NEW METHODS

OF THOUGHT
AND PROCEDURE
NEW METHODS
OF THOUGHT
AND PROCEDURE
Edited by
F. Zwicky
and A. G. Wilson
Contributions to the Symposium on

METHODOLOGIES

Sponsored by the
Office for Industrial Associates
of the California Institute of Technology
and the Society for Morphological Research
Pasadena, California, May 22-24,1967
ISBN-13: 978-3-642-87619-6 e-ISBN-13: 978-3-642-87617-2
DOl: 10.lO07/978-3-642-87617-2

© 1967 by Springer-Verlag New York Inc.

Softcover reprint of the hardcover 1st edition1967
Library of Congress Catalog Card Number 67-31398

The use of general descriptive names, trade names, trade marks, etc. in this pub-
lication, even if the former are not especially identified, is not to be taken as a
sign that such names, as understood by the Trade Marks and Merchandise Marks
Act, may accordingly be used freely by anyone.
Title No. 1479
 LIST OF CONTRIBUTORS

BELLMAN, RICHARD, Departments of Mathematics,

Engineering and Medicine, University of
Southern California, University Park,
Los Angeles, California

ERNST, MARTIN L., Arthur D. Little, Inc.,

Cambridge, Massachusetts

FLAGLE, CHARLES, Department of Public Health

Administration, Johns Hopkins School of
Hygiene and Public Health,
Baltimore, Maryland

GILLETTE, DEAN, Transmission Systems Engineering

Division, Bell Telephone Laboratories, Inc.,
Holmdel, New Jersey

KALABA, ROBERT, The RAND Corporation, Santa

Monica, California

LUCKY, R. W., Data Theory Department, Bell

Telephone Laboratories, Inc.,
Holmdel, New Jersey

MORGENSTERN, OSKAR, Department of Economics,

Princeton University, Princeton, New Jersey

PAGE, THORNTON, Department of Astronomy, Wesleyan

University, Middletown, Connecticut
PIERCE, JOHN R., Research Communications Sciences
Division, Bell Telephone Laboratories, Inc.,
Murray Hill, New Jersey

SHAPLEY, LLOYD S., The RAND Corporation, Santa

Monica, California

SHEA, JOSEPH F., Polaroid Corporation,

Cambridge, Massachusetts

v
SHUBIK, MARTIN, Economic Growth Center, Yale
University, New Haven, Connecticut

WILSON, ALBERT G., Advanced Research Laboratories,

Douglas Aircraft Company, Inc., Huntington
Beach, California

ZWICKY, FRITZ, Department of Physics, Mathematics

and Astronomy, California Institute of
Technology, Pasadena, California

VI
Table of Contents

LIST OF CONTRIBUTORS V

PROLOGUE 1

SECTION I OPERATIONS RESEARCH S

Chapter 1 The Nature of Operations Research 7

and its Beginnings
Thornton Page
Chapter 2 A Decade of Operations Research in Health 33
Charles Flagle
Chapter 3 Operations Research Applied to City Planning 46
Martin Ernst

SECTION II SYSTEMS ENGINEERING 71

Chapter 1 Systems Engineering-Planning 73
Dean Gillette
Chapter 2 Systems Engineering-Implementation 8S
Joseph F. Shea

SECTION III DYNAMIC PROGRAMMING 97

Chapter 1 Dynamic Programming: 99
A Reluctant Theory
Richard Bellman
Chapter 2 Dynamic Programming's Progeny 123
Robert Kalaba

VII
 SECTION IV INFORMATION THEORY 133
Chapter 1 A Survey of Information Theory 135
John R. Pierce
Chapter 2 Information Theory and Modern 163
Digital Communication
R. W. Lucky

SECTION V GAME THEORY 201

Chapter 1 Game Theory: A New Paradigm 203

of Social Science
Oskar Morgenstern
Chapter 2 Welfare, Economic Structure and 228
Game Theoretic Solutions
Martin Shubik
Chapter 3 On Committees 246
Lloyd Shapley

SECTION VI MORPHOLOGICAL RESEARCH 271

Chapter 1 The Morphological Approach to 273

Discovery, Invention, Research
and Construction
Fritz Zwicky
Chapter 2 Morphology and Modularity 298
Albert G. Wilson
Chapter 3 Morphology of Multilanguage Teaching 314
Fritz Zwicky

EPILOGUE 333

The format followed at the conference and in this book is that the first paper
of each section presents a general review of the subject area and subsequent
papers within the section provide examples of specific applications of the
methodology.

VIII
 PROLOGUE

There have been numerous attempts throughout the

recorded history of man to review, classify and systematize
the basic methods of thought and of procedure which are
being used to deal both with the practical and the theoretical
problems of life. To mention a very few of these there are
the ORGANON of Aristotle's, the NOVUM ORGANUM by
Francis Bacon and the Discours de la Methode by Rene"
Descartes. These works were, of course, based on knowledge
available at the tirne, which was both lirnited and in rnany
cases false. Since the treatise of Descartes was written,
science, technology and life in general have becorne so corn-
plex that renewed rneditation on the essential aspects of
fundarnental constructive thought and procedure is in order.
The necessity for such rneditation has obviously been recog-
nized in rnany countries and in rnany quarters and has in sorne
instances led to a successful reevaluation of old principles
and procedures as well as to the developrnent of new thoughts,
while again in other cases a lack of perspective resulted in
rnore confusion. To achieve technically and hurnanly satis-
factory results three prerequisites rnust be fulfilled, narnely
first unbias, that is absolute detachrnent frorn bias and pre-
valuations, second, sufficient knowledge about the true
nature of the world and third, freedorn of action. These
three conditions have seldorn been fulfilled in the past,
today we have hopes of achieving thern.

Perhaps the rnost far reaching and successful effort

of all tirnes which did satisfy all of the rnentioned prerequi-
sites is the idea which was prornoted by Henri Pestalozzi
(1746-1827) and first practiced by hirn -- that knowledge rnust
be rnade available to every child and adult. Pestalozzi is

1
thus the initiator of general education, so:me of whose ideas
were so profound and far reaching that even today they have
not yet been sufficiently taken advantage of inas:much as we
have not yet achieved the training of the whole of :man which
he visualized.

To go back to World War I, those who lived through it,

well re:me:mber the general gloo:m all over the world and
also the ardent desire by :many of the best :minds to under-
stand what had gone wrong and to develop universal vistas
and to devise :methods of thought and procedure which would
enable :man to deal :more effectively than before with the
ever :multiplying co:mplexities of life. As a result the
world was co:mpletely re:modeled through the planned actions
of co:m:munis:m, of the Versailles Treaty, of the League of
Nations and the advents of Fascis:m and Naziis:m. The
nations which re:mained free likewise started on :many
planned develop:ments of their various potentials and they
in particular recognized the need of organizing and inte-
grating their scientific and technological capabilities.
Institutions like the brain trust of the New Deal thus ca:me
into being and were subsequently diversified and expanded
:manyfold during the Second World War in order to insure
the defeat of the dictators.

Since World War II enor:mous efforts have been :made

to develop general :methods of thought and procedure which
would allow us to deal efficiently with all of the co:mplex
proble:ms of the world and which would eventually enable us
to reach that :most desired goal, a unified world based on
:mutual respect and estee:m of all :men and of all nations.
The United Nations, the agree:ments about Antarctical
about nuclear testing and :most recently the Outer Space
Treaty resulted fro:m these efforts. In addition, literally
hundreds of groups all over the world were established for
the purpose of applying their technical and hu:man knowledge
toward the construction of a sound and stabilized world.

By way of illustration I shall :mention a few groups of

which I have. so:me personal knowledge since it has been :my
good fortune to have been associated with the:m. For instance

2
experts from all fields were brought together by the ingenious
founder of the Pestalozzi Foundation of America. Mr. H. C.
Honegger. to establish war orphan villages on all continents
and to deal effectively with the problems of destitute children
all over the world. In way of large scale constructive actions
for adequate housing the efforts of Profes sor Constantinos
Doxiades in Athens are outstanding, partly because of the
yearly Delos Conference which he has organized and in which
experts in absolutely every field of human endeavor partici-
pate. The establishment of the Cit~ de G~n~ralisation du
Canisy near Deauville, which is being promoted by the
French. is intended for occupation by experts and men of
universal outlook who will deal with large scale problems,
one after another. The Conf~rence des Sommets (Cultural
Top Conference) in Brus sels in 1961 should also be men-
tioned. At the invitation of the King and the Belgian Govern-
ment, outstanding representatives of all sciences. tech-
nologies and the arts were invited to attempt an integration
of all essentials of present-day knowledge. The organizer
of this conference, Francois Le Lionnais. pres ident of the
French association o~ scientific writers. had previously
edited a book, LA METHODE DANS LES SCIENCES
MODERNES. which may be regarded as a sequel to
Descartes' "Discours de la M~thode" and which contains
articles by s orne forty authors.

Finally, we organized the Society for Morphological

Research. one of whose purposes it is to bring all new
methods and procedures to the attention of a larger public.
In this endeavor two major projects have been started.
namely

1. A series of some two dozen comprehensive volumes on

new methods of thought and procedure in the sciences.
enginee ring, medic ine. law, and the arts and so on.

2. To arrange conferences periodically at which the ex-

perts of the different methods and procedures will be
brought together for discussions. Along this line a
first proposal was made several years ago to Dr. A. H.
Warner, then Director of the Office for Industrial

3
 Associates at the California Institute of Technology.
After Dr. Warner retired, the project had to be post-
poned but is now being realized in this sym.posium.
through the cooperation of Richard P. Schuster, the
present Director of the Office for Industrial Associates.

The purpose of the present sym.posium. was to awaken

a universal self-awareness of m.ethodology as a discipline.
Dr. Sim.on Ram.o in his inaugural address expressed the
belief that such a trend in thinking is already discernible
and will develop m.ore or less autom.atically. However this
m.ay be, those am.ong us who are active in the invention and
application of new m.ethods of thought and procedure want to
m.ake sure that all knowledge gained is effectively integrated
and widely dissem.inated. As Professor Henry Borsook of
the Biology Departm.ent cf C. Ie T. stated in his intro-
duction of the session on m.orphological research there has
been nothing like this since the latter part of the 5th century
B. C. in Greece. At that tim.e there was a great outburst
and activity in the subjects of logic. the nature of knowledge,
its transm.ission, exercise of power and the use of it. While
the Greek atom.ists were intensely interested in the facts of
nature, the sophists taught techniques, how to be successful
politicians, lawyers. generals, ignoring. however, m.oral
considerations. Justice for them. was nothing m.ore than
the interest of the stronger. Thus knowledge without wisdom.
produced som.e m.onstrous consequences. On the other hand
the wisdom. without em.pirical knowledge of Plato's Academ.y
could be nothing but ineffectual.

Today, after a period of m.ore than 2000 years of

accum.ulation of disconnected thoughts and procedures we are
attem.pting to integrate them. and to m.ake them. available to
every m.an. wom.an and child for the purpose of training the
whole of m.an.

F. Zwicky

4
 SECTION I

OPERATIONS RESEARCH
 CHAPTER 1

THE NATURE OF OPERATIONS RESEARCH

AND ITS BEGINNINGS

THORNTON PAGE
]V' esleyan Universzty

ABSTRACT

The development of military operations research in

World War II is outlined, with examples of several specific
studies. This historical account is followed through post-
war applications in U. S. industry, the organization of pro-
fessional societies and the recent applications by govern-
mental agencies to problems in the public interest. The
common features of these studies are identified as charac-
teristic of 0 - R, although the problems vary widely in
subject matter and scope. Programs of courses for 0 - R
training are offered in more than 20 universities and mili-
tary schools.

Further examples of time studies and interdiscip-

linary studies show a trend toward broader coverage of
significant factors in the operation of organizations involv-
ing men and machines. and of interactions between several
systems. The problem of information handling is considered
to be a basic one in future operations research.

7
 A. HISTORICAL INTRODUCTION

10 Definitions

Operations research, first called operational re-

search by P.M. S. Blackett (1) in Britain about 1938, has
a broad zneaning illustrated by several exaznples in this
section. Briefly, it znay be defined as the study of znan-
znachine systezns that have a purpose. In znore descriptive
terzns, operations research involves the application of
physical, biological, and social sciences in the znost quanti-
tative way possible. It thus draws on the disciplines of
znedicine, psychology, and all forzns of engineering (2). As
the following historical exaznples will show, an essential
step in every 0 - R study is recognition of the purpose of an
ope ration, an organiz ation, or a systezn. This often leads
to a zneasure of znerit or a value parazneter by which
operational results can be coznpared; for exaznple, the
fraction of approaching aircraft shot down by an air-defense
systezn, or annual sales of products by an industrial con-
cern, or gross national product of a country.

Predictions of such operational results, and quanti-

tative coznparisons between different systezns, require
znatheznatical znodels of each systezn and its operation. In
sozne cases--such as the accuracy of anti-aircraft fire, or
the cost of producing a znachined product, or transportation
costs of delivery- -the znodel can be znatheznatically precise
or deterzninistic. More generally, operational results are
probabalistic and require a stochastic znodel. The parazn-
eters to be used are based on statistical studies of znany
data- - such as ranges of radar detection of approaching
aircraft, or znonthly sales of a product, or the crizne rates
in znany different cities. When the znodel can be reduced
to a finite nuznber of rules so that operational results can
be calculated by a coznputer, or "gazned" by hand, the
process is called siznulation.

The success of a siznulation, or of an 0 - R study,

depends on taking all significant factors into account, and
this is the unique characteristic of operations research,

8
distinguishing it frOITl the various branches of engineering.
or froITl other bounded disciplines. For this reason, 0 - R
studies are often undertaken by teaITlS of specialists froITl
various disciplines (statistician. physicist. cheITlist, air-
craft des igner. ITledical doctor. and psychologist. for
instance). Thus "teaITl res earch" and synthesis are often
cited as characteristic of operations research.

2. Pre-1930

As "organized COITlITlon sense. " or "the scientific

ITlethod." operations research has been carried on for cen-
turies. In this sense. it is closely associated with the
developITlent of science; the beginnings of ITlechanics.
cheITlistry. geology. and psychology (to naITle a few) were
efforts to take account of ITlore factors than in earlier
natural philosophy. One interesting exaITlple is in Galileo's
"Two New Sciences" (3) where he points out that shipbuilders
could scale up the designs of ships. but had in fact found by
experience that larger ships would not hang together in
heavy seas--a result of the liITlited strength of the wooden
tiITlbers used.

More recently, in the late 1920's, an astronOITler at

Princeton (J. Q. Stewart) developed a ITlodel of hUITlan inter-
actions he called "Social Physics" (4). He used the Bell
Telephone SysteITl I s statistics on long-distance phone calls.
which then showed that the annual nUITlber of calls between
two cities is accurately proportional to the product of the
nUITlbers of telephone subscribers in the two cities divided
by the distance between theITl. He found that the saITle pro-
portion applied to out-of-state auto tags viewed at one point
on a highway. and that it ITlight also apply to "ITlilitary in-
fluence" and the outcOITle of wars taking place on a battlefield
distant D 1 froITl a country of population P and distant D
froITl an eneITlY country of population P 2' .As discussed 2
below, a battle ITlodel of siITlilar ITlatheITlatical nature had
been previously developed by Lanchester (5). By 1930,
several ITlilitary systeITls had been studied this way; naval
"strength" was set by the nUITlber and tonnage of battleships j

arITlY strength by the nUITlber of battle- ready divisions, etc.

9
Specific designs had received a good deal of attention, both
in military systems (such as fortifications. small arms,
artillery. tanks, submarines, and bombers) and in indus-
trial systems (such as steel mills, oil refineries, and rail-
roads). But a number of innovations such as radar. sonar.
aircraft-dropped depth charges ~ and torpedoes were being
developed. and there was no clear assessment of how these
new devices would affect military-system effectiveness.

3. Early World War II

With the start of war in Europe (and before) many

military problems became urgent ones. and groups of
scientists were formed in several countries to work on
them. Academic scientists. presented with such problems
as where to site radars in Britain, or how to counter bomber
attacks. or how to defend ships against submarines and mag-
netic mines. were able to take a broader approach to the
operational results of large systems. In many cases. there
was no time to modify existing equipment, so their efforts
were directed toward using it more effectively.

Two of the best examples, largely due to Blackett (1)

were the speedier action after radar warnings and the more
shallow settings of aircraft depth charges. It required little
mathematics to calculate the brief time between radar de-
tection and arrival of attacking bombers. Since the purpose
of the R. A. F. Coastal and Fighter Commands was obviously
to intercept the bombers, it was necessary to arrange rapid
communications between radar and fighter-aircraft bases;
the more fighters airborne on the bomber approach path,
the better the outcome.

Similarly, the rapid approach of aircraft toward

diving submarines meant that the submarine was usually
just below the surface when depth charges were dropped--
not as deep as when destroyers attacked in the more con-
ventional manner. Blackett was forceful in presenting his
conclusions (he became known as "the terrible profes sor"
to the R. A. F. and R. N.), and the results in increased

10
bomber and submarine kills proved him correct. At this
point he coined the term "operational research" for his
systematic review of repeated military operations. The
changes in procedure or strategy were later called "deci-
sions" by the military high command, and operations re-
search is sometimes called the "science of decision. "

4. U. S. Wartime Studies

Ve ry similar developments took place in the U. S.

armed forces just before we entered the war. Two large
groups of academic scientists (including myself) worked for
the Navy on underwater mines and anti-submarine warfare.
At first this effort was mostly engineering design. The
Germans had produced an effective weapon in the aircraft-
delivered magnetic mine J the British had countered with
magnetic sweepers and "degaussing coils" on ships, and
the U. S. Navy first tried to match these innovations. Be ..
cause the same group of U. S. scientists worked on both the
offensive (mines) and defensive (sweeping and degaussing).
it was natural to consider the interaction; that is. the
effectiveness of mines in the face of minesweeping and
degaussing. I participated in one of the first "diagnostic
war garnes" in early December g 1941. when we tried (on
paper) mines of various designs against the best possible
U o S. degaussing of naval vessels and sweeping of channels
in Pearl Harbor (6). The results showed that aircraft-laid
mines would have been almost as effective as the bombs
and torpedoes actually dropped there a few days later.

With the shift of military interest to the Pacific.

economic and geographic factors were added in a refinement
of this mine-warfare gaming; i. e. J the number and sizes of
ships of all types needed by the Japanese. and the depths of
water along ship lane s leading to Japan. The mine blockade
(7) carried out by the 21 st U. S. Bomber Command in 1945
confirmed the predictions of this study.

The anti-submarine warfare group (later called the

Operations Evaluation Group in the Navy Dept.) concentrated
on effective aircraft search for submarines. and devised

11
search patterns maximizing the probability of locating a
submarine after one position report. Another group in
London studied German submarine tactics from the many
torpedoings of Allied convoys, from daily radio fixes, and
sightings. Statistics gave the probability of submarines
sighting convoys (as a function of speed and size of convoy)
and the probability of penetrating destroyer-escort screens
(as a function of number of destroyers and number of subma-
rines}. An optimized strategy of convoying was derived,
based on shipping requirements and the number of destroyer-
escorts and search aircraft available. All this (and other
studies of a possible aircraft blockade of the German sub-
marine bases} involved a complex model of the submarine-
ship battle in the Atlantic. The "pay-off" {measure of
merit} was clear: maximize ship deliveries to Britain
during 1943 and 1944.

A group of operations analysts with the U. S. Air

Force bombers based in Britain studied bombing effective-
ness based on photographs taken from the bombing aircraft
and in later air reconnaissance. This led to quantitative
comparisons of planned bombing strikes in terms of damage
achieved per aircraft lost. Similar studies were carried
out in 1944 by an 0- R group with the 21st U. S. Bomber
Command in Guam, showing that low-level attack with fire-
bombs and sea-mines was more effective than high-level
attack with explosives.

The pattern of all these wartime studies {and many

others} was about the same: collection of operational data a
statistical analysis, agreement on the purpose (measure of
merit), picking the significant parameters (strategic deci-
sions), and predicting the optimum strategy, often by very
c rude quantitative anal ys is.

5. Post-war Military Studies

The wartime experience had proved the worth of

operations research, and U. S. military agencies retained
several semi-academic groups: Army's Operations Research
Office (O.R.O.)>> Navy's O.E.G ... Air Force's Office of

12
Operations Analysis (0. O. A.} and RAND Corp. and the
Joint Chiefs' Weapons Systems Evaluation Group (W.S.E.G.).
They conducted many studies in the ensuing decade. and
developed the techniques of quantitative operations research.
particularly "gaming" and mathematical simulation. The
diversified-team attack led to much broader studies. such
as the logistics requirements for overseas wars, the use and
suppression of guerrilla forces. the best use of manpower
in the U. S. Army ("integration" of Negroes started there).
and the economic effects of warfare.

The gaming and simulation required data on the ef-

fectiveness of men, machines 9 and weapons in various
combinations. There were fairly complete records of
battles and other military action to be analysed. One
example is the battle for Iwo Jima, a small island south of
Japan. In this isolated action, there were complete records
of forces on both sides, the casualties day by day, and the
outcome. The data were fitted well (8) by a deterministic
model involving Lanchester' s equations:

-dM/dt = Ai2M2

-dM 2 /dt = A2i Mi

where Mi and M2 are the opposing forces. and the A's

are "effectiveness constants. "

Fairly reliable data from military maneuvers. tests

and training exercises have been used in developing stochas-
tic models and the technique of "diagnostic gaming." One
such effort. carried out at O. R. O. (now renamed the
Research Analysis Corporation) determined the most effec~
tive design of tanks for a certain type of land warfare. The
game was played many times, allowing different tank maneu-
vering and the chance of lucky shots. Average outcomes {in
terms of tanks destroyed and positions over-run} indicated
the most effective tank. Of course~ not every possible com-
bination of men» tanks, artillery, and terrain could be
gamed. Operations analysts soon recognized the danger of
"sub-optimization" (9)--the best results in selected circum-

13
stances not being the best overall. Such reasoning led to
broader studies; one of the broadest I know concerned
nuclear warfare. a "game" carried out jointly by O. R. O.
and RAND from 1954 to 1957.

By this time the effects of thermonuclear weapons

were fairly well understood. The "nuclear-exchange game"
depended on combining these data with others on methods
of delivery. possible defenses. time intervals. political
limitations. and the economic effects. Of course. the cost
and effectiveness of defenses were involved. and the stra-
tegic decisions of surprise attack. size of nuclear stockpile.
choice between strike forces. defenses g and industrial
targets. and the use of radioactive fallout versus shock and
fire damage. Most of these considerations have been re-
counted in popular books and article s (1 O). and all have been
used over the past ten years in formulating U. S. national
policy.

In matters of this breadth. it is difficult to distin-

guish between operations research and a host of other
established activities such as good executive thinking.
administrative study, or "plain common sense." However,
I am sure that the systematic trial of various courses of
action. together with attempts to predict quantitative out-
comes, was highly developed for military purposes during
and after World War II. The method, the quantitative tech-
niques. and the search for all significant factors, are
properly called operations research.

6. Industrial Developments

It is not surprising that American industry developed

an early interest in operations research; the purpose of
industrial activity is easily defined, and the opportunity to
improve profits was universally attractive. Early attention
was given to inventory control- -the smoothed purchase and
production rates required to meet peak demands without
excessive storage costs. This soon expanded to studies of
factory and warehous e locations. chosen to minimize de-
livery time and costs. Many industrial production systems

14
involve several interrelated products, and decisions must
be made on the proportional output of each. One well-known
example is a study of the optimum outputs of a chemical
plant (11) where these proportions can be controlled to match
market demand and price.

Another topic studied early was on the handling of

empty freight cars by railroads (12). It was simple to show
that large sums of money were tied up unproductively in idle
freight cars, but more difficult to devise a car-handling
strategy that reduces such loss to a minimum in a variety
of traffic and scheduling patterns.

7. O.R.S.A. and I.F.O.R.S.

By 1952, the practice of operations research was

recognized and valued in American industry and military
agencies, but there was no professional structure. Three
universities (Case Institute, Johns Hopkins and M. I. T.)
offered seminars and training programs composed of courses
in economics, math, and engineering, but there was no gen-
eral recognition of operations research as a profes sion. In
May 1952, a group of 71 met for two days at Columbia Uni-
versity's Arden House; both industrial and military 0 - R
were represented in organizing the Operations Research
Society of America (13). Officers were elected (Po M.
Morse, R. F. Rinehart, A. N. Watson, and John Lathrop).
and various standing committees were established. It was
my job to produce a professional Journal. It started in
Nov. 1952 with a circulation of less than 500. Today, re-
named Operations Research, it has a circulation of 6000.
The Society has grown from 300 to 5000 members; it holds
two national meetings each year and several regional meet-
ings. A prize is awarded annually for the best 0 - R papers,
and a second periodical, "International Abstracts in 0 - R,"
was started in 1961.

Similar organizations were founded elsewhere: the

Operational Research Club in London (probably the earliest),
the Institute of Management Sciences in Cleveland, and 0 - R
societies in European countries, India. and Japan. By 1957

15
there were 15 separate nations with 0 - R societies. and a
committee was formed (including 7 British and 10 Americans.
under the joint chairmanship of Sir Charles Goodeve of
London and myself) to arrange an international conference
(14). This took place at Oxford in September. 1957, and was
attended by 300 people from 17 countries. It led to the Inter-
national Federation of 0 - R Societies, which has since held
three more conferences (Aixen Provence in 1960, Oslo in
1963, Cambridge, Mass. in 1966). Papers presented at
these meetings covered topics ranging from the operation of
coal mines to the flow of intelligence messages in military
maneuvers in Germany. They have established some unity
of the new discipline ~ and stimulated broader and broader
applications to problems of transportation, urban planning,
medical services, and national economy.

8. Operations Research in Government

Governmental agencies other than the military were

10 to 15 years behind American industry in organizing 0 - R
teams. Recently this field has developed rapidly. One of
the obvious applications is to the Post Office Department's
problems of sorting mail; others were found in agriculture,
social security. and air traffic control. During the early
'60 's advancement in Civil Service began to depend on 0 - R
or Systems Analysis experience. and a number of short
courses were offered in the Washington area. One series
of 2 to 5-day sessions, "Techniques and Methods of Opera-
tions Research," is offered by the Civil Service Commis-
sion's Management Training Center. (More advanced pro-
grams of courses are offered at the universities and schools
listed in Table 1.)

A conference in April, 1966, revealed that there

were then 52 government offices and agencies with 0 - R
studies underway in the Washington area, as listed in Table
2. In many of these, the purpose of the activity (measure
of merit) is poorly defined. That is. the public value of
rapid transportation, or convenient air schedules, or clean
air, or other public service. is not easily quantified. How-
ever, cost and time scheduling can be studied for several

16
 Table 1
Universities Offering Advanced 0 - R Training, 1966

University of California, Berkeley

University of California, Los Angeles
Cas e Institute of Technology, Cleveland, Ohio
Carnegie Institute of Technology, Pittsburgh, Penn.
Cornell University, Ithaca, N. Y.
The Johns Hopkins Univers ity, Baltimore, Md.
University of Maryland, College Park
Massachusetts Institute of Technology, Cambridge, Mass.
University of Michigan, Ann Arbor
U.S. Naval Postgraduate School, Monterey, Calif.
New York University, New York City
Northwestern University, Evanston, Ill.
University of North Carolina, Chapel Hill
University of Pennsylvania, Philadelphia
Stanford Univers ity, Calif.

17
 Table 2
Government Agencies with 0 - R Studies

Agriculture Dept. (Consumer, Marketing, Econ. Res.,

Forestry, Research and Dev., Conservation
Service)
Air Force (0. O.A., Scientific Research)
Armed Forces Industrial College
Atomic Energy Commission {Operations Analysis, BioI.
and Med.}
Army (R. A. C. , Combat Dev. Comd., Limited War,
Pers onnel, Civil Defense, Materiel Comd.)
Bureau of the Budget
Census Bureau
Civil Aeronautics Board
Civil Service Commission
Coast and Geodetic Survey
Coast Guard
Commerce Dept. (Business Service, Area Redevelopment,
Econ. Dev., Transportation. Plans Analysis,
Public Roads, Operations Research)
Defense Dept. (WSEG, Advanced Res. Projects, Res. and
Engineering, Supply, Systems Analysis, Cost
Analysis, Documentation Center, Communications)
Office of Economic Opportunity
Office of Emergency Planning
Export-Import Bank
Federal Aviation Agency
Federal Communications Commis sion
Federal Horne Loan Bank
Federal Trade Commission

18
Food and Drug Administration
General Accounting Office (Labor Statistics)
Health, Education and Welfare Dept. (Science Communica-
tion, Ed. Statistics, Program Planning)
Housing and Horne Finance Agency
Housing and Urban Development
Interior Dept.
Internal Revenue Service
Interstate Commerce Commis sion
Labor Dept. {Automation Manpower, Policy Planning,
Statistic s}
Library of Congress (Science Policy. Legislative Refer-
ence Service)
Maritime Administration
National Academy of Sciences
National Aeronautics and Space Administration (Resource
Anal. )
National Institutes of Health
National Institute of Mental Health
National Science Foundation
National Security Agency
Navy Dept. (O.E.G., Office of Nav. Res •• Nav. Res. Lab ••
Model Basin, Bureau of Weapons. Manpower.
Operations, Weather, Ships)
Naval Academy
Naval Medical Center

Patent Offic e

Peace Corps
Post Office Dept. (Transportation Research)

19
President's Council of Economic Advisors
Small Business Administration
Smithsonian Institution
Bureau of Standards {Technical Analysis Div.}
State Dept. (Arms Control, Overseas Program, Intelligence,
Technical Cooperation)
Office of Transportation
Treasury Dept. {Balance of Payments}
Veterans Administration
Weather Bureau

20
alternative programs. providing a basis for executive de-
cision in standard 0 - R manner (15). In fact. the Bureau
of the Budget now requires such studies. known as planned
programming or PPB.

Similar applications of 0 - R in State governments

are just beginning, following the pattern of the "California
Studies" initiated in 1964 with a budget of about $10 million
per year. The local problems of crime. waste disposal,
and transportation revealed the need for state-wide. con-
sistent data in very large amounts that can now be handled
with high-speed computing equipment.

It is, of course, too early to evaluate these recent

public service 0 - R studies. but I am confident that most
of them will payoff handsomely. in the same way that mili-
tary and industrial studies did from 1940 onward. This
brief history serves to define "operations research" and to
show its rapid spread to other activities after its develop-
ment for military purposes in World War II. The rest of
this chapter concerns broad changes in techniques. and the
two chapters following give modern examples of specific
0- R studies.

B. TIME STUDIES

One broad set of changes in 0 - R techniques has taken

place in studies of timing and scheduling. Early industrial
studies were centered on an individual's actions in various
production processes. The "efficiency experts" of thirty
years ago devised operational schemes requiring less
effort, and started the rise in American worker productivity.
More complex 0 - R time studies are outlined below.

1. Logistics

Typical inventory-control problems include the mili-

tary requirement for spare parts and ammunition for over-
seas units. The probability of need varies widely from one
spare part to another. and minimum supply levels can be
computed from these values and delivery times. Large

21
reserve supplies cut the mobility of land forces - -a serious
limitation in modern warfare. Therefore, army logistics
are organized in a heirarchy of supply bases fed by a "pipe-
line" from home industry. Changes are frequently neces-
sary when new equipment is introduced, all delayed by the
pipe-line flow time. Heavy military action increases the
requirement for most spare parts and replacements; ad-
vances by military units increase the delivery delay. and
the logistics planner must take such possibilities into ac-
count, as well as living within his budget. A fairly accurate
mathematical model is pos sible, if failure data and usage
data are available.

For such expendables as ammunition, there is a very

natural human reaction demonstrated in records of past
action: the more ammo on hand, the more rapidly it is ex-
pended. One of the Army's studies during the war in Korea
attempted to establish a norm of ammunition expenditure
and established that a good deal of artillery fire there pro-
duced almost no phys ical damage.

2. Scheduling

Introduction of new military machines or weapons

requires a complex schedule of subs idiary actions such as
(for helicopters) training pilots and mechanics. building up
supplies of fuel and spare parts, phasing out obsolescent
material, and foreseeing new operational requirements.
In some cases there are good reasons to time a military
innovation with some other planned action, and detailed
study is generally neces sary as to what schedule is feas ible
and at what cos t.

The best known example of an 0 - R scheduling study

was conducted in preparation for deploying submarines
equipped to fire long-range U.S. POLARIS missiles with
nuclear warheads. The resulting computerized P. E. R. T.
system (16) kept track of missile production, warhead pro-
duction, and submarine construction day by day, and indi-
cated which items were lagging behind schedule.

22
3. Queueing

A different type of tiITle study developed in the

Scandinavian countries and Britain. (It was naITled for the
queues of patient Britons waiting for bus or theatre.) When
individuals Or iteITls arrive in randoITl fashion at a proces-
sing point and have to wait in line (queue), there is loss of
ITloney, tiITle and teITlper. The length of queue obviously
depends on the chance spacing of recent arrivals, and the
length of wait depends on this and the processing tiITle.
There are significant applications to ITlessage handling,
traffic control. and toll booths on expressways (17).

Reco rds of cars arriving (in 1953) at the toll booths

for the Lincoln Tunnel and the George Washington Bridge
in New York showed a good fit to a Poisson distribution for
low traffic voluITle (less than 500 vehicles per hour) and to
a norITlal distribution for high voluITle. The average ser-
vice tiITle per vehicle was 11.1 sec. , but queueing generally
doubled the average vehicle delay. Results of this siITlple
study were the nUITlber of lanes (toll collectors) needed at
rush hours, and the superiority of left-side toll collection
ove r right - side.

More cOITlplex studies we re ITlade of queues where

ITlultiple services we re involved, or a wide distribution of
service tiITles (as in long-distance telephone calls), or
whe re prioritie s can be as signed, as in the handling of
ITlilitary ITlessages. Statistical analysis was shown to pre-
dict accurately the delay tiITles for various priorities and
servicing requireITlents at peak voluITles. One recent appli-
cation was to dock space in over-crowded ports, and the
distribution of shipping points neces sary to ITliniITlize delays
in loading and unloading.

4. Long-terITl DevelopITlent

In addition to randoITl arrivals at a toll gate, port,

or other processing point. there are discoveries and inven-
tions occurring at randoITl tiITles that bring radical changes
to warfare, to autoITlobile des ign. radio, TV, and the

23
electronic computer. Until the last minute, there is little
chance of predicting such "break-throughs." but it is well
known that the average rate has been increasing for several
decades in the U. S •• and that it peaks just after a war. An
0- R study made by the National Science Foundation (18)
correlated the invention rate with money spent on research
and development. a conclusion so widely accepted that the
go,:ernment and most large industrial concerns now budget
an appreciable fraction of their income for Rand D. expect-
ing a long-term pay-off of several hundred per cent.

One disadvantage of the R- and D-innovation sequence.

however, is obsolescence. A general trend noted in several
military 0 - R studies is the longer development-time re-
quired for complex modern weapons systems (such as long-
range missiles and nuclear powered submarines). The
useful life of such systems appears to be diminishing;
battleships were a prime weapon for 50 or 60 years. air-
planes lasted 30 to 40 years (one type now lasts 4 or 5 years).
The time may be approaching when a new weapon system
will be obsolete before it can be manufactured and ready for
use (19).

A similar pattern was to be expected in competitive

industry; even though it does not reach the absurd extreme.
rapid obsolescence has often made it impossible to payoff
development costs before the new product became obsoles-
cent. Such a pattern may possibly lead to an advanced form
of operations research study dealing with many whole systems.
their expected life-times. and scheduled replacements.
Several such changes have taken place in the past few decades:
in public transportation from railroads to buses and aircraft.
in army communications from field phones and messengers
to radio. etc. These examples themselves show that a new
system seldom replaces an older one entirely. Combinations
have some advantages; moreover, various systems interfere
with one another and may interact in other ways. Since past
0- R studies have generally been concerned with one system.
multiple-system studies should broaden the field considerably.

24
 C. INTERDISCIPLINARY STUDIES

It has already been stressed that operations research

is not limited to one discipline. Of course, there are differ-
ent degrees of synthesis. as the following examples show.

1. Human Engineering

In any man-machine system, there is an important

interface in the controls used by the men. Early design
work on aircraft cockpits brought together medicine. psy-
chology and engineering to aid pilots in seeing, hearing and
comprehending the many meters and other indicators, and
to provide controls easily reached and manipulated. The
resulting speciality, human engineering, can be a part of
operations research. Similar studies have been made of
many items of military equipment, all with the purpose of
increased effectiveness. After World War II an extensive
study was made of the rifle and other small arms used by
army infantrymen, including earlier designs as far back as
the Civil War (20). Factors considered were accuracy and
rate of fire, weight to be carried, effect on visibility. use-
fulnes s for other purposes, and psychological aspects.

Current human engineering studies sponsored by

N.A.S.A. concern the design of manned space probes and
the reactor-jet devices used for space walks.

2. Psychological Warfare

Team research by psychologists. sociologists and

military engineers produced several methods of inducing
enemy soldiers to surrender or defect. In addition to radar
broadcasts and dropping leaflets, there were loudspeakers
and searchlights used at the battlefront. 0 - R studies were
made to determine the relative effectiveness of these tactics
in actual warfare (21).

3. Manpower

Psychologists and sociologists also cooperated in a

25
1950 study of the Army's use of Negroes (22). The con-
clusion was that better use could be made if Negroes were
integrated rather than segregated. Quantitative data were
obtained on the proportion of Negro recruits, their perfor-
mance on intelligence tests, and their schooling. Opinion
polls showed that white officers who had served with
Negroes found them individually courageous, although
segregated regiments had poor morale and poor battle
records.

More recently. the industrial problem of techno-

logical unemployment has been studied by an 0 -R team
in the Dept. of Commerce who worked out a program of
retaining laborers in plants where automated equipment
is installed (23).

4. Public Interest

With the formation of 0 - R teams in many non-

military government agencies, operations research has
been started on several problems of public interest. Pos-
sibly the greatest synthesis is represented in the North-
east Corridor Transportation Study at several government
agencies, including the Technical Analysis Division,
National Bureau of Standards (24). Computer simulation
of the transportation net (railroads, buses, aircraft and
private cars) has been improved by five changes of the
model in an effort to fit population habits and needs. Esti-
mates of population and industry growth are used, and
various changes in the net tried (including high-speed rail-
road cars). The interdisciplinary combination of economics,
various types of engineering, and computer mathematics
has failed so far to provide an adequate measure of merit
that takes account of convenient schedules and little waiting
time in queues. Estimates of cost and the scheduling re-
quirements are of course a major part of the study. In
some ways it is a multi- system study of the broad type
considered in B.4 above.

26
 D. INFORMATION FLOW

Insofar as operations research is the science of

decision, and rational decisions must be based on informa-
tion, the subject of information itself, its flow in a man-
machine system, its amount and value, must be of signi-
ficance. In the early '50's Shannon (at Bell Labs), Wiener
and Bavelas (at MIT), worked on information theory (25) ,
cybernetics (26), and communication chains (27). The
disciplines involved were electronics, computer mathema-
tics, physics, and psychology. Amount of information was
(and is) measured in bits (binary digits that can be stored
in a computer memory); the analogy with negative entropy
sets an upper limit to the capacity of an electronic circuit
depending on the frequency band-width used; and the per-
formance of groups of people in solving problems jointly
depends on the number of communication links and their
capacity.

These concepts were soon used in 0 - R studies of

which the following are two military example s.

1. Intelligence Messages

In a military system "intelligence" refers to infor-

mation about the enemy. It is used here in a broader sense:
all information needed by a military commander in making
decisions. This obviously includes information on the loca-
tions of his subordinate units, their readines s for battle,
and instructions from higher command, as well as informa-
tion on the enemy. The difficulty is that relevant informa-
tion is transmitted step by step very slowly in the field army,
causing such command errors as artillery fire or air strikes
on positions already occupied by subordinate units.

During several army exercises and maneuvers in

Europe between 1954 and 1957. I measured message delays
and found them to be very large: median values of many
hours. In addition, there was further delay before the
commander and his staff could comprehend the info rrnation
{usually from a "situation map" where locations and activity

27
are plotted). Of course, a large proportion of the infor-
mation was irrelevant to anyone decision, but the effective
use of artillery, aircraft, and missiles required accurate.
up-to-date info rmation.

In a large army maneuver in northern Germany in

1955 a 5-man 0 - R team tested a more rapid radio com-
munications system linking the front lines directly to the
army general's headquarters and carrying selected cate-
gories of important information (28). The flow was faster
but it was not clear that the information was always rele-
vant to the commander's decisions. Nevertheless. this
realistic maneuver showed the phys ical poss ibility of such
communication links, and from the data collected we de-
rived probable errors, omissions 9 and outage times.

2. System Control

Further study in 1957 (29) produced a possible

design for a computer-linked army in which coded messages
were to be stored in rapid-access memories at all unit
command posts. The smaller, frontline units were to
store only immediate information limited to the unit's
sector. new information replacing the old. At higher eche-
lons the memory storage was to be larger, and retain a
j

longer time span as well as covering a larger sector.

Access to the stored information was to be by means of an
electronic plotter showing one or more categories of infor-
mation (front line pos itions. ammunition supplies. road
conditions. air- strike targets. enemy activity. etc.) on
demand. In case of loss of radio contact, the computer
was to be up-dated from higher echelon computers when
radio contact was reestablished.

The memory capacitie s and programmed output

at each command post seemed feasible. and the rapid access
to information would undoubtedly be valuable. but no quanti-
tative comparison with conventional army communications
has yet been made.

Similar applications in industrial concerns are also

28
feasible and, in liIT1ited forIT1" are already in use. (Many
IT1ajor airlines use a single cOIT1puter IT1eIT1ory to keep track
of reservations, and IT10st banks are installing cOIT1puterized
accounting systeIT1s.) In the years since this early work
was done, autoIT1atic data proces sing (A. D. P.) has developed
rapidly. I recently witnessed the trial of a rapid-access de-
vice at the United Aircraft Corp. Research Labs. where a
year's worth of tiIT1e-sheet records, pay checks. publica-
tions. and assignIT1ents for hundreds of scientists and
engineers have been entered on tape. PrograIT1IT1ed for
tiIT1e-sharing on a large cOIT1puter. the alpha-nuIT1eric.
office control panel allows the Lab Director (after he punches
a confidential control input) to see on a video screen the
sUIT1IT1ary of all overtiIT1e work during the past week (or
IT1onth, or year), all expenditures on anyone of several
scores of projects. the list of projects overdrawn on their
planned budgets. or any other sUIT1IT1ary of the input data.
He has found such rapid access of great value in IT1anaging
research activity--allowing hiIT1 to shift personnel to lagging
projects. prevent overexpenditures. and evaluate profes-
sionals for pay raises. The hardware design and cOIT1puter
prograIT1IT1ing are fairly siIT1ple. but the selection of infor-
IT1ation stored in the rapid-access IT1eIT1ory, and the types
of sUIT1IT1ary possible. resulted froIT1 an 0- R study siIT1ilar
to the IT1ilitary studies a decade earlier.

******
This sketchy history of operations research is by no
IT1eans cOIT1plete. but serves to illustrate the rapid develop-
IT1ent and wide applicability. The next two chapters cover
two recent studies in greater detail.

29
 REFERENCES

1. p. M. S. Blackett: "Studies of War," Hill and Wang,

N. Y., 1962.

2. Thornton Page: "A Survey of 0 - R Tools and Tech-

niques," Chapter 6 in "Operations Research and
Systems Engineering," The Johns Hopkins Press,
Baltimore. 1960.

3. Galelio Galilei. "Two New Sciences," translated by

Crew and de Salvio, Northwestern Univ. Press,
1946.

40 J. Q. Stewart, "Social Physics," Scientific American

1 7 8, 20, 1948.

50 F. W. Lancheste r, "Aircraft in Warfare," Constable,

London, 19160

60 Naval Ordnance Lab. Memos #1588: "The Desirable

Characteristics of a Mine," May, 1942; #1598.
"Proposed Mine Attack on Japanese Merchant
Shipping." May. 1942.

7. Strategic Bombing Survey Report, 1950.

8. J o H. Engel: "A Verification of Lanchester's Law."

O. R. S. Ao Journal ~, 163. 1954.

9. C. J. Hitch: "Sub-optimization in Operations Prob-

lems," OoRoSoA. Journal..!., 87.1953.

10. Herman Kahn: "On Thermonuclear War," Princeton,

1960; Business Week, 14 Jan 1967, p. 76.

11. R. L. Ackoff: "Production Control in a Chemical

Process," O.R.S.Ao Journal 1. 319,1955 0

12. Johnson and Kovitch: "Freight Car Distribution,"

Jour. Canadian O. R. Society..!.. 27, 1963.

30
13. P. M. Morse and Thornton Page, O.R.S.A. Journal
.!' 1, 18, 1953.
140 Proceedings of the First International Conference in
0- R, John Wiley, N. Y., 1958.

15. Roland McKean: Efficiency in Government through

Systems Analysis.

16. Charles Clark: "The PERT Model," O.R.S.A.,

Journal 10, 405, 1962; R. W. Miller: "How to
Plan andcontrol with PERT," Harvard Business
Review 40, 93, 1962.

17. L. C. Edie: "Traffic Delays at Toll Booths,"

O. R. S. A. Journal ~, 107, 1954.

18. NSF. "Government Research," circa 1960.

19. Thornton Page, "The Army and National Policy,"

ARMY Magazine, June 1956.

20. O. R. O. Report: "The Rifle," 1954.

21. O. R. O. Report: "Psychological Warfare," 1956.

22. A. H. Hausrath: "Integration of Negro Manpower in

the U.S. Army," O.R.S.A. Journal3,17, 1954.

23. N. B. S. Technical Analysis Division Annual Report

for 1966.

24. Aronoff and Levin: "Computer Simulation of Inter-

urban Travel," Technical Analysis Division N. B. S.
Greyback Report No. 9024, Sept., 1966.

25. C. E. Shannon, Bell System Tech. Jour. 27, 379,

623, 1948; Shannon and Weaver: "The Mathematical
Theory of Communication," U. of ill. Press, 1949.

26. Norbert Wiener: "Cybernetics," John Wiley and Sons,

N. Yo, 1948.
31
27. Alex Bavelas: "Communication Patterns in Task-
Oriented Groups," Jour. Acoustical Soc. 22,725.
1950.

28. O. R. O. (USARENR) Report No. S-65: "The PHANTOM

High-speed Repo rting System," 1955; Thornton
Page: "The Value of Information in Decision Making."
O. R. O. pamphlet, 1956.

29. O.R.O. PISGAH IV Report, 1958; O.R.O. STP-1:

"Intelligence Evaluation and Command Control in
Future Wars," May, 1958.

TEXTBOOKS

1. Morse and Kimball: "Methods of Operations Research."

John Wiley and Sons, N. Y., 1951.

2. Churchman and Ackoff: "Introduction to Operations

Research," John Wiley and Sons. No Y., 1958.

3. Flagle. Huggins and Roy: "Operations Research and

Systems Engineering," The Johns Hopkins Press,
1960.

4. Ackoff: "Progress in Operations Research, Vol. 1,"

John Wiley and Sons, N. Y., 1962.

5. Hertz and Eddison: "Progres s in Operations Research.

Vol. 2," John Wiley and Sons, N. Y. , 1964.

6. Macall: "Systems Engineering Handbook." McGraw-

Hill, 1964.

7. Enrick: "Management Operations Research." Holt,

Rinehart and Winston, 1965.

32
 CHAPTER 2

A DECADE
OF OPERATIONS RESEARCH IN HEALTH

CHARLES FLAGLE
The Johns Hopkim University

INT RODUCTION

Some of the roots of operations research have been

identified by Dr. Page in his paper. Three major ones
are apparent from the outs et in studies of the health s er-
vices, with somewhat different emphasis than in military
and industrial applications. First, the models of systems
behavior, particularly stochastic systems, were applied;
indeed, the role of the early operations researchers in
hospitals was almost a missionary one of revealing to
physicians and administrators the dominant effects of
chance on the time patterns of demands for available
resources. More will be said of this later.

The second stream of thought or body of techniques

to appear were the formal decision proces ses; optimization
proces ses such as mathematical programming. game
theory, inventory theory and decision theory. The optimi-
zation processes have been slow in making their effect in
health services research in spite of the apparent appropri-
ateness of decision theory to problems of medical diagnosis
and therapy. The daily administrative problems of assign-
ment of scarce skilled resources in hospitals has counter-
parts in the military and industrial situations that had

33
produced the necessity for invention of the techniques of
mathematical programming. Perhaps it has been the
absence of competition in the health services, or perhaps
it is that optimization techniques are refinements of
existing rational and developed systems. Health service
systems in the large lack these properties.

The third root is that of work study and the body of

pragmatic methods improvement techniques labeled, in the
past, Scientific Management. Many thought that these pro-
cedures could make immediate and useful contributions to
health services, being simply a carry-over from the in-
dustrial community to another form of human activity.
Generally, such hopes have been frustrated» perhaps be-
cause of the random, nonrepetitive nature of central actions
in medical care, or perhaps again becaus e of the absence
of competitive forces and specific incentive for dollar eco-
nomies. Under the impact of Federal legislation of the
recent years, a renewed interest in the substance and
history of scientific management is justified. In the history
of the scientific management movement in this country,
and the concomitant development of the behavioral science
approaches to the work situation, one can read the evolution
of the continuous proces s of adoption of technology, and the
achievement of a tenuous balance between man and machine
in the production proces s. Now it appears that hospitals
and health services must concern themselves with similar
problems. As noted by Feldman (1), expenditures on health
have doubled in the past ten years. and an increasing pro-
portion of labor costs indicates that the system is not mov-
ing or cannot move toward a less labor intensive technologi-
cal balance.

But a decade ago, although people spoke of taking

the long look, it was clear that the urgent problems of the
day were survival and the best use of the resources one
had at hand. Hospital structures were dilapidated; the
shortage of nurses forced hospitals to close some of their
beds. Physicians complained of overburden, and paradoxi-
cally expres sed a fear of los s of what they liked to call
"patient material. "

34
 BEGINNINGS OF OPERATIONS RESEARCH IN HEALTH

Two developments in the mid 1940' s had great s ig-

nificance for operations research in health. The National
Health Service was created in Great Britain, and among
the profound effects of this was the freeing up of funds of
the larger philanthropic agencies such as the Nuffield
Provincial Hospitals Trust and the King Edward's Fund for
Hospitals. Money that had gone directly to support patient
care now became available for other purposes and research
was chos en as an important area for development. The
counterpart in the United States was the passage of the Hill-
Burton Act for hospital construction. This act provided
some funds for research and demonstration in the areas of
hospital construction and management. Within a few years
res earch project groups began to appear, mainly in uni-
versities, since it was a constraint on the forms of financial
support then existing that they be granted to nonprofit insti-
tutions. Published results began to appear in the middle
1950's, and it is with these and the efforts that preceded
them that part of the rest of this paper will be concerned.
I have not tried to be definitive in listing projects. but have
been selective to show the nature of the work and its rela.;.
tionship to the real world problems that inspired it. Although
I have chosen to organize this review of work along the lines
of the disciplines or analytical approaches. it should be
remembered that the particular illustration was most often
embedded in a complicated real world problem. One did
not start out to apply queuing theory or decision theory.
but to understand and analyze and help change a perplexing
problem in human organization. It is a curious sidelight
that, being university-based, and staffed by faculty and
graduate students. much of the published work reflected
thesis research rather than problem solving effort. Studies
have had more apparent effect on the teaching of hospital
administration rather than the practice of it. Now a new
breed of administrators is beginning to appear. trained in
the recognition of systems and their analysis. Neverthe-
less the earliest efforts did have specific impact here and
there, and some interesting examples are to be found in
the results of viewing hospitals as stochastic and cybernetic

35
systems.

STOCHASTIC SYST EMS

Perhaps it was because the early operations re-

searchers were attached to large. busy, urban teaching
hospitals. perhaps by coincidence that their backgrounds
were in statistics g but for whatever reason, they quickly
noted that a familiar. erratic pattern of a random process
characterized the flow of patients into and through the
clinic wards. This was not just an academic observation.
The random demand processes produced congestion and
delays. and inte rmittentl y. idlenes s. This was to be ex-
pected. So many events occurring independently about a
large number of patients could be expected to produce some
classical queuing patterns 9 and indeed this was the case.

One important example is the pattern of census of

intensive care patients. These are patients who need a
great deal of attention and service. They are often bedfast
and critically ill. Work studies of patient care by Connor
(2) revealed several loads as well as a wide variation in
the distribution of aggregate patient care times. This led
to a somewhat arbitrary classification of patients into
categories of intensive, intermediate, and self care,
guided by the modes of the distribution of care times.
Figures 1 and 2 show how closely a Poisson distribution
of intensive care patients occur on a single ward.

Simultaneously, a similar classification system

was developed for a scheme called Progressive Patient
Care in which intensive care patients are recognized.
located in an area designed and staffed to provide the
specialized operations and treatments 0 The law of large
numbers provided a rationale for progressive care, for
by congregating the intensive care patients the variation
relative to average census could be reduced and operations
stabilized to some extent.

These studies of wards and clinics and analysis of th

probabilistic aspects led to some notions of organizational

36
design and administrative processes. The motivation for
a new synthesis was not to achieve some optimum, but to
design an operating system compatible with the nature of
the load placed upon it. Progressive patient care has
already been mentioned as one alternative form of hospital
organization. Here it was recognized that random vari-
ation in census of patients in the various classifications
was inevitable. and that through built-in flexibility "gray
areas" could be designed to care for several classes of
patients. The magnitude of the gray areas is shown in
Figure 3.

Akin to the notion of a broadened base of operations,

Thompson (3) and Blumberg (4) demonstrated the costs of
small specialized subsystems.

Another way of gaining the beneficial effect of a

broad base is through the use of short term forecasting
devices and allocation of resources from a centralized
pool. This was in effect the outgrowth of the previously
mentioned work of Connor.

From the examples cited above and similar work

two general models have evolved. The first of these was
the cybernetic system. As shown in Figure 4 we may
imagine many facets of the health service system to be
operating continually in a sequence of observation, analysis,
decision g action and feedback. For overall hospital adminis-
tration it is convenient to view the system as in Figure 5.
Here the world of events is the many (sometimes thousands)
of physician-patient relationships simultaneously in pro-
gress. Each generates independently a variety of demands
of the hospital systems in the form of physician's orders
for ancillary services 9 nurse's observation of patient needs.
Elaborate analysis and decision procedures are already
programmed into the hospital administrative process to
permit response to physician's orders without high level
administrative decision. All this is carried out through a
costly communication process. It is estimated by Gross (5)
that about one-fourth of current hospital operating costs is
in the formal communications process.

37
 Another ITIodel has eITIerged tiITIe and again in
health services research, a descriptive one dealing with
the flow of patients and personnel £rOITI one category to
another in the systeITI. The ITIodel appeared first in the
study of progressive patient care, where patients flow
froITI the outside world to and aITIong the three care cate-
gories within the hospital. The saITIe pressures that pro-
duced the interest in progressive patient care reflect a
broader probleITI, that of linking the ITIany subsysteITIs of
the health services. In addition to the categories of hos-
pital patients, such naITIes as "hoITIe care patients, "
"nursing hOITIe patients," "rehabilitation" and a variety of
other extended care resources ITIust be considered to
picture definitively the process of patient care in a COITI-
ITIunity. To be ITIeaningful the broadened picture, now
known as "coITIprehensive patient care," ITIust be concerned
with the decision rules, the adITIinistration processes for
transfer of patients froITI one facility to another. Res earch
in this area now involves probleITIs of the definition cate-
gories, and probleITIs of analysis of ITIultistate systeITIs
where flow is in part stochastic and in part deterITIinistic.
Figure 6 shows the cOITIprehensive care systeITI scheITIati-
cally. Note that the first state is the healthy population
period. The randoITI events producing flow £rOITI that popu-
lation into the health service systeITIs are the phenoITIena
of illness and accident. In several respects these phenoITI-
ena are age-dependent. The incidence, or rate of occur-
rence, of transition £rOITI health to one of the ill states is
dependent upon age. So also is the duration of stay in the
state. Both tend to increase with age. Since age changes
deterITIinistically the ITIodel ITIust sOITIehow take into account
the changing population in age cohorts and hence changing
incidence and prevalence in the various care states.
DECISION PROCESSES
One of the exaITIples of the cybernetic systeITI is
the physician-patient relationship, particularly the processes
of screening, diagnosis, and therapy. In association with
physicians the probleITIs posed to operations analysts have
largely concerned these ITIatters. Essentially the physician
is engaged in a gaITIe against nature. He knows sOITIething

38
of the prevalence of various diseases, that is to say, the
a priori probability that by chance the next patient he con-
fronts will have a particular disease. He has a battery
of screening and diagnostic tests, virtually all of which
have some degree of imprecision. One speaks of sensi-
tivity of tests, the power to detect the disease when it
exists, and the specificity of the same tests, their power
to detect the absence of the same disease. These numbers
are expressed as conditional probabilities and are almost
always less than one. The most difficult aspect of ex-
pressing diagnostic process in formal decision theory terms
is quantifying the loss function. The nature of the costs
of errors is well known. Scheff (6) has cited the high costs
usually placed by physicians on failure to detect a disease
when it is present.

An early example of a diagnostic or screening pro-

cedure expressed in decision theory terms is the attempt
by Flagle and Lechat (7) to select an optimal diagnostic
test and therapeutic strategy for control of leprosy.

A somewhat more complicated problem has been

treated by Parker and Lincoln (8) in differential diagnosis
of liver disease. Unlike the leprosy example, in which
very few test results were available, the differential diag-
nosis problem contains large amounts of data of mixed
kinds. The number of symptom complexes is large,
relative to the total number of patients in available samples.
and the problem is to detect relevant patterns of few symp-
toms.

This can be stated as one of the general aspects of

analytic processes in diagnosis and screening. With the
advent of automated laboratories, multiphasic screening,
and linked medical records, it is possible to assemble
large amounts of information about a single patient. Pro-
cesses of pattern recognition and discriminant analysis of
an almost endless variety appear to face the medical pro-
fession and all those who would assist it.

39
 WORK STUDY--TECHNOLOGY AND MANPOWER

It was noted earlier that the area of classical

organization theory would be considered last in this paper.
For several reasons its importance is greater now than in
the past. Perhaps it has appeared from the examples that
the health services organizationally are in an undeveloped
state- -or as some put it, are not rationalized. The formal
decision processes and the techniques of work measure-
ment methods improvement are usually schemes of refine-
ment of a system already functioning reasonably well. The
fact that stochastic processes are of such great interest in
understanding hospitals is that events are chance controlled.
To some extent this remains inevitable, and in retrospect
one can recall that many of the studies now identified as
queuing problems began as straightforward attempts at
work study and only took their final form when the dominant
facts of chance were noted.

There is a second explanation. historical rather

than inherent. for the lack of concern with optimization.
Historically health services have been bypas sed by the
stream of events. the constructive conflicts of other in-
dustries. The confrontation of legalistic descending
authority and the democratic notions of human rights
interacting with technological developments simply has
not occurred in the health industry. Health service re-
sources have sprung voluntarily from religious and chari-
table organizations and the involvement of staff has been
largely a matter of moral commitment to use the term
proposed by Etzioni to distinguish it from alienative or
calculative involvement. The way of life in the health
services has been reminiscent of medieval guilds and
religious orders. Not being involved in interstate com-
merce. hospitals have been delayed in accepting the work
practices that have justified increas ing technological de-
velopments in the production industr ies.

With the passage of Federal legislation in 1966 the

health services now face a necessity to bring work practices
into line with the rest of society. Increased wages and

40
time and a half for overtime have all become watchwords
within the past few months. A s a trans ient phenomenen
hospitals have responded to the increased demand placed
upon them by the new legislation by increasing personnel.
They have been called upon to explain the paradoxical
behavior of becoming increas ingly labor- intens ive with
rising costs. a pattern exactly opposite to other industries.
The explanation lies in the notion of the underdeveloped
system. There is no well developed labor saving technology
to draw upon. There has not been access to capital for
investments in labor saving equipment if it existed. The
task for administrators. researchers and participants
in the health services system is to direct the field into one
of a developed character where there is a balance between
man and machine. The challenge is to avoid the contro-
versies and antagonisms that have characterized industrial
technological advances.

41
 BIBLIOGRAPHY

(1) Feldman, P., "A Proposed Research Program for

Hospital Medical Care," U. S. Medicine, Vol. II,
No.8, July, 1966.

(2) Connor, R. J., A Hospital Inpatient Clas sification

System, Doctoral Dis s ertation, The Johns Hopkins
University, 1960.

(3) Thompson, J., et aI, "Computer Simulation of the

Activity in a Maternity Suite," (Actes de la 3 eme
Conference de Recherche Operationelle, Oslo,
1963, Dunod, Paris, 1964).

(4) Blumberg, M., "DPF Concept Helps Predict Bed

Needs," The Modern Hospital, December, 1961.

(5) Gross, M. and Jydstrup, R., "Cost of Information

Handling in Hospitals," Health Services Research,
Vol. 1, No.3, Winter g 1966.

(6) Scheff, T. J., "Decision Rules, Types of Errors and

their Consequences in Medical Diagnosis,"
Behavioral Science, Vol. 8,1963, pp. 97-107.

(7) Flagle, C. D. and Lechat, M •• "Statistical Decision

Theory in the Selection of Diagnostic and Thera-
peutic Strategies," Health Services Research,
Vol. 1, No.3. Winter, 1966.

(8) Lincoln, T. L. and Parker, R. D., "Medical Diagnosis

Using Bayes Theorem," Health Services Research,
Vol. 2g No.1, Spring, 1967.

42
DISTRIBUTION OF POPULATION OF INTENSIVE CARE PATIENTS
(Johns Hopkins Hospital Osler Medical Clinic-6th Floor 29 Beds)

Total Patient Sample 3308

(f) 20
I ntensive Care 908
~
Cl 18
I
lJ.J 16
u
~ 14
a::
::>
u 12 THEORETICAL (POISSON)
u
o DISTRIBUTION
lL. 10
o
8
>-
u
Z 6
lJ.J
::>
t1 4
a::
lL. 2

o 0 2 4 6 8 10 12 14 16
NUMBER OF INTENSIVE CARE PATIENTS
Fig. 1

PERCENTAGE OF INTENSIVE CARE PATIENTS

OSLER ME DlCAl CLiN IC - 6 TH FLOOR
(CAPACITY 29 BEDS)
60

50
>-
~
U 40
<t:
!L-

~ 30
lL.
0
~ 20
z
lJ.J
U
a:: 10
w
!L-

0
OCT
1957 1958
Fig. 2
43
 EXAMPLE OF BED ALLOCATION FOR PROGRESSIVE CARE
Interchangeab.
Fixed Intensive a Fixed Interchangeable Fixed
Intensive Care Intermediate Intermediate Intermediate a Self Self Core
a:X
W
IIl Z
~<I
::>:1: 1.00
zl-
(f)
u.(f) 0.80
OW
..J
>- 0.60
1-1-
~~ 0.40
1Il-
<II-
0.20
o1Il~
a:u.
0..0 O.OO~~~~~~--~~--~~--~~--~~--~~--~~

o 10 20 30 40 50 60 70 80 90 100 120 140 160

NUMBER OF BEDS - X
Fig. 3

CYBERNETIC MODEL
Fig. 4

44
 HIERARCHICAL MODEL

SHORT TERM HOSPITAL OPERATIONS

Fig. 5

STATE 0
EMIGRATION - DEATH

STATE
WELL

--
.,,-
".
----.--

ECOLOGICAL MODEL
(Unconstrained State Populations)

nj=f(Aijk' fLijk)
Fig. 6
45
 CHAPTER 3

OPERATIONS RESEARCH
APPLIED TO CITY PLANNING

MARTIN L. ERNST
Arthur D. Little, Inc.

ABSTRACT

Although interest in the application of operations

research to urban planning arose shortly after World War
II, few opportunities for practical work were available
until a decade later. At that time intense activity in the
area of transportation planning led to participation by
operations research workers in large scale studi:es of land
use in and around urban centers. More recent events,
such as the success of PPBS in Federal Government plan-
ning J have widened the spectrum for study of day-to-day
operations of cities. At the present time, however, we
have had only limited experie_lce, and this has been con-
fined almost entirely to a few 1ong- range planning areas.

In this paper a brief review is provided of some of

the areas of city operations where success in analagous
industrial work suggests that operations research could
make a contribution. Primary attention then is focused
on long-range planning, where an example of work is pro-
vided through a description of the San Francisco Commu-
nity Renewal Program simulation model. This model was
devoted to the study of the housing market in San Francisco J
and is intended as a tool for analysis of various city plans

46
to improve the hous ing of its inhabitants.

A brief discus sion is given of the role of operations

research in work of this type. Attention is then paid to a
variety of problems which make difficult the application of
rational planning techniques on a wide scale. These in-
clude: the lack of an appropriate political unit; inadequate
goals and objectives for our urban areas; decentralization
of political authority; difficulties in obtaining continuity of
effort; and difficulty in the integration of the many special-
ists involved.

47
 Operations research--in the modern formal sense
of the term--got its start under government auspices in the
military organizations. It should be no surprise, then,
that on return to civilian life many of the early practitioners
made an attempt to apply their knowledge and experience to
non-military problems of local and national government.
While a wide variety of possibilities were recognized, or-
ganization and support for such work was soon found to be
lacking. Military acceptance of OR had been born in an
atmosphere of emergency, a great leveler of fears of inno-
vation. No corresponding sense of crisis existed in our
urban centers during the 1940's and most of the 1950's, so
little willingness arose to try something new.

Accordingly, early activity was very limited, and

tended to be confined to academic institutions. The poten-
tial scope of operations research for urban problems was
noted at a number of universities and early courses taught
on this subject usually contained references to possibilities
in these areas. Lacking practical means for support, how-
ever, interest declined and the potential use of OR in urban
activities went into a period of hibernation.

The crisis of our cities began building up in the

late 1950 's and simultaneously interest in the subject
was renewed among operations research workers.
Papers on urban planning, and eventually full sessions on
the subject, began to appear at meetings of the Operations
Research Society of America. Most such work, however,
was concerned with the relatively long-range planning
aspects of certain urban problems - -usually originating in
the requirements to improve transportation from the
rapidly growing suburbs. The availability of funds from
the Bureau of Public Roads to develop highway networks
around and into cities furnished the spur for this type of
planning, which was devoted mainly to land -use problems.
These studies in general did not bring the operations re-
search analysts into intimate contact or conflict with the
authorities responsible for the working functions of a city,

48
so there was little immediate impact on city inhabitants or
government.

More recently. a variety of events have broadened

the spectrum of OR applications. The heavy us e of systems
analysis in the Defense Department. for example. has
brought recognition that improved and more formal struc-
turing of planning and budgeting procedures can lead to
beneficial results for a wide range of government agencies.
Action resulting from the President's Law Enforcement
Assistance Act. raised the possibility of applying OR to
assist in improving the effectiveness of our police forces.
Problems arising from delays encountered in Our court
system have produced studies aimed at better scheduling
procedures. In this atmosphere, operations research (and
the nearly synonymous activity. systems analysis) are be-
ginning to take root and flourish widely.

In this paper. I would like to briefly review some of

the potential areas where past experience suggests that
significant gains in effectiveness and economy of city opera-
tions can be achieved. For the major portion of the chapter.
I will then turn to longer range planning for cities. where
we have had more direct experience with both the problems
and the opportunities involved. Throughout the discussion
I will try to supply perspective through adequate emphasis
on the difficulties we face- -technical. political. and eco-
nomic - -as well as on needs and potential accomplishments.

B. THE VARIETY OF CITY PROBLEMS

Our cities provide. on a small scale. almost the

full range of problems encountered in our whole society.
One would expect. therefore. that those areas where opera-
tions research has already been successful elsewhere should
offer sources for potentially valuable work. The range of
activities meeting this criterion is far too broad to be
covered in a brief account such as this. but some sense of
the possibilities can be provided quite simply. Among the
subjects that corne immediately to mind are the following:

49
 Purchas ing and Maintenance

All large cities are involved in an im.m.ense am.ount

of purchasing. They require equipm.ent for operating de-
partm.ents such as police, fire. snow and waste rem.oval
and hospitals. and an im.m.ense variety of supplies for the
general support of civic services. Most of the equipm.ent
purchased will, in due course, require m.aintenance.
Clearly analytical techniques that have been applied to
parallel industrial activities should be able to furnish guid-
ance which can im.prove the effectiveness of city operations.
Inventory control techniques are entirely appropriate for
som.e of the purchasing activities 0 Cost/Effectiveness
justification and capital investm.ent and depreciation analyses
can be applied in m.any of the situations where cities are
purchasing expensive equipm.ent. Maintenance studies can
result in the developm.ent of m.ore econom.ical procedures,
the better recording and budgeting of m.aintenance costs,
and im.proved evaluation of equipm.ent alternatives.

To at least som.e extent, OR already has been applied

to all of these subjects on at least a narrow basis. However,
a better indicator of long-term. possibilities can be developed
from. industrial experience. The total expenditures by cities
in relevant areas are enorm.ous, and the level of im.prove-
m.ent in parallel industrial operations derived from. use of
analytical techniques has been sufficiently high to encourage
the belief that large savings to the taxpayer are theoretically
achievable. There are. however, a variety of practical prob-
lem.s due to the existing decentralization of authority am.ong
the various political institutions; this subject will be covered
m.ore extensively later.

Service Activities

A prim.ary function of m.ost cities is to provide a

variety of services; water supply, trash collection, fire and
police protection, etc. Many of the problem.s encountered
in the operations of these activities also have industrial
analogues that have been studied with s ucces s. The routing
and scheduling of trash collection equipm.ent has {m.ore

50
complex} parallels in the operation of the terminal phase of
many physical distribution systems. Fire station locations
and numbers could be subject to analysis on a cost/benefit
basis, first to determine the relative desirability of many
smaller units versus fewer large units, and then to select
optimum locations; in most respects, the work would paral-
lel procedures used in studying highly responsive physical
distribution systems. The importance of rapid response in
the availability of police personnel on the scene of crimes
is currently being studied; this can lead to a variety of
improvements in law enforcement, derived from better
communications and perhaps better allocation of resources
such as police cars and patrolmen.

Transportation

The flows of personnel and materials into and within

cities leads to the problem of urban transportation- -one
which we all recognize as increasingly critical these days.
Operations research clearly cannot, by itself, provide a
solution to a problem of this magnitude. However. the
techniques of operations research have already been applied
and are finding increased use for the study of this subject.
Modern traffic analyses could quite appropriately be con-
sidered as forms of operations research studies. and the
computer simulations increasingly being employed find
their parallel in many other areas of OR.

Health

Civic problems of health are largely concerned at

present with the provision of hospital services. Since we
are having a separate chapter on this subject, I will not
devote further coverage to this matter now. However. OR
is making contributions to public health in other areas of
potential interest to civic authorities. For example,
epidemiological analyses of diseases, such as tuberculosis.
are leading to a capability to map out public health programs
in a more optimum manner by leading to a better under-
standing of the diffusion of the disease and the costs and
benefits of various measures of control. There is even

51
evidence that sociological problems. such as drug abuse
and addiction, have epidemiological aspects which can be
studied effectively by OR techniques.

This list could be continued almost indefinitely. OR

contributions to the process of education and to the analysis
of educational effectiveness are just beginning. The whole
subj ect of welfare appears to be wide open to the use of
improved analytical procedures. It is clear that there is
much dissatisfaction with present methods and approaches,
and that there is great uncertainty in the evaluation of the
success of existing programs; under these conditions it is
quite natural to anticipate- -although without any guarantee
of success--the application of techniques that have been
found useful in other areas.

The types of problems discussed above correspond

to IItactical problems" in the military analogue. They are
largely concerned with the allocation of immediately availa-
ble resources to meet direct and immediate needs. There
is also a set of urban problems that could be termed
"strategic." They are by far more fundamental and diffi-
cult, and involve questions such as: Where are our cities
going? What should they look like 10, 20, or 50 years from
now? What steps can be taken to improve the likelihood of
reaching desired long term objectives? Operations research
does not provide answers to these questions; it does, how-
ever, provide procedures which can help us understand the
questions better, and which can keep us from making certain
types of mistakes. This area will be the main subject of the
remainder of this paper, but first I believe it desirable to
briefly describe the general role which OR can play in deal-
ing with these subjects. There is perhaps an intimation in
the earlier material that these techniques can do more than
should be claimed for them, so it is appropriate to describe
their limitations at this point.

S2
 C. LIMITATIONS IN THE ROLE OF OPERATIONS
RESEARCH IN PLANNING

To describe the roles of the various participants in

the development of plans, it is us eful to define a number of
components which must exist and be suitably integrated.
While the definitions given below are somewhat arbitrary,
they do provide a basis for delineating the necessary func-
tions.

1. Creative

If planning is to be a useful activity, it must involve

the choice of some course or courses from among a family
of alternatives. In situations in which no alternatives exist,
there is no requirement for a plan. The development of a
portfolio of possible paths of action is a creative function,
without which no meaningful progres s can be made. In
many areas of current difficulty we could be said to lack
suitable portfolios of choice, and suffer accordingly.

Although any member of society can participate in

the creative function, this activity is the primary domain
of a variety of specialists; economists, city planners,
architects, transportation planners, public health personnel,
sociologists, etc. The results of their work should be the
availability of choices for the later stages of planning.

OR analysts can contribute to the creative function

as individuals. and to some extent by assisting the primary
workers in this area through helping to frame appropriate
questions. However. their involvement here is a by-product
of their activity in other aspects of the planning process,
rather than a primary function.

2. Analytical

Once a portfolio of opportunities has been developed,

it is necessary to reach a decision concerning the allocation
of resources among these alternatives. This is the domain
of the OR worker or systems analyst, assisted by appropri-

53
ate specialists- -such as statisticians, com.puter program.-
m.e rs, architects and contractors ~ etc •• - - organized for a
team. effort. It should be em.phas ized ~ however, that the
purpose of the OR analysis is not to m.ake the final decision,
but to determ.ine the im.plications of following each of the
alternative paths of action. Although operations research
is vitally concerned with the decision-m.aking processes,
its goal should be to assist the decision m.aker. rather than
seek to m.ake the final determ.ination directly. sim.ply be-
cause not all relevant factors can be m.ade part of the
analytic proces s.

The analysis, therefore, is concerned with the costs

of each of the alternatives and the tim.es at which these
costs will be incurred; with the results expected to be
achieved, in term.s of who will be helped, how m.uch. and
in what way, and who will be hurt, how m.uch. and in what
way; with the extent to which a program. is apt to retain
flexibility for the future. or to determ.ine a direction of
m.ovem.ent which will be difficult to alte r late r, etc. The
presentation of results in this form. m.ay occasionally m.ake
the following decision a foregone one; m.ore often in com.-
plex situations. a further step is necessary.

3. Evaluation

While OR can describe m.any of the results of a

possible program. in quantitative term.s , there will always
be a series of assum.ptions involved in the process ~ a set
of non-quantitative outputs which m.ust be subject to judg-
m.ent. and a dependence of the quantitative results on cer-
tain policy decisions. In industry, therefore. OR analyses
m.ust be reviewed by senior m.anagem.ent so that these fac-
tors can be evaluated, and the experience and judgm.ent of
those bearing final respons ibility can be incorporated in the
eventual decision. In the field of urban studies. this step
is a function of the elected officials. Hopefully, these m.en
will be tuned to the desires of a city's inhabitants, and can
supply the political judgm.ent and policy decisions necessary
to select and develop an appropriate program.. The OR
worker should provide them. with a statem.ent of the near-

54
end far-term implications to be expected from following
each possibility, and the sensitivity of these results to
critical parameters and policies. The final evaluation
will incorporate these factors, plus the many which our
analytical techniques are completely inadequate to deal
with at the present time.

4. Commitment

The final step in implementing a plan, and one

which is often overlooked during the earlier stages, in-
volves commitment to the concepts on which the plan is
bas ed. The requirement for commitment is one reason
why political judgment is so vital if a feasible plan is to be
developed. The history of OR is strewn with cases of valid
technical work which was never implemented--Iargely be-
cause those whose support was needed for the program were
not willing to participate in the follow-through stage. The
OR analyst can be very heavily involved at this point,
largely in an educational role wherein he explains and
seeks to justify his earlier work. However, the main re-
sponsibility falls again on the political leaders. who must
rally a community in support of the plan.

I believe this framework provides a suitable per-

spective on the activities appropriate for operations re-
search (and for other groups of social scientists) concerned
with large scale social planning. Although the role may
appear to be quite constrained. it is a very vital one. Our
planning has often been ineffective in the past. because we
did not understand the implications of programs that had
great initial appeal. We have been slow to derive valid
goals and objectives for our urban areas, since the difficult
choices which would help sharpen and define these goals
usually have not been presented in a form which made clear
exactly what choices were being made. There are many
problems in political organization and administration where
our current structure is at best obsolescent, but the price
paid for our failure to recognize and adapt to the environ-
ment of modern technology has not been portrayed with
sufficient clarity to encourage change. In the long run,

55
as our techniques improve and our understanding grows,
it may be possible to expand the role of OR; but for the
moment, it appears desirable to accept limitations which
match our capabilities.

D. OPERATIONS RESEARCH AND LONG RANGE URBAN

PLANNING--A CASE EXAMPLE

1. Background

Major urban planning activities are primarily con-

cerned with land use and physical facilities, with the re-
quirements for these items strongly influenced by economic
and demographic considerations. There is a long history
in all parts of the world of efforts to improve the aesthetic
and physical environments of cities, both by wealthy indi-
vidual residents and by civic authorities. In the United
States, the Federal Government began to become an active
participant in the proces s during the 1930 IS. The incentive
for this action lay in the steady growth of sub-standard
housing in most of our major cities, and in strong evidence
that local authorities lacked the resources to cope with this
situation. Early efforts were directed towards solution of
what appeared to be the direct problem i. e. , the tearing
down and rebuilding of specific slum areas.

As experience was gained, more sophistication was

introduced into renewal programs, but they still tended to
be justified on the basis of ratings of the degree of physical
deterioration in different sectors of the cities involved. As
time passed it became apparent that projects based on this
approach were not broadly successful, and that the rate of
deterioration in our major urban areas had not been signifi-
cantly slowed. Meanwhile, a number of developments
offered the hope that a new basis for planning--modeling--
could be employed. The developments which made useful
models possible included the storage and processing capa-
bility of modern computers, the collection of improved
data on urban populations and physical facilities, and the
development of systems analysis techniques. Gradually
these led to a series of very large scale analyses, incor-

56
porating a multitude of models to describe the operations
and interactions of key parameters characterizing an urban
area.

The first and the largest of these efforts was the

Penn-Jersey Transportation Study, which has since been
continued by the Delaware Valley Regional Planning Com-
mis sion. A variety of smaller models have been developed
for planning in Baltimore, Boston, Pittsburgh, San Fran-
cisco and several other cities. Although these models have
not demonstrated any dramatic successes to date, they
have shown a potential for significantly improving urban
planning.

Existing studies have also demonstrated an enor-

mous degree of diversity. This is perhaps inherent in the
problems of our cities; although they share many attributes
in common, each city operates in a relatively individual
environment which will influence the background of planning
efforts on its behalf. Thus:

The Penn-Jersey Study was primarily concerned

with transportation planning. The analyses therefore
stressed accessibility to opportunities of interest to urban
dwellers (such as employment, shopping, recreation, etc.)
as a major factor determining the pattern of growth for
urban housing. Residential development in the more recent
versions of this work are based on a combination of neigh-
borhood characteristics and accessibility, and this approach
implies that the planning of transportation- -which is under
considerable government control- -can be used as a mecha-
nism to guide the character of land development.

For the study of Pittsburgh, particular emphasis

was placed on opportunities for industrial employment. The
primary allocation of land was based on the requirements
of those types of industry that are dependent on specific site
characteristics. This formed the starting point in the
modeling, with other forms of land use heavily dependent
on accessibility relative to industrial locations.

57
 The city of San Francisco is a cOITlpact residential
and cOITlITlercial city, with alITlost all of its available land
already in active use. The ITlodel developed for its study
took cognizance of these characteristics in that it was con-
fined to residential housing and was concerned with estab-
lishing the "ITlarket" operations which influence the quantity,
quality and costs of housing available. The object was to
study the influence of feasible local governITlent actions on
this ITlarket. Accessibility, industrial growth and other
factors of iITlportance in the other ITlodels were considered
only indirectly.

To illustrate the characteristics of a typical large

scale ITlodel I will describe briefly the San Francisco ITlodel
--the one with which I aITl best acquainted. Thereafter~ I
will turn to certain probleITls COITlITlon to all of our ITlodel
building efforts to date.

2. The San Francisco CRP Model

The background for developITlent of the San Francisco

ITlodel was partly described above. Since the city was es-
sentially fully developed, and sufficiently cOITlpact so that
internal accessibility was not a critical ITlatter, ITlany of
the actions possible in other cities to control land develop-
ITlent were not applicable. City capabilities to influence
future events were liITlited to a rather sITlall group of actions:
zoning law changes, iITlproveITlents in building code stan-
dards and enforceITlent, liITlited provis ion of iITlproved
aITlenities (parks, schools ~ civic services, etc.) and a
variety of forITls of selectively subsidizing housing costs.
Many of these actions will cost ITloney, and to the extent
that they are not supported by federal funds will result in
tax increases; this in turn will influence the housing rentals
charged, and in due course will lead to population ITlove-
ITlents.

With this environITlent~ it seeITled ITlost appropriate

to develop a ITlodel of the private housing ITlarket, so that
iITlplications of the liITlited pos sible civic actions on this
ITlarket could be studied. The resulting ITlodel could be

58
described as a si:mulation in which the interaction between
population and housing characteristics is studied in an
effort to deter:mine the factors that lead to increasing and/or
i:mproving hous ing provided and to rais ing or lowering the
rentals charged. The concept of e:mploy:ment of the :model
involves co:mparison of a base run. in which a city takes no
actions other than those already underway. with results
fro:m a series of runs in which specific plans for future
action are s i:mulated. To develop the :model. the following
steps were taken.

a. Land Division

Th¥ area of San Francisco was split into 106 neigh-

borhoods; these were each ho:mogeneous in ter:ms of their
residential housing, and had boundaries deter:mined by co:m-
patibility with 1960 census data. Each of the neighbor-
hoods was described in ter:ms of one of 14 locational cate-
gories. These categories were based on accessibility, the
general level of aesthetic values and a:menities, re lative
housing rental characteristics. and si:milar attributes. The
locational categories, therefore, furnish a general descrip-
tion of neighborhood "quality. "

Each neighborhood was further defined in ter:ms of

units of about two acres (one to two typical city blocks).
These units. called "fracts," represent collections of
s i:milar dwelling units. The dwelling units in a fract are
not neces sadly contiguous, and a fract is therefore not
geographically locatable within a neighborhood. However,
a neighborhood can be described in ter:ms of its locational
category and the nu:mber of fracts of each type that it has ..

*As will be described later, the San Francisco :model has

been unde rgoing :modification and imp rove:ment; the des-
cription provided herein relates to the first operable
version rather than to the current :model. The :mecha-
nis:ms, however, have been changed only slightly; :most of
the alterations have been in the for:m of aggregating inputs
and outputs.

59
The fracts formed the "quanta" in terms of which all
changes in type or condition of phys ical hous ing space in
the city could be made.

b. Land Use

Twenty-two basic land use types were considered in

this study, defined in terms of the type of dwelling unit in-
volved, the size of the typical units, and the tenure (owned
or rented). In addition, one of four states of physical con-
dition--ranging from sound to dilapidated--was associated
with each type of housing (except vacant land and govern-
ment buildings). Every residential fract in the city was
therefore classified in terms of the locational category of
its neighborhood, the type of land use involved, and the
condition of its housing.

c. User Classification

The population of San Francisco was divided into

114 household types, based on household size, race, income,
and rent-paying ability. Rent-paying ability was described
in terms of a distribution, based on census data, rather
than in te rms of fixed amounts. For each hous ehold type,
a ranked preference list for housing was developed, specify-
ing the desirability of economically feasible types of hous ing
sought by that group. Some 50 combinations of dwelling
types, conditions, and locational categories were included
for each household type.

d. Other Data Items

A wide variety of other data items we re employed

to support component models incorporated in the simulation,
and to permit various civic actions to be taken. For ex-
ample, the costs of all feasible transitions among housing
types {i. e. , costs of improvement or construction} were
employed as part of the sub-model wherein the incentives to
make transitions were evaluated. Zoning laws mechanisms
were developed, making it possible to restrict the housing
transitions which could take place in some areas. Similarly,

60
code enforcement rules were constructed, capable of
forcing the renovation of a specified fraction of the sub-
standard housing in particular neighborhoods. These and
a host of other elements led to an eventual requirement
for some 15,000 data input items.

e. Simulation Structure

A brief examination of the simulation structure will

further indicate the many areas of analysis and data col-
lection which had to be undertaken. This is shown in
Figure 1.

1. The simulation moves in two year time peri-

ods. This period was s elected as an appropriate one for
major steps to be taken in a development project involving
at least one fract or area. In other words, two years is an
appropriate period for acquisition of land by the city, or for
clearing land once acquired. or for construction on cleared
land.

2. At the start of each new period. the housing

stock of the city was "aged." A separate component study
had developed a model of aging characteristics of San Fran-
cisco. based on data over an extended time period. Special
provisions were made for a form of accelerated aging,
sometimes called "blight" production, which arises when
economic conditions clearly make it profitable for a land
owner to allow his property to deteriorate rapidly.

3. Population data for each time period were

introduced exogeneously, based on either estimates of
expected popUlation growth. or on a desired target popula-
tion. A more sophisticated approach would involve use of
a feedback loop, wherein the housing and economic charac-
teristics of the city, relative to those available outside the
city» could influence the in or out-migration. This is one
of many pos sible future improvements 0

4. Specific planned public actions can be put into

effect at the start of each time period. These can include

61
direct physical changes. such as public construction; or
they can represent regulatory changes. such as those
effecting zoning laws. code enforcement. etc. Finally.
various forms of subsidization. changes in taxes. etc ••
can also be scheduled in and out of the model at the start
of any given time period.

5. The model then proceeds through an alloca-

tion process. in which user groups select desired space
primarily on the basis of their relative economic power.
This allocation is not intended in any way to represent real
population movement. The many frictional forces lowering
mobility within a city operate to prevent attainment of the
equilibrium condition suggested by this allocation procedure.
The allocation instead is oriented towards determining how
much unfilled demand there is for a given housing type.
based on user preferences within their rent-paying ability.
This demand is developed in the form of a pressure which
can result in rent increases--or in decreases if the pressure
is less than unity. The rent changes themselves can in-
fluence the later steps. wherein the economic incentives
towards changes in land us e are simulated.

6. Based on the pressures developed in the pre-

ceeding steps. a list is developed of possible transitions of
fracts into the types for which there is high demand. This
list is checked for feasibility (i. e. , units will not necessarily
be aged rapidly to fulfill a demand for older houses. etc.)
and for compatibility with regulations such as zoning laws.

7. In the next step. the economic yields to owners

resulting from possible transitions are determined. based
on the costs of effecting the transition changes and the
rentals which are available before and after the change.
These yields are then compared with standards. which can
vary in accordance with locational category. building size.
etc •• at the end point of the trans ition.

8. A limited number of acceptable transitions

are then executed. and a recycle takes place. The recycle
performs the allocation again. determines new pressures.

62
recomputes yields, etc. Successive recycles characteris-
tically result in smaller numbers of transitions being made.
and the total transitions can be limited to represent a
variety of facto rs. such as: unavailability of mortgage
funds; unavailability of labor without use of overtime and
resulting in high costs; lack of information by owners.
resulting in slower changes and construction than would
arise in a perfect market.

9. The entire model then proceeds to the next

two year cycle.

3. Model Evaluation and Use

The most common mechanism for evaluation of

models of this type involves "tuning " the various model
parameters from data applicable to part of a past time
period on which information is available. and then con-
ducting runs to determine how well the model duplicates
actual results occurring after the tune-up period. In the
case of the San Francisco model this was not possible.
since the detailed census data necessary for tuning the
model were not available for other than the 1960 census.
This data. plus information on construction during the
period 1960-1964, had to be used for tuning purposes. As
a result. no claim can be made that the model is validated.

In this respect the San Francisco model represents

an extreme case of a fairly general problem. Large scale
models of social activities can seldom be validated effec-
tively; they can only be adjusted to duplicate events of the
past in a reasonable manner. In almost every situation,
the variety of possibilities for the future defy our ability
to construct the component interactions necessary to be able
to forecast with assured accuracy.

This deficiency is not as severe as it first sounds

0

To date the San Francisco model has received employment

and has proven useful without making any strong claims to
validity. Typically a pair of runs, as described earlier
{i. e •• a base run and a trial run}, will produce a series of

63
res ults. Many of the s e will match the expectations of com-
mon sense and experience, but a few surprises will arise.
When these surprises occur, it is necessary to trace them
back through the model interactions to establish why they
arose. In some situations (particularly in the early stages,
when programming and input data errors were common) it
could be determined that the surprises were due to faults
in the model. In other situations, however, it turned out
that the surprises represented the results of a series of
interactions that had not been anticipated, but which once
demonstrated turned out to make very good sense. In this
manner, the validity of the model has been of less impor-
tance than its educational function.

We anticipate that the future pattern of model use

will be similar to that which I have just described. The
model, after careful checking of its results and the estab-
lishment of reasons for departures from anticipations ~
presents tentative implications; it thereby provides a more
concrete basis for discussion of specific policies and their
likely results than any other tool we have available.

The original model was used to check the validity of

a Community Renewal Program which had been developed
on a more traditional basis. In general it substantiated this
program, although for SOme elements it raised questions
for further checking. At the present time the model is being
revised, primarily to aggregate inputs and outputs for eas ier
employment. It is expected that in corning months it will be
employed to test specific possible plans for the city of San
Francisco. If it proves useful in this role, it can be antici-
pated that it will gradually bec orne a standard working tool
of the municipal government. The success of the model,
therefore, is still in question; the next few months will deter-
mine it.

E. BASIC PROBLEMS OF APPLYING ANALYTICAL

METHODS TO URBAN AFFAIRS

The description of the San Francisco CRP Model

provides an illustration of the type of analysis that has be-

64
corne practical in recent years It must be recognized,
0

however, that this type of study- -and urban OR studies in

general--must be conducted in an environment that places
severe limits on the applicability and utility of their re-
sults. In this section we will consider some of the prob-
lems which make difficult the employment of analytical
techniques, regardless of the validity of their results.

1. Selection of Appropriate Political Unit

The political unit on which a study is based is nor-

mally established by the nature of the assignment. If work
is being done for the city of San Francisco, it is difficult
to devote significant effort- -or to count on necessary assist-
ance- -for the analysis of events or characteristics in the
region outside the city. This situation is a particularly
unfortunate one, since it is obvious that the future of all
our cities is critically dependent on events beyond their
political boundaries. Inhabitants of the San Francisco
area, for example, will make a choice between residing
in the city or in the suburbs based on a variety of circum-
stances over which the city has only limited influence.
Substantial improvement in the quality of low cost housing
in San Francisco would undoubtedly give rise to in-migra-
tion from the surrounding areas, and there will be a ten-
dency to move in the direction of an equilibrium condition
throughout the entire Bay area. The mobility of population
in the United States is sufficiently high that no limited area
can be studied effectively without questions of this type
arising.

Unfortunately, we are at a severe disadvantage in

that there is no political unit corresponding to a metro-
politan area. Although special metropolitan authorities
are often formed. usually in connection with transportation
planning and operations. even these normally are quite
limited in their effective power. Our cities and their
hinterlands are in relatively violent conflict on many criti-
cal issues. such as housing, schools. transportation and
taxes. Although analytical models might assist in resolving
some of these conflicts. there is little realistic hope of

65
their near-term use in this manner. As a result, plans
which are developed for a suitable political unit- -the city--
tend to be greatly weakened by their limited treatment of
the hinterlands; while plans developed for a more natural
unit--such as a metropolitan area--simply cannot be imple-
mented because of lack of political cohesiveness.

2. Goals and Objectives

Until a great deal of competent basic research has

been performed, we are unlikely to be able to establish
rational goals and objectives for our cities. We can recog-
nize that the purposes of cities have changed from those of
the past: cities are no longer important for defense reasons,
or as storage points for local inventories, or as trading
posts. They are serving less efficiently as transportation
centers than formerly. Their role as cultural centers is
still important, but even this is weakening in the face of
mass communications mechanisms such as television.
They are still important for personal contact, but this in
time may decrease--again through improvements in com-
munications methods. We can re co gniz e that the role s of
cities are changing, but we do not yet have a firm basis for
saying what they should be in the future.

For the purpose of OR applications to urban planning,

this lack of accepted goals is not critical. Although it limits
our ability to establish formal measures of effectiveness,
it does not prevent the study of implications of possible
actions--which I consider an appropriate role under the
circumstances. Although the most difficult problems are
not directly encountered in this role, OR analyses can be
important in aiding the gradual development of goals and
objectives. The very process of examining the implications
of feasible civic actions, and making a clear selection, will
in itself assist in developing a rationale from which goals
and objectives may eventually be better defined. At the
present time there are few mechanisms for restricting the
pursuit of apparently desirable objectives through deter-
mining their feasibility, and this restriction is necessary
for intelligent discussions of goals. There are a host of

66
controversial measures related to the subsidization of
housing and education which inspire conflict of opinion in
urban areas. To at least some extent rational analyses of
feasibility and costs can assist in resolving some of this
conflict. and making more clear what true choices are
available to urban dwellers and their political authorities.
The process will be a slow. educational one, but an impor-
tant one for the future.

3. Decentralization of Authority

Both near term and long term analytical work for

cities suffers from the decentralization of political author-
ity characteristics of these organizations. In the long-range
planning field. most cities have a variety of concerned de-
partments. with overlapping authority and responsibilities.
To a certain extent the overall planning authority of any
city is relatively restricted. and often highly dependent
upon the availability of federal funds to be effective. The
planning proces s is becoming inc reasingly complicated.
bec aus e of the many partic ipants • The alte rnati ve - -plac ing
a far higher level of authority in a single body--also has its
drawbacks; it must simply be recognized that when planning
affects large numbers of people. we must look for com-
plexity in the planning operation. One result is that a con-
siderable portion of the total effort of analysts must be
devoted to non-analytical work. dealing with and educating
a variety of participants in the overall program.

In the area of tactical studies, the decentralization

and autonomy of city departments is perhaps even more
severe. In the area of purchasing, for example, authority
is apt to be widely split and in the hands of a host of indi-
vidual departments. To a certain extent, the ability to
purchase items up to a given amount is practically a pre-
requisite of certain political jobs. and will not be sur-
rendered easily to a more organized structure which can
make use of analytical studies. Many of the city depart-
ments. such as police and fire departments. have had a
high degree of autonomy for a long time. and have acquired
some of the characteristics of a very exclusive club. The

67
participation of civilians of any sort in some of these de-
partments. much les s analysts who are studying the details
and efficiency of current operations is not apt to be welcome.
The city's services are developed by a host of specialists.
and each of these has been faced with practical problems.
limited budgets. and lack of recognition and response for
many years. Except in a period of crisis ~ it will be difficult
to achieve change. As suggested earlier. the barriers are
beginning to break down, but progress will not be rapid.

4. Continuity of Effort

It is difficult to achieve continuity of research effort

in studies of our cities. As has already been suggested.
there is great diversity in their environments, so that
studies usually must be tailored to individual cities. The
funding mechanism for studies also encourages this process,
since the majority of funds tend to be devoted to specific
projects related to individual cities. At the present time
there is no simple method for achieving general continuity.

In recent months some interest has shown in the

development of an organization. such as a non-profit insti-
tute, for the specific purpose of developing analyses and
studies of urban problems. The Department of Housing and
Urban Development also has available funds for research
in a variety of important areas. It is still uncertain, how-
ever, what fraction of thes e funds can be devoted towards
generalized studies. or to making techniques useful in one
city applicable in others. In general there are a lot of in-
centives and motivations to continue the past policy of
funding studies primarily devoted to one city at a time.

5. Integration of Specialists

Study of either daily operations or long range plans

for urban centers requires a host of specialists, as noted
earlier. Historically, these specialists have had few op-
portunities to work together in an integrated fashion. As
the latest intruder on the scene. the systems analyst tends
to be viewed wi th particular suspicion; his role and his

68
capabilities are not widely understood, while his publicity
in recent years has been almost too effective. Most large
scale efforts, therefore. have tended to develop consider-
able internal friction.

Earlier in this chapter. I presented a structure of

roles which I believe can assist the various specialties in
working more harmonously together. It will take time,
however. before the different activities can develop an
understanding of one another. and an ability to coordinate
their efforts in a comfortable fashion.

These environmental difficulties quite properly can

be considered an inherent part of the problem of working on
urban affairs. While they are restrictive in nature they do
not destroy the possibility of effective work. Over the long
term. we can hope that changes in the environment will be
encouraged by analyses which indicate the price paid for
these environmental features and suggest more suitable
alternatives.

In summary then, we have had some experience with

the large. longer range problems of urban areas o This
experience has been beset with a number of difficulties but
has had at least marginal success. Some significant possi-
bilities for the future have been demonstrated, but expecta-
tions may be higher than past performance merits. For a
variety of reasons, opportunities for the application of
operations research appear to be increasing, particularly
in the analysis of day-to-day operations of cities. The
primary reason for expectation of continued growth, lies
les s in the effectivenes s of past performance than in the
fact that our problems are growing. and that better means
for handling them have not yet come to light. On this basis,
we can look forward to continuing interest and opportunities,
and a great deal of hard and challenging work.

69
 SIMULATION OPERATION

Initiate a New Time Step;

"Age" Space Stock

l
Insert New Population Data;
Execute Planned Public Actions

~
Allocate Available Users to Space;
, Compute "Pressures" of Unfilled
Demands and Resulting Rent Changes

J
List Potential Trans itions of Space Types
and Check for Zoning Restrictions, etc o ,

Compute Yields of Trans itions From

Costs of Changes, Changes of Rent, etc o ;
Compare Yields with Yield Standards

l
Execute Acceptable Transitions up
to Specific Limits

~
Summarize Results and Move to
New Time Steps

Fig. 1

70
 SECTION II

SYSTEMS ENGINEERING
 CHAPTER 1

SYSTEMS ENGINEERING-PLANNING

DEAN GILLETTE
Bell Telephone Laboratories, Holmdel, N. J.

INTRODUCTION

The term "Systems Engineering" means something

different to almost every writer and lecturer on the subject;
*
indeed, the authors of this and the companion paper do not
agree, in detail, on a single definition. Consequently, the
viewpoint that we have adopted is so broad that every practi-
tioner should be able to find a niche for his particular model.
Throughout, we will assume that a "system" is a collection
of equipment and a plan for its operation that will make the
complex serve a definite function. Systems Engineering.
whatever it is. will be assumed to permeate the creative
proces s that starts with a concept and ends with the release
of the system to a user community as a tested, material
means of fulfilling a need or providing a service.

The origins of the proces s lie in the recognition of

a way to assemble equipment, and to power and operate it,
to serve some purpose. As the purpose is more clearly
identified and made quantitative, it becomes dignified enough
to be labeled an "objective." The equipment concept is an

*J. F. Shea, "Systems Engineering--Implementation,"

this volume.

73
idea of how to use materials and methods to meet the ob-
jectives, and can be only as realistic as is the creator's
understanding of the underlying technology. Broadly,
Systems Engineering is the art in turning such a concept
into a reality. There are many activities in realizing a
concept, including mathematical modeling. preparation of
manufacturing info rmation. development of amain tenanc e
plan. tooling. installation and testing. The project systems
engineer will help to plan the details in many of these areas.
but, most importantly, he is uniquely responsible for under-
standing the relationships among them and for making sure
that the process is orderly and complete.
This description is recognized to be only a discussion
of the place of Systems Engineering in the innovative process,
and not to provide a definition that could satisfy the mathe-
matician or pedagog. However, just as the systems engineer
must often be satisfied, in early design stages, with describ-
ing the operation of the system in terms of the functions, or
outputs, of subsystems, so will the authors in these two
papers attempt to describe Systems Engineering in terms of
the actions in the output of various aspects of the work.
The two talks are titled. "Planning" and "Implemen-
tation." These phases of Systems Engineering overlap sig-
nificantly in time; further, it is usually impossible. in a
productive organization, to identify one group of people as
the planners and another as the implementers. There may
be organizational or administrative advantage in charging one
group with the planning function and another with implemen-
tation, but if the planners lack a sound knowledge of the basic
technology or of the vicissitudes of implementation, or if the
implementers are not adept at task organization or in attaining
a balance between performance and cost. the system may be
a disaster. either functionally or economically. We can agree
on an operational division. however: planning results in in-
formation, in the broad, not technical, sense, and implemen-
tation in operating equipment. The bulk of the remainder of
this pape r is devoted to the products of the planning function:
information for management, for the implementation function,
and for additional planning. Since several of the papers to be
presented deal with mathematical modeling, the author will

74
take the opportunity to display some of his prejudices on
the value and use of such models in planning. Finally.
because this symposium is being held within academic
chambers, and is attended by educators. the author cannot
res ist making a few concluding remarks about the education
of a systems engineer.

INFORMA TION FOR MANAGEMENT

The cost of supporting a planning activity is small in

comparison with that of implementing a system. Planning
may involve, in addition to straight manpower. exploratory
development- -trials of application of a technological ap-
proach--or the use of large-scale computing equipment in
analysis. but the manpower and physical equipment needed
in implementation, even before full manufacture. is much
greater. Consequently. a deliberate decision must be made
before proceeding beyond the paper stage of a project.
"Management. " here, is taken to be that authority that pro-
vides the funds for the continuation of the project, and ulti-
mately is responsible for the risks associated in such ex-
penditure. In the next paper, Shea will use the term
"Management" in a different sense--that identifying the
group respons ible for seeing that the project program is
carried to completion. The first product of the planning
function is the information for Management that will provide
the technical and economic data needed for its decision to
commit funds for implementation.

In a way. the simplest part of the material to pre-

pare for Management is the desc ription of the system: what
it is intended to do, how it is to be operated. what new
functions it is to perform. what service it will supply. and
what it may replace. Technical people involved with the
project will have no difficulty in supplying details here; the
biggest problem may be to write a succinct. but complete,
description.

Since complex systems rarely arise as unique inde-

pendent entities, it is necessary to describe the compatibility

75
of the system with the technical environment within which
it is to operate. When the system is to be a replacement
for an existing one, for example to achieve greater economy,
a plan for introduction must be developed. When the system
is an addition, for example to provide growth or greater
capability, it must be de signed to op erate with exis ting
facilities and equipment. Management must be informed of
any administrative problems. as well as costs, that can
arise with installation of the system.

The proposal to Management should also include

schedule data: milestones for critical steps in implemen-
tation, system trials, installation, personnel training.
test and completion. Where many installations are planned,
manufacturing and installation rate data are required.

All of these items are fairly obvious. and are the

sorts of material that the technical fraternity can supply
readily. Actually, such a narrative only provides a back-
ground for the essential data needed by Management: how
much will the system cost, in terms of manpower, develop-
ment expense and product manufacture; what will be the
return on investment, in terms of quantity or quality of
services provided or functions fulfilled, or profit; and what
are the risks, in terms of technical achievement. time and
money.

The essence of the discussion on planning is that the

development of this essential information includes, but is
not restricted to. quantitative comparisons of alternatives.
Given a set of objectives, a selected base of technology.
and a single line. or concept, of design, a point in the cost-
return-risk space can be found. The task of the systems
engineer at this stage of the process is to explore enough of
this space so that he can get, and give to Management, a
perspective in terms of the relative value of alternatives of
action.

The task first involves study of a variety of alterna-

tives of system design to meet the stated objectives. each
going far enoug h into detail to develop realistic costs. (Here

76
is the first example of a reason why planning involves indi-
viduals with a sound understanding of technology and of the
design and manufacturing process.) The task is complicated
by the need to explore alternatives of system function. The
objectives of the system include a quantitative description of
the capability, capacity, or quality of the system. Variations
in these, with consequent modifications of system design,
have an influence on the cost and risk. Finally, there is a
need to investigate alternative system concepts that may
meet the same general system objectives, perhaps with
greater excursions from the initial quantitative objectives
than are possible simply with alternate designs. If the new
system is proposed as a replacement or an addition to
existing facilities for increased capability, an important
alternative in this last category is simply more of the old
system or a modification of it. A tested system in manu-
facture is often difficult to beat on economic grounds simply
because of the initial costs of a new system including those
of tooling, stockpiling, maintenance instructions, and
operational training.

The term "cost-effectiveness" has been used to

describe the aspect of system planning in which quantitative
comparisons are made between alternatives in fulfilling a
set of objectives. There is no single measure to be used
for either of the variables; it is possible that each itself is
multidimensional, and, further, represents incommensurate
quantities. "Cost" may include risk and skilled manpower
as well as dollar investment; "effectiveness" may mean
either short or long term profit, grade-of-service or attain-
ment of knowledge. The system planner is responsible for
quantifying the objectives and costs. where possible. and
for deriving the functional relationships between them.
Quantitative cost-effectiveness analysis. however, often
does not provide a complete description, particularly of the
value of a system. The depth of knowledge attainable, or
the quality of a type of service, for example, must be judged
within the context of the social, economic and political struc-
ture within which the system is to operate.

The remaining question is that of timing: when

77
should the planners propose iTIlpleTIlentation. The answer
is siTIlpler to state than to provide in practice: when a
useful service or function can be attained econoTIlically.
SOTIletiTIles technology suggests the functions. It is argued
that Bell discovered telephony rather than discovering a
way to provide audio cOTIlTIlunication at a distance. Edison,
however. actively sought and found the TIlechanisTIl of the
incandescent light. SOTIletiTIles technology lags. In 1945,
A. C. Clarke in a science-fiction story, installed a COTIl-
TIlunication relay on an earth satellite. At that tiTIle. arti-
ficial satellites were still only described in advance planning
docuTIlents; the technology to iTIlpleTIlent theTIl was, at best,
iTIlperfectly understood. However, the idea of a cOTIlTIluni-
cation satellite was attractive, if it could be built econoTIli-
cally. Soon. however, technology had caught up with the
concept, and, in 1954, J. R. Pierce described the COTIlpO-
nents of a cOTIlTIlunication satellite, and cOTIlputed t~e per-
forTIlance required of it. The Echo and TELSTAR experi-
TIlents followed.

INFORMATION FOR IMPLEMENTATION

Suppose, now, that ManageTIlent has decided to pro-

ceed. The planners TIlust provide inforTIlation that will be
the basis for continuation of the project. SOTIle of the
TIlaterial will be that already developed: the description of
the systeTIl functions and op erating plan, the technological
base assuTIled in cOTIlputing costs and in preparing schedules;)
and the cOTIlpatibility of the new systeTIl with its technical
environTIlent.

The additional details given to ManageTIlent illuTIli-

nated the interaction between the systeTIl and the user COTIl-
TIlunity- -the effectivenes s or iTIlpact of the systeTIl on the
outside world. The additional data supplied for the iTIlple-
TIlentation process are a quantitative description of the
interactions between the cOTIlponents, or subsysteTIls of the
systeTIl.

Any systeTIl can be described in terTIlS of cOTIlponent

parts and their interactions, and a podoskeptic systeTIls

78
engineer- -one who designs while conteTIlplating his feet on
his desk- -will stop with such a qualitative description.
The job of the planner that truly contributes to the creative
proces s only starts with identification of subsysteTIls. He
TIlust continue treating each of the subsysteTIls as a cOTIlplete
systeTIl in itself by describing its functions, and by quanti-
fying its objectives, now called "perforTIlance requireTIlents. II

This subdivision of a systeTIl is sOTIletiTIles easy, and

follows a natural path. For exaTIlple, a nUTIlber of TIlodern
systeTIls include a real-tiTIle cOTIlputing function as an ele-
TIlent. Certainly the cOTIlputer can be identified as a sub-
systeTIl. and its functions and output data requireTIlents can
be identified quantitatively. Actually, developTIlent of the
cOTIlputer will involve people who, properly. consider theTIl-
selves to be systeTIls engineers.

There is a TIloral here. and perhaps a clue as to how

to fragTIlent a systeTIl: recognize that each subsysteTIl will
be treated as a systeTIl itself, and TIlust be designed to TIleet
a set of quantitative objectives. Natural fracture planes.
representing faces of TIliniTIluTIl interaction, are TIlost useful
in that the fewest nUTIlerical values for subsysteTIl perfor-
TIlance need to be derived. Shea. in the next paper. will
discuss in SOTIle detail the value, in controlling systeTIl
iTIlpleTIlentation, of establishing siTIlple subsysteTIl inter-
faces.

The process of subdivision in the planning stage can

be carried too far. If subsysteTIls are fragTIlented to sub-
subsysteTIls, then fragTIlented again and again iteratively.
soon the analytical effort w ill appear to result in a cOTIlplete
systeTIl design, down to the level of priTIlitive cOTIlponents.
ITIlplicitly, such planning has left the iTIlpleTIlenters only an
equipTIlent asseTIlbly function. Besides being presuTIlptuous.
this analytic process denies the creative designers scope
for synthesis of new TIlethods of application of technology.
There is a delicate borderline between the functions of
planning and iTIlpleTIlentation which the planners should not
cross. The planner is responsible for definition of sub-
systeTIls and for establishTIlent of quantitative perforTIlance

79
requirements for them, but he must stop short of detailed
design. Perhaps the stopping point is at a subsystem
level at which all details are comprehensible to a single
man--he who has the responsibility for implementing
that particular subsystem. The complexity of today's
systems, and the scope of today's technology are so great
that the observation is not vacuous. The single man that
can con.ceive, plan and implement a modern, complex
system is only the hero of a science-fiction novel.

We see again that although the planning function

is described here as resulting in a paper product, it must
be carried out, in part, by individuals who can recognize
the relevance of technology, and can identify the magni-
tude of various tasks. Further, although the planners
have left the job at a quantitative subsystem definition
level. the implementation proces s itself must involve,
at the subsystem level, individuals who can study and
select the alternatives of design while meeting specified
performance requirements, and, at the overall system
level, individuals with a complete understanding of the
nature of the interaction of subsystems and of relevance
of the subsystem performance requirements to the broad
system objectives.

INFORMATION FOR PLANNING

All of the discus sion so far has been concerned

with near-term planning: investigation of systems which
are, or are about to be, in active development. Even in
these, however, the compatibility of a proposed system
should be described not only with the existing technical
environment, but also with the projection of the environ-
ment. Establishment of a view of the future, in terms
of needs or markets, and of the application potential of
technology currently in the laboratory stage. is a respon-
sibility of system planning. The objectives of such
analyses may be variously stated: to provide a goal.
perhaps several development cycles in the future. toward
which an immediate project is a step; to reduce the
economic impact of further advances in technology by

80
anticipation and pre-design for compatibility; or to identify
a service or function that cannot be fulfilled economically
with current technology. but which can with potential ad-
vances.

Such planning must be carried on continuously;

those who forecast needs and technology are constantly
embarrassed either by their temerity or by their optimism
when reading a document several years old. They should
not be discouraged in reviewing past errors. but should
remind themselves that the pace and directions of social.
scientific and technological achievement must be constantly
monitored in order to understand underlying trends that
point toward future attainments.

One example of technology catching up with a system

concept was mentioned earlier: that of the communication
satellite. Another example. with a longer history. is that
of videotelephony. In 1927 a primitive. but operating.
form of a visual adjunct to audible communication was
demonstrated at Bell Laboratories. At that time. the
methods used in the experiment and the restrictions on
long-distance transmission were so great as to prohibit
implementation. Now (1967) it appears that technology
has caught up with the objective. in the sense that it is
pos sible to develop an economic PICTUREPHONE ®
service which allows video and audio communication. and
which can be provided with all the services currently
expected of the telephone: direct distance dialing. auto-
matic charging. complex intraoffice arrangements and
minimum maintenance of the equipment on the subscriber's
premises.

System planners have kept videotelephony in mind

for forty years. and now have turned their attention to
providing the requirements fo r implementation. They
must keep the future in mind. however. For example.
the picture display parameter values--the number of
lines. the frame rate. and so on--that are selected for
the first commercial set will influence improvements

81
twenty years from now, when applicable technology may be
based on laboratory discoveries not yet made, but when it
is likely that economic pressures in evolution will be such
that the customer's display tube must have, for compati-
bility. the same bas ic parameters as will the first.

A useful tenet is apparent here: in long-range plan-

ning. concentrate on the fundamental principles rather than
on details. In the PICTUREPHONE service example, the
principle is that of providing a subjectively satisfying two-
dimensional display. In twenty years, technology may
advance to the point that the display device is no longer a
cathode ray tube, but the principle. hence basic picture
standards are expected to remain invariant. The author's
temerity may show in twenty years if holography, for
example, can be used as the basis of a three-dimensional
display.

MATHEMATICAL MODELS IN SYSTEM PLANNING

In each of the information-producing areas ~ the

need for numerical results has been stres sed. However,
the perceptive reader may have noticed the lack of the
term "extremum" and of its various analogs. including
"optimum." Rather, such terms as "influence" and "vari-
ation" have been used. This usage reflects the author's
distrust of the absolute values than can result from
anal ys is and computation of mathematical repre sentations
of physical systems, of their internal interaction, and
of their relationship to the real world. This heretical
view should not be misinterpreted as denying the need
of such models; far from it. The issue is simply one of
the author having faith in absolute values. which he doesn't,
or his drawing conclusions on the basis of insight gained,
in part, from analysis, which he does.

A mathematical model is valuable when it permits

identification of a breakpoint, a value of a parameter
beyond which no system design is possible when the assumed
technology, or of a region of balance, in which an inter-

82
action is insensitive to a parameter value change. The
purist may say that such points and regions are associ-
ated with extrema. Such extrema may fas cinate the
mathematician, but to the systems engineer, they repre-
sent regions in which alternatives of action, perhaps not
quantified in the model, are most restricted or most free.

A mathematical model is particularly useful for

examining the sensitivity of conclusions, hence recom-
mendations, to primitive assumptions, and in showing
the relative effect of wide excursions in system param-
eter values. In developing performance requirements,
it is useful to have a collection of models, one for each
of the subsystems, rather than a single grandiose,
detailed, representation of the total system. Such a
collection more readily permits study of the effect of
varying subsystem design variables, and identification
of those performance factors that are most sensitive
to des ign as sumptions. With an understanding of the
essential factors that describe subsystem capability, it
is usually possible to construct a simplified system model
that permits exploration of overall relative system effective-
ness. Too often, however, there is debate as to the validity
of the absolute results of such computation, and too little
attention paid to the primitive comparison of capability
of alternate configuration. In the author's opinion, an
absolute computation of complex system effectiveness is
doomed simply because of an inability to quantify all the
influences. However, insight into system significance
can be gained through study of mathematical models, if
the analyst concentrates on the effect of changing basic
assumptions and system design parameter values.

THE EDUCATION OF A SYSTEMS ENGINEER

The logic leading to the conclusions concerning

education is clear. Practitioners of Systems Enginee ring
cannot agree on a concise, to say nothing of complete,
definition of their field. It follows that Systems Engineer-
ing is not a dis cipline. Since it is not a dis c ipline. it
cannot be codified. hence cannot be taught in an orderly.

83
logical fashion.

What then should a potential systems engineer

learn? Throughout the discus sion here, the words
"objectives," "technology," "quantitative" and "relation-
ship" have been used, and used rather broadly rather than
being restricted to any discipline or craft. This suggests
that before becoming a systems engineer, a candidate
should have familiarity with the sciences related to his
general field of interest, and with the mathematical tools
of application. He must also have a sense of the scope,
pace and utility of technology.

But since the authors of these two papers believe

that Systems Engineering, although undisciplined, in-
volves more than an understanding of fundamentals, there
must be more to the educational process. It appears to
the author that the extra is development of a sense of the
importance of interactions, which can be gained only by
practice. It follows that the academic education of a
systems engineer must be in the context of the case
method, with the neophytes, usually in a group effort,
attacking realistic problems. These are probably identi-
fied in terms of objectives. an environment and a base of
technology. (For realism, the objectives should be very
loosely stated.) In the usual academic environment, the
students must constrain themselves to the planning stage,
with the product of their effort a report. The report may
take the form of a proposal to Management, or it may be
the set of performance requirements that form part of the
information for the implementation function. The students
should, however, have the opportunity to formulate quanti-
tative objectives, to identify the essential factors that
govern system performance, and to study the relationships
that represent the application of technology in meeting a
need or providing a service for a user community.

84
 CHAPTER 2

SYSTEMS ENGINEERING-IMPLEMENTATION

JOSEPH F. SHEA
Office of Manned Space Flight, NASA Headquarters

There is a phrase used by Montaigne which. perhaps,

sums up the inherent problem between systems engineering
planning and implementation: "Saying is one thing, and
doing is anothe r. "

To put the implementation phase of our joint discus-

sion into perspective. I would like to make three general
observations gleaned from participating in large research
and development programs and attempting to reconcile the
types of problems encountered today with the problems
solved by technology as it has evolved in the service of
mankind through the generations of recorded history.

The first is that progress is never made without

difficulty. The history of technology is studded with pro-
jects which never succeeded, or were ultimately successful
only after suffering significant setbacks for one of perhaps
several reasons:

A. The body of knowledge- -call it scientific principle.

design know-how, materials. organization- -available
at the time of the initiation was not sufficient to per-
mit attainment of the partic ular goal intended. In
a phras e, invention or a scientific breakthrough

85
 was required.
B. The body of knowledge was sufficient--in other words.
the fundaITlentals were there--but the understanding
of that available inforITlation was not fully utilized.
In other words. either the systeITl concept or the
detailed design was not as good as it ITlight, in retro-
spect' have been.
C. The iITlpleITlentation of the design- -in other words.
the translation into hardware--did not faithfully
reproduce the design intent.
D. The checks and balances which should be applied in a
reasonable ITlanageITlent structure to guard against
any of the above three were either inadequate or
iITlperfectlyapplied.

A second observation is that there is an additional

challenge added to today's type of prograITl. At least in
technology. the projects--and in general. even the funda-
ITlental research- -have progressed froITl sOITlething which
one individual did cOITlpletely by hiITlself or carried out in
sITlall laboratories with closely knit teaITls. As the industrial
revolution and the deITlands of an increasingly organized and
interdependent society began to require technological advance.
a draITlatic expansion. beginning sOITletiITle in the 19th Cen-
tury. forced the size of laboratories to increase. COITlpanies.
industries. then nations, and perhaps. ultiITlately the world--
sponsor increasingly cOITlplex projects which still have. as
their purpose, the attainITlent of a specific goal.

The increased scope of activity and the cOITlpression

of tiITle in a technologically exploding society have transforITled
such large projects frOITl the efforts of a single individual to
the organized contribution of hundreds. thousands and perhaps
ultiITlately even ITlillions of people who are often geographically.
as well as intellectually. separated. The challenge this repre-
sents for responsible ITlanageITlent is perhaps obvious--but is
the ITlajor reason why what we today terITl "systeITls engineer-
ing" is required--not as a specific discipline. but, rather as
an "intellectual glue" which is required to guard against the

86
type of pitfalls outlined above and to provide an atmosphere
in which specific decisions can be considered in the total
context of the goals of the project.

Kant characterized "mind as the coordination of ex-

perience." A good systems engineer (and there is absolutely
no use in having a bad one) should have a mind which relates
all relevant past experience and present (often conflicting)
considerations to make a decision, or provide to the decision
maker the possible alternatives and the consequences (trade
offs) associated with such choices.

The third point is that simplicity is a virtue. Any

new undertaking should be based to the greatest extent
pos sible on existing technology- -with the stretches of "state
of the art" restricted to those portions of the total system
which demand new technology to meet ultimate economic
or performance goals.

With these thoughts as a background, we shall discuss

the implementation phase.

Partiall y becaus e of the type of programs on which I

have served during the past dozen years. and partially be-
cause I believe they represent a typical cross section of
problems encountered in either government- sponsored or
industry-sponsored programs. the discussion will implicitly
assume a "large program" (hundreds of millions of dollars
per year). It will be carried out by many independent or-
ganizations working toward the same goal but with. perhaps.
somewhat mutually antagonistic interests. This situation
requires an independent arbitrator to settle differences of
opinion (or fact), as well as a technical conscience to assure
compliance with overall intent, or to pass on the acceptability
of failure to comply.

In the implementation phase, no matter who does it»

these functions are the essence of systems engineering.

To initiate such a project t the ponderous mechanics

of "Research and Development," as we today know it in this

87
country, must be set in motion. Inevitably it will involve
almost every element of our society- -from Capital Hill to
the least supplier of raw materials or piece parts. The
program will be so large that government funding is manda-
tory, and the complex process by which such vast enter-
prises get started must work itself out. The program will
probably follow the awkward, organizationally difficult pat-
tern of management by a government technical agency and
implementation by several tiers of industrial contractors.

Barring a minor miracle, it will progress down the

well-worn path of such programs--initial optimism border-
ing on euphoria, the hard time of development difficultie s ,
cost escalation. schedule slippage (or adjustment, to use
the more popular euphemism) and ultimate accomplishment
in which the sins of the development program are washed
away in the flush of success and the golden glow of national
publicity.

The question of whether the seemingly inevitable

mid-program problems can be avoided is of more than
casual interest--although, by and large, they have become
so commonplace in today's aerospace industry that many
people, both in government and industry, feel that some
schedule slippage and cost overrun is, in fact. inevitable.

Perhaps I'm naive, but I refuse to accept such a

conclusion. A program--even a very large one--soundly
planned and intelligently executed. should be "doable II within
commitments- -the costs. schedule and performance pro-
jections which formed the basis for the decision to proceed
with the program and which should be accepted by both govern-
ment and industry participants as moral, as well as con-
tractual obligations.

Unfortunately, the potential solution to the problem

of poor program performance involves no startling new
management technique, no revolutionary set of principles
or actions which will, if only invoked with suitable magic
incantations, instantly solve all problems.

88
 Quite the contrary. the most promising is probably
a return to oldfashioned common sense in both the planning
and implementation of the program; above all, a recognition
that. all fads notwithstanding. the best management system
is competent people motivated to consider what can be
referred to as the "total management II problem. Again 9
another definition of systems engineering.

Dr. Gillette has described a well ordered process

by which such research and development programs can be
planned. justified and initiated in such a way to help achieve
the utopia of "successful project. "

Ideally, the planning process has

A. Convinced the ultimate customer of the need for the
development.
B. Synthesized a system concept achievable "within the
state of the art. II
C. Provided a set of performance (as distinct from
detailed design) specifications for the individual
elements of the system with "clean" interfaces
between such elements so that they can be assigned
to different groups for design and manufacture with
a minimum overlap or interaction required between
such groups.
D. Enunciated a clear statement of program policy to
provide a framework within which a cons istent
decision making process can be developed.
E. Established realistic cost and schedule estimates
for the development.
F. Provided a set of competent personnel who understand
the system, are conversant with the technology to
be employed, and who have some analytic tools
available to assist in the decision making process
as the program progresses.

In the implementation phase, the system engineering

tasks become increasingly interwoven with the primary

89
management task.

Since performance specifications rarely translate

into hardware which meets all the initial requirements--
though it may still be completely usable in attaining the
ultimate goal- -the systems engineering tasks es sential to
the implementation phase may be characterized as:
A. Review of the design approach to verify that it is in
accordance with good practice and has a reasonable
probability of meeting the objective. These reviews
can never assure that the specific des ign will meet
the specifications. Design analysis techniques are.
in general, not capable of predicting all problems
which might arise when hardware is actually fabri-
cated and tested.
The reviews can, however, spot unnecessarily com-
plex approaches or attempts to stretch the state of
the art unnecessarily. By thus implicitly stressing
simplicity in each area» the probability of a success-
ful system development can be significantly enhanced.
The proper division of responsibility at this level is
one of the keys to an effective relationship between
the system people and the development team. In
addition, the fact that the system group does not
specify design details provides a measure of im-
partiality which must be maintained by the system
group to perform its other function; a technical con-
sulting service and problem audit capability for
management.
Technical impartiality is one of the major keys to
effective operation of a systems engineering organi-
zation; it is also one of the hardest states of mind to
achieve. In an era when any individual function can
be implemented in a number of different ways, it is
difficult, if not impossible, to specify uniquely which
approach is best. The proper frame of mind is to
ask whether the approach selected can do the job
within the system constraints, not whether it will do
the job in accordance with an individual's preconceived
notion of the proper implementation.

90
 B. Design a test program which assures that all elements
of the system meet the requirements specified--and
monitor the results of this program to validate com-
pliance or note deviations which must be corrected
or compensated for elsewhere in the system.
C. Maintain the system specification. both performance
and interface, to pass on and control all recommended
changes so that overall system integrity and compati-
bility are assured. This area ultimately involves
everything from overall performance to minute details
of electrical or structural design which can affect
compatibility between individual systems.
D. Plan and observe the ultimate use of the system to
certify adequacy or spotlight areas where improve-
ments in design or operational concept must be
achieved.

These general areas must provide a continuity from

concept to achievement. and hence require some continuity
of personnel or else major shifts in approach or philosophy
can occur. (Sometimes this is desirable if the initial con-
cept leaves much to be desired, but we are discussing here
how a program can be run succes sfully. not how it all too
frequently is done.)

The emphasis. as you may have noted. is on changes

from agreed upon concepts or deviations from desired re-
sults. Since these are the major initiators of the need for
decision. the systems engineering organization is, or should
be, the major focus for such decisions. Hence the require-
ment that the systems engineers be the "most competent"
across the board technical people on the project- -and trusted
by the ultimate decision makers. If this is not the case--and
it often is not- -the function degrades to an out of line con-
sulting service and nagging technical conscience, which is
perhaps a useful function, but one frustrating to and not
worthy of the quality individual who should be assigned to
the function.

It has become the vogue for major national programs

91
to attempt to solve some of the fundamental problems which
plague large numbers of people attempting to work toward
a complex end by inventing management systems which are
intended to compensate for individual competence by forcing
stereotyped decision processes and standard, fool-proof
data gathering methods.

In the technical area, the "in" documents these days

are the Air Force 375 series documents and the NASA-500
series. In the main they call for early definition of program
requirements. development of specifications and timely
review of design.

Although these are essential to program success,

there are at least two problems. One is a presumption that
every requirement can be generated from a logical base.
Although in principle this should be pos sible. in practice
the use of "reasonable" requirements based on judgment
or proven "intuition" is often as good as the results of the
most elaborate studies- -and in some instances, the only
way to reach definition in a reasonable period of time.

The second problem is that, as the specification

tree progresses from overall program requirements to
lower level hardware specifications, mO:t'e and more num-
bers appear. Some of them represent key program require-
ments» but many have been selected somewhat arbitrarily
and are not really significant when considered from a total
system performance point of view. There is an unfortunate
tendency, however, to treat specification numbers as sacred,
particularly in lower tier contracts.

For instance, it has been my experience that some-

thing over half of the "failures" listed in the qualification
programs for subsystems on a major program represent
not an outright failure (no go), but rather a failure to meet
one of these arbitrary numerical requirements. which has
insignificant effect on the intended system use.

The solution to this problem is not to go back and be

sure that every numerical requirement is properly validated

92
before it is set. Such a process would extend the definition
phase unnecessarily. Fortunately, the number of problems
which have to be cons ide red in this area is usually small
enough to be handled efficiently at the time of qualification
if sufficient visibility and understanding exist in the program
structure.

In circumstances where convenient requirements

are not met- -either in qualification test or in acceptance
test on individual flight units--but where there is no failure
to perform functionally. it is usually in the best interest of
the program to accept the as des igned or as built condition
of the hardware. rather than force either redesign or re-
work, with attendant cost and schedule problems. The key
to implementing this general philosophy is an awareness at
appropriate levels in the program of the "real requirements"
inherent in the program goal and an open technical communi-
cation channel from s ubcontracto r to the prime, and from the
prime to the government. This information flow, prompt
awarenes s of problems, and a recognition that waivers or
specification relief can really be in the best interest of the
program are es s ential to the kind of quick reaction capability
needed to help a program through the several months of
critical testing and initial manufacturing which prove so
difficult for most research and development programs.

Of equal importance with the specifications is the

world of quality control, particularly as it interacts with
the manufacturing proces s. For example, the NASA has
issued a set of documents--the so-called 200 series--which
delineates the general requirements of a quality program.
The documents permit a relatively wide spectrum of inter-
pretation. and we have noted a tendency for contractors
to overreact. creating cumbersome inspection systems.
awkward decision-making processes and much unnecessary
paper.

Here again, interpretation must hark back to the

real requirements - -which in this can be stated as:
A. Assure that the hardware is made in conformance
with the drawings.

93
 B. Assure that good practices are used in processes.
C. Assure that deviations in configuration or perfor-
mance are noted and either corrected or deemed
acceptable by responsible personnel.
D. Provide a record of the deviations for future review.

Key to efficient operation in this area is the proper

balance of things like mandatory inspection points, the
working relationship between company and government in-
spection personnel, and, particularly, an effective material
review procedure in which decisions are made, whenever
pos sible, when the problem is first discovered; the paper
cycle becoming a documentation of action taken rather than
an in-line element between problem identification and ulti-
mate correction. To this end, we encourage the use of
engineering dispositions, "on-the-spot" engineering orders,
waivers of unimportant non-compliances, standard repair
techniques, and reas onable interpretation of detailed process
requirements - - in effect, a system engineering approach to
problems of detailed manufacturing and quality control.

I would emphasize that this process does not ask for

lower quality hardware- -rather it seeks a focus on those
elements of quality control which do, in fact, contribute
to the ability of the hardware to perform its intended mission
reliably, rather than the trivia which all too often creep into
a program with attendant focus on cosmetic details, redundant
paper work. and ultimately create a system with much form
but little content.

The above statements are very general, but arise

from examination of many real problems which need not
plague a program if they are handled rationally. I would
emphasize that what is needed can only corne from personnel
with an open mind on the interpretation of requirements or
procedures which are restricting performance without con-
tributing to the improvement of the program. By review
at each level this type of problem can be flushed to the sur-
face so that top management can assure that all effort spent
is truly effic iently directed toward the ultimate goal.

94
 The above philosophy can be summed up in a phrase
which appears. at least in Voltaire. and can probably be
traced back much further in the evolution of human thought:
"The better is the enemy of the good. "

This statement. without sufficient amplification by

example. can perhaps too easily be misinterpreted as not
asking for the very best.

Some examples are so obvious that no questions

arise--perhaps the wrong shape of paint or some minor
scratch.

But there are many subtler things--instrumentation

off by a few percent; engines which don't quite meet the
original performance specifications but which will still do
the mission if the overall system design is rebalanced by
modifying trajectories or weight allocations or carrying
more fuel.

Such utilization of the products of the development

process by rebalancing the system design is an essential
element if a program is to meet its objectives economically
and when required.

This type of rebalancing is the real systems engineer-

ing job in the implementation phase- -and might be character-
ized by the term "dynamic systems engineering. "

In the discussion period after the presentation of the

paper. there appeared to be a general feeling of disappoint-
ment that neither Dr. Gillette nor I had presented systems
engineering as a formal discipline- -a modern methodology
which can be accomplished by a somewhat stylized approach.

But equating. as we do. system engineering with a

mental attitude which considers all pertinent aspects of a
problem and attempts to find a solution by balancing all
relevant factors. it can never be stylized effectively- -any
more than common sense can be legislated. I'm sure there
are many individuals who instinctively "system engineer"

95
each problem they are faced with solving, but who would be
surprised to have so formal a name applied to their habits
of thought. The situation might be compared to that of a
gentleman who, at a somewhat advanced age, took a course
in English literature and was amazed to find he had been
"speaking prose" all his life.

Another of the questions wondered why we insisted

dn "clean interfaces"--"weren't there some interfaces that
just had to be dirty?"

The answer to that question also helps remake the

point that one of the tasks of the systems engineer is to
define the elements of a large program so that the individual
design and development tasks have a minimum overlap.
Such "clean interfaces" permit the total task to be accom-
plished by a number of elements working in parallel with
little requirement for interdependence.

But in so structuring the program, the system

engineer must have considered all of the "dirty interfaces rr
and becomes the custodian of all significant interrelationships
because he does work across all the technical disciplines
involved in a program.

The reason for system engineering, as well as the

mental attitude required, is summed up in a passage from
Toynbee. in which he was discus sing "Inadequacy in our
means of thought":
"The conventional academic dismemberment of a
vast subject into disciplines is a convenient, and
perhaps unavoidable, educational device, but it is
an arbitrary surgical operation, and this makes it a
serious impediment to the gaining of knowledge and
understanding. It is true that anyone mind can make
itself familiar with no more than some patch of the
great forest. Yet, unless it also dares to venture
out into the surrounding stretches that, for it, happen
to be terra incognita, it cannot hope to understand
the nature even of its own narrow beat. "

96
 SECTION III

DYNAMIC PROGRAMMING
 CHAPTER 1

DYNAMIC PROGRAMMING:
A RELUCTANT THEORY*

RICHARD BELLMAN
Departments of Mathematics,
Engineering and Medicine, USC

Let me begin by saying that I really enjoyed coming

to this conference. Ordinarily~ I attend conferences ~ con-
gresses and symposia as a convenient way of getting to
some part of the world that I want to visit. If Stockholm,
Tokyo, Moscow or Dubrovnik seem alluring at the moment,
I look through the announcements of scientific meetings and
try to make a reasonable fit between the topics to be dis-
cussed and what I can plausibly pretend to talk about. I
think, however, we can agree that Pasadena isn't the kind
of place that anyone would want to visit. It's all right to
live here, but we wouldn't want to visit. Consequently, I
am here because I am enthusiastic about the subject of
methodology, which is part and parcel of the field of
scientific philosophy. I am sure that we all feel that this
basic domain is almost completely neglected in the univer-
sities. Scientific philosophy is essentially a hard-nosed
examination of what you are doing, why you are doing it,
and what you really want to get out of this activity.

*This work was supported by the National Institutes of

Health under Grant No. GM-14633-01 and by the National
Science Foundation under Grant No. GP- 6154.

99
 There is a large group of people, of course, who
call themselves "philosophers" without any stated occu-
pation. My feeling is that no one should be allowed to
practice under the title of philosopher unless he is a
professional in some field. To engage in philosophy with-
out actually being a physicist, mathematician, artist.
engineer, businessman or politician is very much like
being the smile of the Cheshire cat. A major need in the
modern world is to have scientific philosophy discussed
by those who have actively engaged in science, and pre-
ferably are still engaged in scientific research.

In this connection there is a very interesting com-

ment by the well-known historian Arthur Schlesinger. As
you know, he received all kinds of awards for historlcal
writing. In particular. he had made a profound analysis
of why Jackson's kitchen cabinet had made certain decisions.
Then, as he says. he became a member of Kennedy's kitchen
cabinet and realized that political decisions were made for
reasons entirely different from those he had contemplated.
In other words, there is an enormous difference between
being a practicing politician and an accomplished political
philosopher.

I feel that very much the same situation holds in

science. You cannot speak meaningfully about scientific
research until you have been in the midst of it, struggling.
groping and hoping. Perhaps the most interesting fact that
emerges about research is that almost all research is
accidental. This doesn't mean that one is not prepared
for these accidents. I think that we must distinguish very
carefully between the so-called research in practically any
field that you can mention that is carried out in a haphazard.
unplanned fashion. and fundamental research that is initiated
as a result of certain accidents that occur. It is necessary
to be well trained. both in a discipline and philosophically.
to convert the fortuitous into the fortunate. As a matter of
fact. this is part of the philosophy of dynamic programming
itself, a philosophy which prepares one to exploit accidents.

100
 In view of the foregoing. I thought that it Inight be
Inore interesting and rewarding to chscribe the develop-
Inent of dynaInic prograInIning as it really occurred~ rather
than to speak apparently sensibly, logically and profoundly
about the concepts of dynaInic prograInIning. The rational,
carefully built-up discussion would be alInost entirely
false. Those of us who teach have a guilty conscience
about this Inost of the tiIne. when we cOInpare the SInooth
flow of a polished lecture given after ten or twenty years
of work in a field with the agonizing effort and the Inany
false steps of our own early work. It is a bit unfair to
students who have a tendency to feel that their abilities
are inferior if they do not grasp the Inaterial the first tiIne
as effortlessly as we present it.

There is not enough honest discussion of how re-

search is done. what Inethods are eInployed. and how
success is achieved. As a contribution I would like to
describe how it was I got in the field of dynaInic prograIn-
Ining. Probably. it is best sUInIned up by the COInInent of
the young lady in the faInous story, "Just lucky. I guess. "

What is rather aInusing is that I found Inyself in

dynaInic pro graInIning, an area I did not want to get into,
as a consequence of getting out of another field that I didn't
want to work in to any extent. When I first started out in
InatheInatics. Iny aInbition was to study nUInber theory. I
had read extensively in this dOInain. and I had SOIne faInous
probleIns picked out for a systeInatic attack. When the
war intervened. By the tiIne I got to Princeton in 1946.
at the age of twenty-five and one-half, I was quite tired of
being a foot-soldier in the cause of science. Although I
could have been an acadeInic scholar (buIn) for another
three years without any strain. I decided that I wanted to
get Iny PhD ritual over with so that I could engage in
serious research. I had done SOIne work in differential
equations in 1943. and had a nodding acquaintance with the
field. With this background and Inotivation. I decided to
write a thesis in stability theory. It was not that I appre-
ciated the field so Inuch at the tiIne, as that I felt that it

101
was rn.uch easier to do a quick respectable thesis in stability
theory than in analytic nurn.ber theory.

I had wanted to stay at Princeton and turn to analytic

nurn.ber theory. However, rn.y professor was a benevolent
tyrant who rn.ade it clear that he wanted rn.e to continue work
in stability theory. It was easier to leave for Stanford than
argue. There was, however, a certain difficulty. My
salary at Princeton ended in June and rn.y salary at Stanford
began in Septern.ber. When I approached rn.y professor with
the request that I be allowed to stay at Princeton during the
surn.rn.er and work on his governrn.ent contract, I received
the expected answer, "No!"

As a last desperate attern.pt to avoid staying with rn.y

wife's rn.other in New York during the surn.rn.er, I wrote to
sorn.eone who had been a fellow graduate student at Princeton,
and had then gone on to Rand, asking hirn. whether there was
anything available during the surn.rn.er. To rn.y delight, it
turned out that there was an opportunity to be a consultant
at the rn.agnificent salary of 35 dollars a day, five days a
week, with 12 dollars a day, seven days a week, thrown in
for living expenses. This was a fabulous arn.ount of rn.oney
by 1948 standards, particularly to sorn.eone who was still
oriented towards wartirn.e prices and salaries.

The result was that I began the surn.rn.er of 1948 with

a very warrn. feeling towards the Rand Corporation, and the
Air Force, and I very rn.uch wanted to do sorn.ething to rn.erit
this bonanza. As rn.entioned above, I would have preferred
to have spent the surn.rn.er playing in the rn.agic fields of
analytic nurn.ber theory, but a feeling of gratitude instilled
rn.e with a passion to be useful.

There were a nurn.ber of rn.ajor intellectual activities

going on at Rand at the tirn.e. As a rn.atter of fact, it was a
very exciting place for sorn.eone accustorn.ed to the faster,
exhilarating pace of wartirn.e research, say at Los Alarn.os,
as corn.pared with the rather slowpaced indolent activities
of either Princeton or Stanford. In particular, there was a

102
great deal of attention being paid to the theory, and possible
application, of an intriguing new theory. the mathematical
theory of games created by Borel and Von Neumann-
Morgenstern. Closely connected with this was the area of
"strategic bombing." This was of major interest at the
time since it was widely felt. (and correctly so in my
opinion), that only SAC, (the Strategic Air Command).
stood between Western Europe and Stalin's troops. My
interest in strategic bombing was natural in view of my
wartime work in the field of electronics with reference to
radar. and my participation in the development of the
atomic bomb at Los Alamos.

My attention was attracted by the theory of games

for a different reason. I had been a minor profess ional
gambler. I don't indulge in card games anymore, but from
the time I was about sixteen on anytime that I needed money,
either for social. or more serious purposes such as paying
medical bills, I used to look around for a congenial group
of people who could afford to lose, and play poker with them.
The easiest way to win money in Los Angeles is to play
poker with doctors. Without any difficulty. one can average
about 500 to 1,000 dollars a week. The underlying reason
is quite simple. It is not that these people are stupid,
quite the contrary. It is just that they have very responsible
jobs during the day, requiring the utmost care and concen-
tration. Consequently, when they play cards at night, they
relax and gamble in rather a reckless fashion. I, on the
other hand, as a mathematician, had no particular respon-
sibilities during the day. When I "gambled." I played to
win. To win in a group of the type I described. it is
sufficient to play the odds» simple. consistent poker. I
could never convince the other players of this. They were
firmly convinced that I used formulas from the mathemati-
cal theory of games to beat them.

With this background I was naturally quite interested

in the idea of using mathematics to analyze gambling systems
and games in general. One feature of the theory that existed
at that date. 1948. was obviously unrealistic. The formu-

103
lation was that of a single-stage process; one bet, one
decision, and the game was over.

Anyone who has ever gambled, or apparently more

respectably, invested in the stock market, knows that
gambling is a multi-stage process. One does not make
one bet, one makes a sequence of bets. Furthermore,
t~e amount of the second bet depends upon the outcome of
the first play; the amount of the third bet depends at least
upon the outcome of the second play. and sometimes on the
first playas well. Another way of putting this is that
gambling is a multi-stage decision process. More pre-
cisely, it is a multi-stage decision process of stochastic
type. It is not pos sible to predict exactly what courses of
action will be followed as the proces s unfolds because what
is done at a particular stage depends critically upon what
has actually happened up to that point. Nowadays, processes
of this nature are constantly being analyzed in analytic terms
and we are accustomed to this sophistication. At the time,
it was very intriguing.

It was my first exposure to the thought that it might

be quite interesting to examine these matters in some detail.
The second way in which this type of process impinged upon
my conscious and subconscious was in connection with stra-
tegic bombing. The strategic bombing that had been done
in World War II was relatively simple because of the dis-
tances involved. Strategic bombing of Russian targets was
an entirely different matter. When account was taken of
the distances involved and the question of flying across
hostile territory in daylight, it began to be realized that
probability of survival for a single aircraft was not high.
This meant that strategic attacks had to be planned very
carefully, and, in particular, that they had to be planned
as multi-stage attacks. Onc e again, one was faced with a
multi-stage decision process of stochastic type in con-
nection with the allocation of bombers to targets.

With thes e two different types of problems before

my eyes, the challenge was to obtain a precise analytic
formulation of the task of determining optimal courses

104
of action. Although. as indicated the ongms of the problem
were accidential. I was quite well prepared to study them
because of my general interest in functional equations.
differential-difference equations and the Abel-Schroder
equation from stability theory. theta functions from analytic
number theory. renewal equations from the theory of
branching processes. Thus. it was very natural to envisage
the use of functional equations to handle multistage decision
processes. It was not difficult to obtain an analytic formu-
lation by means of functional equations of unconventional
type~

(1) f(p) = max[g(p,q) + f(T(p,q))]

q
There was certainly no desire to start a new theory,
or even a major new line of investigation. I was deeply
committed to analytic number theory, specifically to the
modular functions of Siegel, and to the elusive square-free
problem. Nevertheless. here was a kind of equation which
was quite challenging. quite different from any equation
that I had ever seen, and as a matter of fact, quite different
from any functional equation which had ever appeared in
analysis.

What was particularly intriguing was that the concept

of "solution" of this equation was different from the usual
concept of "solution" in classical analysis. There one be-
gins with an equation for an unknown function and ends up
with an analytic expression for the function. or. more
generally, an algorithm for calculating the function. As
far as a decision proces s is concerned. "solution" is quite
different. Certain questions concerning the process can
be translated into equations for unknown functions. How-
ever, if one thinks in terms of the actual process, it is
easily seen that what is required is a rule for making
decisions s a prescription for action, a "policy." A solution
is a set of instructions telling one what to do in terms of
what has happened. We want to know the decision to make
in terms of where we find ourselves. This new approach
to the solution of equations is fascinating to a mathematician.
The possibility of a new approach to one type of equation

105
implies the possibility of a new approach to many types of
equations.

Perhaps a major reason for this fascination is that

a concentrated education has the effect of dulling or de-
stroying common sense. In a certain way, it must. After
all. one of the reasons for learning general methods is to
avoid the necessity for sustained thought in individual situ-
ations. Theories in physics» engineering. chemistry,
biology. and in all intellectual areas, provide general
methods which can be used in specific situations in a turn-
of-the-crank fashion. The student is taught to recognize
that a problem is susceptible to the calculus of variations,
or to the theory of stochastic processes, or to game theory.
or to quantum mechanics and so on. He is not encouraged
to invent new theories f and. certainly. it is not profitable
for him to spend his time rediscovering old theories.

Consequently, when a new type of problem arises.

there is a strong temptation, particularly for the well-
educated person, to fit it into one of the known slots.
Despite the fact that this fit is often quite Procrustean, it
takes a tremendous effort of will to face the fact that this
is a new problem requiring a new type of formulation, and
even a new philosophical orientation. This is one of the
reasons why it has taken dynamic programming about
fifteen years to penetrate the engineering ranks. particu-
larly in control theory where earlier formulations of certain
limited utility are available.

My training in classical analysis oriented me

towards a thoroughly conventional attitude concerning the
solutions of functional equations. Howeve r. fortunately.
the analysis of simple models of gambling and strategic
bombing forced me to think in terms of a different concept
of solution. After playing with these equations for a bit it
occurred to me to see whether there were any other prob-
lems in the real world which led to equations of the fore-
going nature with solutions once again in terms of policies.

What transpired is a nice example of what I feel is

106
the proper point of view of the mathematician. I was no
longer primarily interested in merely being useful, despite
my gratitude. I sensed the possibility of an entirely
different approach to all kinds of interesting processes
with the reward of being able to treat new types of functional
equations. I was not asking the usual question: "Do you
have equations for me to work on and possibly solve?"
I was asking a new que s tion, "Is it pos sible that many of
your fundamental problems involve multi-stage decision
processes that I can formulate in a new. and possibly.
more effective fashion?"

What was amazing, and remains amazing, is that as

soon as you look around with this type of question in mind»
it turns out almost every problem involving decision-making
is a problem involving multi-stage decision-making. The
reason for this is simple, a question of time. The scalar
variable, time, keeps increasing. In all processes. it
becomes a matter of making sequences of decisions over
time. As was pointed out a long time ago. it is never
possible to step into the same river twice.

In some carefully selected situations we can get

away with an approximation of stationarity or periodicity.
In many cases. steady state approximations are quite
useful, but they are always approximations. Constant
attention is required to make sure that the conditions that
permitted the original approximations are still valid.
Realistic decision processes are always multi-stage
decision processes. and indeed. decision processes under
uncertainty.

When all of this finally penetrated the upper levels

of my mind in 1950- 51 I had to make a crucial decision.
Should I gamble a number of years on the pursuit of these
new ideas at a critical stage of my career. or should I
cautiously work the promising claims I had already staked
out in stability theory, differential equations and analytic
number theory? As I pointed out above, I have certain of
the instincts of a gambler. Consequently. it was not too
difficult to push the blue chips into the middle. and make

107
the fateful decision to leave the safe confines of academe
for the more exciting halls of the Rand Corporation in
Santa Monica.

Here, as an aside to the reader. particularly the

younge r reader. let me warn agains t the advice of friends.
It is very easy to ignore the comments of enemies since
we know how to discount them suitably. But it is very much
more difficult to override the carefully thought-out advice
of people who are deeply concerned with your future. Oddly.
people who themselves were rebels and made their repu-
tations by working in unfashionable fields at the time rarely
see the analogy between their past and the present. At the
time, dynamic programming was barely known to a few
people; control theory was not the fashionable field it is
now; differential-difference equations did not occupy the
important position that they do now, and so on. Yet it was
obvious, (to me 9 at least). that these fields had fantastic
intellectual potential, with rich untapped mathematical veins
to prospect. In my own mind. it seemed like a sure thing--
which is the only way to gamble professionally.

Thus it was in 1952 that I found myself at Rand.

busily looking through the worlds of economics and opera-
tions research, identifying multi-stage decision processes
and busily formulating problems of determining optimal
policies in terms of functional equations. This was a
reasonably entertaining and instructive pastime and I soon
became proficient both in scenting a multi-stage decision
process 9 in writing down functional equations. and in pro-
viding existence and uniquenes s theorems. At this point 9
methodology entered again. Sooner or later it was neces-
sary to ask the point of it all. Could I obtain analytic
expressions for the solutions of the functional equations;
could I obtain analytic representations of the optimal
policy? Was I making a contribution to the solution of the
original decision proces ses 9 or was I muddying the waters
even more?

It turned out in retrospect, that I spent {wasted} too

much time trying to find explicit solutions. Fin all y. I was

108
forced to accept the fact that apparently simple decision
proces ses lead to intractable analytic equations. I should,
of course, have expected this On the basis of my experience
in the theory of differential equations and in the field of
mathematical physics in general. Simple physical processes
lead to nonlinear differential equations of simple form which
cannot be readily handled analytically.

In order to make any progress. it is necessary to

think of approximate techniques and~ above all, of numeri-
cal algorithms. Finally, having devoted a great deal of
time and effort, mostly fruitless, to the analysis of many
varieties of simple models. I was prepared to face up to
the challenge of using dynamic programming as an effective
tool for obtaining numerical answers to numerical questions.
A considerable part of the motivation in this direction at
that time was the continuing development of the digital com-
puter. Before it was freely available. it was not very
interesting to conjure up hypothetical algorithms. Once
there g it was challenging to utilize this Sorcerer's
Apprentice.

Once again g what became a major endeavor. of

mine. the computational solution of complex functional
equations. was entered into quite diffidently. I had never
been interested in numerical analysis up to that point.
Like most mathematicians of my generation, I had been
brought up to scorn this utilitarian activity. Numerical
solution was considered the last resort of an incompetent
mathematician. The opposite 9 of course. is true. Once
working in the area, it is very quickly realized that far
more ability and sophistication is required to obtain a
numerical solution than to establish the usual existence
and uniqueness theorems. It is far more difficult to obtain
an effective algorithm than one that stops with a demonstra-
tion of validity. A final goal of any scientific theory must
be the derivation of numbers. Theories stand or fall ~
ultimatelyg upon numbers. Thus I became interested in
computers 9 not as electronic toys. but rather because of
what they signified mathematically and scientifically. This
interest led in many unexpected directions. as I will indicate

109
subsequently. This is a significant part of the story of
scientific methodology. It is usually, if not always, im-
possible to predict where a theoretical investigation will
end once started. But what one can be certain of is that the
investigation of a meaningful scientific area will lead to
meaningful mathematics. Inevitably, as soon as one pur-
sues the basic theme of obtaining numerical answers to
numerical questions, one will be lead to all kinds of inter-
esting and significant problems in pure mathematics.

It is important to emphasize that throughout the

major part of this intensive search of the domain of
economics, engineering and operations research, the
development of new and interesting mathematics remained
my paramount goal. I was not concerned with applications
per se t important as they were. Rather my endeavor was
to accomplish what intellectuals have always strived to do
through the ages 9 namely to escape the gossamer chains
of boredom. Boredom is the principal enemy of the over-
active intellect. Yet even with the aid of an intellectual
obsession, it is very hard to pursue the proper line be-
tween boredom and frustration. If the problems attached
are too easy, we become bored; if the problems are too
difficult, frustration sets in.

Most mathematicians exist in one or the other of

these states at any particular time with an occasional brief
transition period during which they are either creating or
viewing with satisfaction. Too soon, the attack begins on
a new problem with the familiar agony of being confus ed
and thwarted. It's a perversion, without doubt9 but an
interesting one. RemarkablY9 it is of advantage to society
to indulge it.

It was my interest in allocation processes which

first stimulated my continuing concern with control pro-
cesses. When I went to RAND in 19529 I had already decided
to study certain mathematical aspects of control theory.
The theory of differential equations was the appropriate tool
to use in the study of descriptive processes. From the
standpoint of human psychology, it was natural to restrain

110
oneself solely to a descriptive theory of celestial mechanics.
However, in connection with investigations in economics,
engineering and operations research it was equally natural
to ask. "How can I make the proces s operate in a better
fashion?"

"Bottleneck processes" in matherna tical economics

first caught my eye. In these processes it is a matter of
coordinating the behavior of a number of interlinking. inter-
dependent industries» such as steel, coal. automobile and
so forth so as to maximize the output of one of them, say
automobiles. over a given time period. The descriptive
aspects had been intensively studied by Leontieff, Von
Neumann, Wald. and others. but not much had been as far
as control and decision-making were concerned- -or. for
that matter. has been done since.

It is clear that certain crucial decisions must be

made from time to time. Should steel be allocated com-
pletely to the utilization of existing production capacity,
not necessarily of the most efficient type, or should it,
and other resources, be used to build newer. more efficient.
factories? This is a constant problem in economics and
industry. How should capital be employed so as to maximize
some utility over a given time period?

A number of mathematical models of this nature

were analyzed using the calculus of variations. The treat-
ment was not routine since we suffered either from the
presence of constraints, or from an exces s of linearity.
An interesting fact that emerged from this detailed scrutiny
was that the way one utilized resources depended critically
upon the level of thes e resources. and the time remaining
in the process. Naturally, this was surprising only to
someone unversed in economics. such as myself. But,
this was my condition with the result that the observation
of this phenomenon came as quite a shock. Again the
intriguing thought: A solution is not merely a set of
functions of time, or a set of numbers. but a rule telling
the decision-maker what to do: a policy.

111
 The mathematical structure of these perplexing

S
analytic problems was quite open. The problems read:
T
(2) Max g(u,v} dt ,
v o
where
du
{3} {a} dt = h(u,v), u{O} =c
(b) 0 .s v :s; k{u} •
What was remarkable was the level of analytic intricacy of
solution introduced by simple constraints. These con-
straints were an essential part of the background» intro-
duced by immediate economic, engineering and military
cons iderations.

A problem of LaSalle I s which caught my attention

at the same time was the "bang-bang" control problem.
This was a question of restoring a system to equilibrium
as quickly as possible subject to constraints in the re-
storing force. This was a problem closely related to
stability theory.

As a result of a detailed study of dozens of vari-

ational problems of the foregoing type, and filling hundreds
of pages with equations and calculations, it became quite
clear that there would never be any elegant, uniform way
of solving problems of this genre in analytic terms. Each
individual problem was an exercise in ingenuity, much like
plane geometry. Change one small feature, and the structure
of the solution was strongly altered. There was no stability i

Consequently, if one really wanted to obtain numeri-

cal solutions to variational problems in an effective fashion,
we needed s orne additional tools. I was reluctant to become
over-involved, since all along I had no desire to work
seriously in the calculus of variations. A course in the
subject in cOllege had given me simultaneously a rather low
opinion of its intrinsic interest, and a healthy respect for
its intricacies. It appeared to be filled with complicated
existence and uniqueness theorems with self-imposed

112
restrictions none pointing in any particular direction. This
is pertinent to a comment made by Felix Klein, the great
German mathematician, concerning a certain type of mathe-
matician. When this individual discovers that he can jump
across a stream, he returns to the other side, ties a chair
to his leg~ and sees if he can still jump across the stream.
Some may enjoy this sport; others, like myself, may feel
that it is more fun to see if you can jump across bigger
streams, or at least different ones.

Despite my personal feelings, the challenge remained.

How did one obtain the numerical solution of optimization
problems? Were there reliable methods? As pointed out
above, I did not wish to grapple with this thorny question,
and I had certainly not contemplated the application of dy-
namic programming to control processes of deterministic
type. Originally I had developed the theory as a tool for
stochastic decision processes. However, the thought was
finally forced upon me that the desired solution in a control
process was a policy: liDo thus-and-thus if you find your-
self in this portion of state space with this amount of time
lift." Conversely, once it was realized that the concept of
policy was fundamental in control theory, the mathema-
ticization of the basic engineering concept of IIfeedback
control," then the emphasis upon a state variable formula-
tion became natural. We see then a very interesting inter-
action between dynamic programming and control theory.
This reinforces the point that when working in the field of
analysis it is exceedingly helpful to have some underlying
physical processes clearly in mind.

What is worth noting about the foregoing development

is that I should have seen the application of dynamic pro-
gramming to control theory several years before. I should
have, but I didn't. It is all ve ry well to s tart a lecture by
saying, "Clearly, a control process can be regarded as a
multi- stage decision process in which ••• ," but it is a bit
misleading. Scientific developments can always be made
logical and rational with sufficient hindsight. It is amazing,
however, how clouded the crystal ball looks beforehand.
We all wear such intellectual blinders and make such inex-
plicable blunders that it is amazing that any progress is

113
made at all.

All this contributes to the misleading nature of con-

ventional history, whether it be analysis of a scientific
dis covery or of a political movement. We are always
looking at the situation from the wrong side, when events
have already been frozen in time. Since we know what
happened it is not too difficult to present convincing argu-
ments to justify a particular course of events. None of
these analyses must be taken too seriously, no more than
Monday-morning quarterbacking.

I strongly recommend the interesting study of these

and related matters by Jacques Hadamard, the great French
mathematician~ in his book, "The Psychology of Invention
in the Mathematical Field," Dover Publications, New York.

As pointed out above» as of 1954 or so I had stumbled

into some important types of problems and had been pushed,
willy-nilly» into answering some significant kinds of ques-
tions. I could handle deterministic control processes to
some extent and stochastic decision processes in economics
and operations research as well. Where next? At this point
I began to think in a logical fashion, using a systematic
methodological approach. From this conscious effort grew
the theories of invariant imbedding» an application of multi-
stage processes to mathematical physics, and quasilineari-
zation, an application of decision processes to purely descrip-
tive theories. These are discussed by R. Kalaba elsewhere.
Here I shall restrain myself to control processes and dy-
namic programming. The point made above about the suitable
philosophy preparing one for the fortunate accident should be
kept in mind.

There are several ways in which a mathematician can

proceed to extend his research efforts, particularly one who
is deeply interested in problems arising from the physical
world. He can~ on one hand, examine the equations he has
been working with and modify them in a variety of ways. Or
he can ask questions that have not been asked before con-
cerning the nature of the solution of the original equations.

114
This is basically a very difficult way to carry out research.
It is very easy to change the forIn of an equation in a large
nUInber of ways. The great Inajority of the new equations
are not Ineaningful. and. in cons equenc e. lead to both diffi-
cult and uniInportant probleIns. SiInilarly. there are Inany
questions that are difficult to answer, but hardly worth
asking. The well-trained InatheInatician does not Ineasure
the value of a probleIn solely by its intractability. The
challenge is there, but even very sInall boys do not accept
all dares.

The trick that one learns over tiIne, a bas ic part of

InatheInatical Inethodology. is to sidestep the equation and
focus instead on the structure of the underlying physical
process. One learns to subInit oneself a catechisIn: "When
I set up these equations originally, I Inade certain assuInp-
tions. How realistic were these assuInptions? What state
variables, and what effects. did I ignore? ••• "

To obtain, in this fashion, a Inore interesting. and

Inore useful, theory of control processes. we observe that
the use of the calculus of variations in control theory pre-
supposes, albeit tacitly. that we have cause and effect under
control. that we know both the objective and the duration of
the control process. As a Inatter of fact, also iInplicit is
the assuInption that one knows what to observe and that the
state variables can be Ineasured with accuracy.

In the real world, none of these assuInptions are

uniforInly valid. Often people want to know why InatheIna-
tics and cOInputers cannot be used to handle the Ineaningful
probleIns of society, as opposed. let us say to the Moon-
Boondoggle and High Energy-High Cost Physics. The answer
lies in the fact that we don't know how to describe the COIn-
plex systeIns of society involving people. we don't understand
cause and effect, which is to say the consequences of deci-
sions, and we don't even know how to Inake our objectives
reasonably precise. None of the requireInents of classical
science are Inet. Gradually a new Inethodology for dealing
with these "fuzzy" probleIns is being developed. but the path
is not easy.

115
 Upon first gazing upon the complexities of the real
world t there is a certain temptation to return to number
theory. Number theory t however. is not rewarding enough
for continual effort. The problems are too difficult. and
the victories too few. Taking up the challenge of complexity.
I felt that the appropriate thing to do was to start with deter-
ministic control processes and to modify them stage by stage
to obtain theories which could be used to deal with basic un-
certainties in a more sophisticated fashion.

To this end, we can begin by introducing well-behaved

uncertainty of the type extensively treated by the clas sical
theory of probability. This leads to the modern theory of
stochastic control processes where uncertainty is represented
by random variables with known probability distributions.
and where the objective is to maximize expected values.
This gives use to an elegant theory with a good deal of
attractive analysis. It is a new part of pure mathematics.

The very success of the theory emboldens us to push

ahead. For it is clear that stochastic control theory has
very little to do with decision-making in the outside world.
The uncertainties there can not usually be well- represented
by random variables with known distributions. A first ap-
proach is to consider random variables with partially known
distributions which must be determined over time on the
bas is of obs ervation and experimentation. What is quite
entertaining about this new field is that apparently simple
questions lead to extraordinary difficulty of formulation.
Methodologically and philosophically. the real shocker is
to discover that there is no standard or unique way of for-
mulating the problem or rational behavior in realistic situ-
ations. There are alternative mathematical approaches.
with the choice of a particular approach quite subjective.
This is pertinent to the area of game theory discussed by
Oscar Morgenstern.

As an example. let me cite a problem posed well

ahead of its time. 1934. which has given rise to what is
now called lithe two-armed bandit ll problem. Suppose that
you are a doctor in a hospital with a large number of patients

116
suffering from a serious disease. You are given two new
wonder drugs which have not been extensively used before,
which means that you possess only fragmentary knowledge
of their efficacy. How do you administer these drugs to the
patients! What procedures do you employ, keeping in mind
the administration of the inferior drug can result in the death
of a human being?

This is no simple experimentation involving merely

a loss of time, equipment or money. This is a situation
where the nature of the testing procedure is crucial. It is
necessary to learn about the uncertainty and to make deci-
sions at the same time. This is a typical question in the
new theory of adaptive control processes, a fascinating
mixture of control theory and statistical decision theory.

In a relatively painless fashion, the analysis of the

structure of the multi-stage decision process leads to ex-
tensive and significant mathematical generalizations.

But more than this happens. Consideration of new

types of decision proces ses automatically lands one in new
areas. For example, as soon as you speak about techniques
for learning and doing at the same time, you have entered
the area of intelligent machines; not necessarily, of course a
the area of intelligent people. I would say that over the last
ten years the number of intelligent people in the area of
intelligent machines were more than outnumbered by the
number of intelligent machines. But this is always one of
the problems in a new domain.

About ten years ago some people whose names I

won't mention because they are gradually sinking into well-
deserved oblivion, decided they were going to stake out the
field of artificial intelligence. They proudly planted their
flag and they loudly claimed that they had computer programs
which would think, create, compose music, play ches s,
invest in the stock market, translate languages t prove
mathematical theorems, and so on and so on. At the time,
you may remember, there was a great deal of Sunday supple-
ment thinking around; magazines and newspapers were filled

117
with speculation about "giant brains." Stuart Dreyfus and I
who had been trying to use computers to solve simple deter-
ministic control processes, as mentioned above were rather
miffed by these extravagant claims. There were only two
possibilities we could think of: Either these individuals
were making wild statements, or they were very much
smarter than we were. We couldn't even use computers
as effectively as we desired to solve these relatively simple
decision processes arising in control theory. Dimensionality
was, and is, a formidable barrier.

Obviously this alternative of intellectual superiority

was unthinkable, so we had to examine the first one carefully.

This has been one of the very interesting aspects of

using a computer, interesting both methodologically and
philosophically. If one were to look at where the great
advances in thought have occurred, the answer would be
that probably the greatest single advance in thought occurred
as the result of an electro-mechanical device ~ namely the
digital computer. With the aid of a digital computer, for
the first time you can tell people, "Fut up or shut up.
Don't tell me that you have a method for understanding how
people recognize pictures g don't tell me that you know how
people make decis ions, don't just tell me about it; show me.
If you really are assured that you have a method for deter-
mining how people recognize pictures, or as to how people
make decisions, or as to how people play chess, write a
computer program." We now have digital computers that
are fast enough and big enough to carry out precise instruc-
tions in a reasonable space of time. Up until very recently
it was easy to say, "Here is my formula for doing it, I'm
awfully sorry, but clearly this might take 30 years to imple-
ment by hand or computer calculation. You can't expect me
to implement this theory. You'll just have to take my word. "
Today we have computers that can perform these giant
operations in a half hour or a couple of hours.

In general, we now possess simple direct ways to

test hypotheses. What is quite interesting is that all of this
is a very natural extens ion of this work in control theory.

118
One gets quickly into the general area of decision processes
because when you have to use a cOlllputer, you can't wave
your hands. You can't say, "Behave in a reasonable fashion.
You know what I lllean." You have to be precise, "What's
the input? What are the data, what are the state variables.
what processing does one do with the state variables. how
doe s one evaluate, and so forth." C on seq uentl y, I feel as
far as scientific lllethodology is concerned, the biggest single
advance has been due to the developlllent of the digital COlll-
puter. For the first tillle we have a good chance of getting
rid of a great deal of hot air. This doesn't lllean that other
evils won't spring up in their place--but at least they'll be
different ones. We can put up with different ones because
as I said before, we can stand allllost anything rather than
be bored. There will be different difficulties and there will
be lllore sophisticated difficulties. It's going to be a great
deal harder to know what to do with the charlatan of five or
ten years frolll now. He's going to be a quite different
charlatan frolll the Morgan character in "THE WIZARD of
OZ." He is going to be a person who has his PhD, all of
the appropriate acadelllic titles, knows how to use a COlll-
puter, and has just a little kink loose SOllle place. But it's
not going to be so easy to find out where that kink is. It's
going to be a llluch lllore difficult task, and the general public
doesn't even realize the existence of the high level charlatan
yet. I could nallle SOllle nallles of people I'lll thinking of,
but you probably have SOllle of your favorites in lllind.

In the preceding pages I have attelllpted to trace briefly

how the continued pursuit of problellls in the area of lllulti-
stage decision processes branches into surprising dOlllains.
Problellls engender new problellls, theories lead to further
theories. Many of the applications of dynalllic progralllllling
to econolllics and engineering, to the identification of systellls
to psychology and psychiatry have assullled independent
existences as theories within these new fields.

Let llle devote SOllle tillle to describing where I think

SOllle of the lllajor challenges are at the lllOlllent. These
are challenges on lllany different levels. In the first place,
there is always that of the educational level. All theories

119
begin at the top. The question is that of distilling and sim-
plifying so that it can be made available at the elementary
levels, in high school and in college. This requires an
entirely different point of view and an entirely different
ability from the person who may have been able to develop
the theory or use the theory. This is a skill all in itself.
Since we don't know how people learn and we don't know how
we teach them what they do learn, it remains a very diffi-
cult problem.

It's a problem that we face as far as the really im-

portant decision makers of our society are concerned. If
it's true» as I mentioned, that the major problems of our
society are those involving sequences of decisions, then the
question is: how can we teach these more sophisticated
techniques I have been describing to people who presumably
are out of our influence, people in Congress, people in the
top echelons of business, people in the military, and so
forth. It's rather a frightening thought if you think of it that
for the first time in history. the individuals in all the major
positions of power are the people who know the very least
about what's going on in the scientific world, the very least
about what one can really do with computers, about decision
making and mathematics. and so forth. Unfortunately» this
holds as far as the universities themselves are concerned.
One of the biggest problems we have in introducing these
new methods is that the deans» the senior professors, the
chairmen of the departments» are all the people who are
almost completely ignorant of the new ideas and the new
tools. There are exceptions here and there. perhaps one
in a hundred. perhaps one in five hundred. not many. The
educational problem remains a major problem inside the
university.

As far as the mathematical problems are concerned.

we can say essentially that we are at the stage the mathe-
matical physics was at, let's say, 100 years ago, in a
classical sense. or that quantum mechanics is at now.
There are many very excellent formulations of processes
which are analytically intractable. Furthermore. they are
of such complexity that they can't be resolved in any easy

120
way, even by computers we can think of ten years from
now. which will be a billion times more powerful. What
we now need is the same type of sustained effort in dynam-
ic programming that is required to do something meaning-
ful let's say, with the many-body problem in quantum
mechanics, with problems in celestial mechanics. and so
forth. We need very sophisticated approximation tech-
niques based upon intimate knowledge of the actual physical
processes. This means that progress will not be dramatic.

Then. off in another direction, one of the interesting

things about the dynamic programming formulation is that
you can extend it to areas which cannot be treated at all by
means of classical mathematical approach. The classical
mathematical approach to optimization requires a criterion
function. If you want to set up a problem within the frame-
work of the calculus of variations, you must have a func-
tional which you can use to minimize or maximize. But if
you think in terms of a policy, you can still think in terms
of what do you do in terms of where you are in state space.
You may not have any quantitative way of evaluating this
policy. but you may have a way of comparing the effects of
different policies. This is the case if simulation methods
are employed. Probably the major problem in the real
world is how to handle these decision processes which
escape classical formulation. How do we handle decision
processes where we don't know how to describe the state of
the system completely, when we have only a partial idea of
cause and effect, and when we have only an imperfect idea
of what we want to do. The average scientist may respond,
"Well why in the world do you want to deal with problems
of that type?" The answer is immediate: the problems of
society are all of this type. Recall what our public school
teacher and the high school teacher used to say when we
got up and fumbled around and finally plaintively murmured:
"Well Ilmow it and I understand it, but I don't know how to
say it." She would say. "If you understand it, you can state
it." Of course, just the opposite is true. There is nothing
of any i.mportance in any area of life which we can state
preci.sely. Nonetheless, we operate very. very efficiently.

121
This leads to the question: What is the explanation of this?
There must be some reason, there must be some way of
processing the information we use for decision-making.
What techniques do we use? Is it an entirely different kind
of mathematics, is it something based upon gestalt rather
than just single bits of numbers? What are the state vari-
ables? Nobody knows. This is one of the outstanding
challenges in the explanation of human behavior. Again the
interesting feature of this study is it widens the horizons
of mathematics. It gets into the areas of biology and medi-
cine, it gets into the areas of psychology and psychiatry,
and it introduces entirely different kinds of problems.

I would like to as sure you that all of the major prob-

lems remain. The theory of dynamic programming is a
fertile field for research by the mathematician who is solely
interested in functional equations and their analysis. by one
who is concerned with applications and the use of computers
and by one who is intrigued with the conceptual bas is of
decision-making. It is an excellent area for the young
mathematician.

122
 CHAPTER 2

DYNAMIC PROGRAMMING'S PROGENY

ROBERT KALABA
The RAND Corporation

I. INTRODUCTION

Dynamic programming was devised to answer cer-

tain optimization problems arising in operations research
and control theory(1). It provides a new analytical and
computational approach to such problems. It may be ex-
pected then, that it will provide new insights into those
branches of physics in which a variational principle is in-
volved. This is indeed the case as is shown in Section II,
employing Fermat's principle of optics. A basic feature
of dynamic programming is the imbedding of a given pro-
cess within a class of processes. This technique, highly
developed in dynamic programming, can be employed in a
variety of ways in applied mathematics. In Section III an
application to the theory of integral equations is given. A
new approach to the computational solution of Fredholm
integral equations is given.

This paper is self-contained, no prior knowledge of

dynamic programming being assumed.

123
 II. DYNAMIC PROGRAMMING AND OPTICS

The ITlode of eITlploYITlent of BellITlan' s P rinc iple of

OptiITlality(1) ITlay be illustrated by considering a classical
probleITl in optics, that of light rays in inhoITlogeneous ITledia.
In a two-diITlensional ITlediuITl let the index of refraction of
the point (x.y) be denoted by n(x,y). According to FerITlat's
principle, in passing froITl one point to another the trajectory
is such as to ITliniITlize J
n ds, where s refers to arc
length along the curve. The typical approach is to regard
the curve as a locus of points and seek to solve the Euler
equation

(1) :y [ n ( x , y 0 ? J - :x {:y' [ n ( x , y ) p ] } = 0,

subject to the conditions that the solution y = Vex) shall pass

through the required end points. FroITl the nUITlerical view-
point the two-point boundary conditions present certain diffi-
culties (2) •

The dynaITlic prograITlITling viewpoint characterizes

the optiITliz ing arc s as envelopes of tangents. Keep the ter-
ITlinal point Q with coordinates (xo ' yo) fixed. Regard
the ITliniITluITl value of J
n ds froITl the general point (x. y)
to the fixed point Q as a function of x and y.
(2) w(x,y) = ITlin Jn ds •
At the point (x,y) the ray ITlust decide on the slope of the
curve so as to arrive at the terITlinal point Q in least tiITle.
Suppose that the tangent to the curve at this point ITlakes an
angle 8 with respect to the x-axis, and let the tangent be
followed by a sITlal1 distance ~. This ITlakes a contribution
of n{x.y)~ to the integral, where terITlS in ~ of power
higher than the first are oITlitted o Then by definition the
ITliniITluITl value of the integral J
n ds froITl the point
(x + ,6 cos 8. y + ~ sin 8) to the fixed point Q is
w{x + ~ cos 8.y + ~ sin 8). This leads to the basic equation
(3) w(x, y) =ITlin { n(x, y)~ + w(x + ~ cos 8, y +~ sin 9)
9 + O{~)} •
where O(~) refers to terITlS m ~ of power at least two.

124
Next we expand in a Taylor series to obtain
(4) w(x,y) = :min {n(x,y).6 + w(x,y) + w (x~y).6 cos a
a x
+ w y {x,y).6 sina + O{.6) }
By subtracting w{x,y} fro:m both sides of the equation
above, dividing by .6, and letting .6 tend to zero. we
obtain the basic relation

(5) 0 = :m~n {n{x,y) + wxcos a + wysin e } 0

This can be rewritten

(6) -n(x,y) = :min {w cos e + wysin a}
e
0

x
This equation is actually a novel for:m of the eikonal
equation{3 ,4) 0To see this note that the :mini:mizing value
of e is obtained by setting the derivative of the su:m in
brackets with respect to a equal to zero. This yields the
opti:mal control law, in auto:matic control theory parlance,
(7) o= - w s in a + w cos a
x y
or
(8) tan e = w y (x.y)/w x (x,y) 0

Equation (8) provides the correct slope of a trajectory at

any point (x,y) in ter:ms of the wave-front function w. It
enables us to eli:minate e fro:m equation (6) by noting that

(9) sin e=w I Iv} + w Y2

YV' x

(10) cos e = w I Iv} + w 2

xy""x y
Equation (6) then assu:mes the for:m of the basic equation of
geo:metric optics
2 2 2
(11) w +w = n •
x y
The slope of a ray at the point (x,y) is given by equation (8).
The slope of a wave-front is given by

( 12) o = dw = w x dx + w dy
y

125
or
(13 ) dy/dx= - w /w •
x y
The fact that the product of these slopes is -1 shows that
the wave-fronts and trajectories are orthogonal.

Finally, let us obtain the equation of the rays from

the optimal control law (8) and the eikonal equation (11).
The slope of a ray is
(14) y' = wy /wx .
Differentiation with respect to x yields the relation

y" = Wx [w yx + Wyyy '] - w

y
[w
xx
+ w xyy']
(15)
2
w
x

To eliminate the function w from the right-hand side of

this equation we first differentiate equation (11) with respect
to x and y ,
(16) nn =ww +ww
x xxx yyx
(17) nn =ww +ww
y x xy y yy
It follows that
2
( 18) nn /w = (w /w) + (w /w )y'
x x xx x yx x
2
( 19) nn /w = (w /w) + (w /w )y'
y x xy x yy x
This implies that
2 2
(20) y" = (nn /w ) - y'(nn /w )
y x x x
or
2 2
(21 ) ny" = (n /w}(n - yIn ) •
x y x
But from the eikonal equation (11) it is observed that

126
(22)

This yields the desired ordinary differential equation for

the rays

which is equivalent to equation (1) above.

In this way we have shown that Bell:man's principle

applied to Fermat's principle quickly leads to both the
eikonal equation for the wave-fronts and the Euler equation
for the rays 0

Applications to the variational principles of mechanics

are given in (5). In particular a method for by-passing
eigenvalue problems in buckling problems is presented.

III. INVARIANT IMBEDDING AND INTEGRAL EQUATIONS

In various branches of mathematical physics, chem-

istry, operations research and engineering, we must inte-
grate a Fredholm integral equation

(1) u(t) = get) + fCo k(t,y)u(y) dy, 0::::: t::::: c,

where the functions get) and k(t,y) are given, and the
function u is to be determined. A direct approach by
dynamic programming is given in reference (6). Here we
shall sketch an approach in which no optimization is in-
vol ved. but we do imbed the original problem, involving
the interval (0 ,c), within the class of problems for all
intervals (0 ,x). The equation (1) is rewritten

(2) u(t ,x) = get) + JX k(t, y) u(y,x) dy 0

o
Furthermore, for ease of exposition. we shall limit our-
selves to the case

(3) J(t,x,z) = e -(x-t)z + IX

o
k( It-y I) J(x,y,z) dy,

127
which is of prime importance in the theories of radiative
transfer(7) and multiple scattering. It is assumed that the
kernel k can be written in exponential form and is even.
rb -rz
(4) k(r) = J a e w(z) dz, r > 0,

(5) k(r) = k(-r) •

The aim is to secure an initial-value formulation for the
integral equation (3).

First equation (3) is differentiated with respect to

x term by term. The result is
-(x-t )z
(6) -ze + k(t-x)J(x,x,z)
+ jXk(t_y)J (y,x,z) dy •
o x
This is regarded as an integral equation for the unknown
function Jx(t,x.z). Using the superposition principle for
linear equations, the solution may be written
(7) J (t,x,z) = -zJ(t ,x,z) + ([>(t,x)J(x»x,z),
x
where the function <f!(t,x) is the solution of the integral
equation
(8) <f!(t,x) = k(t-x) + JX k(t-y)([>(y,x) dy.
o
Equation (7) is the basic differential equation for J at the
fixed point t as a function of x. In this equation z is
merely a parameter. We now deal with the two final factors,
([>(t,x) and J(x,x,z).

In view of the fact that the function q, satisfies

equation (8) and k is represented in exponential form in
equation (4), we see that

=f
b
(9) ([>(t,x) J(t,x,z)w(z) dz •
a
In deriving equation (9) use has also been made of equation (3)
for the function J and the evenness of the kernel k. Thus
([>(t,x) is conveniently expressed in terms of J(t,x,z).

128
Finally, we explain how to determine J(x,x,z). According
to equation (3) for t = x, we have

(10) J(x,x.z) = 1 + Jox k(x-y)J(y,x,z) dy •

Upon using the representation for the kernal k in equa-
tion (4) this becomes

(11) J(x,x,z) = 1 + J0 X[Jb

a e
-(x-y)v ]
w(v) dv J(y,x,z) dy.

We int roduc e the "reflection function" R by means of the

formula

(12) R(v,z,x) = fX e-(x-y)vJ(y,x,z) dy,

o
and equation (11) takes the form

=1 + J
b
(13) J(x,x,z) R(v.z,x)w(v) dv •
a
We now obtain a differential equation for the function R.
Differentiate equation (12) with respect to x to obtain

(14) R (v,z,x) = J(x,x,z) + JX [e-(x-y)v J (y,x,z)

x 0 x
-(x-y)v]
- vJ(y,x,z)e dy.
This becomes

(15) R (v ,z ,x) = J(x,x,z)

x
+ Jx e -(x-y)vrrzJ(y,x,z)
o
+ <I>(y.x)J(x,x,z)] dy - vR(v,z,x)

= J(x,x,z) - zR(v,z,x) - vR(v,z,x)

+ J(x,x,z) f ox e -(x-y)v <I>(y,x) dy •

To evaluate the integral in the last equation, we use equa-
tions (9) and (12) to obtain

(16) JXe-(x-y)v<I>(y,x) dy = J x e - (x-y)vJ b J(y ,x,z)w(z) dz

o o a
=f
b
R(v,z ,x)w{z} dz
a
129
This results in the differential equation for R:

(17) Rx(v,z,x) = -(z+v)R(v,z~x) + [1 + {bR(v,Z,X)W(v) d~

X [1 + Jab R(v,z,x)w(z) dZ]

The initial condition is

(lS) R(v,z,O) =0 ,
which follows from equation (12).

The scheme for determining J at the fixed point t

for all z, a:$ z::S b, and x, O::S x:$ t, is this. Using
the initial condition for R in equation (lS) and the Riccati
differential equation for R in equation (17) (replacing the
integrals by sums using a quadrature formula) the function
R(v,z,x) is determined for
(19) a::Sv, z::Sb
and
(20) 0 !S x :s t •

In particular R(v,z,t) is known. Using formula (13),

(21) J(t.t,z) = 1 + fab R{v,z,t)w{v) dv,

J{t,t,z). a::S; z !S b is evaluated. This provides the initial

conditions for J(t ,x,z) at x = t. The differential equa-
tion (7) is then appended to the differential equation (13). In
this way, both J(t,x,z) and R(v,z,x) are computed for
t:$ x:S xl' In conjunction with equation (7) use is made of
relations (9) and (13).

Successful calculations using this scheme are re-

ported in references (S,9). An extended discussion of the
method is given in reference (10).

130
 REFERENCES

1. R. Bellman, Adaptive Control Processes, Princeton

University Press, Princeton, 1961.

2. R. Bellman and R. Kalaba, Dynamic Programming

and Modern Control Theory. Academic Press,
New York, 1965.

3. R. Kalaba, "Dynamic Programming, Fermat's

Principle, and the Eikonal Equation," J. Opt. Soc.
America, v. 51 (1961), pp. 1150-1151.

4. R. Luneburg, Mathematical Theory of Optics, Univ.

of California Pres s, Los Angeles, 1964.

5. R. Kalaba, "Dynamic Programming and the Vari-

ational Principles of Classical and Statistical
Mechanics," pp. 1-9 of Vol. I of Developments in
Mechanics» Plenum Press. New York. 1961.

6. R. Bellman, "Functional Equations in the Theory of

Dynamic Programming- - VII: A Partial Differential
Equation for the Fredholm Resolvent," Proc. Amer.
Math Soc. v. 8» 1957, pp. 435-440.

7. V. V. Sobolev, A Treatis on Radiative Transfer.

van Nostrand, New York, 1963.

8. R. Bellman, R. Kalaba and M. Prestrud. Invariant

Imbedding and Radiative Transfer in Slabs of Finite
Thicknes s, American Elsevier, New York, 1963.

9. R. Bellman, H. Kagiwada and R. Kalaba, Numerical

Results for the Auxiliary Equation of Radiative
Transfer. J. Quant. Spectrosc. and Rad. Transfer,
v. 6 (1 966), pp. 2 91 - 3 1 0 •

10. H. Kagiwada, R. Kalaba, and A. Schumitzky, "An

Initial- Value Method for Fredholm Integral Equations,"
J. of Math. Anal. and Applications to appear in July
1967 issue.
131
 SECTION IV

INFORMATION THEORY
 CHAPTER 1

A SURVEY OF INFORMATION THEORY

J. R. PIERCE
Bell Telephone Laboratories, Murray Hill, New Jersey

ABSTRACT

Most of the profitable work on information theory

has been on codes for correcting errors in digital trans-
mis sion, and various practical and effective codes are now
available. Considerable attention has been given to encoding
for efficient transmis sion over analog channels. A bound on
the information rate imposed by quantum effects has been
found. Scheme s have been devis ed for reducing the redun-
dancy of signals from various sources. such as text, fac-
simile, and TV, but these have not found practical use.

Exploitation of the fidelity criterion of the message

destination offers the greatest possibility for signaling
over a channel of much reduced capacity. Color TV effec-
tively exploits the poor spatial resolution of the eyes for
color. The vocoder. which antedates information theory,
can transmit a very economical description of speech from
which synthetic speech fairly satisfactory to the ear can be
constructed.

Attempts to apply information to cryptology.

gambling 9 linguistics. psychology and art have been de-
creasingly productive in about that order. Other attempted
applications are pretty empty.

135
 "The Mathematical Theory of Communication,"
which Claude E. Shannon published in the Bell System
Technical Journal of July and October, 1948, was a re-
markably prescient paper. It gave us information theory
in the form in which we know it today. It proved the funda-
mental theorems, and it pointed out the areas in which
future problems lie.

All of this is most easily discussed with the aid of

the diagram which Shannon used as his Fig. 1. Here we
have at the end of the communication process an informa-
tion source which might. for example, be a man speaking,
or a man in front of a television camera. At the other end
we have an information destination, which might be a hearer
or a viewer. Intermediate between the source and the desti-
nation we have a signal which travels through a communica-
tion channel. The signal will be imperfectly received.
Thus. Shannon shows a noise source, which makes the
received signal differ from the transmitted signal in a ran-
dom, that is, in an unpredictable way.

A communication channel is merely the medium

used to transmit the signal from the transmitter to the
receiver. It may be a pair of wires, a coaxial cable, a
band of radio frequencies, a beam of light.

The signal is produced when the information source

acts on a transmitter, which we may think of as an encoder.
The transmitter may produce an encoding or signal which
is an exact electrical analog, of say, the acoustic pressure
of a sound wave, or it may produce a string of pulses which
represents the message in an encoded form. The transmitter
may even produce what we would consider a very indirect
description of the output of the message source. This des-
cription or encoding may be continuous (analog) or digital.

The signal, altered unpredictably by noise, goes to

the receiver, or decoder, and this supplies a message to
the message destination. This need not necessarily be the
exact information generated by the source; it can be anything
that meets the fidelity criterion of the hearer or the viewer.

136
For exaITIple, the sender ITIay hand in a longhand ITIessage
at a telegraph office; the recipient will receive a ITIessage
ITIechanically printed in block letters.

In his general theory of inforITIation. Shannon

assuITIed the output of the ITIessage source to be ergodic;
this is a ITIatheITIatical restriction. There is every reason
to believe. however. that an ergodic source approxiITIates
actual ITIessage sources, such as text. speech or TV sig-
nals. Shannon dealt both with the case of discrete ergodic
ITIessage sources (text is an approxiITIate exaITIple) and
continuous ergodic ITIessage sources (a voice signal is an
approxiITIate exaITIple). It is interesting to note that
Shannon's work applies to linear and nonlinear processes.
The idea of linearity or nonlinearity siITIply does not appear
in his theory of cOITIITIunication.

This picture of cOITIITIunication is a very general

one--it is surprising that Shannon could say sOITIething that
is both true and contentful concerning it.

Shannon was able to give a definition of inforITIation

in terITIS of choice at the inforITIation source and resolution
of uncertainty at the inforITIation destination. In the bit or
binary digit he gave a quantitative ITIeasure of this choice
or uncertainty. A bit of inforITIation is the aITIount of choice
between two equally probable possibilities, such as heads
or tails, or the resolution between two equally uncertain
pos sibilities. such as the answers y.!:.!?. or ~.

A fair coin ITIay fall heads or tails. SOITIebody can

tell us which way it fell with one bit of inforITIation. by
saying heads or tails. or y.!:.!?. or ~. or ~ or one. Yes
or ~. ~ or ~ can serve as an encoding of heads or
tail s.

We should note that in flipping a coin with heads on

both sides. there is no uncertainty as to the outcOITIe. The
"ITIessage" produced, that is, heads, need never be sent,
for it can be guessed with perfect accuracy at the destina-
tion. Such a ITIessage conveys no inforITIation. In Shannon's

137
ter:ms. it is co:mpletely redundant.

If the :message source has an equiprobable choice

of s ending anyone of eight distinct :mes sages, three bits
of infor:mation, that is. three binary choices. can specify
which of the eight is chosen. We can regard these three
choices as branching left or right in a tree of decisions
which will designate which a:mong the eight :messages we
:mean. This is illustrated in Fig. 2. The sa:me principles
can be applied in the case of redundant :messages that are
not equally probable. and in the case of :many, :many:mes-
sages.

Shannon defined the infor :mation rate or entropy rate

of an infor:mation source as the nu:mber of bits per second
necessary to describe or encode the :messages generated
by the source. This definition applies to both discrete and
continuous sources.

Let us now turn our attention to the co:m:munication

channel. Shannon showed that the co:m:munication capacity
of a co:m:munication channel could also be :measured in bits
per second. We can, of course, send no :message over the
channel si:mply by disabling the trans:mitter or encoder. It
is plausible to believe that the outputs of so:me trans:mitters
or encoders are better suited to a particular channel than
the outputs of other trans:mitters. Shannon observes that
there is so:me best for:m of encoding which will send the
largest possible nu:mber of bits per second over a particular
channel. The largest nu:mber of bits per second that can be
sent over the channel is defined as the capacity of the chan-
nel. This is so for both discrete channels (a data channel
is an exa:mple), and continuous or analog channels.

We :might think that a noisy channel would not have

a distinct capacity. Shannon showed that a string of sy:m-
boIs can be trans:mitted through a noisy channel with an
error less than any preassigned level. This can be done
by :means of elaborate encoding. An appropriate code will
enable us to correct the signal e:merging fro:m the noisy
channel and to reconstruct the :message with errors below

138
the presc ribed level.

Such correction of errors is possible, but an accu-

rate reconstruction of the message at the destination is
pos sible only if the entropy rate of the message source is
less than or equal to the capacity of the channel.

Most of Shannon's readers found it astonishing that

we can reconstruct from a noisy signal an accurate mes-
sage, but Shannon was able to prove this. He also showed
that this is pos sible only if we delay the receipt of the
message. That is, we must examine a large amount of a
noisy signal before we can reconstruct the transmitted
message correctly.

In the description of information theory which I

have given, I have mentioned much in a little time. I did
this in order to give a background for discussions of parti-
cular points. With this background in mind, let us ask,
what elements of Shannon's mathematical description of
the communication process do workers in the field of infor-
mation theory strive to elucidate?

Those concerned with information theory may con-

centrate their work on what is called the discrete noisy
communication channel. This is an idealized channel which
transmits characters such as zero or one, or the ten digits,
or the 26 letters of the alphabet and the space, but transmits
them imperfectly, so that sometimes one character is trans-
mitted and another is received. The discrete noisy channel
has a reasonable physical counterpart in the channels pro-
vided for data communication.

Information theorists have also considered the con-

tinuous channel, in which we have a band limited signal
containing only certain frequencies. The amplitude of such
a signal may vary smoothly with time in any way that is
consistent with the bandwidth of the signal and the average
power or the peak power of the transmitter. The continuous
channel has an analog in the channel provided by a pair of
telephone wires or in a radio channel between broadcast

139
transmitter and receiver or between a directive microwave
transmitter and a directive microwave receiver on a hilltop
tens of miles distant.

Some workers have asked, what is the information

capacity of a communication channel in which quantum
effects are important? A channel using the coherent light
of a laser is such a channel.

A worker can concentrate on the nature of the mes-

sage source. That is, just how redundant and predictable
is the message produced by some message source- -a man
writing English, perhaps, or a man talking.

Finally, a worker in communication theory may

concentrate on the nature of the mes sage destination and
what its fidelity criterion is. It is futile to supply to a
man's ear either sounds that he cannot hear, or distinctions
among sounds that he cannot sense. It is futile to supply
the sight with a picture of greater detail than the eye can
descry. Indeed, the standards of quality in transmission
of both sound and sight are set by the imperfections and
fallibility of the human senses.

Let us now consider some work which has been done

in each of these areas. We will start with the discrete
channel.

As we have noted, a discrete noisy channel can be

cons ide red as merely a distortion of the symbols of text,
whether the text consists of zeros and ones, or of decimal
numbers, or of the letters and spaces and punctuation
marks of natural language. The errors in such a system
of communication are much like the errors that a typist
makes when she substitutes one symbol for another. How
are we to infallibly detect and correct such errors? We
can do so only if we send over the noisy channel more sym-
bols than the number of symbols in the message we wish to
convey.

How we should use such extra symbols depends on

140
the nature of the errors in the discrete channel. Shannon's
work tells us that errors can be corrected without trans-
mission of information from the message destination back
to the message source. However. if the communication
channel is merely interrupted infrequently for short periods
of time, a very effective means of avoiding error is to send
a message back to the transmitter. saying that a certain
stretch of the received characters had been lost or badly
garbled, and asking for a retransmission of this portion of
the message. Infrequent but powerful bursts of noise do
indeed garble transmission in just this way in some trans-
mission systems, such as local telephone lines. In such
cases, retransmis sion is often used in correcting errors.

Sometimes there is a small probability that any

character may be received in error. Figure 3 shows one
means for correcting such infrequent errors. In this
figure a block of 16 successive characters of the message
is written in successive rows of four characters each.
Then, zeros or ones are written in the circles to the left
of the rows and above the columns. A zero or a one is
chosen so as to make the total number of ones in a row or
a column. including the circle. even. The "check digits"
in the circles are transmitted before or after the 16 mes sage
digits.

If there is an error in one message digit there will

be an odd number of ones in both a row and a column. and
this serves to locate the digit in error. If there is an error
in a check digit there will be an odd number of ones in a
row or a column. This tells us that it was a check digit
that was in error, and not a message digit.

The error correcting codes that are actually used

are more complicated than that which I have used as an
example.

What do we want in an error correcting code? Of

course we want it to correct errors. but we want it to cor-
rect errors efficiently and with a resonable amount of
apparatus. Some sorts of error correcting codes would

141
require a huge dictionary, in which we would look up each
received block of digits in order to find the shorter trans-
mitted block. This is impractical. Other codes use parity
checks of the sort I have described in connection with Fig. 3.
Some of the best of these codes are called Bose-Chandhuri-
Hocquenghem codes.

The known families of error correcting codes have

a common fallibility. If we try to use them to approach
the channel capacity of the noisy channel, we fail. When
the block length is made very long, in the limit, either the
signaling rate goes to zero or the error probability becomes
large.

That is all I am going to say about error correcting

codes. I will turn next to making the most of the continu-
ous channel.

I first heard of communication theory while Claude

Shannon was working on it. I was then surprised to find
that it was possible to encode messages so as to transmit
them over channels of smaller bandwidths than the band-
width of the message source. It has in most cases proved
more practical to go in the oppos ite direction. Even at that
time. broadband frequency modulation was already being
used to achieve low noise in the high fidelity transmis sion
of music by ultrahigh frequency radio.

Sometimes a great deal of bandwidth is indeed

available, as it is (subject to regulation by the FCC) in the
radio spectrum, and as it would certainly be in signaling
by means of coherent light waves generated by lasers. If
much bandwidth is available, then it is advantageous to use
a broadband transmission system. One may swing the fre-
quency of a carrier wave wildly back and forth. Such high
deviation frequency modulation is a good transmis s ion sys-
tern. although it does not realize quantitatively the advantages
pointed out by Shannon.

Consider a continuous signal of bandwidth B. Let

us sample this signal. that is. measure its amplitude» 2B

142
times a second. It can be shown that this many samples
completely and perfectly describe the signal. Knowing the
sample amplitudes, we can reconstruct the signal exactly.

We can represent the amplitudes of such samples

approximately by means of decimal digits, as 27, 48, etc.,
or by binary digits, as 01011010, 11010110, etc. Such
approximate representation is called quantization. The
representation by binary digits is particularly appealing,
because we can send the digits as off-on or plus-minus
pulses over a digital transmission system.

Such digital representation and transmis sion of

analog signals is called pulse code modulation, or pcmo
Pulse code modulation is an advantageous form of broad-
band communication. In transmitting voice by binary
pulses, pcm takes ten times or more the bandwidth of a
voice signal.

Various complicated frequency spreading techniques

have been devised, principally at the Lincoln Laboratory of
MIT. These encode the voice or other signal into what
seems to an observer to be a broadband spectrum of noise.

Sometimes the bandwidth of the available transmis-

sion channel is limited. For example, the attenuation or
los s in pairs of wires or in coaxial cables goes up rapidly
with increasing frequency. In transmission over such
circuits we may wish to conserve bandwidth. If we wish
to transmit many voice channels over one coaxial cable,
we may merely shift the frequencies of different voice
signals by different amounts, and so stack the voice chan-
nels one above another in frequency, allotting to each only
the bandwidth of one voice signal. In transmitting digital
signals over such a cable, it will prove advantageous to
send, not binary, off-on pulses, but to send pulses having
perhaps 16 different amplitudes. One pulse which may
assume any of 16 amplitudes conveys as much information
as four off-on or binary pulses. The pulses need be sent
at only a quarter of the rate that we would have to send
binary pulses. and the bandwidth needed is only a quarter

143
as great.

The word encoding may conjure up the digits or bits

that are used to measure information rate and channel capa-
city. But encoding need not be a digital operation. Fre-
quency modulation is an analog, not a digital process. One
has a wide choice in using analog or digital encoding.

Let us consider transmission over a noisy analog

channel of bandwidth B as an example. We have noted
that any signal sent over such a channel can be represented
by 2B samples per second. Let us assume that a message
source produces B samples per second which we wish to
transmit over the channel.

If these are analog samples, we can encode as shown

in Fig. 4. Message sample amplitude is represented as
distance along the double spiral, measured from the center;
negative samples will lie on one arm, positive samples on
the other. The amplitudes of the two signal samples which
will represent a given message sample are the abscissa and
the ordinate of the point on the spiral which represents the
message sample.

In decoding the noisy received signal, we choose as

the decoded message sample the point on the spiral nearest
to the point determined by the two received signal samples.
Because the spiral is long compared with the signal samples,
a small fractional error in either signal sample will lead to
a much smaller fractional error in the corresponding mes-
sage sample. The encoding of the message doubles the
bandwidth required for transmission but reduces the signal-
to-noise ratio required for transmission with a given quality.

We should note that with the encoding of Fig. 4, a

large noise in the signal samples may get us onto the wrong
arm of the spiral. The encoding fails disastrously if too
much noise is added in the transmission of the signal sam-
ples. This is analogous to the phenomenon of breaking in
FM transmission.

144
 Now consider the same transmission circuit of band-
width B, but assume that we wish to transmit symbols from
an alphabet of 37 symbols at a rate of B symbols per sec-
ond. We can represent each message symbol by a point, as
shown in Fig. 5, and encode it by means of the two signal
samples which are given by its abscissa and ordinate. We
decode the noisy received signal as the point nearest that
given by a pair of received signal samples. As in the ana-
log case, the decoding fails disastrously if the noise added
in transmission is too large.

The analog and digital encodings of Figs. 4 and 5

both double the bandwidth required. I believe that they
have comparable efficiencies. We could encode each mes-
sage sample or symbol by means of more signal samples.
The diagrams corresponding to Figs. 4 and 5 would then
have a higher dimensionality. And the ratio of signal
bandwidth to message bandwidth would be greater.

We usually think of a radio channel as a continuous

channel. Suppose that we go to electromagnetic waves of
higher and higher frequency, so that we are using electro-
magnetic waves which we call light waves rather than
electromagnetic waves which we call radio waves. Our
"continuous channel" now exhibits quantum behavior. In-
deed, if the signal is weak, and if we use a photosensitive
device as a receiver, we can watch the discrete photons
of the signal arrive.

Often the noise in a communication system is noise

added to a signal. But, Fig. 1 should be interpreted as
meaning that noise influences the transmission on the signal.
Shannon's work is not limited to additive noise. Indeed,
J. P. Gordon has used his approach in deriving a bound to
the channel capacity in signaling with electromagnetic waves
of very high frequency.

For strong signals, the bound can be approached by

amplifying the signal with an ideal amplifier (a laser can be
an almost ideal amplifier) to such a strength that quantum
effects are negligible in the output. In this case, an added

145
"quantum noise" appears in the output of the amplifier.

An amplifier proves to be a poor way to detect an

exceedingly weak signal. It is better to count arriving
photons. Indeed, if no extraneous noise is added to the
signal, Gordon shows that in the case of a very weak sig-
nal, counting photons enables us to approach the quantum
bound on channel capacity. But this is not so in the presence
of added noise (nonmessage photons).

I have noted that another possible area of investiga-

tion is that of evaluating the entropy of the source (finding
out how redundant the messages it produces are). In a
practical way this goes back to the encoding of English in
Morse's telegraph code. Morse shrewdly observed that !:.
occurs more frequently than say ~, and so he used a short
representation, one dot, for!:. and a longer representation
for z.

Table I, which I have taken from a paper by E. N.

Gilbert, shows three encodings of the letters of the English
alphabet. These all have the prefix property; that is, no code
for a letter forms the leading digits of a code for another letter.
Even though the digits representing successive letters are strung
together without separation, an examination of the message
makes it possible to decode the message unambiguously.

Code 1 of Table I uses the same number of binary

digits for all letters, and so it takes as many digits to
transmit !:. as it takes to transmit~. This cannot be an
efficient code. Code 2 of the figure is the Morse code,
with 1, Oland 00 substituted for dot, dash and letter space;
then 100 identifies the end of a letter. Code 3 is the optimal
code, as devised by David Huffman, for English encoded
letter by letter.

It is interesting to compare the efficiency of these

three codes. In encoding English text code 1 will take five
bits per letter. Code 2 will take 4.868 bits per letter and
code 3 will take 4.120 bits per letter.

146
 Table I. Binary codes for English text

Li p.
1
Code 1 Code 2 Code 3

.1859 00000 00 000

A .0642 00001 10100 0100
B .0127 00010 0111100 011111
C .0218 00011 01101100 11111
D .0317 00100 011100 01011
E .1031 00101 100 101
F .0208 00110 1101100 001100
G .0152 00111 0101100 011101
H .0467 01000 111100 1110
I 00575 01001 1100 1000
J .0008 01010 101010100 0111001110
K .0049 01011 0110100 01110010
L .0321 01100 1011100 01010
M .0198 01101 010100 001101
N .0574 01110 01100 1001
0 .0632 01111 01010100 0110
p .0152 10000 10101100 011110
Q .0008 10001 010110100 0111001101
R .0484 10010 101100 1101
S .0514 10011 11100 1100
T .0796 10100 0100 0010
U .0228 10101 110100 11110
V .0083 10110 1110100 0111000
W .0175 10111 1010100 001110
X .0013 11000 01110100 0111001100
Y .0164 11001 011010100 001111
Z .0005 11010 01011100 0111001111

147
 In order to encode text with greater efficiency we
must know much more about English text than the proba-
bility of occurrence of the individual letters. We could
achieve a greater efficiency by encoding text word by word,
using Huffman's procedure to obtain codes for the various
words. This is not the limit. By an ingenious procedure,
Shannon estimated that in principle only about one bit per
letter should be required to encode English text.

Shannon did this by asking people to guess what

would be the next letter in a message written in English.
The people could use all their knowledge, conscious and
unconscious. of English words and of grammar and of the
meaning which the message might plausibly express.
Figure 6 shows the results Shannon obtained, when he
showed the person who was doing the predicting from one
to 100 preceding letters of the message. Without the use
of a human being, we do not know any way of encoding
English text with anything like this efficiency.

Using his model of the message source as an ergodic

source, Shannon gives a way which in principle should lead
to an efficient encoding of messages from the source. Mes-
sages are broken up into very long blocks of symbols. The
probability of each block which can appear is evaluated by
examining a very long stretch of message. If we know the
probability for each pos sible block of mes sage, we can in
principle encode such blocks of mes sage efficiently by using
the Huffman code.

In practice, it is impractical to find the probabilities

of long blocks of Inessage. For example, in any feasible
examination of English text, many long blocks of letters will
not appear, including blocks from books which have not yet
been written. One must use insight and judgment in order
to arrive at practical and efficient means for encoding
messages.

In facsimile we transmit a few dark lines or letters

against a uniform white background. Surely there must be
some more efficient way of sending facsimile than by

148
specifying whether each tiny area of the mes sage is black
or white. Indeed. one system was devised as early as
1957 by W. S. Michel, W. O. Fleckenstein and E. R.
Kretzmer. This would reduce the number of digits re-
quired for transmission to around a tenth the number
required in conventional systems. Such systems have as
yet found no commercial acceptance.

Many people have speculated that because brightness

tends to change slowly and uniformly in television pictures,
except in regions of fine detail, it should be pos sible to
encode television signals so that they could be transmitted
over a channel of low information capacity. In digital trans-
mission of television we can encode the difference in bright-
ness between adjacent picture elements rather than the
brightness itself. It appears that if we do this. 3 or 4 bits
per picture element give a satisfactory picture. while 7 are
required if we encode the brightness directly. Another
promising technique is to transmit frames of television
more slowly than we now do. store them, and flash each
frame onto the screen several times in order to avoid
flicker. This is how flicker is avoided in motion pictures.
Much more complicated proposals have been made. but they
do not appear to be practical at the present time.

In all, little use has been made of the order or

redundancy of the messages produced by communication
sources. Perhaps this is because communication channels
are getting cheaper and cheaper. while the equipment that
would be needed for efficient encoding of the messages is
still very expensive. The ideal of today is the data channel
which blithely transmits a succes sion of binary digits.
even though these may be all zeros or all ones and hence
completely redundant and of zero entropy rate.

The limitation of the mes sage destination, its

fidelity criterion, its ability to receive information, has
had a far stronger influence on communication systems.
Standards of noise and bandwidth in telephony are set by
what noise will annoy a human being and what bandwidth
he needs in order to understand speech. The detail and

149
noisiness of television pictures is set by the sharpness of
the eye.

Further, in color television ingenious use is made

of the fact that the eye is sensitive to fine patterns of
brightness but not to fine patterns of color. Hence, the
color television signal consists of two parts, a brightness
signal, which produces fine detail and uses up most of the
television channel, and a chroma signal, which uses a
small part of the channel. The chroma signal is used to
paint the color in in broad detail. Little is lost because of
the insensitivity of the eye to fine detail in color.

Both science and common sense tell us that our

senses and our brains are limited. There is a limit to the
ability of our senses to transmit to the brain rapid changes
or great detail in the world about us. There is a limit to
the ability of the brain to appreciate and make use of such
detail. Some experiments. which are in a way misleading.
would seem to indicate that the human being can at the most
use about 50 bits of information per second, a speed that
is comparable to teletypewriter transmission. More
refined experiments, while they show that the eye can see
more than we can remember and reproduce, still indicate
that the brain is not conscious of very much of the detail
that the eye glimpses.

The eye can look where the brain chooses. In tele-

vision transmission it will not do to transmit merely a
small part of the picture. because the eye may look at some
other part. Television transmission would certainly be
more economical if the eye could direct the televis ion
camera to the part of the scene at which it is looking, and
this has been proposed. This would require two-way
transmission, so that it would not be feasible for broadcast
television. But even with the eye fixed in a given direction,
the brain can explore this or that part of what the eye sees.
A reasonable conclusion is that we are forced in pictorial
transmission to transmit far more than the eye can see or
the brain can appreciate.

150
 Are we being misled? Suppose, for instance, that
we consider the use of television in conversing with another
person. Suppose that we don't try to transmit a continuous
picture at all. Before the conversation, we will transmit
a description of the talkers, a description detailed enough
so that we can make little models of the talkers' heads and
shoulders, and clothe and light these models properly. A
model will be so constructed that the head can move, the
eye swink, and the lips and jaws move, with the s arne con-
straints as those of a human being. In order to transmit
a satisfactory picture we need merely find out how a person
who is talking moves the various parts of his face and head,
and have the model at the receiving end reproduce these
motions. This would seem to reduce the channel capacity
needed to transmit all the info rrnation about the person
talking to around telegraph speeds, rather than television
speeds, which are perhaps 10,000 times as great.

This may seem ridiculous, partly because it isn't

part of our life, and partly because we realize that we may
want to look at something other than a head. However, in
listening to human beings we listen to nothing but the human
voice. Indeed, a close analog of what I have just been des-
cribing was invented ove r 30 years ago and is now in limited
practical use, in the transmission of encoded speech.

The device to which I refer is the vocoder. At the

transmitting end of a vocoder transmission system, a voice
analyzer arrives at a description of the human speech in
terms of its pitch, loudness, the strengths of different bands
of frequencies, and whether the sound is a voiced sound such
as a vowel, or fricative sound such as an "f." The speech
itself is never transmitted. Instead, it is encoded as sig-
nals which describe the speech, and these are transmitted.

The decoder at the receiving end is a speech synthe-

sizer ~ an artificial talking machine, which is controlled by
the signals derived by the analyzer. To the ear, the syn-
thesized speech closely resembles the speech of the message
source. A careful analysis will show that it is very differ-
ent, that its waveform resembles very little the waveform

151
of the talker's voice. The important thing is that the
fidelity criterion of the message destination is satisfac-
torily met by this synthetic speech. Unfortunately. the
vocoder tends to have an electrical accent which. along
with its great cost. militates against its general use.

In a practical way there is a cost race between

complicated terminal equipment and broadband transmis-
sion facilities. Inasmuch as it is cheap to send broadband
signals. it is uneconomical to provide fancy transmission
schemes which will make use of the low entropy of the
source or the low ability of the message destination to dis-
criminate among alternatives. Monolithic integrated cir-
cuits may make complicated transmitting and receiving
equipment practical; this may lead us to more efficient
encoding of messages from various message sources.

I have saved until the last a few words about appli-

cation of information theory in broader fields. In the early
days of information theory. people's attitudes were guided
by a feeling that information is a very general term and
that a theory must be a very powerful thing. Accordingly.
they sought to apply information theory to large areas of
human lif e.

In order to achieve powerful results. Shannon

defined information in the narrow and special way. As
Newton's mass has little to do with mass movements. and
as the stres s of mechanical enginee ring has little to do with
the stress of circumstances. so Shannon's information has
little to do with the fields in which many people have tried
to apply. One appears to have a choice. He can either
get meaningful and powerful results using information in
Shannon's narrow sense. or he can generate nonsense and
no results using information in a broader sense. Attempts
have been made to somehow extend Shannon's concept of
information to embrace meaning. but I don't believe that
they have been successful.

Nonetheless. information theory has had some

meaningful impact in a few fields" One of these is

152
cryptography.

In 1949 Shannon published a paper. "Conununication

Theory of Secrecy Syste:ms.'1 Unhappily. it is difficult to
discuss this in a practical way because so :much crypto-
graphic infor:mation is classified. The concepts of infor:ma-
tion theory. and entropy as a :measure. have :markedly
influenced thought and work in cryptography. As in co:m-
:munication in general. info r:mation theory has not :made
cryptanalysis suddenly easy.

In 1956 J. L. Kelly successfully applied infor:mation

theory to dishonest ga:mbling. He showed that the a:mount
of infor:mation received concerning the outco:me prior to
betting gives the exponent of the possible gain which can be
:made.

Atte:mpts have been :made to apply infor:mation theory

in psychology. There are a nu:mber of experi:ments. old and
new, which indicate that the ti:me it takes a person to respond
to a signal is proportional to the infor:mation, in Shannon's
sense, that the signal conveys. This is so. for instance,
when the signal is the visual presentation of one of several
:monosyllabic 'I na:mes ," and the subject's response is speak-
ing the "na:me." Unfortunately, this see:ms to be true only
in the early stages of learning to :make such responses.
Most of our practical life. including our ability to use
language, is highly overlearned. Thus, the relation between
response ti:me and the infor:mation of the sti:mulus see:ms to
apply only to a s:mall part of our behavior.

Psychologists have also tried to bring infor:mation

theory and linguistic behavior together. Indeed it is possible
to indicate infor :mation theoretic reasons for the probability
of distribution of words (known as Zipf's law). But these
are not in fashion.

Another field which has been at least influenced by

infor:mation theory is linguistics. This influence is directly
traceable to Shannon. Shannon generated sequences of
letters by choosing letters and the space at randorn with

153
equal probabilities. As we see in Fig. 7, the groups of
letters are larger than English words, and they do not
resemble English words.

A faint resemblance to English appears and in-

creases as we make the symbol probabilities approach
those of English. If one chooses a letter on the basis of
its probability of following the preceding one or two
letters; English words appear in the text, and the words
that are not English resemble English words (Fig. 7).

By choosing words at random on the basis of their

probability, or on the probability of their following the
preceding word, one can obtain a sort of pseudo-English
which sometimes seems to say something (Fig. 8).

Chomsky and other modern linguists point out

somewhat bitterly that this is not a royal road to the under-
standing of English or of any natural language. The con-
straints of grammar are not constraints of a word following
other words with a certain probability. E". N. Gilbert gave
a very simple proof of this, shown in Fig. 9. The text
reads like ordinary English, yet it contains no ~, which is
very improbable in English text.

The broad pattern of language is consistent with

info rrnation theory as explicated by Shannon. Only certain
utterances are possible, because the constraints of gram-
mar rule out many sequences of words. Other sequences
of words are very unlikely because mo stly people talk sense,
or what sounds like sense. Because of the constraint of
grammar and the constraint of saying meaningful things
about our common life and environment, the number of
pages that will be written is far smaller than the number
of pages that ,!!light be written if any letter could follow
any other letter or if any word could follow any other word.

Linguists such as Chomsky and his followers have

revolutionized grammar. Indeed, they have tried to incor-
porate into grammar enough constraints to rule out im-
plausible sentences such as "golf plays John." In so doing

154
they may have tried to include in grammar something which
would be easier to include in the general idea of an orderly
world and sanity and sense in speaking about it. Informa-
tion theory and linguistics may yet corne meaningfully
together in discussing the probabilistic constraints of
grammar and a meaningful environment on the utterances
of man.

Information theory has also been evoked in the fields

of art and music. Indeed, in the past I have written on this
at some length. I found some of the words and passages
that Shannon constructed by probabilistic methods affecting
and charming. Around 1950 I constructed more, and I also
constructed what I called stochastic music. Such music has
since corne to be called aleatory music--I believe through
an uninformed translation of the French al~atoire.

You may expect to hear some of my ideas about

information theory and music, but you won1t. Instead, you
will hear an expression attributed to Beethoven by my friend
Gerald Strang. According to Strang, Beethoven said that
every part of a composition should be equally surprising
and expected. I believe that I understand this well. If
something isn't surprising, it will be boring; if it isn't
expected, it will be confusing, and one will soon lose all
s ens e of continuity or meaning. This simple statement
seems to me to say all I can validly say about information
theory and music. I refer those who wish to go further to
various issues of the Gravesaner Bl'atter.

155
 REFERENCES

1. Claude E. Shannon and Warren Weaver, "The Mathe-

matical Theory of Communication," The University
of Illinois Press, Urbana, Eighth Printing, 1959.

2. E. N. Gilbert, "Information Theory After Eighteen

Years," Science, Vol. 152, pp. 320-326, April 15,
1966. (This is an excellent summary, especially
of coding, and has a good bibliography.)

3. J. R. Pierce, Symbols, Signals and Noise, Harper and

Brothers, New York, 1961. (This treats a large
variety of topics at a popular or semipopular level.)

4. J. P. Gordon, "Quantum Effects in Communication

Systems," Proceedings of the IRE, Vol. 50,
pp. 1898-1908, September, 1962.

5. Co E. Shannon, "Prediction and Entropy of Printed

English," Bell System Technical Journal. Vol. 30.
#1, pp. 50-64, January, 1951.

6. A number of interesting papers on transmission of

pictures with reduced channel capacity, and one on
transmission of English text with reduced channel
capacity. appear in Proceedings of the IEEE,
Vol. 55, March, 1967.

7. W. S. Michel, W. O. Fleckenstein and E. R. Kretzmer,

"A Coded Facsimile System," IRE WESCON Conven-
tion Record, Part 2, pp. 84-93, 1957.

8. George Sperling, "Successive Approximations to a

Model for Short-term Memory," Acta Psychol..
Vol. 27, pp. 285-292, 1967 (based on experiments
in visual perception and recall).

9. J. L. Flanagan, "Speech Analysis, Synthesis and Per-

ception." Springer-Verlag, Berlin-Heidelberg-
New York, 1965 (about vocoders and related matters).

156
10. C. E. Shannon, "Communication Theory of Secrecy
Systems, II Bell System Technical Journal, Vol. 28,
pp. 656-715, October, 1949.

11. J. L. Kelly, Jr., "A New Interpretation of Information

Rate." Bell System Technical Journal, Vol. 35,
pp. 917-926, July. 1956.

157
INFORMATION TRANSMITTER RECEIVER DESTINATION
SOURCE (CODER) (DECODER)

SIGNAL RECEIVED
SIGNAL
MESSAGE MESSAGE

NOISE
SOURCE

Fig. 1. Schematic diagram of a general communication system.

5 7 6 0 4 2 3

Fig. 2.
0 0

0 0

0 0 0
Fig. 3.

OUI
01lJ z.J
z.J oD..
on. ()~
u:::t UI<I
IIJ« mm
"'''' • • •
• • •
• • 0
• • 0

FIRST
SAMPLE
0 0
• • •
• •
0 0 0 0

Fig. 5.
Fig. 4.
158
 5

~
t(
,

4
\\
Io-UPPER
\\ ~
'~ BOUND
3 \
\ I. ~
f!!
I-' CD ~
l ,
lJ1
~ 2
~
"" ....
~
,
LOWER .........
.. ...,A ) l
BOUND
""" r----. - ~
N """'" . . r---- ) l
(
-
I (

'------ ,--, <- --

o, 15 100
o 2 3 4 5 6 7 IS 9 10 II 12 13 14
NUMBER OF LETTERS

Fig. 6. Upper and lower experimental bounds for the entropy of 27-letter English.
10 Zero-order approxiInation (symbols independent and
equi-probable) •
XFOML RXKHRJFFJUJ ZLPWCFWKCYJ FFJEYV-
KCQSGHYD QPAAMKBZAACIBZLHJQD.

2. First-order approximation (symbols independent but

with frequencies of English text).
OCRO HLI RGWR NMIELWIS EU LL NBNESEBYA
TH EEl ALHENHTTPA OOBTTVA NAH BRL.

3. Second-order approximation (digram structure as in

English) •
ON IE ANTSOUTINYS ARE T INCTORE ST BE S
DEAMY ACHIN D ILONASIVE TUCOOWE AT
TEASONARE FUSO TIZIN ANDY TOBE SEACE
CTISBE.

4. Third-order approximation (trigram structure as in

English).
IN NO 1ST LAT WHEY CRATICT FROURE BIRS
GROCID PONDENOME OF DEMONSTURES OF THE
REPTAGIN IS REGOACTIONA OF CRE.

Fig. 7

160
5. First-order word approximation. Rather than continue
with tetragram ••••• n-gram structure is easier and
better to jump at this point to word units. Here words
are chosen independently but with their appropriate
frequenc ies.
REPRESENTING AND SPEEDIL Y IS AN GOOD APT
OR COME CAN DIRRECT NATURAL HERE HE THE
A IN CAME THE TO OF TO EXPERT GRAY COME
TO FURNISHES THE LINE MESSAGE HAD BE
THESE.

6. Second-order word approximation. The word transition

probabilities are correct but no further structure is
included.
THE HEAD AND IN FRONTAL ATTACK ON AN
ENGLISH WRITER THAT THE CHARACTER OF
TillS POINT IS THEREFORE ANOTHER METHOD
FOR THE LETTERS THAT THE TIME OF WHO
EVER TOLD THE PROBLEM FOR AN UNEXPECTED,

Fig. 8

161
 ABSTRACT

If noisy digits occur in bursts, a calculation of

capacity is tricky. A difficulty is that a transition diagram
for such a link may fail McMillan's unifilar condition. In
a burst situation of utmost simplicity a capacity is found.

Our work also has cryptographic and linguistic

applications. It shows that writing may avoid basic statis-
tical constraints without sounding odd. In particular our
writing will adopt an unusual monogram probability distri-
bution.

Fig. 9

162
 CHAPTER 2

INFORMATION THEORY AND MODERN

DIGITAL COMMUNICATION

R. W.LUCKY
Bell Telephone Laboratories, Holmdel, New Jersey

ABSTRACT

The fundamental theorem of information theory states

that digits can be transmitted through a noisy channel with
an arbitrarily small probability of error at any rate less
than a certain limit known as channel capacity. For the
past two decades this theorem has presented a constant
challenge to communication theorists. In spite of a great
deal of theoretical and applied research, practical com-
munication systems realize only a fraction of the capability
promised by the bounds of information theory. A discussion
of one of these bounds, obtained by random coding, points
out some of the difficulties in the application of information
theory. The probability of error decreases exponentially
with message length, but the system complexity increases
exponentially at the same time. Furthermore information
theory gives no constructive procedure by which its ideals
can be achieved.

Present research in communication theory centers

on two distinct areas - -modulation and coding. Although
each of these areas is discussed, the bulk of the paper is

163
devoted to an exposition of current coding techniques. In
these techniques extra, redundant, digits are added to the
message before transmission. At the receiver this redun-
dancy is used to correct errors which may have occurred
during transmission. The use of coding represents a
practical approach toward the ideals of information theory.
The rapidly decreasing cost of physical implementation and
the increasing need for digital communication promise new
uses of infQrmation theory in communications technology in
the near future.

164
 Information theory started about two decades ago as
a technical tool for the communication engineer. In the
early years it received a great deal of publicity as scientists
attempted the application of information theoretic concepts
to many seemingly-alien fields. Although some of these
applications have withstood the test of time, many more
have proved to be only superficial. In the late 1950's a
journal devoted to information theory, decrying the flood of
worthIes s interdisciplinary paper s submitted to it. editori-
alized about the gene ric paper entitled "Information Theory,
Photosynthesis and Religion." In this type of paper the
author describes the close relationship between the concep-
tual framework of information theory and. e. g •• psychology.
"Having placed the discipline of psychology for the first time
on a scientific base. the author modestly leaves the filling
in of the outline to the psychologists. "

Much of this fanfare subsided and engineers settled

down to the real business of applying information theory to
the study of communication. J. R. Pierce in his companion
paper. "A Survey of Information Theory," has described
many of the areas in which information theory has had an
impact on modern technological thought. In the present
paper we shall concentrate on that area in which Shannon's
famous paper of 1948 presented its most well defined chal-
lenge- -the field of digital communication. With this re-
striction we shall not be concerned with the various formi-
dable problems of source encoding, that is, with the trans-
formation of the information to a suitable digital form.
Rather we shall deal with the efficient transmission of the
digits themselves from the source to the destination.

In the paper "The Mathematical Theory of Communi-

cation," Shannon(1) proved that digits could be transmitted
through a noisy channel with an arbitrarily small probability
of error at any rate less than a certain limit he termed
channel capacity. When the values of channel capacity were
computed for many of the common communication media it
was discovered that they were very much greater than the
data rates of existing transmission systems. In the twenty
years since Shannon's publication both theoretical and

165
practical aspects of digital communication have undergone
an enormous expansion. Yet a large gap exists between
the theoretically possible and the practically achievable.
In fact. it is still difficult to point to substantial contribu-
tions of information theory to the practice of digital com-
munication.

In subsequent sections of this paper we will discuss

the present status of information theory in digital communi-
cation both on the theoretical and applied levels. In order
to understand the problems involved in approaching Shannon I s
ideals we shall begin with a brief discussion of some of the
relevant theory. We will see that, although the theorems of
information theory tell what ~ be done, they do not indicate
explicitly how to go about doing it. The search for theoreti-
cal and practical techniques of implementing the ideals of
information theory is still going on. Much of this work
centers on coding theory, and thus the greater portion of
the remainder of this paper will be devoted to a heuristic
explanation of some of the coding techniques which are now
beginning to diffuse through communications practice.

Before delving into the theory, it should be pointed

out that the slowness of application of information theoretic
concepts to practice has not been due solely to technical
reasons. Economic necessity has been a very great factor.
While there has always been a large gap between the practi-
cally achievable and the theoretical ideals. there has not
been much of a gap between what could be done and what was
needed. Only in recent years has this latter gap been
growing. The growth of computer technology and the conco-
mitant need for machine -to -machine communication, the
widespread acceptance of pulse code modulation, and the
so-called "information explosion" all have created an ex-
ponentially- growing need for data communication. At the
same time the new device technology is enabling the imple-
mentation of increasing complex systems. The combination
of these factors promises in the near future the long-awaited
breakthrough for information theory in communication tech-
nology.

166
 The Theory

Most of the results in information theory have been

applied to the Gaussian channel. In this mathematical
model the channel accepts samples xn at discrete times
tn. The channel adds a noise sample Sn to each input
sample ~ where the samples Sn are considered to be
independent Gaussian random variables with variance N.
If the input samples are average power limited t i. e. t
K
lim ~ ~ x~= S~ then Shannon proved that there was an
K-a:> i=1 1
upper limit to the amount of information which could be
conveyed through the channel by each sample x n •

In order to provide a bridge between the time-

discrete~ Gaussian channel and the more conventional sort
of channel which accepts time-continuous waveforms t
Shannon invoked the sampling theorem. This theorem
states that a signal bandlimited to W cps can be completely
specified by time samples taken at 1/
2W second intervals.
Another way to look at this theorem is that a bandlimited
waveform has 2WT independent samples. or "degrees of
freedom t " at a T second interval. After using this method
of representing bandlimited waveforms by their time
samples Shannon obtained his celebrated equation
C = W log (1 + SiN) for the information capacity of the
bandlimited channel.

Although Shannon I s original paper proved that

virtually error-free information could be transmitted
through the Gaussian channel at any rate less than capacity.
an infinite signal processing time was required and the
proof gave no hint as to the error performance of systems
employing finite processing times. Shannon himself clari-
fied the situation and set a new trend in communication
theory with the publication of a paper(2) in 1959 in which
he obtained upper and lower bounds to the probability of
error achievable on a Gaussian channel with finite signal
processing or "coding delay." It will be instructive at this
time to consider the method by which Shannon derived the

167
upper bound to the achievable error probability.

Suppose as a concrete exa:mple that we wish to

trans:mit a :mes sage of 3 bits through the Gaussian channel
during a ti:me interval equivalent to, say. two signal sa:m-
pIes. Equivalently we :might think of a long :message being
trans:mitted in blocks of three bits at a ti:me using two
channel sa:mples for each block. There are eight possible
:messages or blocks (000,001,010,011,100,101,110, and
111) each of which :must be put into correspondence with a
particular selection of the two sa:mples xi and x Z " We
:might think of these sa,E'lples as being the co:mponents of a
Z-di:mensional vector Xo Since there is an average power
li:mitation on the channel input, the lengths of the 8 vectors
chosen :must be such that all lie on or within a circle of
radius ill. More generally if we had wanted to trans:mit
at an average rate of R bits per sa:mple for a duration of
M sa:mples. then we would have had to select ZRM vectors
in an M-di:mensional space. It is i:mportant to realize that
the nu:mber of signal vectors grows exponentially with the
length of the :message or block to be trans:mitted.

In trans:mission through the channel two noise co:m-

ponents. constituting a noise vector ~, are added to the
signal co:mponents to produce a disturbed vector y. The
receiver has a duplicate of the code book used by the trans-
:mitter listing the eight possible :message vectors. On the
basis of the received vector y the receiver :makes a
decision as to which of the eight :message vectors was :most
likely to have been trans:mitted. If all :messages were
initially equally likely. this :merely a:mounts to testing each
of the eight pos sible :message vectors and selecting that
:message vector closest to y in Euclidean nor:m. The
geo:metric interpretation of the trans:mission syste:m is
shown in Figure 1. Notice that the receiver decides all
three bits constituting the :mes sage at once. rather than by
:making bit-by-bit decisions ~ and that the nu:mber of alter-
natives which :must be tested grows exponentially with
:message length. Although the exa:mple here is si:mple and
intuitively obvious, the decision rule applied by an ideal
receiver has been given :much study in the field of statistical

168
decision theory.

For a given selection of the set of 8 signal vectors

the probability of error could be corne. uted
. as the proba-
bility that the Gaussian ~ise vector S is such that the
x
received vector y = + S lies closer to one of the 7 in-
correct signal vectors than to x. Although this is con-
ceptuall y calculable. even in two diITlens ions an exact
solution is all but iITlpossible to obtain. FurtherITlore.
what we really need to know is the ITliniITluITl probability
of error which can be obtained when using the best possible
set of 8 input vectors. Fortunately all the difficulties are
circuITlvented by an ingenious arguITlent called the randoITl
coding bound. Suppose that each cOITlponent of each of the
eight input vectors is selected subject to the average power
constraint. but otherwise at randoITl. We could iITlagine
spinning pointers or drawing saITlples froITl a Gaussian
population or using SOITle other randoITl ITleans of selecting
the set of vectors. Then surely the best set of vectors
ITlust have a probability of error at least as sITlall as this
randoITlly selected set. so the probability of error for the
randoITl set serves as an upper bound to the achievable
probability of error on the channel.

It is ironic that. even though the probability of error

cannot be cOITlputed for any given choice of the input vector
set. the average probability of error over a randoITlly
chosen enseITlble of signal sets can nevertheless be tightly
bounded. In the interpretation given to the randoITl coding
bounds in ITlore recent papers {especially noteworthy is that
of Gallager} (3) for a given rate R and block length M the
bound is put in the forITl
P < e -ME(R)
e
where for large M the exponent E(R) is as shown in
Figure 2. This bound shows that aSYITlptotically the proba-
bility of error decreases exponentially with the ITlessage
(or block) length M. So long as the rate is such that E{R)
is positive, the probability of error can be ITlade arbitrarily
sITlall by increasing the ITlessage length. The rate C for
which E{R) goes to zero is of course called the channel

169
capacity.

A number of other bounds, both upper and lower. to

the probability of error at rate R bits per symbol for a
message of length M symbols have also been obtained.
The asymptotic exponential decay of error rate with M is
now fairly well determined~ but in the region between Rcrit
(the "critical" rate) and capacity shown in Figure 2 the decay
exponent for the best code is known to coincide with the E(R)
from the random coding bound. The implication is that
signal sets selected at random achieve on the average the
best obtainable asymptotic performance.

The info rrnation theoretic bounds point out the practi-

cal dilemma. The probability of error at any rate below
capacity can be made exponentially smaller by increasing
the number of input bits which we encode into a channel
vector at a time. However the number of vectors required
in the transmitter and receiver code books. the amount of
searching required to make a decision at the receiver, and
generally the complexity of the system all increase expo-
nentially with the block length M. To make matters worse
the decay of error rate with M for small values of M
(which are the only values presently feasible) is not parti-
cularly rapid. For an example we have shown in Figure 3
a curve computed by D. Slepian(4) showing the additional
signal power in decibels above the minimum power S given
in the capacitl formula needed to obtain a probability of
error of 10- as a function of the block length M. Observe
that at M = 20 where transmitter and receiver would have
to contend with 220R different waveforms the transmission
system requires 7 db. more (about 5 times as much) power
than the ideal system given by the capacity formula and in
which M =00. It is evident that the approach to the per-
formance promised by the capacity formula is slow and
paved with exponentially increasing system complexity.

170
 Quantized Signal Samples - Modulation and Coding

The direct method of mapping digital input sequences

into signal sample sequences involves too many vectors to
be practically useful. If instead the signal sample values
xn are only allowed to assume a discrete set of levels
(usually 2, 4 or 8 for simplicity of instrumentation) then the
coding process can be made entirely digital. For example
the signal samples might be quantized to four poss ible values.
each of which could be represented by two binary digits as
shown in Figure 4. Each of the eight vectors in the previous
example could then be specified by 4 bits (two components.
each given by two bits).

A code would consist of a mapping of each possible

three bit information block into a four bit signal vector,
e.g. 000 - 0000, 001 - 0011, 101 - 10to, etc. This
particular code would be said to have an efficiency (also
called rate) of 0.75 since only 3/4 of the bits specifying the
transmitted waveform are used for information transmission.
This redundancy in the selection of transmitted signals
enables us to choose signal vectors which are relatively
far apart and hence less susceptible to confusion by noise.
Since 4 bits are used to specify each signal vector there
are 24 = 16 possible vectors, but only 8 of these vectors
need be designated in the code set. Presumably we would
choose those 8 of the possible 16 which were in some sense
"maximally separated. "

Obviously the capacity of the quantized system is

less than the capacity of the original channel, but if the
quantization is comparable to the standard deviation of the
noise then the loss in capacity is small. The choice of
quantization is crucial and delicate. If too few bits are
used to represent each sample {component of the signal
vector} then the coarseness of the quantization results in a
serious capacity loss. On the other hand if too many bits
are used then the probability of deciding the components
incorrectly at the receiver is large. While this poses no
theoretical problem, it does mean that much more compli-
cated codes are required to ensure that the probability

171
of error for the vector itself is small.

Because of the quantization of the signal sample

values 9 the communication problem is divided into coding
and modulation aspects. The coding cons ists of a set of
rules by which the input sequence of digits is mapped into
a longer sequence of digits. The modulation portion of the
system takes the coder output digits 9 uses these digits to
specify sample values of the transmitted waveform and
performs whatever operations are necessary on the wave-
form to prepare it for transmis s ion over the communication
medium. Not only does this philosophy dichotomize the
system, but it also has served to dichotomize the set of
engineers dealing with communication theory. Those who
deal with coding deal with digits (usually binary) and em-
ploy the tools of modern algebra. Those who specialize in
modulation and detection employ Fourier analysis or statis-
tical decision theory. The degree of specialization required
in each of these two fields has practically precluded knowl-
edge of the other.

In the remainder of this paper we will discuss briefly

current research in modulation techniques and in coding
techniques. We will devote greater space here to the coding
problem since this aspect of communication has been more
influenced by information theory than has modulation.
Indeed coding owes its very existence to information theory,
while modulation techniques have been relatively unaffected.

Real Channels and Modulation Techniques

Modulation techniques were fairly well understood

before the advent of communication theory. The basic
methods of amplitude, phas e a and frequency modulation
were all in practical use in the 1940' s. Until this time
engineers had not taken the effects of random noise explic-
itly into consideration in their calculations and designs.
During the war it became necessary to deal with random
waveforms in control and communication systems. The
most influential work of this time was Wiener's book
Extrapolation, Interpolation and Smoothing of Stationary

172
Time Series (5) Although this report was affectionately
0

termed the "yellow peril" becaus e of its color and mathe-

matical difficulty. it soon became the cornerstone of
statistical communication theory. In it Wiener derived
certain linear filter configurations which optimally sepa-
rated des ired signals from unwanted noise.

After Wiener I s work engineers adopted statistical

decision theory and derived mathematically optimum
receiver structures for most common modulation techniques
and situations. (6) These results have not yet had a great
effect on digital communication practice. It turned out that
many of the optimum receiver configurations were already
in practical use when they were derived theoretically. Also
the solutions were for the most part quite insensitive to the
exact filter response used. so that practically optimum
performance was obtained from receivers which sometimes
bore little resemblance to the theoretical ideals.

We do not mean to imply that modulation techniques

are well understood and are near-optimum. Efficient
digital communication depends on a close relationship
between modulation and coding. The relatively poor effi-
ciencies obtained by current transmission systems can be
blamed more on inefficient modulation than on poor coding
practice. As an example of this situation let us consider a
typical voice telephone channel.

A good quality voice channel has a usable bandwidth

of about 2400 cps and a signal-to-noise ratio of approxi-
mately 30 db. The capacity of such a channel is on the
order of 24,000 bits per second. Using practical coding
techniques and straightforward amplitude modulation we
can on paper send at a rate of about 20,000 bits per second
with a very low probability of error. Until recently the
highest commercially obtainable rate was 2400 bits per
second--only 10% of capacity! Why is there such a big
difference between analytical and practical results?

The major reason for this difference is that the

Gaussian channel does not adequately model the voice

173
telephone channel. Similar gros s inadequacies of the
Gaussian model are found on high frequency radio and
tropospheric channels. On the other hand the deep space
and, to a lesser extent, the satellite channel are fairly
well characterized by the Gaussian model. In the case of
the telephone and h. f. radio media the real channels do
something other than just add Gaussian noise to the
transmitted signal. Probably their most striking charac-
teristic is that they also distort the signal in a linear (but
possibly time varying) manner. This distortion is a form
of time dispersion created by the arrival at the receiver of
various frequency components of the signal at different
times. In the radio channel the signal may also arrive via
a number of diverse electrical paths - -the same phenomenon
which produces ghosts in television pictures.

Of course no mathematical model perfectly charac-

terizes a real channel, but we would at least hope that the
dominant attributes of the real channel were included.
That this is not so for the radio channel is demonstrated in
Fig. 5 where we have drawn a curve of probability of error
for a particular modulation system versus average signal
power. Also shown is a theoretical curve based on a
Gaussian channel model. Naturally this latter curve shows
that we can reduce the probability of error to an arbitrarily
small value by increasing the signal power and simply
"overpowering" the channel noise. However the actual
curve shows a "bottoming" effect. Regardle s s of signal
power the probability of error cannot be reduced below a
certain minimum value caused by self-noise of the signal
due to the distorting effect of the channel.

A linear operation on the transmitted signal can be

appended to the mathematical model. This has been done
and in recent years certain theoretical results have been
obtained as to channel capacity, coding bounds, and opti-
mum signal processing methods. {7} These latter methods
are critically dependent upon the exact signal distorting
characteristics of the channel. The characteristics are
not known to the communication system designer a priori,
therefore a recent trend has been towards adaptive receiver

174
structures. (8) In these systems the receiver attempts to
learn the channel characteristics as closely as possible at
any given time and to redesign itself for optimum reception
on the particular channel it has identified. Adaptive methods
have considerably improved practical system performance
on h. f. radio and telephone channels. In the telephone
example which we mentioned previously data rates of 10 ,000
bits per second are obtained with the aid of adaptive receiver
circuitry.

Another big difference between theory and practice

has been in the actual application of the sampling theorem.
It has proved impossible to actually transmit independent
samples at intervals of 1/ZW seconds. As this sampling
rate is approached the problems of filter design and system
sensitivities to perturbations in timing and other parameters
become insurmountable. Only about half of the promised
Nyquist sampling rate is usually obtained~ in which case
the capacity formula is reduced proportionally to
(W /Z) log (1 + S/N). Furthermore in many cases the modu-
lation technique has been frequency or phase modulation,
which use roughly twice as much bandwidth as should be
required for a given sampling rate- -another factor of two
loss in capacity. The reason why such inefficient systems
are used is because they are inexpensive, and because they
possess an inherent rigidity against the sort of channel
impairments which are not included in mathematically-
tractable channel models.

Block Codes

It has been said that Shannon I s theorems tell us in

effect. "Good codes exist, now you go find them." The
random coding and other bounds of information theory give
no constructive procedure by which codes can be selected.
Just picking codes at random and listing the code words in
dictionaries may be theoretically acceptable. but the enor-
mous number of words involved {more than Z100 in a typical
situation} makes such a procedure practically unthinkable.
The only way this large number of code words can be handled
is to introduce some mathematical structure into the code

175
so that code words can be calculated from the input infor-
mation digits. For this reason coding theorists have turned
to the theorem'S of modern algebra dealing with groups,
rings, and fields. Recently, codes have also been formed
based on projective geometry, perfect difference sets and
graph theory. In certain situations the best existing codes
have even been found by computer-aided trial and error or
search procedures. Although many portions of coding
theory possess a beautiful mathematical framework, the
field as a whole seems to have developed in patchwork
fashion when compared with the original foundations of
information theory.

The earliest and best known of the codes originated

in the information theory era are the Hamming codes. (9)
As an example of this type of code let us consider the
Hamming code which maps blocks of four information
digits into blocks of seven digits for transmission over the
channel. Such a code is designated a {7,4} code; for each
four information digits. three check digits are computed
and appended to the end of the information digits to form a
7-bit block. Let d1' d Z ' d 3 , and d 4 be the data, or infor-
mation,bits and P1' PZ' and P3 represent the parity, or
check, bits. Then the rules by which the check digits are
computed for the Hamming (7,4) code are:

P1 = d 1 + d Z + d 3
Pz =d Z + d 3 + d4
P3 = d 1 +d Z +d 4
Since we are talking about binary digits {bits}, the arith-
metic used for the three parity check equations is modulo-Z,
=
i. e. , 0 + 0 =
0, 0 + 1 1, 1 + 1 =o. The check digits are
called parity bits because they check on the evenness or
oddness of the sum of the number of 1's in a certain subset
of the information digits. For example, in the (7,4) code
the first parity bit p 1 is 0 if the sum of l' s in data bits
d 1 , d Z ' and d 3 is even and is 1 when this sum is odd.

The 16 words in this code are listed below:

176
 0000 000 0100 111 1000 101 1100 010
0001 011 0101 100 1001 110 1101 001
0010 110 1001 110 1010 011 1110 100
0011 101 0111 010 1011 000 1111 111
We can demonstrate many of the properties of more com-
plicated codes with this simple example. We have already
seen that the process of encoding consists of calculating
the three parity check bits from the four input data bits.
Practically this can be accomplished with an extremely
simple 3-stage feedback shift register circuit. This is a
circuit with three binary storage elements and means of
shifting the contents of one storage unit to the next at
regular bit intervals. Two feedback connections enable
the circuit to calculate the required parity check bits on
successive shifts of the contents after the data bits have
been loaded into the stores.

In transmission through the channel errors may

occur at various positions within the code word. The effect
of an error at a given position is to add a binary 1 at that
point in the word. Ideally the receiver would decode the
sequence by selecting the most likely transmitted word.
given the particular received bit sequence. If all errors
and code words were equally likely {an assumption which
is invariably made} optimum decoding would consist of
selecting the code word which is closest to the received
bit sequence in the sense that it differs in the fewest bit
pos itions.

To illustrate decoding let us as sume that the trans-

mitted word is the all-zero sequence 0000 000 and that an
error occurs in the third bit. We receive 0010 000. This
sequence differs from the transmitted word in only one
pos ition; thus. it is said to be at distance one from the
correct code word. An examination of the table of 16 code
words shows that 0010 000 is at least distance two from
every other code word. Thus. 0000 000 is unambiguously
the closest code word and is correctly selected by the
receiver. In fact we can see that this word would have been
received correctly regardless of the position of this single

177
error. Each nonzero word in the code is at least distance
three from the all-zero code word since every othe r code
word has at least three l' s. This holds true not only for
the all-zero code word, but between any two code words.
The code is said to have minimum distance 3 and can clearly
be used to correct all single errors. Indeed the rationale
for the code construction rules was to create a set of 16
code words which were as far apart as the given amount of
redundancy would allow- -in this case a minimum distance
of three between any two code words.

In practice true optimum decoding cannot be imple-

mented for codes of any reasonable length, but something
nearly equivalent is accomplished instead. If a code has
the ability to correct all patterns of t or fewer errors it
is said to be a t-error correcting code. Usually the code
can also be used to correct some, but not all. patterns of
greater number of errors. (The (7,4) Hamming code is
one of a small class of exceptions called perfect codes. It
can correct all single errors, but this exhausts its capability
and no other error patterns are correctable.) Generally,
the decoding procedure ignores the possibility of correcting
those patterns which contain more errors than the code's
guaranteed capability.

At the receiver the first step in decoding is to recal-

culate the check bits based on the bits actually received in
the information positions of the code word. The recalculated
check bits are compared with the received check bits to see
if there is agreement. If there is, then the word is accepted
as correct as it stands. If not, then one or more errors have
occurred. The error pattern mayor may not be correctable
at the receiver, but its existence has been detected. The
knowledge that errors exist in the given block may be used
to signal the transmitter, via a feedback loop, that the word
should be repeated. Codes used in this manner for detection
and retransmission are simply-implemented and quite power-
ful. For instance the (7,4) Hamming code can only correct
single errors, but it can be used instead to detect the pres-
ence of all double error patterns. This latter is true since
detection only fails when one code word is converted into

178
another code word by the channel--and this requires at
least three errors in the (7,4) code.

In an elaborate experiment(10) a (31,l1) code

operating in a detection-retransmission mode was tested
on typical data calls over the switched telephone network.
This code has a minimum distance of 5 between code words.
Thus all patterns of 4 or fewer errors in each block of 31
bits are detected. (In addition about 99.9% of all other
error patterns are detected.) The decoder employed 357
transistors for logic operation and required a 540- bit
binary storage unit. Without the error· control unit the
probability of error at a data rate of lOOO bits per second
was measured at 3.7 X 10- 5 • With error control at a speed
of 1100 bits per second (allowing for the code's redundancy
and for the retransmission times) the measured error rate
was 7 X 10- 9 • This represented an improvement in accuracy
of about 5000. Notice that detection-retransmission is
particularly powerful and efficient when the input error
rate is low, as it was in this experiment.

By far the majority of coding systems which have

been constructed are of the detection-retransmission variety.
However, this is not the solution to all coding problems. A
feedback channel may not be available, retransmis sions may
be too time consuming, or errors may be so frequent that
nearly every word would have to be repeated and very little
information would be delivered to the sink. On the other
hand, forward-acting correction without the aid of feedback
is usually very much more complicated than simple error
detection. In an (n,k) code there are In-k different ways
that the recalculated check bit sequence can differ from the
actual received check bit sequence (including the all-zero
sequence indicating agreement). Decoding entails the deter-
mination of which of the correctable error patterns is associ-
ated with the particular pattern of disagreements between
received and recalculated check sequences (called the syn-
drome). Once the error pattern has been determined, it
can be added to the received pattern to produce the corrected
code word. (Notice that in binary arithmetic addition and
subtraction are equivalent.)

179
 Once again the spectre of exponential growth in
complexity with block length afpears, for it seems neces-
sary to use a table listing Zn- different syndromes and
their associated error patterns. Actually for some short
codes a similar procedure is used in which all correctable
error patterns are added in turn to the received sequence.
The syndrome of the resultant is calculated in each case
=
if it is a code word. In the (7,4) code there are only Z3 8
correctable error patterns, each of which could easily be
tried and tested by the decoder. However, for longer and
more powerful codes so many patterns would have to be
tried that it would be physically impossible for the decoder
to complete all the calculations before the arrival of the
next code block. For these longer codes it is necessary
to have some way of calculating the error pattern directly
from the syndrome. Just how this is done depends upon the
structure of the particular code. In some cases it is re-
markably simple; in others no constructive procedure is
known.

Nearly all coding systems which have been imple-

mented have us ed binary parity check codes similar in
many ways to the (7,4) Hamming example. It has been
shown by various bounding arguments that the class of
parity check codes contains good codes. (11) This means
that there do exist parity check codes for which the proba-
bility of decoding error can be made exponentially smaller
by increasing the block length, while maintaining a nonzero
rate less than channel capacity. However, with one excep-
tion, none of the known procedures for constructing parity
check codes yields asymptotically good codes, i. e. , either
the codes have zero rate or high probability of error in the
limit of large block length. It is curious that the average
performance of codes picked at random is asymptotically
better than the best of the known families of codes. Perhaps
it is the absence of mathematical structure which results
in the good average performance of the random codes.

The Hamming codes are a subclass of the Bose-

Chaudhuri-Hocquenghem (BCH) codes. {iZ} Some of the
basic properties of these codes are evident in the 16 code

180
words of the {7,4} example. The code words form a group
under the operation of addition. (The sum of any two code
words is another code word. the addition is associative»
the all zero code word is the identity element, and each
word has an inverse within the code.) Also, the code is
cyclic. Any cyclic shift of a code word results in another
code word. For example. a one unit right shift of the code
word 0100 111 gives 1010 011, which is als 0 a code word.

The BCH codes are best described as polynomials

with coefficients in the binary field. If the seven bits of the
example code are c6' cs. c4' ••• ,co then we associate a
polynomial c 6x 6 + cSxS + c4x4 + ••• + Co with each code
word. The word 0100 111 is the polynomial x S + xZ+ x + 1,
etc. We can show that every word of a cyclic code of length
n must be a multiple, modulo x n + 1, of a generator poly-
nomial g(x} which divides x n + 1. For the (7,4) code the
generator polynomial is g(x} = x 3 + x + 1. As an example,
it might be verified that the code word x S + xZ+ x + 1 is
(x Z +l)g{x), and that x7+ 1= (x4 +x Z+x + l)g(x).

The theorem basic to BCH codes gives the minimum

distance of the code in terms of the roots of the generator
polynomial. To construct a BCH code the generator poly-
nomial is selected which contains the appropriate sequence
of roots and the code is determined by all possible multiples
(modulo x n + 1 for a code of length n) of g{x). Extensive
tables have been compiled which make the determination
of the generator polynomial relatively straightforward.

The BCH codes constitute the largest and generally

most useful family of codes known. They can be used to
correct t errors in a block of length Zm- 1 digits with at
most mt parity check digits. Asymptotically they are not
good codes, but up to lengths in the thousands, far beyond
the limits of practicality, their performance is still rela-
tively close to that given by theoretical bounds.

When the BCH codes were discovered, there was no

general procedure for decoding. Finding the positions of
the errors from the syndrome involved the solution of a set

181
of simultaneous nonlinear equations. Late r a method us ing
Newton's identities was discovered which required the in-
version of a matrix and a relatively simple substitution
process. A number of modifications of the decoding pro-
cedure have been proposed in recent years. Still, however,
decoding BCH codes of even modest lengths and error cor-
recting abilities is not easy from an equipment complexity
viewpoint. While many coding systems have been con-
structed which use BCH codes to detect errors, only a very
few systems have been designed for forward-acting error
correction.

One of these systems, designed at Bell Telephone

Laboratories, (13) used a (200, 175) BCH code with a mini-
mum distance of 8. The decoder corrected any patterns of
3 or fewer errors in a block of 200. If more than 3 errors
occurred and the error pattern was detectable (but not cor-
rectable) a retransmission was requested. The decoder
used a logic unit of 900 transistors operating at a speedup
factor of 25 times the input data rate and a 400-bit store.
This error control system was used with an experimental
high speed data terminal operating at 9600 bits per second
on voice telephone lines. Here the probability of error
without error control was 5 X 10- 3 • Observe that this
error rate is too large to rely upon detection-retransmis-
sion solely since on the average about every other block
contains an error. After error control the probability of
error at a data speed of 8400 bits per second was 5 X 10- 5 •
Here we have an example of an error correcting code used
to enable high speed, accurate data transmis sion.

The difficulties involved in decoding BCH codes (wit-

ness the 900-transistor logic unit of the previous ~xample)
have led to a search for efficient codes which are more
easily decoded. The usual procedure of first finding codes
and then determining decoding methods was reversed. A
very simple decoding alrorithm called majority-logic de-
coding was invented. {14 In majority-logic decoding there
is a particular set of parity check relations which applies
to the decoding of each bit. When there are errors in the
received data some of thes e equations will be satisfied and

182
some will not. A "vote" is taken to see if the majority are
satisfied. If so. the bit is accepted as correct; otherwise,
it is changed. Unfortunately. the families of codes which
have been found to be majority-logic decodable are sparsely
populated and generally less efficient than comparable
BCH codes.

Real Channels and Burst- Correcting Codes

The coding philosophy discus sed thus far is pre-

dicted on the assumption that the channel-induced errors
are independent events. If this were not the case the de-
coding rule of selecting the code word closest to the
received sequence would not be appropriate. To make
matters difficult there is considerable evidence that on
many channels. particularly the telephone and h. f. chan-
nels. errors are far from independent events.

For independent errors. if the probability of an

individual bit being in error is P. then the probability of
m errors occurring in a block of n bits is given by the
binomial distribution
F{m .n) = (;) p m{l _ p)n-m

In Figure 6 we have plotted an example case of P(m.3l) as

given by this equation and as actually measured over a
number of calls on the telephone network. (10) It can be
seen that the measured probability of a large number of
errors occurring in a given block is very much greater
than that predicted by the binomial distribution. In other
words errors on the real channel tend to occur in clumps
or bursts rather than as independent events.

Information theory would tell us that the clumping

together of errors should be to our benefit. This should
imply a predictability about errors which is not present
when they are independent events. In actuality. however.
these bursts are quite difficult to handle; much more so
than independent random errors. Uncorrectable error
patterns occur on the burst channel frequently when coding

183
techniques designed for random error correction are used.

In theory a code and decoding procedure could be

specified which corrected those error patterns which oc-
curred most often on the real channel. Developing such a
code appears to be an extremely difficult problem about
which little is known. However, a great deal of work has
been done on what are known as burst-correcting codes.
In coding theory an error burst of length k is defined as
a sequence of k bits in which the first and last bits (at
least) are in error. It can be shown that a parity check
block code cannot correct all bursts of length greater than
half the number of check bits. Many codes are known which
have a burst correcting ability close to this limit. Burst
correction is surprisingly easy to implement and in addition
most codes have the power to correct a large majority of
bursts greater than their guaranteed ability, up to the num-
ber of check bits.

The simplest method of constructing a very long

code capable of correcting large bursts is by the inter-
leaving of a shorter code. Interleaving is a method in
which the data are rearranged before transmission and
then reordered at the receiver. In this way error bursts
are broken up and made to appear as more is olated events.
In Figure 7 we have shown how the (7,4) code can be inter-
leaved to degree 5 in order to provide a burst-correcting
ability of 5. Five consecutive words of the code are stored.
Then the 35 bits involved are read out into the channel in
the order shown. A burst of length 5 or less, an example
of which has been indicated. will cause no more than one
error in each of the five code words. Since the (7,4) code
is single-error-correcting» all of the burst errors are
corrected.

As a more realistic example of interleaving we

might mention the use of a (15,9) code interleaved to degree
200 and tested{i5) on the voice telephone network. The
(15,9) code is a perfect burst-correcting code which cor-
rects bursts of 3 or less. Thus the interleaved code cor-
rects bursts of 600 or less in a block of 3000 bits. This

184
systeITl was iITlpleITlented with 210 logic transistors and a
2800-bit store. In tests it reduced the error rate of a 2000-
bit per second data systeITl froITl 1.2 X 10- 5 to 8.5 X 10 -7 •
The latter figure is sOITlewhat deceptive since decoding
errors. when they do occur, tend to occur in large bursts.
Actually the ITlean tiITle between these error bursts at the
decoder output was about 12 hours.

Convolutional Codes and Sequential Decoding

Another iITlportant class of codes is cOITlposed of

convolutional codes. Convolutional codes differ froITl block
codes in that the check bits within a given block check not
only the inforITlation bits in that block, but also inforITlation
bits in previous blocks. Typically the basic block length
itself is quite sITlall. For exaITlple. there is a rate 1/2
convolutional code in which each block consists of one
info rITlation and one check bit. The check bit in each block
is the ITlodulo - 2 s UITl (parity che ck) of the info rITlation bit
in the given block and the inforITlation bits in the second,
fifth and sixth preceding blocks. We can visualize the en-
coding being done by a sliding window as shown in Figure 8.
This particular code is said to have a constraint length of
14 since the window length, or ITleITlory, of the encoder
encoITlpasses 14 bits. Half of these 14 bits are inforITlation
bits, thus the c ode is called a (14.7) code. Its ITliniITluITl
distance is 5 and it is capable of correcting any two errors
which occur within a span of 14 bits.

Encoding convolutional codes is quite siITlple and

is accoITlplished with shift register circuits just as in the
case of block codes. However, decoding presents an inter-
esting probleITl because all the previous bits at any given
tiITle are tied up with the decoding of a particular inforITla-
tion bit. While it ITlight seeITl that only the previous 13 bits
were involved. these bits theITlselves depend on the decoding
of previous bits. etc. There are two basic decoding ITlethods
used, one of which we shall terITl algebraic and the other
sequential.

In algebraic decoding the decoder uses a 14-bit (for

185
our example code) sliding window. Each sub block of two
bits is corrected based on the information within the de-
coding window. With these bits corrected the window moves
two more bits to the right and corrects the next subblock,
etc. This method of decoding ignores information which
is outside the window at any given time. Also, because
previously corrected bits are used in the decoding of sub-
sequent bits. any errors made in the decoding process tend
to propagate, i. e. , to make subsequent decoding errors
more probable.

Like many of the more popular convolutional codes,

the (14,7) example code is one that can be decoded simply
using majority-logic circuitry. (16) Each bit is decoded
based on a "vote ll of a few simple parity check relations.
This democratic procedure is possible because the example
code was shown from a class of codes called "self-orthogo-
nal. II In these codes the parity check equations used in
decoding a particular bit do not involve any other bit more
than once. The "recipe" for constructing such codes in-
volves the determination of certain perfect difference sets
of integers.

Convolutional codes of this type have become quite

popular because of their simplicity. One company markets
a unit for telegraph speeds (75 bits per second) with a code
rate of 1/2 and constraint length of 22. This code corrects
bursts of up to 6 bits (a telegraph character) and any two
random errors within a constraint length. The size of the
unit is roughly that of a loaf of bread. although the use of
integrated circuitry would lessen this by a considerable
factor.

Information theoretic bounds can be established for

convolutional codes. just as with block codes. Although a
direct comparison is difficult, it does appear that better
asymptotic performance can be obtained with the convolu-
tional codes than with block codes. The performance of
shorter codes also tends to slightly favor convolutional
codes. However, with known codes and algebraic decoding,
arbitrarily small error probabilities cannot be obtained at

186
nonzero code rates. On the other hand. the next method of
decoding convolutional codes which we shall describe,
sequential decoding, enables the attainment of arbitrarily
small error rates at significant fractions of channel
capacity. (17)

In describing the sequential decoding algorithm,

the code itself is best characterized by the tree structure
shown in Figure 9. In this rate 1/2 code two channel bits
are emitted for each information bit. If the information bit
is a "1" we use the two channel bits in the upper branch of
the coding tree; for a "0" we use the lower branch emanating
from the given node. A given message traces a particular
path through the coding tree. (An example of such a path
is shown on the diagram.) It should be clear that this is
just another way of describing a convolutional code. The
structure of the tree will repeat itself after a constraint
length of the given code.

In sequential decoding the decoder attempts to trace

a path through a duplicate tree available at the receiver.
Because of errors the received bit sequence will not be
identical to any of the pos sible tree paths, but the receiver
tries to find that path which is closest (relative to some
appropriate metric) to the given bit sequence. While
attempting to do this the decoder occas ionall y makes mis-
takes and chooses an incorrect branch. Like the driver
who has made an incorrect turn the decoder begins to
notice that the new path seems unfamiliar. i. e. , the distance
between its present path and the received data sequence is
growing much faster than it should. In this case the decoder
backs up and changes the earlier decision to see if a better
path can be found. Thus the decoder moves in an erratic
fashion, sometimes moving along quite easily when the
channel introduces few errors and at other times doing con-
siderable backtracking in its confusion over the correct
path.

Because the amount of computation the decoder needs

to do varies with the level of the channel noise, this type of
decoding is frequently called probabalistic decoding. In

187
theory the decoder never really makes a decision at all.
since later it may have to change its mind about an earlier
decision. In practice of course the decoder must turn over
decisions to the user after it has proceeded well enough
past the given bit to feel reasonably sure about its decis ion.
While the bit is in limbo waiting to be released it must be
placed in a storage unit in the decoder known as a buffer.

In practice sequential decoders almost never make

a mistake. Instead they infrequently experience the cata-
strophic event of buffer overflow in which a bit required
for reexamination has already been pushed out of the buffer.
This happens when the channel noise is high over an appre-
ciable time interval. The decoder works frantically to keep
up with the received data, but finds itself losing ground as
the front end of the buffer is filled with new data bits it has
not yet had the chance of examining. Finally the buffer
cannot hold both the incoming bits and the bits needed for
its tree search.

Since the probability of buffer overflow is the deter-

mining parameter in the design of a sequential decoder, the
statistics of the computation process assume paramount
importance. The problem is quite difficult, but it has been
shown that the average computation per decoded bit is
bounded. independent of the constraint length of the code,
if and only if the code rate is less than a rate called R comp .(18)
This rate is related to the random coding bound in the man-
ner shown in Figure 2. For high signal-to-noise channels
Rcomp is close to capacity; for very noisy channels it is
approximately half of channel capacity. Obviously a sequen-
tial decoding system cannot operate above Rcomp.

Below Rcomp the distribution of computation is

quite interesting. For white, Gaussian channel noise the
probability of the decoder having to backtrack decreases
exponentially with distance back into the tree. However the
number of necessary computations increases exponentially
with this distance. Thus it is not surprising that it has
been found that the overall computation is a Paretian random
variable. (19) i. e •• the probability of having to do N calcu-

188
lations to decode a given bit decreases only as N-f3.
Therefore the buffer ITlUst be extremely long (or the arith-
metic unit very fast) to keep the probability of buffer over-
flow small. The sequential decoding method solves one
problem of exponential growth, but introduces another
event (buffer overflow) which cannot be made exponentially
smaller by additional complexity.

Although sequential decoding is theoretically the

most efficient method known, it is also an inherently expen-
sive technique to implement. Thus its use seems reserved
for those occasions when the cost of the terminal itself is
small compared with the cost of maintaining the communi-
cation medium. A deep space probe would be an example
of a situation in which sequential decoding would be eco-
nomically viable.

Only two sequential decoders have been constructed;

both at MIT's Lincoln Laboratories. An earlier model was
designed for telephone line transmission while a more
recent terminal employing se~uential decoding is intended
for satellite transmis sion. (20 In this latter system the
decoder ope rates at rates 1/2 or 1/4 corresponding to
information rates of 10,000 and 5,000 bits per second
respectively. The code has a constraint length of 60 infor-
mation bits and the buffer storage at the decoder is 35,000
bits. In addition to the sto rage unit the equivalent of abo ut
4,000 transistors are required for the various logic opera-
tions. The system is designed to operate near 0.9 Rcomp
at which point the input error rate would be in the range
0.1 - 0.2. (Notice that a sequential decoding system
operates at a much higher input error rate than the typical
algebraic decoding system.) The output error rate is deter-
mined chiefly by the probability of buffer overflow, which
would typically be about 10- 6 • The decoder is resynchro-
nized periodically so that the effect of buffer overflow is
restricted to causing a burst of errors of about 0.2 seconds
duration.

189
 Conclusion

The challenge that information theory poses to the

communication system designer is to reduce the probability
of error to an arbitrarily small value at a rate which is a
reasonably large fraction of capacity without using an
exponentially-growing amount of complexity. We have seen
that sequential decoding is the only known method of ac-
complishing this. but only at rates smaller than Rcomp'

Between Rcomp and channel capacity there is

recent work» particularly in the U. S. S. R. , which indicates
that the concatenation of se~uential decoding and algebraic
decoding may be effective. ( 1) By "concatenation" we mean
that the output of one coder is in turn coded by another
coder. The second coder in such a system cannot be binary.
but must deal with digits representing entire blocks of the
first coder.

Such systems as sequential decoding and concatenated

coding seem impossibly complex by present equipment stand-
ards. Yet for certain applications. such as space probes.
satellite transmission, and undersea cables. the economic
factors are very much in favor of the more efficient com-
munication systems in spite of their inherent complexity.
The introduction of integrated circuitry makes these systems
even more promising for the future. We have all seen
electronic circuit sizes shrink by orders of magnitude in
the past few years. Now it is necessary to scale up our
design thinking by this same amount. The unthinkably com-
plex system of today may be commonplace in the near future.

Another facet of device technology which is affecting

coding and communication theory is the rapidly decreasing
cost of digital storage. We have seen that all coding systems
rely on the processing of large blocks of data at one time in
order to overcome the effects of noise. The cost of the
required digital storage is decreasing so fast that in the very
near future very long block length (or constraint length) codes
will be attractive. Of COurse this will mean a greater delay
in message reception- -a factor thus far almost totally

190
ignored by theo rists and syste:ms designers. This factor
:may beco:me i:mportant in future co:mputer-to-co:mputer
co:m:munication.

In the i:m:mediate future we can expect the s i:mpler

coding syste:ms, algebraic codes with forward-acting
correction and/or detection-retrans:mission, to gain in-
creasing acceptance. This acceptance should co:me about
both because of their decreasing cost and because of the
increased de:mands for accurate trans:mission as relatively-
infallible co:mputers phase out hu:mans in the co:m:munication
link. In theoretical work we should expect an increased
e:mphasis on the role of syste:m co:mplexity vis -~-vis per-
for:mance.

191
 REFERENCES

1. Shannon, C. E., "A Mathematical Theory of Communi-

cation," Bell System Technical Journal, Vol. 27,
July and October, 1948.
For more material on information theory the reader
is referred to: Ash, R. B., "Information Theory,"
Interscience Publishers, New York, 1965.
Fano, R. M., "Transmission of Information,"
MIT- Wiley, New York, 1961.

2. Shannon, C. E., "Probability of Error for Optimal

Codes in a Gauss ian Channel," Bell System Techni-
cal Journal, Vol. 38. 1959.

3. Gallager, R. G., "A Simple Derivation of the Coding

Theorem and Some Applications," IEEE Trans. on
Information Theory, January, 1965.

4. Slepian, D., "Bounds on Communication," Bell System

Technical Journal, Vol. 42, May, 1963.

5. Wiener, N., "Extrapolation, Interpolation and Smoothing

of Stationary Time Series," MIT Press, Cambridge,
Mass., 1949.

6. The following two books are good general references on

statistical decision theory in communication:
Helstrom. C. W., "Statistical Theory of Signal
Detection," Pergamon Press, New York, 1960.
Hancock, J. C. and Wintz, P. A., "Signal Detection
Theory," McGraw-Hill, New York, 1966.

7. Holsinger, J. L •• "Digital Communication over Fixed

Time-Continuous Channels with Memory- -with
Special Applications to Telephone Channels,"
Lincoln Laboratory Technical Report 366, Oct.
1964, Lexington, Mass.

192
 Tufts, D. W., "Nyquist's ProbleIl1--The Joint
OptiIl1ization of TransIl1itter and Receiver in Pulse
AIl1plitude Modulation," Proc. of the IEEE~
March, 1965.

8. ExaIl1ples of Adaptive Receivers are:

Price, R. and Green, p. E., "A COIl1Il1unication
Technique for Multipath Channels ,II Proc. IRE,
March, 1958.
Lucky, R. W., "Techniques for Adaptive Equiliza-
tion of Digital COIl1Il1unication SysteIl1s," Bell
SysteIl1 Technical Journal, February, 1966.

9. HaIl1Il1ing, R. W., "Error Detecting and Correcting

Codes," Bell SysteIl1 Technical Journal, Vol. 29,
1950.
Slepian, D •• "A Class of Binary Signalling Alpha-
bets," Bell SysteIl1 Technical Journal, Vol. 35, 1956.
As a general reference on coding and for additional
bibliography, see
Peterson, W. W., "Error-Correcting Codes,"
MIT- Wiley, New York, 1961.

10. Townsend, R. L. and Watts, R. N., "Effectiveness of

Error Control in Data COIl1Il1unication over the
Switched Telephone Network," Bell SysteIl1 Tech-
nical Journal, Vol. 42, 1963.

11. Many of the bounds of cOIl1Il1unication theory are given

in Wozencraft, J. M. and Jacobs. I. M., "Principles
of COIl1Il1unication Engineering," John Wiley, New
York,1965.

12. Bose, R. C. and Ray-Chaudhuri, D. K., "On a Class of

Error Correcting Binary Group Codes." Inf. and
Control,Vol. 3, 1960.
HocquengheIl1, A., "Codes Correcteurs d'erreurs,"
Chiffres, Vol. 2, 1959.

193
13. Burton, H. O. and Weldon, E. J., Jr •• "An Error
Control System for use with a High Speed Voiceband
Data Set." National Communication Conference.
Boulder, Colorado, June, 1965.

14. Massey, J. L •• "Threshold Decoding," MIT Press,

1963.
Hagelbarger, "Recurrent Codes for the Binary
Symmetric Channel," Lecture Notes from U. of
Mich. Conf. on Theory of Codes, 1962.

15. Weldon, E. J., Jr., "Performance of a Forward-

Acting Error- Control System on the Switched
Telephone Network," Bell System Technical
Journal, May-June, 1966.

16. Convolutional codes were first des cribed in Elias. P. ,

"Coding for Noisy Channels," IRE Conv. Record,
part 4, 1955.
The decoding procedure was due to Massey. £E. cit.

17. Wozencraft, J. M. and Reiffen. B., "Sequential

Decoding." MIT Press, 1961.
Fano. R. M. , "A Heuristic Discus sion of Proba-
balistic Decoding," IEEE Trans. on Information
Theory, Vol. IT-9. 1963.

18. Wozencraft,~. cit.

19. Savage, J. E., "Sequential Decoding--The Computa-

tion Problem," Bell System Technical Journal.
Vol. 45, 1966.

20. Lebow, J. L •• "Sequential Decoding for Efficient

Channel Utilization," National Communications
Con£.. Boulder. Colo., June, 1965.

21. Pinsker, M. S •• "On the Complexity of Decoding,"

Problems of Information Transmission, Vol. 1,1965.
Forney, G. D., "Concatenated Codes ," MIT--
R. L. E. Technical Report 440, 1965.

194
a. Message
101

b. Corresponding waveform
formed from 2 samples

c. Transmitted vector

d. Received vector

t Noise

e. Received Vector Detected as Nearest

Signal Vector

DIGITAL COMMUNICATION SYSTEM

Figure 1
195
E(R)

I ,
I
I ""
I

TaE RANDOM CODING BOUND

Figure 2

10

co
rl
ill
6
~
u

'""
C
-rl
rl
4
'"
ill
'd -----------
H
ill
>
0

"'<>: 2
~
<Il

25 50 75 100
M

SIGNAL-TO-NOISE RATIO REQUIRED ABOVE

THEORETICAL MINIMUM OF CAPACITY FORMULA
FOR A PROBABILITY OF ERROR OF 10- 6 AS A
FUNCTION OF BLOCK LENGTH M.

Figure 3

196
 101

Coded Message

o 1 ----> 10 11

1 ________ _

,,-£1)-
01-------------

00-

Transmi tted Waveform with

Quantized Samples

QUANTIZED SYSTEM

Figure 4

Signal-to-Noise Ratio - db.

10 20 30

Measured
Pr~b. of Error

Theoretical

BOTTOMING EFFECl' IN HIGH FBE'~UENCY RADIO CHANNELS

Figure 5

197
 I~
\\
~

"" \
P(m, 31)
Einom al
f\.
\
o 4 8 12 16 20 24 28
m

P(m,31) DISTRIBUTION FOR TELEPHONE CHANNEL (T)

COMPARED WITH BINOMIAL DISTRIBUTION WITH SAME
AVER~GE BIT ERROR RATE.

Figure 6

Five Blocks of the (7,4) code are

stored anj then bits are read out
in the order shown.

d1 d2 PI P2 P3
6 21 26 31

7 22 27 32

8 23 28 33

4 9 24 29 34

5 10 25 30 35

The Shaded Area Represents a Burst

of 5 Errors) No M8re than 1 Error
Occurs in each Word.

INTERLEAVING

Figure 7

198
 Infol"Jlation bIt

C Check bjt;

l' \'
I C I C

CONV;:-r;~J'I'TONAl
l' il
,
I

;~:'JCUDJNG
c

r!Jod-2
SUM

Figure 8

214
10

11
10

11

OJ

r
r----- . . Example Path,
Message 0111

0 r ---
11
10

I
L ___
00

00
11

CODE TREE

Figure 9

199
 SECTION V

GAME THEORY
 CHAPTER 1

GAME THEORY:
A NEW PARADIGM OF SOCIAL SCIENCE

OSKAR MORGENSTERN
Princeton University

ABSTRACT

Game Theory is essentially a mathematical disci-

pline which aims at explaining optimal strategic behavior
in social games as well as in economic and political situ-
ations. The theory is normative in that it gives advice;
it is descriptive in that it shows social and economic
phenomena as being strictly identical with suitable games
of strategy. Sharp, quantitative concepts are formed for
hitherto only vaguely known events. The role of chance
devices for rational behavior is established. Notions of
equilibria, different from those in the physical sciences.
are developed; these, called" standards of behavior. "
determine zones of uncertainty found to be inherent in
social processes. Game Theory emerges as a fundamen-
tally new approach to the social sciences. has stimulated
many mathematical researches, initiated experimental
work in new areas and illuminated large parts of the social
world. The Theory is in the stage of active development
carried on in many parts of the world.

203
 I. Bas ic Concepts *
One of the principal problems in science is the
development of proper concepts. This task assumes
particular significance in the social sciences where there
is a great wealth of phenomena that are difficult to describe,
to order and to analyze. Only proper, sharply defined con-
cepts can give exactness; data, no matter how good, never
can do this.

The difficulty of studying the social world is further

augmented by the fact that, partly due to its comparative
youth, partly because of the lack--until recently--of strictly
planned experiments, standards of observations and measure-
ments--comparable to those in astronomy and physics--are
very slow in evolving. As long as proper concepts are lack-
ing it is impossible to provide a suitable mathematical
structure for the social sciences and there can be no doubt
that the complex relations to be described demand intensive
mathematical analysis and ultimately axiomatic treatment.

Since the human, social world is quite obviously

very different from the physical world in which it is em-
bedded it is to be expected that the mathematical structures
of these two sciences will also differ from each other. It
is furthermore clear that in the first attempts to mathe-
matize, the social sciences have borrowed heavily from the
mathematics used earlier with spectacular success in
physics. In the same vein it is not surprising that this
effort has not been very successful. The reason is that
fields so widely different from each othe r require mathe-
matics that can cope with the specific characteristics of
their basic problems. As a consequence calculus, de-
veloped for mechanics, plays a negligible role in Game
Theory; instead there is use of combinatorial methods,
of set theory and other branches of mathematics until perhaps
someday a mathematical discipline will arise that is as

*The preparation of this paper was supported in part by

the Office of Naval Research thr ough the Econometric
Research Program at Princeton University.

204
typical for the social sciences as the great and powerful
tool of analysis.

The difference between the two fields is that in

games one is confronted with purposeful actions of the
players. Each is seeking to obtain his maximum advantage
while faced with opposition from other players. Sometimes
he is given the opportunity to cooperate with them part of
the way. until the opposition of interest re-emerges which
is when it comes to the division of the joint spoils in a
manner that allows us to speak of a certain kind of stability
in these arrangements. Clearly. nothing of this sort hap-
pens in any physical domain: particles. atoms. molecules.
planets and stars may collide t explode or coalesce. but to
the best of our knowledge they do not suffer conscious an-
tagonism, nor do they from time to time cooperate in order
to obtain joint goals. It is therefore strange that the idea
to model the social sciences after the physical world should
have prevailed so long, even though the fact that social
behavior cannot be described in purely physical terms
should have been obvious. So it happened that such theories
had to allow for all kinds of curious exceptions. special
cases and the like and that no general. rigorous theorems
have been found until 1928, when John von Neumann proved
the crucial minimax theorem of the differently oriented
Theory of Garnes. In spite of its profound importance and
wide applicability the theory went practically unnoticed
until in 1944 when a book. Theory of Garnes and Economic
Behavior appeared, in which Game Theory was extended in
scope and various applications were made. Since then a
great literature has sprung up and Game Theory is in a
state of active development in many parts of the world.

II. The Notion of Rational Behavior

As its name indicates Game Theory attempts to

explain games of strategy such as chess or poker. In these
the outcome, which must be sharply defined, depends on
the action of each player and each is affected by every other
player. No individual alone controls his own fate nor that
of the community, nor does it depend, as in games of chance,

205
on the arbitrary. seemingly indifferent behavior of nature.
Consequently no statistical assumptions can be made re-
garding lithe others II as one can in the case of nature. nor
can such assumptions guide the participant who is bent on
obtaining a maximum advantage. One cannot play chess
successfully by attributing statistical ideas to the other
player.

The question is whether there is an optimal behavior

for each participant. whether the players discover and adopt
it. and whether the theory can advise them how to behave
optimally. Rational behavior is easy to describe when the
individual faces fixed conditions. including those that are
only fixed probabilistically. as e.g. the weather: maximize
your return in terms of utility or profit. both given numeri-
cally. But the idea of rational behavior in the other. normal
situation is elusive. We ask: where are the social and
economic situations in which the individual. isolated from
all such others. faces rigidly given fixed conditions and can
do nothing but passively adapt himself to them? In that case.
and in that alone. would he control all the variables on which
the outcome of his actions hinges. Yet this is the-false-
conceptual basis on which all of modern economics rests:
there is nothing to solve but a variety of maximum problems.
big and small. with or without side conditions. And all this
is supposedly valid uniformly for the single individual or
for society as a whole.

Reality is quite different: while the individual can

fix the value of some variables. other individuals. with
different interests fix those of other variables. those they
control. but on all of them together depends the result. To
explain the results of these interactions and to give mathe-
matically satisfactory explanation of optimality is no mean
task but this is what Game Theory was designed to do.

Man is what Aristotle said he was: a "zoon politicon";

Donne's "island" notion of man may refer to his soul, but
definitely is not true with respect to his social condition.

Here it is necessary to avoid a frequent misunder-

206
standing: Game Theory does not assume rational behavior
but wishes to explain it. to give it precise mathematical
meaning. How could it assume something that has neither
intuitive meaning nor is otherwise established! Game
Theory merely assumes that each individual strives for
his greatest advantage and will follow- -within the unalterable
rules of the game- -every lead that he believes will secure
it for him, no matter how small the advantage may be. We
can safely claim that outside of Game Theory general talk
about some behavior as being "rational" or not in situations
that can be and must be modeled by games of strategy has
no meaning within the constraints set by scientific standards.

Once it is realized that the explanation of rational

behavior under conditions of less than complete control of
the variables is far from trivial, a veritable shift of point
of view has occurred. We now look at seemingly well-known
phenomena in a new mood- -as has often happened in the
history of science. Instead of developing further the same
method, applying it still more, in greater strength. but
without any hope for fundamentally new insights a shift of
position occurs. The same evidence before one's eyes
yields different information. It is as in those puzzle pictures
we use to entertain children with: "Where is the man in the
tree?" Suddenly you see him and the whole perception is
changed, a new vision is there.

With Game Theory it is similar: the essential fact

is that decisions have to be made (under the generally
accepted ideas of wanting to obtain a maximum of utility.
advantage or profit, and moved by the smallest differential
advantage within reach) but that one does not alone control
the outcome. This has to be reiterated because once an-
nounced it is immediately obvious; yet this view demands
a fundamental, and perhaps painful departure from conven-
tional approaches.

The consequences of this shift reach very far. Not

only is the logical-mathematical structure of any theory
built upon the recognition of this situation widely different
from the current one, but other factors follow wh ich have

207
the addition of another player to a given number possibly
transforming the game in a substantial manner. E. g. ,
when n> 2, coalitions among players are possible and
will happen if their members derive advantages from com-
bining. There are games where the sum of gains and
los ses add up to zero; and there are those where that sum
may be positive or negative. constant or variable. Non-
zero-sum games are models for social and economic
relations since they may show productivity or loss. These
games are an extension of zero-sum games to which they
can be reduced by introducing a (fictitious) player who
makes contributions to. or withdrawals from. the game.
as the case may be. a fact that unifies the theory for all
pos sible cas es.

Game Theory is descriptive. It certainly is designed

to explain the occurrences in a game of chess, or poker and
to determine where. in the view of the players. and accord-
ing to the inviolate rules of the game, the stable solution-
outcome rests.

Game Theory is also normative in that it aims at

advising a player how to behave optimally in the face of
opponents and/or collaborator s. Since he cannot s imply
maximize the question of choosing the best strategy arises.
A strategy is a complete plan of action. considering all
contingencies arising from nature and especially from the
behavior of "the others." There are games where the
player can identify and safely pick--and even announce--
a particular "pure" strategy he is going to play. that is
better than any other, and where the opponent can do like-
wise. If the one player deviates from his optimal pure
strategy the other can only profit from such deviation.
These "strictly determined" games are a rarity. As a rule
a player will have to hide his choice. By announcing it he
would induce his opponent to pick a strategy which would
reduce him to the minimum he could ever obtain from that
strategy. The same holds for the opponent. Strictly deter-
mined games have a saddle point in pure strategies in the
payoff matrix; i. e. in the configuration of all payoffs re-
sulting from respective choices of the players, there is

208
one which is both the maximum of the minima which the first
player can ever obtain and also the minimum of the maxima
the second player may ever have to pay. Not strictly deter-
mined games have no saddle point in pure strategies and
there is doubt what optimal play should consist of in such
cases. Von Neumann proved in his famous minimax theorem
that if, however, each player, instead of choosing a pure
strategy, uses his strategies with certain specifiable proba-
bilities, then even these games have saddle points. Thus
there is always a safe way of playing and the correct proba-
bilities can be computed. This is the advice to be given to
the player.

We shall pause here to consider the situation. It

was stated earlier that a player could not treat the actions
of his opponent statistically. Yet now it is claimed that he
should choose his ~ strategies with probabilities only!
There is no contradiction, however. He is not making
statistical or probabilistic assumptions about the other
player. The reason for such probabilistically "mixed"
strategies is that only thereby can he assure himself of his
greatest advantage which critically depends upon not reveal-
ing information about his own procedure of play. The
safest way not to pass on information is not to have any.
Consequently if a random mechanism chooses my strategy
over a specified set of feasible strategies, then I cannot
inadvertently disclose this choice to my opponent. My own
strategy is then to determine the probabilities.

Consider matching pennies. On matching player 1

obtains one unit from player 2; on not matching he pays him
one unit. To put the coin down openly first is a sure way
of losing. Thus in practice the players toss the coins simul-
taneously as the best way of playing. The chosen probabilities
for the only two strategies--heads or tails--are therefore
1/2, 1/2 and the fair coin itself is a proper random mecha-
nism for obtaining in a simple way the desired probabilities.
It is obvious, that if on matching heads a premium is paid
over matching tails, the 1/2, 1/2 probabilities for the two
strategies will no longer work and as a consequence the
coin, which can represent no other probabilities than these,
is unsuited to produce the correct probabilities. Another

209
chance device will have to be used.

Matching pennies. frequently believed to be a game

of chance. is in fact a prototype of the simplest forms of
games of strategy. clearly exhibiting the role of mixed or
statistical strategies for optimal play. On that foundation.
given by the minimax theorem, the entire theory can be
built.

An important development has sprung up related to

Game Theory which must be seen in proper perspective:
linear programming and its derivatives. Here the task is
to maximize a function under the constraint of linear in-
equalities. A famous theorem, due to Gale-Kuhn-Tucker.
has proved the duality of such programs with zero- sum
two person games. This is a matherm. tical property of as
much interest as importance, but it is primarily a mathe-
matical matter. In other words. linear programming is
not application of Game Theory when the latter is viewed
as a model for social behavior. Linear programming is
concerned with straight maximization, precisely what is
not occ.urring in 2 person Game Theory. Linear program-
ming applies therefore only to cases where there is control
over all variables e. go within an organization, firm. etc.
It deals with allocation of resources in short supply accord-
ing to a specified goal and determines the most efficient
procedure. In other words linear programming is (because
of the consideration of inequalities and of discrete rather
than continuous functions. etc.) a substantial and much more
realistic revision of conventional, rna rginal economics but
it is still hemmed in by the idea of complete control (even
when it turns stochastic or dynamic). Where it is truly
applicable it is at least not in conflict with Game Theory
but it just goes as far as the requirement of control of
variables is met and it was shown above where the critical
point lies. I add. that these considerations are concerned
with the logical situation primarily not with the frequency
of empirical occurrence of one case or another.

210
an immediate, intuitive appeal, A maximum problem is
deterministic; even when probabilities are introduced, they
are fixed, But the uncertainties associated with the actions
of others who control vital variables in their own interest
are of a different order, and as said before, they cannot be
treated in a conventionally probabilistic manner Yet we
0

shall presently see that probabilities enter into the picture

again- -however in a significantly novel manner, Similarly
the notion of a solution necessarily assumes entirely new
aspects, In the maximum case there is no logical probl~m:
the maximum is the solution, It is unique, It is completely
described by a number or by a single set of numbers, The
idea that a problem is only solved when a single unique
number or set of numbers is found is deeply ingrained upon
much of modern science although even in physics through
the appearance of quantum mechanic s a break has been made
with this belief. It was not easy for physicists to accept the
new views but one clearly could not close one's eyes to
reality, unhappy as one might be 0 Thus it is not surpris ing
that social scientists look upon the new concept of a solution
of an n-person game which models economic and political
phenomena- -to be described below- -with discomfort and
suspicion,

Neither is it surprising that soc ial scientists are

inclined to take it as a deficiency of Game Theory that often
alternatives are given as solutions or only probabilities for
the choice of actions. Quite on the contrary these so-called
"indeterminacies." far from being a weaknes s of Game
Theory to be overcome, express a fundamental property of
social phenomena, i, e. of organizations. of the formation
of prices and incomes, as well as the division of the spoils.
It would take radically new ideas and quite different as-
sumptions or characterizations of human behavior in order to
arrive--perhaps !--at the umqueness that has for so long
been considered the criterion of a satisfactory answer,

III. Probabilistic Strategy Choices

Turning now to games proper we observe the need

for classification: there are games of 1, 2, •• n-persons,
0

211
 IV. Randomization and Reality

Since Game Theory demands randomization the

question arises whether man can randomize properly in all
critical situation. He does not carry urns around in which
he always has the right admixtures of balls. Can he intui- t
tively make the correct choices? Emile Borel has asserted
and--I think--believes to have proved mathematically that
man is congenitally unable to imitate chance. There is no
strict mathematical proof of this assertion possible since
it is not a mathematical proposition. A random sequence
has to be infinite in order to be truly random and man can-
not make an infinite sequence. He can only specify a
generating function or mechanism. But there clearly is a
difficulty. It is not the objective complication of equipping
oneself with physically suitable random devices. Man
certainly can construct them, which in a certain sense takes
care of Borel's--alleged--difficulty. Rather it is man's
apparent, instinctive desire for order and certainty that
runs counter to the idea that randomness in his behavior
should be preferable to the clear. concise and definite
choice of a course of action.

It was noted above how difficult it has been even in

the natural sciences to make room for the uncertainty
associated with quantum mechanics. Einstein's aversion
against this property of quantum mechanics is well known.
The craving for determinateness is deeply imbedded in the
human mind. It will take a long time for social scientists
to adjust to the new situation which is that the advice they
have to tender government and businessmen for optimal
behavior involves the indication of "mere" probabilities to
be chosen for action. Add to this the fact that there exist
no precise, simple, determinate equilibria as solutions
for those social situations as are modeled by games and
you have the picture of a truly difficult situation.

t E . Borel: Sur 1 'imitation du hasard, Compte-Rendus de

l'Academie des Sciences, Paris, Jan. 25, 1937, pp. 203-
205. lowe this reference to Camilo Dagum.

212
 If Game Theory is descriptive then we would expect
to find randomization in actual, individual economic behavior
since it seems reasonable to assume that individuals have
learned to adapt themselves to the underlying game situations
and have discovered that mixed strategies are to be used to
their advantage. They may not have arrived at the precisely
computed probabilities which the complete theory yields, but
they would at least have dis covered the principle. In answer
we note that matching pennies is played according to the
theory and it is a prototype for large classes of games.

It is, of course, odd, in a certain sense, that it

should be "rational" to introduce chance and chance devices
into one's behavior when one has the possibility of selecting
a pure strategy, i. e. a definite course of action. When
playing roulette, probability has to be accepted. Only
cranks would try to overcome chance by a "system," i.e.,
by some hidden regularity. And now here is Game Theory
and does the oppos ite: chance is being introduced where it
does not a priori exist!

In business one of the most outstanding features is

the attempt to hide information. to create uncertainty in the
mind of competitors about one's prices. costs, inventories,
reserves, etc. In military affairs the hiding of information,
the desire to achieve surprise, are as old as armed con-
flict. However. secrecy is only one way in which a game
theoretical element manifests itself in actual behavior.
Another one is not to choose except at the last moment and
with a strong component of uncertainty and uneasiness. Such
dark, purely subjective feelings can, of course, be made
clear. can be illuminated by the proper use of the theory
which will pinpoint the precisely computed probabilities
with which alternative courses of action should be chosen.
And. indeed, there may be an infinity of strategies to choose
from and no one can ever determine and find a point in a con-
tinuum. More is needed and probability of choice is the only
answer.

It is an interesting and important task to study actual

business situations, typical decisions made by boards of

213
directors, in order to discover to what extent randomization
has taken place. This is not easy to do. The critical phases
may be submerged in the subconscious--especially if Borel
is right in seeing difficulties which he clearly takes to pre-
sent themselves in the conscious domain. This is part of a
wider task for descriptive work to be done: social reality
described so far is known to us by the use of the old concepts
discussed at the beginning of this paper. But they differ
radically from those of Game Theory. Consequently also
the account of what really goes on according to Game Theory
in social and economic life differs in significant part from
what is currently taken to be important.

An interesting is sue is the following: If Game Theory

is absolutely convincing (in a normative way. giving advice
as to how to behave optimally) and if at present only a few
act according to the theory, but as the latter becomes known
will be widely applied, certain changes in the structure of
the market and the society will occur. It will become more
"rational" since its components will act more rationally in
a now well defined sense. The questions whether this pro-
duces greater stability for society as a whole, or more
randomness, and whether there would be greater social
justice or equity are intriguing. It would lead too far to
discuss them here; they are real and important. In another
context, these que stions are of interest from the point of
view of philosophy of science, since the question is raised
again as to what extent the knowledge of a science, or of
parts of it (among the individuals making up that society),
is a prerequisite for further development stages of that
same science and what the consequences are of different
degrees of such knowledge upon the processes to be
described. 1

11 have treated this problem a long time ago in O. Morgen-

stern: Logistik und Sozialwissenschaften, Zeitschrift
fur NationalOokonomie, Vol. VII. 1936, pp. 1-24.

214
 v. Rules of the Game

The concept of a game determines the rules of the

game. A closer look at these is indicated. They are
assumed to be inviolate, as, indeed, when playing chess
there is strict observance by both players of the way the
figures can be moved, when moves must occur, which
figure eliminates which under what circumstances 9 etc.
The same is, mutatis mutandis, neces sarily true of any
other parlor game. When the socio-economic world is
modeled by suitable games the question arises how well
the rules are established and how well they are known.
These are possibly two separate matters. First of all,
it is important not to identify the behavior of players as
expressing "rules," no matter how standardized or con-
ventional the behavior may have become. The rules in the
social world are the physical. legal and moral constraints
within which the actions of the players ~ take place» as
far as the physical possibilities are concerned; and should
tak.e place, as far as the legal-moral laws prescribe. Now
these rules may be imperfectly known- -that applies now
certainly to the laws of the physical world and they may
never become fully known. Or that knowledge may be
differently distributed among the various players. But
known or not, the physical laws make themselves felt at
any rate. The other, social laws may also be known in
various degrees only, and they can be broken--even if
known- -in varying degree and manner. 1 This introduces
an element of uncertainty that the theory of games has 9 as
yet. not much explored but which deserves analysis.
Clearly, the situation is as described, no matter what
analytical approach is made of the social world. Conven-
tional theory has completely neglected this issue, confined,
as it is, to a largely deterministic attitude and procedure.
It is not clear how those theories could be modified to take
account of these phenomena. But this is hardly interesting

1Cf • my paper: "L 1Attitude de la Nature et Ie Comportement

Rationnel" in E. M. Claas sen (Ed.) Les Fondements
Philosophiques des Systemes Economiques. Textes de
J. Rueff. Paris 1967, pp. 131-141.

215
since their basic assumptions are at any rate and for other
reasons in more profound conflict with reality.

The imperfect knowledge and poss ible, partial, un-

intentional and sometimes deliberate violation of the rules
of the game introduces an uncertainty which must reflect
in the payoffs to the players. Since there are also other
reasons why payoffs may be known only stochastically we
face a difficult. open problem. The payoffs may be affected
because of lack of information, of random variations in
nature, etc. which account for the presence of stochastic
elements. Here, however, the uncertainty rests in addition,
in the improper choice of strategies (i. e., in violation of a
moral or legal restraint). A strategy may be chosen by
one player which the other does not contemplate because he
will abide by legal and moral law.

Game Theory correctly eliminates these matters

at first, since it deals with enormous and novel difficulties
even under very rigorous assumptions. It is always easy'
to foul up a theory or a problem by introducing additional
questions, assumptions. etc. in order to make them more
"realistic." The great art is to know where to draw the
line. Among others Kant knew this very well.

VI. The Solution Concept

Characterizing more specifically what is meant by

a game-theoretical solution we turn once more to the 2
person zero-sum situation. There the opposition of interest
is absolute once the play starts. The payoff matrix dis-
closes a number for the intersection of the respective strate-
gies chosen by the two players. It can refer to utilities or
to payments, say in money. Game Theory has, as a side
product, produced a revolution in utility theory which in-
volves the discarding of the conventional clumsy indifference
curve method. The analysis of the two person zero-sum
game» with a view of finding optimal strategies for each
player and thereby assuring some kind of stability of be-
havior, runs along lines completely different from the way
it might be expected. There is no "I think he thinks that I

216
think ••• " variety of argumentation still often encountered
but which leads nowhere. Instead t the minimax principle t
being abs olutel y convincing. is the only one to offe r the
way out. It gives advice to each player since it first assures
him his optimal result regardles s of the actions of the other
and lets him only profit should the latter deviate from his
optimal strategy, and second it defines the value of the
game, thereby imparting full information about the objective
possibilities residing in the situation.

We saw earlier that the minimax principle holds for

all games provided properly mixed strategies are used.
Optimal strategies may be very difficult to compute--even
for pure strategies. For chess t for example, they cannot
be calculated because there are too many pure strategies t
even given the capacity of the most modern computers,
though it is known that ches s has a saddle point in pure
strategie s. The fundamental theorem is an existence
theorem. But this is the nature of things: constructive
methods follow later and for many games they have been
found.

A solution for an n-person game has a different

structure: The task is, as for the two person game ~ to
find the optimal strategies for each player and to determine
the distribution of proceeds among all players, some of
which- -by forming coalitions- -may have cooperated with
each other in order to win. The point is that infinitely
many ways of dividing the proceeds may be possible t but
only some will prove to be acceptable to the players. Each
possible division is called an "imputation. " Among these a
"domination" relationship is defined such that for a domi-
nating imputation a group of players can be identified which
is capable of enforcing that imputation and for which it is
individually better than the imputation it dominates. So
far this would not be too different from any maximum situ-
ation: there all others are dominated by the sole maximum.
However, a "solution" of an essential n-pers on game con-
tains never only ~ imputation but several, and among
them the domination relationship fails. The several impu-
tations stand in delicate relationship to each other: each

2.17
one, taken singly, is dominated by an imputation not in the
solution set, but this dominating one is in turn dominated
by another one belonging to the solution set. The imputations
belonging to the solution set are each separately ringing for
acceptance by the players, but once one of them is chosen
the others protect it by their "virtual" existence. They
offer acceptable alternatives to the players. The solution
set represents a "standard of behavior." All this is des-
cribed by the so-called characteristic function of the game,
which is a numerical set function derived from the applica-
tion of the minimax theorem to the n-person case.

Finding solutions is obviously very difficult. Many

games have several. some infinitely many; each is com-
posed of several, sometimes infinitely many imputations.
As yet there is no general proof of the existence of a solu-
tion for any game of arbitrarily large n; but in each specific
case investigated one has been found. I would not be sur-
prised--largely on the basis of some very recent work--if
the general theorem were proved in the near future. It will
be an event of prime importance. Other solution concepts
have been suggested, mostly springing from consideration
of special cases but they are all related to the initial one
described here which occupies a central position and fully
meets the logical requirements to be imposed.

The fact that the trans ition from 2 to 3 to 4

players each time introduces novel features requiring
careful analysis. with no guarantee that under all circum-
stances the increase in the numbers will bring about
asymptotic behavior of the solutions is methodically very
important. In economics such transitions to large numbers,
from "micro" to "macro" theories are usually made without
much ado; but aggregation is a serious matter. In physics
we have learned that theories valid for certain magnitudes
cannot be extended to very different magnitudes. The dif~
ferences between an individual firm and an entire economy,
however organized, are enormous and it is a priori doubtful
that the same principles apply equally to both. Game Theory,
forced to meet this issue head on, has begun to reveal there-
by manifold structural phenomena of social and economic

218
organizations. In particular the troublesome old problem
of complementarity of economic goods is now adequately
taken care of by the superadditivity property of the charac-
teristic function mentioned above.

VII.

Between the development of a theory and its applica-

tion lies often a span of several scientific generations. The
question of applicability is, however, nevertheless, quickly
raised, especially in our impatient age. It is dr6le, though.
when it comes largely from those whose early theories,
now being superseded by game theory, have not been con-
spicuous of widespread application. But the demand is
legitimate and raises interesting issues: Game Theory is
only applicable so far to simple cases of economic, political
and military matters. This is to be expected since the
theory starts with analyzing simple cases in which the full
complexity of the logical situation is revealed. To consider
a lack of immediate, wide-flung application as a deficiency
creates an interesting methodological situation: Assuming
that Game Theory correctly describes and treats the given
problems while other approaches fail to deal properly with
the underlying logical structure, how could the problems
get solved which are not (yet) accessible to Game Theory?
By what mathematical or logical procedure? How are solu-
tions even defined. let alone computed? Certainly false
theories cannot be used with confidence though they some-
times give correct answers as is later discovered when
more complete theories are available. Astronomy offers
some striking examples: Ptolemaian theory allowed the
prediction of eclipses of the sun and other events. That
theory breaks down for more complicated problems and is
at any rate fundamentally false. A similar situation pre-
vails as far as current economic theory is concerned: it
can predict certain phenomena but the theory is shown to
rest on a false logical basis as demonstrated by its identify-
ing the economic problem, caused by the interaction of men,
as one of finding mere maxima or minima.

The solutions to be found in applications involve

219
actions and interactions of persons and their consequences.
If these cannot be discovered by a scientifically acceptable
procedure- - i. e. Game Theory- -then there are only un-
scientific methods and this will not do. There is no knowl-
edge outside scientific knowledge. There is "art," "intui-
tion," "faith" and other- -so far- -unanalyzed mental activity;
all may from time to time prove successful. The normative
element of the theory comes through, however: advice is
to be given how to behave optimally. The mere fact that
individuals, corporations and governments take action in
situations for which there is as yet no complete scientific
understanding and for which science can give no correct
advice, does not mean that these men and agencies some-
how, mysteriously, have discovered the secret. They have
learned from experience, of course, but men cannot com-
municate experience easily whether from one generation to
another or to an outside observer. We each live but are
unable to say how we do it, except in the grossest terms.
It takes biology and medical science to tell. And govern-
ments are very hard to describe; the fact that interactions
among individuals take place makes communication even
more difficult. So there is no way out: either we have
scientific knowledge or we don ft. If a theory carries us
only up to a certain point, that is precisely as far as our
knowledge will reach.

There are, of course, results of Game Theory;

indeed they are numerous and widespread. Some are pre-
cisely as expected: They confirm what also common sense
would tell us. If there were a conflict we would lose confi-
dence in the theory. E. g., it is easy to see that the price
of an object will be greater or equal to what the seller wants
to s ell it for J and smaller or equal to the maximum bid the
buyer can afford. The theory confirms this and other simi-
lar cases. But soon the divergences set in and then we
trust the rigorous, mathematical theory rather than our
instinct or common belief. Most of these results- -though
discovered mathematically--can be restated in ordinary
language. At one point, however J the results will no longer
allow translation from mathematics into ordinary language
and can only be stated in mathematical form. This point

220
too is reached and more and more results fall into this
category. In the present paper I can, unfortunately, give
no discussion of these cases. Between these fall numerous
others, all of interest and some surprising in nature. The
following is a partial list, the order in which the items are
mentioned is of no consequence.

I emphasize that the statements to be made below

summarize in each case mathematical theorems whose
proofs are completely established. They are due to a
number of authors. Some proofs are lengthy and difficult;
all exhibit the mathematical structure of Game Theory,
i.e •• they use essentially combinatorial, set theoretical,
topological methods; as mentioned earlier calculus plays a
negligible role, if any.

(1) If among two traders bargaining with each other

the first has a finer utility scale than the second.
the advantages of the trade will go to the first.

(2) In certain bargaining situations it may be advan-

tageous to disclose one I s own utility function and
in others not. It is possible to distinguish these
two classes.

(3) There exist 6 person zero-sum games (and pre-

sumably others with more players) where a player
can only be beaten by members of his own "class."
indicating social stratification.

(4) The re are two kinds of dis crimination by some

players against others: discrimination may be
either due to the behavior of the players. 1. e ••
some showing a "tabu" against cooperating with
others. the tabu based on arbitrary grounds; or
it may be in the characteristic function of the
game, i. e. due to the objective situation desc ribed
by the characteristic function.

(5) Discrimination, while unfavorable to the player

who is being discriminated against, need not be

221
 as disadvantageous for a player as being treated
an equal by the other players, but be ing pushed
by them into a completely losing position.

(6) Complete symmetry among players does not

necessarily produce symmetric solutions. Thus,
if symmetric arrangements are thought to be
"just," in whatever sense, such belief is not in
general warl'anted. This obviously has implica-
tions for the "laissez faire" principle.

(7) Perfect substitutability among players (in a

winning coalition) does not necessarily entail the
assignment of equal values for these individual
players. These can be viewed as representing
factors of production. (Current economic theory
would assign to all factors that are substitutable
equal values. This is one of the cornerstones of
contemporary economic theory.) De facto in
some cases no values at all can be assigned;
what values these players have they only derive
from their relation to other players. These
relations can be of a very complicated nature.

(8) If a player is assigned a privileged position by

the rules of the game provided he is among the
winners, he may, in order to be considered at
all for inclusion in a coalition, have to give up
his privilege. This he does not by breaking the
rules of the game but by offering compensatory
payments to the other coalition partners which
will effectively annihilate his privilege. Related
to this result is the observation that democracy
will only work when "sufficiently large" numbers
of players are involved.

(9) It is possible for an outside source to offer con-

tributions to the proceeds of a game. They will
be paid to coalitions forming within it. If these
contributions exceed certain precisely determined
limits it will no longer be possible to form coali-

222
 tions strong enough to exhaust the contributions
offered. In that case the social organization of
that game. as represented by alternative coali-
tions, will break down completely. (This may be
taken to be a mathematical description of the
corrupting influence certain forms or amounts
of foreign aid have on underdeveloped. primitive
societies. Similar events may be observed also
for parts of highly organized societies.)

(10) Bluffing in games where this is permissible with-

in the rules of the game is a defensive, not an
offensive component in a strategy. Bluffs, in
orderto be effective must from time to time,
during the course of a play, be discovered as
such by the other players. (These findings run
counter to common expectations but they are
firmly established by the rigorous, mathematical
theory.)

(11) There exist coalitions of different strength. In

some games an inner core of a coalition forms
first which admits new members under less
favorable conditions they obtain for themselves.
All these can be quantitatively formulated. It
is pos sible to give precis ion to the notion of the
"power" of a coalition.

(12) The weight of a voter, say in a committee or in

the various bodies of the legislature, as des-
cribed by rules or the constitution may differ
significantly from his "influence" when it comes
to actual decision making, or the passage of a
bill. Clearly the originators of a proposed bill
have a different--lesser--influence upon its ulti-
mate passage than the last committee members
who have to be persuaded to throw their weight
in favor of the passage. Again all this is subject
to quantification and proof.

These illustrations exhibit some of the range and

223
scope of game theory beyond those commonly given. They
are but a small segment of what has been discovered. I
use this word intentionally: few of thes e res ults could have
been anticipated. Prior to Game Theory none could have
been proved mathematically or otherwise; most of them
are the consequence of investigating the mathematical
properties and implications of comparatively few and
intuitively transparent assumptions. Let us also recall
that without exception there is behind all of them one uni-
fying theorem, that of the minimax. This is a remarkable
property of the theory, probably of considerable interest
also to philosophers of science.

Specific economic applications exist in increasing

numbers: to duopoly, oligopoly, the theory of the firm.
the theory of location of industry. inflation. etc. Doubtless
the applications will increase in scope and number. But
conventional economics is well entrenched. It took marginal
analysis many decades to overcome labor value theory.
Game Theory being mathematically so different from mathe-
matical marginal theory not only threatens the logical,
conceptual make-up of present day economics; it also changes
the tools which are to be used. The challenge therefore
reaches far. Political and sociological theory being much
less developed presents in some way less of an obstacle. On
the other hand scholars in those fields have hardly any
acquaintance. let alone tradition with rigorous mathematical
analysis so that for them the mathematization is in itself a
novum. One recalls how long it took even for some rudi-
mentary mathematics to take root in economics. Many
processes have been speeded up in our time but an intrinsic
difficulty remains.

Even so: looking at current literature in the social

sciences there is now an abundance of problems newly seen:
there is a whole field of "decision making." of deciding under
uncertainty, of the role of threats, of their costs and credi-
bility, of the role and cost of information. and so on. One
would look in vain at publications 30 or more years old in
which these terms occur as frequently as now. There is an
awarenes s that these things exist and that they must be

224
captured by precise concepts. Of course, the mere use of
a new terminology is not equivalent to application of the
theory. The frequent occurrence in many wr itings of the
word "strategy" is not by itself an indication of a proper
use of Game Theory.

VIII. Outlook

I have referred at the beginning of this paper to the

comparative lack of experiments in the social sciences and
the ensuing difficulty of developing adequate scientific
standards as in physics regarding the evaluation of evidence.
Under the influence of Game Theory a change is occurring
spurred also by the almost simultaneous development of
computers. Simulation of reality, carried out with their
aid. though often unsupported by theory, can be a very im-
portant step toward the development of the latter. But it
is the actual physical. i.e. real life, experiment that is of
prime importance. The new ideas concerning a numer ical
utility--a side-issue for Game Theory, although originating
with it--have produced countless, rigorously controlled
experiments aiming at determining the utility functions of
different groups of individuals. Many experiments have
been made now also on the actual playing of games in order
to discover under what circumstances persons will form
coalitions, how they distribute the proceeds, whether these
imputations conform to the predictions by the theory, whether
and how players learn from repeated plays, whethar confi-
dence and trust increases or whether double-crossing be-
comes more frequent, whether men or women are more
likely to double -cros s, etc. There are experiments modeling
oligopoly situations and political procedures, not to mention
many military conflicts.

These experiments serve in the accustomed manner

to test and "confirm" the theory. Of course, no experiment
can ever prove a theory; it sometimes can provide new
material to incorporate into it. Confirmation- -which is
never more than giving a better description- -correction and
advancement, this is what good experiments ought to accom-
plish. It is gratifying that those generated by Game Theory

225
appear to work in all three directions.

In the social sc iences we will not discover new

phenomena hidden to our senses such as radiation. quasars.
antimatter with all their startling consequences for science.
It is not likely that there are mysterious. so far unnoted
occurrences. invisible little men mixed up with ourselves
or the like. What can be expected is that we will discover
subtle interrelationships among men that have so far escaped
observation and. a fortiori, analysis. It is possible that we
may from time to time be confronted with surprises 7 pleasant
and unpleasant. At any rate the social sciences have over
the last few decades definitely entered into a new phase of
s ophis tic ation.

There are many unsolved problems in Game Theory;

this is probably one of the attractions it exercises because
these problems are mathematically interesting and empiri-
cally significant. I shall point out only one. that of the
"integral game J" as I shall call it. This is the realization
that an economic individual is engaged simultaneously in
playing several different games, that, for him, the outcomes.
and therefore the strategie s, of thes e games are interrelated.
that he has a problem of allocating resources among strategies
in different games (not only among pre-determined uses as
in conventional theory) and that individuals are interrelated in
complex manners by the games they are variously playing with
each other--these games in addition being non-zero sum. A
comprehensive theory of the integral game will be very diffi-
cult to establish but is called for. At first we will have to
be satisfied with preliminary descriptions setting up the prob-
lem.

A final word about the role of the social sciences for

mathematics. It is no vain hope, I am sure, that mathe-
matics will receive as many stimuli from this field as it has
time and again from physics. Evidence is already on hand.
It suffices to point to the new developments in the field of
linear inequalities J convex bodies. fixed point theorems,
etc. The more intimate the contacts will bec ome between
social scientists and pure mathematicians the faster this

226
process will develop. It is important to realize that the
raw social evidence does not automatically reveal the under-
lying mathematical structure and that propositions from
pure mathematics are only seldom directly useful and
applicable to concrete empirical problems. The "inter-
face." to use a word beloved by engineers. between the
two worlds is where the main difficulty lies. But if that
situation is recognized by both sides it will surely lead to
further interesting and. we hope. practically important
(agreeable!) consequences.

227
 CHAPTER 2

WELFARE, ECONOMIC STRUCTURE AND

GAME THEORETIC SOLUTIONS

M. SHUBIK
Economic Growth Center, Yale University

ABSTRACT

The concept of solution to an n-person garn.e is dis-

cussed. Six solutions are defined and interpreted in terrn.s
of social desiderata. These properties include efficiency,
decentralization, social stability, fair division, and lirn.i-
tations on the power of groups in a society.

Sorn.e observations are rn.ade on the applicability of

the solutions to rn.odels of a national econorn.y. It is seen
that if certain physical and legal conditions were to hold it
is possible to satisfy rn.any different desiderata sirn.ultane-
ously. If the conditions do not hold then we are faced with
rn.aking a choice arn.ong rn.utually incorn.patible goals. It is
observed that a society such as the United States or any
other rn.odern nation state with large arn.ounts of social
capital does not have the conditions necessary for the sirn.ul-
taneous satisfaction of the desiderata noted.

Sorn.e further observations are rn.ade on econorn.ic

rn.odels with governrn.ent and political choice.

228
 A. Economics and Game Theory

What is the problem, and what constitutes a solu-

tion to the problem? Both of thes e questions are fundamental
to understanding the role of mathematical model building in
the behavioral sciences. The fo11owing discussion is devoted
to posing some questions concerning problems in economics
.and more particularly welfare economics. The methods of
mathematical economics and more specifically the theory
of games are used both to formulate and answer them.

Game theory provides a method for the study of

decision making in situations involving competition and/or
cooperation. A "game" involves decision-makers with
possibly different goals or objectives in which their fates
are intertwined. Each individual may have some influence
over the outcomes, but, in general, no one has complete
control.

"Game theory man" is a c10s e relative of the

rational utilitarian man who may be found in much of
economic theory. His desires are assumed to be we11
known to him, his abilities to calculate are often implicitly
(or explicitly) assumed to be high. He knows what are the
alternatives that confront him. He is aware of the powers
and limits of his actions.

Obviously the above caricature is a very stark

abstraction of man which for many purposes is plainly
inadequate. He is often unsure, vague, vain, pleasant.
generous. and so forth. A11 of these dimensions appear to
be removed from our description. Yet. for our purposes.
we argue that in economic life. and especia11y in actions
such as trading. setting prices, or paying taxes. the ab-
straction is a good first order approximation. Given that
we are wi11ing to assume that economic man knows more
or less what are his preferences and powers. we may then
be in a position to ask questions concerning the economy of
a society comprised of many more or less economic men.

How does a society distribute its goods and services?

229
How should it distribute its goods and services? What is
meant by economic efficiency? What limits are there or
should there be upon the economic powers of the individual
or the group? These and many more are natural economic
or politico- or socio-economic questions. The simplified
model of man combined with game theoretic analysis will
help to give some insights into and partial answers to the
questions. In particular an interpretation is given of the
virtues and limitations of the price system in the freely
competitive market guided by the "invisible hand."

B. Players and Preferences

Although the mathematical content of our dis cus s ion

is to be kept minimal J a ce rtain amount is useful as an aid
to exposition. We consider an economy as consisting of n
decision-makers or "players." They could be firms.
individuals, or institutions. Each player i has a set of
strategies Si which describe the actions that he is able to
take. A strategy may be a complex decision involving
pricing. advertising. distribution. and so forth. The pre-
cise form of the individual strategy is of considerable im-
portance in the analysis of a specific market. but at the
level of generality of this discussion detailed specification
is not necessary.

Associated with every set of n strategies (one for

each player) is an outcome. For example. in a Poker game
after all players have followed their strategies they expose
their cards and the outcome is the distribution of the money
at stake. Each player is assumed to have a preference
system which enables him to evaluate all outcomes of the
game. We may say that for each player i there exists a
payoff function Pi (s l' s 2' s 3' ••• ,sn) which is a funct ion
of the strategies of all of the players. This evaluates for
each player the worth to him of a particular outcome.

The rules of the game are a description or set of

instructions which specify the strategies which can be
chosen, the outcomes and their values. For example. in
chess the rules of the game describe the board. the way in

230
which any piece can be moved; they define checkmate and a
dra wand in the cas e of a tournament they as sign a point
value to win, lose, or draw. In an economy obviously the
rules are not so well defined. Nevertheless the technology,
laws, and institutions of society provide the bas is for the
construction of models sufficiently relevant to merit study.

The preferences of individuals for penknives, public

parks, space programs, automobiles, or soft drinks are
not fixed over time, are not necessarily well known and are
possibly subject to the persuasion and influence of others.
However, as a first approximation, in the short run we take
them as given. Even with fixed preferences there are still
several qualifications which must be made.

Some of the more important technical as sumptions

concerning preferences» are noted, then explained. Can
the individual's preferences be best represented by a com-
plete ordering, a partial ordering, or a lexicographic
ordering? If his preferences ~re completely ordered, do
we have just an ordering (indifference curves. for instance)
or is there a cardinal measure (a utility function fixed up
to a linear transformation). If a cardinal meas ure exists,
can one compare utilities between people? Does there exist
any commodity which can serve as a transferable utility in
the sense that a transfer of any number of units from one
individual to another will always cause the relative gain or
loss of one to change in a fixed proportion to the loss or gain
of the other.

The usual assumptions made in much of economic

theory are that individual preferences are best represented
by a complete ordering. Comparisons of welfare cannot be
made between individuals and no transferable commodity
in the sense noted above exists (in some cases, for small
amounts it may be argued that money serves as this com-
modity).

The assumption of completely ordered preferences

is often illustrated in terms of indifference curves as
follows. Suppose that an individual is confronted with a

231
choice between two bundles of commodities consisting (for
the sake of the example) of xi apples and Yi oranges or
X z apples and yZ oranges. The first alternative Ai or
(x1'Yi) will be preferred or indiffere:i,t to the second alter-
native A Z Ot' {xz,yZ}' or vice-versa. If we consider all
pairs and we as sume a condition of trans itivity (i. e. , if A 1
is preferred or indifferent to A Z and A Z to A3 this im-
plies that A1 is preferred or indifferent to A 3 ) we may
draw a map as is shown in Figure 1 , where each line is the
contour of alternatives to which the individual is indifferent.

ORANGES

APPLES
Figure i

Given only a complete ordering we can say that any

alternative on curve u1 is preferred to any alternative on
Uz but we cannot say by how much. An important contri-
bution of von Neumann and Morgenstern was to suggest how
one could obtain a cardinal measure for an individual's pre-
ferences. (1) This means that one can fix a scale specified
up to a linear transformation (i. e., with a free zero point and
scale factor) much in the same way as we have a temperature
scale.

It might be that an indi vid ual cannot 0 rde r all of his

choices. His preferences could be defined by a partial
ordering in which between any two alternatives not only is
there a possibility of preference or indifference but of
noncomparability. This is shown in Figure Z. Here armed
forces ranks have been used for the example where the
relations, "superior," "of equal rank," or "not comparable"
hold. The president is a superior of an admiral or general,
but in the military heirarchy an admiral is not equivalent
to a gene ral.

232
 ADMIRAL GENERAL

NAVAL CAPT AIM COLONEL

Figure 2

A further pos sibility is that in the m.aking of choices

alternatives are judged according to the fulfillm.ent of vari-
ous properties in order. For exam.ple, when selecting a
secretary, the individual m.ay rate typing speed, responsi-
bility. and good looks in that order. Rating each, choice is
made on the higher score. For exam.ple, if Hortense rates
(6,10,10), Mabel (8,3,10) and Fanny (8.4.1), Fanny is
chosen. This is a lexicographic ordering.

All of the above com.m.ents dealt with individual pre-

ferences. Even though we m.ight be able to find a m.easure-
m.ent scale for individual preferences this does not m.ean
that we can com.pare preferences am.ong individuals. Thus.
we m.ay not be able to say that som.ething is worth twice as
m.uch to Jack as it is to Joe. Why should we want to be able
to m.ake such a com.parison? It depends upon our needs to
resolve problem.s of welfare, involving taxation, subsidies.
and the provision of joint services. In log-rolling in
Congress and in m.any other situations involving the actions
of trustees for groups of people. assum.ptions concerning
equality. equity, or fairnes s are m.ade and they often are
used to establish som.e sort of approxirnate basis for inter-
personal com.parison of welfare in determ.ining joint action.

The existence of a possibility to com.pare welfare

does not necessarily m.ean that the m.eans to carry out all
policies exist. For exam.ple. suppose that two poor indi-
viduals jointly inherit a painting. If they try to divide it
between them. the worth of a fraction of the painting is as is

233
 A
GAIN OF 2

GAIN OF I
8 c
Figure 3

shown by AB in Figure 3. It is obvious that to cut it

destroys almost all of its value to either. We may be able
to say that the painting is worth twice as much to 2 as to 1,
but how are we to divide it? There are many schemes
pos sible. In particular, if there existed a means of side-
payment any division along the line AC could be achieved.
Is E the point of fair division or should some recognition
be given to the "unfairness" to 2 just because he valued
the painting more? We do not propose to discuss fair divi-
sion schemes further at this point but wish to stress that
the feasibility of any of the schemes will depend upon the
assumptions we make concerning individual preferences,
interpersonal comparisons and the existence of media for
transfer. These are not purely questions of experimentation
and logic. We are a long way from a unified general theory
of political-economy. Some of the different assumptions we
have noted may be useful and good approximate representa-
tions when we wish to deal with one set of problems but not
another.

C. Solutions and Their Interpretation

A popular mis conception concerning the applications

of n-person game theory is that they depend upon the
assumptions that individual's preferences can be compared
and that there exists a side-payment mechanism. These
assumptions do make much of the mathematics far less
complex problems; however, for the problems to be dis-
cussed, substantially the same results are obtained(2) even

234
if we dispense with interpersonal cOITlparisons and trans-
ferability.

Bya solution to an n-person gaITle, we ITlean an

iITlputation or outcOITle or set of outcOITles satisfying certain
properties. These properties ITlay arise froITl norITlative
or behavioral considerations. We ITlay believe that people
do or should behave according to theITl.

Five solutions are noted, discussed. and interpreted

in terITlS of the overall econoITlic context of the price systeITl
in an enterprise econoITlY. Many virtues have been sug-
gested, of the econoITlY run by the unfettered interplay of
ITlarket forces. Its proponents suggest that it is efficient,
fair, just, and so forth. It is possible to exaITline SOITle of
these c1aiITls in the light of the various solution concepts of
gaITle theory as is shown in Table 1 below:

Solution Social Relevance

(1) Pareto optimality econoITlic efficiency
(Z) Core power of coalitions
(3) Noncooperative equilibriuITl power of the individual
(4) The cOITlpetitive equilibriuITl decentralization
(5) Value fairness or equity

Table 1

1. Pareto OptiITlality

A division A of the proceeds froITl an econoITlic

systeITl aITlong its ITleITlbers is said to be Pareto optiITlal if
there does not exist any other distribution B such that all
ITleITlbers would obtain at least as much in B as in A and
at least one would have his position iITlproved. Figure 4
illustrates this for two people; in this diagram all points
on the surface Pi P z are Pareto optimal. In particular B
and C are Pareto optimal but A is not. Any outcome in

235
 Figure 4

the triangle ABC is preferred to A in the sense that

either one can gain at no cost to the other or both can gain.

A Pareto optimal outcome is efficient inasmuch as

no reallocation of resources can result in an increase of
welfare to any individual without a decrease in welfare to
some other individual. Apart from this weak welfare con-
dition the Pareto optimal condition does not tell us very
much. In particular, the distribution may be highly biased
but still efficient (for example, an outcome which gives
everything to one individual, such as P 2 is Pareto optimal).

2. The Core

The core is the set of outcomes such that no group

of individuals acting together can obtain by themselves
more than they are offered in any of the outcomes in the
core. (3)

Two very simple examples will serve to illustrate

a game without and a game with a core. They are illus-
trated below in characteristic function form:

v(9)=0 v(9)=0

v(1)=O v(2)=0 v(3)=0 v(1 )=0 v(2)=0 v(3)=0

-- 1 -- 1 -- 1 --
v(1,2)=2"2 v(1 ,3)=2"2 v(2,3)=2"2 v(1 ,2)=1 v(1 ,3)=1 v(2,3)=1

v(1,2,3)=3 v(1,2,3)=3

236
 The characteristic function specifies the worth that
a coalition can achieve if its members decide to act in
unison by themselves. For a three person game there are
eight coalitiions (count~the one with no members). In the
two examples above v(1.Z) =
zt means that the coalition of
players 1 and Z in the first game are able to obtain zt
by independent action. v( 1 .3) = 1 means that the coalition
of players 1 and 3 in the second game can obtain 1 by
indepe ndent action.

An imputation of wealth is a vector a = (a 1 ,a Z ' ••• ,an)

where ai is the share obtained by the ith individual and
a is a point on the Pareto optimal surface (if there exists
a transfer or sidepayments mechanism, then f
i= 1
a i equals
the joint maximum, a single number).

The two games above can be described as follows:

if any two individuals form an agreement, the referee gives
them a sum a to share (a = zt or 1). the excluded indi-
vidual obtains nothing. If all three form a coalition they
obtain 3.

In both games, when viewed from society as a

whole, it is evident that cooperation is the most profitable
as they can obtain 3 to divide among them. In the first
case, however? no matter how they divide the wealth there
will always be a potential coalition that could have obtained
more by failing to cooperate. This can be seen by trying
to solve the following inequalities:

ai - Z-2.1
+ aZ:::::
a. az + a3 2: zt

a1 + a3 2: zt

b.

Adding the three inequalities together we obtain:

Z(a1 + a Z + a 3 ) = 7t
or a1 + a Z + a3 = 3-3/4

237
which is inconsistent with condition (b). This means that
the core is empty. In the second game, however, we have:

a1 + aZ 2: 1

c. aZ +a3 2: 1

a1 + aZ 2: 1

d. a1 + aZ + a3 =3 •
Adding the three inequalities together we obtain:
1
a1 + aZ + a 3 2: 12

which is consistent with (d). There will be many impu-

tations in the core such as (3/4,3/4,d-) or (1,1,1) or
1
( 2,1 1
2 ,1).

The lack of existence of a core implies a great

potential for social instability. Any division of the wealth
of society will always be threatened by some group.

3. The Noncooperative Equilibrium

A comp lete(41 different concept of solution has been

provided by Nash. ) The core is a cooperative solution
inasmuch as the individuals are presumed to negotiate and
coordinate their actions to obtain jointly as much as possible
while using the coalitional power to decide upon the division
of proceeds. The spirit behind the noncoope rative equili-
brium is that all exercise their individual power in an intro-
spective manner without any attempt at coordination or co-
operation.

Conside.r a game with n players. Let the set of

strategies for player i be Si where si denotes a parti-
cul~r ~tr~egy. '!:.he payoff to the ith player is
Pd s 1,sZ,s3'··· ,sn)· A set of strategies ls1'sZ"" ,sn)
is said to constitute an equilibrium point if for all i. the payoff

Pi (s 1 ' s Z ' •• 0 , s i -1 ' s i ' s i +1 ' • 0 • , -; n)

is maximized by setting si = sio In words, a set of

238
strategies, one for each player, forms an equilibrium point
if each player, knowing the strategies of the others. is not
motivated to change.

Noncooperative sOlutiO~s form the basis for much of

the work in oligopoly theory. (5 They are in general not
optimal. The individual exercise of large economic powers
may easily cause an inefficient allocation of resources.

4. The Competitive Equilibrium

The competitive equilibrium market model of the

price system contrasts with both cooperative and nonco-
ope rati ve game solutions. It is bas icall y a mechanistic
solution which will (under the appropriate conditions) satisfy
Pareto optimality and some important properties linked
with decentralization.

Suppose there are a number of individuals trading

in k commodities. Each individual i has a preference
system which can be represented by a function
iii i
ui(x i ,x 2 ' ••• ,xk ) where x k stands for the amount of the
kth commodity held by the ith trader. The existence of
a price system in competitive equilibrium amounts to there
being a set of prices (Pi ,P2' ••• ,Pk) such that if each
individual merely accepts the prices as given and each
tries to maximize his welfare subject to the budget con-
straint that income and expenditures must balance (this can,
of course be modified for tax and credit conditions); then
supply will exactly equal demand in all markets. Further
more the result will be Pareto optimal.

The impact of the existence of a price system is that

a complex multi-person optimization problem can be replaced
by a host of individual decentralized optimization problems,
all coordinated through the mechanism of prices.

5. The Value

There are several different but highly related value

239
solutions which hav~ been suggEjsted such as those of
Shapley. (6) Nash. (7) Harsanyi .l8) and Selten. (9) They are
all cooperative solutions. i. e •• all participants will eventu-
ally cooperate but use the solution to determine their
shares of the final proceeds). These solutions are based
upon the axiomatization of concepts of symmetry. fairness
or equity.

The simplest solution to explain is the Shapley value.

It can be calculated directly from the characteristic function
of a game.

We consider every way in which an individual can

enter every coalition and we credit him with his incremental
contribution to the coalition. In terms of a coalition s
containing player i this is v(S) - v{S - {i}) whe re the
symbol s - {i} stands for the set of players S with player
i removed. Adding all of his contributions together we
average them and award him that amount as his "fair share."
(This takes into account both his needs and his contribution
to society.) If 4;. is the fair share for the ith player
then it can be sho\.vn that:

~. = (s-1) !(n-s) i [v(S) _ v(S _ {i})]

1 n !
over all
S

D. Solutions and Their Application to Markets

1. The "Right Structure" but Few Individuals

There are two major sources of difficulty which

may interfere with the functioning of a market system.
One is fewness of individuals in the markets. "Thin mar-
kets" tend to be imperfect. The other concerns the
structure of preferences and technology. *
*If preferences can be represented as convex sets and pro-
duction conditions can be modeled as a cone containing
the origin. no difficulties will be encountered.

240
 To begin with, we assume that the structure of pre-
ferences and technology cause no problems. We examine
the solutions to a simple market model of an economy by
applying all five solutions suggested in Section C. Suppose
there were three traders in an economy, Figure 5 shows
what the solutions might look like:

u,

Figure 5

The Pareto optimal surface is bounded by the lines

P 1 ,P Z ' P 1 'P 3 and PZP 3. The core lies on the Pareto opti-
mal surface and is shown as the triangular area C 1 CZC •
It is a remarkable fact that it can be proved that the core
will always be non-empty if the sufficient conditions fOf
t h e eXlstence 0 f a compe toltlve equl°lob
0 0 0
1 num are met. (10
Thus the competitive equilibrium exists and is shown
as a point E on the Pareto optimal surface and within

241
the core.

The value V will also lie on the Pareto optimal

surface and mayor may not be contained within the core ,(11)
the more symmetric the game the more likely it will be in
the core.

The noncooperative equilibrium N will lie below

the Pareto optimal surface. There are some difficulties
encountered in formulating a model of an economy in such
a way that it is fossible to apply the noncooperative equili-
.
b rlum so lutton.
· 12)

When numbers are few, each solution concept gives

rise to a different outcome or set of outcomes. In parti-
cular, one's criterion of fairness may be at odds with the
power of coalitions.

2. Many Individuals: Convergence Theorems

Suppose that the structure of preferences and

technology were the same as before but that there were
many traders in all markets. It can be shown that (using
the appropriate concept of limiting process) the core, value
and noncooperative solutions all proceed towards the same
limit solution(13) and that limit is the competitive equili-
brium.

The core shrinks down; the value "chases" the core

and for any size may be in or out of it. The noncooperative
equilibrium approaches the competitive equilibrium from
below the Pareto optimal surface.

The meaning of these results is that the virtues

which may have been sung of the free enterprise price
system rise from many and not one solution concept. If
the conditions in the economy are right, then it may be
possible to satisfy one's desires for efficiency, decentrali-
zation» fair division, and protection against the bargaining
power of groups or the noncooperative behavior of the
individual, all at the same time.

242
3. The Wrong Structure

Unfortunately in the type of economy that typifies

most of the world the conditions required for the concur-
rence of the solutions noted above do not exist. In parti-
cular. the presence of economies or diseconomies external
to the firm may destroy the convergence. Examples are:
the improvement of transportation by one firm may improve
transportation facilities for others as well; the burning of
soft coal by one firm can deposit soot on delicate processes
being run by another.

Around thirty per cent of the assets of U. S. society

are owned by the state, municipal and Federal governments.
Joint assets are important and are becoming more so. They
do not fill the conditions indicated in D.i. and D.2. Indivisi-
bility of various resources also introduces problems.
Ownership is a far more complex problem than most eco-
nomic models indicate and it too may cause the failure of
convergence.

4. Conclusions

If this were the best of all possible worlds the in-

visible hand would appear in the best of all possible ways
to guide the economy. The price system would not only
exist but have many properties beyond decentralization and
efficiency. Unfortunately conditions are not such. This
means that sometimes price systems do not exist.

The game theory approach has been used by construct-

ing solution concepts and formalizing different properties
deemed desirable in a society. In a very simple instance.
it appears that a unification of the different approaches is
feasible. Beyond that the aims of a society in pursuing
different desiderata may be inconsistent. The price of
control over the power of groups could be inconsistent with
a fair division scheme, efficiency, decentralization, or
control of economic power.

243
 References

(1) von Neumann. J •• and O. Morgenstern. Theory of

Garnes and Economic Behavior, Princeton University
Press. Princeton. New Jersey. 1944.

(2) Shapley. Lloyd S. and M. Shubik, Competition. Welfare

and the Theory of Garnes (Forthcoming).

(3) • "Concepts and Theories of Pure Compe-

tition." Chapter 6 in M. Shubik (Ed.). Essays in
Mathematical Economics In Honor of Oskar
Morgenstern. Princeton. Princeton University
Press. 1967.

(4) Nash. J. F •• "Non-Cooperative Garnes." Annals of

Mathematics, Volume 54, pp. 286-295. Princeton:
Princeton University Press. 1951.

(5) See Shubik, M. , Strategy and Market Structure, John

Wiley» New York, 1959.

(6) Shapley. L. S., "A Value for N-Person Garnes,"

Annals of Mathematics. Study 28. pp. 307-317.
Princeton: Princeton University Press. 1953.

(7) Nash, J. F., "The Bargaining Problem, "

Econometrica, 18. pp. 155-162.1950.

(8) Harsanyi, J. C., "Approaches to the Bargaining Prob-

lem Before and After the Theory of Garnes."
Econometrica. 24. pp. 144-157.1956.

(9) Selten, R •• "Valuation of N-Person Garnes," in

Drescher, Shapley, and Tucker (Eds.), Advances
in Game Theory, Annals of Math. Studies, Study 28
(1953), pp. 307-317, 1953. Princeton: Princeton
University Press.

(10) Scarf. H •• and G. Debreu. "A Limit Theorem on the

Core of an Economy." Intern. Econ. Rev. 4,
pp. 235-246, 1963.
244
(11) Shapley. L. and M. Shubik. "Pure Co:mpetition.
Coalitional Power and Fair Division." Intern.
Econ. Rev. (Forthco:ming); also RM-1-PR.
RAND Corporation. Santa Monica. March. 1967.

{12} • "Concepts and Theories of Pure

Co:mpetition." {see Reference 3 above}.

(13) • Co:mpetition, Welfare and the Theory

of Ga:mes (see Reference 2 above).

245
 CHAPTER 3

ON COMMITTEES

L. S. SHAPLEY
The RAND Corporation

INTRODUCTION

In this paper we offer the reader an opportunity to

inspect, at close hand, the substance and methodology of a
special corner of descriptive game theory. The "simple
games II that populate this area are finite. combinatorial
structures that are not only amusing to mathematicians but
can serve as abstract representations of voting systems or
other group-decision procedures. As such, they have found
applications in political science and organization theory.
as well as in certain branches of pure mathematics.

The substantive theme of the paper will be the analysis

of the structural role played by "committees." The method-
ological theme will be the intensive use of the language and
logic of Boolean algebra and elementary set theory- -almost
to the exclus ion of any other mathematical apparatus. In
this respect, the theory of simple games provides a striking
example of the trend away from the techniques of classical
analysis, as mathematical theories and mathematical model
building invade ever-wider areas of the nonphysical sciences.

This paper is not intended as a survey of the theory of

simple games, except incidentally. Rather. it works toward

246
a specific goal, a "unique factorization" theorem that de-
scribes how a simple game is decomposed into a hierarchical
arrangement of committees, subcommittees, and individual
agents. Since this is a new theorem, we give it a fully
rigorous treatment, and almost half of the following text is
taken up with the sometimes complicated (but entirely
elementary!) proofs that are required to "keep us honest. "
These proofs may be skipped on first reading without loss
of continuity.

The study of simple games was initiated by von Neumann

and Morgenstern in their epochal book on game theory, first
published in 1944.* Since then many authors have made many
contributions not only to the descriptive theory but to various
solution theories and to several domains of application. **

*Theory of Garnes and Economic Behavior, Princeton

University Press. 1944, 1947, 1953.

**Some recent examples: J. R. Isbell. "Homogeneous

Garnes TIr" and L. S. Shapley, "Compound Simple Garnes,"
in Advances in Game Theory: Annals of Mathematics Study
No. 52, Princeton University Press. 1964; W. H. Riker and
L. S. Shapley, "Weighted voting: a mathematical analysis
for instrumental judgments ~" in Nomos X: Representation
(Yearbook of the American Society for Political and Legal
Philosophy). Atherton Press, New York. 1967. For a short
expository article with a complete bibliography to date. see
this author's Simple games: an outline of the descriptive
theory. RAND Corporation Paper P-2277. April 1961.
(Also published without bibliography in Behavioral Science 7
(1962). pp. 59- 66.)

247
 1. SIMPLE GAMES

Th~ theory of simple games is primarily oriented

towards problems of organization and control, rather than
payoff and strategy. Its point of departure is the primitive
notion of "winning coalition." It makes no attempt to treat
situations where the costs of winning are significant or
partial victories are pos sible, or where the fruits of
victory are not freely divisible within a winning coalition.
A simple game may be thought of as an idealized power
structure, a voting system, a legislature, or indeed any
constituted procedure for arriving at group decisions.

If one wishes to relate simple games to the game-

theory models more commonly found in economics or
operations research, with their explicit strategies and
their numerical payoff functions, one may imagine that
there is a lump sum of money to be won, and that the
strategic opportunities open to the players are such that
it takes coordinated action by the members of some winning
coalition in order to capture and divide the prize. To re-
late simple games to the numerical characteristic-function
form introduced by von Neumann and Morgenstern~. cit ••
it suffices to assume that the characteristic function takes
on only two values: "1" for the winning coalitions and "0"
for the others.

The notation r(N, W) will be used to denote a simple

game. Here N is a finite set, the players; and W is a
collection of subsets of N, the winning coalitions. We lay
down three conditions on W:
(la) N E:: W;
(1 b) cjJ (" W; and

(1 c) if SST S Nand S E:: W, then T E:: w.

The first two merely suppress certain extreme cases, for
technical reasons. The third condition, however, expresses
a fundamental monotonicity property. inherent in the notion
of "winning": any coalition containing a winning coalition

248
must also win. A sort of converse would also be a natural
condition:
(1d) if SnT =cf> and S E::W. then T tW.
i. e., any coalition completely disjoint from a winning coali-
tion must lose. But we do not impose this condition a priori.
Garnes that satisfy (1d) are called proper; all others, im-
proper. Tho\:lgh rarely found in application. improper
games play an important role in the structural theory,
somewl;tat analogous to that of imaginary numbers in algebra.

Our notation for simple games is deceptively concise.

The double abstraction "w" (a set of sets) embodies in a
single symbol a possibly intricate web of voting rules. We
have made exhaustive counts of the different simple-game
structures that are possible on small sets of players.
Excluding games with dummies (see Section 3 below) and
games that are me rely permutations of game s already
counted. we found the following:
No. of players. 1 2 3 4 5 6
No. of games 1 2 5 20 179 1

In the face of this explosive growth, we can hardly

expect to explore the possibilities in an effective way without
the aid of patterns and symmetries and special classes of
games having "nice II properties. It would be helpful to
know how to detect and exploit substructures within a game
{i. e •• "committees "} that allow it to be decomposed into
smaller games. That this happens often enough to be worth-
while is shown by the following data:
No. of players. • • • 1 2 3 4 5 6
No. of decomposables. 0 2 4 14 78 1210

No. of "primes" • • • 1 o 1 6 1 01 1

In this paper we shall achieve a complete analysis of the

structural possibilities for decomposition. However, the
"detection" problem remains in an unsatisfactory state; an
efficient algorithm is sorely needed for finding committees

249
in larger ga:rnes.

2. COMMITTEES

In analyzing si:rnple ga:rnes for deco:rnposability, the

funda:rnental idea is to identify certain subsets of individuals,
called "co:rn:rnittees," that can be treated en bloc. Then
we can separate the internal politics of the co:rn:rnittee fro:rn
the external politics of the rest of the ga:rne. In deter:rnining
whethe r a given large coalition can win, in a ga:rne where
there are co:rn:rnittees, we do not have to ask which particular
co:rn:rnittee:rnen are party to the coalition, but only whether
they have the controlling voices in their respective co:rn-
:rnittees. Thus, an ancillary notion of "winning" co:rnes into
play, defined inside each co:rn:rnittee. In fact, a co:rn:rnittee
is a si:rnple ga:rne in its own right, e:rnbedded in the larger
ga:rne.

For:rnalizing these ideas, we define a co:rn:rnittee of

the si:rnple ga:rne T(N, W) to be another s i:rnple ga:rne r(C, W C},
with cf> c C ~ N, which is related to the first as follows: for
every coalition S ~ N such that:
{2a} sUc E:. Wand S-C ¢ W,
we have:
{2b} S E: W if and only if SnC E: W C •
Condition (2a) expresses that the participation of :rne:rnbers
of the co:rn:rnittee C is crucial to the success of S. Con-
dition (2b) expresses that the effect of their participation
is entirely deter:rnined by the rules of the co:rn:rnittee ga:rne
r(C, W C).

By extension, the word "co:rn:rnittee" will also be used

for the set C, whenever a ga:rne r(c. W C) exists that is a
co:rn:rnittee in the sense just defined. (No confus ion should
result fro:rn this; see the corollary at the end of the next
section. )

A ga:rne always has certain extre:rne co:rn:rnittees,

na:rnely the "co:rn:rnittee of the whole" and the co:rn:rnittees

250
consisting of single individuals. By a proper committee we
shall mean one of intermediate size. Obviously. only proper
committees can lead to significant decomposition of the
game structure.

3. MINIMAL SETS AND DUMMIES

If W is any collection of subsets of N, we shall de-

note by Wm the collection of minimal elements of W
i. e., those sets in W that have no proper subsets in W.
If W is known to be monotonic {condition {1c} }, then W
can easily be reconstructed from a knowledge of Wm •
Accordingly, we shall often use just the minimal winning
coalitions to identify a particular game.

The abbreviation Wm{C) will be much used in the

sequel, standing for the collection of all coalitions in Wm
that "meet" C, io e •• that have a nonempty intersection with
Co Wm(C) should not be confused with W~, the set of
minimal elements of W Co

A player belonging to no minimal winning coalition is

called a dummy, since he never makes any difference to
the status of a coalitiono Any set of dummy players is
vacuously a committee by our definition, since the hypothe-
sis {2a} is never met. Such an all-dummy committee is
called inessential; all others essential. The internal rules
of an inessential committee are quite arbitrary, being ir-
relevant to the game as a whole.

A player who is a dummy in the committee game is

obviously a dummy in the full game too. Conversely, a
dummy in the full game is a dummy in any essential com-
mittee to which he may happen to belong. If a dummy is
dropped from, or added to, any committee, the resulting
set remains a committee.

Our first theorem gives the relation between the

minimal winning coalitions of a committee and the minimal
winning coalitions of the full game.

251
 THEOREM 1. Let r(C, W C} be a
committee of r(N. W}. Then for every
S E: Wm(C) there exists B E: We such
that SnC =
B. Conversely, for every
B E: We there exists S E: Wm such
that S nC =
B, unle s s the committee is
inessential.

Proof. (a) Given S E: Wm(C), we have S UC E: W

by (1c) and S-C K.
W by the minimality of S. Hence
(Za) applies II and SnC E: W C by (Zb). Hence there
exists B ~ SnC with B E: We. Let T =
(S-C) UB
(see the diagram); then {Za} holds for T. and T E: W
by (Zb). But T ~ S; hence T =
S by the minimality
of S. Hence SnC = T nC =
B, as required.

(b) Unless C is inessential there is a nondummy

in C, and hence a set Q E: Wm(C}. We have QUC E: W
by {1c} and Q-C f:.
W by the minimality of Q. Given
B E: We. define R =
(Q-C) UB (see diagram). Then
RUC = QUC and R-C = Q-C; hence {Za} holds for R,
and R E: W by (Zb}o Hence there exists S ~ R with
S E: Wm. By {tc} we see that sUC (: Wand also
S-C ~W; hence SnC E: W C ' by (Za), (Zb). But
SnC ~ RnC =
B; hence SnC =
B by the minimality
of B. This completes the proof.

c_-----_ c_-----_

T shaded R shaded
(a) (b)

252
 Theorem 1 reveals that except for the inessential
case the elements of We are precisely the intersections
of the elements of Wm{C) with C. Thus the following
corollary holds, justifying our double use of the term
"committee" :

COROLLARY. Each committee set C,

unles s it cons ists entirely of dummie s, cor-
responds to a unique committee game
nC, WC}.

4. THE SUBSTITUTION PROPERTY

The connection between Wm{C} and We proves to

be even closer than Theorem 1 would indicate. Indeed, the
next theorem asserts that we may take any element of We
and adjoint to it any set of the form S-C with S E:: Wm{C)
and the result will be an element of Wm •

THEOREM 2. Let r{C, W C) be a com-

mittee of r(N. W}. Then
{4a} Wm{C) ={BU{S-C) I B E:: W~ and S E:: Wm(C)}.

Proof. Theorem 1 tells us at once that "<;;" holds

in (4a); it remains to show that ";:?" holds in order to
establish equality. In other words. we must consider
an arbitrary B E:: We and an arbitrary S E:: Wm{C}
and show that BU{S-C) E:: Wm{C). We shall do this
in two steps.

(a) Write T for BU(S-C). (See the diagram.)

We have TUC C W by (tc); also T-C t.
W by the
minimality of S. Hence (2a) holds for T. and we
have TC W if and only if TnC E:: W Co But
TnC = Be We; hence T E::. W.

(b) To show that T is minimal, let S' S; T with

S' (Wm. Then we must have _Sf E:: Wm{C), since
S'-C S T-C ~ W. Write B' for SnC (see the dia-
gram); by Theorem 1 we have B' E: W~. We can

253
 therefore repeat the argument of (a), with primed
letters, and obtain T' E:: W. where T' = B' U(S'-C).
But T' S S; hence T' =
S by the minimality of S.
Thus we have
S'-C =T'-C =S-C = T-C.
Also, we have S'nC C We by Theorem 1. But
s' nc S; B; hence
S'nC = B = TnC,
by the minimality of B. The two displayed equations
establish that S' =
T, and hence that T E:: W m •
Finally. since T ~ B ~ cf>, we have T C Wm(e).
This completes the proof.

c c
~----- ......

,,--_~~O~""""'S'
T shaded T' shaded

(a) (b)

Theorem 2 enables us to "substitute" the portion of

any S E:: Wm that lies within a committee for the portion
of any other S' E: Wm that lies within that committee
(provided that both portions are nonempty), with the
assurance that the resulting coalition is also :r;ninimal
winning. According to the next theorem, this doesn't work
for any set that is not a committee. In other words, the
substitution property is a sufficient as well as a neces sary
condition for committeehood.

THEOREM 3. Let r(N, W) be a simple

game, and let C <; N. Then C is a com-
mittee if and only if, for every S, S' C Wm(C),

254
(4b) (sOC)U(S'-C) E: W m •

Proof. If C is a committee, the result is im-

mediate from Theorem 2. Conversely, suppose that
(4b) holds for all S, S' E: Wm{C). Possibly Wm{C)
is empty; in that case the players in C are all dum-
mies, and C is a committee (inessential). Other-
wise, let Q be a fixed element of Wm(C) and define
the collection W C to consist of all sets B S C
suchthat BU{Q-C)E:W. Clearly r{C,W C ) isa
simple game, in the sense of (1a), (1b), (1c); we
must verify that it is a committee of r{N, W).

Referring to the definition of committee, we see

that we must show that "T E: W" is equivalent to
"TOC E: W C," for every T with T Uc E:. Wand
T-C r..W. Thus, suppose T E: W. Then we can
find S € Wm with seT. S must meet C, since
S-C S T- C ¢. W. Applying (4b), we substitute S for
Q in C and obtain (SOC) U(Q-C) E: Wm • Hence
(TOC)U(Q-C) E:: W. Hence by the definition of WC'
we have TOC E:: W C.

For the other direction, suppose TOC E: WC.

Then (TOC)U{Q-C) E:: W, and we can find S E: Wm
with S S{TOC)U(Q-C). Since Q is minimal, S
must meet C. Similarly, since TUC C W, we can
find S' E:: Wm such that S' C TUC; and since
S'-C S T-C ¢ W, S' must ~eet C. Applying the
substitution (4a) , we obtain (SnC)U{S'-C) E: Wm.
But
(SnC)u(S'-C) =2 ([ (TOC)U(Q-C)] nC)U{[ TUC]-C)
= (TOC)U(T-C)
= T.
Hence T E: W. This completes the proof.

Note that Theorem 3 makes no mention of the com-

mittee game r(C, W C). This eliminates much of the

255
clumsiness involved in testing committeehood directly from
the definition.

5. SUBCOMMITTEES

THEOREM 4. Let C be an essential

committee of r(N, W) wi th winning sets
W C, and let D S C S N. Then D is a
committee of r(N, W) if and only if D is
a committee of r(C, W C).

Proof. (a) Suppose D is a committee of r(N, W).

Take any B, B' C W~ with BOD"* cP, B'OD"* cPo
By Theorem 1 there are S, S' C W m with sOC = B,
S' Oc = B' and we have S, S' C Wm(D). Let
T = (SOD)U(S'-D). By Theorem 3 we have T C Wm,
By Theorem 1 we have TOC C We. But TOC =
(BOD)U(B'-D). (See the diagram.) By Theorem 3,
D is a committee of r(c, W C).

(b) Conversely, suppose D is a committee of

r(C,W C )' Take any S, S' ~ Wm(D) and define
B=SOC, B'=S'OC. By Theorem 1, BE::We and
B' t: W~. By Theorem 3, (BOD)U(B'-D) E: W~.
By Theorem 1, there is T C wm(c) with T Oc =
(BOD)U(B'-D). By Theorem 3, we have
(TOC) U(S'-C) E:: Wm. But

c,-______________ ~

T nC shaded

256
 (TnC)U(S'-C) = (BnD)U(B'-D)U(S'-C)
= [(SnC)nD] U[ (S'nC)-D] U(S'-C)
= (SnD)U(S'-D)
(same diagram illustrates). A final application of
Theorem 3 shows that D is a committee of r(N, W).

6. CONTRACTIONS

By the "contraction" of a game on one of its com-

mittees. we shall mean the game obtained by coalescing
the committee into a single player. More specifically:
(a) the members of the committee are dropped from the
game; (b) a single new player is introduced; (c) coalitions
containing the new player win if and only if, with that
player replaced by the whole committee, they won in the
original game; and (d) coalitions not containing the new
player win if and only if they won in the original game.

Two simple theorems describe a contraction's effect

on committees that either are disjoint from or include the
committee on which the contraction takes place. (See the
figure.) We omit the simple proofs.

THEOREM 5. Let C be a committee of

r(N. W) and let D S N-C. Then D is a
committee of r(N, W) if and only if D is
a committee of the contraction of r{N, W)
on C.

THEOREM 6. Let C be a committee

of r(N, W) and let C S D S N. Let D(C)
denote the set (D - C) U{i C }' where iC is
the new player introduced by contracting
on C. Then D is a committee of r(N, W)
if and only if D{C) is a committee of the
contraction of T(N, W) on C.

257
 0-
c

o
c

258
 If DeC, then of cours e D would dis appear in any
contraction on C, and we can infer nothing about D's
conunitteehood. There remains the case where D and C
"overlap," i. e., where the sets CnD, C-D. and D-C are
all three nonempty. This is a more complex situation.
since contraction on C now violates the integrity of D--
some of the players in D are eliminated, others remain.
It is not clear intuitively how the committeehood of D
is related, if at all, to the committeehood (in the contracted
game) of either D-C or (D-C) U{i C }' the two most likely
candidates for comparis on.

It might be hoped that the problem does not really

arise. i. e •• that the committees of a simple game can never
overlap (at least if we ignore dummy players). Then the
committees. ordered by inclusion. would form a tree-like
hierarchy. and we could determine the "prime" game played
by any committee by contracting on all of its proper sub-
committees. With no overlapping present, Theorems 5 and
6 would assure that the result would be independent of the
order in which the contractions were carried out.

Unfortunately, overlapping does occur. without the

*
aid of dummies. and we have a real problem. A unique
decomposition based solely on the contraction principle is
not attainable, since we would sometimes be forced to
decide whether to contract on C and spoil D or to con-
tract on D and spoil C. Some further structural con-
cepts must be introduced before we can cope with this
difficulty.

7. COMPOUND SIMPLE GAMES

A new notational device will be useful at this point.

Let Hi' ••• ,Hm be simple game s having disjoint sets
of players. and let K be another simple game having exactly
m players. indexed by the numbers from 1 to m. Then
we shall let the symbol

*For example, in the game r(N{N}) every subset of N

(except c/» is a committee.

259
 K[ Hi' ••• , Hm]
respresent the compound game, defined by taking as players
all the players of the component games Hi' and by taking
as winning coalitions all sets that include winning contingents
from enough of the components to spell out, by their indices,
a winning coalition of the quotient game K.

Repeating this definition more formally, let

I
N = {i, ••• , m}; let {Ni i (" M} be a collection of dis-
joint sets with union N; let Hi =
T(N i , Wi)' all i E:: M;
and let K = T(M, U). Then K[ Hi' ••• , Hm] is a~­
pound representation of the game r(N, W), where W is
defined as the collection of all SeN such that
I n
{i S N i E:: Wi} E:: U.

If the quotient is a one-person game, or if all of the

components are one-person games, then "compounding" is
a trivial operation. The compound is the same as the lone
component in the first case, and the same as the quotient in
the second case. We say that a game is decomposable only
if it possesses a nontrivial compound representation. We
say that a game is prime if and only if it is not decomposable.

What is the connection between compounds and com-

mittees? Clearly, each component in a compound represen-
tation is a committee of the compound game. Conversely,
any committee of a simple game can be made a component
of some compound representation of the game. In fact,
we can use the contraction on the committee for a quotient,
and let all the other components be one-person games. Of
course, if we start with a trivial committee (one-player or
all-player), then we get the trivial compound representations
just described. Thus, the decomposable games are exactly
those that possess proper commiftees.

In the case of a one-person component, we shall often

write the name of the player rather than the name of the
game in the compound- repres entation symbol. This will
give us, in particular, a way of displaying the names of the
players in any game, thus: G[P1' ••• ,Pm]'

260
 Subcommittees can be displayed by using compound
rep res entations as components. For example, the symbol
K[H 1 [G 1 ,G Z ,p,q] ,H Z ,H 3 ,r,s]
reveals that G 1 and G Z are committees of H 1[ G 1 ,GZ ,p,q] ,
which is in turn a committee of the full game.

For a further example, let us consider the following

five-person compound simple game:
M 3 [p,M 3 [q,r,s] ,t],
Here "M3" denotes the simple majority game on three
players. The reader will verify that there are exactly
seven minimal winning coalitions, namely:
pqr, pqs, prs. pt, qrt, qst, rst.
This is of course a decomposable game, since qrs is a
proper committee. But might there be other proper com-
mittees, not revealed by the given compound representation?
The game M3 is prime, but how can we be sure that there
are not other compound representations, in this and similar
situations, that distribute the players into components in a
completely different way?

This is the question that motivated the present study.

Compound simple games were introduced many years ago,
and their solutions have been extensively studied. *
But the
question of the uniqueness of a compound representation
with prime quotients has been elusive, and can apparently
be resolved only by going to the more fundamental notion
of "committee" that we have introduced in this paper.

8. SUMS AND PRODUCTS

As our notation indicates, a quotient is essentially a

function. Since both the arguments and the values are games,
however, it can also be regarded as an operation, acting on

*References will be found in the papers cited near the

beginning of this article.

261
games in much the same way as Boolean operations act on
sets. or as logical operations act on truth values. Two
extreme cases. among the possible quotients, playa special
part in the theory. and it will be convenient to represent
them as operations rather than functions. The first,
denoted by G:>, corresponds to quotients having the maxi-
mum possible number of winning coalitions; the second,
denoted by ®. corresponds to quotients having the least
pos sible number of winning coalitions.

Specifically, we define the ~ of m games:

m 2: 2,
to be the compound game K[ Hi' •••• Hm 1 where K =
r(M,{S S M Is* cP}). That is, a coalition wins in a sum of
games whenever it contains a winning contingent from at
least one of them. In similar fashion, we define the product
of m games:
m 2: 2,

to be the compound game L [Hi' ••• , Hml where L =

T(M, {M}). That is, a coalition wins in a product of games
only if it contains winning contingents from all of them.

These operations are obviously associative and com-

mutative, provided that we prop'3rly identify the players
after any re-ordering. A distributive law relating them
cannot even be stated. however, so long as we require the
player sets of different components in a compound to be
disjoint. But there is a duality that can be developed between
<!> and ® analogous to the Boolean duality between U
and n and the logical duality between "or" and "and."

It is worth noting that one-person games are not trivial

building blocks in the formation of sums and products. In
fact, there is an interesting class of games that is generated
by repeated applications of the operations ED and ®. in
which all the components are one-person games.

A sum is distinguished by the fact that its minimal

262
winning coalitions are precisely the minimal winning coali-
tions of its components. No minimal winning coalition
meets more than one component. *
Conversely, any game
whose minimal winning coalitions "fit" within a partition
in this manner is decomposable as a sum. It is easy to
see that such a game has a unique representation as a sum
of games that are not themselves sums--we merely take
the finest partition that does not split any minimal winning
coalition.

A similar unique dec ompos ition hold s for products.

(This holds from the duality between the two operations g
mentioned above, or can be proved directly.) Since no
game can be both a sum and a product, there is even a
unique decomposition of any game into a polynomial expres-
sion in (!) and ®, the components of which are indecom-
posable with respect to both operations.

For later reference, we note that there is a simple

test that tells whether a given committee gives rise to a
product decomposition of its parent game. that is, whether
r{N, W} has a compound representation of the form
r(C,WC>®H for some H. where C is the given committee.
=
In fact, such a decomposition exists if and only if Wm(C) W m
and there is a nondummy in N -C. The proof is a simple
application of Theorem 2. ** A committee that passes this
test will be called a factoring committee. If C is a factor-
ing committee, then obviously so is its complement N -C.

*It follows that a sum is always an improper game.

**For sums, thecorresponding condition is Wm(C} = w mC

and a nondummy outside C; or, more succinctly,
Wm ::J We {note the strict inclusion}.

263
 9. OVERLAPPING COMMITTEES

The seeming digression of the last two sections has

equipped us to deal with the problem of overlapping com-
mittees. In fact, the next theorem reveals that when com-
mittees overlap it is either due to the presence of "floating"
dummy players, who can safely be disregarded, or it is
due to the associativity of the operations (!) and ~,
since when three or more components are added or multi-
plied together the partial sums or products will involve
overlapping sets of players. Except for these cases p
committees do not overlap.

We shall say that C and D "overlap essentially"

if each of the sets CnD, C-D, and D-C contains a non-
dummy.

THEOREM 7. Let C and D be com-

mittees of r(N, WL that overlap essentially.
Write E for CUD, and write E l , E 2 • E3
for CnD, C-D, and D-C respectively
(see the diagram). Then E is also a com-
mittee of r(N, W) p and the committee game
r(E, WE) is e ithe r the ~ or the product
of the games r(Ei' WE.)' i = 1, 2, 3,
1
where WE. denotes the set of nonempty
intersectio~s of W with E i •

c E3

El
Ez

264
 Proof. (a) Assume first that no element of Wm
meets more than one of E l • E Z' Er To show that E
is a committee we must show that {S n E} U(S ,_ E) e Wm
for all S, S' e Wm o Without loss of generality, S'
meets Co

Subcase i: If S also meets C. then we can sub-

stitute S for S' in C and obtain
{SnC} U{S'_C} e Wm •
But by our initial assumption Sand S' cannot meet
E3; hence SnC =SnE and S'-C =
S'-E, and we have
(SnE}U(S'-E) eWm.

Subcase ii: If S does not meet C then S must

meet D. Find aTE: W m that meets El (such a
set exists because E 1 contains a nondummy). Then
T meets C i and we can substitute T for S I in C
and obtain
{TnC}U{S'-C} E: Wm •
But this set--call it R--meets El and hence D.
and so we can substitute S for R in D and obtain
{SnD} U(R-D) E: W m •
=
But SnD SOE and R-D S'-C = = S'-E. and we again
have (SnE) U(S'-E) e W m • This shows that E is a
committee. Moreover. by our initial assumption, the
minimal winning coalitions of r(E. WE) "fit" within
the partition {E l .E Z .E 3 }. so that by the remarks in
Section 8 we have r{E , WE) = r{E l • W El} GJ r{E Z ' W EZ)
ED r(E 3 , WE3).
{b} Now suppose that the initial assumption of (a)
is not satisfied. so that there exists a set Q E: Wm
that meets at least two of E l , E Z ' E 3 • Without loss
of generality, Q meets E3 and also meets C. Let
S1 E: Wm {E 1 ), Sz E: Wm(E Z) (such sets exist by the
nondummyassumption). Substituting Sz for Q in
C, we obtain

265
Substituting S1 for this set in D, we obtain
(S1nE1)U(SZnEZ}U(S1nE3}U(Q-E) E: W m •
In the committee C z E1 UE Z ' we therefore have

(S1nE1}U(SZnEZ) E:W~ ,
as well as

and

Sznc = (SZnE 1 }U(SZnE Z) E: W~ ,

all by Theorem 1. Consider these three elements of
W~. Since both S1nE1 and SZnE Z arenonempty,
we must also have SZnE 1 and S1 nE Z nonempty,
to avoid contradicting minimality. Thus we have
shown that every S E: Wm that meets E1 also meets
E Z and vice versa.

Q was chosen unsymmetrically, as between C

and D, but we now know that the re are sets in Wm ,
for example SZ' that meet both E Z and D. Using
such a set in place of Q, we can repeat the above
argument and establish that every S E: Wm that
meets E1 also meets E 3 , and vice versa. In other
words, every element of Wm{E) meets all three of
E 1 , E Z' E 3 •

It is now easily shown by a substitution argument

that E is a committee. C is therefore a committee
of r(E, WE}' by Theorem 4. Applying the test at
the end of Section 8 we find that C is a factoring
committee of r(E, W E). Hence the complementary
set E-C =E3 is also a factoring committee. Simi-
larly E Z ' the complement of D in E, is a factoring
committee; hence also E Z UE 3 and finally E-(E Z UE 3 ) =
E1 are factoring committees. Thus we have
=
r{E,W E } r{E 1 ,W E1 ) ® r{EZ'W EZ ) ® r{E 3 ,W E3 }·
This completes the proof.

266
 COROLLARY. Let G be a simple
game without dummies. Then either
G is a sum or product. or the maximal
proper committees of G are disjoint.

Proof. Let C and D be maximal proper com-

mittees g C =1= D. and assume that CnD =1= cf>. We
have C ¢ D and D ¢ C by maximality. Hence C
and D overlap essentially. there being no dummies.
Hence E:: CUD is a committee. But since E
strictly contains the maximal proper committees C
and D, it can only be an improper committee. i. e ••
the all-player set of G. But this means that G
itself is a sum or product.

10. THE UNIQUE PRIME DECOMPOSITION

Theorem 7 and its corollary pave the way for a syste-

matic determination of the entire decomposition pattern
of a simple game, once the committees are known. *
The
process starts with the full game G (from which we shall
assume the dummies have been eliminated). and works
down the hierarchy of committees until the individual
players are reached.

At the first step there are four mutually exclusive

possibilities: G is a sum; G is a product; G is decom-
posable but not a sum or product; G is prime.

(a) G is a sum. Then in the next level we install

I
the components {Gi i = 1 •••• , m} of the finest sum

*In principle thisis no problem. since the number of poten-

tial committee games to be checked out is finite. However.
it would be desirable to have an efficient combinatorial
algorithm that would discover committees, given a list of
the minimal winning coalitions or other description of the
game. At present, this can only be done for sums and
products.

267
decomposition of G. Thus, none of the Gi's are them-
sel ves sums.

It is necessary to prove here that we have not chopped

too fine, i.e., that no committee of G overlaps any of the
player-sets {N) of the {G) 0 Suppose therefore that C
is a committee that overlaps N 1 , and hence also overlaps
the set NZU ••• UNm • By Theorem 7, the set E =
CU{N Z U ••• UNm ) is a committee. Applying the theorem
again to N1 and E, we see that the whole game G is the
sum or product of games on the three sets N1 nC, N 1 -C,
and NZU ••• UNm • In facta it is the sum, since no game
is both a sum and a product. But this means that the com-
ponent G 1 decomposes into a sum, contrary to hypothesis.
Hence an overlapping committee like C is imposs ible j

and we can assert that every committee of G that is not a

Gi or a sum of Gi IS is actually a subcommittee of one of
them.

(b) G is a product. Then the components of the finest

product decompos ition comprise the next level. As above,
we can assert that the committees of G that are not products
of these components must be committees of the individual
components.

(c) G is decomposable, but not a sum or product.

Then the corollary to Theorem 7 tells us that the maximal
proper committees of G are disjoint, so we can install
them at the next level of the hierarchy, along with any
" unaffiliated " individual players that belong to no proper
committee of G. Again we can assert that all committees
of G not yet represented are subcommittees of the com-
ponents we have just installed.

(d) G is prime. Then all players are "unaffiliated .. "

so we list them in the next level and stop.

The process then continues by analyzing in the same

way each game that appears on the second level, and so on
until every branch of the tree has terminated with an

268
individual player.

The accompanying diagram is virtually self-explana-

tory. The game y is a sum of three components
G 1 (!} G 2 ® G 3 ; the first of these is a product of two
components Gil ® G i2 ; the second happens to be a 6-
person game having a 3-person committee; and so on. The
only committees not represented directly in the diagram
are the partial sums G 1 (!) G 2 , G 1 ®G 3 , and G 2 (i:) G y

The compound representation of y written beneath

the diagram is more concise. and in its own way just as
descriptive. Note that only quotients and individual players
are named. "A" is some five-person game, "D," "F."
and "G" are four-person games. and so on. (A minor
notational change would save us the trouble of writing down
the one-person games.)

Committee decomposition diagram:

0= (A[a,b,c,d,e)0(B[f,g,h) (i)c[i)))
@D[ E[ j ,k,.€ ) m, n, 0]
@F[G[p,q,r,s] ,(H[t) ®I[u] )0J[v) ,w,K[x,y,z) )

269
 The i:mportant point is that the quotients that appear
in the representation are all pri:me. For when a quotient
has a proper co:m:mittee. then players in the corresponding
co:mponents of the co:mpound for:m a proper co:m:mittee of
that ga:me. which necessarily includes the players of at
least two co:mponents in the representation. In the present
case, however, the co:mponents are :maxi:mal proper co:m-
:mittees (or unaffiliated players). so that this cannot happen.

We have thus established our final result:

THEOREM 8. Every si:mple ga:me as a

co:mpound representation that uses nothing
but pri:me quotients and the associative
operations G} and @ and that is unique
except for arbitrarines s in the ordering of
playe rs and co:mponents and in the disposition
of du:m:my players.

270
 SECTION VI

MORPHOLOGICAL RESEARCH
 CHAPTER 1

THE MORPHOLOGICAL APPROACH

TO DISCOVERY,
INVENTION, RESEARCH AND CONSTRUCTION

F. ZWICKY
California Institute of Technology

ABSTRACT

The morphological approach to discovery, invention,

res earch and construction has been conceived and developed
for the purpose of dealing with al1 situations in life more
reasonably and more effectively than hitherto. This is
achieved through the study of all relevant interrelations
among objects. phenomena and concepts by means of methods
which are based on the utmost detachment from prejudice
and carefully refrain from all prevaluations. Applications
of the morphological methods of the total field coverage, of
negation and construction, of the morphological box and
others to technical problems in particular and to human
problems in general are described. These not only illustrate
how discovery. invention and res earch can be conducted
most effectively but also how the morphological approach
makes possible the clear recognition of those fatal aberra-
tions of the human mind which must be overcome if we are
ever to build a sound world.

273
 A. The Origin and Development of the
Morphological Approach

During the past five thousand years men have fought

five thousand wars. more or less. In spite of all of the
talk about progres s we are still at it. fighting. murdering
and destroying. Many other ills, some major and very
many minor ones, confront and bother us in addition and
seem to get more difficult to handle as time goes on. Grave
problems related to the growth of the world 1 s population
remain largely unresolved. In spite of the fact that many
of the plagues which formerly decimated humanity have
been checked, the overall state of the physical health of
men hardly shows much improvement. The mental aber-
rations of individuals and of peoples clearly seem to be on
the increase. Generally life has become so complex that
we find less and less time for deep contemplation and
happy creation and recreation. Regrettably, science and
technology, in the hands of irresponsible, weak or selfish
men and exploiters have vastly contributed to the ills of
the world, rather than to their elimination or sublimation.

Under these circumstances a thorough reappraisal

of the overall conditions of the world and of man IS place in
it appears imperative to all who hold that a more satisfactory
state of affairs can only be reached if each individual and
each race has the opportunity and the means to realize its
own inherent genius without having to fear the ever con-
tinuing inhibitive or destructive interference of other indi-
viduals and races. Two principal goals must be kept in
mind if we are ever to achieve a sound society living in a
sound and beautiful world. First we must see to it that con-
ditions do not further deteriorate. To achieve this, a uni-
versal holding action is necessary in order to preserve at
least the most satisfactory aspects of the status quo.
Secondly our mental world image must be set straight and
enriched, so that we may clearly visualize the true inter-
relations among all things, material and spiritual. And
thirdly we must proceed with planning and construction
which is inspired and guided by truths thus visualized rather
than by disregarding them.

274
 In order to achieve all of these goals we must con-
ceive of an approach which allows us to implement and to
integrate our knowledge of all of the essential interrelations
among the physical objects of the world and the physical,
chemical and biological phenomena which govern their
interaction. And further we must clearly understand the
nat ure and interplay of the ideas, concepts and actions of
men. With these goals in mind, the morphological approach
was cori.ceived which concerns itself with the development
and the practical application of basic methods which will
allow us to discover and analyze the structural or morpho-
logical interrelations among objects, phenomena and con-
cepts. and to explore the results gained for the construction
of a sound world.

To give a concrete illustration for these thoughts,

a complex task may be mentioned with which we concern
ourselves for instance within the International Academy of
Astronautics. The legal-scientific committee of this
Academy is studying, the morphological aspects of Justice
in the Space Age,(l){6) including both outer and inner space,
which latter is the space below the surface of the Earth,
both its inte rior and the ocean depths. Our ultimate goal
is the formulation of a treaty which will be adopted by all
nations. Before we can hope to achieve this, a universal
holding action must be organized, agreed upon and enforced.
For instance. we must prevent all and any possible large
scale contamination of the oceans and of all natural life
within them. Nor can we tolerate any explorations and
operations within the interior of the Earth which would
endanger any stretches of the land on the surface. Likewise.
in outer space we must see to it that no irresponsible dam-
age is done to the Moon or other bodies of the solar system.
Both the holding action and the formulation of a successful
treaty for inner and outer space require a comprehensive
integration and evaluation of all present knowledge about the
physical world as well as about the positive and negative
intentions of men. We are furthermore badly in need of
observations and analyses of basic facts as yet unknown but
suspected, and we must in partic ular try to visualize the
possibility of new devices, the invention, construction and

275
use of which might completely alter the course of the
destinites of men and of nations. For instance if nuclear
fusion of any common materials, such as ordinary rocks
were achieved by a group of scientists in some specific
country, the rest of us might be forced to accept a treaty
about the future of inner and outer space as dictated by
those in power in that particular country. Obviously. the
members of the International Academy of Astronautics. as
well as all men of good will and sound mind will spare no
effort to keep themselves informed on all scientific and
technological progress. or, as we morphologists express
it, on all new pegs of knowledge. Such info rmation will be
indispensable if we wish to construct for our good use a
world image which is true in the sense that it will enable
us to fully comprehend and control the happenings around
us as well as those within us. The visualization of this
true world image can guide us on sure paths to discoveries
and to basic inventions and. last but not least it will enable
us to predict correctly the course of future events and to
adjust our actions accordingly.

It is my considered opinion that among all of the

methods of thought and of procedure the morphological
approach offers us the greatest guarantee for success in
our endeavor to construct a true world image and to make
effective use of the information embodied in it for the
elimination of all human aberrations. These, in the past
have been the cause of most major human tragedies and
they have prevented us from recognizing and constructively
promoting the genius of man.

In order to demonstrate the unique usefulness and

power of the morphological approach we now proceed to
describe a few of the most effective methods which have
been developed and applied. making use of morphological
thinking. In each case we shall briefly mention some of
the specific results achieved. For further details we must
occasionally refer the reader to the existing literature.

276
B. Some Characteristics of the Morphological Approach

1. Totality Research and Unbias *

As we have stated already, the morphological ap-
proach aims at an understanding of all of the interrelations
among objects, phenomena and ideas and concepts in ever
widening complexes of human life and of the world as a
whole. It insists on visualizing all of these interrelations
without prejudging the values of any of them. Utmost
detachment from prejudice is the first and foremost require-
ment of the morphologist. It must be emphasized that such
freedom from prejudice is far more difficult to achieve and
to live up to than most men might think. Anyone who has
tried it will certainly have found out that, in order to re-
main free and to continue to act truly independently, he
must be determined to go it alone if necessary and he must
be able to persist» even if no recognition is forthcoming and
even though abuse and criticism may be reaped instead of
rewards.

2. The Scope of Morphological Research

Since the morphological approach uncompromisingly

proceeds to explore all phases of life and all aspects of the
external world it does not take anything for granted which
has been done in the past. The morphologist always starts
from the beginning. no matter what the problem or the
task at hand may be. On the other hand. regardless of how
thorough any previous study, investigation or action appears
to have been, he will always go beyond the stage where
others have stopped. As one French writer expressf2\ it,
the morphologist is the specialist of the impossible. Let
me briefly sketch two examples which illustrate these
aspects of the morphological approach.

*Detachment from Prejudice is the indispensable prerequi-

site of the Morphological Approach. Since the English
language does not have any single words for the opposite
of bias and prejudice I propose that the word unbias be
introduced.

277
 a. The Conversion of Chemical Energy into Kinetic
Energy of a Missile

For decades, and even for centuries projectiles and

mis siles have been launched through the release of chemical
ene rgy, making us e of either guns and rockets, although the
thermopropulsive efficiency of both devices is miserably
low when compared with what should idea11y be pos sible.
According to the fundamental principles of thermodynamics.
if we let the chemicals A. B, C.etc. react to produce the
substances AI, BI, C',etc. the difference ~<I>= <I> - cp'
released by this reaction at constant external pressure can
be extracted as useful mechanical or electrical work or
energy. Here cp is the total free enthalpy of a11 of the
reagents A, B, C,etc. and cpl is the free enthalpy of the
reaction products. By definition it is <I> = E - TS + pV,
where E and S are respectively the total energy and
entropy of the system, while T, P and V are the tem-
perature, the external pr6ssure and the volume. If, as an
=
example, we cons ide r the well known reaction ZHZ + 0z ZHZ °
we find that per mole of HZO formed the free enthalpy
=
6<1> 56.7 kcal is liberated at 8PT. This amounts to
3.15 kcal/gram = 13.1 X 10 10 ergs per gram of the original
prope11ant mixture being available for transformation into
useful mechanical or electrical energy. This means that
by reacting one gram of the gaseous mixture of ZHZ + 0z
to HZO at constant pressure, we should be able, theoreti-
cally, to accelerate a projectile which weighs one gram to
a speed of 5.1 kilometers per second. Or, using n grams
of one of the conventional chemical solid or liquid propellant
mixtures or explosives the free enthalpy released (which is
of the order of Z kilacalories per gram) should allow us to
launch a one gram projectile with a velocity of about 4.fn
km/sec. Now it is a well known fact that the maximum
muzzle velocities of guns are only about one km/sec, ~
matter how much gun powder we use for each shot, that is
no matter how great the number n is.

And the thermopropulsive efficiency of rockets,

especially that of multistage rockets is poorer yet. For

278
instance. to get one gram of a payload off the Earth into
interplanetary space requires theoretically 6 X 10 11 ergs
(plus an additional amount of a few per cent to compensate
for the frictional losses of the large rockets in traversing
the atmosphere). m present day practice of launching
spacecraft as the final stages of multiple rockets. per gram
of payload 100 grams of propellant are needed which release
a useful energy of 100 X 1011 ergs. Over 90 per cent of
this available energy is therefore wasted in the conventional
types of rocket launchings of artificial satellites and space
capsules.

Engineers and scientists who have occupied them-

selves with the theory and practice of launching missiles
either from guns or by means of rockets will no doubt object
that. although the principles of thermodynamics admit of the
full conversion of the free enthalpy released into useful work.
this can be achieved in practice only under very special con-
ditions. For instance. if a chemical reaction is reversibly
carried out in an electrolytic cell (power cell, propellant
flow battery) the energy of the generated current will be very
nearly equal to the full value of 6<1> released. provided that
the current intensity is very small. This current, or the
electrostatic potential energy of a condensor which is
chal'ged up by the reaction in the cell can be used subse-
quently to accelerate a small "miss ile, " for instance a
charged colloidal particle with a minimum loss of energy.
Most specialists in the field, however, seem to be under
the impression that the thermopropulsive efficiency of
large scale conversions of chemical energy into kinetic
energy will. in practice.never approach the theoretically
possible values and must actually remain at the miserably
low level of the rockets of today. They consequently appear
to have given up all hope and have made no efforts to think
any further.

But the morphologist will not let himself be stopped

by any of the conventional objections. He will emphasize
the importance of one of the most fundamental directives of
the morphological mode of life which holds that nothing
should be accepted as impossible unless it is clearly proved

279
to be impossibleo As a consequence he will attempt to
visualize all of the possible ways of converting chemical
energy into kinetic energyo In parenthesis I like to point
out that comprehensive reviews of this type are seldom
given in the textbooks or in lecture courses. although they
would be highly instructive. The teaching profession in
science and engineering should seriously occupy itself with
this subject and explore it in all of its vast aspects
0 Here
I must restrict myself to a brief sketch of some typical
considerations which have led to rather startling results.

Returning to the fundamental aspects of our problem

of the conversion of chemical energy into kinetic energy we
must first attempt to achieve a total perspective over the
geometrical, dynamical and chemical characteristics which
may be involved in the construction of launching devices.

Starting with the geometry of poss ible launchers we

see that there are two principal bodies involved, that is the
missile M and the propellant or explosive P. A tube T
may be needed in addition to act as a container or as an
initial guide. We further designate with V the arrow which
indicates the direction of the velocity with which the mis sile
is to be launched. Considering only M and P, three typical
locations of these two bodies with respect to the arrow V
are possible. These are shown inthe diagrams la, lb and
1 co

Diagram

a)
®® - - V ---+-

b)
© - - V ---+-

c)
@® --V--

The first arrangement a) schematically represents

the initial locations of the propellant and the missile within

280
a gun, the missile being in front. This method of launching
has the drawback tha t no muzzle velocity can ever be achieved
which is materially greater than the average thermal velo-
cities of the molecule s of the reaction products at the peak
temperatures reached after the explosion of the pr opellant
charge. For molecules such as N 2 , H 2 0, CO 2 as they
are relased by the common chemical explosives these
velocities are of the order of 2 km/sec. Once the pro-
jectile has reached this speed it will not be accelerated
any further since the molecules are too slow to follow and
cannot transfer any more forward momentum to the missile.

The second arrangement b) schematically represents

a rocket, the propellant being located within the missile
itself. Rockets can be accelerated with optimum propulsive
efficiency relative to any given inertial reference system ~
only at the moment when their forward velocity with respect
to ~ is exactly equal and opposite to the velocity u e of
their exhaust gases with respect to the same frame ~.
For smaller and for greater values of v the propulsive
efficiency decreases rapidly and becomes zero for both
v = 0 and for v = very large compared with u e •

If we designate with Mo the mass of an empty

rocket and with Mp the mass of the propellant loaded in
it, we may accelerate the rocket in vacuum from standstill
to a maximum velocity v max which is given by the
equation (1)
v = u log (M + M )/M ( 1)
max e e p 0 0

For a high energy chemical propellant the exhaust velocity

is of the order of u e = 2.5 km/sec. To achieve therefore
in vacuum a terminal rocket velocity of v max = 11.2 km/sec.
which is equal to the escape velocity from the Earth we
must react 88 grams of the said propellant per each gram
of the empty rocket. Ideally, however, less than 20 grams
should be necessary if we also remember that the com-
bustion of the propellant would actually produce a higher
exhaust velocity than u e = 2.5 km/sec if the whole of the
available free enthalpy had been transformed into kinetic

281
energy of the exhaust gases. Under the best conditions,
and using a high energy propellant capable of producing
the assumed exhaust velocity we therefore can achieve
escape velocity only with a thermo propulsive efficiency
of less than 20 per cent. For multi-stage rockets the
losses are still greater since the empty cases of the initial
boosting rockets will be jettisoned into space with con-
siderable kinetic ene rgies with respect to the frame .6,
relative to which the missile was launched. This energy
is entirely wasted, in addition to that carried away uselessly
by the exhaust gases of the various stages.

A thorough scrutiny of the potentialities of both of

the arrangements a) and b); that is of the gun and the rocket
does not leave us with any hope that a particularly more
complete conversion of chemical energy into kinetic energy
of high velocity miss iles can be accomplished with their aid.
There thus remains only the possibility c) of positioning
propellant and missile, as shown in diagram 1.

The third arrangement c) of placing the propellant

in front of the missile would at first sight appear to be com-
pletely hopeless if not utterly stupid. This is no doubt the
reason why it has never been even given any thought.
Remarkably enough, however, once one decides to explore
its possibilities seriously he not only begins to visualize
an unexpected number of devices but, among them also
various types of launchers which are potentially capable of
working with the greatest thermopropulsive efficiency.

Some devices of the type c) are as follows.

a) The propellant P may be placed in front of a missile M
which lies on a disk D fixed to a spring S which is attached
to a solid wall representing the reference system L:.
(Figure 2)
On explosion of P the spring will be compressed and on its
rebound propels M in the direction V. This launching
device works in vacuum, in air or under water. Many vari-
ations are imaginable inasmuch as the energy released by
the explosion can be stored temporarily as gravitational,
electric or magnetic energy, rather than by a spring.

282
 Wall

~) If our device is in the atmosphere we can dispense with

the disk and spring and put M directly on the wall. The
explosion of P will be followed by an overexpansion of the
air in front of M which thus will be propelled in the direc-
tion V by suction.

I leave it to the reader to amuse himself with the

invention of many other pos sible launchers in which P is
placed in front of M. Among these one of the most promis-
ing is the following.

,,) The tube-duct launcher

To achieve maximum propulsive efficiency the pro-

pellant P, which is at rest with respect to M at the start
must be picked up and accelerated by the missile itself in
such a way that the work W done in this process is stored
temporarily, through stagnation compression for instance.
On reaction of the compres sed propellant W will then be
released together with the originally available free enthalpy
~cp from P. The liberation of ~cp + W will generate an
exhaust velocity u e which increases .with the forward
velocity v of the missile, rather than remaining constant
throughout the flight as in ordinary rockets. An aeroduct
(ramjet) or an aeropulse(3) which can slide along a closed
launching tube filled with a gaseous propellant is suitable
as a mis s ile which will "swallow" up P and sto re the
energy thereby expended through compression in the stag-
nation cross section. The exhaust gases will remain at
rest with respect to the launching tube (.6) and will not
carry away any kinetic energy as in the case of rockets. (4)

283
 b. Nuclear Fusion Ignition

Much work has been done on the possibility of nuclear

fusion within constricted plasmas. Although the liberation
of nuclear ene rgy through the fus ion of light elements into
iron in ordinary rocks for instance would perhaps represent
the most important technical achievement of all times; no
concerted effort seems to have been made by anybody to
review all imaginable processes for igniting nuclear fusion
in common materials. This task. which obviously neces-
sitates a morphological sCllutiny of a totality of phenomena
has deeply interested me for many years. Of the results
I have arrived at I can mention here only the idea of ac-
celerating small solid particles to velocities of up to 1000
km/sec. Such particles, impacting on rocks for instance,
could generate temperatures of hundreds of millions of
degrees and readily ignite nuclear fusion. This suggests
an auxiliary morphological study. analogous to the one
presented in the preceding section a) of the various devices
which might be capable of producing very high speed par-
ticles. This study is now so far advanced that I am inclined
to predict that the ignition of nuclear fusion in common
mater ials will be achieved in the not too distant future.

c. The Morphological Methods

As I have already stated, the morphological approach

is based on an attitude which uncompromisingly and inexorably
pursues its quest for all of the interrelations among objects,
phenomena and concepts without permitting the interference
of any prejudices or rash prevaluations. In order to achieve
these goals morphologists have developed and applied a num-
ber of powerful methods of thought and of practical procedure.
Among these we mention the methods of the morphological
box, of the systematic field cove rage, of negation and con-
struction. of the method of the extremes, the reasoning and
action by analogy, the various procedures of generalization
and of successive approximation, the principle of the flexi-
bility of scientific truth, the principle of perfection and
imperfection and others. In the following we briefly sketch
some of these methods and a few of the results achieved.

284
1. The Method of the Morphological Box

We proceed as follows.

First Step. The probleTIl which is to be solved TIlust be

exactly forTIlulated.
Second Step. All of the paraTIleters which TIlight enter into
the solution of the given problelTI TIlust be
localized and characterized.
Third Step. The TIlorphological box or TIlultidiTIlensional
TIlatrix which contains all of the solutions of
the given probleTIl is constructed.
Fourth Step. All of the solutions which are contained in the
TIlorphological box are closely analyzed and
evaluated with respect to the purposes which
are to be achieved.
Fifth Step. The best solutions are being selected and are
carried out, provided that the necessary
TIleans are available. This practical appli-
cation requires an additional TIlorphological
study.

The TIlethod of the TIlorphological box has been used

with great success for a cOTIlprehensive review of propulsive
power plants and propellants. (3) for the enuTIleration and
analysis of all of the processes that can be applied to indi-
vidual nylon fibers (in connection with the invention of the
curly yarn and the faTIlous Hellanca fabrics), to all of the
processes that can be used to produce the curly yarn, to
the const ruction TIlethod s fo r straight line s and plane sur-
faces, for the invention of cOTIlposite analytical tf1hotograPhy.(5)
the totality of all possible energy conversions,{ ) the collec-
tion and shipTIlent of scientific journals to warstricken
scientific libraries, (6) the problem of bothersoTIle inquiries ,(5)
rock and ice cliTIlbing, shorthand systeTIls, the systematics
of nOTIlenclature, the construction and dissection of regular
polyhedra and other geometrical probleTIls. (6) A TIlost use-
ful classification of scientific procedures was developed(5)
through the characterizations of diTIlensionless TIlorphology,

285
phenomenological morphology and absolute morphology.
Also, the system of multi-language teaching which will be
discussed in another report to this symposium was in part
developed with the aid of the method of the morphological
box.

Historically, the first large scale task which has

been solved with the method of the morphological box is that
of the totality of all possible propulsive power plants which
can be activated by chemical energy. In 1939 some of my
colleague s of the California Institute of Technology in
Pasadena, California, under the direction of Professor
Th. von K~rm~n and Drs. F. Malina and M. Summerfield,
with the support of the US Air Force established the
Guggenheim Aeronautics Laboratory of the California
Institute of TechnOlogy (GALCIT), now called the Jet Pro-
pulsion Laboratory (JPL), where they occupied themselves
with the problem of jet assisted take off rockets (JATOS)
for airplanes. The Aerojet Engineering Corporation (now
Aerojet-General Corporation) was subsequently charged in
194Z with the production of the solid and liquid propellant
rockets which had been developed at GALCIT. The engineers,
however, found it difficult to scale up the GALCIT models.
As director of research of the Aerojet Company I therefore
started a comprehensive study of all possible jet engines
which can be activated by chemical energy and I explored
their various characteristics in terms of the following six
parameters P z.
P1 = The medium through which the jet engine moves. Four
components P 11 , P 1Z ' P 13 and P 14 were chosen to
de s ignate vacuum, the atmosphere, large bodie s of
water and the interior of the solid surface strata of
the Earth.
The type of motion of the propellant relative to the jet
engine. P Z1 ' P ZZ ' P Z3 ' P Z4 thus designate rest,
translatory, os cillatory and rotatory motion.
Physical state of the propellant. This parameter has
three components P 31 = gaseous, P 32 =
liquid,
P 33= solid.

286
P4 = Type of thrust augmentation. P 41 = None, P 42 = inter-
nal, P 43 = exte rnal.
P5 = Type of ignition. P 51 = self igniting, P 52 = external
ignition.
= Sequence of operations. P 61 = continuous, P 62 = inter-
mittent.

Any of the 576 symbolic chains (P 1 i' P 2j , P 3k' P 41 '

P 5m • P 6n ) where i and k can assume the values 1 to 4,
j and 1. the values 1 to 3 and m, n the values 1 and
2 represent a possible propulsive power plant which can
be activated by chemical propellants. (3) At the time of my
study only five of these, namely the solid and liquid propel-
lant rockets, the aeroduct (ramjet), the reciprocating engine-
propellar combination and the Campini piston engine-jet
were known. The morphological review had therefore re-
vealed the possibility of an additional 571 propulsive power
plants which had apparently escaped the attention of the
professional engineers. *
Unknown to us, however, the
British and Germans were working on the aeroturbojet
and Schmidt in Munich on the aeroresonator (buzz bomb
engine) •

With hundreds of new jet engines in view it was ob-

viously not pos sible to carry out step number four of our
program with the aid of any conventional procedure. With
regard to the evaluation of the propulsive efficiency. another
pleasant surprise was in store for us which is characteristic
of the morphological approach. Under neglect of irrelevant
details and complications it proved pos sible to derive a good
approximation for the specific impulse of any jet engine by

*In addition to theUS patents Nos. 2 426 526, 2 433 943,

2 811 431,2 815 271,2 821 838,2 914 913,2 932 943,
2 974 626,3 044 252,3 044 253,3 048 007,3 121992,
3 135 205, 3 137 994, 3 253 511, and the British Patent
No. 602 807 which were granted the author, his associates
at the Aerojet-General Company were issued many more
patents On the new jet engines and propellants.

287
using only first principles of mechanics and of thermody-
namics. I proposed to call the resulting expression the
universal thrust formula(3}
U 2 1/2
I :: U /g::-f3o[f3-1+(1-f3+2f3LH:jU)] (2)
sp e 0

where U is the velocity of the missile with respect to the

o
surrounding medium. .6€ is the chemical energy available
per gram of fuel (hydrofuel, terrafuel) and f3:: Mf/(M f + M m },
where M f and Mm are the mass flows per second of fuel
and external medium (air. water. earth) through the exhausts
of the propuls ive power plant.

From 1943 on we proceeded at Aerojet to build as

many as possible of the newly invented engines. Some of
them proved right away to be surprisingly successful.
Many, however. for instance the hydropulse and the hydro-
turbojet will only corne into widespread use when cheap and
efficient hydrofuels. that is water reactive chemicals such
as Al(BH 4 }3 become available (or once they are combined
with nuclear reactors). In order to implement the knowledge
gained from the morphological box of the propulsive power
plants we therefore had to initiate a systematic study of all
pos sible stable and pseudostable chemicals. This led to the
introduction of many new chemicals and opened up the very
pro(~fing fields of fragment chemistry and of metachemis-
try.

The study of morphological boxes also suggests the

introduction of systems of nomenclature which at the present
time is in such a state of confusion that much time is being
wasted by librarians. engineers and scientists in locating
needed information. The box of the jet engines suggests for
instance that one should designate all devices moving through
the air. water or the Earth respectively by the prefixes aero,
hydro and terra; thus: aeroresonator, aeropulse. aeroturbo-
jet. aeropistonjet and hydroresonator. hydropulse. hydro-
turbojet, hydropistonjet. and terraresonator. terrapulse,
terraturbojet, terrapistonjet.

In expansion of the studie s on poss ible propulsive

288
power plants and propellants we started at the end of
World War II work on various vast projects of space re-
search. The first one among these was an attempt on
December 17. 1946 at the White Sands Proving Grounds
to fire artificial meteors from shaped charges mounted on
a V-2 rocket. (3) (7) The first man-made objects (slugs of
fused TiC and A1 2 0 3 ) were subsequently successfully fired
into interplanetary space and free of the gravitational pull
of the Earth from an Aerobee rocket which was launched at
Alamogordo. New Mexico,on the night of October 16. 1957.
These may now appear to be modest efforts when compared
with later spectacular achievements, but they nevertheless
remain the first practical attempts to start the SPACE AGEJ3)

2. The Method of the Systematic Field Coverage

If we start from given pegs of knowledge and infiltrate

the surrounding fields until we arrive at other known pegs,
we may expect to make discoveries and prove at the same
time that nothing has been overlooked. This is really the
procedure used by a good detective in tracking down crimi-
nals. During the past thirty-five years I have applied this
method successfully in the search for new cosmic objects
and phenomena.

The main guiding principles which I used are the

fOllowing:
a) All objects may be expected to form families whose
members exhibit continuous sequences of characteristics.
If some types appear to be missing. they most likely are
short-lived objects. but it must nevertheless be possible
to find them. If, however. some objects are really non-
existent. such as for instance, stable and electrically
neutral nuclei. or clusters of clusters of galaxies this
must be due to some very fundamental reasons.
~) If the general principles of statistical mechanics are
applicable to the large scale distribution of cosmic
matter we must expect much of this matter in resting
places and configurations which are long lived, that is
in very compact and in very dispersed states. Actually

289
 there is a close connection between these two inasmuch
as during the formation of compact bodies energy must
be liberated in the form of electromagnetic and gravi-
tational waves, corpuscular radiation such as the cosmic
rays and through the ejection of various types of material
bodies.
y) Compact bodies may be formed both slowly or rapidly
(implosively) , the terms slow and fast being defined by
the statement in the following paragraph.

6) When we talk about the evolution or the history of a sys-

tern (or of the world) we mean that certain pure numbers.
which are dimensionless ratios of significant physical
parameters of the system change their values as functions
of tiT, where t is the time and T is a characteristic
time within the original system.
E) For any large scale search to be effective, the proper
instruments must be available. For my purposes I
promoted the construction of the 18-inch and the 8-inch
Schmidt telescopes on Palomar Mountain, which were
followed thirteen years later with the installation of the
48-inch Schmidt. In addition I composed mechanically
the first full size mosaic echelette transmission ratin 3)
fo r the 18- inch Schmidt telescope with the aid of 5 inch X
7 inch replicas kindly sent me by Professor R. W. Wood
who later on constructed three additional and more per-
fect objective gratings. Unfortunately no full size prism
or grating has ever been built for the 48-inch Schmidt
telescope, a neglect which almost certainly deprived us
of the early discovery of objects like 3C 273 and of a
much more effective search for flare stars, supernovae
and compact galaxies of all types (BCO = BSO = blue com-
pact extragalactic objects, RCO = RSO = red compact
extragalactic objects; ECO = emission-line compact
objects ,etc. )

The discoveries(5) which I succeeded in making as a

consequence of the program of the morphological field cover-
age for new cosmic objects include among others the first
dwarf, pygmy and gnome galaxies, the first compact galaxies

290
of all types, including the BCO, (which are called BSO or
blue stellar objects by some) and the more numerous red
and infrared RCa (RSO) and ICO (ISO), many groups of
compact galaxies, as well as the apparently rare large
clusters of compact galaxies. I could also show that inter-
galactic matter, the existence of which had been vehemently
denied by E. P. Hubble, W. Baade and other experts in the
field(3) is distributed throughout all of cosmic space between
the commpn galaxies and that this matter consists of stars,
dust, gases and plasmas of protons and electrons, in addi-
tion to the cosmic rays. New classes of galactic objects
such as the Humason-Zwicky stars, pygmy stars and certain
variables were also discovered.

Some of the hypothetical cosmic bodies whose existence

I predicted theoretically on the basis of principles developed
by the method of the systematic field coverage, however,
still remain to be discovered or identified. These include
the neutron stars, (8) the nuclear goblins, (9) the neutron
star studded gaseous compact galaxies, (10) the pure cosmic
balls of light P 1) (12) and the neutron star studded cosmic
balls of light. These ultracompact objects which contain
mostly neutron stars and either gases or trapped radiation
are in my opinion the most likely candidates for models of
the quasistellar radio sources and certain blue, red and
infrared radioquiescent compact galaxies whose large red-
shifts are probably explainable as Einstein gravitational
shifts.

3. The Method of Negation and Construction

This method is based on the principle of the flexibility

of truth as I formulated it some thirty-five years ago. (13)
In particular, any communicable statement, which of neces-
sity must be formulated in finite terms cannot be absolute.
Thus, for instance, the uncertainty principle in quantum
mechanics cannot really represent an ultimate truth because
its claim that products of the uncertainties in the simultane-
ous determina,tion of coordinates and momenta, or of energies
and life times can never be smaller than Planck's constant
h is too inflexible. I predict that a determined effort

291
to negate the absolute certainty expres sed by the uncertainty
principle, (3) followed by the formulation of a law which is in
better accord with the principle of the flexibility of scientific
truth will lead to new fundamental insights in theoretical
physics.

The method of negation and construction is also most

helpful in many rather prosaic circumstances of daily life
and it can be particularly useful in breaking down prejudices
and beliefs which so often impede scientific and technological
progress. As an example among hundreds I mention the pro-
blem of the photographic plate whose quantum efficiency in
recording optical images is miserably low inasmuch as it
takes a thousand light quanta or more to produce one chemi-
cally developable grain of silver bromide for instance. It
seems that most scientists have tacitly accepted this as a
fact about which nothing can be done. No morphologist. how-
ever. will agree to any such defeatist conclusion without
being shown an unambiguous proof of its validity. Actually.
however. such a proof cannot be given at all and the possi-
bility of recording each individual light quantum photographi-
cally remains a certainty. Once one has adopted this attitude
many solutions suggest themselves. One of the most obvious
consists in the idea of dispersing colloidal droplets of an
explosive. for instance. properly colored and doped nitro-
methane (Jet-X) throughout the emulsion. Every quantum
of light will then initiate a microscopic explosion and thus
release many times its own energy which can be used for the
production of photographic image s.

4. The Method of the Extremes

Truth can at best be approximate and "osculate" reality

closely only in certain limited ranges of parameters. just
as a tangent or a circle of curvature osculate a general
curve only in the neighborhood of some chosen point. At
far away points departures of the curve from the tangent or
the circle become obvious. To discover deviations from any
apparently absolute truth we must therefore consider extreme
sets of circumstances. For instance. we may ask if Newton's
law holds at distances which are either very small or very

292
large cOITlpared with those for which we have tested it. As
a result of this inquiry I found that there are no clusters
of rich clusters of galaxies. (5) The ITlost direct interpre-
tation of this fact leads to the conclusion that the gravitational
attraction between bodies which are separated by ITlore than
fifty ITlillion parsecs decreases ITlore rapidly than Newton's
law would indicate. Likewise» Euclidian space ITlay be a
good approxiITlation for those liITlited regions within which
we usually operate but it probably does not represent real
space any ITlore if we deal with nuclear diITlensions or the
universe as a whole.

5. Confrontation of Perfection and IITlperfection

Although the ITlorphological approach generally aiITls

at perfection it also is deterITlined not to neglect anything.
In this spirit .it is interested in the evaluation of all iITlper-
fections. This outlook leads often to surprising results.
Thus. a telescope whose iITlage forITlation of celestial objects
is badly affected by chroITlatic errors. cross diffraction
patterns» and ITlultiple reflection ghosts is generally con-
sidered a partial failure. On closer scrutiny it ITlay be found
that the ITlentioned defects are useful and sOITletiITles even
necessary for ascertaining the reality of objects, for the
discovery of stars of certain colors, for the calibration of
apparent ITlagnitudes, for the siITlultaneous ueterITlination of
the lUITlinositie s of bright and of faint objects» for the dis-
tinction of stars and compact galaxies and many other prob-
leITls.

As to the value of shortcoITlings in daily life we only

need to reITleITlber that ITlany a ITlarried couple has broken
up because of what seeITled to be perfection in the character
or the behavior of one of the partners becoITling unbearable
to the other.

A ITlorphological study of the characteristics and the

pos sible values of iITlperfections of physical objects and
devices» of various types of phenoITlena and of certain ideas,
concepts and procedures is under consideration by SOITle of
the ITleITlbers of the Society of Morphological Research.

293
6. The Method of Generalization

Generalization is one of the most useful procedures

practiced by the morphological approach. It helps us to
achieve broad vistas and to derive solutions of specific
problems more easily by starting from more general ones.
What is most important. however. is that bold generalization
enormously stimulates the imagination and often yields un-
expected results. The idea of simultaneous multilanguage
instruction, which I shall discuss in a next chapter is a case
in poil1:t. Here I restrict myself to a brief sketch of one im-
portant is sue for the future of man and his society.

It is generally assumed that there are only relatively

few men and women of genius. As soon as one starts ponder-
ing this statement seriously he finds that it is difficult to
say exactly where genius starts and ends. In way of a vast
generalization I therefore became interested to evaluate the
merits of the universal proposition that EVERYONE IS A
GENIUS. (5) (6) which I intend to expound soon in a book of
the same title. Be it said here only that I consider it as the
most important task for morphological research to explore
this is sue thoroughly with a view to reshaping all of our
sterile ideas about education and the construction of a har-
monious human society.

D. The Choice of Tasks

Life confronts us with many problems which we cannot

disregard. The morphological approach promises to be
helpful in solving any of them. Even if we are successful
in dealing with these unavoidable probl ems. however, the
question remains as to what our destiny really is and what
ultimate values we should strive for. In order to make the
correct choice of the goals which one should try to achieve
he must first recognize what his genius really is and he
must also clearly know what all of the possible human values
and valuations really are. One of the major tasks of mor-
phological research therefore is to establish the morpholo-
gical box of all human values and to recognize their various
characteristics. Everyone who has been lucky enough to

294
realize his very particular genius can then choose from this
box those tasks to whose solutions his genius can contribute
most. If I were asked as to what I personally have in mind
I should say that I am most interested in the elimination of
all human aberrations. On the way to the realizati on of a
sound and harmonious human society. however. all men and
women will have to make their own unique contributions.
and nobody else will be able to take their place if each one
puts his very own genius to work.

295
 REFERENCES

(1) F. Zwicky, "The Morphology of Justice in the Space

Age," Inaugural address delivered at the Meeting of
the International Astronautical Federation in
Washington, D.C •• October 3, 1961. Published in
Proceedings of the Fourth Colloquium on the Law of
Outer Space, Oct. 3-4, 1961, Washington. D.C.
Univ. of Oklahoma Research Institute, 1963, pp. 1-4.

(2) Pierre de Latil et Jacques Bergier, QUINZE HOMMES

UN SECRET. Paris, Gallimard, 1956.

(3) F. Zwicky. MORPHOLOGY OF PROPULSIVE POWER.

Monograph No. 1 of the Society for Morphological
Research. Pasadena, 1962.

(4) F. Zwicky, US Patent No 3 253 511, "Launching Pro-

cess and Apparatus. "

(5) F. Zwicky. MORPHOLOGICAL ASTRONOMY. Berlin,

Springer, 1957.

(6) F. Zwicky. ENTDECKEN, ERFINDEN. FORSCHEN

IM MORPHOLOGISCHEN WELTBILD, Munich,
Knaur-Droemer. 1966.

(7) A. G. Wilson, "Anniversary of a Historic Failure,"

Engineering and Science Magazine. Calif. Institute
of Technology, Pasadena, December, 1966.

(8) The Physical Review~ 55, 726-743 (1939).

(9) Max Planck Festschrift 1958. 243 -251.

(10) Padua Symposium of the Giornate Galileiane (G. Barbera.

Firenze 1966).

(11) "Compact and Dispersed Cosmic Matter" to appear in

Vol. V of ADVANCES IN ASTRONOMY AND ASTRO-
PHYSICS, ed. by Z. Kopal, New York. Academic Press.

296
(12) Proceedings of the lAU Sym.posium. at Byurakan 1966
(to appear). Also Com.ptes Randus 261, 649 (1965)
and 262, 1566 (1966).

(13) The Physical Review, 43,1031 (1933).

297
 CHAPTER 2

MORPHOLOGY AND MODULARITY

A. G. WILSON
Douglas Advanced Research Laboratories
Huntington Beach, California

The morphological approach is not only a method-

ology for solving problems, it is an attitude toward problems.
It is an attitude that demands that no problem be considered
in is olation of all relevant contexts. It is an attitude that
would try to take off our customary blinders before looking
at the problem. It tries to obtain an unfiltered view by
comparing views through as many different filters as
possible. It looks for all possible solutions by also looking
at many of the impossible ones. It attempts fresh views of
the problem by looking at similar problems. In short. the
morphological approach uses whatever methodologies are
available to arrive at the most complete and unbiased
representation of the structure of the problem and its solu-
tions as is possible.

This attitude will be recognized as basic not only

to the morphological method but to some of the other
methodologies described in this symposium. In order that
ground previously covered not be repeated t this paper will
restrict discussion to examples of two important method-
ological procedures not hitherto considered. The first of
these is an exercise toward the development of useful non-
mathematical modeling. The second is an example of

298
hypothesis generation through morphological analogy.

10 NON-QUANTIFIED MODELS

One of the extens ions of problem solving capability

needed for many important problems today is the develop-
ment of methodologies for handling problems not easily
quantified.

It frequently happens that many of the parameters

that we know have relevance to a problem are not readily
amenable to measurement or quantification. There is a
tendency to concentrate on those parameters for which
numerical values are obtainable and to neglect those
parameters which are not measurable even though their
relative weight in the problem may be high. To offset this
tendency a methodology is required by which we are some-
how able to incorporate the effects of those parameters
which cannot be quantifiably represented. Examples are
esthetic, ethical and moral values, psychological factors»
and unquantifiable requirements of the future.

When standard numerical methodologies fail, or

when non-quantifiable factors must be taken into account,
a relevance type morphological procedure proposed by
Alexander and Manheim may be applicable. The idea is
based on the predicationthat any form or structure may be
thought of as resulting from the interaction of a set of
abstract forces or tendencies. These are general, not
merely physical~ forces. They may be quantifiable or
unquantifiable~ with no restriction on their variety. The
totality of these forces generates a solution that reflects
the contribution of each. The problem is to find a repre-
sentation of the forces that allows them to be combined.
Said in another way the problem is posed in an abstract
space in which the representative elements are the gener-
alized forces. The aggregate of such elements defines a
form. If the aggregate is complete and in balance, the
form becomes a stable object or solution.

299
 A concrete example of this approach attempted by
Alexander and Manheim may be found in the MIT report
entitled, "The Use of Diagrams in Highway Route Location."
Alexander and Manheim I s problem was to locate the route
for a freeway covering a 20 mile stretch in Massachusetts
starting at Springfield and ending somewhere near North-
hampton. They first morphologically derived all of the
individual abstract forces whose interaction would deter-
mine the path which the freeway should take. Shown on
Table 1 of the freeway design parameters ~ is the goal or
objective of the study. which was a freeway to meet major
current traffic desires. In this case the aggregate solution
was restricted to be a new freeway. rather than a morpho-
logical examination of all possible solutions to meet current
traffic requirements. This new freeway had to be considered
in the context of its interaction with existing freeway systems
and in support of the competition with other transportation
systems. Future transportation systems as visualized also
had to be given representation. However. the largest num-
ber of constituent forces fall into two classes; those which
determine the internal structure and behavior of the free-
way. and those reflecting the interaction of the freeway
with the environment. Table 2 of freeway design param-
eters shows the decomposition of the internal and environ-
mental parameters into their different values. Under
internal parameters. are first the construction parameters
including earthwork costs. bridge costs. pavement and sub-
grade costs. and construction interference. Secondly.
there are economic factors: land costs. public financial
losses. user costs. obsolescence; and thirdly, operational
factors: travel time, local accessability, safety, mainte-
nance, and self-induced congestion.

The environmental parameters may be divided into

physical. economic and esthetic. The physical environment
includes questions of drainage patterns and catchment areas.
effects of weather. air pollution. The economic environ-
mental factors include the effect of the freeway on regional
and local land development. public and private losses» such
as the obliteration of historical. commercial. or other struc-
tures due to the routing of the freeway. Finally, esthetic con-

300
siderations such as eyesores and noise must be considered.

Certainly not all of these parameters are easily

measured, nor is it even possible to assign numerical
values to some of them. Alexander and Manheim developed
a method by which each factor would be reflected in the
overall selection of the freeway route. They employed a
modification of the method Zwicky has termed compos ite
analytical photography. Each of the modular forces listed
on the charts by itself favors a particular location for the
highway. For example, consider earth work costs. The
requirement to minimize earth work favors the location
of the freeway in areas where the land is relatively flat.
A transparent map is made in which the flat portions are
rendered dark and the hilly portions light, the degree of
hilliness and flatness can be represented by a correspond-
ing density or opacity on the map.

Thus the tendency or force to locate the freeway in

accordance with the minimization of earth work costs is to
put the path in regions of maximum density on the map.
Similarly for each of the other forces. If a separate
transparent map of each of the forces which contributes
to the location of the freeway is made so that the dark area
favors location and the light area rejects location; if the
forces are then combined through the process of composite
photography t the resulting density on the photograph made
from superimposing all the individual photographs would give
the location that all the forces in combination tend to favor.
The darkest strip would mark the best route.

By using this method, those parameters or forces

which cannot be quantified can be weighted either through
the density used on their representative maps or through
the way in which the maps are superimposed. A subset of
three or four parameters given equal weight and dens ities
can be combined to produce a composite density which might
then be reduced in order to adjust the joint weight of the set
before combining with the maps of other parameters or sets.
The structuring of the combinations thus provides the ability
to weight the various factors.

301
 II. MORPHOLOGICAL ANALOGY

a) Artifical and Natural Systezns

An interesting eznergent property of recent tiznes

is that the source of new concepts and basic scientific
knowledge is not only the natural order but also the struc-
tures and organizations created by znan. The collection
of concepts which are called "cybernetics" were derived
jointly frozn the study of aniznal nervous systezns and znan_
znade control systezns. The idea of inforznation cazne frozn
the study of special coznznunication networks but znerged
with the concept of entropy. Today the iznportant basic
concepts underlying structure and organization are being
brought to light by the designer of coznplex systezns as well
as by the observer of the natural order. In the sense of
discovery vis-a-vis application the historic distinction
between science and technology is thus tending to disappear.

As a result of the abstract paralleliszns between

natural and artifical systezns we are able to create objects
for study which provide us with the equivalent of new views
of the natural order with teznporal and spatial resolving
powers hitherto unavailable. For exaznple. the freeway
provides us with a new type of fluid. called traffic. whose
properties can be znade useful to us in developing znore
coznprehensive theories of fluid dynaznics. extending to
new realzns the laws of fluids as observed in nature. The
growing saznple of such structures and organizations which
have been znade available for study as a result of our own
creations provides still another positive feedback contribut-
ing to the accelerated developznent of science and technology.
In effect we are creating another powerful episteznological
znethodology siznply through constructing and studying sys-
tezns that occupy sozne of the gaps in the natural order.
(Even the reasons for the natural gaps znay be learned in
tizne if our creations prove unstable. )

b) Hierarchical Modular Structures

One pos sible source of inforznation on systezns of

302
com.plex structures is in the analysis of how complexity
and bigness are treated in the natural order. We observe
throughout nature that the large and complex is constructed
in a hierarchical modular manner from the small and
simple. Direct confrontation of the large and small is
avoided, a hierarchical linkage is always interposed.
Bigness is avoided in the sense that the ratio between the
size of any structure and the modules out of which it is
built is functionally bounded. If there are demands for a
structure to continue to grow in size or complexity. then
a new level in the hierarchy and a new module are intro-
duced so that aggregate to module ratios may remain
bounded.

Formally, by a hierarchical modular structure we

shall mean an aggregate or organization of modules that
are in turn hierarchical modular structures. Such a
structure may be closed in the sense that there is ulti-
mately a lowest level whose modules are not decomposable.
Examples of hierarchical modular structures are ubiquitous:
in the macrocosmos, there is the grouping of stars into
galaxies, galaxies into clusters, etc.; in the microcosmos,
the grouping of atoms into molecules. molecules into
crystals, etc.; in the mesocosmos, there are the organiza-
tions and structures of man, armies. hospitals and hier-
archical coding models.

What can we learn through comparing the properties

of these hierarchical modular structures, artificial and
natural t that will be useful in deriving a syntax to structure
and increasingly complex systems of today's world. or
that will be useful in understanding the limitations of our
own organizations and structures? As an example of the
method of morphological comparison we propose to look
at two hierarchical systems--one social. one physical.

Martin Ernst's paper in this volume on city planning

from the operations research point of view discusses the
modular parameters bas ic to urban structure and evolution.
The paper elaborates on one model, affording techniques
through whkh planners and city officials could control the

303
direction of changes in an urban complex. Ernst's approach
might be called a reductionist approach, decomposing the
city into components and sub- components, and looking at
the "portfolio of possibilities." This is an important part
of the analysis of any complex problem. However, the
morphologist wants to add something. There may exist
some parameters which place limits on the portfolio of
possibilities but which are not evident in the reductionist
approach. I would like to look at the city in this alternate
manner. For this purpose the important properties of
hierarchical modular structures to abstract are the bounds
or limits to which the modules and the aggregates may be
subject.

There are indications that our cities may be approach-

ing some kind of critical limits. What kind of limits might
these be, and how may we avoid difficulties without having
to test to destruction to see where the failure occurs? To
do this, let us compile sufficient modular forces to close
the form we call a city, and see what the limitat ions on
that form might be.

First, human beings as modules are subject to

aggregating forces as are other so-called social creatures.
These forces tend to draw people into physically compact
aggregates. Historically, humans aggregated into towns
and walled cities for trade and physical security. Today
natural gregariousness is still very much a force bringing
men together for physical, economic, and emotional security
and growth.

Next, there are density limits governing how closely

people may satisfactorily live together. These limits de-
pend on the amount of freedom of movement and privacy
we require. The higher densities in prisons and concen-
tration camps are possible because of the restriction of
movement and loss of privacy. Without knowing the value
of the density limit, we can definitely assert that such a
limit exists. (If you want an absolute limit, you may take
the value of one person per 1.83 sq. ft., provided by
Surajah Dowlah's experiment in close packing of humans

304
in Calcutta in 1756.) However, we :must bear in :mind that
in the :modern city for purposes of density li:mits, the real
inhabitants are :motor vehicles, not people. The :maxi:mu:m
density is deter:mined by the :mini:mu:m space needed for
:maneuvering, parking, and servicing auto:mobiles.

A second li:mit exists in city life. This is the li:mit

on the ti:me required to be in :move:ment to transact the
city's business, or the bound on the :maxi:mu:m fraction of
the day that the average co:m:muter can tolerate spending
in co:m:muting. Doxiadis' studies show in cities of the
past, the :maxi:mu:m distance fro:m their centers was ten
:minutes by walking. We have certainly :moved a long way
fro:m this value toward the co:m:muting ti:me li:mit. Three
hours, or one-eighth of the day is not unco:m:mon although
the average is still considerably less than one hour per
day. Both the city and the hu:man :modules which co:me
together to :make it are governed by the characteristic
ti:me period of 24 hours. This is an "absolute" value that
is not at our disposal appreciably to :modify. It is :more
basic than the day-night cycle i:mposed by the earth's
rotation since this period is also set by the biological
clock in each inhabitant. Even though adjust:ments in
bas ic co:m:muting proble:ms can be :made by s o:me people,
such as going to work on Monday, living near their work,
and returning ho:me on Friday for the week end, such
practices can not alter the basic 24-hour period set by the
needs of the city and its population. With present work
and sleep require:ments co:m:muting ti:me :must be no greater
than 1/3 of 24 hours.

These li:mits :may readily be co:mbined sy:mbolically

to define a closed entity. Let iT be the density bound and
T the co:m:muting ti:me bound. (The latter :may be expressed
in ter:ms of the natural period of the city T = 24h by T :5 z;, T
where z;, < 1.) For a si:mplified :model of a two di:mensional
city, N = a.O=R 2 where R is the :maxi:mu:m length path
through the city and ~ is a shape factor. A li:miting veloc-
ity which depends on the state of the art will be des ignated
by c. The realizable co:m:muting velocity will be less than
c.

305
 Since R::; c7' ::; C7 and a: < iT, where barred 2
quantities are m.ean values, we have N =
aO'R2 < ac 2a:'T <
s
ac 2 iT 7 2 :s ac 2 iT 2 T2. In a three dim.ensional m.odel we
m.ay introduce the m.ean height, n, of the city and use
three dim.ensional densities, p and p, giving
2-~ ~2
N<a'c hpT
If we designate the absolute lim.it p7 2 by l/H, then
HN < a' (1)
c2n

These particular lim.its com.bined with an aggregating

force m.ay indeed have som.e significance with regard to
cities, for it is interesting that a sim.i1ar relation obtains
in cosm.ic aggregates.

In 1907 before the developm.ent of m.odern cosm.ologi-

cal theories and before the establishm.ent of the existence of
white nebulae as external galaxies. the Swedish m.athe-
m.atician C. V. L. Charlier showed that in a universe con-
taining an infinite num.ber of stars the sum. of gravitational
forces acting at every point would still be finite provided
the universe were structured in a hierarchical m.odular
m.anner. Quite independently of possible relevance to
cosm.ology, Charlier's inequalities showed in general that
a hierarchical m.odular structure could be used to bound
density and inverse square type forces.

Under assum.ptions of uniform. density and spherical

syrn.m.etry, Schwarzschild showed that the field equations
of general relativity predicted the existence of a bound on
the gravitational potential
GM <:: 1
(2)
2 - 2
c R
where M is the m.ass and R the radius of the gravitating
sphere. Under the assum.ption of uniform. density this lim.it
dem.ands the existence of hierarchical m.odular structure.
If the equation is written in the form.
p R 2 :s B

306
where p is the density and B is a fixed bound (w, 'ssume
that G and c are constants), we see that for a given
den s ity- - as for example, mean stellar dens ity- -the maxi-
mum possible radius of a star is determined. Such an
inequality not only defines a limit to stellar size but forbids
close packing of stars in space. Stars can be organized
together into a larger aggregate only if a lower value of p
obtains. If p assumes the mean value of galactic density
the argument may be repeated. The maximum size of a
galaxy is determined by the same bound but with a lower
value of p. The repeated application of a potential bound,
like in the Schwarzschild inequality, can account for the
levels in the hierarchical modular structure observed in
the universe. However, the inequality does not explain
the particular set of pIS which are observed in the universe
nor does it indicate at what level the hierarchical modular
structure may terminate. Potential bounds like the
Schwarzschild li.Init may also be interpreted as bounding
the maximum velocity a module may possess in a coordinate
system at rest with respect to the aggregate. With this
last interpretation, we see from Figure 1 that cosmic
bodies are either "density limited" or "velocity limited. "
The "slope 3" line represents the limiting density of matter
in a non-degenerate form. Solid cosmic bodies lie on or
to the right of this line. (On the logarithmic scales used
in the diagram, the planetary bodies appear to have
ess entially the same dens ities.) The" slope 1" line repre-
sents the observed location of the velocity limited bodies,
Le •• the star, galaxy, cluster, and derived super cluster
having the largest potentials or escape velocities. (This
is an observed potential bound and differs in numerical
value from the theoretical Schwarzschild bound. The
objects falling on the observed bound, like those on the
density bound, are non-degenerate.) The inequalities (1)
and (2) may be put in the respective forms,
PT 2 < B* and pR 2 < B
These inequalities have the same ingredients and we might
expect them to have the same significance even though the
values of the coupling constants are quite different.

307
 On the basis of these similarities we might propose
a theorem of the form:
Given
1. The existence of an aggregating force tending to
bring modules into a condition of maximum com-
pactness, (gravity in the case of cosmic bodies.)
2. The existence of a maximum limiting density, (the
limit set by non-degenerate matter in the cosmic
example. )
3. The existence of a potential bound or its equivalent,
(such as the Schwarzschild Limit, in the gravita-
tional cas e. )
then hierarchical modular structures provide a way for
accommodating indefinite size while satisfying these
intrinsic limitations. ¥ecifically we are led to inequalities
of the p R2 < B or p'T < B* type. If we assume we may
apply such a theorem to a city. then frcim 0: R2 S ac 2 ;.;:2.
we see that for a given density, the size depends on a bound
set by the effective velocity of travel and the maximum
acceptable commuting time. The bound may be satisfied
as N increases by increasing c, or alternatively the
solution may be found in hierarchical structure.

If a polynucleated city develops on hierarchical lines,

it will be stable so long as each nucleus and the complex of
all the nuclei (with an overall lower density) satisfy the
inequality, ITR2 < B. However, the nuclei will not close
pack, which means that if subsequent urban development
fills in the areas between the nuclei bringing the mean
density up to the level obtaining within a nucleus, the com-
plex will surpass the limit. This sort of "filling in" process
is occurring in the "megapolis" areas of the Eastern
United States and Southern California. If these derived
inequalities are valid, we will not escape with impunity
the destruction of our open spaces or the low density back-
ground between present cities.

Since no physical restrictions governing the distri-

bution of dens ity in the city exist as in the cosmic case.
there are other possible solutions. It can be shown that

308
the bound Illay be satisfied by selecting a density distri-
bution o-(r)- r-(Y+l}where '{ > 1. In this case, the city
Illay grow and still satisfy the bound if it is built in a ring
shape. Several suggestions of this sort have been Illade
including a city which is nothing but a series of linear
structures several stories high with freeways on top.

Additional liIllit theoreIlls on the structure of cities

Illay be derived. However, these require more sophisti-
cated models and exceed the parallelisms in the hierarchi-
cal modular analogy given. Since our purpose here is not
to develop a general theory of urban structure, but to
illustrate the method of morphological parallelism, this
one analogy will suffice. The method of morphological
analogy does not per se generate valid theories. It pro-
duces hypotheses and ideas on which models may be con-
structed. These must then be tested by the usual canons
of scientific verification.

309
 BIBLIOGRAPHY

Alexander, C. The Man-Made Objects t G. Kepes, Ed.,

1966 (New York: G. Braziller), From a Set of
Forces to a Form, p. 96.

Alexander, C. and Manheim, M. L. The Use of Diagrams

in Highway Route Location (1962). Report R 62-3
Civil Engineering Systems Laboratory, M. I. T.

Charlier, C. V. L. 1922 HowanInfiniteWorldMaybe

Built Up, Arkiv for Math. Ast. , och Fysik, Bd. 16,
No. 22, 1.

Doxiadis, C. A. The Coming Era of Ecumenapolis,

Saturday Review, March 18.1967. p. 11.

Zwicky, F. Morphological Astronomy 1957 (Berlin:

Springer).

310
 FREEWAY DESIGN PARAMETERS
I

MAJOR CURRENT
TRAFFIC DESIRES

LV OTHER
l OTHER
I-' NEW
I-' TRANSPORTATION 4 ~ 4 ~ FREEWAY
FREEWAY
SYSTEMS SYSTEMS

STRUCTURE L ENVIRONMENT
I ~ FUTURE
TRANSPORTATION
SYSTEMS
--_._- -
 FREEWAY DEffiGN PARAMETERS
II

Internal Environmental
Construc tional Physical
Earthwork costs Catchment areas
Bridge costs Drainage patterns
Pavement and subgrade costs Weather effects
Construction interference Air pollution

W Economic Economic
I-'
N Land costs Regional land development
Public financial los ses Local land development
User costs Non - recompensable public
Obsolescence and private loss

Ope rational Esthetic

Travel time Eyesores
Local accessibility and Noise
integrity
Safety
Maintenance and services
Self induced congestion
 T30 ~~------------------------------~
SUPER·CLUSTER
CLUSTER
~20

GALAXY-_~

+ 10
SEQUENCE
'"
'"
t.)

en JUPITER
:::> SATURN
c EARTH
c:(
a::: MARS
...J
c:(
-10 MOON
z
0
~
c:(
~
-20
>
c:(
a:::
c.!l
C>

<.::J
a -3D
DENSITY LIMITED
AND VELOCITY
..J

LIMITED BODIES
-40

-so

-10 +lU + 20 + 30 +40

LOGIO (RADIUS) (CGS)

313
 CHAPTER 3

MORPHOLOGY OF
MULTI LANGUAGE TEACHING

F. ZWICKY
California Institute of Technology

ABSTRACT

The building of a sound world depends on the COITl-

ITlunication between ITlan and ITlan. Since this cOITlITlunica-
tion cannot reITlain in the abstract but ITlust be alive. it is
iITlperative that all ITlen of good will and especially scien-
tists and other professionals ITlaster the tongues of other
peoples. The deterITlination to occupy oneself with
languages. however. is not sufficient if ITlethods of
instruction are not developed which are far ITlore effective
than those used in the past. For this purpose a ITlorpho-
logical study has revealed that the tiITle honored ITlethod
of learning only one language at a tiITle is too inefficient.
Multilanguage teaching ITlust be introduced instead which
is to be conducted in environITlents and by ITlethods which
will be described, but which ITlust, in the course of tiITle,
be iITlproved as new experiences are being gained with this
entirely novel approach.

314
 A. Problems of Communication

During the past few decades tremendous progres s

has been made with the invention of new techniques and
with the installation of efficient devices for fast world-wide
communication. We have even successfully reached out far
into interplanetary space and we shall no doubt soon be able
to send mes sages throu,gh inner space, that is through the
solid Earth and to all points within the oceans. At the same
time, however, we have lost out on the communication from
man to man. One of the reasons for this is that, although
people travel all over the world more than ever before they
do not speak the languages of the peoples they meet. And
what is worse, if one speaks only one language, he really
knows none at all, as the great German poet Goethe once
said. As a consequence many have even lost the human
contacts with the members of their own families.

The greatest problem before us therefore is not the

development of ever more sophisticated means for trans-
mitting mes sages to all points of the solar system and per-
haps beyond. It is far more important but also very much
more difficult to reestablish the human contacts between
man and man. This. I am afraid,will remain impossible if
we cannot dispense with the dragomans and speak the
languages ourselves.

Essentially we have two choices. Either we must

develop a universal language or we must learn many. The
first choice has not proved successful and probably never
will be. A newly created language may serve as a means of
formal exchange but it is of little value humanly since it is
schematic and disregards the cultural and emotional heritage
invested in the many individual languages.

We therefore have no choice at all but must learn

the tongues of other peoples if we wish to make our proper
contribution towards the construction of a sound and united
world. Diplomats, businessmen and scholars of many
countries have long recognized this fact. It is mostly the
English speaking peoples who are dangerously dragging their

315
feet. Fortunately in the United States some commendable
efforts are now being made to remedy this situation.

Since scientists as a body probably command the

most extensive knowledge on technical and on human issues
and since they travel the most, it will be up to them, before
all others to become multilingual. But the determination to
start learning languages is not enough and it will soon
weaken unless the present system of education is radically
changed. Indeed, as the result of an almost incredible
oversight, language studies have been a torture to countless
children, students and adults alike and have become utterly
ludicrous in our time. It remained for morphological re-
search to point out that the teaching of only one language at
a time, as it has been practiced for so long, has its serious
and even fatal drawbacks and that the possibility of multi-
language instruction must be studied and developed instead.

B. Advantages of Teaching Several Languages Simultaneously

If we try to study only one foreign language at a time

we must essentially learn it by heart. The simultaneous
occupation with several languages is very much more effi-
cient, since interrelations among words, expressions and
concepts are easily remembered, especially so if we avail
ourselves of the morphological approach. At the same time
the student's imagination is constantly being stimulated and
he will automatically think of further similarities, analogies
and associations. He will also notice certain irregularities
and contradictions and, exploring their origin he will deepen
his understanding of the structures of his own as well as of
other languages.

In illustration of Goethe's statement I may in this

connection mention one of the many observations which I
made during the past few weeks. This happened while I was
struggling to translate my recent book ENTDECKEN,
ERFINDEN, FORSCHEN 1M MORPHOLOGISCHEN WELTBILD
(Knaur-Droemer Verlag, Munich 1966) into English. The book
deals with the fundamental principles of the MORPHOLOGICAL
APPR OACH, the prime requirement for which is detachment

316
from all prejudice, bias and prevaluations. I therefo re was
searching for a noun which expresses the opposite of bias,
a word which I had naturally used extensively in the German
edition and which I needed very badly for the translation.
I discovered to my dismay that no such single word exists
in English. a fact which probably has occurred to very few
individuals who speak only this one language. Under these
circuIpstances I therefore propose to create a new word,
that is the expression "unbias" which will be exceedingly
useful if the morphological approach and morphological
research will have the great future which I venture to pre-
dict for them.

There is also no English word which simply trans-

lates the beautiful German word DICHTEN= compose poems.
We have poetry. poems, poets but no single verb which
expresses poetically what a poet does. On the other hand
I had invented the "morphological approach" in English and
I have since encountered great difficulties to translate
"approach" succinctly into German or French, a fact which
rendered many passages in my German book rather awkward.
I should indeed be most grateful to hear from a linguist who
knows more than I, how the expressions "scientific approach."
"practical approach," "artistic approach," "morphological
approach," etc. can be accurately and succinctly translated
into German or French.

c. The Morphological Box of Multilanguage Instruction

1. Formulation of the Problem

Our problem consists in reviewing the various

methods which can be used for the simultaneous study of
two or more, say p languages. Of the many factors,
which we shall describe by the parameters Pi we can only
enumerate here a few of the most important ones. With
these we shall construct a preliminary morphological box
which contains the various approaches and aspects of multi-
language teaching.

317
2. The ParaITleters of the ProbleITl

First ParaITleter P 1 :

We take P 1 to denote all of the known p languages.

P 1 therefore has the cOITlponents P l1 , P 12 •••• , P 1p •
Thus for instance P l1 =
English~ P 12 =
French. P 13 =
Russian, •••• P 1 ,p-1 = Chinese. P 1p =
Sanskrit.

Second ParaITleter P 2 :

P 2 characterizes the professions and the backgrounds

of the people who are to be taught. For ins tance P 21 =
engineers. P 22 = scientists. P Z3 = businessITlen. P Z4 =
diplomats J P 25 = airline stewardes ses, Ii P 2 q -1 =
0 0 ,

children. P 2q =
ordinary travellers. •

P 3 designates the ITlethods of teaching. Thus for

instance P 31 = language class. P 32 =
teaching in conjunction
with professional instruction, P 33 = correspondence ITlethod.
P 34 =through translation and cOITlposition, •••• P 3r = in
conjunction with travels.

Fourth ParaITleter P 4:

P 4 characterizes the psychological types of the

pupils: P 41 = practical ITlen of action. P 42 =
iITlpractical
individuals. P 43 = frustrated. inhibited individuals •••••
P 4, s -1 = concrete thinkers, P 4s = abstract universal
thinkers (ITlorphologists).

3. The Morphological Box

In order to appraise and evaluate the probleITls of

ITlultilanguage teaching we ITlust study all the possibilities
contained in the following s cheITle or four diITlensional
ITlorphological box.

318
 Morphological Box of Multilanguage Teaching

Essential Parameters Parameter Components

Pi :: Language (Set 1)

Pi :;; Language (Set 2)

Pi :: Language (Set 3)

Pi :: Language (set p) P 11 P 12 P 13 P 14 ••• P 1 ,p-1 Pi

P 2 :: Background
of pupil

P 3 :: Method of
teaching

P 4 :;; Psychological P 41 P 42 """.00 •• "000<100

type

One further most important parameter of course is the

native language which the student already knows, as well
as the number of languages which he understands to start
with. Fo r simplicity we shall think of the s e qualifications
as being contained in the components of the parameter P z
which describes the background of the student. Or we can
simply establish a separate Morphological Box or Chart
for every native language we start out with.

Suppose then that our chart refers to a Chinese.

The chain of circles marked on our chart, that is the chain

[P 11 ,P 12 ,Zero, ••• ,Zero'P22,P31,P4)

thus means that a Chinese scientist who is a universal

thinker wishes to attend a language class in which Fl'ench
and English are being taught simultaneously. The problem
for the language teacher and the morphologist of ed ucation

319
therefore is to evaluate all of the possible chains and to
establish through successive approximation and back feeding
of experience gained what the most effective means of
language teaching are.

D. Evaluation and Implementation of Some of the Aspects

Involved in the Morphological Box

We proceed here to give some brief illustrations

as to the bearing of the various parameters P on multi-
language teaching.

1. Linguistic Aspects of Multilanguage Teaching: Simi-

larities, as sociations, differences. irregularities

Thes e characteristics refer to the parameter com-

ponents P 1 i.

a, International Expressions

We here must consider several well known examples

for words and expressions which, mostly for historical
reasons and because of convenience of use thread their way
through many languages. Among these there are first the
thousands of internationally current technical words, second
the many Latin words and phrases and third certain famous
proverbs and songs. Many of these may have been twisted
more or less in the course of their adoption by different
languages. but. with some practice most of the alterations
can be guessed at.

a. Technical Expressions

The number of words of this type used simultaneously

in many languages probably runs into the thousands so that
their knowledge. implemented by rudimentary familiarity
with additions such as prefixes and endings should go a long
way to help both ordinary travellers and professionals on
their journeys. Such words, which I choose entirely at
random are. for instance: active, analysis. apparatus. atom,
automobile. autobus. author. bank, blockade, bomb. botany,

320
canon, chell1istry, Chri stian, COll1ll1une, conse rve, coopera-
tive' dell1on, despot, discuss. edict, effect, engineer.
finance, forll1, general, genius, gigolo, hell1isphere, hotel,
intill1ate, interpret, locoll1otive, ll1achine, ll10ral , ll1otor,
negro, observatory, pill, popular, quartz, register, spec-
trull1, taxi, technical, t ele scope, telephone, telegraph,
universal, university, ventilate, wine, xylophone, zone,
etc. , etc.

As to the endings of the international technical verbs

such as to "telefone I. they run in general consistency as
telefonieren (Gerll1an), telephoner (French). telefonare
(Italian). telefonear (Spanish). telefonirovat (Russian). SOll1e
sll1all irregularities are often easily rell1ell1bered because
of eas e of pronunciation such as in Spanish "telegrafiar"
instead of telegrafear.

For scientists and engineers it is of course of the

greatest ill1portance that designations of devices, engines,
instrull1ents. ll1aterials, processes. ll1ethods and physico-
chell1ical phenoll1ena be used in the different languages which
leave no doubt as to what one is dealing with. For this pur-
pose it has proved necessary to construct ll1ultilanguage
technical dictionaries. For instance a special cOll1ll1ittee
of the International Acadell1y of Astronautics with head-
quarters in Paris has been hard at work for several years
on the cOll1pilation in six languages of all of the essential
terll1S used in aeronautics and in astronautics.

Beyond that, new words and expressions are con-

stantly being introduced by scientists. engineers. physicians,
philosophers. writers and for that ll1atter by ll1en froll1 all
walks of life. Such new creations are often adopted un-
changed by other languages. A few exall1ples relating to
sports, fashions. novels and non-fiction are the words ski,
football, goal, skiff, bobsleigh, golf, karate, Ju-ji-tsu,
ll1ukluk, corsage, decollete, dernier cri de Paris, pajall1as,
ll1iniskirt, gaucho, jockey, goulash, bar, tearooll1, sand-
wich, picnic, dacha, dragoll1an. bandit, guerilla, cOll1puter,
ll1aser, laser, taxi, pilot, stewardess--Weltanschauung,
Gell1uet, Gell1uetlichkeit, etc.

321
 As to the introduction into any language of new tech-
nical expres sions they should be chosen so as to be usable
in all languages and designations such as ramjet or athodyd
should be avoided and the universal word aeroduct should be
used instead. In this connection I have proposed to the
International Academy of Astronautics that in our journal.
the Astronautica Acta. every author, in a brief appendix to
his article should suggest translations of new technical terms
he might have introduced. Reviewing my own activities in
science, engineering and philosophy, I was surpris ed to dis-
cover that I am responsible for having coined over a hundred
new special words, most of which are generally understand-
able all over the world while others require expert transla-
tions. To mention some of my creations, they are. in
astronomy: neutron stars. pygmy stars. whittled down
stars, nuclear goblins {Kernkobolde, lutins nucleaires.
duendes nucleares, folletti nUcleari, iadrenie tshortiki},
compact galaxies, neutron star studded compact galaxies.
cosmic balls of light, neutron star studded cosmic balls of
light, symbolic velocity of recession {relating to the cosmo-
logical redshift}, or in engineering: aeroduct, aeroresonator,
{aeroturbojet}, aeropulse. aeropistonjet, and the correspond-
ing propulsive power plants which travel under water or
traverse the solid Earth, such as hydropulse, the terrapulse
and the "fuels" they need, that is the hydrofuels and the
terrafuels which are the analogs to the aerofuels or ordinary
conventional fuels; and in Physics: cooperative phenomena,
secondary structure of crystals. mosaic structure of
crystals {now called dislocations}, gravitons, mosaic
gratings; in chemistry: fragment chemistry, metachemistry,
monopropellants, coruscatives {heat detonators} and finally
in methodological phil os ophy: the principle of the flexibility
of scientific truth, morphological research, the method of
morphological field coverage, the method of the morpho-
logical box and so on.

13. Latin Expressions

Whenever we read inscriptions, titles of international

journals, academic diploma, articles in ordinary magazines,
books and advertisements. court proceedings or pronounce-

322
ments of the Pope we encounte r a multitude of Latin words
or phrases. Starting from such pegs of knowledge acquired
from everyday reading we cannot only learn to understand
quite a bit of Latin but, by tracing Latin words through the
Western languages we can retain parts of these as well.

For instance, in scientific articles we find these

expressions: etc. (et = and, cetera = all the other), et al
(alius = another), sic (in this way), a priori, a posteriori,
a fortiori, ipso facto, ad infinitum, ad nauseam, sine qua
non, in toto, in situ. per se, versus, Helvetica Physica
Acta, Astronautica Acta, Astronomica Acta, and so on;
or in the general literature: sub poena, habeas corpus, in
memoriam, dies academic us , per diem, pro rata, semper
fidelis (US Marines), mea cupla, per capita, lapsus linguae,
exit, fiat, status quo, summa summarum, per annum, bona
fide, excels ior. ultimatum. and so on. Furthermore. there
are all of the various sayings and proverbs such as: mens
sana in corpore sano; vox populi. vox Dei; veni, vidi. vici;
vincit omnia veritas; facta, non verba; veritas vas liberabit;
sic semper tirannis; suum cuique; ad maiorem Dei gloriam;
requiescat in pace (notice that the conjunctive is used by the
ancients. meaning "may you rest in peace," while we ~­
mand the dead to rest in peace!). Well known are also the
mottoes of the US (E Pluribus Unum), of the District of
Columbia (Justicia Omnibus) and of the various States like
Arizona (Ditat Deus). New York (Excelsior). Oklahoma
(Labor Omnia Vincit) and Montana (Montani Semper Liberi).
Some of the encyclicals of the Pope were Popularum Pro-
gressio (1967); Mater et Magistra (1961); Pacem in Terris
(1963); Rerum Novarum (1891).

While some of the Latin words are still in use directly

or only slightly altered in many languages others are found
only in a few of the modern tongues or they may have disap-
peared entirely. As far as I know. this is the case with
oppidum (town) while urbs has been retained in urban, suburb
and so on. Ambo = both is still used in Spanish in this sense
while other languages have it incorporated in ambivalent.
ambidextrous. ambiguous and so on. Sin = without has been
changed to sans. senza and sin in French, Italian and Spanish

323
but is used only in combinations with, for instance,
cura = care in German and English as Sinekure and sine-
cure; also sine die = indefinitely (without naming the day).
Tangere = touch survives as tangent (in Russ ian tangens),
tangible and the dance tango; canis = dog in canine and in
.-
canicular days, journees caniculaires, giornate caniculare,
and in Russian kanikuli = vacation, dogs days; vox = voice
and vocare = call, in voice, voix (FR), voce (IT), voz (SF)
and in vocation or advocate, while lawyer in French,
German, Italian, Spanish and Russian is 1 'avocat, Advokat,
avocato, abogado, and advokat. Avus--grandfather seem~
to live only in Spanish as abuelo, domus = house remains
as an ordinary house (dom) only in Russian while it has
suffered a substitution of meaning in all other languages
where it has become a cathedral (or a dome of an observa-
tory, etc.). The same is true of the Latin sinister = left,
which it still is in Italian (sinistra) while sinister in all
other languages means baleful or foreboding evil.

The designations for the months have been taken

over from the Latin into many languages, including the
Russian, often with considerable twists which, however
can easily be learned. The same is true for the cardinal
and ordinal numbers which have suffered still more severe
alte rations.

Astronomers are immediately familiar with many

Latin words in the nominative and the genitive since the
stellar constellations are named that way, thus Virgo, Leo,
Serpens, Aquila, Cygnus and so on, while stars in these
constellations are a, ~, y and soon Virginis, Leonis,
Serpentis, Aquilae, Cygni.

y. International Words in Music

Some expressions in the world of music may also

be regarded as being fairly universally used, for instance
those which designate the types of works, instruments and
so on such as opera, operetta, concert, symphony, libretto,
violin, piano, saxophone, guitar, etc., or the Italian words
which are used as annotations of how a piece should be

324
played such as alto, andante, allegro, basso, crescendo,
forte, finale, piano, sotto voce, vivace and soon. Mus icians
therefore have at their disposal SOTIle words with which they
can cOTIlTIlunicate inte rnationall y.

b. The Indo-GerTIlanic Heritage

Chains of sound transitions which lead froTIl the Indo-

GerTIlanic into dozens of the Western languages are the TIlost
nUTIlerous and probably the best known. They should be tabu-
lated by experts and then be specifically adapted for the
purpose of TIlulti-language teaching. I TIlention here only a
few of the words with equal roots, for instance those for the
principal TIleTIlbers of a faTIlily. faTIlilia (L), FaTIlilie (G),
faTIlille (Fr), faTIliglia (It). faTIlilia (Sp). faTIliliia (R), that
is the following:
TIlother father brother sister son daughter (E)
Mater pater frater soror [ filius filia ] (L)
TIl~re p~re fr~re soeur [ fils fille ] (Fr)
TIladre padre fratello sorella [ figlio figl ia ] (It)
TIladre padre herTIlano herTIlana [ hijo hija ] (Sp)
Mutter Vater Bruder Schwester Sohn Tochter (G)
TIlat atiets brat siestra sIn dotch (R)

It is seen that the words in brackets for son and daughter in

the rOTIlance languages TIlatch aTIlong theTIlselves but derive
froTIl another root. Notice also that the consonant f ill
Latin, French and Italian is changed often into an h in
Spanish, thus also the French words for farine (flour), faiTIl
(hunger), feuille (leaf), faire (to do), fil (a thread), fuir (to
escape) in Spanish are harina, haTIlbre. hoja, hacer,hilo
and huir.

In SOTIle cases it pays for TIlneTIlotechnical purposes

to pursue even the TIlinutest details. Fo r instance, in the
word for TIlother the letter r appears in TIlost of the Western
languages, but it is TIlissing in the Russian TIlat. On closer
study, however. one discovers that it reappears in the plural
(TIlatieri) and, on declination in all six cases, singular and
plural, except in the nOTIlinative and accusative singular.

325
 c. Regularities Within Certain Groups of Languages

a. SiInilar Words in the Romance Languages

I mention here only one of many analogous simi-

larities. Consider for instance the words which in Spanish
start with esc, esp, est. Many of the corresponding words
in French have either the same combination of letters or
the consonant s is omitted. In Italian, almost consistently,
the initial letter e is omitted, while in English, German
and Russian, if they use the respective word at all, either
the Spanish or the Italian patte rns are followed, thus:
Escuela (Sp) €cole (Fr) scuola (It) school (E) Schule (G) Shkola (R)

esposa €pouse sposa spouse

escudo ecus
~
scudo (kudos?)

espalda €paule spaldo (shoulder) (Schulter)

estudiar €tudier studiare study studieren shtudirovat

Estado Etat Stato State Staat Shtat

estrella €toile stella star Stern

(extranjeroJ €tranger straniero stranger

and, with the substitution of a different meaning in English,

German, Russian:
estar etre stare stay stehen stoiat
espUIna €cume spuma scum Schaum shum

The following words start out the same way in Spanish and
in French:
esperar esp€rer sperare
espanol espagnol spagnuolo spanish spanisch ispanski

espirito esprit spirito spi rit

esclavo esc1ave schiavo slave Sklave

Finally some words retain the vowel e only in Spanish

while starting with s in all others:
especial sp€c ial spec iale special spez ial spets ialni
escena sc~ne szena scene Szene stsena

326
 j3. Relative Strength of the Words in the Romance
Languages

It is of interest for teaching purposes to recognize

rules about the relative strength of the pronounciation of
consonants and vowels in related languages. Here are a
few examples of how from the Italian to the French to the
Spanish consonants and vowels soften continually:
Citt~ (It) Cit~ (Fr) Ciudad (Sp)
Universit~ universit~ universidad
libert~ libert~ liberdad
cinque cinque cinco
cento cent cien
cuccina cuisine cocina
lunedf lundi lunes
viaggio voyage viaje

y. Much Used Words

These can be found in booklets issued by airlines.

travel agencies. in multilanguage menus. laundry lists.
advertisements and so on. They contain much useful
material for establishing daily contacts with all strata of
people and should therefore be made extensive use of. The
expressions found in these sources refer to travel by plane.
rail or autobus J to passports. custom regulations. greetings.
meals. tips, physicians. medicine. time of the day and so
on and they can be learned when there is not much else to do.

d. Grammar

a. Nouns and Adjectives

These can in principle be masculine. feminine or

neuter as well as in the singular or the plural. and they may
in some cases all have to be declined. In multilanguage
teaching the respective patterns should be reviewed right
from the start. I can here give only a few hints.

In English for instance all distinctions between

327
masculine, feminine and neuter have practically disappeared,
except in references to objects by means of the pronouns he.
she or it. Declension is achieved through the use of prepos i-
tions such as with, for, of, from, by and so on. but without
any change in the endings of the nouns or adjectives, except
for the occasional addition of a terminating letter s in the
genitive. Most other languages, however, use far more com-
plicated patterns. In German for instance, the nouns. ad-
jectives and associated definite and indefinite articles are
all being declined. German on the other hand allows us to
combine nouns, adjectives and prepositions freely and to use
most verbs as nouns which enormously facilitates the ease
of expression and allows one to state in one word thoughts
which need a sentence in other languages.

13. Verbs

Here intercomparisons must be made from the begin-

ning as to the formation of the tenses. as to whether the per-
fect and imperfect past tense and the past partic iple all are
used, as to whether the future and conditional tenses are
expressed by using different endings or if the auxiliary verbs
are used for the purpose and so on. Also, how many auxilary
verbs are being used, for instance in Spanish for "to be" both
ser and estar, for "to have" both tener and habere In German
the disconcerting and confusing habit of splitting the verbs
must be paid particular attention to.

'I. Phrases and Expressions

It is also most profitable that one familiarize him-

self with all possible variations of greetings, welcome and
wishes for good health, birthdays, Christmas and New Year.
Much can be learned from tracing proverbs through the
various languages because of the wealth of the words involved
and the deeper acquaintance which one gains with the thinking
and the habits of other peoples. The same is true if one
starts learning songs, especially if this is done among con-
genial groups of friends, music societies, in camps, or in
the army.

328
Z. Background of Pupils, Methods of Teaching, Psycho-
logical Types

a. General Cons ide rations

All of these aspects which are described by the

parameters P Z ' P 3 and P 4 of our morphological box
render multilanguage teaching a very complex but mo st
exciting and rewarding project. Its realization is best
started with a few general principles in mind, but with pre-
severing determination to accumulate experience as one
goes along. Among these principles I mention some ex-
ceedingly simple ones.

For instance, one's parents, relatives or friends

may know other languages. Such invaluable opportunities
and assets should never be wasted as it is being done today
in a most thoughtless way, especially in the U.S. where
there live millions of parents of foreign birth who neglect
to pass their languages of origin on to their children.

The same wasteful habits apply to much of the free

time which we do not make any use oL I am thinking of
the short intervals when one has to wait for one's wife,
husband or children, or the hours wasted at airports, rail-
road stations, in the waiting rooms of dentists, hospitals,
courts and so on. These hours can all be used to peruse
small booklets for words or articles written in other
languages.

The most important problem, however, is to bring

the proper people together and to conduct the multilanguage
teaching in the proper places and at the proper time. These
problems must obviously be studied by an intensified morpho-
logical approach. If they are solved correctly one of the
greatest obstacles of today,that is,inhibitions and inferiority
complexes of so many individuals, are automatically eliminated,
as I have learned from long experience in teaching. Indeed
when instruction is conducted through the use of the morpho-
logical approach there is no coercion of any kind and infor-
mation is picked up in a free interplay of all of the partici-

329
pants each of which spends effort and tiITle as he chooses
and as he feels theITl to be adequate for the purposes which
he has in ITlind. In way of illustration one of the ITlany
observations which I have ITlade along these lines ITlay here
be presented.

b. Specific Experiences with Scientists

The experiences which I have in ITlind refer to groups

of scientists.

During the war. I conducted two seITlinars on ITlorpho-

logical research, one at the California Institute of Technology
on the subject of astrophysics and another at the Aerojet
Engineering Corporation on jet propulsion. Both of these
seITlinars were attended by graduate students, assistants.
professors and engineers whose experience. taken alto-
gether. included the knowledge of about six languages. While
discussing various scientific and technical probleITls we there-
fore jotted down on the blackboard and in our notebooks words
and expres sions in all of these languages and we conversed
in theITl as far as we could. A further extension in versa-
tility was gained when we went to lunch together. which gave
us the opportunity to exchange SOITle ITlore of our knowledge
of words and phrases as they are useful for daily cOITlITluni-
cation. Also, whenever possible. assignITlents were ITlade
for hOITle study of the various topics in foreign language
journals and texts. Multilanguage dictionaries were likewise
ITlade extensive use of.

The whole procedures described proved very effective

and the results achieved becaITle invaluable to several of us
who. at the end of the war went to GerITlany and to Japan to
work on the scientific and technical evaluation of the GerITlan
and the Japanese war efforts.

Short articles in various languages were also written

by ITleITlbers of our groups and I have continued this habit on
a sITlall scale ever since by publishing scientific notes in
French, GerITlan, Italian, Spanish and Russian and by speak-
ing on radio and television in France, GerITlany, BelgiuITl,

330
Russia and Switzerland.

E. Conclusions and Suggestions

From what has been said above we conclude that,

with some moderate effort and imagination it should prove
profitable to practice multilanguage teaching first in an
informal way within families or in association with general
courses and seminars on all sorts of subjects, professional
and recreational. In the second place, however. the inter-
relations among languages which are well known to the philo-
logists must be reviewed and formulated in a way suitable
for promoting the introduction of multilanguage teaching.
I therefore suggest that universities, high schools, trade
schools as well as large private concerns institute formal
courses for the purpose. A new class of teachers will of
course also have to be educated I especially if the ideas
here presented are to be applied to instruction in the grade
schools already. And finally, the knowledge gained must
be properly applied "in the field ll so that no talent or
opportunity is wasted which might contribute to the strength-
ening of the bonds of mutual respect and friendship among
the nations of the world.

331
 EPILOGUE

METHODOLOGY--A DISCIPLINE

A primary purpos e of this conference has been to

consider whether the various methodologies employed in
sOlving problems when taken together constitute in them-
selves a useful scientific and technological discipline. The
descriptions of the several approaches to problems that
have been presented here- - Operations Research. Systems
Engineering, Morphological Analysis. etc. --have made
visible some common principles which have been inde-
pendently developed for structuring. analyzing, and solving
complex problems of many types. Though using different
names and terminologies, the identities and overlaps con-
tained in these approaches, taken with the fact of their
independent discovery in many diverse contexts, strongly
suggest the developability of a useful discipline that we
may call IImethodology. II Although the presentations during
this conference have only partially defined the subject area
of methodology, they have demonstrated that it would now
be meaningful to take steps toward systematic definition
and organization of the concepts so far developed and
establish a formal discipline.

Specific problem areas from hospitals to codes to

jet engines have been treated at this conference. However,
in all the variety of problems discussed, almost nothing has
been said concerning how to select which problems to solve.
It seems most important that any discipline of methodologies

333
for problem solving be concerned not only with the definition
and solution of specific problems but also with the totality
of that growing complex consisting of the set of problems
competing for our attention. The discipline of methodology
should investigate criteria by which to assign priorities t
the appropriate levels of resources - -funds and talent- -to
be thrown against a problem, the nature of the interrelated-
ness of problems, the consequences of solutions to problems
and the anticipation of derivative problems.

Neglecting an overview of the interrelated complex

of problems has given rise to some serious unbalances in
our culture. Dr. Ramo, in his introduction, pointed out
a few of these unbalances. In 90 minutes we can travel
around the earth in Gemini while in 90 minutes in our cities
we sometimes can travel only a few blocks. We can provide
pure breathable air 100 miles above the earth for our astro-
nauts, but not within a hundred surface miles of our major
cities. We have developed remote sensing equipment that
can tell us everything going on inside a space capsule. but
have not equipped the physician with comparable equipment
for monitoring what is going on inside his patient. There is
no need to enumerate our disparate and desperate social
unbalances. We might now add that a conference on
methodologies for solving problems without consideration
of how to choose which problems to solve in itself consti-
tutes an unbalance.

In addition to unbalances. there are other short-

comings inherent in our present approach to the growth
and application of scientific and technological knowledge.
For example. early this year. the world's largest oil
tanker of 120,000 tons was wrecked off the east coast of
England, releasing thousands of tons of crude oil which
floated ashore and polluted hundreds of miles of shore
line. This developed into a tragedy that assumed national
proportions in England. It is estimated that extensive
portions of beach will be polluted for decades, perhaps
even permanently; and since the feedback on the ecology
of major environmental alterations of this sort are some-

334
times delayed, the full extent of the damage created by the
pollution probably will not be evident for some years. As
expected, there was widespread comment on this disaster.
However, criticism did not focus on the navigational situ-
ation which was the immediate cause of the wreck, nor on
the structural feas ibility of large tankers (they are quite
feasible- -there is a tanker of 300.000 tons currently under
construction and one of 500,000 tons on the drawing boards),
rather comment focussed on the defects in a technology that
could blindly and blandly create the set up for this sort of
disaster. This isolated example made some of the blind
spots of technology visible to many for the first time. One
of our own cabinet officers commented, "The environmental
backlash we confront today cannot be eliminated just by
applying more of the ~ science and technology that put
us in our present predicament. "

There is growing feeling in some quarters that the

time has come to ring the bell on applying technology with-
out re sponsibility to the environment or to the future; on
synthesizing complexity without regard for social and human
consequences; on continuously injecting change into society
without direction or evaluation. We must now face the great
responsibilities of what We choose or do not choose to do
with our technological capabilities. We have reached the
precarious level of technological development in which we
have the power significantly to alter our environment without
having either the power totally to control the means by which
we effect the alterations. or an understanding adequate to
predict the properties of the environmental states we bring
about. Not only must the proposed discipline of methodology
be able to derive knowledge concerning the limits to the
controllability and predictability of specific applications of
technology but also be able to derive the summary conse-
quences resulting from the piecewise solutions of the various
portions of the total problem complex.

Some of the methodologies reviewed at this conference

pointed to the importance of the elimination of prejudice as
basic to the problem solving process. Prejudices are often

335
habits of thought that we unconsciously carry to new situ-
ations in which they are no longer applicable. An example
of such a habit of thought that affects our application of
technology is the making of decisions primarily on the
basis of feasibility. One of the severe deficiencies in the
present use of technology is the failure to note that at some
level of the state of the art the answers to the two questions:
how big can we build a tanker. and how big should we build
a tanker, begin to diverge. For decades technology has
been primarily concerned with finding ways to do things
hitherto impossible. The emphasis has been on pushing
back the limitations of nature and ignorance in order to
make more products and activities feasibile and broaden
our spectrum of choice. In an increasing number of tech-
nological areas we have recently moved from the regime
of finding a way to the regime of choosing the best way.
The task is no longer to remove natural limitations but to
set up limitations of our own, to define the constraints
and restraints which are prerequisite to sensible choice.
In a regime of limited capability, choice is usually properly
made for the limit of feas ibility- - build a plow that will cut
as many furrows simultaneously as possible. However,
the habit of thinking developed in this regime tends to carry
over into the second regime; the difficult problems of choice
being ignored and option being made simply for the limit
of feasibility. For example, in typical past wars the level
of tolerance to destruction and ability to recover was higher
than the level of any enemy's capabilities to destroy. How-
ever, in the past two decades. this inequality has been
reversed. It is now possible to destroy beyond any nation's
tolerance to absorb. We have entered the regime of choice.
There is the neces sity for limited and restrained actions,
but some spokesmen still adhere to first regime thinking.

Although this phenomena of regime change seems

tautological to many, and is well understood by many
business and government leaders, the oil on the beaches
bears witnes s that one of our urgent problems is to spread
more broadly the awareness of the regime change and re-
place feasibility thinking with some of the new methodologi-
cal tools that are now available for making difficult decisions.

336
We had best rapidly acquire the techniques essential for
decisions in a choice regime. The new developments in
biology, for example, are leading us to a capability level
where we may shortly be able to determine the sex of our
offspring. extend our life spans indefinitely. and even
create new varieties of organisms. Clearly the responsi-
bilities of choice imposed by such developments are likely
to be as demanding as any ever faced by man. The tempta-
tion to be guided purely by feas ibility. say in producing
selective viruses. could put an end to the human experiment.

In a choice regimei) it becomes necessary to formulate

every problem. not only in terms of the internal capability
parameters, but also in terms of the contextual parameters.
considering environmental effects and interrelationships
and possible synergistic developments. Our failure to do
this reveals another prejudice - -the prejudice to settle for
the reductionist factors and ignore the holistic ones. This
is a pattern of thought which derives partially from the past
successes of reductionism, especially in physics. and
partially from the unwarranted association of holistic
effects with supernaturalism.

Besides facing up to these and other prejudices such

as fadism, the proposed discipline of methodology must
derive techniques for treating the increasing complexity
of our problems and systems, complexity leading to such
occurrences as regional power blackouts or postal service
breakdowns. Oftimes feedback signals from complex systems
cannot be interpreted promptly. The signals may be delayed
or lost in other effects. Pollution is an example of a prob-
lem area whose feedback signals have been unheeded until
the environmental backlash has reached proportions whose
correction will require major technological and social sur-
gery. Development of techniques for prompt interpretation
of feedback signals is an urgent problem area of the disci-
pline of methodology.

Other new problem situations are on the horizon. The

trend toward longer development times and shorter life times

337
for new systems with the impossibility of paying off
development costs before obsolescence may place us in
the same situation as an organism whose life span drops
below its gestation period.

There are many other aspects of the subject of how

to select, define, and solve problems which will concern
the methodologist. If the future comes to be dominated by
unknown and uncontrolled parameters arising from the
interaction of the random application of technology to
specific problems in agriculture. medicine, manufacture,
space, defense, etc •• then planning becomes illusory and
the course that our civilization will take is that of a car
without a driver. It will be useless to construct one of
our usual "good guy--bad guy" explanations for the situation.
There is no villain, only complexity, and it is not too
early to bring out best research talents to grips with it.

Albert Wilson

338