ffc team mailing list archive

[Jed Brown <jed@xxxxxxxx>]

Andy Ray Terrel wrote:
> Wouldn't that ignore the coupling in solving Stokes to begin with?
> 
> You want to be able to add an integral operator perhaps something like
> the mass matrix (m) with a constant function m(1, p). Added to the
> problem.

I suppose you could do this, but it adds a lot of nonzeros (a dense row
and column) and it's really not necessary.  This leaves two choices:

1. "Pin" pressure at one node and "normalize" (peculiar word choice)
the result.

2. Inform the Krylov solver that constant pressure is in the null space
of the operator.  See KSPSetNullSpace in PETSc, Trilinos must support
this as well.  This option is usually preferable unless you need to use
a direct solver.

Jed

Attachment: signature.asc
Description: OpenPGP digital signature