Formulation of LPs

Hi, everyone

I'm currently working on a two-stage program. I use Benders decomposition to decompose my problem into two smaller ones and I further decompose the resulting subproblem in scenarios.
These scenarios are linear programming problems. I've shared the primal and dual versions of one of these scenarios, respectively, primal.mps and dual.mps.
Along with the mps files, I've attached the gurobi log for both problems.

As you can see by the log, despite being two versions of the same problem, one takes considerably more time to solve than the other.

I guess this is an expected behaviour since, from a computational viewpoint, the primal and the dual are actually two different problems.

However, this behaviour intrigues me and prompts my questions.

Taking this computational point of view, how could I improve the formulation of my LP problem, either the dual or the primal, so as to better exploit the potential of Gurobi?

My idea is to reformulate my problem to get the best possible result from Gurobi.

As of today, since I do not need an exact solution to the LP and thus I turn off the crossover iterations, my version of choice is the dual one because Gurobi's barrier method solves it quicker than it solves the primal.

I appreciate any suggestions and look forward to what you have to say.

Thanks in advance,

Bruno

https://drive.google.com/open?id=1AwiGOhW0hgLqV-yl8nxjQgNGbWYA41tQ

Formulation of LPs

Comments

Didn't find what you were looking for?