An efficient nonsymmetric Lanczos method on parallel vector computers (Q1200180): Difference between revisions

The purpose is to reduce the number of synchronization points, when executing the nonsymmetric Lanczos algorithm on a shared memory parallel computer. This is effected by doing the same operations as in \(s\)-step look ahead Lanczos every \(s\)-th iteration. Break down can then occur only every \(s\)-th step, and running time gets shorter on an \(s\)-processor parallel Cray2 computer.