Sunday, November 22, 2009

Some GAMS/Gurobi message issues

A client of mine purchased GAMS/Gurobi, and tried his license on a small LP problem and immediately sent me an email with “help my license does not work”. The output showed:

Gurobi demo license.
Gurobi library version 2.0.1

That looks like the license was not recognized. Turns out that for small models (within the demo limits) this message is issued. This is really confusing.

And what about this message: when solving an unbounded model, the screen log mentions:

Solved in 0 iterations and 0.00 seconds
Infeasible or unbounded model
LP status(4): Model was proven to be either infeasible or unbounded.

Come on, make up your mind! For a modeler it makes a lot of difference if the model is infeasible or unbounded and when writing algorithms it is even more important to know if the sub-model was infeasible or unbounded. I would prefer the solver (by default) spends some more time (e.g. by restarting using a primal algorithm) to give a proper diagnostic message.

Even worse, the listing file mentions:

**** SOLVER STATUS     1 Normal Completion        
**** MODEL STATUS      19 Infeasible - No Solution
**** OBJECTIVE VALUE                0.0000

which is wrong: the model is unbounded! Also if the model were infeasible it would be good to report a solution with minimum infeasibility or some other hints: just “no solution” is not very helpful for a modeler.

Note: these issues are mostly not related to Gurobi per se, but rather to the GAMS link implementation.

1 comment:

  1. Reporting a solution with minimum infeasibility for infeasible problems is quite hard.

    Moreover, I will even claim that in most cases it is not what you want. For instance
    how do you use that in column generation algorithm.

    Indeed what you want is the Farkas certificate of infeasibility. This tells you a about
    how to repair the infeasibility. Also you can use the certificate to verify the optimizer
    is correct.

    Shameless advestisement:

    I suggest you also try MOSEK for unbounded and infeasible problems. We have work very hard
    on dealing with these situations in a stringent manner. The

    describes the idea.

    The above does not applies to MIPs of course.