Anthony J. Kearsley
(Department of Mathematical Science, Carnegie Mellon University and
Math and Computational Science Division, NIST)
On the relaxation of constraints in a particular SQP algorithm
Abstract
In this talk, we will discuss a sequential quadratic programming (SQP)
algorithm developed to solve general nonlinear programming (NLP) problems
(minimization of a function subject to non-linear equality and inequality
constraints). With an eye towards a particular class of application motivated
problems, the algorithm solves a sequence of `relaxed' quadratic programming
(QP) problems. The relaxation guarantees that the linearization of the
constraints yields a consistent set of linear equations and thus the resulting
relaxed QP is solvable. Numerical performance on a collection of test problems
will show that inconsistent linearizations were encountered by the SQP
algorithm and that the relaxation strategy appears to be effective at overcoming
the difficulty. We conclude the talk with a short demonstration of the
solution to one particular test problem (optimal control fluid flow). |
|