Hi Chris,

Thank you for your bug report and patch.

> There is a bug in the dual simplex transition from phase I to phase II
> when perturbation was applied - it may be that the perturbed problem
> is dual infeasible while the unperturbed one is feasible, a case that
> is not checked in the code. This happens with stat96v1 (downloaded
> from http://www.sztaki.hu/~meszaros/public_ftp/lptestset/misc/ ).
> 
> The attached patch is an approach to fix this, but disables
> perturbation for phase II in this case, which may not be desirable.
> 

I changed the code according to your patch. However, this case needs to
be analyzed more carefully, because perturbation should only relax the
(dual) constraints, so the original unperturbed lp cannot be dual
feasible if the perturbed (thus, relaxed) lp is dual infeasible. I think
this happens due to free (unbounded) variables.


Andrew Makhorin


_______________________________________________
Bug-glpk mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-glpk

Reply via email to