Laboratoire de Physique Théorique
Toulouse - UMR 5152

Rare events are of importance in many branches of science, for example in the form of heat waves or storms in climate dynamics and meteorology or complex reactions in chemistry. Although these events are rare, they can have a large impact. A reliable estimation of their probability is therefore necessary.

Recently, a class of rare event simulation algorithms has been introduced that could yield significant increases in performance with respect to brute force estimation of rare event probabilities (see e.g. [1]). In these algorithms an ensemble of simulations is evolved in parallel, while at certain interaction times ensemble members are killed and cloned so as to have better statistics in the region of phase space that is relevant to the rare event of interest.

I will discuss issues in implementation and performance gains of the algorithms. I also present results on a first application of a rare event simulation algorithm to a toy model for chaos in the atmosphere, the Lorenz 96 model. I demonstrate that for the estimation of the histogram tail of the energy observable, the algorithm gives a significant error reduction. I will furthermore discuss first results on the application of rare event simulation algorithms to a complex climate model.

[1] Del Moral, P. & Garnier, J. "Genealogical particle analysis of rare events." The Annals of Applied Probability 15, 2496–2534 (2005).
[2] Wouters, J. & Bouchet, F. "Rare event computation in deterministic chaotic systems using genealogical particle analysis" arXiv:1511.02703