Conjugate gradient and Lanczos methods for sparse matrices on distributed memory multiprocessors (Q1373792)

The paper presents algorithms for iteratively solving sparse systems of linear equations arising by the discretization of differential equations and corresponding eigenproblems on parallel systems with distributed memory. The conjugate gradient and Lanczos methods have been parallelized. The approaches of flexible data distribution, overlapping the communication and computation are applied. The algorithms developed have been tested successfully for large finite element matrices on the Paragon XP/S 10 system.