Discretization errors in molecular dynamics simulations with deterministic and stochastic thermostats

TL;DR

This paper examines how numerical discretization errors affect average measurements in molecular dynamics simulations of water using different thermostats, proposing practical methods to estimate and mitigate these errors.

Contribution

It introduces simple approaches for estimating discretization errors in molecular dynamics simulations with various thermostats, enabling accurate measurements at larger step sizes.

Findings

Accurate measurements possible with step sizes up to 70% of stability threshold.

Discretization errors can be effectively estimated and corrected.

Step size of about 7 fs is feasible for TIP4P water simulations.

Abstract

We investigate the influence of numerical discretization errors on computed averages in a molecular dynamics simulation of TIP4P liquid water at 300 K coupled to different deterministic (Nos\'e-Hoover and Nos\'e-Poincar\'e) and stochastic (Langevin) thermostats. We propose a couple of simple practical approaches to estimating such errors and taking them into account when computing the averages. We show that it is possible to obtain accurate measurements of various system quantities using step sizes of up to 70% of the stability threshold of the integrator, which for the system of TIP4P liquid water at 300 K corresponds to the step size of about 7 fs.