0

I would like to use SCIP (or really, any solver, SCIP just seems the most likely candidate for this to work at all) to solve a MIP using a pure cutting plane approach -- no branching. Using just Gomory cuts will probably work fine.

Then, I want to read out the constraints that show up in the final LP relaxation, the one whose optimal solution is integer feasible.

Do the features to do this exist at all in SCIP or any other solver? If so, where in the API should I start digging?

(I am trying to avoid generating the cuts myself as I have heard this can be numerically tricky.)

Michael Curry
  • 183
  • 1
  • 5

1 Answers1

3

What have you tried so far? You should disable all cut limiting parameters, i.e. increase the number of cutting rounds and cuts per round (for all cut generators you are interested in) and decrease parameters like minorthogonality.

But even in this setting, SCIP may start branching at some point, so you might have to disable some checks in the code to keep on cutting.

Fair warning, though: You are most likely not going to solve your instance(s) due to numerical troubles. Too many cuts almost always deteriorate numerical stability of the LP solutions - this is one of the reasons why we use branch-and-bound.

mattmilten
  • 6,242
  • 3
  • 35
  • 65