Unlocking Large Scale Uncertainty Quantification with In Transit Iterative Statistics

Multi-run numerical simulations using supercomputers are increasingly used by physicists and engineers for dealing with input data model uncertainties. Most of the time, parameters a simulation modeled as random variables, then run (possibly large) number times varied according to specific design experiments. Uncertainty quantification is hard computational problem, currently bounded large size produced results. This book chapter about in situ techniques enable scale uncertainty studies. We provide comprehensive description Melissa, file avoiding, adaptive, fault-tolerant, elastic framework that computes transit statistical quantities interest. Melissa implements on-the-fly computation statistics necessary realization studies: moment-based (mean, standard deviation, higher orders), quantiles, Sobol’ indices, threshold exceedance.