Neural Network Solution of Non-Markovian Quantum State Diffusion and Operator Construction of Quantum Stochastic Process
Jiaji Zhang, Carlos L. Benavides-Riveros, Lipeng Chen · 2025
Read Neural Network Solution of Non-Markovian Quantum State Diffusion and Operator Construction of Quantum Stochastic Process by Jiaji Zhang, Carlos L. Benavides-Riveros, Lipeng Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-01049.txt.
Introduction
Jiaji Zhang, with two co-authors, introduces a neural approach to non-Markovian quantum state diffusion. That framework models open quantum systems with wavefunctions. The environment retains memory in such systems. Standard machine learning merely approximates wavefunctions or mean values. Their operator-based approach aims for broader use and clearer meaning.
An operator construction algorithm uses a neural network as a generator. It rebuilds the stochastic time evolution operator from many trajectories. The method is benchmarked on the spin-boson model. Diverse spectral densities are covered in those tests. Accuracy is shown across those cases in the paper. The operator also proves useful for absorption spectra. It can rebuild reduced density matrices at long timescales as well.
Catalog record
| CREATOR | Jiaji Zhang, Carlos L. Benavides-Riveros, Lipeng Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-01049
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Neural Network Solution of Non-Markovian Quantum State Diffusion and Operator Construction of Quantum Stochastic Process Jiaji Zhang, Carlos L. Benavides-Riveros, Lipeng Chen Abstract Non-Markovian quantum state diffusion provides a wavefunction-based framework for modeling open quantum systems. In this work, we introduce a novel machine learning approach based on an operator construction algorithm. This algorithm employs a neural network as a universal generator to reconstruct the stochastic time evolution operator from an ensemble of quantum trajectories. Unlike conventional machine learning methods that merely approximate time-dependent wavefunctions or expectation values, our operator-based approach yields broader applications and enhanced interpretability of the stochastic process. We benchmark the algorithm on the spin-boson model across diverse spectral densities, demonstrating its accuracy. Furthermore, we showcase the operator’s utility in calculating absorption spectra and reconstructing reduced density matrices at extended timescales. These results establish a new paradigm for the application of machine learning in quantum dynamics. I Introduction Multi-dimensional spectroscopy has emerged as a powerful technique for elucidating the structural and dynamical details of chemical reactions. Mukamel 2000 ; Pios et al. 2024a ; Ruetzel et al. 2014 ; Kowalewski et al. 2017 ; Loe et al. 2024 ; Panman et al. 2017 ; Bressan et al. 2024 ; Bangert, Stienkemeier, and Bruder 2022 Theoretical descriptions of these spectra require simulating the system’s response function under an external laser field. Mukamel 1995 ; Cho 2019 ; Oliver 2018 ; Gelin, Chen, and Domcke 2022 ; Dorfman, Schlawin, and Mukamel 2016 While spectroscopic simulations in gas-phase systems are often performed by directly solving the time-dependent Schrödinger equation, Pios et al. 2024b ; Gelin et al. 2021 ; Sun et al. 2020 ; Cho and Fleming 2020 the increasing number of degrees of freedom (DOFs) in condensed-phase systems makes a full quantum mechanical treatment prohibitively challenging. In such applications, the theory of open quantum systems provides an efficient strategy. Weiss 2012 ; Vacchini 2024 This framework treats DOFs of lesser theoretical interest as a heat bath, which can be analytically traced out using path integral influence functional theory. The dynamics of the reduced system are then governed by quantum master equation (QME). Well-known examples employing the Markovian approximation for the bath include the Lindblad master equation and the multi-level Redfield equation. Becker, Schnell, and Thingna 2022 ; Kuhn and Tanimura 2003 ; Wang et al. 2023 Non-Markovian effects, when necessary, can be incorporated via numerically rigorous approaches like the hierarchical equations of motion (HEOM). Tanimura 2020 ; Yan 2014 These advanced methods have become essential tools for characterizing quantum dynamics in spectroscopic simulations. Zhang, Borrelli, and Tanimura 2021 ; Takahashi and Tanimura 2023 ; Ikeda, Dijkstra, and Tanimura 2019 ; Duan et al. 2022 ; Hein et al. 2012 ; Polley and Loring 2020 Despite their success, a central difficulty with various QMEs is their density-matrix-based formulation, which scales quadratically with system size, making it computationally prohibitive for large systems. To alleviate this limitation, an alternative wavefunction-based approach represents dynamics of open quantum systems by modeling bath effects as quantum stochastic processes. Weiss 2012 ; Milz and Modi 2021 In the Markovian case, this leads to the quantum state diffusion (QSD) equation, which can be derived from the stochastic unraveling of the corresponding master equation. Gisin and Percival 1992 ; Christie et al. 2022 ; Pažėra et al. 2024 This approach has been generalized to the non-Markovian case, the non-Markovian QSD (NMQSD) equation, by incorporating an additional memory integral characterizing the bath effects. Diósi, Gisin, and Strunz 1998 ; Strunz, Diósi, and Gisin 1999 ; Link, Luoma, and Strunz 2023 However, this integral involves the functional derivative \delta|\psi(t,{\bm{z}})\rangle/\delta{\bm{z}} , where |\psi(t,{\bm{z}})\rangle and \bm{z} denote the time-dependent wavefunction and the stochastic noise, respectively. Direct evaluation of this term is only feasible in specific cases, such as harmonic oscillators or via perturbative expansions. Zhou, Gao, and Shuai 2024 ; Yu et al. 1999 ; Zhong and Zhao 2013 Recently, the hierarchy of pure states (HOPS) method provided an indirect solution by encoding the memory integral into a set of auxiliary equations solved similarly to the HEOM. Suess, Eisfeld, and Strunz 2014 ; Yu et al. 2021 ; Hartmann and Strunz 2021 ; Gao et al. 2022 ; Citty et al. 2024 Consequently, while HOPS avoids direct computation of the functional derivative, it inherits HEOM’s computational scaling challenges, due to the growth in hierarchical elements. In recent years, advances in artificial intelligence (AI) have opened new avenues for applying AI methods, particularly deep neural networks, to scientific research. Levine and Tu 2024 ; Krenn et al. 2022 ; Dral 2024 ; Keith et al. 2021 ; Baum et al. 2021 Within the field of open quantum systems, neural network applications primarily fall into two categories. The first category focuses on function learning, where neural networks approximate time-dependent functions such as density matrix elements and expectation values. Herrera Rodríguez and Kananenka 2021 ; Ullah and Dral 2022a ; Mazza et al. 2021 ; Vicentini et al. 2019 ; Luchnikov et al. 2020 By incorporating quantum mechanics rules, these networks can be trained using short-time input data and subsequently predict long-time dynamics beyond the training time window. To reduce data preparation costs, completely physics-driven and data-free models have been developed by leveraging the time-dependent variational principle. Norambuena et al. 2024 ; Herrera Rodriguez and Kananenka 2024 ; Ullah and Dral 2022b While these approaches have been extensively applied to population and coherence dynamics under time-independent Hamiltonians, their exploration for time-dependent systems with external fields remains limited. The second category involves functional learning, where neural networks approximate functionals describing linear or nonlinear mappings between functions under different conditions. Chen and Chen 1995 ; Kovachki, Lanthaler, and Mishra 2021 A prominent example is the surrogate solver for dynamical equation. Li et al. 2021 ; Brunton and Kutz 2024 ; Lu et al. 2021 Such solvers directly generate solutions to target equations (e.g., Schrödinger or master equations) from input conditions such as the initial state and external field parameters. Functioning as universal solvers, they are applicable to diverse computational scenarios, including multi-dimensional spectroscopy. Zhang, Benavides-Riveros, and Chen 2024 ; Zhang, Benavides-Riveros, and Chen 2025 In this work, we aim to develop a neural network architecture capable of solving NMQSD while directly calculating stochastic functional derivatives. To achieve this, we adopt the surrogate model previously developed for solving driven-dissipative quantum dynamics. Zhang, Benavides-Riveros, and Chen 2025 Before proceeding, however, we first summarize key limitations of our previously developed architecture. In earlier work, the mapping from input to output was solely determined by the neural network’s intrinsic structure. This approach is inherently black-box, lacking interpretability, and consequently makes it difficult to enhance the surrogate solver’s performance by integrating advantages from other algorithms. Improvements were largely restricted to pre-processing input conditions or post-processing output solutions. Furthermore, the network employed a strictly sequential layer structure with a single pathway from input to output. This design proves inefficient for approximating functionals involving multilevel correlations, where highly interwined local (short-range) and global (long-range) correlations coexist. These shortcomings have hampered the model’s further development and practical utility. To address these limitations, we propose a novel surrogate solver algorithm grounded in operator construction. Operators constitute the fundamental building blocks of quantum mechanics; for the NMQSD