[Dune] Boundary conditions, segments and indexing

[Firmbach, Max]

Maybe a short addition.

In 3D I depend on volume elements like tetraeders and due to

the geometry i need to read the mesh in via the gmsh-reader.

So i can't really check the implementation with other grid-types,

like AlbertaGrid. I might could set up a simple case with a

YaspGrid, but that wouldn't help me in the long term. So the only

option is the UGGrid-implementation.


For the simple one element, hexaeder case the tags are read into

the boundaryEntity vector like [0, 0, 3, 0, 1, 0]. So on four faces

there is just a 0, which means no B.C. is applied. For front and back

either B.C. 1 or 3 should be applied. Looping over the boundary

intersections my insertion index for the segments is something like

[0, 1, 2, 4, 3, 5], which is identical to the boundarySegmentIndex()

method (somewhere it's said that they don't need to be identical

somehow, for this case they are).

________________________________
Von: Dune <dune-bounces at lists.dune-project.org> im Auftrag von Firmbach, Max <max.firmbach at tum.de>
Gesendet: Freitag, 21. August 2020 20:20:43
An: Dedner, Andreas; dune at lists.dune-project.org
Betreff: Re: [Dune] Boundary conditions, segments and indexing


Hello Andreas,


thanks for the quick response.

I followed the "get-started" tutorial, so I'm using the UG-Grid implementation.

Like this, where UG-Grid  depends on the dimension and a LeafGridView

is later generated:


    using Grid = UGGrid<Parameters::dim()>;
    using GridView = Grid::LeafGridView;

So I'm also not sure if either the indexing of the boundary segments might
be wrong in my code, so that the wrong array-entry and thus wrong tag is
chosen for a specific intersection/boundary segment. Or the calculation of
the row, where the entry is placed/modified might be false.

Hope that helps.

Best regards,
Max

________________________________
Von: Dedner, Andreas <A.S.Dedner at warwick.ac.uk>
Gesendet: Freitag, 21. August 2020 19:33:37
An: Firmbach, Max; dune at lists.dune-project.org
Betreff: Re: Boundary conditions, segments and indexing

Hi Max.
Looks alright to me but I might have missed something.
Could you please add which grid manager you are using since it might be a bug in the grid implementation you are using.
Best
Andreas
________________________________
From: Dune <dune-bounces at lists.dune-project.org> on behalf of Firmbach, Max <max.firmbach at tum.de>
Sent: 21 August 2020 17:58
To: dune at lists.dune-project.org <dune at lists.dune-project.org>
Subject: [Dune] Boundary conditions, segments and indexing


Hello DUNE-developers,


I have a question regarding the boundary segment indexing and

connecting tags corresponding to the segments with boundary conditions.


Right now, I load a gmsh-file and obtain a vector with boundary tags.

Those tags should trigger some kind of boundary condition for example

set some values in a block-vector or sparse-matrix.

After reading the book from Mr. Sander and some how-to's I think I

understood the principle concept of intersections and so, but still

something is not working right for me.

My actual code loops over all elements of the gridview and over

the corresponding intersections. Inside the loop i check if

the intersection is in contact with the boundary (because I

want to set boundary conditions). If that is true, a switch

should select the right b.c. code, depending on the tag that is

stored inside the boundaryEntitity vector I obtain from loading the

gmsh-file. The array-entry of the boundaryEntitiy vector is

selected via the intersection boundary segment index (is this the right

approach ?).


And here starts my actual question ... in two dimensions that is working

fine (even for complex geometries and several tags), but for three dimensions

it's not working properly even for a single element (for example a hexa-element

with different tags on each of the boundary faces).

So is my description the common approach to handle the boundary situation

or is there a more efficient or better way ? Or am I missing something ?


My code looks like this (the 3 lines after the double loop and the

code inside case 1 should set the corresponding entry in a block-

vector, by looping over the vertices of the intersection geometry and

getting the "global" index for that position):


for( const auto& element : elements(gridView)) {
    for( const auto& isect : intersections(gridView, element)) {


      GeometryType elementGeometry = element.type();
      using Factory = ReferenceElements<double, dim>;
      const auto &ref = Factory::general(elementGeometry);


      if(isect.boundary()) {

        switch(boundaryEntity[isect.boundarySegmentIndex()]) {
          case 0:
            ...
            break;
          case 1:

            for(int i=0; i<ref.size(isect.indexInInside(), 1, dim); i++) {
              int row = indexSet.subIndex(element, ref.subEntity(isect.indexInInside(), 1, i, dim), dim);
              loadVector[row] = Load;
            }

           ...
        }
      }
    }
  }

As written before, for 2d meshes it works pretty well, but for 3d it's the complete
opposite and I don't really understand why.

Thanks for your help and best regards,
Max

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20200824/30b7a73c/attachment.htm>