You are on page 1of 1

BOOK REVIEWS

593

siderable appeal to the desired audience, though it may be argued that ric Programming appeared in 1971 with the intention of giving en-
sometimes the author has carried his simplification too far. For instance,gineers a brief overview of GP procedures.
in the chapter on the inverse z transform, only the case of real poles has It is only recently that one finds GP discussed in texts on math-
been considered. This may mislead the beginning reader, since, in the ematical programming [6]-[12]. A notable exception is Foundations of
following chapter, some examples with complex poles have also been Optimization, by Wilde and Beightler [13], which already in 1967
discussed. included a well-written chapter on this topic, incorporating further
On the whole, it can be said that the author has succeeded in writing work by Wilde, Avriel, and Passy. Now that the literature on GP is
a book suitable for nonspecialists, who will really appreciate the easy- growing steadily, special sessions on GP are held at meetings of the
going manner in which the subject has been developed, since it does Operations Research Society of America (ORSA) and at the Math-
require advanced mathematical preparation. ematical Programming Symposia. Of special interest to readers of these
transactions should be some recent work with GP in modern control
[14], [15], which also makes it timely to provide a glimpse of GP and
a review of the books that will be quoted ever more frequently.
Geometric Programming gets off to a good start. In the fifteen pages
of the first chapter the authors manage to convey an overall view of
Control System Theory--Sushil Das Gupta (Delhi: Khanna 1973, GP, and in keeping with that spirit a quick glimpse will be given here.
1018 pp.). Reviewed by Dr. Sachindra M. Sen, Faculty of Engineering Of the various approaches open to them (Legendre transformation,
and Technology University of Baroda, India. Lagrange multipliers), the authors choose the geometric-mean in-
This book deals with control system theory in a rather comprehensive equality to develop a dual problem. The arithmetic and geometric
way. There is some additional information on linear algebra, which is means of positive numbers U1,,-*,U form the inequality
useful in the modern mathematical approach to system theory, Ul + U2 + + Un2 (Ul U2 ...* nI
although its actual application is not elaborated in the book.
The book, as Dr. Das Gupta says in the preface, is meant for under- n
graduate studies and it has enough material to more than meet this with equality applying when U, = U2= = Un. Consider, for
requirement. It is amply interspersed with a number of illustrative example, the function g(t) = 4t + IIt, which consists of two positive
examples that should prove very helpful to the first-course engineering terms for positive t. Hence (4t + 1/t)/2 > (4t- Ilt)''2 or g(t) =

student taking the subject. 4t + (l/t) 2 4. Strict equality is attained at t = with g(t) = 4,
There has been a strong emphasis in deriving the basic structure of which, therefore, must be the optimum.
the relevant control system out of various physical systems, such as Instead of minimizing g(t) we can now set ourselves to the task of
electrical, mechanical, thermal, hydraulic, chemical, economic, and maximizing a dual function, which is defined through the geometric
hybrids. These examples are indeed very illuminating when focusing inequality. Introduce weights ...'£5 that satisfy the normality
upon the motivation involved. condition 61 + 62 + + £5n= 1 such that at the optimum (de-
Treatments of sampled-data systems (particularly their classifi- signed by '), we have
cation), the state-variable approach, compensation, etc., have been dis-
cussed from quite a few angles and contain some materials not presen- c. 17 tj'aii
ted in any other standard text on this subject, but there ought to have
been a chapter on optimal control and stochastic control, which are g(t')
more or less two standard topics. However, as an undergraduate-level Then the geometric inequality can be generalized to read
text this book provides so much additional discussion on system
concepts, real systems, and control system components that this 61 Ul + 62 U2 + * * + 6n U, . U1 1 U262 ... Un6n.
deficit is overwhelmingly compensated. The primal variables t do not always disappear as nicely from the
geometric means as they do in the preceding example. However, by
imposing orthogonality conditions, one for each tj, the weights £ will
be the only variables in the dual

Geometric Programming-R. J. Duffin, E. L. Peterson, and C. M.


51aij + 62a2J + *-- + 6,a,j = 0, j = 1,.
Zener (New York: Wiley, 1967, 278 pp.). A dual to the objective function g(t) can then be developed to be the
Engineering Design by Geometric Programming-Clarence Zener (New maximand
York: Wiley-Interscience, 1971, 98 pp.). Reviewed by R. Schinzinger,
School of Engineering, University of California, Irvine, Calif 92664.
(61 6
(2 ) 6n)
Geometric programming (GP) is an optimization method par- If the primal were of the form
ticularly well-suited to handling problems that can be modeled as
polynomials, i.e., functions of the form ming(t) = 0-
(t1 t2t3) + 40t2t3 + 20t1t3 + 10t1t2
n m

g(t) E C, ^1 tjaij, Ci > 0, for alli the dual problem would read
i=1 j=l
in nonnegative variables t1, 140~61 60 62 20\ 63 1 0 654
,tm. Constraints are to be expressed as maxv(£5) = (y)46)) (=) (
inequalities gk(t) 1, k = - *,p.
Utilizing various transformations, a surprising number of functions subject to
will fit this prototype. A refreshingly novel approach to the solution of
problems of this sort was initiated by Clarence Zener in the early (51 +62 + £53 + £54 = 1, normality condition
1960's [1], [2]. Richard Duffin pointed out the inherent duality relation- -1 + 53 + 64 = 0, orthogonal condition t1
ships and formulated an approach via a Legendre transformation [3]
and also via the arithmetic-geometric-mean inequality [4]. Elmor -
61 + 62 + 64 = 0, orthogonal condition t2
Peterson joined Duffin in extending this work to cover polynomially-
constrained polynomials [5]. In 1967 these three authors published a
-
£1 + 62 + £53 = 0, orthogonal condition t3
book, Geometric Programming. Zener's Engineering Design by Geomet- 6j > , j =1, - --,4.

You might also like