New conditions for non-stagnation of minimal residual methods (Q929360)

The authors deal with the problem of conditions guaranteeing that a minimal residual Krylov subspace method makes progress, i.e. that it does not stagnate. In the literature a condition was already proved which establishes that, for a given linear system, the symmetric part of the coefficient matrix has to be positive definite. However this condition provides a bound for convergence which appears too pessimistic. This remarkable paper presents new more general conditions for non-stagnation which involve both the symmetric part of the coefficient matrix and its skew-symmetric part. The effectiveness of these conditions are enlightened by some significant examples.