On Fri, 2 Jul 2010 14:44:25 -0400, Xuan YU <xxy113 at psu.edu> wrote: > > On Jul 2, 2010, at 2:21 PM, Jed Brown wrote: > > > On Fri, 2 Jul 2010 14:08:14 -0400, Xuan YU <xxy113 at psu.edu> wrote: > >> If I don't provide Jacobian, it will be: type: euler > >> total number > >> of nonlinear solver > >> iterations=0 > >> total number > >> of linear solver iterations=0 > > > > The default TS is the simplest possible integrator, forward Euler. Of > > course this doesn't do any solves (no SNES) and is not appropriate for > > stiff systems. Run with -ts_type theta for something more likely to > > work (and which will use SNES). > > > >> [0]PETSC ERROR: Floating point exception! > >> [0]PETSC ERROR: User provided compute function generated a Not-a- > >> Number! > > > > Euler is unstable for your problem, this should have nothing to do > > with > > the way you specified the Jacobian. > > > > > I tried beuler and pseudo, and I got the errors:
If you have petsc-dev, run in a debugger with -fp_trap to see where the NaN is first computed. Some possibilities include domain error (like negative density), nonlinear instability, and inadequate solution of the linear system. Jed
