3

We have a complex optimization problem which includes several quadratic terms with integer and continous variables (using Anaconda Python / Pycharm with Gurobi 6.0.2). We applied the setPWLObj function to apprixmate the quadratic objective components. The code for this is as follows:

    m.addConstr(l1[t] == 1/2.0 * (hsqrt[t]+hQ[t]))
    m.addConstr(l2[t] == 1/2.0 * (hsqrt[t]-hQ[t]))
    hlx1 = linspace(-10, 10, 50)  
    hlx2 = linspace(-10, 10, 50)
    h1y1 = [0]*50
    hly2 = [0]*50

    for i in range(len(hlx1)):
        h1y1[i] = hlx1[i] * hlx1[i] * 7.348 / 1000.0
        hly2[i] = -hlx2[i] * hlx2[i] * 7.348 / 1000.0

    m.setPWLObj(l1[t], hlx1, h1y1)
    m.setPWLObj(l2[t], hlx2, hly2)

With l1 and l2 being continous variables.

The problem behaves inconsistently. Running it on a Mac mostly delivers the exit codes 138 and 139 (correspondent to Bus Error 10), sometimes the same problem a solution can be calculated. This is particularly the case when starting the optimization several times in a row. This appears to be random.

On Windows machines either Python crashes, or the exit code "-1073741819" is delivered. Further searches for this exit code did not really help us.

double-beep
  • 5,031
  • 17
  • 33
  • 41
Alexander Schuller
  • 59
  • 5

1 Answers1

1

Sorry for taking so long, but we fixed the issue.

Actually we found out that the python crash is or was due to a bug in Gurobi. Following a request we filed with them the bug was removed.

If Gurobi 6.0.3. or above is used the error is not there anymore.

Alexander Schuller
  • 59
  • 5