[petsc-dev] How do you get RIchardson?

[Vijay S. Mahadevan]

> Also, it's extremely common to just assemble the Picard linearization and
use it for preconditioning MFFD Newton. Use -snes_mf_operator for Newton,
drop that option to use the Picard linearization.

Jed, this is exactly what I do and have done for more than 3 years with
PETSc. Also calling newton even without the matrix free jacobian with just
Picard linearization with an algebraic preconditioner is sometimes more
efficient. And all of them do fit in the implemented newton framework. It is
both elegant and flexible.

Vijay

On Sep 16, 2011 5:52 PM, "Jed Brown" <jedbrown at mcs.anl.gov> wrote:
>
> On Sat, Sep 17, 2011 at 00:48, Barry Smith <bsmith at mcs.anl.gov> wrote:
>>
>> Should we have support for this in SNES as a particular class since it
comes up fairly often (in certain literature): perhaps for short we could
call it SNESPICARD :-)
>
>
> Haha, but no. I think we should document that this is how to do
Picard-with-solve, but we can't do anything automatically because the user
has to provide the linearization they want to use. Also, it's extremely
common to just assemble the Picard linearization and use it for
preconditioning MFFD Newton. Use -snes_mf_operator for Newton, drop that
option to use the Picard linearization.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110916/8a0edf01/attachment.html>