A variant of importance splitting for rare event estimation

摘要: Importance splitting is a simulation technique to estimate very small entrance probabilities for Markov processes by sample paths at various stages before reaching the set of interest. This can be done in many ways, yielding different variants method. In this context, we propose new one, called fixed number successes. We prove unbiasedness and some known variants, because papers, proof based on an incorrect argument. Further, analyze its behavior simplified setting terms efficiency asymptotics comparison standard variant. The main difference that it controls imprecision estimator rather than computational effort. Our analysis examples show robust parameter choice present two-stage procedure which also yields confidence intervals.