Rare-event sampling applied to the simulation of extreme mechanical eorts exerted by a turbulent ow on a blu body

This study evaluates the relevance of rare-event sampling techniques to accelerate the simulation of extreme mechanical eorts exerted by a turbulent ow impinging onto a blu body. The main idea is to replace a long simulation by a set of much shorter ones, running in parallel, with dynamics that are replicated or pruned in order to sample large-amplitude events more frequently. Such techniques have been shown to be ecient for a wide range of problems in statistical physics, computer science, biochemistry, enabling the simulation of rare events otherwise out of reach by direct sampling. This work is the rst application to uid-structure interaction problems. The drag experienced by a squared obstacle placed in a turbulent ow (in two dimensions) is taken as a representative case study to investigate the performance of two major rare-event sampling algorithms, namely the Adaptive Multilevel Splitting (AMS) and the Giardina-Kurchan-Tailleur-Lecomte (GKTL) algorithms. Practical evidence is given that the fast sweeping-time of uid structures past the obstacle has a drastic inuence on the eciency of these two algorithms. While it is shown that the AMS algorithm does not yield signicant run-time savings, the GKTL algorithm appears to be ecient to sample extreme uctuations of the time-averaged drag and estimate related statistics such as return times. Beyond the study of applicability of rare-event sampling techniques to a uid-mechanical problem, this work also includes a detailed phenomenological description of extreme-drag events of a turbulent ow on a blu body.