GAMS takes a long time to return infeasibility for a certain parameter

[Michael He]

Hello everybody,

currently, I am running a MILP in GAMS with different input parameters, just to check, for which values the model is feasible. Most of the time, infeasibility is found very fast, in a few seconds, but for one particular value it takes a lot, >240 seconds.

I am confused about it, since it only takes that long that one random value, while the duration is low for all other values, higher and lower ones. In general, I have always experienced that GAMS is very fast in finding the infeasibility of a model (since it doesn't solve the model). What could be a reason to take that long for GAMS, even before actually starting the optimisation?

Thank you in advance,
Michael

[bussieck]

A MIP solver can often detect the infeasibility of an instance in the presolve step. This goes usually very quickly. If the infeasiblity cannot be determined by presolve and the underlying LP relaxations are feasible than B&C kicks in and this can take a long time. Theoretically, determining if a MIP is infeasible is as computationally expensive as finding the optimal solution (separation vs optimization).

-Michael

[Michael He]

Hello Michael,

Thank you for your answer.

Best