[Bug-glpk] Re: mip bug

[Andrew Makhorin]

>I seem to have run into a mip bug.
>
>For the attached problem, the LP optimal is 0,
>which is confirmed by cplex.  cplex also finds
>an integer solution with objective=0.  but glpsol
>claims to find an optimal integer solution with
>objective=290.

Thank you very much for your report.

Your problem is very interesting. It is hard for glpk mip solver, so I
could not solve it to optimality using default settings. However, when
I specified 1.0 as incumbent objective value, the solver found an
integer optimal solution (which was 1.0) for a few seconds, so in your
run the result was erroneous (I was using glpsol 4.4, not 4.3). On the
other hand, when I specified 1e-3 as incumbent value, the solver could
not find any integer feasible solution for a very long time. All this
puzzles me.

>I have a few others similar cases with slightly
>different data.  Do you guys want to see the others
>as well?

If you have similar instances of smaller size, please post them to me.
This would help me to understand the bug.

Andrew Makhorin