Low-lying Wilson Dirac operator eigenvector mixing in dynamical overlap Hybrid Monte-Carlo

TL;DR

This paper introduces a method to reduce large fermionic forces caused by near-zero eigenvector mixing in dynamical overlap fermion simulations, enhancing algorithm stability and efficiency.

Contribution

A novel technique that eliminates large forces and reduces action jumps from eigenvector mixing, improving the stability and performance of hybrid Monte Carlo simulations.

Findings

Large fermionic forces are mitigated.

Acceptance rates improve with larger time steps.

Algorithm stability is significantly increased.

Abstract

Current dynamical overlap fermion hybrid Monte Carlo simulations encounter large fermionic forces when there is mixing between near zero-eigenvectors of the kernel operator. This leads to low acceptance rates when there is a large density of near zero eigenvectors. I present a method where these large forces are eliminated and the large action jumps seen when two eigenvectors approach zero are significantly reduced. This significantly increases the stability of the algorithm, and allows the use of larger integration time steps.