[Dune] RemoteIndices with nonconforming ALUGrid

[alkaemper]

Hi Simon,

the important difference in your case between conforming and
non-conforming refinement strategy for ALUGrid is that for conforming
refinement there are no ghost elements implemented.

So the problem lies somewhere within the ghost elements of ALUGrid. I
think it may be that ghost elements are attached to boundary faces, and
so may be present twice (@Robert: please correct me, if im wrong and
this has been changed).

Consider the grid elements E1,E2,E3 and Ghost element G as follows:

-----------------
|   E1  |  G    |
-----------------
|   E2  |   E3 |
-----------------

In this example the Ghost element G is created twice once from E1 and
once from E3.

Best,
Martin

On 23.07.19 17:04, Simon Praetorius wrote:
> Hi community,
>
> I'm trying to implement and test the ParallelIndexSet and RemoteIndices
> from dune-common with ALUGrid (dim=2/3, elType=simplex/cube,
> refineType=nonconforming!)
>
> Therefore, I build up a parallelIndexSet for Vertex DOFs (identified by
> its global entity ID) and then I create the RemoteIndices with the
> ParallelIndexSet as source and target.
>
> My expectation is the following: When I count the number of non-owner
> DOFs it should be equal to the number of remote DOFs with myAttribute !=
> owner && remoteAttribute == owner, that is, for each local non-owner DOF
> there is exactly one remote owner DOF in the RemoteIndices set. An
> owner-DOF is defined as a DOF on an interior entity or on an
> overlap/border entity for the process with the highest rank.
>
> Is this assumption correct, also for the case that I have nonconforming
> refinement in ALUGrid?
>
> The test passes for YaspGrid, UGGrid, and conforming ALUGrid (simplex
> and cube) for grids created using a StructuredGridFactory. (should be
> conforming by construction)
>
> I have setup a minimal test-project on
>
>> https://gitlab.mn.tu-dresden.de/spraetor/remoteindices_error
> The code is compiled with
>
> 1.) dune release/2.6, GCC 8.2, OpenMPI 3.1.3,
> 2.) dune-git, GCC 7.3, OpenMPI 3.1.1
>
> run with 3 cores, e.g.
>
>> mpirun -np 3 ./build-cmake/src/ParallelIndexSetTest_2 # ERROR
>> mpirun -np 3 ./build-cmake/src/ParallelIndexSetTest_4 # ERROR
>> mpirun -np 3 ./build-cmake/src/ParallelIndexSetTest_5 # NO-ERROR
> Is my assumption wrong, or do I need to consider something special for
> nonconforming ALUGrid?
>
> Best,
> Simon
>
>
> _______________________________________________
> Dune mailing list
> Dune at lists.dune-project.org
> https://lists.dune-project.org/mailman/listinfo/dune
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pEpkey.asc
Type: application/pgp-keys
Size: 2476 bytes
Desc: not available
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20190726/ca064b90/attachment.key>