Hi all! I am quoting an almost three year-old-e-mail of this list about the deletion of active constraints in a LP, keeping the current basis in order not to solve the model from scratch. I used this method in my algorithm (removing the constraint bounds: GLP_FR), but I don't understand a point: when I solve the LP again, the first solution (at iteration 0) has a lower objective (all right) but the value of the infeasibility increases. How is it possible that the infeasibility increases although I removed a constraint? I am (still) using the GLPK version 4.38. Best regards,
Sylvain Fournier If you delete some active row, i.e. the row, whose auxiliary variable > is non-basic, the number of rows is decreased by one while the number > of variables marked as basic remains the same, and the basis becomes > invalid. Similarly, if you delete some basic structural variable, the > number of rows remains the same while the number of basic variables > is decreased by one, so the basis again becomes invalid. > > If a row is active, deleting it invalidates the basis as was explained > above. However, you can attain the same effect by *freeing* the row, > i.e. by changing its bounds to -inf and +inf, resp., in which case the > basis remains valid. Analogously, if a column is basic, you can change > both its bounds to 0, i.e. *fix* the structural variable at zero, rather > than delete it, in which case the basis again remains valid. > >
_______________________________________________ Help-glpk mailing list [email protected] http://lists.gnu.org/mailman/listinfo/help-glpk
