Presolve disabled leads to constraint violation

Answered

Maximilian Roth

• February 24, 2022 11:36

Dear Gurobi Team, 

I have a very large MILP model with various linearizations, both univariate and bivariate. When I solve the problem - presolve enabled - I get a plausible solution. But when I deactivate presolve, the solution for the system variables as well as the objective function value deviates. It should be mentioned that the solution in this deactivated context is clearly invalid because constraints are not respected. 
My guess is that this could be due to my big-M, which is quite large and applies to all constraints and presolve cleans that up and thus no numerical errors show up.
Any ideas? 

Best regards
Max 

 

0

• 

  Jaromił Najman

  Gurobi Staff

  • February 25, 2022 12:25
  • Edited

  Hi Max,

  My guess is that this could be due to my big-M, which is quite large and applies to all constraints and presolve cleans that up and thus no numerical errors show up.

  Yes, your guess is very likely correct. Presolve does not only reduce the number of columns and rows but it often improves the behavior of numerically difficult models by, e.g., scaling. Obviously, it is best to try to formulate a well-scaled model but Presolve can often handle some numerical issues.

  Our Guidelines for Numerical Issues are a good starting point to improve the numerical properties of a model.

  Best regards, 
  Jaromił

  0
• 

  Maximilian Roth

  • February 25, 2022 12:24

  

  

  

  Thanks a lot Jaromił! :)

  0

Please sign in to leave a comment.