[Cbc] Unbounded MIP problem returned as infeasible by CBCSolver

[John Forrest]

Kish,

Thanks for finding all these outlier problems.

I did not notice secondaryStatus_==1 case (infeasible on LP solve), so 
if changing CbcModel::isProvenInfeasible() to

   if (!status_ && (bestObjective_ >= 1.0e30
            && (secondaryStatus_==0||secondaryStatus_==1)))

fixes then I will commit change.

John

On 08/03/14 21:35, Kish Shen wrote:
> John,
>
> Thanks for the fix. I am unfortunately running into a new problem:
>
> One of our unit tests for our CBC interface code is now aborting with 
> the new CBC 2.8 that I downloaded, and I suspect it is related to the 
> fix for the unbounded problem -- for the following problem:
>
> Minimize
> obj: x0
> Subject To
> cons0:  x0 + x1 >= 3
> cons1:  x0 - x1 = 0
> cons2:  x0 + x1 <= 3.5
> Bounds
>   x0 Free
>   x1 Free
> Integers
> x0
>
> which is infeasible, but my code now report this as aborted, because 
> isProvenInfeasible() now returns false for this problem:
>
> >> if (model->isInitialSolveProvenOptimal()) {
> >>     if (model->isProvenInfeasible()) {
> >>         } // else assume MIP aborted
> >> //<1>
>
> I assume it previously returned true, and the change is related to the 
> previous fix?
>
> Cheers,
>
> Kish
>
> On 06/03/2014 20:18, John Forrest wrote:
>> Kish,
>>
>> Have changed so (model->isInitialSolveProvenOptimal()) returns false (as
>> it should).
>>
>> Checked in to 2.8.  Hope that fixes it.
>>
>> John
>>
>> On 06/03/14 19:32, Kish Shen wrote:
>>> Hi John,
>>>
>>> Thank you very much for your reply.
>>>
>>> > However when I try model->isContinuousUnbounded() it returns true
>>>
>>> I reran my program and you are correct -- isContinuousUnbounded() does
>>> return true for the problem. I am not sure what happened previously,
>>> sorry for the misreporting.
>>>
>>> >    (model->isProvenInfeasible()) returning true is a bug - the fix is
>>> > modifying code in that function to
>>> >
>>> >   if (!status_ && (bestObjective_ >= 1.0e30 && !secondaryStatus_))
>>> >
>>>
>>> I made the above change in CbcModel.cpp, and isProvenInfeasible()
>>> returns false as expected. Thanks! Will this change be put into the
>>> Cbc 2.8 branch?
>>>
>>> However, my result status checking code now reports the problem slove
>>> as aborted, because in my code:
>>>
>>>
>>> if (model->isInitialSolveProvenOptimal()) {
>>>     if (model->isProvenInfeasible()) {
>>>         } // else assume MIP aborted
>>> //<1>
>>> ...
>>> }
>>> ...
>>> if (model->isContinuousUnbounded()) {
>>> // problem is infeasible or unbounded
>>> ....
>>> }
>>>
>>> so if isProvenInfeasible() is false, the problem is assumed to have
>>> aborted, and isContinousUnbounded() is only called if
>>> isInitialSolveProvenOptimal() is false.
>>>
>>> Should the isContinousUnbounded() test be moved into <1> above, or
>>> should the test be performed at both places (i.e. at <1> and where it
>>> is now)?
>>>
>>> Thanks and cheers,
>>>
>>> Kish
>>>
>>>
>>>
>>>
>>> On 06/03/2014 10:08, John Forrest wrote:
>>>> Kish,
>>>>
>>>>    (model->isProvenInfeasible()) returning true is a bug - the fix is
>>>> modifying code in that function to
>>>>
>>>>   if (!status_ && (bestObjective_ >= 1.0e30 && !secondaryStatus_))
>>>>
>>>> However when I try model->isContinuousUnbounded() it returns true
>>>>
>>>> If I have
>>>>
>>>> printf("status %d secondary %d infeasible %s unbounded %s\n",
>>>> model.status(),model.secondaryStatus(),model.isProvenInfeasible()
>>>>       ? "true" : "false",model.isContinuousUnbounded() ? "true" :
>>>> "false");
>>>>
>>>> in my code I get
>>>>
>>>> status 0 secondary 7 infeasible false unbounded true
>>>>
>>>> with above fix and
>>>>
>>>> status 0 secondary 7 infeasible true unbounded true
>>>>
>>>> without fix.
>>>>
>>>> Can you check isContinuousUnbounded() and if it fails send me some 
>>>> code
>>>> to reproduce?
>>>>
>>>> John Forrest
>>>>
>>>>
>>>> On 06/03/14 01:17, Kish Shen wrote:
>>>>> Hi,
>>>>>
>>>>> For the following (very simple) problem:
>>>>>
>>>>> Minimize
>>>>> obj: x0
>>>>> Subject To
>>>>> cons0:  x0 + x1 <= 3
>>>>> cons1:  X0 = X1  = 0
>>>>> Bounds
>>>>>  x0 Free
>>>>>  x1 Free
>>>>> Integers
>>>>> x0
>>>>> End
>>>>>
>>>>> which is unbounded, but when we use CBCSolver to solve this problem,
>>>>> we get the result that the problem is infeasible, specifically,
>>>>>
>>>>> CbcModel* model;
>>>>> .....
>>>>> model->isProvenInfeasible() returns 1 (true)
>>>>>
>>>>> Here is the relevant code we use to determine the status of a problem
>>>>> after calling CbcSolver (via CbcMain0 and CbcMain1) to solve the
>>>>> problem:
>>>>>
>>>>>
>>>>> if (model->isInitialSolveProvenOptimal()) {
>>>>>    if (model->isProvenInfeasible()) {
>>>>>        ....
>>>>>
>>>>> For the above problem, the above if .. succeeds, which I think should
>>>>> only be the case if the problem is proven infeasible at the root 
>>>>> node,
>>>>> but the root node solve should be unbounded (and in fact if I solve
>>>>> the linear problem with CLP, the result is indeed unbounded), but the
>>>>> test
>>>>>
>>>>> We actually have code to test if the root node solve is unbounded (in
>>>>> which case we classify the solve result as "Unknown", i.e. either
>>>>> infeasible or unbounded, which was the result I expected for the 
>>>>> above
>>>>> problem), but
>>>>>
>>>>> model->isContinuousUnbounded() returns 0 for the above problem, 
>>>>> rather
>>>>> than 1 as I expected.
>>>>>
>>>>> Is there something I am doing wrong, or is there a problem?
>>>>>
>>>>> I am using Cbc 2.8, which was downloaded last month (14 Feb).
>>>>>
>>>>> Thanks in advance for any help.
>>>>>
>>>>> Cheers,
>>>>>
>>>>> Kish
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Cbc mailing list
>>>>> Cbc at list.coin-or.org
>>>>> http://list.coin-or.org/mailman/listinfo/cbc
>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> Cbc mailing list
>>>> Cbc at list.coin-or.org
>>>> http://list.coin-or.org/mailman/listinfo/cbc
>>>
>>> _______________________________________________
>>> Cbc mailing list
>>> Cbc at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/cbc
>>>
>>>
>>
>> _______________________________________________
>> Cbc mailing list
>> Cbc at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/cbc
>
>
>