A posteriori error estimates for the solution of nonlinear ill-posed operator equations (Q5946008)

The author derives a posteriori error estimates of the form \[ \|u_h-u^\dag \|X\leq f\biggl(\bigl \|F(u_h)\bigr \|_Y\biggr)\quad \text{with}\quad u_h\in U_{a-p} (u^\dag)\tag{1} \] for approximate solutions \(u_h\) of inverse and ill-posed problems written as operator equations \(F(u)=0\), where the nonlinear Fréchet differentiable operator \(F:D\subseteq X\to Y\) maps between Hilbert spaces \(X\) and \(Y\). In this context, the elements \(u_h\) come from a neighbourbood \(U_{a-p}\subset D\) of a solution \(u^\dag\in D\) with \(F(u^\dag)=0\). A first part of the paper is devoted to the study of a posteriori estimates of general type based on the degree of nonlinearity. For different such degrees specific subsets \(U_{a-p}\) are presented that provide error estimates (1) with appropriate functions \(f(t)\). In particular, the cases \(f(t)=t^p\) \((0<p\leq 1)\) for mildly and moderately ill-posed problems and \(f(t)=(-\ln (t))^{-p}\) \((p>0)\) for severely ill-posed problems are of interest. The main objective of the paper is the introduction of a posteriori estimates into the theory of regularization methods, which allows the development of new techniques like two-level methods for the solution of inverse problems. A second part of the paper shows that the general results can also be applied to derive a posteriori error estimates for nonlinear differential equations.