Originally posted by: story_

Hi,

I am using Java API to solve an NP-hard, iterative MIP problem. I have set the MIP tolerance to a greater value (MipGap=10%) in earlier iterations to obtain quicker results, however, I need to solve its last iteration (which is determined through the optimal values of the decision variables) to optimality.

I was wondering if CPLEX is able to continue reducing the MIP gap when the predefined MIP tolerance is reached, but the optimal solution with MipGap=10% meets some criteria (instead of terminating when MipGap=10% is reached). 

Many thanks in advance.

Sorry, it is not quite clear what your problem is.

If I understand correctly then you call IloCplex.solve() in a loop. For all iterations but the last one you want to solve with MipGap=10%. Then why don't you just change the MipGap for the last iteration?

Or do you want to continue the previous search that stopped due to a gap of 10% but now want to continue with a tighter gap limit?

One way to implement complicated termination criteria is to use an info callback. In such a callback you can query the current incumbent solution and dual bound and decide whether you want to stop or not.

Originally posted by: story_

Hi,

Many thanks for your response. My main objective here is to continue the previous search that stopped due to the MIP tolerance gap and continue with a tighter one. I should have clearly indicated that it is not possible for us to know which iteration is the last one, as the terminating condition depends on the optimal objective value of the problem.

This should work as expected. If CPLEX stops because of a relative MIP gap then you can tighten the parameter and solve again. CPLEX should start the solve from where the previous one stopped.