Some results about GMRES in the singular case (Q1969411)

The author presents new results concerning the use of the generalized minimal residual (GMRES) method for solving linear systems of equations with singular coefficient matrices. He deals with the QR-factorization of the Hessenberg matrix which appears in GMRES and introduces some new results about the rotations of this factorization and their use when the coefficient matrix is singular or nonsingular. The breakdowns occurring in GMRES are discussed and some connections with the necessary and sufficient conditions to obtain a Krylov subspace solution are made. Thanks to some results which may be applied in exact or in finite precision arithmetic the convergence properties of both GMRES and restarted GMRES are analyzed. Results from the numerical experiments when these algorithms are applied to some singular systems occurring in several practical situations are discussed.