Why does MIPFocus value change the optimal solution?

Vincent

• April 13, 2020 18:51

Hi,

- When I select MIPFocus=1, the optimal solution is equal to 1.79542 with 0.0000% gap.
- When I select MIPFocus=3, the optimal solution changes to 1.75932 with 0.0000% gap.

I cannot understand why changing the value of this parameter should change the optimal solution of my model by almost 2%!

 

Thanks!

0

• 

  Sonja Mars

  Gurobi Staff

  • April 15, 2020 08:16

  Hi Vicent,

  Can you share (if they are not too long) the full log files of these runs? It would be interesting to understand a bit more about the model structure and how Gurobi works on it.

  Thanks,

    Sonja

  1
• 

  Eli Towle

  Gurobi Staff

  • April 21, 2020 21:17

  Hi Vincent,

  This behavior is often indicative of numerical issues (e.g., a large relative range of matrix coefficients, or large RHS values). It may help to reformulate the model to reduce the range of matrix coefficients. For more information, see the Gurobi Guidelines for Numerical Issues section of the documentation.

  As a workaround, you could try some of the following parameters mentioned in that section of the documentation (e.g., NumericFocus, Method, Aggregate=0). That said, the most reliable fix is to reformulate the model.

  Thanks,

  Eli

  1

Please sign in to leave a comment.