Should I get the same optimal solution when the initial point is different?

Answered

Arezoo Jafari

June 16, 2022 15:51 Edited

• 
• 
• 

Hi everyone,

I designed a two-stage stochastic programming model and solved it using Benders Decomposition. When I forced the model to have a different initial point, the optimal solution and objective value became different. 

I found this post (link: https://support.gurobi.com/hc/en-us/articles/360031636051-Is-Gurobi-Optimizer-deterministic- )

Does it mean that different initial points may change the optimal solution?

 

0

• 

  Maliheh Aramon

  • Gurobi Staff

  June 18, 2022 14:28

  Hi Arezoo, 

  If you are solving the exact same model with the exact same input data, starting from different initial solutions should not change the optimal solution. Depending on the parameters used, the solution path taken to prove optimality might vary. 

  Strange solver results such as reaching different optimal solutions  by just using different initial solutions might be a sign of having a model with numerical issues. You can follow the steps outlined in our documentation of Does my model have numerical issues? to determine whether your model has numerical issues or not.

   

  Best regards,

  Maliheh

   

  0

Please sign in to leave a comment.