Issue with dual values for continuous model

Answered

Simon Dieck

• June 10, 2022 11:31

I have a continuous model with a linear objective function with a minimization sense.

My expectation of duality would be that given:

total = 0
for c in rm.getConstrs():
    total += c.RHS * c.Pi

"total" would be equal to the objective value, if strong duality holds, or at least less than the objective value, if weak duality holds.

However,

print(total - rm.ObjVal)

751391.7872288637

Which has me pretty confused. The Objective value of my model is:

Optimal objective  1.099028532e+07

I'm optimizing with default parameters. Is Gurobi maybe not updating the dual values to the final value after finishing optimization? And if this is the case, can I force this update? Or am I missing something else entirely?

0

• 

  Jaromił Najman

  Gurobi Staff

  • June 10, 2022 14:11
  • Edited

  In theory your code is correct. However, in practice, models are not in standard form, i.e., there are also variable bounds and objective constants you have to consider . The following code should do the trick

  total = m.ObjCon
  for c in m.getConstrs():
    total += c.RHS * c.Pi
  
  for v in m.getVars():
    if v.VBasis == -1:
      total += v.RC * v.lb
    if v.VBasis == -2:
      total += v.RC * v.ub

  Best regards, 
  Jaromił

  1
• 

  Zhiyuan

  • December 19, 2024 16:32

  

  I also had this problem, Jaromił  is correct. Please use GRB_INFINITY to define the UB of the variable if it is implicitly bounded by some constraints elsewhere.

  0

Please sign in to leave a comment.