Coarse grained models for variance reduction in particle methods
Monday, April 9, 2018 - 4:00pm - 5:00pm
Interacting particle methods are splitting methods that can be used to sample from high dimensional distributions, corresponding to, for instance, stochastic MD trajectories. Such methods do not alter or bias the underying dynamics. Instead a population of the MD trajectories evolve, and periodically some favorable trajectories are copied while others are killed. This resampling procedure leads to unbiased estimates of observables that can have lower empirical variance compared to naive Monte Carlo. We show how, in the rare event setting, coarse-grained models can be used to guide this resampling procedure to minimize the variance. In contrast to many rare event methods in the literature, this technique does not require reaction coordinates or a variable measuring progress towards the rare event. Moreover, it is optimal in the limit where the coarse model becomes exact and the population becomes infinite.