A Budgeted Multi-Level Monte Carlo Method for Full Field Estimates of Multi-PDE Problems

Abstract: We present a high-performance budgeted multi-level Monte Carlo method for estimates on the entire spatial domain of multi-PDE problems with random input data. The method is designed to operate optimally within memory and CPU-time constraints and eliminates the need for a priori knowledge of the problem's regularity and the algorithm's potential memory demand. To achieve this, we build on the budgeted multi-level Monte Carlo framework and enhance it with a sparse multi-index update algorithm operating on a dynamically assembled parallel data structure to enable estimates of the full field solution. We demonstrate numerically and provide mathematical proof that this update algorithm allows computing the full spatial domain estimates at the same CPU-time cost as a single quantity of interest, and that the maximum memory usage is similar to the memory demands of the deterministic formulation of the problem despite solving the stochastic formulation in parallel. We apply the method to a sequence of interlinked PDE problems, ranging from a stochastic partial differential equation for sampling random fields that serve as the diffusion coefficient in an elliptic subsurface flow problem, to a hyperbolic PDE describing mass transport in the resulting flux field.