Predict open quantum dynamics with data-informed quantum-classical dynamics
Pinchen Xie, Ke Wang, Anupam Mitra, Yuanran Zhu, Xiantao Li, Wibe Albert de Jong, Chao Yang · 2025
Read Predict open quantum dynamics with data-informed quantum-classical dynamics by Pinchen Xie, Ke Wang, Anupam Mitra, Yuanran Zhu, Xiantao Li, Wibe Albert de Jong, Chao Yang (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-2508-17170.txt.
Introduction
Xie and co-authors write a research paper on predicting how open quantum systems change over time. Such systems leak energy and information into their surroundings, so their motion is harder to compute than that of a closed system. The authors propose a method they call data-informed quantum-classical dynamics.
The method blends measured or simulated data with a standard equation for open systems, the Lindblad equation, so the model learns corrections instead of relying on fixed assumptions. Tests show the approach tracks the true dynamics more closely over longer times. Better tools of this kind help physicists design quantum devices whose behavior stays predictable outside ideal laboratory conditions.
Catalog record
| CREATOR | Pinchen Xie, Ke Wang, Anupam Mitra, Yuanran Zhu, Xiantao Li, Wibe Albert de Jong, Chao Yang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-17170
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Predict open quantum dynamics with data-informed quantum-classical dynamics Preprint: APS/123-QED Pinchen Xie, Ke Wang, Anupam Mitra, Yuanran Zhu, Xiantao Li, Wibe Albert de Jong, Chao Yang Abstract We introduce a data-informed quantum-classical dynamics (DIQCD) approach for predicting the evolution of an open quantum system. The equation of motion in DIQCD is a Lindblad equation with a flexible, time-dependent Hamiltonian that can be optimized to fit sparse and noisy data from local observations of an extensive open quantum system. We demonstrate the accuracy and efficiency of DIQCD for both experimental and simulated quantum devices. We show that DIQCD can predict entanglement dynamics of ultracold molecules (Calcium Fluoride) in optical tweezer arrays. DIQCD also successfully predicts carrier mobility in organic semiconductors (Rubrene) with accuracy comparable to nearly exact numerical methods. Despite the inevitable coupling between quantum devices and noisy environments containing vast degrees of freedom, device performance can be characterized based on the statistical and dynamical behaviors of a few observables. Dynamical modeling can focus on a subsystem of the device-environment duo, treating the rest of the system as an effective bath that perturbs the subsystem in a way that yields similar statistics and dynamics of observables relevant to device metrics. Such dimension-reduction ideas evolve into a sequence of numerical approaches for modeling equations of motion (EOM) governing open quantum system dynamics, with a tradeoff between accuracy and efficiency Lindblad 1976 ; Gorini et al. 1976 ; Garg et al. 1985 ; Tanimura and Kubo 1989 ; Garraway 1997 ; Diósi and Strunz 1997 ; Tanimura 2006 ; Xu and Yan 2007 ; Suess et al. 2014 ; Tamascelli et al. 2018 ; Wang and Li 2024 . For example, on the high-accuracy end, pseudomode methods Garraway 1997 ; Tamascelli et al. 2018 ; Li 2021 ; Park et al. 2024 ; Park et al. 2024 ; Huang et al. 2025 enlarge the system Hilbert space to incorporate a finite set of effective modes that encode the correlation function of a Gaussian bath. On the high-efficiency end, the plain Lindblad equation sketches a dissipative environment without introducing auxiliary degrees of freedom, similar to how the Langevin equation sketches a classical environment. In between, mixed quantum-classical dynamics (MQCD) may strike the balance of accuracy and efficiency. With additional classical degrees of freedom, MQCD can describe coherent fluctuations in quantum evolution, which is otherwise reduced to mean-field dissipators in the plain Lindblad equation. Well-known phenomenological MQCD approaches such as Ehrenfest dynamics Ehrenfest 1927 and surface hopping Tully 1990 have been intensively used for modeling non-adiabatic molecular dynamics. There are concerns about unsystematic control of error regarding the exact quantum dynamics Subotnik et al. 2016 . Uncontrolled error hinders accurate predictions of material properties or device metrics. Moreover, for systems where the environment does not have a particle-based semiclassical representation, phenomenological MQCD becomes irrelevant. For example, one needs to devise special-purpose, ab initio quantum-classical models for cold-atom systems Shaw 2024 . In this work, we place phenomenological MQCD within a general context by introducing variational capacity on the classical dynamics and the quantum-classical coupling. The variational capacity can be understood as a mechanism to renormalize MQCD, thereby circumventing the limited applicability of non-variational MQCD approaches. We develop the algorithm and the software for optimizing a general MQCD EOM, parameterized as a time-dependent Lindblad equation, on time series data of measurements on observables relevant to device metrics. We call our method Data-Informed Quantum-Classical Dynamics (DIQCD). The approach emphasizes practicality for dealing with experimental or simulation time-series data. DIQCD requires only the data that represents measurements performed on the quantum subsystem of interest, without requiring the time-correlation function of the quantum environment. DIQCD works well with sparse and noisy data due to its restricted variational flexibility, which reduces generality but avoids overfitting. These features allow the application of DIQCD to various real-world challenges. We will demonstrate the applications of DIQCD for both artificial quantum systems and condensed matter systems, with data from either experiments or first-principles-based simulations. The first case study is on controlled dynamics of Calcium-fluoride (CaF) molecular qubits in optical traps Holland et al. 2023 . We train DIQCD-EOM on sparse experimental data collected from a single CaF molecule. We show that DIQCD can capture quantum decoherence across multiple time scales and predict two-qubit dynamics of Bell state generation. The second case study is on quantum transport in Rubrene crystal ( \mathrm{C_{18}H_{8}(C_{6}H_{5})_{4}} ). We train DIQCD-EOM on simulation data collected for a single Rubrene molecule. With the cost of simulating a tight-binding model, DIQCD predicts carrier mobility with similar accuracy as nearly exact time-dependent density matrix renormalization group (TD-DMRG) simulation Li et al. 2020 of the whole system (carrier and phonons). The study brings a new perspective for overcoming the accuracy-efficiency tradeoff in modeling band-like transport. Methods – In the following, we call the quantum subsystem of interest the “system”, and the rest of the degrees of freedom the “environment”. The DIQCD approach is made up of a flexible EOM for describing system evolution, a loss function for optimizing the EOM on data, and the corresponding training and simulation algorithms. The EOM in DIQCD is a Lindblad equation with a time-dependent Hamiltonian \hat{H}_{\bm{\epsilon}}(t) : \small\frac{d\hat{\rho}_{\bm{\epsilon}}(t)}{\partial t}=-i[\hat{H}_{\bm{\epsilon}}(t),\hat{\rho}_{\bm{\epsilon}}(t)]+\sum_{k}\gamma_{k}(\hat{L}_{k}\hat{\rho}_{\bm{\epsilon}}(t)\hat{L}_{k}^{\dagger}-\frac{1}{2}\{\hat{L}_{k}\hat{L}_{k}^{\dagger},\hat{\rho}_{\bm{\epsilon}}(t)\}). (1) \hat{\rho}_{\bm{\epsilon}}(t) is the density matrix of the system. \{\hat{L}_{k}\} is a finite set of static jump operators. \hat{H}_{\bm{\epsilon}}(t)=\hat{H}_{0}+\hat{H}_{c}(t)+\sum_{j=1}^{M}f_{j}(\bm{\epsilon}(t))\hat{S}_{j} combines the static system Hamiltonian H_{0} , the external control \hat{H}_{c}(t) , and perturbing Hermitian operators \hat{S}_{j} . f_{j}(\bm{\epsilon}(t)) is a scalar function of the multidimensional, classical dynamical processes \bm{\epsilon}(t) that encodes information about the environment. As a special case, Eq. ( 1 ) becomes Ehrenfest dynamics when \bm{\epsilon}(t) represents atomic nuclei evolving under the average potential energy surface \mathrm{Tr}(\hat{H}_{\bm{\epsilon}}(t)\rho_{\bm{\epsilon}}) . In DIQCD, we allow \bm{\epsilon}(t) to be generic dynamical processes that can be integrated by an explicit Markovian integrator with or without auxiliary degrees of freedom \bm{\xi}(t) . This means \bm{\epsilon}(t+\delta t) and \bm{\xi}(t+\delta t) are determined by \bm{\epsilon}(t) and \bm{\xi}(t) in an integration scheme with \delta t as the time step. These requirements are imposed such that \bm{\epsilon}(t) can evolve concurrently with the structure-preseving integration Cao and Lu 2025 of the Lindblad equation. And eventually, the parameters controlling the EOM of \bm{\epsilon}(t) can be optimized together with other system parameters (such as \gamma_{k} ) through chain rules, such that the behavior of \hat{\rho}_{\bm{\epsilon}}(t) can match time-series data of the system. A Langevin process \epsilon_{0}(t) is an example of a flexible classical process. The EOM of \epsilon_{0}(t) is parameterized by \tau and A with the scheme Zhang et al. 2019 \epsilon_{0}(t+\delta t)=e^{-\delta t/\tau}\epsilon_{0}(t)+A\sqrt{1-e^{-2dt/\tau}}\xi_{0}(t) . \xi_{0}(t) is randomly sampled from a standard normal distribution. Simpler examples of flexible processes include periodic signals and Gaussian white noise. Complex examples include isothermal molecular dynamics on a parameterized potential energy surface. In DIQCD, the expectation of any system observable \hat{O} at time t is given as O(t)=\langle O_{\bm{\epsilon}}(t)\rangle_{\bm{\epsilon}}=\langle\mathrm{Tr}(\hat{O}\hat{\rho}_{\bm{\epsilon}}(t))\rangle_{\bm{\epsilon}} , where \langle\cdot\rangle_{\bm{\epsilon}} represents the average over realizations of \bm{\epsilon}(t) with given probability distribution of \bm{\epsilon}(0) and possible auxiliary variables \bm{\xi}(0) . DIQCD is trained on time-series data at discrete time points \{t_{1},t_{2},\cdots\} with the mean-squared loss \mathcal{L}=\sum_{ij}(O_{i}(t_{j})-O_{i}^{*}(t_{j}))^{2} . The data O_{i}^{*}(t) comes from measurements performed on either an experimental or simulated system. For example, for an atom-based quantum computing platform, \hat{O_{i}} can be a