[Coin-discuss] quadratic integer solver?

[Stefan Vigerske]

Hi,

> what are my options to solve a quadratic mixed integer (binary) problem? 
>   (I mean a problem with quadratic elements in the objectives and only 
> linear constraints.)

Is the objective function convex?

> 1) calling from AMPL
> It looks like the choice is between Bonmin and LaGo - or is there 
> anything else?

If it's convex, then Bonmin is a good choice, I think.
If it's not too large or too dense, then also LaGO could work. In that 
case you do not need a convex problem.

> 2) calling from C++ - this is the preferred option, as I will be using 
> it inside another alg. I have some experience with using LP solvers via 
> OSI, but have never been into NLP problems before. Can I use OSI and 
> just add info about which variables have quadratic terms in the 
> objective, or do I have to use some completely different interface?

As far as I know, Cbc cannot handle quadratic mixed-integer problems, 
but maybe I'm wrong.
For Bonmin you can use their C++ Interface. It shouldn't be difficult to 
use since your functions are relatively "easy" (constant jacobian and 
hessian).

Best,
Stefan

-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan