dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Aug 13, 2008 at 09:20:17PM +0000, Jed Brown wrote:
> On Wed 2008-08-13 18:31, kent-and@xxxxxxxxx wrote:
> > 
> > > Kent-Andre Mardal wrote:
> > >> BTW: the Assembler code that computes A, b and enforces BC
> > >> simultaneously in a symmetric way,
> > >
> > > Very nice. We can now implement some symmetric linear solvers.
> > >
> > > Do you keep the 'Dirichlet' dofs and zero the row and column, or do you
> > > eliminate them from the global system?
> > 
> > The Dirichlet dofs are kept. I thought that was simplest to do..
> 
> It is certainly more work at this stage, but there are significant
> benefits to actually removing the degrees of freedom.  In particular, it
> allows more preconditioners to be effective and makes related matrices
> (i.e. the Schur complement) better conditioned.
> 
> Jed

Do you have any numbers/references to support this? I was under the
impression it didn't matter that much.

It would be a complication to remove dofs (and then put them back in).
There may be an elegant solution, but it would break the simple

  A = assemble(a, mesh)
  b = assemble(L, mesh)
  solve(A, u.vector(), b)

-- 
Anders

Attachment: signature.asc
Description: Digital signature