Regularization of exponentially ill-posed problems

From MaRDI portal





Ill-posed problems (1) \(Tx= y\) are considered, where \(T: X\to Y\) is a linear operator between Hilbert spaces \(X\) and \(Y\). It is supposed that equation (1) has a minimum norm solution \(x^\dagger\) that satisfies a logarithmic source condition of the form (2) \(x^\dagger= f_p(T^*T) w\), where \(f_p(\lambda)= (-\ln \lambda)^{-p}\) for \(0 < \lambda \leq \exp(-1)\) with some real parameter \(p > 0\), and a scaling \(\|T \|\leq \exp(-1/2)\) is assumed. For the stable solution of equation (1) with a minimum norm solution that satisfies such a logarithmic source condition (2), the author considers a priori and a posteriori parameter choice strategies for regularization methods of the general form (2) \(x_\alpha^\delta = g_\alpha(T^*T) T^* y^\delta\). Here \(y^\delta \in Y \) is a perturbation of the exact right-hand side \(y\) with \(\|y^\delta- y \|\leq \delta\). Moreover it is supposed that \(g_\alpha\) (\(\alpha > 0\)) are given functions such that, for each \(\lambda > 0\), we have \(g_\alpha(\lambda) \to 1/ \lambda\) as \(\alpha \to 0\). Under certain additional conditions on the mappings \(g_\alpha\), order optimal convergence rates of the form \(\|x^\dagger -x_\alpha^\delta \|\leq c_p \rho f_p(\delta^2/\rho^2)\) are obtained for each considered parameter choice \(\alpha= \alpha(\delta)\), where \(\rho \geq \|w \|\) holds. The considered class of regularization methods contains well-known regularization methods like the Tikhonov regularization and the Landweber iteration, for example. This class of regularization methods is considered also to solve a class of nonlinear ill-posed problems. It is supposed here that the underlying nonlinear operator is given approximately by some linear operator, and moreover it is supposed that the solution to this nonlinear equation exists and satisfies a generalized logarithmic source condition. For a second class of nonlinear ill-posed problems the iteratively regularized Gauss-Newton method is considered. Finally it is shown that the obtained results can be applied to specific problems like the backwards heat equation and the sideways heat equation, respectively.




Cited in
(only showing first 100 items - show all)








This page was built for publication: Regularization of exponentially ill-posed problems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4508645)