A modified Landweber iterative method for solving a class of parameter identification problems (Q3022221)

The paper concerns an ill-posed parameter identification problem in an abstract setting. The corresponding direct problem has the form \(Au+L(N(u,g))=f\), where \(A\), \(L\) are linear operators considered in certain Hilbert spaces, \(A\) is invertible and \(N\) is a bilinear operator. The problem is to reconstruct \(g\) from inexact data \(\widetilde{u}\). The Landweber iteration method is applied for solving this problem. Following the idea introduced by \textit{H. W. Engl} and \textit{J. Zou} [Inverse Probl. 16, No. 6, 1907--1923 (2000; Zbl 0968.35124)], source assumptions are formulated for an approximate Fréchet derivative. The author shows that the modified version of Landweber iteration with a discrepancy principle as a stopping criteria is convergent under this weaker form of source conditions, similarly as it was established by \textit{M. Hanke, A. Neubauer} and \textit{O. Scherzer} [Numer. Math. 72, No. 1, 21--37 (1995; Zbl 0840.65049)] under stronger assumptions on the Fréchet derivative. Numerical examples illustrate the method for the parameter identification for the boundary value problem for second order differential equations.