Open halpernf opened 6 years ago
Thanks for your comment, sounds like a good addition!
@halpernf @swarren I wanna know the reason
@pyramidpoint please, create separate issues for your questions. As to errors you are seeing - you should add "store_res_history=1" parameter your solver's config
AMGX ships with many nice sample programs, including simple implementation of Poisson-like equations. Currently, the charge function (e.g. in A x = b, the array b) is set to 1.d0 everywhere.
This is a poor choice, since integrating a constant over larger and large volumes gives you an exponentially growing solution. As a consequence, the sample cases have terrible convergence rates and do not scale as the problem size is increased.
I suggest a better choice: set b = sin( 2 pi x / Lx ) sin( 2 pi y / Ly ) sin( 2 pi z / Lz ), which satisfies either periodic or Dirichlet boundary conditions. This is more representative of the use of a Poisson solver.