A parallel preconditioned block conjugate gradient method for solving large systems of linear equations on a MIMD supercomputer (Q805145)

To solve a large symmetric positive-definite system of N linear algebraic equations using the MIMD parallel multiprocessing environment, the authors propose a method being a combination of: (a) the incomplete Cholesky factorization (used in the ``preconditioning step'' to transform the original system to one having the matrix close to the identity matrix), (b) an appropriate block partitioning of the modified system to enable solving it on a MIMD computer, and (c) the conjugate gradient method to solve the block-partitioning system. The presented algorithm gives a good, fast convergence and achieves the speedup \(2S/3+O(S/N)\) and efficiency \(>2/3+O(1/N)\) (i.e. 66\% asymptotically) when S processors \((S<N)\) are used.