[petsc-dev] How do you get RIchardson?

[Jed Brown]

On Sat, Sep 17, 2011 at 00:58, Matthew Knepley <knepley at gmail.com> wrote:

> Sure you can, just solve with the identity (i.e. short-circuit the solve).
>> See earlier emails in this thread.
>>
>
> Okay, to summarize
>
>   1) You think we should have this class,
>

How do you get this? I'm not convinced it's necessary since it's really the
same defect-correction procedure where you correct by the identity. I don't
see how "skip the solve" is something that needs a full-blown implementation
unless it is taking advantage of that restriction for some other purpose.

>
>   2) you agree that it is a Picard iteration,
>

Yes, everything that doesn't use old state can be naturally phrased as a
fixed point iteration. NGMRES and BFGS are notable exceptions.

While it is a fixd point iteration, It is not the "Picard linearization"
that is commonly described in the literature.


>   3) but you want it called something else
>

If it will stay around (not necessarily bad), then I don't want it to be
called Picard.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110917/47bc4381/attachment.html>