Linear Functional Strategy and the Approximate Inverse for Nonlinear Ill-Posed Problems

Abstract

Abstract This article generalizes the results of the so-called linear functional strategy [R. S. Anderssen, Inverse Problems (Oberwolfach, 1986)], used for fast reconstruction of some particular feature of interest in the solution of a linear inverse problem. Two versions are proposed for nonlinear problems. The first one applies to differentiable forward operators and is based on the One-Step Newton method. The second one, in turn, uses a linearization of the forward operator obtained by the employment of basic Machine Learning techniques, being applicable to non-differentiable operators. As a byproduct of the proposed methods, we derive two variants of the so-called approximate inverse method [A. K. Louis, Inverse Problems, 1996] for nonlinear inverse problems. Numerical tests, using electrical impedance tomography applied to a biphasic flow problem, are presented to test the efficiency of the proposed methods.