zbMATH — the first resource for mathematics

The preconditioned conjugate gradient method on the connection machine. (English) Zbl 0725.65033
This is a report on numerical experiments where more than ten preconditioners were compared in their performance on a quarter of the connection machine (which means \(2^{14}\) processors, connected in hypercube mode). The problem is the solution of the 5-point approximation to the Poisson and the anisotropic Poisson equation in the unit square; by computing time the simple 4-step Jacobi preconditioner turned out to be best. Interestingly, the unpreconditioned conjugate gradient method needed less than twice as much time.
There is a broad discussion of several aspects of the experiments, and a list of computing times measured for basic arithmetical operations (showing rather unexpected time relations, from a sequential point of view).

65F10 Iterative numerical methods for linear systems
65F35 Numerical computation of matrix norms, conditioning, scaling
65N22 Numerical solution of discretized equations for boundary value problems involving PDEs
65Y05 Parallel numerical computation
65N06 Finite difference methods for boundary value problems involving PDEs
35J25 Boundary value problems for second-order elliptic equations
Full Text: DOI