Large-Scale Nonlinear Optimization

10:45 AM-12:45 PM

*Room: Capitol South*

Optimization problems may be arbitrarily large and challenging regardless of computing speeds. Successful solution involves modeling languages and network systems, apart from the expected improvements to algorithms.

GAMS is one of the languages that has helped make nonlinear optimization a truly practical tool. Some important topics from the GAMS environment are addressed here, including mixed-integer NLPs and extremely large complementarity problems.

MetaNEOS represents a radically new concept in large-scale optimization: the use of many computers across a network. Key questions to be discussed are whether an algorithm can be made {\em fault tolerant\/} and {\em preemptive}. (Can it proceed when computing resources suddenly fail or become available?).

SQP methods have brought reliability and efficiency to nonlinear programming, but few large-scale implementations exist. Two new SQP algorithms will be presented, both using interior methods to solve the QP subproblems.

**Organizer: Michael Saunders**

*Stanford University, and University of Auckland, New Zealand*

**10:45-11:10 Practical Issues in Nonlinear Optimization**- Steven P. Dirkse, GAMS Development Corporation
**11:15-11:40 Large-Scale Optimization Using Meta-Computing Platforms**- Golbon Zakeri, MCS Division, Argonne National Laboratory
**11:45-12:10 Implementation of a Filter Algorithm for Regularized SQP Solution of Nonlinear Programs**- John Tomlin, IBM Almaden Research Center
**12:15-12:40 The Solver Omuses/HQP for Structured Large-Scale Constrained Ooptimization: Algorithm, Implementation and Example Application**

*Ruediger Franke*, ABB Corporate Research, Heidelberg, Germany; and Eckhard Arnold, Technical University of Ilmenau, Germany

*MMD, 3/12/99*