Shortcutting Monte Carlo Uncertainty Propagation Plans With Matched Digital Nets

Abstract Planned uncertainty propagation tasks with Monte Carlo simulation often involve a set of predefined input samples to be propagated through a given model. This article presents a digital net matching scheme that sequentially selects low-discrepancy point sets that are subsets of the predefined input sample set. This technique is shown to generally outperform Monte Carlo simulation up until the point where all of the predefined input samples are used. The result is an uncertainty propagation methodology with the potential to enable confident decision-making prior to the execution of the entire predefined input sample set. The approach is demonstrated on randomly drawn input distributions of varying dimensions and randomly drawn functions from a class of Gaussian process prior distributions.