Originally posted by: SystemAdmin

[zhzhang said:]

I built a mixed integer quadratic model by concert. there is a error: cplex error 3019 when it is running.
However, if I export the model to a file and use interactive optimizer to solve it, it is no problem. Why?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[prubin said:]

Error 3019 means CPLEX was unable to solve one of the subproblems.  Either you bumped into a limit (time limit, iteration limit, ...) or else there was some numerical precision/stability problem.  I suspect the latter.  If you export the problem to a text file (MPS or LP format) and then read it into the interactive optimizer, you may not get exactly the same problem, because double-precision values are rounded/truncated during the conversion to/from text.  Also, unless the parameter settings are identical between the two runs, CPLEX might follow a different search path in the interactive optimizer, and might never trip over the subproblem causing the error.

You might try loosening tolerances and see if that helps.  Also check your model for bad scaling.

/Paul
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[zhzhang said:]

Thank. But how to do it. I am a newcomer.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[EdKlotz said:]

If you export the problem to a text file (MPS or LP format) and then read it into the interactive optimizer, you may not get exactly the same problem, because double-precision values are rounded/truncated during the conversion to/from text.

IloCplex::exportModel also allows you to export a binary SAV file, which does give you the same problem as in your program.  I suspect you exported an MPS or LP file;
try a SAV file and see if you can reproduce the behavior with interactive CPLEX on the SAV file.  While you cannot read the SAV file in a text editor, you can write out
an LP file from the SAV file if you need to do that (or just call exportModel twice).

Look at the section on numerical issues in the CPLEX user guide for more information on numerical problems.  Also, look at the FAQs at support.ilog.com.  Also, try
setting the scale parameter to 1; that may help on models with numerical problems
(although if you can address the numerical issues in the model formulation itself,
that is better).


#CPLEXOptimizers
#DecisionOptimization