A INTERIOR POINT ALGORITHM FOR CONVEX NONLINEAR PROGRAMMING PROBLEMS

This paper describes a primal-dual interior paint algorithm for convex nonlinear programming problems subject to linear constraints. The algorithm is based on the path following idea. Each iteration updates a penalty parameter and finds a Newton step associated with the simplified Karush-Kuhn-Tucker system of equations which characterizes a solution of the logarithmic barrier function problem for that parameter. It is shown that the duality gap if reduced at each iteration by a factor of (1 - δ / n;), where S is positive and depends on some parameters associated with the objective function.