Smoothed Circulant Embedding with Applications to Multilevel Monte Carlo Methods for PDEs with Random Coefficients

Anastasia Istratuca,Aretha Teckentrup

2024-06-06

Abstract:We consider the computational efficiency of Monte Carlo (MC) and Multilevel Monte Carlo (MLMC) methods applied to partial differential equations with random coefficients. These arise, for example, in groundwater flow modelling, where a commonly used model for the unknown parameter is a random field. We make use of the circulant embedding procedure for sampling from the aforementioned coefficient. To improve the computational complexity of the MLMC estimator in the case of highly oscillatory random fields, we devise and implement a smoothing technique integrated into the circulant embedding method. This allows to choose the coarsest mesh on the first level of MLMC independently of the correlation length of the covariance function of the random field, leading to considerable savings in computational cost. We illustrate this with numerical experiments, where we see a saving of factor 5-10 in computational cost for accuracies of practical interest.

Numerical Analysis

What problem does this paper attempt to address?

The paper is primarily dedicated to addressing the computational efficiency issues encountered when using Monte Carlo (MC) methods and Multilevel Monte Carlo (MLMC) methods in random coefficient partial differential equations (PDEs), particularly when dealing with highly oscillatory random fields (i.e., random fields with short correlation lengths).

Specifically, the paper focuses on the problem of uncertainty quantification in groundwater flow models. In this context, the unknown parameter is typically the hydraulic conductivity, which describes the ability of water to permeate through underground rock layers, and this is a random field. Since actual measurements can only be taken at a limited number of locations, and numerical simulations require estimating these parameter values across the entire computational domain, this introduces significant uncertainty.

To effectively handle this uncertainty, the paper proposes an improved circulant embedding method for sampling from highly oscillatory Gaussian random fields. This method involves smoothing the circulant embedding process, allowing these random fields to be represented on coarser grids, thereby significantly reducing computational costs. Moreover, this method is integrated into the Multilevel Monte Carlo framework to further enhance computational efficiency.

Through theoretical analysis and numerical experiments, the paper demonstrates that the proposed smoothed circulant embedding method, when combined with the Multilevel Monte Carlo method, significantly improves computational efficiency compared to traditional Monte Carlo methods, especially when dealing with random fields with short correlation lengths. Specifically, in some instances, the computational cost can be reduced by 5 to 10 times.

Therefore, the problem the paper attempts to solve is how to effectively reduce computational costs when dealing with random coefficient partial differential equations, particularly in the case of highly oscillatory random fields. By introducing smoothing techniques to improve the circulant embedding method and combining it with the Multilevel Monte Carlo method, the paper provides an efficient approach to address this issue.