I think that it would be quite appropriate for VI to politely suggest that one uses the unconstrained Newton solver by error message when it can't find constraints explicitly put on it. Having VI try to slorp the bounds of the DM after failing in the solver is just another level of confusing indirection for the user. The first step we discussed with Jason with respect to PETSc's assimilation of TAO was the addition of a constraint object. Should we see how that goes before fixing behavior?
- Peter On Fri, Oct 4, 2013 at 3:46 PM, Jed Brown <[email protected]> wrote: > Barry Smith <[email protected]> writes: > > > On Oct 4, 2013, at 8:51 AM, Jan Blechta <[email protected]> > wrote: > > > >> On Fri, 04 Oct 2013 08:01:47 -0500 > >>> \ > >> > >> I see. I'm currently quite confused about a terminology. Is > >> the semi-smooth Newton method merely the Newton method (with linesearch > >> bussiness) when used on unconstrained problem? > > > > Yes > > Do we want the VI solvers to automatically fall back to solving an > unconstrained when no bounds are set? This seems error-prone since it's > more confusing to find out that your bounds are somehow being ignored > than to get an error that none were set. We could provide a more > instructive error message. >
_______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
