Scribd
Upload
117 views

Convex Analysis and Nonlinear Optimization

Optimization

Uploaded by

Mayank
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
117 views

Convex Analysis and Nonlinear Optimization

Optimization

Uploaded by

Mayank
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

92 BOOK REVIEWS

CONVEX ANALYSIS AND NONLINEAR A recurring example is the case of spectral


OPTIMIZATION: THEORY AND EXAM- functions on symmetric matrices, which allows a
PLES. Jonathan M. Borwein and Adrian very nice treatment of the currently exciting topic
S. Lewis, Springer, New York, 2000. of semidefinite programming.
Besides the expected material in such a book,
This is an excellent modern book on convex and there is a section on convex process duality; a
non-smooth analysis, suitable as a text for a chapter on fixed-point theorems and variational
graduate class or for reference. In a relatively analysis; and a chapter discussing infinite-dimen-
short space (the main part of the text is just sional extensions as well as characterizations of
208 pages), the authors include a great deal infinite dimensionality via properties of convex
of interesting material. Of course, their coverage sets and functions.
is by necessity selective, as opposed to the One remarkable feature is the requirement for
more encyclopedic Variational Analysis by substantial reader involvement: indeed the ex-
Rockafellar and Wets (Springer, 1998), but by ercises are an integral part of the book. The
an extensive set of exercises a very nice picture, preface claims that the main text is self-contained
with many glimpses of intriguing related areas, and the exercises optional, but a few times the
is painted. (Other related and recommended exercises introduce definitions or results that are
books are the classic Convex Analysis of later used in the main text. Moreover, the
Rockafellar (Princeton, 1970), the more algorith- exercises are very rich. Many sections are in fact
mic Convex Analysis and Minimization shorter than the set of exercises they precede!
Algorithms by Hiriart-Urruty and Lemarechal Exercises contain proofs for (parts of) many of the
(Springer, 1993), and Clarke et al. Non- results, and address such basic notions as reces-
smooth Analysis and Control Theory (Springer, sion cones, relative interiors, and infimal convolu-
1998).) tion, but also singular values, von Neumann’s
The unifying theme is optimality conditions, minimax theorem, Kirchoff’s law, Schur and
and these are treated in a succession of increas- order convexity, minimum volume ellipsoids, etc.
ingly complicated settings, from smooth inequal- Because of the wealth of material in the
ity-constrained optimization to convex, non- exercises and the revisiting of many issues in more
smooth, and inequality and equality-constrained general contexts, easy navigation is important.
problems. The variety of techniques used include Indeed, several of the more general optimality
max-functions and directional derivatives, value conditions appear in sections labelled ‘Value
functions, and contingent cones. Indeed, the Functions’ or ‘Generalized Derivatives’. The
authors take care to point out the different authors have provided an extensive and very
approaches possible, and the interplay between helpful index, and have supplemented it with a
algebraic/geometric and topological viewpoints. final chapter listing all named results (important
One of their goals is to introduce the reader to theorems, in the main text and in the exercises, are
modern ideas in analysis through a study of all named for future reference) and notation.
optimization, and in this they succeed wonder- My criticisms are few: the authors explain that
fully. Besides developing general tools which are their concern is the analysis of optimization
then used in optimization, they also apply problems, not computational algorithms, which is
optimality conditions to prove results that see- fine; but their statement in the commentary on
mingly have nothing to do with minimizing or Section 7.1 that the use of penalty functions is now
maximizing. standard practice in computational optimization is
In a similar development, convex functions are somewhat misleading. Penalty functions are often
first introduced in the real-valued case, and then used in a supplemental way, as merit functions, but
more general extended real-valued convex func- penalty methods, where constrained problems are
tions appear, motivated by their relationship to reduced to a sequence of unconstrained problems,
optimization problems or analytic techniques are not: a distinction could have been made
already seen. between these and interior-point methods which

Copyright # 2003 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2003; 13:91–93
BOOK REVIEWS 93

use barrier functions. The authors introduce the those who want to study the recent advances in
core of a set in Section 3.1, and stress that it is convex and non-smooth analysis in an engaging
different from the interior. But at this point, almost and stimulating text.
all sets appearing are convex, and it is not until
Section 4.1 that the authors explain (and prove)
that the two notions coincide for convex sets. The Michael J. Todd
book is very well prepared, and there seem to be School of Operations Research
very few typos, but one is noteworthy: the Rhodes Hall, Cornell University, Ithaca
definition of the lim inf notation below the growth NY 14853-3801, USA
condition in Section 1.1 is inaccurate. E-mail: [email protected]
Despite these minor complaints, this is a very
rewarding book, and I highly recommend it to (DOI: 10.1002/rnc.701)

Copyright # 2003 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2003; 13:91–93

You might also like