[Ipopt] Cutting back alpha due to evaluation error

[Stefan Vigerske]

Hi,

> although it seems to be a little bit off topic: Is there a possibility
> in GAMS to detect the place of interest in the model?

If you could make Ipopt stop at the point where it gets into evaluation 
errors, you should get the primal solution values back in Ipopt as level 
values. The solution status should be something like 
intermediate-nonoptimal.
Maybe just run it once to see at which iteration the problems come up, 
and then again with a proper iteration limit.

Stefan

>
> Best regards,
> Martin
>
> Am 08.06.2011 12:04, schrieb Stefan Vigerske:
>> Hi,
>>
>> that the point cannot be evaluated is already recognized in the function
>> evaluation code (probably to GAMS AD library in your case), the Ipopt
>> interface then just signals to Ipopt that the given point could not be
>> evaluated.
>> Unfortunately, I do not see a place in the code that would print the
>> point before or after calling the evaluation routines.
>> If you can modify Ipopt code, then try to activate the
>> DBG_PRINT_VECTOR's in
>> IpoptCalculatedQuantities::trial_primal_infeasibility().
>>
>> Stefan
>>
>>
>>> Hi,
>>>
>>> I have an NLP for which I often get the message
>>>
>>> "Warning: Cutting back alpha due to evaluation error".
>>>
>>> I know that this has something to do with constraints that are not well
>>> defined at the current point (like sqrt of negative numbers etc.).
>>>
>>> Is there any chance to analyze at which constraint in the model the
>>> problems occurs? I thought I could find some "inf" or "nan" values in
>>> the higher output levels but I didn't find anything.
>>>
>>> Thanks in advance,
>>> Martin
>>>
>>> _______________________________________________
>>> Ipopt mailing list
>>> Ipopt at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/ipopt
>>>
>>
>>
>
>


-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan