We present a multi-index quasi--Monte Carlo method for the solution of elliptic partial differential equations with random coefficients. By combining the multi-index sampling idea with randomly shifted rank-1 lattice rules, the algorithm constructs an estimator for the expected value of some functional of the solution. The efficiency of this new method is illustrated on a three-dimensional subsurface flow problem with lognormal diffusion coefficient with underlying Matérn covariance function. This example is particularly challenging because of the small correlation length considered, and thus the large number of uncertainties that must be included. We show numerical evidence that it is possible to achieve a cost inversely proportional to the requested tolerance on the root-mean-square error, for problems with a smoothly varying random field.
Civil engineering applications are often characterized by a large uncertainty on the material parameters. Discretization of the underlying equations is typically done by means of the Galerkin Finite Element method. The uncertain material parameter can be expressed as a random field represented by, for example, a Karhunen–Loève expansion. Computation of the stochastic responses, i.e., the expected value and variance of a chosen quantity of interest, remains very costly, even when state-of-the-art Multilevel Monte Carlo (MLMC) is used. A significant cost reduction can be achieved by using a recently developed multilevel method: p-refined Multilevel Quasi-Monte Carlo (p-MLQMC). This method is based on the idea of variance reduction by employing a hierarchical discretization of the problem based on a p-refinement scheme. It is combined with a rank-1 Quasi-Monte Carlo (QMC) lattice rule, which yields faster convergence compared to the use of random Monte Carlo points. In this work, we developed algorithms for the p-MLQMC method for two dimensional problems. The p-MLQMC method is first benchmarked on an academic beam problem. Finally, we use our algorithm for the assessment of the stability of slopes, a problem that arises in geotechnical engineering, and typically suffers from large parameter uncertainty. For both considered problems, we observe a very significant reduction in the amount of computational work with respect to MLMC.
Random fields are used to represent spatially-varying uncertainty, and are commonly used as training data in uncertainty quantification and machine learning applications. Gaussian-RandomFields.jl is a Julia (Bezanson et al., 2017) software package to generate and sample from Gaussian random fields. It offers support for well-known covariance functions, such as Gaussian, exponential and Matérn covariances (Bishop & Nasrabadi, 2006; Chiles & Delfiner, 2012; Montero et al., 2015), as well as user-defined covariance structures defined on arbitrary domains. The package implements most common methods to generate samples from these random fields, including the Cholesky factorization, the Karhunen-Loève expansion, and the circulant embedding method (Lord et al., 2014). GaussianRandomFields.jl makes use of Plots.jl (Christ et al., 2023) to quickly visualize samples of the random fields.