On steepest-descent-Kaczmarz methods for regularizing systems of nonlinear ill-posed equations (Q942384)

A new method for obtaining regularized approximations of systems of nonlinear ill-posed operator equations is proposed. The inverse problem consists of determining an unknown physical quantity \(x\in X\) from the set of data \(y_1,\dots, y_N\in Y^N\), where \(X\) and \(Y\) are Hilbert spaces. In applications, we do not know the data exactly, but only approximate measured data \(y^\delta_i\). The quantities (parameters) \(x\) are solutions of \(F_i(x)= y_i\), \(i= 1,\dots, N\). Standard methods to solve the equations are based on the iterative-type regularization or Tikhonov-type regularization. It is known, that these methods become inefficient if \(N\) is large or the evaluations of \(F_i(x)\), \(F_i'(x)\) are expensive. Therefore, a loping steepest-descent method of Kaczmarz-type for solving ill-posed problems is used in the present paper. For this method a convergence analysis is developed and proved. Then the numerical performance of this method is compared with other standard algorithms to solve inverse problems. Numerical tests are presented for a linear problem related to photoacoustic tomography and a nonlinear problem related to the testing of semiconductor devises. They are illustrated by impressive figures.