Performance consideration: constraining a boolean variable vs not having it in the model at all

[Mapiarz]

Hi! I'm working on a scheduling problem. My solution is basically a big matrix of boolean variables. Some of those variables must have value 0 and I know that beforehand (e.g. some employees are not qualified to work certain shifts so they will never be assigned such a shift). I achieve that by adding a constraint ensuring their value will always be 0 (solver.Add(variable == 0)).

Is there performance to be gained by not adding those variables to the solver in the first place, if I know they have to be 0 before hand?

Thanks

[jmarca]

I’ve noticed that the cpsat solver presolve step seems to eliminate redundant constraints. So I have always kept in all the constraints and variables I need for the logic to be clear, and let the solver winnow those down to a more concise set.

But I’m very interested in a more definitive answer.

[lperron]

It takes memory and time, but the solving time after presolve should be the same.