dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, Feb 27, 2006 at 05:27:44PM +0100, Garth N. Wells wrote:
> On Mon, 2006-02-27 at 08:32 -0600, Anders Logg wrote:
> > Garth, I think the nonlinear solver and the nonlinear PDE class look
> > very good now.
> > 
> > I just have one remaining question. It doesn't seem like F() and J()
> > are used by the Newton solver, only form(). Is that right?
> > 
> 
> They're not used yet. They will be used later when the Newton solver can
> be tuned. For example, a modified Newton method might call J() only at
> the first iteration and F() at every iteration. The extra functions
> could be removed until I get around to adding functionality to
> NewtonSolver.

ok.

> > Maybe one can put something in NonlinearProblem::form() that calls the
> > user-supplied F() and J() functions if form() has not been
> > implemented? The NewtonSolver would thus call form() and which in turn
> > calls F() and J(), unless form() has been overloaded by the user
> > (which would be more efficient).
> > 
> 
> Sure.  I'll take a look at it.

Maybe it's not critical to fix before we make the new release. I was
just curious.

> Any ideas about NonlinearPDE having NewtonSolver as a private
> variable?

I guess it may be more efficient to have it as a member if you are
doing multiple solves and call the Newton solver multiple times.

> It's possible that it could be removed once NonlinearPDE has more
> sophisticated solve options, like doing Newton-Raphson itself rather
> than single Newton solves as it does now.

Looks good for now.

/Anders