Big M problem in Python-Gurobi

Bob Pay

unread,

Mar 26, 2016, 7:27:47 PM3/26/16

to Gurobi Optimization

Dear all,

I am solving this MIP model with gurobi-python interface. I had a binary-continuous multiplication, so I linearized it and in result, I have a big M in my model that is upper bound for the continuous variable. Although I found a reasonable upper bound for big M ( that is less than 1) and it sounds that everything is working, when I try with much larger values (like 100 and 1000), to make sure that everything else is right, the model crashes and the final solution is fractional for binary variables. Now my question is, how is it possible? Because even if there is a problem, it should happen for smaller value not larger value. do you think this problem is coming from formulation side or gurobi side?

I really appreciate your help.

Cheers,

Babak.

Kostja Siefen

unread,

Mar 30, 2016, 4:52:36 AM3/30/16

to Gurobi Optimization

Can you be more specific about your observations? What does "model crash" mean? Is the fractionality you see within integer tolerances (IntFeasTol)?

Kostja

Hugh Yu

unread,

Apr 2, 2016, 12:11:50 PM4/2/16

to Gurobi Optimization

I have the same problem. Finally, I found the cause. Gurobi cannot handle a big M that is too large in a linear constraint. For example,I found that Gurobi could accept a big M less 1e15 but could not handle a big M of 1e20. So defining M as a smaller number such as 1e10 would make Gurobi work well.