Dimension-Independent MCMC Sampling for Inverse Problems with Non-Gaussian Priors

Abstract

The computational complexity of Markov chain Monte Carlo (MCMC) methods for the exploration of complex probability measures is a challenging and important problem in both statistics and the applied sciences. A challenge of particular importance arises in Bayesian inverse problems where the target distribution may be supported on an infinite-dimensional state space. In practice this involves the approximation of the infinite-dimensional target measure defined on sequences of spaces of increasing dimension bearing the risk of an increase of the computational error. Previous results have established dimension-independent bounds on the Monte Carlo error of MCMC sampling for Gaussian prior measures. We extend these results by providing a simple recipe for also obtaining these bounds in the case of non-Gaussian prior measures and by studying the design of proposal chains for the Metropolis–Hastings algorithm with dimension-independent performance. This study is motivated by an elliptic inverse problem with non-Gaussian prior that arises in groundwater flow. We explicitly construct an efficient Metropolis–Hastings proposal based on local proposals in this case, and we provide numerical evidence supporting the theory.

Type
Publication
SIAM/ASA J. Uncertainty Quantification