Help me in Solving a large scale model..

[Sizhe Zhang]

 I have some questions regarding the solver CPLEX in the AIMMS platform. I hope you can help me to clarify this.

I am running my model with totally the number of decision variables of more than 40,000 in 200 instances.

The results show that most instances (176) can be solved efficiently, within about 200s. However, for the other instances, the optimal cannot be achieved after acceptable computation time (2 hours), i.e., the gap still exists which cannot be reduced anymore.

I wonder why this is the case. That is to say, for such large scale examples, under what conditions CPLEX can efficiently generate global optimal or cannot achieve global optimal within acceptable time.

Is that because of the parameter setting? I am not quite sure about the algorithm used in CPLEX. It will add cutting planes at each node, right? Is the efficiency dependent on the tightness of the cut? 

If I want to explore the reasons why CPLEX performs differently for different instances, do you have any suggestions? Furthermore, where I can find the log file of how CPLEX solves the model?

Thank you very much for your time. 

[Marcel Hunting]

Hi,

CPLEX uses a branch-and-cut algorithm to solve MIP problems, so CPLEX will indeed (try to) add cutting planes at each node.

In general you cannot predict whether CPLEX will be able to solve a MIP problem efficiently or not. It helps if the LP relaxation is tight.

To print the CPLEX log file you have to set the general solvers option 'Solver Listing Messages' to 'All' and the CPLEX option 'MIP Display' to a non-default setting, e.g., 'Display each nth node'. The CPLEX log file will be printed in the log folder in the AIMMS project folder.

Best regards,

Marcel Hunting

AIMMS Optimization Specialist

[Sizhe Zhang]

Hi Marcel,

Thanks a lot for your reply. That's really helpful.

Best regards,

Sizhe