This is what I have done to fix the worst:

1) The original idea of setting the tolerance to 10^-8 instead of
10^-15 times cell diameter caused some more roundoff problems, so I
set it to 10^-12. This should be fixed to a reasonable value at some
point

2) To cope with some convergence problems, I increased the maximum
number of steps to 50.

Since some tests still don't compile, I think we can fairly assume
that Newton does not converge in these cases. On the other hand, the
implementation does not even use the most basic stepsize control.

Since Seba is working on a fix, I suggest to let the corresponding
tests fail, because that reflects correctly the status of the library.
Depending on the timeline for this fix, we probably should add a bug
warning to the documentation.

What do you think?
Guido
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to