Originally posted by: orange2005

Dear all,

I am coding a problem with CPLEX12.2 and c++.

I have two real number three dimensional variable matrices x,y, both can take values between 0 and 1.

One of my constraints is

model.add( (x[0][0][3]+x[0][1][4])<=y[0][0]24);

But the LP result gives x[0][0][3]=0.0001, x[0][1][4]=0.0001, but y[0][0]24=0.

Could you please tell me what is wrong with that constraint?

Thanks a lot,
Juzi
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

> model.add( (x[0][0][3]+x[0][1][4])<=y[0][0]24);

That is strange. The constraint does look fine. Is the value of y being printed with the same level of precision as x? (to make sure that its not a display issue)

> But the LP result gives x[0][0][3]=0.0001, x[0][1][4]=0.0001, but
> y[0][0]24=0.
> Could you please tell me what is wrong with that constraint?

Can you check for the result status at the end of solve() by printing out the value of 'cplex.getStatus();'. It is possible that the status was infeasible or unbounded in which case the values are not valid.

The other thing to check would be to see if you increased the simplex feasibility tolerance (EpRHS) from its default value of 1e-6 to 1e-4 or something like that. This parameter would allow the solution to violate the variable bounds by a certain degree and increasing the tolerance could explain the results.

Numerical trouble while solving the model could explain the bound violations as well. This would only be applicable if the model is ill-conditioned (have very high or very low co-efficients/bounds etc), since the limitations of finite precision calculations could lead to significant round off errors during the computations.

I hope these tips help you diagnose further.
#CPLEXOptimizers
#DecisionOptimization