Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Hybrid Lanczos-type product methods

Conference ·
OSTI ID:433396
 [1]
  1. Swiss Center for Scientific Computing, Zuerich (Switzerland)

A general framework is proposed to construct hybrid iterative methods for the solution of large nonsymmetric systems of linear equations. This framework is based on Lanczos-type product methods, whose iteration polynomial consists of the Lanczos polynomial multiplied by some other arbitrary, {open_quotes}shadow{close_quotes} polynomial. By using for the shadow polynomial Chebyshev (more general Faber) polynomials or L{sup 2}-optimal polynomials, hybrid (Chebyshev-like) methods are incorporated into Lanczos-type product methods. In addition, to acquire spectral information on the system matrix, which is required for such a choice of shadow polynomials, the Lanczos-process can be employed either directly or in an QMR-like approach. The QMR like approach allows the cheap computation of the roots of the B-orthogonal polynomials and the residual polynomials associated with the QMR iteration. These roots can be used as a good approximation for the spectrum of the system matrix. Different choices for the shadow polynomials and their construction are analyzed. The resulting hybrid methods are compared with standard Lanczos-type product methods, like BiOStab, BiOStab({ell}) and BiOS.

Research Organization:
Front Range Scientific Computations, Inc., Lakewood, CO (United States)
OSTI ID:
433396
Report Number(s):
CONF-9604167--Vol.1; ON: DE96015306
Country of Publication:
United States
Language:
English

Similar Records

Projection preconditioning for Lanczos-type methods
Conference · Mon Dec 30 23:00:00 EST 1996 · OSTI ID:440726

An iterative method for the solution of nonlinear systems using the Faber polynomials for annular sectors
Conference · Fri Dec 30 23:00:00 EST 1994 · OSTI ID:223850

Commentaries of three papers of Cornelius Lanczos
Technical Report · Tue Oct 01 00:00:00 EDT 1991 · OSTI ID:677182