Gauge-equivariant neural networks as preconditioners in lattice QCD
This work addresses computational bottlenecks in lattice QCD simulations for physicists, presenting an incremental improvement with practical implementation benefits like communication avoidance.
The authors tackled the challenge of efficiently learning multi-grid preconditioners for lattice QCD by using gauge-equivariant neural networks, achieving minimal re-training on different gauge configurations and maintaining efficiency under modest parameter changes.
We demonstrate that a state-of-the art multi-grid preconditioner can be learned efficiently by gauge-equivariant neural networks. We show that the models require minimal re-training 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.