On Dec 17, 2012, at 4:59 PM, Stefano Zampini <stefano.zampini at gmail.com> 
wrote:

> I need a 1-coloring algorithm to implement a Deluxe version of BDDC, where 
> the partition of unity is realized by solving coupled subdomain problems. I 
> completely understand Barry's point about adding a different kind of coloring 
> to a library devoted to Nonlinear solvers, but using MatColoring only for 
> Jacobian computations is at least ambiguous; MatColoringType could carry 
> information about the type of coloring (distance-1, distance-2 star, acyclic 
> etc), MatColoringOrderingorWhatever contain the kind of ordering (ID, SL, LF 
> ..)

   We use the term Type in PETSc to represent subclasses (different 
implementations to achieve the same task)*, thus currently MatColoringType is 
sl, sf etc.  It would be confusing to change it to mean something else so we 
need another term. Maybe MatColoringProblemType() ? Other suggestions?

  And I totally agree we should have these other colorings available to PETSc. 

   Barry

* One could argue that the current model of XXXSetType() is stupid and we 
should change it to XXXSetSubClasse() or something, but I doubt that will 
happen :-)

> 
> 
> 2012/12/17 Barry Smith <bsmith at mcs.anl.gov>
> 
> On Dec 17, 2012, at 2:56 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> 
> > Stefano wants it to solve face-centered subdomain problems. The same 
> > structure also shows up in local processing for multiscale finite elements.
> 
>    Like in a multiplicative Schwarz kind of way? That is fine but completely 
> different than what 2 colorings are for.
> 
>    Barry
> 
> >
> >
> > On Mon, Dec 17, 2012 at 1:16 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> >
> > On Dec 17, 2012, at 1:49 PM, Jed Brown <issues-reply at bitbucket.org> 
> > wrote:
> >
> >    What is the 1 coloring used for? The 2-coloring is for computing 
> > Jacobians, not for other stuff. It is not clear to me that both colorings 
> > should be hidden behind the same method names, calling sequence etc if they 
> > are used for very different purposes.  So think before blindly implementing 
> > a 1 coloring algorithm onto of what is already there.
> >
> >     Barry
> >
> > > --- you can reply above this line ---
> > >
> > > New issue 25: Add support for 1-coloring instead of only 2-coloring
> > > https://bitbucket.org/petsc/petsc-dev/issue/25/add-support-for-1-coloring-instead-of-only
> > >
> > > Jed Brown:
> > >
> > > `MatGetColoring()` computes a coloring of `A^T A`, which is a 2-coloring 
> > > of the graph `A`. Support for a 1-coloring would also be useful.
> > >
> > >
> > > --
> > >
> > > This is an issue notification from bitbucket.org. You are receiving
> > > this either because you are the owner of the issue, or you are
> > > following the issue.
> > >
> >
> >
> >
> 
> 
> 
> 
> -- 
> Stefano

Reply via email to