Unbiasedness of some generalized adaptive multilevel splitting algorithms (Q511481)

A generalization of the adaptive multilevel splitting (AMS) algorithm in the discrete time dynamic setting when it is applied to sample rare events associated with paths of Markov chains is discussed. An estimator of the rare event probability (and of any nonnormalized quantity associated with this event) is constructed. This is unbiased, whatever the choice of the importance function and the number of replicas may be. Numerical experiments which illustrate the unbiasedness result and the efficiency of the AMS algorithm to sample rare events are given.