Complex Langevin simulations with a kernel

We discuss recent developments regarding the use of kernels in complex Langevin simulations. In particular, we outline how a kernel can be used to solve the problem of wrong convergence in a simple toy model. Since conventional correctness criteria for complex Langevin results are only necessary but not sufficient, the correct convergence of complex Langevin simulations is not always straightforward to assess. Hence, we furthermore discuss a condition for correctness that we have recently derived, which is both necessary and sufficient. Finally, we outline a machine-learning approach for finding suitable kernels in lattice gauge theories and present preliminary results of its application to the heavy-dense limit of QCD.

💡 Research Summary

The fundamental challenge in studying Quantum Chromodynamics (QCD) under extreme conditions, such as high baryon density, is the “sign problem,” which renders traditional Monte Carlo importance sampling methods inapplicable due to the complex nature of the action. The Complex Langevin (CL) method has emerged as a promising alternative to bypass this obstacle by complexifying the field variables. However, the CL method is notoriously susceptible to “wrong convergence,” where the stochastic process converges to an incorrect expectation value, often due to the distribution of complexified variables leaking into regions that do not represent the original physics.

This paper presents a multi-faceted advancement in stabilizing and validating Complex Langevin simulations. The primary technical contribution is the implementation of a “kernel” within the Langevin equation. By introducing a kernel function, the researchers modify the drift term of the stochastic process, effectively guiding the trajectories to remain within a controllable region of the complex plane. This approach is demonstrated to successfully mitigate the wrong convergence issue in a simplified toy model, providing a structural remedy to the instability of the complexified dynamics.

A significant theoretical breakthrough presented in this work is the derivation of a new correctness criterion. Historically, the criteria used to verify the validity of CL simulations were merely “necessary” conditions; passing these tests did not guarantee that the simulation had reached the true physical result. The authors introduce a newly derived condition that is both “necessary and sufficient,” providing a mathematically rigorous framework to certify the accuracy of the simulation results. This advancement is crucial for establishing the reliability of CL-based computations in high-energy physics.

Furthermore, the paper addresses the practical difficulty of designing effective kernels for high-dimensional lattice gauge theories. Since manual kernel design is computationally prohibitive in complex systems, the authors propose a machine-learning-based approach. This method utilizes ML algorithms to autonomously discover optimal kernel functions that enhance convergence stability. The paper presents preliminary applications of this ML-driven kernel optimization to the heavy-dense limit of QCD, demonstrating the potential of integrating artificial intelligence with computational physics to tackle the most daunting problems in the Standard Model. In summary, the research provides a robust roadmap for more reliable and automated complex Langevin simulations in the presence of the sign problem.