[petsc-dev] [petsc-users] Multigrid with defect correction

[Jed Brown]

Matthew Knepley <knepley at gmail.com> writes:
> I think that would be more convincing with a concrete use case
> example.  Right now it sounds more like a additional complication for
> not much extra functionality, 

I think it is less complication.  We nest preconditioners all the time,
sometimes several levels deep, and need various approximations within
FieldSplit and levels of MG.  LSC is a total hack, picking up auxiliary
operators and scaling via PetscObjectCompose.  Of course all this stuff
should ideally be computed as needed, but it's often more efficient to
assemble multiple operators in one traversal of elements.  I'm not
saying we should change this now, or necessarily ever, but I think the
system with exactly one Pmat is awkward.

> but maybe that is because I cannot think of when I would use it.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20170225/dae19429/attachment.sig>