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

[Andrew Makhorin]

Thank you for your bug report.

>I have a problem when calling lpx_simplex(lp).
>It loops infinitely on some data sets in the function lpx_simplex with the
>path:
>L1360 in glplx6a.c    feas:
>    ret = LPX_OK
>    goto opt
>    ...
>L1392 in glplx6a.c    opt:
>    ret = LPX_E_INSTAB
>    goto feas 
>Hence, it oscillates between the first step (find a feasible primal
>solution) and the second (find an optimal basis solution).
>
>Is it possible ? Is it a bug ?

If it happens, it is obviously possible :+) However this is not a bug,
it is a defect of the implementation. In your problem an optimal point
found by the primal simplex slightly violates the primal feasibility
conditions (the maximum relative primal infeasibility at the final point
is about 1e-6 while the default tolerance is 1e-8), so the primal
simplex switches back to the phase I. I need a time to understand why
this involves cycling.

You can increase the primal tolerance (before call to lpx_simplex) as
follows:

      lpx_set_real_parm(lp, LPX_K_TOLBND, 1e-5);

This will help (probatum est), but primal values will be computed less
accurately (approx. 5-6 decimal places).