Re: [Help-glpk] [Fwd: mip behavior]

[Andrew Makhorin]

> I'm a bit puzzled about the behavior of glpk when solving pure integer
>  programs. All my variables are binary and all the non-zero constants
>  are 1. I'm running the stand-alone version and here is what I get in
>  the late stage of the run:
> 
> +3032261: mip =   6.000000000e+00 >=   5.644706378e+00   5.9% (39730; 54174)
> +3037891: mip =   6.000000000e+00 >=   5.644706378e+00   5.9% (39645; 54442)
> +3044704: mip =   6.000000000e+00 >=   5.644706378e+00   5.9% (39579; 54703)
> 
> My interpretation of this output (documentation?) is that glpk found an
>  integer solution with objective 6, and found a lower bound of 5.644...
>  My question: why does it keep going -- sometimes for a very long time?
>  We know the objective is an integer and greater than 5, and a solution
>  with value 6 has already been found -- why doesn't glpk terminate?
> 

Currently the objective granularity check is disabled because of the new
mip preprocessor, which is still incomplete. Sorry.

If you know that your objective is integer, you may specify an
appropriate mip gap as Marcelo suggested (say, 0.06 for your example) to
stop the search once the optimum has been found.