Rob Falgout

The only choice you have is to use a 7-point stencil everywhere. So, outside the domain you need to set a 1 on the diagonal and zeros everywhere else as...

If your problem is isotropic, then semi-coarsening is not needed. Both solvers use semi-coarsening regardless. My guess is that the issue with PFMG is interpolation along the stair-stepped boundary. I...

Hi Olivier. I have not had a chance to look into this yet. Regarding your question, can you elaborate? I don't follow what you mean by "periodicity of 1" or...

Okay. There is nothing wrong with this approach, but you don't need to use periodicity here. Just make sure that the size of the grid in the unused dimensions is...

Hi Olivier, I took a look at the matrix/rhs you sent, and I am not having a problem getting (fast) convergence for PFMG either standalone or as a preconditioner for...

Hi Olivier, What do you mean by "re-setting it up at each new time step"? Are you calling `PFMGSetup` each time or just changing the system at each time step...

The process you describe is supposed to be okay, but we don't really test it. Also, it would cost almost nothing to destroy the matrix and solver and rebuild them...

If the matrix doesn't change, there is no need to destroy the solver or call PFMGSetup. Just update the rhs vector and call PFMGSolve with the new rhs.

I'll take a look at your new problem a little later today. Thanks!

You don't need to re-call Assemble, just SetValues. BTW, you were using the interface as it was originally designed to be used, but since we are not testing the multiple-solve...