Page 1 of 1

CPLEX stuck "Removing shift" on MILP

Posted: Tue Feb 14, 2017 7:19 pm
by cladelpino
Hi all. We are trying to solve a MILP, but a particular instance gets consistently stuck at a point in the solve.

Looking at the detailed log we can see that the process is stuck while performing the "Removing shift" step after achieving dual infeasibility for the node. It just keeps writing this to the screen.

Is there any known bugs with this symptom ? Any way we can generate more detailed information ?

Best regards!!!
Claudio

PS: I will crosspost this to the CPLEX forum. Sorry for that !

Re: CPLEX stuck "Removing shift" on MILP

Posted: Wed Feb 15, 2017 3:04 pm
by dirkse
Claudio,

If this is something you can conveniently pack up and send to GAMS Support, that would be my recommendation. This sounds like something CPLEX would want to address, but the behavior may depend on the GAMS/CPLEX link so I wouldn't recommend passing this to them directly.

I was a little confused though by your description. You mention the solver achieves dual infeasibility for the node. If the child nodes are formed by adding bounds to a feasible parent node, how could the nodes be dual infeasible?

-Steve

Re: CPLEX stuck "Removing shift" on MILP

Posted: Thu Feb 16, 2017 3:30 pm
by cladelpino
Steve: Thanks for your reply, we will see if we can manage the time to get this to a reproducible case but I'm guessing this will be difficult since its a very hard problem and solve takes very long so it will be hard to pinpoint the problem.

I'm sorry for my conceptual mixup. I had misunderstood the purpose of the perturbation done by cplex. I see now that it is to circumvent stalling caused by degeneracy. Thanks for pointing it out. If you have a reference for this "shifts" cplex removes, I would be grateful.

I attach the log of the exact moment of the issue.
log.txt
(5.45 KiB) Downloaded 366 times
Thanks !!
Claudio

Re: CPLEX stuck "Removing shift" on MILP

Posted: Thu Feb 16, 2017 3:37 pm
by bussieck
This truly looks like a Cplex bug or numerical issue. If you add options

writesav prob.sav
writeparam param.prm

to your GAMS/Cplex option file you could submit these two files together with the entire log to support@gams.com or directly to the Cplex folks. Please also provide information about the platform (Windows/Linux/... 32bit/64bit/, CPU type, ...) so it becomes easy to reproduce.

-Michael