Boundary conditions besides Dirichlet BC for Multigrid

[yy.wayne]

Hi guys, I have a question on setting the boundary constraints for mg_constrained_dofs object.

In many problems where the only boundary conditions(for AffineConstraints) are Dirichlet BC and hanging nodes, mg_constrained knows them by MGConstrainedDoFs::make_zero_boundary_constraints(the Dirichlet boundary id, not necessarily zero) and MGConstrainedDoFs::initialize(dof_handler)(the hanging nodes), respectively.

However, in case other BC like compute_normal_flux_constraints or compute_no_normal_flux_constraints exists, mg_constrained_dofs has no way to include it. Therefore the MGLevelObject mg_matrices is intrinsically different to global system_matrix. Then mg_matrices[0] cannot be a good approximation of coarsest grid. 

Is there some hint for such cases?

[yy.wayne]

An alternative choice is applying the boundary conditions weakly(during assemble). Will weakly implemented boundary condition behaves worse than explicitly through AffineConstraint class, when it's not Dirichlet BC?

[Peter Munch]

Hi Wayne,

you can attache your own constraints as shown here: https://github.com/dealii/dealii/blob/2376c62a4b89953b74801852983eb8556930d54d/tests/numerics/no_flux_18.cc#L1136-L1146

Hope this helps!

Peter

[yy.wayne]

Wow that's wonderful.

Thank you Peter. You save me all the time.