Originally posted by: MichaelPoss

Dear all,

I am solving a LP with CPLEX 12.2 concert technology for JAVA.
If I understood well, a slack variable for a constraint "c^t x <= b" should always be negative. Hence I don't understand how the following can happen.
The following code

IloRange test = cplex.addLe(lin,0);
cplex.solve();
System.out.println(test);
System.out.println("last cut slack: "+cplex.getSlack(test));

returns

IloRange : -infinity <= 1.0*pi[0]28 + 1.0*pi[0]46 + 4.0*pi[3]22 + 4.0*pi[3]28 + 3.0*pi2228 + 1.0*pi2246 - 14.0*alpha <= 0.0
last cut slack: -17.699452416918497

Thanks in advance,

Michael.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: MichaelPoss

> MichaelPoss wrote:
> If I understood well, a slack variable for a constraint "c^t x <= b" should always be negative.

I meant non-negative of course.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

That indeed looks suspicious. What is the return value of cplex.solve()? Does it return true or false?
Did you change any parameter settings? Is this the first solve you do on the model or did you do other solves before?
What if you fetch the x vector from cplex and calculate the slack yourself?
Could you show us the log output of CPLEX?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: MichaelPoss

Thank you for your answer.
The output of cplex.solve() is FALSE.
Then, cplex.isPrimalFeasible() also returns FALSE, while cplex.isDualFeasible() returns TRUE.
Hence, I imagine that the "weird" slack values simply arise because the primal is not feasible, isn'it ?

Michael.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Indeed. If the solution is not primal feasible, then rows can be violated and slacks can be negative.

So, the question is: why is CPLEX not able to solve this problem to optimality?

Posting a log output could shed some light on this issue. Be sure to enclose the log output in "code" quotes.

Tobias
#CPLEXOptimizers
#DecisionOptimization