Data-driven basis for reconstructing the contrast in inverse scattering: Picard criterion, regularity, regularization, and stability

TL;DR

This paper introduces a data-driven basis using generalized prolate spheroidal wave functions for inverse scattering, providing a new regularization and stability framework applicable to various data configurations.

Contribution

It develops a novel basis for inverse scattering problems based on eigenfunctions of a Fourier integral operator, enabling improved reconstruction and stability analysis.

Findings

Established a Picard criterion for contrast reconstruction.

Derived spectral cutoff regularization strategies for noisy data.

Provided stability estimates for contrast in different data scenarios.

Abstract

We consider the inverse medium scattering of reconstructing the medium contrast using Born data, including the full aperture, limited-aperture, and multi-frequency data. We propose data-driven basis functions for these inverse problems based on the generalized prolate spheroidal wave functions and related eigenfunctions. Such data-driven eigenfunctions are eigenfunctions of a Fourier integral operator; they remarkably extend analytically to the whole space, are doubly orthogonal, and are complete in the class of band-limited functions. We first establish a Picard criterion for reconstructing the contrast using the data-driven basis, where the reconstruction formula can also be understood from the viewpoint of data processing and analytic extrapolation. Another salient feature associated with the generalized prolate spheroidal wave functions is that the data-driven basis for a disk is also a basis for a Sturm-Liouville differential operator. With the help of Sturm-Liouville theory, we estimate the $L^2$ approximation error for a spectral cutoff approximation of $H^s$ functions. This yields a spectral cutoff regularization strategy for noisy data and an explicit stability estimate for contrast in $H^s$ ($0<s<1/2$) in the full aperture case. In the limited-aperture and multi-frequency cases, we also obtain spectral cutoff regularization strategies for noisy data and stability estimates for a class of contrast.