[petsc-users] More stable solver

[Jed Brown]

paul zhang <paulhuaizhang at gmail.com> writes:

> Hello,
>
> I am modelling a hypersonic computational fluid dynamic problem, where the
> matrix A is quite sparse and stiff. The solution is x=[rho_1, rho_2, ..,
> rho_11, u, v, w, T], where rho_s is the density for each of the species,
> some of which can reach 1.E-5, while some others can be below 1.E-20.
> As I ran my case, I found the number of iterations oscillated after being
> steady for a long time. For instance, the number of iterations was like 10
> for 2000 steps, then it goes up to 50 for no reason. 

Computers usually don't do things "for no reason".  I suggest that you
write out the problem state immediately before and after the problem
gets more difficult.  It can be useful to visualize that state, but also
to debug convergence using -ksp_monitor_true_residual,
-ksp_monitor_singular_value and similar.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20141002/163a392a/attachment.pgp>