A-posteriori QMC-FEM error estimation for Bayesian inversion and optimal control with entropic risk measure

Marcello Longo,Christoph Schwab,Andreas Stein

2023-07-06

Abstract:We propose a novel a-posteriori error estimation technique where the target quantities of interest are ratios of high-dimensional integrals, as occur e.g. in PDE constrained Bayesian inversion and PDE constrained optimal control subject to an entropic risk measure. We consider in particular parametric, elliptic PDEs with affine-parametric diffusion coefficient, on high-dimensional parameter spaces. We combine our recent a-posteriori Quasi-Monte Carlo (QMC) error analysis, with Finite Element a-posteriori error estimation. The proposed approach yields a computable a-posteriori estimator which is reliable, up to higher order terms. The estimator's reliability is uniform with respect to the PDE discretization, and robust with respect to the parametric dimension of the uncertain PDE input.

Numerical Analysis

What problem does this paper attempt to address?

The paper primarily focuses on addressing two types of high-dimensional numerical integration problems related to parameterized partial differential equations (PDEs), specifically including the Bayesian inverse problem (BIP) and the PDE-constrained optimal control problem (OCP) with entropy risk measures. The goal of the paper is to effectively estimate errors when numerically approximating these complex problems and to ensure solutions are obtained within a given numerical tolerance without wasting computational resources.

Specifically, the main contributions of the paper can be summarized as follows:

1. **Proposed a new posterior error estimation technique**: This technique is particularly suitable for high-dimensional integration problems constrained by PDEs, where the quantities of interest are in the form of ratios of high-dimensional integrals.

2. **Combined two error estimation methods**: This method integrates the posterior error analysis of Quasi-Monte Carlo (QMC) with the posterior error estimation of the Finite Element Method (FEM) to obtain a computable and reliable posterior error estimator.

3. **Addressed error estimation for parameter uncertainty and spatial discretization**: The proposed error estimator is reliable for PDE discretization and robust with respect to the dimensionality of parameter uncertainty.

4. **Scope of application**: This method is applicable to linear elliptic PDEs dependent on parameters, where uncertainty is introduced through affine parameterization of the diffusion coefficient.

5. **Numerical validation**: The paper also provides numerical experiments to validate the effectiveness of the proposed method.

In short, the core objective of the paper is to provide an efficient posterior error estimation technique for two important scientific computing problems, which helps ensure the accuracy of numerical solutions and provides guidance for practical applications.