[Trilinos-Users] ML - mlguide example

Burkhard Doliwa doliwa@temf.tu-darmstadt.de
Tue, 14 Sep 2004 21:13:00 +0200

Dear users,

I have built the ML package as a stand-alone tool and am now trying to 
learn more about its capabilitites. Since good documentation of ML had 
encouraged me, I tried out the mlguide.c example from the tutorial 
(unfortunately I am no AMG expert). Well, the convergence rate was 
impressive so I used - instead of 129 unknowns - 1000000. No 
convergence. I then adjusted mlguide.c so that more than two levels were 
initialized. Also I used the Krylov-CG solver instead of the Jacobi 
smoother which also did not cause converge at large number of unknowns.

So my question to the experts would be whether it is possible by simple 
means to make mlguide.c work at N=1000000 or larger (the overall 
RELATIVE residuum should be smaller than 1e-8 or so). Other examples 
using the AZTEC package (e.g., ml_exampled2s.c) work quite well, however 
are too complex for a newbie.