Comparative Benchmarks of full QCD Algorithms

TL;DR

This paper provides performance benchmarks for various algorithms used in simulating two-flavor dynamical quarks in full QCD, including hybrid Monte Carlo methods with preconditioning, and discusses handling autocorrelations.

Contribution

It offers a comparative analysis of multiple algorithms for full QCD simulations, highlighting their performance and a method to manage autocorrelations in complex observables.

Findings

Hybrid and polynomial hybrid Monte Carlo algorithms show different performance profiles.

Preconditioning improves simulation efficiency.

A new method addresses autocorrelations in nonlinear observables.

Abstract

We report performance benchmarks for several algorithms that we have used to simulate the Schr"odinger functional with two flavors of dynamical quarks. They include hybrid and polynomial hybrid Monte Carlo with preconditioning. An appendix describes a method to deal with autocorrelations for nonlinear functions of primary observables as they are met here due to reweighting.