Applications of flow models to the generation of correlated lattice QCD ensembles

Ryan Abbott,Aleksandar Botev,Denis Boyda,Daniel C. Hackett,Gurtej Kanwar,Sébastien Racanière,Danilo J. Rezende,Fernando Romero-López,Phiala E. Shanahan,Julian M. Urban

2024-05-28

Abstract:Machine-learned normalizing flows can be used in the context of lattice quantum field theory to generate statistically correlated ensembles of lattice gauge fields at different action parameters. This work demonstrates how these correlations can be exploited for variance reduction in the computation of observables. Three different proof-of-concept applications are demonstrated using a novel residual flow architecture: continuum limits of gauge theories, the mass dependence of QCD observables, and hadronic matrix elements based on the Feynman-Hellmann approach. In all three cases, it is shown that statistical uncertainties are significantly reduced when machine-learned flows are incorporated as compared with the same calculations performed with uncorrelated ensembles or direct reweighting.

High Energy Physics - Lattice,Machine Learning

What problem does this paper attempt to address?

This paper discusses how to use machine learning flow models in lattice gauge theory of quantum chromodynamics (QCD) to generate correlated lattice gauge field configurations and reduce statistical errors in computing observables. The authors propose to use residual flow architectures for the flow models, which can map and generate correlated lattice field configurations under different action parameters. This can effectively reduce statistical uncertainties when computing physical quantities such as continuum limits, mass dependence of QCD observables, and meson matrix elements based on Feynman-Hellmann method.

Specifically, the paper mentions three concepts:
1. **Flow models**: These are mappings that use trainable parameters to transform samples from a base distribution to model distributions with different densities. They can be used to accelerate the sampling of lattice gauge field configurations.
2. **Correlated lattice gauge field configurations**: By mapping field configurations using flow models between different action parameters, correlated configurations can be generated to reduce uncertainties in computing expectation values.
3. **Residual flow architecture**: A specific flow model structure composed of a series of trainable layers that directly act on the lattice field. These layers are reversible and can effectively construct complex transformations.

The paper demonstrates the advantages of this approach through three applications:
1. **Continuum limits**: By using flow models, the continuous limit of the gradient flow scale can be computed to reduce statistical errors.
2. **Feynman-Hellmann technique**: The flow models are used to compute the gluonic momentum fraction of mesons, showing a significant reduction in statistical uncertainties when computing matrix elements.
3. **Mass dependence of QCD observables**: The dependence of observables on the mass in Nf=2 QCD is studied, indicating that flow models can effectively reduce the number of configurations required to achieve the same statistical accuracy.

In summary, the paper aims to improve the efficiency of lattice QCD calculations and reduce statistical noise by utilizing machine learning flow models, leading to a better understanding of phenomena in the strong interaction regime.