[Ipopt] abort trap

[Sylvain Auliac]

I just discovered that when I compile my interface with the HSL subroutines , IPOPT do have the expected behaviour (that is: EXIT: Converged to a point of local infeasibility. Problem may be infeasible.). But troubles come when I compile with sequential mumps (which is the solver we can provide by default). The point is that compilation options are different in the two cases, so I guess it may be due to one of them. Getting closer...

Le 17 avr. 2012 à 13:46, Sylvain Auliac a écrit :

> Hi,
> Thanks for the answer. I did try to catch any exceptions that could have been thrown by the optimization process, with code like that
> try
> {
>   status = app->OptimizeTNLP(optim);
> }
> catch(...)
> {
>   cout << "error caught" << endl;
> }
> App is just a SmartPointer to an IpoptApplication. Nothing is caught...
> I've just tried a simpler problem, which obviously can't be solved, consisting in minimizing the norm of a vector with the constraint that the norm must be equal to 1 and 2 at the same time. Then, the result is always the following :
> 
>  ......
>  197r 1.5022719e+00 5.02e-01 3.13e-13  -9.0 3.80e-05    -  1.00e+00 1.00e+00h  1
>  198r 1.5022719e+00 5.02e-01 2.73e-13  -9.0 2.57e-05    -  1.00e+00 1.00e+00h  1
> Abort trap
> 
> I have seen in the current issues section that ADD_CXXFLAGS="-mmacosx-version-min=10.4" under macOS 10.6 could lead to "problems with try catch causing ABORT TRAP errors". We do not compile with this option, so there may be some other cases where this happens... Gonna open a ticket then when I'll be sure it's not a side effect of something else.
> 
> Sylvain.
> 
> 
> 
> Le 16 avr. 2012 à 17:46, Stefan Vigerske a écrit :
> 
>> Hi,
>> 
>> Ipopt usually does not just aborts.
>> In some very bad situations, it may end by throwing an exception.
>> You should add a try/catch clause around your call to Ipopt's Optimize() function and see if you can catch an IpoptException, or something else.
>> If that does not give further insight, open a ticket at the Ipopt webpage with information like the backtrace from the debugger.
>> 
>> Stefan
>> 
>> 
>>> Hello,
>>> I'm testing an interface i have written for ipopt in our software. Toy-test problems work fine, we also have a medium scale quadratic one who's running with no problem. I'm also trying to debug with a minimal surface problem with volume and simple bounds constraints. This one is nonlinear on every sides, IPOPT often succeed in finding a local minima (~half of the times) but on the other half, it crashs with no indication except an "abort trap" console message. When it happens, I see it has some difficulties because before stopping, it tries to perform a lot of line searchs and the optimization variables start oscillating around the last best solution found. Then the "r" appears right to the iteration number... and then the dreadfull message. On some very rare attempts, I get a restoration phase related error message.
>>> 
>>> The last function called in the debugger was in IpRestoMinC_1Nrm.cpp:354...
>>> 
>>> 
>>> Is there a way to catch the error and prevent the program to stop? I'd love to be able to get back the last values of the variables or things like that ...
>>> 
>>> Best regards,,
>>> 
>>> 
>>> Sylvain Auliac
>>> Laboratoire Jacques-Louis Lions
>>> Université Pierre et Marie Curie
>>> mail: auliac at ann.jussieu.fr
>>> tel : 01 44 27 71 70
>>> 
>>> 
>>> 
>>> 
>>> 
>>> _______________________________________________
>>> Ipopt mailing list
>>> Ipopt at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/ipopt
>>> 
>> 
> 
> Sylvain Auliac
> Laboratoire Jacques-Louis Lions
> Université Pierre et Marie Curie
> mail: auliac at ann.jussieu.fr
> tel : 01 44 27 71 70
> 
> 
> 
> 
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt

Sylvain Auliac
Laboratoire Jacques-Louis Lions
Université Pierre et Marie Curie
mail: auliac at ann.jussieu.fr
tel : 01 44 27 71 70




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20120417/0b7954e5/attachment-0001.html>