Topic
4 replies Latest Post - ‏2013-01-30T03:14:52Z by SystemAdmin
xhan
xhan
7 Posts
ACCEPTED ANSWER

Pinned topic Adding constraints improves the optimal solution?

‏2013-01-29T11:42:12Z |
Hi, everyone.
I am using Cplex 12.4 to solve my MIP model with minimization objective. The optimal solution is 203.
I then add several constraints to the model (they acutally indicate values for some decision variables), and the optimal solution is 200.

It seems impossible theoritically, but Cplex solved both models to optimality without reporting error. I am confused...

I have uploaded log files of these two runs. Thanks for your help!

Attachments

Updated on 2013-01-30T03:14:52Z at 2013-01-30T03:14:52Z by SystemAdmin
  • SystemAdmin
    SystemAdmin
    7929 Posts
    ACCEPTED ANSWER

    Re: Adding constraints improves the optimal solution?

    ‏2013-01-29T22:50:39Z  in response to xhan
    If you add the number of columns eliminated by presolve and the number of columns in the reduced MIP, you get 12,222 in the first model and 12,230 in the second model. In the process of adding constraints, did you also add variables? That could explain the improved objective value.

    Paul

    Mathematicians are like Frenchmen: whenever you say something to them, they translate it into their own language, and at once it is something entirely different. (Goethe)
    • xhan
      xhan
      7 Posts
      ACCEPTED ANSWER

      Re: Adding constraints improves the optimal solution?

      ‏2013-01-30T01:12:13Z  in response to SystemAdmin
      Thank you, Paul!

      In the added constraints, I just specify the values for some existing variables in the first model. I didn't introduce new variables
      Could this difference between 12,222 and 12,230 caused by the "substitutions" did by "Aggregator"?

      Besides, I am not sure, but shouldn't adding variables also degrade the objective value?
      • SystemAdmin
        SystemAdmin
        7929 Posts
        ACCEPTED ANSWER

        Re: Adding constraints improves the optimal solution?

        ‏2013-01-30T03:14:52Z  in response to xhan
        I'm not sue about the substitutions. Adding a variable never hurts the objective value (set it to zero and the previous solution remains feasible) and may help.

        Paul

        Mathematicians are like Frenchmen: whenever you say something to them, they translate it into their own language, and at once it is something entirely different. (Goethe)
    • xhan
      xhan
      7 Posts
      ACCEPTED ANSWER

      Re: Adding constraints improves the optimal solution?

      ‏2013-01-30T01:59:17Z  in response to SystemAdmin
      Problem solved. I tried the latest version of 12.5, and the first model is optimized to 200. Please see the attached log file.
      Thanks again for your help!