An element-by-element preconditioned conjugate gradient method implemented on a vector computer (Q1180617)

For large linear algebraic systems with a sparse symmetric positive definite coefficient matrix arising e.g. from the finite element discretization of elliptic boundary value problems the preconditioned conjugate gradient method is known as an efficient iterative solver. Among the various preconditioners, the Crout EBE (element-by-element) preconditioner is quite popular and has already been implemented in available finite element codes. It is further known that this preconditioner is highly vectorizable when using an appropriate colouring of the graph associated with the triangulation. For 2D and 3D elastic and thermal problems the authors compute the speed- up gained by vectorization (on a CRAY-2) and compare the performance of the EBE preconditioner with other standard preconditioners. It is not so surprising that EBE is much faster than diagonal preconditioning but inferior to the preconditioner based on incomplete Cholesky decomposition.