1: \begin{abstract}

2:   The recent statistical finite element method (statFEM) provides a coherent

3: statistical framework to synthesise finite element models with observed

4: data. Through embedding uncertainty inside of the governing equations, finite

5: element solutions are updated to give a posterior distribution which quantifies

6: all sources of uncertainty associated with the model. However to incorporate all

7: sources of uncertainty, one must integrate over the uncertainty associated with

8: the model parameters, the known \textit{forward problem} of uncertainty

9: quantification. In this paper, we make use of Langevin dynamics to solve the

10: statFEM forward problem, studying the utility of the unadjusted Langevin

11: algorithm (ULA), a Metropolis-free Markov chain Monte Carlo sampler, to build a

12: sample-based characterisation of this otherwise intractable measure. Due to the

13: structure of the statFEM problem, these methods are able to solve the forward

14: problem without explicit full PDE solves, requiring only sparse matrix-vector products. ULA is

15: also gradient-based, and hence provides a scalable approach up to high

16: degrees-of-freedom. Leveraging the theory behind Langevin-based samplers, we

17: provide theoretical guarantees on sampler performance, demonstrating

18: convergence, for both the prior and posterior, in the Kullback-Leibler

19: divergence, and, in Wasserstein-2, with further results on the effect of

20: preconditioning. Numerical experiments are also provided, for both the prior and

21: posterior, to demonstrate the efficacy of the sampler, with a \texttt{Python}

22: package also included.

23: \end{abstract}

24: