Reminder: Mike Powell

[Ben Herbst]

Just a friendly reminder of Mike Powell's talk today. 

Please note that the talk in in the Mechanical Engineering building (details below).

SEMINAR: Friday 02 November 2012 from 13h00 to 14h00 in Room M604
                 in the M&M Building.

RSVP: Tuesday 30 October 2012 at 17h00


          On derivative-free optimization with linear constraints

                                    by

                              M.J.D. Powell

A version of the symmetric Broyden updating method is highly successful
for the construction of quadratic models of nonlinear objective functions
when derivatives are not available. It is applied in the NEWUOA and UOBYQA
software of the author for unconstrained and simple bound constrained
optimization, respectively. These packages are now in general use for
calculations with up to about 500 variables. Further, they are being
extended to a new package, namely LINCOA, which allows general linear
constraints on the variables. The progress of this work will be described
with some preliminary numerical results. An important feature is that on
average the routine work of each iteration is only of magnitude n squared,
where n is the number of variables. It is achieved by solving trust region
subproblems approximately by truncated conjugate gradients, by updating
the inverse matrix of a linear system of equations, and by employing only
of magnitude n values of the objective function on each iteration.


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Mike Powell graduated from the University of Cambridge in 1959, when he
moved to the Atomic Energy Research Establishment, Harwell to assist with
numerical calculations. He moved back to Cambridge in 1976, becoming
Professor of Applied Numerical Analysis until his official retirement in
2001, but he still has an office at the university and does much research.
He is a Fellow of the Royal Society, a Foreign Associate of the National
Academy of Sciences and a Corresponding Member of the Australian Academy
of Science. He is a pioneer of several optimization algorithms, including
the DFP method, the augmented Lagrangian method and sequential quadratic
programming. His theoretical work is not only on practical algorithms
for optimization but also on the radial basis function method for
multivariable approximation. He has published about 180 papers but only
one book, namely Approximation Theory and Methods.

--
Ben Herbst
Applied Mathematics
University of Stellenbosch
Matieland 7602
South Africa

Phone: +27 (0)21 808-4217 (office)
             +27 (0)21 939-9705 (home)
Fax: +27 (0)21 808-3778
Cell Phone: +27 (0)83 566-4466
URL: http://dip.sun.ac.za/~herbst/