Stochastic Speculative Computation Method and its Application to Monte Carlo Molecular Simulation

Monte Carlo (MC) molecular simulation has significant computational complexity, and parallel processing is considered effective for computation of problems with large complexity. In recent years, multicore or many-core processors have gained significant attention as they enable computation with a large degree of parallelism on desktop computers. However, in conventional parallel processing, processes must be synchronized frequently; thus, parallel computing is not necessarily efficient. In this study, we evaluate the effect of applying MultiStart-based speculative parallel computation to MC simulations. Using probability theory, we performed theoretical verification to determine if speculative computation is more effective than conventional parallel computation methods. The parameters obtained from the theoretical calculations were observed in experiments wherein the speculative method was applied to an MC molecular simulation. In this paper, we report the results of the theoretical verification and experiments, and we show that speculative computation can accelerate MC molecular simulations