Empirical convergence algorithm for successive overrelaxation

The algorithm concerns the computing by successive overrelaxation (s.o.r.) of the magnetic vector potential at the nodes of a rectangular grid of graded meshes covering the pole pitch of a homopolar or of an heteropolar electrical machine. It has been applied to several cases with grids having up to 1400 nodes, and has operated practically without oscillations. The algorithm has been developed empirically starting from one demonstrated by F.de la Vallee Poussin and A. Lion for linear media. Computational experience published in two papers by E.A.Erdelyi and E.F.Fuchs concerning the use of the latter for nonlinear media has also been considered. Two changes have been made in the above algorithm. For the two particular problems being solved, these changes have simplified the computing program, and have nevertheless resulted in good convergence. The two types of problems solved with the new algorithm are described, and a synopsis of numerical results is given