Reduced-order modeling for nonlinear Bayesian statistical inverse problems (1909.02539v1)

Abstract: Bayesian statistical inverse problems are often solved with Markov chain Monte Carlo (MCMC)-type schemes. When the problems are governed by large-scale discrete nonlinear partial differential equations (PDEs), they are computationally challenging because one would then need to solve the forward problem at every sample point. In this paper, the use of the discrete empirical interpolation method (DEIM) is considered for the forward solves within an MCMC routine. A preconditioning strategy for the DEIM model is also proposed to accelerate the forward solves. The preconditioned DEIM model is applied to a finite difference discretization of a nonlinear PDE in the MCMC model. Numerical experiments show that this approach yields accurate forward results. Moreover, the computational cost of solving the associated statistical inverse problem is reduced by more than 70%.