Issue with corner value using adaptive mesh

[David Montiel Taboada]

Hello, 

I am working on PRISMS-PF (which is deal.II based) and trying to find an issue with a corner value not being properly set. 

The problem happens when the code is finding the solution of a poisson type equation iteratively, e.g.

\nabla^2 \phi = f(\phi)

I set up a 2D system, and I am using adaptive mesh. We set the initial guess of the field \phi to a constant value everywhere. After the solution is found, the x=y=0 node becomes zero (this is not the correct solution), whereas the rest of the system looks fine (sea image below).

This only happens when we use adaptive mesh (not uniform). And we have noticed that it happens if we use Neumann boundary conditions on both sides of the corner, or a combination of Neumann and periodic BCs (we have not tried other combinations).

Does anyone know what may be causing this problem?

Thank you!

David

[Bruno Turcksin]

David,

Hard to tell what's the problem here. Might be a problem with the AffineConstraints. Are they correct? Do you distribute them when you should?

Best,

Bruno