[Bug-glpk] Re: feasible LP declared infeasible by GLPK

[Andrew Makhorin]

>Thank you very much. It works if I call lpx_adv_basis before lpx_simplex.
> I wonder why
>the default settings are not being used automatically. Perhaps it's a
> quirk of the java
>interface (I should have mentioned, I'm using the Java interface).
>I'll look into raising the lower bound on some variables as you suggest.

It is not a quirk of the java interface; in C you would have the same
effect. The point is that lpx_simplex can be used for re-optimization
and therefore it does not construct an initial basis expecting it to be
specified on entry. On the other hand, this feature is not needed in
glpsol, where lpx_simplex is called only once.

I would like to note that yet another way to avoid the error you
reported in your previous mail is to enable the lp presolver before the
first call to lpx_simplex as follows:

      lpx_set_int_parm(lp, LPX_K_PRESOL, 1);

(by default it is disabled). The presolver is able to improve bounds of
columns, so the error must disappear. Note that if the presolver is
enabled, lpx_simplex additionally performs automatic scaling and
constructs an advanced initial basis for an internal lp problem which is
actually solved. So, in this case you do not need to call lpx_adv_basis.