Large-Scale Quadratic Programming

SE27.1 An Interior Point Algorithm for Linearly & Quadratically Constrained Quadratic Programming

• Dave Shanno; Rutgers University, RUTCOR, New Brunswick, NJ 08903-5062; shanno@rutcor.rutgers.edu

We describe the performance of the LOQO algorithm on a variety of quadratic programming problems, including standard convex quadratic programming, nonconvex quadratic programming and quadratically constrained quadratic programming. Comparative numerical testing with other codes will be given.

SE27.2 An Interior Point Method for Nonconvex Nonlinear Programming

• Robert Vanderbei; Princeton University, ACE-42 E-Quad, Princeton, NJ 08544; rvdb@teal.princeton.edu

We describe the performance of LOQO on a variety of nonconvex NLP problems. A large collection of AMPL models have been collected and serve as a base for comparing LOQO against other well-known algorithms such as MINOS and LANCELOT.

SE27.3 Extrapolation Matters(!)

• Ariela Sofer; George Mason University, ORE Dept., Fairfax, VA 22030; asofer@osfl.gmu.edu

Barrier methods went out of fashion around 1970, because they were thought ill-conditioned. Now we know that (if Newton's method is used) this may not matter. But without extrapolation (alas), the Newton step may be infeasible. We show that with extrapolation, however, barrier methods are loaded with good properties.