[petsc-dev] question about Krylov solves, symmetric matrices, and ICC

[Jed Brown]

On Mon, Mar 4, 2013 at 12:43 PM, Paul Mullowney <paulm at txcorp.com> wrote:

> I'm guessing this question has been asked and answered before on this
> list. Apologies for the repeat. Anyway, suppose I have a symmetric matrix
> stored as only L in a Matrix Market file. Then, I read in the L part with
> MatSetValues() and then do:
>
> MatSetOption(A,MAT_SYMMETRIC,**PETSC_TRUE);
>
> Will PETSc then do a KSP solve (with say an ICC preconditioner) on the
> full matrix even though the upper part hasn't been explicitly
> created/inserted via MatSetValues()? The example below suggests not but I
> want to be absolutely sure.
>
> http://www.mcs.anl.gov/petsc/**petsc-current/src/ksp/ksp/**
> examples/tutorials/ex2.c.html<http://www.mcs.anl.gov/petsc/petsc-current/src/ksp/ksp/examples/tutorials/ex2.c.html>
>
> Is there any way to work with just L?
>

Insert it as the upper triangular part in a MATSBAIJ, then everything you
want will work.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130304/cac9f088/attachment.html>