[Paper Review] Multi-level quasi-Monte Carlo finite element methods for a class of elliptic partial differential equations with random coefficients

Frances Y. Kuo, Christoph Schwab+1 more|arXiv (Cornell University)|Aug 1, 2012

Probabilistic and Robust Engineering Design17 references17 citations

TL;DR

This paper introduces a multi-level quasi-Monte Carlo (QMC) finite element method for solving elliptic PDEs with random coefficients, combining multi-level FE discretizations with randomly shifted lattice rules. It achieves an optimal convergence rate of O(h²) in root-mean-square error with total work comparable to a single fine-level FE solve, significantly reducing computational cost compared to standard Monte Carlo methods.

ABSTRACT

Quasi-Monte Carlo (QMC) methods are applied to multi-level Finite Element (FE) discretizations of elliptic partial differential equations (PDEs) with a random coefficient, to estimate expected values of linear functionals of the solution. The expected value is considered as an infinite-dimensional integral in the parameter space corresponding to the randomness induced by the random coefficient. We use a multi-level algorithm, with the number of QMC points depending on the discretization level, and with a level-dependent dimension truncation strategy. In some scenarios, we show that the overall error is $\mathcal{O}(h^2)$, where $h$ is the finest FE mesh width, or $\mathcal{O}(N^{-1+δ})$ for arbitrary $δ>0$, where $N$ is the maximal number of QMC sampling points. For these scenarios, the total work is essentially of the order of one single PDE solve at the finest FE discretization level. The analysis exploits regularity of the parametric solution with respect to both the physical variables (the variables in the physical domain) and the parametric variables (the parameters corresponding to randomness). Families of QMC rules with "POD weights" ("product and order dependent weights") which quantify the relative importance of subsets of the variables are found to be natural for proving convergence rates of QMC errors that are independent of the number of parametric variables.

Motivation & Objective

• To reduce the computational cost of computing expected values of linear functionals of solutions to elliptic PDEs with random coefficients.
• To extend the single-level QMC finite element approach of [24] to a multi-level framework for improved efficiency.
• To analyze the error and work complexity of the multi-level QMC FE method under general assumptions on the random coefficient's parametric structure.
• To derive new POD (product and order dependent) weights tailored for the multi-level setting that ensure convergence independent of the number of parametric variables.

Proposed method

• Combines multi-level finite element discretizations with randomly shifted lattice rules for quasi-Monte Carlo integration in the infinite-dimensional parameter space.
• Uses a hierarchical decomposition of the solution into level-wise corrections: $ u^{s_\ell}_{h_\ell} - u^{s_{\ell-1}}_{h_{\ell-1}} $, enabling variance reduction.
• Applies a level-dependent dimension truncation strategy to balance accuracy and computational cost.
• Employs product and order dependent (POD) weights in the QMC rules to quantify variable importance and ensure convergence rates independent of parametric dimension.
• Derives error bounds by decomposing the total error into truncation, QMC, and FE discretization components.
• Uses a cost model assuming linear complexity solvers for FE problems and exact stiffness matrix assembly.

Experimental results

Research questions

• RQ1 Can the computational cost of computing expected values of functionals of the solution to random elliptic PDEs be significantly reduced using a multi-level QMC finite element approach?
• RQ2 What is the optimal balance between QMC sampling points, FE mesh refinement, and parametric truncation levels to minimize total work while maintaining high convergence rates?
• RQ3 How should the POD weights in the QMC rules be chosen in the multi-level setting to ensure convergence independent of the number of parametric variables?
• RQ4 What is the convergence rate of the multi-level QMC FE method in terms of the finest mesh width h, and can it achieve optimal complexity?
• RQ5 How does the multi-level QMC method compare in efficiency to single-level QMC and multi-level Monte Carlo methods for the same class of problems?

Key findings

• For spatial dimension d = 2 with linear elements, the total work of the multi-level QMC FE method is essentially of the order of one single PDE solve at the finest FE discretization level.
• The root-mean-square error of the expected value estimate is O(h²) or O(N⁻¹⁺δ) for arbitrary δ > 0, where N is the maximal number of QMC sampling points.
• In Scenario 1 with k-orthogonality and appropriate parameter choices, the method achieves optimal convergence rate O(h²) with work O(h⁻²/(¹⁻⁵)) for δ = 0.5, matching the best possible bound for deterministic H²-regular problems.
• The method requires stronger regularity assumptions on the parametric functions ψj than in the single-level case, specifically ∑j≥1 ∥ψj∥qW¹,∞(D) < ∞ with q = p/(1−p) ≤1.
• The POD weights used in the QMC rules are different from those in the single-level setting and are specifically tailored to the multi-level structure to maintain convergence.
• The analysis extends to weaker regularity assumptions such as Hölder continuity, with error bounds depending on ∥ψj∥C⁰,ᵣ(D) instead of ∥ψj∥W¹,∞(D).