A generalization of Arcangeli's method for ill-posed problems leading to optimal rates (Q1803936): Difference between revisions

The author discusses a strategy of parameter choice for the regularization of ill-posed problems, leading to a near-optimal convergence rate. Let \(X\) and \(Y\) be Hilbert spaces and \(T:X\to Y\) be a bounded linear operator. Let \(y^ \delta(\delta>0)\) be inexact data such that \(\| y-y^ \delta\|\leq\delta\) and \(x^ \delta_ \alpha=(T^*T+\alpha I)^{-1}T^*y^ \delta\) for \(\alpha>0\). Similar to \textit{R. Arcangeli}'s method [C. R. Acad. Sci. Paris, Sér. A 263, No. 8, 282-285 (1966; Zbl 0166.411)], a generalized ``discrepancy principle'' is suggested for the choice of \(\alpha=\alpha(\delta)\). It is shown that the strategy leads to a near-optimal convergence rate of \(\{x_ \alpha^ \delta\}\) as \(\delta\to 0\). The results constitute an improvement compared to the work of \textit{E. Schock} [J. Optimization Theor. Appl. 44, 95-104 (1984; Zbl 0531.65031)].