770 BOOK REVIEWS

11. M. Olinick, An introduction to mathematical models in the social and life sciences, Addison-
Wesley, Reading, Mass., 1978.
12. W. H. Riker and S. J. Brams, The Paradox of vote trading, American Political Science
Review 67 (1973), 1235-1247.
13. F. Roberts, Discrete mathematical models: with applications to social biological, and environ-
mentalproblems, Prentice-Hall, Englewood Cliffs, N. J., 1976.
14. T. L. Saaty and F. J. Wegl, The spirit and the uses of the mathematical sciences,
McGraw-Hill, New York, 1969.
15. E. M. Uslaner and J. R. Davis, The paradox of vote trading', effects of decision rules and
voting strategies on externalities, American Political Science Review 69 (1975), 929-942.
DANIEL P. MAKI
BULLETIN (New Series) OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 3, Number I, July 1980
© 1980 American Mathematical Society
0002-9904/80/0000-0317/$01.50

Mathematics offinite-dimensionalcontrol systems. Theory and design, by David

L. Russell, Lecture Notes in Pure and Applied Mathematics» Vol. 43,
Marcel Dekker, New York and Basel, 1979, viii + 553 pp., $45.00.
Control theory was brought into existence during the second half of the
eighteenth century by the development of complex machinery such as the
steam engine. Since that time until about 1900 it was primarily concerned
with elimination of undesirable traits (chiefly instability) by means of feed-
back devices, the Watt governor being a notable example; design was mainly
the result of intuition and empirical insights. The beginning of the theory can
be traced to J. C. Maxwell's celebrated paper on governors [1]. Progress was
slow during the nineteenth century but became faster after 1900 due to the
development of power transmission, communications and complex processing
plants and some mathematical techniques (such as the Routh-Hurwitz stabil-
ity criteria) began to be systematically used. Growth was enormous during
and after the second World War and many other mathematical tools like
Laplace transforms and probability theory found applications. In the late
fifties and early sixties, starting with the work of Bellman, Glicksberg and
Gross [2], Bellman [3], Pontryagin, Boltyanskiï, Gamkrelidze and Mischenko
[4], Kalman [5], Kalman and Bucy [6] and others, control theory began to be
accepted as a respectable mathematical discipline. It also started to absorb
relatively sophisticated "modern" mathematics into its language (for instance
measure theory, elementary functional analysis, abstract algebra and
Liapunov stability theory) and brought to the forefront the idea of quality of
control: if the control engineer was content in the past, say, with rendering
stable the operation of a machine by means of a feedback device his modern
counterpart would try to achieve the same effect in a suitably optimal way
(for instance, minimizing the stabilization time, the cost of the control device,
the strain on the machinery, etc.). Finally, concepts like controllability,
observability and stabilization by feedback, until then living in a latent state
in the literature were given precise formulations.
Although many of the initial contributions to the mathematical theory of
control were firmly rooted in reality (for instance, the influence of [3] and [6]
in modern technology was and is enormous) control theory tended to develop
along two parallel lines since the early sixties. The first is practiced by
 BOOK REVIEWS 771

mathematicians, relies on mathematics that the average engineer considers

esoteric and (at least in some cases) has tended to produce more general
theory than solutions of particular problems. On the positive side, it has
brought to light connections (such as that between optimal control theory and
classical variational calculus), it has clarified many concepts and eliminated
duplication. The second line inherits the classical approach and stresses the
need for solving real problems, some times neglecting the search for general
principles and ignoring the difference between an heuristic argument and a
proof. It must be recognized of course that here, as in almost all areas of
applied science, mathematics lags behind the applications. Although the
overlap of these two fields of research is considerable, a great part of the
literature can be easily placed in one (and only one) of them.
This is not the case with the book under review. In spite of its title, I
believe it will be of interest both to mathematicians and engineers. The author
is a mathematician and a practitioner of control theory and both qualifica-
tions are obvious throughout the book. Proofs are complete and correct (as
well as elementary and self-contained when possible) but concrete problems
are never out of sight and the actual design and implementation of control
devices, including computation of all parameters involved is stressed at all
times. The mathematical demands on the reader do not go much beyond
linear algebra and multivariable calculus; all that is needed from the theory
of differential equations can be found in Chapter I and the Lebesgue integral
can be dispensed with if one is willing to accept on faith a few existence
statements. Some acquaintance with probability may be necessary in Chapter
VI but all nonelementary facts needed (such as those from the theory of
stochastic differential equations) are carefully introduced and explained.
The book considers exclusively control systems described by systems of
ordinary differential equations. For the most part these systems are determin-
istic, but stochastic systems (in particular, deterministic systems where the
state is corrupted by white noise) are studied at length in Chapter VI. The
equations are mostly linear, but there are several exceptions; controllability of
a nonlinear system near an equilibrium point is found in Chapter II and some
results on global controllability are given in Chapter IV. The treatment of
optimal control problems is chiefly restricted to linear systems with quadratic
performance criteria, both in finite and infinite time intervals for determinis-
tic and stochastic systems, although the dynamic programming method for a
general optimal control problem is briefly expounded in Chapter IV. A great
deal of attention is given to computational methods, especially in Chapter V
where numerical solution of the quadratic matrix equations arising from
linear-quadratic problems is discussed at length.
An especially attractive feature of this book is the nice balance between the
classical and the abstract. For instance, after controllability and observability
are given a suitably elementary treatment an abstract version of these
concepts (due to Doleckiï and Russell) is introduced; the reader who knows
(or is willing to learn) a few elementary facts on operators in Hubert space
will be able to understand the subject in a deeper and more transparent
fashion. Another is the awareness, obvious in many places but especially in p.
93 of the divergent points of view of the engineer and the mathematician
772 BOOK REVIEWS

about what is meant by "optimal". Finally, mention should be made of the

excellent collection of exercises, some of them challenging numerical projects
involving access to a high speed computer.
There are not many references in the literature where one can learn of real
control problems without undue strain on credibility. This book is one of
them.

REFERENCES
1. J. C. Maxwell, On governors, Proc. Royal Soc. 16 (1868), 270-283.
2. R. Bellman, I. Glicksberg and O. Gross, On the "bang-bang" control problem, Quart AppL
Math. 14(1956), 11-18.
3. R. Bellman, Dynamic programming, Princeton Univ. Press, Princeton, N. J., 1957.
4. L. S. Pontiyagin, V. G. Boltyanskiï, R. V. Gamkrelidze and E. F. Mischenko, The
mathematical theory of optimal processes, Gostekhizdat, Moscow, 1961. English translation:
Wiley, New York, 1962.
5. R. E. Kalman, Contributions to the theory of optimal control, Bol. Soc. Mat Mexicana 12
(1960), 102-119.
6. R. E. Kalman and R. S. Bucy, New results in linear prediction and filtering theory, J. Basic
Eng. (Trans. ASME, Ser. D) 83 (1961), 95-107.
H. O. FATTORINI
BULLETIN (New Series) OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 3, Number 1, July 1980
© 1980 American Mathematical Society
0002 9904/80/0000-0318/$01.7S

Basic set theory, by Azriel Levy, Springer-Verlag, Berlin, Heidelberg, New

York, 1979, xiv + 391 pp., $24.90.
Perhaps the greatest obstacle in teaching elementary set theory is the
mathematical logic needed to formalize the axioms. There is nothing inher-
ently difficult about the basic material-the theory of ordinal and cardinal
numbers, and the axiom of choice-which every mathematician is expected to
know. And most of the axioms of ZF, the system of Zermelo and Fraenkel
used most frequently nowadays, can be stated easily and understandably in
English. The exception is the axiom scheme of replacement, which when
formalized looks like this
V*! • • • Vxn(Vx Vy \/z(<p(x,y, xv...9 xn) A <p(x, z,xl9..., xn) -*y - z)
-» Va 3b >fw(w G b ++3x(x G a A <p(x> w> xv • • • > *«))))•
Here <p stands for a formula of the first-order language of set theory; each
such <p yields a new instance of the axiom scheme, so there are infinitely
many axioms.
Now the idea behind the replacement scheme is quite simple: any corre-
spondence carries sets to sets. The problem is how the "correspondence" is to
be specified. The solution, of course, is that the correspondence must be
definable from parameters in a way which could be formalized in first-order
logic. Unfortunately, many students do not find this completely clear.
Nor is this the only such problem. To take another example, each instance
of the principle of definition by transfinite recursion is usually quite clear, yet
the formalization of the principle itself (as a theorem scheme) is often
confusing.