Hi Ryan,
If your Poisson problem is periodic in all directions, then the
constant vector will likely be in the nullspace of the matrix, i.e.,
you have a singular matrix, A. This would explain very slow
convergence. Essentially multigrid can't reduce error, if that error
e satisfies A*e = 0. To check this, see if
>>> A*numpy.ones((A.shape[0],0))
gives you a vector of essentially zero values. What to do, is to use
the callback parameter offered by
pyamg.krylov.cg() or
smoothed_aggregation_solver.solve(). This parameter lets you apply a
user-defined function to the current solution guess every iteration.
Set callback to be a function that will project out the constant
vector every few iterations. Convergence should happen quickly.
Hope this helps,
Jacob
> --
> You received this message because you are subscribed to the Google Groups
> "pyamg-user" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to
pyamg-user+...@googlegroups.com.
> To post to this group, send email to
pyamg...@googlegroups.com.
> Visit this group at
http://groups.google.com/group/pyamg-user.
> For more options, visit
https://groups.google.com/groups/opt_out.