Originally posted by: m_carv

Hi everyone

I am solving the following mixed integer quadratic problem

Maximize
 obj: - 16 y1 + 13 q1 + [ - 2 q1 ^2 ] / 2
Subject To
 bigM: - 50 y1 + q1 <= 0
Bounds
 0 <= y1 <= 1
Binaries
 y1
End
 

for which the optimal solution is y1=1, q1 = 6.5 with objective value equal to 26.25.

However, when the optimization executed by CPLEX 12.6.0.0 is y1=0 and q1=0.

Attached it is the python file to create and solve the above optimization problem.

Am I doing something wrong? Is this a bug?

I changed the quadratic coefficient to be 0.5 and, in this case, the correct optimal solution is returned by CPLEX....

Thanks in advance.

Hi

this looks like a presolve issue.

Can you try with

parameters.preprocessing.presolve

set to 0 ?

regards

Originally posted by: m_carv

Adding

cpx.parameters.preprocessing.presolve.set(0)

to the python file in fact solves the issue. Cplex is able to return the optimal solution.

Thank you very much for the suggestion.

best

Originally posted by: EdKlotz

Yes, something looks wrong with presolve here, and we will investigate.   Meanwhile, you can work around the problem without giving up all of presolve by setting the presolve reduce parameter to only do primal reductions.    From Python I think this would be cpx.parameters.preprocessing.reduce.set(1).