sections III and IV. The derivation in Section II A clarifies the key steps of our time-delay scheme to propagate y(t)y(t). In Section IV C, we revisit and refine this derivation for the quantum chemistry context. The derivation in Section II B frames our problem as a partially observed linear system, enabling a comparison with Mori-Zwanzig theory. We begin with the discrete-time linear system

z(t+1)=A(t)z(t)(2)z(t+1) = A(t)z(t) \tag*{(2)}

with initial condition z(0)z(0). Here tt is a nonnegative integer, A(t)A(t) is an n×nn \times n complex, unitary matrix, and z(t)z(t) is a complex n×1n \times1 vector. We take A(t)A(t) to be unitary as this corresponds to quantum systems of interest. More generally, the derivations in this section hold when A(t)A(t) is invertible, a condition always satisfied when (2) arises by discretizing a linear ordinary differential equation with a sufficiently small time step. When AA is unitary, A†=A−1A^\dagger= A^{-1}; in case A(t)A(t) is invertible but not unitary, one can replace A(t)†A(t)^\dagger with A(t)−1A(t)^{-1}.

Suppose we are given a complex m×nm \times n matrix RR with m≪nm \ll n that enables one to compute, at each time tt, a reduced quantity of interest:

y(t)=Rz(t).(3)y(t) = Rz(t). \tag*{(3)}

Here y(t)y(t) is a complex m×1m \times1 vector. Starting with y(t+1)y(t+1) and using (2-3), we have

y(t+1)=Rz(t+1)=RA(t)z(t).(4)y(t+1) = Rz(t+1) = RA(t)z(t). \tag*{(4)}

To close this equation, and thereby enable propagation of y(t)y(t), we must express z(t)z(t) in terms of {y(t−ℓ),…,y(t−1),y(t)}\{y(t-\ell), \ldots, y(t-1), y(t)\} (for some ℓ≤t\ell\le t).

Self-Contained Propagation of Reduced Quantities

From (2), we can derive

z(t−ℓ)=∏j=0ℓ−1A(t−ℓ+j)†z(t).(5)z(t-\ell) = \prod_{j=0}^{\ell-1} A(t-\ell+j)^\dagger z(t). \tag*{(5)}

Multiplying both sides by RR and using (3) at time t−ℓt-\ell yields

y(t−ℓ)=R∏j=0ℓ−1A(t−ℓ+j)†z(t).(6)y(t-\ell) = R \prod_{j=0}^{\ell-1} A(t-\ell+j)^\dagger z(t). \tag*{(6)}

Stacking column vectors vertically and applying both (3) and (6), we obtain

References

  1. [1]J. Jortner, M. Bixon, and M. E. Michel-Beyerle, “Electron transfer dynamics in photosynthesis,” in Current Research in Photosynthesis: Proceedings of the VIIIth International Conference on Photosynthesis Stockholm, Sweden, August 6-11, 1989, edited by M. Baltscheffsky (Springer Netherlands, Dordrecht, 1990) pp. 11–18.
  2. [2]M. F. Kling and M. J. Vrakking, “Attosecond electron dynamics,” Annual Review of Physical Chemistry 59, 463–492 (2008).
  3. [3]M. R. Provorse and C. M. Isborn, “Electron dynamics with real-time time-dependent density functional theory,” International Journal of Quantum Chemistry 116, 739–749 (2016).
  4. [4]X. Li, N. Govind, C. Isborn, A. E. I. DePrince, and K. Lopata, “Real-time time-dependent electronic structure theory,” Chemical Reviews 120, 9951–9993 (2020).
  5. [5]I. Shavitt, “The method of configuration interaction,” in Methods of Electronic Structure Theory, edited by H. F. Schaefer (Springer US, Boston, MA, 1977) pp. 189–275.
  6. [6]R. McWeeny, Methods of Molecular Quantum Mechanics, 2nd ed., Chemistry, Physical and Theoretical (Academic Press, 1989).
  7. [7]A. Szabo and N. S. Ostlund, Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory (Dover Publications, 2012) “Unabridged, unaltered republication of the First Edition, Revised, originally published by McGraw-Hill, New York in 1989”.
  8. [8]I. Shavitt, “The history and evolution of configuration interaction,” Molecular Physics 94, 3–17 (1998).
  9. [9]H. Gao, S. Imamura, A. Kasagi, and E. Yoshida, “Distributed implementation of full configuration interaction for one trillion determinants,” Journal of Chemical Theory and Computation 20, 1185–1192 (2024).DOI
  10. [10]P.-O. Löwdin, “The historical development of the electron correlation problem,” International Journal of Quantum Chemistry 55, 77–102 (1995).DOI
  11. [11]C. Ullrich, Time-Dependent Density-Functional Theory: Concepts and Applications, Oxford Graduate Texts (Oxford University Press, Oxford, 2012).
  12. [14]P. Elliott, J. I. Fuks, A. Rubio, and N. T. Maitra, “Universal dynamical steps in the exact time-dependent exchange-correlation potential,” Physical Review Letters 109, 266404 (2012).
  13. [15]D. Rappoport and J. Hutter, “Excited-State Properties and Dynamics,” in Fundamentals of Time-Dependent Density Functional Theory, edited by M. A. Marques, N. T. Maitra, F. M. Nogueira, E. Gross, and A. Rubio (Springer Berlin Heidelberg, Berlin, Heidelberg, 2012) pp. 317–336.
  14. [16]B. F. Habenicht, N. P. Tani, M. R. Provorse, and C. M. Isborn, “Two-electron Rabi oscillations in real-time time-dependent density-functional theory,” The Journal of Chemical Physics 141, 184112 (2014).DOI
  15. [17]M. R. Provorse, B. F. Habenicht, and C. M. Isborn, “Peak-shifting in real-time time-dependent density functional theory,” Journal of Chemical Theory and Computation 11, 4791–4802 (2015).DOI
  16. [18]L. Lacombe and N. T. Maitra, “Developing new and understanding old approximations in TDDFT,” Faraday Discussions 224, 382–401 (2020).DOI
  17. [19]K. Ranka and C. M. Isborn, “Size-dependent errors in real-time electron density propagation,” The Journal of Chemical Physics 158, 174102 (2023).DOI
  18. [20]Y. Suzuki, R. Nagai, and J. Haruyama, “Machine learning exchange-correlation potential in time-dependent density-functional theory,” Phys. Rev. A 101, 050501 (2020).DOI
  19. [21]H. S. Bhat, K. Collins, P. Gupta, and C. M. Isborn, “Dynamic learning of correlation potentials for a time-dependent Kohn-Sham system,” in Proceedings of The 4th Annual Learning for Dynamics and Control Conference, Proceedings of Machine Learning Research, Vol. 168, , edited by R. Firoozi, N. Mehr, E. Yel, R. Antonova, J. Bohg, M. Schwager, and M. Kochenderfer (PMLR, 2022) pp. 546–558.
  20. [22]N. T. Maitra, K. Burke, and C. Woodward, “Memory in time-dependent density functional theory,” Phys. Rev. Lett. 89, 023002 (2002).
  21. [23]R. Zwanzig, “Memory effects in irreversible thermodynamics,” Phys. Rev. 124, 983–992 (1961).
  22. [24]H. Mori, “Transport, Collective Motion, and Brownian Motion,” Progress of Theoretical Physics 33, 423–455 (1965).
  23. [25]A. J. Chorin, O. H. Hald, and R. Kupferman, “Optimal prediction and the Mori-Zwanzig representation of irreversible processes,” Proceedings of the National Academy of Sciences 97, 2968–2973 (2000).
  24. [26]E. Darve, J. Solomon, and A. Kia, “Computing generalized Langevin equations and generalized Fokker–Planck equations,” Proceedings of the National Academy of Sciences 106, 10884–10889 (2009).DOI
  25. [27]A. J. Chorin and O. H. Hald, Stochastic Tools in Mathematics and Science, 3rd ed. (Springer New York, New York, NY, 2014).
  26. [28]F. Takens, “Detecting strange attractors in turbulence,” in Dynamical Systems and Turbulence, Warwick 1980 (Springer, 1981) pp. 366–381.
  27. [29]J. Stark, “Delay embeddings for forced systems. I. Deterministic forcing,” Journal of Nonlinear Science 9 (1999-6), 10.1007/s003329900072.DOI
  28. [30]H. Broer and F. Takens, Dynamical Systems and Chaos, Applied Mathematical Sciences, Vol. 172 (Springer, New York, 2010).
  29. [31]A. Mauroy, I. Mezić, and Y. Susuki, The Koopman Operator in Systems and Control: Concepts, Methodologies, and Applications, 1st ed., Lecture Notes in Control and Information Sciences, Vol. 484 (Springer, Netherlands, 2020).
  30. [32]S. E. Otto and C. W. Rowley, “Koopman operators for estimation and control of dynamical systems,” Annual Review of Control, Robotics, and Autonomous Systems 4, 59–87 (2021).
  31. [33]S. L. Brunton, M. Budišić, E. Kaiser, and J. N. Kutz, “Modern Koopman Theory for Dynamical Systems,” SIAM Review 64, 229–340 (2022).
  32. [34]Y. T. Lin, Y. Tian, D. Livescu, and M. Anghel, “Data-driven learning for the Mori-Zwanzig formalism: A generalization of the Koopman learning framework,” SIAM Journal on Applied Dynamical Systems 20, 2558–2601 (2021).arxiv.org/abs/2101.05873
  33. [35]I. Mezić and A. Banaszuk, “Comparison of systems with complex behavior,” Physica D: Nonlinear Phenomena 197, 101–133 (2004).
  34. [36]J. N. Kutz, S. L. Brunton, B. W. Brunton, and J. L. Proctor, Dynamic Mode Decomposition: Data-Driven Modeling of Complex Systems (SIAM, 2016).
  35. [37]M. Kambe, E. Kaiser, S. L. Brunton, and J. N. Kutz, “Time-delay observables for Koopman: Theory and applications,” SIAM Journal on Applied Dynamical Systems 19, 886–917 (2020).
  36. [38]G. Stefanucci and R. Van Leeuwen, Nonequilibrium Many-Body Theory of Quantum Systems: A Modern Introduction (Cambridge University Press, 2013).
  37. [39]L. P. Kadanoff, Quantum Statistical Mechanics (CRC Press, 2018).
  38. [40]J. W. Negele, “The mean-field theory of nuclear structure and dynamics,” Reviews of Modern Physics 54, 913 (1982).
  39. [41]G. Onida, L. Reining, and A. Rubio, “Electronic excitations: density-functional versus many-body Green’s-function approaches,” Rev. Mod. Phys. 74, 601–659 (2002).DOI
  40. [42]F. Aryasetiawan and O. Gunnarsson, “The GW method,” Reports on Progress in Physics 61, 237 (1998).
  41. [43]L. Reining, “The GW approximation: content, successes and limitations,” Wiley Interdisciplinary Reviews: Computational Molecular Science 8, e1344 (2018).DOI
  42. [44]L. Hedin, “Correlation effects in electron spectroscopies and the GW approximation,” Journal of Physics: Condensed Matter 11, R489 (1999).
  43. [45]H. Bassi, Y. Zhu, S. Liang, J. Yin, C. C. Reeves, V. Vlček, and C. Yang, “Learning nonlinear integral operators via recurrent neural networks and its application in solving integro-differential equations,” Machine Learning with Applications , 100524 (2023).arxiv.org/abs/2310.09434
  44. [46]Y. Zhu, J. Yin, C. C. Reeves, C. Yang, and V. Vlcek, “Predicting nonequilibrium Green’s function dynamics and photoemission spectra via nonlinear integral operator learning,” (2024), arXiv:2407.09773.arxiv.org/abs/2407.09773
  45. [47]B. Schäfer-Bung, C. Hartmann, B. Schmidt, and C. Schütte, “Dimension reduction by balanced truncation: Application to light-induced control of open quantum systems,” The Journal of Chemical Physics 135, 014112 (2011).
  46. [48]P. Benner, T. Breiten, C. Hartmann, and B. Schmidt, “Model reduction of controlled Fokker–Planck and Liouville–von Neumann equations,” Journal of Computational Dynamics 7, 1–33 (2020).DOI
  47. [49]A. D. McLachlan and M. A. Ball, “Time-dependent Hartree-Fock theory for molecules,” Reviews of Modern Physics 36, 844 (1964).
  48. [50]H. S. Bhat, K. Ranka, and C. M. Isborn, “Machine learning a molecular Hamiltonian for predicting electron dynamics,” International Journal of Dynamics and Control 8, 1089–1101 (2020).arxiv.org/abs/2007.09814
  49. [51]P. Gupta, H. S. Bhat, K. Ranka, and C. M. Isborn, “Statistical learning for predicting density-matrix-based electron dynamics,” Stat 11, e439 (2022).arxiv.org/abs/2108.00318
  50. [52]A. J. Coleman, “Structure of fermion density matrices,” Reviews of Modern Physics 35, 668–686 (1963).
  51. [53]D. A. Mazziotti, “Contracted Schrödinger equation: Determining quantum energies and two-particle density matrices without wave functions,” Phys. Rev. A 57, 4219–4234 (1998).DOI
  52. [54]L. M. Mentel, Reduced Density Matrix inspired approaches to electronic structure theory, PhD Thesis, Vrije University, Amsterdam (2015), available at https://research.vu.nl/en/publications/7d553654-df2a-45d7-a460- 018f3cbbe127.
  53. [55]K. Pernal, O. Gritsenko, and E. J. Baerends, “Time-dependent density-matrix-functional theory,” Phys. Rev. A 75, 012506 (2007).
  54. [56]K. J. H. Giesbertz, Time-Dependent One-Body Reduced Density Matrix Functional Theory, PhD Thesis, Vrije University, Amsterdam (2010), available at https://research.vu.nl/en/publications/da4297a6-0542-4c98-bc20-ade87d85e4fd.
  55. [57]K. J. H. Giesbertz, O. V. Gritsenko, and E. J. Baerends, “The adiabatic approximation in time-dependent density matrix functional theory: Response properties from dynamics of phase-including natural orbitals,” The Journal of Chemical Physics 133, 174119 (2010).
  56. [58]A. Akbari, M. J. Hashemi, A. Rubio, R. M. Nieminen, and R. van Leeuwen, “Challenges in truncating the hierarchy of time-dependent reduced density matrices equations,” Physical Review B 85 (2012), 10.1103/PhysRevB.85.235121.DOI
  57. [59]N. Ferré, M. Filatov, and M. Huix-Rotllant, “Reduced Density Matrix Functional Theory (RDMFT) and Linear Response Time-Dependent RDMFT (TD-RDMFT),” in Density-Functional Methods for Excited States, Topics in Current Chemistry, Vol. 368 (Springer International Publishing AG, Switzerland, 2016) pp. 125–183.DOI
  58. [60]J. M. Herbert and J. E. Harriman, “Comparison of two-electron densities reconstructed from one-electron density matrices,” International Journal of Quantum Chemistry 90, 355–369 (2002).DOI
  59. [61]K. J. H. Giesbertz, E. J. Baerends, and O. V. Gritsenko, “Charge transfer, double and bond-breaking excitations with time-dependent density matrix functional theory,” Phys. Rev. Lett. 101, 033004 (2008).
  60. [62]D. B. Jeffcoat and I. DePrince, A. Eugene, “N-representability-driven reconstruction of the two-electron reduced-density matrix for a real-time time-dependent electronic structure method,” The Journal of Chemical Physics 141, 214104 (2014).
  61. [63]F. Lackner, I. Brezinová, T. Sato, K. L. Ishikawa, and J. Burgdörfer, “Propagating two-particle reduced density matrices without wave functions,” Phys. Rev. A 91, 023412 (2015).DOI
  62. [64]F. Lackner, I. Brezinová, T. Sato, K. L. Ishikawa, and J. Burgdörfer, “High-harmonic spectra from time-dependent two-particle reduced-density-matrix theory,” Phys. Rev. A 95, 033414 (2017).DOI
  63. [65]J. Fosso-Tande, T.-S. Nguyen, G. Gidofalvi, and A. E. DePrince, “Large-scale variational two-electron reduced-density-matrix-driven complete active space self-consistent field methods,” Journal of Chemical Theory and Computation 12, 2260–2271 (2016).DOI
  64. [66]D. A. Mazziotti, “Quantum many-body theory from a solution of the N-representability problem,” Phys. Rev. Lett. 130, 153001 (2023).DOI
  65. [67]S. Pan and K. Duraisamy, “On the structure of time-delay embedding in linear models of non-linear dynamical systems,” Chaos: An Interdisciplinary Journal of Nonlinear Science 30 (2020), 10.1063/5.0010886.
  66. [69]B. O. Roos, P. R. Taylor, and P. E. Siegbahn, “A complete active space SCF method (CASSCF) using a density matrix formulated super-CI approach,” Chemical Physics 48, 157–173 (1980).DOI
  67. [73]R. Zwanzig, “On the identity of three generalized master equations,” Physica 30, 1109–1123 (1964).
  68. [74]Y. Ohtsuki and Y. Fujimura, “Bath-induced vibronic coherence transfer effects on femtosecond time-resolved resonant light scattering spectra from molecules,” The Journal of Chemical Physics 91, 3903–3915 (1989).DOI
  69. [75]V. May and O. Kühn, Charge and Energy Transfer Dynamics in Molecular Systems, 4th ed. (Wiley-VCH, Weinheim, Germany, 2023).
  70. [76]F. Zhang, The Schur Complement and Its Applications, First ed. (Springer US, New York, NY, 2005).
  71. [77]https://docs.nersc.gov/systems/perlmutter/architecture/.
  72. [78]https://github.com/hbassi/1rdm_memory_model.
  73. [79]J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, and Q. Zhang, “JAX: composable transformations of Python+NumPy programs,” (2018), http://github.com/google/jax.
  74. [80]R. Okuta, Y. Unno, D. Nishino, S. Hido, and C. Loomis, “CuPy: A NumPy-Compatible Library for NVIDIA GPU Calculations,” in Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing Systems (2017).
  75. [81]Y. Kurzweil and R. Baer, “Time-dependent exchange-correlation current density functionals with memory,” The Journal of Chemical Physics 121, 8731–8741 (2004).
  76. [82]S. Hirata, M. Head-Gordon, and R. J. Bartlett, “Configuration interaction singles, time-dependent Hartree-Fock, and time-dependent density functional theory for the electronic excited states of extended systems,” The Journal of Chemical Physics 111, 10774–10786 (1999).DOI
  77. [83]J. B. Foresman, M. Head-Gordon, J. A. Pople, and M. J. Frisch, “Toward a systematic molecular orbital theory for excited states,” The Journal of Physical Chemistry 96, 135–149 (1992).DOI
  78. [84]C. Chicone, Ordinary Differential Equations with Applications, Texts in Applied Mathematics (Springer, New York, 2024).

Paper details

vlnc:W_4vw94x56
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.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents