Solve MINLP with SCIP

[uvutna]

Describe the bug I'm using CVXPY with solver SCIP to solve mixed integer nonlinear programming problems (MINLPs). I have a problem with a constraint that contains log function.

My constraint is as below:

for k in range(A):
   temp1 = 0
   for b in range(B):
       temp2 = 0
       for i in range(C): 
           temp2 += p[i,k,b]*z[(i,k,b)]
       temp1 += W*(cp.log(1+ temp2))
   constraint.append(temp1 >= R * x)

Expected behavior If I comment that constraint, it can compile, solve my problem and show status whether it is feasible or not

If I keep it, CVXPY doesn't compile my problem and shows non_feasible immediately

I also test with another solver and it is feasible

System

• OS: Window 10
• SCIP version: 8.0
• How did you install pyscipopt? I use pip install PySCIPOpt

[Joao-Dionisio]

Good morning, @uvutna!

Have you imported the log function from pyscipopt? Usually you would have something like this in your file:

from pyscipopt import Model, log

Just be careful that this is the natural logarithm, and you might need to do some basis change.

I've never done constraints the way you are doing (with constraint.append), so I'm unsure whether that might also be a problem.

EDIT: Oh, my bad, I misread your issue, you aren't actually using pyscipopt. I think this is more of a SCIP issue than a PySCIPOPt issue. I recommend sending an e-mail here: http://listserv.zib.de/mailman/listinfo/scip/

If you could, maybe attaching your problem in the e-mail would definitely be helpful. Without knowing more about it, maybe it's a problem with floating point arithmetic? Sorry for not being able to help out more, but the SCIP guys definitely will be able to.

[Opt-Mucca]

I don't think this is a PySCIPopt issue. @uvutna should raise this as an issue with cvxpy.

If I've understood the output correctly, then SCIP has not even been called at at the time of failure. The failure may also be because the instance is simply infeasible.