Hi James, hi Wolfgang, Not a problem to have a look at this one, as it concerns anisotropic refinement which was basically my work anyway. I'm always glad to hear that someone wants to use the anisotropic refinement feature.
The answer to your problem, James, is actually quite simple: For continuous FEs in 3D of polynomial degree 2 or higher hanging node constraints are not supported for general anisotropic refinement cases in the publicly available deal.II version. You can use DG or you can use 2D or you can use degree 1 elements, but the case above is not implemented. I actually have implemented it for the new version of creating hanging nodes that was introduced in the hp context, at least for nodal finite elements of globally constant degree. However, this is in a local version. Due to some arguments I do not want to go into on this mailing list I do not sync my version with the svn anymore, so there might be quite a few diffs. Unfortunately, I do not have the time to look into this now or in the near future. However, the principal authors of deal.II (Wolfgang, Guido and Ralf) have access to my version and if they want to they can look at my changes and incorporate them into the public library. In case one of the authors wants to tackle this problem: have a look at the changes in revision 6225 of our local version. I'm sorry that I cannot give you a more satisfying answer, but anisotropic refinement is not really a core feature of deal.II and corresponding problems are usually not tackled by anyone but me, and due to the development model of deal.II I'm not an active developer anymore. Best, Tobias -----Ursprüngliche Nachricht----- Von: Wolfgang Bangerth [mailto:[email protected]] Gesendet: Do 13.05.2010 20:17 An: James Avery Cc: [email protected]; Leicht, Tobias Betreff: Re: [deal.II] Measuring aspect ratios of cells for anisotropicrefinement Tobias, I'll be out of the loop for a while. Would you possibly be willing to take a look? Thanks Wolfgang On Thursday 13 May 2010, James Avery wrote: > I'm reposting, since I accidentally only sent this to Wolfgang on the > first try: > > On Wed, May 12, 2010 at 6:00 AM, Wolfgang Bangerth > > <[email protected]> wrote: > >> I'll have to compile deal.II with debug symbols to give more > >> information. I'll also try to reproduce the problem with a small > >> example program -- I'll do both tomorrow. > > > > Go do that. In general, while you develop new code, running in debug > > mode is what you should always do. > > All right, here we go: When running in debug mode, updating hanging > nodes doesn't blow up, but exits with an exception telling me that I'm > not using isotropic refinement. I've made a small test case that > provokes this behaviour. The following happens: > > - When I generate the mesh from scratch using > GridGenerator::hyper_rectangle, nothing bad happens. > - When I load a mesh from a GMSH file, but only do isotropic > refinement, everything is also fine. > - When I load a mesh from file and do anisotropic refinement, I get: > > -------------------------------------------------------- > An error occurred in line <2244> of file > </store/build/deal.II-svn/deal.II/source/dofs/dof_tools.cc> in > function > void > dealii::internal::DoFTools::make_hp_hanging_node_constraints(const DH&, > dealii::ConstraintMatrix&) [with DH = dealii::DoFHandler<3, 3>] The > violated condition was: > > cell->face(face)->refinement_case()==RefinementCase<dim-1>::isotropic_r >efinement The name and call sequence of the exception was: > ExcNotImplemented() > Additional Information: > (none) > > Stacktrace: > ----------- > #0 /store/build/deal.II-svn/lib/libdeal_II_3d.g.so.6.3.pre > [0x1925ef2]: /store/build/deal.II-svn/lib/libdeal_II_3d.g.so.6.3.pre > [0x1925ef2] > #1 ./aspect_test [0x804e7e3]: ./aspect_test [0x804e7e3] > #2 ./aspect_test [0x804da0e]: ./aspect_test [0x804da0e] > #3 /lib/tls/i686/cmov/libc.so.6: __libc_start_main > #4 ./aspect_test [0x804d781]: ./aspect_test [0x804d781] > -------------------------------------------------------- > > The comment immediately preceding the exception in dof_tools.cc states > that, for all faces with children, there must either be no dofs on the > face, or the face must not be anisotropically refined. Why is this > necessary? And can you see any way around this problem? > > I also can't easily see why this works with GridTools-generated meshes > (or perhaps it only worked in this particular case) but not with the > GMSH mesh. > > I've placed the source code for the small example (aspect_test.cc) as > well as the input mesh and outputs at: > > http://diku.dk/~avery/aspect/aspect_test.tar.gz > > Compiling aspect_test.cc with -DLOAD_MESH=1 shows the problem, and > without it shows it working. > > Any ideas? > > Thanks a bunch for the help! > > > -- > Med venlig hilsen, > James Avery <[email protected]> -- ------------------------------------------------------------------------- Wolfgang Bangerth email: [email protected] www: http://www.math.tamu.edu/~bangerth/ _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
