Originally posted by: HendrikEwe

Hi,

we just came across a curious case of a constraint which is violated using Cplex 12.1 and which is not violated using various older versions (e.g., Cplex 11.2).
You can see from the attached LP-file that

IloC1249: y0 = 1

Hence, by constraint k0, at least one of y1, ..., y190 has to be 1. This is not the case with the Cplex 12.1 solution.
Is this a bug or are we missing something here..?

Thanks for your help,

Hendrik Ewe
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: T_O

At the moment, I do not have 12.1, but 12.4. returns the following:
Cover cuts applied: 1
Zero-half cuts applied: 31
Gomory fractional cuts applied: 6

Root node processing (before b&c):
Real time = 1.81
Parallel b&c, 32 threads:
Real time = 1687.35
Sync time (average) = 0.34
Wait time (average) = 0.01

---

Total (root+branch&cut) = 1689.16 sec.

Solution pool: 10 solutions saved.

MIP - Integer optimal, tolerance (0.0001/1e-06): Objective = 1.3416217275e+05
Current MIP best bound = 1.3417020319e+05 (gap = 8.03044, 0.01%)
Solution time = 1689.56 sec. Iterations = 13595703 Nodes = 135012 (4)
Deterministic time = 903161.37 ticks (534.55 ticks/sec)

CPLEX> dis sol var *
Incumbent solution
Variable Name Solution Value
x37 1.000000
x69 1.000000
x102 1.000000
x136 1.000000
x168 1.000000
x202 1.000000
x234 1.000000
x266 1.000000
x298 1.000000
x330 1.000000
x362 1.000000
x394 1.000000
x426 1.000000
x458 1.000000
y10 1.000000
y21 1.000000
y0 1.000000
y491 1.000000
All other variables matching '*' are 0.

Best regards,
Thomas
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: T_O

Here is the result from another computer (also CPLEX 12.4):
Cover cuts applied: 1
Zero-half cuts applied: 23
Gomory fractional cuts applied: 7

Root node processing (before b&c):
Real time = 1.63
Parallel b&c, 4 threads:
Real time = 3396.71
Sync time (average) = 0.03
Wait time (average) = 0.00

---

Total (root+branch&cut) = 3398.33 sec.

Solution pool: 14 solutions saved.

MIP - Integer optimal, tolerance (0.0001/1e-06): Objective = 1.3404697305e+05
Current MIP best bound = 1.3406034899e+05 (gap = 13.3759, 0.01%)
Solution time = 3398.70 sec. Iterations = 8765254 Nodes = 148807 (215)
Deterministic time = 2015711.14 ticks (593.08 ticks/sec)

CPLEX> dis sol var *
Incumbent solution
Variable Name Solution Value
x35 1.000000
x67 1.000000
x99 1.000000
x135 1.000000
x167 1.000000
x202 1.000000
x234 1.000000
x266 1.000000
x298 1.000000
x330 1.000000
x362 1.000000
x394 1.000000
x426 1.000000
x458 1.000000
y2 1.000000
y15 1.000000
y0 1.000000
y491 1.000000
All other variables matching '*' are 0.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: T_O

Here is a third (and my last) result from a third PC:

Cover cuts applied: 1
Zero-half cuts applied: 25
Gomory fractional cuts applied: 8

Root node processing (before b&c):
Real time = 0.88
Parallel b&c, 8 threads:
Real time = 4022.78
Sync time (average) = 0.05
Wait time (average) = 0.01

---

Total (root+branch&cut) = 4023.66 sec.

Solution pool: 19 solutions saved.

MIP - Integer optimal, tolerance (0.0001/1e-06): Objective = 1.3405977270e+05
Current MIP best bound = 1.3407313647e+05 (gap = 13.3638, 0.01%)
Solution time = 4023.91 sec. Iterations = 20429166 Nodes = 86777 (392)
Deterministic time = 2712435.16 ticks (674.08 ticks/sec)

CPLEX> dis sol var *
Incumbent solution
Variable Name Solution Value
x26 1.000000
x72 1.000000
x104 1.000000
x136 1.000000
x168 1.000000
x201 1.000000
x233 1.000000
x266 1.000000
x298 1.000000
x330 1.000000
x362 1.000000
x394 1.000000
x426 1.000000
x458 1.000000
y1 1.000000
y17 1.000000
y63 1.000000
y0 1.000000
y491 1.000000
All other variables matching '*' are 0.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: HendrikEwe

Hi Thomas,

thanks for your help.
The results of Cplex 12.4 are a bit different, but also in all three cases you described constraints seem to be violated. For example, in your last post you y63==1 and hence by constraint k63 (see mipex1.lp) one of y64,...,y263 should be set to one, too.
Does that make sense? :-)

Best regards,

Hendrik.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: T_O

Hmm, I guess, there are numerical problems. It also failed with numerical emphasis enabled. (cf. MILP slack bound error)

CPLEX> dis sol qual
Incumbent solution:
MILP objective 1.3420057275e+05
MILP solution norm |x| (Total, Max) 1.90000e+01 1.00000e+00
MILP solution error (Ax=b) (Total, Max) 7.32015e-14 2.65413e-16
MILP x bound error (Total, Max) 1.57697e-13 2.09589e-14
MILP x integrality error (Total, Max) 3.15399e-13 2.09832e-14
MILP slack bound error (Total, Max) 1.00000e+00 1.00000e+00

Branch-and-cut subproblem optimization:
Max condition number: 6.8972e+18
Percentage (number) of stable bases: 88.29% (3241)
Percentage (number) of suspicious bases: 3.16% (116)
Percentage (number) of unstable bases: 6.56% (241)
Percentage (number) of ill-posed bases: 1.99% (73)
Attention level: 0.039896
CPLEX encountered numerical difficulties while solving this model.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Strange. Using CPLEX 12.4, I get a solution that seems to be okay.

With the solution that violates a constraint, what output did you get with "display solution quality" in the interactive?
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: HendrikEwe

This is really strange indeed: I re-ran the problem last night and now got different results!

CPLEX> dis sol var *
Incumbent solution
Variable Name Solution Value
x22 1.000000
x54 1.000000
x86 1.000000
x118 1.000000
x160 1.000000
x192 1.000000
x224 1.000000
x256 1.000000
x288 1.000000
x330 1.000000
x362 1.000000
x394 1.000000
x426 1.000000
x458 1.000000
y1 1.000000
y11 1.000000
y151 1.000000
y321 1.000000
y0 1.000000
y491 1.000000

Which is totally fine. Also the errors are different:

CPLEX> display solution quality
Incumbent solution:
MILP objective 1.3272857279e+005
MILP solution norm |x| (Total, Max) 2.00000e+001 1.00000e+000
MILP solution error (Ax=b) (Total, Max) 0.00000e+000 0.00000e+000
MILP x bound error (Total, Max) 0.00000e+000 0.00000e+000
MILP x integrality error (Total, Max) 0.00000e+000 0.00000e+000
MILP slack bound error (Total, Max) 0.00000e+000 0.00000e+000

I think last time I only used one thread but this time eight. I will try again with only one...
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: HendrikEwe

I don't get it - here are the results for one thread:

CPLEX> dis sol var *
Incumbent solution
Variable Name Solution Value
x22 1.000000
x54 1.000000
x70 -0.000000
x77 -0.000000
x78 0.000000
x86 1.000000
x96 0.000000
x102 -0.000000
x109 -0.000000
x118 1.000000
x134 0.000000
x160 1.000000
x192 1.000000
x224 1.000000
x256 1.000000
x288 1.000000
x328 -0.000000
x330 1.000000
x362 1.000000
x394 1.000000
x426 1.000000
x458 1.000000
y1 1.000000
y70 0.000000
y101 -0.000000
y142 -0.000000
y151 1.000000
y159 -0.000000
y320 -0.000000
y321 1.000000
y322 0.000000
y333 0.000000
y343 0.000000
y0 1.000000
y491 1.000000
All other variables matching '*' are 0.

CPLEX> dis sol qual
Incumbent solution:
MILP objective 1.3272857279e+005
MILP solution norm |x| (Total, Max) 1.90000e+001 1.00000e+000
MILP solution error (Ax=b) (Total, Max) 8.33579e-014 4.95635e-016
MILP x bound error (Total, Max) 8.75369e-011 2.38737e-011
MILP x integrality error (Total, Max) 1.63137e-010 2.38737e-011
MILP slack bound error (Total, Max) 2.38740e-010 2.38738e-011

Are there some other settings that I might have changed unintentionally? I will further investigate...
#CPLEXOptimizers
#DecisionOptimization