Originally posted by: alib

Hi,

I used OPL 12.6 solver for a linear program which all decision variables are fractional. All the parametres of .ops file are switched to default.

I noticed that when adding redundant constraints, the running time is reduced drastically especially for large scale ones (the number of constraints is much more larger than the number of variables).

I used to know that adding valid inequalities speeds up mixed integer programs because it strenghthens the formulation and improves the quality of lower bounds used by the branch-and-bound algorithm. But in my case, the linear program has only fractional variables.

The engine log displays that cplex ran dual simplex, then barrier.

Is there any explanation on how redundant constraints (implied by other constraints) reduces the computational time for pure linear programs?

Originally posted by: RWunderling

By and large you are correct that a larger problem tends to take longer to solve. That said, any particular case may be different and without looking at the details there is no way of telling what could be causing this.  It could be:
- Just random luck
- The added constraints give better guidance to the algorithms or improve its numerical properties.
- The added constraints reach a threshold by which CPLEX may decide to dualize the problem.  Try setting the presolve dual parameter by hand.

- In general the different problem statistics may change decisions CPLEX takes internally.

Roland