[Help-glpk] Re: glpk mip

[Andrew Makhorin]

>A solved a scheduling problem with glpk that had binary variables like
>x[I,j,k] = 1 if the course i takes place in the timeslot j and room k .
>It has 100000 variables.

I think the glpk mip solver is *not* suitable for your problem due to
its very large size. Besides, problems of this class are intractable in
their nature and therefore, as a rule, very hard even for much more
powerful mip solvers.

> After pre_solve it has 25000 variables.

The presolver is used only on solving LP relaxation. It does not affect
the integer optimization phase, so the mip solver deals with all 100000
variables.

> In a
>few minutes it finds a solution saying optimal solution found, but then
>it goes to the second part , when it tries to find mip solution. There
>it doesn't find any solution(or at least it didn't in the last 30
>minutes,it is still running). Can I somehow speed up the process? Can I
>find any partial solution? If I tell it to run for x seconds, then it
>will put all variables to zero. 

Not seeing your problem it is difficult to say something reasonable.
If exact solution is not critical for your needs, you might try to solve
your problem with some heuristic methods (unfortunately, glpk does not
have such methods).

Andrew Makhorin