Adaptive eigenvalue computation: Complexity estimates (Q958152)

For a linear symmetric bounded and strongly positive operator in Hilbert space, an eigenvalue problem in the weak sense is considered. By discretizing this problem using finite elements or finite differences, an equivalent finite dimensional problem can be obtained. The authors apply the analysis to the finite dimensional problem, but in the same time an abstract iteration scheme that solves the original infinite dimensional problem is proposed. The authors are focused on developing and analyzing a convergent adaptive algorithm and yield complexity estimates to prove the (asymptotic ) optimality of the scheme. The basic algorithm MINIT, that is used in analysis of the adaptive algorithm, is proposed. This algorithm can be viewed as a preconditioned steepest descent method for minimizing the Rayleigh quotient. In fact, this basic algorithm resembles the preconditioned Richardson iteration. The main adaptive algorithm named MINIEIG, solves iteratively the eigenvalue problem. As the authors show, the results proved in this paper can be applied to solve the Schrödinger equation.