Questions about Gurobi's solution

回答済み

Ahhyun Kim

• 2024年09月08日 08:02

I ran the very simple problem below in an interactive shell (it's an excerpt from a larger problem). I expected the result to be infeasible, but it yielded a solution, and when I ran it with presolve=0, the result was infeasible.
What could be the problem?
Any help would be appreciated.

 

Thank you.

 

===========

Maximize
    C0
Subject To
 budget_sum: C0 <= 17065
 budget_media_sum: C0 = 17065
 revenue_sum: C5 >= 2.23116e+05
Bounds
 15358.5 <= C0 <= 18771.5
General Constraints
 revenue_PWLConstr: C5 = PWL ( C0 ) : (15358.5, 26482.68874) 
  (18771.5, 62763.75945)
End
  

==========

関連するエンティティ

• Infeasible

0

• 

  Riley Clement

  Gurobi Staff

  • 2024年09月08日 23:50

  Hi Ahhyun,

  The value of C0 is fixed in your model:

   budget_media_sum: C0 = 17065

  The value of C5 will therefore fixed by the PWL constraint:

  revenue_PWLConstr: C5 = PWL ( C0 ) : (15358.5, 26482.68874) 
    (18771.5, 62763.75945)

  You can calculate this (eg with Python):

  >> x1=15358.5
  >> y1=26482.68874
  >> x2=18771.5
  >> y2=62763.75945
  >> x = 17065
  >> (x-x1)/(x2-x1)*(y2-y1)+y1
  44623.224095

  So if C0 = 17065 then C5 = 44623.224095.  But you also have a constraint

  revenue_sum: C5 >= 223116

  and this is not compatible with the calculated value of C5.  So, the model you have presented is infeasible.  Perhaps there is a mistake in your data or your model if you were expecting it to be feasible.

  - Riley

   

   

  0
• 

  Ahhyun Kim

  • 2024年09月10日 11:28

  

  

  Thank you for your kind response.

  
  You are correct that the problem with the attached LP file is INFEASIBLE.
  My question is that when I solve this lp file in interactive shell, feasible comes out.
  I have attached the gurobi result below.
  Sorry for the inconvenience, but I would appreciate if you could take another look.

  Thank you.

  

  0
• 

  Riley Clement

  Gurobi Staff

  • 2024年09月11日 13:29
  • 編集済み

  Hi Ahhyun,

  This may be a bug.  I suspect this issue only arises if the model, minus general constraints, can be solved entirely by presolve (which is certainly the case here).

  As a workaround, please use the ConstrVio attribute to check the violation in addition to Status, and in the meantime we'll work on resolving the issue.

  Edit: there should be no difference between the interactive shell and using an LP file, and I don't observe one.  The only difference I see is when presolve=0.  In this case we do not run the normal presolve routine but we do convert general constraints into linear constraints, which is then why we do not encounter the (assumed) bug.

  - Riley

   

  0
• 

  Riley Clement

  Gurobi Staff

  • 2024年09月12日 03:59

  Hi Ahhyun,

  I can confirm it is a bug, and will be fixed in our upcoming release V12 which will be published later this year.  Thanks for bringing it to our attention!

  - Riley

  0

サインインしてコメントを残してください。