Re: [Help-glpk] Infinite loop in glplx6a.c

[Andrew Makhorin]

>Continuing my work with GLPK, I felt again over the infinite loop.
>FYI, when debugging the solver, I see that one of the variables goes out of
>his bound (1 <= 1.000000451...), hence when checking the primal feasibility,
>it reports a numerical instability (as normally, the solution must always be
>primal feasible).
>Playing (increasing or decreasing) with LPX_K_TOLBND "solves" the problem.

In your problem (I mean the mps file you sent me a month ago) there are
almost linearly depent constraints. Though LP formulation allow such
constraints, they worsen numerical properties of the problem and may
cause instability. So, increasing the primal feasibility tolerance is a
reasonable way to help the solver. Improving accuracy wouldn't help
because the basis matrix at the optimal point is ill-conditioned in this
case.