Fix essential bc with meshvar #255
Conversation
|
Glad to see you've figured this out - Is this tested in parallel ? |
|
Yes, I tested it in parallel on mac, and everything looks fine. I still need to run tests on NCI with multiple nodes. |
|
Please hold off on merging this PR for now. I’m currently running some tests on the annulus with an internal boundary and checking if the code handles pressure boundary conditions without breaking. |
|
I am not sure that we really can do pressure boundary conditions ? Are you talking about essential boundary conditions on the constraint equation ? Don't the other assumptions we make rule out using any discontinuous functions for the pressure term if we want to have boundary values imposed this way ? Have you checked with @knepley whether the field split machinery is set up to work this way ? |
|
In the spherical benchmarks, applying essential boundary conditions on pressure appears to eliminate the null space in the pressure solution. Since I am still testing these benchmarks, I thought it would be useful to keep the pressure boundary conditions for now, even though we don't fully understand their effects during the solve. I made a small change at the solution separation stage to return all indices in the index set for the pressure field, which allows us to separate the velocity solution by taking the complement of that index set. The current implementation works correctly, whether or not we apply pressure boundary conditions, so things should work as expected. |
|
@gthyagi An alternative for removing the null space due to the pressure is to solve in a multigrid iteration, and on the coarse grid use SVD. This can often work because the smoothers and prolongation to not reintroduce components in the nullspace. You can set the coarse solve with |
|
this is same as #278, which is already merged. |
Here is the fix for using mesh variable in essential boundary condition. #221