1: \begin{abstract}

2: A randomized misfit approach is presented for the efficient

3: solution of large-scale PDE-constrained inverse problems with

4: high-dimensional data. The purpose of

5: this paper is to offer a theory-based framework for

6: random projections in this inverse problem setting. The stochastic approximation to

7: the misfit is analyzed using random projection theory. By expanding

8: beyond mean estimator convergence, a practical characterization of

9: randomized misfit convergence can be

10: achieved. The theoretical results developed hold with any valid random

11: projection in the literature. The class of feasible distributions is broad yet simple to characterize compared to previous stochastic misfit methods. This class includes very sparse random

12: projections which provide additional computational benefit. A different proof for a variant of the

13: Johnson-Lindenstrauss lemma is also provided. This leads to a different intuition

14: for the $\mc{O}(\varepsilon^{-2})$ factor in bounds for

15: Johnson-Lindenstrauss results. The main contribution of this paper is a theoretical result

16: showing the method guarantees a valid solution for small reduced

17: misfit dimensions. The interplay between Johnson-Lindenstrauss theory and Morozov's

18: discrepancy principle is shown to be essential to the

19: result. The computational cost savings for large-scale

20: PDE-constrained problems with high-dimensional data is discussed.

21: Numerical verification of the developed theory is presented for

22: model problems of estimating a distributed parameter in an elliptic partial

23: differential equation.  Results with different random projections are presented to demonstrate the viability and accuracy of the proposed approach.

24: \end{abstract}