Imprecision in Gurobi

Answered

Namrita Varshney

• August 12, 2024 09:31
• Edited

gbm.addConstr(f <= t + (1 - d) *M - e, f 'B1')

gbm.addConstr(f>=t-d*M, f 'A1')

 

this create equation
B1: f + 1e+09 d
   <= 1.0000000003888891e+09
A1: f + 1e+09 d >= 0.388889015
I have taken 

M=1e9

e=1e-12

Do you have any suggestions on resolving this, or am I wrong somewhere?

 

f=0.3888890743255615

t=0.388889015

and gurobi is setting it to 
d = 0.9999999999999999

Also I noticed that , the equation set to correctly for one variable i.e d=0 , and then for second variable , having same values , same equation , It is setting 0.9999999 

0

• 

  Jaromił Najman

  Gurobi Staff

  • August 13, 2024 07:31

  Hi Namrita,

  This is usual behavior for large values and finite precision. Short answer: you should try to reduce the value of your big-M. For the full answer, please have a look at

  • Why does Gurobi sometimes return non-integral values for integer variables?
  • Dealing with big-M constraints
  • Gurobi tolerances and the limitations of double-precision arithmetic

  Best regards, 
  Jaromił

  0

Please sign in to leave a comment.