Optimization of approximate solution of integral equations of several variables (Q1862901)

Let \((X, \|\cdot \|)\) be a linear normed space, \(\mathcal F\) a subset of \(X\), and \(\mathcal H\) a set of continuous linear operators \(H\), acting from \(X\mapsto X\) such that \[ z =Hz + f \tag{1} \] is uniquely solvable for any \(H\in\mathcal H\), \(f\in\mathcal F\). A direct method of solving (1) is a rule \(A\), according to which, with an operator \(H\in\mathcal H\) there is associated a subspace \(F_m\subset X\) of dimension \(m\), and an operator \(H_m\) from \(X\) into \(F_m\) such that the equation \[ z_A =H_mz_A +f \] is uniquely solvable with \(z_A\) being taken as the approximate solution of (1). The authors study the problem of optimization of adaptive direct methods for the approximate solution of some classes of Fredholm integral equations of the second kind with anisotropic smooth kernels and free terms belonging to anisotropic Sobolev classes. Adaptivity of a direct method is understood in the sense that the subspace on the basis of which it is constructed is chosen depending on the operator of the concrete equation, which would essentially let one increase the precision. The exact order of the error of adaptive direct methods is determined, and an optimal algorithm is given.