1: \begin{abstract}

2:   We show that the effects of finite-precision arithmetic in forming

3:   and solving the linear system that arises at each iteration of

4:   primal-dual interior-point algorithms for nonlinear programming are

5:   benign, provided that the iterates satisfy centrality and

6:   feasibility conditions of the type usually associated with

7:   path-following methods. When we replace the standard assumption that

8:   the active constraint gradients are independent by the weaker

9:   Mangasarian-Fromovitz constraint qualification, rapid convergence

10:   usually is attainable, even when cancellation and roundoff errors

11:   occur during the calculations.

12: % This conclusion holds for all three

13: % of the standard formulations of the linear system that is solved at

14: % each iteration of a primal-dual method. 

15:   In deriving our main results, we prove a key technical result about

16:   the size of the exact primal-dual step. This result can be used to

17:   modify existing analysis of primal-dual interior-point methods for

18:   convex programming, making it possible to extend the superlinear

19:   local convergence results to the nonconvex case.

20: \end{abstract}

21: