On the regularization of nonlinear ill-posed problems via inexact Newton iterations (Q4237183)

The author considers the stable solution of the nonlinear problem \[ F(x)= y^\delta,\tag{1} \] where \(F: D(F)\subset X\to Y\) operates between the Hilbert spaces \(X\) and \(Y\). Here, \(D(F)\) denotes the domain of definition of \(F\). In (1) \(y^\delta\) is a perturbation of the exact but unknown data \(y= F(x^*)\) satisfying \[ \|y- y^\delta\|\leq\delta\tag{2} \] with the a priori known noise level \(\delta\geq 0\). We call (1) ill-posed if \(x^*\), the solution of (1) with exact data, does not depend continuously on \(y\). Any algorithm for solving (1) has to take care of this instability. Algorithms computing approximations from \(y^\delta\) to \(x^*\) in a stable way are named regularizations. It is supposed, throughout the paper, that \(F\) is compact and continuous and \(D(F)\) is finite-dimensional. Then (1) is ill-posed (essentially). The author treats with inexact Newton methods for the stable solution of nonlinear ill-posed problems (1). The corresponding inner scheme is chosen to be any linear regularization with a sufficient modulus of convergence. The regularization property of these Newton-type algorithms is verified, that is, the iterates converge to a solution of the nonlinear problem with exact data when the noise level tends to zero. Moreover, convergence rate are given. Finally, implementation issues are discussed and the algorithm is applied to a parameter identification problem for an elliptic partial differential equation. The numerical results reproduce the theoretical predictions. It is shown that the efficiency of the algorithm depends on the choice of the tolerances. In section 6 the author proposes a dynamic selection strategy based on the convergence analysis. The relation between the inexact Newton method and other iterative techniques to regularize (1) is discussed in the final section.