dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, Dec 05, 2005 at 10:30:40AM +0100, Johan Hoffman wrote:

> > A good approach would be to implement our own Newton procedures, and set
> > things
> > up such that users can use the PETSc nonlinear solvers without too much
> > difficulty if they are willing to call PETSc functions directly.
> >
> > An interesting feature that PETSc does have the is the computation of
> > approximate Jacbian matrices. This feature seems to be under development
> > still.
> 
> Ok. So let us go with our own Newton procedures for now. And then possibly
> add a wrapper for computing approximate Jacobian matrices?

Sounds good.

Do we need any approximate Jacobians for solving nonlinear PDEs since
we assemble the Jacobian directly (or its action)?

Eventually, FFC will compute the Jacobian for you so then you can just
specify the (nonlinear) form and have everything else generated
(Jacobian, dual, a posteriori error estimate).

/Anders