Solving large elliptic difference equations on CYBER 205 (Q1096359)

Presented is a numerical algorithm for the solution of systems of linear equations arising from high-order finite difference methods. An iterative procedure based on the method of conjugate gradient is applied. Preconditioning techniques based on incomplete LU-factorization and matrix polynomial expansion are considered. The idea of an approximate polynomial preconditioning is introduced for equations resulting from high order discretization methods. The major advantage of this approach is that a matrix consisting of far fewer non-zero elements than the original one can be used as an effective preconditioning operator. Consequently, the computational work in solving the preconditioner system, which is needed in each cg-iteration, is substantially reduced. The iterative procedure is suitable for implementation on vector processors such as the CYBER 205.