Autocorrelations in Hybrid Monte Carlo Simulations

TL;DR

This paper investigates the critical slowing down in Hybrid Monte Carlo simulations of QCD, especially in topological charge, and introduces a new error analysis method to better estimate slow mode contributions.

Contribution

It presents a novel error analysis technique that accurately assesses the impact of slow modes on simulation errors in HMC QCD simulations.

Findings

Critical slowing down significantly affects all observables in HMC simulations.

The new error analysis method provides more realistic error estimates.

Slow modes contribute substantially to the overall errors in high-statistics runs.

Abstract

Simulations of QCD suffer from severe critical slowing down towards the continuum limit. This problem is known to be prominent in the topological charge, however, all observables are affected to various degree by these slow modes in the Monte Carlo evolution. We investigate the slowing down in high statistics simulations and propose a new error analysis method, which gives a realistic estimate of the contribution of the slow modes to the errors.