An iteratively regularized Gauss–Newton–Halley method for solving nonlinear ill-posed problems

In this paper we analyze a second order method for regularizing nonlinear inverse problems, which comes from adapting Halley’s method to the ill-posed situation. The method is related to the iteratively regularized Gauss–Newton method and enhanced by the use of second derivatives of the forward operator. For parameter identification problems in PDEs, these second derivatives often require not much additional computational cost as compared to first order derivatives. We prove convergence and convergence rates under certain conditions on the forward operator in Hilbert spaces. Numerical tests for a problem of identifying a distributed coefficient in an elliptic PDE illustrate the performance of the method.