Trajectory-guided sampling for molecular dynamics simulation

Direct molecular dynamics simulation for processes dominated by rare events is still challenging. Here we propose a trajectory-guided sampling technique which generates configurations for molecular dynamics (MD) simulation in a Monte Carlo (MC) procedure based on a set of configurations from a reference MD trajectory. Combining the advantages of the controllable accessibility of MC samplings to infrequently visited domains in phase space and the faithful representability of MD simulations for the collective fluctuations, the new scheme allows for an efficient sampling for hydrogen diffusion on the copper (001) surface, a benchmark system of rare event-dominated dynamics. TGS produces results for mean square displacement functions in quite good agreement with those obtained from regular MD simulations but at a less expensively computational cost. It is suggested that TGS may have great potential to be applied to general molecular dynamics simulations for rare events.