On 2014-01-29 12:11, Patrick Farrell wrote:
On 28/01/14 19:49, Garth N. Wells wrote:
A plus of this approach is that if your system is symmetric you can
use CG or
MINRES, which you can't with NonlinearDiscreteProblem.
I don't really understand your comment. Perhaps I've misunderstood.
You can use
CG or MINRES perfectly well without subclassing NonlinearProblem
directly. Try
it out in the nonlinear-poisson demo. That's using
NonlinearDiscreteProblem
under the hood. Am I missing something?
NonlinearDiscreteProblem breaks symmetry when applying bcs.
I don't have an objection to it being spun out, but it is of limited
practical
use.
I think it would be quite useful, as it enables people whose codes
currently use
the solve(F == 0) interface to easily set up custom SNES solvers
without copying
and pasting the same code everywhere.
If a user wants lower-level control, we already provide a lower level
interface in the form of NonlinearProblem.
If you're pasting the same code everywhere, you need to look at how you
write your code.
On 28/01/14 21:13, Anders Logg wrote:
So in your case, either pick (1) or (2). I don't see that there should
be an
interface in between.
I wasn't suggesting to create a new interface in between. I was
suggesting that
the way (3) is turned into (1) should be accessible to users, so that
codes that
currently use (3) can easily use (1) themselves.
You are asking for a new public interface which I don't think is
necessary but does add to the public interface that needs to be
maintained.
Garth
Cheerio,
Patrick
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
