False Positive of Model being Infeasible, maybe about exponential constraints

Awaiting user input

Lei Huang

• February 03, 2025 22:32
• Edited

Summary: 

 

This model, involving exponetial constraints, should have obvious feasible solution according to my observation. However, my Gurobi solver said it "Model is Infeasible". Furthermore, if I remove the only general constraint at the end of the model file, GC60: C1575 = EXP ( C1574 ) , the Gurobi solver does not say "model is infeasible" anymore. Here, C1575 is free and C1574 has feasible assignment.

 

My solver version is gurobi11.0.0_linux64.tar.gz

 

Model file:

(computeIIS() has applied)

https://drive.google.com/file/d/1G1PfF_jcMPWvu798lTudGfR_9xfRQZno/view?usp=sharing 

 

Reproduce code:

from gurobipy import GRB
import gurobipy as gp

def solve_from_ilpfile(filename):

    # Create a new model

     m = gp.read(filename)

    # Optimize model

    m.optimize()

    # Print solution

    if m.status != GRB.INFEASIBLE:

        print('Optimal objective: %g' % m.objVal)

        m.write("output.sol")

    else:

        print('Optimization was stopped with status %d' % m.status)

    return m

if __name__ == "__main__":

   import argparse

   parser = argparse.ArgumentParser()

   parser.add_argument("filename", help="input ilp file", default="iis.ilp.txt")

   args = parser.parse_args()

   res = solve_from_ilpfile(args.filename)

   print(res)

Related to

• General Constraints

0

• 

  Jaromił Najman

  Gurobi Staff

  • February 04, 2025 14:31
  • Edited

  Hi Lei,

  Could you please update to Gurobi version 12 and try again.

  If you are not able to upgrade to version 12, please try setting the FuncNonlinear parameter to 1.

  This will tell Gurobi to treat the \(\exp\) function as a nonlinear term and not via a piecewise-linear approximation. Fore more details, please refer to the documentation about Solution strategies for nonlinear models.

  This model, involving exponetial constraints, should have obvious feasible solution according to my observation.

  Are you able to construct one such feasible solution? If yes, then please try providing it via a MIPStart, see How do I use MIP start?

  You could try removing portions of other constraints (not the \(\exp\) one) and see whether the model still remains infeasible. This would be useful to construct a smaller version of your infeasible model, which might be easier to analyze.

  If nothing of the above helps, could you please share your original model?

  Best regards, 
  Jaromił

  0

Please sign in to leave a comment.