[petsc-dev] DA refine/coarsen with stencil_width=0 for nproc>1

[Jed Brown]

On Mon, Feb 21, 2011 at 21:39, Lisandro Dalcin <dalcinl at gmail.com> wrote:

> Perhaps I'm missing something obvious, but DA refine/coarsen with
> stencil_width=0 for nproc>1 is failing, see the example below
>

Lisandro and I chatted about this earlier. The problem is that getting
interpolation will fail because a coarse basis function always crosses the
process boundary. You can have the locality if you use Q0 interpolation
(maybe that should be set as default for stencil_width=0?), but that will
fail even in serial unless you set periodic (which can be used to emulate
cell-centered).

There is a degenerate case that I think should work in parallel, but does
not: Q0 interpolation and periodic refinement. The constraint equations I
wrote do not have a solution in this case. I'm not sure if it requires
handling as a special case or not.

I put the consistency check in because I didn't like needing to wait for a
confusing error in DMGetInterpolation when trying to coarsen in an
incompatible way. But perhaps it is useful to refine or coarsen with no
intention of having an interpolation operator, in which case the check can
be removed.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110223/19e0e274/attachment.html>