On the optimality of regularization methods for solving linear ill-posed problems

Publication date 26 November 1995

Published in Zeitschrift für Analysis und ihre Anwendungen (Search for Journal in Brave)

Hilbert spaces ill-posed problem optimal estimates regularization methods

Equations and inequalities involving linear operators, with vector unknowns (47A50) Numerical solutions to equations with linear operators (65J10) Numerical solutions of ill-posed problems in abstract spaces; regularization (65J20)

Review text

Let \(X\), \(Y\) be infinite-dimensional Hilbert spaces, and \(A \in {\mathcal L} (X,Y)\). The authors consider the ill-posed problem (1) \(Ax = y\). Given noisy data \(y^\delta \in Y\) such that \(|y - y^\delta|\leq \delta\) one needs to approximate an exact solution \(x\) which is assumed to exist and to belong to the correctness class \(M_{p,E} = \{x \in X : x - \overline{x} = (A^* A)^{p\over 2} v,\;|v|\leq E,p > 0\}\) (here \(\overline{x}\) is a suitable initial approximation for (1)). The authors introduce the family of regularization methods \[ x^\delta_\alpha = \overline {x} + g_\alpha ((A^* A)^{s + 1}) (A^* A)^s A^* [y^\delta - A \overline{x}] \] where \(s \geq - {1\over 2}\), \(g_\alpha (\lambda) : [0,a] \to \mathbb{R}\) is a family of continuous functions depending on a regularization parameter \(\alpha > 0\). They consider different functions \(g_\alpha\) and discuss the question how to choose \(s\) and \(\alpha = \alpha(\delta,E,p)\) to obtain optimal estimates \[ \sup_{{x \in M_{p,E}, y^\delta \in Y}\atop {|Ax - y^\delta|\leq \delta}} |x^\delta_\alpha - x|\leq E^{{1\over (p+1)}} \delta^{{p\over (p + 1)}}. \]

Recommendations

Cited in

(44)

This page was built for publication: On the optimality of regularization methods for solving linear ill-posed problems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1340797)