[Help-glpk] Re: The GLPK questions

[Andrew Makhorin]

Thank you for your interest in glpk.

>I'd like to express my feeling from using GLPK. I've been testing the soft 
>for a while mostly on big SetCovering/SetParting problems. It seems the 
>GLPK has quite stable and performing LP solver (good job :-) ), but MIP is 
>a bit simplistic and I feel there is a potential for improvement. 

Yes, you are right. The mip solver currently implemented in glpk is
based on the branch-and-bound method with an easy branching heuristic,
so it is inefficient for most interesting mip problems. Besides, many
useful features (preprocessing, presolving, etc.) are not implemented
yet.

>Commercial solvers use quite some techniques for making MIP perform well, 
>namely lifted cover cuts, Gomory cuts. According to reports these improved 
>performance dramatically. Do you think there will be some effort for 
>making these being part of GLPK ? Would you be able to accept any form of 
>help implementing these techniques ?

glpk has a component, which allows generating cutting planes on
solving mip problems (in particular, the b&b mip solver is implemented
using that component; however, cutting planes are not used there). So,
if you are interested in implementing some classes of cutting planes
(or other things related to mip), please inform me, and I will provide
you with a necessary information.

Andrew Makhorin