[dune-pdelab] Parallel non-overlapping transient DG method on UG

Lukas Riedel mail at lukasriedel.com
Tue May 9 00:28:23 CEST 2017


Hi guys,

Sorry to bother you again :-) I’m trying to set up my transient DG solver in parallel on both UG and YaspGrid. The latter works out well. I create an OverlappingEntitySet, and a QkDGFEM with P0ParallelConstraints, and I use the OVLP_AMG_4_DG linear solver.

Now I want to find a setup for the unstructured UG grid with ghosts. I use grids with simplices or rectangles exclusively, which means that I either use a PkFEM or a QkDGFEM. I try to follow the Changelog instructions of PDELab v2.4, and the non-overlapping solver examples provided in the ’dune/pdelab/test’ directory, but it doesn’t work out.

I want to use the NOVLP_BCGS_AMG_SSOR solver. As already noted in [MR !251](https://gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/251 <https://gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/251>), the OneStepGridOperator does not supply the required makeConsistent() function. So I plug in my first GridOperator instead.

In the Changelog, it’s stated that ‘Nonoverlapping computations should now use `ConformingDirichletConstraints` instead of `NonoverlappingConformingDirichletConstraints`, as there are no ghost DOFs that need to be constrained anymore.’.
Is this also true for a DG setup? The PDELab non-overlapping tests all assemble a GFS with QkFEM and no constraints. So does this FEM handle restricting the DOFs at processor boundaries itself?

I use the NonOverlappingEntitySet, as that's explicitly necessary for non-overlapping computations.

PkFEM works fine when run on one process.
However, the linear solver throws an error if run on more than one process. The backtrace is bt_simplex_novlp_ghostcc.txt.

QkDGFEM throws the same segfault if either executed sequentially or in parallel. The backtrace is bt_cube_novlp_ghostcc.txt.

Both errors are independent of whether I use NoConstraints, P0ParallelGhostConstraints or P0ParallelConstraints.

Can you help me out on this one?
Thanks and cheers,
Lukas







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20170509/8ee9dda0/attachment.htm>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: bt_simplex_novlp_ghostcc.txt
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20170509/8ee9dda0/attachment.txt>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20170509/8ee9dda0/attachment-0001.htm>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: bt_cube_novlp_ghostcc.txt
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20170509/8ee9dda0/attachment-0001.txt>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20170509/8ee9dda0/attachment-0002.htm>


More information about the dune-pdelab mailing list