On the behavior of Lagrange multipliers in convex and non-convex infeasible interior point methods∗

This paper analyzes sequences generated by infeasible interior point methods. In convex and nonconvex settings, we prove that moving the primal feasibility at the same rate as complementarity will ensure that the Lagrange multiplier sequence will remain bounded, provided the limit point of the primal sequence has a Lagrange multiplier, without constraint qualification assumptions. We also show that maximal complementarity holds, which guarantees the algorithm finds a strictly complementary solution, if one exists. Alternatively, in the convex case, if the primal feasibility is reduced too fast and the set of Lagrange multipliers is unbounded, then the Lagrange multiplier sequence generated will be unbounded. Conversely, if the primal feasibility is reduced too slowly, the algorithm will find a minimally complementary solution. We also demonstrate that the dual variables of the interior point solver IPOPT become unnecessarily large on Netlib problems, and we attribute this to the solver reducing the constraint violation too quickly.