Performance of several Lanczos eigensolvers with HISQ fermions

TL;DR

This paper compares the performance of various Lanczos eigensolvers from Grid and QUDA libraries on the HISQ Dirac operator, highlighting optimization techniques like Chebyshev acceleration.

Contribution

It provides a comprehensive performance analysis of state-of-the-art Lanczos eigensolvers for HISQ fermions, including optimization strategies.

Findings

Implicitly Restarted Lanczos performs efficiently

Thick-Restart Lanczos offers competitive convergence

Chebyshev acceleration improves solver performance

Abstract

We investigate the state-of-the-art Lanczos eigensolvers available in the Grid and QUDA libraries. They include Implicitly Restarted Lanczos, Thick-Restart Lanczos, and Block Lanczos. We measure and analyze their performance for the Highly Improved Staggered Quark (HISQ) Dirac operator. We also discuss optimization of Chebyshev acceleration.