Stochastic molecular dynamics in systems with multiple time scales and memory friction

Abstract

The generalized Langevin equation (GLE) has been used to model a wide variety of systems in which a subset of the degrees of freedom move on a potential of mean force surface subject to fluctuating forces and dynamic friction. When there is a wide separation in the time scales for motion on the potential surface and for relaxation of the friction kernel, direct integration of the GLE is very costly in CPU time. In this paper we introduce an integrator based on our previous work using numerical analytical propagator algorithm (NAPA) and reference system propagator algorithm (RESPA) that greatly accelerates such simulations. We also discuss sampling methods for the random force. Accuracy of this algorithm is assessed by comparisons with an analytically solvable example. Introducing dynamic friction kernels determined from full molecular dynamics (MD) simulations allows us to compare the accuracy of the GLE simulations with full scale molecular dynamics simulations.