Re: [Help-glpk] Asking information on modelling conditional constraints

[john tass]

Finally, I found the answer to the problem.

In general, both [1] and [2] are necessary to ensure the correct modelling of [Rel 1].

Nevertheless, the fact is that in my problem, the cases that all follow the form of [Rel 1] are not independent. So, by forcing both [1] and [2] to only one of them ensures that both [1] and [2] are set to all cases, even if [2] is not set explicitly.

In other words, the fact of dependency of the cases in my model, allows to set only the [1], while [2]  implicitly holds, provided that both [1] and [2] are set to only one of the total cases.

Hope that the information on this topic  will be helpful to some one and I ma sorry if I waste your time.

2016-02-22 1:44 GMT+02:00 ΤΑΣΣΟΠΟΥΛΟΣ ΙΩΑΝΝΗΣ <address@hidden>:

Good evening to everyone.

I am facing a rather peculiar situation which is as follows:

I have a MIP problem with several constraints (about five). All of them have the same type (with different parameters, though).

The form of these constraints is:

Let U denotes some integer variables (indices not included for simplicity reasons), where 

      - Lower <= U <= Upper (where -Lower and Upper depend on the variable U)

Let D be a binary variable.

I mast model the following:

D = 1 if U > 0 (i.e. 1 <= U <=Upper) and D = 0 otherwise (i.e. -Lower <= U <= 0).   [Rel 1]

I have tested the following model:

[1]  U <= Upper*D

[2] U + Lower >= (Lower + 1)*D

As it is easily verified these two constraints ensure the relation [Rel 1].

Now, I noticed that if I  exclude the second form of  constraints ([2]) from my model at all, an error occurs . To be more exact, D takes wrong value, when -Lower <= U <= 0.

At the other hand, if I include both [1] and [2], the solution is correct but the execution time is too much while the value of objective function is high. (but correct).

To make things even more complex, I noticed that when I exclude again the form [2] from all my constraints, but I keep it in only one of them, the model works perfectly. The execution is done fast and the objective value is almost optimal.

Now the question is: in order to model a condition like [Rel 1] are both [1] and [2] constraints necessary or [1] should be enough?

Thanks in advance for spending your time on my issue and for any answer.

_______________________________________________
Help-glpk mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/help-glpk

--

Ioannis X. Tassopoulos