Analysis of an interacting particle method for rare event estimation (Q351492)

Among the various numerical schemes based on the construction of branching processes that have been proposed for problems of rare event estimation the present paper focuses on a procedure where nested sets are introduced with a rule for branching when these sets are entered. This class of schemes have recently been considered and called interacting particle systems or genealogical particle methods. Here the large deviation analysis for the performance of the method is developed in a one-dimensional setting. Dependence at each branching time is reintroduced by a resampling and this makes the analysis more difficult than in the independent scheme. It is an occupation measure that is now relevant and within this framework neither stationarity nor Feller property can be assumed. Large deviation upper and lower bounds are yet obtained. For any fixed number of particles, the method has a suboptimal rate of convergence of the second moment of the estimation to zero.