Originally posted by: yanzhiping

Excuse me,

I have a MIP problem as attached in mps,lp,sav for the same problem.

the .out file shows the process when solving the problem with cplex. it took around 2 hours to solve it. the problem size is

Reduced MIP has 308429 rows, 297532 columns, and 1093399 nonzeros.
Reduced MIP has 46989 binaries,

it shows some number issues found from data check, such as Detected righthand side <= CPX_FEAS_TOL at constraint, Decimal part of coefficient for variable in constraint looks like 5/77 in single precision..

Does the following warning indicates the problem is suffering from performance variability?

CPLEX Warning  1033: Detected  99.10% (2209) suspicious condition number(s) >= 1e+07.
CPLEX Warning  1034: Detected   0.90% (20) unstable condition number(s) >= 1e+10.

I also observed that it took a long time for the gap to go from 3.23% to 0.17% .

Is it important to fix these warnings? would it help improve the solver performance?

Thanks,

Originally posted by: yanzhiping

Here is an updated log file corresponding to the previous attached lp/mps/sav problem.

The data check warning about feasibility tolerance and the 1033/1034 warnings are concerning. They might be linked to performance variability, but the bigger concern for me would be the possibility that incorrect solutions could result. Search this forum, or the CPLEX documentation, for "numerical instability" to learn more.

Numerics can definitely be an issue here. Look at your problem statistics:

Problem name         : Mixplan1.sav
Objective sense      : Minimize
Variables            :  572544  [Nneg: 345053,  Box: 1143,  Free: 102004,
                                 Binary: 124344]
Objective nonzeros   :       2
Linear constraints   :  754218  [Less: 329970,  Greater: 70968,  Equal: 353280]
  Nonzeros           : 2166889
  RHS nonzeros       :  215813

Variables            : Min LB: 0.000000         Max UB: 1.000000       
Objective nonzeros   : Min   : 1.000000         Max   : 1.000000       
Linear constraints   :
  Nonzeros           : Min   : 4.000000e-06     Max   : 3914.298       
  RHS nonzeros       : Min   : 4.000000e-08     Max   : 48655.47

The very small coefficients/right-hand sides and the wide range of them will most likely cause numerical trouble.