Quadrature based collocation methods for integral equations of the first kind (Q421380)

The paper is concerned with solving integral equations of the first kind with a non-degenerate kernel. To solve such ill-posed problems a stable approximation method based on noisy data is needed. In Section 1 the noisy data is defined and the author states that they are seeking the minimum norm solution. A discretized form of the original equation is used, an assumption about the collocation points is stated and a weighted inner product is considered. This approach was adopted by the author and \textit{S. V. Pereverzev} [J. Complexity 23, No. 4--6, 454--467 (2007; Zbl 1131.65113)] while analyzing a regularized collocation method. In Section 2 a representation for the adjoint of the approximation operator \(K_n\) is obtained, and it is shown that \(\| K^\ast_n K_n-K^\ast K\| \rightarrow 0\) as \(n \rightarrow \infty\) whenever the nodes and weights are associated with a convergent quadrature rule. A Nyström approximation of \(K^\ast\) is used. In Section 3, under stated assumptions, the method of finding \(x_n^\ast\), the minimal-norm solution of the discretized form of the equation, is discussed and its properties are used to show convergence to \(x^\ast\), the minimal-norm solution of the equation. Error estimates are derived, including estimates for the case when \(y\) is replaced by a noisy data. A regularization is then used in Section 4. It is shown that the derived error estimates are of the same order as in the case of Tikhonov regularization when no approximation of the integral operator is involved.