A Comparison of Three Basic Conjugate Direction Methods

From MaRDI portal





The authors focus on the system of linear equations \(Ax=b\), where \(A\) is a large sparse real non-symmetric non-singular matrix and \(b\) is a real vector. To solve the system, three particular realizations of the block conjugate gradient algorithm of \textit{D. P. O'Leary} [Linear Algebra Appl. 29, 293-322 (1980; Zbl 0426.65011)] are considered. Namely the biconjugate gradient (bcg) method, the Galerkin algorithm of Hegedüs (Heg) and the biconjugate residual (bcr) method [\textit{C. G. Broyden}, Comput. Math. Appl. 31, No. 4-5, 7-17 (1996; Zbl 0874.65024)]. The respective algorithms are presented. Some of the linear systems used in tests are taken from the Harwell-Boeing collection. The others are artificially generated. The typical behaviour of the above-mentioned algorithms is illustrated by tables and graphs. Results are discussed and conclusions are made. In brief, the bcg method is highly erratic, generally unreliable though it can surprise by a quick convergence (e.g. for diagonally dominated matrices), the bcr method exhibits monotone but slow convergence, the performance of Heg falls roughly between those of bcg and bcr.











This page was built for publication: A Comparison of Three Basic Conjugate Direction Methods

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4220443)