Originally posted by: jacobmays

I'm using AMPL and CPLEX 12.6.3.0 and have noticed a strange pattern. First I solve a MILP (a maximization) and get something like 

CPLEX 12.6.3.0: optimal integer solution within mipgap or absmipgap; objective 860919080.1

Then I fix the binary variables and resolve the resulting LP and get something like

CPLEX 12.6.3.0: optimal solution; objective 837455991.9

The surprising feature is that the objective shown for the LP is substantially worse than for the MIP. However, if I use the AMPL display command after solving the LP, it returns the higher value. Furthermore, I have compared the solutions to the two models and confirmed they are identical.

Any ideas as to what might be going on? 

Originally posted by: PierreBonami

Hi,

What is the value of the objective reported inside AMPL?

If your binary variables have non-zero objective coefficients, it might be that AMPL removes them from the problem it passes to CPLEX to solve. This would explain that the log of CPLEX reports a different value.

Best,

Pierre