1: \begin{abstract}

2: A nonlinear optimization method is proposed for the solution of inverse medium problems

3: with spatially varying properties.

4: To avoid the prohibitively large number of unknown control variables resulting from

5: standard grid-based representations, the misfit is instead minimized in a small subspace

6: spanned by the first few eigenfunctions of a judicious elliptic operator, which itself depends

7: on the previous iteration. By repeatedly adapting 

8: both the dimension and the basis of the search space, regularization is inherently incorporated

9: at each iteration without the need for extra Tikhonov penalization.

10: Convergence is proved under an angle condition, which is included

11: into the resulting \emph{Adaptive Spectral Inversion} (ASI) algorithm. 

12: The ASI approach compares favorably to standard grid-based inversion using $L^2$-Tikhonov regularization when applied to an elliptic inverse problem. The improved accuracy

13: resulting from the newly included angle condition is further demonstrated via numerical experiments from

14: time-dependent inverse scattering problems.

15: 

16: \bigskip

17: \noindent

18: \textbf{Keywords:} Inverse problem, inverse scattering problem, adaptive eigenspace inversion, nonlinear optimization

19: 

20: \end{abstract}

21: