Pore-scale uncertainty quantification with multilevel Monte Carlo

Abstract

Computational fluid dynamics (CFD) simulations of pore-scale transport processes in porous media have recently gained large popularity. However the geometrical details of the pore structures can be known only in a very low number of samples and the detailed flow computations can be carried out only on a limited number of cases. The explicit introduction of randomness in the geometry and in other setup parameters can be crucial for the optimization of pore-scale investigations for random homogenization. Since there are no generic ways to parametrize the randomness in the porescale structures, Monte Carlo techniques are the most accessible to compute statistics. We propose a multilevel Monte Carlo (MLMC) technique to reduce the computational cost of estimating quantities of interest within a prescribed accuracy constraint. Random samples of pore geometries with a hierarchy of geometrical complexities and grid refinements, are synthetically generated and used to propagate the uncertainties in the flow simulations and compute statistics of macro-scale effective parameters.