Originally posted by: bocaboca

Hi all!

I experience a very strange behavior of the solver and I'd like to ask for some help. Let me first explain my problem:

I implemented the heuristic approach to solve something, and it finds a solution (this is not related to the CPLEX). Additionally, I formulated the problem in CPLEX. However, for the same inputs, the solver doesn't find it. In the beginning I thought that I had a mistake in the formulation of the problem, so I was checking it step by step, and didn't find any mistake. Then I tried to "help" the solver a bit, by giving him some parts of the heuristically-produced solution. And when I fix some values, (e.g. cplex.addEq(someBinary, 1);), the solver manages to find a solution.

To me, it is very strange that putting additional constraints (in the form of some equalities), leads to a feasible solution, while omitting them causes the solver to fail. It doesn't make any sense.

Did anybody have a similar experience? Any advices?

Thanks in advance!

This is a guess, but I wonder if the problem is numerically unstable.  Did you try using the numerical emphasis parameter?  You can find other suggestions, for example, here.

You may also want to export the original CPLEX model and the one with "help" as SAV files and attach them here (you can add some code to export them just before calling solve).