Re: [Help-glpk] to whom can help me

[Andrew Makhorin]

> And now is the first time I got a problem. Doing a branch cut
> exploration, I need to remove a row of the problem whenever I go up a
> level in my exploration tree. And this seems not work. I do the branch
> on variables called zk. If start with z3 = 1 for instance, I have to
> add a row when testing z3 = 1 childs, as can be z3 = 1 and z4 = 1. It
> keeps working. z3 = 1 and z4 = 0. It continues ok. And after these
> tests, I remove the row with z4 = 0 and change the previous to z3 = 0.
> And GLPK gives me back a "solution is undefined" code. If I start with
> z3 = 0, the solution exists.

Make sure that you do not remove active row, i.e. constraint whose
auxiliary variable is non-basic. Otherwise glp_del_rows invalidates
the current basis factorization, so subsequent call to glp_simplex
will lead to the "bad basis" error, in which case the basic solution
status will be set to "undefined".

> I have extracted the problem in any case, with the lpx_print_prob.
> The files are exactly the same when I try z3 = 0 at the begining
> than when I try firstly z3 = 1, then z3 = 1 and z4 = 1, then z3 = 1
> and z4 = 0, and finally z3 = 0. The files written by GLPK for the
> case z3 = 0 are exactly the same, I insist, even thought the value
> of the solution in the first case is 7.00 (as it is supposed to be),
> ans in the second "solution is undefined". So, I think the
> instruction that is not working is lpx_del_rows. Furthermore, after
> three years having  been using GLPK, is the first time I need it.