swMATH ID: 11516
Software Authors: Brezinski, C.; Redivo Zaglia, M.; Sadok, H.
Description: Avoiding breakdown and near-breakdown in Lanczos type algorithms. The paper deals with methods which the authors have developed to avoid breakdown and near-breakdown due to division by a scalar product whose value is zero or is different from zero but small in Lanczos type algorithms for solving linear systems.par In particular, the bulk of the paper concentrates on a method called by the authors the method of recursive zoom (MRZ) and its variants: SMRZ, BMRZ, GMRZ and BSMRZ, where S, B, G stand for symmetric, balancing and general, respectively. It is shown that a breakdown can be avoided by considering only the existing orthogonal polynomials in the Lanczos type algorithms. The methods described in the paper are able to detect if such a polynomial does not exist in order to jump over it.par Pseudo-codes for MRZ and BSMRZ are given and some numerical results are presented. (netlib numeralg na1)
Homepage: http://www.netlib.org/numeralgo/index.html
Keywords: Lanczos method; biconjugate gradient; projection; orthogonal polynomials; methods of moments; method of recursive zoom
Related Software: CGS; na5; BiCGstab; Harwell-Boeing sparse matrix collection; Regularization tools; GpBiCg; QMRPACK; LSQR; LAPACK; blgaussexp; mftoolbox; ABLE; OPQ; Eigtool; mctoolbox; JDQZ; JDQR; eigs; CMRH; LINPACK
Cited in: 55 Publications

Citations by Year