Galerkin eigenvector approximations (Q4501033)

Eigenvalue problems \(Av=\lambda v\) are considered where the closed linear operator \(A\) is defined via a quadratic form in a Hilbert space setting. The following question is treated. How close is the Galerkin approximation to the best approximation available out of the trial space? Under some mild conditions it is shown that the Galerkin method gives an approximate eigenvector that approaches asymptotically the projection of the exact eigenvector onto the trial subspace -- and this occurs more rapidly when compared with the underlying rate of convergence of the approximate eigenvectors. Both orthogonal-Galerkin and Petrov-Galerkin methods are considered here with special emphasis on nonselfadjoint problems, thus extending earlier studies by \textit{F. Chatelin} [Spectral approximation of linear operators (1983; Zbl 0517.65036)], \textit{I. Babuška} and \textit{J. E. Osborn} [Math. Comput. 52, No. 186, 275-297 (1989; Zbl 0675.65108)] and \textit{A. V. Knyazev} [ibid. 66, No. 210, 985-995 (1997; Zbl 0870.65045)]. Consequences for the numerical treatment of elliptic partial differential equations discretized either with finite element methods or with spectral methods are discussed. New lower bounds for the sep of a pair of operators are established as well.