Overcorrection strategy for faster solving of Laplace

[brettviren]

There is a trick of over correcting to speed up FDM solving.

Selecting the correction factor is itself a (meta) optimization problem. An "under-over-correction" or an "over-over-correction" can be slower than the nominal unity factor and the best factor is somewhere in between. The linked blog finds an optimum over-correction factor or 1.94 but cautions that this optimum is a function of the boundary conditions, grid spacing (and presumably also the dimension of the domain).

The speed up can be substantial so even a guess which is not too far from optimal would be worth taking.

pochoir should allow a configurable correction factor, with a default of 1.0. A stretch-solution will provide support for performing the meta-optimization.