Skip to main content

Gurobi reports a bounded model as unbounded

Answered

Comments

3 comments

  • Official comment
    Simranjit Kaur
    • Gurobi Staff Gurobi Staff
    This post is more than three years old. Some information may not be up to date. For current information, please check the Gurobi Documentation or Knowledge Base. If you need more help, please create a new post in the community forum. Or why not try our AI Gurobot?.
  • Jaromił Najman
    • Gurobi Staff Gurobi Staff

    Hi Bruno,

    It is hard to explain it briefly in a community post, thus I will directly point you to the publication explaining what is happening here. While the whole paper is definitely worth a read, you are most likely interested in Section 2.3 of Klotz 2014.

    What happens in your case is that the round-off errors of the coefficients and bounds in the MPS file come into play, e.g., the coefficient 2.0215719780000002e-01 should most likely read 2.021571978e-1 or 1.6581263629999998e-01 should rather be 1.658126363e-1. Note that when using the LP file format, you don't observe the behavior you describe, because the LP file format simply cuts off the numerical representation at some fixed decimal. Already these small round-off errors are enough to result in this rather unexpected behavior. You will also observe that if you would use the primal simplex algorithm (Method=0) instead of the dual simplex algorithm (Method=1), the situation you describe does not occur. This is because the optimality tolerance can be seen as the feasibility tolerance of the dual simplex method. Thus, playing with the optimality tolerance affects the feasibility numerics of the dual simplex algorithm.

    I hope that the paper and the above explanation help in the understanding of your problem.

    Best regards,
    Jaromił

    2
  • Bruno Colonetti
    • Gurobi-versary
    • First Question
    • First Comment

    Hi, Jaromił

    That's a great explanation! Thank you. And I'll definitely take a look at that paper.

    Thanks,

    Bruno

    0

Post is closed for comments.