A parallel balance scheme for banded linear systems

From MaRDI portal





A new algorithm is proposed for the solution of a non-symmetric banded linear system that is sparse within the band. The system matrix is partitioned into blocks of rows in a such way that there is a small number of unknowns common to multiple blocks. The balance scheme algorithm consists of two main steps. In the first step, the underdetermined system is solved while in the second step the solution of a small reduced system is obtained. The reduced system with matrix \(M\) can be solved using either direct or iterative method. The upper bound of the condition number of \(M\) is derived and storage requirements are estimated. Another variant is proposed in which the matrix \(M\) is not expressed explicitly. The use of QR factorization to solve the underdetermined system is replaced by projection-based approach. The parallel implementation is examined and the results of experiments for narrow banded non-symmetric systems are presented. The experiments shows that the balance scheme is a competitive alternative to preconditioned Krylov subspace methods, especially for matrices with strongly indefinite symmetric part.





Describes a project that uses

Uses Software






This page was built for publication: A parallel balance scheme for banded linear systems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4828881)