Bootstrapping time correlation functions of molecular dynamics

TL;DR

This paper introduces a bootstrap method to efficiently estimate uncertainties in time correlation functions from molecular dynamics simulations, addressing error quantification and finite size effects.

Contribution

It applies bootstrap techniques to molecular dynamics data for the first time, providing a low-cost alternative to traditional error estimation methods.

Findings

Bootstrap method effectively quantifies uncertainties in correlation functions.

Comparison shows bootstrap matches replica method in accuracy.

Bootstrap helps analyze finite size effects in simulations.

Abstract

Molecular dynamics is often considered as a numerical experiment. The error bars on the results are therefore mandatory, but sometimes difficult to determine and computationally demanding. As a low-cost approach, we describe the application of the bootstrap (BS) method to the quantification of uncertainties pertaining to the time correlation functions. We chose the autocorrelation functions of velocity and interdiffusion current for a binary ionic mixture as a test bed, and we assessed the merit of the Darken approximation relating both of them. The intrinsic errors related to phase space sampling is investigated comparing the BS method with the reference method of replica. We also study how the BS method can assist in addressing the finite size effects.