Lattice real-time simulations with learned optimal kernels

Daniel Alvestad,Alexander Rothkopf,Dénes Sexty
DOI: https://doi.org/10.1103/physrevd.109.l031502
IF: 5.407
2024-02-06
Physical Review D
Abstract:We present a simulation strategy for the real-time dynamics of quantum fields, inspired by reinforcement learning. It builds on the complex Langevin approach, which it amends with system-specific prior information, a necessary prerequisite to overcome this exceptionally severe sign problem. The optimization process underlying our machine-learning approach is made possible by deploying inherently stable solvers of the complex Langevin stochastic process and a novel optimality criterion derived from insight into so-called boundary terms. This conceptual and technical progress allows us to both significantly extend the range of real-time simulations in 1+1d scalar field theory beyond the state of the art and to avoid discretization artifacts that plagued previous real-time field theory simulations. Limitations of and promising future directions are discussed. https://doi.org/10.1103/PhysRevD.109.L031502 Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article's title, journal citation, and DOI. Funded by SCOAP 3 . Published by the American Physical Society
astronomy & astrophysics,physics, particles & fields
What problem does this paper attempt to address?
This paper attempts to solve the problem of real - time dynamics simulation in quantum field theory, especially the challenges encountered when dealing with the so - called sign problem. Specifically, the paper proposes a machine - learning - based method to optimize the kernel in the Complex Langevin (CL) method in order to overcome the convergence and stability problems that occur during real - time dynamics simulation. ### Core Problems of the Paper 1. **Requirements for Real - Time Dynamics Simulation**: - Many unsolved key problems in modern physics require the study of the dynamic properties of strongly - correlated quantum many - body systems in Minkowski time. This includes the transport properties of high - temperature and high - density nuclear matter, high - energy parton scattering, the real - time transport of non - relativistic fermions and their interactions with gauge fields, etc. 2. **Limitations of Existing Methods**: - Current analytical and numerical Monte Carlo methods are difficult to directly handle these problems, especially when dealing with the sign problem. The sign problem causes the effective signal of the traditional Markov chain sampling method to be flooded by a large number of almost mutually canceling phases, making these methods ineffective in practical applications. 3. **Improvement of the Complex Langevin Method**: - The Complex Langevin method is a strategy to try to solve the sign problem, but it faces problems such as converging to wrong solutions and numerical instability in practical applications. By introducing system - specific prior information and using machine - learning techniques to optimize the kernel, the paper aims to improve the stability and accuracy of the Complex Langevin method. ### Main Contributions - **Optimizing the Kernel**: Through machine - learning techniques, the paper proposes a system - specific method for optimizing the kernel, which can significantly improve the performance of the Complex Langevin method, especially in the real - time dynamics simulation of 1 + 1 - dimensional scalar field theory. - **Avoiding Discretization Errors**: The optimized kernel not only expands the scope of simulation but also avoids the discretization errors that existed in previous methods. - **Verifying the Effectiveness of the Method**: The paper verifies the correctness and effectiveness of the proposed method by comparing the optimized results with the known Hybrid Monte Carlo (HMC) simulation results. ### Conclusion The paper shows how to effectively solve the sign problem in real - time dynamics simulation in quantum field theory by combining machine - learning techniques and the Complex Langevin method, providing new ideas and a technical basis for future practical applications in higher dimensions (such as 3 + 1 - dimensional).