Adaptive polynomial preconditioning for Hermitian indefinite linear systems (Q1263933)

The paper deals with the solution of Hermitian indefinite linear systems in the context of conjugate gradient methods combined with the various polynomial preconditioners. An adaptive procedure is devised for estimating dynamically the extreme eigenvalues of a given indefinite matrix that is equivalent to computing dynamically the optimum polynomial preconditioner. Finally, the efficiency of polynomial preconditioning is demonstrated by means of some numerical experiments executed on a single processor of a Cray X-MP/48. A better performance on parallel architectures is expected.