[For part I see SIAM J. Numer. Anal. 37, 1909-1929 (2000; Zbl 0961.65112).]
The approximate inverse is a numerical scheme for solving operator equations of the first kind in Hilbert spaces. In this paper, the authors further develop its analytic foundation, relying on their previous results. This general setting allow them to investigate the discrete version of the approximation inverse which actually under lies numerical computations. The convergence is shown where the discretization parameter tends to zero. The authors also prove stability by showing the regularization property. Finally they apply the results to the filtered back projection algorithm in 2D-tomography to obtain convergence rates. One of the theorems is given below:
Theorem. Let \(\alpha>\frac 12\) and let \(f\) be in \(H_0^{\alpha}(\Omega)\) with supp \(f\Subset\Omega\). The radially symmetric mollifier \(e\) is assumed to be in \(H_0^{\alpha+1}(\Omega)\). Let \(\tilde d=\tilde d(f)\) be the smallest positive integer such that supp \(f\) is contained in \(B_{1-\frac 1{\tilde d}}(0)\), the ball about the origin with radius \(1-\frac 1{\tilde d}\). If \(d\geq \tilde d\), then \[ ||\tilde R_{nl,d}^{(l)} \psi_{q,p}^{(l)} Rf||_{L^2(\Omega)}\lesssim(d^{-\min\{2,\alpha\}}+h^{\min\{\alpha,\alpha+\frac 12\}}d^{\alpha+2})||f||_{H^{\alpha}(\Omega)}. \]