Originally posted by: Falcon_G

Hi everybody

I have a branch-and-cut where in I separate cuts several cuts. The polytope description is not complete (some of the cuts are added in the lazy constraints callback).

for none of the solutions reported by the heuristic (with * besides) the cutcallback or lazycallback are not invoked. Therefore, some of the incumbents are feasible solutions when taking into account the cuts that are must be separated in the lazyconstraints callback. Even the optimal solution is not  feasible in such cases.

The only thing I can imagine is that I try to reject the proposed incumbent, separate the cuts in the incumbent callback and add them via lazy constraints callback or cutcallback.

is there any intelligent way to do it in a more straightforward way?

Thanks a lot in advance.

*    38     5      integral     0   468940.0000   463436.5517     3882    1.17%
     38     4   468940.0000     0   468940.0000   463436.5517     3882    1.17%       z(18)(10) N     38     37     21
Elapsed time = 130.95 sec. (3493.16 ticks, tree = 0.01 MB, solutions = 2)
incumbent written
*    45     5      integral     0   466910.0000   464006.0000     4613    0.62%

Zero-half cuts applied:  12
Lift and project cuts applied:  4
Gomory fractional cuts applied:  6
User cuts applied:  122

The '*' in the log indicates that the solution was found from an integral node, not from a heuristic.

But anyway, it is a bug in CPLEX if the lazy constraint callback is not invoked for such a solution. You should get a chance to cut off that solution.

What version of CPLEX do you use?

The workaround with incumbent+lazy callback is the best thing you can do, I think.

Originally posted by: Falcon_G

Thank you.

I use 12.6.2 64Bit

best

There used to be a bug that showed exactly this symptom but this bug is fixed in your version.

I carefully looked at the source code but found nothing that would explain the behavior you observe. Do you have an example code that you could provide here or to daniel(dot)junglas(at)de(dot)ibm(dot)com? I would like to get to the bottom of that.