I need a feasible solution, an optimal solution is just nice-to-have

Answered

Dániel Varga

November 16, 2023 10:03

• 
• 
• 

I have a really large integer program with linear constraints. It will probably run for weeks, extrapolating from smaller exploratory runs. I really need to find a feasible (integer) solution. We expect the optimal solution to have nicer properties, but that's a secondary concern at this point. Is there a way to tell Gurobi to emit any feasible solution it finds, and then keep working towards an optimal solution?

Related to

• MIP
• Solutions

0

• 

  Mario Ruthmair

  • Gurobi Staff

  November 16, 2023 16:48

  Hi Dániel,

  There are several solver parameters that focus on finding feasible solutions while proving optimality is only the secondary goal:

  • MIPFocus=1: search focus on finding good solutions
  • NoRelHeurTime: activation of optional NoRel heuristic to find a first feasible solution
  • Heuristics: time slice for heuristics
  • ZeroObjNodes, MinRelNodes: activation of special heuristics if no feasible solution can be found
  • PartitionPlace: use partition heuristic if your variables have a natural partition structure

  You could also solve a feasibility relaxation if your constraints are so restrictive that finding a solution is extremely difficult.

  Best regards,
  Mario

  1
• 

  Dániel Varga

  November 17, 2023 19:05

  • 
  • 
  • 

  Thanks a lot, Mario! It will take significant time to try out all these suggestions, but I am optimistic that it will lead to results.

  0

Please sign in to leave a comment.