Originally posted by: GlcBrero

Dear CPLEX support,

I have noticed that varying the relative MIP gap tolerance in a QP has only effect when the QToLin parameter is set to 0 (which, according to some old documentation I have found here https://www.ibm.com/support/knowledgecenter/SSSA5P_12.7.1/ilog.odms.cplex.help/CPLEX/Parameters/topics/QToLin.html should mean "not linearize"). Is this normal? Changing the relative MIP gap tolerance has no effect when I set the QToLin parameter either to 1 or to -1.

Thanks,

Gianluca

P.s. I am currently using Cplex Studio 12.9

The MIP gap parameter should always be respected, no matter what presolve settings you have. How exactly do you conclude that the parameter is ignored? Can you show the full logs for a run in which it is respected and for one in which it is supposedly not respected?

Originally posted by: GlcBrero

Hi Daniel,

Thank you for your answer. I now believe that the MIP gap parameter is always working. In the linearized version, the final MIP gap happened to be zero most of the times, even under a large MIP gap tolerance. However, I have just found out a case where the final gap was positive even under the linearized version.

Thanks again,

Gianluca