Tolerances and warm-starts

Awaiting user input

Philipp Brendel

• May 04, 2021 08:59

Hi all,

I noticed something which I'm not sure whether its intentional.

a) I solve a MIP only for feasibility (obj=0) with MIPGap = 1e-4 and default values for OptimalityTol, IntFeasTol etc...output leads to e.g.:

Optimal solution found (tolerance 1.00e-04)

b) I use the results from a) as warm-start for another optimization of the same MIP but with a non-zero  objective this time, output:

User MIP start did not produce a new incumbent solution
User MIP start violates constraint R1161 by 0.000050000

 

Two questions regarding this:

1. Why could MIPstart violate some constraint by 5e-5 when the tolerances in a) had been of size 1e-6 (FeasibilityTol, OptimalityTol)

2. Would this violation lead to Gurobi not using (all of) the warm-start values?

0

• 

  Matthias Miltenberger

  Gurobi Staff

  • May 04, 2021 12:58

  Hi Philipp,

  This might be a numerical issue. Can you please share the output of printQuality() after both optimizations? You could also share the entire log for additional information.

  Thanks,
  Matthias

  0
• 

  Dominik Zehetner

  • June 28, 2022 10:18

  Hello Matthias,

  I have the exact same problem. Is there already a solution for this issue?

  Thank you very much!

  Br,
  Dominik

  0
• 

  Matthias Miltenberger

  Gurobi Staff

  • June 28, 2022 12:25

  Hi Dominik!

  We need more information to work on this, for example, log files, model files, or anything else to reproduce this issue. This is not supposed to happen, of course, and is most likely a numerical problem.

  You can also open a Support request if you cannot share the details here.

  Thanks,
  Matthias

   

  0

Please sign in to leave a comment.