1: \begin{abstract}

2: 	\noindent

3: 	Gaussian random fields play an important role in many areas of science and engineering. In practice, they are often simulated by sampling from a high-dimensional multivariate normal distribution which arises from the discretisation of a suitable precision operator on a finite grid. Existing methods such as Cholesky factorisation and Gibbs-sampling become prohibitively expensive on fine meshes due to their high computational cost or the fact that they do not explore the probability space efficiently. In this work we revisit the Multigrid Monte Carlo algorithm of Goodman \& Sokal (\emph{Physical Review D} 40.6, 1989),  which can overcome these issues. The novelty of our work consists in the application of the method to linear Bayesian settings where the prior is constrained by a finite set of observations. For this, we develop a bespoke random smoother which takes care of the low-rank updates that arise in Bayesian inference. We provide a rigorous analysis of the method based on the link between linear solvers and samplers for multivariate normal distributions, drawing on standard multigrid convergence theory in a finite element setting. In particular, we prove that Multigrid Monte Carlo is algorithmically near-optimal in the limit of the grid-size going to zero. These theoretical results are confirmed by numerical experiments which demonstrate that Multigrid Monte Carlo can be significantly more efficient than alternative methods when applied in a Bayesian setting.

4: \end{abstract}

5: