Learning thermodynamic master equations for open quantum systems

Peter Sentz, Stanley Nicholson, Yujin Cho, Sohail Reddy, Brendan Keith, Stefanie Günther · 2025

Read Learning thermodynamic master equations for open quantum systems by Peter Sentz, Stanley Nicholson, Yujin Cho, Sohail Reddy, Brendan Keith, Stefanie Günther (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-2506-01882.txt.

Introduction

Peter Sentz and co-authors learn the governing equations of open quantum systems from data while respecting heat laws, in work from twenty twenty-five. Open systems leak energy to their surroundings, and finding their Hamiltonians matters for quantum computing. Most machine learning fits are nonlinear, while the true models are linear, and the fit often ignores physics.

Their data driven model holds learnable terms that keep heat behavior consistent. The trained model stays readable, since it directly estimates the Hamiltonian and the linear coupling to the environment. Tests on synthetic two and three level systems worked well. So did tests on real data from a quantum device at a national laboratory.

Catalog record

CREATORPeter Sentz, Stanley Nicholson, Yujin Cho, Sohail Reddy, Brendan Keith, Stefanie Günther
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-01882

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Learning thermodynamic master equations for open quantum systems Peter Sentz, Stanley Nicholson, Yujin Cho, Sohail Reddy, Brendan Keith, Stefanie Günther Abstract The characterization of Hamiltonians and other components of open quantum dynamical systems plays a crucial role in quantum computing and other applications. Scientific machine learning techniques have been applied to this problem in a variety of ways, including by modeling with deep neural networks. However, the majority of mathematical models describing open quantum systems are linear, and the natural nonlinearities in learnable models have not been incorporated using physical principles. We present a data-driven model for open quantum systems that includes learnable, thermodynamically consistent terms. The trained model is interpretable, as it directly estimates the system Hamiltonian and linear components of coupling to the environment. We validate the model on synthetic two and three-level data, as well as experimental two-level data collected from a quantum device at Lawrence Livermore National Laboratory. 1 Introduction Characterizing quantum dynamical processes is central to the advancement of quantum computing devices, with significant impacts on quantum control, compilation, and error mitigation [ 1 ] . Accurate dynamical models are essential to address two challenges: 1) optimized quantum compilation, enabling optimal quantum gate sequences and resource allocation, and 2) quantum error mitigation and correction to achieve overall improved system performance [ 2 ] . A precise model of the underlying quantum dynamics requires accurate knowledge of system parameters such as qubit frequencies, coupling strengths, and noise channels. Extracting such information from measurement data is inherently challenging, yet recent advances in data-driven methods incorporating machine learning (ML) have led to significant progress in this area. A comprehensive overview of ML for quantum systems can be found in [ 3 ] , featuring recent developments for the tasks of learning quantum states, quantum dynamics, and quantum measurements. Hamiltonian dynamics can be reconstructed using observed time series of a complete set of measurements [ 4 ] , and quantum process tomography allows for the identification of linear non-unitary dynamics governed by the Lindblad master equations [ 5 ] . However, the number of measurements required to fully characterize the dynamics of a quantum system grows exponentially with the number of levels. Therefore, a variety of approaches have been developed that pursue estimates of the unknown parameters of the Hamiltonian and Lindblad operators using limited measurements [ 6 , 7 , 8 , 9 ] . Nevertheless, a growing array of approaches incorporate deep neural networks in the characterization process, introducing nonlinear transformations of the quantum state into the dynamical model [ 10 , 11 , 12 , 13 , 14 ] . For instance, recurrent neural networks (RNNs) have been employed to predict measurement outcomes under dissipative dynamics [ 15 ] , and to reconstruct quantum dynamics using trajectory data from superconducting qubits under microwave control sequences [ 16 ] . These methods, which leverage machine learning to approximate quantum states and dynamical processes, often fall under the category of “black-box” approaches, where the learned model lacks explicit physical interpretability, hence requiring post-processing of the learned agent. On the other hand, “gray-box” approaches combine data-driven learning models with physically understandable dynamical structures, incorporating known physics directly into the machine learning agent to improve accuracy and interpretability. For example, Youssry et al. [ 17 ] combine the output of a black-box machine learning agent with numerically evolved quantum dynamical processes to address uncertainties in the Hamiltonian model and capture noise characteristics. More intrusively, machine learning models based on physics-informed neural ordinary differential equations [ 18 ] , also termed universal differential equations (UDEs) [ 19 ] , use trajectory data to train a correction to an underlying approximate physical model described by a differential equation. Such learning approaches often require smaller training data sets to achieve accurate model fidelity and are more resilient to noisy training data [ 20 , 21 ] . The UDE approach was recently applied to the problem of Hamiltonian learning in [ 22 ] , where a nonlinear correction to the Hamiltonian dynamics is learned from trajectory data. An application of UDEs to learning open quantum systems is proposed in [ 23 ] , where a parameterized augmentation of the Lindblad equation was trained to identify latent dynamics in a single qubit superconducting quantum processing unit using microwave-driven control pulses. There, a structure-preserving ansatz to learn corrections to approximate Hamiltonian and Lindblad decoherence operators is compared to a nonlinear augmentation using a nonlinear feed-forward neural network. While the former guarantees completely-positive trace-preserving (CPTP) maps [ 5 ] , the latter has demonstrated higher model accuracy after training, indicating the need for nonlinear augmentations to the linear Lindblad equation. However, ensuring that a learned nonlinear dynamical model yields physically consistent dynamics when evaluated outside of the training data remains a crucial challenge. Approaches that incorporate physical laws as inductive biases [ 24 , 25 , 26 , 27 ] , have demonstrated improvements in both accuracy and generalization. Thus, incorporating nonlinearity into quantum master equations in a physically consistent manner has the potential for more robust learned models. In particular, a recent family of ML techniques that embed the laws of thermodynamics into the learnable model has been successfully applied to dissipative dynamical systems [ 28 ] . Among these methods, the so-called metriplectic [ 29 ] neural networks are based on a unifying formalism for reversible and irreversible dynamics known as the generic equation for non-equilibrium reversible-irreversible coupling (GENERIC) [ 30 , 31 , 32 ] . Thermodynamics-informed neural networks (TINNs) parametrize the GENERIC structure of dynamics as neural networks, and weakly enforce the necessary symmetry and degeneracy conditions through penalization terms in the loss function [ 33 , 34 , 35 ] . In contrast, GENERIC formalism-informed neural networks (GFINNs) employ architectures that directly enforce the properties of GENERIC for every configuration of the neural network parameters [ 36 ] . Other works strongly enforce the necessary conditions by learning structure-preserving brackets parametrized by learnable tensors [ 37 , 38 ] . The recent port-metriplectic approach [ 39 ] decomposes dissipative systems into local subsystems, reducing the computational complexity of training the model. In this work, we extend the use of ML enforcing thermodynamic consistency to the context of open quantum dynamical systems. 1.1 Contributions and highlights We present a thermodynamically consistent, data-driven model for learning dynamics of open quantum systems. In particular, the model incorporates nonlinear terms based on physical principles via the GENERIC formalism, in contrast to methods that augment linear master equations with nonlinear transformations in an ad hoc manner. This yields dynamical equations that guarantee valid quantum states for two-level systems. The learned model is interpretable, directly providing an estimate of the Hamiltonian and the operators that govern the coupling to an external environment. We describe this model within the density operator formalism and detail its construction using the density operator’s representation in the Pauli or Gell–Mann basis. Next, we test the accuracy and generalizability of the model on two (qubit) and three (qutrit) level synthetic quantum systems. Finally, we validate its efficiency on noisy, two-level experimental data taken from Lawrence Livermore National Laboratory’s (LLNL) QuDIT device. 1.2 Outline We continue by discussing the shortcomings of the linear Lindblad

Read the complete text