An efficient discretization scheme for solving ill-posed problems (Q819641)

For the stable numerical solution of linear ill-posed operator equations \(Ax=y\) with some bounded linear forward operator \(A: X \to X\) mapping in the Hilbert space \(X\) a finite-dimensional approximation scheme of Tikhonov's regularization method with some new aspects is dicussed. The author uses orthogonal projections and refers with respect to some related ideas to a paper by \textit{P. Maaß, S. V. Peraveszev, R.Ramlaw} and \textit{Soladky} [Numer.~Math.~87, 485--502 (2001; Zbl 0974.65055)]. First an analysis of convergence rates is presented for an appropriate a priori choice of the regularization parameter in combination with a corresponding choice of the discretization level. In this context, smoothness properties \(S_1\) and \(S_2\) for the forward operator \(A\) are assumed which are partially related to closed linear differentiation operator and associated Hilbert scales. As an example the linear Fredholm integral operator in \(L^2(0,1)\) and the orthonormal basis of Haar wavelets are considered with focus on \(H^1\)-type smoothness assumptions. Unfortunately, the role of assumptions \(S_1\) and \(S_2\) is not verified explicitly by the author for that example. As second the efficiency of the discretization scheme is proposed by some complexity studies. Numerical experiments complete the paper.