Re: [Help-glpk] numerical instability

[Andrew Makhorin]

> I do not really understand what Andrew means.
> Perhaps that rescaling necessarily throws away
> valuable solution information.  Ask him if you need
> to know.
> 

Changing row/column scale factors may change the basis matrix that would
invalidate the current basis factorization stored in glp_prob by the
simplex solver, because the factorization is computed for the scaled
matrix, not for the original one. However, this affects only efficiency,
because if the basis factorization is invalid, the simplex solver will
compute it.