Originally posted by: anahana

Hello everyone,

I have a MIP with a maximization objective. CPLEX terminates at the root node with an infeasible or unbounded solution. I added a constraint to the objective to be greater than or equal to zero and CPLEX found the optimal solution within minutes. The solution makes sense and is consistent with other instances I solved. Any idea why adding a lower bound on the objective works? I'm guessing it has to do with the dual problem, but not sure if that is the case.

Regards,  

Originally posted by: EdKlotz

Please include the CPLEX node logs of the two runs in question.   That will give the readers of this thread a better chance of providing an explanation.   Meanwhile, the first thing you should check is the accuracy of the solution CPLEX found.    From interactive CPLEX, use the 'display solution quality' command.   If the solution quality is marginal, your model might be right on the edge of feasibility/infeasibility, which would explain the different results you reported.

Originally posted by: anahana

I'm using CPLEX 12.6.2 through AIMMS, I should be able to print the node log, but I have to figure it out. For now, I used a newer version of AIMMS that uses CPLEX 12.6.3 and I did not run into this problem, so for now things are fine for me, but it is worth understanding why this happened in 12.6.2 and if indeed it was a bug that was fixed in 12.6.3.

Regards, 

Originally posted by: EdKlotz

Agreed; given the nature of how branch and bound can be sensitive to path changes associated with different versions, we typically do not assume a bug or other issue went away just because it no longer occurs with a later version.   Googling for "AIMMS cplex output" turfed up

https://groups.google.com/forum/#!topic/aimms/z2vRrRijp2s

which appears to describe how to generate CPLEX output to a file.   I also found

http://techblog.aimms.com/2011/12/

that describes how to export an MPS file of the model from AIMMS.   Presumably AIMMS offers a similar option to generate a SAV file of the model,

which is what we would prefer.  If you upload the SAV file here, I will have a look and hopefully verify that the issue is either a bug that has been fixed,

or some sort of numerical/ill conditioning issue that can be remedied by an adjustment to the model formulation.

Originally posted by: anahana

Thanks! Much appreciated. I'll upload the model in .sav or .mps if I can export it. 

Regards,