cplex 12.6.3 vs 12.7 for a MISOCP model

[hosein.h...@gmail.com]

hi,

I am running a MI-SOCP model with cplex 12.6.3 on my PC and it solve in 5 sec with "optimal integer solution" message. however, when I run the model on Neos (with Cplex 12.7) it gives an infeasible problem message. 

Which output I can trust? optimal or infeasible? is there any setting that I can make sure I get a true optimal solution?

[Robert Fourer]

I always start troubleshooting a problem like this by double-checking that the same problem is being solved. Set "option show_stats 1;" and check that the stats displayed by AMPL are the same on both your PC and NEOS. Also set "option cplex_options 'mipdisplay=2';" (or add mipdisplay=2 to your cplex_options string) and check that at the beginning of the CPLEX log, the same problem sizes are reported; also verify that the same option selections are listed at the beginning of the CPLEX listing.

If that checks out then I would try running the same problem through NEOS using the Gurobi and Xpress solvers, to see whether they agree with CPLEX 12.6.3 or CPLEX 12.7 (or neither). If you have access to Gurobi and Xpress on your PC then try them there, too. Then post your findings which should give some clue as to where we should be looking for the cause of the discrepancy.

Bob Fourer
am...@googlegroups.com

=======