Monday, May 10

Primal-Dual Methods for Nonconvex Nonlinear Programming

10:45 AM-12:45 PM
Room: Atlanta 3

Primal-dual interior methods are typically derived from either properties of the logarithmic barrier function or perturbed complementarity. Beyond these common foundations, recently proposed primal-dual approaches to nonconvex nonlinear programming differ from one another in almost every aspect. This session provides a broad view of current research in primal-dual techniques by describing the motivation and implementation of four promising methods. Some of the issues to be discussed are the form of the Newton equations, convergence properties, initialization and adjustment of the barrier parameter, incorporation of equality constraints, estimation of Lagrange multipliers, treatment of indefiniteness, and strategies for infeasible starting points.

Organizers: Philip E. Gill
University of California, San Diego
Margaret H. Wright
Bell Laboratories, Lucent Technologies

10:45-11:10 Global and Local Convergence of a Primal-Dual Barrier Algorithm for Nonlinear Optimization
Andrew R. Conn, IBM T. J. Watson Research Center; Nicholas I. M. Gould, Rutherford Appleton Laboratory, Oxfordshire, United Kingdom; and Dominique Orban, CERFACS, Toulouse Cedex, France
11:15-11:40 Primal-Dual Algorithms for Nonlinear Programming
Anders Forsgren, Royal Institute of Technology, Stockholm, Sweden; and Philip E. Gill, Organizer
11:45-12:10 An Interior-Point Algorithm for Nonconvex Nonlinear Programming
David Shanno, Rutgers University; and Robert J. Vanderbei, Princeton University
12:15-12:40 NewPrimal-Dual Methods: Theory and Practice
Margaret H. Wright, Organizer

OP99 Home


Program Updates

Speaker Index




MMD, 3/26/99