HMC with Normalizing Flows

TL;DR

This paper introduces a novel Hamiltonian Monte Carlo method that incorporates trainable Normalizing Flows to improve sampling efficiency and scalability in large lattice systems.

Contribution

It presents a new approach integrating Normalizing Flows into HMC, enabling better dynamics simplification and improved performance over traditional methods.

Findings

Outperforms traditional HMC in generating independent configurations

Easily scalable to large lattice volumes with minimal retraining

Open-source implementation available online

Abstract

We propose using Normalizing Flows as a trainable kernel within the molecular dynamics update of Hamiltonian Monte Carlo (HMC). By learning (invertible) transformations that simplify our dynamics, we can outperform traditional methods at generating independent configurations. We show that, using a carefully constructed network architecture, our approach can be easily scaled to large lattice volumes with minimal retraining effort. The source code for our implementation is publicly available online at https://github.com/nftqcd/fthmc.