On Sat, 5 Jun 2010, Wolfgang Bangerth wrote:
It still might be related to the hanging nodes, because the same
method is used to read the mesh in whether your use dg or normal. I
believe that the method used to read the mesh does not support hanging
nodes. I.e. if you have a cell refined further than its neighbors it
will have a node on the face of its neighbors and I don't think this
is supported when the mesh is read.

Please EMail me some of the problem mesh files.
I have been trying to develop code to handle complex boundaries.
This includes more than one "hole" and holes that may have
different properties or materials, including anisotropic.
[email protected]



Indeed, and the GridIn class in fact says so:
.............................
Since the coarse mesh fed into a @p Triangulation object cannot have
hanging nodes, strange things will happen if the input file has them. This
is due to the fact that most mesh description formats do not store
neighborship information between cells, so the grid reading functions have
to regenerate it. They do so by checking whether two cells have a common
face. If there are hanging nodes in a triangulation, adjacent cells have no
common (complete) face, so the grid reader concludes that the adjacent
cells have no neighbors along these faces and must therefore be at the
boundary. In effect, an internal crack of the domain is introduced this
way. Since such cases are very hard to detect, the library does not make
any attempt to abort on such situations, and you will get a triangulation
that probably does not do what you want.
.............................

This is a frequently asked question, so if anyone has suggestions on how to
improve this text I'd be happy to change it!
Best
W.

The text is fine.
We may be able to develop the code to extract the intended
geometry in a usable form. Fallback would be to ask the user for
specific additional information.
Jon Squire   [email protected]

_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to