Gauge-equivariant neural networks as preconditioners in lattice QCD

INSPIRE: 2631052
arXiv: 2302.05419
DOI: 10.1103/PhysRevD.108.034503

Authors: Lehner, Christoph, Wettig, Tilo

Submitted: 10 February 2023

Subjects:

• hep-lat
• cs.LG
• cs.NA
• math.NA
• Lattice
• Computing
• Math and Math Physics

Journal reference: Phys.Rev.D 108 (3) 034503 (2023)

Citations: 25

Abstract

We demonstrate that a state-of-the-art multigrid preconditioner can be learned efficiently by gauge-equivariant neural networks. We show that the models require minimal retraining on different gauge configurations of the same gauge ensemble and to a large extent remain efficient under modest modifications of ensemble parameters. We also demonstrate that important paradigms such as communication avoidance are straightforward to implement in this framework.