BARON is stuck

GAMS
1

Hello

I am solving a NLP problem with around 300 constraints (linear and nonlinear) in BARON. The problem is that BARON seems to get stuck when I ask for tighter constraint tolerance (AbsConFeasTol). After 24 hrs of running, no improvement in the upper and lower bounds. Nothing in the problem formulation seems to be infeasible. In fact I used the output of BARON with more relaxed constraint tolerance (AbsConFeasTol=1e-5) as an initial guess but still the upper bound is 1e+52. Here is a sample of the output

Iteration Open nodes Time (s) Lower bound Upper bound
1 1 86.10 0.00000 0.100000E+52
5 3 127.89 0.00000 0.100000E+52
6 4 167.82 0.00000 0.100000E+52
12 7 209.01 0.00000 0.100000E+52
14 8 265.94 0.00000 0.100000E+52
61 31 313.80 0.00000 0.100000E+52
79 40 379.44 0.00000 0.100000E+52
103 52 412.32 0.00000 0.100000E+52
108 55 458.68 0.00000 0.100000E+52
109 55 501.58 0.00000 0.100000E+52
121 61 531.93 0.00000 0.100000E+52
157 79 581.79 0.00000 0.100000E+52
166 84 624.81 0.00000 0.100000E+52
193 97 655.01 0.00000 0.100000E+52



16990 8255 171793.04 0.00000 0.100000E+52
16999 8260 171941.71 0.00000 0.100000E+52
17017 8269 172099.75 0.00000 0.100000E+52
17026 8273 172249.53 0.00000 0.100000E+52
17035 8278 172335.77 0.00000 0.100000E+52
17044 8282 172487.45 0.00000 0.100000E+52
17053 8287 172641.90 0.00000 0.100000E+52
17062 8291 172792.15 0.00000 0.100000E+52
17071 8296 172941.27 0.00000 0.100000E+52
17080 8301 173001.49 0.00000 0.100000E+52
17098 8310 173151.41 0.00000 0.100000E+52
17107 8314 173302.94 0.00000 0.100000E+52
17116 8319 173371.63 0.00000 0.100000E+52


what could be happening? How do you get a more accurate solution beyond the default AbsConFeasTol = 1e-5?

2

What is the tighter constraint tolerance that you are trying? Tolerance cannot be less than 1e-12.

1e-5 is already a good enough tolerance. But if a solution is feasible for 1e-5 and infeasible for lower tolerances (which means it is truly infeasible), then it tells me that your model is poorly scaled. Provide better bounds, make sure that variables are in similar ranges, avoid having astronomical numbers in your model by using better units and so on…
This should fix the problem.

  • Atharv