Using "addGenConstrPow" but model is infeasible

Answered

Jack Kaz

November 17, 2022 11:06 Edited

• 
• 
• 

Hello together,

I am trying to solve an easy power function for "x = 20 + A"  with A=200, but it is infeasible. I do not understand why. If I set my variable A = 0 it is feasible. Thank you.

Here is my example:

model = gp.Model("test")

A = 200

x0 = model.addVar(vtype=GRB.CONTINUOUS, name="x0")
x = model.addVar(vtype=GRB.CONTINUOUS, name="x")
x_pow4 = model.addVar(vtype=GRB.CONTINUOUS, name="x_pow4")

model.addConstr(x0 == 20)
model.addConstr(x == x0 + A)

model.addGenConstrPow(x,x_pow4,4)

model.optimize()

Related to

• Numerics
• General Constraints

0

• 

  Jaromił Najman

  • Gurobi Staff

  November 17, 2022 11:42

  Hi Jack,

  The issue is caused by the default value of the FuncMaxVal parameter which is 1e+6. This limits the bounds on variables participating in general constraints, but \(220^4 > 10^6\). Thus, your model is declared as infeasible.

  To avoid this, you have to set the parameter to a larger value, e.g.,

  model.setParam("FuncMaxVal",10e+20)

  Please note that working with such large values is not recommended. So, if you want to extend your model and indeed work with the value of \(\texttt{x_pow4}\) in an extension of your model, then this might very likely lead to numerical issues.

  Best regards, 
  Jaromił

  0
• 

  Jack Kaz

  November 17, 2022 12:34

  • 
  • 
  • 

  Hi Jaromil,

  thank you very much for your advice!

  Unfortunately, we have this big values in thermodynamics when working with absolute temperatures.. Is there a suggestion to reduce the possibility of numerical issues? I was thinking of a way to limit the value range of the function.. or is it done when defining upper and lower bounds to the variables?

  0
• 

  Jaromił Najman

  • Gurobi Staff

  November 17, 2022 13:42

  Hi Jack,

  Unfortunately, we have this big values in thermodynamics when working with absolute temperatures.. Is there a suggestion to reduce the possibility of numerical issues? I was thinking of a way to limit the value range of the function.. or is it done when defining upper and lower bounds to the variables?

  I understand. You could re-scale the orders of magnitude, e.g., say that 1 unit for some variable equals 1000 degrees instead of 1. This would already scale the values by 4 orders of magnitude. For more detailed tips, please have a look at our Guidelines for Numerical Issues.

  0

Please sign in to leave a comment.