Continuous normalizing flows for lattice gauge theories

Mathis Gerdes,Pim de Haan,Roberto Bondesan,Miranda C. N. Cheng

2024-10-17

Abstract:Continuous normalizing flows are known to be highly expressive and flexible, which allows for easier incorporation of large symmetries and makes them a powerful tool for sampling in lattice field theories. Building on previous work, we present a general continuous normalizing flow architecture for matrix Lie groups that is equivariant under group transformations. We apply this to lattice gauge theories in two dimensions as a proof-of-principle and demonstrate competitive performance, showing its potential as a tool for future lattice sampling tasks.

High Energy Physics - Lattice,Statistical Mechanics,Machine Learning,High Energy Physics - Theory

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the sampling efficiency and reduce the autocorrelation time in lattice gauge theory. Specifically, the paper proposes a new continuous normalizing flow framework for generating high - quality samples in two - dimensional pure gauge theory. Traditional computational methods such as Markov Chain Monte Carlo (MCMC) will encounter serious slow - convergence problems near phase transitions and in the continuous limit, leading to significant computational challenges. As a machine - learning model, the normalizing flow can sample more effectively by learning a bijective transformation from an easily - sampled distribution to the target distribution.

The main contributions of the paper are:
1. **Proposing a new continuous normalizing flow architecture**: This architecture is defined on matrix Lie groups and is equivariant to group transformations. This enables the model to generate samples efficiently while maintaining gauge symmetry.
2. **Application to two - dimensional lattice gauge theory**: As a proof of principle, the authors apply this method to two - dimensional SU(2) and SU(3) lattice gauge theory and demonstrate its competitiveness in terms of the effective sample size (ESS).
3. **Efficient gradient calculation method**: The authors implement the adjoint sensitivity method, which is an efficient gradient - calculation technique applicable to any neural network architecture. In addition, they also use the Crouch - Grossman Runge - Kutta integration method to ensure numerical stability.

Through these improvements, the paper demonstrates the potential of continuous normalizing flow in lattice gauge theory, especially its efficiency and accuracy in handling large - scale lattice simulations.