BOOK REVIEWS 97

Operator Inequalities. By JOHANN SCHROEDER. Academic Press, New York, 1980.

xvi + 367 pp. $39.50.
Downloaded 11/29/14 to 129.120.242.61. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

This book is relevant for numerical analysts who are interested in boundary value
problems for (second order ordinary) differential equations and their discretizations.
It is not a book on (linear or nonlinear) functional analysis as its title may suggest. It
is concerned with linear and nonlinear operators on ordered vector spaces. But there
are essentially no topological hypotheses imposed and, consequently there are almost
no advanced results or methods from functional analysis being invoked or developed.
It is the concern of the author to deduce a priori bounds for possible solutions
of linear and nonlinear equations, if it is known that these equations satisfy certain
inequalities. For example, if it is known that Mv <-Mu, one wants to deduce that
_-< u. If such an "inverse positivity" is known to hold, then it is often possible to
construct approximate solutions by monotone iterations. In addition, one often gets
bounds for the error, which, according to the chosen ordering, can be quite effective
numerically.
It is clear that these methods are closely related to the maximum principle in the
theory of elliptic differential equations and to the well-known comparison techniques
for ordinary differential equations, based upon differential inequalities. Hence their
applicability is quite restricted, and there are many solutions to boundary value
problems, say, which cannot be obtained or estimated by the relatively simple-minded
methods of the book.
The book contains five chapters with the following heads: I. Some Results on
Functional Analysis, II. Inverse-Positive Linear Operators, III. Two-sided Bounds
for Second Order Differential Equations, IV. An Estimation Theory for Linear and
Nonlinear Operators, Range-Domain Implications, V. Estimation and Existence
Theory for Vector-Valued Differential Operators. Each chapter is supplemented by
"Notes" containing references to the literature and to related work. Throughout the
text there are many concrete examples illustrating the abstract results.
HERBERT AMANN
Mathematisches Institut
University of Zurich

Practical Methods of Optimization, Vol. 1: Unconstrained Optimization. By R.

FLETCHER. Wiley, Chichester, 1980. viii + 120 pp. $24.50.
Roger Fletcher is one of the best researchers in practical methods of optimization,
and so we are lucky that he is writing this long needed two-volume set of the same
name. This is a review of Volume I, entitled Unconstrained Optimization. The authors
preface, dated July 1979, indicates that Volume II on Constrained Optimization
"should be finished within the year."
The emphasis in this book is on the types of algorithms one finds implemented
in such subroutine libraries as HARWELL, IMSL, MINPACK, and NAG. These
algorithms are of two basic types: conjugate directions and Newton-like. The author
also mentions briefly the continuation methods useful for some problems, and he gives
a lucid description of the simplex algorithm (not related to the linear programming
algorithm) which is so effective for noisy low-dimensional problems.
The book is organized in a thoughtful and efficient way to solve the problem of
dealing with both conjugate direction (Chapter 4) and Newton-like (Chapter 3)
methods by making use of their superficial similarity as search direction generators
underpinned by the notion of a quadratic model. He deals with their different line
search requirements at appropriate places (Chapter 2) as a part of an overview of
 98 BOOK REVIEWS

algorithmic analysis, implementation, and testing. He resolves the pedagogical problem

of which to do first, line searches or direction generators, through the clever device
Downloaded 11/29/14 to 129.120.242.61. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

of considering directions generated by the simple and intuitive coordinate search and
simplex algorithms. He introduces these in a historical perspective called ad hoc
methods.
Such thoughtful organization is an ndication that it would probably be easy to
lecture from this book. Indeed, I think it would be if one presented exactly this
material, but I would not. I feel a need for more motivation and more rigor in a text
than there is room for in 120 pages including text, exercises, references and index.
The lack of rigor is sometimes a lack of specific reference to definition of terms. For
example, stability is used in the traditional sense and also to mean something like
"proceeding steadily toward a solution". We are left to infer both meanings from
context, and an advanced student certainly will. In Theorem 2.4.1 on page 21, the
same notation is used first for a sequence, then a subsequence, then for the whole
sequence again in a completely ambiguous way.
Chapter 5 is a lucid exposition of modern Levenberg-Marquardt-type algorithms
for varying search direction along with step length, and Chapter 6 briefly treats the
nonlinear least squares and nonlinear simultaneous equations problems. The index is
quite good, but the references are somewhat eccentric. There is no reference to
Ortega-Rheinboldt or to Wolfes important paper on Goldstein-Armijo-type step
acceptance conditions, but there is a reference to Hardys analysis book.
This is a well-written compact book with helpful numerical examples and useful
figures. I look forward to Volume IL and I only hope that the publisher will constrain
the price a bit in keeping with the topic.
JOHN E. DENNIS, JR.
Rice University

Computability: An Introduction to Recursive Function Theory. By N. J. CUTLAND.

Cambridge University Press, Cambridge, 1980. ix / 251 pp.
Professor Cutland has. written a delightful primer on recursive function theory.
As was the goal of the author, the fundamental results of classical recursion theory
are presented in a style which is accessible to undergraduates with a minimal mathe-
matical background. A salient feature of the book is the appropriate blend of intuition
and rigor used in the exposition of proofs. Examples and exercises are abundantly
scattered throughout the text. This reviewers reading uncovered very few typo-
graphical errors.
As with any text, no matter how lucid the prose, the size of the intended audience
must be considered. Cutlands book would make an excellent text for an introductory
course in recursion theory for mathematics students. The author claims that his book
is also useful for computer scientists pursuing knowledge in abstract computability.
Unfortunately, there is no mention of any of the work done in the last decade by
researchers taking a recursion theoretic approach to computer science. Hence, any
instructor must be prepared to substantially augment the material in Cutlands book
when it is used as a text for computer scientists. This reviewer has used portions of
the book (Chapter 9 on Reducibility and Degrees) as source material for lectures.
Preparation was straightforward.
CARL H. SMITH
Purdue University