We are interested in sampling the stationary distribution of a class of Markov jump processes that are multiscale and irreversible. Based on the parallel replica method, we propose an accelerated algorithm that are suitable for massively parallel implementation. Theoretical results regarding numerical consistency and convergence will be presented. If time permits, the parametric sensitivity problem will be discussed. Joint work with Petr Plechac and David Aristoff.
Copyright. All Rights Reserved.