Originally posted by: shchwang

I am using the ILOG CPLEX Interactive optimizier 12.5.1.0 to solve a MIP problem. Then, I got the following information:

CPLEX> optimize
Row 'c6' infeasible, all entries at implied bounds.
Presolve time = 0.00 sec. (0.40 ticks)

Root node processing (before b&c):
  Real time             =    0.01 sec. (0.52 ticks)
Parallel b&c, 32 threads:
  Real time             =    0.00 sec. (0.00 ticks)
  Sync time (average)   =    0.00 sec.
  Wait time (average)   =    0.00 sec.
                          ------------
Total (root+branch&cut) =    0.01 sec. (0.52 ticks)


MIP - Integer infeasible.
Current MIP best bound = -1.3125424058e+06 (gap is infinite)
Solution time =    0.01 sec.  Iterations = 0  Nodes = 0
Deterministic time = 0.52 ticks  (45.38 ticks/sec)

I tried to find the problem, so I enter "conflict" try to invoke the conflict finder, then I got the following information:

CPLEX> conflict

Refine conflict on 1194 members...

 Iteration  Max Members  Min Members
         1          896            0
         2          878            0
         3          874            0
         4          867            0
         5          866            0
         6          758            0
         7          752            0
         8          749            0
         9          655            0
        10          644            0
        11          638            0
        12          637            0
        13          635            0
        14          596            0
        15          587            0
        16          582            0
        17          580            0
        18          579            0
        19          437            0
        20          420            0
        21          411            0
        22          409            0
        23          310            0
        24          260            0
        25          235            0
        26          223            0
        27          217            0
        28          214            0
        29          212            0
        30          111            0
        31          108            0
        32          107            0
        33           12            0
CPLEX Error  1720: Conflict detection could not reproduce previously found infeasibility.
Failed to compute conflict.

Could anybody give me some suggestion how to solve this problem? Or how can I find the conflict constraints in the LP model. Thanks a lot. I also upload the input file as  an attachment, if you need it.

Originally posted by: T_O

dis prob stat yields:

Linear constraints   :
  Nonzeros           : Min   : 0.2033840        Max   : 1.000000e+014
  RHS nonzeros       : Min   : 4.312240e-008    Max   : 1.000000e+014

So you have coefficients in the range of 10^-1 to 10^14 and an even worse right hand side range. I would not rely on any results for this model. Can you reformulate your model so that there is a smaller range of coefficients?

Best regards,
Thomas

Originally posted by: shchwang

Thank you for the reply. Now I have reformulated my model, and now the dis prob stat yields:

Problem name: b17_10.lp
Variables            :     387  [Nneg: 60,  Box: 300,  Binary: 27]
Objective nonzeros   :      76
Linear constraints   :     534  [Less: 126,  Greater: 108,  Equal: 300]
  Nonzeros           :    1951
  RHS nonzeros       :     342

Variables            : Min LB: 0.000000         Max UB: 1.000010       
Objective nonzeros   : Min   : 38.50360         Max   : 207048.0       
Linear constraints   :
  Nonzeros           : Min   : 0.2033840        Max   : 1.000000e+07   
  RHS nonzeros       : Min   : 4.312240e-08     Max   : 1.000000e+07  

Is it better now? Or I need  a much smaller range of coefficients? (if it is possible, please give me a speicific value for the range.) The input file has been uploaded as attachment. Thanks a lot

Originally posted by: T_O

Though I think these ranges are still very large, at least the conflict refiner seems to work now.

I cannot give you a specific value, but typically, a small range behaves much more stable than a huge range.

Maybe you find a completely different model formulation that does not need such huge ranges.

Best regards,
Thomas