Complementary Slackness: Issues with Big-M

回答済み

Zhisheng Xiong

2025年08月09日 18:44

• 
• 
• 

Hi,
I’m using the Big-M technique to linearize the complementary slackness conditions from the KKT system, which turns my optimization model into a MIP (the only integer variables come from these complementary slackness constraints).

When I set M too large, I end up getting only a relaxed solution; but if I set it smaller, Gurobi fails to converge and keeps iterating without termination.

Is there a smarter way to handle this? For example, can using addGenConstrIndicator avoid having to manually choose a Big-M value? Or do you have any other suggestions?

Thanks a lot!

関連

• Bilinear
• Big-M

0

• 

  Riley Clement

  • Gurobi Staff

  2025年08月11日 00:32

  Hi Zhisheng,

  You can use addGenConstrIndicator to avoid choosing a Big M value, but I don't expect this would resolve your problem.

  Your best bet is to try and tune the solver.  I would first try with different combinations of MIPFocus and Presolve and if you want to dive deeper then the following video may help: Introduction to Performance Tuning.

  - Riley

  0
• 

  Zhisheng Xiong

  2025年08月14日 08:08

  • 
  • 
  • 

  Hi Riley,

  Thanks for your valuable information. I have checked the code examples and video, and now have some ideas in mind. I will try running my optimization again. 

  Since the solution of the KKT system is used to generate cuts in a decomposition algorithm, if the solution quality is poor, the cuts will be weak—or worse, the whole algorithm may converge prematurely.

  0

サインインしてコメントを残してください。