cbc team mailing list archive

[Harish Narayanan <harish.mlists@xxxxxxxxx>]

On 3/20/10 9:14 AM, Anders Logg wrote:
> Interesting, how does it compare to the existing solver?
> 
> I expect it to converge slowly (in particular the solution of the
> linear saddle-point problem in each iteration).

Which is what I expected given the folklore on this matter, but I was
surprised by what I saw. It converges quickly with two iterations of the
Newton solver (with two iterations of the linear solver) while the
problem is time dependent, and one iteration thereafter.

My deeper plan was to pose the monolithic problem defined in section 3.4
of the FSI paper and compare it to the existing operator-split primal.
This was an attempt to make the solid and fluid solvers look as close as
possible (and note from trial.py and trial2.py in sandbox/cg1 that they do.)

Harish