[mesa-users] variability in runs

[Bill Paxton]

On Apr 23, 2012, at 12:11 PM, andres suarez wrote:

> I don't know if mesa has any non-deterministic part in it... or if rounding errors could lead to such varying evolution.

Hi Andres,

The non-determinism comes from bugs -- typically from uninitialized variables that get different values on different runs when you are using multiple threads.   The compilers help some in finding these, but in other cases it is just a matter of a long hard slog to track down the problem.  

You could help a lot if you could find a spot in the run where the variability shows up with a high probability.
Save a model a few 10's of steps before that.  
E.g. use something the following:
      save_model_number = 401
      save_model_filename = 'test.mod'

Change your inlist to load the model you just saved.
e.g., do something like
      load_saved_model = .true.
      saved_model_name = 'test.mod'

Rerun with the saved model to check that it shows the variability.
Assuming that it does, send me both your inlist and the saved model so I can try it.

Of course, if you enjoy debugging, I'd be delighted for you to track down the bug and tell me how it fix it!!!!! ;-)

Cheers,
Bill