dolfin team mailing list archive

[kent-and@xxxxxxxxx]

Hi,

Here are my rough thoughts on it.

In FunctionSpace there is a function restriction which creates a
(restricted) DofMap based
on a mesh, a ufc:.dofmap and a MeshFunction. The way I have implemented,
this function is such that the MeshFunction contains booleans that
indicates which cells
that should be used in the DofMap. It could equally well be vertices or other
mesh entities, but the way restriction is implemented now, it has to be
cells.
It should however be relatively easy to change it to other entities (I
think).
Changing this would give a loc2glob array which contains -1 for the entries
that should not be used. As I understand, petsc will then avoid these entries
so it might work, at least for petsc, with only this alteration in the
function restriction.

Kent

> Hi
>
> Since I last rasied this issue there have been a lot of changes (great
> ones
> I must add) to the fenics codebase, so I thought I would ask again.
>
> I am interested in removing the rows and columns of the finite element
> matrices (for vector electromagnetic problems) associated with zero valued
> Dirichlet boundary conditions.  It is important that the original degrees
> of
> freedom (or a mapping between the two sets) be available so that the
> solution of a system of equations or an eigenvalue problem can be used to
> visualize the solution, for example.
>
> The removal of the degrees of freedom has two advantages that I can think
> of
> (for my applications at any rate).  The first is that the systems being
> solved will be smaller.  The second is of particular concern to me as I am
> often solving eigenvalue problems and the unity eigenvalues that are
> introduce when applying the Dirichlet BCs in their current form
> contaminate
> the spectrum in many of my examples.
>
> Any thoughts or assistance would be greatly appreciated.
>
> Thanks
> Evan
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev
>