Tobias Achterberg
unread,Jul 17, 2016, 3:44:27 PM7/17/16Sign in to reply to author
Sign in to forward
You do not have permission to delete messages in this group
Sign in to report message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to gur...@googlegroups.com
Your model seems to include severe numerical challenges. The values in the order
of 1e-9 in your coefficient matrix are most likely the reason for this behavior.
As you can see, the crossover step struggles a lot to find an initial basis (for
example, it says it needed to restart crossover, and that 3 variables dropped
from the basis).
In theory, after the push phase is done you should have reached an optimal basis
and the solve should be complete. But in practice, this is often not the case
due to numerical difficulties, and a clean-up phase is executed. In your case,
this clean-up struggles again a lot, even losing primal feasibility. You can see
that the primal infeasibility suddenly goes from zero to 1e+3. Most likely,
Gurobi switched from primal to dual simplex at this point because primal simplex
could not terminate for numerical reasons.
Overall, I doubt that there is much for you to do in terms of tuning the
algorithm (e.g., by means of parameter settings). Instead, you need to work on
your model formulation to come up with a model that is numerically more stable.
Tobias