Obtaining a well-converged ensemble of structures from molecular dynamics (MD) simulations is extremely desirable as it allows determination of equilibrium properties and thermodynamic quantities, which are in turn important for the evaluation of force field accuracy. Despite advances in hardware (e.g. the advent of GPU-accelerated MD code), obtaining a converged ensemble of structures remains challenging with standard MD for all but the simplest biomolecules. Enhanced sampling techniques such as replica exchange MD (REMD) can be used to improve convergence speed when compared to MD. However, use of an enhanced sampling technique does not automatically guarantee convergence. Here we present several analysis methods for examining the convergence of multiple independent REMD simulations, including cluster analysis of combined coordinate trajectories and time-dependent Kullback-Leibler analysis which measures the evolution of overlap between probability distributions of various derived properties. We apply these analyses to REMD simulations in both 1 and 2 dimensions (Hamiltonian and Hamiltonian plus temperature), and show that well-converged ensembles can be obtained from the 2-dimensional REMD simulations. In addition, we show that the rate of convergence can be improved in a single dimension by simply increasing the number of replicas without increasing the level of bias.

Evaluating convergence in replica exchange molecular dynamics simulations

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.