I actually prefer the model where a geometric object (e.g., a DA) controls the restriction (e.g., DAGlobalToLocalBegin, etc).
On Tue, Dec 7, 2010 at 2:29 PM, Dmitry Karpeyev <dkarpeev at gmail.com> wrote: > Yes, but how is the local form defined? Do you suggest that every Vec > carry a scatter from the global to the local form? > That appears to be a necessarily geometric (maybe in the abstract sense) > notion of restricting to some subdomain, which > might involve degrees of freedom not currently owned locally, and which > become "ghosts". > > > On Tue, Dec 7, 2010 at 2:24 PM, Jed Brown <jed at 59a2.org> wrote: > >> On Tue, Dec 7, 2010 at 21:20, Barry Smith <bsmith at mcs.anl.gov> wrote: >> >>> In other words all parallel Vecs could have related local forms and the >>> model is "give me local form Vec", scatter to/from local form vec etc? >> >> >> The only consequence I can think of is that simultaneously manipulating >> the global and local forms could not be allowed (in a manageable way). But >> I think any need to do that is an obscure edge case. >> >> Jed >> > > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101207/f96959de/attachment.html>
