1: \begin{abstract}

2: We present a randomized maximum {\em a posteriori} (rMAP) method for

3: generating approximate samples of posteriors in high dimensional

4: Bayesian inverse problems governed by large-scale forward problems. We

5: derive the rMAP approach by: 1) casting the problem of computing the

6: MAP point as a stochastic optimization problem; 2) interchanging

7: optimization and expectation; and 3) approximating the expectation

8: with a Monte Carlo method. For a specific randomized data and prior

9: mean, rMAP reduces to the maximum likelihood approach (RML). It can

10: also be viewed as an iterative stochastic Newton method. An analysis

11: of the convergence of the rMAP samples is carried out for both linear

12: and nonlinear inverse problems. Each rMAP sample requires solution of

13: a PDE-constrained optimization problem; to solve these problems, we

14: employ a state-of-the-art trust region inexact Newton conjugate

15: gradient method with sensitivity-based warm starts. An

16: approximate Metropolization approach is  presented to reduce the

17: bias in rMAP samples. Various numerical methods will be presented to

18: demonstrate the potential of the rMAP approach in posterior sampling 

19: of nonlinear Bayesian inverse problems in high dimensions.

20: \end{abstract}

21: