Originally posted by: giask

I use cplexbilp , by changing the order of the constraints (i.e. moving up or down one constraint line and its corresponding rhs) I get different results.

These are always satisfactory because my problem is very underdetermined but I like to have a tech explanation of this behavior

thank you in advance

giask

You get different x-vectors but the same objective function value, right?

The order of constaints or variables may influence the path the solver takes through the search tree. This in turn may result in different optimal x vectors if there is more than one optimal solution.

Originally posted by: Javier Nieto

I have the same case for a MIP model, changing the order of the constraints in the OPL code, the time to find a feasible solution considering an order of the constraints, CPLEX lasts longer than one hour and changing the order CPLEX finds a feasible solution in less than half hour.

I would have a technical explanation of this behavior and know what can be done to improve this behavior in CPLEX

ATT

Javier Nieto

This is expected and is known as "performance variability". The order of constraints or variables can make a big difference in solution time to MIP solvers.

Consider a problem with only binary variables and a heuristic that scans through the variables and sets a variable to 1 if that is still feasible and 0 otherwise. If you provide the variables in the "right" order then this heuristic will immediately find the optimal solution. If you chose a "bad" ordering then the heuristic may not even find a feasible solution. MIP solvers contain a lot of sub-procedures like the above that depend on the ordering of constraints and variables.