Parameter choice by discrepancy principles for the approximate solution of ill-posed problems (Q762741): Difference between revisions

Let \(T: H_ 1\to H_ 2\) be a continuous linear operator between Hilbert spaces. If Range T is not closed, then an equation \(Tx=y\) is said to be ill-posed, and small perturbations in the right hand side y may cause large errors in the solution. For the approximate solution of such an equation Tikhonov regularization is widely used, i.e. the solution of the well-posed equation \((T^*T+\alpha I)\times (\alpha,\delta)=T^*y(\delta),\) where \(\alpha\) is a positive real parameter and y(\(\delta)\) is a perturbation of \(y\in Range T\) with \(\| y-y(\delta)\| \leq \delta.\) For a properly chosen dependence of the regularization parameter \(\alpha\) and the error level \(\delta\) the approximate solution x(\(\alpha\),\(\delta)\) converges to the exact solution of \(Tx=y\). The choice of the regularization parameter \(\alpha\) for a given error level \(\delta\) is done by ''discrepancy principles'': Let \(r(\alpha,\delta)=y(\delta)-Tx(\alpha,\delta).\) Arcangeli \((p=1\), \(q=0.5)\) and Morozov \((p=1\), \(q=0)\) have shown that for each \(\delta >0\) there is a unique \(\alpha >0\), such that \(\| r(\alpha,\delta)\| =\delta^ p\alpha^{-q}\) and that for this choice x(\(\alpha\),\(\delta)\) converges to the solution of \(Tx=y\). In this paper I have shown, that for each \(q>0\) there is a unique \(p>0\) such that for each \(\delta >0\) there is an \(\alpha >0\) with \(\| r(\alpha,\delta)\| =\delta^ p\alpha^{-q}\) and x(\(\alpha\),\(\delta)\) converges to the solution of \(Tx=y\) and that the rate of convergence is nearly optimal if the solution is sufficiently smooth. In a recent paper of \textit{H. Engl} ''On the choice of the regularization parameter for iterated Tikhonov regularization of ill-posed problems'', J. Approximation Theory, to appear, it is shown, that the equation \(\| T^*r(\alpha,\delta)\| =\delta^ p\alpha^{-q}\) leads to optimal rates of convergences.