Unbiased Simulation of Rare Events in Continuous Time

Abstract For rare events described in terms of Markov processes, truly unbiased estimation the event probability generally requires avoidance numerical approximations process. Recent work exact and $$\varepsilon$$ ε -strong simulation diffusions, which can be used to almost surely constrain sample paths a given tolerance, suggests one way do this. We specify how such algorithms combined with classical multilevel splitting method for simulation. This provides estimations question. discuss practical feasibility algorithm reference existing methods provide proof-of-concept examples.