Torquil, I ran your example with Roy's PR, and it worked fine for me for 1 up to 8 MPI processes. Note that there are some issues with your example:
- You need to use CondensedEigenSystem, not EigenSystem. You used CondensedEigenSystem in the first version of your example, I'm not sure why you switched that? - You need to call eigen_system.initialize_condensed_dofs(); after equation_systems.init(), as mentioned in my previous email. After making those changes, it worked fine for me. I have merged Roy's PR now. Let me know if you have any further issues with this. Thanks, David On Fri, Mar 18, 2016 at 10:55 AM, David Knezevic <david.kneze...@akselos.com > wrote: > On Fri, Mar 18, 2016 at 10:41 AM, Torquil Macdonald Sørensen < > torq...@gmail.com> wrote: > >> On 17/03/16 21:48, Roy Stogner wrote: >> > >> > On Thu, 17 Mar 2016, David Knezevic wrote: >> > >> >> Torquil: Once this change is in, it'd be good if you can modify your >> >> example to use DirichletBoundary and re-run it to check if it fixes >> >> your problem. >> > >> > If you could try that out *before* the change is in, that would be >> > helpful. >> > >> > https://github.com/libMesh/libmesh/pull/872 >> > >> > I don't know if there's any MOOSE test coverage for >> > CondensedEigenSystem and libMesh only has the one example IIRC. >> > --- >> > Roy >> >> Hi! >> >> After merging the change, and changing to use DirichletBoundary, the >> results seem to be independent of the number of MPI nodes, as one would >> expect. >> >> The output of my program is now: >> >> 1 1 1 1 1 1 1 1 1 1 1 1 1 32.9925 73.0811 73.0811 73.0811 120.25 120.25 >> 120.25 175.419 175.731 >> >> where the unit values seem to be spurious eigenvalues. Is there a >> convenient way in libmesh for projecting out those parts of the vector >> space? >> >> The program is available here: >> >> >> https://drive.google.com/file/d/0B5vWRGzG5kNlbHhWN1M3WHVDeU0/view?usp=sharing >> >> Best regards, >> Torquil Macdonald Sørensen >> > > > It sounds like the Dirichlet dofs aren't getting condensed out, since they > produce these spurious eigenvalues. You need to > call initialize_condensed_dofs() (no argument required) after > equation_systems.init(). > > David > > > ------------------------------------------------------------------------------ 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