Re: [Bug-glpk] Incorrect result from glpk-4.39

[Andrew Makhorin]

> I am writing to report what looks like a bug in glpk-4.39, that,
> surprisingly, is not there in glpk-4.26.

> Below, I first describe the problem, and its output when run with
> glpk-4.39 on a pc running fedora-11. It is solved incorrectly by
> glpk-4.39. Then, I will give the output of glpk-4.26, which solves it
> correctly Would you please let me know if there is a bug, or if I am doing
> something wrong. I do however know that I am doing the same thing in
> invoking glpk-4.39 and glpk-4.26.

The error appears due to very small constraint coefficient in your
model (of order 1e-32). Glpk 4.26 successfully solves the model, because
by default it performs only equilibration scaling while glpk 4.39 also
performs geometric mean scaling to minimize the ratio max|aij|/min|aij|
that in your case does not help.

You need to either disable automatic scaling with the option
'--noscale' or, that is much better, remove tiny constraint coefficients
from the model.