On Dec 3, 2012, at 12:40 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> Fine with me, but it might be nice if the name could somehow convey that this
> is just a defect-correction iteration and need not be Newton (e.g., it's
> Picard if a Picard linearization is used).
Picard is a weird beasty the way I implemented it. Instead of having a
"SNESType picard" one uses SNESType ls and then provides SNESSetPicard() (or
local version of snessetpicard) instead of SNESSetFunction/Jacobian.
We could change things to have a SNESType Picard and then have the user
call SNESSetFunction() and SNESSet"Jacobian"() but the meaning of function and
Jacobian are not the same as for Newton so that is not great.
We could add a SNESType Picard and then require the user to also call
SNESSetPicard() to provide the Picard function and matrix. Then there is no
confusing of setting a type of Newton but actually doing Picard. Internally the
Picard snes type would behave just as the ls one does now handling Picard as a
special case; in other words SNESType Picard is just an alias for Newton
internally. But user is clearly using picard type. What do you think?
Barry
> OTOH, the name "Newton" is much more recognizable...
>
>
> On Sun, Dec 2, 2012 at 5:51 PM, Matthew Knepley <knepley at gmail.com> wrote:
> On Sun, Dec 2, 2012 at 6:27 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> >
> > Can we rename
> >
> > #define SNESLS "ls"
> > #define SNESTR "tr"
> > #define SNESVIRS "virs"
> > #define SNESVISS "viss"
> >
> > to something like newtonls newtontr vinewtonrs vinewtonss? Suggestions
> > for names?
>
> +1
>
> Matt
>
> > Now that we have many other methods than Newton these names are terrible
> > and confusing.
> >
> > Thanks
> >
> > Barry
> >
>
>
>
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which
> their experiments lead.
> -- Norbert Wiener
>
