Exploring the HMC trajectory-length dependence of autocorrelation times in lattice QCD

TL;DR

This study investigates how the length of molecular dynamics trajectories in hybrid Monte-Carlo algorithms affects autocorrelation times in lattice QCD, finding longer trajectories can improve efficiency and thermalization.

Contribution

It demonstrates that longer trajectory lengths can reduce autocorrelation times and accelerate thermalization in lattice QCD simulations, challenging conventional practices.

Findings

Autocorrelation times vary by a factor of two with trajectory length.

Longer trajectories outperform conventional lengths in efficiency.

Faster thermalization observed with longer trajectories.

Abstract

We study autocorrelation times of physical observables in lattice QCD as a function of the molecular dynamics trajectory length in the hybrid Monte-Carlo algorithm. In an interval of trajectory lengths where energy and reversibility violations can be kept under control, we find a variation of the integrated autocorrelation times by a factor of about two in the quantities of interest. Trajectories longer than conventionally used are found to be superior both in the Nf=0 and Nf=2 examples considered here. We also provide evidence that they lead to faster thermalization of systems with light quarks.