[Couenne] false infeasible

[Pietro Belotti]

Hi Abdel,

Baron and Couenne attempt at solving the same class of problems, so if
Baron finds an optimal solution Couenne should find the same solution.
Obviously for large and/or difficult problems it isn't always possible
to find a global optimum, hence some solvers resort to heuristics. The
fact that Baron here finds a feasible solution and Couenne doesn't
might indicate that the heuristics in Baron were more successful in
finding a solution. Since CLP finds a solution with the same objective
function value, it might also mean that the feasibility tolerances are
larger than in Couenne. Try setting

feas_tolerance 1e-4

in the option file couenne.opt. This option is the tolerance for
feasibility of a solution, and it's 1e-6 in Couenne.

Pietro

On Wed, May 20, 2015 at 10:34 AM, ALI ZAZOU Abdelkrim
<abdelkrim.ali-zazou at ensma.fr> wrote:
> Hi,
>
> thank you very much for your fast reponse and sorry for my late one.
>
> so i tried the options you adviced me but i couldn't get no changement.
>
> If i understand well this "objectiv value" is not   necesseraly feasible but
> in that case the
> "optimal" value i get from NEOS/BARON is exactely the one that CLP
> displayed. So why for baron it's a feasible solution and not for Couenne?
>
> regards
> abdelkrim
>
>
>
>
> Le 14/05/2015 12:16, Pietro Belotti a écrit :
>>
>> Hi Abdel,
>>
>> the "objective value" strings you see are from CLP, the LP solver used
>> by Couenne to find lower bounds. The solutions found by LP are not, in
>> general, feasible for the problem itself. In order to filter out the
>> output for less clutter, I would suggest adding the following options
>> to the option file:
>>
>> bb_log_level 1
>> lp_log_level 0
>> nlp_log_level 0
>>
>> Also, I'm not sure about the time limit and I don't know why no
>> solution is found, but you could try to restrict the bound tightening
>> by adding one or more of the following options:
>>
>> feasibility_bt no
>> optimality_bt no
>> aggressive_fbbt no
>>
>> Hope this helps,
>> Pietro
>>
>>
>> On Wed, May 13, 2015 at 10:37 AM, ALI ZAZOU Abdelkrim
>> <abdelkrim.ali-zazou at ensma.fr> wrote:
>>>
>>> Hi everyone,
>>>
>>> i'm having a problem to get results for my optimization problem that i
>>> run
>>> through NEOS.
>>> When i look at the result, i can see that the solver find the optimal
>>> solution, because it's the same objectif value that i get from another
>>> solver that finished with "optimal", but it ends on a "Couenne finished.
>>> No
>>> feasible solution found.".
>>>
>>> here is the text from the solver where i can see that "optimal value":
>>>
>>> "Clp0006I 7269  Obj 831436.86
>>> Clp0000I Optimal - objective value 831436.86
>>> Clp0032I Optimal objective 831436.862 - 7269 iterations time 0.522,
>>> Presolve
>>> 0.
>>>
>>> 08
>>> Clp0006I 0  Obj 831436.86 Primal inf 161.58395 (12)
>>> Clp0006I 4  Obj 831436.86
>>> Clp0000I Optimal - objective value 831436.86
>>> NLP0014I             2        TIME 0        0 0.035995
>>> Clp0000I Optimal - objective value 831436.86
>>> Cbc0020I Exiting on maximum time"
>>>
>>> but the solver ends with that message:
>>>
>>> "**** SOLVER STATUS     3 Resource Interrupt
>>> **** MODEL STATUS      14 No Solution Returned
>>> **** OBJECTIVE VALUE               NA          "
>>>
>>>
>>> So i wanted to how can i make the solver stops when it reach this
>>> "Optimal -
>>> objective value"??
>>>
>>> Thank you very much
>>> Abdel
>>> _______________________________________________
>>> Couenne mailing list
>>> Couenne at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/couenne
>
>