Accelerating the Dynamics of Infrequent Events: Combining Hyperdynamics and Parallel Replica Dynamics to Treat Epitaxial Layer Growth

Abstract

AbstractDuring the growth of a surface, morphology-controlling diffusion events occur over time scales that far exceed those accessible to molecular dynamics (MD) simulation. Kinetic Monte Carlo offers a way to reach much longer times, but suffers from the fact that the dynamics are correct only if all possible diffusion events are specified in advance. This is difficult due to the concerted nature of many of the recently discovered surface diffusion mechanisms and the complex configurations that arise during real growth. Here we describe two new approaches for this type of problem. The first, hyperdynamics, is an accelerated MD method, in which the trajectory is run on a modified potential energy surface and time is accumulated as a statistical property. Relative to regular MD, hyperdynamics can give computational gains of more than 102. The second method offers a way to parallelize the dynamics efficiently for systems too small for conventional parallel MD algorithms. Both methods exploit the infrequent-event nature of the diffusion process. After an introductory description of these methods, we present preliminary results from simulations combining the two approaches to reach near-millisecond time scales on systems relevant to epitaxial metal growth.