A posteriori parameter choice for general regularization methods for solving linear ill-posed problems (Q1104704)

The purpose of this paper is to present an a posteriori parameter choice strategy for regularization methods that leads to optimal convergence rates and that does not require knowledge of the smoothness of the exact solution. It is proved that optimal convergence rates are possible with a parameter choice criterion depending only on the given perturbed right- hand side - i.e., only the actual data are required. It is also proved that the convergence rate is optimal in the sense that there can be no parameter choice rule yielding faster convergence rates. The theory applies to continuous as well as iterated regularization. In the last section, the theory is used to derive parameter choice rules for the following specific regularization methods: ordinary and iterated Tikhonov regularization, modified spectral cut-off and modified truncated singular value expansion, Landweber-Fridman iteration, and the three methods of Showalter, Lardy and Schulz.