Re: [Help-glpk] Hard or very large scale MIP problems

[Brady Hunsaker]

On Thu, 2004-09-23 at 10:22, Bernardo Machado Reis wrote:
> 
> 
> Hello.
> 
> I'm working with glpk and trying to resolve very large scale MIP problems.
> I've alredy done some tests with many integer variables (about 26000 ). It
> takes about an hour but it allways gives a solution.
> In your documentation, in page 51, it says "Hard or very large scale MIP
> problems cannot be solved by this routine".
> May you say why? Are the solutions that I have incorrect? Are they not
> optimal solutions?
> Can I keep going work with glpk?
> 
> Thank you very much,
> Bernardo Reis
> 

If GLPK is indicating that it has found an optimal solution, then your
solutions are optimal.  I believe that the point of the disclaimer is
that GLPK does not yet implement some advanced techniques to deal with
large-scale MIPs, so many large MIPs will take much longer than they
would on other (commercial) solvers.  When the documentation says
"cannot be solved" it would probably be more accurate to say "may take a
really long time to solve or may run out of memory before solving". 

So you shouldn't *expect* GLPK to solve large MIPs quickly.  It sounds
like your instances are reasonably "nice", however, so that you do get
solutions.  If it's working well for you, then there's no reason not to
continue using it. 

Brady

-- 
Brady Hunsaker
Assistant Professor
Industrial Engineering
University of Pittsburgh
http://www.engr.pitt.edu/hunsaker/