[petsc-dev] bad advice? TSStep has failed due to DIVERGED_NONLINEAR_SOLVE, increase -ts_max_snes_failures or make negative to attempt recovery!

[Barry Smith]

On Oct 22, 2013, at 11:58 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> Barry Smith <bsmith at mcs.anl.gov> writes:
> 
>> Begin forwarded message:
>> 
>>> From: Christophe Ortiz <christophe.ortiz at ciemat.es>
>>> 
>>> [0]PETSC ERROR:   !
>>> [0]PETSC ERROR: TSStep has failed due to DIVERGED_NONLINEAR_SOLVE, increase -ts_max_snes_failures or make negative to attempt recovery!
>>> 
>> 
>>   Do we really want this as a our "helpful message" here? Usually the
>>   problem is a bad Jacobian etc which the person needs to track
>>   down. It would be better if this instead listed an FAQ question
>>   with information (and links there) to all possible causes of the
>>   problem and how to track it down.  Just allowing failure here is
>>   not the first step one should take I think.
> 
> Hmm, this is not uncommon if solving a strongly nonlinear problem,
> especially when using crude error tolerance.  So for the expert user,
> this error is usually offering the right guidance.

    What expert users do we have ?  90+% of all failures of Newton are due to bad Jacobians so the suggestion here is useful less than 10% of the time and sends all non-experts down the wrong road.

   BTW: we need to finish up the refactorization of -snes_type test to the continuous test that runs for all solves. We have two competing fragments that need merging.

   Barry


>  To accommodate the
> error-prone user, it would be fine to use a link to a web page
> explaining the steps to debug.