Originally posted by: MarcoMojana

Hi everybody,

When I solve the attached problem with the interactive optimizer with the following commands:
read original.lp
set emphasis mip 3
set mip strategy probe 3
mipopt

it gives me:
MIP - Integer optimal solution: Objective = 7.2890000000e+03

Now, repeating the process adding:
set mip tolerances uppercutoff 7289

gives me:
MIP - Integer infeasible.

The same happens using a cutoff of 7290. With values of 7291 or higher the solver is able to find the solution. Setting the cutoff to 7289 and removing the emphasis and probe settings results in the correct solution. My question is: is there a tolerance parameter that can be used to solve the problem (I have already tried "set mip tolerances absmipgap 0" and "set mip tolerances mipgap 0")? Otherwise, how can I trust the obtained results? I would like to avoid doing some sort of cutoff = (1.0 + 1.0e-3)(realCutoff + 2)

Marco Mojana
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

This sounds strange. As far as I know we are not exploiting the user cutoff in presolve, so I do not think that this is related to presolve or probing. Are you getting the infeasibility message already after presolve, or is CPLEX solving the root node (or even some more nodes)?

Could you please post the relevent parts of the log files?
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: MarcoMojana

Here are the logs. I'm using the CPLEX interactive Optimizer 12.1.0 on a linux 32bit machine.

Best Regards,

Marco Mojana
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Looks like something is going wrong with the numerics, either in probing or in the LP solve. Does this still happen with a newer version of CPLEX (12.2.0.2 or 12.3)?

Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: MarcoMojana

With CPLEX 12.2 everything works as expected... I cannot find the version 12.3

Thank you for your time

Marco Mojana
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

We have released 12.3 a few weeks ago, but I think that the Academic Initiative version is not yet online.
#CPLEXOptimizers
#DecisionOptimization