UMFPACK options not exposed to user

[Garth Wells]

Hi @all

I tried to solve the stokes example with the direct solver umfpack.
With the backend PETSc everything is fine, using MTL4 or uBLAS, umfpack reports a singular matrix. I think this happens because of the pivoting strategy of umfpack.

After a short look in the dolfin source file "UmfpackLUSolver.cpp", I think, that dolfin fixes the strategy to zero and it cannot be change using the parameters, could I?

The matrix template library has a direct interface to umfpack in boost/numeric/mtl/interface/umfpack_solve.hpp and allows to set the umfpack strategies from a double.

Should I implement this at my own, or are there plans to implement the strategy selection in dolfin?
Currently I use dolfin version 0.9.10.

Regards,
Andreas Naumann

[Anders Logg (logg)]

Will this be fixed before 1.0? It looks more like a feature request than a bug fix to me. Could it be moved to trunk?

[Garth Wells (garth-wells)]

I had already re-targeted it to 1.1.

Garth

On 15 Nov 2011, at 23:02, Anders Logg <email address hidden> wrote:

> Will this be fixed before 1.0? It looks more like a feature request than
> a bug fix to me. Could it be moved to trunk?
>
> --
> You received this bug notification because you are a member of DOLFIN
> Core Team, which is subscribed to DOLFIN.
> https://bugs.launchpad.net/bugs/806896
>
> Title:
> UMFPACK options not exposed to user
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/dolfin/+bug/806896/+subscriptions

[Anders Logg (logg)]

ok, strange that it showed up in my list then. Maybe you did it just
while I was reading it.

--
Anders

On Tue, Nov 15, 2011 at 11:28:42PM -0000, Garth Wells wrote:
> I had already re-targeted it to 1.1.
>
> Garth
>
> On 15 Nov 2011, at 23:02, Anders Logg <email address hidden> wrote:
>
> > Will this be fixed before 1.0? It looks more like a feature request than
> > a bug fix to me. Could it be moved to trunk?
> >
>