A new algorithm for molecular dynamics simulations in the grand canonical ensemble

Abstract

We present an algorithm for implementing molecular dynamics simulations in the grand canonical ensemble that takes advantage of parallelism. The algorithm is an extension of the one presented recently for performing Monte Carlo simulations in the same ensemble. In contrast to most commonly used algorithms for open systems, instead of physically adding or deleting molecules to generate concentration fluctuations, parallel sets of trajectories are generated using molecular dynamics simulations in the canonical ensemble, corresponding to various compositions. Appropriate combinations of chains of configurations are selected according to the prescription of the grand canonical probability distribution. The method is illustrated for a test case of the isotopic Lennard-Jones mixture. We compare the thermodynamic properties obtained with this parallel method to those obtained from the Adams algorithm for performing Monte Carlo simulations in the same ensemble, observing a faster convergence to equilibrium and smaller errors with our method. Comparisons with the parallel Monte Carlo algorithm are also made.