Goal oriented adaptivity in the IRGNM for parameter identification in PDEs: II. all-at-once formulations

In this paper we investigate adaptive discretization of the iteratively regularized Gauss–Newton method IRGNM. All-at-once formulations considering the PDE and the measurement equation simultaneously allow to avoid (approximate) solution of a potentially nonlinear PDE in each Newton step as compared to the reduced form Kaltenbacher et al (2014 Inverse Problems 30 045001). We analyze a least squares and a generalized Gauss–Newton formulation and in both cases prove convergence and convergence rates with a posteriori choice of the regularization parameters in each Newton step and of the stopping index under certain accuracy requirements on four quantities of interest. Estimation of the error in these quantities by means of a weighted dual residual method is discussed, which leads to an algorithm for adaptive mesh refinement. Numerical experiments with an implementation of this algorithm show the numerical efficiency of this approach, which especially for strongly nonlinear PDEs outperforms the nonlinear Tikhonov regularization considered in Kaltenbacher et al (2011 Inverse Problems 27 125008). (paper)