On Thu, 17 Mar 2016, David Knezevic wrote: > I'd be interested to see the code, if you can send it through. In > particular, I'd be interested to see how you're imposing the > boundary conditions. Are you using CondensedEigenSystem? If so, > perhaps there is an issue with how the "indices to condense" are > being communicated in parallel.
There's one obvious bug I can see: If a Dirichlet node is owned by a processor which doesn't own any elements with a Dirichlet side containing that node, then that dof won't be condensed by get_dirichlet_dofs, because the processors that try to add it to the condensed set don't own it and the processor that owns it won't try to add it. IIRC I had to fix a similar bug when putting together the DirichletBoundary support. We really ought to be adding DirichletBoundary support to CondensedEigenSystem and not forcing the user to try to do this themselves. Hmmm... if constrained dofs were treated properly in CondensedEigenSystem, both DirichletBoundary and AMR would work automatically, right? I was also about to say that there's no way this can happen on a cube, but actually it ought to happen semi-frequently on a *Tet* cube, like in the example code. Check the partitioning and my guess should be easy to confirm or disprove in this case. --- Roy ------------------------------------------------------------------------------ Transform Data into Opportunity. Accelerate data analysis in your applications with Intel Data Analytics Acceleration Library. Click to learn more. http://pubads.g.doubleclick.net/gampad/clk?id=278785231&iu=/4140 _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
