[petsc-dev] PCFieldSplitSetFields() API change

[Dmitry Karpeev]

On Thu, Jun 7, 2012 at 6:24 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> On Thu, Jun 7, 2012 at 6:17 PM, Dmitry Karpeev <karpeev at mcs.anl.gov>wrote:
>
>> DMCreateFieldDecomposition(DM dm, PetscInt *numsplits, char **splitnames,
>> IS *rowislist, IS *colislist, DM *dmlist)
>
>
> How can this be right? Shouldn't I be getting arrays back (rather than
> having to create my own)?
>
Sorry, I missed an extra '*' on all the arguments but the first.

>
> Does the PCFieldSplit have any say over the way this decomposition is
> pivoted?
>
Well, the above defines the "pivots" -- in the linear case the submatrices
that will be solved with as part of the PCFIELDSPLIT application.  I don't
see any way around this.  The extra flexibility can be achieved by having
DMs encapsulating different field decompositions, which by definitions are
lists of named "pivots" (with optional DMs attached to each pivot).  In the
case of stride-based fields we already have that functionality.

>
> How is one name enough? Take Euler with a residual expressed in
> conservative variables, but state expressed in primitive variables. How
> would you name the splits?
>
Yeah, I missed the extra '*'.

Dmitry.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120607/f1f215c3/attachment.html>