CO

66986 [<OU_1

DO
^)))

Call No.\"7
Author
Title))

Osmania UniversityLibrary

'

'

\"

\"*

Accession o. G? N

*>

This book shouldbe returnedon or before thedate

lastmarked below.)))

THEORY OF GAMES
AND

ECONOMIC BEHAVIOR)))

THEORY OF
GAMES
AND ECONOMIC BEHAVIOR))
By

JOHN VON NEUMANN, and OSKARMORGENSTERN))

PRINCETON PRINCETONUNIVERSITY PRESS

1953)))

Copyright
PRINTED

1944, y Princeton University Press b

IN

THE UNITED STATESOP

EDITION),

AMERICA

Second printing (SECOND Third printing,

Fourth printing, Fifth printing (THIRD EDITION), Sixth printing, 1955))

1948 1950

1947

1953

LONDON:

GEOFFREY

CUMBERLEGE

OXFORD

UNIVERSITY

PRESS)))

TO PREFACE FIRSTEDITION
This book contains an expositionand various applicationsof a mathematical theory of games. The theory has been developedby one of us a since 1928nd is now publishedfor the first time in its entirety. The applicationsareof two kinds:On the one hand to gamesin the propersense, a on the other hand to economicnd sociological problemswhich, as we hope to show,arebest approachedfrom this direction. The applicationswhich we shall make to gamesserve at leastas much the to corroborate theory as to investigatethesegames. Thenature of this clearas the investigation proceeds. relationship will become reciprocal direction. a i Our major interestis, of course,n the economicnd sociological Herewe can approachonly the simplestquestions. However,thesequeso tions are of a fundamental character.Furthermore, ur aim is primarily to show that there is a rigorousapproach to thesesubjects, involving, as they do, questionsof parallelor oppositeinterest,perfect or imperfect inforinfluences. mation, free rational decisionor chance
JOHN VON
PRINCETON,

NEUMANN

N.

J.

OSKAR MORGENSTERN.

January,

1943.))

PREFACE SECOND TO EDITION

The second edition differs from the first in some minor respectsnly. o have carriedout as complete elimination of misprintsas possible, nd an a wish to thank several readers ho have helpedus in that respect. We have w added an Appendix containing an axiomatic derivation of numerical utility. This subjectwas discussedin considerable detail, but in the main qualita3. tively, in Section A publication of this proof in a periodical as promised w in the first edition,but we found it more convenient to add it as an Appendix. Various Appendices applicationsto the theory of location of industries on and on questionsof the four and five persongames were also planned,but had to be abandonedbecause the pressure other work. of of Sincepublication of the first edition several papers dealing with the subjectmatter of this book have appeared. The attention of the mathematically interestedreadermay be drawn to the following: A. Wald developeda new theory of the foundations of statistical estimation which is closelyrelated and draws on, the theory to,
We
of)))

vi

PREFACE TO SECONDEDITION

the zero-sum two-person game (\" StatisticalDecisionFunctionsWhich Minimize the Maximum Risk,\" Annals of Mathematics, Vol. 46 (1945) the He pp. 265-280). also extended main theorem of the zero-sumtwoto (\" continuous-infinite-cases, Generalization persongames(cf. 17.6.)certain Zero-Sum of a Theorem von Neumann Concerning Two-PersonGames,\" by A Annals of Mathematics,Vol. 46 (1945), p. 281-286.) new, very simple p and elementary proof of this theorem (which coversalso the more general theorem referred to in footnote 1on page 154)was given by L. H.Loomis, N of (\"Ona Theorem von Neumann,\" Proc. at.Acad.,Vol. 32 (1946) p. 13p 2 results concerning role of pure and of mixed the 215).Further,interesting in strategies the zero-sum two-persongame wereobtained by Kaplanski, Annals of Mathe(\"A Contributionto von Neumann'sTheory of Games,\" b p matics, Vol. 46 (1945), p. 474-479). We alsointend to come ack to vario ous mathematical aspects f this problem. The group theoretical roblem p mentioned in footnote 1on page 258 was solvedby C.Chevalley. interestedreader The economically may find an easier pproach to the a book in the expositionsof L. Hururicz, (\"TheTheory of problemsof this American Economic eview, Vol. 35 (1945), p. 09EconomicBehavior,\" R p 9 a of Marschak(\"Neumann'snd Morgenstern's ew Approach N 925) and Journal of Political conomy, Vol. 54, (1946), to Static Economics,\" E

/.

J.

pp.97-115).
PRINCETON,

JOHN

N. September, 1946.)))

J.

VON NEUMANN OSKAR MORGENSTERN

TO PREFACE THIRDEDITION
TheThird Editiondiffers from the Second Edition only in the elimination such further misprints as have come to our attention in the meantime, and we wish to thank several readers ho have helpedus in that respect. w Since publicationof the Second the Edition,the literature on this subject has increased very considerably. A completebibliographyat this writing includesseveralhundred titles. We are therefore not attempting to give one here. We will only list the following bookson this subject: \" Contributionsto the Theory (1) H.W. Kuhn and A. W. Tucker (eds.), of Games, I,\" Annals of Mathematics Studies,No.24, Princeton(1950),
of

containing fifteen articlesy thirteen authors. b \" Contributionsto the Theory (2) H.W. Kuhn and A. W. Tucker (eds.),
of

Annals of MathematicsStudies,No. 28, rinceton P (1953), Games, II,\"

containing twenty-one

(1950). (4) J. C.C. McKinsey,Introduction to York (1952).

(3) J'. McDonald,Strategy

articlesy b

in

twenty-two authors. Poker,Businessand War, New York the Theory of Games, New

N (5) A. Wald, StatisticalDecisionFunctions, ew York (1950). o (6) Williams, The CompleatStrategyst,Beinga Primern the Theory of Gamesof Strategy, New York (1953). o Bibliographies n the subjectarefound in all of the above booksexcept (6). Extensivework in this field has beendone during the last years by the staff of the RAND Corporation,Santa Monica,California. A bibliography of this work can be found in the RAND publicationRM-950. In the theory of n-persongames,therehave beensomefurther developments in the direction of \" non-cooperative\"games. In this respect, \" particularly the work of J.F. Nash, Non-cooperative Games,\"Annals of Vol. 54, (1951), 286-295, be mentioned. Further must Mathematics, pp. references this work arefound in (1),2),and (4). to ( Of various developments economics mention in particular \"linear in we and the \" assignmentproblem\"which also appear to be programming\" will find with the theory of games. The reader increasinglyconnected indicationsof this $gain in (1),2),and (4). ( in Thetheory of utility suggested Section and in the Appendix to the

J.

SecondEdition has undergoneconsiderabledevelopmenttheoretically,as the well as experimentally, nd in various discussions. In this connection, a

1.3.,

reader ay consult in particular the following: m M. Friedman and L.J. Savage,\"TheUtility Analysis of ChoicesInvolving Risk,\" Journalof Political Economy,Vol. 56,(1948), p. 79-304. p 2
vii)))

viii

PREFACE TO THIRDEDITION

J. Marschak,\"Rational Behavior, UncertainProspects, Measurable and Vol. 18, 1950), 111-141. Utility/' Econometrica, ( pp.
Journalof Political Economy,Vol. 59, (1951), 371-404. pp. M. Friedman and L. J. Savage, \"TheExpectedUtility Hypothesisand
the Measurability of

F.Mostellerand P. Nogee, \" An

ExperimentalMeasurementof Utility,\"

E. Malinvaud, \"Noteon von Neumann-Morgenstem's StrongIndeA pendence xiom.\" with the methodological ritique exercised some of the In connection by c

A. S. Manne, \"The Strong Independence Assumption Gasoline Blendsand Probability Mixtures.\" P.A. Samuelson,"Probability,Utility, and the Independence Axiom.\" \

Journal of Political Economy, Vol. 60, Utility,\" (1952), 463-474. pp. Seealso the Symposiumon CardinalUtilitiesin Econometrica, 20, Vol. (1952): H. Wold, \"OrdinalPreferencesr Cardinal Utility?\" o

contributors to the last-mentionedsymposium,we would like to mention that we applied the axiomatic method in the customaryway with the customary precautions. Thus the strict,axiomatic treatmentof the concept a of utility (in Section3.6. nd in the Appendix) is complementedby an The latter'sfunction is to heuristic preparation (in Sections the the to the reader viewpoints to evaluate and to circumscribe convey I of the subsequent axiomatic procedure. n particular our disvalidity coverswhat of cussionand selection \"natural operations\"in those sections \"indeseemsto us the relevant substrate of the Samuelson-Malinvaud

3.1.-3.5.).

axiom.\" pendence
JOHN
PRINCETON,
VON NEUMANN OSKAR MORGENSTERN

N. J.

January,

1953.)))

TECHNICAL OTE N
in this

a remark in A.3.3. makeuse of somesimpleintegrals.) Conceptsriginating o b in settheory, linear geometry and group theory play an important role, ut and of they areinvariably taken from the early chapters those disciplines are moreover analyzed and explainedin special expositorysections.Neverthelessthe bookis not truly elementary because mathematical deductions the arefrequently intricate and the logical possibilities extensivelyexploited. are Thus no specificknowledgeof any particularbody of advancedmathematics is required. However,the readerwho wants to acquaint himself more thoroughly with the subjectexpoundedhere,will have to familiarize himself with the mathematical way of reasoning definitely beyond its w o routine, primitive phases. The characterf the proceduresill be mostly that of mathematical logics, et theory and functional analysis. s We have attempted to presentthe subjectin such a form that a reader who is moderatelyversedin mathematicscanacquire necessarypractice the in the course this study. We hope that we have not entirely failed in of
what it would be in a mathematical treatise. All definitions and deductions are constrictly t siderably broaderhan they would be there. Besides, purely verbal discussionsand analyses take up a considerablemount of space. We have a in particular tried to give, whenever possible,a parallel verbal exposition for every major mathematical deduction. It is hoped that this procedure will elucidate unmathematical language what the mathematical technique in and will also show where it achieves more than can be done signifies

that no advancedalgebra,or calculus, occurs. (With two, rather unimin et sequ.and Part of portant, exceptions: of the discussion an example

The nature of the problemsinvestigatedand the techniquesemployed a book necessitateprocedure hich in many instancess thoroughly w i mathematical.Themathematical devicesused areelementaryin the sense

etc.,

19.7.

this endeavour.

In accordance this, the presentationis not with

without

In this, as well as in our methodological stand, we are trying to follow the best examples theoretical hysics. of p The reader is not specifically who in interested mathematicsshould at first omit thosesections the book which in his judgment are too matheof matical. We prefer not to give a definite list of them, sincethis judgment marked with an must necessarilybe subjective. However,those sections asteriskin the table of contentsaremost likelyto occur the average reader to will little in this connection. At any rate he will find that theseomissions interfere with the comprehension the early parts, although the logical))) of

it.

TECHNICAL NOTE
chain may in the rigorous sense have suffered an interruption. As he the a proceeds omissionswill gradually assumea more seriouscharacternd the lacunaein the deduction will become ore and more significant. The m reader then advised to start again from the beginningsincethe greater is is familiarity acquired likely to facilitate a betterunderstanding.

ACKNOWLEDGMENT
this publicationpossible. which They arealso greatly indebted to the Princeton University Press has made every effort to publish this book in spiteof wartime difficulties. The publisher has shown at all times the greatest nderstanding for the u

The authors wish to express thanks to Princeton their University and to the Institute for Advanced Study for their generous elp which rendered h

authors'wishes.)))

CONTENTS
PREFACE

TECHNICAL

NOTE

ACKNOWLEDGMENT

v ix x

CHAPTER I FORMULATIONOF THE ECONOMIC PROBLEM

1.THE MATHEMATICAL METHOD IN 1.1.ntroductory remarks I

1.2.Difficulties of the application 1.3.Necessaryimitations of the l 1.4.Concluding remarks 2. QUALITATIVE
IOR

ECONOMICS

of the mathematical method

objectives
BEHAV-

2 6
7

1 1

DISCUSSION THE PROBLEM OF RATIONAL OF

2.1.The problem of rational

2.4.

2.2. 2.3.
2.5.

behavior \"Robinson Crusoe\" economy and social exchangeeconomy The number of variables and the number of participants Thecaseof many participants: Freecompetition The \"Lausanne\" theory

12 13 15
15
20

8 8 9

3.THE NOTION OF UTILITY P and 3.1. references utilities

3.4.

3.2.Principles of measurement: Preliminaries 3.3.Probability and numerical utilities

3.5.

15 16
17

3.6. 3.7. 3.8.

4. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7. 4.8.

discussion Principles of measurement: Detailed Conceptual structure of the axiomatic treatment utilities The axioms and their interpretation General emarks concerning the axioms r o The role of the concept f marginal utility

of

numerical

24 26 28 29
OF

STRUCTURE BEHAVIOR

OF

THE

THEORY: SOLUTIONSND A
for

STANDARDS

4.1.The simplest conceptof a solution

Extension

The solution as a set of imputations The intransitive notion of \"superiority\" or \"domination\" The precise definition of a solution
of our definition Interpretation Gamesand socialorganizations Concluding remarks
in

to all participants

one participant

31 31
33

34

terms of \"standards

of behavior\"

39
40
43 43

37

CHAPTER II O GENERAL FORMAL DESCRIPTIONF GAMES OF STRATEGY

5. INTRODUCTION
S 5.1.hift

46

5.2.

from economics games to General rinciples of classification and of procedure p

of emphasis

46
46)))

CONTENTS

6. THE SIMPLIFIED CONCEPT A GAME OF 6.1.Explanation of the termini technici 6.2.The elementsof the game 6.3.Information and preliminary
6.4.
Preliminarity, transitivity, and signaling GAME

48 48 49
55

51 51
55
57 60 60 63 66

7. THE COMPLETECONCEPT A OF

of the characteristicsf eachmove o 7.2. Thegeneral description

7.1.Variability

8. SETS AND PARTITIONS 8.1.Desirability of a set-theoretical description of a game 8.2.Sets,their properties, and their graphical representation 8.3.Partitions, their properties, and their graphical representation
8.4. Logisticinterpretation
of sets and
partitions

61
67 67
7

*9. THE SET-THEORETICAL DESCRIPTIONF A GAME O The *9.1. partitions which describea game *9.2. Discussionof thesepartitions and their properties

*10. AXIOMATIC FORMULATION *10.1. axioms and their interpretations The *10.2. discussion of the axioms Logistic *10.3. remarks concerning the axioms General
*10.4.Graphical
representation
AND

73 73 76 76
77

11.TRATEGIES THE FINAL SIMPLIFICATION OF THE DESCRIPTION S OF A GAME 79 of a strategy and its formalization 11.1. concept The 79 11.2. The 81 simplification of the description of a game The 84 11.3. role of strategies the simplified form of a game
final

The 11.4. meaning

in

of the zero-sum

restriction

84

12. RELIMINARY SURVEY P 12.1. viewpoints General

CHAPTER III ZERO-SUMTWO-PERSONGAMES: THEORY

85

The 12.2. one-personame g and Chance probability 12.3. The 12.4. next objective

85 85
87 87 88

F 13.UNCTIONAL CALCULUS Basic 13.1. definitions

The 13.2. operations Max and Min 13.3. Commutativity questions The 13.4. mixed case. Saddlepoints Proofsof the main facts 13.5.

88 89 91
93

95

14.STRICTLY DETERMINED GAMES 141. Formulation of the problem 14.2. minorant and the majorant games The 14.3. Discussionof the auxiliary games

98 98 100
101)))

CONTENTS
14.4. Conclusions

14.6. 14.8. *15.AMES WITH G

*15 1

of strict determinateness The interchange of players. Symmetry 14.7. Non strictly determined games Program of a detailed analysis of strict determinateness

14.5.Analysis

105 106 109

110

111
112 112 114 116
117 120 123 124 126 128 128 129 134 138
143

PERFECT INFORMATION Statement of purpose. Induction *15.2. heexactcondition (First step) T

..

*15.3. heexactcondition (Entire induction) T inductive step *15.5. xact discussion of the inductive step (Continuation) E *15.6. heresult in the caseof perfect information T *15.7.Application to Chess T *15.8. he alternative, verbal discussion
*15.4.Exact discussion of the

16. INEARITY AND CONVEXITY L 16.1. Geometricalbackground 16.2. operations Vector 16.3. theorem of the supporting The 16.4. theorem of the alternative The
17.MIXED

hyperplanes for matrices GAMES

17.2.Generalization of this viewpoint 17.3.Justification of the procedure applied to an individual play as 17.4.The minorant and the majorant games. (For mixed strategies) 17.5.Generalstrict determinateness 17.6.Proof of the main theorem 17.7.Comparison of the treatment by pure and by mixed strategies 17.8.Analysis of general strict determinateness 17.9.Further characteristics good strategies of

17.1. Discussionof two

STRATEGIES.

THE SOLUTION ALL FOR

elementary examples

17.10. and their consequences.Permanent Mistakes 17.11. interchange of players. Symmetry The

optimality

143 145 146 149 150 153 155 158 160 162 165

CHAPTER IV ZERO-SUMTWO-PERSONGAMES: EXAMPLES

18. OME ELEMENTARY GAMES S 18.1. simplest games The

Detailedquantitative discussion of thesegames 18.2. 18.3. Qualitative characterizations forms of Matching 18.4. Discussionof some specificgames. (Generalized Pennies) 18.5. Discussionof some slightly more complicatedgames 18.6. Chance imperfect information and

169 169
170 173 175 178

18.7.Interpretation

of this result

P *19.OKER AND BLUFFING *19.1. Descriptionof Poker *19.2. Bluffing *19.3. Descriptionof Poker (Continued)

*19.4.Exact formulation of the rules

182 185 186 186 188 189

190)))

CONTENTS

*19.6. Statement of the problem *19.7. assagefrom the discreteto the continuous problem P Mathematical determination of the solution *19.8. Detailedanalysis of the solution *19.9. *19.10. of the solution Interpretation *19.11. general forms of Poker More 19.12. hands Discrete *19.13. bids m possible Alternate *19.14. bidding Mathematical description of all solutions *19.15. *19. 16.nterpretation of the solutions. Conclusions I
CHAPTERV
G ZERO-SUM THREE-PERSONAMES

of *19.5. escription the D

strategy

191 195 196 199 202

204

208 209 211 216 218

207

20.PRELIMINARY SURVEY 20.1. eneralviewpoints G 20.2. oalitions C

21. HE SIMPLE MAJORITY GAME T 21.1. Definition of the game

21.2. nalysis A 21.3. nalysis A
of the of the

OF THREE PERSONS

of game:Necessity \"understandings\" game:Coalitions. The role of symmetry

220 221 222 222 223

224
225

220

22.FURTHER EXAMPLES 22.1. nsymmetric distributions. Necessity of compensations U 22.2. oalitionsof different strength. Discussion C A 22.3. n inequality. Formulae 23.THE GENERAL CASE 23.1. etaileddiscussion. Inessentialand essentialgames D 23.2. omplete formulae C
24. DISCUSSION AN OBJECTION OF

225
227

229 231 231 232 233

235
233

24.1. he caseof perfectinformation T

and its significance 24.2. Detaileddiscussion. Necessity of compensations between three or more players

CHAPTERVI
FORMULATION OF THE GENERAL THEORY: ZERO-SUM n-PERSONGAMES

25. THE CHARACTERISTIC FUNCTION

25.1. otivation M

and definition 25.2.Discussionf the concept o

25.3.Fundamental
25.4. Immediate

properties
mathematical
GAME

238 240 24 1
242
243
A

238

consequences
WITH GIVEN

26.CONSTRUCTIONOF A FUNCTION 26.1. he construction T 26.2.ummary S

CHARACTERISTIC

243
245)))

CONTENTS 27. STRATEGIC EQUIVALENCE.

27.4. Various criteria.
INESSENTIAL
AND

ESSENTIAL

27.1.Strategic equivalence. Thereducedform

27.2. Inequalities. The quantity y 27.3. Inessentiality and essentiality 27.5. The inequalities in the essentialcase f 27.6. Vector operations on characteristicunctions
Non additive utilities

GAMES

245 245 248 249

28.GROUPS,SYMMETRY AND FAIRNESS 28.1.ermutations, their groups and their effect on a game P 28.2.ymmetry and fairness S 29.RECONSIDERATION OF THE ZERO-SUM THREE-PERSON GAME 29.1. ualitative discussion Q 29.2. uantitative discussion Q 30.THE EXACT FORM OP THE GENERAL DEFINITIONS T 30.1. he definitions 30.2. iscussionnd recapitulation D a

250 252 253 255 255 258

260 260 262 263 265 266

271
272 272 277
263

31.

*30.3. The conceptof saturation 30.4. Threeimmediate objectives FIRST CONSEQUENCES

31.2. he system of all imputations. Oneelement solutions T t 31.3. he isomorphism which correspondso strategic equivalence T

31.1. Convexity, flatness,

and some criteria for

domination

281

32.DETERMINATION OF ALL SOLUTIONS THE ESSENTIAL ZERO-SUM OF THREE-PERSON GAME 282 32.1.ormulation of the mathematical problem. The graphical method 282 F 32.2. etermination of all solutions D 285 33.CONCLUSIONS 288 33.1. he multiplicity of solutions. Discrimination and its meaning T 288 33.2. taticsand dynamics S 290
CHAPTERVII GAMES ZERO-SUMFOUR-PERSON

34.PRELIMINARY SURVEY 34.1. eneralviewpoints G

34.2. Formalism of the essential erosum z 34.3. Permutations of the players

four

person games THE CUBE Q

294

291 291 291

295 295

35.DISCUSSION SOME SPECIAL POINTSIN OF The corner /. (and V., VI., VII.) 35.1.

35.2.The corner VIII.(and //.,///.,7F.,). The three person game and a \"Dummy\" 299 35.3.Someremarks concerning the interior of Q 302
307
312)))

36.DISCUSSION THE MAIN DIAGONALS OP 36.1. he part adjacentto the corner VIII.' Heuristic discussion T 36.2. he part adjacentto the corner VIII.Exact discussion T :
*36.3. Other parts of the
main diagonals

304 304

CONTENTS

313 313 37.2. 315 37.3. 316 37.4. Thesecondalternative at the center 317 37.5. Comparison of the two central solutions 318 319 37.6. Unsymmetrical central solutions J8.A FAMILY OPSOLUTIONS A NEIGHBORHOOD THE CENTER 321 FOR OP *38.1. Transformation of the solution belonging to the first alternative at 321 the center 322 *38.2. Exact discussion
37. THE CENTER AND ITS ENVIRONS

37.1.First

orientation about the conditions around the The two alternatives and the roleof symmetry The first alternative at the center

center

*38.3. Interpretation

of the

solutions

327

CHAPTERVIII

39.THE NUMBER OF PARAMETERS T 39.1. he situation for n - 3, 4 T 39.2. he situation for all n ^ 3
T 10. HE SYMMETRIC
F 40.1.ormalism

n SOME REMARKS CONCERNING ^ 5 PARTICIPANTS

IN VARIOUS

OF CLASSES GAMES

FIVE PERSON GAME

40.2. The two extreme cases 40.3. Connectionbetween the symmetric

symmetric four person game

of the symmetric

five

person game
five

person game and

the

1,2, 3-

330 330 330 332 332 332

334

CHAPTERIX

11. OMPOSITION C

AND OF COMPOSITION DECOMPOSITION GAMES

AND

T 41.2. he first type. Composition 41.3. xact definitions E 41.4. nalysis of decomposability A 41.5. esirability of a modification D

41.1. for n-person games for which Search

DECOMPOSITION
all solutions

can be determined

and decomposition

12.MODIFICATION OF THE THEORY N 42.1. o completeabandonment of the zerosum restriction

42.2. Strategicequivalence. Constant sum games function in the new theory 42.3. The characteristic 42.4. Imputations, domination, solutions in the new theory 42.5. Essentiality, inessentiality and decomposability in the

350
new theory

343 345 345 345 346 348

339 339 340 341

351
353 353 353

PARTITION T 13. HE DECOMPOSITION 43.1.plitting sets. Constituents S

43.2. Propertiesof the system of all splitting sets 43.3. Characterization of the system of all splitting sets. The decomposition partition 354 357 43.4. Propertiesof the decomposition partition 44.2. Composition and decompositionof imputations and of setsof imputations

14.DECOMPOSABLE AMES. FURTHER EXTENSION OF THE THEORY 358 G 44.1.olutions of a (decomposable) and solutions of its constituents 358 S game
359)))

CONTENTS
44.3. Composition and decomposition of solutions.
and surmises Extension of the theory.

The main

possibilities

363 Outsidesources 44.4. 364 44.5. Theexcess 44.6. Limitations of the excess.The non-isolated characterof a game in
the new

361
367

44.7. Discussionof the 45. LIMITATIONS OF

THEORY THE

setup

366
368

new

setup. E(e )t F(e )

Q

EXCESS.STRUCTURE OF

THE

EXTENDED

45.2. 45.3. 45.4.

T 45.1. he lower

45.5. 45.6. 46.2.

limit of the excess The upper limit of the excess.Detachedand fully detachedimputations Discussionof the two limits, 2 Their ratio Detachedimputations and various solutions. The theorem connecting E(e ), F(e ) Proof of the theorem Summary and conclusions

|r|i,r|. |

368 369
372 375 376

46.DETERMINATION 46.1.lementary E

46.10. Imbedding of a game 46.11. Significance of the normal zone o 46.12. occurrencef the phenomenon First
47. THE
ESSENTIAL

46.3. Continuation 46.4. Continuation 46.5. The complete result in F(e ) result in E(e ) 46.6. The complete 46.7. Graphical representation of a part of the result 46.8. Interpretation: The normal zone. Heredity of various properties 46.9. Dummies
of

OF OF ALL SOLUTIONS A DECOMPOSABLE GAME of decompositions properties Decompositionand its relation to the solutions: First results con384 cerning F( )

380 381 381 386 388 390 393 394 396 397 398 401
402
403

transfer: n
IN

-6
THEORY

GAME THREE-PERSON this discussion

THE NEW

47.2. Preparatory considerations 47.3. The six casesof the discussion. Cases(I)-(III) 47.4. Case(IV) :First part 47.5. Case(IV): Second part 47.6. Case(V) 47.7. Case(VI) 47.8. Interpretation of the result: The curves (one dimensional parts) 47.9.
the solution Continuation:

47.1.Needfor

403 403 406 407 409

in

Theareas (two

dimensional

parts)

in the solution

413 415 416 418

CHAPTER X SIMPLEGAMES
A 48.WINNING AND LOSINGCOALITIONSND GAMES OCCUR Decision coalitions T 48.1. hesecondtype of 41.1. by

WHERE

THEY

48.2. Winning

420 420
421)))

and Losing

Coalitions

CONTENTS 49. CHARACTERIZATION OP THE SIMPLE GAMES '49.2. The specialrole of one element sets 49.3. Characterization of the systems TF, L of actual games 49.4. Exact definition of simplicity 49.5. Some elementary properties of simplicity 49.6. Simple gamesand their W, L. The Minimal winning coalitions: m W 49.7. The solutions of simple g^mes

49.1. eneralconceptsof winning G

423

and losing coalitions

423 425 426 428 428 429 430

50. THE MAJORITY 50.3. 50.4. 50.5.

50.2.

433 435 436 438 50.6. 440 50.7. 442 50.8. 443 51. ETHODSOR THE ENUMERATION OF ALL SIMPLE GAMES M F 445 51.1. 445 Preliminary Remarks 51.2. hesaturation method: Enumeration by means of W T 446 51.3. easonsfor passing from W to Wm . Difficulties of using W m R 448 51.4.Changed Approach: Enumeration by means of Wm 450 51.5.implicity and decomposition 452 S 51.6.nessentiality, Simplicity and Composition. Treatment of the excess454 I 51.7.A criterium of decomposability in terms of W m 455 52. THE SIMPLE GAMES FOR SMALL n 457 o 52.1.rogram, n = 1,2 play no role. Disposal f n = 3 457 P f 52.2.Procedureor n ^ 4: Thetwo element setsand their role in classify458 ing the W m 52.3.Decomposability of casesC*, Cn _ 2, Cn_i 459
Homogeneity A more direct use of the concept f imputation in forming solutions o Discussionof this direct approach Exact formulation Connections with the general theory. Reformulation of the result of the result Interpretation Connection with the Homogeneous Majority game.

50.1.xamples of simple games:Themajority E

GAMES AND THE MAIN

SOLUTION
games

431 431

n 52.4. The simple games other than [1, 2]* (with dummies): , 1, TheCasesCk , k 0, 1, 461 ,n 3 o 52.5.Disposal f n = 4, 5 462 OF 53. THE NEW POSSIBILITIES SIMPLE GAMES FOR n ^ 6 463 The Regularities observedfor n ^ 6 53.1. 463 53.2.The six main counter examples (for n * 6, 7) 464

54.

DETERMINATION

54.1. easonsto consider other R

games

IN OF ALL SOLUTIONS

SUITABLE

GAMES

470 470

solutions than the main solution in simple solutions

54.2. Enumeration of those games for which all to 54.3. Reasons considerthe simple game [1,

*55.THE SIMPLE GAME [1, *55.1.reliminary Remarks P

1,n

are known n , 1, 2]
A

2]h

*55.2. Domination. The chief player. Cases(I) and (II) o *55.3. Disposal f Case(I) D *55.4. Case(II):etermination of Y D *55.5. Case(II):etermination of V a *55.6. Case(II): and S*

472 473 473 473 475 478

471

481
484)))

CONTENTS
*55.7. Case(II')and (II\.") Disposal f Case(IF) o *55.8^Case(II\:") and V. Domination a *55.9. Case(II\:") etermination of V D *55.10. o Disposal f Case(II\") *55.11. of the complete Reformulation result

*55.12. Interpretation

of the

result

485 487 488 494 497 499

CHAPTERXI
GENERAL NON-ZERO-SUMGAMES

56.EXTENSION OF THE THEORY 56.1.ormulation of the problem F

504

56.2.The fictitious player. Thezerosum extension r 56.3.Questionsconcerning the characterof r 56.4. Limitations of the use of r 56.5.The two possible procedures 56.6.The discriminatory solutions 56.7. Alternative possibilities 56.8.The new setup 56.9.Reconsiderationof the casewhen T is a zerosum game

504

505 506 508

.

56.10. of the conceptof domination Analysis 56.11. Rigorous discussion 56.12. new definition of a solution The

510 511 512 514 516

520 523 526

57. THE CHARACTERISTIC FUNCTION AND RELATED TOPICS 527 57.1.The characteristic function: The extendedand the restrictedform 527 57.2. Fundamental properties 528 57.3. Determination of all characteristicunctions f 530 57.4. Removable sets of players 533 57.5. Strategicequivalence. Zero-sum and constant-sum games 535 58.INTERPRETATION OF THE CHARACTERISTIC FUNCTION 538 58.1. nalysis of the definition 538 A 58.2.Thedesireto make a gain vs. that to inflict a loss 539 58.3.Discussion 541 542 59. GENERAL CONSIDERATIONS 59.2.Thereducedforms. 59.3.Various topics
o D 59.1. iscussionf the program

The inequalities
GENERAL GAMES WITH

O 60.THE SOLUTIONSF ALL 60.1. he casen - 1 T T 60.2. he casen - 2 T 60.3. he casen = 3

^3

542 543 546 548 548 549 554

61.

zerosum games 2 FOR n = ECONOMICINTERPRETATION OF THE RESULTS Thecasen 1 T 61.2. hecasen = 2. The two person market f of 61.3. iscussion the two person market and its characteristicunction D 61.4.ustification of the standpoint of 58 J 61.5. ivisiblegoods. The \"marginal pairs\" D T 61.6. he price. Discussion
with

60.4. Comparison

550

the

61.1.

1,

555 555 555 557 559 560

562)))

CONTENTS
INTERPRETATION 62.ECONOMIC
F OF THE RESULTSOR n

= 3:SPECIAL
564

CASE

62.1. hecasen = 3, specialcase. The three person market T 62.2.reliminary discussion P 62.3. he solutions: First subcase T
62.4. The solutions:Generalform

564
569

566 566
570

62.6. iscussion D
CASE

62.5.

Algebraical

form of the result

571 = 3: GENERAL
573 573 577 577

63.ECONOMICINTERPRETATION 63.1. ivisible goods D

OFTHE RESULTSOR n F

63.2. Analysis of the inequalities 63.3. Preliminary discussion 63.4. The solutions 63.5. Algebraical form of the result 63.6. Discussion

575

580
583 583 584

581

64. THE GENERAL

64.2.

64.1.ormulation F

MARKET of the problem Somespecial roperties. Monopoly and monopsony p

CHAPTER XII

EXTENSIONOF THE CONCEPTS DOMINATION OF

AND SOLUTION

65. THE EXTENSION.

65.1.ormulation F

65.2.Generalemarks r 65.3.Orderings, transitivity, acyclicity 65.4. The solutions: For a symmetric relation. For a completeordering 65.5.The solutions: For a partial ordering 65.6.Acyclicity and strict acyclicity 65.7. The solutions: For an acyclic relation 65.8.Uniqueness 'of solutions, acyclicity and strict acyclicity 65.9.Application to games:Discreteness continuity and

SPECIAL CASES of the problem

588 589 591 592

594
597

587 587

600 602 603 604 606

607
608 603

66.GENERALIZATION OF THE CONCEPT UTILITY OF T 66.1. he generalization. The two phasesof the theoretical 66.2. Discussionof the first phase 66.3. iscussionof the secondphase D
66.4. Desirability of unifying the O 67. DISCUSSIONF AN EXAMPLE
two

treatment

phases

67.1. escriptionof the D

67.2. The solution

example

67.3. Generalization:Different discreteutility scales

APPENDIX:
INDEX

and its interpretation

67.4. Conclusions concerning bargaining THE AXIOMATIC TREATMENT OF UTILITY

OF FIGURES OF NAMES OF SUBJECTS

608 611 614 616

617
633 634
635)))

INDEX
INDEX

CHAPTER I FORMULATION THE ECONOMICPROBLEM OF

The purposeof this bookis to presenta discussionof some fundamental questions of economicheory which requirea treatmentdifferent t from that which they have found thus far in the literature. The analysis is concerned with some basic problemsarising from a study of economic behavior which have beenthe center attention of economists a long of for time. They have their origin in the attempts to find an exactdescription of the endeavorof the individual to obtain a maximum of utility, or,in the caseof the entrepreneur, maximum of profit. It is well known what a considerableand in fact unsurmounted difficulties this task involves in given even a limited number of typical situations, as, for example, the of caseof the exchange goods,director indirect,between two or more persons,of bilateral monopoly, of duopoly,of oligopoly,and of free competition. It will be made clear the structureof theseproblems,familiar that is to every student of economics, in many respectsuite different from the q way in which they are conceivedat the present time. It will appear, furthermore, that their exact positingand subsequentsolution can only be achievedwith the aid of mathematical methods which divergeconsiderably from the techniques applied by olderor by contemporary mathematical
lead to the applicationof the mathematical o of \"games f strategy\" developedby one of us in severalsuccessive theory 1 After of and stagesin 1928 1940-1941. the presentation this theory, its problems in the sense indicated above will be application to economic t undertaken. It will appearhat it providesa new approachto a number of economicuestionsas yet unsettled. q We shall first have to find in which way this theory of games can be t brought into relationshipwith economicheory, and what their common are. This can be done bestby stating briefly the nature of some elements fundamental economic problems so that the common elementswill be seenclearly. It will then becomepparent that there is not only nothing a artificial in establishing this relationship but that on the contrary this
1The first phasesof this- work were published: von Neumann, \"Zur Theorie der M Gesellschaftsspiele,\" ath. Annalen, vol. 100 (1928),pp. 295-320. The subsequent completion of the theory, as well as the more detailed elaboration of the considerations of loc.cit. above,are published here for the first time.
1)))

1.1.1.

1.The Mathematical Method in Economics 1.1. Introductory Remarks

economists.

w 1.1.2. considerations ill Our

J.

FORMULATIONOF THE ECONOMIC PROBLEM

theory of gamesof strategy is the proper instrument with which to develop b a theory of economicehavior. One would misunderstandthe intent of our discussions interpreting by them as merely pointing out an analogy between thesetwo spheres.We a hope to establishsatisfactorily,after developing few plausibleschematizabehavior become that the typical problems of economic strictly tions, identicalwith the mathematical notions of suitable gamesof strategy.

1.2. Difficulties
It 1.2.1.may

of the Application

of the Mathematical Method

a of the facts with which economists re dealing. Only those wlio fail to this appreciate condition arelikelyto attempt the constructionof universal which are far more advanced than economics, systems. Even in sciences like physics,thereis no universal system available at present.
To continue the simile with physics: It happens occasionallythat a particular physical theory appears to provide the basis for a universal h up system, but in all instances to the presenttime this appearance as not at lasted more than a decade best. The everyday work of the research such high aims, but rather is conphysicist is certainly not involved with \" cernedwith special problemswhich are mature.\" Therewould probably at be no progress all in physicsif a seriousattempt weremade to enforce that super-standard.The physicist works on individual problems,some of great practicalignificance, others of less.Unifications of fields which s were formerly divided and far apart may alternatewith this type of work. a However,such fortunate occurrencesrerareand happen only after each field has beenthoroughly explored.Consideringthe fact that economics is much more difficult, much less understood,and undoubtedlyin a much earlier tageof its evolution as a science physics,oneshouldclearlynot than s more either. expect than a developmentof the above type in economics Secondwe have to noticethat the differences in scientific questions make it necessaryto employ varying methods which may afterwards have to be discardedif betteronesoffer themselves. This has a doubleimplicaI tion:n somebranchesof economics most fruitful work may be that of the this may be by far the largestdomain careful, patient description; indeed for the presentand for some time to come. In othersit may be possible to developalready a theory in a strictmanner, and for that purpose the use of mathematicsmay be required.)))

the be opportuneto begin with someremarks concerning and to discuss briefly the question of the role nature of economic theory which mathematicsmay take in its development. Firstlet us be aware that thereexists at presentno universal system of economicheory and that, if one should ever be developed,it will very t probably not be during our lifetime. The reasonfor this is simply that to is economics far too difficult a science permit its construction rapidly, especiallyin view of the very limited knowledgeand imperfect description

THE MATHEMATICALMETHOD IN ECONOMICS

in an

Mathematicshas actually beenused in economic theory, perhaps even m exaggerated anner. In any caseits use has not been highly successful. This is contrary to what one observes in other sciences: There and mathematicshas beenapplied with great success, most sciences could hardly get along without it. Yet the explanationfor this phenomenon is
fairly

there exists fundamental reason why matheany The matics should not be used in economics. argumentsoften heard that f or o o becausef the human element, f the psychologicalactors etc., because there is allegedly no measurement of important factors, mathematics will find no application,can all be dismissedas utterly mistaken. Almost all these objections have been made, or might have been made, many centuriesago in fields where mathematicsis now the chief instrument of L analysis. This \" might have been\"is meant in the following sense: et the us try to imagine ourselvesin the period which preceded mathematical or almost mathematical phase of the developmentin physics, that is the 16th century, or in chemistry and biology, that is the 18th century. Taking for granted the skeptical attitude of those who objectto mathein matical economics principle,the outlook in the physical and biological at these early periods can hardly have been betterthan that in sciences economicsmutatis
mutandis

1.2.2. is not that It

simple.

at

As to the lack of measurement of the most important factors, the of example the theory of heat is most instructive ; before the developmentof the mathematical theory the possibilitiesof quantitative measurements

present.

The precise they are now in economics. of measurements the quantity and quality of heat (energyand temperature) of were the outcomeand not the antecedents the mathematical theory. This ought to be contrasted with the fact that the quantitative and exact notions of prices, oney and the rate of interestwere already developed m
were

lessfavorable there than

further group of objections against quantitative measurementsin the lack of indefinite divisibility of economic quantities. This is supposedlyincompatible with the use of the infinitesimalcalculusand hence( !) of mathematics. It is hard to seehow such in c objectionsan be maintained in view of the atomic theories physicsand the theory of quanta in electrodynamics, tc., the notorious e and chemistry, of and continuedsuccess mathematical analysiswithin thesedisciplines. At this point it is appropriate to mention another familiar argument of literature which may be revived as an objectionagainst the economic mathematical procedure. the w In orderto elucidate conceptions hich we are applying to we have given and may give again some illustrations from economics, s physics. There are many social cientistswho objectto the drawing of such parallels on various grounds, among which is generally found the that assertion economic theory cannot bemodeledafter physicssinceit is
A

centuries ago.

centersaround economics,

1.2.3.

a)))

FORMULATIONOF THE ECONOMICPROBLEM

doubt reasonable to discoverwhat has led to progressn other sciences, to investigate i and whether the application of the same principlesmay not lead to progress in economics also.Shouldthe needfor the applicationof different principles arise,it could be revealed only in the courseof the actual development of economic theory. This would itself constitutea major revolution. But sincemost assuredlywe have not yet reached such a state and it is by no means certain that there ever will be need for entirely different scientific principles it would be very unwise to consider anything else than the pursuit of our problemsin the manner which has resulted in the establishmentof physicalscience. 1.2.4* reason why mathematics has not been more successfulin The economics must, consequently, be found elsewhere.The lack of real is success largely due to a combination of unfavorable circumstances, some of which can be removed gradually. Tobeginwith, the economicroblems p were not formulated clearlyand areoften stated in such vague terms as to make mathematical treatmenta prioriappear hopeless ecause is quite it b uncertain what the problemsreally are. Thereis no point in using exact methods where there is no clarity in the concepts nd issuesto which they a areto beapplied. Consequentlythe initial task is to clarify the knowledge of the matterby further careful descriptive work. But even in those where the descriptiveproblem has been handled more parts of economics mathematical tools have seldom been used appropriately. satisfactorily, a They were eitherinadequatelyhandled, as in the attempts to determine economicquilibrium by the mere of numbersof equations e counting general and unknowns,or they led to meretranslations from a literary form of into expression symbols,without any subsequentmathematical analysis. the empiricalbackgroundof economiccience definitely inades is Next, Our knowledge the relevant facts of economics incomparably of is quate. smaller than that commandedin physics at the time when the mathematization of that subjectwas achieved. Indeed, decisivebreak the which camein physics in the seventeenth century, specificallyin the field of was of mechanics, possibleonly because previousdevelopmentsin astronIt was backed several millennia of systematic,scientific,astroby omy. nomical observation, culminating in an observer of unparalleled caliber, in science.It Tycho deBrahe. Nothing of this sorthas occurred economic have been absurd in physicsto expect eplerand Newton without would K Tycho, and there is no reasonto hope for an easierdevelopment in

etc. Suchstatements areat leastpremature. It is without

into science social, human phenomena,has to takepsychology account, of of

economics.

These obvious comments should not be construed,of course, s a a of research which holdsthe realpromise disparagement statistical-economic in of progress the proper direction. It is due to the combination of the above mentioned circumstances has that mathematical economics not achievedvery much. Theunderlying)))

THE MATHEMATICALMETHOD IN ECONOMICS

vaguenessand ignorance has not been dispelledby the inadequate and inappropriate use of a powerful instrument that is very difficult to

handle. our Inthe light of theseremarkswe may describe own positionas follows: The aim of this book lies not in the direction of empiricalresearch.The on advancement of that side of economiccience, anything like the scale s which was recognizedbove as necessary,is clearly a task of vast propora tions. It may be hoped that as a result of the improvements of scientific gained in other fields, the developmentof techniqueand of experience economicsill not take as much timeas the comparisonwith w descriptive t astronomy would suggest. But in any casethe task seemso transcend the limits of any individually plannedprogram. concernWe shall attempt to utilize only somecommonplace xperience e human behavior which lends itself to mathematical treatment and ing which is of economic importance. We believe that the possibilityof a mathematical treatmentof these r phenomenarefutes the \"fundamental' objectionseferredto in 1.2.2. o It will be seen,however, that this processf mathematization is not m at all obvious. Indeed,he objections entioned above may have their t
1

roots partly in the rather obvious difficulties of any directmathematical approach. We shall find it necessaryto draw upon techniquesof matheand maticswhich have not beenusedheretofore in mathematical economics, result in the future in the creation it is quitepossible that further study may

of new mathematical disciplines. To conclude, may alsoobservethat part of the feeling of dissatisfacwe tion with the mathematical treatment of economic theory derives largely not proofs but mere assertions from the fact that frequently one is offered which are really no betterthan the sameassertionsgiven in literary form. a Very frequently the proofs are lacking because mathematical treatment has beenattempted of fields which areso vast and so complicatedthat for is a long time to come until much more empiricalknowledge acquired more mathematico. there is hardly any reasonat all to expectprogress in The fact that thesefields have beenattacked this way as for example the theory of economic fluctuations, the time structureof production,etc. indicates how much the attendant difficulties are being underestimated. They areenormousand we arenow in no way equippedfor them. We have referred to the nature and the possibilitiesof those in changes mathematical technique in fact, in mathematicsitself which a successfulapplication of mathematics to a new subjectmay produce. It is important to visualize thesein their proper perspective. It must not be forgotten that thesechanges be very considerable. may The decisivephase of the applicationof mathematicsto physics Newton's creationof a rational discipline of mechanics brought about, and can from, the discovery of the infinitesimal calculus. hardly be separated but areseveralother examples, none strongerthan this.)))) (There

1.2.6.

FORMULATIONOF THE ECONOMIC PROBLEM

The importance of the social henomena,the wealth and multiplicity p of theii manifestations, and the complexityof their structure,aie at least equal to thosein physics. It is therefore to be expectedor feared that will mathematical discoveriesof a stature comparableto that of calculus be needed ordei to producedecisivesuccessn this field. (Incidentally, in i it is in this spirit that our present efforts must be discounted.) A fortiori it is unlikely that a mere repetition of the tricks which servedus so well in physicswill do for the socialphenomenatoo. The probabilityis very slim indeed,sinceit will be shown that we encounter in our discussionssome mathematical problemswhich arequite different from those which occurin

physicalscience. Theseobservationsshouldberemembered connectionith the current in w a overemphasison the use of calculus, differential equations, etc., s the main tools of mathematical economics.

Objectives We have to return, therefore, to the position indicated earlier: It is necessaryto begin with those problemswhich aredescribedlearly, c even if they should not be as important from any other point of view. It shouldbeadded, moreover, that a treatment of thesemanageableroblems p may lead to results which are already fairly well known, but the exact proofs may neverthelessbe lacking. Before they have been given the respective theory simply does not existas a scientific theory. The movements of the planets were known long before their courses beencalcuhad lated and explainedby Newton's theory, and the sameapplies in many smaller and less dramatic instances. And similarly in economic theory, certainresults say the indeterminatenessof bilateral monopoly may be known already. Yet it is of interestto derive them again from an exact theory. The same could and should be said concerning practically all
of the

1.3.1.

1.3. Limitations Necessary

establishedeconomicheorems. t 1.3.2. might be added finally that we do not propose to raisethe It question of the practicalignificance of the problemstreated. This falls s in line with what was said above about the selection fields for theory. of The situation is not different herefrom that in othersciences.Theretoo the most important questionsfrom a practical point of view may have been completelyout of reachduring long and fruitful periods of their development. This is certainly still the casein economics, where it is of utmost the importance to know how to stabilize employment, how to increase national income, or how to distribute it adequately. Nobody can really answer these questions, and we need not concern ourselveswith the pretension that there can be scientific answersat present. Thegreatprogressin every scienceame c when, in the study of problems which were modest as comparedwith ultimate aims, methods weredevelfurther and furthei. The free fall is a very oped which couldbe extended trivial physicalphenomenon, ut it was the study of this exceedingly b simple)))

THE MATHEMATICALMETHOD IN ECONOMICS

mechanics.

fact and its comparison with the astronomical aterial, which brought forth m
1

It seemso us that the samestandard of modestyshould be applied in t It economics. is futile to try to explain and \" systematically' at that T everything economic. he sound procedureis to obtain first utmost and to precision masteryin a limited field, and then to proceed another, someo theoriesto economicr social reform where o practicef applying so-called they arein no way useful.
what wider one,and

so on. This would also do away

with

the unhealthy

We believethat it is necessaryto know as much as possibleabout the behavior of the individual and about the simplestforms of exchange. This standpoint was actually adopted with remarkablesuccess y the founders b of the marginal utility school, ut neverthelessit is not generallyaccepted. b \" Economists m \" frequently point to much larger, ore burning questions,and brush everything aside which prevents them from making statements about these. The experience more advanced sciences, example for of indicatesthat this impatience merely delays progress,ncluding physics, i that of the treatment of the \" burning \" questions. Thereis no reason to assumethe existence shortcuts. of

It is essentialto realize that economistsan expect o easierate c n f than that which befell scientistsin other disciplines.It seemseasonable r to expect they will have to take up first problemscontainedin the very that facts of economic and try to establish theorieswhich explain life simplest them and which really conform to rigorous scientific standards. We can have enough confidence that from then on the sciencef economicsill o w than of more vital impoitance matters grow further, gradually comprising 1 those with which one has to begin. The field covered in this book is very limited, and we approach it in this senseof modesty. We do not worry at all if the results of oui study conform with views gained recently or held for a long time,for what is important is the gradual development of a theory, based on a careful f analysis of the ordinary everyday interpretation of economicacts. This t preliminary stage is necessarilyheuristic, i.e.he phase of transition from unmathematical plausibility considerations to the formal procedureof mathematics.The theory finally obtained must be mathematically rigorous and conceptually general. Its first applications are necessarily to elementary problems where the result has never been in doubt and no theory is actually required. At this early stage the applicationselvesto corroborate theory. Thenext stagedevelopswhen the theory is applied the
national

1.4.

1.4. Concluding

Remarks

1The beginning is actually of a certain significance, because forms of exchange the a few individuals are the same as those observed someof the most important on markets of modern industry, or in the caseof barter exchangebetween states in interbetween
trade.)))

FORlVgJLATION OF

THE ECONOMICPROBLEM
which

certainextent beyond the obvious and the familiar. Heretheory and applicationcorroborateeachother mutually. Beyondthis liesthe field of

to somewhat more complicated situationsin

it may

already lead to a

realsuccess: genuine prediction by theory. It is

well known that all mathematized scienceshave gone through these successive phases of evolution.

The subjectmatter of economic theory is the very complicated mechanism of prices and production,and of the gaining and spendingof i incomes. In the courseof the development of economicst has been t found, and it is now well-nighuniversally agreed,hat an approachto this vast problemis gainedby the analysis of the behavior of the individuals which constitutethe economic community. This analysishas been pushed and much disagreement fairly far in many respects, while therestill exists the significance of the approach cannot be doubted, no matter how great its difficulties may be. The obstaclesareindeed considerable, ven if the e shouldat first be limited to conditionsof economics a investigation statics, s well must be. One of the chief difficulties lies in properly describing they the assumptions which have to be madeabout the motives of the individual. This problemhas beenstated traditionally by assumingthat the consumer desiresto obtain a maximum of utility or satisfaction and the entrepreneur a maximum of profits. The conceptualand practical difficulties of the notion of utility, and it particularly of the attempts to describe as a number, arewell known and their treatment is not among the primary objectives this work. We shall of neverthelessbe forced to discussthem in some instances,in particular in and Let it be said at once that the standpoint of the presentbook on this very important and very interesting questionwill be mainly opportunistic. We wish to concentrate one problem which is not that of on the measurement of utilities and of preferences and we shall therefore as attempt to simplify all other characteristics far as reasonablypossible. We shall therefore assumethat the aim of all participants in the economic system, consumersas well as entrepreneurs,is money, or equivalently a singlemonetary commodity. This is supposedto be unrestrictedlydivisible and substitutable,freely transferable and identical,even in the quantitative sense,with whatever \" satisfaction\"or \" utility\" is desired by each paro ticipant. (Forthe quantitative character f utility, cf. quoted above.) It is sometimesclaimedin economiciterature that discussionsof the l notions of utility and preference altogether nnecessary, incetheseare are s u purely verbal definitions with no empirically observableconsequences, to entirely tautological. It does not seem us that thesenotions are qualiinferior to certainwell establishedand indispensablenotions tatively

2.1.1.

2.Qualitative Discussionf the Problem Rational Behavior o of 2.1. Problem of Rational Behavior The

3.3. 3.5.

3.3.

i.e.,
in)))

Thus the notion of utility is raised above the status of a tautology by such economic a theories s makeuse of it and the resultsof which canbecompared with experience at least with common sense. or 2.1.2. individual who attempts to obtain theserespectivemaxima The is also said to act \"rationally.\" But it may safely be stated that there exists,at present,no satisfactory treatment of the question of rational behavior. Theremay, for example, several ways by which to reach exist the optimum position; hey may depend upon the knowledgeand undert standing which the individual has and upon the paths of action open to him. A study of all thesequestionsin qualitative terms will not exhaust them, because they imply, as must be evident, quantitative relationships. It would, therefore, be necessaryto formulate them in quantitative terms so that all the elements the qualitative descriptionare taken into conof sideration. This is an exceedinglydifficult task, and we can safely say that it has not been accomplishedin the extensive literatureabout the topic. The chief reasonfor this lies,no doubt, in the failure to develop and apply suitable mathematical methods to the problem; this would have revealedthat the maximum problemwhich is supposedto correspond to the notion of rationality is not at all formulated in an unambiguous way. a reveals Indeed, more exhaustive analysis (to be given in 4.S.-4.5.) that the significant relationshipsare much more complicatedthan the popular and the \" philosophical\" se of the word \" rational\" indicates. u A valuable qualitative preliminary description of the behavior of the individual is offered by the Austrian School, articularly in analyzing the p to economyof the isolated \" Robinson Crusoe.\"We may have occasion note also some considerationsof Bohm-Bawerkconcerning exchange the betweentwo or more persons. The more recentexpositionof the theory of i the individual'schoicesn the form of indifference curve analysisbuilds up on the very samefacts or alleged facts but usesa method which is often held this i to be superiorin many ways. Concerning we refer to the discussionsn 2.1.1. 3.3. and We hope,however, to obtain a realunderstanding of the problem of d exchange y studying it from an altogether ifferent angle;this is, from the b of a \"gameof strategy.\" Our approach will become clear perspective after some ideaswhich have been advanced, say by presently, especially Bohm-Bawerk whoseviews may be considerednly as a prototype of this o
theory

THE PROBLEM OF RATIONAL BEHAVIOR 9 physics, like force, mass, charge,etc. That is, while they are in their immediate form merely definitions, they become subjectto empiricalcontrol through the theorieswhich are built upon them and in no other way.

aregiven correct uantitative q

formulation.

SocialExchange Economy Let us look more closelyat the type of economy which is representedby the \"RobinsonCrusoe\" odel,that is an economyof an isolated m singleperson or otherwiseorganizedunder a singlewill. This economy
Crusoe\" Economy and

2.2. \"Robinson

2.2.1.

is)))

10

FORMULATIONOF THE ECONOMIC PROBLEM

with

and certainquantities of commodities a number of wants satisfy. Theproblemis to obtain a maximum satisfaction. This is consideringin particular our above assumptionof the numerical characterf utility indeed an ordinary maximum problem, its difficulty o dependingappaiently on the number of variablesand on the nature of the function to be maximized;but this is more of a practical difficulty than a theoretical ne. If one abstracts from continuous production and from o 1 the fact that consumption too stretchesover time (and often uses durable consumers'goods),one obtains the simplest possible model. It was thought possibleto use it as the very basis for economic theory, but this attempt notably a feature of the Austrian version was often contested. The chief objection againstusing this very simplified model of an isolated individual for the theory of a socialexchangeconomy is that it does not e to a represent n individual exposed the manifold socialinfluences. Hence, it is said to analyze an individual who might behave quite differently if his w to choices ere made in a social world where he would be exposed factors of imitation, advertising,custom,and so on. Thesefactors certainly make the a greatdifference, but it is to be questionedwhether they change formal propertiesof the processof maximizing. Indeedthe latterhas never been w implied,and sincewe areconcernedith this problemalone,we can leave out the above socialconsiderations of account. \" Someother differences between\" Crusoe and a participant in a social us of e exchangeconomy will not concern either. Suchis the non-existence in where thereis only a standard money as a meansof exchange the first case for of calculation, which purposeany commodity can serve. This difficulty a quantitative indeed has beenploughedunder by our assumingin 2.1.2. and even monetary notion of utility. We emphasizeagain:Our interest lies in the fact that even after all these drastic simplifications Crusoeis confronted with a formal problemquite different from the one a participant confronted
which they may in a

physical data (wants and commodities) and his task is to combineand apply them in such a fashion as to obtain a maximum resultingsatisfaction. Therecan be no doubt that he controls exclusively all the variables upon which this result depends say the the allotting of resources, determination of the uses of the same commodity 2 for different wants, etc. Thus Crusoefaces an ordinary maximum problem, the difficulties of which areof a purely technical and not conceptual nature, as pointedout. Considernow a participant in a social xchange e economy. His i m problemhas, of course, any elementsn common with a maximum prob-

2.2.2. Crusoeis given certain

socialconomy faces. e

2.2.3.

1It is not

all its

2 Sometimes uncontrollable factors also intervene, e.g.he weather in agriculture. t Thesehowever are purely statistical phenomena. Consequently they can be eliminated o by by the known procedures f the calculus of probabilities: i.e., determining the prob\" abilities of the various alternatives and by introduction of the notion of mathematical in Cf however the influence on the notion of utility, discussed expectation.\"

aspects.

important

for the following

to determine

whether its theory

is completein

3.3.)))

THE PROBLEM OF RATIONAL BEHAVIOR

11

lem. But it also contains some, ery essential,elementsof an entirely v different nature. Hetoo tries to obtain an optimum result. But in order to achieve this, he must enter into relations of exchange with others. If two or more personsexchange goods with eachother, then the result for each one will depend in generalnot merely upon his own actions but on thoseof the others as well. Thus each participant attempts to maximize a function (his above-mentioned o \"result\")f which he does not control all variables. This is certainly no maximum problem,but a peculiarand dis-~ m concerting ixture of severalconflicting maximum problems. Every partia cipant is guided by another principleand neither determines ll variables which affect his interest. This kind of problem is nowhere dealt with in classical mathematics. at We emphasize the risk of beingpedanticthat this is no conditional maximum problem, no problem of the calculus of variations, of functional analysis, etc. It arises in full clarity, even in the most \" elementary\" when situations,e.g., all variablescan assumeonly a finite number of values. A particularly striking expression of the popular misunderstanding about this pseudo-maximum to problemis the famous statementaccording which the purpose of social effort is the \"greatest ossiblegood for the p greatest possible number.\" A guiding principle cannot be formulated by the requirementof maximizing two (or more) functions at once. Such a principle,taken literally, is self-contradictory, (in generalone function will have no maximum where the other function has one.) It is that a firm should obtain maximum prices no better than saying, e.g., at maximum turnover, or a maximum revenue at minimum outlay. If someorderof importanceof these principlesor some weightedaverage is meant, this shouldbe stated. However,in the situation of the participants in a social conomy nothing of that sort is intended, but all maxima are e at desired once by various participants. One would be mistaken to believe that it can be obviated, like the in the Crusoe casementioned in footnote 2 on 10, y a mere b difficulty to the devicesof the theory of probability. Every participant can recourse his the determine variableswhich describe own actions but not thoseof the f \"alien\" variablescannot, rom his point of view, those others. Nevertheless be describedy statistical assumptions. This is becausethe others are b guided,just as he himself, by rational principles whatever that may mean w not can and no modus procedendi be correct hich does attempt to understand thoseprinciplesand the interactionsof the conflicting interestsof all

p.

participants. s Sometimesome of these interests run more 01 less parallel then we are nearerto a simple maximum problem.But they can just as well be opposed.The generaltheory must cover all these possibilities,all intermediary stages,and all their combinations. The 2.2.4. difference between Crusoe's erspectiveand that of a parp
ticipant in a socialeconomy can also be illustrated in this way: Apart
from)))

12

FORMULATIONOF THE ECONOMIC PROBLEM

C those variables which his will controls, rusoeis given a number of data which are \"dead\"; they are the unalterable physicalbackground of the situation. (Even when they are apparently variable, cf. footnote 2 on p. 10,they are really governed by fixed statistical laws.) Not a single w datum with which he has to deal reflectsanother person'sill or intention k nature as his own. A of an economicind based on motives of the same economy,on the other hand, faces data participant in a social exchange of this last type as well:they arethe product of other participants' ctions a and volitions (likeprices). Hisactionswill be influenced by his expectation of these,and they in turn reflect the other participants' xpectation his of e
Thus the study of the Crusoeeconomy and the use of the methods t applicableto it, is of much more limited value to economicheory than has beenassumedheretofore even by the mostradicalcritics. Thegrounds for this limitation lie not in the field of those social relationshipswhich we have mentionedbefore although we do not questiontheir significance d but rather they arise from the conceptual ifferences between the original maximum problemand the more complex a problemsketched bove. (Crusoe's) will be convinced by the above that we face We hope that the reader here and now a really conceptual and not merely technical difficulty. And it is this problemwhich the theory of \" games of strategy\" is mainly devisedto meet.

actions.

The 2.3.1. formal set-upwhich we used in the preceding paragraphs to indicatethe events in a social xchange conomymade use of a number of e e \" the of variables\"which described actions the participantsin this economy.
h e p togethercompletelydescribeis actions,i.e. xpressreciselythe manifestations of his will. We call thesesetsthe partial setsof variables. The partial setsof all participants constitute togetherthe set of all variables,to the becalled total set. Sothe total number of variablesis determinedfirst o by by the numberof participants,i.e.f partial sets,and second the number
Thus every participant is allotted a set of variables,\"his\" variables,which

The 2.3. Number

of Variables

and the

Number of Participants

of variablesin every partial set. From a purely mathematical point of view there would be nothing in objectionable treatingall the variables of any onepartial set as a single \"the\" variable of the participant correspondingto this partial variable, which we are going to use frequently in set. Indeed,his is a procedure t i our mathematical discussions;t makes absolutely no difference conceptually,and it simplifies notations considerably. Forthe moment, however, we proposeto distinguishfrom eachotherthe m variables within each partial set. The economic odels to which one is thus led suggestthat procedure; it is desirable to describeor f naturally the quantity of every particulargoodhe wishesto acquire every participant v by a separateariable, etc.)))

technically, but only technically. Thus in a Crusoeeconomy where there exists oneparticipant and only one partial set which then coinonly cideswith the total set this may make the necessarydeterminationof a maximum technicallymore difficult, but it will not alterthe \"puremaximum \" character f the problem.If, on the other hand, the number of o participants

THE PROBLEM OF RATIONAL BEHAVIOR 13 2.3.2. we must emphasize that any increase the number of Now of variables inside a participant'spartial set may complicate problem our

of the partial setsof variables is increased, something of a very different nature happens. To use a terminology which will turn in out to be significant, that of games,this amounts to an increase the a number of players in the game. However,to take the simplest cases, is very fundamentally different from a two-persongame, three-personame g a four-person game from a three-person game,etc. The combinatorial of the problem which is, as we saw, no maximum problem complications in at all increase tremendouslywith every increase the number of players, will as our subsequentdiscussions amply show. in We have gone into this matter in such detail particularly because a mostmodelsof economicspeculiarmixture of thesetwo phenomenaoccurs. of participants in a socialeconomy, Whenever the number of players, the too; increases, complexityof the economic system usually increases o e.g.the number of commoditiesand services exchanged,processesf productionused, etc. Thus the number of variablesin every participant's of partial set is likely to increase.But the number of participants, w h t partial sets, as increasedoo. Thus both of the sources hich we discussed i contribute paripassu to the total increasen the number of variables. It is in t essentialo visualize eachsource its proper role.

i.e.,

i.e.

i.e.

2.4.The Caseof Many Participants :Free Competition In 2.4.1. elaborating the contrast between a Crusoeeconomy and a we emphasized those features socialexchangeeconomy in 2.2.2.-2.2.4., more prominent when the number of participants of the latterwhich become s while greaterthan 1 is of moderateize. The fact that every particiis influenced by the anticipated reactionsof the others to his own pant and measures, that this is true for eachof the participants,is most strikingly in a the cruxof the matter (as far as the sellers re concerned) the classical of duopoly, oligopoly,etc. When the number of participants problems that becomes really great, some hope emerges the influence of every parand ticular participant will become egligible, that the above difficulties n recedeand a more conventional theory become possible.These may t t conditionsof \"free competition.\"Indeed,his of course,he classical are, was the starting point of much of what is best in economicheory. Comt o of pared with this case greatnumbers free competition the cases f small numbers on the side of the sellers monopoly, duopoly, oligopoly were to a even considered be exceptions nd abnormities. (Even in these cases number of participants is still very largein view of the competition))) the

14

FORMULATIONOF THE ECONOMIC PROBLEM

Thereexists the literaturea considerablemount of theoretical isin a d cussionpurporting to show that the zonesof indeterminateness ratesof (of exchange) which undoubtedly existwhen the number of participants is small narrow and disappear as the number increases.This then would of providea continuous transition into the ideal case free competition for a very great number of participants where all solutionswould be sharply and uniquely determined. hile it is to be hopedthat this indeedturns out W to be the casein sufficient generality, one cannot concede that anything like this contention has been established conclusively thus far. There is no getting away from it: The problem must be formulated, solved and understoodfor smallnumbersof participants before anything canbe proved about the changes its character any limiting caseof largenumbers, of in suchas free competition. A 2.4.3. really fundamental reopeningof this subject is the more b it desirableecause is neithercertainnor probable that a mereincrease in the number of participants will always lead in fine to the conditions
of)))

satisfied.

two oligopolies, etc.) In 2.4.2. all fairness to the traditional point of view this much ought I to be said:t is a well known phenomenon in many branches of the exact to and physical sciences very great numbers are often easier handle that than thoseof medium size. An almost exacttheory of a gas, containing 26 t is about 10 freely moving particles, incomparablyeasierhan that of the and solar system, made up of 9 major bodies; still more than that of a mulo size. This is, of course, tiple star of threeor four objectsf about the same due to the excellent ossibilityof applying the laws of statistics and probp abilitiesin the first case. This analogy, however, is far from perfectfor our problem. Thetheory f bodies is well known, and in its general of mechanicsor 2, 3, 4, and theoretical distinguishedfrom its special computational) form is the (as foundation of the statistical theory for great numbers. For the social f e exchangeconomy i.e.or the equivalent \" gamesof strategy\" the theory of 2, 3, 4, participants was heretofore lacking. It is this need that our previous discussions were designedto establishand that our subsequent will endeavor to satisfy. In other words, only after the investigations theory for moderatenumbersof participants has beensatisfactorilydevelw oped will it be possibleto decide hether extremelygreat numbers of parthe situation. Let us say it again:We share the hope ticipants simplify of chiefly because the above-mentioned analogy in other fields! that such free simplifications will indeed occur. The current assertions concerning competition appear to be very valuable surmisesand inspiringanticipations of results. But they arenot results and it is scientifically unsoundto treat them as such as long as the conditionswhich we mentionedabove arenot

i among the buyers. The casesnvolving really small numbersare those of b bilateral monopoly, of exchange etween a monopoly and an oligopoly,or

that if certain whatsoever greatgroupsof participantswill for any reason act together,then the great number of participants may not become effective; the decisive exchanges ay take placedirectly between large m \" \" coalitions, l few in number, and not betweenindividuals, many in number, actingindependently. Our subsequentdiscussionof \" gamesof strategy\" \" will show that the roleand size of \" coalitions is decisivethroughout the entire subject. Consequentlythe above difficulty though not new still remains the crucial problem.Any satisfactory theory of the \" limiting transition \" from small numbersof participants to largenumbers will have to explainunder what circumstances big coalitionswill or will not be such formed when the largenumbers of participants will becomeffective e and lead to a more or less free competition.Which of thesealternativesis will likelyto arise dependon the physicaldata of the situation. Answering this questionis, we think, the real challenge any theory of free competition. to

THE NOTION OF UTILITY 15 free competition. The classical definitions of free competitionall involve further postulates besideshe greatness that number. E.g., is clear t of it

i.e.

2.5. he \"Lausanne\" T

Theory

T 2.6. his section should not be concludedwithout a reference the to a equilibrium theory of the LausanneSchool nd also of various othersystems which take into consideration \" individual planning \" and interlocking individual plans. All thesesystems pay attention to the interdependence of the participants in a socialeconomy. This, however, is invariably done f under far-reaching restrictions. Sometimesree competition is assumed, after the introduction of which the participants face fixed conditionsand act like a number of RobinsonCrusoes solely bent on maximizing their w individual satisfactions, hich under theseconditionsareagain independent. other restricting devices are used, all of which amount to In other cases \" the excluding free play of \" coalitions formed by any or all types of parhidden, assumpticipants. There are frequently definite, but sometimes the tions concerning ways in which their partly paralleland partly opposite or interestswill influence the participants,and causethem to cooperate not, a case be. We hopewe have shown that such a proceduremounts as the may to a petitio principii at least on the plane on which we shouldlike to put the discussion. It avoids the real difficulty and dealswith a verbal problem, we which is not the empiricallygiven one. Of course do not wish to questhese investigations but they do not answer our tion the significance of

queries.
way we wish to describe the fundamental conceptof individual preferencesby the use of a rather 1 Such as trade industrial cartels,and conceivably unions, consumers' cooperatives, someorganizations more in the political sphere.)))

3.The Notion of Utility and Preferences Utilities 3.1. in We 3.1.1. have stated already in 2.1.1.what

16

FORMULATIONOF THE ECONOMICPROBLEM

far-reaching notion of utility. Many economistswill feel that we are assumingfar too much (cf the enumeration of the propertieswe postulated from the more cautious in and that our standpoint is a retrogression modern techniqueof \" indifference curves.\" Before attempting any specificdiscussionlet us state as a general excuse our procedureat worst is only the application of a classical that to preliminary device of scientific analysis: To divide the difficulties, concentrate n one (the subjectproper of the investigation in hand), and o to reduce othersas far as reasonablypossible, y simplifying and schemaall b tizing assumptions. We shouldalso add that this high handed treatment of preferences and utilities is employedin the main body of our discussion, which an but we shall incidentally investigate to a certainextentthe changes in avoidance of the assumptionsin questionwould cause our theory (cf. 66.,

2.1.1.),

i.e.

67.).

radical as is often assumedin the literature. We shall attempt to prove this particular point in the paragraphswhich follow. It is hoped that the reader forgive us for discussingonly incidentally in a condensedform will a subjectof so greata conceptualimportanceas that of utility. It seems however that even a few remarks may be helpful, becausethe question t of the measurability of utilities is similar in charactero corresponding questionsin the physicalsciences. 3.1.2. Historically,utility was fiist conceivedas quantitatively measuri a able, .e.s a number. Valid objectionscan be and have beenmade against that every measurement this view in its original, naive form. It is clear or rather every claim of measurability must ultimately be based on some immediate sensation, which possibly cannot and certainly need not be analyzed any further. 1 In the caseof utility the immediatesensation of of as preference of one objector aggregate objects against another providesthis basis. But this permits us only to say when for one person oneutility is greaterthan another. It is not in itself a basis for numerical comparison of utilities for one person nor of any comparisonbetween different persons. Sincethere is no intuitively significant way to add two iUtilities for the same person, the assumption that utilities are of noneven seems lausible. The modern method of indifferp Jnumerical character encecurve analysisis a mathematical procedureo describe situation. t this this is strongly reminiscentof the conditionsexistantat the of the theory of heat:that too wasbased on the intuitively clear beginning o concept f one body feeling warmer than another, yet therewas no immediate way to expressignificantly by how much, or how many times, or in s
light, All 3.2.1.

We feel, however, that one part of our assumptionsat least that of treating utilities as numerically measurable quantities is not quite as

3.2. Principles of Measurement :Preliminaries

what sense. 1Such as the sensations of branches of physics.)))

heat,

muscular

effort,

i etc.,n

the corresponding

THE NOTION OF UTILITY

what the ultimate

17

with heat alsoshowshow little onecanforecasta priori This comparison shape of such a theory will be. Theabove crudeindications do not disclose all what, as we now know, subsequentlyhappened. at It turned out that heat permitsquantitative descriptionnot by one numbei but by two: the quantity of heat and temperature. he former is rather T directly numerical becauseit turned out to be additive and also in an with mechanical energy which was numerical unexpected connected way anyhow. The latter is also numerical, but in a much more subtle way; it is not additive in any immediatesense, ut a rigid numerical scale it for b from the study of the concordantbehavior of ideal gases, nd the a emerged role of absolutetemperaturein connectionith the entropy theorem. w The 3.2.2. historicaldevelopmentof the theory of heat indicates that one must be extremely careful in making negative assertions about any w concept ith the claim to finality. Even if utilitieslook very unnumerical the history of the experiencethe theory of heat may repeattself, in today, i and nobody can foretell with what ramifications and variations. 1 And it should certainly not discourage theoreticalexplanations of the formal of possibilities a numerical utility.

We can go even one step beyond the above double negations which were only cautions against premature assertionsof the impossibility can beshown that under the conditionson which of a numerical utility.

3.3.1.

3.3. Probability It

and Numerical

Utilities

the indifference curve analysisis based very littleextraeffort is needed to reacha numerical utility. It has beenpointedout repeatedlythat a numerical utility is dependent upon the possibilityof comparing differences in utilities. This may seem and indeedis a more far-reaching assumptionthan that of a mere ability that the alternatives to which ecoto statepreferences.But it will seem must be appliedaresuchas to obliterate distinction. this nomic preferences t ILetus for the moment accepthe pictureof an individual whose system ol preferencesis all-embracmg^md-eemplete, i^e^who^forany two a clear or rather for any two imaginedevents, possesses intuition of objects

3.3.2.

More preciselywe expectim, for any two alternative events which are h t put before him as possibilities,o be able to tell which of the two he prefers. It is a very natural extensionf this picture to permit such an individual o to comparenot only events, but even combinationsof events with stated
2 probabilities. L By a combination of two events we mean this: et the two events be denotedby B and C and use, for the sakeof simplicity,the probability

preference.

1 A goodexample of the wide variety of formal possibilities is given by the entirely All thesenotions a different development of the theory of light, colors, nd wave lengths. too became numerical, but in an entirely different way. 2 Indeedthis is necessary he is engaged in economicactivities which are explicitly if Of.the example of agriculture in footnote 2 on p. 10.))) dependent on probability.

18

FORMULATIONOF THE ECONOMIC PROBLEM

and also to C, then he will prefer it to the above combination as well; similarly, if he prefersB as well as C to A, then he will prefei the combination too. But if he shouldprefer A to, say B,but at the same time C to A, then o any assertionabout his preference f A against the combination contains I fundamentally new information. Specifically: f he now prefers A to the 50-50ombination of Band C, this providesa plausiblebase for the numerc ical estimate his preferencef A over B is in excessf his preference that o of o C over A. 1-2 If this standpoint is accepted, then there is a criterionwith which to the preferenceof C over A with the preferencef A over B. It is o compare well known that thereby utilities or rather differencesof utilities become numerically measurable. That the possibilityof comparisonbetween A, B, and C only to this extent is already sufficient for a numerical measurementof \" distances\" was first observed in economicsy Pareto. Exactly the same argument b has beenmade, however, by Euclid for the positionof points on a line in fact it is the very basis of his classical derivation of numerical distances. The introduction of numerical measures can be achieved even more Indeed: Consider directly if use is made of all possible probabilities. three events, C, A, B, for which the orderof the individual's preferences is the one stated. Let a be a realnumber between and such that A is exactlyequally desirablewith the combinedevent consistingof a chance of probability 1 a for B and the remaining chance probability a.for C. of Then we suggestthe use of a as a numerical estimateor the ratio of the f 8 o preference f A over B to that of C over B. An exactand exhaustive

50%-50%.Then the \" combination\"is the prospect of seeingB occur with a probabilityof 50%and (if B does not occur) with the (remaining) C probability of 50%. We stressthat the two alternatives are mutually exclusive,so that no possibility of complementarity and the like exists. o Also, that an absolute certainty of the occurrencef eitherB or C exists. Torestateour position. We expecthe individual under consideration t to possessa clearintuition whether he prefers the event A to the 50-50 combination of B or C, or conversely. It is clear that if he prefersA to B

1,

1To give a simple example: ssume that an individual prefers the consumption of a A and to glass of tea to that of a cup of coffee, the cup of coffee a glass of milk. If we now the want to know whether the last preference exceeds former, difference in utilities it suffices to placehim in a situation where he must decide this: Doeshe prefer a cup of coffee a glass the content of which will be determined by a 50%-50 chance to device as % tea or milk. 1Observehat we have only postulated an individual intuition which permits decision t as to which of two \"events\" is preferable. But we have not directly postulated any intuitive estimate of the relative sizesof two preferences in the subsequent terminology, of two differences of utilities. This is important, sincethe former information ought to be obtainablein a reproducible way by mere \"questioning.\" 1This offers a good opportunity for another illustrative example. Theabovetechnique permits a direct determination of the ratio q of the utility of possessing unit of a certain good to the utility of possessing units of the samegood. The individual must))) 2

i.e.,

i.e.

THE NOTION OF UTILITY 19 elaborationof theseideas requiresthe useof the axiomatic method. A simple treatment on this basis is indeed possible.We shall discuss it in
avoid misunderstandingslet us state that the \" events\" which were used above as the substratum of preferencesre conceivedas a future events so as to make all logically possible alternatives equally admissible. However, it would be an unnecessarycomplication,as far as our presentobjectivesareconcerned, get entangledwith the problems to of the preferences etweenevents in different periods of the future. 1 It b seems,however, that such difficulties can be obviated by locating all i \"events\"n which we are interestedat one and the same, standardized, moment, preferably in the immediatefuture. The above considerationsare so vitally dependent upon the numerical the conceptof probability that a few words concerning latter may be

3.5-3.7.

To 3.3.3.

Probability has often been visualized as a subjective conceptmore the nature of an estimation. Sincewe proposeto use it in constructing an individual, numerical estimation of utility, the above view of is, probability would not serve our purpose. Thesimplestprocedure thereof fore, to insist upon the alternative, perfectly well founded interpretation as frequency in long runs. This gives directly the necessary probability
or

appropriate. lessin

numerical foothold.2

f of This procedureor a numerical measurement the utilities of the in individual depends, of course, pon the hypothesisof completeness the u 8 of individual preferences.It is conceivable and may even in a system where the individual is neither way be more realistic to allow for cases able to statewhich of two alternatives he prefersnor that they areequally desirable.In this case the treatment by indifference curves becomes

3.3.4.

too. impracticable 4 How realthis possibilityis, both for individuals and for organizations, t seemso be an extremelyinteresting question, but it is a question of fact. It certainly deservesfurther
At any

it study. We shall reconsider briefly in 3.7.2. the treatmentby indifference if curves implieseithertoo much or too little: the preferencesof the indi-

ratewe hope we have shown that

to with certainty or of playing the chance get two If he prefers the units with the probability <*, or nothing with the probability 1 former, then a < 5; if he prefers the latter, then a > g; if he cannot state a preference either way, then a = q. 1It is well known that this presents very interesting, but as yet extremely obscure, connectionswith the theory of saving and interest, etc. 2 If one objectsto the frequency interpretation of probability then the two concepts c (probability and preference) an be axiomatized together. This too leadsto a satiso o factory numerical concept f utility which will be discussedn another occasion. 8 We have not obtained or qualitatively, any basis for a comparison, quantitatively of the utilities of different individuals. 4 Theseproblems belong systematically sets. in the mathematical theory of ordered The abovequestion in particular amounts to asking whether events, with respectto form preference, a completely or a partially orderedset. Cf. 65,3.))) of be given the choice obtaining 1 unit

a.

20

FORMULATIONOF THE ECONOMIC PROBLEM

1 vidual are not all comparable,then the indifference curves do not exist. If the individual'spreferences all comparable, we can even obtain a are then (uniquely defined) numeiical tility which renders the indifference curves u

superfluous. All this becomes, course,pointless for the entrepreneur who can of calculate terms of (monetary)costs profits. in and 3.3.5. objectioncouldbe raised that it is not necessaryto go into The all these intricate details concerning measurability of utility, since the evidently the common individual, whose behavior one wants to describe, doesnot measurehis utilities exactly but rather conducts his economic activitiesin a sphereof considerable haziness. Thesameis true, of course, for much of his conduct regardinglight, heat, musculareffort, etc. But in orderto build a sciencef physics these phenomenahad to be measured. o And subsequently the individual has comeo use the resultsof such measuret ments directly or indirectly even in his everyday life. The samemay obtain in economicst a future date. Once a fuller understanding of a economicehavior has beenachievedwith the aid of a theory which makes b use of this instrument, the life of the individual might bematerially affected. It is, therefore, not an unnecessarydigressionto study theseproblems. 3.4.Principles of Measurement :DetailedDiscussion 3.4.1. readermay feel, on the basis of the foregoing, that we The obtained a numerical scale utility only by begging principle,i.e. y of the b of really postulating the existence such a scale. We have argued in 3.3.2. that if an individual prefers A to the 50-50 combination of B and C (while preferring C to A and A to JB), this providesa plausiblebasis for the numerical estimate that this preference f A over B exceeds that of C over A. o Are we not postulatinghere or taking it for granted that one preference t may exceed another, i.e.hat such statementsconvey a meaning? Such a view would be a complete of misunderstanding our procedure. 3.4.2. arenot postulating or assuming anything of the kind. We We have assumedonly one thing and for this thereis goodempiricalevidence namely that imagined events can be combinedwith probabilities. And therefore the same must be assumed for the utilities attachedto them, whatever they may be. Or to put it in more mathematical language: There frequently appear in science uantities which are a priorinot q mathematical, but attached to certain aspectsof the physical world. Occasionallythese quantities can be grouped togetherin domains within which certain natural, physically defined operations are possible.Thus the physicallydefined quantity of \"mass\"ermitsthe operationof addition. p 2 Thephysico-geometrically defined quantity of \"distance\" permitsthe same
1Points on the same indifference curve must be identified and are therefore no instances of incomparability. f Let a us, for the sake of the argument, view geometry as a physical discipline, tenable viewpoint. sufficiently By \"geometry\" we mean equally for the sake of the
argument Euclidean
geometry.)))

THE NOTION OF UTILITY

21

1 of position\"does not permit this operation,but it permits the operation \" 2 of forming the center f gravity\" of two positions. Again other physicoo \" geometrical concepts, usually styled vectorial\" likevelocity and acceleration permit the operationof \" addition.\" In all these cases where such a \" natural\" operation is given a name which is reminiscentof a mathematical operation like the instances of \" addition\"above one must carefully avoid misunderstandings. This nomenclature is not intended as a claim that the two operationswith the i same name areidentical, this is manifestly not the case;t only expresses the opinion that they possesssimilar traits, and the hope that somecorbetweenthem will ultimately be established. This of course respondence when feasible at all is done by finding a mathematical model for the physicaldomain in question, within which those quantities are defined by the numbers, so that in the model the mathematical operation describes \" natural\" operation. synonymous \" b n To return to our examples: energy\"and \"mass\" ecameumbers in a the pertinent mathematical models,\"natural\" ddition becomingordinary addition. \"Position\" well as the vectorial quantities became as triplets3 of called coordinatesor componentsrespectively. The \"natural\" numbers, o o x%, 2 concept f \"center f gravity\" of two positions {#1, x 3) and \\x' ly x', z'a}/ a (cf. footnote 2 above),becomes with the \"masses\" 1 a,

defined quantity operation.On the other hand, the physico-geometrically \"

3.4.3.

5 {ax,+ (1 a)x(,ax, + (1 a)*J,ax, + (1 <*X). x of The\"natural\" operationof \"addition\" vectors {zi, 2, x*\\ and [x(,z, x' z becomesxi+ x[,x 2 + x 2, x* + ZgJ. 6 {

\\

and mathematical operations was said above about \"natural\" to natural and mathematical relations. The various conapplies equally which occurin physics greater energy, force, heat, ceptsof \"greater\" etc. aregoodexamples. velocity, These\"natural\" relationsare the best base upon which to construct 7 the mathematical modelsand to correlate physicaldomain with them. '8
What
1 We are i thinking of a \"homogeneous\" Euclidean space,n which no origin or frame of i references preferred aboveany other. 2 With respectto two given masses a, occupying those positions. It may be con*. venient to normalize so that the total mass is the unit, i.e. ** 1 3 We are thinking of three-dimensional Euclidean space. 4 We are now describing them by their three numerical coordinates. 8 This is Cf.(16:A:c)16.2.1. in 2,z xj, usually denoted by a (xi,z 8 1 + (1 a)js|, *',) 8 This is usually denoted by (xi, x*, x s -f (z'i, zj, xj|. Cf.the beginning of 16.2.1. 7 Not the relaonly one. Temperature is a good counter-example. The \"natural\"

\\

tion
8

of

model,
picture tant

We do not want to give the misleading impression of attempting here a complete of the formation of mathematical models, of physical theories. It should be w remembered that this is a very varied process ith many unexpectedphases. An importhe one is, e.g., disentanglement

Cf. 3.2.1.

\"greater,\" would not have sufficed to establish the present day mathematical the absolute temperature scale. Thedevices actually used were different.

i.e.

i.e. of concepts: .e. i splitting

up something

which

at)))

22

FORMULATIONOF THE ECONOMIC PROBLEM

1 systemsis groups. conof such situations arenumerous. Thus the geometrical Examples of distance is a number, up to multiplication by (positive)constant cept 2 factors. The situation concerning physical quantity of mass is the the The physicalconcept f energyis a number up to any linear transsame. o a formation, i.e. ddition of any constant and multiplication by any (posi8 o tive) constant. Theconcept f positionis defined up to an inhomogeneous

3.4.4. a further remarkmust be made. Assume that a satisfactory Here mathematical model for a physical domain in the above sensehas been found, and that the physical quantities under considerationhave been correlatedith numbers. In this caseit is not true necessarilythat the w description (of the mathematical model) provides for a unique way of the correlating physicalquantities to numbers; it may specifyan entire family of such correlationsthe mathematical name is mappings any from oneof which can beused for the purposesof the theory. Passage one of thesecorrelations another amounts to a transformation of the numerical to data describingthe physicalquantities. We then say that in this theory the physicalquantities in question aredescribedy numbers up to that b systemof transformations. Themathematical name of suchtransformation

i.e.,

orthogonal linear transformation. 4- The vectorial conceptsare defined t up to homogeneousiansformationsof the samekind.5 6 3.4.6. is even conceivablethat a physicalquantity is a number up to It any monotone transformation. This is the casefor quantities for which \" \" t only a \"natural\" relation greater exists and nothing else. E.g. his for was the case temperature long as only the concept f \" warmer \" was as o 7 i of known; it appliesto the Mohs' scale hardnessof minerals;t appliesto))
B

'

to mathematical notions. superficial inspection seems be one physical entity into several Thus the \"disentanglement\" of forceand energy, of quantity of heat and temperature, in were decisive their respective fields. It is quite unforeseeablehow many such differentiations still lie ahead in economic theory. 1 We shall encounter where referenceso the t groups in another context in literature arealsofound. * there is nothing in Euclidean geometry to fix a unit of distance. 3 there is nothing in mechanics to fix a zeroor a unit of energy. Cf with footnote 2 above. Distancehas a natural zero, the distance of any point from itself. 4 X||are to bereplaced {xi*, xa *, x9 *\\ where)) by

28.1.1,

I.e. I.e. I.e.*i, , | x

*i*

^i* -

0*1X1 031X1

-f OU.TI 4- 013X3 + 61, -f 022X1 -f 023X3 -f 62, -f as2X S + 033X3 +

&3,))

the a</, bi being constants, and the matrix (a,/) what is known as orthogonal. there is nothing in geometry to fix either origin or the frame of reference when a when vectors are positions are concerned;nd nothing to fix the frame of reference

I.e.

in footnote 4 above. Sometimes a wider conceptof matrices is the bi We need not discussthesematters here. permissible, all those with determinants ^ But no quantitatively reproducible method of thermometry
f

concerned.

'

I.e.

0.

.)))

THE NOTION OF UTILITY

the notion of utility
in when

23

ence. In thesecases nemay be tempted to takethe view that the quantity o

this is based on the conventional idea of prefer-

l geneous rthogonal linear transformations in space,inear transformations o 1 of one numerical variable, multiplication of that variable by a constant. In fine, the caseeven occurswhere the numerical descriptionis absolutely 2 rigorous, where no transformations at all needbe tolerated. 3.4.6. a physical quantity, the system of transformations up to Given w which it is described numbers may vary in time, i.e. ith the stageof by was originally a number developmentof the subject. Thus temperature only up to any monotone transformation. 8 With the development of i thermometry particularly of the concordantdeal gas thermometry the to transformations were restricted the linear ones, only the absolute zero and the absolute unit were missing. Subsequent developmentsof thermodynamicseven fixed the absolute zero so that the transformation consistsonly of the multiplication by constants. system in thermodynamics t couldbe multipliedbut thereseemso be no needto go into this Examples further. subject to For utility the situation seems be of a similar nature. One may take the attitude that the only \"natural\" datum in this domain is the I o relation \"greater,\" the concept f preference. n this caseutilities are numerical up to a monotone transformation. This is, indeed,the generally

how questionis not numerical at all, considering arbitrary the description to by numbers is. It seems be preferable,however, to refrain from such qualitative statements and to stateinstead objectivelyup to what system of transformations the numerical description is determined.The case when the system consistsof all monotone transformations is, of course, a rather extreme ne;various graduations at the other end of the scaleare o the transformation systems mentioned above:inhomogeneousor homo-

i.e.

i.e.

i.e.

of indifference curves. To narrow the system of transformations it would be necessaryto discover further \"natural\" operations or relations in the domain of utility. 4 Thus it was pointed out by Pareto that an equality relationfor utility differences would suffice; in our terminology it would reducethe transfor6 mation system to the linear transformations. However,sinceit does not 1Onecould alsoimagine intermediate casesof greater transformation systems than Various forms of the theory of thesebut not containing all monotone transformations.

in l s acceptedtandpoint in economiciterature,best expressed the technique

relativity give rather technical examplesof this. 2 In the usual language this would hold for physical quantities where an absolute zero the casefor the absolute value as well as an absolute unit can be defined. This is, (not the vector!)of velocity in such physical theories as those in which light velocity r electrodynamics, specialelativity. plays a normative role:Maxwellian \" 8 As a \"natural\" relation \"greater\" was o long as only the concept f warmer\" known. We discussed in extenao previously. this 4 V. Pareto, Manuel d'Economie olitique, Paris, 1907, . 264. p P is exactly what Euclid did for position on a line. The utility conceptof \" \" \" \" to preference corresponds the relation of lying to the right of there, and the (desired) relation of the equality of utility differences to the geometrical congruenceof intervals.)))

e.g.,

i.e.

'This

24

FORMULATIONOF THE ECONOMIC PROBLEM

o o seemthat this relation is really a \"natural\" ne i.e. ne which can be d interpretedby reproducibleobservations the suggestionoes not achieve the purpose. t Thefailure of oneparticulardeviceneednot excludehe possibility of achieving the same end by anotherdevice. Our contention is that the domain of utility containsa \"natural\" operationwhich narrows the system of transformations to preciselythe same extent as the other device would have done. Thisis the combination of two utilities with two given alternain The tive probabilities a, 1 a, (0 < a < 1) as described of i processs so similar to the formation of centers gravity mentionedin that it may be advantageous to use the same terminology. Thus u r we have for utilities u, v the \"natural\"elation u > v (read: is preferable to v), and the \"natural\" operation an + (1 a)v, (0 < a < 1),(read: of or:combinacenter gravity of u, v with the respective weightsa, 1 tion of u, v with the alternative probabilities ). If the existence and reproducibleobservability of these concepts conceded,hen our t is W betweenutilities and numbers way is clear: e must find a correspondence the relation u > v and the operation au + (1 a)v for which carries utilities into the synonymousconceptsor numbers. f Denote correspondence the by

3.6.1.

3.6. onceptual Structure C

of the

Axiomatic Treatment

of Numerical

Utilities

3.3.2.

3.4.3.

,!

a;

u u

->p = v(w),

to

attaches being the utility and v(u) the number which the correspondence it. Our requirementsarethen:
u

(3:l:a) (3:l:b)
(3:2:a) (3:2:b)
(3:3)

v(au

>v + (1

> implies - a)v) = av(u)v(u)(1- a)v(y). +

v(v),

If two such correspondences

u-+p = v(u),
u

'))

p' = v'(u),

should exist,hen they setup a correspondence t betweennumbers

p+ P',

for which we may also write

(3:4) P Since(3:2:a), fulfill (3:2:b) (3:1 (3:1 the correspondence i.e. :a), :b), (3:3), the function 0(p)in (3:4)must leave therelation > 2 and the operation p
cr

that in in each case the left-hand ^Observe utilities, and the right-hand sidethe conventional 1Now theseare applied to numbers p,
o-l)))

side has the \"natural\" onesfor numbers.

conceptsfor

THE NOTION OF UTILITY

ap

25

+ (1

)<r unaffected

(cf footnote 1 on p.24).

I.e.

(3:5:a) (3:5:b)
Hence </>(p)
(3:6)
must

p
<t>(ap

> > <(<r), implies + (1 a)r) = a*(p) + (1 a)0(<r).

<r

<f>(p)

'

be a linear function, i.e. p' = <(p) es p +

o>

i,

where w , i arefixed numbers (constants) ith w > 0. w So we see:f such a numerical valuation of utilities1 exists all, then I at it is determined up to a linear transformation. 2 8 I.e.then utility is a number up to a linear transformation. In orderthat a numerical valuation in the above senseshould existit is necessaryto postulate certainproperties the relation u > v and the of o operation au + (1 ct)v for utilities. The selectionf these postulates or axioms and their subsequent analysis leadsto problems of a certain mathematical interest. In what follows we give a generaloutline of the a situation for the orientation of the reader; complete discussion found in is the Appendix. A choice axiomsis not a purely objective of task. It is usually to expected achieve some definite aim some specifictheorem or theorems are to be derivable from the axioms and to this extent the problemis exactand objective.But beyond this there are always other important o desiderataf a less exactnature:The axiomsshould not be too numerous, their system is to be as simpleand transparent as possible,and eachaxiom should have an immediateintuitive meaning by which its appropriateness 4 may be judged directly. In a situation like ours this last requirementis we particularly vital, in spite of its vagueness: want to make an intuitive amenable to mathematical treatment and to see as clearly as concept

3.5.2.

possiblewhat hypothesesthis requires. The objectivepart of our problem is clear:the postulates must imply with o the properties (3:l:a), the existencef a correspondence3:2:a) ( The as described in 3.5.1. further heuristic,and even esthetic (3:l:b) indicated above, do not determinea unique way of finding desiderata, this axiomatic treatment. In what follows we shall formulate a set of t axiomswhich seemso be essentiallysatisfactory.
1

3 Remember the physical examplesof the same situation given in 3.4.4. (Ourpresent discussion is somewhat more detailed.) We do not undertake to fix an absolute zero and an absolute unit of utility. 4 The first and the last principle may represent at leastto a certain extent opposite influences: If we reducethe number of axioms by merging them as far as technically we possible, may losethe possibility of distinguishing the various intuitive backgrounds. in the Thus we could have expressed group (3:B) by a smaller number of axioms, the but this would have obscured subsequent analysis of is To strike a proper balance a matter of practical and to someextent even esthetic

I.e. correspondence which a (3:2:a) I.e.ne of the form (3:6). o

fulfills

(3:1 (3:1 :a), :b).

3.6.1.

3.6.2.

judgment.)))

26

FORMULATIONOF THE ECONOMICPROBLEM

3.6. Axioms The

Our 3.6.1. axioms arethese:
We

and

Their Interpretation
v, w,

a, (0 < a < 1),an operation v, au + (1 a)v = w. Theseconceptsatisfy the following axioms: s

given, u

considera system U of entities1 u,

>

and for any number

In

a relation is

W This means: rite u

A) (3:

>v

is a completeordering of f/. 2

< v when v > u. Then:

two u
y

(3:A:a)

relationsholds:
u

Forany

one and only one of the threefollowing

(3:A:b)

(3:B) (3:B:a) (3:B:b) (3:B:c) (3:B:d) (3:C) (3:C:a) (3:C:b)

with (3:2:a)

> v, u < v. u > v, v > w imply u > w.z 4 Ordering and combining. u < v impliesthat u < au + (1 a)v. u > v impliesthat u > au + (1 a)v. u < w < v impliesthe existence an a with of au + (1 a)v < w. u > w > v impliesthe existence an a with of au + (1 a)v > w.
u Algebra of combining.

= vt

where 7 = aft.

a(ftu

+ (1 a)v = (1 a)v + au. + (1 fiv) + (1 a)v = yu + (1

au

- -

- -

y)v

a linear transformation.

Onecan show that theseaxioms imply the existence a correspondence of Hence the properties (3:1 (3:1 as describedin 3.5.1. :b) :a), the conclusionsof 3.5.1. good: system U i.e.n our present hold i The interpretation, the system of (abstract)utilities is one of numbers up to

The construction of (3:2:a) (3:1 (3:1 by means of the (with :a), :b) axioms (3:A)-(3:C))s a purely mathematical task which is somewhat i lengthy, although it runs along conventional lines and presents no par1 This is, of course, eant to be the system of (abstract) utilities, to be characterized m by our axioms. Concerning the general nature of the axiomatic method, cf. the remarks and references the last part of in * For a more systematic mathematical discussion of this notion, cf. The of o was equivalent concept f the completeness the system of preferences previously consideredat the beginning of and of 8 These in conditions (3:A:a), Aft) correspondto (65:A:a), 65:A:b) ( 4 Remember that the a, 0, y occurring here arealways > 0, <

10.1.1.

65.3.1.

3.3.2. 3.4.6.
(3:

65.3.1.

1.)))

THE NOTION OF UTILITY

27

ticular difficulties. (Cf. Appendix.) It seems quallyunnecessaryto carry out the usual logisticdiscussion e 1 of theseaxioms on this occasion. We shall however say a few more wordsabout the intuitive meaning the justification of eachone of our axioms(3:A)-(3:C). The analysisof our postulatesfollows:

i.e. 3.6.2.
(3:A:a*)

This is the statement of the completeness the system of of individual preferences.It is customary to assumethis when utilities or preferences, in the \" indifferencecurve discussing e.g. in analysismethod.\" Thesequestionswere already considered and 3.4.6. This is the \" transitivity \" of preference, plausible and a (3:A:b*) generally acceptedroperty. p (3:B:a*) We statehere:If v is preferableto u, then even a chance 1 a of v alternatively to u is preferable. Thisis legitimate sinceany kind of complementarity (or the opposite) as been h cf. the beginning of excluded, with i (3:B:b*) Thisis the dual of (3:B:a*), \"less referable\"n placeof p

3.3.4.

3.3.2.

\"

(3:B:c*)

w is preferableto u, and an even more is also given, then the combination of u with a preferable chance1 a of v will not affect w'& preferability to it if this chanceis small enough. I.e.: oweverdesirablev may be in H itself, one can make its influence as weakas desired by giving it a sufficiently small chance. This is a plausible\" continuity\" assumption. \" :B Thisis the dual of (3 :B with \"less referable in place. of (3 :d*) :c*), p

preferable.\"
We
v

statehere:If

This is the statement that it is irrelevant in which order the constituentsu, v of a combination arenamed. It is legitimate, particularly sincethe constituents are alternative events, cf. above. (3:B:a*) (3:C:b*) This is the statement that it is irrelevant whether a combination of two constituents is obtained in two successive steps, first the probabilitiesa, 1 a, then the probabilities 0, 1 /J; or in one operation, the probabilities7, 1 y where 2 The same 7 = things can be said for this as for (3:C:a*) above. It may be, however, that this postulate has a deeper below. significance, to which one allusion is made in 3.7.1.

(3:C:a*)

\"preferable.\"

a.

1A similar situation is dealt with more exhaustively in those axioms describe a subject which is more vital for our main objective. The logistic discussion is indicated of there in Some the general remarks of apply to the present casealso. 2This is of course a the correct rithmetic of accounting for two successive admixtures of v with

10.;

10.2.
u.)))

10.3.

28

FORMULATIONOF THE ECONOMICPROBLEM

3.7.GeneralRemarks

combine (with probabilities)like mathematical expectations! And utility has yet the conceptof mathematical expectation been often questioned, and its legitimatenessis certainly dependent upon some hypothesiscon1 the cerning the nature of an \" expectation.\" Have we not then begged Do not our postulates introduce, in some oblique way, the question? hypotheseswhich bring in the mathematical expectation? More specifically: ay there not exist in an individual a (positiveor M act o negative) utility of the mere of \" taking a chance,\"f gambling, wRich the use of the mathematical expectation bliterates? o Howdid our axioms (3:A)-(3:C) around this possibility? get As far as we can see, ur postulates(3:A)-(3:C) not attempt to avoid o do it. Even that one which gets closest excludinga \"utility of gambling\" to seems (3:C:b) its discussionin 3.6.2.), to be plausibleand legitimate, (cf. unlessa much more refined system of psychology used than the one now is available for the purposesof economics. fact that a numerical utility The with a formula amounting to the use of mathematical expectationscan be built upon (3:A)-(3:C), to seems indicate this: e have practically W defined numerical utility as being that thing for which the calculus of 2 mathematical expectationss legitimate. Since(3:A)-(3:C)ecurethat i s the necessary construction can be carried out, concepts like a \" specific of gambling\" cannot be formulated free of contradiction on this utility

3.7.1. this point it may be well to stop and to reconsider situaAt the tion. Have we not shown too much? We can derive from the postulates t and (3:A)-(3:C)he numerical characterof utility in the senseof (3:2:a) and (3:1 (3:1 in 3.5.1.; (3:1 statesthat the numerical values of :a), :b) :b)

Concerning the Axioms

1 Thus Daniel Bernoulli's well known suggestion to \"solve\" the \"St. Petersburg the use of the so-called \"moral expectation\" (instead of the mathematical expectation) means defining the utility numerically as the logarithm of one'smonetary
Paradox\" by

3.7.2. we have stated,the last time in As our axioms arebased on the relation u > v and on the operation au + (1 a)v for utilities. It seemsoteworthy that the lattermay be regardedas more immediately n O given than the former: ne can hardly doubt that anybody who could imagine two alternative situations with the respectiveutilities u, v could not also conceive the prospect of having both with the given respective On the other hand one may questionthe postulate probabilities of axiom (3:A:a) u > v, i.e.he completeness this ordering. for t of Let us considerthis point for a moment. We have concededhat one t doubt whether a person^can w decide hich of two alternatives may always 1 Cf. in der Wertlehre, Zeitschrift ftir Karl Menger: Das Unsicherheitsmoment vol. 5, (1934) p.459ff. and Gerhard Tintner: A contribution to the National6konomie, p non-static Theory of Choice, uarterly Journal of Economics, vol. LVI, (1942) p.274ff. Q p

3 level.

3.6.1.,

,! .

possessions. 9

This may seemto bea paradoxicalassertion. But anybody to axiomatize that elusive concept, ill probably concur with w

who

has seriously

tried

it.)))

THE NOTION OF UTILITY

with

29

6 is still possible. It leads to what may be described a many-dimensional as o vector concept f utility. This is a more complicatedand less satisfactory set-up,but we do not proposeto treatit systematicallyat this time. 3.7.3. brief expositiondoesnot claim to exhaust the subject,but This we hope to have conveyedthe essentialpoints. To avoid misunderstandings, the following further remarks may be useful. that (1) We re-emphasize we are consideringnly utilities experienced o the d oneperson. Theseconsiderations o not imply anything concerning by of the utilities belonging different individuals. to comparisons (2) It cannot bedeniedthat the analysisof the methodswhich makeuse of mathematical expectation footnote 1 on p.28 for the literature)is (cf. far from concludedat present. Our remarks in 3.7.1. in this direction, lie but much more shouldbe said in this respect. Therearemany interesting questions involved, which however lie beyond the scopeof this work. Forour purposesit sufficesto observethat the validity of the simpleand in for plausibleaxioms (3:A)-(3:C) 3.6.1. the relation u > v and the operation au + (1 a)v makesthe utilitiesnumbersup to a linear transformation in the sensediscussedin thesesections.

1 he prefers. But, whatever the merits of this doubt are, this possibility i.e. he completeness the system of (indit of vidual) preferences must beassumedeven for the purposesof the \"indifferencecurve method\" (cf. our remarks on (3:A:a)in 3.6.2.). if this But property of u > v 2 is assumed,then our useof the much less questionable 4 au + (1 ot)v * yieldsthe numerical utilities too! If the general comparabilityassumptionis not made,5 a mathematical theory based on au + (1 <x)v togetherwith what remains of u > v

the utilities u,

nessof preferences.

the postulates (3 (3:C)ogether with the obvious postulate (3:A:b). t 4 At this point the readermay recallthe familiar argument according to which the unnumerical (\"indifference curve\") treatment of utilities is preferable to any numerical one, becauseit is simpler and based on fewer hypotheses. This objection might be legitimate if the numerical treatment were basedon Pareto'sequality relation for utility This relation is, indeed, a stronger and more complidifferences (cf. the end of cated hypothesis, added to the original ones concerning the general comparability of of utilities (completeness preferences). ) instead, and we hope that the reader However, we used the operation au + (1 a will agreewith us that it representsn even safer assumption than that of the complete-

o Concept f Marginal Utility clearthat we feel free to make of use of a numerical conception utility. On the other hand, subsequent 1Or that he can assertthat they are preciselyequally desirable. 2 I.e. completeness the postulate (3:A:a). 1

3.8.1. preceding The analysis made it

I.e.
:B),

The 3.8. Role of the

3.4.6.).

as We think therefore that our procedure, distinguished from Pareto's,is not open to the objections asedon the necessityof artificial assumptions and a lossof simplicity. b 6 This amounts to weakening (3:A:a) an (3:A:a')y replacing in it \"one and only to b then correspondto (65:B:a), one\" by \"at most one/' The conditions (3:A:a')(3:A:b)

In this necessary.)))

(65:B:b). 6

case some modifications

in the groups

of postulates

( (3:B),3:0)are also

30

FORMULATIONOF THE ECONOMIC PROBLEM

will discussions show that we cannot avoid the assumptionthat all subjects of the economy under considerationare completelyinformed about the of and physicalcharacteristicsthe situation in which they operate areable to perform all statistical, mathematical, etc., perationswhich this knowlo edgemakes possible.The nature and importanceof this assumptionhas been given extensiveattention in the literatureand the subjectis probably very far from being exhausted. We propose not to enter upon it. The questionis too vast and too difficult and we believethat it is best to \" divide w difficulties.\" I.e. e wish to avoid this complication which, while interesting in its own right, should be considered separately from our present

that our investigations although they assume without any further discussion do make a contribution to the study of this subject. It will be seenthat many economic and social henomenawhich areusually ascribed the individual'sstateof to p \" information\" make their appearance our theory and can be in incomplete satisfactorilyinterpreted with its help. Sinceour theory assumes \" completeinformation,\" we concludefrom \"this that those phenomena have information.\" Some nothing to do with the individual's incomplete particularly striking examples of this will be found in the conceptsof \" i in and discrimination\"n of \"incomplete xploitation\" 38.3., of the e i \"transfer\"or \"tribute\"n 46.12. On the basis of the above we would even venture to questionthe impor1 tanceusually ascribedto incompleteinformation in its conventional sense a in economicnd socialheory. It will appear that some t phenomenawhich would prima facie have to be attributed to this factor, have nothing to do
\"

problem.
Actually

we think

information\" complete

33.1., 46.11.,

with

definite physical definite quantities of goods at his disposal.In view of what wassaid above, he is in a positionto determine maximum the which can be obtained in this situation. Sincethe maximum is a utility when a well-definedquantity, the sameis true for the increasehich occurs w unit of any definite goodis added to the stock all goodsin the possession of of the individual. This is, of course, he classical t notion of the marginal 8 of a unit of the commodity in question. utility Thesequantities are clearly of decisiveimportance in the \"Robinson Crusoe\" conomy. The above marginal utility obviously correspondsto e 1We shall seethat the rules of the games considered that may explicitly prescribe 6.4. certain participants should not possessertain piecesof information. Cf. 6.3., c i and of and (Gamesn which this doesnot happen are referredto in 14.8. in (15:B) 15.3.2., arecalled We shall recognizeand utilize this kind of gameswith \" perfect information.\") \"incomplete information\" (according to the above, rather to be called \"imperfect But we reject all other types, vaguely defined by the use of concepts information\.") like complication, intelligence, etc.

and characteristics

Let 3.8.2. us now consideran isolated individual with

with

2 it.

2 Our theory attributes these phenomena to the possibility of multiple standards of behavior\" cf 4.6.and the end of 4.7. * More precisely:he so-called t utility.\) \"indirectly dependent expected

\"stable

A SOLUTIONS ND STANDARDSOF BEHAVIOR

31

the maximum effort which he will be willing to make if he behavesaccordto ing to the customary criteria of rationality in order obtain a further of that commodity. all, however, what significance it has in determining the behavior of a participant in a socialexchangeconomy. We saw that e the principlesof rational behavior in this case still await formulation, and that they are certainly not expressed a maximum requirement of the by Crusoetype. Thus it must be uncertain whether marginal utility has any 1 meaning at all in this case. Positivestatements on this subjectwill be possibleonly after we have in succeeded developinga theory of rational behavior in a social xchange e economy, that is, as was stated before, with the help of the theory of o \"games f strategy.\" It will be seen that marginal utility does,indeed, b play an important role in this casetoo, ut in a more subtle way than is usually assumed.
unit

It is not clearat

We have now reachedthe point where it becomesossibleto p T give a positive description of our proposed procedure. his means primarily an outline and an accountof the main technicalconceptsand

4.1.1.

4. Structure of the Theory : Solutionsand Standards of Behavior 4.1. Simplest Conceptof a Solution for OneParticipant The

devices.

of economy, and to derive from them the generalcharacteristics that behavior. And while the principles ought to be perfectly general valid in all situations we may be satisfiedif we can find solutions,for the cases. moment, only in some characteristicpecial s a First of all we must obtain a clear notion of what can be accepteds a solution of this problem; what the amount of information is which a solution must convey, and what we should expect egarding its formal r structure, A precise p analysis becomesossible only after these matters have been clarified. Theimmediateconcept f a solutionis plausiblya setof rules for o eachparticipant which tell him how to behavein every situation which may a conceivably arise. One may object t this point that this view is unnecesinclusive. Since want to theorize about \" rational behavior,\" there we sarily to seems be no need to give the individual advice as to his behavior in situations other than those which arisein a rational community. This would justify assumingrational behavior on the part of the othersas well, that. Such a procedure in whatever way we are going to characterize o would probably lead to a unique sequence f situations to which alone our theory needrefer.

As we stated before, we wish to find the mathematically complete principleswhich define \" rational behavior\" for the participants in a social

i.e.,

i.e.,

4.1.2.

1All
they

are relaxed, then

this

is understood

within the domain of our severalsimplifying various further difficulties ensue.)))

assumptions.

If

32

FORMULATIONOF THE ECONOMIC PROBLEM

This objection seemso be invalid for two reasons: t t First,the \"rulesof the game,\" i.e.he physical laws which give the factual backgroundof the economicctivities under considerationmay be a explicitlystatistical The actions of the participants of the economy may the with events which depend on determine outcome only in conjunction I chance(with known probabilities),cf. footnote 2 on p. 10and 6.2.1.f then this is taken into consideration, the rulesof behavior even in a perfectly rational community must providefor a greatvariety of situations someof 1 which will be very far from optimum. and Second, this is even more fundamental, the rulesof rational behavior must provide definitely for the possibilityof irrational conducton the part of others. In otherwords: Imagine that we have discovereda set of rules for all participants to be termed \"optimal\"or \"rational\"each of as which is indeed optimal provided that the other participants conform. Then the questionremainsas to what will happenif someof the participants do not conform. If that shouldturn out to be advantageousfor them and, quite particularly, disadvantageousto the conformists then the above \"solution\" would seem very questionable.We arein no positionto give a discussionof these things as yet but we want to make it clear positive that under such conditionsthe \"solution,\" r at leastits motivation, must o be considereds imperfect and incomplete. n whatever way we formulate a I the guidingprinciplesand the objectivejustification of \"rational behavior,\" provisoswill have to be made for every possibleconduct of \"theothers.\" Only in this way can a satisfactory and exhaustive theory be developed. But if the superiorityof \"rational behavior\" over any other kind is to be t established,hen its description must include rules of conduct for all conceivable situations including those where \"the others\" behaved of irrationally, in the sense the standards which the theory will setfor them. 4.1,3. this stagethe readerwill observea great similarity with the At o everyday concept f games. We think that this similarity is very essential; a indeed,that it is more than that. For economicnd social roblemsthe p gamesfulfill or shouldfulfill the samefunction which various geometricomathematical modelshave successfully performed in the physicalsciences. Suchmodelsare theoretical onstructs with a precise, c exhaustive and not too complicateddefinition; and they must be similar to reality in those respectswhich are essentialin the investigation at hand. To recaand exhaustive in pitulate in detail: The definition must be precise orderto make a mathematical treatment possible.The constructmust not be unduly complicated, that the mathematical treatment can be so the mereformalism to the point where it yields complete brought beyond numerical results. Similarity to reality is neededto make the operation to significant. And this similarity must usually berestricted a few traits
1That a unique optimal behavior is at all conceivable spite of the multiplicity in of the possibilities determined by chance,s of course i due to the useof the notion of \"mathe1
matical

expectation. Cf.loc.cit. above.)))

'

SOLUTIONSAND STANDARDSOF BEHAVIOR

33

deemed \"essential\" tempore sinceotherwisethe above requirements pro 1 would conflict with eachother. It is clearhat if a model of economicctivities is constructedaccording a t
striking in the formal

to theseprinciples,the descriptionof a game results. This is particularly descriptionof markets which are after all the core of the economicystem but this statement is true in all cases nd without a s qualifications. 4.1.4. describedin a We a what we expect solution i.e. characterization of \" rational behavior \" to consistof. This amounted to a complete set of rules of behavior in all conceivablesituations. This holds equivalently for a social economy and for games. The entire result in the above senseis thus a combinatorial enumeration of enormous complexity. a But we have accepted simplified concept f utility accordingto which all o the individual strives for is fully describedby one numerical datum (cf. and Thus the complicatedcombinatorial catalogue which f we expectrom a solution permitsa very brief and significant summarizacan t tion:he statement of how much 2 3 the participant under consideration p get if he behaves \" rationally. \" This \"canget\"is, of course, resumedto be a minimum; he may get more if the others make mistakes (behave

4.1.2.

2.1.1. 3.3.).

irrationally).

It ought to be understood that all this discussionis advanced, as it should be, preliminary to the building of a satisfactory theory along the lines indicated. We formulate desiderata which will serve as a gauge of i with successn our subsequentconsiderations; it is in accordance the but usual heuristic procedure reason about these desiderata even before to we are able to satisfy them. Indeed,his preliminary reasoning is an t essentialpart of the process finding a satisfactorytheory.4 of
4.2.Extension to All Participants 4.2.1. have consideredso far only what the solution ought to be for We one participant. Let us now visualize all participants simultaneously. l I.e.,et us considera social economy, or equivalently a game of a fixed information which a solution number of (say n) participants. The complete should convey is, as we discussedit, of a combinatorial nature. It was indicated furthermore how a single quantitative statement contains the decisivepart of this information, by stating how much each participant
description of the solar system by a small number of \"masspomts.\" attract eachother and move like the stars;this is the similarity in the essentials, while the enormous wealth of the other physical features of the planets has beenleft out of account. 2 Utility; for an entrepreneur, gain or loss. profit ; for a player, 8 We t mean, of course,he \"mathematical expectation,\" if there is an explicit element of chance. Cf.the first remark in and alsothe discussionof 4 Thosewho are familiar with the development of physics will know how important such heuristic considerations can be. Neither general relativity nor quantum mechanics ;/ Hiannaainn nf thft HpaidemtA concerncould have beenfound without a \"T>rA-thpnrptip*i the theory-to-be.))) ing

Thesepoints

1E.g., ewton's N

4.1.2.

3.7.1.

34

FORMULATIONOF THE ECONOMIC PROBLEM

we (Of course have not yet formulated theserequirements.Foran exhausc c b tive discussion, f. loc. it. elow.) The structure of the societyunder considerationwould then be extremelysimple: Therewould existan absolute state of equilibrium in which the quantitative shareof every participant would be preciselydetermined. It will be seenhowever that such a solution, possessingall necessary properties, does not exist in general. The notion of a solution will have to be broadenedconsiderably,and it will be seenthat this is closelyconnectedwith certaininherent features of social organization that are well \" known from a \" common sense point of view but thus far have not been viewed in proper perspective. (Cf. 4.6.nd 4.8.1.) a 4.2.3. mathematical analysis of the problem will show that there Our exists, indeed,a not inconsiderable family of gameswhere a solution can be defined and found in the above sense: .e. s one singleimputation. In i a such cases obtains at leastthe amount thus imputed to every participant him by just behaving appropriately, rationally. Indeed, e gets exactly h this amount if the other participants too behave rationally; if they do not, he may get even more. Thesearethe gamesof two participants where the sum of all payments is zero. While these gamesare not exactly typical for majoreconomic processes, contain some universally important traits of all gamesand they the resultsderivedfrom them arethe basis of the generalheory of games. t We shalldiscussthem at length in Chapter III.

obtains by behaving rationally. Considertheseamounts which the several \" participants''obtain. If the solution did nothing more in the quantitative with the well sensethan specify these amounts, 1 then it would coincide of i known concept imputation:t would just statehow the total proceeds 2 areto be distributed among the participants. We emphasize that the problem of imputation must be solved both are when the total proceeds in fact identicallyzero and when they arevariable. This problem,in its general orm, has neither been properlyformuf latednor solvedin economic literature. 4.2.2. can seeno reason why one should not be satisfied with a We a solution of this nature, providing it can be found: i.e. singleimputation which meetsreasonable requirements for optimum (rational) behavior.

1And of course,n the combinatorial sense,as outlined above,the procedure i how to obtain them. *In games as usually understood the total proceedsare always zero; one participant can gain only what the others lose. Thus there is a pure problem of distribution and absolutely none of increasing the total utility, the \"social imputation In all economicquestions the latter problem arisesas well, but the question product.\" of imputation remains. Subsequently we shall broaden the concept f a game by dropo ping the requirement of the total proceedseing zero(cf. Ch. XI).))) b

4.3.The Solution as a Set of Imputations 4.3.1.eitherof the two above restrictionss dropped,the situation is If i altered materially.

i.e.

i.e.

A SOLUTIONS ND STANDARDSOF BEHAVIOR

35

The simplest gamewhere the secondrequirement is oversteppedis a two-persongame where the sum of all payments is variable. This corresponds to a social economy with two participants and allows both for 1 and their interdependence for variability of total utility with their behavior. As a matter of fact this is exactly the caseof a bilateral monopoly (cf. The 6L2.-61.6.). well known \"zone of uncertainty \" which is found in current efforts to solve the problemof imputation indicatesthat a broader c conceptof solution must be sought. This casewill be discussedloc. it. above. Forthe moment we want to use it only as an indicatorof the diffiwhich is more suitableas a basis for a first culty and pass to the other case

the above two-persongame, this does not correspondto any fundamental n a economicroblembut it represents evertheless basicpossibilityin human p relations. The essentialfeature is that any two players who combineand against a third can thereby securean advantage. The problem cooperate is how this advantage shouldbe distributed among the two partners in this combination. Any such schemeof imputation will have to take into t accounthat any two partners can combine; while any onecombination of is in the process formation, eachpartner must considerthe fact that his could break away and join the third participant. prospectiveally h of Of coursethe rules of the game will prescribeow the proceeds a discoalition should be divided between the partners. But the detailed shows that this will not be, in general,the cussion to be given in 22.1. a final verdict. Imagine game (of three or more persons) in which two can form a very advantageouscoalition but where the rules participants of the gameprovide that the greatestpart of the gain goesto the first participant. Assume furthermore that the second participant of this coalition can also entera coalition with the third one,which is lesseffective in toto but promises him a greaterindividual gain than the former. In this situation it is obviously reasonablefor the first participant to transfer a part of the gains which he could get from the first coalition to the second participant in order to save this coalition. In other words: One must t expecthat under certainconditionsone participant of a coalition will be willing to pay a compensationto his partner. Thus the apportionment within a coalition depends not only upon the rules of the game but also upon the above principles, under the influence of the alternative

4.3.2. simplestgamewhere the first requirementis disregardedis a The g three-personame where the sum of all payments is zero. In contrast to

positive step.

i.e.

2 coalitions.

that one cannot expect ny theoretical tateCommon sensesuggests s a ment as to which alliance will be formed3 but only information concerning 1It will be remembered that we make useof a transferable utility, cf. * This doesnot mean that the rules of the game are violated, sincesuch compensatory payments, if made at all, are made freely in pursuance of a rational consideration. 1Obviously three combinations of two partners eachare possible. In the example w to be given in 21., preference ithin the solution for a particular alliance will be any

2.1.1.

a)))

36

FORMULATIONOF THE ECONOMIC PROBLEM

how the partners in a possiblecombination must divide the spoilsin order to to avoid the contingency that any oneof them deserts form a combination in with the third player. All this will be discussed detailand quantitatively
in

It suffices to state here only the result which the above qualitative make plausibleand which will be establishedmore rigorously considerations o c loc. it. A reasonableoncept f a solution consistsin this case a system of c of three imputations. These correspond to the above-mentionedthree combinationsor alliancesand express division of spoilsbetweenrespecthe
tive

Ch.V.

We shall seethat a consistent theory will for solutions which are not single imputations, but

situation.

4.3.3. last result will The

allies.

turn

out to be the prototype of the general

result from looking rather systems of

imputations. It is clearthat in the above three-person no single imputation game from the solution is in itself anything like a solution. Any particular alliance describes one particular considerationwhich entersthe minds only of the participants when they plan their behavior. Even if a particular is alliance ultimately formed, the division of the proceeds betweenthe allies will be decisivelyinfluenced by the other alliances which each one might alternatively have entered. Thus only the three alliancesand their imputations togetherform a rational whole which determines all of its a stability of its own. It is, indeed,this whole which detailsand possesses is the really significant entity, more so than its constituent imputations. i Even if one of these is actually applied, i.e.f one particular allianceis actually formed, the others arepresent in a \"virtual\" existence: Although they have not materialized, hey have contributedessentiallyto shapingand t determiningthe actualreality. In conceiving of the general roblem,a socialconomy or equivalently p e a gameof n participants, we shall with an optimismwhich can be justified t only by subsequentsuccessexpecthe same thing:A solution shouldbe a 1 in system of imputations possessing its entirety some kind of balanceand stability the nature of which we shall try to determine.We emphasize that this stability whatever it may turn out to be will be a property of the system as a whole and not of the singleimputations of which it is T composed. hese brief considerationsregarding the three-person game have illustratedthis point. w a system of imputationsas a 4.3.4. exactcriteria hich characterize The solution of our problem are,of course,of a mathematical nature. For a t a precisend exhaustive discussionwe must therefore refer the readero the d mathematical developmentof the theory. Theexact efinition subsequent
the game will be symmetric with respect o all three t limine excludedby symmetry. Of.however participants. 1They may again include compensations between partners in a coalition, asdescribed in 4.3.2.)))

I.e. 33.1.1.

A SOLUTIONS ND STANDARDSOF BEHAVIOR

37

itself is stated in undertaketo give a prelimiWe shall nevertheless nary, qualitative outline. We hopethis will contributeto the understanding the of the ideas on which the quantitative discussionis based. Besides, placeof our considerationsin the generalframework of socialheory will t

30.1.1.

becomelearer. c

4.4.The Intransitive Notion of \"Superiority\" or \"Domination\" 4.4.1. us return to a more primitive conceptof the solutionwhich we Let know already must be abandoned. We mean the ideaof a solution as a it singleimputation. If this sort of solution existed would have to be an imputation which in some plausiblesensewas superiorto all otherimputations. This notion of superiority as between imputations ought to be formulated in a way which takesaccountof the physicaland socialstructure of the milieu. That is, one should define that an imputation x is superior to an imputation y whenever this happens:Assume that society, t i.e.he totality of all participants, has to considerhe questionwhether or t not to \"accept\"staticsettlement f all questionsof distribution by the a o imputation y. Assumefurthermore that at this moment the alternative T settlement y the imputation x is also considered. hen this alternative x b will suffice to exclude of acceptance y. By this we mean that a sufficient number of participants prefer in their own interestx to i/, and areconvinced or can be convinced of the possibilityof obtaining the advantages of x. In this comparisonof x to y the participants should not be influenced by w of the consideration any third alternatives (imputations). I.e. e conceive the the relationship of superiority as an elementary one,correlating two imputations x and y only. The further comparisonof three or more ultimately of all imputations is the subject of the theory which must now follow, as a superstructure erected upon the elementary conceptof superiority. Whether the possibilityof obtaining certainadvantagesby relinquishing in y for x as discussed the above definition, can be made convincing to the interestedpartieswill depend upon the physical facts of the situation in the terminology of games, n the rules of the game. o a We prefer to use,instead of \" superior\" ith its manifold associations, w word more in the nature of a terminus technicus. When the above described 1 relationship between two imputations x and y exists,then we shall say
y

that x dominates y. If one restates little more carefully what should be expected from a a S solution consistingof a single imputation, this formulation obtains: uch an imputation should dominate all others and be dominated by

none.

30.1.1.)))

as formulated or rather indicated above is clearly in the nature of an ordering, similar to the question of 1 That is, when it holds in the mathematically precise form, which will be given in

4.4.2. notion of domination The

38

FORMULATIONOF THE ECONOMIC PROBLEM

or preference, of size in any quantitative theory. The notion of a single w imputation solution1 correspondsto that of the first element ith respect 2 to that ordering. w Thesearch such a first element ould be a plausibleone if the orderfor our notion of domination, possessedthe important ing in question, property of transitivity ; that is, if it were true that whenever x dominates one y and y dominatesz, then alsox dominatesz. In this case might proceed as follows:Starting with an arbitrary x, look for a y which dominatesa:;if such a y exists, hoose and look for a z which dominatesy\\ if such a z one c choose ne and lookfor a u which dominatesz, etc. In most practical o exists, e problemsthere is a fair chancethat this processitherterminates after a o finite number of steps with a w which is undominatedby anything else, r but that these that the sequence y, z, u, , goeson ad infinitum, x, tend to a limiting positionw undominatedby anything else. x, y, z, u, And, due to the transitivity referredto above, the final w will in eithercase dominateall previously obtained x, y, z, w, We shall not go into more elaborate etails which could and should d t be given in an exhaustive discussion.It will probablybe clearo the reader that the progress through the sequence y, z, u, #, correspondsto \" t successive improvements \" culminating in the \" optimum,\" i.e.he \"first\" w element which dominatesall others and is not dominated. All this becomesery different when transitivity does not prevail. v In that caseany attempt to reachan \"optimum\" by successiveimprovements may be futile. It can happen that x is dominatedby y y by z, and

i.e.

4.4.3. the notion of domination on which we rely is, indeed,not Now w transitive. In our tentative descriptionof this concept e indicatedthat x dominatesy when there exists group of participants eachone of whom a prefers his individual situation in x to that in y, and who are convinced that they are able as a group i.e. s an alliance to enforce their prefera ences. We shall discuss these matters in detail in 30.2.This group of participantsshall becalledthe \"effectiveset\"for the domination of x over y. Now when x dominatesy and y dominatesz, the effective sets for thesetwo can dominationsmay be entirely disjunct and therefore no conclusions be drawn concerning relationshipbetween z and x. It can even happen the that z dominates x with the help of a third effective set, possiblydisjunct' from both previous ones.
1We continue to use it as an illustration although we have shown already that it is a forlorn hope. Thereason this is that, by showing what is involved if certain complicafor tions did not arise, we can put these complications into better perspective. Our real interest at this stagelies of course thesecomplications, which are quite fundamental. in 1Themathematical theory of ordering is very simple and leadsprobably to a deeper of these conditions than any purely verbal discussion. The necessary understanding mathematical considerations will be found in 65.3. 8 In the caseof x never transitivity this is impossible because if a proof be wanted dominates itself. Indeed,if y dominates x, z dominates y, and x dominates z, then we can infer by transitivity that x dominates

z in turn

by x.8

e.g.

x.)))

A SOLUTIONS ND STANDARDSOF BEHAVIOR

may

39

Thislack of transitivity, especially the above formalistic presentation, in and desirable appear to be an annoying complication it may even seem to make an effort to rid the theory of it. Yet the reader ho takes another w look at the last paragraph will notice that it really containsonly a circumlocution of a most typical phenomenon in all social organizations. The domination relationshipsbetween various imputations z, i/, z, i.e. betweenvarious statesof society correspondto the various ways in which these can unstabilize i.e.upset each other. That various groups of participants acting as effective sets in various relations of this kind may d bring about \"cyclical\" ominations e.g., over x, z over y, and x over z y is indeed one of the most characteristicifficulties which a theory of these d phenomenamust face. 4.5.ThePreciseDefinition of a Solution 4.6.1. our task is to replacehe notion of the optimum i.e.f the o Thus t first element by something which can take over its functions in a static equilibrium. This becomesecessary becausethe original concepthas n instance become untenable. We first observedits breakdownin the specific of a certainthree-personame in 4.3.2.-4.3.3.now we have acquired But g o a deeper of i insight into the cause its failure:t is the nature of our concept f domination, and specifically its intransitivity. This type of relationshipis not at all peculiar to our problem. Other that instancesof it are well known in many fields and it is to be regretted have never receiveda generic mathematical treatment. We mean all they w of those concepts hich arein the general ature of a comparison preference n the or \"superiority,\" of order, but lack transitivity: e.g., strength of or e 1 form\" in sports and races, tc. chessplayersin a tournament, the \"paper indicated 4.5.2. discussionof the three-personame in 4.3.2.-4.3.3. The g a that the solution will be,in general, set of imputationsinstead of a single w imputation. That is, the conceptof the \"first element\" ill have to be that of a set of elementsimputations)with suitable properties. ( replacedy b In the exhaustive discussionof this game in 32.(cf. also the interpretawhich tion in 33.1.1. calls attention to somedeviations)the system of three in which was introducedas the solutionof the three-person game imputations, will be derived in an exact ay with the help of the postulates w 4.3.2.-4.3.3., Thesepostulates will be very similar to those which characterof 30.1.1. ize a first element.They are,of course, equirementsfor a setof elements r (imputations),but if that set should turn out to consistof a singleelement only, then our postulates go over into the characterization of the first element(in the total system of all imputations). We do not give a detailedmotivation for those postulatesas yet, but we w shall formulate them now hoping that the reader ill find them to besome1Someof these problems have beentreated mathematically by the introduction of Without denying that this approach has a certain justification, and chance probability. even in those cases. It we doubt whether it is conducive to a complete understanding would be altogether inadequate for our considerations of socialrganization.))) o

40

FORMULATIONOF THE ECONOMIC PROBLEM

reasonsof a qualitative nature, or ratheronepossible what plausible. Some interpretation, will be given in the paragraphsimmediatelyfollowing. The 4.5.3. postulatesare as follows:A set S of elements(imputations) these is a solutionwhen it possesses two properties:

(4:A:a) (4:A:b)

No y contained in S is dominatedby an x containedin S. Every y not containedin S is dominated by some x contained in S. can and (4:A:a) (4:A:b) be stated as a singlecondition:
which of The elements S are preciselythose elements undominated by elements f 1 o

(4:A:c)

S.

are

in who is interested this type of exercise ay now verify The*reader m x our previous assertionthat for a set S which consistsof a singleelement the above conditionsexpressreciselythat x is the first element. p 4.5.4. of the malaisewhich the preceding Part postulates may causeat first sight is probably due to their circular character.This is particularly w obvious in the form (4:A:c), here the elements f S arecharacterizedy a o b It is important not to relationship which is again dependent upon misunderstandthe meaning of this circumstance. Sinceour definitions (4:A:a)and (4:A:b),or (4:A:c), re circular a an implicit for S, it is not at all clearthat there really exists S which fulfills them, nor whether if there exists one the S is unique. Indeed thesequestions,at this stagestill unanswered,arethe main subjectof the subsequenttheory. What is clear, owever, is that these definitions tell h unambiguously whether any particular S is or is not a solution. If one insists on associating with the conceptof a definition the attributes of a existencend uniquenessof the objectdefined, then one must say: We have not given a definition of S, but a definition of a property of S we have not defined the solution but characterized possible solutions. all Whether the totality of all solutions, thus circumscribed, contains no S, oneS,or several<S's,is subjectfor further inquiry.2 exactly

S.

i.e.

s concerningocialphenomena.
1

Termsof \"Standards of Behavior\" The singleimputation is an often used and well understood conceptof economicheory, while the setsof imputations to which we have t beenled are rather unfamiliar ones. It is therefore desirable to correlate them with somethingwhich has a well established placein our thinking
of Our Definition
in

4.6.1.

4.6. Interpretation

Thus (4:A:c)s an exactequivalent of (4:A:a) nd (4:A:b)ogether. It may impress i a t the mathematically untrained reader as somewhat involved, although it is really a straightforward expressionof rather simple ideas. 2 It should be unnecessary to say that the circularity, or rather implicitness, of or does (4:A:a)and (4:A:b), (4:A:c), not at all mean that they are tautological. They o a express, f course, very serious restriction of
S.)))

SOLUTIONSAND STANDARDSOF BEHAVIOR

ing

41

organization. Let us examine this assertionmore closely. Let the physical basis of a social economy be given, or, to take a 1 broader view of the matter, of a society. According to all tradition and human beingshave a characteristicay of adjustingthemselves w experience to such a background. This consistsof not setting up one rigid system of apportionment, i.e.of imputation, but rather a variety of alternatives, which will probably all expressome generalprinciplesbut nevertheless s 2 differ among themselvesin many particular respects. This system of \" the establishedorder of society\" or \" accepted imputations describes 1 standard of behavior/ \" Obviously no random grouping of imputations will do as such a standi ard of behavior \":t will have to satisfy certain conditionswhich charactero ize it as a possible order of things. This concept f possibilitymust clearly for conditions of stability. The readerwill observe,no doubt, provide that our procedurein the previous paragraphsis very much in this spirit: The setsS of imputations x, y, z, correspondto what we now call a o \"standardof behavior/'and the conditions(4:A:a) nd (4:A:b), r (4:A:c), a stability in the above the which characterize solution S express, indeed,

i Indeed,t appearsthat the sets of imputations S which we areconsider\" with a social correspondto the \"standard of behavior connected

sense. ate in

i a (4:A:a) nd (4:A:b)s particularly appropriinstance. Recall that domination of y by x means that the o imputation x, if taken into consideration,excludesacceptancef the
this

The 4.6.2. disjunctioninto

imputation y (this without forecasting what imputation will ultimately be the Thus and cf. accepted, 4.4.1. 4.4.2.). (4:A:a)expresses fact that the No standard of behavior is free from inner contradictions: imputation y standard of behavior\" c belongingto S i.e. onforming with the \"accepted d can be upset i.e. ominated by another imputation x of the samekind. that On the other hand (4:A:b) expresses the \"standardof behavior \" can be used to discredit any non-conforming procedure: Every imputation y dominated by an imputation x not belonging to S can be upset belongingto S. that Observethat we have not postulated in 4.5.3. a y belonging to S if shouldnever be dominatedby any x.3 Of course, this shouldhappen,then x would have to be outside of S, due to (4:A:a). In the terminology of An socialorganizations: imputation y which conforms with the \"accepted 1 In the caseof a game this means simply as we have mentioned before that the rules of the game are given. But for the present simile the comparison with a social economy is more useful. We suggest therefore that the readerforget temporarily the o analogy with gamesand think entirely in terms of social rganization. 2 Theremay be extreme, or to use a mathematical term, \"degenerate\" special ases c can where the setup is of such exceptional simplicity that a rigid single apportionment be put into operation. But it seems legitimate to disregard them as non-typical. 8 It can be that :M)in 31.2.3., such a postulate cannot be fulfilled shown, cf. (31 it in general;i.e. that in all really interesting cases is impossible to find an Swhich satisfies

i.e.

it together with our other requirements.)))

42

FORMULATIONOF THE ECONOMICPROBLEM

standard of behavior\" may be upset by another imputation x, but in this it case is certain that x doesnot conform.l It followsfrom our other require-

ments that then x is upset in turn by a third imputation z which again conforms. Since and z both conform, z cannot upset y a further illustray \" \" tion of the intransitivity of domination. Thus our solutions S correspondto such \" standards of behavior as have an inner stability: oncethey are generally accepted they overrule elseand no part of them can be overruled within the limits of everything standards. This is clearly how things are in actual social the accepted of the organizations, and it emphasizes perfect appropriateness the circular of character our conditionsin 4.5.3. to We 4.6.3. have previously mentioned, but purposelyneglected disThat n an important objection: neither the existenceor the uniqueness cuss, o of a of a solution S in the sense the conditions(4:A:a) nd (4:A:b), r (4:A:c), is in 4.5.3.evident or established. as I There can be, of course,no concessions regards existence. f it a turn out that our requirementsconcerning solution S are,in any should a u specialcase, nfulfillable, this would certainly necessitate fundamental of in the theory. Thus a general proof of the existence solutionsS change 2 casesis most desirable.It will appear from our subsefor all particular quent investigations that this proof has not yet been carried out in full c generality but that in all cases onsideredso far solutionswere found. d the situation is altogether ifferent. The often As regardsuniqueness \" of our requirements makes it rather mentioned circular\"character u probablethat the solutionsarenot in general nique. Indeedwe shall in 3 what we have o most cases bservea multiplicity of solutions. Considering saidabout interpretingsolutionsas stable \" standards of behavior\" this has a simple and not unreasonablemeaning, namely that given the same \" \" physicalbackgrounddifferent establishedorders of society\"or accepted those characteristics of standards of behavior\" can be built, all possessing inner stability which we have discussed.Sincethis conceptof stability is admittedly of an \"inner\" nature i.e.perative only under the hypothesis o of of general cceptance the standard in question these different standards a may perfectly well be in contradictionwith eachother. Our 4.6.4. approach should be compared with the widely held view that a social theory is possibleonly on the basis of some preconceived principlesof social purpose. Theseprincipleswould include quantitative b statements concerningoth the aims to be achieved in toto and the appora tionments betweenindividuals. Oncethey areaccepted,simplemaximum

'

problemresults.

1 We use the word \"conform\" (to the \"standard of behavior'')temporarily as a and the word \"upset\" asa synonym for dominate. synonym for being contained in *In the terminology of games:for all numbers of participants and for all possible rules of the game. 1An interesting exceptionis 65.8.)))

S,

SOLUTIONSAND STANDARDSOF BEHAVIOR

43

Let us note that no such statement of principlesis ever satisfactory per se, and the arguments adduced in its favor are usually eitherthose of inner stability or of less clearly defined kinds of desirability,mainly concerning distribution. Little can be said about the lattertype of motivation. Our problem is not to determine what ought to happen in pursuance of any set of necessarily arbitrary a priori principles, but to investigate where the equilibrium of forceslies. As far as the first motivation is concerned, has been our aim to give it a b just those arguments precisend satisfactoryform, concerningoth global aims and individual apportionments. This made it necessaryto takeup the entirequestionof inner stability as a problemin its own right. A theory which is consistentat this point cannot fail to give a preciseccount f the a o entireinterplay of economic i interests,nfluence and power. 4.7.Gamesand SocialOrganizations 4.7.It may now be opportuneto revive the analogy with games,which we purposelysuppressed in the previous paragraphs (cf. footnote 1 on on p. 41). The parallelismbetween the solutionsS in the senseof 4.5.3. one hand and of stable \" standards of behavior \" on the other can be used i theseconceptsn both directions. for corroboration assertionsconcerning of At leastwe hope that this suggestionwill have some appeal to the reader. We think that the procedure of the mathematical theory of games of which exists strategy gains definitely in plausibility by the correspondence between its conceptsand those of social organizations. On the other hand, almost every statement which we or for that matter anyone else social organizations, runs afoul of some existing ever made concerning the very nature of things, most opinionsthus far could opinion. And, by hardly have been proved or disproved within the field of social theory. It is therefore a greathelpthat all our assertionscan be borneout by specific from examples the theory of gamesof strategy. Suchis indeed one of the standard techniques of using models in the brings out a significant funcphysical sciences.This two-way procedure in tion of models,not emphasizedin their discussion 4.1.3. To give an illustration:The question whether several stable \" orders of society\" or \" standards of behavior \" based on the same physical background are possibleor not, is highly controversial.There is little hope of that it will be settled the usual methods because the enormouscomby plexity of this problem among other reasons.But we shall give specific several of examples gamesof threeor four persons,where one gamepossesses A of will solutionsin the senseof 4.5.3. nd some these examples be seen to be modelsfor certainsimpleeconomic roblems.(Cf. 62.) p

4.8.1.
nature.)))

4.8. Concluding

Remarks

it In conclusion

remainsto make a few remarks of a more formal

44

FORMULATIONOF THE ECONOMIC PROBLEM

Our considerations started with single We begin with this observation: imputations which were originally quantitative extractsfrom more to detailed combinatorial sets of rules. From these we had to proceed setsS of imputations,which under certainconditionsappearedas solutions. Sincethe solutionsdo not seem to be necessarilyunique, the complete answer to any specific problem consists not in finding a solution, but in determiningthe set of all solutions. Thus the entity for which we look in of any particularproblemis really a setof sets imputations. This may seem in to be unnaturally complicated itself; besidesthereappears no guarantee w that this processill not have to be carried further, conceivably because of laterdifficulties. Concerningthese doubts it sufficesto say: First,the mathematical structure of the theory of gamesof strategy providesa formal the connections S justification of our procedure. econd, previously discussed \" to standards of behavior \" (corresponding setsof imputations) and with of the multiplicity of \" standards of behavior \" on the same physicalbackto ground (corresponding setsof setsof imputations)makejust this amount d of complicatedness esirable. Onemay criticize our interpretationof sets of imputationsas \" standards and 4.1.4. introduced a more elewe of behavior.\" Previously in which may strike the reader s a direct formulation of a a mentary concept, \" t standard of behavior\":his was the preliminary combinatorial concept of a solution as a setof rules for eachparticipant, telling him how to behave in every possible situation of the game. (From these rules the single as imputations were then extracted a quantitative summary, cf. above.) Sucha simple view of the \" standard of behavior\" could be maintained, between however, only in gamesin which coalitionsand the compensations coalition partners (cf. 4.3.2.) no role,sincethe above rules do not play i provide for thesepossibilities. Gamesexistn which coalitionsand compenbe disregarded: the two-person sations can e.g. game of zero-summentioned \" and in 4.2.3., more generally the \" inessential games to be discussedin and But the general,ypical game in particular 27.3. in (31 of :P) t of all .significant problems a socialexchange witheconomy cannot be treated out thesedevices. Thus the samearguments which forced us to consider sets of imputations instead of singleimputations necessitate abandonment the o of that narrow concept f \" standard of behavior.\" Actually we shall call thesesetsof rules the \" strategies\"f the game. o 4.8.2. next subjectto be mentioned concernshe staticor dynamic The t nature of the theory. We repeatmost emphatically that our theory is thoroughly static. A dynamic theory would unquestionably be more and complete therefore preferable. But thereis ample evidencefrom other branchesof science it is futile to try to build one as long as the static that side is not thoroughly understood. On the other hand, the readermay t objecto somedefinitely dynamic arguments which weremade in the course of our discussions.This appliesparticularly to all considerations concernthe interplay of various imputations under the influence of \"domina-))) ing

4.1.2.

31.2.3.

SOLUTIONSAND STANDARDSOF BEHAVIOR

tion,\" cf.

45

4.6.2. e think that this is perfectly legitimate. A static W 1 of theory dealswith equilibria. The essentialcharacteristic an equilibrium i t is that it has no tendencyto change, .e.hat it is not conducive to dynamic developments. An analysis of this feature is, of course,inconceivable the use of certain rudimentary dynamic concepts.The important without point is that they arerudimentary. In other words:Forthe real dynamics which investigatesthe precise otions, usually far away from equilibria,a m 2' of much deeper nowledge thesedynamic phenomenais required.3 k 4.8.3. let us note a point at which the theory of socialphenomena Finally take a very definite turn away from the existingpatterns of will presumably mathematical physics. This is, of course, nly a surmiseon a subject where o much uncertainty and obscurityprevail. s Our static theory specifies equilibria i.e.olutionsin the senseof 4.5.3. which aresetsof imputations. A dynamic theory when one is found the of will probablydescribe changesin terms of simplerconcepts: a single valid at the moment under consideration or something imputation of similar. Thisindicatesthat the formal structure this part of the theory betweenstatics and dynamics may be generically different the relationship 4 theories. from that of the classical physical All these considerationsillustrate once more what a complexity of in f theoreticalorms must be expected social theory. Our static analysis of the and alone necessitated creation a conceptual formal mechanism which i is very different from anything used, for instance,n mathematical physics. Thus the conventional view of a solution as a uniquely defined number or of aggregate numberswas seento be too narrow for our purposes,in spite in of its success other fields. The emphasis on mathematical methods and settheory and away to be shifted more towardscombinatorics seems of differential equationswhich dominate mathematical from the algorithm
physics.
even if they are sometimes called theory dealsalsowith inequilibria equilibria.\" 1Theabovediscussion of statics versus dynamics is, of course, ot at all a construction n ad hoc. The readerwho is familiar with mechanics for instance will recognizein it a m reformulation of well known features of the classical echanical theory of statics and dynamics. What we do claim at this time is that this is a general characteristic of and changesin structures. scientific procedure involving forces 3 Thedynamic concepts hich enter into the discussion of static equilibria are parallel w mechanics. Thereader to the \"virtual displacements \" in classical may alsoremember at 1 this point the remarks about \"virtual existence in 4 Particularly from classical echanics. The analogies of the type used in footnote 2 m above,ceaseat this point.)))
\"dynamic

1Thedynamic

' 4.3.3.

CHAPTER II OF GENERAL FORMAL DESCRIPTION GAMESOF STRATEGY

of discussions Chapter I that a theory It of the foundations of economics of the main and of rational behavior o mechanisms f socialorganization requiresa thorough study of the \" games we of strategy/' Consequently must now takeup the theory of gamesas an independent subject. In studying it as a problem in its own right, our shift. In Chapter I our point of view must of necessityundergoa serious It primary interest lay in economics. was only after having convinced ourselvesof the impossibilityof making progressin that field without a previous fundamental understanding of the games that we gradually approached the formulations and the questions which are partial to that v subject. But the economiciewpoints remainedneverthelessthe dominant ones in all of Chapter I. From this Chapter II on, however, we shall have to treat the games as games. Therefore we shall not mind if some points c taken up have no economiconnections whatever, it would not be possible to do full justiceto the subjectotherwise. Of coursemost of the main a of l concepts restill those familiar from the discussions economiciterature but the details will often be altogether alien to it (cf. the next section) a and details, s usual, may dominate the expositionand overshadowthe
from the shouldbe clear

5.1.

5.1. Shift

5.Introduction
of Emphasis
from

Economicsto Games

i.e.

guiding

principles.

I An important viewpoint in classifying games is this: s the sum of all o by payments received all players (at the end of the game)always zero; r is this not the case? If it is zero, then one can say that the playerspay only to eachother,and that no productionor destruction of goodsis involved. are All gameswhich areactually playedfor entertainment of this type. But a the economically significant schemes remost essentiallynot such. There the sum of all payments, the total social product, will in generalnot be
46)))

o prominent in the last sections f Chapter I will not appear in the beginning stages of the discussionswhich we are now undertaking. Specifically: There will be at first no mention of coalitionsbetween players and the these, cf. compensationswhich they pay to each other. (Concerning w 4.3.3., 4.3.2., in Chapter I.) We give a brief accountof the reasons, hich of will alsothrow somelight on our general isposition the subject. d

5.2. eneralPrinciples of Classification and of Procedure G Certain aspectsof \" games of strategy\" which were already 5.2.1.

INTRODUCTION

47

i zero, and not even constant. I.e.,t will depend on the behavior of the players the participants in the socialeconomy. This distinction was particularly in footnote 2, p.34. We shall call already mentioned in a gamesof the first-mentioned type zero-sumgames, nd those of the latter type non-zero-sumgames. We shall primarily constructa theory of the zero-sum games,but it will be found possibleto dispose,with its help, of all games,without restriction. in We Precisely: shall show that the general(hence particular the variable game. sum) n-person game can be reduced to a zero-sumn + 1-person Now (Cf. 56.2.2.) the theory of the zero-sumr?-persongame will be based Hence of on the specialcase the zero-sum two-persongame. (Cf. 25.2.) w will begin with a theory of thesegames, hich will indeed our discussions out be carried in Chapter III. Now in zero-sum two-person games coalitions and compensations 1 can play no role. The questionswhich are essentialin thesegames are How does each of a different nature. These are the main problems: o h his course i.e. ow does one formulate an exact concept f a player plan What information is available to each player at every stage strategy? of the game? What is the roleof a playerbeing informed about the other player'sstrategy? About the entire theory of the game? All 5.2.2. these questionsare of courseessentialin all games,for any i h number of players,even when coalitions and compensations ave comento their own. But for zero-sum two-persongames they are the only ones which matter, as our subsequentdiscussionswill show. Again, all these but as questionshave beenrecognized important in economics, we think that of games they arise in a more elementary as distinguished in the theory from composite fashion. They can, therefore, be discussedin a precise way and as we hope to show be disposedof. But in the processof this analysisit will be technically advantageous to rely on picturesand examples from the field of economicsroper, and belong which are rather remote p strictly to the field of games of the conventional variety. Thus the discussions which follow will be dominated by illustrations from Chess, \" a B P Matching Pennies,\" oker, ridge, etc., nd not from the structure of

4.2.1.,

etc. markets,oligopolies, cartels,

At this

t point it is also opportune to recallhat we considerall transt actions at the end of a game as purely monetary ones i.e.hat we ascribe to all playersan exclusively monetary profit motive. The meaning of this in Chapter I. Forthe w in terms of the utility concept as analyzed in \" present particularly for the zero-sum two-persongames\"to be discussed

2.1.1.

1The only fully satisfactory \"proof\" of this assertion lies in the construction of a completetheory of all zero-sum two-person games, without use of those devices. This result being contained in 17. It ought to be the decisive will be done in Chapter clearby common sense,however, that \"understandings\" and \"coalitions\" can have no role here:Any such arrangement must involve at leasttwo players hencein this caseall th re are no opponents for whom the sum of payments is identically zero. players

III,

I.e.

left and no

o possiblebjectives.)))

48
first

DESCRIPTIONF GAMES OF STRATEGY O

task is to give an exact efinition of what constitutesa d as the conceptof a game has not been described with long game. absolute mathematical combinatorial precision, we cannot hope to and exhaustive answersto the questionsformulated at the end give exact Now while our first objectiveis as was explainedin 5.2.1. of 5.2.1. the of zero-sum two-persongames,it is apparent that the exact escriptheory d to tion of what constitutesa game neednot be restricted this case. Consequently we can begin with the descriptionof the generaln-persongame. In giving this descriptionwe shall endeavorto do justiceto all conceivable w nuancesand complications hich can arisein a game insofar as they are not of an obviously inessentialcharacter.In this way we reach in several successive steps a rather complicatedbut exhaustive and mathematically scheme.And then we shall seethat it is possibleto replacehis t precise generalschemeby a vastly simpler one, which is neverthelessfully and t rigorously equivalent to it. Besides,he mathematical device which this simplification is also of an immediate significance for our permits It o problem: is the introduction of the exactconcept f a strategy. understoodthat the detour which leads to the ultimate, It shouldbe simple formulation of the problem, over considerablymore complicated ones is not avoidable. It is necessary to show first that all possible complicationshave been taken into consideration,and that the mathethe matical devicein question does guarantee equivalence of the involved setup to the simple. be done for all games,of any number of playAll this can and must ers. But after this aim has been achieved in entire generality, the next objectiveof the theory is as mentioned above to find a completesolution for the zero-sum two-persongame. Accordingly, this chapter will deal b with all games, ut the next one with zero-sumtwo-person gamesonly. After they aredisposedof and someimportant exampleshave been discussed,we o the shall begin to re-extend scope f the investigation first to zero-sum na person games, nd then to all games. w Coalitionsand compensations ill only reappear during this latterstage.
As

it (cf. the discussionof 5.2.1.) is an absolutely necessarysimplifiwe cation. Indeed, shall maintain it through most of the theory, although i variants will be examinedlateron. (Cf. Chapter XII,n particular 66.)

Our 6.2.3. first

6.The SimplifiedConceptof a Game 6.1. Explanation of the Termini Technici

be given, we must first clarify the use of some termini. Thereare some notions which are quite fundamental for the discussionof games, but the use of which in everyday languages highly ambiguous. Thewords i which describe them areusedsometimes one sense, ometimes another, in s in and occasionally worst of all as if they were synonyms. We must)))
can

6.1. eforean exactdefinition of the combinatorial conceptof a game B

SIMPLIFIED CONCEPT F A GAME O

49

therefore introduce a definite usage of termini technici, and rigidly adhere to it in all that follows. First,one must distinguish between the abstract conceptof a game, and the individual plays of that game. The game is simply the totality of the rules which describet. Every particular instance at which the i is played in a particular way from beginning to end,is a play.1 game t Second,he correspondingdistinctionshould be made for the moves, which are the componentelements the game. A move is the occasion of of a choice between various alternatives, to be made eitherby one of the players, or by some device subjectto chance,under conditionsprecisely prescribedby the rules of the game. Themove is nothing but this abstract \" a component with occasion/' the attendant details of description, of the game. The specificalternative chosenin a concrete instance in a concrete play is the choice. Thus the moves are relatedto the choices the same way as the game is to the play. The game consists in 2 of a sequence f moves, and the play of a sequence f choices. o o the rules of the game shouldnot be confusedwith the strategies Finally, of the players. Exactdefinitions will be given subsequently, but the distinction which we stress must be clearfrom the start. Each player h selectsis strategy i.e.he general rinciplesgoverning his choices freely. t p While any particular strategy may be good or bad provided that these and can concepts be interpreted in an exactsense(cf. 14.5. 17.8-17.10.) it is within the player's discretionto use or to reject it. The rules of the game,however, are absolutecommands. If they are ever infringed, then to the whole transaction by definition ceases be the game described those by 3 i rules. In many casest is even physicallyimpossibleto violate them.

i.e.

i.e.

Let us now considera game F of n players who, for the sake of , n. The conventional pictureprovides brevity, will be denotedby that this game is a sequencef moves, and we assumethat both the number o o and the arrangement f thesemoves is given ab initio. We shall seelater that these restrictions are not really significant, and that they can be removed without difficulty. For the present let us denote the (fixed) The moves number of moves in F by v this is an integerv = 2, themselveswe denoteby 3TCi, , 3TC,, and we assume that this is the order chronological in which they areprescribedto takeplace.

6.2.1.

The 6.2. Elements

of the Game

1,

1,

1In most games everyday usage callsa play equally a game ; thus in chess,in poker, to In Bridge a play corresponds a \"rubber,\" in Tennis to a \"set,\" in many sports, but unluckily in thesegames certain components of the play are again called\"games.\" The French terihinology is tolerably unambiguous: \"game\" * \"jeu,\" \"play\" \"partie.\" 1In this sensewe would talk in chessof the first move, and of the choice \"E2-E4.\" 1 In Chess rules of the game forbid a player to move his king into a position the in of \"check.\" This is a prohibition in the sameabsolute sense which he may not move a pawn sideways. But to move the king into a position where the opponent can \"checkmate\" him at the next move is merely unwise, but not forbidden.)))

etc.

: E.g.

50 Every

alternatives, among which the choice which constitutes the move 9TC, takes place. Denote the number of these alternatives by a< and the alternatives themselvesby GL (l), , The moves are of two kinds. A move of the first kind, or a personal made by a specific move, is a choice player, i.e.ependingon his free decision d and nothing else. A move of the secondkind, or a chance move, is a choice fortuitous dependingon somemechanical device,which makesits outcome 1 with definite probabilities. Thus for every personal move it must be determines move, whose move it is. specifiedwhich player's decision this We denote the playerin question (i.e. number)by k . Sofc, = 1, his , n. For a chancemove we put (conventionally) = 0. In this casethe probabilitiesof the various alternatives , &(<*)must be given. 2 We denotetheseprobabilitiesby p(l), , p(a)respectively. 6.2.2. a move 3TC, the choiceconsists of selectingan alternative In i (*(!), ' , Ct(<O>i.e.ts number We denotethe number so chosenby <r,. Thus this choiceis characterized y a number <r = 1, b , a,. And the completelay is describedy specifyingall choices, p b to it corresponding all moves OTli, , 9R,. I.e. is described a sequence by
K

DESCRIPTIONF GAMES OF STRATEGY O move 3TC,, = 1, , v, actually consists of a number of

K

&()

!,,.
K

a(l),

fc<

Now the rule of the game T must specifywhat the outcome the play of , n, if the play is described a given sequence by what payments every player receives ov I.e. <TI, when the play is D completed. enotethe payment to the player k by $k ($k > if k receives a payment, $ < if he must make one, $* = if neitheris the case). k Thus each$* must be given as a function of the <TI, , <r,:

is for eachplayerk =

!,-

$k

SF*(<ri,

, <r v),

1,

n.

$k(*i)

, <r,, is no part of the game T. It is, as we pointed out above, the definition of a play. 1E.g., shuffled deck,throwing dice, etc. It is dealing cardsfrom an appropriately even possibleo include certain gamesof strength and skill, where \"strategy\" plays a role, t e.g. ennis, Football, etc. In thesethe actions of the players are up to a certain point T personal moves i.e. dependent upon their free decision and beyond this point chance moves, the probabilities being characteristics the player in question. of Sincethe p(l), , p (a ) are probabilities, they are necessarilynumbers *z 0. Sincethey belong to disjunct but exhaustive alternatives, their sum (for a fixed K) must be one. I.e.
K K

' emphasizeagain thatfunction, 3 i.e.he abstractdependencef each ' ' , only as a the rules of tthe gameT specifythe function o 5u on the variables <TI, . But all the time eachcr, is a variable, , with the domain of variability A specification particular of numerical values for the i.e.he selectionf a particular sequence t o <n,
We
<r>) <r <r<,

!,-,,.

3 For

a systematic

exposition

of the conceptof a function cf. 13.1.)))

OF SIMPLIFIED CONCEPT A GAME

51

Our descriptionof the game F is not yet complete.We have failed to include specificationsabout the state of information of every w player at each decisionwhich he has to make, i.e. henever a personal move turns up which is his move. Therefore we now turn to this aspect of the matter. This discussion best conductedby followingthe moves 9fTli, is , 2fTl,, as the corresponding choices remade. a Let us therefore fix our attention on a particular move 3fH. If this is the 3TC< is a chance move, then nothing more needbesaid: choice decided will and nobody's knowledgeof other things can n by chance; obody's influence it. But if $TC,is a personalmove, belongingto the player fc, then it is quite important what k 'a state of information is when he forms his decisionconcerningO, i.e. is choice <r. h of T The only things he can beinformed about arethe choices orresponding c h to the moves preceding 9fR the moves SfTCi, , 9fTC_i. I.e. e may know the values of <TI, <r But he need not know that much. It is an , , (TI, important peculiarityof F, just how much information concerning We <r the player k is granted, when he is called upon to choose shall soon show in severalexamples what the nature of such limitations is. k The simplesttype of rule which describes stateof information at 3fn, a is this: set A consisting somenumbersfrom among X = of ,K is given. It is specifiedthat k knows the values of the a\\ with X belongwith any otherX. ing to A, and that he is entirely ignorant of the In this casewe shall say, when X belongsto A,, that X is preliminary to /c. This impliesX = X < K, but neednot be implied ,K it. Or, if we consider, instead of X, K, the corresponding moves SfTlx, 3Tl: by 1 but neednot be impliedby it. Preliminarity impliesanteriority, In spite of its somewhat restrictive character, conceptof this deservesa closernspection. In itself, and in its relationship preliminarity i to anteriority (cf. footnote 1 above), it gives occasion various combinato torial possibilities. Thesehave definite meaningsin thosegamesin which of they occur,and we shall now illustrate them by some examples particuinstances. larly characteristic
K K

6.3.1.

Information 6.3.

and Preliminarity

-.i.

-i

<? K

's
<r\\

1,

1,

1,

1,i.e.

6.3.2.

6.4.Preliminarity, Transitivity, and Signaling We 6.4.1. begin by observing that there existgames in which prew liminarity and anteriority are the same thing. I.e., here the players k m who makes the (personal) ove 9TC, is informed about the outcome of the o choicesf all anterior moves 3Tli, , 9Tl<_ i. Chessis a typicalrepresentaof tive of this class games of \"perfect\" information. They are generally to considered be of a particularly rational character.We shall seein 15., in how specifically 15.7., this can be interpreted in a precise ay. w 1 In time, X < K means that 9H\\ occursbeforeWL
K * K
.)))

52

is possible to conserve the first-mentioned property the equivalenceof moves. preliminarity and anteriority even in gameswhich contain chance o Backgammonis an examplef this.1 Somedoubt might be entertained

DESCRIPTIONF GAMES OF STRATEGY O Chesshas the further feature that all its moves are personal. Now it

not of moves does vitiate the \" rational charwhether the presence chance w o acter\" f the gamementioned in connectionith the preceding xamples. e in that this is not so if a very plausibleinterpretation We shall see i of that \" rational character\"s adhered to. It is not important whether f all moves are personal or not; the essential act is that preliminarity and anteriority coincide. 6.4.2. us now consider games where anteriority doesnot imply Let m where the player who makesthe (personal) ove 9TC, preliminarity. is not informed about everything that happened previously. There is a large family of gamesin which this occurs. Thesegamesusually contain chancemoves as well as personal moves. General opinion considers w them as being of a mixed character: hile their outcomeis definitely dependent on chance,they are also strongly influenced by the strategic abilitiesof the players. and Poker Bridgearegoodexamples.Thesetwo gamesshow,furtherwhat peculiar features the notion of preliminarity can present, more, onceit has beenseparated from anteriority. This point perhaps deserves a little more detailed consideration. the chronological ordering of the moves, possesses Anteriority, the property of transitivity. 2 Now in the present case,preliminarity n a need not be transitive. Indeedit is neither in Poker or in Bridge, nd the conditionsunder which this occurs quite characteristic. are L Poker: et 3TC be the deal of his \"hand\" to player 1 a chancemove; the first (subsequent) 3Tlx the first bid of player 1 a personalmove of 1 ; bid of player 2 a personalmove of 2. Then and is preliminary to to 3fll but is not preliminary to 3TC,.3 Thus we have intransitivity, 3Tlx i but it involves both players. Indeed, t may first seemunlikely that could in any game be intransitive among the personalmoves preliminarity of one particularplayer. It would require that this player should \"forget\" and 3TC, the outcome the choice of with betweenthe moves connected and it is difficult to seehow this \"forgetting\" couldbe achieved,and 3TV

15.7.1.

I.e.,

fc

i.e.
M

9fTC

3flfl M

9fTlx

2fTl M

9fTl\\

1Thechance moves in Backgammon are the dicethrows which decide total numthe ber of stepsby which eachplayer'smen may alternately advance. The personal moves are the decisionsy which eachplayer partitions that total number of stepsallotted to b

him among his individual men. Also his decisionto double the risk, and his alternative to acceptor to give up when the opponent doubles. At every move, however, the outcomeof the choices f all anterior moves are visible to all on the board. o 2 If 3TC M is anterior to 9Tlx and 311 to 9K then 3TC M is anterior to 9TC. Special ituas x tions where the presence absence transitivity was of importance, were analyzed in or of of Chapter in connection with the relation of domination. 3 1 makes his first bid knowing his own \"hand\"; 2 makes his first bid knowing \"hand.\" (preceding) first bid; but at the same time 2 is ignorant of 4 We assumethat 9TC is M preliminary to 9fR\\ and 3Tlxto 3TZ* but 9TC M not to

I.e. : 4.6.2. 4.4.2., I.e., 1's

1's

3TC*.)))

OF SIMPLIFIED CONCEPT A GAME just this.

53

even enforced1 Neverthelessour next exampleprovides an instance of

i as A A,B,C,D,t should be classified a two-persongame. Indeed, and C form a combination which
A Bridge: lthough

Bridge is played by 4 persons, to be denoted by

is more than a voluntary coalition, and so do B and D. ForA to cooperate B (or D) instead of with C would be with \" in in to cheating,\" the same sense which it would be \"cheating\" lookinto i JB's cards or failing to follow suit during the play. I.e.t would be a violation of the rules of the game. If three (or more)personsplay poker, then for it is perfectly permissible two (or more) of them to cooperate against another player when their interests are parallel but in Bridge A and C while A and B areforbiddento (and similarly B and D) must cooperate, this cooperate.The natural way to describe consistsin declaringthat A and that B and D are really one player 2. and C are really one player 1, Or, equivalently:Bridgeis a two-persongame,but the two players 1and 2 A do not play it themselves. 1actsthrough two representatives and C and 2 through two representativesB and D. of A Considernow the representatives 1, and C. Therulesof the game of t restrictcommunication, i.e.he exchange information, between them. let E.g.: 9TC be the deal of his \"hand\" to A a chancemove; 9TCx the first card played by A a personalmove of 1; 3TI, the card playedinto this trick by C a personalmove of 1. Then 9TI is preliminary to 3fllx and 2fllx to 9TC, 1 but 2(TI is not preliminary to 9TC,. Thus we have again intransitivity, but this time it involves only one player. It is worth noting how the necessary \"forgetting\" of 3TI between 3Tlx and 3Tl was achieved by \"splittingthe personality\"of 1into A and C. The 6.4.3. above examplesshow that intransitivity of the relation of preliminarity correspondsto a very well known componentof practical I is strategy: to the possibility of \"signaling.\"f no knowledgeof at available at 3fTC,but if it is possibleto observe9Hx'soutcome 3TC,and 3Tlx t has been influenced by (by knowledgeabout 9TC/S outcome),hen OTx is really a signal from 9TI to 3TC, a device which (indirectly)relays to information. Now two oppositesituations develop,according whether and 9TC,aremoves of the same player, or of two different players. In the first case which, as we saw, occursin Bridge the interestof t the player (who is x = fc,).lies in promoting the \"signaling,\" i.e. he of information \"within his own organization.\" This desire spreading finds its realization in the elaborateystem of \"conventional signals\"in s 2 These areparts of the strategy, and not of the rules of the game Bridge.
M M M M 3fTC M 9fTl M M 3flflx fc

1 A plays his first cardknowing his own \"hand\"; C contributes to this trick know1 ing the (initiating) cardplayed by A ; but at the sametime Cis ignorant of A's \"hand/ 1Observe to that this \" signaling \" is considered be perfectly fair in Bridge if it is it is correct carriedout by actions which are provided for by the rules of the game. cf. \" to agree beforethe play begins! for A and C (the two components of player \" indicates a weakness of the other suits. But that an \"original bid\" of two trumps 11 to indicate a weakness by an inflection of the voiceat it is incorrect \"cheating bidding, or by tapping on the table, etc.)))

I.e.

1, 6.4.2.)

E.g.

i.e.

54 (cf.

O DESCRIPTIONF GAMES OF STRATEGY and 6.1.), consequentlythey

may vary, 1 while the game of

In the secondcase which, as we saw, occursin Poker the interest of the player (we now mean k\\, observethat herek\\ ^ k ) lies in preventing this \"signaling,\" the spreading of information to the opponent (&<). This is usually achieved by irregular and seemingly illogical behavior at (when making the choice 3Tlx) this makes it harder for the opponent to draw inferencesfrom the outcome of the (which he sees)concerning outcome 91Z (of which he has no direct news). I.e.his procedure akes of t m the \"signal\" uncertain and ambiguous. We shall seein that this is 2 indeed the function of \"bluffing\" in Poker. direct and inverted signaling. It ought We shall call thesetwo procedures to be added that inverted signaling i.e. isleadingthe opponent occurs m in almost all games, ncluding Bridge. This is so sinceit is based on the i intransitivity of preliminarity when several players are involved, which is P s easy to achieve. Direct ignaling,on the other hand, is rarer;e.g. oker as contains no vestigeof it. Indeed, we pointed out before,it impliesthe it intransitivity of preliminarity when only one player is involved \" a well-regulated forgetfulness\" of that player, which is obtainedin requires Bridge the deviceof \"splitting the playerup\" into two persons. by At any rate Bridge and Pokerseemto be reasonably characteristic of directand of inverted instances of these two kinds of intransitivity

remainsthe same.

Bridge

i.e.

3fTlx

19.2.1.

i.e.

signaling,respectively. Both kinds of signalinglead to a delicateroblemof balancingin actual p o in \"rational\" laying. playing, i.e. the processf trying to define \"good,\" p than \"unsophisticated\" Any attempt to signalmore or to signal less playing i this is usually possibleonly at a definite cost,i.e.ts direct playing. T are s consequences losses. hus the problemis to adjust this \"extra\"ignaling so that its advantages by forwarding or by withholding information w overbalancethe losses hich it causes irectly. Onefeelsthat this involves d for somethinglike thesearch an optimum, although it is by no meansclearly defined. We shall seehow the theory of the two-persongametakes care already of this problem,and we shall discussit exhaustively in one characteristic instance. (Thisis a simplifiedform of Poker. Cf. 19.) o Let us observe, finally, that all important examplesf intransitive moves. Thisis peculiar, because containing chance preliminarity aregames 34 between these two phenomena. Our there is no apparent connection
And
-

would involve,

deviations from necessitates

the \"unsophisticated\" of way

they must agree. 1And that \"bluffing\" is not at all an attempt to secureextra gains in any direct sense when holding a weak hand. Cf.loc.cit. * Cf the corresponding question when preliminarity coincides ith anteriority, and w in As mentioned there, the presence absence or thus is transitive, as discussed of chancemoves is immaterial in that case. 4 \"Matching pennies\" is an example which has a certain importance in this connection. This and other relatedgames will be diacussedin 18.)))

1They may even be different for the two players, for A and C on one hand and B and D on the other. But \"within the organization' 1 of one player, e.g.or A and C, f

i.e.

6.4.1.

(COMPLETECONCEPT OF A GAME

55

o o subsequentanalysiswill indeedshow that the presencer absence f chance moves scarcely influences the essential aspects of the strategies in this situation.

7. The CompleteConceptof a Game

7.1.ariability V

of the

Characteristicsof Each Move

the a* alternatives &*(!), We introduced in , Gt<(a) of the move STl*. Also the index which characterizedthe move as a one,and in the first casethe player whose move it is; personalor chance the and in the secondcase probabilities *(l), , P (<X ) of the above alterp the conceptof preliminarity with the help natives. We describedin of the sets A,, this beingthe set of all X (from among the X = ,* which are preliminary to K. We failed to specify,however, whether all these objects a*, k f , A and the Ct(<r), p*(a) for o- = , a* depend on K or also on other things. These\" other things \" can, of course, solely to c only be the outcome of the choices orresponding the moves which are t anterior to 3fn. I.e.he numbers <n, <r_i. (Cf. 6.2.2.) , This dependence requiresa more detaileddiscussion. First,the dependenceof the alternatives (%((?) themselves(as distinon <n, , o-,_i is immaterial. We may guishedfrom their number to as well assumethat the choice corresponding the move 311,is made not betweenthe Ct,(<r) themselves,but betweentheir numbersa. Infine, it is the c d only the <r of 311*,i.e.r, which occursin the expressions escribing outk = i of the play, i.e.n the functions SF*(<M, come , cr*), , n.1
fc* * *

7.1.1.

6.2.1.

6.3.1.

1,

!)

1,

a!)

a Second, ll dependences(on en, w move i.e. hen out to be a chance

(Cf. 6.2.2.)

1,

causeno (cf. the end of do not interfere with our analysisof the behavior of complications. They the players. This disposes,in particular, of all probabilitiesp(<r), since moves. (TheA,, on the other they occuronly in connection with chance moves.) hand, never occurin chance , <r_i) of the Third, we must considerthe dependences(on <n, k , A, when 2KI, turns out to be a personalmove.2 Now this possibility 3 is indeeda sourceof complications. And it is a very real possibility. The
kK
K

(7<_i)

which

arise when

6.2.1.)

3TI* turns

reasonis this.
to
<TX,

1 Theform and nature of the alternatives GLK (<r) offered at 3TI,might, of course,onvey c the player kK (if WL K is a personal move) some information concerning the anterior , <r_i values, if the a(<r) depend on those. But any such information should be specifiedseparately, as information available to A; at 2fTl*. We have disfcussed the and shall completethe c simplest schemesoncerning the subject of information in The discussion of , kK, A*, which follows further below, is characdiscussion in o s teristic also as far as the role of the a,t(<r) as possibleources f information is concerned. 8 Whether this happens for a given K, will itself depend on kK and hence indirectly on <n, , <7-i sinceit is characterizedby kK & (cf.the end of 3 In Chess number of possiblelternatives the at 9R K depends n the positions o a the previous course the play. In Bridge the player who plays the of of the men,

7.1.2.

6.3.1.,

E.g.: i.e.

6.2.1.).

first)))

56
A*

O DESCRIPTIONF GAMES OF STRATEGY

to seehow conflicts can be avoided. To be precise: Thereis no conflict in this special case:Let A, be independentof all (T\\, - - - , <7\\_], and let , k dependonly on the with X in A*. Then the player k can certainly not get any information from a , k , A the beyond what he knowsanyhow (i.e. values of the with X in A*). If w this is the case, e say that we have the special orm of dependence. f But do we always have the specialorm of dependence?To take an f c W k expected be completely extremease: hat if A is always empty to uninformed at 3TC, and yet e.g.* explicitlydependent on some of the a , <r._i! <n, This is clearly inadmissible. We must demand that all numerical conclusionswhich can be derived from the knowledgeof a*, & A , must be explicitlyand ab initio specifiedas information available to the player k at 3TC,. It would be erroneous, however, to try to achieve this by including in A* the indicesX of all theseo-x, on which fc,, A, explicitlydepend. In caremust be exercised orderto avoid circularity in in the first place great 1 this requirement,as far as A is concerned. But even if this difficulty does not arise, ecause dependsonly on K and not on cri, b if A, , o-^i i.e. the information available to every player at every moment is independentof of the previous course the play the above procedure ay still be inadmism sible. Assume, e.g.,hat a depends on a certain combination of some t from among the X = 1,and that the rules of the gamedo indeed providethat the player k at y(l should know the value of this comthe bination, but that it doesnot allow him to know more (i.e. values of the individual ai, Hemay know the value of cr + where , ov-i). M, X are both anteriorto K (ju, X < K), but he is not allowed to know the v separatealues of and a\\. One could try various tricks to bring back the above situation to our w fc/s simpler,scheme, hich describes stateof information by means earlier, the set A,. 2 But it becomes of completelyimpossibleto disentanglethe various componentsof fc/s information at 3TI,, if they themselvesoriginate from personal moves of different players, or of the same player but in
K K
<r\\

sincethesearenow part of the rulesof the game which he must observe. Insofaras they depend upon ai, , <r_i, he may draw from them certain conclusionsconcerninghe values of <TI, t , <7_i. But he is supposed to know absolutelynothing concerning the with X not in A ! It is hard
<r\\

The 7.1.2. player k

must

be informed at 311,of the values of

a,,k ,
K

<r\\

i.e.

a,,

l,

-,*
K

<r\\

E.g.:

<r\\

<T

kK at 911*, is the one who took the last trick, next trick, again dependent of upon the previous course the play. In some forms of Poker, and some other related A* at 9HK , games,the amount of information available to a player at a given moment, on depends what he and the others did previously. 1The on which, among others, A, depend are only defined if the totality of all A, - , <r<_i, is considered. Should every A* contain theseX? for all sequences <n, 2 In the above the example one might try to replace move 9TIM by a new one in which not o> is chosen, but -f *\\. SHI* would remain unchanged. Then k* at 9R K would be informed about the outcome of the choice connected the new 3TCu onlv.))) with

card to the

i.e.

i.e.

i.e.

<r\\

<?

COMPLETE CONCEPT OF
fc M fc

GAME

57

different stages of information. In our above examplehis happens if t 5^ fc\\, or if fc^ = x but the stateof information of this player is not the 1 same at 311,, nd at a
SfTCx.

7.2.1. arestill various, more or lessartificial, tricks by which one There couldtry to circumvent thesedifficulties. But the most natural procedure t seemso be to admit them, and to modify our definitions accordingly. This is done by sacrificing the A, as a means of describingthe stateof information. Instead, describe stateof information of the player fc, we the at the time of his personalmove 311, xplicitly By enumerating thosefunc: e tions of the variable anterior to this move i.e. f the <TI, o , <r_i the numerical values of which he is supposedto know at this moment. This is a systemof functions, to be denotedby <,. So$,is a setof functions))
<r\\

7.2.TheGeneralDescription

the o Since elementsf

<

is fixed, i.e. ependingon K only.2 d

on describe dependence <TI, the

, k may
K K

, <r_i,

sincetheir values areknown to k at

must

91Z,,

thesefunctions))

dependon <n,

so <,itself
, <r_i, and

Of course,whenever it turns out that k = (for a is a chanceone (cf. set , cr_i values), then the move special of <ri, but this does matter. not above),and no use will bemade of Our previous mode of description,with the A*, is obviously a special caseof the present one,with the $*.8 At 7.2.2. this point the reader ay feel a certaindissatisfactionabout m has was the turn which the discussion taken. It is true that the discussion deflectedinto this direction by complicationswhich arosein actual and typical games (cf. footnote 3 on p. 55). But the necessityof replacing the A by the <$ originated in our desireto maintain absolute formal (mathematical) generality. These decisive difficulties, which caused us to take this step (discussedin particularly as illustrated by the I t footnotes there)werereally extrapolated. .e.hey werenot characteristic))

belong to

$.

3fTl

<!>*

7.1.2.,

1 In the instance of footnote 2 on p.56,this means If fr M ?* fc\\, there is no player to whom the new move 3!Z M (where <r M 4-<rx is chosen, and which ought to be personal) can be If & M = k\\ but the state of information varies from 3TI to $TCx, then no state attributed. M for of information can be satisfactorily prescribed the new move 9Tt M 1This arrangement includes nevertheless the possibility that the state of information all functions dependson <n, , <rK _\\. This is the case if, by expressed o h(vi, , <r_0 of show an explicit dependencen o>for one set of values of <r\\, while is fixed. being independent of o> for other values of <r\\. Yet 1If ** happens to consist of all functions of certain variables <r\\ say of those for which X belongs to a given set M* and of no others, then the $* description specializes But we have seenthat we cannot, in back to the A one: A, being the aboveset o general, count upon the existence f such a set.)))

e.g.,

<J>

M.

DESCRIPTIONF GAMES OF STRATEGY O of the original examples, which areactual games. (E.g. Chessand Bridge
58

by undue specialization. it w Besides, will be seenthat all the formal elements hich we areintronow into the discussiondo not complicate ultima analyst. I.e. it ducing they complicate only the present, preliminary stage of formal description. The final form of the problem turns out to be unaffected by them.
bility

But in Thereexist ames which require discussion means of the g by most of them one could revert to the A, by means of various extraneous tricks and the entiresubjectrequiresa ratherdelicatenalysisupon which a 1 it does not seem worth while to enterhere. Thereexist unquestionably 2 economic odelswhere the $ arenecessary. m The most important point, however, is this. In pursuit of the objectiveswhich we have setourselveswe must achieve in the certainty of having exhaustedall combinatorialpossibilities connection with the entireinterplay of the various decisionsof the players, their changing statesof information, etc. Theseareproblems,which have been dwelt upon extensivelyin economic literature. We hope to show that they can be disposed of completely. But for this reasonwe want to be safe of p against any possibleaccusation having overlookedsome essential ossi-

can be describedwith the help of the A*.)

$.

assumptionformulated at the very start of this discussion(at the beginning that of 6.2.1.) both the number and the arrangement the moves aregiven ab initio. We shall now seethat this restriction not essential. is (i.e. fixed) first Consider the \"arrangement\" the moves. The possiblevariaof full o has already received bility of the nature of eachmove i.e. f its in 7.2.1.). orderingof the moves SHI,, k = 1, considerationespecially The ( one. Thus there is , v, was from the start simply the chronological nothing left to discusson this score. Considernext the number of moves v. This quantity too could be variable, i.e. ependent upon the courseof the play.3 In describingthis d variability of v a certainamount of caremust beexercised.
of
fc

(Cf.11.2.) The There 7.2.3. remains only one more point to discuss: specializing

choices.) 1Let a participant be ignorant of the full details of the previous actionsof the others, but let him be informed concerning certain statistical resultants of those actions. 3 It is, too, in most games: Chess, ackgammon, Poker,Bridge. In the caseof Bridge B to this variability is due first to the variable length of the \"bidding\" phase,and second the changing number of contracts needed make a \"rubber\" (i.e. play). Examples a to of gameswith a fixed v are harder to find: we shall seethat we can make v fixed in every game by an artifice, but games in which v is ab initio fixed are apt to be monotonous.)))

t With referenceo that description: belongsin this class. (For its description cf. w Each player knows about the \" possibility\" of the other's anterior choices, ithout themselves and this \"possibility\" is a function of all anterior knowing those choices

1We mean card games where players may discard some cards without uncovering later. them, and are allowed to take up or otherwise use openly a part of their discards \" Thereexists alsoa game of double-blind Chess sometimes called\" Kriegsspiel which

9.2.3.

COMPLETECONCEPT OF

GAME

59

*i,' ' ' ,

<r Now one cannot state simply that v may be a (cf. 6.2.2.). function of the variables <TI, t , a, <TI, , <r,, becausehe full sequence cannot be visualized at all, without knowing beforehand what its length v 1 is going to be. Thecorrect formulation is this:magine that the variables I 2 arechosenone after the other. If this succession choices of 0*1,0*2,ffs, is carried on indefinitely, then the rules of the game must at some place v stop the procedure. Then v for which the stop occurs d will, of course, epend on all the choices p to that moment. It is the number of moves in that u particular play. Now this stop rule must be such as to give a certainty that every conceivable play will be stopped sometime. .e.it must be impossibleto I in such a manner (subject choices f a\\, <r 2, <r 3, o arrange the successive to the restrictions of footnote 2 above) that the stop never comes.The obvious way to guarantee this is to devise a stop rule for which it is certain that the stop will come t before a fixed moment, say *>*. I.e.hat - , it is sure to be v ^ v* where v* while v may depend on o-i, a*, a^ does not depend on <TI, <r 2, <r 3, If this is the casewe say that the rule is boundedby v*. We shall assumefor the gameswhich we constop sider that they have stop rules bounded by (suitable,but fixed) numbers
*

The courseof the play is characterizedby the sequence choices) (of

* 3,4

1 one cannot say that the length of the game depends all choices ade in conon m nection with all moves, sinceit will depend on the length of the game whether certain moves will occurat all. The argument is clearly circular. * The domain of of is , on. The domain of variability of <TI is variability 1, , at, and may depend on <n: a 2 = az(<ri). The domain of variability of is , ai, and may depend on <TI, z i ai(<n, <rt). Etc., 8 This stop rule is indeed an essential part of every game. In most games it is easy to find v's fixed upper bound v*. Sometimes, however, the conventional form of the rules of the game does not excludethat the play might under exceptionalconditions go on ad infinitum. In all these casespractical safeguards have been subsequently incorof porated into the rules of the game with the purpose of securing the existence the bound v*. It must be said, however, that these safeguards are not always absolutely effective although the intention is clearin every instance, and even where exceptional infinite plays exist they are of little practical importance. It is nevertheless quite instructive, at least from a mathematical point of view, to discussa few typical examples. We give four examples,arranged according to decreasingeffectiveness. ficarte\": A play is a \"rubber,\" a \"rubber\" consistsof winning two \"games\" out of three (cf. footnote 1 on p. 49), a \"game\" consistsof winning five \"points,\" and each a \"deal\" gives one player or the other one or two points. Hence \"rubber\" is complete after at most three \"games,\" a \"game\" after at most nine \"deals,\" and it is easy to v* 3 ^9 18 486. verify that a \"deal\" consistsof 13, 14or 18moves. Hence It is Poker:A priori two players could keep \"overbidding\" eachother ad infinitum. therefore customaiy to add to the rules a proviso limiting the permissible number of \"overbids.\" (Theamounts of the bids are also limited, so as to make the number of a secures finite v*. This of course alternatives a at thesepersonal moves finite.) Bridge:The play is a \"rubber\" and this could go on forever if both sides(players) failed to make their contract. It is not inconceivable that the sidewhich is in invariably danger of losing the \"rubber,\" should in this way permanently prevent a completion of b the play by absurdly high bids. This is not done in practice, ut there is nothing explicit in the rules of the game to prevent it. In theory, at any rate, some stop rule should be introduced in Bridge. of Chess:It is easy to construct sequences choices(in the usual terminology:)))
<r\\

I.e.

1,

1,

<r

etc.

<r

60
ity of v.

DESCRIPTIONF GAMES OF STRATEGY O

Now we can make useof this bound v* to get entirelyrid of the variabilv* moves 2fEi, , 9TI,*. For every sequenceTI, <r, 0-3, < everything is unchangedup to the move 3TI,,and all moves beyond 2fTl, are i \"dummy moves.\" I.e.f we considera move 3TC,, K = , v*, for a for which v < K, then we make 9TC a chance move o sequence-i, era, <TS, with one alternative only 1 one at which nothing happens. Thus the assumptions made at the beginning of particularly that v is given ab initio arejustified ex post.

are always

the of This is done simply by extending scheme the gameso that there

1,

i.e.

6.2.1.

8.Setsand Partitions 8.1. Desirability

It is worth
of

a Set-theoretical Descriptionof a Game

W 8.1. e have obtained a satisfactory and generaldescription of the with axiomatic precision conceptof a game,which could now berestated and rigidity to serveas a basis for the subsequentmathematical discussion.

tion. This formulation is exactly equivalent to the one which we reached b in the preceding sections, ut it is more unified, simpler when stated in a to and transparent notations. general form, and it leads more elegant In orderto arrive at this formulation we must use the symbolismof the theory of sets and more particularly of partitions more extensively a than we have doneso far. This necessitatescertainamount of explanation and illustration, which we now to proceed give.
\"moves\") particularly in the \"end game\" which can go on ad infinitum without ever ending the play producing a \"checkmate\.") The simplest onesare periodical, b indefinite repetitions of the samecycleof choices, ut there exist non-periodical onesas well. All of them offer a very real possibility for the player who is in danger of losing to sometimes a \" tie.\" For this reason various \"tie rules \" secure stop rules arein use just to prevent that phenomenon. Onewell known \"tie rule\" is this: Any cycleof choices \"moves\,") when three m times repeated,terminates the play by a \"tie.\" This rule excludes ost but not all and henceis really not effective. infinite sequences, Another \"tie rule\" is this: If no pawn has beenmoved and no officer taken (these are \"irreversible\" operations, which cannot be undone subsequently) for 40moves, then the play is terminated by a \"tie.\" It is easyto seethat this rule is effective, although the

while, however, before doingthat, to pass to a different formula-

(i.e.

i.e.

i.e.

(i.e.

v* is enormous. 4 From a purely mathematical point of view, the following question could be asked Let the stop rule be effective in this senseonly, that it is impossible so to arrange the c that the stop never comes. let there always be a successivehoicesTI, at, at, <
finite

directly. (Cf., however,

mathematically.

v dependent upon <n, <rj, <rj, Doesthis by itself securethe existence a of v*? such that all v fixed, finite v* bounding the stop rule? s The question is highly academicinceall practical game rules aim to establish a v*

. I.e.

I.e.

footnote

3 above.) It is nevertheless

quite

interesting

weiseaus dem

v* always exists. Cf. D.K&nig: tJber eine SchlussEndlichen ins Unendliche, Acta Litt. ac Scient.Univ. Szeged, Sect.Math. Vol. (1927) p. p particularly the Appendix, pp. 12&-130. 1This means, of course,hat = kK = 0, and p*(l) *= t

The answer is \".Yes,\"

i.e.

e.g.

III/II

121-130; 1,

1.)))

SETS AND PARTITIONS

61

to A. 1
We

t being placed on the nature and number of these objects,he elements of the set in question. Theelements constitute and determine set as the without any ordering relationshipof any kind betweenthem. I.e. or such, if two setsA, B aresuch that every element f A is also one of Band vice o A then they are identical in every respect, = B. The relationship versa, of a being an element f the setA is also expressed sayingthat a belongs o by

8.2.1.set is an arbitrary collectionf objects, A o absolutelyno restriction

8.2. Their Properties,and Sets,

Their Graphical

Representation

s i.e.etsconsistingof a finite number of elements.

shallbe interested chiefly, although not always,in finite setsonly,

We re-emphasize any objects an be elements a set. Of course that of c we shall restrictourselves to mathematical objects. ut the elements B for instance,perfectly well be setsthemselves(cf. footnote 3), thus can, e latteraresometimes by some called other leadingto setsof sets, tc. These o or aggregates f sets.But this is not s equivalent name, e.g.ystems

3 and should never be confused.

Given any objects , 0,7, we denotethe set of which they arethe a It is also convenient to introducea setwhich elements y (a, 0,7, b ) 2 a containsno elements t all, the empty set. We denotethe empty setby . We can,in particular, form setswith preciselyone element, ne-element ets. o s Theone-element (a), and its unique elementa, arenot the same thing set

necessary.

8.2.2. main The

(8:A:a)

a with c concepts nd operations onnected setsarethese: of A is a subset of B, or B a superset A, if every element f o B or B 2 A. A\\s A is also an element f B. In symbols:A o a proper subset of B,or B a propersupersetof A, if the above is which are not elements A. of true, but if B contains elements In symbols:A c B or B A. We see:f A is a subsetof B and I B is a subset of A , then A = B. (This is a restatement f the o
=>

Also: A is a principle formulated at the beginning of 8.2.1.) subset of B if and only if A is a subset of B without proper
A

= B.

literature of the theory of sets is very extensive. We make no beyond what will be said in the text. The interested reader will find more A. Fraenkel: Einleitung in die Meninformation on set theory in the good introduction: and technically excellent Hausdorff: Mengenc genlehre, 3rd Edit. Berlin 1928; oncise lehre, 2nd Edit. Leipzig 1927. * If two sets A, B are both without elements, then we may say that they have the there exists only one same elements. Hence,by what we said above, A B.
1

use of it

The mathematical

:F.

empty
8

This reasoning may sound odd, but it isneverthelessfaultless. Thereare some parts of mathematics where (a) and a can be identified. This is then occasionallydone, but it is an unsound practice. It is certainly not feasible in a l general. E.g.,et a be something which is definitely not a one-element set, Then (a) and a must be distinguished, since two-element set 0), or the empty set (a) is a one-elementset while a is not.)))

set.

I.e.

(,

Q.

i.e.

62
(8:A:b) (8:A:c) (8:A:d) (8:A:e)

(8:A:f) (8:A:g)

1 sums of more than two setsareformed. The product, or intersection, of two setsA, B is the set of all common elementsof A and of B, to be denoted by A n B. Similarlythe products of more than two setsareformed.l The differenceof two setsA, B (A the minuend, B the subtrao hend) is the setof all those elementsf A which do not belongto 1 B, to be denoted by A B. B the compleWhen B is a subset of A, we shallalsocall A it ment of B in A. Occasionally will be so obvious which set B and talk about the A is meant that we shall simply write complementof B without any further specifications. i Two setsA, B aredisjunct if they have no elementsn comif mon, i.e. A n B = . is A system (set) & of sets said to be a systemof pairwise diso junct setsif all pairs of different elements f Q,aredisjunct sets, i i.e.f for A, B belongingto a,A ^ B impliesA n B = . larly the

o with together all elements f B, to be denotedby A u B. Simi-

O DESCRIPTIONF GAMES OF STRATEGY The sum of two sets A, B is the set of all elements A of

point some graphicalillustrationsmay be helpful. o w the objects hich areelements f setsin theseconsiderations We denote the dots (elements))) We denote sets by encircling by dots (Figure

At 8.2.3. this

1).

Figure

1.

which belong to them, writing the symbol which denotesthe set across line the encircling in one or more places (Figure1). The setsA, C in this figure are,by the way, disjunct, while A, B arenot. 1This nomenclature of sums, products, differences, is traditional. It is based on algebraic analogies which we shall not use here. In fact, the algebra of these interest of its own. operations U, n, also known as Booleanalgebra, has a considerable N Cf.further Garrett Birkhoff: A Cf. e.g. . Tarski:Introduction to Logic, ew York, LatticeTheory, New York 1940. This book is of wider interest for the understanding of the modern abstract method. Chapt.VI. dealswith BooleanAlgebras. Further literature is given there.)))
certain

1941.

SETS AND PARTITIONS))

With

63))

this devicewe can also represent ums,productsand differencesof s neither A is a subset of B nor B one of A, henceneither the difference A B nor the difference JB A is a complement. In the next figure, however, B is a subset of A, and so A B is the 3).)) complementof B in A (Figure

sets(Figure2). In this figure

Figure

2.))

Figure

3.))

is a partition in

8.3.1. a set Let

ft if

8.3. Partitions,

a Their Properties, nd Their Graphical

ft

Representation

(8:B:a) (8:B:b)

the two following requirements: A Every element of is a subset of ft, and not empty. ft is a system of pairwisedisjunct sets.
it
fulfills

and a system of sets Ct be given. We say that a

1 t This conceptoo has been the subjectof an extensiveliterature. We say for two partitions ft, (B that a is a subpartition of (B, if they fulfill this condition: (8:B:c) Every elementA of a is a subset of some elementB of (B.2 Observethat if a is a subpartition of (B and (B a subpartitionof

(8 :B :d)
1

a,then a = (B.3 Next we define:

(8:B:a):

A intersections n B

Given two partitions

a, (B, we form the system of all those

over all elements f ft and B over o are (8:B:b) not exactly the (8:B:a),

A running

Cf. G. Birkhoff loc. cit. Our requirements customary ones. Precisely: Ad It is sometimes not required that the elements A of be not empty. Indeed,we shall have to make one exceptionin (cf.footnote 4 on p. 69). It is customary to require that the sum of all elementsof d be exactly Ad the set ft. It is more convenient for our purposesto omit this condition. 2 SinceCt, (B are also sets, it is appropriate to compare the subset relation (as far as w Oneverifies immediately that if Ct Ct, (B are concerned) ith the subpartition relation. is a subset of (B then Q, is alsoa subpartition of (B, but that the conversetatement is not s (generally) true. 3 a It must be subset of an element B of (B, and Proof:Considern element A of B in turn subset of an element Ai of Ct. So A, A\\ have common elements all thoseof the not empty set A are not disjunct. Sincethey both belong to the partition d, A A this necessitates *= AI. So A is a subsetof B and B oneof A ( = AI). Hence = J3, and thus A belongs to (B; is a subset of (B. (Cf.footnote 2 above.) Similarly (B is a subset of

(8:B:b):

9.1.3.

a.

i.e.

Hencea =

I.e.: a

a.

(B.)))

64))

of partition, the superposition Ot, (B. Finally, we also define the above relationsfor two partitionsCt, (B within a given set C. is a subpartition of (B within C, if every A belongingto ft (8:B:e) to which is a subset of C is also subset of someB belonging (B which is a subset of C. (8:B:f) Ot is equal to (B within C if the samesubsetsof C areelements of and of (B. Clearly footnote 3 on p. 63 applies again, mutatis mutandis. Also, w the above concepts ithin ft arethe sameas the original unqualified ones.))

O DESCRIPTIONF GAMES OF STRATEGY)) all those of (B which are not empty. This again is clearly a
1

Figure

4.))

.S3v))
8.3.2. give again somegraphicalillustrations,in the senseof 8.2.3. We
We
Figure

5.

of the partition clingline

We

begin by picturing a partition. We shall not give the elements which aresets names, but denoteeachone by an encir-

1It is easyto show that the superposition of (Bis a subpartition of both and (B and that every partition C which is a subpartition of both and (B is also one of their Hence name. Cf O.Birkhoff, loc.cit. Chapt.I-II.))) the superposition.

them by markingthe picture next two partitions a,(B distinguishing lines of the elements f a by o and of the elements f (B by o encircling

(Figure 4).

a,

SETS AND PARTITIONS))

65))

5). (Figure In this figure a is a subpartitionof (B. In the following oneneithera is a subpartition(B nor is & oneof a (Figure 6). We leave it
of t to the readero determinethe superposition Ot, (B in this figure.))

Figure

6.))

Figure

7.))

Figure

8.))

Figure

9.
a)))

o representationf partitions obtains by repreAnother, more schematic, the set 8 by one dot, and every element f the partition which is o senting

66

O DESCRIPTIONF GAMES OF STRATEGY

12

subset of

b Figure5 will be represented y a much simplerdrawing (Figure7). This d within the elements the of representation oesnot indicatethe elements s partition, and it cannot be used to represent everal partitions in 12 simultaneously, as was done in Figure 6. However, this deficiency can be as removed if the two partitions ft, (B in 12 arerelated in Figure5 : If Q is a subpartition of (B. In this casewe can represent12 again by a dot at the bottom, every elementof (B by a line going upward from this dot as in Figure7 and every elementof ft as another line going further upward, t o beginningat the upper end of that line of (B, which representshe element f o t (B of which this element f ft is a subset. Thus we can representhe two is partitions ft, (B of Figure5 (Figure8). This representation again less one revealing than the corresponding of Figure5. But its simplicitymakes it 4-6 it possibleto extend further than picturesin the vein of Figures could W of practically go. Specifically: e can picture by this device a sequence , ft , where eachone is a subpartition of its immediate partitions fti, = 5 (Figure with We 9). predecessor. give a typical example Configurations of this type have beenstudied in mathematics,and are
M

by a line going upwardfrom this dot. Thus the partition

a of

ju

known

as trees.

8.2.2.

8.4.LogisticInterpretation of Setsand Partitions in will be The 8.4.1. notionswhich we have described 8.2.1.-8.3.2.useful of in the discussion gameswhich follows, because the logisticinterpretaof tion which can be put upon them. s Let us begin with the interpretation concerningets. o If 12 is a set of objects f any kind, then every conceivableproperty which some of these objects may possess,and others not can be fully of characterizedy specifyingthe set of thoseelements 12 which have this b in i property. I.e.f two properties correspond this senseto the same set o elements f 12 will possessthesetwo subset of 12),then the same (the same they properties,i.e. areequivalent within 12, in the sensein which this term is understoodin logic. of Now the properties (of elements 12) are not only in this simple corwith respondence sets(subsetsof 12),but the elementarylogical operations involving propertiescorrespondto the set operationswhich we discussedin

Thus the disjunction of two properties the assertion that at least one of them holds corresponds obviously to forming the sum of their sets, the assertion the operation A u B. The conjunction of two properties to that both hold corresponds forming the product of their sets, the operthe assertion ation A n B. And finally, the negation of a property of the opposite correspondsto forming the complementof its set, the

i.e.

i.e. i.e.

operation A. 1
1Concerning

loc.cit. Chapt.VIII.)))

the connection

of set theory

and

of formal logic cf., e.g., Birkhoff, O.

SET-THEORETICAL ESCRIPTION D

67

Insteadof correlating subsetsof ft to propertiesin ft as done above the we may equallywell correlate them with all possible bodiesof information a o concerningn otherwiseundetermined element f ft. Indeed, such any information amounts to the assertion that this unknown elementof ft a possesses certain specified property. It is equivalently represented to of by the set of all those elements ft which possessthis property; which the given information has narrowed the range of possibilitiesfor the unknown element f ft. o to Observe,in particular, that the empty set corresponds a property which never occurs, .e.o an absurd information. And two disjunct sets i t correspondto two incompatibleproperties, to two mutually exclusive bodiesof information. 8.4.2. now turn our attention to partitions. We in and by restatthe By reconsidering definition (8:B:a), (8:B:b) A ing it in our present terminology, we see: partition is a systemof pairwise an mutually exclusivebodiesof information concerning unknown element of ft none of which is absurd in itself. In other words: A partition is a preliminary announcement which states how much information will be o to what element f ft; a given laterconcerningn otherwiseunknown w f extentthe range of possibilitiesor this element ill be narrowed later. But the actual information is not given by the partition, that would amount to i an o selecting element f the partition, sincesuch an elements a subset of ft, i.e. ctual information. a We can therefore say that a partition in ft is a pattern of information. that they correspondto definite As to the subsets of ft: we saw in informati6n. In order to avoid confusion with the terminology used for f partitions, we shall use in this case i.e.or a subset of ft the wordsactual information. it in and relate to our present Considernow the definition (8:B:c) This expresses two partitions d, (B in ft the meaning of ft for terminology. being a subpartition of (B: it amounts to the assertionthat the information announcedby d includesall the information announcedby (B (and possibly t more) ; i.e.hat the pattern of information includesthe pattern of information (B. in 4-9 Theseremarks put the significance of the Figures in 8.3.2. a new It appears,in particular,that the tree of Figure9 picturesa sequence light. of continually increasingpatterns of information.

i.e.

i.e.

8.3.1.,

i.e.

8.4.1.

8.3.1., a

We assumethe number of moves as we now know that we may to be fixed. Denotethis number again by v, and the moves themselves , SKI,. again by 3TCi, Considerall possibleplays of the game F, and form the set ft of which s they arethe elements.If we use the descriptionof the precedingections,)))

9.1.1.

9. The Set-theoretical Descriptionof a Game 9.1. Partitions Which Describea Game The

of by describingeach play as the sequence the v + 1 consecutive 8 positions which ariseduring its course. In general,of course,a given position may not be followed by an arbitrary position, but the positions which are possibleat a given moment arerestrictedy the previousposib 4 tions, in a way which must be preciselydescribedby the rules of the game. Sinceour descriptionof the rulesof the game beginsby forming ft, it may be undesirableto let ft itself dependso heavily on all the details of thoserules. to We observe,therefore, that thereis no objection including in ft absurd 6 even of sequences positionsas well. Thus it would be perfectly acceptable o to let ft consistof all sequencesf v -f 1successive positions,without any restrictionswhatsoever. Our subsequentdescriptionswill show how the really possibleplays from areto beselected this, possiblyredundant, set ft. v and 9.1.2. ft being given, we enterupon the more elaborateetails of d the course a play. of Consider a definite moment during this course,say that one which a immediately precedes given move 3TI,. At this moment the following m s general pecifications ust be furnished by the rulesof the game. to Firstit is necessaryto describe what extentthe events which have 6 led up to the move 9TC. have determinedthe courseof the play. Every o particular sequencef theseevents narrows the set ft down to a subset A : of this beingthe setof all those plays from ft, the course which is, up to 3^,, of the particular sequence events referred to. In the terminology of the the earliersections,ft is as pointed out in 9.1.1. set of all sequences - , a, for , a,}then A t would be the set of those sequencesn, <TI, <
form it
K

O DESCRIPTIONF GAMES OE STRATEGY l then all possible <n, playsaresimplyall possiblesequences , *. There 2 and so ft is a finite set. exist nly a finite number of such sequences, o Thereare,however, also more directways to form ft. We can, e.g.,
68

which the

<TI,

But from our presentbroader point of view we need only say that A must bea subset of ft. the Now the various possiblecourses game may have taken up to 3TC, if must berepresentedy different setsA . Any two such courses, they are b different from each other, initiate two entirely disjunct sets of plays; i.e. run no play can have begun (i.e. up to 9TC,) both ways at once. This means that any two different setsA x must be disjunct.
K K

<r,_i

have given numerical values (cf.footnote 6 above).

on

p.59. 1Verification I.e. I.e.

1Cf

The . in particular^ 6.2.2. range

by means

of

the

<TI,

in , <r v is described

footnote

cal values

9R 2 and 9flli, 3H,, after 3Hp. 4 This is similar to the development of the sequence <n, footnote 2 on p. 59. * oneswhich will ultimately be found to be disallowed rules of the game. c with the anterior moves 9Tli, the choices onnected
<ri,)))

1Before between 3Jli,

of the

footnote

referredto aboveis immediate. ,

<r,,

9lli and 9Rj, between

e b etc., tc.,etween

9TC,>_i

and

as described in

by the fully formulated

, 9TC*-.i

i.e. numerithe

SET-THEORETICAL DESCRIPTION

69

Thus the complete formal possibilities the courseof all conceivable of plays of our game up to 9TC,are described a family of pairwisedisjunct by subsets of ft. This is the family of all the setsA mentionedabove. We denotethis family by The sum of all setsA containedin a, must contain all possibleplays. But sincewe explicitlypermitted a redundancyof 12 (cf. the end of 9.1.1.), this sum neednevertheless be equal to ft. Summing up: not (9: ) ft is a partition in ft. A We couldalso say that the partition &< describes pattern of informathe of tion of a personwho knowseverything that happenedup to 3NI,;1 e.g. an the umpirewho supervises courseof the play.2 9.1.3. i Second,t must be known what the nature of the move 3fll is k , n if the going to be. This is expressed the k< of 6.2.1.:= 1, by move is personaland belongsto the player fc,; fc, = if the move is chance. kt may dependupon the courseof the play up to 3TI,, i.e.pon the informau 3 tion embodiedin a,. This means that must be a constant within each

a.

fc

setA
K

this family by (B,. (B, is again a partition in ft. Since (9:B) every A of ft, is a subset of someB (k) of (B, therefore & is a subpartition of (B*. But while there was no occasiono specifyany particular enumeration t of the setsA of Q , it is not so with (B,. (B*consistsof exactly n + 1sets B (k), k = 0, 1, , n, which in this way appear in a fixed enumeration 4 by means of the k = 0, 1, , n. And this enumeration is essential
K K K K K

of Gfc,, but that it may vary from one A to another. Accordingly we may form for every = 0, , n a set B*(k), which containsall setsA with k = k, the various B (k) beingdisjunct. Thus the B (k),k = 0, , n, form a family of disjunctsubsetsof ft. We denote
K K fc

1,

1,

sinceit replaces function k (cf. footnote 3 above). the with 9.1.4. the conditionsunder which the choiceconnected the Third,
K

must be described detail. in move 311, to take place is Assume first that 311, s a chancemove, i.e.hat we are within the set i t Then the significant quantities are: the number of alternatives a jB(0). and the probabilitiesp*(l), , p*(ctt) of these various alternatives (cf. item the end of As was pointed out in (this was the second

6.2.1.).

7.1.1.

1 c with the moves 9Ri, the outcome of all choices onnected , 9TC-i. In our earlier terminology: the values of <n, , <T*_I. 1It is necessary introduce such a person since,in general, no player will be in to o possessionf the full information embodiedin 8 In the notations and in the sense of the preceding footnotes:k* of , <r,(_i). fc(<ri, 4 Thus (B is really not a set and not a partition, but a more elaborate i concept:t con= 0, sists of the sets(B,c(fc), , n, in this enumeration. a which characterize of It possesses, (8:B:b) however, the properties (8:B:a), Yet even there an exceptionmust be made:among the sets (B(fc) there can partition.

I.e.

7.2.1., 1,

a.

A;

8.3.1.,

be empty

ones.)))

70

O DESCRIPTIONF GAMES OF STRATEGY

K

of the discussionthere), all these quantities may depend upon the entire information embodiedin GL (cf. footnote 3 on p. 69), since3TI* is now a move. I.e. and the a chance , />(<*) must be constant within eachsetA of Q l but they may vary from one A to another. Within each one of these A the choice among the alternatives &(!), of i t , a (cf. 6.2.2.). , &(*)takes place, .e.he choice a a = This can be describedby specifyinga disjunct subsets of A which corof expressed A , plus the choice a which has by respond to the restriction taken place. We call thesesetsC , and their system consistingof all C in all the A which aresubsetsof B (0) e(0). Thus 6^(0)is a partition in therefore B (0). And sinceevery C of G(0) is a subset of some A of i C(0)s a subpartition of Q, The a aredeterminedby (());2 hencewe neednot mention them any more. For the , p(cOthis descriptionsuggests itself:with every C of C^O) a number p (C^) (its probability) must be associated, 8 subjectto the equivalentsof footnote 2 on p.50. is a personalmove, say of the player Assume,secondly,that k = 1, that we are within the set B (k). In this casewe , n, this must specifythe stateof information of the player k at 311*. In was describedy means of the set A, in b by means of the family of a the latterdescriptionbeingthe more general nd the final one. functions the values of all functions According to this description k knows at h(<n, , 0\"ic-i) of $ and no more. This amount of information operates a subdivisionof B (k) into several disjunct subsets, correspondingto the various possible contents of fc's information at ?RI We call these sets D,and their system > (k). Thus ><(&)is a partition in B (k). Of course fc'sinformation at STC* part of the total information existing is at that moment in the senseof which is embodiedin ft, Hence in an A of &, which is a subset of B (k), no ambiguity can exist, .e.his i t A cannot possess common elements ith mt>re than one D of ><(&). This w means that the A in questionmust be a subset of a D of ><(&). In other words:within B (k) Q is a subpartition of >(&) In reality the courseof the play is narrowed down at am, within a set A of GL But the player k whose mo've 9TI is, does not know as much: as far as he is concerned, play is merely within a set D of >(fc). He the must now make the choice among the alternatives &*(!), , d(cx)> the choicef a <r = o As was pointed out in and , at the end of a may well be variable, but it can only (particularly i dependupon the information embodiedin ><(&). I.e.t must be a constant within the set D of 3D(fc) to which we have restricted ourselves. Thus the choice a = of can be described specifyinga disjunct , a, by subsets of D,,which correspond the restriction to expressed D , plus the by 1 We are within B (Q),henceall this refers only to AJs which are subsetsof #(()). 1 is the number of those C of e(0)which aresubsets the given A of * I.e. very p(C*) ;> 0, and for each A , and the sum extended over all C of e,t(0) e which aresubsets A t we have Sp.tC*) = of
K

p(l),
K

1,

a,,

p(l), i.e.

9.1.5.

9ffl

<,

7.2.1.
3fTC

6.3.1.

9.1.2.
K

1,

a,. 7.2.1.),
K

i.e. 7.1.2. 7.2.1.

<T

1,

1.)))

SET-THEORETICAL DESCRIPTION
K K K

71

of choice <r, which hastaken place. We call thesesetsC,and their system consistingof all C in all the D of )*(fc) <3(fc). Thus G (k) is a partition in B (k). And since every C of Q (k) is a subset of someD of >*(fc),therefore e(fc) is a subpartition of >(&). 1 The a* are determined by e(fc); hencewe need not mention them any more, a* must not be zero, i.e., given a D< of $> (k),someC of C(fc), 2 which is a subset of D,,must exist.
K K K K K

and Their Properties have completely described in the preceding sectionsthe situation at the moment which precedes move 3TC,. We proceed to the now discuss what happens as we go along these moves K = 1, -,*>. It is convenient to add to these a K = v + 1,too, which correspondsto the conclusion the play, i.e.ollowsafter the last move 311,. of f ForK = , v we have, as we discussedin the preceding sections, the partitions

9.2. o Discussionf ThesePartitions

9.2.1. We

1,

a.,(B, = (B,(0),B.(l),
All with

B.(n)),C.(0),C.(l),

, C.(n), >,(!),

of these, the sole exception &, refer to the move 3TC,, of But &+!has a perthey need not and cannot be defined for K = v + shows: It represents the fectly good meaning, as its discussionin full information which can conceivably exist concerning play, a the individual identity of the play.3 At this point two remarkssuggestthemselves In the sense the above of observations Oti correspondsto a moment at which no information is available at all. HenceCti shouldconsistof the one set 0. On the other hand, Ot,+i correspondsto the possibilityof actually identifying the play which has taken place. Hence sets. <$+!is a system of one-element We now proceed describe the transition from K to K + to when

1. 9.1.2. :

3>.(n). hence

i.e.

Nothing can be said about the changein the (B, e(fc), 5)(fc) our previous discussionshave shown that when K is replacedy K b is when this replacement made anything may happen to those objects, to what they represent. is possible,however, to tell how Q +i obtains from The information embodied in &,+) obtains from that one embodied connected the move with in by adding to it the outcome of the choice 4 This Thus the SHI*. ought to be clearfrom the discussionsof

9.2.2.
It

1,

1,

, v.

+ 1,

i.e.

6t.

9.1.2. 1.

1 is the number of those C of e(fc) which aresubsetsof the given A*. 1We required this for k = , n only, although it must be equally true for Jc = with an A , subset of #(0),in placeof our D of 3)*(fc). But it is unnecessary to of it state it for that case, because is a consequence footnote 3 on p. 70; indeed, if no of kind existed,the Sp(C) loc.cit, would be and not 0* of the desired 8 In the senseof footnote 1 on , <r,. And the sequence p. 69,the values of all <n, the play itself. as <TI, , <r, characterizes, stated in 4 In our earlier terminology: the value of

1,

6.2.2.,
<r.)))

72

O DESCRIPTIONF GAMES OF STRATEGY

restate this:

information in Ot,+i which goes beyondthat in ft, is preciselythe information embodiedin the e(0), (!), , C,(n). & the This meansthat the partitions a+i obtainsby superposing partition a. with all partitions 6(0), (1), , <S>*(k). I.e. y forming the interb 6 of section every A , in a,with every C< in any C^O),C(l), , C(n), and then throwing away the empty sets. Owing to the relationshipof ft, and of the e(fc) to the setsB (k) as sections we can say a little more about this discussedin the preceding of process superposition. Hence in In B(0), is a subpartitionof (cf.the discussion 9.1.4.). C(0) w there Q +i simply coincides ith G(0). In (&), k = 1, , n, G(fc) Hence in and ft, areboth subpartitionsof )(&) (cf.the discussion 9.1.5.). there &<+i obtains by first taking every D of 3)(fc),hen for every such D, t all A of &x and all C of C,(fc)which aresubsets of this D , and forming all A intersections n C . those plays which arise when the Every such set A n C represents player fc, with the information of D beforehim, but in a situation which is C really in A (a subset of D),makes the choice at the move 9TC,so as to restricthings to C. t Sincethis choice, to according what was said before, is a possibleone, there existsuch plays. I.e.the set A n C must not be empty. We
K Ofc K K K K K K K K K K K

(9:C)
ment

If A of Ot and C of C(fc)aresubsetsof the sameD of >(&), A then the intersection n C must not be empty.
K K K

betempted to set this requireplayer may make a legitimate choice which turns out subsequentlyto be a forbiddenone; the e.g. doubleblind Chessreferred to in footnote 1on p.58: erea player can make an h on apparentlypossiblechoice (\"move\") his own board, and will (possibly) be told only afterwardsby the \"umpire\" that it is an \"impossible\" one. This example is, however, spurious. The move in question is best resolvedinto a sequence f severalalternative ones. It seems est to give o b in the contemplatedrules of double-blind Chess full. \" Thegameconsistsof a sequencef moves. At eachmove the \"umpire o move was a \"possible\" announcesto both players whether the preceding one. If it was not, the next move is a personalmove of the same player as the preceding one;if it was, then the next move is the other player's personalmove. At eachmove the player is informed about all of his own o anteriorchoices, o about the entiresequence f \"possibility\" r \"impossibilof all anterior choices both players,and about all anterior instances of ity\" where either player threatenedcheckor took anything. But he knows of the identity of his own lossesnly. In determiningthe course the game, o the \"umpire\"isregardsthe \"impossible\" oves. Otherwisethe gameis m d played like Chess,with a stop rule in the senseof footnote 3 on p. 59,))) aside. Theseare games in
which a

9.2.3. aregamesin which onemight There

AXIOMATICFORMULATION

73

amplified by the further requirement that no player may make (\"try\ the samechoice twice in any one uninterruptedsequencef his own personal o of moves. (In practice, course,he players needtwo chessboards out of t \" eachother's view but both in the \" umpire's view to obtain theseconditions of information.) At any rate we shall adhere to the requirementstated above. It will beseenthat it is very convenient for our subsequentdiscussion(cf. 9.2.4. one thing remains: reintroducen our new terminology, to i Only the quantities SF*, k = of S , n, of 6.2.2.F* is the outcome the play for the player k. $ must be a function of the actual play which has taken k 1 place. If we use the symbolir to indicate that play, then we may say: $k is a function of a variable IT with the domain of variability 12. I.e.

11.2.1.). :

1,

$k

= ^W,

v in

12,

1,

n.

the new the use of sets and of partitions, is now complete. technique involving All constructionsand definitions have been sufficiently explained in the and to past sections, we can therefore proceed a rigorousaxiomatic definition of a game. This is, of course,only a concise restatementof the things s which we discussedmore broadly in the precedingections. 2 We give first the preciseefinition, without any commentary: d the complete An n-person game F, system of its rules,is determined of the specification the following data: by (10:A:a) A number v.

10.1.1. descriptionof the generalconceptof a game,with Our

10. xiomatic Formulation A 10.1. Axioms and Their Interpretations The

i.e.

(10:A:b) (10:A:c) (10:A:d) (10:A:e)

(10:A:f)

set 12. Forevery k = 1, , n:A function TTin 12. 3k = $*(*), Forevery = 1, A , v, v + 1: partition & in 12. For every = 1, v: A partition (B, in 12. (B con, in sists of n + 1 sets B (k), k = 0, 1, , n, enumerated
A finite
K
K

this way.

(10:A:g) (10:A:h)

1, partition Forevery * = 1,
K

partition e(fc) in

For every = 1,
K

For every =

B (k).
K

, ,

and every and every

k k

= 0, 1, =

>(/;)in J3,(fc).
,
v

1,

n: , n: A
,

A and every C* of C(0): number))

Theseentitiesmust satisfy the following requirements: ft, is a subpartition of (B,. :a) (10:1 e(0)is a subpartition of :b) (10:1
In the old terminology, accordingly, 2For \"explanations\" cf. the
1

a. we end of 10.1.1.

had ff* g*(ai, and the discussion

, o>). Cf. of 10.1.2.)))

6.2.2.

74

(10:1 :c) (10:1 :d) (10:1 :e) (10:1 :f) (10:1 :g) (10:1 :i) (10:1 :j)

For every = 1, , ? and every .A* of Ot which is a F subset of 5(0): or all C of C(0) which are subsets of this over <4,P*(C*) ^ 0,and for the sum extended them Sp^C,)= 1. Cti consistsof the one set Q.
K
K

ft(*).

O DESCRIPTIONF GAMES OF STRATEGY Fork = 1, , n:C,(fc)is a subpartition of ><(fc). Fork = 1, , n:Within B<(fc), 6t is a subpartition of

(10:l:h)
with

For = 1,
K K

Ot.,+1 consistsof

sets. one-element
K K

sectionA
K

For = 1, a, , n are subsets of the same D of

K

k all e,(*),

= 0, 1,

it , v. Q +i obtains from Q by superposing , n. (Fordetails, cf. and C of e*(fc), = , v: If A of

K

9.2.2.)
fc K

For =

C, must not be empty.

>*(fc), then

the inter-

1,

>,(&)

:SomeC,(fc)of e,,which is a subset of D , must exist.

K

1,

and

fc

= 1,

, n and every

D of

This definition shouldbe viewed primarily in the spirit of the modern axiomatic method. We have even avoided giving names to the mathei matical conceptsntroducedin (10:A:a)-(10:A:h)in order to estababove, of no correlation with any meaning which the verbal associations names lish can In this absolute \"purity\" these concepts then be the may suggest.
m o objectsf an exact athematical

This procedureis best suited to develop sharply defined concepts. The application to intuitively given subjectsfollows afterwards, when C the exactanalysis has been completed. f. also what was said in 4.1.3. I about the role of modelsin physics: The axiomatic models in Chapter to for intuitive systemsareanalogous the mathematical modelsfor (equally intuitive) physicalsystems. Once this is understood, however, there can be no harm in recalling that this axiomatic definition was distilled out of the detailed empirical which precedet. And it will facilitate its use, i of discussions the sections, and make its structure more easilyunderstood,if we give the intervening conceptsappropriate names, which indicate,as much as possible, the in intuitive background. And it is further useful to express, the same i.e.he t spirit, the \"meaning\" of our postulates (10:l:a)-(10:l:j) intuitive from which they sprang. considerations All this will be,of course, erely a conciseummary of the intuitive cons m w siderationsof the precedingections, hich lead up to this axiomatization. s o 10.1.2. statefirst the technicalnamesfor the concepts f (10:A:a)We
i (10:A:h)n
1

investigation. 1

10.1.1.

This is analogous to the present attitude in axiomatizing such subjects as logic, geometry, etc. Thus, when axiomatizing geometry, it is customary to state that the intuinotions of points, lines, and planes are not to be a priori identified with anything in tive, they are only notations for things about which only the properties expressed the axioms are assumed. Cf., D. Hilbert: Die Grundlagen der Geometrie,Leipzig 1910.))) 1899,2rd Engl. Edition Chicago

e.g.,

AXIOMATIC FORMULATION

75

(10:A:a*) (10:A:b*) (10:A:c*) (10:A:d*) (10:A:e*)

(10:A:f*)

of assignment, a B*(k) of (B, is the actual assignment, of the move 9TC,. <3*(fc) is the pattern of choice, a C, of 6<(fc) is the actual Of choice, of the player & at the move 3TZ*. (For =
(B*is the pattern
A;

a the immediately preceding) umpire's ctual information at (i.e. move 3TC. (For = v + 1 At the end of the game.) :
K

is the length of the game F. is the set of all plays of T. ^(TT)is the outcome of the play w for the player fc. Ct is the umpire's attern of information, an A of ft, is the p
v ft

0:

3)(A;) is the player k'spattern of information, a D of >*(&) the player k's actual information, at the move 3TC. (10:A:h*) p*(C*) is the probability of the actual choiceC at the m (chance) ove 3TI,. We now formulate the \"meaning\" of the requirements (10:1 :a)of with the useof discussion :j) in the senseof the concluding the above nomenclature. (10:l:a*) The umpire'spattern of information at the move 3TC includesthe assignmentof that move. (10:l:b*) The pattern of choiceat a chancemove includes the umpire'spattern of information at that move. at Thepattern of choice a personalmove 3TC,of the playerk (10 :c*) includesthe player k'spattern of information at that move. (10:l:d*) The umpire's pattern of information at the move 9Tl includes to the extentto which this is a personalmove of the player k the player k's pattern of information at that move. (10:l:e*) The probabilitiesof the various alternative choicesat a to chance move 9TC< behave like probabilities belonging disjunct but exhaustive alternatives. (10:l:f*) The umpire'spattern of information at the first move is

(10:A:g*)

chance.)

(10:1

10.1.1
9fTl

:1

Theumpire's pattern of information at the end of the game the play fully. determines (10:l:h*) The umpire'spattern of information at the move ST^+i (for = v: at the end of the game)obtains from that one at
K

(10:1 :g*)

void.

(10:l:i*)

a the move SfTl, by superposingit with the pattern of choicet the move 311,. Let a move 2(11,be given, which is a personalmove of the player k, and any actual information of the player k at that move also be given. Then any actual information of the of umpire at that move and any actualchoice the player k at r that move, which are both within (i.e.efinements of) this actual (player's)information, are also compatiblewith each Other T +.VIPV nnniir in flftiifll nlnvs)))
**

76

O DESCRIPTIONF GAMES OF STRATEGY Let a move SHI, be given, which is a personalmove of the player A;, and any actual information of the player k at that move also be given. Then the number of alternative actual available to the player k, is not zero. choices,

(10:1 :j*)

of s o This concludesur formalization of the general cheme a game.

We 10.2. have not yet discussedthose questionswhich areconventionin associated

10.2. Logistic

Discussion the of

Axioms

freedom from formal logics with every axiomatization: 1 and of (completeness), independence the axioms. contradiction,categoricity t Our systempossesseshe first and the last-mentioned properties,but not the secondone. Thesefacts are easy to verify, and it is not difficult to see that the situation is exactlywhat it shouldbe. Insumma: from Therecan be no doubt as to the existence Freedom contradiction: f of games,and we did nothing but give an exact ormalism for them. We t shall discussthe formalization of several games later in detail, cf. e.g.he of 19. From the strictly mathematical logistic point of examples view, even the simplest game can be used to establishthe fact of freedom from contradiction.But our real interestlies, of course,with the more 2 involved games,which arethe really interestingones. This Categoricity (completeness): is not the case,sincethere exist different games which fulfill these axioms.Concerning effective many cf. examples, the precedingreference. The readerwill observethat categoricityis not intended in this case, a sinceour axiomshave to define a classof entities(games) nd not a unique
ally

18.,

entity.8

Thisis Independence: easyto establish,but

we do not enterupon it.

the Axioms

There 10.3. aretwo more remarkswhich ought to be madein connection

with

10.3. Remarks General

Concerning

follows the classical lines of obtaining an exact First,our procedure formulation for intuitively empirically given ideas. The notion of a in i e game existsn general xperiencea practicallysatisfactory form, which is to neverthelesstoo loose be fit for exacttreatment. The reader who has followedour analysiswill have observedhow this imprecision gradually was 1 Cf. D. loc.cit.; .Veblen & J. W. Young: Projective Geometry, New York 0 Hilbert,

this axiomatization.

1This is game: v 0, 12 has only one element, sayiro. Consequently no (B, C(fc), >(fc),exist, while the only O is di, consisting of 12alone. Define$(iro) 5 for k = , n. An obvious description of this game consistsin the statement that nobody doesanything and that nothing happens. This alsoindicates that the freedom from contradiction is not in this casean interesting question, 8 This is an important distinction in the general logistic approach to axiomatization. a Thus the axioms of Euclidean geometry describe unique object while those of group theory (in mathematics) or of rational mechanics (in physics) do not, sincethere exist many different groups and many different mechanical systems.)))

H. 1910;Weyl: Philosophic der Mathematik 1927.

Philosophic, Munich, the simplest

und Naturwissenschaften,

in Handbuch

der

I,

AXIOMATIC FORMULATION

77

o a formularemoved,the\"zone f twilight \" successively reduced, nd a precise tion obtained eventually. i o Second,t is hoped that this may serveas an examplef the truth of a That it is possible to describe discuss much disputed proposition: and mathematically human actions in which the main emphasis lieson the psychologicalside. In the present casethe psychologicalelementwas the brought in by the necessityof analyzing decisions, information on the o basis of which they are taken, and the interrelatednessf such sets of information (at the various moves) with eachother. This interrelatedness
in of originates the connection the various sets of information in time, causation,and by the speculativehypotheses of the players concerning eachother. Thereareof course o many and most important aspects f psychology which we have never touchedupon, but the fact remains that a primarily g psychological roup of phenomenahas beenaxiomatized.

10.4. Graphical

Representation

a we had to use to represent gameis not easy. We shall not attempt to treat this matter systematically: even relatively simple gamesseemto and lead to complicated confusing diagrams,and so the usual advantagesof d graphicalrepresentationo not obtain. of restrictedossibilities graphicalrepresentaThereare,however, some p tion, and we shall say a few wordsabout these.

o 10.4.1. graphical representationf the numerous partitions which The

\"meaning, that a,+i is a \") in of , ft,, ft,+i subpartition of ft,. I.e. the sequence partitions Cti, eachoneis a subpartition of its immediatepredecessor. Consequentlythis i.e. much can be pictured with the devicesof Figure9 in 8.3.2., by a tree. in (Figure9 is not characteristiconeway: sincethe length of the gameF is of assumedto be fixed, all branches the treemust continue to its full height. Cf. Figure10in 10.4.2. below.) We shall not attempt to add the ,(&), Cic(fc), 3D(fc)to this picture. o Thereis, however, a class f gameswherethe sequenceti, C , ft,,ft,+i tells practically the entire story. This is the important class already and discussed in 6.4.1., about which more will be said in 15.where find and anteriority are equivalent. Its characteristics a preliminarity in our presentformalism. simpleexpression 10.4.2. Preliminarity and anteriority areequivalent as the discussions of show if and only if every 6.4.2. of 6.4.1., and the interpretation 6.4.3. who makesa personalmove knowsat that moment the entireanterior player history of the play. Let the player be fc, the move 9TC,. The assertion that 9TI, is k'a personalmove means,then,that we arewithin B (k). Hence the assertion is that within B (k) the player fc's pattern of information that with coincides the umpire's pattern of information; i.e. 2D(fc)is equal
K K
to)))

i.e. :h*) (10:1 in 10.1.2.,by rememberingthe

In the first placeit is clearfrom

(or (10:1 in 10.1.1., equally by :h)

78))

DESCRIPTIONF GAMES OF STRATEGY)) O

K K

Q, within B (k). But >(&) is a partition in B (k); hencethe above ment meansthat >,(&)simplyis that part of a* which liesin B (k).
K

state-

We

this: restate

Preliminarity and anteriority coincide i.e.very player who e makes a personalmove is at that moment fully informed about the entire anterior history of the play if and only if > (k) is that part of (t which lies in B (k). If this is the case,hen we can argueon as follows: y (10:1 in 10.1.1. t B :c) and the above, Q (k) must now be a subpartition of Q . This holds for f , n, but for k = it follows immedi-)) personalmoves, i.e.or k = 1,

(10:B)

from

Now (10:1in 10.1.1. (10:1 in 10.1.1. the inference :b) :h) permits this (for details 9.2.2.) a+i coincidesith Q (k) in B (k) for cf. that w all k = 0, 1, , n. (We could equally have used the corresponding i.e. t points in 10.1.2.,he \" meaning \" of theseconcepts.We leave the verbal o expressionf the argument to the reader.) But Q (k) is a partition in B (k) ; hencethe above statementmeans that C<(k) simply is that part of G^+i which lies in B^k). ately from
K K K K

Figure

10.

We

restate this:

(10:C)

If the condition of (10:B) fulfilled, then Q (k) is that part is of ft*+i which lies in B^k).
K

Thus when preliminarity and anteriority coincide,hen in our present t formalism the sequence and the sets Bx (k), k = 0, tti, , a,,a,+i the t , n, for eachK = , v, describe gamefully. I.e.he picture)))

1,

1,

AND FINAL SIMPLIFICATION STRATEGIES

K

79

of Figure9 in 8.3.2. be amplified only by bracketing must togetherthose o elements f each&, which belongto the same set(& (k). (Cf. however, the remark made in We can do this by encircling them with a line, across hich the number k of B (k) is written. SuchB (k) as areempty can w be omitted. We give an exampleof this for v = 5 and n = 3 (Figure 10). In many games of this class even this extradevice is not necessary, for because every K only one B (k) is not empty. I.e.he character each of t move 311,is independent of the previous courseof the play.1 Then it sufficesto indicateat eachOt, the character the move of the unique k = 0, , n for which B (k) j

10.4.1.)

1,

3fTC*

i.e.

11. trategiesnd the Final Simplification of the Descriptionof a Game a S 11.1. Conceptof a Strategy and Its Formalization The 11.1.1. us return to the courseof an actual play TTof the game F. Let The moves follow eachotherin the order = 1, , v. At each is move 9Tl a choice made,eitherby chance the play is in B<(0) or by a the play is in B (k). Thechoice consistsin the ,n player k = 1, of t selection a C from Q (k) (k = or k = 1, , n, cf. above), o which
SfTC*

if

if

is restricted.If the choice made by a player fc, then precautions must be taken that this player's pattern of information shouldbe at this moment >*(&), as required. (That this can be a matter of some as d practicalifficulty is shown by such examples Bridge[cf. the end of and double-blindChess[cf. 9.2.3.].) 6.4.2.] now Imagine that eachplayer k = 1, , n, instead of making each m decisionas the necessityfor it arises, akes up his mind in advance for all i.e. possiblecontingencies; that the player k beginsto play with a complete w h situa$ plan which specifies hat choicese will makein every possible plan: for every possibleactual information which he may possessat that tion, moment in conformity with the pattern of information which the rules of the gameprovide for him for that case. We call such a plan a strategy. Observethat if we requireeachplayer to start the game with a complete w h plan of this kind, i.e. ith a strategy, we by no means restrict is freedom of action. In particular, we do not thereby force him to make decisions information than therewould be available for him in each on the basis of less instancein an actual play. This is because strategy is supthe practical as a function of just that to specify every particular decision only posed amount of actual information which would be available for this purposein an actualplay. The only extraburden our assumptionputs on the player o is the intellectual ne to be prepared with a rule of behavior for all eventualities, although he is to go through one play only. But this is an innocuous assumption within the confines of a mathematical analysis. (Cf. also 4.1.2.)
the play is then
1This is
true for

Chess. The rules of

Backgammon permit interpretations

both)))

80

O DESCRIPTIONF GAMES OF STRATEGY 11.1.2. chancecomponent of the game can be treatedin The

the same

It isindeedobvious that it is not necessaryto makethe choices hich are w left to chance, those of the chance moves, only when thosemoves come their along. An umpire could make them all in advance, and disclose outcome to the players at the various moments and to the varying extent, as the rulesof the game provide about their information. It is true that the umpire cannot know in advance which moves will be chance ones,and with what probabilities;this will in general depend upon the actual course the play. But as in the strategies hich we considered of w above he could provide for all contingencies: could decide advance He in in what the outcome of the choice every possiblechance move shouldbe, for f every possibleanterior course of the play, i.e.or every possibleactual information at the move in question. Under theseconditionsthe umpire's probabilitiesprescribedby the rules of the game for eachone of the above instances would be fully determined and so the umpire could arrange for t eachone of the necessarychoiceso be effected by chance, ith the approw

way.

i.e.

priate probabilities. The outcomescouldthen be disclosed the umpire to the players at by the propermoments and to the properextent as describedabove. We call such a preliminary decisionof the choices all conceivable of chance moves an umpire'schoice. We saw in the last sectionthat the replacement the choices f all of o personalmoves of the player k by the strategy of the player k is legitimate ; i.e.that it does not modify the fundamental characterof the game F. o Clearly our present replacementof the choices f all chancemoves by the is choice legitimate in the same sense. umpire's 11.1.3. It remains for us to formalize the concepts f a strategy and of o the umpire'schoice.The qualitative discussionof the two last sections makesthis an unambiguous task. A strategy of the player k doesthis: onsidera move 3fTC,. C Assume that it has turned out to be a personalmove of the player &, i.e.ssumethat a the play is within B(fc). Considera possibleactual information of the c player k at that moment, i.e. onsidera D* of ><(&). Then the strategy in question must determinehis choice this juncture, i.e. C of Q (k) at a which is a subset of the above D .
K K

Formalized:

(11:A)

A strategy of the player defined for every * = whosevalue

!,,*/
A;

is a function 2*(/c;D ) which is and every D of >,(k), and

K K

2*(*; ,) = C, D
K

That strategies i.e.unctions S*(K;D<) fulfilling the above requirement f existat all,coincides preciselywith our postulate (10:1in 10.1.1.))) :j)

ofD,.

has always theseproperties: belongsto 6(fc) and is a subset C

AND FINAL SIMPLIFICATION STRATEGIES

An

81

does choice this: umpire's Considera move 9TC. Assume that it has turned out to be a chance move, i.e.assume that the play is within J3(0).Consider a possible i c actual information of the umpire at this moment; .e.onsideran A of ft, in which is a subset of B(0). Then the umpire'schoice question must a at determinethe chancechoice this juncture, i.e. C of <3(0) which is a subset of the above A . : Formalized
K K

(11 :B)

A umpire'schoiceis a function 2o(*; ) which is defined for = 1, v and every A of QL which is a subset of every , B (0) and whosevalue An
K

A So(*;

= C,

of A
fulfilling

has always theseproperties: belongsto <B(0) and is a subset C

K

the above requirement cf. the remark after (11:A) above, and footnote 2 on p.71. of Sincethe outcome the umpire'schoicedepends on chance,the coris choice an respondingprobabilitiesmust be specified. Now the umpire's of independent chanceevents. There is such an event, as aggregate in for every K = 1, described , v and every A of G, which is a for subset of J3(0).I.e. every pair *, A< in the domain of definition of the A SO(K; t). As far as this event is concerned probabilityof the particular outcome2o(*;A ) = C is p*(C^). Hencethe probability of the entire representedy the function 2 (*;A ) is the product of the b umpire'schoice, individual probabilitiesp (C ).1

the of choices i.e. f functions o Concerning existence umpire's

2 (*;^)

11.1.2.,
K

: Formalized

(11:C)

The probability of the umpire'schoice, represented the by function o(*; A ) is the product of the probabilities p(C), and K, A run over the entire domain of where 2 (*; A ) = A definition of 2o(*; ) (cf. (11 :B)above).
K K

C,
K

in If we considerthe conditionsof for all thesepairs and multiply them all with eachother, then thesefacts result: he T K, of (11:C) above areall ^ 0,and their sum (extended over all probabilities i umpire'schoices)s one. This is as it should be, sincethe totality of all choicess a systemof disjunct but exhaustive alternatives. i umpire's
A K,

(10:l:e)10.1.1.

then these , n, and if a definite umpire'schoicehas been selected, determinethe entire courseof the play uniquely, and accordinglyits
1

ll.).The Simplification of the Description of a Game 11.2.1.a definite strategy has been adopted by each player k = 1, If
Final

Thechanceevents in

question must

be treated as independent.)))

82

b the verbal descriptionof all theseconcepts, ut an equally simple formal proof can be given. i Denotethe strategiesn questionby 2*(/c;D ), k = , n, and the umpire'schoiceby 2 (/c; A ). We shall determine the umpire'sactual In orderto avoid information at all moments K = , v} v + it confusing it with the above variable A^, we denote by A A i is, of course, qual to 12 itself. (Cf. (10:1in :f) e A n Consider ow a K = , v, and assumethat the corresponding is already known. Then A is a subset of preciselyone B (k), k = 0, - , n. (Cf.(10:1 in If k = 0,then 3TC,is a chance move, and :a) is so the outcome the choice 2o(*;A ). Accordingly A f +i = 2o(*;A ). of I and the details in 9.2.2.)f k = , n, then (Cf. (10:1 in :h) one D of 3TC,is a personalmove of the player k. A is a subset of precisely is of So the outcome the choice Z*(K; D ). :d) >,(fc). (Cf. (10:1 in and the details :h) AccordinglyA f +i = A n S*(K; D ). (Cf. (10:1 in
K

O DESCRIPTIONF GAMES OF STRATEGY from outcome for eachplayer k = 1, , n. This shouldbe clear too,

1,
K

1,

1, 10.1.1.) 10.1.1. 10.1.1.)

K K K

1. . 10.1.1.)
K

1,

1,

in

a Consider given player k = , n. Form all possiblestrategies of his, 2*(*; D ), or for short 2*. While their number is enormous, it is obviously finite. Denoteit by Pk, and the strategies themselvesby ZJ, , 2&Form similarly all possibleumpire'schoices, (*;A ), or for short 2 2 their numberis finite. Denote by /3o, and the umpire's it choices Again by Denotetheir probabilities by p1, 2j, , 2?o. , p^o respectively. All theseprobabilitiesare ^ and their sum is one. (Cf. (11 in :C) the end of (Cf. A definite choice all strategies nd of the umpire's hoices, 2J* for of a c say k = n and for k = respectively,where ,

the gameF.

Thus we determineinductively A\\, A^ A^ - , A A v+i in succession. denote But A 9+i is a one-element (cf. (10:1 in 10.1.1.); its unique set :g) element y if. b 1 This TTis the actual play which took place. Consequentlythe outcome of the play is *(*) for the playerk = 1, , n. 11.2.2. fact that the strategiesof all players and the umpire's The choicedeterminetogetherthe actual play and so its outcomefor each of player opens up the possibilityof a new and much simplerdescription
vy

9.2.2.)

10.1.1.

1,

'

11.1.3.) 11.1.3.) 1,
O

1,

T* =

/fo

for

= 0, 1,

, n,

determineshe play TT (cf. the end of 11.2.1.), its outcome^(TT) for t and n. Write accordingly eachplayer k = 1, , = S*(T , TI, k = 1, for (11:1) IF*(T) , r n) , n.
1Theaboveinductive derivation of the Ai, ^2,^j, , Av, A+i is just a mathematical reproduction of the actual course the play. Thereadershould verify the parallelof ism of the stepsinvolved.)))

AND FINAL SIMPLIFICATION STRATEGIES

83

i.e. number a

The entire play now consistsof each player k choosinga strategy SJ, 7 of choice T = , 0*; and of the chance umpire's , 0o, with the probabilitiesp\\ , /A respectively. h The playerk must chooseis strategy, his r*, without any information of the concerning choices the other players, or of the chanceevents (the choice). This must be so sinceall the information he can at any umpire's in time possessis alreadyembodied his strategy 2* = SJ* i.e. the function in Even if he holds definite of 2jk = *(*; D ). (Cf. the discussion views as to what the strategies the other players are likely to be, they of must be already containedin the function 2* (K; D). All this means, however, that F has beenbrought back to the very simplestdescription,within the least complicated original framework of the sections We have n + I moves, one chanceand one personalfor eachplayer k = , n eachmove has a fixed number of move and 0i, alternatives, /3 for the chance , ft for the personalones and every player has to make this choice with absolutelyno information 1 the concerning outcome of all otherchoices. of Now we can get rid even of the chancemove. If the choices the t players have taken place,he player k having chosen T*, then the total influenceof the chance o move is this Theoutcome f the play for the playerk may be any one of the numbers
fc

=!,

1,

i.e.

11.1.1.)

11.2.3. 6.2.1. -6.3.1.

1,

9*(T0, TI,
with
\"

, Tn ),

=
is))

1,' ' ' , 00,

T lf

the probabilities p 1, , yA respectively. Consequently his \" mathematical expectation of the outcome
JC*(TI,

(11:2)

,r ) = -lP .S*(T,
n

, r n ).

The player's judgment must be directed solelyby this \" mathematical expectation/'becausethe various moves, and in particular the chance 2 move, are completelyisolated from each other. Thus the only moves n. which matter arethe n personalmoves of the playersk = 1, The final formulation is therefore this : t (11:D) Then persongame T, i.e.he complete of its rules,is system

(ll:D:a) (ll:D:b)
1

determinedby the specification of the following data:

Forevery k = Forevery k = 1, 3C* = OC*(ri, r, = 1,

!,

, , , ft

n:A number /3*. n:A function

for

, r n ),

j = 1,

n.

of complete disconnectedness the n -f 1 moves, it doesnot matter order they are placed. 2 We are entitled to use the unmodified \"mathematical expectation\" sincewe are satisfied with the simplified conceptof utility, as stressedat the end of 5.2.2 This excludes particular all those more elaborateconceptsof \"expectation,\" which are in D o really attempts at improving that naive concept f utility. (E.g. . Bernoulli's \"moral expectation\" in the \"St. Petersburg Paradox.\))
Owing

to

this

in what chronological

84

O DESCRIPTIONF GAMES OF STRATEGY

of Thecourse a play of T is this: a Each player k chooses number r = 1, ,/}. Each o in player must make his choice absoluteignorancef the choices of the others. After all choiceshave been made, they are of that submittedto an umpire who determines theoutcome the rn). for the player k is 3C*(ri, , play

strategy.\" Each player has one move, and onemove only; and he must 1 o make it in absolute ignorance f everything else. This completerystalc lization of the problem in this rigid and final form was achieved by our
11

11.3. of Strategiesin the Simplified Form of a Game TheRole no is 11.3. Observethat in this scheme space left for any kind of further 11.1.1.

order.

f on, in which the transition from manipulations of the sectionsrom the original moves to strategieswas effected. Sincewe now treat these of themselvesas moves, there is no needfor strategies a higher strategies

11.4. Meaning The

of the Zero-sum

Restriction

11.4. concludeheseconsiderationsby determiningthe placeof the We t within zero-sumgames (cf. 5.2.1.) our final scheme. That T is a zero-sum this: game means,in the notation of 10.1.1.,
(11:3)
becomes
Jb-l , r n ), in the

5k(*)

for all w of fl. then senseof 11.2.2., this

If we pass from ^(TT)to 8*(TO, r,,

n

(11:4)
And if

*-i

9*(T ,Ti,

, rn)

=0
,

for all TO,TI,

,r .
n

we finally introduceJC*(TI,
n

we r), in the senseof 11.2.3.,obtain for all n,

, rn

(11:5)
defined in

5) *-i

OC*(ri,

rn) =

t :5) Conversely,it is clearhat the condition(11 makesthe gameF, which we

one 11.2.3.,of zero sum.

1Reverting to the definition of a strategy as given in In this game a player has one and only one personal move, and this independently of the courseof the play, the move 311*. And he must make his choiceat Sdl* with nil information. So his for strategy is simply a definite choice the move 9R*, no more and no less; precisely r , 0*. We leaveit to the readerto describe game in terms of partitions, and to compare this the abovewith the formalistic definition of a strategy in A) in 11.1.3.)))
A;

11.1.1.:

- 1,

i.e.

: (11

CHAPTER III ZERO-SUMTWO-PERSONGAMES: THEORY

12.1.1.the precedingchapter we obtained an all-inclusive formal In o characterizationf the general ame of n persons (cf. 10.1.). followed We g an o up by developing exact oncept f strategy which permittedus to replace c the rather complicatedgeneral cheme a gameby a much more simple of s special ne, which was neverthelessshown to be fully equivalent to the o former (cf. 11.2.). the discussionwhich follows it will sometimesbe In more convenient to use one form, sometimes other. It is therefore the desirable give them specifictechnical ames. We will accordinglycall to n them the extensiveand the normalizedform of the game, espectively. r Sincethesetwo forms are strictly equivalent, it is entirely within our province to use in eachparticular casewhichever is technicallymore convenient at that moment. We propose, indeed,to make full use of this and must therefore re-emphasize this does in the least that not possibility, affect the absolutelygeneral alidity of all our considerations. v Actually the normalized form is better suited for the derivation of w t general heorems, hile the extensiveform is preferablefor the analysisof cases; .e., former can be used advantageouslyto establish proi the special w perties hich arecommon to all games,while the latterbrings out characteristicdifferences of games and the decisive structural features which determinethesedifferences. (Cf. for the former 14., and for the latter 17.,
Sincethe formal descriptionof all gameshas been completed, that we must now turn to build up a positivetheory. It is to be expected from simplergames to a systematicprocedure this end will have to proceed to more complicated games. It is therefore desirable to establish an o for all gamesaccording their increasing to degreef complication. ordering to We have already classifiedgames according the number of partician pants a game with n participants being called n-person game and to also according whether they areor arenot of zero-sum. Thus we must distinguish zero-sum n-persongames and general n-persongames. It will to be seenlaterthat the generaln-person gameis very closelyrelated the zero-sum (n + l)-persongame, in fact the theory of the former will case obtain as a special of the theory of the latter. (Cf. 56.2.2.) t someremarks concerninghe one-person ame. g In the normalized form this game consists of the choiceof a number
85)))

12. reliminarySurvey P 12.1. Viewpoints General

15.) e.g.

12.1.2.

12.2.1. begin with We

The 12.2. One-personGame

86
r

ZERO-SUMTWO-PERSONGAMES: THEORY

8 are the known forms of \"solitaire\" not such. This possibility is neverthelessa practical ne for certaineconomic o A setups: rigidly establishedcommunistic society, in which the structure i where thereis no exchange, of the distributionschemes beyonddispute (i.e. but only one unalterable imputation) would be such sincethe interests 4 of all the membersof such a societyarestrictly identical this setup must be as treated a one-personame. But owing to the conceivable g imperfections of communications among the members,all sorts of incomplete information can occur. u Thisis then the case o which, by a consistent se of the concept f strategy of (i.e. planning),is naturally reducedto a simplemaximum problem. On the basis of our previous discussionsit will therefore be apparent now at We Cf. (ll:D:a), (ll:D:b)the end of 11.2.3. suppressthe index 1 8 Then3C(r)= 0, cf. 11.4. \" 8 The existing \"double solitaires are competitive games between the two particitwo-person games. pants, i.e. 4 The individual members themselves cannot be considered players, since all as possibilities of conflict among them, as well as coalitions of some of them against the others, are excluded.)))

and the amount of information available to the player at any particular personalmove may vary in any prescribedway. of Numerousgood examples many complications subtleties 12.2.2. and in which may arise this way aregiven by the various gamesof \"Patience\" Thereis, however, an important possibilityfor which, to or \"Solitaire.\" the best of our knowledge,examples are lacking among the customary of o one-person ames. This is the case incomplete g information, i.e. f nonequivalence of anteriority and preliminarity of personal moves of the Forsuch an absence equivalenceit would be of unique player (cf. 6.4.). necessarythat the player have two personalmoves 311*and 3n\\ at neither of which he is informed about the outcomeof the choiceof the other. Sucha stateof lackof information is not easy to achieve, but we discussed in 6.4.2. it can be brought about by \"splitting\" the player into two how or more persons of identical interestand with imperfect communications. c o We saw loc. it. that Bridgeis an examplef this in a two-persongame; it would be easy to construct an analogousone-personame but unluckily g

i zero-sum caseis obviously void and there is nothing to say concerningt. The generalcasecorrespondsto a generalfunction 3C(r) and the \"best\" o or \" rational\" way of acting i.e. f playing consists obviously of this: r The player 1will choose = 1, , so as to make JC(r) a maximum. This extremeimplification of the one-person s game is, of course,due not a choice(in a move) but to the fact that our variable r represents it the his the player's strategy; i.e., expresses entire\"theory\"concerning situations which may occurin the course the of handling of all conceivable that g play. It shouldbe remembered even a one-personame can be of a I moves as well as personal pattern: t may contain chance very complicated moves (of the only player), eachone possiblywith numerous alternatives,

1,

1 (only) player 1 getsthe amount 3C(r). The 0,after which the 2

PRELIMINARY SURVEY

87

that this and this only is the case which the simplemaximum formuin lation the \"RobinsonCrusoe\"orm of economics appropriate. f is These considerationsshow also the limitations of the pure maximum i.e.the \"RobinsonCrusoe\" approach. The above example of a society of a rigidly establishedand unquestioneddistribution scheme showsthat on this plane a rational and critical ppraisalof the distribution a scheme itself is impossible. In order to get a maximum problem it was the of necessaryto place entirescheme distribution among the rules of the game, which are absolute, inviolable and above criticism.In order to the strategy bring them into the sphere of combat and competition of the game it is necessaryto considern-persongames with n ^ 2 and o thereby to sacrifice the simplemaximum aspect f the problem.

i.e. 12.2.3.

i.e.

Before going further, we wish to mention that the extensive literature of \"mathematical games\" which was developed mainly in the 18th and 19thcenturies deals essentiallyonly with an aspect f the o matter which we have already left behind. This is the appraisal of the influence of chance. This was, of course,effected by the discoveryand appropriate application of the calculus of probability and particularly I of the conceptof mathematical expectations.n our discussions,the 12 operationsnecessaryfor this purposewere performedin Consequentlywe are no longerinterestedin these games,where the mathematical problem consistsonly in evaluating the roleof chance S in computing probabilitiesand mathematical expectations. uch games in lead occasionallyto interesting exercises probability theory;8 but we hope that the readerwill agreewith us that they do not belong in the theory of games proper.

12.3.

12.3. and Chance

Probability

11.2.3.
*

i.e.

2 Concerning the important connection between the use of mathematical expectation it. and the considerations which precede o and the concept f numerical utility, cf 8 Some games like Roulette are of an even more peculiar character. In Roulette the mathematical expectation of the players is clearly negative. Thus the motives for in that game cannot be understood if one identifies the monetary return participating

Objective to the analysis of more complicated games. proceed The generalone-person game having been disposed of, the simplest one of the remaining games is the zero-sumtwo-person game. Accordingly we aregoing to discussit first. Afterwards there is a choiceof dealing either with the generaltwog persongame or with the zero sum three-person ame. It will be seenthat necessitates up the zero-sumthree-person our techniqueof discussion taking 1We do not in the leastintend, of course, o detract from the enormous importance t of of those discoveries.It is just because their great power that we are now in a position of to treat this sideof the matter as briefly as we do. We are interested in those aspects the problem which are not settled by the conceptof probability alone;consequently theseand not the satisfactorily settled onesmust occupy our attention.

The 12.4. Next

12.4. We

now

.3.7.

with

utility.)))

88

ZERO-SUMTWO-PERSONGAMES: THEORY

game first. After that we shall extendthe theory to the zero-sumn-person ) and only subsequentlyto this will it be game (for all n = 2,3, found convenient to investigatethe general -person n game.

1,

Our 13.1.1. next objective is as stated in 12.4.the exhaustivedisto cussionof the zero-sumtwo-persongames. In order do this adequately, of it will be necessaryto use the symbolism the functional calculus or at leastof certainparts of it more extensivelythan we have done thus far. w The concepts hich we need arethose of functions, of variables, of maxima A and minima, and of the use of the two latteras functional operations. ll a this necessitates certainamount of explanation and illustration, which will be given here. After that is done,we will prove some theorems concerning maxima, minima, and a certaincombination of these two, the saddlevalue. These will theorems play an important part in the theory of the zero-sumtwopersongames. w is a dependence hich stateshow certainentities A function 13.1.2.
y, value of
%,

F 13. unctional Calculus Basic 13.1. Definitions

'''
</>.

</>

Thus u is determined by and by the x, y, , and this i.e.ependencewill be indicatedby the symbolicequation determination d
<f>

the called

variables of
u

<t>

determine entity an ).

the called

= </>(*, y,

In principleit is necessaryto distinguishbetween the function itself o which is an abstractentity, embodying only the generaldependence f and its value <t>(x, y, u = <t>(x, y, - - - ) on the x, y, - ) for any In practical mathematical usage,however, it is often specificx, y, to write <t>(x, y, - - - ) but with x, y, - - - indeterminate convenient
<t>

footnote 1below). the In orderto describe

instead of

(c)-(e)below;(a), (b) are even worse, cf. (cf. the examples

<

function it is of coursenecessary among the number of its variables x, y, - - other things to specify Thus thereexistone-variablefunctions <t>(x), two-variable functions <t>(x, y), etc.

is v (c) For any fixed k the *(*) of 9.2.4. a one- ariable function (of w). But it can also be viewed as a two-variable function (of fc, w). is a (d) Forany fixed k the S*(K, DJ of (11 in 11.1.3. two-variable :A)
2 function (of ic, D). is a n-variable function (e) Forany fixed k the 3C*(ri, , r 5) of , r n ).1 (of n, 1 Although they do not appearin the abovecanonicalforms <t>(x, y). 1 We couldalsotreat k in (d) and k in (e)like k in (c),i.e. <t>(x), asa variable.)))

and xy, aretwo-variable functions. 1

Someexamples: l functions. (a) Thearithmetical operationsx + 1and x 2 areone-variable (b) The arithmetical operationsof addition and of multiplication x + y

11.2.3.

FUNCTIONALCALCULUS
<

89

x, y,

13.1.3.is equally necessary, in orderto describe function to It a the value specify for which specific choicesof its variables x, y, ' ' ' ) is defined at all. Thesechoices i.e.hesecombinations of t <t>(%> y>
Theexamples f (a)-(e)show someof the many possibilitiesor the domains of functions: hey may consist of arithmetical or of analytical entities,as T
form the domain of
<t>.

well as of others. Indeed: (a) We may considerthe domain to consist of all integernumbers, or equally well of all realnumbers. (b) All pairs of eithercategoryof numbersused in (a),form the domain. t * (c) The domain is the set Q of all objects which representhe plays of the gameT (cf. and 9.2.4.). (d) The domain consistsof pairs of a positiveintegerK and a set (e) The domain consistsof certain systemsof positiveintegers. A function is an arithmetical function if its variables are positive it is a numerical function if its variables arerealnumbers;t is a i integers; if its variablesaresets(as,e.g., in (d)). D< set-function Forthe moment we aremainly interested arithmetical and numerical in functions. We conclude this section an observationwhich is a natural conseby of our view of the concept f a function. This is, that the number o quence of variables, the domain, and the dependencef the value on the variao if two functions <, have the constitute the function as such: bles, and the same domain, and if <t>(x, y, same variables x, y, ) = 1 i this domain, then <, areidenticaln all respects. y> ) throughout

9.1.1.

D.

<f>

i.e.,

\\l/

'

\\l/

13.2.1. a function Consider

s chosen, ay as x = z so that say that has the maximum
<t>

The 13.2. OperationsMax and

<t>

Min

which

has realnumbersfor values))

Assumefirst that

is a one-variable function. If its variable can be

<f>(x

and assumes at x = x . it Observe that this maximum </>(z ) is uniquely determined; the i.e., maximum may be assumedat x = XQ for several z , but they must all furnish the same value <t>(xo).2 We denotethis value by Max <t>(x), the maxi<t>(x<>)

<t>(x')

for all otherchoices ',then we x

mum value

of <t>(x). then the concept f </>'s minimum, <A(x ), obtains, o If we replace by ^ it. and of XQ where assumes Again theremay beseveralsuch x , but they must all furnish the same value <t>(x ). We denotethis value by Min <t>(x), the minimum value of

^,

</>

</>.

viewed in parallel with the exposition of * Proof:Considertwo such a? , say x'and

The conceptof a function is closelyallied to that of a set, and

8.2.

the

aboveshould be
<K*'o')

Hence

') x'J. Then 4(si) *(x' and

4>(x'>))))

90
Min
<(>(x)

ZERO-SUMTWO-PERSONGAMES: THEORY Observe that there is no a priori guaranteethat either Max

1 exist.
<t>(x)

or

limitations of of consequence their continuity togetherwith the geometrical 8 to our their domains. At any rate we arerestricting considerations such functions, for which Max and Min exist. Let now have any number of variablesx,y,z, By sin, as y gling out oneof the variables,say x, and treatingthe others, z, w constants, e can view <t>(x, y, z, - - - ) as a one-variablefunction, of the w variable x. Hence e may form Max <t>(x,y,z, ) ), Min (z,t/, 2, with respect this x. to of course as in But sincewe couldhave donethis equally well for any one of the other it becomes variables y, z, necessary to indicatethat the operations t Max, Min were performed with respecto the variable x. We do this by ) instead of the incom), Min x <(z,y, z, writing Max* </>(z, y, z, Thus we can now apply to the function Max Min pleteexpressions ) anY ne of the operators ax*, Min z, Maxy , Min y , Max*, M <t>(x> y, z Min,, They areall distinctand our notation is unambiguous. This notation is even advantageousfor one variable functions, and we will use it accordingly; we write Max* <(#), Min* <t>(x) instead of the Max (x), Min 4>(x) of i Sometimest will be convenient or even necessary to indicate the domain S for a maximum or a minimum explicitly. E.g. hen the funcw x tion <t>(x) is defined alsofor (some) outsideof S,but it is desiredto form the maximum or minimum within S only. In such a casewe write

o t consists only of a finite number of elements,hen the existencef both for Max <t>(x) and Min <t>(x) is obvious. This will actually be the case most 2 functions which we shall discuss. For the remaining ones it will be a

If, however, the domain of

</>

over which the variable x may

run

13.2.2.

<t>

13.2.1.,

</>

''

>

' ' .
*

<t>,

</>.

i.e. 13.2.1.

Maxxin5 0(x),

Minxin5

<t>(x)

instead of Max* <(#),Min x <(z). In certainothercasest may be simplerto enumeratehe values of <f>(x) i t than to expresst>(x) as a function. We may then write < say o,6,
exist.
8

1E.g.f i

<t>(x)

ss x with

all

real numbers as domain,

3C*(,n,
> >
17

then neither Max

0(x) nor

Min

<f>(x)

2 Typical examples: The functions function 3C(ri,TZ) of

14.1.1.
0i

, r n ) of
*

11.2.3. (e)in 13.1.2.), the (or of

> >

T Typical examples: he functions K( ,

functions Min Tl

), Max-> K(

17

), Min- K( ,
*

>
17

) in

the 17.4.,

ablesof all thesefunctions are or j or both, with respectto which subsequent maxima and minima are formed. Another instance is discussed 46.2.1. in f espec.ootnote 1 on p. 384,where the mathematical background of this subject and its literature are considered. It seems unnecessary to enter upon thesehere, sincethe aboveexamplesare entirely elementary.)))

^/ ^(n, T2) v r^l^

Max fi

rt

^ 3C(n, r -l

ft,

t )ij r in

17.5.2. variThe

FUNCTIONALCALCULUS Max (a,

91
[Min*))

&,-)>[Min (a, 6,
</>(x,

)] instead of Max* <(z),

) is a function of the variables y, z, - - - ), Min* <f>(x, y, arestill funcx, y, z, , Max* <t>(x, y, z, tions, but of the variables y, z, only. Purely typographically,x is still presentin Max,</>(#, y, z, - ), Min* <t>(x, y, z, - - - ), but it is no longera variable of these functions. We say that the operations Max*, 2 Min* kill the variable x which appearsas their index. SinceMax* <t>(x, y, z, ) are still functions ), Min* 4>(x, y, z, of the variablesy, z, ,3 we can go on and form the expressions Maxy Min* <f>(x, y, z, Maxy Max* <t>(x, y, z, ), ), - - ), Miny Max* <(z,y, z, Min y Min* <t>(x, y, z, ), We

13.2.3. that while Observe

*,)
y,

could equallyform
Max* Maxy
4
<j>(x,

y,

z,

),

Max* Miny
x,
y

<t>(x,

z,

etc. or use two other variablesthan ;

a at all, i.e. constant.

variablesthan two (if thereareany). Infine, after having appliedas many operations ax or Min as thereare M in any order and combination,but precisely variablesof <t>(x, y, z, ) we obtain a function of no variables one for eachvariable x, y, z, -

(if

there are any) ; or use more

entirely as functional operations, each Max,, Min*, a one of which carries function into anotherfunction. 5 We have seenthat it we can apply several of them successively. In this lattercase is prima a t operations reapplied. facie relevant, in which orderhe successive Two But is it really relevant? Precisely: operationsaresaid to commute t of a object),he orderin if, in case their successivepplication (to the same which this is donedoes not matter. Now we ask Do the operationsMax*, all commute with eachotheror not? M Min*, Maxy, Min y , Max,, in,, We shall answerthis question. Forthis purpose we needuse only two variables,say x,y and then it is not necessarythat be a function of further
Min*, Maxy, Min tf ,

The o 13.3.1. discussions f 13.2.3. the basisfor viewing the Max*, provide

13.3. Commutativity

Questions

<f>

variablesbesides , y.6 x
1A

which kills a variable x is the definite integral: is a constant. 4>(x)is a function of x, but / <j>(x)dx 8 We treated But now that x has been as constant parameters in fy, z, the variables y, z, killed we release 4 Observe that if two or more operations are applied, the innermost appliesfirst and kills its variable; then the next one follows suit, etc 6 With one variable less,sincetheseoperations kill one variable each. 8 Further variables of <, if any, may be treated as constants for the purpose of this

1Of courseMax (a, b, ) [Min (a, b, among the numbers a, b,

well known operation

)] is simply

the

greatest [smallest]one

in analysis

f\\

13.2.2.

analysis.)))

92

ZERO-SUMTWO-PERSONGAMES: THEORY

a So we consider t Wo-variable function <t>(x, y). The significant questions of commutativity arethen clearlythese: Which of the threeequationswhich follow aregenerally true: Max,Maxy 0(x,y) = Maxy Max,<t>(x y), (13:1) Min, Miny <Kz, y) = Min y Min, <(x,y), (13:2) Max,Min y </>(x, y) = Miny Max,<t>(x, y).1 (13:3)
f

is shall seethat (13:1), aretrue, while (13:3) not; i.e., two any (13:2) Max or any two Min commute, while a Max and a Min do not commute in general. We shallalso obtain a criterion which determinesin which special Max cases and Min commute. This question of commutativity of Max and Min will turn out to be and decisivefor the zero-sum two-persongame (cf. 14.4.2. 17.6.). It 13.3.2. us first consider(13:1). ought to beintuitively clearthat Let Max,Maxy <f>(x, y) is the maximum of <(x,y) if we treatx,y togetheras one t variable; i.e.hat for somesuitable x , 2/0, </>(x , 2/o) = Maxx Maxy 0(x,y) and that for all x', y', <f>(x , y ) ^ <t>(x', y'). If a mathematical proof is neverthelesswanted, we give it here:Choose xo so that Maxy <f>(x, y) assumesits x-maximum at x = x , and then choose j/ so that <t>(xQ, y) assumesits y-maximum at y = y . Then))
We
()

/o)

= Maxy

</>(x

, y)

= Max,Maxy <(x,y),
</>(x',

and for all x', y 1

o, 2/0)

= Maxy </>(x

, y)

^ Maxy

y)

<t>(x', y').))

the This completes proof. Now by interchanging x, y we seethat Maxy Max,<(x,y) is equally the maximum of </>(x, y) if we treat x, y as onevariable. Thus both sides of (13:1) the same characteristic roperty, and have p therefore they areequal to eachother. This proves (13:1). of Literally the same arguments apply to Mm in place Max:we need use ^ consistentlyin placeof ^ . This proves(132). only : This device of treating two variables x, y as one,is occasionallyquite in convenient in itself. When we use it (as,e.g., 18.2.1., n,r 2, 3C(ri,r2) with in place our present x, y, <(x, z/)), we shall write Max,, <(x, y) and of y
At this point a graphical illustration may be useful. Assume that the domain of for x, y is a finite set. Denote, or the sake of simf plicity, the possiblevalues of x (in this domain) by , t and thoseof Then the values of </(x, y) corresponding all x,y in this to y by domain to all combinationsof x = , /, y = , $ can in a rectangular cheme: use a rectangle t rows and s be arranged s of We
<f>

Min,. <(x,2/). y

13.3.3. 1,

,$. i.e.

1,

1,

1,

1The combination Min, Max,, requires no treatment the above Max, Min y by interchanging x,
y.)))

of

its own, sinceit

obtains from

FUNCTIONAL CALCULUS))

93))

the former and columns,using the number x = 1, , t to enumerate the number y = 1, the , s to enumerate latter. Intothe fieldof intersectionof row x and column y to be known briefly as the field x, y we write the value <t>(x t y) (Fig.11).This arrangement, known in mathematics as a rectangular matrix, amounts to a complete of characterization the function <t>(x, y). The specific values <(x,y) arethe matrix elements.))
1)

2)

y)

s)

1)

1)) 0(1,

2)) $(1,

*>(!,y))
<fr(2, y})

^(l,
0(2,

))

2)

d(2, 1))

0(2, 2))

))

x)

4>(x, 1))

<t>(x

t 2))

0(z, y})

0(s,s])
-)

t)

d(J, 1))

4>(l

2)) Figure

<t>(t

y})

4>(L

a))

11.
of <(#, y) in the row x.

Now

MaXy

<t>(x,

y)

is the

maximum

Max is therefore the

maximum

x))

of the row

maxima.On the otherhand,

Max* <t>(x, y) is the maximum of <f>(x, y) in the column y. Maxy Max* <(#, y) is therefore the maximum of the column maxima. Our assertionsin concerning can now be stated thus: The maximum of the row maxima is the sameas the maximum of the column maxima; arethe absolute maxiboth mum of <t>(x, y) in the matrix. In this form, at least,hese t assertionsshould obtain be intuitively obvious. The assertions concerning (13:2) in the of same way if Min is put in place Max.

(13:1)

13.3.2.

The 13.4. Mixed Case. SaddlePoints the left-hand side of

consider (13:3). Using the terminology of 13.3.3. is the maximum of the row minima and the (13:3) two right-handsideis the minimum of the column maxima.These numbers are neitherabsolute maxima, nor absolute minima, and there is no prima t fade evidencewhy they should be generallyequal. Indeed,hey arenot.

13.4.1. us now Let

Two functions for which they

are different are given in Figs.12,13.

A)))

94))

ZERO-SUMTWO-PERSON GAMES: THEORY))

function for which they are equal is given in Fig. 14. (All these figures and of Fig. shouldbe readin the senseof the explanationsof Thesefigures as well as the entirequestion of commutativity of Max r and Min will play an essential olein the theory of zero-sumtwo-person i games. Indeed,t will be seenthat they representcertain games which are typical for some important possibilitiesin that theory (cf. w But for the moment we want to discussthem on their own account, ithout t any referenceo those applications.))

13.3.3.

11.)

18.1.2.).

t
1)

-s -

2))

t
row)

-a 2)

3))

2)

row
3)

minima)
j) 1)

1)

minima)
1)

1)

1)

j)

-1)
1)

2)

-1)
1)

1)

-1)

2)

-1)
1)

column) maxima) Maximum Minimum

3)
1)

-1)
1)

-1) -1) -1)

of row minima of column maxima

Figure
12.))

1 1

column) maxima) Maximum Minimum t

1)

1)

1)

of row minima of column maxima

Figure
13.))

1 1

-*-2))
row)
2)

minima)

1)

2)

-2) -1) -1)

1)

2)

-2) -1)

column)
maxima)

2)

Maximum of row minima Minimum of column maxima Figure

1 1

14.

to desirable discussthe relationshipof its two sides

is Since 13.4.2. (13:3) neither

Max* Miny
</>(#,

generally true, nor generallyfalse, it is

Miny Maxx </>(x, y), which illustrated the behavior of (13:3) a more fully. Figs. 12-14, to certain degree, ive somecluesas to what this behavior is likely to be. g Specifically:
y) y

(13:4)

(13 :A)

In all three figures the left-hand sideof (13:3) the first (i.e. o is expressionf (13:4)) ^ the right-hand side of (13:3) the (i.e. in second expression (13:4)).)))

FUNCTIONALCALCULUS

95

: (13B)

and the maximum of its column. (This happens to be the ele1 the left lower cornerfield of the matrix.) In the ment other figures 12,13, here (13:3) not true, there existsno is w such field. It is appropriate to introduce a generalconceptwhich describes the behavior of the field mentioned in (13B). We define accordingly: : Let <f>(x, y) be any two-variable function. Then x , 3/0 is a saddlepoint of if at the sametime </>(x, y ) assumesits maximum at x = x and <(x , y) assumesits minimum at y = 3/0The reason for the use of the name saddlepoint is this:maginethe I matrix of all x, y elements(x = y = a; cf. Fig.11) as an oreographical map, the height of the mountain over the field x, y being the value of <t>(x, y) there. Then the definition of a saddle point x , 2/0 describes indeedessentiallya saddleor pass at that point (i.e. the over field xo, 2/0) ; the row XQ is the ridge of the mountain, and the column i/o is the road (from valley to valley) which crosses ridge. this 1 in The formula (13:C*) also falls in with this interpretation. Figs.12,13show that a may have no saddle points at all. that possesses On the otherhand it is conceivable severalsaddle points. But all saddle points x , 2/0 i? they existat all must furnish the same value 0(x, 2/0)-2 We denotethis value if it exists all by Sax/v <(x,y), at the saddlevalue of </>(x, y).8 We now formulate the theorems which generalize the indications of (13 (13B). We denote them by (13A*), (13 and emphasize :A), : : :B*), that they arevalid for all functions <(x,y).
<f>

In Fig.14 where a is true there exists field in the matrix which contains simultaneously the minimum of its row

(13:3)

1,

,,

1,

13.4.3.

13.5.2.
<f>

<

(13 :A*) (13 :B*)

Always

Max,Min y <(x,y) ^

Min y

Max,#(x, y).

13.5.1.
Miny
<t>(x,

1All this more general mathematical theories involving extremal problems, calculus of variations, etc. Cf. M Morse: he Critical Points of Functions and the Calculus of Variations in T J the Large,Bull. Am. Math. Society,an .-Feb. pp.38cont.,and What is Analysis in 1929, Am. Math. Monthly, Vol. XLIX, 1942, p.358cont. the Large?, p 1This follows from (13 in Thereexists an equally simple direct proof: :C*) Considerwo saddlepoints , y o, say yj and t yi'. Then Min, *(*' '*) Max, *(*, y'o) fc *(*\", vl) y +(*;,yi) <^x' j/ Similarly *(*' y'') +(*;, j). Hence yi) yi'). 1Clearlythe operation Sa/ 0(x, y) kills both variablesx} y. Cf. 13.2.3.)))

We have Max* Min y <(x,y) = Mmy Maxz </>(x, y) if and only if a saddle point x , t/o of exists. Proofs the 13.5. o! Main Facts We define first two sets A*, B+ for every function <(x, 2/)2/) is a function of x; let A+ be the set of all those x for which c with is closelyconnected although not preciselya special aseof; certain
<#>

13.5.2.

i.e.:

a?

4,

4',

0(4,

- 0(4',

', ').

',

: ',

y)))

96

ZERO-SUMTWO-PERSONGAMES: THEORY
maximum

at x = XQ. Max,4>(x, y) is a function of y ; let B+ bethe setof all thosey for which this function assumesits minimum at y = 2/0. We now prove (13:A*), 13:B*). ( X :A*) Choose in A+ and y in B*. Then)) Proof of (13

this function assumesits

Max* Min y

: i.e.Max,Min

<f>(x,

Proof of the Assume that

= Min y <(zo,y) ^ <(zo,2/0) g Max, 2/0) = Min y Max* <t>(x, y), y) ^ Min y Max,<t>(x, y) as desired. o necessity of the existencef a saddle point in (13:B*):
<l>(x,

y)

<f>(x,

Max,Miny <t>(x,
Choose
XQ

y)

= Miny Max,<t>(x, y).

y) y)

in A*

and yo in

5*;then we have))
<f>(x,

Max,<(z,2/0) = Min y Max,<t>(x, = Max,Miny

f Henceor every xf

= Min y
y))

g Max,

<l>(x,

y<>)

= Min))
assumes its
maximum

i.e.
And

<t>(x

, i/o)

</>(x', 2/0)

so

</>(z, 2/o)

at x

=x.

for every 2/'

y') ^ Min y ^(a: , y) = Max,</>(x, 2/0) ^ *(^o,2/o),)) i.e. (x, 2/0) ^ *(^o,2/0 so *(^o,2/) assumesits minimum at y = 2/0. 0 ConsequentlytheseXD, 2/0 form a saddle point. o Proof of the sufficiency of the existencef a saddle point in (13:B*): Let XQ, 2/0 be a saddle point. Then)) Max,Min y </>(z, y) ^ Miny </>(z , 2/) = *(^o,2/o),
Min y

Max,

<f>(x,

y)

^ Max,<t>(x,

2/0)

= </>(z

, t/o),

hence

Max,Miny <^(a:,y) ^ *(z , 2/0) ^

Min y

Max,0(z,2/).

Combiningthis with (13 gives :A*) Max,Miny <t>(x, y) = <t>(x , 2/0) = Miny Max,<t>(x, y), and hencethe desiredequation. 13.5.2. considerationsof 13.5.1. some further results which The yield are worth noting. We assume now the existencef saddle points, i.e. o the validity of the equation of (13 :B*). Forevery saddle point a; ,
yo))

(13:C*)0(x, 2/0) (13:B*)

= Max,Miny <t>(x,

y)

= Miny Max,<t>(x, y).

sufficiency proof of

T w Proof: his coincides ith the last equation of the

in 13.5.1.)))

FUNCTIONALCALCULUS

97

x , 2/0 is a saddlepoint if and only if x belongsto A* and yo belongsto B*.1 Proof of sufficiency:Let x belongto A* and y* belongto B+. Then the in shows necessityproof of (13:B*)13.5.1. exactlythat this x , yo is a saddle point. L : Proof of necessity: et xo, 2/0 be a saddle point. We use (13C*).For

(13D*) :

every

x'

Min y

i.e. in M
at x

0(s', ) y

Similarly for every y' Max* <(>(x, i/O ^ Miny Max* <(x,y) = ^(x0| j/o) = Max* 0(x,2/0), Maxx <(x,2/0) ^ Maxz <t>(x, y') so Max* 0(x,y) assumesits minimum the at y = I/O- Hence belongsto B*. This completes proof. y the way, the limitations The theorems(13:C*) indicate,by ; (13:D*) of the analogy describedat the end of they show that our conidea of a saddle point is narrower than the everyday (oreographical) cept statesthat all saddles provided of a saddle or a pass. Indeed, 13 ( :C*) that they exist areat the samealtitude. And (13 states if we depict :D*) 2 are the setsA*, B+ as two intervals of numbers that all saddlestogether 8 an areawhich has the form of a rectangular lateau. p this sectionby proving the existence a saddle of We conclude casewill be seen kind of x, y and <(x,y). This special for a special point to be of a not inconsiderablegenerality. Let a function ^(x, u) of two variables x, u be given. We consider all functions /(x) of the variable which have values in the domain of u. Now we keepthe variable x but 4 of in place the variable u we use the function / itself. The expression is determinedby x,/;hence may treat^(x,/(ar)) a function of as we lK*/(z)) the variablesx, and let it take the place <(x,y}. of of We wish to prove that for these x, and ^(x,/(x)) in place x, y i t and <f>(x y) a saddle point exists;.e.hat

= x . Hence belongsto A+. x

<(x , t/)

= Min y 4>(z<>,y), ^ Max,Min y <t>(x, y) = <t>(x , ^ Min y <(x',i/) so Min y <(x,y) assumesits maximum
Q

y<>)

i.e.

i.e. 13.4.2.;

13.5.3.

(13E) :

Maxx Min/ ^(x,/(x)) = Min/ Maxx f(x,/(x)).))

UQ

a F Proof: orevery x choose with ^(x, u ) = Min u ^(x, w). This w depends on x, hence we can define a function / by w = /o(z). Thus ^/(x, /o(z)) = Min ^(x,u). Consequently Max* ^(x,/o(x))= Max* Min u ^(x,u).
M

integers, then this can certainly be brought about by two appropriate permutations of their domain. 3 The general mathematical concepts alluded to in footnote 1 on p. 95are free from theselimitations. They correspondpreciselyto the everyday ideaof a pass. 4 Thereaderis askedto visualize this: Although itself a function, / may perfectly well be the variable of another function.)))

no

saddlepoints at all. 1If the x, y are positive

Only

under our hypothesis

at

the beginning

of this

section! Otherwise there exist

98
A

ZERO-SUMTWO-PERSONGAMES: THEORY

fortiori,

Min/ Max, f(x,/(*)) ^ Max* Min u f(x, u).)) Now Min/ ^(z,f(x)) is the same thing as Min tt \\(/(x u) since entersinto this expression x, f(x), for which we only via its value at the one place i.e. may write u. So Min/ \\l/(x, f(x)) = Min ^(x, w) and consequently,)) Max* Min/ $(x, (#)) = Maxx Min ^(z, u). / (13G) : The (13:G) (13:F), togetherestablishthe validity of a ^ in (13:E). ^ in (13 holds owing to (13 :A*). Hencewe have = in (13E), .e.he : i t :E) proof is completed.
y

(13F) :

14.1.1. now proceedto the considerationof the zero-sum two-person We

we beginby using the normalized form. to this the game consistsof two moves: a According Player 1 chooses 2 chooses number r2 = number TI a 2, each , 0i, player ,# choicebeing made in completeignorance of the other, and then players 1 1and 2 get the amounts 3Ci(ri,r 2) and 3C2(Ti, T2), respectively. the Since game is zero-sum,we have, by

14.Strictly Determined Games 14.1. Formulation of the Problem

game.

Again

=!,

!,

11.4.
= 0.

3Ci(Ti, I) + JC(T,, I) T T
We

prefer to expressthis by writing

T2). 3Cl(Ti,T2) 35 3C(Ti,Tl), JCl(Ti, T2) = We shall now attempt to understand how the obvious desires of the T players 2 will determinethe events, the choices I, r 2 o must again be remembered, f course,hat TI, r 2 stand ultima analysi t not for a choice a move) but for the players' strategies; their entire (in \"theory\"or \"plan\" concerning the game. For the moment we leave it at that. Subsequentlywe shall also go the of \"behind\" TI, T2 and analyze the course a play. of The desires the players 2, are simpleenough. I wishes to make 5Ci(ri, T2) s= JC(TI, T2) a maximum, 2 wishesto make 3C2(Ti, T2) wants to maximize and 2 wants to minimize #C(TI, T2) a maximum;

-JC^i,

It

1,

i.e.

i.e.

14.1.2.

1,

1 i.e.

3C(n, Tt ). So the interests of the two playersconcentrate n the same object: o the one function 3C(Ti,T2). But their intentions are as is to be expected a in zero-sum two-person game exactlyopposite 1wants to maximize, 2 wants to minimize. Now the peculiardifficulty in all this is that neitherplayer has full control of the objectof his endeavor of 3C(ri, T2) i.e. f both o its variables TI, T2 1wants to maximize, but he controlsonly TI; 2 wants to minimize, but he controlsonly T2 What is going to happen? l Cf. (ll:D)in 11.2.3.)))

STRICTLYDETERMINED GAMES))

99))

The difficulty is that no particular choice say TI, needin itself make of, 3C(ri,r 2) eithergreator small. Theinfluence of r\\ on 3C(ri,r2) is, in general, of with the choice that no definite thing; it becomes only in conjunction the other variable, in this caser2 (Cf. the correspondingdifficulty in
a Observe that from the point of view of the player 1 who chooses as variable, say n, the other variable can certainly not be considered a chanceevent. The other variable, in this caser2, is dependentupon the \" in will of the otherplayer, which must be regarded the same light of ration\" own. (Cf. also the end of 2.2.3. 2.2.4.) and ality as one's At this point it is convenient to make use of the graphical We represent C(ri,r2) by a rectangular d 3 representationevelopedin of W matrix: e form a rectangle pi rows and 2 columns,using the number n= , #1to enumerate the former, and the number r 2 = , 2 t to enumeratehe latter;and into the field n,r 2 we write the matrix element The x, y, t, s therecorrespond 3C(ri,r 2). (Cf. with Figure in to our OC,TI, r2, 01, 2 (Figure 15).)))

as economics discussedin

2.2.3.)

14.1.3. 1,

13.3.3.

1,

11 13.3.3.

</>,

1)

2)

7>2)

1)

3C(1,1))
3C(2, 1))

JC(1,2))
3C(2,2))

3C(1 rj)) 3C(2,T2

))

3C(1. 00)
3C(2,0))

2)

7-1)

3C(ri, 1))

3C(n, 2))

3C(T1,T))

3C(n, 2

))

0i)

C(0i,1))

JC(0i, 2))
Figure

3C(#i.n))

3C(0i,0i))

15.

It ought to be understoodthat the function 3C(ri, r 2) is subjectto no restrictions w whatsoever;i.e., e are free to chooseit absolutely at will. 1
in the

Indeed, ny a

given function 5C(ri, r 2) definesa zero-sumtwo-persongame senseof (11 of :D) by simply defining
T 3Ci(T!, 2)

11.2.3.
JC(TI,r 2),

3C2(ri, r2)

s -3C(ri, r

2)

o above in the The (cf. 14.1.1.). desires f the players 1,2, as described last section,can now be visualized as follows:Both players are solely))
1Thedomain, of course,s prescribed: consistsof all pairs n, TI with n i It , , 0t. This is a finite set, so all Max and Min exist, cf. the end of 13.2.1.))) ;n =

1,

1,

100

ZERO-SUMTWO-PERSONGAMES: THEORY

in interested the value of the matrix elementJC(ri, r 2). Player 1 tries to t maximize it, but he controls only the row, i.e.he number n. Player 2 triesto minimize it, but he controlsonly the column, i.e.he number r2. t

of come this peculiartug-of-war. 1

We must now

attempt to

find

a satisfactoryinterpretation for the outand the Majorant

Gaifces 14.2. Insteadof attempting a directattackon the gameF itself for which we arenot yet prepared let us considerwo othergames, hich are t w of w closelyconnectedith F and the discussion which is immediatelyfeasible. The difficulty in analyzing F is clearlythat the player in choosingn r doesnot know what choice 2 of the player 2 he is going to face and vice F versa. Let us therefore compare with other gameswhere this difficulty doesnot arise. with F in every detail except We define first a gameFi,which agrees that of player 1 has to make his choiceof TI before player 2 makes his choice in T2, and that player 2 makes his choice full knowledge the value given of 2 T by player 1to TI (i.e. smove is preliminary to 2's move). In this game to Fi player 1 is obviously at a disadvantage as compared his position in the original gameF. We shall therefore call Fi the minor ant game of F. w We define similarly a second game F 2 which again agrees ith F in every t o detail excepthat now player 2 has to make his choicef r 2 beforeplayer 1 makes his choiceof r\\ and that 1 makes his choicein full knowledge 3 2's of the value given by 2 to r 2 (i.e. move is preliminary to move). In this game F 2 the player 1 is obviously at an advantage as compared to his positionin the gameF. We shall therefore call F 2 the majorant game of F. The introduction of thesetwo games FI, F 2 achievesthis: t ought to I be evident by common sense and we shall also establish it by an exact discussion that for FI, F 2 the \"bestway of playing\" the concept f o rational behavior has a clearmeaning. On the other hand, the game F f the liesclearly \"between\" two gamesFI,F 2 ; e.g.rom point of view FI is always lessand F 2 is always more advantageousthan F.4 Thus FI, F 2 to may be expected provide lower and upper bounds for the significant F d concerning . We shall, of course, iscussall this in an entirely quantities form. A priori,these \"bounds\"could differ widely and leave a precise considerable uncertainty as to the understandingof F. Indeed, prima fade this will seemto be the casefor many games. But we shall succeed in in this technique such a way by the introduction of certain manipulating 1Thepoint is, of course,hat this is not a tug-of-war. t Thetwo players have opposite interests, but the means by which they have\" to promote them are not in opposition to the o eachother. On the contrary, these \" means i.e. choices f n, TJ are apparently the independent. This discrepancycharacterizes entire problem.

14.2. Minorant The

1,

1's

i.e. 1's

1Thus Ti while extremely simple is no longer in the normalized form. * Thus TI while extremely simple is no longer in the normalized form. 4 Of course,o be precise should say \"lessthan or equal to\" instead of we t \"less,\" \"more than or equal to\" instead of \"more.\)

and

STRICTLYDETERMINED GAMES
further

101

answersto all questions. complete

devices as to obtain in the end a precise theory of F, which gives

14.3.1. us first considerthe minorant game IV After player 1 Let has made his choice the player 2 makes his choice 2 in full knowledge T TI of the value of n. Since desires to minimize JC(ri, T2), it is certain 2's i that
T he will choose2 so as to make the value of 5C(ri,r2) a minimum for this TI. In otherwords:When 1chooses particularvalue of TI he can alreadyforesee a with certaintywhat the value of 3C(ri,r2) will be. Thiswill be Min TjX(TI,r 2).l This is a function of TI alone. Now 1 wishesto maximize 3C(ri,r 2) and sincehis choice f TI is conducive to the value Min Ta 3C(ri,r 2) which depends o on TI only, and not at all on T2 so he will choose so as to maximize TI Min Tj JC(TI,T2). Thus the value of this quantity will finally
be))

14.3. Discussion the of

Auxiliary

Games

Tf

MinTj JC(TI,T2).2

Summingup:

(14:A:a)

MaxTj Min,t OC(TI,T2). I The good way (strategy) for 2 to play is this:f 1 has (14:A:b) a chosen definite value of TI,S then T2 shouldbe chosen belonging to the set Br ^ BTi being the set of those T2 for which 3C(ri,T2) assumesits minimum value Min Tj 3C(n, T2).4 On the basis of this we can statefurther:

The good way (strategy) for 1to play the minorant game Ti is to choose belongingto the set A, A being the set of TI, thoseTI for which Min Tj 3C(n, T2) assumesits maximum value

(14:A:c)

If both players 1 and 2 play the minorant game Fj well, i i.e.f TI belongsto A and T2 belongsto BTi then the value of JC(TI,T2) will be equal to
vi

= MaxTj MinTj JC(TI,T2).

all

TZ may not be uniquely 3C(ri, T2) may assume its Tj-mmimum for

1Observe that

will, however, Minr t

2 For the samereason as in footnote 1 above,the value of n may not beunique, but the value of Min Tf 3C(ri, ri) is the same for all n in question, namely the uniquely-defined maximum value Max Tl Min Tl 3C(n r 2).
8

3C(n,ri). (Cf. 13.2.1.)

be the samefor

theseT2, namely

determined:For a given n the T2-function severalvalues of T^ The value of 3C(n, T2)

the uniquely defined minimum

value

2 is informed of the value of n when calledupon to make his choice r 2, this is of of PI. It follows from our conceptof a strategy (cf. and end of that at this point a rule must be provided for 2'schoice T2for every value of n, irreof of whether or not the value chosenbelongs spective whether 1 has played well or not, to A. 4 In all, this n is treatedasa known parameter on which everything depends, including the set Brt from which n ought to be chosen.))) the
rule

4.1.2.

11.1.1.)

i.e.

102

ZERO-SUMTWO-PERSONGAMES: THEORY

The truth of the above assertionis immediatelyestablishedin the mathematical senseby remembering definitions of the setsA and BV and by the substituting accordinglyin the assertion. We leave this exercisewhich
common sense.
Fi

is nothing but the classical operation of \" substituting the defining for the defined\" to the reader. Moreover,the statement ought to be clearby

Theentirediscussionshouldmake it clear that every play of the game has a definite value for eachplayer. This value is the above Vi for the player 1and therefore Vi for the player 2. An even more detailedidea of the significance of Vi is obtained in this way: (14:A:d) Player 1 can, by playing appropriately,securefor himself a gain ^ Vi irrespective of what player 2 does. Player 2 for can,by playing appropriately,secure himself a gain ^ Vi, of what player 1does. irrespective of T (Proof: he former obtains by any choice TI in A. The latterobtains of t by any choice T^ in BT .* Again we leave the details to the reader;hey

arealtogetherrivial.) t

The above can be stated equivalently thus:

Player 2 can, by playing appropriately,make it sure that the gain of player 1 is ^ Vi, prevent him from gaming > Vi irrespectiveof what player 1does.Player 1can,by playing appropriately, make it sure that the gain of player 2 is g Vi, i.e. revent him from gaining > Vi irrespectiveof p what player 2 does.

(14:A:e)

i.e.

14.3.2.\" have carried the discussionof FI in rather profuse detail out We although the solution\"is a rather obvious one. That is, it is very likely vision of the situation will easily reachthe same that anybody with a clear of conclusions \"unmathematically,\" just by exercise common sense. Neverthelesswe felt it necessaryto discussthis caseso minutely because it is a prototype of severalothers to follow where the situation will be much less open to \" unmathematical \" vision. Also, all essential elementsof complicationas well as the bases for overcoming them are really present in this very simplest case. By seeingtheir respective positions clearly in this case,t will be possibleto visualize them in the subsequent, more i ones. And it will be possible,in this way only, to judge complicated, how much can be achievedby every particular measure. precisely Let us now considerhe majorant game F2. t of F2 differs from Fi only in that the roles players 1 and 2 are interr Now player 2 must make his choice 2 first and then the player 1 changed makeshis choice r } in full knowledge the value of r2 of of

14.3.3. :
Recall that of
n.)))

n must be chosenwithout

knowledge

of r 2,

while

r 2 is chosen in

full

knowledge

STRICTLYDETERMINED GAMES
But in sayingthat
it must

be remembered these players conservein the processheir that t functions 3Ci(ri, r2), aC (rj, r2), i.e. C(ri,r 2), 3C(ri,r2). That 2 respective 3 to to is, 1still desires maximize and 2 still desires minimize X(TI, r2). Thesebeing understood,we can leave the practically literal repetition

103 F 2 arisesfrom FI by interchanging the players1and 2,

of the considerationsof to the reader. We confine ourselvesto the significant definitions, in the form in which they apply to F 2 restating (14:B:a) The good way (strategy) for 2 to play the majorant game F 2 is to choose 2 belongingto the set JB, B being the set of r thoser2 for which MaxTf 3C(ri,r2) assumesits minimum value Min Tj MaxTj X(TI, r 2). I (14:B:b) The good way (strategy) for 1 to play is this:f 2 has a definite value of T2,* then TI should be chosen chosen belonging to the set A Ti, A T being the set of those r\\ for which 3C(ri,r2) assumesits maximum value MaxTi 3C(ri,r2).2 On the basis of this we can statefurther:

14.3.1.

(14:B:c)

i.e.if r2 belongsto
v2

If both players 1 and 2 play the majorant game F 2 well, B and TI belongsto A Tj then the value of 5C(ri,T2) will be equal to
).))

= Min tt MaxTi 5C(ri,r 2 Theentirediscussionshouldmake it clearthat

In orderto stressthe symmetry of the entirearrangement, repeat, we mutatis mutandis, the considerationswhich concluded 14.3.1. now They idea serve to give a more detailed of the significance of v 2. (14:B:d) Player 1 can, by playing appropriately,securefor himself a gain ^ v 2, irrespective of what player 2 does. Player 2 can, by playing appropriately, securefor himself a gain o i v 2, irrespectivef what player 1does. of T (Proof: helatterobtains by any choice r2 in B. Theformer obtains choice TI in A T^ Cf. with the proof, loc. it.) of c by any The above can again be stated equivalently thus :
Player 2 can, by playing appropriately, make it sure that the gain of player 1is g v 2, prevent him from gaining 1 1 is informed of the value of of T*when called upon to make his choice n this is the rule of r a (Cf.footnote 3 on p. 101). 2 In all this r a is treated as a known parameter on which everything depends,ncluding i the set A T|from which n ought to be chosen. 8 Remember that rj must be chosen without any knowledge of n, while n is chosen

F 2 has a definite value for eachplayer. This value player and therefore v 2 for the player

2.

every play of the game is the above v 2 for the

(14:B:e)

i.e.

with full knowledge

of

TI.)))

104

ZERO-SUMTWO-PERSONGAMES: THEORY

> v 2, irrespectiveof what player 1does.Player 1 can, by playing appropriately,make it sure that the gain of player 2 is ^ v 2, i.e. revent him from gaining > v 2, irrespective p of what player 2 does. and 14.3.4. discussionsof Fi and F 2, as given in The respectively,are in a relationship of symmetry or duality to each other; they obtain from eachother,as was pointed out previously (at the beginof by interchanging the roles the players 1and 2. In itself ning of 14.3.3.) to neither game FI nor F 2 is symmetricwith respect this interchange; indeed, of o this is nothing but a restatement f the fact that the interchange the modifies the players 1and 2 also interchanges two games Fi and F 2, and so both. It is in harmony with this that the various statements which we and t made in concerninghe good strategiesof FI and F 2, loc. (l4:A:b), (14:B:a), (14:B:b), cit. were respectively i.e.(14:A:a), t not symmetricwith respecto the players 1 and 2 either. Again we see: the An interchangef the players 1and 2 interchanges pertinent definitions o for Fi and F 2, and so modifies both.1 o It is therefore very remarkable that the characterizationf the value and of a play (vi for Fi,v 2 for F 2), as given at the end of 14.3.1

14.3.1. 14.3.3.,

14.3.1. 14.3.3.

said above, this is the same thing as assertingthat thesecharacterizations arestatedexactlythe sameway for Fi and F 2.2 All this is, of course, qually e clear immediateinspectionof the relevant passages. by in Thus we have succeeded defining the value of a play in the sameway for the gamesFI and F 2, and symmetrically for the players 1 and 2:in in and (14:A:c), (14:A:d), (14:A:e), (14:B:c), (14:B:d) (14:B:e) 14.3.1. and in 14.3.3., in spite of the fundamental difference of the individual this roleof each player in these two games. From this we derive the hope that the definition of the value of a play may be used in the same form for other gamesas well in particular for the game F which, as we know, occupiesa middle position between FI and F 2. This hope applies, of o course, nly to the concept f value itself, but not to the reasoningswhich o leadto it;thosewere specificto Fi and F 2, indeed different for Fi and for F 2, and altogether for f w impracticable F itself; i.e., e expector the future

loc. (14:A:d), (14:A:e), (14:B:c), (14:B:d), (14:A:c), (14:B:e), cit.(except for the formulae at the end of (14:A:c) of (14:B:c)) fully and are t symmetric with respecto the players 1 and 2. According to what was

i.e. 14.3.3.

morefrom (14:A:d), than (14:A:e), (14:B:d), (14:B:e)

held fixed, cf. 2 This careful consideration point deserves Naturally these two characterizations must obtain from eachother by interchanging the rolesof the players 1 and 2. But in this casethe statements coincide lsodirectly when no interchange of the players is made a at all. This is due to their individual symmetry.)))

1Observe that the original game T was symmetric with respectto the two players 1 and 2, if we let eachplayer take his function 3Ci(n, r a), 3Cs(n, r 2) with him in an interi change; the personalmoves of 1 and 2 had both the same charactern P. For a narrower concept of symmetry, where the functions 3Ci(n, T), JCs(ri, are

(14:A:b), (14:B:a), (14:B:b).

from

(14:A:a),

i.e.

14.6.

r)

STRICTLYDETERMINEDAMES G

105

These are clearly only heuristic indications. Thus far we have not even attempted the proof that a numerical value of a play can be defined in this manner for F. We shall now begin the detaileddiscussion which by this gap will be filled. It will be seen that at first definite and serious to but difficulties seem limit the applicabilityof this procedure, that it will be possibleto remove them by the introduction of a new device(Cf.

and 17.1.-17.3., respectively). 14.4.Conclusions 14.4.1. have seen that a perfectly plausibleinterpretation of the We value of a play determinesthis quantity as Vi = Max Min Tj OC(ri,r 2), T( v 2 = Min Tj Max 3C(ri,r2), Ti 1 for the games FI, F 2, respectively,as far as the player 1is concerned. Sincethe game FI is less advantageousfor 1 than the game F 2 in FI he must make his move prior to, and in full view of, his adversary,while in conclusionthat the value F 2 the situation is reversed it is a reasonable not greaterhan) the value of F 2. of Fi is less than, or equal to (i.e. certainly t One may argue whether this is a rigorous\" proof.\" The questionwhether but it is, is hard to decide, at any ratea close analysisof the verbal argument involved showsthat it is entirely parallel to the mathematical proof the of the same propositionwhich we already possess. Indeed, proposition in question,
w coincides ith (13 in :A*) T2.) 3C,TI, Insteadof ascribingVi,

14.7.1.

^ V2 13.4.3. (The
Vi

tf>,

x, y there correspondto

our

v 2 as values to two games FI and F 2 different them with F itself, under suitable from F we may alternatively correlate

the \" o assumptionsconcerning intellect\" f the players 1and 2. that the rules of the game F prescribe eachplayer must make Indeed, of of choice(his personal move) in ignorance the outcome the choice his his adversary. It is neverthelessconceivablethat one of the players, of \" that say 2, finds out\" his adversary; i.e., he has somehowacquired the 2 as to what his adversary's strategy is. The basis for this knowledge not concern it may (but neednot) be experience from us; knowledgedoes At any rate we assume that the player 2 possesses this previous plays. knowledge. It is possible,of course,that in this situation 1 will change his strategy; but again let us assume that, for any reasonwhatever, he 8 does not do it. Under theseassumptionswe may then say that player 2 \" has found out\" his adversary.
1For player 2 the values are consequently vi, the strategy is just the actual which is in the normalized form at choice the unique personalmove of the player. Remember how this normalized form was derived from the original extensive form of the game; consequently it appearsthat this choice equally to the strategy in the original game. corresponds 1For an interpretation of all theseassumptions! cf. 17.3.1.)))
* In the game T

v.

106

ZERO-SUMTWO-PERSONGAMES: THEORY

Similarly, we may visualize the oppositepossibility,that player 1 has found out\" his adversary. Then conditionsin F become exactlythe same of as if the gamewere F 2; and henceall discussions 14.3.3. literally. apply In the light of the above we can say: Thevalue of a play of the game F is a well-definedquantity if one of the \" a following two extremessumptionsis made:Either that player 2 finds \" out\" his adversary, or that player 1 finds out\" his adversary. In the case first case the value of a play is Vi for and vi for 2;in the second the value of a play is V2 for 1and V2 for 2. 14.4.2. discussionshows that if the value of a play of F itself This without any further qualifications or modifications can be defined at all, then it must lie between the values of Vi and v 2 (We mean the values i for the player I.e.f we write v for the hoped-forvalue of a play of F t itself (for player 1),hen theremust be
\"

In this case, onditionsin F become c exactlythe sameas if the game were of apply Fi,and henceall discussions 14.3.1. literally.

1,

1.)

Vi

5jj

V2

The length of this interval,

which

is still available for v, is

A = v2 Vi ^ 0. the At the same time A expresses advantage which is gained (in the gameF) by \"finding out\" one'sadversary instead of being \" found out\"

Now the game may be such that it doesnot matter which player \" finds that the advantage involved is zero. According out\" his opponent; if to the above, this is the case and only if

1 by him.

i.e.,

= = V2
).))

or equivalently
Vi

V Or, if we replacei, v 2 by their definitions: MaxTj MinTj 3C(ri,r 2) = Min T2MaxTj JC(TI,r 2

that the game F is strictly determined if and only if a saddle point of 3C(ri,T2) exists. 14.5. Analysis of Strict Determinateness Let us assume the game F to be strictly determined; that a i.e. saddle point of 3C(ri,rj) exists. 1Observe that this expressionfor the advantage in question applies both players: for Theadvantage for the player 1 is v vi; for the player 2 it is ( ~vi) ( v 2) and these two expressions equal to eachother, i.e. are to

these then possesses properties, we call it strictly determined. in Thelast form of this conditioncallsfor comparisonwith (13:3) 13.3.1. of 4.1. 5.2. and with the discussions 13. -13. (The 0,x, y there again correin t spond to our 3C, 7i, r2). Indeed,he statementof (13:B*) 13.4.3. says
If the game F

14.5.1.

A.)))

STRICTLY DETERMINEDAMES G
it
will

107

it In this case is to be hoped considering analysisof 14.4.2. the that

be possibleto interpret the quantity

v

= Vi = v 2

as the value of a play of r (for the player 1). Recalling the definitions of and in Vi, v 2 and the definition of the saddlevalue in 13.4.3. using (13:C*) we see 13.5.2., that the above equationmay alsobe written as v = Max Min Tj 3C(7i,r 2) == Min Tj Max^ OC(TI,r 2) Ti

3C(ri,T2). and at the endof 14.3.3. the By retracing steps madeat the end of it is indeed not difficult to establishthat the above can be interpreted as the value of a play of F (for player

= Sar/rj

14.3.1.

(14:A:d),(14:A:e), (14:B:c), Specifically:(14:A:c), (14:B:d), (14:B:e) and where of 14.3.1. 14.3.3. they apply to FI and F 2 respectively,can now first the equivalent of (14:A:d) nd a be obtained for F itself. We restate
Player 1 can,by playing appropriately,securefor himself a gain ^ v, irrespectiveof what player 2 does. Player 2 can, by playing appropriately,securefor himself a gain ^ v irrespectiveof what player 1does. In order to prove this, we form again the set A of (14:A:a)n i in Theseareactually the setsA+, B+ of and the set B of (14:B:a)14.3.3. to (the corresponds our 5C). We repeat: (14:D:a) A is the setof thoseTI for which Min Ta 3C(ri,r 2) assumesits f maximum value; i.e.or which Min Ti JC(ri, T2) = Max Min rj 3C(n, r 2) = v. Tj B is the set of those r 2 for which MaxTj 3C(ri,r 2) assumes (14:D:b) f its minimum value; i.e.or which MaxTi 3C(ri,T2) = Min Tj MaxTi JC(TI,n) = v. is easy: Now the demonstrationof (14:C:d) Let player 1 choose from A. Then irrespectiveof what player 2 TI does, for every r 2, we have X(TI, r 2) ^ Min Tj JC(TI,r 2) = v, gain is ^ v. r Let player 2 choose 2 from B. Then, irrespectiveof what player 1 does, for every n, we have JC(n,r 2) ^ MaxTj 3C(ri T2) = v, i.e. gain is ^ v and so 2*s gain is ^ v. the This completes proof. and We pass now to the equivalent of (14:A:e) (14:B:e). Indeed, as (14:C:d)formulated above can beequivalently formulated thus: (14:C:e) Player 2 can, by playing appropriately,make it sure that the gain of player 1is v, prevent him from gaining > v of what player 1does.))) irrespective

1).

(14:B:d):
(14:C:d)

14.3.1.

13.5.1.

<t>

i.e.

1's i.e.,

i.e.

1's

: i.e.

108

ZERO-SUMTWO-PERSONGAMES: THEORY Player 1 can, by playing appropriately, make it sure thai

the gain of player 2 is ^ v i.e. resenthim from gaining p o > v irrespectivef what player 2 does. establishsatisfactorilyour interpretation v as the of and (14:C:d) (14:C:e) value of a play of T for the player and of v for the player 2. We n 14.5.2. consider ow the equivalentsof (14:A:a), (14:B:a), (14:A:b),

1,

in it is reasonable define a good way for 1 to Owing to (14:C:d) him a gain which is greater to play the gameF as one which guarantees o than or equal to the value of a play for irrespectivef what 2 does; a choiceof TI for which X(TI, T2) ^ v for all r2 This may be equivalently statedas MinTj 3C(rj,r2) ^ v. Now we have always Min Tf 3C(ri,T2) ^ MaxTi Min Tj 3C(ri,r 2) = v. the Hence above condition for TI amounts to Min T 3C(ri, T2) = v, to r\\ being in A. in (by (14:D:a) in it is reasonable define the good way for to Again, by (14:C:d) 2 to play the game T as one which guarantees him a gain which is greater than or equal to the value of a play for 2, irrespectiveof what 1 does; i.e. a choice r 2 for which 3C(ri,T2) ^ v for all TI. That is, JC(TI,T2) ^ v of for all TI. This may be equivalently stated as MaxTj 5C(Ti,TZ) ^ v. Now we have always MaxTj 3C(Ti, T2) ^ MinTj MaxT OC(TI, T2) = v. the Hence above conditionsfor T2 amounts to MaxTi OC(TI,T2) = v, i.e.by ( in to T2 being in B. (14:D:b) So we have: (14:C:a) The good way (strategy) for 1 to play the game F is to choose TI belongingto A, A beingthe set of (14:D:a) in any

(14:B:b).

14.5.1.

1,

i.e.

14.5.1.) 14.5.1.

i.e.

14.5.1.) 14.5.1.

(14:C:b)

The good way (strategy) for 2 to play the game F is to chooseany T2 belongingto B, B being the set of (14:D:b)
in 1 14.5.1.

Finally our definition of the good way of playing, as stated at the beginning of this section,yields immediately the equivalent of (14:A:c) or

both players 1 and 2 play the game F well i.e.f r\\ i to belongsto A and T2 belongs B then the value of 3C(n, T2) will be equal to the value of a play (for 1), .e. v. i to and We add the observationthat (13 :D*)in 13.5.2. the remark concerning in the setsA, B before (14:D:a), (14:D:b)14.5.1. give this: together T (14:C:f) Both players 1and 2 play the gameF well i.e.I belongs to A and T2 belongsto B if and only if TI, T2 is a saddle point
T of 3C(T,, 2). 1 Sincethis is the game F each player must make his choice(of knowledge of the other player's choice(of TI or n). Contrast this in for Ti and with (14:B:b) fo* Tf

(14:B:c): (14:C:c) If

or r t )

without

14.3.1.

14.3.3.

with

in (14:A:b)

.)))

GAMES STRICTLYDETERMINED
The 14.6. Interchange of Players. Symmetry in 14.5.1. settle 14.6. (14:C:a)-(14:C:f) and 14.5.2. everything

109

as far as the strictly determined two-person games are concerned.In this 14.3.3. Ti, T2 we derived for let connection us remark that in 14.3.1.,

than those of 14.5.2. in were of a much more heuristic character favor of (14:C:d), (14:C:e). a Our use of the function 3C(ri,r 2) = 3Ci(ri,r 2) implies certain asymmetry a specialrole. It ought the of the arrangement; player 1is thereby given to be intuitively clear, owever, that equivalent results would be obtained h r if we gave this specialole to the player 2 instead. Sinceinterchanging the players 1and 2 will play a certain rolelater,we shall nevertheless a give of brief mathematical discussion this point also. t Interchanginghe players 1 and 2 in the game T of which we need not assume now that it is strictly determined amounts to replacingthe 1 functions 5Ci(ri, r2), 3C (ri, r 2) by 3C2(r 2, TI), JCi(r , ri). 2 It follows,there2 2 the means replacing function 3C(r i, r 2) by 3C(r 2, r i) fore, that this interchange Now the changeof sign has the effect of interchanging the operations the Max and Min. Consequently quantities

(14:B:e) (14:B:d), from (14:A:a), (14:A:b),(14:B:a), (14:A:e), (14:A:d), for 14.5.2. T itself we obtained (14:C:a), while in 14.5.1., (14:B:b) This from (14:C:e). is an advantage sincethe argu(14:C:b) (14:C:d), in favor of (14:A:a), 14.3.3. ments of 14.3.1., (14:A:b), 14:B:a), ( (14:B:b)

14.5.1.,

? MaxTi Min Tj OC(T], 2) = Vi, Min rf Max 5C(ri,T2) = v 2, Ti n becomeow as defined in 14.4.1. = -Min MaxT|JC(r2, TI) MaxTi Min Tj [-3C(> TI)] 2, Ti = Mint| MaxTi JC(TI,r 2) 8 = 2 Min Tj Max [ 3C(T, TI)] = Max Min Ti OC(r, TI) 2 Tj Ti

v2

= -Max Min Tj JC(TI,T2) 8 = -VL Ti

So Vi,
1

v 2 become v 2,
A

Vi.

= v2

Hencethe value of
Vi

= (-Vi)

- (-v
r

2)

There This is no longer the operation of interchanging the players used in we were only interested in the arrangement and the state of information at eachmove, and the players 1 and 2 wereconsideredas taking their f unctions 3Ci(n, r 2) and 3Cj(ri, TI) unaffected by with them (cf.footnote 1 on p. 104). In this sense was symmetric, that interchange At present we interchange the rolesof the players 1 and 2 completely, even in their f unctions 3Ci(n, r a ) and 3C (Ti, r 2 2 * We had to interchange the variables r t sincer\\ represents the choice player 1 of of and T2the choice player 2. Consequently it is now T which has the domain , 0i. Thus it is again true for 3C*(r2,TI) as it was before for3C*(ri, TJ) that the variable before the comma has the domain 1, , ft and the variable after the comma, the domain , 023 This is a mere change of notations: Thevariables TI, r* are changed around to r s, TI. 4 This is in harmony with footnote 1 on p. 105, s it should a

14.3.4.

(id.).

i.e.

).

n,

1,

1,

be.)))

110

ZERO-SUMTWO-PERSONGAMES: THEORY

it s is unaffected, 1 and if F is strictly determined, remainsso, incethis propv erty is equivalent to A = 0. In this case = Vi = v 2 becomes
V

Vi

= ~ V2.

remain 14.5.2. the same when the players 1and 2 areinterchanged.

N 14.7. on -strictly

It is now easy to verify

in that all statements(14:C:a)-(14:C:f) Determined Games

14.5.1.,

All 14.7.1. this disposescompletelyof the strictly determined games, but of no others. Fora game I which is not strictly determined we have in A > i.e. such a game it involves a positive advantage to \"find out\" one's adversary. Hencethere is an essentialdifference between the t results, i.e.he values in FI and in F 2, and therefore also betweenthe good of 14.3.3. ways of playing thesegames. Theconsiderations 14.3.1., provide 14.5.2. therefore no guidance for the treatment of F. Thoseof 14.5.1., of do not apply either,sincethey make use of the existence saddle points

of 3C(ri,r2) and of the validity

of))

MaxTj Min Tj JC(71, r 2) = Min Tj MaxTi 3C(ri,r 2

),))

in the inequality at the beginningof 14.4.2. According to this, the value v c of a play of F (for the player 1) if such a concept an be formed at all in this generality,for which we have no evidenceas yet 2 is restrictedy b
Vi

o i.e. f F being strictly determined. Thereis, of course,someplausibility

g V g V2.

But this still leaves an interval of length A = v 2 Vi > open to v; the and, besides, entiresituation is conceptuallymost unsatisfactory. Sincethere is a positive One might be inclined to give up altogether: in \"finding out\" one'sopponent in such a game F, it seems advantage to plausibleto say that thereis no chance find a solution unlessonemakes 3 somedefinite assumptionas to \"who finds out whom,\" and to what extent. t a We shall seein 17.hat this is not so, nd that in spite of A > a solution can be found along the same lines as before. But we propose first, certain gamesF with A > 0, without attacking that difficulty, to enumerate and others with A = 0. The first which are not strictly determined will be dealt with briefly now; their detailed investigation will be underT taken in 17.1. he second which are strictly determined will be anadetail. lyzed in considerable 14.7.2. Sincethere existfunctions 3C(ri, r 2) without saddle points (cf. 13.4.2.; 13.4.1., the 4>(x, y) there, is our 3C(ri, r 2)) there exist not strictly those examplesi.e. determinedgamesF. It is worth while to re-examine 1This is in harmony with footnote 1 on p. 106,as it should be. 2 Cf. however, 17.8.1. * In plainer language:A > means that it is not possible this game for eachplayer in
simultaneously know just how than to be cleverer clever eachparticular his opponent. Consequently player
is.)))

it

seemsdesirable to

STRICTLYDETERMINED GAMES

111

the functions describedby the matrices of Figs.12, 3 p. 94 in the 1 on light of our presentapplication. That is, to describe explicitlythe games to which they belong. (In eachcase, eplace <t>(x, y) by our JC(TI, r a ), r 2 r beingthe column number and n the row number in every matrix. Cf. also Fig.15on p.99). Fig. This is the game of \"MatchingPennies.\"Let for TI and for T2 1be \"heads\"nd 2 be \"tails,\" then the matrix element as the value 1 a h if n, r 2 \"match\" are equal to eachother and if they do not. So player 1\"matches\" a player 2:Hewins (oneunit) if they \"match\" nd he loses (oneunit), if they do not. S Fig.13Thisis the game of \"Stone, Paper, cissors.\"Let for n and for r 2 1be \"stone,\" be \"paper,\" 2 The distribution of and 3 be \"scissors.\" 1 elements and 1over the matrix expresses \"paper\"efeats\"stone,\" that d 1 \"scissors\" defeat \"paper,\"\"stone\" defeats \"scissors.\"Thus player 1 wins (one unit) if he defeatsplayer2, and he losesone ( unit) if he is defeated. Otherwise(if both playersmake the same choice)he gameis tied. t 14.7.3. two examplesshow the difficulties which we encounter These in a not strictly determinedgame, in a particularly clear form; just because of their extremeimplicity the difficulty is perfectly isolated here,in vitro. s The point is that in \"MatchingPennies\"nd in \"Stone, aper, cissors,\" a P S any way of playing i.e. ny TI or any T2 is just as good as any other: a o Thereis no intrinsic advantage or disadvantagein \"heads\"r in \"tails\" nor in \"stone,\" or \"scissors\" se. Theonly thing which perse, per \"paper\" matters is to guesscorrectlywhat the adversaryis going to do; ut how are b we going to describe without further assumptionsabout the players' that

12: :

i.e.

1,

2 \"intellects\"? There are, of course,more complicatedgames which are not strictly determined and which are important from various more subtle, technical viewpoints (cf. 18., 19.).But as far as the main difficulty is concerned, a the simplegamesof \"MatchingPennies\"nd of \"Stone, aper, cissors\" S P

areperfectly characteristic.

14.8. Analysis Program of a Detailed

of Strict Determinateness

14.8. the strictly determined games T for which our solution While case is valid arethus a special only, oneshould not underestimatethe size of the territory they cover. The fact that we are using the normalized It form for the game F may tempt to such an underestimation: makesthings look more elementarythan they really are. One must remember the that w in T2 represent trategies the extensiveform of the game, hich may be TI, s a of a very complicatedstructure, s mentionedin 14.1.1. In orderto understand the meaning of strict determinateness, is it therefore necessaryto investigate it in relation to the extensiveform of the the game. This bringsup questionsconcerning detailednature of the moves,
1\" Papercovers stone, scissorsut the paper,stone grinds the scissors.\" c the 2 As mentioned before, e shall show in that it can be done.))) w

17.1.

112
in

ZERO-SUMTWO-PERSONGAMES: THEORY

chanceor personal the stateof information of the players,etc.; i.e. e w on cometo the structural analysisbased the extensiveform, as mentioned
in We areparticularly interested thosegamesin which eachplayer who makes a personal move is perfectly informed about the outcomeof the o choicesf all anteriormoves. These games were already mentioned in and t 6.4.1. it was stated therethat they aregenerallyconsideredo be of a rational character.We shall now establish this in a precise particular sense, y proving that all such games are strictly determined. And this b will be true not only when all moves are personal,but also when chance

12.1.1.

moves too

15.1. Statement of Purpose. Induction 15.1.1. wish to investigate the zero-sum two-persongamessomewhat We further, with the purpose of finding as wide a subclass among them as w possible in which only strictly determined gamesoccur;i.e. here the
quantities
vi v2

arepresent. 15. ames ith Perfectnformation w G I

= MaxTj MinTt 3C(ri,r2), = Min Tj Max JC(TI,r 2)

Ti

of

14.4.1. which
fulfill

turned out to be so important for the appraisal of the

Vi

game

= V 2 = V.

We shall show that when perfect information prevails in F i.e. hen w and the end of 14.8.) preliminarity is equivalent to anteriority (cf. then F is strictly determined. We shall also discuss the conceptual igs nificance of this result (cf. w Indeed, e shall obtain this as a special r caseof a more general ule concerningi, v 2, (cf. v We begin our discussionsin even greater generality,by consideringa perfectly unrestrictedgeneraln-person game T. The greatergenerality will be useful in a subsequentinstance. Let T be a generaln-person game,given in its extensiveform. We shall consider ertainaspectsf F, first in our original pre-set-theoretical c o and then translateeverything into the par7., terminology of 6., (cf. tition and set terminology of 9.,10.cf. et sequ.).The reader will ( probably obtain a full understandingwith the help of the first discussion a w alone; nd the second, ith its ratherformalistic machinery, is only undertaken for the sakeof absolute rigor,in orderto show that we are really s proceedingtrictly on the basis of our axiomsof We considerhe sequence all moves in F:SfTli, 3T1 , t of 2 , 3TI,,. Let us fix our attentionon the first move, 9Tli, and the situation which exists at the moment of this move. Sincenothing is anteriorto this move, nothing is preliminary to it of either; the characteristics this move depend on nothing, they are constants. This appliesin the first place the fact, whether 9fRi is a chance))) to

6.4.1.

15.8.).

15.5.3.).

15.1.2.

15.1.),

15.2.,

10.1.1.

i.e.

GAMES WITH PERFECT INFORMATION

113 6.2.1.

move or a personalmove ; and in the latter to case, which player9fTli belongs, to the value of fci = 0,1, , n respectively,in the senseof And it appliesalso to the number of alternatives a\\ at and for a chance move (i.e. when fci = 0)to the values of the probabilities pi(l), , p\\(ct\\). The result of the choice Sflli chance personal is a <TI = at or , a\\. Now a plausible step suggests itself for the mathematical analysis of the game T, which is entirely in the spirit of the method of \"complete induction\" widely used in all branches of mathematics. It replaces, if the analysis of T by the analysis of other gameswhich contain successful, one move lessthan T.1 This step consistsin choosinga *i = , cm and denotingby T9i a game which agrees that with F in every detail except the move 9Tli is omitted, and instead the choice is dictated the rules <n (by of the new game)the value ai = 9\\* T9i has, indeed, nemove less than o \" \" 3 F Itsmoves are9TC 2, , 3TI,. And our inductive step will have been successfulif we can derive the essentialcharacteristics F from thoseof of all rfi , *i = , It must be noted, however, that the possibilitiesof forming F i are dependent upon a certain restriction on F. Indeed, very player e who makes a certainpersonalmove in the gameT9i must be fully informed about the rulesof this game. Now this knowledge consistsof the knowledge at choice of the rules of the original gameF plus the value of the dictated 91. Hence 9i can be formed out of F without modifying the rules T SfTli, which govern the player's state of information in F only if the outcome at of the choice SfTli, by virtue of the original rules F, is known to every at any personalmove of his 3TC2, 9fRi must be prelimplayer , 9TI,; this: We inary to all personalmoves 9dl 2, , 311,. restate

i.e.

3fTCi

1,

1,

1, 15.1.3. i.e.

i.

i.e.

(15:A)

property: m SfTli is preliminary to all personal oves 9Tl2, (15:A:a)

T9i can be formed without essentially modifying the the structureof F for that purpose only if F possess following
*

9Rr.

1 I instead of v. Repeatedapplication of this \"inductive\" step if have v the feasibleat all will reduce game Tto one with steps; to oneof fixed, unalterable in a outcome. And this means, of course, completesolution for (Cf.(15:C:a) * E.g. is the game of Chess, nd a\\ a particular opening move F at choice Dili of a Then r^ is again Chess, ut beginning with a move of the charb \"white,\" player acter of the secondmove in ordinary Chess a \"black,\" player 2 and in the position createdby the \"opening move\" a\\. This dictated \"opening move\" may, but neednot, be a conventional one (like E2-E4). The same operation is exemplified by forms of Tournament Bridge where the

I.e.

i.e. r.

i.e.

1.

i.e.

15.6.1.)

in (This is done, e.g., Duplicate Bridge.) was originally personal (of \"white,\" In the first example, the dictated move in example it was originally chance(the \"deal\. player 1); the second In some games occasionally\"handicaps\" are used which amount to one or more
9TCi

\"umpires\"

assign

the

players

definite

known

and

previously

selected

\"hands.\"

such

o , v 1 and indicate the dependencen 9\\\\ , STl*. , 3Hji,. But we prefer the simpler notation 3fRj, writing SJTlf 1, in f we use the special orm of dependence the This is the terminology of ; like we must state (15:A:a) this: senseof Using the general description of For every personalmove 31k, * 2, , r, the set contains the function b e.g. y 4

* We should really use the indices

operations.

7.2.1.

1, i.e. 6.3.

7.2.1. *

<n.)))

114

ZERO-SUMTWO-PERSONGAMES: THEORY
The 15.2. Exact
Condition

(First

Step)

15.1.3. We 16.2.1. now translate 15.1.2., into the partition and set The of 9.,10., also the beginning of 15.1.2.). notations of (cf. terminology be used. 10.1. therefore will and di consistsof the one set 12 ((10:1in 10.1.1.),it is a subpartition :f) hence of :a) ((10:1in 10.1.1.); (Bi too consistsof the one set (the others being empty) .** That is :
(&i

ft

12

for preciselyone fc, say k for all ?^ fci.

fc

= fci,

This fci = 0,1,

If
fci

fc

tition but Bi(ki) = 12). HenceSDi(fci) too consists of the one set 12. And for i 7* fc], the >i(fc) which is a partition in Bi(fc) = ((10:A:g)n must be empty. So we have preciselyone AI of Cti, which is 12, and for fci = ,n one Di in all SDi(fc), which is also 12;while for fci = thereareno precisely Di in all >j(fc). The move 3TZi consistsof the choiceof a Ci from Ci(fcj); by chanceif = by the player fci if fci = fci , n. C\\ is automatically a subset of the unique A\\( = 12) in the former case,and of the unique DI(= 12) in the latter. The number of these Ci is c*i (cf. particularly footnote 2 on 70); and sincethe AI or DI in question is fixed, this a\\ is a definite

i i.e.f the move is personal then &i is also a subparThis of a)i(fci),((10:1in 10.1.1. was only postulatedwithin 5i(fci), :d) 3

1,

,n

determinesthe character OTijit is the fci of 6.2.1. of

,n

10.1.1.)

1,

0;

1,

constant, ai is the number of alternatives at 'JTli, the a\\ of 6.2.1. 15.1.2. and TheseCi correspondto the a\\ = 1, and , a\\ of 15.1.2., we denote 4 them accordingly Ci(l), shows by , Ci(ai). Now (10:1 in 10.1.1. :h) as is readily verified that Ct 2 is also the set of the Ci(l), , Ci(ai),i.e. equal to So far our analysishas been perfectly general, valid for m\\ (and to a s certain extentfor ffil 2) of any game F. The reader hould translate these into everyday terminology in the senseof 8.4.2. 10.4.2. and properties We pass now to iy. This should obtain from F by dictatingthe move = as describedin 15.1.2. putting At the same the Sflli by ' ,9\\. This meanstime the moves of the game arerestricted 9fTl2, to that STC,.

p.

9.1.5.,

d.

<j\\

This (Bi is an exceptionfrom (8:B:a) in cf.the remark concerning this (8:B:a) 1 on p. 63,and alsofootnote 4 on p. 69. 8 Proof:ft belongs to Q,\\, which is a subpartition of (Bij hence ft is a subset of an element of (Bi. This element is necessarilyequal to ft. All other elements of (Bi are therefore disjunct from (cf. empty. s Cli, S)i(fci)unlike i (cf.above) must fulfill both (8:B:a), in hence (8:B:b) both have no further elements besides ft. 4 and 9.1.5.))) , Cli(ai) of They represent the alternatives Cti(l),
1

in footnote

8.3.1.;

i.e. 8.3.1.),

8.3.1.;

6.2. 9.1.4.,

GAMES WITH PERFECT INFORMATION

115

element w
but

And the partitions enumeratedin 9.2.1. are ft, 1 (and restricted thosewith * = 2, to v + 1for Ct). , v, 15.2.2. now come the equivalent of the restrictionof 15.1.3. We to

which

is restricted Ci(ffi). to

t representshe actual play can no longervary over all

K

Ci(fri). Therefore all those sets which figured in the description of F and which were subsets of Q, must be made over into subsets of C}(9\\) and the partitions into partitions within Ci(ffi) (or within subsets of Ci(frj)). Howis this to

15.2.1. is dependentupon a certain restriction F. on As indicated, e wish to restricthe play i.e. within v w t

The possibilityof carrying out the changesformulated at the end of

be done? fall The partitions which make up the descriptionsof F (cf.9.2.1.) into
two
K

assume,of course ^ 2 (cf. the end of 15.2.1.). In the first class of partitions we need only replace eachelementby its its intersectionwith Ci(frj). Thus (B* is modified by replacing elements B,(0), n B,(l), , B (n) by C,(*,) B.(0), n B.(l), , Ci(*i) I even this is not necessary: t is a subpartition of Ci(fri) n B (ri). In &2 (since *t 2, cf. 10.4.1.),of the system of pairwisedisjunct sets i.e. o hence (Ci(l), , Ci(i))(cf. 15.2.1.); we keeponly thoseelementsf a, which are subsets of Ci(fri), i.e.hat part of ft* which lies in Ci(fri). The t
We
K
K K

classes: which represent bjective those facts the Ctc, the (B = (S(0), o B (n)) and the C<(fc),k = 0, 1, , n and those which J3,(l), , k the player's stateof information, 2 the ><(&), = 1, ,n. o represent nly

Ct<

In the second class of partitions i.e.or the )*(&) we cannot do anyf w like it. Replacingthe elements 3)(fc) by their intersectionsith of thing
Ci(fri) would involve a modification of a
K

Q*(k)

shouldbe treated (B, but we prefer to postponethis discussion. like

player's stateof information 3 and would be should therefore be avoided. The only permissibleprocedure that which was feasiblein the caseof Ct : replacement of >(&) by that part of itself which lies in Ci(fri). But this is applicableonly if 33<(fc) like & before is a subpartition of ($2 (for K ^ 2). So we must postulate this. Now e(fc) takescareof itself:It is a subpartition of ^> (k) ((10:1 :c) in 10.1.1.), of dz (by the above assumption); hence and so we can replace it by that part of itself which lies in Ci(fri). T So we see: he necessary restrictionof F is that every D^fc) (with be a subpartition of ($2. Recallnow the interpretation of K ^ 2) must in and 8.4.2. of (10:A:d*), (10:A:g*) 10.1.2. give to this restriction They the meaning that every player at a personal move 9n 2, , 9TC r is fully
K

on

p. 113. 2

1We do not

wish

to change the

enumeration

to

= 1,

1,cf. footnote

events

extent 3 Namely, giving him additional

the fact: the represents umpire's state of information, but this is an objective to that moment have determined the courseof the play precisely to that (cf.9.1.2.). up
information.)))

116
9Hlj)

ZERO-SUMTWO-PERSONGAMES: THEORY
SfTli

informed about the

beforethe move stateof things after the move (i.e. in expressed &2. (Cf. also the discussionbefore (10:B) 10.4.2.) by That is, 3Tli must be preliminary to all moves 2, , 3TI,. We o Thus we have again obtained the condition (15:A:a)f 15.1.3. leave to the reader the simple verification that the game T9i fulfills the
Sfll

requirementsof 10.1.1.

induction.\" cessful would be a typicalstep of a \"complete Forthe moment, however, the only classof gamesfor which we possess c any kind of (mathematical) characteristicsonsists of the zero-sumtwoLet for thesewe have the quantities Vi, v 2 (cf. 15.1.1.). us persongames: therefore assumethat F is a zero-sum two-persongame. with Now we shall seethat the Vi, v 2 of F can indeedbe expressed the This circumstance of thoseof the rfj , 9i = 1, , i (cf. 15.1.2.). help makes it desirable to push the \"induction\"further, to its conclusion: i.e.,to form in the same way I\\, s, F^.^,^, , r it s 9,.1 The successively point is that the number of steps in these games decreases
from
v
jf

we 15.3.1.indicated at the end of 15.1.2.,wish to obtain the charAs acteristics T from thoseof all F^, = 1, of , i, sincethis if sucfri

The 15.3. Exact

Condition

(Entire Induction)

i.e.F*

(for F), v
f

amount $*(fri, This is the terminology of of 6.,7. In that of of 9.,10. we would say that ft (for F) is gradually restricted a Ci(fri) of 2 (for rf|), a C2(fri, fr 2) of ft 8 (for F^,^),a C8 (fri, fr 2, 9*) to of 4 (for rf f ,f|), etc., and finally to a C,(*i,2, fr etc., , 9,)of a,+i(for rt |(f t 9,). And this last set has a unique element((10:1 in :g) of the say it. Hence outcome the game r 9, is fixed: The player k getsthe fixed amount $*(*). ,9y is trivially clear; Consequently,the nature of the gameF^,^ it is clear what this game's alue is for every player. Therefore the process v which leads from the F^ to F if established can be used to work backwards from T9l .9t to F,ir 9 9 t to rf|ffi 9 ,_t etc., tc.,o e t to T9l and finally to F. F,i i But this is feasibleonly if we areable to form all games the sequence of rV T'i'*i> *X.it , rfi ,f t 9 , if the final condition of or is fulfilled for all thesegames. This requirementmay again be formulated for any general -persongameF; so we return now to thoseF. n The requirement then is, in the terminology of o (i.e.f 6.,7.) that 3Tli must be preliminary to all 9fR 2, SflZa, , 9ffl,; that

, is a \"vacuous\" game (like the one mentioned in the footnote 2 on p. 76). There are no moves; the player k gets the fixed
9

- 1(for F^),over - 2, - 3,
v v

to

(for rfi ,fi

,);

-,#).

15.2.2. 15.2.1., i.e. a a

if f

15.1.3., 15.1.2., i.e.

10.1.1.),

,,

,
>

V_

15.2.2. 16.3.2.
1,

. '''
$

i.e.

15.1.3.

15.1.3. 15.1.2.,
, ai
where

, on; 1 *t); etc.,

etc.)))

- 1,

, at

where a>

* a (*i); *i - 1,
f

GAMES WITH PERFECT INFORMATION

3TC2must

117

liminarity

* All is obtained: > (k), K ^ 2 must be subpartitionsof a2; all SD(fc), ^ 3 i a must be subpartitionsof Ct$, etc., etc.;.e.ll 5)(fc) must be subpartitions it of a\\ if K ^ X. 1 SinceOt< is a subpartition of Q,\\ in any case 10.4.1.), (cf. However a, is a suffices to require that all >(&) be subpartitions of :d) consequentlyour subpartition of >(/0 within & (k) ((10:1 in 10.1.1.); in requirementis equivalent to saying that ) (k) is that part of & which lies 2 in this B(fe). By (10:B) 10.4.2. means preciselythat preliminarity and
K

etc.; t etc., i.e.hat prethe 15.2.2. o In the terminology of 15.2.1., i.e.f 9.,10. of course same
be preliminary to all STls, 3TZ 4, w must coincide ith anteriority.
,
3fTC,;

a.

in anteriority coincide F. we By all theseconsiderations have establishedthis: to o In order be able to form the entiresequencef games (15B) :

(15:1)

F,

of

ly, rf it F^,,^,' ' r j, t v, v 1,v 2, ,

f f>
*

- -

>

*9

i moves respectively,t is necessaryand sufficientthat in the game F preliminarity and anteriority should coincide, i.e.hat pert and the end of 14.8.) fect information shouldprevail. (Cf. If F is a zero-sum two-persongame,then this permits the backelucidation of F by going through the sequence (15:1) wards from the trivial game F^,^, , r to the significant game F performing eachstep with the help of the devicewhich leads from the T9i to F as will be shown in

6.4.1. ...

15.6.2.

Inductive Step to carry out the announcedstep from the T9g9 proceed to F, the \"inductive step.\" F need therefore fullfill only the final condition but it must be a zero-sumtwo-persongame. or of w Hence ecan form all F^, <r i = 1, two, a , and they alsoarezero-sum 1 i persongames. We denotethe two players'strategiesn F by Sj, , S? and the \"mathematical expectation\" the outcome of and 2, of , the play for the two players, if the strategiesSJi, 2j areused,by

15.4. Discussionf the o Exact

15.4.1. 15.1.3. 15.2.2.,

We now

\\

2'

2?;

(cf.
by

11.2.3. 14.1.1.). , S^r and s; , z; i/2

t/1>

3Ci(n, rt) and

OC(TI,TI),

T2) T 3Ci(T,, I) -3C(r!, We denotethe corresponding quantities in

F,,
1

zj^ , and

if

the strategies

S^/2*areused,by
flCfj/lOlVj/lj

S^,

Te^/l) 30^(^/1,T^/j), ^Crj/aCTcj/l, T^/j) T^/|). 1We stated this abovefor X = 2, 3, ; for X = 1 it is automatically true: every :f)in partition is a subpartition of Oti sinceCti consistsof the oneset 1For the motivation if one is wanted cf the argument of footnote 3 on p. 63. 1From now on we write <n, **, no , <r* instead of 9\\, ft, , S9 because mis-

S JC^^/i, ((10:1 10.1.1.).

understandings

will

be possible.)))

118
We form

ZERO-SUMTWO-PERSONGAMES: THEORY

caseby v^/i, v, i/2. So

and))

the vi,

v 2 of

for 14.4.1.T and for F^ denotingthem

in the

latter

= MaxT| MinT|3C(ri,7 2), v 2 = Min MaxTi 3C(n, r 2

Vi
Tf
),))

v,

/2))

the Our aim is to express vi, v 2 in terms of the v r /i, v^/j. which determine character the move 3Ej the Thefcj of of n will play an essentialole. Since = 2,its possible r values arek\\ = 0, 2. We must considerthesethreealternatives separately. 15.4.2. Considerfirst the case = 0; let 3Tli be a chance move. The k\\ , i arethe pi(l), , PI(I) probabilitiesof its alternatives <TI in of mentionedin withCi = Ci(<7i) (pi(<n) is thepi(Ci) (10:A:h)

15.2.1. 15.1.2.,

1,

Now a strategy of player 1in T, S^, consistsobviously in specifyinga strategy of player 1 in F^, S^// for every value of the chancevariable

in

15.1.2. 15.2.1.). 1,
, i, 1

=!,

i.e.

10.1.1.

ai =

for all possiblecombinations TI/I, , r ai /i. a strategy of player 2 in F, 2J consistsin specifying strategy a Similarly

the i.e., 2I correspondto the aggregates 1/7, 2

, Z^ //

variable <TI = of player 2 in F^, 2^/2 for every value of the chance , 1 the 22 correspondto the aggregatesV/ 2, 2 , ^a\" ^ for all possiij ble combinations Z, TI/ , r aj /2. in of Now the \" mathematical expectations\" the outcomes F and in F areconnected the obvious formula by

i.e.

' '
*

1,

ff|

JC(TI,T2)

= J
rt

-l

our Therefore formula for vi gives

Vi

= MaxTi Min Tj 3C(n, ?2) = Maxvi

..... ',/,
V

Min v,))

r,-l
extremeight-hand r
side))

The <ri-term of the sum

on the

view by applying in 15.2.1.)))

1This is clearintuitively. The readermay verify it and (11:A) in the definitions of

11.1.1.

point of 11.1.3.formal isticdescribed to the situation from the

GAMES WITH PERFECT INFORMATION contains only the two variablesr^/i, T,i/2. Thus the variable pairs
TI/I,Ti/ 2;
; Tai /i, r ttj / 2

119

occurseparately,in the separate-i-terms o

in Hence forming

the Min Vj

Ta

arately, and in forming the MaxVi /v we can again maximize each b <ri-term separately. Accordingly, our expressionecomes
r<x <*t

,, we can minimize

each<ri-term sep-

pi(cri) MaxT<Vi Min rvt JCri (Tr|/i, rV2 )

<*i

pi(n)vri /i.

Thus we have shown

(15:2)

Vi

piOrOv^/i.

If the positions of Max and Min are interchanged, then literally the same argument yields

(15:3)
The 15.4.3. caseof
fci

v2

pi(cri)v<rj/2

imagination by showing that it is the formal statementof an intuitively g w plausiblefact concerning ames. This will also make it clearer hy this result must play a roleat this particular juncture. which we arenow going to give to the result of Theinterpretation is based on our considerationsof 14.2.-14.5. particularly those of and 14.5.2. for this reasonwe could not proposeit in a Forthis purposewe shall consider zero-sumtwo-persongameF in its and alsoits minorant and majorant games Fi, normalized form (cf.

= 1 comes ext, and in this casewe shall have n the to makeuse of the result of 13.5.3. Considering highly formal character d of this result,it seems esirableto bring it somewhatnearerto the reader's

13.5.3.

13.5.3. 14.5.1.,

to If we decided treatthe normalized form of F as if it were an extensive e and introduced strategies tc. with the aim of reachinga (new) form, of then nothing would normalized form by the procedure in and particularly in footnote 1on p.84. The as described happen, situation is different, however, for the majorant and minorant gamesFI,Fa; thesearenot given in the normalized form, as mentionedin footnotes 2 and 3 on p.100.Consequentlyit is appropriateand necessaryto bring them into of their normalized forms which we are yet to find by the procedure 11.2.3.)))

F2

(cf. 14.2.).

14.1.1.) 11.3.

11.2.3. 11.2.2.,

11.2.2.,

120

ZERO-SUMTWO-PERSONGAMES: THEORY

solutionsof FI,F 2 were found in 14.3.1., it is to Sincecomplete 14.3.3., 1 that be expected they will turn out to be strictly determined.
We use the notations TI, r 2, 3C(n, r 2) and vi, v 2 for F and we denotethe r f corresponding conceptsor FI by r(, r 2, 3C'( i> r i) and vi, v 2. A strategy of player 1 in FI consists in specifying a (fixed) value r a TI(=1, , 0i) while a strategy of plaj er 2 in FI consistsin specifying on value of r2(= 1, , 2) depending TI for every value of TI(= 2 0i). So it is a function of TI:TI = 32(ri). Thus T'J is TI, while TZ correspondsto the functions 32, andSC'^i, j) to r

to proceed do.

and It sufficesto considerFI (cf. the beginning of 14.3.4.), this

we now

!,-,

3C(ri,32(ri)).

Accordingly)) Tj = Min 3f MaxT|3C(Ti,32(ri)).

va

Min 3| 3C(Ti,32(rj)),

Hencethe assertion that FI is strictly determined, i.e.the validity of with (13 in 13.5.3. we needonly replace precisely :E) ; there vi = vi coincides
to This equivalenceof the result of 13.5.3.the strictly determined o character f Fi makes intelligible why 13.5.3. play an essential role will in the discussion which follows. FI is a very simpleexamplef a gamein o which perfect information prevails, and thesearethe gameswhich arethe ultimate goal of our present discussions(cf. the end of 15.3.2.). the And first move in Fi is preciselyof the kind which is coming up for discussion now:It is a personalmove of player 1, i.e.i = 1. k
Exact 15.5.

the X, U, /(*), ifr(x, f(x)) by our n, T2, 32(r,),JC(TI,32(ri)).

Discussionf the o the

Inductive

Step (Continuation)

of the player
A

c strategy of player 1in F, 2{i onsistsobviously in specifyinga (fixed) 1 value <rj( = 1, , on) and a (fixed) strategy of player 1 in F,,Zj1 8 ; i.e. /^
to the Z5> correspond the pairs <rj, T^/I.
merely argument, principles as 14.3.3. disposed 14.3.1., arebasedare not entirely the samethe those by which wewerea of the former principles 14.5.2., could be made stepping strictly determined casein 14.5.1., although to the latter. It is true that the an

Consider now 15.6.1. 1.

casefci =

l 1;i.e.et 3Tli be a personalmove

1This is

a heuristic

sincethe

on which the \"solutions\"

of

argument pretty convincing by We prefer to settle the matter \"unmathematical,\" purely verbal, amplification. the reasons being the same as given in a similar situation in mathematically, * This is clearintuitively. The readermay verify it from the formalistic point of in the partition and set view, by reformulating the definition of } as given in in and terminology, and then applying the definitions of The essentialfact is, at any rate, that in Ti the personal move of player 1 is preto that of player liminary 1Cf.footnote 1 on p. or footnote 2 above.)))

stone

14.3.2.

11.1.1. (11:A) 11.1.3.

14.2.

2. 118

GAMES WITH PERFECT INFORMATION

A
f

121

strategy of player 2 in F, 2^, on the otherhand, consistsin specifying a strategy of player 2 in IV, 2^' , for every value of the variable <rj = /2 1 the 1, , ai. Sor^/2 is a function of <r J :r^ /2 = 3s(<7j); i.e. S5 correspond to the functions 32 and clearly))
Therefore our formula for Vi gives:
vi

= Max,;. Min^OC^r,;,!,2(or)) 3 TrJ/1 = Max Max,j in^JC^j/i, (<rJ)). 32 M

r<r?/i

Now
3 Max,,;Min^JC^Ovj/i, s(<rJ)) = Max,; in^SC,;^,;/!, r,;/i) M owing to by our erj,

t there (13G)in 13.5.3.; we needonly replacehe x, u, f(x), \\l/(x, : 2

r,;

r /2, 32(cr?),3C,;(T,j/i,
vi
T<rJ/l

u)

ff

j/ ).
2
ff

Consequently))

= Max Max ; MinTr,/t JC^r,;/!, /t ) T,; = Max^j Max Minv/t JCr j(r r j/i, r cj/t) = Max,; r j/i v
T<r0/i

And

8 our formula for v 2 gives:

v2

= Min 3j Max^o/^ajCraJ/i, = Min 3j Max,; Max JC,;(rj/ r

rr;/1

Now
Min 3l Max^j Max rr!/I

= Max,j 3j Max ^jKj/i,32(<r5)) Min r<r;/l = Maxa j Min Max^o/.X^^j/i, Trj/,)

Tff;/1

JC^j/i,3t (aJ))

owing to

f(x),\\fr(x,

(13 and (13G)in :E) :

u) by our v2
a\\,
=))

4 2 r,;/, 32(<r?), Maxv/i acr j(r rj/i,r^/).

t there 13.5.3.; we needonly replacehe x, u,

Consequently

Summingup (and writing <n instead of crj) : 1Cf. footnote 1 on p. 118 footnote 2 on p. 120. or 1
TrJ/i must

be treated in this caseas a constant. This stepis of course rather trivial one, cf the argument loc.cit. a 8 In contrast to d there is now an essential ifference between the treatments of vi and vj. 4 rrj/i is killed in this caseby the operation Maxr a0/1 of the essential result in This step is not trivial. It makes use of (13:E) of that paragraph, as stated in 15.4.3.)))

15.4.2.,

i.e. 13.5.3.,

122 (15:4) (15:5)

ZERO-SUMTWO-PERSONGAMES: THEORY

c (fci this Interchanginglayers1and 2 carries into the precedingase = 1). p v 2, v 2 by this discussedin 14.6., interchange Vi V replaces i, and henceequallyv^/i, v, i/2 by v^/j, v^/i. Substitutingthesechanges t it becomes clearhat theseformulae into the above formulae (15:4), (15:5), Max in them by Min. So we have: must be modified only by replacing
As

16.6.2. Considerfinally 2. player

= Max,, ^/i, = Max^ v i/2 v, i.e. the case = 2; let fci

vi v2
.))

SfTli

be a personalmove of

(15:6)
(15:7)

vi v2

= Min^ v, t/ i, = Min i v ri /i.

<

of 15.4.2., 16.6.3. may sum up the formulae (16:2)-(15:7) We 15.5.1., as follows: 15.5.2., For all functions /(<TI) of the variable <TI(= 1, , i) define three = 0,1, as follows: 2 operationsMjj,
fci

for

fci

(15:8)
for))

= 0, = 2.

for Then
v*

fci

= Mk\\\\9i/k

for

fc

= 1, . 2

We wish to
M\".\\.

t emphasizesome simple facts concerninghese operations

integral used as an illustration in footnote 2 on p. k Second,M dependsexplicitlyon the gameF. This is evident since in has the range fci occurs it and , a\\. But a further dependence i of is due to the use of the pi(l), , PI(I),n the case fci = 0. the dependencef v* on v /k is the same for = 2 for each o Third, value of fci. We conclude y observingthat it would have beeneasy to make these b
\\ <r\\

M First,M*\\ kills the variable a\\ ; i.e. 9 \\f(ai) no longerdepends on <n. f For Forfci = 1,2 i.e.or Max^, Min^ this was pointed out in 13.2.3. it is obvious; and this operation is, by the way, analogousto the fci =

91.

1,
ff

fc

1,

formulae involving the average

maximum

f PI(^I)/(^I)or a chancemove, the

for a personal move of the first player, and the minimum for one of his opponent plausible by a purely verbal (unmathematical) n m argument. It seemedeverthelessnecessaryto give an exact athematical treatmentin orderto do full justiceto the preciseositionof Vi and of v 2 p A purely verbal argument attempting this would unavoidably becomeo s involved if not obscure as to be of little value.)))

GAMES WITH PERFECT INFORMATION

123

at We return now to the situation described the end of w and make all the hypothesesmentioned there;i.e. e assume that perfect information prevailsin the game F and also that it is a zero-sum two-person indicated loc. it.,ogetherwith the formula (15:8) of c t game. Thescheme of which takescare the \"inductive\" step,enableus to determinethe of essential properties F. We prove first without going any further into details that such a F induction \" with is always strictly determined. We do this by \" complete t This consistsof proving respecto the length v of the game (cf.

15.6.1.

The 16.6. Result

in

the Caseof PerfectInformation

15.3.2.

15.5.3.

two things

15.1.2.).

it is true for all gamesof length 1,for a given 1,2, , then it is also true for all gamesof length v. : If the t Proof of (15:C:a) length is zero,hen the gamehas no moves at 2 all;it consistsof paying fixed amounts to the players 1, , say the amounts 0 w, w.1 Hence i = 2 = 1 so TI = T = 1, C(n, r 2) = w, 2 and so 3
v

(15:C:a) That v = 0. (15:C:b) That

=

this
if

f is true for all games of minimum length;i.e.or

v

= v 2 = w; F i.e. is strictly determined,and its v = w.3

vi

Proof of
v

v 9i/i

1;henceby assumption every F^ is strictly determined. Therefore 4

ss v^/2. Now the formula
of shows (15:8) 15.5.3. that
Vi

Let (15:C:b):F be of length

v.

Then every

F,t is of length = V2 = v

We shall now go more into detail and determinehe Vi t of F explicitly. Forthis we do not even needthe above result of o the sequence f games We form, as at the end of

F Hence is also strictly determinedand the

= vt .

15.6.2.

proof is completed.

(15:9)

15.3.2., r, rv r,rv - ,r,,

it

15.6.1.

of the respectiveengths l
v, v

,/ ......
,

- 1, - 2,
v

0.

Denotethe

Vi,

v 2 of thesegamesby))

1Cf. the game in footnote 2 on p. 76 or In the partition and F^,^ i v in show that 12 has only one eleset terminology: For v = (10:1 (10:1in :f), :g) ment, say TT ft = (TT). Sow = $I(TT), w = 5i(ir) play the role indicated above. 8 I.e. eachplayer has only one strategy, which consistsof doing nothing. * This is, of course,ather obvious. Theessentialstep is U5:C:b). r 4

15.3.1. ..... 10.1.1.

for

eachvalue of k\\. 5 Cf.footnote 3 on p. 117.)))

I.e. fact mentioned the

at the end of

that 15.5.3.,the formula

is the same for

A;

= 1,2

124

ZERO-SUMTWO-PERSONGAMES: THEORY

a for Let us apply (15:8) 15.5.3. the \" inductive \" step describedt of the end qf 15.3.2.; et us replace <TI,T, T^of 15.5.3. the i.e. l by<r, t ^_ = then T9i v. The ki of 15.5.3. refers _itt , for each * t to the first move of T0i i.e.o the move 311*in F. It is therefore j ' , ow). (Cf. 7.2.1.) ccordingly convenient to denoteit by fc(*i, A k* '\"-1 In we form the operation M the ', replacing Mk of 15.5.3. this way we obtain

OK

!,,
* *

r,

if

(ffl

\\

(15:10) v.,

w * = M*,f ''' '

\"

v, t

K/ *

for

fc

1,2.

o the Considernow the last element f the sequence 15:9), game F^ ( v of in 15.6.1.; no moves at all. it has This falls under the discussion (15:C:a) 2 its Denoteits unique play 1 by if = ff(<ri, , v). Hence fixed w is , <r,)). So we have: equal to (Fi(((ri,
<?

(15:11)
K

v^
-

%/i

= v^
K

=s

with Now apply (15:10)

v

,2,1successively. In this manner

k

, <r,)). ,,/t= ffi(ir(<ri, to (15:11) then to the result,with and 9 M**

*

v, (15:12) = v 2 = v = Mk\\M
9

\"'*Si(*(ffi,

O).

obtains.
a This proves oncemore that T is strictly determined, nd also gives an formula for its value. explicit

15.7. Application

to Chess

and character, we have now given this vague view a precise eaning m that the games In question are strictly determined. nd we A by showing
rational

The and 15.7.1. allusions of 6.4.1. the assertionsof 14.8. concerning those zero-sumtwo-persongames in which preliminarity and anteriority w coincide i.e. here perfect information prevails arenow established. We there to the general pinion that thesegamesareof a particularly referred o

have also shown a fact much less founded on any \" generalopinion\" moves. that this is also true when the gamecontains chance Examplesof gameswith perfectinformation were already given in Chess (without chancemoves) and Backgammon (with chancemoves). o Thus we have established for all these games the existencef a definite value (of a play) and of definite beststrategies.But we have established w their existence in the abstract, hile our method for their construction only 3 t is in most casesoo lengthy for effective use. it In this connection is worth while to consider Chess in a little more

: 6.4.1.

detail.

1Cf.the remarks concerning r^ ^ in 1Cf. (15:C:a) in particularly footnote 1 on p. 123. 8 This is due c mainly to the enormous value of v. For Chess, f footnote 3 on p. 59. (Thev* there is our K, cf.the end of 7.2.3.))))

15.6.1.,

15.3.1.

.the pertinent

part

of

GAMES WITH PERFECT INFORMATION

125
SF*

6.2.2. 9.2.4.is restricted the numbers 1,0, I.1 or to

9* of

Theoutcome a play in Chess of

i.e.very value of the functions e

of

11.2.2. the samevalues, and sincethereareno chancemoves in have 2 In t Chess,he sameis true for the function OC*of 11.2.3. what follows we = 3Ci of 14.1.1. shall usethe function 3C Sincere only the values, 1,0, 1, he number t has
(15:13) v = MaxTi Mm Tj3C(n, r 2) = MinTj MaxTj 3C(r,,r 2)
has necessarilyoneof thesevalues
We leave to the

Thus the functions

v- 1,0,-1.
of irrespective
what

(15:D:a)

which he

readerhe simplediscussionthat (15:13) this: t means If v = 1then player 1(\" white \") possesses a strategy with
\"wins,\"

(15:D:b)

does.
If v

player 2

(\"black\

each one can \"tie\" (and possibly \"win\,") irrespective of what the otherplayer does. a (15:D:c) If v = 1 then player 2 (\"black\") ossessesstrategy p o with which he \"wins,\" irrespectivef what player 1(\"white\

then both players possessa strategy with which

does.
8

there would be nothing three possibilities(15:D:a), (15:D:c) (15:D:b), actually holds, and accordThe b ingly the play would be decidedefore it starts: decisionwould be in

15.7.2. showsthat This

the theory of Chess were really fully known left to play. The theory would show which of the
if

for case(15:D:a) \"white,\" in case(15:D:b) a \"tie,\" in case(15:D:c) for for \"black.\" But our proof, which guarantees validity of one (and only one) the of thesethreealternatives,gives no practicallyusablemethod to determine the the true one. This relative, human difficulty necessitates use of those heuristicmethods of playing, which constitute \"good\" hess; C incomplete, in o and and without it there would be no element f \"struggle\" \"surprise\" that game.
1This is the simplest way to interpret a \"win,\" \"tie,\" or \"loss\" of a play by the player k. * Every value of 9* is one of ff *; every value of 3C* in the absence f chance o moves is l c cit. If there were chancemoves, then the value of 3C*would be the one of 9*> a number which may lie anywhere probability of a \"win\" minus that of a \"loss,\" between 1 and 8 When there are chance probability of a \"win\" moves, then 3C(ri, 2) is the excess over a \"loss,\" cf. footnote 2 above. The players try to maximize or to minimize this abovedoesnot, in general, number, and the sharp trichotomy of

1.

i.e.

obtain.

(15:D:a)(15:D:c)

Although Backgammon is a game in which completeinformation prevails, and which contains chance moves, it is not a goodexamplefor the abovepossibility; Backgammon the is played for varying payments, and not for simple \"win,\" \"tie\" or \"loss,\" values of the $* arenot restrictedto the numbers

1,0,

i.e.

1.)))

126

ZERO-SUMTWO-PERSONGAMES: THEORY
The 15.8. Alternative,
Verbal

Discussion

which perfect information prevails,arestrictly determined. It can be questionedwhether the argumentation which follows is reall a proof; we prefer to formulate it as a plausibilityargument by which value can be ascribed eachplay of any gameF of the above type, but thi to is still open to criticism.It is not necessaryto show in detailhow thos value v of a play c can criticisms be invalidated,sincewe obtain the same T as in and there we gave an absolutely rigorous proc using precisely defined concepts.The value of the present plausibilit f t argument is that it is easiero grasp and that it may be repeatedor othe in which perfect information prevails, which are not subjectto th games, zero-sum two-personrestriction.The point which we wish to bring on a is that the same c criticismsapply in the general asetoo, nd that they ca no longerbe invalidated there. Indeed,he solution there will be foun t (even in gameswhere perfect information prevails) along entirely differen lines. This will make clearerthe nature of the difference between th zero-sum two-person case and the general case. That will be rathe important for the justification of the fundamentally different method which will* have to be used for the treatment of the general cas

isticapproachto our main result,

16.8.1. concludehis chapterby an alternative, We t i.e.,

simpler,less forma that all zero-sumtwo-persongames,i

15.4.-15.6.,

a Consider zero-sumtwo-persongameF in which perfectinforma tion prevails. We use the notations of in all respects: or th F

(cf.24.).

16.8.2.
,

'9' ''-''jth the probabilities; operators h Af*; , M.\\, the , M*?' derivedfrom F;and the function &I(*(<TI, - - ,0-,)) sequence 15:9)games ( of We proceed discuss the gameF by starting with the last move 3M to and then goingbackwardfrom therethrough the moves 9fTC,_i, 3TC,_2, Assumefirst that the choices o-2, <TI, , -\\ (of the moves 9Tli, 3fE 2, ' a < 3fll,_i) have already beenmade, nd that the choicer, (of the move 9TC,)i
(<ri>

2 STCi, 3TC,

the 311,; <n,

cr

2,

t a,]he fci,

15.6.2.
fc

2(<n),

, k,(<n, **,
1

.....
vv
wil

, <r,_0

<r

the respective robabili p tiesp,(l),*(2), p ,p9 (cL 9 ((r\\, , o-,_i)). Sothemathematical expec tation of the final payment (for the player 1) 9 i(*(o-i, , <r,_i, <7,))
with
r
i))

now to be made. If 9TC, is a chancemove, if k,(<n, 0-2, have the values 2, , a,,(<ri, , v v-\\)

1,

i.e.

, <r,_i)

= 0, then

if k 9 (<r\\, 3TI, is a personal move of players 1or 2, , <r v-\\) = then that player can be expected maximize or to minimize to 2, the outcome $i(*(<n, , <r,-i,<r*)) by his choiceof <r,; or Min^ $i(ff(<ri, , <r,-i, <r,)) , <r,_i, <r,)), respec Max,,2Fi(lf(<ri, tively is to be expected.)))

If

*-

or

i.e. i.e.

GAMES WITH PERFECT INFORMATION

the for I.e., outcome to be expected the play
v-\\

127

have beenmade is at any

after the -choices<TI,

rate))

a,_i; Mj; ;('> \"-'> Atf'/' \"-> , a,)). to for Consequentlythe outcome be expected the play after the choices <n, , cr,_2 have beenmade
,
* *
is))

Assume next that only the choices<n, , a r _2 (of the moves have beenmadeand that the choicer,_i (of the move 9K,_i) < , 9TC,_i) is now to be made. Sincea definite choiceof <r,_i entails, as we have seen,the outcome *'~ (r| l) &i(*(<ri, , 0v)) which is a function of <n, , Afjj '\"** kills cr, we can a only, sincethe operationMj proceeds above. v '\"- friOKn, l} We needonly replace; <n, Mk 1 F , or,)) by

.....

v- 1;en,

..... ..... ,*,; .*' *,(*(, : ..... .....

(ffl
is))

a,-i

<n,

for Similarly the outcometo be expected the play after the choices <r,_s have beenmade ,

has begun
And this

Finally, the outcometo

is))

be expectedor the play f

outright

before it

1 is preciselythe v of (15:12) in 15.6.2. 15.8.3. objectionagainst the procedureof 15.8.2.that this The is b approach to the \"value\" of a play of T presupposes\"rational\"ehavior of all players; i.e. layer 1's p strategy is based upon the assumption that player 2's strategy is optimal and vice-versa. Specifically:Assume fc,_i(<n, , <r,_2) = 1,k,(*i, , <r,_i) = 2. Then player 1, hose personalmove is 9H v_i choosesis <r,_i in the convicw h tion that player 2, whosepersonalmove is 9fR, chooses is <r, \"rationally.\" h t f Indeed,his is his soleexcuseor assumingthat his choiceof <r v-\\ entails

the outcome i.e. ^'1 Min,,SiOr(en, M of of the play. (Cf. the discussion 3TC,_i in 15.8.2.)

,O),

..... '\"-'^i^^i,

,*,)),

1 In imagining the application of this procedureto any specific game it must be rememberedthat we assume the length v of T to be fixed. If v is actually variable and it is so in most games (cf. footnote 3 on p. 58) then we must first make it constant, by at I the device adding \"dummy moves\" to T as described the end of 7.2.3. t is only of after this has beendone that the aboveregression through 3TC,,9M*-i, , 9Ri becomes feasible. For practical construction this procedure is of course better than that of 15.4.no

15.6.

either

Possibly some very simple games, like Tit-tat-toe,could be effectively

manner.)))

treated

in

128

ZERO-SUMTWO-PERSONGAMES: THEORY

to we came the conclusionthat the Now in the secondpart of of \" rationality \" in others must be avoided. Theargumentation hypothesis did not meetthis requirement. of It is possibleto arguethat in a zero-sumtwo-persongamethe rationality the of the opponent can be assumed,because irrationality of his opponent s can never harm a player. Indeed, incethere are only two players and sincethe sum is zero, very loss which the opponent irrationally inflicts e 1 an upon himself, necessarilycauses equal gain to the other player. As it c but stands, this argument is far from complete, it couldbe elaboratedonwith its stringency: siderably. However,we do not needto be concerned 2 which is not open to thesecriticisms. We have the proof of 15.4. But the above discussionis probably neverthelesssignificant for an essentialaspectof this matter. We shall seehow it affects the modified c conditionsin the more general ase not subjectto the zero-sumtwo-person restriction referredto at the end of

4.1.2.

15.8.2.

-15.6.

15.8.1.

16. inearity and Convexity L

confronts us next is that of finding a solution which meetsthe comprises all zero-sum two-person games, i.e. difficulties of the non-strictlydeterminedcase. We shall succeed doing in this with the help of the same ideaswith which we mastered strictly the determined : It will appear that they can be extended as to cover all case so zero-sumtwo-persongames. In orderto do this we shall have to make use of certain possibilitiesof probability theory (cf. 17.1., And 17.2.). it will be necessaryto use some mathematical eviceswhich are not quite d the usual ones. Our analysis of 13. for providesone part of the tools; the remainderit will be most convenient to fall back on the mathematicoon theory of linearity and convexity. Two theorems convex geometrical 3 bodies will be particularly significant. Forthesereasonswe arenow going to discuss to the extentto which o they areneeded the concepts f linearity and convexity. 16.1.2.is not necessaryfor us to analyze in a fundamental way the It notion of n-dimensionallinear (Euclidean) pace. All we need to say is s is that this space describedy n numerical coordinates. ccordingly we b A define for each n = 1,2, l , the n-dimensionalinear space Ln as the setof all n-uplets of realnumbers [x\\, - - ,xn }. Thesen-uplets can also belookedupon as functions i of the variable i,with the domain (1, , n)
which
1

16.1.1. task which The

Geometrical 16.1. Background

than two

20.1., 58.3. 24.2.2., Cf. in this respectparticularly (14:D:a), in (14:C:e) 14.5.1. (14:D:b), (14:C:d), in and (14:C:a), (H:C:b) 14.5.2. 1Cf. T. Bonessenand W. Fenchel:Theorieder konvexen Korper, in Ergebnisseder Vol. III/l, erlin 1934. Further investigations in Mathematik und ihrer Grenzgebiete, B H. Elementare Theorie konvexen der Commentarii
p tici, Vol. VII, 1935, p.290-306.)))
Weyl:

This is not necessarilytrue if players. For details cf.

the sum

is not

constantly

zero,or if there are more

Polyeder.

Mathematici

JGEelve-

LINEARITY AND CONVEXITY

in

129
with

not usage call i an indexand not a variable;but this does alterthe nature of the case. In particular we have

the sense of

13.1.3. 13.1.2., 1 We
{xi,
, xn }

shall in conformity

general

jtfi,

y n]

and only if x One , n (cf.the end of 13.1.3.). couldeven take the view that Ln is the simplestpossible o space f (numerical) functions, where the domain is a fixed finite set the set (1, , n).2 We shall also call thesen-uplets or functions of Ln pointsor vectorsof Ln and write
if

= yi

for all i = 1,

(16:1)
The Xi for the specifici =

\"? =

{xi,,n

,x*\\.
the values of the function x

1,

are

the componentsof the vector x. We mention although this is not essentialor our further work f that Ln is not an abstract Euclideanspace,but one in which a frame of 3 h reference (system of coordinates) as already beenchosen. This is due to the possibilityof specifyingthe origin and the coordinatevectors of Ln numerically (cf. below) but we do not proposeto dwell upon this aspect of the matter. The zero vector or origin of Ln is

16.1.3.

0*= {0, The n coordinatevectorsof Ln arethe 7= {0, ,0) =

- -

,0}. ,6,l j = 1,
n

,1,

{!,,
.

, n,

where for for After these preliminarieswe can now ations and propertiesof vectorsin Ln

i = j,4-5 i ? j.
the describe fundamental

oper-

16.2. Vector

operations involving vectors are those of scalar > t multiplication, i.e.he multiplication of a vector x by a number t, and of
effective enumeration

16.2.1. main The

t I.e.he rMiplets

Operations

vector.
6 in

The [x\\, , xn are not merely sets in the senseof of the x, by means of the index i = , n is just as essential as the aggregateof their values. Cf the similar situation in footnote 4 on p. 69. 1 Much in modern analysis tends to corroborate attitude. this 8 This at leastis the orthodox geometrical standpoint. 4 Thus the zero vector has all components 0, while the coordinatevectors have all and its index for thej-th coordinate that one component being components but one
\\

1,

8.2.1.

1,

is the

\"symbol

of Kronecker and

Weierstrass,\"

which

is quite

useful in many

respects.)))

130

ZERO-SUMTWO-PERSONGAMES: THEORY

vector addition,

i.e.ddition of two vectors. Thetwo operationsaredefined a i m by the correspondingoperations, .e. ultiplication and addition, on the componentsof the vector in question. More precisely:
-

Scalar ultiplication: t{x\\, m

Vector addition

- , xn } ,

= [txi,

tx n

}.
, xn

{zi,

, xn }

+ [yi,

y n]

= [xi + t/i,

+ y n \\.

The algebraof these operations is so simple and obvious that we forego o its discussion.We note, however, that they permit the expressionf any - - , x n ] with the help of its components and the coordivector x = [x\\, nate vectors of L
n))

Someimportant subsets of Ln : e (16:A:a) Considera (linear,inhomogeneous) quation

n

(16:2:a)
(ai,
, an ,
b

i-i

ax
>

>

=l

areconstants). We exclude =a = ai = since in that casethere would be no equation at all. All points (vectors)x = {zi, , xn which fulfill this equation, 2
\\

form a

(16:A:b)

hyperplane. Given a hyperplane

n

(16:2:a)
it

X a '*'= b definestwo parts of Ln .

n

-i

>

It cuts Ln into thesetwo parts:

(16:2:b)
and

J) t-i

OM > 6,

(16:2:c)
1

^ t-i
and

a lxi

< b.
>

Thesearethe two half-spaces producedby the hyperplane.

Thex,arenumbers, hencethey act mx } 8 'as scalarmultipliers.
n

analogues;hencethe

summation. * For n in ordinary (3-dimensional Euclidean) space, these are just the 3, 1)-dimensional) ordinary (2-dimensional) planes. In our general casethey are the ((n

y-i

2^ is a vector

i.e.

name.)))

LINEARITY AND CONVEXITY))

131))

a Observethat if we replacei, , a n , 6, then , a n , b by ai, the hyperplane(16:2 remains unaffected, but the two half-spaces (16:2:b), :a) we are (16:2:c) interchanged.Hence may always assume a half spaceto be given in the form (16:2:b).

(16:A:c)

>

>

x Given two points (vectors) , then the center of gravity of x, y I,

andaJ ^ Owithl

0;

with

the respectiveweights

1
>

in

the senseof mechanics is / x

+ (1

t) y.

The equations
X

*7 + (1

- 07 = Itei+ (1- Ol/i,

{Xl,

Xn

),

(2/1,

' , 2/n),
ten

- - ,

+ (1

t)y n ]

shouldmakethis amply clear. of A subset,C, of Ln which containsall centers gravity of all its points i.e. hich contains with x, y all x + (1 f) y , Q t 1 is convex. w
3
this
t

o is the customary concept f cont vexity. Indeed,he set of all points

x

i.e.n the ordinary planeor space i

+ (1 y].
y

Thereader ill note that for n = 2, w

the linear (straight) interval connecting the points x and y , the interval
[x ,
And

^ t ^ 1 is precisely

so a convex set is one its points x ,

>

which,
y

with

any two of

x interval Figure

showsthe conditionsfor n = 2, i.e. the plane. in Figure16 16.2.2. the intersection any number of convex setsis again of Clearly if convex. Hence any number of points (vectors)x ', , x p is given, a there exists smallest convex set containing them all:he intersection t of
all convex setswhich contain x
>

, alsocontainstheir interval [ x ,

].

16.

i*

Cf. Fig.17,where p = points (vectors)))

by x

',,
x
p

>

>

. It is again useful to visualize the casen = 2 (plane).

6. It is easy to verify
,
tp

',

, x

Thisis the convex set.spanned

that this

set consistsof all

/,

(16:2:d)))

(/

*''f rally^ 0, o

with))

1.))

'

Proof:The points
is such a

form (16:2:d) a set containing

0.)))

all x

',

point: tj = 1and all otherti put

132

ZERO-SUMTWO-PERSONGAMES: THEORY

The points(16:2form a convex set If :d) then * z

Any

+ (1

p
t*v

y-i

'

with

of

(16:2:d):

convex

set, D, containing

We prove

this by induction for all p =

ti

a;

point of

F Proof: or p = 1it is obvious;sincethen

',,
w/

x = 2}J/ * ; and

>

>

= **,-+ (1

O*/-

z p containsalso all \"points 2,

(16:2:d).
=
<

. = 1and so x ' is the only 1,

If 2)

Assume that it is true for p

jj

1. Considerp itself.

P-I
<;

then
If))

p_ 1

= 0,the point of (16:2 is x p and thus belongsto D. :d)

Shaded

area: Convex

spanned

by

7-,....,?
17.))
5\" 8

Figure

P-I
Put

p-l p-i)) p 2 *, > 0,then put * = y-i *,,sol y-i y-i ^. 5) -J^JJ^-S^ y-i
5,

HenceO<

^ 1.

<7

/< for

j = 1,

,
>

1. So

P-I
y-i
2)

Sj
?

1. Hence,by our

assumptionfor p

- 1,P-I _
%
8jX
y-i)) y-i))

is in D. D is convex, hence

is also in

D;but this vector is equal to))

5;
y-i)))

which thus belongsto

Z>.))

LINEARITY AND CONVEXITY))

133))

The proof is therefore completed. The h, , t p of (16:2:d) themselvesbe viewed as the commay
ponents of a vector
and)) t

to give a name to the

= {t\\, set to which

ti

tp ]

they

inLp. It is therefore appropriate defined by arerestricted,

tp 0,))

0,

- - ,

1.))

r,-txit))

Figure

18.

Zt-axis))

Figure

19.))

?))

Figure

20.

Figure

21.

shall denoteit by Pp. Both setsSp, PP areconvex. , tp Let us picture the cases = 2 (plane)and p = 3 (space). P2 is the p 18). positive quadrant, the areabetweenthe positivex\\ and x* axes(Figure betweenthe positivex\\, x* and x 8 axes, P8 is the positiveoctant, the space i.e. etweenthe plane quadrants limited by the pairs x\\, #2; x\\, x8 ; x 2, x 8 of b P these(Fig.19). /S2 is a linear interval crossing 2 (Figure 18).58 is a plane P likewisecrossing 8 (Figure19). It is useful to draw Si,Sisep-))) triangle,
*

shall denoteit by Sp It is also convenient to give a name to the set which is described the first line of conditionsabove alone, by t\\ ^ 0, by
We

0.

i.e.

We

134

ZERO-SUMTWO-PERSONGAMES: THEORY

arately, without the Pa, P8 (or even the La, L8) into which they arenaturally immersed(Figures 21).We have indicated on thesefigures thosedis20, tanceswhich areproportionalto Xi, Xa or x\\, x 2, x 8, respectively. marked x\\, x 2, x 8 in Figures 21 Thedistances 20, are (We re-emphasize: not the coordinates\\, xa, x 8 themselves.Theselie in La or L8 outside of x Sa or S8 , and therefore cannot be pictured in Sz or 83} but they are easily seento be proportionalto thosecoordinates.) Another 16.2.3. important notion is the length of a vector. The length
of x

= {xi,

xn

is))

= 1*1 A

' t-i))
:))

i The distanceof two points (vectors)s the length of their difference

from the origin Thus the length of x is the distance

sets:

16.3. shall now establish an important We

16.3. TheTheorem the of

Supporting

Hyperplanes

generalproperty of convex

(16:B)

Let p vectors x l , - - - , x p be given. Then a vector y > either belongs to the convex C spanned by x x p (cf. in or there a (16:A:c)16.2.1.), exists hyperplanewhich contains in such (cf. (16:2:a) 16.2.1.) that all of C is contained in y one half-spaceproduced by that hyperplane (say (16:2:b) in

',,

This is true even if the convex spannedby x ', , x p is replacedy b any convex set. In this form it is a fundamental tool in the moderntheory

16.2.1. (16:A:b) ;cf. id.).

picturein the casen = 2 (plane)follows:Figure22 uses the convex set C of Figure 17 (which is spannedby a finite number of points, as in 2 the assertionabove),while Figure23 showsa general onvex set C. c Before proving (16 we observe that the secondalternative clearly :B), t excludeshe first, since y belongsto the hyperplane,hencenot to the half it and in (16:A:b) space. (I.e. fulfills (16:2:a) not (16:2:b) above.)
A

of convex sets.

proof: A a Proof: ssumethat y doesnot belong to C. Thenconsider point of C which liesas near to y as possible, i.e.or which f
We now give the 1The Euclidean
*For the

meaning of thesenotions is immediate. topology, we add: To be exact,this sentence should be qualified the statement is meant for closed convex sets. This guarantees the existence the minimum that we use in the proof that follows. of Regarding these)))

readerwho is familiar

Pythagorean

with

LINEARITY AND CONVEXITY

135))

|7- 7l
assumesits minimum
value.))
The hyperplane))

=
t-l))

ViY))

The half

space))

Figure 22.)) The hyperplane)) Thehalf

space))

Figure

23.

Considerany other points u of

tu

+ (I

t) z

C. Then for every t with ^ t ^ 1, also belongsto the convex C. By virtue of the minimum

property of z (cf.above) this necessitates))

i.e.)) i.e.))

- y)+t(u

-))

By elementaryalgebrathis means
2
<*))

^
t-l)))

0.))

136
So for t

ZERO-SUMTWO-PERSONGAMES: THEORY

>

t (but of course ^ 1) even

2 5) (ft

t-i

~ Vi)(Ui

ft)

t-i(Ui

t ft)

z<)*t

* 0.
n
~~ &

to then the left-hand sideconvergeso 2 J) (ft t If t converges 0,

ft).))

Hence

(16:3)
As
w,
yi

(*
=
(w t
-

y,)(t*<

0.

Zi)

+ (ft ~ 2/.),this means))

t'-l))

2
i
.))

t-1))

Now 2; 7^ y (as z belongsto C, but y does not); hence Sothe left-hand sideabove is > 0. I.e.))
\\

|2

> 0.

(16:4)))
-i))

(ft

t/)2/.))

Put

a<

y, then

ai))

= a n = is excluded by n above). Put also b = a^,.

z j*
Thus))

(cf.

(16:2
;s.

*)))

defines a hyperplane, to which 1

Next))

clearly be-

(16:2:b*)))
Figure 24.))

<z,

>))

statesprecisely is a half spaceproduced by this hyperplane, and (16:4) that u belongsto this half space. o the Since u was an arbitrary element f C this completes proof. can alsobe statedin the geometrical This algebraic proof language. n first. Thesituationis pictured Letus do this for the case = 2 (plane) in Figure z is a point of C which isas nearto the given point y as possible; 24:
for of i.e. which thedistance y and z ,

y\\

assumesits minimum

value.)))

LINEARITY AND CONVEXITY

137

Sincey ,

_*

>

a hyperplane and

arefixed,and

w is a variable point (of C),therefore (16:3) defined one of the half spaces roduced by it. And it is easy p

_+

to verify that z belongsto this hyperplane,and that it consistsof those points u for which the angle formed by the three points is a right-angle for (i.e. which the vectors z
y

and u

areorthogonal).Thismeans,))

Interval

(x , u J
to y
than
*))

Part of the interval nearer

Hyperplane

of

(1 3)
Figure 25.))
Hyperplane

of

(16:4)
of
(16:3)))

Hyperplane

Figure
n

26. of C must lie on this

indeed,that

JJ (zi

-i

yi)(ui

z>)

= 0. Clearly all

hyperplane, or on that side of it which is away from y . If any point u of C did lie on the y side,hen somepoints of the interval [ z , u ] would be t

Thecomputation on pp. 135-136 properly interpreted shows precisely this.) SinceC contains z and u , and so all of [ z , u],this would contradicthe statement that z is as near t nearerto
y

than z

is. (Cf. Figure 25.

>

to

as possiblein

C.

Now our passage from this hyperplanefrom z to

to amounts (16:3) (16:4)

y

to a parallel shift of b the (parallel, ecause coefficientsa =

y)))

138

ZERO-SUMTWO-PERSONGAMES: THEORY

of Uiy = , n are unaltered). Now y lies on the hyperplane,and all of C in onehalf-spaceproducedby it (Figure 26). n couldbe visualized in a similar way. The case = 3 (space) n manner. It is even possibleto accountfor a general in this geometrical n-dimensional\"geoIf the reader can persuade himself that he possesses metrical intuition \" he may accepthe above as a proof which is equally t valid in n dimensions. It is even possible to avoid this by arguing as follows:Whatever n, the entireproof deals with only three points at once, to p namely y , z , u Now it is always possible lay a (2-dimensional) lane through three given points. If we consideronly the situation in this a 24-26nd the associated plane, then Figures argument can be used without any re-interpretation. Bethis as it may, the purely algebraic proof given above is absolutely rigorous at any rate. We gave the geometrical analogiesmainly in the that they may facilitate the understandingof the algebraic perations hope o performedin that proof.

i 1,

fundamental for our subsequentwork. We start by considering rectangular a matrix in the senseof with n rowsand m columns,and the matrix element (t, (Cf. Figure a in The <, z, y, t, s there correspondto our a, n, m.) I.e. a(t,j) is a perfectly arbitrary function of the two variables = , n\\ vectorsin Ln Foreach = , m. Next we form certain j , j'

The 16.4. Theorem of the Alternative for Matrices 16.4.1. theorem (16B) of 16.3. an inference which The : permits

will

be

13.3.3. 1,
, n the

j). i,j, i 1, : 1,

13.3.3. 11

the vector x 1 = we have

16.1.3.;

theorem(16 of :B)

7=7.
with

x m,

Theconvex C spannedby x x m, 6 If this is the case,hen we can conclude t from

',, ',, ',,',, ',,

{x\\,

coordinate vector
the replaced
n

j there by our 1.)

x' ] with x{ = a(i, j) and for each I = 1, n l = 5 l {B i}. (Cf. for the latter the end of
Let us now apply the to thesen + m vectors x
x
p

16.3. p = n + m for 6 . (They replacehe x t

loc. it.) We put c

>

(16:2:d)16.2.2.

6 n may contain that in

0.

i>7< ; - 1 */7;>-+f-1
(16:5) (16:6)
*! ^

=K
- ,

0,

tm

0, 81 ^ 0, Sl = t,: +
^

sn ^

0.

1.)))

LINEARITY AND CONVEXITY

139

<!,*

componentsthis means

, Jm,

i,

' ,s

the replace t\\ t

m n

, tp

(loc. it.).In terms of the c

y=i

2)

*x*> j)

+ 2}*/** = o.

j-i

Thesecond term on the left-hand sideis equal to

s,so we write

(16:7)
m

y-i
tj

a(i,j)ti=

-,.
=
m

If we had %

y-i

= 0,then ti =

= t m = 0,hence (16:7) = sl by

sn = 0, thus contradicting Hence2) */ > - We replace (16:6). (16:7) y-i by its corollary

(16:8)

aftJH-SO.))

(16:5) x\\ gives

(16:9)

0,
x

j-i

2^ ^ for j
, xm

=
^

1,
, xm
\\

, m.

Thenwe have 2}*/ =

y-i

0. Hence
to belongs Sm

= jxi,

and (16:8) gives

m

U6:10)

^ 2 *(*'/)*/

for

1,

n.))

on not Consider, the otherhand, the possibilitythat C does contain . of us to infer the existencef a o Then the theorem (16:B) 16.3. permits
in such hyperplane which contains y (cf. (16:2:a)16.2.1.), that all of C is containedin one half-spaceproducedby that hyperplane (cf. (16:2:b) in Denote 16.2.1.). this hyperplaneby

2}a tXi = 6. t-i

Since belongsto
it, therefore 6
n

= 0. So the half space questionis in

(16:11)

o,^>

0.)))

140
x
5

becomes a ',(16:11) -io $,j > 0,i.e.i > 0.

t

',*, ',*,
x m,
d

ZERO-SUMTWO-PERSONGAMES: THEORY
6n

belong to this half space. Statingthis for

So we have

(16:12)
Statingit for
x J,

ai > 0, becomes (16:11)

n

, an

> 0.

(16:13)

i-i

J) a(t,j)a< > 0.
n

(16:12) wi gives

/n-i > 0,
m t

2) 0* for i = 1,
,
wn

, n. Then we have

> 0. Hence
,
wn
\\

t-i

ti\\

= 1and

w = {wi, (16:14) And (16:13) gives

belongsto

Sn .

(16:15)
%

- i a(i j)Wi ^

for

j = 1,

, m.

Summingup

(16:9), (16:14), we may state: (16:10), (16:15),

(16 :C)

Let a rectangularmatrix with n rows and m columns be , n; given. Denoteits matrix elementby a(i, j), = = , m. Then there existseither a vector x =

i 1,

[xi, -

1, m

xm ] inS with m

(16:16:a)

/-i
n

a(t, j>,

^0
{101,

for

i = 1,
,
wn]

, n,

or a vector w =

in

<S n

with

(16:16:b)

-ia(i,j)w<

>

for

j = 1,

, m.

We observefurther: The two alternatives (16:16:a), exclude (16:16:b) eachother. Assumeboth (16:16:a)(16:16:b). and Proof: Multiplyeach(16:16:a) by
Wi

and sum over i =

1,

,n;this gives %

a(i,j)w&j^ 0.

Multiply)))

LINEARITY AND CONVEXITY

141

each (16:16:b) and sum over j = by x,

1,

w; this gives))

t-iy-i
Thus we have a contradiction.

i.e. e denotethe columns (and not, as before,the rows)by i = 1, w and the rows (and not, as before, the columns)by j = 1, , m. ' 16.4.1.
- -

16.4.2. replacehe matrix a(i,j) by its negative transposedmatrix; t We

,n
And

b we let the matrix element e a(t,j) (and not, as before a(i,j)). (Thus m too areinterchanged.) n, We restate now the final results of as applied to this new matrix. But in formulating them, we let x = {x\\, m , x' play the rolewhich
\\

w
[x\\y

=
-

[wi,
,
m}

matrix.

had.

wn ]

had, and w' = {w(,

And

, w'n the rolewhich x =* we announce the result in terms of the original

\\

Then we have:

(16 :D)

Let a rectangular matrix with n rows and m columnsbe Denoteits matrix elementby a(i, j), i = 1, given. , n;

j = 1,
(x'j,
m

, m. Then in Sm with ,

x'J

there existseither a vector

x' =

(16:17:a)

y-i

a(i,j)x' < ; w' = {w(, -

for
- - , f
r))

1,with

, n,

or a vector
n

u^) in Sn

:b) (16:17
And

i-l & JX ^0

eachother. the two alternatives exclude

or or and (16:17:a),(16:16:b),(16:16:a) (16:17:b)

and We 16.4.3. now combinethe results of 16.4.1. 16.4.2. imply They

that we must have

each exclude other. simultaneously;and also that thesethree possibilities matrix a(t,j) but writing x , w , x\\ w for the vectors Usingthe same 16.4.2. x', w x , w' in 16.4.1., we obtain this:

'

1>

and not only

m

<

0. Indeed,
1.)))

would

necessitate\\ = x

xm

which

it impossible since J) *,

142 (16:E)

ZERO-SUMTWO-PERSONGAMES: THEORY
either a vector Thereexists
m

x t

\\x\\,

- - - , zm j , n,

in

Sm with

(16:18:a)

>-i
n

a(t, ;>,<
w

for
,

= 1,
in

or a vector

[wi,

wnj

Sn with
, m,
w
f

(16:18:b)
in

;=

a(i,j)wi >
i

for
-

j = 1, ,x' inS
m
\\

ortwovectors x' = [x(, -

and

= \\w' lt

& with))

,u/n

(16:18:c) '\"'

for

i = 1, j = 1,

, n, , m.

2) t-ia(i,j)^

^0

for

Thethree alternatives

other.

(16:18:a), (16:8:c) each (16:8:b), exclude

1

and on one and By combining (16:18:a) (16:18:c) hand and (16:18:b) '2 on the (16:18:c) other,we get this simplerbut weakerstatement.

(16:F) (16:19:a)

e Thereexists ithera vector

m

x i

= [xi, =
in

- - , xm , n,

\\

in

Sm with

y-i

a(i,j)xf
w

^0
{wi,

for
- - ,
wn

1,
Sn with

or a vector
n

\\

(16:19:b)

J) i-ia(ij)w,^0

for

j = 1,

, m.

w coincidesith its negative transposedin the senseof 16.4.2. n = m and ; i.e.

16.4.4. Consider now

ofej) =

a skew symmetric matrix for

o a(i, j), i.e. ne

,

which

-aQ',

i,j = 1,

n.

1The two alternatives do not exclude eachother:Their conjuncis precisely 8 This result could also have beenobtained directly from the final result of is precisely is a weakened form of there, and there. We gave the abovemore detaileddiscussion becauseit gives a better insight into the entire situation.)))
tion

(16:19:b) (16:19:a), (16:18:c). (16:16:a) (16:19:b) (16:19:a)

(16:16:b)

16.4.1.:

MIXED STRATEGIES.THE SOLUTION is thing:Indeed, 16:19:b) (

143

Then the conditions(16:19:a) (16:19:b) in 16.4.3. and the express same

n

2) a(i,j)w<^

0;
n

this may be written

n

We

need only write jf, i for i,j l so that this becomes a(i,j)Wj ^

0,and

._

then

x for

l 10,so that

we have V

a(i, j)xj g 0.

And

this is precisely))

Therefore we can replace disjunction of (16:19:a) the and either one of them, say by (16:19:b). we obtain: So by

(16:19:b)

(16 :G)
cf.

Sn with))

a(i,j) is skew-symmetric therefore n = m (and a above), then there exists vector w = {wi, , w n } in
If the matrix

2) t-i

a(i,j)wi^0

for

j = 1,

n.

17.Mixed Strategies.The Solutionfor All Games

of 17.1. Discussion Two Elementary

Examples

of simplestexamples this phenomenon. Thesearethe gamesof Matching Since Penniesand of Stone, Paper,Scissors(cf. 14.7.2., 14.7.3.). an t attitude with respecto the \"problems\" these of common-sense empirical, we may hope to get a clue for the solution of non-strictly gamesexists,

case

In t 17.1.1.ordero overcomethe difficultiesin the non-strictlydetermined

which we observedparticularly in

14.7.it

is best to reconsiderhe t

determined (zero-sumtwo-person)gamesby observingand analyzing these

i that, e.g.n Matching Pennies,no particular way of playing i.e.neither playing \"heads\"nor playing \"tails\" is any i betterthan the other,and all that matters is to find out the opponent'sntento tions. This seems block the way to a solution, sincethe rules of the gamein questionexplicitlybar eachplayer from the knowledgeabout the actions,at the moment when he has to make his choice. But opponent's
1

attitudes.

It was pointed out

that Observe

now, with n

= m this is only a change in

notation

!)))

144

ZEROSUM TWO-PERSONGAMES: THEORY

o the above observationdoes not correspondfully to the realitiesf the case: In playing Matching Penniesagainst an at leastmoderately intelligent i opponent,the player will not attempt to find out the opponent'sntentions on but will concentrate avoiding having his own intentions found out, by and \" tails\" in successive ames. Sincewe g playing irregularly \"heads\" wish to describe strategy in one play indeedwe must discussthe course the o in one play and not that of a sequence f successivelays it is preferable p this to express as follows:The player's strategy consistsneither of playing but w n \"tails\" or of playing \"heads,\" of playing \"tails\" ith the probability w of and \"heads\"ith the probabilityof to One might imagine that in order play Matching Pennies a in rational way the player will before his choicein each play decide by 1 or c some 50:50hancedevice whether to play \"heads\" \"tails.\" The him w protects from loss.Indeed, hatever point is that this procedure the opponent follows, the player's expectation the outcome for strategy 2 of the play will be zero. This is true in particular if with certainty the a and opponentplays \"tails,\" nd also if with certainty he plays \"heads\"; if he like the player himself may play both \"heads\" nd a also,finally,

i17.1.2.

to Thus, if we permit a player in Matching Pennies use a \"statistical\" t i.e.o \"mix\" the possibleways of playing with certain probastrategy, h bilities (chosen him), then he can protect imself against loss.Indeed, by we specifiedabove such a statistical strategy with which he cannot lose, o irrespectivef what his opponentdoes.Thesameis true for the opponent, t i.e.he opponent can use a statistical strategy which prevents the player 4 o from winning, irrespectivef what the player does. w The reader ill observethe greatsimilarity of this with the discussions f o 5 14.5.In the spirit of those discussionsit seems to consider legitimate and the 50 :50 statistical zero as the value of a play of Matching Pennies a a mixture of \"heads\"nd \"tails\" s a goodstrategy. is The situation in Paper,Stone,Scissors entirely similar. Common sensewill tell that the good way of playing is to play all threealternatives 6 with the probabilitiesof each. The value of a play as well as the inter1E.g. e could throw a die of coursewithout letting the opponent seethe result h and then play \"tails\" if the number of spotsshowing is even, and \"heads\" if that number is odd. 1 his probability of winning equals his probability of losing, because under these conditions the probability of matching as well as that of not matching will be J, whatconduct. ever the opponent's 3 Say p, I p. For the player himself we used the probabilities J, J. 4 All this, of i t course,n the statistical sense:hat the player cannot lose,means that his probability of winning. his probability of losing is That he cannot win, means that the former is 2 to the latter. Actually eachplay will be won or lost, sinceMatching Pennies knows no ties.

3 w \"tails,\"ith certainprobabilities.

I.e.

chancedevicecould be introduced as before. Thedie mentioned one. E.g.he player could decide\"stone\" t above,would be a possible show, \"paper\" is 3 or 4 spotsshow, \"scissors\" if 5 or 6 show.)))
A

We mean specifically

in (14:C:d), (14:C:e) 14.5.1.

in footnote if 1 or 2 spots

1,

MIXEDSTRATEGIES.THE SOLUTION

145

pretation of the above strategy as a good onecan be motivated as before, 1 again in the senseof the quotation there,

17.2.1.is plausibleto try to extendthe results found for Matching It and Pennies Stone,Paper, cissorso all zero-sumtwo-persongames. t S We use the normalized form, the possiblechoices the two players of for being n = 1, , Pi and r 2 = 1, , 2, and the outcome player 1 3C(ri,r2), as formerly. We make no assumptionof strictdeterminateness. i Let us now try to repeathe procedure hich was successfuln 17.1. t w ; i.e. let us again visualize players whose\"theory\"of the gameconsistsnot in the choice definite strategies ut rather in the choice severalstrategies of of b 2 with definite probabilities. Thus player 1 will not choosea number TI = 1, , pi i.e.the correspondingstrategy 2^ but 0i numbers i, , iflj the probabilitiesof thesestrategiesSi, , 2J?, respeca tively. Equally player 2 will not choose number TJ = 1, , 2 i.e. the corresponding , ifo s the probastrategy S but /9 2 numbers iji, bilities of these strategies2', ' 2 respectively. Sincetheseprob* *
>

17.2. Generalization

of

This Viewpoint

abilities belongto disjoint but exhaustive alternatives,the numbers T|, *?r, aresubjectto the conditions

2|,

(17:l:a)
(17:1 :b)
and to no others.
We form the

^ fe 0, ri-i{T| = 1; all^^O, r,-l^ = 1.

all

to S0t Then rj in the senseof 16.2.2. In this setup a player does not, as previously, choose strategy, but his with which he plays all possible and choosesnly the probabilities strategies o t he is going to play them respectively. This generalization meetshe major of the not strictly determinedcaseto a certain point:We have difficulty of seenthat the characteristic that casewas that it constituted a definite 3 for each disadvantage player to have his intentions found out by his 4 on p. 144. 8 That these in the examor probabilities were the same for all strategies (i, 1 J, that this ples of the last paragraph) was, of courseaccidental. It is to be expected in equality was due to the symmetric way in which the various alternatives appeared now on the assumption that the \"appearance of probabilities those games. We proceed a strategy was the essential thing, while the particular values -were in formulating accidental. 'TheA > Oof 14.7.1.)))
1In Stone,Paper,Scissors there existsa tie, but of losing is ^ the probability of winning, and no
no lossstill means that the probability gain means the reverse. Cf footnote

= {1, , &J and ty = {771, the above conditionsstatethat { must belong to Sp^ and
vectors

,
>

J?0

}.

|,i

146

ZERO-SUMTWO-PERSONGAMES: THEORY

of ties are determined, is a very effective way to achieve a degree such By protection: this devicethe opponent cannot possiblyfind out what the2 player's strategy is going to be,sincethe player does not know it himself. information i Ignorances obviously a very goodsafeguardagainstdisclosing directly or indirectly.

1 opponent. Thus oneimportant consideration for a player in such a game h is to protect imself againsthaving his intentionsfound out by his opponent. at Playing severaldifferent strategies random, so that only their probabili-

player's freedom of action. It may happen, after all, that he wishes to o o play one definite strategy to the exclusionf all others; r that, while desirto use certainstrategies with certainprobabilities,he wants to exclude ing 3 absolutelythe remaining ones. We emphasizethat thesepossibilitiesare of not perfectly within the scope our scheme.A player who does wish to at certainstrategies all will simply choose them the probabilities for play of zero. A player who wishes to play one strategy to the exclusion all others will choosefor this strategy the probability 1 and for all other the strategies probabilityzero. Thus if player 1wishesto play the strategy 2\\i only, he will chooseor f
v the coordinate ector 6 Ti (cf. 2 and the vectors and 6 T
T?

17.2.2. may It

now

seemthat

we have incidentally

the restricted

16.1.3.). Similarly for player 2, the strategy .

>

In view of all theseconsiderations call a vector we

ry

of S$ or a vector i

a statistical or mixed strategy of player 1or 2, respectively. The > coordinatevectors 5 Ti or 6 T correspond, we saw,to the original strategies as or r 2 i.e. ^i or S^ of player 1or 2, respectively. We call them strict TI 2 or pure strategies. of
S/3t
*

17.3. Justification

of the

Procedure As

a contradictionbetweentwo vital throughout our discussions.On the one hand we have always insisted that our theory is a static one (cf.4.8.2.), that we analyze the course and 1But not necessarily only one. the \" \" 2 If the a opponent has enough statistical experiencebout the player's style, or if he is very shrewd in rationalizing his expected t behavior, he may discoverhe probabilities whether and how this frequencies of the various strategies. (We need not discuss But may happen. Cf. the argument of 17.3.1.) by the very conceptof probability and randomness nobody under any conditions can foresee what will actually happen in case. (Exception must be made for such probabilities as may vanish; any particular cf. below.) 8 In this casehe t clearly increaseshe danger of having his strategy found out by the opponent. But it may be that the strategy or strategies in question have such intrinsic advantages over the others as to make this worth while. This happens, e.g.in an extreme form for the \"good\" strategies of the strictly determined case(cf.14.5., particularly

17.3.1.this stagethe readermay At

Applied

to an

Individual

Play

becomeneasy and perceive u a which we have stresseds equally viewpoints

have

in (14:C:a), (14:C:b) 14.5.2.).)))

MIXED STRATEGIES.THE SOLUTION

147

be used. How arethey to be solved? Our 17.3.2. answeris this: a To beginwith, the ultimate proof of the resultsobtainedin 14.nd 17. and i.e.he discussionsof 14.5. of 17.8.do not contain any of thesecont flicting elements.So we could answer that our final proofs are correct even though the heuristic procedureswhich lead to them are questionable. c But even these proceduresan be justified. We make no concessions: Our viewpoint is staticand we are analyzing only a singleplay. We are stage for the zero-sumtwotrying to find a satisfactory theory, at 'this we arenot arguing deductivelyfrom the firm person game. Consequently t basis of an existing theory which has already stood all reasonableests for but we aresearching such a theory.2 Now in doingthis, it is perfectly a for legitimate us to use the conventional tools of logics, nd in particular that of the indirect proof. This consistsin imagining that we have a satisfactory theory of a certaindesiredtype,3 trying to picture the consequences of this imaginary intellectual ituation, and then in drawing conclusions s from this as to what the hypotheticaltheory must belike in detail. If this i for is process appliedsuccessfully,t may narrow the possibilities the hypothetical theory of the type in question to such an extent that only one
will

of one play and not that of a sequence f successivelays (cf. o But p on the otherhand we have placed considerations the concerning danger of one'sstrategy being found out by the opponent into an absolutely central 14.7.1. again the last part of 17.2.). can and How position (cf. 14.4., the strategy of a player particularly one who plays a random mixture of several different strategies be found out if not by continuedobservation! We have luled out that this observationshould extendover many plays. Thus it would seem necessaryto carry it out in a singleplay. Now even if the rules of the game shouldbe suchas to make this possible if they lead to long and repetitiousplays the observationwould be effected only in of gradually and successively the course the play. It would not be available at the beginning. And the whole thing would be tied up with various dynamical considerations, while we insisted on a statictheory Besides, 1 the rulesof the gamemay not even give such opportunitiesfor observation; a of they certainly do not in our original examples Matching Pennies, nd S Stone,Paper, cissors.Theseconflictsand contradictionsoccurboth in the with the discussionsof 14. where we used no probabilitiesin connection of o w choice a strategy and in our presentdiscussions f 17. here probabilities

17.1.).

i.e.

cannot imagine (yet) what it would be like, if we had one. no worsethan any other indirect proof in any part All this is in its own domain science the per absurdum proofs in mathematics and in physics).)))

1 \"gradual,\" \"successive\" observations of the behavior of the opponent within one play. 1Our method is, of course,he empirical one: We are trying to understand, formalize t This is, and generalizethose features of the simplest games which impress us as typical. w after all, the standard method of all sciences ith an empirical basis. 3 This is full cognizanceof the fact that we do not (yet) possessne, and that we o

I.e.

(e.g.

of

148

ZERO-SUMTWO-PERSONGAMES: THEORY

down to nothing i.e. that it and cessful,\" that it narrows the possibilities 2 demonstratesthat a consistenttheory of the desiredkind is inconceivable. Let 17.3.3. us now imagine that there exists a complete of the theory zero-sumtwo-persongamewhich tells a player what to do, and which is absolutelyconvincing. If the playersknewsuch a theory then eachplayer would have to assumethat his strategy has been \" found out\" by his opponent. Theopponentknowsthe theory, and he knowsthat a playerwould be 8 of unwisenot to follow it. Thus the hypothesisof the existence a satisour investigation of the situation when a playfactory theory legitimatizes er's strategy is \" found out\" by his opponent. And a satisfactory theory 4 can existonly if we are able to harmonize the two extremes and F 2, FI of player 1\"found out\" or of player 2 \"found out.\" strategies Forthe original treatment free from probability (i.e. pure stratewith the extentto which this can be done was determinedin 14.5. gies) is We saw that the strictly determinedcase the one where thereexistsa on that basis. We arenow trying to push further, by theory satisfactory w using probabilities(i.e. ith mixed strategies). The same device which when will we usedin 14.5. therewereno probabilities do again, the analysis of \"finding out\" the strategy of the other player. It will turn out that this time the hypotheticaltheory can be determined (not merely for the strictly determinedcase completelyand in all cases cf. the theory is found we must justify it independentlyby a direct 6 This was done for the and strictly determined casein argument. in 17.8. we shalldo it for the present complete theory
After

1 i device. Of course,t can happen that the applicationis even more \" suc-

possibilityis left,

i.e.hat t

the theory is determined, discoveredby this

17.6.). 17.5.1.,

14.5.,

1 Thereare several important examplesof this performance in physics. Thesuccest sive approacheso Specialand to General Relativity or to Wave Mechanicsmay be of viewed as such. Cf. A. D'Abro: The Decline Mechanism in Modern Physics, New

1This too occurs physics. The N. Bohr-Heisenberg in analysis of \"quantities which are not simultaneously observable\" in Quantum Mechanicspermits this interpretation. Cf N. Bohr:Atomic Theory and the Descriptionof Nature, Cambridge 1934 P.A.M. and Dirac:ThePrinciples of Quantum Mechanics,London Chap. 8 Why it would be unwise not to follow it is none of our concernat present;we have assumed that the theory is absolutely convincing. That this is not impossible will appearfrom our final result. We shall find a theory which is satisfactory ; neverthelessit implies that the player's strategy is found out by his opponent. But the theory gives him the directions which permit him to adjust himself no so that this causes loss. (Cf.the theorem of 17.6. the discussion of our complete and

York

1939.

1931, I.

solution in 4 a theory using our present devices only. Of coursewe do not pretend to be ableto make \"absolute\" statements. If our present requirements should turn out to be unfulfillable we should have to look for another basisfor a theory. We have actually done this onceby passing from 14.(with pure strategies) to (with mixed strategies). * The indirect argument, as outlined above,gives only necessary conditions. Hence it may establish absurdity (per absurdum proof), or narrow down the possibilities to one; but in the latter caseit is still necessary show that the one remaining possibility is to

I.e. 17.8.)

17.

satisfactory.)))

MIXED STRATEGIES. THE SOLUTION

149

17.4. Minorant The

17.4.1.
element

Games(For Mixed Strategies) a Our present picture is then that player 1 chooses n arbitrary
and the Majorant

sf

Thus if player 1 wishesto play the strategy 2^ only, he will chooseor f the coordinateector d Ti (cf. 16.1.3.) v ; similarly for player2, the strategy T. ZS and the vectors rj and 6 of in ignorance f o We imagine again that player 1 makes his choice

from

a S$ and that player 2 choosesn arbitrary element from

i?

of choice y and viceversa. player 2's h The meaning is, of course,that when these choicesave been made 1 will actually use (every) TI = 1, , ft\\ with the probabilities player = 1, , 0iwith the probabilities TI and the player 2 will use (every) TI Since their choicesare independent, the mathematical expectation iyv is of the outcome

(17:2)

K(

>

0i

0t

n-lr.-l

3C(n,

n)^,.

the In other words, we have replaced original gameT by a new oneof essentiallythe same structure,but with the following formal differencei: o Thenumbersn, r 2 the choices f the players arereplacedy the vectors b o The function 5C(ri,r 2) the outcome, r rather the \" mathematical \" o b expectation of the outcome f a play is replacedy K( , ). All these considerationsdemonstrate the identity of structureof our presentview o the of T with that of 14.1.2., soledifference being the replacement f above. This isomorphism , , K( , ), described TI, T2, 3C(n,TJ) by the application of the samedeviceswhich we used on the original suggests T, the comparisonwith the majorant and minorant games FI and F* as
,
q
>
17
t\\ t\\

in described

and player2 choosesis ij h afterwards in full knowledgeof the chosen his opponent. In Tj the by orderof their choicesis reversed. So the discussionof 14.3.1. applied t literally. Player 1,choosinga certain , may expecthat player 2 will o his choose /j , so as to minimize K( , 17 ) ; i.e. player Ts choice f leadsto > h the value Min~* K( , ). This is a function of alone; enceplayer 1
first
17

14.3.1., 14.2., 14.3.3. h Thus 17.4.2. in TI player 1choosesis

his so as to maximize Min-+K( , should choose a play of TI is (for player

1))))

ij

). Thus the value of

150

ZERO-SUMTWO-PERSONGAMES: THEORY
vj

= Max- Min-> K(T,7). = Min-> Max-* K(

*?

Similarly the value of a play of F 2 (for player 1) turns out to be v'2 really of
f

rj

).

(The apparent assumptionof rational behavior of the opponent does not (14:B:a)-(14:B:e) matter,sincethe justifications (14:A:a)-(14:A:e),

for player 1than F 2 constitutesa proof

vi))

14.3.1. 14.3.3. apply literally.) and again we can argue that the obvious fact that As in 14.4.1.
of))

FI

is less favorable

and that if this is questioned,a rigorousproof is containedin (13 in :A*) to The x, y, therecorrespond our , , K.l If it shouldhappen

13.4.3.

17

that))
v'i))

then the considerationsof and and

14.5. literally. The arguments (14:C:a)apply o (14:D:b) c d (14:D:a), loc. it., eterminethe concept f a \" good\" { (14:C:f),
i?

fix

the \"value\" of a play of (for the player 1) at v' = v { = v'2 2

All

this happens by
</>

. in if and (13:B*) 13.4.3. only

> rj

if

a saddle point of K

exists. (Thex, y, therecorrespondto our , , K.) 17.5. GeneralStrict Determinateness We t 17.5.1. have replacedhe YI, v 2 of (14:A:c)and (14:B:c)our by present v{, v'2, and the above discussionshowsthat the lattercan perform the functions of the former. But we are just as much dependent upon v[ = v 2 as we were then upon v l = v 2. It is natural to ask, therefore,

whether thereis any gain in this substitution. Evidently this is the caseif, as and when there is a betterprospect of having vj = v'2 (for any given F) than of having vi = v 2 We called F preferableto make a disstrictly determined when Vi = v 2 ; it now seems F tinction and to designate for Vi = v 2 as specially strictly determined, and for vi = v 2 as generally strictly determined. This nomenclature is justified only providedwe can show that the former impliesthe latter.

are vectors, sequences real numbers of and ) it is perfectly admissible to view eachas a single variable in the maxima and 10 Their domains are, of course,the sets 8ft , Sft minima which we are now forming. which we introduced in 17.2. 8 For an exhaustive repetition of the arguments in question cf. 17.8.)))
'nit

'''

1Although
1

>

->

17

i.e.

(1, ,$

MIXED STRATEGIES. THE SOLUTION

151

: This implication is plausibleby common sense Our introduction of mixed h t strategies as increasedhe player'sability to defendhimself against having his strategy found out; so it may be expected vj, v 2 actually lie between that one may even asserthat t Vi, v 2. Forthis reason V! ^ vl g v'2 g v 2. (17:3) of t (This inequality secures, course,he implication just mentioned.) To excludeall possibilityof doubt we shall give a rigorous proof of
I (17:3). t is convenient to prove this as a corollary of another lemma.
First 17.5.2. we prove this lemma:

(17:A)

Forevery { in S^

-* Min-K( f ,

r,

= Min= Min Tj

ft

*
3C(n,))

r^l
0i

JC(n,

Forevery

77

in
>

Sp t
77

Max- K(

= Max= MaxTi

ft

n-lr.-l

rz

^ JC(TI,r . -l
2)r/ Tj
>

/5i

the

17.2.) gives
0!

is W Proof: e prove the first formula only ; the proof of the second exactly same, nly interchanging Max and Min as well as ^ and ^ . o = 6 Ti, (cf. 16.1.3. the end of and vector Considerationof the special
>
r?

Sincethis is true for all r', so 2

Minrcd^rOfc^g TS^i>^)Mv;=Zl X (T1 r ')k'r, = T!-!= l rj-lTt -l

>

ft

01

ft

^1

(17:4:a)Min- ^
1))

01

ft

JC(n,njfc^ g Minr ;

^ 3C(n, r;)fv

01

On the otherhand, for all r 2

ft))

ac(n,))

Given any

77

in

Since

r.-l)))

S0,
8

multiply

this by

r\\

T^

and sum over r 2 =

1,

y Ta

= 1,herefore t

152

ZERO-SUMTWO-PERSONGAMES: THEORY
ti

ft

r,-l T!-!
(17:4:b)Min0i

3C(r,,rt

)^ ,
r

fi

Min,,

fi-1

r, JC(n r,){
f

results. Sincethis is true

0t

for all i; , so 3C(n,

^ r.)^*, Min ,
r

% OC(n,r)

fti

v))

If we combinethe above formulae with the definition of v'^ v'2 in then we obtain

(17:4:b) togetherthe desiredrelation. (17:4:a), yield

= Max- MinTt = Min-MaxT
0i

17.4.,

(17:5:a) (17:5:b)

vi v

K(rl9 T,) V
?))

In Theseformulae have a simple verbal interpretation: computing v't we needonly to give player 1the protection against having his strategy found out which liesin the useof (insteadof TI); player 2 might as well proceed a in the old way and use r 2 (and not ). In computingvj the roles reinterThis is plausibleby common-sense:belongsto the gameFI v{ changed. there and after player 1 and is fully (cf. 17.4. 14.2.); player 2 chooses of he no informed about thechoice player 1, hence needs protectionagainst having his strategy found out by player 1. For vi which belongsto the are gameTj (cf.id.) the roles interchanged. t Now the value of v( becomes if we restricthe variability of in ^ i the Max of the above formula. Let us restrictt to the vectors = d r cf. 16.1.3. the end of 17.2.). and Since (rj =
17

!,-,/>!,

'\\

0i

TI-I our this replaces expressiony b

fri,

lf ri)*r/i = 3e(r' ri),

T Max/ Min Tj $(/!,) = VL

Z

So we have shown that

vi

v;.

Similarly(cf. the remark at the beginningof the proof of our lemma above)
of restriction
77

to the

rj

8T

establishes
v
vj.)))

MIXED STRATEGIES.THE SOLUTION

with Together

163

v{

(17:3)
as desired.

these (cf. 17.4.), inequalitiesprove Vl g v; ^ vi g v f ,

17.6. of the Proof

Main

Theorem

17.6. have establishedthat generalstrictdeterminateness = vj) We (v( of holds in all cases special trict determinateness = v 2) us is to be s (vi t expected.That it holds in some further casesas well i.e.hat we can have Vi = v'2 but not Vi = v 2 is clearfrom our discussions Matching of 1 Penniesand Stone,Paper,Scissors. Thus we may say, in the senseof

17.5.1. the passage special generalstrictdeterminateness from to does that constitutean advance. But for all we know at this moment this advance it may not cover the entireground which shouldbe controlled; could hapthat certaingames F are not even generallystrictly determined, i.e. pen we have not yet excluded possibility the
vl

< v 2.

If this possibility should occur,then all that was said in would and to an increasedxtent finding out one's e s apply again opponent'strategy would constitute a definite advantage

14.7.1.

A'

additional hypothesesas to structedwithout some

and it would be difficult

to

= v2 seehow

vj

> 0,

a theory of the game should be con\"who finds out whose

strategy.\" The decisivefact is, therefore, that it can be shown that this never

happens. Forall gamesF

i.e.

v;

= v2

(17:6) Max- Min-K(7,\"7 ) = Min-Max- K(7,7), in or equivalently (againuse (13:B*)13.4.3.x, y, therecorresponding the
to our

K): A saddle point of K( ( , ) exists. This is a general heorem valid for all functions K( { , t
, q,
rj

(17:2)

K(7,7) =

Ti-lT,-!

can be interpreted

The coefficients3C(ri,r t ) areabsolutelyunrestricted; form, as described they * * in 14.1.3. a perfectly arbitrary matrix. The variables $ , are really 1In both games YI 14.7.3.), 1,vi 1 (cf. 14.7.2., while the discussion of 17.1.
17

<t>

??

) of the form

W(TI,

. r,) r>i

as establishing v{

vj

0.)))

154

ZERO-SUMTWO-PERSONGAMES: THEORY
,

o sequences f real numbers {i,

^ and

rji,

<]^\\

their domains
>
1?

being the setsS^, S0 (cf. footnote 1on p.150).The functions K( , ) t of the form (17:2) calledbilinear forms. are With the help of the results of 16.4.3. proof is easy.1 This is it: the in 16.4.3. the We apply (16:19:a), (16:19:b) replacing i,j, n, m, a(i j) thereby our TI, r 2, 0i, 2,3C(ri,r 2) and the vectors w , x thereby our , .
y ry

If

holds, (16:19:b) then we have a

0i

in

Sp with i
r2

i.e.ith w

n-i

T 2j 3C(Ti,TI){I ^

for

1,

|8,,

0i

Min Tj

JC(n,r 2)

0.))

Therefore the formula

of (17:5:a)17.5.2. gives vl ^ 0.
rj

If

9 :a) (16:1 holds, then we have an

0*

in

Sp with
2

2) r,-l

3C(ri,rz)^ ^

for

n = 1,

0,,

1This theorem occurred was proved first in the original publication of one of the and authors on the theory of games: von Neumann: \"Zur Theorie der Gesellschaftsspiele,\" Math. Annalen, Vol. 100(1928), p.295-320. p A slightly more general form of this Min-Max problem arises another question of in in mathematical economics connection with the equations of production von Neumann: \"tlber ein okonomisches Gleichungssystem und eine Verallgemeinerung des Brouwer'schen Fixpunktsatzes,\" Ergebnisseeines Math. Kolloquiums, Vol. 8 (1937), p.73-83. p It seemsworth remarking that two widely different problems related to mathematical economics although discussed entirely different methods lead to the same by mathematical problem, and at that to one of a rather uncommon type: The \" Min-Max type.\" Theremay be some deeperformal connections here, as well as in some other directions, mentioned in the second paper. The subject should be clarified further. The proof of our theorem, given in the first paper, made a rather involved use of sometopology and of functional calculus. Thesecond papercontained a different proof, which was fully topological and connected the theorem with an important deviceof \" that discipline: so-called Fixed Point Theorem\" of L.E. Brouwer. the This aspect was further clarified and the proof simplified by Kakutani: \"A Generalization of rtrouwer's Fixed Point Theorem,\" Duke Math. Journal, Vol. 8 pp.457-459. All theseproofs are definitely non-elementary. The first elementary one was given Ville in the collectionby E. Boreland collaborators,\"Trait6 du Calcul desProbby abilit& et de sesApplications,\" Vol. IV, \"Applications aux Jeux de Hasard,\" Paris N des (1938), ote by Ville: \"Sur la TheorieG6ne>ale Jeux ou intervient THabilete* des Joueurs,\" pp. The proof which we are going to give carriesthe elementarization initiated by Ville further, and seems be particularly to simple. The key to the procedure of is, and particularly with the course,the connection with the theory of convexity in results of 16.4.3.)))

/.

J.

S.

J.

(1941),

J.

105-113.

J.

2:

/.

16.

MIXEDSTRATEGIES.THE SOLUTION

155

i.e. ith w
ft*

MaxT
Therefore the formula

3e(n,r 2)i7 T ^

0.))

of (17:5:b)17.5.2. gives v'2 ^ 0. So we see: ither vi ^ or v'2 ^ 0,i.e. E Never vi < < v'2. (17:7) Now choose arbitrary number an

byJC(n, r 2)

w. 1

and
w

t replacehe
0i r,

function 3C(ri, r 2) that is as

> >

This replaces , K(
01

T;)

_
by

>

K(
02
77^

T;)

^ =

lie in S^, fi^ and so

vj, v 2 arereplacedy vi b w, v'2 thesevi w, v'2 w gives

= ^ TI T!-! r,-l

=1

_ _ by K( , )
TJ

] l r, =

0*

TI TJ

17

w. Consequently

w.2 Therefore applicationof (17:7)o t

(17:8)

Never vi

< w < v'2.

Now w was perfectly arbitrary. Hence vi < v'2 it would be possible for to choosew with vi < w < v'2 thus contradicting (17:8). o vi < v'2 is S impossible,and we have proved that vi = v 2 as desired. This completes the proof.

17.7. omparison of the C

the result of

Treatments

by

Pure and

by

17.7.1. going further Before

Mixed Strategies

let us oncemore considerhe meaning of t

Theessentialeature of this is that we have always v{ = v 2 but not always f = v 2, i.e. lways general trict determinateness, not always special but s Vi a strict determinateness the beginning of 17.6.). (cf.
it Or, to express mathematically : We have always

(17:9)

Max- Min-K(7,7) = Min- Max-K(7,7), n v , 6

than in T. 2 This is immediately footnote.)))

1 the game Tis replaced a new one which is played in preciselythe same way as by T excepthat at the end player 1 gets less(and player 2 gets more) by the fixed amount w t

I.e.

clear if

we remember the interpretation

of

the

preceding

156

ZERO-SUMTWO-PERSONGAMES: THEORY

i.e.
M (17:10) ax-Min-

SC^rj)^
0i

Min-+Max->V n *
w Using (17:A) e may even write for this

ft*

(17:11) Max^Minrt
(17:12)

2) 3C(n,r

ri

2 Min^ Max,t % 3C(n,T)i?v

But we do not alwayshave

r MaxT| Min^^COn,2) = Min Ti MaxTj 3C(ri,r2).

Let us compare17:9) (17:12): is always true and (17:12) and (17:9) ( is not. Yet the differencebetweentheseis merelythat of , ij , K and n,n, 3C. Why doesthe substitution of the former for the latter convert the into untrue assertion (17:12) the true assertion (17:9)? r is Thereasonis that the5C(ri, a ) of (17:12)a perfectly arbitrary funcwhile is tion of its variables n, r2 (cf. 14.1.3.), the K( , rj ) of (17:9) an
o f ,& extremelyspecialunction of its variables , ij i.e.f the i, ' t *?/v namely a bilinearform. (Cf.the first part of 17.6.)Thus Vi; the absolute generalityof JC(TI,r 2) rendersany proof of (17:12) impossible, while the special bilinearform nature of K( , rj ) providesthe basisfor l as the proof of (17:9), given in 17.6.
* *

>

>

that While 17.7.2. this is plausibleit may seemparadoxical K( , q) t should bemorespecialhan 3C(ri, r2), although the former obtained from w the latter by a process hich bore all the marks of a generalization: We o obtained it by the replacement f our original strict conceptof a pure in i.e. as strategy by the mixed strategies, described 17.2.; by the replacement of n, r2 by
,
17

.
i\\

But a closernspectiondispelsthis paradox. K( , ) is a very special i function when comparedwith 3C(ri,r2) ; but its variableshave an enormously
1

ways. Its significance from particularly in 3.7.1.)))

i with concept s connected the this opensup a rather interesting perspective: nemight investigate O Mathematically 1 i which other concepts,n placeof \"mathematical expectation, would not interfere with for i.e. the result of 17.6. zero-sum two-person games. with our solution, The conceptof \"mathematical expectation\" is clearly a fundamental one in many

tion

\"

That the K(
wherever

17

) is a bilinear

probabilities

intervene.

is due to our useof the \" mathematical expectaIt seemssignificant that the linearity of this o existence f a solution, in the sensein which we found one.
form

'

the point of view of the theory of utility

was brought forth

MIXEDSTRATEGIES.THE SOLUTION

157

wider domain than the previous variablesn, r 2. Indeed had the finite T\\ varies over the setS^, which is a set (1, , pi) for its domain, while l surface in the 0i-dimensional inear space (cf.the (Pi l)-dimensional S^ > and for r 2 and rj .l end of 16.2.2. 17.2.). Similarly

There are actually among the in S special points which correspond to the various TI in (1, , 0i). Given such a r\\ we can form (as in
fti

v and the 16.1.3. at the end of 17.2.) coordinate ector = 6 Ti, expressing of the choice the strategy S^ to the exclusionof all others. We can correlatespecial in S^ with the r 2 in (1, , fit) in the sameway: Given such = 6 r , expressing choice the of the v a T2 we can form the coordinateector o strategy 22 to the exclusionf all others.
rj > rj

Now clearly:))

/-ISCW, T'I) *,**,

, r 2).2

Thus the function K( , T;) contains, in spite of its special the character, entirefunction 3C(ri,r 2) and it is therefore really the more general oncept c of the two, as it ought to be. It is actually more than 3C(ri,r 2) sincenot all
,
77

form areof the special

17

6 r , 5 %,

not all mixedstrategies pure.* are

Onecould say that K( ,

o is the extensionf 3C(ri,r 2) from the narrower

5T domain ofthe ri,r i.e.ofthei, 6 r to the widerdomain ofthe , rj 2 t to all of Sp, Sp from the pure strategieso the mixedstrategies.Thefact t
TJ

i.e.

is that K( , ) is a bilinear form expresses merely that this extension carried w out by linear interpolation. That it is this process hich must be used, is 4 of due to the linear character \" mathematical expectation/* of course
1

that Observe
r\\
17

contains

; but

= i, , there is a fundamental

t l with

difference.

the components TI , = , 0i, also In 3C(n,TJ),r\\ itself is a variable. In

!,
17

is ), is a variable, while n is, so to say, a variable within the variable. and this function as such is the variable a function of TI (cf.the end of of K( , q ). Similarly for r 2 and T; > Or, in terms of n, r 2 3C(n, r*) is a function of n, TJ while K( , ) is a function of functions of n, *2 (in the mathematical terminology a functional). 2 The meaning of this formula is apparent if we consider hat choice of strategies w K( ,
actually

16.1.2.)

d r i,
8

>

\"

be used effectively with positive probabilities. o connection between the concept f numerical utility and the linear mathematical expectation\" was pointed out at the end of 3.7.1.)))
4 Thefundamental

I.e. strategiesmay several

5 r represent.

>

158
truth

ZERO-SUMTWO-PERSONGAMES: THEORY
the we can express of

we see 17.7.3, Reverting to (17:9)-(17:12), now that

his strategy found out by his opponent if he can use the mixed strategies states n, , ij instead of the pure strategies r 2. (17:11) that this remains true if the player who finds out his opponent'sstrategy uses the T\\, r2 while only the player whose strategy is being found out enjoys the pro7 2) of tection the , r; . Thefalsity of ( 1 :1 , finally, showsthat both players and particularly the player whosestrategy happens to be found out may
not forego with impunity the protection the of

as and the (17:9)-(17-11) untruth of (17:12)follows: a (17:9), express eachplayer is fully protectedgainst having (17:10) that

{,

>

> rj

17.8. Analysis

of GeneralStrict Determinateness

a We shall now reformulate the contentsof 14.5. s mentioned at the end of 17.4.with particular considerationof the fact established in 17.6. every zero-sumtwo-persongame F is generally strictly deterthat mined. Owing to this result we may define:

17.8.1.

= v' = Max-Min- K(7,7) Min-Max- K(7,7)

Let us form two setsA, B subsets of S^, S/^,respectively in analogy of These to the definition of the sets A, B in (14:D:a), (14:D:b) to arethe setsA+, B+ of (the corresponding our K). We define:

7). Sa-|-K(7, and (Cf.also (13 in 13.5.2. the end of 13.4.3.) :C*)
=

13.5.1.

14.5.1.
*>

</>

(17:B:a)

is the set of those (in Sp) for which MinK( , f assumesits maximum value, i.e.or which
A

Min-K(7, 7) = Max- Min-K(7,7) = V. * * { (17:B:b)

B is the set of those
77

(in

Sp) for which Max- K( , t

7)

assumesits minimum value,

f i.e.or which

= Min-Max- K(7, = V. Max-^K(7, It is now possibleto repeatthe argumentation of 14.5.

1 (14:C:a)-(14:C:f)asinl4.5.
1

7)

7)

e In doingthis we shalluse the homologous numeration for the assertions (a)-(f)will therefore appearin an order different from the natural one. This was sincethe enumeration there was basedupon that of 14.3.1., 14.5., those 14.3.3., in paragraphs followed a somewhat different route.)))

equally true in and the argumentation

MIXEDSTRATEGIES.THE SOLUTION
We

159

observefirst:
for Player 1 can, by playing appropriately,secure himself a gain ^ v', irrespectiveof what player 2 does. f Player 2 can, by playing appropriately,secureor himself a gain ^ v', irrespectivef what player 1does. o

(17:C:d)

L Proof: et player 1choose{ from

for does,i.e. every 2 chooserj from
T; wehave

K(

irrespectiveof what player 2 ) ^ Min->K( { , rj ) = v'. Let player

A

; then

i?

Then irrespectiveof what player 1does, for every the , we have K( , ) g Max->K( { , rj ) = v'. This completes proof. is clearlyequivalent to this: ( Second, 17:C:d)

B.
77

i.e.

(17:C:e)

Player 2 can, by playing appropriately, make it sure that the gain of player 1 is ^ v', i.e. prevent him from gaining > v', irrespectiveof what player 1does. Player 1 can, by playing appropriately, make it sure that the gain of player 2 is ^ v', prevent him from gaining > v', irrespectiveof what player 2 does.

i.e.

17.8.2. we may now assert on the basisof (17:C:d) (17:C:e) and Third, in and of the considerations the proof of (17:C:d) that:
(17:C:a) (17:C:b)
The good way (combination of strategies)for 1 to play to the game F is to chooseny a belonging A, A beingthe set
of

The good way (combinationof strategies)for 2 to play to the gameF is to chooseny a belonging 5, B beingthe set
rj

above. (17:B:a)

of

above. (17:B:b)

of those of

Fourth, combination of the assertions of (17:C:e)ives: g

o (17:C:d)r

equally well

(17:C:c)

i If both players 1 and 2 play the game F well i.e.f belongsto A and ij belongsto B then the value of K( , to v'. will be equal to the value of a play (for 1),

{
rj

i.e.

in and We add the observationthat (13:D*)13.5.2. the remark concerning the setsA, B before (17:B:a), above give togetherthis: (17:B:b)

(17:C:f)

Both players 1and 2 play the game F well

to A and
t;

i.e.{ belongs
is a saddlepoint

belongsto B if and only

if

rj

of K(7,7)-!)))

160))

ZERO-SUMTWO-PERSONGAMES: THEORY))

All this should make it amply clear that v' may indeedbe interpreted a as the value of a play of F (for 1), nd that A, B contain the good ways of playing T for 1,2, respectively. Thereis nothing heuristic or uncertain about the entire argumentation (17:C:a)-(17:C:f). We have made no \" extra hypotheses about the \" intelligence of the players, about \"who has found out whose strategy\" etc. Nor are our results for one player based upon any belief in the rational conduct of the other, a point the importanceof which we have repeatedly stressed. (Cf. the end of 4.1.2.; also 15.8.3.) Further 17.9.

in and 17.9.1. last results (17:C:c) (17:C:f) 17.8.2.ive also a The g o of characterizationf the elements our present solution, simple explicit i.e.f the number v' and of the vector setsA and B. o

of Characteristics GoodStrategies

B,and we shall do this by means of

loc. A By (17:C:c) cit., , B determine hencewe needonly study A, v'; id. (17:C:f)
t\\

Accordingto that criterion, belongsto A, and belongsto B if and only if , i? is a saddle point of K( , rj ). This means that

Max- K(7',7

Min-K(T,7'
We

o make this explicit y using the expression (17:2)f b

,
17

17.4.1. and

of for of ), and the expressions the lemma (17:A) 17.5.2. K( , y ') Then our equationsbecome: Max-*,K( ', if) and MinI'

17.6. K( for

V))

Max
T))

r.-lr.-l))
that Considering

Min T

'

ac(r

lf))

^ r,-!
r; r

TI
'))

, we can . ^^
*
fin

also write for these

t _
\\

Tj

[Max
1))

T t

X
1))

It

^s

r,-l))

r, 1

; - [-Min {r ^))1 t
T7

Now on the left-hand side of theseequations the v tj Tf have coefficients these which are all ^ O.1 The TI, Tf themselvesare also ^ 0. Hence 1Observe how the Max and Min occurthere!)))

MIXED STRATEGIES.THE SOLUTION

181

equationshold only when all terms of their left hand sidesvanish separately. w I.e. hen for eachn = 1,. , $\\ for which the coefficientis not zero, we have r = 0;and for eachr* = 1, , fa for which the coefficientis not
zero, we have
Summing
T/

up;
belongsto
A

TI

= 0.

(17:D)

and

rj

aretrue:

belongsto B
,

if

and only if these

JC(TI, TZ)rj Tl

For each n = 1,
not assumeits maximum

0i,for

0t

which

(in

n) we have

TI

r,-l =
0i

does

0.
OC(ri, T2)
ri

For each r 2 = 1,
not assumeits minimum

, ft, for which (in r 2) we have

r;Tj

n-i =

does

0.

It is easy to
,
77

formulate these principlesverbally. They expresshis: If t then excludesll strategies\\ which are a aregoodmixed strategies, T
t\\

a r not optimal (for player 1)against , and rj excludesll strategies2 which are not optimal (for player 2) against ; i.e. , rj are as was to be expectedoptimal against eachother. Another remark which may be made at this point is this: 17.9.2.

(17:E)

exists eachplayer a goodstrategy which is a pure strategy. for

The game is speciallystrictly determinedif and only if there

a In view of our past discussions, nd particularly of the processf geno to eralization by which we passed from pure strategies mixed strategies, this assertion may be intuitively convincing. But we shall also supply a mathematical proof, which is equally simple. This is it that We sawin the last part of 17.5.2. both Vi and v\\ obtain by applying

fe Max->to Min Tj 2) & (TI> T*)^* only *

\\-i
1,

with

different domainsfor

-*

:The set i.e.

the of all 5 Ti (TI = , 00for YI, and all of 8^ for vi; i.e. purestrategies a the first case, nd the mixed ones in the second.Hence = v\\, in Vi the two maxima areequal if and only if the maximum of the seconddomain is assumed(at leastonce)within the first domain. This meansby (17:D) b above that (at least)one pure strategy must belongto A, i.e. e a goodone.

I.e.

(17:F:a)

if and only if there exists for which is a pure strategy.))) strategy

Vi

= vi

player

1a

good

162
Similarly:

ZERO-SUMTWO-PERSONGAMES: THEORY

(17:F:b)

i.e.i v

if and only if there existsfor player 2 a good strategy which is a pure strategy. Now vj = v'2 v' and strict determinateness eans Vi = V2 = v', m
V2

=v

= v; and v 2 = v. So (17:F:a), (17:F:b) together(17:E). give

Mistakesand 17.10.
Their Consequences.Permanent
Optimality

discussionshave made clearwhat a good mixed is. Let us now say a few wordsabout the othermixedstrategies. strategy from \" goodness\" those strategies the for We want to express distance (i.e. and vectors {, rj ) which arenot good; to obtainsomepicture of the conseo o quences f a mistake i.e.f the use of a strategy which is not good.Howw ever, we shallnot attempt to exhaustthis subject, hich has many intriguing
ramifications.

Our 17.10.1. past

Forany

in

S0i and any

??

in

S0 we form the numerical functions

2

(17:13:a) (17:13:b)
(17:13**) (17:13:b*)
The definition
v'

- Min-K(T,7),)) 0()= Max- K(, - V.

a(7) = V
) (
ft(

equally :A) By the lemma (17 of 17.5.2.

-*) =

-> O

Min Tj % 3C(n,T,)$ TI>

Tt

-l

Max,,

r,-l

* V

jc(n,

r^ -

v;

= Max- Min-K(7,7) = Min-Max- K(7,7)

t guaranteeshat always

(7) ^ 0, 0(7)^ 0. in and And now (17:B:a), (17:C:b)17.8. that 7 imply (17:B:b) (17:C:a), is goodif and only if a( ) =0, and is good if and only if 0( ) =0.
ij
77

v t expressingheir distancefrom goodness. The expliciterbal formulation even more plausible: The of what a( ),0( ) are,makesthis interpretation
>

for Thus a( ^)jft(r)) areconvenient numerical measures the general{,

t\\

>

>

>
77

formulae

above or (17:13:b*) make (17:13:b)(17:13:a*), (17:13:a),

clear)))

MIXED STRATEGIES.THE SOLUTION

163

2 of the worst that can happen under the given conditions. > It must be understood,however, that a( ( ),0( ) do not disclosewhich strategy of the opponentwill inflict this (maximum) lossupon the playerwho > is using { or rj It is, in particular,not at all certain that if the opponent
i?

how much of a lossthe player risks relative to the value of a play for him 1 by using this particular strategy. We mean here \"risk\" in the sense

usessomeparticulargoodstrategy,i.e.n 17 in 5or a o in A, this in itself a > impliesthe maximum lossin question. If a (not good) or rj isusedby the for player,then the maximum losswill occur those 17 ' or { of the opponent,

'

for which))

K(,
17

') = Min 7

K(,

),

K(7,7) = Max K(7,7),))

optimal againstthe given 17 > And we have never ascertainedwhether any fixed o or { can be optimal ^ against all or
17 77

i i.e.f ' is optimal againstthe given

, or

'

17.10.2. us therefore call an ' or a which is optimal againstall Let or (17:14:b) in 17.10.1. 7, 7~ for all 7 or 7 i.e.which fulfills (17:14:a),
TJ

'

permanently optimal. 1

is necessarily permanently optimal 17 or this shouldbe clearconceptuallyand an exactproof is easy.8 But good;
Any we I.e. mean by lossthe value of the play minus
> >

'

'

for player 1 and , 77 )) = K( 2 Indeed,using the previous footnote and

'

2. (-v') - (~K( , ) - v' (17:13:b) (17:13:a), (7) - - Min-.K(T, 7) = Max- (v' - K(7,7)1, 1 - = Max- |K(7,7) - v'|. 3(7) = Max- K(7, 7)
>
17

the

v' actual outcome:

for player

K( ,

1?

e I.e.achis a maximum
8
77

loss.

is analogous. Proof:It suffices to show this for rj ; the proof for * which is a with Let 'be permanently optimal. Choose optimal against 17',i.e.

'

'

By definition

Thus by

in (17:C:f) 17.8.2.)))

|
*,

= K(T',V) Max-K(7,7)

7 K(7*. ) = Min- K(7*. ). 7

If

> 17'

is a saddle point of K( ,

17

) and

therefore

V belongs

to B

i.e.is good it

164

ZERO-SUMTWO-PERSONGAMES: THEORY

also permanently optimal? A the question remains: re all good strategies exist? And even:Do any permanently optimal strategies In generalthe answeris no. Thus in Matching Penniesor in Stone, the Paper, cissors, only good strategy (for player 1 as well as for player 2)is S
1 i}, respectively. If player 1playeddifferently 2 heads\" 2 or always \" stone\" then he would loseif the oppoe.g.lways a 3 3 nent countered playing \" tails\" or \"paper.\" But then the opponent's by is not good i.e. ?} or (i, },respectively either. If the strategy the good strategy, then the player's mistake would not opponent played f

rj

= (i,i) or {i,
\"

|,

{,

i,

4 matter.

We shall get another w way in connectionith

this may be summed up by saying that while our good strategies not areperfect from the defensivepoint of view, they will (in general) get the maximum out of the opponent's(possible)mistakes, i.e.hey arenot t
All

19.10.3.

of example this in a more subtle and complicated and and Poker the necessityof \" bluffing,\" in 19.2

It should be remembered, are however, that our deductions of 17.8. a neverthelesscogent;i.e. theory of the offensive, in this sense,is not possible without essentially new ideas. The readerwho is reluctant to acceptthis, ought to visualize the situation in Matching Penniesor in simplicity of these two Stone, Paper,Scissorsonce more;the extreme gamesmakes the decisivepoints particularly clear. Another caveat against overemphasizingthis point is:A great deal u goes,in common parlance, nder the name of \"offensive,\"which is not at w all \"offensive\"in the above sense, i.e. hich is fully coveredby our present theory. This holds for all gamesin which perfectinformation prevails, 10.3. as will be seenin 17. 6 Also such typically\"aggressive\" operations(and 6 which arenecessitated imperfect information) as \"bluffing\" in Poker. by We 17.10.3.conclude remarkingthat thereis an important class of by (zero-sumtwo-person) games in which permanently optimal strategies exist. Theseare the games in which perfect information prevails, which and a we analyzed in 15. particularly in 15.3.2., 15.7.Indeed, small 15.6., modification of the proof of special strictdeterminateness thesegames, s of a given loc. it., ould suffice to establishthis assertion too. It would give c w permanently optimal pure strategies.But we do not enterupon theseconsiderationshere.
w Cf.17.1.ny other probabilities would leadto losses hen \" found out.\" Cf.below. A = * or (1, , 0),respectively. 'Thisis 0 (1, ) 0 *This is S\" or {0,1, ),respectively. 0 (0, 1) \" 4 I.e. bad strategy of \" heads (or \"stone 1') can be defeated the only by \"tails\" (or \"paper\,") which is just as bad in itself. 1Thus Chessand Backgammon are included. The preceding discussion appliesrather to the failure to \"bluff.\" Cf. 19.2. and
1
i?

for calculated the offensive.

- -

'

19.10.3.)))

MIXED STRATEGIES. THE SOLUTION

165

Sincethe gamesin which perfect information prevails are always spea cially strictly determined(cf.above),onemay suspect more fundamental b connectionetweenspeciallystrictly determinedgamesand those in which e permanently optimal strategies xist(for both players). We do not intend to discuss thesethings here any further, but mention the following facts which arerelevant in this connection: :a) (17:G determined.

e It can be shown that if permanentlyoptimal strategies xist (for both players) then the game must be speciallystrictly

(17:G:b) (17:G:c)

is It can be shown that the converseof (17:G:a)not true. strictdeterCertain refinements of the conceptof special minatenessseemto bear a closerelationshipto the existence r

of permanently optimal strategies. 17.11. Interchange of Players. Symmetry The Let us considerthe roleof symmetry, or more generally the effects of interchanging the players 1 and 2 in the game F. This will naturally be a continuation of the analysisof 14.6. the As was pointed out there, this interchange f the players replaces o r function OC(T!,2) by -3C(r n). The formula (17:2) 17.4.1. 17.G. and of 2,

17.11.1.

it ) is to replace by K( r; , { ). In the terminology of 16.4.2.,replace matrix (of 5C(ri,r 2) cf. we the by its negative transposedmatrix. Thus the perfect analogy of the considerationsin 14. ontinues; gain c a we have the same formal results as there, provided that we replace

showsthat the effect of this for K( {,

>

TJ

14.1.3.)

o TI, r 2, 3C(ri, T2) by , rj , K( f , T;). (Cf. the previous occurrencef this in and We saw in that this replacement 3C(ri,r 2) by 3C(r2, n) carries of v 2, VL A literal repetitionof those considerationsshows Vi, v 2 into

17.4. 17.8.)

14.6.

carriesj, v 2 into v'2, vj. Summing up:Interchanging players1, , the v 2 carries i, v 2, v'lt v'2 into v 2, Vi, v'2, v't . V The result of 14.6. establishedfor (special) trictdeterminateness was s into v = Vi = v 2. In the absenceof that v = Vi = v 2 is carried that property no such refinement of the assertion was possible. At the present we know that we always have generalstrictdeterminateness,so that v' = v'i = v 2. Consequently this is carriedinto))
in S we Verbally the contentof this resultis clear: ince succeededdefining a satisfactoryconcept f the value of a play of F (for the player 1), ', it is o v that this quantity should changeits sign when the roles only reasonable of the players areinterchanged. We can also state rigorously when the game F is symmetric. This is the casewhen the two players 1 and 2 have preciselythe same)))

now that the

o correspondingreplacement f K( ,

77

) by

K(

17

17.11.2.

166
rolein
from it

ZEROSUM TWO-PERSONGAMES: THEORY

i it, i.e.f the game F is identical with that gamewhich obtains 2 by interchanging the two players 1, . According to what was said above, this means that JC(n,TI) = -JC(r, I), T

or equivalently that

K(T,7)= -x(7,7).
This property of the matrix 3C(ri,r 2) or of the bilinearform K( { , rj ) was 12 and introducedin 16.4.4. called skew-symmetry. ' v 2, v 2, In this caseVi, v 2 must coincide with Vi; hence Vi = and sinceVi ^ v 2, so Vi ^ 0. But v' must coincide with v'; therefore
t we can even asserthat v'

= O.8

fair.

T So we see: he value of eachplay of a symmetricalgame is zero. It shouldbe noted that the value v' of eachplay of a game F couldbe zero without F being symmetric. A gamein which v' = will be called

i S of The examples 14.7.2., illustrate this: tone,Paper, cissorss 14.7.3. S without hencefair); Matching Penniesis fair (cf. 17.1.) symmetric (and
4 beingsymmetric.
1 >

matrix is defined by

For a

3C(n, r 2) or for the corresponding 3C(n, Tj)

bilinear form

K(

77

) symmetry

= X(r 2, TI),

or equivalently

by))

It is remarkable that symmetry of the game F is equivalent to skew-symmetry, and not to symmetry, of its matrix or bilinear form. 1Thus skew-symmetry means that a reflection of the matrix scheme of Fig. 15 in carriesit into its on its main diagonal (consisting of the fields (2, 2), own negative. (Symmetry, in the senseof the preceding footnote, would mean that it carriesit into itself.) Now the matrix scheme Fig. 15is rectangular; it has 2 columns and /3i rows. In of Henceit the case under consideration its shapemust be unaltered by this reflection. must be quadratic, 0i = #2 This is so, however, automatically, sincethe players 2 are assumed to have the same role in 3 This is, of without course,due to our knowing that v\\ = v 2 Without this the general theorem of we should assertfor the vj, v 2only the same which we obtained above the Vi, v 2:vj = vj and since ^ v^, sov\\ ^ 0. for vj 4 The players 1 and 2 have different rolesin Matching Pennies:1 tries to match, and 2 tries to avoid matching. Of course, nehas a feeling that this difference is inesseno tial and that the fairness of Matching Penniesis due to this inessentiality of the assymeThis could beelaborated try. upon, but we do not wish to do this on this occasion. A better exampleof fairness without symmetry would be given by a game which is grossly but in which the advantages and disadvantages of each player are so unsymmetric, value v' = results. judiciously adjusted that a fair game A not altogether successfulattempt at such a game is the ordinary way of \" Rolling Dice.\" In this game player 1 the \" player \" rolls two dice,each of which bear the numbers Thus at eachroll any total 2, may result. Thesetotals))) ,

14.1.3.

(1,1),

etc.)

1,

i.e.

r.

(16:F) 16.4.3.

i.e.

i.e.

1,

6.

,12

MIXED STRATEGIES. THE SOLUTION

167

In a symmetricalgame the setsA, S of (17:B:a), in are (17:B:b)17.8. = in the final criterion A obviously identical. Since = B we may put
of it (17:D) 17.9.We restate for this case: In a symmetrical game, belongsto A (17:H)
T\\

if
(*i

and only if this is

3C(ri, T2)
ri

true:Foreachr 2 =

1,

0* for which (in r 2 ) we have

T

not assumeits minimum

^ r,-l =

does

0.

remark of 17.9., seethat the we Usingthe terminology of the concluding above conditionexpresses this:{ is optimal against itself. 17.11.3. results of 17.11.1., that in every symmetrical The 17.11.2. can be combinedwith (17:C:d) 17.8.Then we obtain in game v' =

this:

(17:1)

ately, avoid loss1 irrespectiveof what the opponent does. We can statethis mathematically as follows: If the matrix 3C(n, r 2) is skew-symmetric, a then thereexists vector))
with))
for))

In a symmetrical game eachplayer can,by playing appropri-

n))

r2

=))

This could also have been obtained directly, because coincides ith w it the last result (16 in 16.4.4. seethis it suffices to introduce there To :G) our presentnotations:Replace i,j, a(i,j) thereby our r\\ t r 2, X(n, r 2) the and the w thereby our { .
have the following probabilities:))

Total

Chanceout of 36)
Probability)

.....)
2)
1)

3) 2)

4) 3)

5) 4)

6)
5)

7)

8)
5)

9)
4)

10)
3)

11)
2)

12)
1)

6)
A)

A)

A)

A)

A)

A)

A)

A)

A)

A)

A))

The rule is that if the \" player \" rolls 7 or (''natural\") he wins. If he rolls 2, 3, or 12 he loses. If he rolls anything else (4, 5, 6, or 8, 9, 10)then he rolls again until he rolls either a repeatof the original one (in which casehe wins), or a 7 (in which casehe loses). Player 2 (the \"house\") has no influence on the play. In spite of the great differences of the rules as they affect players 1 and 2 (the
we do not detail, shows out of a total of 495;
\"player\"

11

are and the \"house\") their chances nearly equal:A simple computation, which that the \"player\" has 244chances against 251for the \"house,\" the value of a play played for a unit stake

i.e.

244 251
495))

---

is))

1.414%.))
it may

more was intended.

1

Thus the approximation

to fairness is reasonably good,and

be questioned

whether

s I.e.ecurehimself

a gain *

0.)))

strictdeterminateness all F can be derivedfrom that oneof the symmetric of ones. The proof has a certaininterestof its own, but we shall not discuss it heresincethe derivation of 17.6. more direct. is

168 ZERO-SUMTWO-PERSONGAMES: THEORY t It is even possibleto base our entiretheory on this fact, i.e.o derive from the theoremof 17.6. the above result. In other words: The general
The possibilityof protecting oneselfagainst loss (in a symmetricgame) exists due to our useof the mixedstrategies , ij (cf.the end of 17.7.). only If the playersarerestricted pure strategies r2 then the dangerof having to TI,

e one'sstrategy found out, and consequently of sustaining losses, xists. To see this it suffices to recallwhat we found concerning Stone,Paper, Scissorscf. 14.7. 17.1.1.). shall recognizehe samefact in conand We t ( nectionwith Poker the necessityof \"bluffing\" in 19.2.1.))) and

CHAPTER IV ZERO-SUMTWO-PERSONGAMES: EXAMPLES

18.Some ElementaryGames
We 18.1.1. have concludedour generaldiscussionof the zero-sum twoto of person game. We shall now proceed examine specificexamples such w games. These examplesill exhibit better than any generalabstract discussionscould, the true significance of the various componentsof our theory. They will show, in particular, how some formal steps which are dictated our theory permit a direct common-sensenterpretation. It by i will appear that we have herea rigorousformalization of the main aspects a of such \"practical\"nd \"psychological\" phenomenaas thoseto be men1 tioned in 19.2., and 19.16. 19.10. t The 18.1.2. size of the numbers0i, 2 i.e.he number of alternatives confronting the two players in the normalized form of the game gives a for of first reasonable estimate the degree complicationof the game F. This Thecase either,or both, of thesenumbersis 1may be disregarded: that at would mean that the player in questionhas no choice all by which he can 2 influence the game. Therefore the simplest games of the class which interestsus arethose with

18.1. Simplest Games The

i is that saw in 14.7. Matching Pennies such a game;ts matrix scheme of in Another instance sucha game is Figure was given in Figure12
We

(18:1)
14,
id.))

0i

= fa = 2.

13.4.1.
1)

2)

1)

1 3C(1, )) 3C(2,1))
Figure

3C(1,2))
3C(2,2)))

2)

27.

i.e. the Let us now consider most generalgame falling under (18:1), if to 27. This applies,e.g., MatchingPennies the various ways under Figure t of matching do not necessarilyrepresenthe same gain (or a gain at all),
1

unfit

zerosum. Cf. 12,2.

1Thus the game would

of We stressthis because the widely held opinion that for rigorous (mathematical) treatment.
really
169)))

thesethings are congenitally

then,

be one of oneperson;but

of course, o n

longer

of

170

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

nor the various ways of not matching the sameloss (or a loss at all).1 We the t proposeto discussfor this case resultsof 17.8.,he value of the gameF h and the setsof goodstrategies , 4 Theseconcepts ave beenestablished (based on the theorem of 17.6.); p by the generalexistential roof of 17.8. but we wish to obtain them again by explicit computation in this special a case, nd thereby gain somefurther insight into their functioning and their

5.

There are certain trivial adjustments which can be made on a gamegiven by Figure27, and which simplify an exhaustive discussion
considerably. First it is quite arbitrary which of the two choices f player 1 we denote o t t by TI = 1 and by TI = 2;we may interchangehese, i.e.he two rows of
the

possibilities.

18.1.3.

matrix.

columns of the matrix. Finally, it is also arbitrary which of the two playerswe call 1 and which 2;we may interchangehese, i.e. eplaceC(ri,r 2) by 3C(ri,r 2) (cf. 14.6. t 5 r This amounts to interchanging the rows and the columns and of the matrix, and changing the sign of its elements esides. b = 8 possible Putting everything together, we have here adjustments, all of which describessentiallythe same game. e

o i Second,t is equally arbitrary which of the two choices f player 2 we t denoteby TI = 1 and by r2 = 2;we may interchangehese, i.e.he two t

17.11.).

2X2X2

to the discussionproper. This will consist considerationof several alternative possibilities,the \"Cases\" be to below. enumerated TheseCases distinguishedby the various possibilitieswhich exist are for the positions of those fields of the matrix where X(TI, r 2) assumesits maximum and its minimum for both variables n, r 2 together. Their delimitations may first appear to be arbitrary; but the fact that they lead to a quick cataloguingf all possibilities o justifies them ex post. Consider accordingly MaxVTj 3C(ri, r 2) and Min VTj JC(ri, r 2). Each one of thesevalues will be assumedat least nceand might be assumedmore o 2 than once;but this does not concern at this juncture. We begin now us
in the
with

We 18.2.1. proceed now

18.2. DetailedQuantitative

Discussionf TheseGames o

is possible to choosea field where MaxVTj is assumedand one where Min Ti ,Tj is assumed,so that the two areneitherin the same row nor in the same column. By interchanging TI = 1,2 as well as r 2 = 1,2 we can make the firstmentioned field (of MaxTi,Tj) to be (1,1). The second-mentioned field
1 Comparison of Figs. 12and 27 shows that in Matching Pennies3C(1, ) 3C(2,2) 1 (gain on matching) ;3C(1, ) 2 3C(2, (losson not matching). 2 In Matching Pennies (cf. footnote 1 above) the Max Tj ,ri is 1 and is assumed and (2, 2), while the Min V r f is and is assumed at 2) and (2,

18.2.2. (A): It Case

the definition of the various

: Cases

1)

- -1

-1
at

(1,1)

-1

(1,

1).)))

SOME ELEMENTARYGAMES
we (of Min ri .r)) must then be (2,2). Consequently have

171

(18:2))) Therefore (1, ) is a saddlepoint.1 2 Thus the game is strictly determinedin this case ajid = v = 3C(1, ). v' (18:3) 2 Case(B):It is impossibleto make the choices prescribed as

above:

18.2.3.

Choosethe two fields in question (of MaxTi ,Tj and Min ri ,Tj); then they the samecolumn. If the former shouldbe the case, t then interchangehe players 1,2, so that thesetwo fields are at any rate in the same column.2 2 2 By interchanging TI = 1, as well as r 2 = 1, if necessary,we can again make the first-mentioned field (of MaxVT2) to be (1, ). Sothe column in 1 f question is r 2 = 1. The second-mentionedield (of Min ri ,Tj) must then be 3 (2, I). Consequentlywe have:))

arein the samerow or in

2 3C(1,1) = JC(1, ) or JC(2,2) = 3C(2,1) areexcludedbecause for the Max ,rj and Min T ,Tj fields they'would permit the alternative choices Tj of (1, ), (2, 1) or (1, ), (2, 2), thus bringing about Case 4 1 (A). 2 to So we can strengthen (18:4)
Actually

(18:5)
We must now

18.2.4. Case(Bi):

make a further disjunction:

3C(1, ) OC(2, 2 2) (18:6) can Then (18:5) be strengthenedto 3C(1, ) > 3C(1, ) ^ OC(2, ) > 3C(2,1). 2 1 2 (18:7)
Therefore (1, ) is again a saddlepoint. 2 Thus the game is strictly determinedin this

case and again too;

1).

(18:8)
1

that Observe we had to take 2) and not (2, Recall 2 This interchange of the two players changes the sign of every matrix element (cf above),hence it interchanges Max Ti ,T2 and Min Ti>Tj . But they will nevertheless be in the same column. 3 To be precise: might also be It But then 3C(n, r,) has the same Max Tl ,Tsand Then we can use (2, 1) also for Min rj>Tj Min r .T , and soit is a constant. <3C(1,1) 3C(2,2) and JC(1,2) 3C(2, are perfectly possible,as the example of Matching Pennies shows. Cf.footnote 1 on p. 170and footnote 1 on p. 172.)))

13.4.2.

= v =3C(1, 2).
(1,

(1,1).

1)

172

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

Case 18.2.6. (B2):

3C(1, ) < 3C(2,2) 2 (18:9) to can Then (18:5) be strengthened (18:10) 3C(1,1) ^ 3C(2,2) > 3C(l,2)
2 The game is not strictly determined.

1 JC(2, I).

It is easy however, to find

satisfying the
We can 2

i.e. goodstrategies, a { in A and an in B,by condition(17:D)of 17.9.We can do even more : characteristic
rj

choose so that
77

->

^ r,-l

3C(ri,TJ)T;TI is the samefor all r\\ and

- so that

for 2} 3C(fi,T2) ri is the same all r*. Forthis purposewe need:))

1 U*' 1;

2) 2){i 1){, (30(1, +3C(2, =5C(1, +3C(2, fa. l){i

) )

This means))

- 3C(2,1

1)-JC(2, -3C(1,2):OC(1, 1).

theseratios, subjectto the permanentrequirements fa + fa = 1 fa 0, fa ^ =1 r/2 ^ r?i + i?i S 0, ratios (i.e. right-hand sides in the t This is possiblebecausehe prescribed
We must satisfy We are (18:12))positiveby (18:10). have \"
> **

and further
171

772

- 3C(2, - 3C(1,21) - 3C(2,1)' ) 2) X(l, 1) -30(1, ^ 1 3C(1, ) + 3C(2,2) - 3C(1, ) - JC(2, 2 OC(2, ) - 3C(1, ) 2 2 3C(1,1) + 3C(2,2) - 3C(1, ) - 3C(2,1)' 2 ge(i,i)-ae(2,) i _ 2 5C(1,1) + 3C(2,2) - 3C(1, ) - JC(2, 1)'
3C(2,2) 1 3C(1, ) + 5C(2,2)
1)*

_ _ _ _
these

172

_ _)) _ _ _
A,

We can even show that

elements.

17

areunique,i.e. that

B possess other no

1This is actually the casefor Matching Pennies. Cf footnotes i on p. 170and 4 on p. 171. * Clearly vi MaxTi Min f| 5C(n, r t ) 3C(1, ), v, MmTj Max Tl3C(r,,r t ) JC(2, 2), 2 80 V t

<

Vj.)))

SOME ELEMENTARY GAMES

Proof If either
77

173

or

rj

were somethingelse than we found above,then

respectivelymust have a component0,owing to the characteristic condition (17:D) of 17.9.But then rj or { would differ from the above I values since in these both componentsare positive. So we see:f either > > or differs from the above values, then both do. And then both must have a component0. Forboth the othercomponentis then 1, .e. are i both
or
77

1 coordinate ectors. Hencehe saddle point of K( , T; ) which they reprev t sent is really oneof 3C(ri,r 2), cf. (17:E) in 17.9.Thus the game would be

but strictly determined; we know that it is not in this case. This completes proof. the
All

four expressions in

are (18:11) now seento

namely

2 JC(1, ) X(2,1) and by (17:5:a), in this (17:5:b)17.5.2. is the value of v'. So we have 2) 3C(1, )OC(2, ) 2 1 1 3C(1, )3C(2, na-nn v' v ^10. 0, 1 ly x(^ ^ + 2) _ 2) _

JC(1, ) + 3C(2,2) 1

1 2) 3C(1, )3C(2,

- 3C(1,2)3C(2,1) - -

have the samevalue,

18.3. Qualitative
which

^ ^ ^
Characterizations

can The 18.3.1. formal results in 18.2. be summarizedin various ways of Thefields (1, ), 2,2) form one diagonal of the matrix scheme Fig.27, 1( the fields (1, ), (2,1) form the otherdiagonal. 2 We say that two setsof numbersE and F areseparatedeitherif every t elementof E is greaterhan every elementof F, or if every elementof E o is smallerthan every element f F. I n Consider ow the Cases (Si),B2) of 18.2. n the first two cases ( (A), the game is strictly determinedand the elementson one diagonal of the 2 from those on the other. In the last casethe matrix are not separated a on gameis not strictly determined, nd the elements one diagonal of the matrix of Thus separation the diagonalsis necessaryand sufficient for the game T not being strictly determined. his criterionwas obtained subject to But the threeprocesses of the usemade in 18.2. the adjustmentsof in affect neither strict determinateiiess of adjustment described 4 is our of nor separation the diagonals. Hence first criterion always valid.
8 from areseparated those on the other.

make their meaning clearer. We begin with this criterion:

18.1.3.

18.1.3.

We

it: restate :

Case(Bi) 3C(1,1) JC(2, 2) > 3C(1, ) OC(2,1)by (18:10). 2 is evident sincetheseare only changesin notation, inessential is Thesecond immediately verified.)))
4 The first

^ 3C(2,2) by (18:7). 8

Ml, 0|or (0,1). 2 Case (A):3C(1, 1)

OC(1,2)

^ 3C(2, 2) by (18:2).Case (Bi):C(1,1)> JC(1,2) J

for the

game.

174

ZERO-SUMTWO-PERSONGAMES:EXAMPLES

(18 :A)

other. In 18.3.2.case(#2), i.e. hen the game is not strictly determined,both w

iy 5**

Thegame is not strictly determinedif and only if the elements on one diagonal of the matrix are separated from those on the

the (unique){of A and the (unique) of B which we found, have both components 0. This, as well as the statement of uniqueness,is unaffected 1 So by adjustmentsdescribedin 18.1.3. we have: (18:B) If the game is not> strictly determined,then thereexistsonly in one good strategy { (i.e. A) and only one good strategy i (i.e.n 5), and both have both their componentspositive. b I.e. oth playersmust really resortto mixedstrategies. :B) According to (18 no componentof or rj ( in A, vj in B) is zero. shows that the argument which preceded Hencethe criterion of 17.9. which was then sufficient without (18:11) being necessary is now must necessary(and sufficient). Hence(18:11) be satisfied,and therefore are all of its consequences true. This applies in particular to the values and given after (18:11), to the value of v' given in (18:13). $1, {2, i?i, All theseformulae thus apply whenever the game is not strictly determined.
77 772

p. 99 (we allow for a moment any 0i, 2) we say that a row (sayT() or a column (say 2) majorizes another row (sayr\") or column (sayr' respectively,f this is true for their i 2 I correspondingelements without exception. .e.if 3C(rJ, r 2) ^ 3C(r7, r 2) for all T2, or if 3C(n, T{) ^ 3C(rj,/,')for all r*. This concepthas a simple meaning:It means that the choiceof r( is at leastas good for player 1as that of r\" or that the choice r( is at of most as good for player 2 as that of r' and that this is so in both cases 2 2 irrespectiveof what the opponentdoes. Let us now return to our present problem (/?i = 2 = 2). Consider I a 2 (A), (J5t ), (/? ) of 18.2. n the first two cases row or a again the Cases 8 column majorizesthe other. In the last case neither is true.4 Thus the fact that a row or a column majorizesthe other is necessaryand sufficient for T being strictly determined. Like our first criterionthis is of subjectto the usemadein 18.2.the adjustmentsmade in 18. And, as of there, those processes adjustment affect neither strict determinateness nor majorization of rows or columns. Henceour present criteriontoo is always valid. We restate The game r is strictly determinedif and only if a row or a : (18C) column majorizesthe other. 1Thesetoo are immediately verified. 'Thisis, of course, n exceptionaloccurrence: general the relative merits of two a In w alternative choices ill depend on what the opponent does. 'Case(A): Column 1 majorizes column 2, by (18:2)Case (Bi): ow 1 majorizes R

In a generalmatrix 3C(ri, r 2) cf. Fig.15on

We 18.3.3. now formulate another criterion:

'), '

r'

1.3.

it:

row

2 by 4 Case(J?t):

(18:7). a excludesll four (18:10)

possibilities,

as is easily

verified.)))

SOME ELEMENTARY GAMES))

175))

That 18.3.4. the condition of ( 18 is sufficientfor strict determinateness :C) is not surprising: meansthat for one of the two playersone of his possible It choicessunder all conditionsat leastas goodas the other (cf. above). Thus i he knowswhat to do and his opponentknowswhat to expect, which is likely to imply strict determinateness. Of coursethese considerations imply a speculationon the rationality of the behavior of the other player, from which our original discussionis free. The remarks at the beginning and at the end of 15.8. to a apply certain extentto this, much simpler,situation. What really matters in this result (18 however is that the necessity :C) of the condition is also established; .e.that nothing more subtle than i outright majorization of rows or columns can causestrict determinateness. It shouldbe rememberedthat we areconsideringthe simplestpossible case: i = 2 = 2. We shall seein 18.5. conditionsget more involved how 0 in all respects hen pi, fa increase. w

18.4. F Discussion SomeSpecific ames. (Generalizedorms of Matching G of

in (a) MatchingPennies its ordinary form, where the JC matrix of Figure on 27 is given by Figure12 p.94. We know that this game has the value v' = and the (unique) goodstrategies

18,3.

18.4.1. following are some applicationsof the results of 18.2. The and

Pennies)

T will, g (Cf. 17.1. he formulae of 18.2. of course, ive this immediately.) w 18.4.2. MatchingPennies, here matching on heads gives a double (b) by premium. Thus the matrix of Figure27 differs from that of Figure12 the doublingof its (1,1) element:))
1)

7=7= {*,*}

2)

(\\) 1)

-1)
1))

2)

-1)

Figure 28o. the and h The diagonalsare separated1 2 are > than 1), ence goodstrate( are unique and mixed (cf. (18: (18:B)). using the pertinent A), By gies we obtain the value formulae of case(#2) in

18.2.5.,

and the goodstrategies))

beobservedthat the premium put on matching headshas increased him value of a play for player 1 who triesto match. It also causes the
to)))

Itwill

176))

ZERO-SUMTWO-PERSONGAMES: EXAMPLES))

h chooseeadslessfrequently, sincethe premium makesthis choice plausible t and therefore dangerous. The direct hreat of extraloss by beingmatched on heads influences player 2 in the same way. This verbal argument has some plausibilitybut is certainly not stringent. Our formulae which yielded this result, however, were stringent. 18.4.3. Matching Pennies,where matching on heads gives a double (c) premium but failing to match on a choice(by player 1) of heads gives a triple penalty. Thus the matrix of Figure27 is modified as follows:))
1)

2)

1)

2)

-3)
1))

2)

-1)
Figure

286.

The diagonalsareseparated and 2, are > than 1, 3),hencethe good (1 are strategies unique and mixed (cf. as before). The formulae used before give the value))
v

-))

and the good strategies))

to of We leave it to the reader formulate a verbal interpretation this in the same senseas before. The construction other examples of result,

of this type is easy along the lines indicated. that these variants of Matching Pennies 18.4.4. We saw in (d) in a way, the simplestforms of zero-sumtwo-persongames. By this are, w circumstance acquirea certaingeneralsignificance, hich is further they and corroborated the results of 18.2. indeed we found there that by this class of games exhibits in their simplest forms the conditionsunder which strictly and not-strictly determined casesalternate. As a further in the samespirit we point out that the relatedness f these addendum o only one particular aspect. Other gamesto Matching Penniesstresses gameswhich appear in an entirely different material garb may, in reality, well belongto this class. We shall give an examplef this: o The gameto be considered an episode the Adventures of Sherlock is from

18.1.2.

18.3.:

1-2 Holmes.

1 Conan Doyle: he Adventures of Sherlock Holmes, New York, T p 550-551. 1The situation in question is of courseagain to be appraised a 1938, p. of many as paradigm conflicts in practicallife. It was expounded as such by possible Morgenstern: Wirtp schaftsprognose,Vienna, 1928, . 98. The author doesnot maintain, however, some pessimistic views expressed or in id.

0.

\" Vollkommene

Voraussicht

our solution also answers doubts in the same vein expressed K. by in den exacten Wissenschaften, Fortschritte \"Einige neuere Fortschritte in der exactenBehandlung Socialwissenschaftlicher Probleme,\" Vienna, 1936, pp. and 131.)))

alokonomie, Vol.
Accordingly

6, 1934.

und wirtschaftliches

Gleichgewicht,\"

Zeitschrift

fur Nation-

Menger: Neuere

117

SOME ELEMENTARY GAMES))

177))

Sherlock t Holmesdesireso proceed from London to Dover and hence to the Continent in order escaperom ProfessorMoriarty who pursues to f him. Having boarded the train he observes,as the train pulls out, the appearanceof Professor Moriarty on the platform. SherlockHolmes takes it for granted and in this he is assumedto be fully justified that his adversary, who has seenhim, might secure special a train and overtake him. Sherlock Holmes faced with the alternative of going to Dover or is of leaving the train at Canterbury, the only intermediate station. His is adversary whoseintelligence assumedto be fully adequateto visualize the thesepossibilities has the same choice. Bothopponentsmust choose of decision. o place their detrainmentin ignorancef the other's corresponding If, as a result of these measures,hey should find themselves,in fine, on t the same platform, Sherlock t Holmesmay with certainty expecto be killed Holmesreaches over unharmed he can make D by Moriarty. If Sherlock
good his escape. What are the good strategies, Holriies? This particularly for Sherlock game has obviously a certain similarity to Matching Pennies,Professor to Moriarty being the one who desires match. Let him therefore be to to a t H player 1, nd Sherlock olmesbe player 2. Denotehechoice proceed Dover by 1and the choice quit at the intermediatetation by 2. (This to s appliesto both r\\ and T2.) Let us now considerhe X matrix of Figure27. The fields (1, ) and t 1 which it Sherlock Holmes, (2,2) correspondto ProfessorMoriarty catching matrix is reasonableo describey a very high value of the corresponding t b s 100.The field (2, 1) signifiesthat SherlockHolmesucelement, say to Dover,while Moriartystopped at Canterbury. This is cessfullyescaped and* Moriarty'sdefeat as far as the presentactionis concerned, shouldbe describedy a big negative value of the matrix element in the orderof b magnitudebut smallerthan the positivevalue mentionedabove say, 50. The field (1,2) signifies that Sherlock Holmes scapes oriarty at the M e intermediatetation, but fails to reachthe Continent. This is best viewed s 0 as a tie,and assignedthe matrix element . The5Cmatrix is given by Figure29:))
1)

2)

1)

100)

2)

-50)
Figure

100))

29.

As in (b), (c) above,the diagonalsareseparated100 > than 0, 50); ( is hencethe good strategiesare again unique and mixed. The formulae used beforegive the value (for Moriarty)))
40)))

178

ZERO-SUMTWO-PERSONGAMES: EXAMPLES
for Moriarty,
rj

and the goodstrategies (

for Sherlock Holmes)

7 = {*,!}, 7 = {*,D.
Thus Moriarty should go to Dover with a probability of 60%, while Holmesshouldstop at the intermediatestation with a probability Sherlock of60 %, the remaining 40 % beingleft in eachcase the otheralternative. l for

18.5. Discussion f SomeSlightly o

The 18.6.1. generalsolutionof the zero-sum two-persongame which we obtained in 17.8. rings certain alternatives and conceptsparticularly b into the foreground:The presence absence strict determinateness, the or of value v' of a play, and the setsA, S of goodstrategies.Forall thesewe obtained very simpleexplicit and characterizations determinationsin 18.2. Thesebecameven more striking in the reformulation of those results in e

More Complicated Games

18.3.

the resultsof 18.2., were obtainedby explicitcomputationsof the most 18.3.

This simplicity may even lead to some misunderstandings. Indeed,

in of elementary sort. The combinatorial criteria (18 (18:C) :A), for strict determinateness were at leastin their final form also considerbefore. ably more straightforward than anything we have experienced This may give occasion doubts whether the somewhat involved considerto ations of 17.8. the corresponding in considerationsof 14.5. the case (and of strict determinateness) were necessary, particularly since they are based on the mathematical theorem of 17.6. our which necessitates analysis of linearity and convexity in 16. If all this couldbe replaced discussions by in the style of then 18.3. our modeof discussionof 16. 17.would be and entirely unjustified. 2 This is not so. As pointed out at the end of the greatsimplicity of the proceduresnd results of 18.2. a and is due to the fact that they the apply only to the simplesttype of zero-sum two-person games: Matching Penniesclassof games,characterized ft\\ = 2 = 2. For the general by casethe more abstract machinery of 16.nd 17. eemsso far indispensable. a s

18.3.

18.2.,

18.3., 18.3.

1The narrative of Conan Doyle excusably disregards mixed strategies and states the actual developments. According to these Sherlock Holmes gets out at the intermediate station and triumphantly watches Moriarty 's specialtrain going on to Dover. Conan Doyle'ssolution is the best possibleunder his limitations (to pure i which we found to be the strategies),nsofar as he attributes to eachopponent the course more probable one (i.e. e replaces60% probability by certainty). It is, however, h somewhat misleading that this procedureleadsto Sherlock Holmes's complete victory, t whereas, as we saw above,the odds (i.e.he value of a play) are definitely in favor of
instead

(Our result for , 17 yields that Sherlock Holmes is as good as 48%dead Moriarty. when his train pulls out from Victoria Station. Comparein this connection the suggestion in Morgenstern, loc. the p. 98,that the whole trip is unnecessary because loser

could be determined beforethe start.) 'Of courseit would not lack rigor,

{ cit.,

but it would
problem.)))

be an

mathematical machinery on an elementary

unnecessary

use of heavy

SOME ELEMENTARY GAMES))

179))

It may help to seethesethings in their right proportionsif we show by some exampleshow the assertions of 18.2., fail for greater values 18.3.
actually suffice to considergames with 0i = fa = 3. will be somewhat related Matching Pennies, more general to only by introduction of a third alternative. Thus both playerswill have the alternative choices 2, 3 (i.e. values the for TI, r 2). The reader ill best think of the choice in terms of choosing 1 w the 2 3 \"tails\"nd the choice as somethinglike a \"heads,\" choice of choosing off.\" off,\" \"calling Player 1again tries to match. If eitherplayer \"calls then it will not matter whether the otherplayer chooses"heads\" \"tails,\" or \ the only thing of importanceis whether he chooses ne of thesetwo at o all or whether he \"calls off\" too. Consequentlythe matrix has now the o appearance f Figure30:)) of 18.

In fact they

18.5.2.will It

1,

\\)
1)

1)

2)

3)

1)

-1)
1)

7) y)

2)

-1)
a)

3)

a)

ft))

Figure

30.
two

arethe familiar pattern of MatchingPennies(cf.Fig.12).The two fields o with a areoperative when player 1 \"callsff\" and player 2 does not. The w with 7 areoperative in the oppositecase. Theelement ith two elements where both players\" call off.\" By assigning refers to the case |8 appropriate
values (positive, negative or zero) we can put a premium or a penalty or on any one of theseoccurrences, make it indifferent. shall obtain all the exampleswe needat this junctureby specializing We this scheme, i.e. y choosingthe above a, 0,y appropriately. b Our purpose is to show that none of the results (18:A), 18:B), ( is generally true. (18C)of : This Ad (18:A): criterion strict determinateness clearly tied to the of is c special ase$\\ = fa = 2:For greatervalues of 0i, 2 the two diagonalsdo o and not even exhaust the matrix rectangle, therefore the occurrencen the a diagonalalonecannot be characteristics before. Ad (18 o W :B): e shall give an examplef a gamewhich is not strictly a but where neverthelessthere exists good strategy which is determined, for one player (but of course for the other). This examplehas the not pure w further peculiarity that one of the playershas severalgoodstrategies, hile the otherhas only one.)))

The four

first

elements i.e.he first t

elements f the first two rows o

18.5.3. 18.3.

180
We

ZERO-SUMTWO-PERSONGAMES: EXAMPLES
in choose the game of Figure30 ,
ft,

as follows:))
3)

\\)
1)

1)

2)

1)

-1)
1)

2)

-1)
a)

3)

a)

-a))

Figure

31.

w f a > 0,6 > 0. Thereader ill determineor himself which combinationsof \" calling off\" areat a premium or arepenalizedin the previously indicated sense. of This is a complete discussionof the game,using the criteria 17.8. w For = {i, 0}always K( { , ) = 0,i.e. ith this strategy player 1 v' cannot lose.Hence jg 0. For = 6 8 = {0,0,1}always K( , ) v w ^ O;1 i.e.ith this strategy player 2 cannot lose.Hence' g 0. Thus we

i,

rj

t\\

rj

have v'

=
TJ

Consequently is a goodstrategyif and only if always K( { , rj ) ^ and is a goodstrategy if and only if always K( , rj ) ^ O.2 Theformer is easily seento be true if and only if
*i

= {2 =
5))

and the latterif and only

172))

{3 = 0,

if

2(a))

Thus the set A of all goodstrategies contains preciselyone element, and this is not a pure strategy. The set B of all goodstrategies , on the and other hand, contains infinitely many strategies, one of them is pure:
77

namely

18.5.4. for a moment any 0i,/32 Thesignificanceof the majorizaAllow a tion of rowsor of columnsby eachotherwas consideredt the end of It was seento mean that one of the players had a simple directmotive
1

sentation of Figure21(cf. Figures 33): 32, Ad (18C): shall give an examplef a gamewhich is strictly detero : We e mined but in which no two rowsand equally no two columnsmajorizeach other. We shall actually do somewhatmore.

= d 8 = {0,0,1}. The setsA, B can be visualized by making use of the graphical reprerj

18.3.

* We leaveto the

It is actually equal to

readerthe simple verbal

interpretation

of thesestatements.)))

SOME ELEMENTARY GAMES))

181))

strict determinateness. Specifically: If the row T\" is majorized by the row r\\ i.e.if 3(T\"> TS) ^ 3CM, T2) for all r2 then player 1 need never consider the choice sinceT\\ is at leastas good for him in every contingency. And: Tit If the column r 2' majorizesthe column 2 i.e. JC(TI,r'') ^ 3C(n,r2) for all if 2 r n then player2 neednever considerthe choice2'',sincer' is at leastas good 2 for him in every contingency. (Cf. loc. it., articularly footnote 2 on c p p. 174. Theseare of courseonly heuristic considerations,cf. footnote 1,
with

for neglecting of his possiblechoices favor of another, and this one in narrowed the possibilitiesin a way which could be ultimately connected

p. 182.)))

33. If Now we may use an even more generalset-up: the row r\", i.e. to the player purestrategycorresponding r\" is majorizedby an average
Figure

1's

of all rows T\\

\"

then it is still plausibleto assume that player 1 neednever conT of the sider the choice r\", since other T\\ areat least s goodfor him in every a o contingency. The mathematical expressionf this situation is this:

T\"

i.e.by

a mixed strategy

with

the component

it T2)

y
n))

0i

3C(ri,r 2)

for all T
0.))

2))

to player 2 s pure strategy corresponding r2 majorizesan average of all > a columns r' ^ //, i.e. mixed strategy with the component r \" = 0. 2 o The mathematical expressionf this situation is this:
??
t\\

if The correspondingsituation for player 2 arises the column rj' ;

'

i.e.

(n, r2;)

0*

Y JC(n,rj)^
in 8^))

for all n))

= 0.))

are The conclusions the analoguesof the above.)))

182))

ZERO-SUMTWO-PERSONGAMES: EXAMPLES))

Thus a game in which (18:14:a)(18:14:b) or occurs,permits of an for immediate and plausible narrowing of the possiblechoices one of the players.1 We 18.5.5. are now going to show that the applicabilityof (18:14:a), is very W (18:14:b) limited: e shall specify a strictly determined gamein nor which neither(18:14:a)(18:14:b) is ever valid. 30. (0i = /32 = 3). Let us therefore return to the classof gamesof Figure
We

choose

< a < 1, = 0,y = -a:))

1)

2)

3)

1)

1)

-1)
1) i)

a) a)

2)

3)

a)

a)

Figure

34.

: This is a discussionof the game The element(3, 3) is clearly a saddle point, so the game is strictly and determined
v

areat a premium or arepenalizedin the previously indicatedsense. = v' = 0.

will The reader determine himself which combinationsof \" calling off\" for

It is not
17

difficult

to

seenow

(with

the aid of the method used in

18.5.3.),

nor is ever that neither(18:14:a)(18:14:b) valid here, that in Figure34 no row is majorized by any averageof the two other rows, and that no column majorizesany average of the othertwo columns.

setB of all goodstratethe gies , contains preciselyone element: pure strategy d 8 = {0,0, 1). w little On the otherhand, the reader ill experience trouble in verifying
that the set A of all good strategies as well as the

i.e.

18.6. and Chance

Imperfect Information

The 18.6.1. examplesiscussedin the precedingparagraphs make it d clearthat the roleof chance more precisely,of probability in a gameis

not necessarilythe obvious one,that which is directly provided for in the in rules of the game. The games described Figures27 and 30 have rules

1This is of course thoroughly heuristic argument. We do not needit, sincewe have a and discussionsof 14.5. of 17.8.But one might suspecthat it can be used the complete t to replace at least to simplify those discussions.The example which we are going to or to give in the text seems dispel any such hope. Thereis another course which might produce results:If or (18:14:b) holds, can then a combination of it with 17.8. be used to gain information about the setsof good A to strategies, and B. We donot propose take up this subjecthere.)))

(18:14:a)

SOME ELEMENTARYGAMES

183

which do not provide for chance; he moves are personal without except tion.l Nevertheless found that most of them arenot strictly determined, we i.e.hat their good strategiesare mixed strategiesinvolving the explicit t

On the other hand, our analysisof thosegamesin which perfectinformation prevails showedthat theseare always strictly determined, that which arepure strategies, they have goodstrategies involving no probabilities at all. (Cf. 15.) of the strateThus from the point of view of the players' behavior is giesto be used the important circumstance whether the game is strictly moves. determinedor not, and not at all whether it containsany chance The results of 15.on games in which perfect information prevails indicatethat thereexists close a b connectionetweenstrict determinateness and the rules which govern the players' stateof information. To establish o this point quite clearly, and in particular to show that the presencef chancemoves is quite irrelevant, we shall now show this:n every (zeroI sum two-person)game any chance move can be replacedy a combination b of personal moves, so that the strategical ossibilities the game remain of p exactlythe same. It will be necessaryto allow for rulesinvolving imperfect information of the players, but this is just what we want to demonstrate: That imperfect information comprises(among other things) all possible

use of probabilities.

i.e.

i.e.

tains only personal moves. 2 A direct way of removing chance moves existsof course after the introduction of the in a Indeed as the last (pure) strategiesand the umpire's choice, s described step in bringing a game into its normalized form we eliminated the remaining chance moves by the explicit introduction of expectationvalues in to But we now propose eliminate the chance moves without upsetting the structure We shall replace eachchancemove individually by personal of the game so radically. rolesin determining the moves (by two moves, as will be seen),so that their respective players'strategies will always remain differentiated and individually appraisable. This idea detailedtreatment is likely to give a clearer of the structural questions involved than m the summary procedure entioned above. 3 For our present purposes it is irrelevant whether the characteristicsof 9Tl depend of upon the previous course the play or not. 4 This is no real loss of generality. To seethis, assume that the probabilities in rational values, say r\\/t, , ra /t (r\\, , ra and / question have arbitrary integers). (Herein lies an actual restriction but an arbitrary small one sinceany degree.) probabilities can be approximated by rational onesto any desired Now modify the chancemove 9fR so that it has r\\ + -f r ajc t alternatives and so that designated by </K , t (instead of , (instead of each of the first n values of n has the same effect on the play as = eachof the Then giving all K next rj values of K the same as *K , t the 2, has the same effect as giving same probability , aK the original probabilities Ti/t, , Tat.)))

( a = , a and assume that their probabilitiesp(l), , p?** are all equal to l/ex<.4 Now replace by two personal moves 2fTl^, 9TC\". 3TI, 1The reduction of all games to the normalized form shows even more:It proves that every game is equivalent to one without chancemoves, sincethe normalized form conK

2 of chance moves. consequences explicit 18.6.2. us consider, ccordingly,a (zero-sumtwo-person)game F, Let a 3 and in it a chancemove 3TC*. Enumerate the alternatives as usual by

1,

11.1.

11.2.3.

a,),

' 'l/l,

a'

J 1,

<r

1,

e etc., tc.

<r

1,

1, *' 1,
<r

a);

184))
Wl(

ZERO-SUMTWO-PERSONGAMES: EXAMPLES))
o-^
K

a,alternatives;we denotethe correspondingchoicesby = 1,- , a = 1, and , a . It is immaterial in which orderthese moves are
<r\"
K

and 9(Tl\" arepersonalmoves of players 1and 2 respectively. Both have

that made, but we prescribe both moves must be made without any information concerninghe outcome any moves (includingthe other move t of 3fl, 9TC\-") We define a function S(<r', <r\") by this matrix scheme. (Cf. i Figure35. The matrix elements 5(a', \-")1)- The influence of 3ffl^, 9TC\" a o i.e. f the corresponding(personal) hoices , <r\" on the outcomeof the c a gameis the same as that of STl* would have been with the corresponding (chance)choicea* = 6(<r, <r\.") We denotethis new game by F*. We claim that the strategical ossibilities F* arethe sameas those of F.)) of p

a))

1))

a*))

-2)) 35.
K

Figure

18.6.3. Let player 1use in F* a given mixedstrategy of F with Indeed: the further specificationconcerning move 9Tl^ 2 to chooseall a' = 1, the
strategy of player 1 will be from player 2'spoint of view the same as F. This is so becauseany choiceof his at 9Tl^ (i.e.ny cr\" = , a producesthe same result as the original chancemove 9R,: One look at Figure35 will show that the a\" = cr\" column of that matrix containsevery that 5(er', <r\ number <r = 6(a',<r\") = precisely once, will assumeevery value (owing to player strategy) with the sameprobability just as 9H, would have done. So from player Ts point of view, F* is at leastas good as F. The sameargument with players 1 and 2 interchanged hence with the rows of the matrix in Figure35 playing the roleswhich the column had above showsthat from player 2'spoint of view too F* is at leastas

-,,

with

the same probability 1 /a*. Thenthe gameF*

!/<,

!,,, !,,
-f *'-*\" 1 + a* +
<r'

with

this

1,

a)

i.e. 1's

goodas
1

F.))

Arithmetically)) a\"

for for

a' ^
<r'

<r'
<r'))

<

Hence5(<r', a\") is always oneof the numbers 1, , ot. 131li is his personal move, so his strategy must provide needfor this in r, since9R, was a chancemove.)))

for it in

r*.

Therewas no

SOME ELEMENTARYGAMES
F* 1 and F areequivalent.

185

Sincethe viewpoints of the

two

players are opposite,this means that

of This Result

18.7.1. moves of F, of the operation Repeated applicationto all chance in described 18.6.2., will remove them all, thus proving the final 18.6.3., contention of 18.6.1. meaning of this result may beeven better underThe

18.7. Interpretation

stood if we illustrate it by somepracticalinstances f this manipulation. o Considerthe following quite elementary \"game chance.\"The of (A) two players decide, a 50%-50% chance who pays the otherone by device, unit. Theapplicationof the device of 18.6.2., transforms this game, 18.6.3. which consistsof precisely one chance into one of two personalmoves. move, = 2 with the 3(</,<r\") values 1, 2 A look at the matrix of Figure for 35 w the actual payments 1, 1 shows that it coincides ith replaced by 14.7.3. Figure12. Remembering14.7.2., we seethat this means what is that this is the game of MatchingPennies. plain enough directly : I.e.MatchingPennies the natural deviceto producethe probabilities is moves and imperfect information. (Recall 7.1.!) i> by personal 1 The b (B) Modify (A) so as to allow for a \"tie\": two playersdecide y a 33%, 3%,3^%chancedevicewho pays the otherone unit, or whether 3 3 18.6.3. nobody pays anything at all. Apply again the device of 18.6.2.,
<*<

Now the matrix of Figure35 with a = 3 with the &(&', a\") values 2,3 1 coincidesith Figure By w replacedy the actual payments0, b 14.7.3. seethat this is the game of Stone,Paper, cissors. we S 14.7.2., i Scissorss the natural device to produce the probaStone,Paper, information. (Recall bilities by personal moves and incomplete
K

1,

1, 13.

35 can be replacedy another func<r\") of Figure b and even the domainsa' = , aand<r\"= , a,by other tion, domains<r' = 1, a'and o-\" = 1, , a\",providedthat the follow, ing remains true:Every column of the matrix of Figure35 contains each 2 number 1, , ac the same number of times, and every row contains 8 t each number same number of times. Indeed,he conthe two properties S(<7<,a\") (and of of siderationsof 18.6.2. use of these made :, *'.')only. It is not difficult to seethat the precautionof \"cutting\" the deckbefore dealingcards falls into this category. When one of the 52 cards has to be chosen a chancemove, with probability this is usually achievedby by move, but if the player \"mixing\" the deck. This is meant to be a chance move of his. who mixesis dishonest,it may turn out to bea \"personal\"
K K

I.e., i,i,i 17.1.!)

18.7.2. The d(<r', (C)

!,-,
K

!,

!,

^,

1We leaveit
and sideration

Theaboveverbal arguments convey the we hope. in a clearer and simpler way * Hence <*i/ times; consequently a' must bea multiple of 1Hencea\"/ctimes; consequently a\" must be a multiple of
K

17.8.: 17.2., This presentsabsolutely

to the

readerto casttheseconsiderations
no

f into the preciseormalism of 1 difficulties, but it is somewhat lengthy. essential easonof the phenomenon under conr

1.

o.
a,.)))

186
As

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

a protectionagainst this, the otherplayer is permitted to point out the in f place the mixeddeck,rom which the card in questionis to betaken,by at \"cutting\" the deck that point. This combination of two^moves even move. if they arepersonal is equivalent to the originally intended chance
K

The lack of information is, of course,he necessaryconditionfor the effect tivenessof this device. = Herea, = 52, a' = 52! the number of possiblearrangementsof the \"

a deck, \" = 52 the number of ways of cutting.\" We leave it to the reader 1 tofill in the detailsand to choose 5(<r^, a\") for this set-up. the

19. okerand Bluffing P 19.1. Description of Poker

It 19.1.1.has been stressed repeatedly that the case = 2 = 2 as 18.3. more specifically in 18.4., and comprisesonly the very some zero-sum We then gave in 18.5. instances simplest two-persongames.
fti
ft

discussed in

of the complicationswhich can arisein the generalzero-sum two-person game, but the understanding of the implications of our generalresult will of (i.e. 17.8.) probablygain more by the detaileddiscussionof a special of the more complicatedtype. This is even more desirablebecause game o with 0i = $2 = 2 the choices f the r\\, r 2, called(pure) stratefor the games deservethis name: ust calling them \"moves\"would have j gies,scarcely

been less exaggerated.ndeed,in these extremely simple gamesthere I

and couldbe hardly any differencebetweenthe extensive the normalized form ; a characteristic normalized of the and so the identity of moves and strategies, in is inescapable thesegames. We shall now considera game in the form, form in which the player has several moves, so that the passage extensive tothe normalized form and to strategiesis no longer a vacuous operation. 2 19.1.2. gameof which we give an exactdiscussionis Poker. HowThe i subjectfor an exhaustive ever,actualPokers really a much too complicated discussionand so we shall have to subjectit to some simplifying modifica\" that the mixing is used to produce only one card. If whole hands\" is not an absolute safeguard. A dishonest mixer can produce correlations within the deckwhich one \"cut\" cannot destroy, and the knowledge of which gives this mixer an illegitimate advantage. 1The general considerations concerning Poker and the mathematical discussionsof the variants referred to in the paragraphs which follow, were carried out by von but Neumann in 1926-28, not published before. (Cf a general reference \" Zur Theorie in M in der Gesellschaftsspiele,\" ath. Ann., Vol. 100[1928]). This applies particular to the the variants (A), (B) of and to the entire symmetric variant of of \"Bluffing\" which dominates all thesediscussions. The unsymmetric interpretation was consideredin 1942 the purposes of this publication. variant (C) of for The work of E. Borel and Ville, referred to in footnote 1 on p. 154,alsocontains considerations on Poker (Vol. IV, \"Applications aux Jeux de Hasard,\" Chap. V: \"Le jeu de Poker\.") They are very instructive, but mainly evaluations of probabilities applied to Poker in a more or less heuristic way, without a systematic use of any underlying general theory of games. A definite strategical phaseof Poker (\"La Relance\" \"The Overbid\") is analyzed on pp. loc.cit. This may be regarded alsoas a simplified variant of Poker,)))

are dealt, \"cutting\"

1We assumed

J.

19.4.-19.10., 19.14.-19.16. /. 2:

19.11.-19.13.,

91-97

POKERAND BLUFFING

187

1 t tions, some of which are,indeed,quite radical. It seemso us, neverthet a less,hat the basicideaof Poker nd its decisivepropertieswill be conserved in our simplified form. Therefore it will be possibleto basegeneral onc clusionsand interpretations on the results which we are going to obtain by the applicationof the theory previously established. 2 To begin with, Pokeris actually played by any number of persons, but sincewe are now in the discussionof zero-sumtwo-persongames,we shall set the number of playersat two. The game of Pokerbeginsby dealing to each player 5 cards out of a 8 deck. The possiblecombinationsof 5 which he may get in this way there are2,598,960 4 arecalled\"hands\"and arrangedin a linear of them i t order, .e.here is an exhaustive rule defining which hand is the strongest 5 of all, which is the second,third, strongest down to the weakest. is Poker played in many variants which fall into two classes: and \"Stud\" \"Draw\"games. In a Stud game the player's hand is dealt to him in its entirety at the very beginning,and he has to keep it unchangedthroughout the entireplay. In \"Draw\"gamestherearevarious ways for a player to a exchangell or part of his hand, and in somevariants he may get his hand in several successivestagesin the courseof the play. Sincewe wish to discussthe simplestpossibleform, we shall examine the Stud game only. In this casethere is no point in discussingthe hands as hands, i.e. s a combinations of cards. Denoting the total number of hands by S a for S = 2,598,960a full deck, s we saw we might as well say that each related to the latter. closely Thereaderwho wishes to comparethese two
tions helpful:

comparableto the

two which we

considerin

19.4.10. 19.14-19.16. It is actually -19. and

variants,

may

find the following

indica-

bid, then our variant provides for a comparison of hands, while that in loc.cit. makes him losethe amount of the \"low\" bid unconditionally. we treated this initial \"low\" bid as \"seeing\" cf. the discussion at the beginning of while in loc.cit. it is treated as \"passing.\" particularly footnote 1 on p. We believe that our treatment is a better approximation to the phase in question in real Poker;and in particular that it is neededfor a proper analysis and interpretation of For technical details cf. footnote 1 on p. \"Bluffing.\" 1 Cf. however and the end of 2 The \"optimum\" to interpret in a sensewhich we do not undertake is supposed
with

If player 1 begins

(I) Our bids a, 6 correspondto 1 + a, 1 loc.cit. and that (II) The difference between our variant of 19.4.-19.10.
a \"low\"

in

loc.cit. is this:

I.e.

19.14.,

211

19.11.

19.16.

219.

specialfunctions, 4
This holds
binatorics 5 out of 52\":))

to be 4 or 5. 3 This is occasionallya full deckof 52 cards,but for smaller numbers of participants only parts of it usually 32 or 28 are used. Sometimes one or two extra cardswith

\"jokers,\" areadded. for a full deck. Thereaderwho is conversant with the elements of comwithout repetitions of will note that this is the number of \"combinations

= ^52.51-50.49.48 U/ * 1.2-3.4.52 598960

/52\\
> >
-))

&

Flush,\"
here.)))

This description involves the \"Four of a Kind,\" \"Full

well known technical terms \"Royal Flush,\" \"Straight House,\" etc. Thereis no needfor us to discussthem

188

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

, S instead. The ideais that s = <S player draws a number s = 1, corresponds to the strongestpossible hand, a = S 1 to the second e a strongest, tc.,nd finally s = 1 to the weakest. Sincea \" square deal\" amounts to assumingthat all possiblehands aredealt with the same probability, we must interpretthe drawing of the above number s as a chance move, each one of the possible values s = 1, , S having the same probability 1/8. Thus the game begins with two chancemoves:The 1 drawing of the number s for player 1 and for player 2, which we denote by 81 and $2. 19.1.3. next phase of the generalgame of Pokerconsists of the The making of \"Bids\"y the players. Theidea is that after one of the players b has made a bid, which involves a smaller or greateramount of money, his o opponent has the choice of \" Passing,\"\"Seeing,\"r \"Overbidding.\" Passing means that he is willing to pay, without further argument, the amount of hislast preceding id (which is necessarily lower than the present b In this caseit is irrelevant what hands the two players hold. The bid). hands are not disclosedat all. \"Seeing\" means that the bid is accepted: the hands will be comparedand the player with the stronger hand receives the amount of the present bid. \"Seeing\" t terminateshe play. \"Overbidmeans that the opponent countersthe presentbid by a higher one, ding\" in which the roles the players are reversedand the previous bidder has of the choiceof Passing, eeing Overbidding, tc. or e 2 S

The point in all this is that a player with a strong hand is likely to make high bids and numerous overbids sincehe has good reason to t expecthat he will win. Consequentlya player who has made a high bid, t or overbid, may be assumed by his opponent a posteriori!o have a hand. This may provide the opponentwith a motive for \" Passing.\" strong of the However,sincein the case \"Passing\" hands arenot compared,even a player with a weakhand may occasionally obtain a gain against a stronger the (false) impressionof strength by a high bid, or opponent by creating by overbid, thus conceivably inducinghis opponent to pass. This maneuver is known as \"Bluffing.\" It is unquestionably practicedby all experienced Whether the above is its realmotivation players. is may be doubted;actually a secondinterpretation conceivable.That is if a playeris known to bid high only when his hand is strong, his opponentis likelyto passin such cases. The player will, therefore, not be able to collect on high bids, or on numerous overbids,in just thosecases where his actual him the opportunity. Hence is desirable him to create for it strength gives 1In actual Poker the secondplayer draws from a deckfrom which the first player's hand has already been removed. We disregard this as we disregard some other minor complications of Poker. 2 This scheme is usually complicatedby the necessity of making unconditional pay\"ante,\" at the start, in some variants for the first bidder, in others for all those who wish to participate, again in others extra payments are required for the privilegeof drawing, etc. We disregard all this.)))

19.2.1.

19.2. Bluffing

ments, the

POKERAND BLUFFING

189

mind as to this correlation, to make it uncertainty in his opponent's known that he does occasionally high on a weak hand. bid To sum up:Of the two possiblemotives for Bluffing, the first is the to desire give a (false)impressionof strength in (real)weakness;thesecond is the desire give a (false) impressionof weaknessin (real)strength. to Both are instances inverted signaling(cf. 6.4.3.), of misleading of the opponent. It should be observedhowever that the first- type of Bluffing is most successfulwhen it \"succeeds,\" when the opponent actually s t \"passes,\"incethis secureshe desired gain; while the secondis most successfulwhen it \"fails,\" i.e. hen the opponent \"sees,\"incethis will s w convey to him the desired confusing information. 1 19.2.2. possibilityof such indirectly motivated henceapparently The irrational bids has also another consequence. uchbids are necessarily S risky, and therefore it can conceivably be worth while to make them riskier

i.e.

i.e.

i.e.

theiruseby the opponby appropriate countermeasures, thus restricting ent. But such countermeasuresare ipso facto also indirectly motivated moves. We have expounded these heuristic considerations at such length our because exacttheory makes a disentanglementof all these mixed
motives

phenomena which surround Bluffing can be understood quantitatively, and how the motives areconnected with the main strategic features of the of game,like possession the initiative, etc.
Poker (Continued) to Let us now return to the technical ules of Poker. In order r 2 the number of bids is usually limited. In order avoid endless overbidding irrational effects to avoid unrealisticallyhigh bids with hardly foreseeable the opponent there are also maxima for each bid and overbid. upon It is also customary to prohibit too small overbids; we shall subsequently indicatewhat appears to be a good reasonfor this (cf.the end of theserestrictionson the size of bids and overbidsin the We shall express form: e shall assumethat two numbersa, b W simplestpossible
of

possible.It will be seenin

and 19.16.2. 19.10. in 19.15.3., how the

19.3.1.

19.3. Description

19.13.).

a >b >
oncemore of disregarding our previously stated At this point we might be accused guiding principle; the above discussion obviously assumes a seriesof plays (so that statistical observation of the opponent'shabits is possible)and it has a definitely that character. And yet we have repeatedly professed our considerations \"dynamical\" must be applicable oneisolated to play and alsothat they are strictly statical. where this apparent contradiction has beencarefully We refer the readerto examined. Thoseconsiderations are fully valid in this casetoo, and should justify our procedure. We shall add now only that our inconsistency the use of many plays and is a merely verbal one. In this way we were ableto make of a dynamical terminology more succinctand more akin to the way that thesethings are talked about our discussions how all thesequestionable pictures it was elaborated But in in everyday language. can be replaced the strictly static problem of finding a goodstrategy. by 1This is the stop rule of 7.2.3.)))
1

17.3.,

17.3.

190

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

The limitation of the number of bidsand overbidswill now be used for a simplification of the entirescheme.In the actual play one of the players a beginswith the initial bid; fter that the players alternate. The advantage or disadvantage contained in the possessionof the of initiative by one player but concurrent with the necessity acting first! constitutes an interesting problemin itself. We shall discussan (unsymBut we metric)form of Pokerwhere this plays a rolein wish at first to avoid beingsaddled^yvith this problemtoo. In otherwords, we wish to avoid for the moment all deviations from symmetry, so as to obtain the other essentialfeatures of Pokerin their purest and simplest form. We shall therefore assume that the two players both make initial bids, eachone ignorant of the other'schoice.Only after both have made this bid is eachoneinformed of what the otherdid, whether his bid was

aregiven ab inilio,and that for every bid there areonly two possibilities: it the bid may be \"high,\" in which case is a;or \"low,\" in which caseit is 6. By varying the ratio a/b which is clearlythe only thing that matters o we can make the gamerisky when a/b is much greaterthan 1, r relatively safe when a/b is only a little greaterthan 1.

19.15. 19.14.,

of by giving to the players only the choice \"Overor i.e. \"Passing\" \"Seeing,\" by excluding\"Overbidding.\" Indeed, a o bidding\"is only a more elaboratend intensive expressionf the tendency we which is already containedin a high initial bid. Since wish to do things as simplyas possible,we shall avoid providing severalchannelsfor the same and 19.15.). tendency. (Cf. however (C) in 19.11. 19.14., Accordingly we prescribe these conditions:Consider the moment bids. If it then develops when both players are informed of eachother's that both bid \"high\" or that both bid \"low,\" then the hands are compared the and the playerwith the strongerhand receives amount a or b respectively from his opponent. If their hands areequal,no payment is made. If on the other hand onebids \"high\" and one bids \"low,\" then the player with or of the low bid has the choice \"Passing\" \"Seeing.\"\"Passing\" eans m that he pays to the opponent the amount of the low bid (without any of m consideration their hands). \"Seeing\" eans that he changesover from his \"low\" bid to the \"high\" one, and the situation is treatedas if they both had bid \"high\" in the first place.
further

\"high

19.3.2. simplify We

\"or\"low.\"

i.e.

19.4. Formulation Exact

We 19.4.

of the Rules

denotethe hands of players 1,2, by 81,82 respectively. this each player will, by a personal move, chooseeither a or b, the \"high\" or \"low\" bid. Each player makes his choice(bid) informed)))
We After

Poker giving an exactstatement of the rules agreedupon: by First, by a chancemove each player obtains his \"hand,\" a number s = 1, S,eachoneof thesenumbershaving the same probability 1/5.

can now sum

up the

preceding descriptionof our simplified

POKERAND BLUFFING

191

''Seeing\" \"Passing.\"
or

about his own hand, but not about his opponent's hand or choice(bid). but Lastly, eachplayer is informed about the other's choice not about his hand. (Eachstill knows his own hand and choice.) If it turns out that one bids \"high\" and the other \"low,\" then the latter has the choiceof This is the play. When it is concludedthe payments are made as follows:If both players bid \"high,\" or if one bids \"high,\" and the other bids
from
\"low\"

but

then for i = $2 player subsequently \"Sees,\" a a

<

1 obtains

player 2 the amount

respectively. If both players bid \"low,\"

b

then for si = s2 player 1obtainsfrom player2 the amount If one player bids then \"Passes,\" the
\"high,\" \"high

> <

and subsequently bidder\"being ^ player 1obtains from player 2

\"low\" of the
\"

and the other bids

-b respectively.

the amount

, o

.
19.6. Description
\"

or a \"low\" bid will be made, and in the lattercasethe further statement whether, if this \"low\" bid runs into a \"high\" bid of the opponent,the player this by a numerical index It o will \"See\" r \"Pass.\" is simplerto describe = 2, t = I meaning a \"high\" bid; = 2 meaning a \"low\" bid with subsequent \"Seeing\" the occasion rises); = 3 meaning a \"low\" t. a (if bid with subsequent\"Passing\" the occasion rises). Thus the strategy a (if of the of is a specification such an index for every s = , S,

A 19.6.1.(pure)strategy in this game consistsclearly of the following : , S whether a \"high\" specificationsTo statefor every hand s = 1,

Strategies

i. 1, 3;i

i,

t sequence'i,

This applies to both players 1and 2. Accordingly we shall denotethe above strategy by 2i(t'i, , js). is) or 22( as Thus eachplayer has the same number of strategies: many as there i.e. recisely3s With the notations of aresequences } i\\ p , is, = 182 = = 3s 0i 1For the sake of absolute formal correctness should still be arranged according this to the patterns of 6. and 7. in Chapter II. Thus the two first-mentioned chancemoves moves 1 and 2; the two subsequent personalnjoves (the dealing of hands) should becalled (the bids), moves 3 and 4; and the final personal move (\"Passing\" or \"Seeing\,") move 5. In the caseof move 5, both the player whose personal move it is, and the number in and of of alternatives, depend on the previous course the play asdescribed 7.1.2. 9.1.5.

, is.

i,

1,

i.e.

ji,

11.2.2.

it

(If both players bid \"high\" or both bid \"low,\" then the number of alternatives is and doesnot matter to which player we ascribethis vacuous personal move. If one bids \"high\" and the other bids \"low,\" then the personalmove is the \"low\" bidder's). A consistent use of the notations loc.cit. would alsonecessitate writing <n, <n for i, I;<TI,<rifor the \"high\" or \"low\" bid;** for \"Passing\" or \"Seeing.\" We leaveit to the readerto iron out all thesedifferences.)))

1,

192))

ZERO-SUMTWO-PERSONGAMES: EXAMPLES))

If we wanted to adhererigorously to the notations of loc. it., e should c w enumerate sequences the fi, , s with a r\\ = , ft and then denotethe (pure)strategies the players 2 by S^, S^. But we prefer of to continue with our present notations. We must now express payment which player 1receives the stratethe if two players. This gies 2i(t\"i, , is), 2j(ji, , js) are used by the 1 is the matrix elementC(i\\, 3 , t'alji, , js). If the players have actually the \"hands\" i, $2 then the payment receivedby player 1 can be expressed this way (using the rules stated in It is 90n s2,2 and where sgn($i s2) is the sign of s\\ above): it where the threefunctions +(i, j), JB (t, j), -(*,) f, = 2, 3. j can be representedy the following matrix schemes: b
now

1,
*

1,

(t-j(i j,)
\\

j 1,
3))

^)
1)

3)
1)

1)

2)
a) a)

3)

2)

3)

\\j)
1)

1)

2)

3)

a)

&)

1)

&)

a)

a)

b)

2) 3)

6) 6)

&)

2)

2)

a)

-6)

-b)
-&))

-6)
Figure

6))

3)

-6)

3)

-6) -b)
Figure

36. Figure 37. Now i, s2 originate from chancemoves, as 8

9

38.

describedbove. Hence a

a i,

, ja)

1
2

i pass to the (mixed) strategiesn the senseof 17.2. Theseare the vectors , rj belongingto Sp. Consideringthe notations 1Theentire sequence'i, t , is is the row index, and the'entire sequenceit j , ja is the column index. In our original notations the strategies were Si,Sj and the
matrix element 3C(n, r 2

19.5.2. now We
>

1
8

I.e. for si

-I-

).

8t respectively. It expresses in

an arithmetical

form which hand

is

stronger.

of the

Thereaderwill do well to comparethesematrix schemes ith w rules, and to verify their appropriateness.
Another circumstance which identities

our verbal statements

to corresponds the
4 Thereader

is worth

observing

is that

the symmetry

of the game

M*.3) m
may
verify))

--(/0.
at
f

o(i, j)

-JBo(j,

as a consequence the of
is skew-symmetric,

relations

expressing

end of footnote 3 above. 5) 3C(ti , ia\\ji, oncemore the,symmetry of the game.)))

the

,j

I.e.

POKER AND BLUFFING

193

which we are now using, we must indexthe componentsof thesevectors also in the new way: We must write fy instead of v i?v i, ig We express17:2) 17.4.1., evaluates the expectation of which value of ( player 1's gain))

, if)

=))

Thereis an advantage K( {,
If we now
77

in interchanging the two

S and writing

gj2}

i.t i

..... .....
2^ /!
Js
*ii

Snri.c,-.,) j'Ofc, (*'.,,

..... ^.....
v

put))

(19:1)
(19:2)
<rj,

i^d

'excluding

,))

=
/IP

J'sexcluding

/.
f))

then the above equation becomes

(19:3) It is worth

K(7,7)
while to
pji

=))

i,i tj

expoundthe meaning of (19:1)-(19:3) verbally. is the probability that player 1,using the mixed showsthat a-J is the i strategy , will choose when his \"hand\" is a\\\\ (19:2) j probabilitythat player2,usingthe mixedstrategy y , will choose when his 1 \"hand\"is S2. Now it is intuitively clearthat the expectationalue v > > K( , rj ) dependson theseprobabilitiesp{,<r$ only, and not on the underly2 can ing probabilities fct a, ^ themselves. The formula (19:3) shows (19:1) that
<

1 means a \"high\" bid, t 2, 3 a \"low\" bid with from 194. that i (the intention of) a subsequent \"Seeing\" or \"Passing\" respectively. 2 This means that two different mixtures of (pure) strategies may in actual effect be
the and

1We know

or;-

same thing. Let us illustrate

a \"low\"

this by

hand.

Consideri

let a simple example. Put S 2, i.e. there be only a \"high let 2, 3 asonething, i.e. there be only a \"high\" and

\"
a)))

194

ZERO-SUMTWO-PERSONGAMES: EXAMPLES
<

i easily be seen to be correctn this directway: It suffices to remember the meaning of the JB w ti t )(i,j) and the interpretation of the pji, <rj. 19.5.3.is clear,both from the meaning of the pji, o-Ji and from their It formal definition (19:1), that (19:2), they fulfill the .conditions

-.

(19:4) (19:5)
On the other hand, any
77

all P ^

; 0, t-i
3
fulfill

p\\>

=1 =1

allcr}.0,
pji,

irj.

aj which

theseconditionscan be obtained

from suitable , This is clearmathematically, 1 and by which intuitively as well. Any such system of pji, <rj is oneof probabilities define a possible modus procedendi, so they must correspond somemixed to

(19:2). (19:1),

strategy.

v (19:5) (19:4), make it opportuneto form the 3-dimensional ectors P' = (Pi , P?SPa }, *'*= {*i , <r'2 , <rj|.
1 1 1
f 2

(19:5) (19:4), statepreciselythat all p*i, a**belongto 53. This shows how much of a simplification the introduction of these vectors is: (or ) was a vector in 50, i.e. ependingon ft 1 = 3s 1 d
Then
77

numerical constants; he p '* (or the a ) are5 vectorsin 58, i.e. achone t e depends on 2 numerical constants;hence they amount togetherto 25 numerical constants. And 3s 1 is much greater than 25, ven for modere
2 ate 5.
\"low\"

i.e.

Bid \"high\" on a \"high\" hand, \"low\" on a \"low\" hand. Bid \"high\" on a \"low\" hand, \"low\" on a \"high\" hand. Then a 50-50mixture of \"Bold\" and \"Cautious\" is in effect the same thing as a 50-50mixture of \"Normal\" and \"Bluff\": both mean that the player will according to chance bid 50-50\"high\" or \"low\" on any hand. Nevertheless theseare,in our present notations, two different \"mixed\" strategies, vectors This means, of course,hat our notations, which were perfectly suited to the general t case,are redundant for many particular games. This is a frequent occurrence mathein matical discussions with general aims. Therewas no reasonto take accountof this redundance as long as we were working out the general theory. But we shall remove it now for the particular game under consideration. 1Put e.g. is the (17:1 P{ p* , i, , t ,a <r, s and verify :a), fy t v of as consequences the above(19:4), :b) (17:1 of (19:5). s 8 S is about 2J millions (cf.footnote 4 on p. 187); 1 and 25 are both so3 Actually great, but the former is quite exorbitantly the greater.)))
\"Normal\":
\"Bluff\":

bid. Then there are four possible (pure) strategies, to which we shall give names* \"Bold\": Bid \"high\" on every hand. \"Cautious\": Bid \"low\" on every hand.

17.2.1.

...

<7}

...

POKERAND BLUFFING

195

19.6. Statement

of the Problem

19.6. we aredealingwith Since

> >

a symmetric game,we can use the char\"\"\"*

of acterization the good (mixed)strategies i.e. f the o o It (17:H)f 17.11.2.stated this:

must
> rj

in

given in

be optimal against itself,

i.e.

Min- K(

ij

) must

be assumedfor
t\\

Now we saw in

that 19.5. K( {,

the states(we rearrange S somewhat)

we may write for it,

K(7s

,7V
,

) dependsactually on the p % a 5). Then 1, in

. So (19:3) 19.5.2.

(19:6)K(7',
And

' ,7V\\
p l,
*

O = 55
, p 8 of a
*

the

of characteristic the

goodstrategy is that

s l 7,K(7, ' ' ' ,7V1, - ,7) conditionsfor this is assumedat a l = p - - , a s = p s. The explicit 1. canbefound in essentiallythe sameway as in the similar problemof 17.9.;

Min^

we will give a brief alternative discussion. a The Min-* . ->8 of (19:6)mounts to a minimum l O
ff

. .,

with

t respecto

, a s separately. Consider therefore restrictednly by the requirementto belongto $3, o 3 all o}i X *'f =

each

>

cr

l,

>

i.e. y b

such a a

. It is

SO,

y-i

1-

in is (19:6) a linearexpression thesethree components^V, it

orj*,

cr|. Hence

to assumesits minimum with respect a '* therewhere all thosecomponents the smallest possiblecoefficient (with respecto j, t which do not have <rj cf. below),vanish. The coefficientof <r* is

^V
,,f

. r.,)(i,
C

n(.

J)P?

to

be denotedby g 7^.

becomes Thus (19:6)

(19:7)

K(7S

75|7s

, 7)))

196
And

ZERO-SUMTWO-PERSONGAMES: EXAMPLES
minimum (with

the conditionfor the

(19:8)

Foreachpair s2, j, for which (in j *), we have crjt = 0.

t respecto a ) is this: not 7} does assumeits minimum minimization at a

>

ofa the Hence characteristic goodstrategy

'=

,))

(19 :A)

a if and a a in p *, , p describe goodstrategy, only if this is true: Foreachpair s2, for which 7j does assumeits minimum not (in l), we have p;* =

i.e.

J!,

0.

o 36-38.They matrix schemes f Figures

7 (19:9:a)*1'

We

finally

statethe explicit for expressions the 7<,of course usingthe by

are))

=*))

-i
(19:9:b)))

(ap\\i

+ ap' 2

(19:9:c)))

Discreteto the Continuous Problem The criterion(19 of :A) togetherwith the formulae (19:7), 2 can (19:9:b), (19:9:c), be used to determineall good strategies. (19:9:a), This discussionis of a rather tiresomeombinatorialcharacter, c involving the analysis of a number of alternatives. The results which areobtained
the

19.7.1.

19.7. from Passage

19.6.,

1We mean in and not in s, j\\ 2 This determination has beencarried out by

oneofus and

will

bepublished

elsewhere.)))

POKERAND BLUFFING))

197))

say more about this in

are qualitatively similar to those'which shall derive belowunder somewe what modified assumptions,exceptor certaindifferences in very delicate f detail which may becalled \"fine structure\" the strategy. We shall the of
\" tion and not in the questionof \"fine structure. We beginby turning our \" \" attention to the granular structureof the sequence possiblehands of

Forthe moment we arechiefly interested the main featuresof thesoluin

19.12.

If we try to picture the strengthof all possible\"hands\"on a scale from or 0% to 100%, rather of fractions from to 1,then the weakestpossible to hand, 1,will correspond 0, and the strongestpossiblehand, 5, to 1. Hence \"hand\" s(= 1, the , S) shouldbe placed at z = ~ __ 1 on this

= 1,-

' - , S.

scale. I.e. e have this correspondence:)) w

Old scale:)
S
SB) 1)

2)

3)

S
Figure

1)

2)
1)

I)

S S -2) S
1) 1)

S)
I)

39.

Thus the values of z fill the interval

(19:10)

g1

continuousone. z t w I.e. e assumethat the chancemove which chooseshe hand s i.e. We may produceany z of the interval (19:10). assumethat the probability that of any part of (19:10) length of that part, i.e. z is equidistributed is the 2 We over (19:10). denotethe \"hands\"of the two players 1,2 by *i, za respectively. t 19.7.2. changeentails that we replacehe vectors This b (0 ^ *i, z* ^ 1); ut (si, ,, = 1, , S) by vectors p , nature as before, they are, of course,still probability vectors of the same the i.e.belonging to S*. In consequence, components (probabilities) p' , *',* (*i> s2 = 1, , S;i,j = 1,2, 3) give way to the components = 1,2,3). imilarlythe y',* (in (19:9:a), 2 ^ (19:9:b), S Pjs<rj(0 *,,z ^
p , a -i
<r 1 >

1 very densely, but they form neverthelessa discretesequence.This is the it s \"granular\" tructure referred to above. We will now replace by a

of 19.6.) (19:9:c) become5*. 7 for We now rewrite the expressions K and the 7}in the formulae (19:7), in 19.6. learlyall sums)) C (19:9:b), (19:9:a), (19:9:c)
1It will be remembered (cf.footnote 4 on p. 187)that.S is about 2J millions. 1This is the so-calledeometrical probability.))) g

l;i,j

198

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

1y
must

be replacedy integrals)) b
/:))
1-1))

sums

by

integrals

./O))

'2 t isolatederms behind a factor 1//Smay be neglected. Thesebeing understood,the formulae for K and the 7}(i.e. become: 7*) K = (19:7*)
while
1

/*

dzi,))

y{> (19:9:a*) = 7 (19:9:b*)J. = (19:9:c*) yl*

[ (-opf(-op[. \"

op}))

bp;>

fepj.)<fci

(opj.+ bp' t

>

And

thecharacterization :A) of 19.6. over into this: (19 goes

a The (0 ^ 2 g 1) (they all belong to 89) describe good strategy if and only if this is true: For each z, j for which 7* doesnot assume its minimum
(in ,;' a )> we have pj

(19:B)

= O.4

carrying out the limiting

8

1Theseterms correspond i = *2, in our present set-up to to j, and sincethe is 2sare continuous variables, the probability of their (fortuitous) coincidence indeed i, one may describethese operations by saying that we are now Mathematically
z\\

Specificallywe mean

the middle terms

bp\\*

and 6pJ in

and (19:9:a)

(19:9:c). 0.
> *

4 Theformulae

We mean

in

,; and not

S process
in z, j\\

>

oo

and (19 (19:9:a*), (19:7*), (19:9:6*),:9:c*) by discussing this \"continuous\"

this criterion
with

couldalsohave
the p f i, p f i

been derived

directly

arrangement,

in placeof the , 17 from the start. We preferredthe lengthier and more explicit procein order to make the rigor and the completeness our procedure followed in of dure apparent. The readerwill find it a good exercise carry out the shorter direct to discussion, mentioned above. It would be tempting to build up a theory of games, into which such continuous in sufficient generality for applications parametersenter, systematically and directly; like the present one,and without the necessityfor a limiting processrom discrete f games. An interesting step in this direction was taken by Ville in the work referredto in loc.cit. Thecontinuity assumptions made there seem, footnote 1on p. 154: p. p in particular for the present one.))) however, to be too restrictive for many applications,

19.4. -19.7.

i.e.

110-113

/.

POKERAND BLUFFING

199

19.8. Mathematical
Assume first that

Determination

of the Solution

19.8.1. now proceed the determination of the goodstrategiesp ', We to i.e.f the solution of the implicit condition(19:B)19.7. o of
Min, 75 = y\\ hencey\\ ^

p| >
y*2

i.e.

1 ever happens.

For such a

necessarily

7*2

y\\

0.
p^dz, g 0.

into Substituting (19:9:a*), (19:9:b*) this gives

(19:11) (a - 6) (

fydz,
Q

- f* jcfei)+ 26
P

Now let z be the upper limit of thesez with p\\ > O.2 Then (19:11) holds by continuity for z = Z too. As fy > does not occurfor Zi > 2 by is now hypothesis so the / p dz\\ term in (19:11) 0. So we may write it
with

+ instead of

becomes:)) (19:11) (a - 6) P pjufei + 26 JP pjufei o

*))

J *o , and

0.

p|iis always ^ and sometimes> 0, by hypothesis;hencethe first term is > O.8 4 The secondterm is clearly g: 0. So we have derived a contradiction.I.e. e have shown w 6 (19:12) pj = O. 19.8.2. eliminated = 2 we now analyze the relationship of Having = I and = 3. Sincep| = so p\\ + pj = 1
But

'

(19:13)
and consequently

PJ

=1
pj

i.e.:

pj,

Now there may existin the interval ^ z g 1 subintervalsin which always p* = or always p\\ = I.6 A z which is not inside any interval of
1 that the good strategy under consideration provides for \"low\" 2, bidding with (the intention of) subsequent \"Seeing,\" under certain conditions. 1 the greatest 2for which pj > occursarbitrarily near to (But we do not Q This 2 existscertainly if the z with pj > exist. require pj > for all z < Z 1Of course a b > 0. 4 It does to not seemnecessary go into the detailed fine points of the theory of integration, measure, etc. We assume that our functions are smooth enough so that a positive function has a positive integral etc. An exacttreatment could be given with easeif we made use of the pertinent mathematical theories mentioned above. 8 The reader-should reformulate this verbally: We excluded \"low\" bids with (the intention of) subsequent \"Seeing\" by analysing conditions for the (hypothetical) upper limit of the hands for which this would be done;and showed that near there, at least,an outright \"high\" bid would be preferable. c This is, of course, onditioned by our simplification which forbids \"overbidding.\" 6 where the strategy directs the player to bid always \"high,\" or where it directs him to bid always \"low\" (with subsequent \"Passing\.)))

(19:14) I.e. I.e.

g 1. j

i.e.

.)

z.

I.e.

200
will

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

eitherkind

i.e. rbitrarily near to which both pj' 5* and pf ^ 1 occur a be called intermediate. Sincep\\' ^ or p\\' 7* 1 (i.e. |'/^ 0) imply p ' B Min, 75' = 7*' or y\\ respectively,therefore we see: oth y[' g ys and for 7*' ^ 7a' occurarbitrarily near to an intermediate2. Hence such a z, 1 y\\ = 7s by continuity, i.e.
f

7' 8

- 7! = 0.
l Js

and (19:12), gives (19:13), (19:9:c*) recalling Substituting (19:9:a*),

(a

i.e. (19:15)

( + 6) JQ p^dz,
Z

(a + 6)

- (a - 6) f p^dz, + 26 f (1p^dz, - Vi^i + 26 1 - =

l
Jt

P\\^dz l

*)

0.))

Consider next two intermediate ', z\". e = z\" and subtract. Then

Apply

to (19:15)z = z' and

2(a + b)
obtains, i.e.

(19:16)))
Verbally:Betweentwo intermediatez',z\" the average of p{ is -r , So neither p* s= nor p\\ = 1can hold throughout the interval
this sincethat would yieldthe average or 1. Hence interval must contain b (at least)a further intermediatez, i.e. etweenany two intermediateplaces of therelies (at least)a third one. Iteration this resultshowsthat between z two intermediate z', z\" the further intermediate lie everywhere dense.

z' g z ^

z\"

But then

the Hence z', z\"

must (19:16)

for which (19:16) lie everywhere densebetweenz',z\" holds hold for all 5', z\" between 2',z\", by continuity. 2
z\"

This leaves no alternative but that pi = a r o everywhere betweenz' -r r

1The7* aredefined by integrals hencethey arecertainly (19 continuous. 1The integral in is certainly continuous. * Clearly isolated exceptionscovering a z area of total probability zero a finite number of fixed z's) could be permitted. They alter no integrals. An exact mathematical treatment would be easy but doesnot seemto becalled in this context for

(19:9:a*), (19:9:b*),:9:c*),

(19:16)

i.e.

(e.g.

(cf.footnote 4 on p. 199).So it seemssimplest to assume pf a -p o in 2' ^ z <: 2\" pr without any exceptions. This ought to be kept in mind when appraising the formulae of the next pages which deal with the interval 2' ^ z ^ 2\" on one hand, and with the intervals z < I1 and I\" < z' g 1 on the other; i.e. which count the points 2',I\" to the first-mentioned interval. This is, of course,irrelevant: two fixed isolated points z' and z\" in this case could be disposed in any way (cf.above). of The reader must observe,however, that while there is no significant difference)))

POKERAND BLUFFING

201

a Now if intermediatez existat all, then there exists smallest one and a largest one;choose 2\" as these. We have z',

19.8.3.

(19:17)

p\\

If no intermediatez exist,hen we must have p\\ = (for all z) or p\\ ss 1 t 1 that neither is a solution. Thusintermediatez (for all z). It is easy to see do exist and with them z',z\" exist nd (19:17) a is valid. The is 19.8.4. left hand side of (19:15)y\\ y\\ for all z\\ hencefor z =

^pj

throughout

ti

z\".

1))

(since i = is excluded). By continuity y\\ y\\ > 0,i.e. < 7!remains 7i pi true even when z is merely near enough to 1. Hencej = 0,i.e. = 1for p* p z\" necessitates < 1. Now no intermediatez exists thesez. Thus (19:17) in 2\" ^ z g 1;hencewe have p\\ = or p\\ = 1 throughout this interval. the r Our precedingesult excludes former. Hence 1 throughout z\" ^ z 1. (19:18) pi
Consider 19.8.5. finally

z'. the lower end of (19:17), If z' > then we z'. This interval contains no intermediatez; have an interval g = 1throughout Thefirst derivative z wehave p\\ = hence is clearly 2(a + 6)pi 26. Hence the left sideof (19:15), of 7S 7ii i-e-of 26 = 26 < if pi as in ^ z < z' this derivative is 2(a + 6) 1 26 = 2a > if p\\ = 1there, 7*3 y\\ fis monotone there,2(a+ 6) or respectively, throughout ^ z < z Sinceits decreasing increasing the upper end (the intermediatepoint z')>we have 73 7* > value is at or < respectively, y\\ < y\\ or 75 < y\\ respectively, throughout The former necessitates = 0, pi ss 1 the latter pi = in z < ^ p\\ werepi as orpi B 1 ^ z < z'\\ but the hypotheseswith which we started

^z or pi

'. - i.e.

'.

i.e.

respectively,there. Sothereis u contradictionin eachcase. Consequently

(19:19)

z' = 0.

now we determinez\" by Then z for the intermediate = z' = 1

And 19.8.6.

0.

becomes (19:15)
=))

the expressing validity of

(19:15)

-(a+ 6) f pi^1 + 26

between a < and a ^ when the z'a themselves are compared, this is not Thus we saw that y' > y\\ implies p\\ 0, while y\\ ^ 71 nas no such (Cf.alsothe discussion of Fig. 41and of Figs. 47, 48.) 1 bidding \"low\" (with subsequent \"Pass\") under all conditions strategy; nor is bidding \"high\" under all conditions.

I.e.

Mathematical
p\\

contradicting contradicting

pj

-1proof: For 1 ^ 0. ^
For
0.)))

p\\

pj m

0: Compute

71 1:Compute 7?7? --6,7i <*,

so for the y'r consequence. is not a good

& &

hence 7? < y\\ hence 7? <

?!

202 ZERO-SUMTWO-PERSONGAMES: EXAMPLES But (19:17), (19:18), (19:19)

give))

Jo
So we have

r^-r.-J^u-n-i + +
a
b))

+6
26))

+6

i.e.
(19:20)

+ 6 2/

=t _
2\"

26

+6

g-6

+ 6'

\"-g z g ^=-))

Combining(19:17), (19:18), (19:20) (19:19), gives:

(19:21)
w Together ith

,1

+
=1

for
b))

for))

the (19:13) (19:12), this characterizes strategy completely.

t strategy which will make it easiero discuss it verbally in what follows. (Cf. Figure 40. The actual proportions of this figure correspond to

19.9. DetailedAnalysis of the Solution one that ascertain thereexists and only one The 19.9.1. results of 19.8. 1 under consideration. It is described good strategy in the form of Poker W (19:13) (19:12), in 19.8. e shallgive a graphicalpictureof this by (19:21),

a/6 ~ 3.) above The line plots the curve p = p\\. Thus the height of = is the probabilityof a \"high\" bid:\\; the height of the line the line p p is the probabilityof a \"low\" bid (necessarily with subp = 1above p sequent\"pass\:")\\ = 1 p\\.

of The 19.9.2. formulae (19:9:a*), (19:9:c*)19.7. (19:9:b*), permit us now to computethe coefficients7*. We give the graphicalrepresentations instead of the formulae, leaving the elementary verification to the reader. 40, a (Cf. Figure41. Theactual proportionsarethoseof Figure i.e. /6 3 cf. there.) The line plots the curve 7 = 7*;the line plots the curve 7 = 7j; the line plots the curve 7 = y\\. Thefigure showsthat

1We have actually shown only that nothing elsebut the strategy determined in can be good. That this strategy is indeed good, could be concludedfrom the established of existence (at least) a good strategy, although our passageto the \"continuous\" case may there createsomedoubt. But we shall verify below that the strategy in question is of good, that it fulfills (19:B) 19.7.)))

19.8.

i.e.

POKERAND BLUFFING
and
in (i.e. and 75)coincide g z g a y\\
~))

203 and that

and))

(i.e. y\\

and 7$) coincide in

1. All

threecurves aremade))

-6
a

Figure 40.))

Figure 41.))

of two linear
7*

pieces ach,joining at e

=))

a
,

6))

The actual values of

the))
1))

at the

criticalpoints))

0,
verification

1 are given

in the figure.
reader.)))

1Thesimple computational

of theseresults is left to the

204

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

19.9.3. Comparisonof Figures40 and 41shows that our strategy is In t indeed good,i.e.hat it fulfills (19 of 19.7.Indeed: ^ z ^ :B) where both p{ 5^ 0, pj 5* both y{ and yl arethe lowestcurves,i.e.qual e

to Min, y'j. In

fl ~\"

lowest curve,

sincealways pj = 0.)

i.e.

only p{ 5* there only 7? is the equal to Min, y]. (The behavior of y\\ does not matter,

< z ^ 1 where

s is easily obtained;and this is the value to be expectedincethe gameis symmetric.

We can also computeK from

the in (19:7*)19.7., value of a play. K =

19.10. Interpretation

of the Solution

The results of although mathematically complete, w call for a certainamount of verbal comment and interpretation, hich we to now proceed give. Firstthe pictureof the good strategy, as given in Figure40, indicates that the player should then that for a sufficiently strong hand p{ = bid \"high,\" and nothing else. This is the casefor hands z > For

19.10.1.

19.8., 19.9.,

1;i.e.

weakerhands, however, p\\ = ^r~r> p\\ = 1 pf = . , ; soboth p{, pj 7* 0; t i.e.he player shouldthen bid irregularly \"high\" and \"low\" (with specified The \"high\" bids probabilities).This is the casefor hands z g

increases.

i this case)shouldbe rarerthan the \"low\" ones, ndeed = o and a > 6. PI This last formula shows toothat the last kind of \"high\" bids become increasinglyrare if the costof a \"high\" bid (relative to a \"low\" one)
(in
-=-

-|

Now these\"high\" bids on \"weak\"hands made irregularly, governed o probabilitiesonly, and gettingrarerwhen the cost f \"high\" by (specified) is increasedinvite an obvious interpretation:These are the bidding \"Bluffs\" of ordinary Poker. Due to the extremeimplifications which we applied to Pokerfor the s u \ purposeof this discussion,"Bluffing\" comesp in a very rudimentary form T only; but the symptoms are nevertheless unmistakable: he player is advised to bid always
\"high\"

on

a stronghand

>

and to bid

on a \"low\" one z < ?-ZjM mostly \"low\" (with the probability but with occasional, irregularly distributed \"Bluffs\" (with the probability)))

jpr-j)

POKERAND BLUFFING

205 z

19.10.2. the conditionsin the zoneof \"Bluffing,\" Second,

throw
from

17.10.2. as discussedin 17.10.1., Assumethat player 2 deviates from the goodstrategy, i.e. sesprobau bilities (Ty which may differ from the pj obtained above. Assume,furthert t more,hat player 1still usesthose p*, i.e.he goodstrategy. Thenwe can in 19.7., usefor the 7' of (19:9a*), (19:9:b*), (19:9:c*) the graphicalreprea the sentation of Figure41, nd express outcome the play for player 1 of in by (19:7*)19.7.
(19:22)
<r* Consequentlyplayer 2's areoptimal againstplayer in is of the condition(19:8) 19.6.fulfilled:

somelight on anothermatter too, the consequences deviating of the goodstrategy, \"permanentoptimality,\" \"defensive,\"\"offensive,\"

1's p*

if

the analogue

(19C) :

Foreachpair z, j for which (in j *) we have a* = 0.

<7*

not y] does assumeits minimum

causeno losseswhen the opponentsticks to the strategy pj if and only if the a* fulfill (19 above. :C) goodstrategy at t Now one glance Figure41 sufficesto make it clearhat (19 means :C) a a but merely a\\ = for z ^ I.e.:19:C) a = = for z > ( a ct
will
\"\"

I.e.(19C)is necessaryand sufficientfor beingjust as goodagainstpj as p* : itself, that is, giving a K = 0. Otherwisea] is worse, that is, giving a K > 0. In other words: A mistake, i.e. strategy a which deviates from the good (19D) :
<r*

f \"high\" bidding,and nothing else,or strong hands ( z > prescribes

it

|
200.)))

ff\\

~^^

J;

forbids \"low\"

bidding with

foils to

of \"high\" bidding and of \"low\" in the zone of for bidding (with subsequent \"Passing\") weak hands, the prescribe probability ratio

for subsequent \"Seeing\" all hands, but it

i.e.

\"Bluffing\ Thus any deviation from the good strategy which involves I leadsto immediatelosses. t suffices more than just incorrect \"Bluffing,\" for the opponent to stick to the good strategy. Incorrect"Bluffing\" \ causesno lossesagainst an opponent playing the good strategy; but the
1Actually even ?* would be permitted at the one placez a b But this and so it can be disregarded. Of. footnote 3 on isolatedvalue of * has probability 1We mean
in

19.10.3.

j, and

not in *,

j!

p.

206

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

by opponent could inflict losses deviating appropriately from the good strategy. I.e.the importance of \"Bluffing\" lies not in the actual play, playedagainst a goodplayer,but in the protectionwhich it providesagainst the opponent'spotential deviations from the good strategy. This is in with particularly with the agreement the remarksmade at the end of 19.2., \" the second interpretationwhich we proposedtherefor Bluffing.\" 1 Indeed, o element f uncertainty created \"Bluffing\" is just that type of constraint by on the opponent's trategy to which we referredthere,and which was anas lyzed at the end of 19.2. Our results on \"bluffing\" fit in also with the conclusionsof 17.10.2. is We seethat the unique goodstrategy of this variant of Poker not perh manently optimal; enceno permanently optimal strategy existsthere. (Cf. the first remarks in 17.10.2., particularly footnote 3 on p.163.)And in half is a defensive measure the sensediscussedin the second \"Bluffing\"
of

17.10.2.

loc. it.,.e.he deviationsfrom goodstrategy by which a player can profit c i t f from his opponent'sailure to \"Bluff\" correctly. use Let the We reverse roles: player 1 \" Bluff \" incorrectly,i.e. p* different is involved, we from thoseof Figure40. Sinceonly incorrect \"Bluffing\"
still

Third 19.10.4. and last,letus takea lookat the offensivesteps indicated

assume

= 1I Pf Pi =
/))

p* 2

==

for all
r all for n

z z

^ >))

a-b))

So we areinterested only

in the

of consequences
z

(19:23)

p\\

^-~

for some

=Z <
Q

^~'

is in Theleft hand sideof (19:15)19.8.still a valid expression y\\ for Considernow a z

it ,
_

7*!.

<z .
y\\

Then ^ in

n , ., . ., increases pi^azi ,henceit decreases ,left ,hand, side of (19:15), ,, ,+* * e i.e. / the ~,
Jz decreases
will
f
\\

leaves / (19:23) Jo

p^dziunaffected, but

increases

.,

7s

s 41),o it

7i- Sincey\\
now

be ^

0. I.e.\\ ^ y
ZQ

would

be without the change(19:23) Figure (cf. Considernext a z in y\\.

/ ^ < ^
Z

-6
a

All this holds for the form of Poker now under consideration. For further viewpoints cf. 2 We need this really for more than one z, cf. footnote 3 on p. 200. The simplest assumption is that theseinequalities hold in a small neighborhood of the ZD in question. It would be easyto treat this matter rigorously in the senseof footnote 4 on p. 199 and of footnote 3 on p. 200. We refrain from doing it for the reasonstated there.)))
1

19.16.

POKERAND BLUFFING Then ^

in
\" 1 eases f* (19:23)\" p^Zl

wn

iie jt

leaves f p{feiunaffected;

'

207

henceit increases left hand side of (19:15),7} y\\. Sincey\\ y\\ the i.e. , would be without the change(19:23) Fig.41),o it will now be ^ 0. s (cf. I.e. f ^ 75. Summing up: 7 The change(19:23) ^ causes with (19 :E)
7s % y\\
for for
z

7' % 7l 3
the Hence opponent can
which differ from

<z , z, < z <z

by

~^K of

gain,

i.e.decrease the

(19:22), using t?y

* at

&*

the present pj: Forz

increasingi

the expense

of

i.e.y increasing from the value of pf,' a + b to the extremealue , v -erf b . jj-i 1 * at the ** i.e. of . And for < z ^ expense by , ^ a by increasing <r| decreasing . to the extremealue n In otherwords: v from the value of pf,
ff

\\, <rj

ZQ

---

<r\\

a\\

(19:F)

for a certain If the opponent\" Bluffs\" too hand ZQ, then little he can be punished by the following deviations from the good)) for hands weakerthan z^ and \"Bluffstrategy:\" Bluffing\"
ing\"

...
ZQ.

imitating his mistake for hands which are stronger and by doing the oppositefor weakerones. \" d Thesearethe preciseetails of how correct Bluffing\" protectsgainst a too much or too little \"Bluffing\" of the opponent, and its immediate even beyond consequences.Reflections in this directioncould be carried but we do not proposeto pursue this subjectany further. this point,

I.e.by
ZQ

for , less

than hands stronger

than

MoreGeneralForms of Poker 19.11. throw a the discussionswhich we have now concluded of s deal of light on the strategical tructureand the possibilities Poker, good simplification of the rules of only they succeeded due to our far reaching the game. Thesesimplifications were formulated and imposed in 19.1., and 19.3. 19.7.For a real understanding of the game we should now make an effort to remove them. of By this we do not mean that all the fanciful complications the game 1 must necessarily be reinstated, have eliminated (cf. 19.1.) which we
1Nor do we
wish,

While 19.11.

yet to consideranything

but

a two-persongame!)))

208
but

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

(Cf. 19.3.) The problemof meeting thesedesiderata (B),(C)simultaneously (A), and finding the good strategies is unsolved. Therefore we must besatisfied for the moment to add (A), (B),(C) separately. solutions for (A) and for (B) are known, while for (C) The complete only a very limited amount of progress has been made. It would lead too far to give all these mathematical deductionsin detail, but we shall (A), report briefly the results concerning (B),(C).

somesimpleand important features of the game were equally lost and would be of greatadvantage. We mean in particular: their reconsideration and The \" hands\" shouldbe discrete, not continuous. (Cf. 19.7.) (A) Thereshouldbe more than two possibleways to bid. (Cf. 19.3.) (B) (C) Thereshouldbe more than oneopportunity for eachplayer to bid, and alternating bids,insteadof simultaneous ones,shouldalsobe considered.

19.12. Hands Discrete

In 19.4-19.7. this casethe solution is in many ways similar to that of 40. Generallypj = and thereexists a certain5 such that pf = 1 Figure j 0,1 for s < s. Also, if we changeto the z scale(cf. for s > s,while so } a ^ Fig.39), then -~ r is very nearly d l Sowe have a zoneof \" Bluffing \" o 1 and above it a zone of \"high\" bids, just as in Fig.40. in But the for s < s,i.e. the zone of \" Bluffing,\" arenot at all equal
p\\
__

hands s =

scale 19.12.1. first (A). I.e.let us return to the discrete of Consider and 1, , S as introduced at the end of 19.1.2., used in

p\\

to or near to the

, , of

Fig.40.2

They

oscillateround this value by a

of amounts which depend on certainarithmetical peculiarities S but do not tend to disappear for S > oo . The averagesof the p\\ however, tend to : r- 3 In otherwords: a The good strategy of the discrete game is very much like the good of the continuous game:his is true for all details as far as the t strategy division into two zones (of \"Bluffing\" and of \"high\" bids) is concerned; also for the positionsand sizesof thesezones,and for the events in the zone of \"high\" bids. But in the zone of \"Bluffing\" it applies only to average several hands of approximatelyequal strength). statements (concerning f The precise roceduresor individual hands may differ widely from those p Precisely: -^-l % fl o 1 b 2 I.e. PI -* not 7 for S
*

+b

- ^^ -

for
<

S -*

.
of

>

whatever the variability for most

* Actually 1 (pf

+ P?*1 )

<

.)))

POKERAND BLUFFING

209

given in Figure40, and depend on arithmetical peculiaritiesof 8 and S 1 to (with respect a/6). Thus the strategywhich corresponds ore precisely Figure m to 40

is not good, and it differs quite considerablyfrom the good one. Neverthelessit can be shown that the maximal loss which can be incurred by playing this \"average\" strategy is not great. More precisely,it tends to for S >.* So we see:n the discreteame the correct ay of \" Bluffing \" has a I g w \"fine structure,\" which however secures nly an extremely very complicated o small advantage to the player who usesit. This phenomenon is possiblytypical, and recursin much more complicated games. It showshow extremelycareful one must be in asserting real or expecting continuity in this theory.3 But the practicalmportance i the gains and lossescaused seemso be small, and the whole thing t is probablyterra incognita to even the most experienced players.
for all a

19.12.2. i.e. here p[ m w

<s

-i.e.

m 19.13.possibleBids

permit biddingin more than two ways.

a

19.13.1. second,B):I.e.etus keepthe hands continuous,but l Consider, (

w I.e. e replacehe two bids t

>6(>0)
>a
m

by a greaternumber, say m, ordered:

ai > a 2 >

_i

> a (>0).
m
Q

w highestbid, and nothing else, hile for z < 2 he shouldmake irregularly various bids (always including the highest bid a, but also others),with specifiedprobabilities. Which bids he must make and with what proba1Thus in
line the equivalent

4 In this case the solution bears a certainsimilarity to that of Figure40. too 6 such that for z > Z the Thereexists a certainz player shouldmake the

of Figure 40,the

left part

of the

figure will not

be a straight
around this

(P average.

\"

TT

in

S2

j, but
the remarks

one which oscillates violently

that in the

(Cf.footnote 4 on p. 187.) 1Recallin this connection

1It is actually of the order 1/S. Remember

real Poker Sis about

millions. 2} on

It has actually beendetermined only under the further restriction of the rules which to eachplayer is expected make his final, highest bid forbids \"Seeing\" a higher bid. if at once,and to \"Pass\" (and acceptthe consequences)the opponent's bid should turn out higher than his.

198. 4

made in

second of footnote 4 part

p.

I.e.

1Analogue of the z

-6

in Figure

40.)))

210

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

bilities,is determinedby the value of z.1 So we have a zone of \"Bluffing\" and above it a zone of \"high\" bids actually of the highestbid and nothing else just as in Figure40. But the \"Bluffing\" in its own zone z g Z has a much more complicatedand varying structure than in Figure40. We shall not go into a detailed analysis of this structure,although it offers some quite interestingaspects. We shall,however, mention one of its peculiarities.
Q

19.13.2.two values Let

>b >

be given, and use them as highest and lowestbids: a m == b. di = dj - , a m-i so that they Now let m > and choose remaining bids a*, the
fill

the interval

(19:24)
with

gx

^a

unlimited increasing density. (Cf. the two examplesto be given in footnote 2 below.) If the goodstrategy described above now tends to a oo limit to an asymptoticstrategy for m then one couldinterpret this as a goodstrategy for the game in which only upper and lower bounds aresetfor the bids (a and 6), and the bids can be anything in between(i.e. in (19:24)). the requirement of a minimum interval between bids I.e. is removed. mentionedat the beginning of Now this is not the case. E.g. e can interpolate a 2, w the , a m _i 2 betweenai = a and am = b both in arithmetic and in geometricequence. s > a In both casesn asymptoticstrategy obtainsfor m but the two strated giesdiffer in many essential etails. If we considerthe game in which all bids (19:24) permitted, as one are in its own right, then a direct etermination of its goodstrategiess possible. i d

i.e.

19.3.

then it

If the bids which he must make are ai, a , a , can be shown that their probabilities must be
p
fl

, ,Ic + a,, can V at a h ca caq if respectively. I.e. a certain bid is to be made at all, then its probability must be inversely proportional to the cost. am actually occurfor a given z is determined by a more comWhich Op, ac, plicatedcriterion, which we shall not discusshere. that the c abovewas neededonly to make the sum of all probabilities Observe The readermay verify for himself that the probabilities in Figure 40 have equal to the abovevalues. 1The first one is defined by

ca\\

111 If ----,_,... 1,1,

,
j

, an (l

<p <q <

1

< n),

1.

a,
the

one second is defined

T i ((m

- p)a + (p - 1)6)
by))

for

- 1,2,
,m

,m

- 1,

ap

~a ~*b*~l
m

for

1,2,

1,

m.)))

POKERAND BLUFFING

211

It turns out that

many

This chowsto what complicationshe abandonmentof a minimum intert val betweenbids can lead: goodstrategy of the limiting case a cannot be an w o approximationfor the goodstrategies f all nearby cases ith a finite numberof bids. The concludingremarksof arethus re-emphasized.

others.

both strategies mentionedabove are good,togetherwith

19.12.
Bidding

and player 2 bids afterwards. Thus the rulesstated in 19.4. modified as follows: are Firsteachplayer obtains,by a chancemove, his hand 8 = 1, , S, eachone of thesenumbers having the same probability l/S. We denote the hands of players 1, ,by Si, 2 respectively. 2 eithera or 6, the After this1 player 1will, by a personalmove, choose 2 or the \"low\" bid. He does this informed about his own hand \"high\" but not about the opponent's hand. If his bid is \"low,\" then the play is concluded. If his bid is \"high,\" then player 2 will, by a personal move, choose eithera or 6, the \"high \" or the \"low \" bid.3 Hedoesthis informed but about his own hand, and about the opponent's hoice, not his hand. c t This is the play. When it is concluded,he payments are made as follows:If player 1 bids \"low,\" then for si = s2 player 1 obtainsfrom player
b

19.14.1. and last, consider(C): The only progressso far made in Third bids of the two players this directionis that we can replace simultaneous the two successive in i.e. y an arrangement which player 1bids first by ones; b

19.14. Alternate

> <

2 the amount

-6

\" respectively. If both playersbid high,\" then for Si = s2

player 1 obtains from player 2 the amount

a a

> <

respectively. If player 1

the amount b.4 19.14.2. discussionof the pure and mixed strategiescan The

bids \"high\" and player 2 bids \"low,\" then player 1obtains from player 2
now

lines of this discussionin a way which will be pero w f clearor the reader ho remembersthe proceduref 19.4.-19.7. fectly
We give the main
A

carried essentiallyas we did for our original variant out,

i of Pokern

19.5.

be

pure strategy in this game consistsclearly of the following specifications: statefor every hand s = 1, , S whether a \"high\" or a \"low\" to t bid will be made. It is simpler to describehis by a numerical index = 1, ;t. = 1meaning a \"high\" bid, i. = 2 meaning a \"low\" bid. Thus i. 2
1We continue from here on as if player 2 had already made the \"low\" bid, and this We disregard \"Passing\" at this stage. were player turn to \"See\"or to \"Overbid.\" 1 \"Overbid\" or \"See,\" cf. footnote 1 above. 8 the \"See\"or \"Pass.\" Observe shift of meaning sincefootnote 2 above. 4 In interpreting theserules, recallthe abovefootnotes. From the formalist ic point m of view, footnote 1 on p. should be recalled, utatis mutandis.)))

I.e. I.e.

1's

191

212))

ZERO-SUMTWO-PERSONGAMES: EXAMPLES))

This applies to both players 1 and 2;accordinglywe shall denotethe above strategy by 2i(i , ja). Thus eachplayer i, , ia) or 2*(ji has the samenumber of strategies,as many as there are sequences 25 i, , <*; i.e. recisely . With the notationsof 11.2.2. p
y

of the strategy is a specification such an index t, for every s = o i.e.f a sequence'i, t , is.

1,

0i

- A
*

ft

= 2*.

(But the gameis not symmetrical ) ! if the We must now express payment which player 1receives the strate, is), *0*i, , ja) areused by the two players. This is gies2i(ti, the matrix elementOC(t'i, , ia\\ji , ja). If the players have actually hands *i, Sjthen the payment receivedby player 1can beexpressed in this way (using the rules statedabove):It is JB^nc^-*,)^, ) where j sgn(si *) is the sign of 81 s and where the threefunctions))
9 f

can be represented the by

1)

following matrix schemes:))

2) 6)
b))

\\j)
1)

2)

\\j
t \\)
1)

1)

2)

t
1)

\\) a)
b) 1)

6)

a)

&)

2)

2) Figure

2)

-b)
Figure

-6))

Figure

42.

43.

44,

Now Si,Sj originate from chance above. moves, as described

S))

Hence:

, ja)))

1))

19.14.3.now passto the mixedstrategiesn thesenseof 17.2.These We i arevectors , belongingto Sa. We must indexthe componentsof these
> 17

Of fe,, 1?r,. We express17:2) of (

gain))

vectors like the (pure)strategies: must write we

..... ^.....
<f>

<f

instead

17.4.1. evaluatesthe expectationf player 1's which o

'tis,
j'i,)))

.....))

POKERAND BLUFFING Thereis an advantage in interchanging the two Z and writing

>

213

K(

, ?)

i-

2,
pj

VA

If we now put

(^25)
(19:26)

P',= t. <=

,ta excluding

t))

......
V
j

J.'

i ja excluding l

then the above equation becomes

(19:27)

K(7,7) = gi

All As 19.14.4.this is preciselyas in 19.5.2. there, (19:25) that shows p*

when

is the probability that player 1, singthe mixedstrategy will choose i u shows his hand is s\\. (19:26) that a',* is the probabilitythat player 2,

using the mixed strategy

that 17 probabilitiesonly, and not on the underlying probabilities{^ s, 17^,. a this themselves. (19:27) expresses and could have easily been derived directly, on this basis. b It is alsoclear, oth from the meaning of the <rj and from their formal that definitions (19:25), (19:26), they fulfill the conditions:
____

clearintuitively

choose when his hand is s2. It is again j the expectation K( , ) depends upon these value
77

will

>

..... ,

p',

(19:28) (19:29)

all p}.

V p}i = 1

-i
2

alUji

;-i

<rj

1,

and that any p's <r} which fulfill these conditions can be obtained from suitable T> V by (19:25), (Cf. step (19:26). the corresponding in 19.5.?. footnote 1on p.194.)It is therefore opportune to form the particularly v 2-dimensional ectors
P*

! - {pfsPi'K

* *'

(*i'> ^J1 )-

state Then(19:28), (19:29) preciselythat all 7\\ ^ belongto S.)))

'

214
Thus

ZERO-SUMTWO-PERSONGAMES: EXAMPLES

was a vector in 8$ i.e. ependingon ft 1 = 2s 1 d * t e constants;he p (or <r areS vectorsin St i.e.achone dependson one to numerical constant, hencethey amount together S numerical constants. 2 Sowe have reduced s 1to S. (Cf.the end of 19.5.3.) 19.14.5.now rewrite (19:27) in 19.6. We as

(or

17

')
,

(19:30) K(7s ' '

with

7V\\

7s) = ^ V 7;-*;',

the coefficients

S' i.e. sing the matrix u

7l (19:31:a) = g
o schemes f Figures 42-44,

,-!

,-i

M)

Sincethe game is no longer symmetric, we need also the corresponding formulae in which the rolesof the two playersare interchanged.This is:

(19:32) K( p ',
with

, p

VS''',

O=g

the coefficients
*

H* o 02 S

.81
8

i.e. sing the matrix u

J (19:33:a) . =
g))

o schemesf Figures 42-44,

i-l

(19:33:b)

-g
will

^^ + MO

+))

are for The criteria good strategies now essentiallyrepetitions of those in

our present criterion

19.6.I.e.due to the asymmetry

of the variant now under consideration be obtained from the generalcriterion (17:D))))

POKER AND BLUFFING

in of 17.9. the same way as that of metrical criterion at the end of

215
the sym-

17.11.2.: I.e.

19.6. couldbe obtained from

(19:G)

to /Si

, 7* they all belong goodstrategieefifand only if this is true: For each t, j, for which 7* does not assumeits minimum l (in j ) we have crj = 0. Foreach i, i for which 6J does not assumeits maximum (in i *) we have pj = 0.

The

7l, , 7* and the 7l, describe

in

Now 19.14.6. we replacehe discreteands *i, *i by continuous ones, t h the senseof 19.7.(Cf. in particular Figure 39 there.) As describedin this t 19.7. replaceshe vectors p \\ (sit t = 1, , S) by vectors ^ w p *, *(0^ *i,2s 1), hich arestill probabilityvectorsofthe samenature to make placefor the as before,i.e. elonging Sj. Sothe components b pji, Similarly the 5*-, 7j become*, 7j. Thesumsin our 5 componentspji, and formulae (19:30), :a), (19:31:b),(19:32), (19:31 (19:33:a), go (19:33:b)
<r <r
<r*-

<rjt.

over into integrals,just as in So we obtain:))

in 19.7. (19:9 (19:9:b*), (19:7*), :a*), (19:9:c*)

(19:30*)
*) (19:31
y\\>

/''(-ap|. bp'.Odz,+ f ' (apj.+ 6p{>)(b ^0

/ *t t|
/o))

:b*) (19:31 T! = f\" 0>til ~ bP'

and
(19:32*)))

' (19:33:a*),.
<; (19:33:b*) =
f\" (hrjt y

bff' t >)dz t

'

f*t

(-bo*? fo;)<bi.

i Our criterion for good strategiess now equally transformed. (This is the to c discreteriterion of 19.6. the continuous sametransition as that from the We criterion of 19.7.) obtain

(19H) :

The p f i and the (0 g * zi g 1)they all belongto if and only if this is true: describeoodstrategies g For each *s, j for which 751 does not assumeits minimum 2 = 0. For each * lf i for which $Ji does not (in j ) we have 2
<r
Ip

'

Si-

<r*

assumeits maximum (in

* We mean in s We mean in

j (t) and j (t) and

not in not in

i ) we have pj = 0. **,,; (t i, i) !
zs,

(ci>

j)

!)))

216))

ZERO-SUMTWO-PERSONGAMES: EXAMPLES))

19.15. Mathematical

Descriptionof All

Solutions

out completely. Themathematfcal methodswhich achieve this aresimilar in the o to thosewith which we determined 19.8. goodstrategies f our original the solutionsof the implicit conditionstatedat the variant of Poker, end of 19.7. We shall not give the mathematical discussion here,but we shall describe

o 19.15.1. determinationof the goodstrategiesp and a *, i.e.f the The can solutionsof the implicit conditionstated at the end of 19.14.,becarried

'

i.e.

the good strategies * and a * which it produces. p Thereexistsoneand only one goodstrategy p * while the goodstrategies
form an extensive <r family. of thesefigures correspondto

a/6 ~ 3.)))

45-46.Theactual proportions (Cf Figures

Figure 45.))
\"

(o

Figure 46.))

b)b

curvesp = p\\ and a = a\\ respectively. Thus the)) above the line p = (a = 0) is the probabilityof a \"high\" height of is the probability bid, p\\ (or}); the height of the line p = 1(<r = 1)above

Thelines

a(a + 36) a + 2a6 6* a (a + 36)

- plot the

of a \"low\" bid, pj = 1 pj fa = 1 <rj). The irregular part of the v = <r* curve (in Figure46) in the interval u ^ z ^ v represents

the multiplicity of the good strategiesa z : Indeed,his part of the o- = a\\ t curve is subjectto the following (necessaryand sufficient) conditions:))
v

b
*<>

I a\\dz J..))

a
b
a))

when
when))

u))

<2 <

V.))

Verbally: Betweenu and v the average of <rj is 6/a, and on any right end of this interval the average of a\\ is j 6/a.)))

POKERAND BLUFFING))
1 threeintervals:

217))

Thus both p * and a * exhibit threedifferent types of behavior on these

First g z < u. Second <> z v. Third:v < z ^ Thelengths of thesethreeintervals areu, v 1 v, and the somewhatcomplicated u, for with expressions M, v can be best remembered the help of theseeasily verified ratios: t; = a +b

:0

:u

1.

tt:l
v

The formulae (19:31:a*), and (19:31:b*) (19:33:a*), (19:33:b*) of 19.14.6. us now to computethe coefficients?J, 3J. We give (asin permit 19.9. Figure41) the graphicalrepresentations, in instead of the formulae, leaving the elementaryverification to the reader. Foridentification of the

19.15.2.

u:l

t;

= a:b.

b:a

this 'differenceis > then p = or <r| = respectively, nd that a

the criterion the end of 19.14. beformulated as stating that whenever at can
whenever))

p *, <r * as goodstrategies the differences, t\\ only

6J, y\\

y\\

matter:ndeed, I

Figure tg

= or <rj = respectively. We give therefore this difference is < the graphs of thesedifferences. (Cf. Figures47, 48. The actual proportions arethoseof Figures 46; i.e. /6 3, cf. there.) 45, a The line plots the curve y = y\\ yj; the line plots the curve 5 = &{ b\\. The irregular art of the d = $5 d' curve p 2 t (in Figure 48) in the interval u ^ z ^ v correspondso the similarly irregularpart of the <r = o-J curve (in Figure46) in the same interval, i.e.t alsorepresentshe multiplicity of the goodstrategies Therestrici t a tion to which that part of the o- = a\\ curve is subjected(cf.the discussion after Figure46) means that this part of the 6 = b\\ 5J curve must lie within the shaded triangle /////////(cf. Figure48). 19.15.3. Comparisonof Figure 45 with Figure47, and of Figure 46 with Figure48 shows that our strategies t are indeed good,i.e.hat they fulfill W t (19:H). e leave it to the readero verify this, in analogy with the comparisonof Figure40 and Figure41in 19.9. 1Concerning the endpoints of theseintervals, etc., footnote 3 on p. cf.

47. a = 2a, then pj

tg

ft

- 26,

tg y

- 2(a -

Figure

48.

b)

'.

200.)))

218

ZERO-SUMTWO-PERSONGAMES: EXAMPLES
in 19.14.6. or (19:32*) (19:30*)

Thevalue of K can alsobe obtainedfrom The result is:))

Thus player 1 has a positive expectationalue for the play, i.e.n advana v the tage2 which is plausiblyimputableto his possessing initiative.

length, but just to make a few remarks on this subject. We seethat instead of the two zones of Figure40 three zones appear in Figures 46. The highest one (farthest to the right) corresponds to 45, i for \"high\" bids, and nothing else, n all these figures (i.e. both players). The behavior of the otherzones,however, is not so uniform. Forplayer 2 (Figure46) the middlezone describes kind of \" Bluffthat ing\" which we had on the lowestzone in Figure40, irregular \"high\" and \"low\" bids on the same hand, But the probabilities,while not entirely arbitrary, arenot uniquely determinedas in Figure40.3 And there exists a lowest zone (in Figure46) where player 2 must always bid \"low,\" where his hand is too weak for that mixedconduct. in middlezone the 7* show the sameindifference Furthermore, player 2's as in Figure41 y\\ y{ = there, both in Figure and in Figure 41 47 so the motives for his conductin this zone areas indirectas thosediscussed in the last part of these\"high\" bids aremore of a defense Indeed, against \"Bluffing,\" than \"Bluffing\" proper. Sincethis bid of player 2 concludes play, thereis indeedno motive for the latter,while thereis a the need to put a rein on the opponent's\"Bluffing\" by occasional"high\" \

19.16.1. results of 19.15. now be discussedin the same way The should as thoseof 19.8., were in 19.10. do not wish to do this at full 19.9. We

19.16. Interpretation

of the

Solutions.

Conclusions

i.e.

19.10.

him. bids, by \"Seeing\" Forplayer 1(Figure the situationis different. Hemust bid \" high,\" 45) and nothing else,n the lowest zone;and bid \"low,\" and nothing else,n i i the middle zone. These\"high\" bids on the very weakesthands while the bid on the medium hands is \"low\" are aggressive\"Bluffing\" in its 1 For numerical orientation: If a/6 = 3, which is the ratio on which all our figures are based,then u J, v J and K =

1For a/6 ~ 3 this is about 6/9 (cf. footnote 1 above), about per cent, of the \"bid. 1Cf.the discussion after Figure 46. Indeed,t is even possibleo meet those i t requirements with o\\ and 1 only; e.g.j in the lower fraction and a\\ * 1 in the a 6 fraction of the middle interval. upper The existence f such a solution (i.e.ever o\\ ^ 0, o n by Figure 45never p[ y* 0, 1 t either) means, of course, hat this variant is strictly determined. But a discussion on that basis(i.e. with pure strategies) will not disclose solutions like the one actually drawn
\"low

1,

i.e.

11

in Figure

46.)))

POKERAND BLUFFING
t>\\

219

purest form. The d* are not at all indifferent in this zone of \"Bluffing\" t (i.e.he lowest zone) : a 6J > there in Figure48. i.e.ny failure to \"Bluff\" under theseconditionsleads to instant losses. 19.16.2. Summing up: Our new variant of Pokerdistinguishes two varieties of \"Bluffing\": the purely aggressiveone practicedy the player b who has the initiative; and a defensive one \"Seeing\" irregularly, even with a medium hand, the opponent who is suspected of \"Bluffing\" the player who bids last. Our original variant where the practiced by initiative was split between the two players because they bid simultanecontaineda procedure hich we can now recognizes a mixture of a w ously
1 thesetwo things.
All

this gives valuable heuristic hints how real Poker with longer of sequences (alternating)bids and overbids ought to be approached. Themathematical problemis difficult, but probablynot beyondthereachof that thetechniques areavailable. Itwill be considered otherpublications. in
way

1The variant of E. Borel, referredto in footnote 2 on p. 186,is treated loc.cit. in a which bearsa certain resemblanceto our procedure. Using our terminology, the as courseof E.Borel can be described follows: The Max-Min (Max for player Min for player 2) is determined both for pure and for mixed strategies. The two are identical, this variant is strictly determined. The good strategieswhich are obtained in this way are rather similar to those of our of Figure 46. Accordingly the characteristics \"Bluffing\" do not appearas clearly as in our Figures 40and 45. Of.the analogous considerations in the text above.)))

1,

i.e.

CHAPTER V GAMES ZERO-SUM THREE-PERSON

of the zero-sum two-person game having been We we takethe next step in the senseof 12.4. shall establish completed, This will bring entirely the theory of the zero-sumthree-person game. new viewpoints into play. The types of games discussedthus far have had also their own characteristic problems.We saw that the zero-sum of one-personame wascharacterized the emergence a maximum problem g by and the zero-sumtwo-persongameby the clearcut oppositionof interest which could no longerbe described a maximum problem.And just as as to the transition from the one-person the zero-sum two-person game from of removed the pure maximum character the problem,so the passage the zero-sum three-personameobliterates g two-person game to the zero-sum the pure oppositionof interest. it Indeed, is apparent that the relationshipsbetweentwo players in a zero-sumthree-person game can be manifold. In a zero-sum twoperson gameanything one player wins is necessarilylost by the other and vice versa, so there is always an absolute antagonism of interests. In a zero-sumthree-personame a particular move of a player which, for the g sakeof simplicity,we assumeto be clearlyadvantageousto him may be disadvantageousto both other players, but it may also be advantageous 1 to to oneand (afortiori) disadvantageous theotheropponent. Thus some a parallelismof interests and it may be experience players may occasionally even whether theory will have to decide imagined that a more elaborate this parallelismis total,or partial, etc. On the other hand, oppositionof interestmust also existin the game (it is zero-sum) and so the theory the complicatedsituations which may ensue. will have to disentangle It may happen, in particular, that a player has a choiceamong various That policies: he can adjust his conductso as to get into parallelismof interest with another player, or the opposite; he can choose ith that w which of theothertwo playershe wishesto establishsuch a parallelism,and (possibly)to what extent.

20.1.1. theory The

20.Preliminary Survey 20.1. Viewpoints General

20.1.2.

1All this, of course, s subjectto all the complications and difficulties which we have i already recognizedand overcomein the zero-sum two-person game:whether a particular move is advantageous or disadvantageous to a certain player may not dependon that move alone, but alsoon what other players do. However, we are trying first to isolate the new difficulties and to analyze them in their purest form. Afterward we shall discussthe interrelation with the old difficulties.
220)))

PRELIMINARY SURVEY

221

20.1.3. soon as there is a possibility of choosingwith whom to As establishparallel interests,his becomescase choosingan ally. When t a of alliances reformed, it is to beexpected some of a mutual undera kind that standing between the two players involved will be necessary. One can also state it this way: A parallelism of interests makes a cooperation and b desirable, therefore will probably lead to an agreement etween the o players involved. An oppositionof interests, n the other hand, requires this presumablyno more than that a player who has elected alternative act independentlyin his own interest. Of all this there can be no vestige in the zero-sumtwo-persongame. Betweentwo players, where neither can win except precisely)the other's ( 1 o loss, agreementsr understandingsarepointless. This should beclear by common sense. If a formal corroboration (proof) be needed, ne can o find it in our ability to complete theory of the zero-sumtwo-person the or game without ever mentioning agreements understandings between players.
zero-sumthree-person game (as against the zero-sumtwo-persongame). Whether it is the only one is a question which can bedecidednly later. o If we succeed completingthe theory of the zero-sumthree-person ame in g establishedthis uniqueness. This will be the case essentiallywhen we reach F 23.1. orthe moment we simply observethat this is a new major element in the situation, and we proposeto discussit fully before taking up anything
without bringing in any further new

20.2.1. have thus recognizeda qualitatively We

20.2. Coalitions

different feature of the

t concepts,hen we can claim to have

else.
with,

o Thus we wish to concentraten the alternativesfor acting in cooperation w or in oppositionto, others, mong which a player can choose.I.e. e a want to analyze the possibilityof coalitions the question between which players, and against which player, coalitionswill form.2
1 This one with variable sum) d is, of course, ifferent in a general two-person game to there the two players may conceivably cooperate produce a greater gain. Thus there is a certain similarity between the general two-person game and the zero-sum three-

(i.e.

person game.
We shall
behind

this:

the

seein Chap.XI, particularly

in

game. 1The following seemsworth noting: coalitions occurfirst in a zero-sum game when three. In a two-person game there are the number of participants in the game reaches not enough players to go around: a coalition absorbsat least two players, and then nobody is left to oppose. But while the three-person game of itself implies coalitions, the scarcity of players is still such as to circumscribe thesecoalitions in a definite way: a coalition must consist of preciselytwo players and be directed against preciselyone (the
remaining) involved, another considerably more players, then the situation becomes eachother, etc. severalcoalitions may form, and thesemay merge or oppose et seq.; et seq., Someinstances of this appearat the end of 36.1.2., the end of 37.1.2.,

general n-person game is closelyrelated to

that 56.2.2., there the

zero-sum n

is a general connection + I-person

If there are four or more allied phenomenon at

player.

the

end of 38.3.2.)))

222

ZERO-SUMTHREE-PERSON GAMES

o Consequentlyit is desirable to form an examplef a zero-sumthreei persongame in which this aspects foremost and all others aresuppressed; a a i.e., game in which the coalitionsarethe only thing that matters, nd the 1 aim of all players. only conceivable 20.2.2. this point we may mention also the following circumstance: At A player can at best choose betweentwo possiblecoalitions,sincethereare two otherplayerseitherof whom he may try to induce cooperate him with to the third. We shall have to elucidate the study of the zero-sum against by and three-personame just how this choice g operates, whether any particular has such a choice all. If, however, a player has only one possiat player bility of forming a coalition (in whatever way we shall in fine interpret this in operation)then it is not quite clear what sensethereis a coalition at all: of moves forced upon a player in a unique way by the necessities the rules of the game aremore in the nature of a (one sided)strategy than of a (coopera ative) coalition. Of coursethese considerations re rather vague and uncertain at the presentstageof our analysis. We bring them up nevertheb thesedistinctionswill turn out to be decisive. less, ecause It may also seemuncertain,at this stage at least,how the possible o choices f coalitionswhich confront one player arerelated those open to to of indeed,whether the existence severalalternatives for one player another; impliesthe same for another.

21. he SimpleMajority Game of ThreePersons T 21.1. Descriptionof the Game

sum

coalitions between the players are the only considerationswhich matter. This is the game in question: the Each player, by a personal move, chooses number of one of the 2 two other players. Each one makes his choiceuninformed about the of choices the two otherplayers. After this the payments will be made as follows:If two players have choseneach other'snumbers we say that they form a couple* Clearly

21.1. now formulate the example entionedabove:a simple zeroWe m in which the possibilitiesof understandings i.e. game three-person

1This is methodically the same deviceas our consideration of Matching Pennies in the theory of the zero-sum two-person game. We had recognizedin that the decisivenew feature of the zero-sum two-person game was the difficulty of deciding which player \"finds out\" his opponent. Matching Pennies was the game in which this \" finding out\" dominated the picture completely, where this mattered and nothing else. 2 Player 1 chooses or 3, player 2 chooses or 3, player 3 chooses or 2. 2 1 1 * It will be seenthat the formation of a couple is in the interest of the players who createit. Accordingly our discussion of understandings and coalitions in the paragraphs which follow will show that the players combine into a coalition in order to be able to o form a couple. Thedifference between the concepts f a coupleand a coalition neverthew less should not be overlooked A coupleis a formal concept hich figures in the set of rules of the game which we define now; a coalition is a notion belonging to the theory concerning this game (and, as will be seen,many other games).)))

14.7.1.

therewill

couple,then the two playerswho belongto it get one-half unit each,while the third (excluded) losesone unit. If there is no player correspondingly couple,then no one gets anything. 8 The reader will have no difficulty in recognizing the actual socialprocesses which this game is a highly schematizedmodel. We shall call it for the simplemajority game (of three players).

THE SIMPLEMAJORITY GAME 223 1 be preciselyone couple,or none at all. 2 If thereis preciselyone
-

21.2. Analysis

21.2.1.

game is played. To begin with, it is clear that there is absolutelynothing for a player to do in this game but to look for a partner, i.e.or another player who f is prepared to form a couplewith him. The game is so simpleand absot lutely devoidof any other strategicpossibilitieshat therejust is no occasion for any other reasoned procedure. Sinceeach player makes his personal move in ignorance those of the others, no collaborationof the players of can be establishedduring the courseof the play. Two playerswho wish to collaborate get togetheron this subjectbefore the play, i.e. utside must o the game. The player who (in making his personalmove) lives up to his agreementby choosingthe partner'snumber) must possessthe conviction ( that the partner too will do likewise. As long as we are concerned only with the rules of the game, as stated above, we arein no positionto judge what the basis for such a conviction may be. In other words what, if anything, enforces the \"sanctity\" of such agreements?There may be games which themselves by virtue of the rules of the game as defined in and and provide the mechanism for agreements for their enforce4 ment. But we cannot base our considerations this possibility,sincea on t game neednot provide this mechanism; he simplemajority game described above certainly does not. Thus there seems be no escape to from the of considering concludedoutside the game. If we do necessity agreements not allow for them, then it is hard to seewhat, if anything, will govern the conduct of a player in a simplemajority game. Or, to put this in a somewhat different form 1 I.e. two different couples. Indeed,two couples there cannot be simultaneously must have one player on common (sincethere are only three players), and the number chosen by this player must be that of the other player in both couples, i.e. two the a couples re identical. 2 It may happen that no couplesexist: if e.g., 1 chooses , 2chooses , and 3 chooses 2 3 3 For the sakeof absolute formal correctness this should still be arranged according to the patterns of 6. and 7. in Chap.II. We leave this to the reader,as in the analogous in situation discussed footnote 1 on p. 191.

Game. Necessityof \"Understandings\" Let us try to understand the situation which exists when the
of the

6.1. 10.1.

1.

4 By providing personal moves of one player, about which only one other player is future informed and which contain (possibly conditional) statements of the first player's policy; and by prescribing for him to adhere subsequently to these statements, or by providing (in the functions which determine the outcome of a game) penalties for the non-adherence.)))

224

ZERO-SUMTHREE-PERSON GAMES

21.2*2. agreements a certain amount of similarity with have These like conventions \" in somegames Bridge with the fundamental difference, \" \" one however, that those affected only one organization (i.e. player split into two \"persons\") while we are now confronted with the relationshipof two players. At this point the readermay rereadwith advantage our and a discussion \"conventions\"nd related of topics in the last part of 6.4.2. especiallyfootnote 2 on p.53. If our theory wereappliedas a statistical analysisof a long series of plays of the same game and not as the analysisof one isolatedplay an alternative interpretation would suggest itself. We should then view as and agreements all forms of cooperation establishing themselves by in o repetition such a long series f plays. It would not be impossibleto derive a mechanismof enforcement from the player's desireo maintain his record to be able to rely on the record t and of his partner. However,we prefer to view our theory as applying to an individual play. But these considerations, nevertheless, possessa certain significance in a virtual sense. The situation is similar to the one which we encountered the analysisof the (mixed) trategies a zero-sumtwoin of s mutatis person game. The readershould apply the discussionsof 17.3. mutandis to the presentsituation.
\"

We aretrying to establisha theory of the rational conductof the participants in a given game. In our considerationof the simplemajority game we have reached point beyond which it is difficult to go in formulating the \"undersuch as \" agreements/' such a theory without auxiliary concepts 1 we standings/ etc. On a lateroccasion proposeto investigatewhat theoreticalstructures are required in orderto eliminate these concepts.For this purposethe entiretheory of this book will berequiredas a foundation, and the investigation will proceed along the linesindicatedin Chapter XII, weakand and particularly in 66. At any rate,at presentour positionis too W our theory not sufficiently advancedto permit this \"self-denial.\" e shall which follow, make use of the possibilityof the therefore, in the discussions t establishmentof coalitionsoutsidethe game;his will includethe hypothesis that they arerespected the contracting by parties.

6.4.3., 21.2.3.

21.3. Analysis

of the Game Coalitions. TheRoleof

Symmetry

offer to anybody opportunitiesfor rational actionof any other kind. The rulesareso elementarythat this point ought to be fully convincing. t Again the gameis wholly symmetricwith respecto the three players. That is true as far as the rules of the gameareconcernedthey do not offer toany player any possibilitywhich is not equallyopen to any otherplayer. What the players do within thesepossibilities of course, nothermatter. a is,

Onceit is conceded agreements ay exist etweenthe players that b m in the simplemajority game, he path is clear. This game offers to players t an who collaborate infallible opportunity to win and the gamedoesnot

21.3.

Theirconductmay be unsymmetric;indeed, inceunderstandings,i.e.oali-))) c s

FURTHEREXAMPLES

225

tions, are sure to arise,it will of necessity be unsymmetric. Among the three players thereis room for only one coalition (of two players) and one player will necessarily left out. It is quite instructive to observehow be the rules of the game areabsolutelyfair (in this case, ymmetric),but the s 1 conduct of the playerswill necessarily be.2 not Thus the only significant strategicfeature of this game is the possibility 8 of coalitionsbetween two players. And sincethe rules of the gameare 4 all three possiblecoalitions must be considered perfectly symmetrical, on the same footing. If a coalition is formed, then the rules of the game provide that the two alliesget one unit from the third (excluded) player
*

eachone getting one-half unit. Which of thesethree possiblecoalitions will

form, is beyond the scope of the theory, at least at the presentstageof its development. (Cf the end of 4.3.2.) e can say only that it would be irrational if no coalitions W were formed at all, but as to which particular coalition will be formed must depend on conditionswhich we have not yet attempted to analyze.

The remarks of the precedingparagraphsexhaust, at leastfor the time being, the subject of the simplemajority game. We must now which begin to remove, one by one, the extremely specializing assumptions characterized game:its very special nature was essential for us in this order to observethe role of coalitions in a pure and isolatedform in vitro There,if
in that no such thing occurs the zero-sum two-person games. in the rules of the game are symmetric, both players get the same amount the value of the game is zero),and both have the same good strategies. there is no reason to expecta difference in their conduct or in the results which they ultimately obtain. It is on emergenceof coalitions when more than two players are present and of the \"squeeze\" which they produce among the players, that the peculiar situation a describedbovearises. (In our present caseof three players the \"squeeze\" is due to the fact that eachcoalition can consistof only two players, lessthan the total number of players but more than one-half of it. It would be erroneous, however, to assume that no such \"squeeze\" obtains for a greater number of players.) * This is, of course, very essentialfeature of the most familiar forms of social rgania o zations. It is also an argument which occursagain and again in the criticism directed against these institutions, most of all against the hypothetical order basedupon \"Jowser It is the argument that even an absolute, formal fairness symmetry of the rules faire.\" of the game doesnot guarantee that the use of theserules by the participants will be fair and symmetrical. Indeed,this \"doesnot guarantee\" is an understatement: it is to be expected any exhaustive theory of rational behavior will show that the particithat pants are driven to form coalitions in unsymmetric arrangements. a To the extent to which an exacttheory of these coalitions is developed, real underc standing of this classical riticism is achieved. It seemsworth emphasizing that this \"social\" phenomenon occurs only in the case of three or more characteristically participants. 5 Such a coalition is in this e simply an agreement to choose achother's game, of course, numbers, so as to form a couple in the senseof the rules. This situation was foreseen already at the beginning of 4 Between 2,3.))) players

22.1.1.

22.1. Unsymmetric

22.Further Examples
Distribution.

Necessityof Compensations

1 We saw

17.11.2.

I.e.

(i.e.

i.e.

4.3.2. 1,3; 1,2;

226

ZERO-SUMTHREE-PERSON GAMES

but now this step is completed. e must begin to adjust our ideasto W

21.1.

The specializationwhich we proposeto remove first is this:n I the simplemajority game any coalition can get one unit from the opponent; the rules of the game provide that this unit must be divided evenly a among the partners. Let us now consider game in which eachcoalition offers the same total return, but where the rules of the gameprovide for a different distribution. Forthe sake simplicityletthis be the case of only in the coalition of players1and 2, where player say,is favored by an amount e. The rules of the modified game aretherefore as follows: in The moves are the same as in the simple majority game described The definition of a coupleis the same too. If the couple1,2 forms, then player 1gets the amount + l, player 2 getsthe amount e, and 3 loses one unit. If any other coupleforms (i.e. ,3 2,3) then or player 1 the two players which belongto it get one-half unit eachwhile the third one (excluded) player loses unit.

more general ituations. s

22.1.2.

1,

What will

To beginwith, it isstill characterized the possibility threecoalitions of by to the three possible couples which may arisein it. corresponding Prima it may seem player 1has an advantage,sinceat leastin his that facie with player 2 he getsmore by than in the original, simplemajority couple

happen in this game?

oneunit. and lose So any attempt of player 1to keep his privilegedpositionin the couple is 1,2 bound to lead to disaster for him. The best he can do is to take steps which make the couple 1,2 as attractive for 2 as the competing just couple2,3. That is to say, he actswisely if, in caseof the formation of a couplewith 2,he returns the extra to his partner. It shouldbe noted an that he cannot keep if any fraction of e;i.e., he shouldtry to keep extra amount e'for himself,2 then the above argumentscouldberepeated literally
with

have the following consequence: couple would never form, because The would the couple is more desirablefrom 1,2 point of view; the couple1,2 the is never form, because couple2,3 more desirablefrom 2's point of view; is but the couple 2,3 entirely unobstructed, sinceit can be brought about who then needpay no attention to 1 and his special by a coalition of 2,3 and desires. Thus the couple2,3 no other will form ; and player 1will not p get + e nor even one-half unit, but he will certainly be the excluded layer

However, this advantage is quite illusory. If player 1 would really insist on gettingthe extrae in the couplewith player 2, then this would

game.

1's

1,3

'

in

of 3 place e.

1 It seems natural to assume < e < 2 We mean of course 8 So the motives for ultimate player 2,3 would be weaker, but the disasterthe on p. 228.))) this connection footnote

<'<. 1's

J.
same and

disaster

the just

certain formation of couple as certain as before. Cf. in

FURTHEREXAMPLES

227

One could try some other variations of the original, simple, s majority game, till always maintaining that the total value of eachcoalition w is one unit. E.g. e couldconsiderrules where player 1 getsthe amount + in eachcouple1,2, while players 2 and 3 split even in the couple with 2,3. In this caseneither 2 nor 3 would careto cooperate 1if 1should try to keep his extra or any fraction thereof. Henceany such attempt of player 1would again lead with certainty to a coalition of 2,3 against him and to a loss of one unit. Another possibilitywould be that two playersarefavored in all couples in with the third: e.g. the couples1,3 and 2,3, layers 1and 2 respectively p both get one-half get + while 3 getsonly ; and in the couple 1,2 both players1and 2 would lose unit each. In this case interestin a coalition with eachother,and player 3 will become desirablepartner for eachof the them. One must expect this will lead to a competitive biddingfor his that This cooperation. must ultimately lead to a refund to player3 of the extra back into the field of advantage e. Only this will bring the couple 1,2

22.1.3.

1,3;

competitionand thereby restorequilibrium. e 22.1.4. leave to the reader the considerationof further variants, We where all three players fare differently in all three couples. Furthermore we shall not push the above analysisfurther, although this couldbe done and would even be desirablein order to answer some plausibleobjections. We aresatisfiedwith having establishedsomekind of a general lausibility p for our present approach which can be summarized as follows:It seems that what a player can get in a definite coalition dependsnot only on what the rules of the game provide for that eventuality, but also on the other o possibilities f coalitions for himself and for his partner. Since (competing) the rules of the game are absoluteand inviolable, this means that under certainconditionscompensationsmust be paid among coalition partners; t i.e.hat a player must have to pay a well-defined price to a prospective will coalition partner. The amount of the compensations depend on what otheralternatives areopen to eachof the players. Our examples above have servedas a first illustration for theseprinciples. This being understood, we shall now take up the subjectde novo and in more generality,and handle it in a more precise anner. l m
Discussion In accordance the above we now take a far reaching step with towards generality. We considera game in which this is the case: If players 1,2 then cooperate, they can get the amount c, and no more, from player if players 1,3 cooperate, can get the amount 6, and no they from player2;if players2,3 more, cooperate, can get the amount a, and they no more, from player

22.2.1.

22.2. Coalitions

of Different Strength.

3;

1.

1This is why
graph

we need not analyze any further the heuristic arguments of this the discussion of the next paragraphs takes careof everything. All these possibilities were anticipated at the beginning of and in 4.3.3.)))

para-

4.3.2.

228

GAMES ZERO-SUMTHREE-PERSON

follows:

further particulars We make no assumptions whatsoever concerning about the rules of this game. So we neednot describey what steps of b what order complication the above amounts aresecured.Nor do we of statehow theseamounts aredividedbetweenthe partners, whether and how eitherpartner can influence or modify this distribution,etc. We shall neverthelessbe able to discussthis gamecompletely. But that a coalition is probably connected it will be necessaryto remember with compensationspassing between the partners. The argument is as

a partner if his x fulfills

count upon obtaining more than the amount c x in a coalition with player Similarly player 3 cannot count on getting more than the Now if the sum of these upper amount 6 x in a coalition with player the amount (c x) + (b x) is less than what players 2 bounds and 3 can get by combining with eachother in a coalition, then we may 1 safely assumethat player 1 will find no partner. A coalition of 2 and 3 we see: player 1desires get an amount x If to can obtain the amount a. So under all conditions,then he is disqualified from any possibilityof finding

with coalitions: player 2 or with player 3. Assume that he attempts to an amount x under all conditions.In this caseplayer 2 cannot retain

22.2.2. Considerthe situation of player 1. Hecan entertwo alternative

1. i.e.

1.

- x) + - x) < a. to I.e.he desire get is unrealisticand absurd unless t (c - x) + - x) ^ a.

(c
(b x (b this restate :

This inequality may be written equivalently as

. **-a +2b + c

We

(22:l:a)

Player 1 cannot reasonablymaintain a claim to get under all conditionsmore than the amount a = ~ 2i

f The sameconsiderations be repeatedor players2 and 3, and they may give: (22:l:b) Player 2 cannot reasonablymaintain a claim to get under all conditionsmore than the amount
ft

^ i

(22:l:c)

Player 3 cannot reasonablymaintain a claim to get under all conditionsmore than the amount 7 = 5 &

1We assume, of course,hat a player is not indifferent to any possiblerofit, however t p small. This was implicit in our discussion of the zero-sum two-person game as well. The traditional ideaof the \"homo oeconomicus,\" to the extent to which it is clearly a conceivedt all, alsocontains this assumption.)))

FURTHEREXAMPLES

229

22.2.3. the criteria Now were only (22:l:a)-(22:l:c) necessaryones,and one could imagine a priori that further considerations could further lower their upper bounds, a, 0, y or lead to some other restrictionsof what the players can aim for. This is not so,as the following simple considerationshows.
One verifies immediately that

a + = c,

a + 7 = 6,

+7 =

<*

In other words: If the players 1,2,3o not aim at more than permittedby d (22:1 (22:1 (22:1 i.e.than a, 0, y respectively,then any two :a), :b), :c), players who combine can actually obtain these amounts in a coalition. Thus theseclaimsarefully justified. Of course only two players the two who form a coalition can actually obtain their \"justified\"dues. The third player, who is excludedrom the coalition, will not get a, ft, y respecf 1 tively, but a, 6, c instead.
point an obvious question presents itself:Any player amount a, 0,y respectivelyif he succeeds enteringa in i he coalition;f he does not succeed, gets instead only a, b, c. This makessenseonly if a, 0,y aregreaterthan the corresponding a, 6, c, sinceotherwise a player might not want to enter a coalition at all, but might find it more advantageous to play for himself. So the question is whether the threedifferences c 1,2,3an get the

22.3.1. this At

22.3. Inequality. An

Formulae

areall ^

It is immediately seenthat
P

0.

a _ (__ a) = a + a, = g (-&) =0 + 6, r = y ~ (-<0= y + c,

0-

they

: areall equal to eachother. Indeed

=9=r =a +b +c
^

----

We

denotethis quantity by A/2. Then our questionis whether))

This inequality can be demonstratedas follows: A 22.3.2.coalition of the players 1,2can obtain (from player 3) the amount c and no more. If player 1 plays alone, then he can prevent players 2,3from reducinghim to a result worsethan a sinceeven a coalition of i can players 2,3 obtain (from player 1) the amount +a and no more; .e. player 1 can get the amount a for himself without any outside help. Similarly,player 2 can get the amount 6 for himself without any outside betweenthem can get the amount help. Consequentlythe two players 1,2
1Theseare indeed the amounts which a coalition of the other players can wrest players 1,2,3respectively. The coalition cannot take more.)))
from

230))

ZERO-SUMTHREE-PERSON GAMES))

fail to cooperate with eachother. Since the maxithey can obtain togetherunder any conditions is c, this implies A = a +6+ c^ b 22.3.3. proof suggestshe following remarks: This t First: e have based our argument on player Owing to the symW with respecto the three players, t metry of the result the same inequality would have obtained if we had analyzed the situation of player 2 or player 3. This indicateshat thereexists certainsymmetry t a in the roleof the three players. A Second: = meansc = a 6 or just as well a = a, and the two correspondingairs of equations which obtain by the cyclicpermutation p of the three players. So in this caseno coalition has a raison d'&tre:Any two players can obtain, without cooperating, same amount which they the can produce perfect cooperation for players 1and 2 this amount is in (e.g. a b = c). Also, after all is said and done, each player who succeeds in joining a coalition gets no more than he could get for himself without outside help (e.g. player 1this amount is a = a). for If, on the otherhand, A > then every player has a definite interestin joining a coalition. The advantage contained in this is the same for all

(a -f 6) even if they

mum

-a - i.e.

c0.

A=a+6+c^O

1.

three players:A/2.

Herewe have again an indication of the symmetry of certainaspects it a of the situation for all players:A/2 is the inducementto seek coalition; is the same all players. for 22.3.4. result canbe expressed the following table:)) Our by
Player)
i)

2)

3)

With coalition) Value

a) a)

ft)

of a play)
Without Figure coalition)
49.))

7)
c))

-b)

If we put))

L, 1 a 1 6' = -6 + -A = /J- g A a-26+ c, g . 1A = 1 = a + b 2c, / c'=-c+

A A A

.,

T-gA
A

then we have

a'+ V + c' = 0,

the and we can express abovetable equivalently in the following manner: the A play has for the players 1,2,3 basicvaluesa', c'respec6', (22:A) is a possible valuation, since the sum of these))) tively. (This

THE GENERAL CASE

231

values is zero, f. above). The play will, however, certainly be c attended by the formation of a coalition. Thosetwo players who form it get (beyondtheir basicvalues) a premium of A/6 and the excluded layersustainsa lossof A/3. p Thus the inducement to form a coalition is A/2 for each player, and always A/2 ^ 0.

Let T be a perfectly arbitrary zero-sumthree-person ame. A simple g considerationsufficesto bring it within the reachof the analysis of 22.2., 22.3.We argue as follows: If two players,say 1and 2, decide cooperate to completely postponing o t temporarily,for a latersettlement,he question of distribution,i.e. f the a compensationsto be paid between partners then F becomes zero-sum two-persongame. The two playersin this new game are:the coalition 1,2 (which is now a compositeplayer consistingof two \"natural persons and the player 3. Viewed in this manner T falls under the theory of the zero-sumtwo-persongame of Chapter III. Each play of this gamehas a well defined value (we mean the v' defined in 17.4.2.). us denoteby c Let the value of a play for the coalition 1,2 (which in our presentinterpretation is one of the players). and Similarly we can assumean absolute coalition between players 1,3 view T as a zero-sum two-person game betweenthis coalition and the player

23.1.1. can now remove all restrictions. We

23.1. Exhaustive

23.The General ase C

Discussion.Inessential nd Essential Games a

\,

2.

We then denote b the value of a play for the coalition by a Finally we can assume an absolute coalition between players 2,3, nd view F as a zero-sum two-person game between this coalition and the We then denoteby a the value of a play for the coalition 2,3. player It ought to be understoodthat we do not yet! assumethat any such coalition will necessarilyarise. The quantities a, 6, c aremerely computationally defined; we have formed them on the basis of the main (matheof matical) theorem of 17.6.(Forexplicitexpressions a, 6, c cf. below.) Now it is clearthat the zero-sum three-person game F falls a coalition of the entirely within the domain of validity of 22.2., can or or players3 or 2 or 1) players 1,2 1,3 2,3 obtain (from the excluded all the amounts c, 6, a respectively,and no more. Consequently resultsof 22.3. in particularthe one formulated at the end which describes hold, 22.2., every player's situation with and without a coalition. Theseresults show that the zero-sumthree-person gamefalls to the possiinto two quantitatively different categories, corresponding bilities A = and A > 0. Indeed: coalitionshave no raison d'etre, A = We have seenthat in this case and each player can get the same amount for himself, by playing a \"lone hand\" against all others,as he couldobtain by any coalition. Inthis case,)))

1,3.

1.

23.1.2.

22.3.:

23.1.3. 0:

232

ZERO-SUMTHREE-PERSON GAMES

importance.

inessentialgame. A > 0:In this case thereis a definite inducementto form coalitions, as discussedat the end of 22.3. hereis no needto repeatthe description T given there; we only mention that now a > a' > a, ft > b' > 6, 7 > c' > c. We shall call a game in this case,in which coalitionsare a essential, n essentialgame. Our above classification,inessential and essential, pplies at present a only to zero-sum three-personames. But we shall see g subsequentlythat it can be extendedto all games and that it is a differentiation of central

and in this case alone,it is possibleto assumea unique value of eachplay for eachplayer, the sum of thesevalues being zero. Thesearethe basic values I the &', c' mentionedat the end of 22.3. n this case formulae of 22.3.how that = a = s V = ft = c1 = y = We shall call a game in this case,n which it is inessentialto consider oalitions, i an c

a',

a!

-a,

-b,

-c.

basedupon them in terms of which our solution was expressed. Assume the zero-sumthree-person game F in the normalized form of 11.2.3. the players 1,2,3hoosethe variablesTI, r2, r 3 respectively There c (eachone uninformed about the two other choices) nd get the amounts a T 3Ci(ri,T2, r 3), 3C2(ri, 2, r3), 3C (n, T2, T3) respectively. Of course 3 (thegame is : zero-sum) 3Ci(n,T2, T8) + JC2(ri, T2, T3) + 3C3(n, r 2, r 3) 0. The domainsof the variablesare:
A

23.2. we analyze this result any further, letus make a few purely Before mathematical remarksabout the quantities a, 6, c and the a, 0,7, a',6', ', c

23.2. CompleteFormulae

T! = T, =

1,2, - - - , 0,, 1,2, , 02, T3 = 1, , 3. 2,

Now in the two-persongame which arises betweenan absolute coalition of and players 1,2, the player 3,we have the following situation: The composite has player 1,2 the variables n, r 2; the other player 3 has the variable r8 The former getsthe amount

T 3Cl(Tl, 2, T3) +3C2(n, T2, T3) the latterthe negative of this amount.

3 -3Cs(Ti,T2, T3),

mixedstrategy of the compositeplayer 1,2 a vector is

of

characterized by

componentsof which we may denoteby {fiiT| .1 Thus the

*V r,

S^,the
fti

of S

pt

are

0,

*!/*

2 *v, = 1-

1The number of pairs n, TJ is, of course,

0ifa.)))

OF DISCUSSION AN OBJECTION
A

233

mixed strategy of the player 3 is a vector

j

17

of Sp the components f o

which we denoteby

v The
Hr,

i\\

of Sft arecharacterized by

0,
*\\

*, = 1.
is therefore 17.4.1.

The bilinear form K( ,

r?

in ) of (17:2)

K(

1?

2) iJCi(n,T2, T3)

+ JC(n,T2, T,)

TI

r^))

and

finally

= Max-* Min-> K(
f n

>

77

. = Min- Max-* K(
n

>

rj

).

f Theexpressionsor 6, a obtain from this by cyclical permutations of the in all details of this representation. players 1,2,3 We repeat the formulae expressing /3, 7, a', c and A: a, 6', = a + b + c necessarily ^ 0, A
f

-a +
+6

+c
c

-2a+ b +
+6
2c

c))

_a
and we have))

,'+,))

a' + b' + c' = 0,))

^ 0,

o 24.Discussionf an Objection TheCaseof PerfectInformation and Its Significance 24.1.

a f accountsor all possibilities nd which indicatesthe directionthat the searchfor the solutionsof the n-persongame must take:the analysisof all
which

We 24.1.1. have obtaineda solution for the zero-sum three-personame g

eachother,

possible coalitions,and the competitive relationshipwhich they bear to that players who which shoulddeterminethe compensations
want to form a coalition will pay to eachother. We have noticedalready that this will be a much more difficult forn ^ 4 playersthan it wasfor n = 3 (cf.footnote 2, 221).)))

p.

problem

234

GAMES ZERO-SUMTHREE-PERSON

1In other words in claiming general validity for a theory one necessarilyassumesthe burden of proof against all objectors. 2 Onemight hope to evadethis issue,by expecting to find A = for all zero-sum threeThis would make coalitions unnecessary. Cf. person games with perfectinformation. the end of Just as games with perfectinformation avoided the difficulties sum two-persongames by being strictly determined (cf. those of the zero-Bum three-person amee by being inessential.))) g

We have adduced already some strong positive argumentsfor it, in our discussionof the zero-sumthree-person game. Our ability to build the theory of the n-persongame(for all ri) on this foundation will, in fine, be the decisive positive argument. But there is a negative argument an which arises connection in with those games objection to be considered, where perfect information prevails. The objectionwhich we shall now discuss applies only to the above mentionedspecial ategory of games. Thus it would not, if found valid, c provideus with an alternative theory that appliesto all games. But since we claim a general alidity for our proposedstand, we must invalidate all v 1 even objections, thosewhich apply only to somespecialcase. Games 24.1.2. with perfect information have already beendiscussedin a 15.We saw there that they have important peculiaritiesnd that their in nature can beunderstoodfully only when they areconsidered the extensive form and not merely in the normalized one on which our discussion chiefly relied also 14.8.). (cf. The analysis of 15. egan by considering -person games (for all n), b n but in its laterparts we had to narrow it to the zero-sumtwo-persongame. At the end, in particular,we found a verbal method of discussing (cf. 15.8.) it which had some remarkable features: irst,while not entirely free from F it w t objections, seemed orth considering.Second,he argumentation used was ratherdifferent from that by which we had resolvedthe general ase c of the zero-sumtwo-persongame and while applicableonly to this special case,it was more straight forward than the otherargumentation.Third, it led for the zero-sumtwo-persongameswith perfect information to the sameresult as our generalheory. t Now onemight be tempted to use this argumentation for n ^ 3 players i does too;ndeeda superficial inspectionof the pertinent paragraph 15.8.2. not immediatelydisclose reason any why it shouldbe restricted (asthere)to n = 2 players (cf.,however, 15.8.3.). this procedure But makes no mention of coalitionsor understandingsbetween players, etc.;o if it is usable s 2 for n = 3 players,then our presentapproachis open to grave doubts. We

o reconsiderur position. In the discussionswhich follow we shall put the between on main stress the formation of coalitionsand the compensations the participants in thosecoalitions, using the theory of the zero-sumtwot person game to determinehe values of the ultimate coalitionswhich oppose eachother after all players have \"takensides\"(cf. 25.1.1., But is 25.2.). o this aspect f the matter really as universal as we proposeto claim?

Before we attackthis question, it is well to pause for a moment to

23.1.

15.6.1.), they would

of the theory of zeronow avoid

DISCUSSION AN OBJECTION OF

235

is inconclusive when proposeto show therefore why the procedureof the number of players is threeor more. To do this, let us repeat omecharacteristicteps of the argumentation s s in question (cf. the notations of which we are also using).

15.8.

15.8.2.,

j(= 1,2,

24.2.Detailed Discussion. Necessityof Compensations between Threeor More Players 24.2.1. Consider accordinglya game F in which perfect information , <r, the choices prevails. Let SfTCi, 3fE 2, , 9(11,be its moves, <TI,(7 2, with these moves, T(<TI, connected , 0-,) the play characterized y b thesechoices, 3v(7r(oi, and , <r,)) the outcome of this play for the player
- -

Assume that the moves 3TCi, 3Tl2, h , 3TC,-i ave already been made, the outcome of their choices eing <TI, <r 2, , cr,_i and considerthe last b move i.e. (en, move 3TC,and its <r,. If this is a chance ft, , <r,~\\) = 0, = values a, h then the various possible , a,-i) ave the , a,(cri, , p,(a,(o-i, , cr,_i)),respectively. If this p probabilities ( 1), pv (2), is a personalmove of player A: i.e. ,<r,-i)= k = , n, fc,(ai, a then player k will choosef so as to make SFjb(ir(ori, , 0v-A| tr,)) a maximum. Denotethis <r, by <r,(<n, , *,_i). Thus one can argue that the value of the play is already known (for eachplayer = n) after as a function of the moves , 2(TC,-i (and before 9ftl,!), 3Tl2, 0*1,0% , <r,-i alone. Indeed: y the above it b
*

,n).

1,2,

1,2,

Sflfcj,

j 1, i.e.

is))

*-!

for

fc,(<7i,

, <r,_i)

= 0,

where <r, = <r,(<n,

S^Or^i,

, <r,-i) maximizes , <r,-i, <r,)) for

Consequentlywe can treat the game T as if it consistedof the moves

9Tli,
9fn

By this device we have removed the last move 9TC,. Repeating it, we the cansimilarly remove successively moves 3fTC,_i,9fH,_a, , 9H a , 9Tli and obtain a definite value of the play (for eachplayerj = 1, , 2 n). finally considerhe last two t 24.2.2. a critical ppraisal of this procedure For a steps 9Tl,_i, 9fll, and assumethat they are personalmoves of two different This, however, is not the case. To seethat, it suffices to modify the rules of the
simple majority the moves (i.e.
informed about three coalitions

2,

, 3Tl,_i only (without 3TC,).

as follows: Let the players 1,2,3make their personal game (cf. 21.1.) choices n, TJ, r* respectively,cf.loc.cit.) in this order, eachone being of the anterior moves. It is easy to verify that the values c, 6, a of tho 1,2,1,3, are the same as before 2,3

to A detaileddiscussion of this game, with particular respect the considerationsof to would be of a certain interest, but we do not propose continue this subject further at present.)))

c-6-a-l,A-a+6+c-3>0.

21.2.,

236

GAMES ZERO-SUMTHREE-PERSON

respectively. In this situation we have assumed that players, say 1,2 cr, , o-,_i, <r y). player 2 will certainly choose so as to maximize $2(0-1, This gives a a> = 0-,(<ri, , <r,_i). Now we have also assumed that that the i < player 1,n choosingr,_i can rely on this;i.e. he may safely replace b , ov-i, 0-,), (which is what he will really obtain), y $I(<TI, 1 , cr,_i)) and maximize this latter quantity. $1(0-1, , (rr _i, OV(<TI, But canhe rely on this assumption? To beginwith, <r y (cri, , <r,,_i) may not even be uniquely determined: , o>_i) , cr,_i, (T,) may assumeits maximum (for given <TI, $2(^1, at severalplacesv In the zero-sumtwo-persongame this was irrelevant: o value to $F 2, also give the there$1= #2, hencetwo o> which give the same samevalue to 9Fi. 2 But even in the zero-sum three-personame,$2 doesnot g o determine due to the existencef the third player and his $3! So it $1, happens herefor the first time that a difference which is unimportant for one player may be significant for another player. This was impossiblein the zero-sumtwo-persongame,where eachplayer won (precisely)what the
if then must player 1 expect two v v areof the same importancefor ? t but not for player 1 One must expecthat he will try to induce player 2, the 2 to choose a, which is more favorable to him. Hecould offer player to pay to player 2 any amount up to the difference this makesfor him. one This being conceded, must envisagethat player 1may even try to a induceplayer2to choose a v which doesnot maximize S^i, ' , ov-i, ov). causesplayer 2 less of a loss than it causesplayer 1 As long as this change a gain,3 player 1 can compensate layer 2 for his loss, and possiblyeven p give up to him some part of his profit. But 24.2.3. if player 1can offer this to player 2, then he must also count t on similar offers coming from player 3 to player 2. I.e.hereis no certainty of at all that player 2 will, by his choice o-,,maximize 2(0-1, , <r,_i, cr v). In comparing two a v one must considerwhether player 2J s loss is overby gain, sincethis couldlead to undercompensated player Ts or player 3's and compensations.I.e. nemust analyze whether a coalition 1,2 o standings would gain by any modification of a v . or 2,3 This 24.2.4. bringsthe coalitionsback into the picture. A closer analysis would lead us to the considerationsand results of 22.2., 23. 22.3., in every detail. But it does not seem necessaryto carry this out herein complete a detail: fter all,this is just a special and the discussionof 22.2., 22.3., case, w 23. as of absolutelygeneralvalidity (for the zero-sumthree-personame) g

otherlost.
What

, a>_2, <r,_i only, of which <TI, , o>_2areknown he is ableto maximize it. He cannot in any sensemaximize $1(0-1, on , 0>-i,ov) sincethat alsodepends <r, which he neither knows nor controls. 2 Indeed, refrained in we from mentioning $2 at all: instead of maximizing $2, we talked of minimizing There was no need even to introduce 9(a\\ t , r-i) and everything was described Max and Min operations on JFi. by 3 when it happens at the expense player of at
9Rr-iy and

1Since is a function of <TI, this <r,,_i is controlled by

player

1,

15.8.2. $1.

I.e.

3.)))

DISCUSSIONF AN OBJECTION O

237

of i.e. providedthat theconsideration understandingsand compensations, of is permitted. coalitions, of We wanted to show that the weakness the argument of 15.8.2., already in 15.8.3., becomes destructive exactly when we go beyond the recognized zero-sumtwo-persongames, nd that it leads preciselyto the mechanismof a coalitions etc.foreseen in the earlierparagraphs of this chapter. This shouldbeclearrom the above analysis,and so we can return to our original f methodin dealingwith zero-sum c three-personames, i.e.laim full validity g for the results of 22.2., 23.))) 22.3.,

CHAPTER VI FORMULATION THE GENERAL THEORY: OF ZERO-SUM n-PERSONGAMES

25.TheCharacteristicunction F 25.1. Motivation and Definition

the that gained in ChapterV concerning casen = 3 suggests experience in of the possibilities coalitionsbetweenplayers will play a decisiverole the generaltheory which we are developing. It is therefore important to evolve a mathematical tool which expresses these \" possibilities\"in a
We 25.1.1. now turn

to the zero-sumn-persongame for general . The n

quantitative way. o Sincewe have an exact oncept f \" value\" (of a play) for the zero-sum c we can also attribute a \" value\" to any given group of two-persongame, of players,providedthat it is opposedby the coalition all the otherplayers. We shall give theserather heuristicindicationsan exactmeaning in what follows. The important thing is, at any rate, that we shall thus reacha o mathematical concept n which one can try to basea generalheory and t that the attempt will, in fine, prove successful. definitions which carry out this Letus now statethe exact athematical m we have a gameF of n players who, for the be denoted by 1,2, , n. It is convenient to introducethe set / = (1,2, , n) of all these players. Without yet

program.

sakeof

25.1.2. then that Suppose

brevity,
will

is likely to

a predictionsor assumptionsabout the course play of this game if we observethis: we group the players into two parties, take, if and treateachparty as an absolutecoalition i.e. we assumefull coopera1 tion within eachparty then a zero-sumtwo-persongame results. PreLet S be any given subset of 7, S its complementin /. We cisely: consider zero-sumtwo-persongamewhich results when all players k the with belongingto S cooperate eachotheron the onehand, and all playersk to S cooperate eachotheron the otherhand. with belonging Viewed in this manner F falls under the theory of the zero-sumtwoperson gameof ChapterIII. Each play of this gamehas a well defined Let value (we mean the v' defined in 17.8.1.). us denoteby v(S) the value i of a play for the coalition of all k belonging S (which, in our present nterto pretation, is oneof the players).
making any
already alluded

1This is exactly what we did in the casen to at the beginning of

24.1.
238)))

3 in

The 23.1.1. generalpossibility

was

THE CHARACTERISTICUNCTION F

239

1 for Mathematicalexpressions v(S) obtain as follows: Assume 25.1.3. the zero-sum n-persongame F in the normalized form of a 11.2.3. Thereeachplayerk = 1, , one , n chooses variable r* (each 2 1otherchoices)nd getsthe amount a uninformed about the n
3C*(Tl, T2,

Of course(the game is zero-sum)

, Tn ).

(25:1)
T* =

*-i

3e*(n,
for

, r n) m

0.
,

The domainsof the variables are:

1,

j8*

= 1,2,

n.

Now in the two-person game which arisesbetweenan absolutecoalition of all playersk belongingto S (player and that one of all players k belonging w to S (player 2'), e have the following situation: of The compositeplayer has the aggregate variablesT* where k runs to o a over all elements f S. It is necessary treatthis aggregates one variable and we shall therefore designate it by one symbol ra The composite of player 2' has the aggregate variablesrk where k runs over all elements too of S. This aggregate is one variable, which we designateby the symbolr~a The player getsthe amount))

1')

1'

1'

s) (25:2) 5c(rV- =

OC*(n, kinS

, r n)

km

-S

3C(n,

, rn

);'

the player 2'getsthe negative of this amount. A mixed strategy of the player I/ is a vector of which we denote by
r

8 of <S0, the components

Thus the

of S0aarecharacterized by

t- ^ o,
A

2) s*a = L
T

mixed strategy of the player 2'is a vector rj of /V,4 the components of which we denote rj T-a. Thus the ij of Sp-aarecharacterized by by
rj

r -8

^0,

r~

5J rj T-a =

1. .

zero-sum property. 8 3 is the number of ft a possibleggregatesTS, the all elements of S. 4 0~ is the number of possible s aggregatesr\" 5, over all elements of
<S.)))

1This is a repetition of the construction of 23.2., applied only to the special which casen 3. 1The TS, r~5 of the first expressionform together the aggregate of the n, , r n of so , the two other expressions; TS r~3 determine those n, , rn The equality of the two last expressionsis, of course,only a restatement of the

i.e. product of all 0*where k runs over i.e. product of all 0*where k runs the

240

GENERAL THEORY:ZERO-SUMn-PERSONS
> >

The bilinearform K(

t?

) of

in is therefore (17:2) 17.4.1.

K(7,7)and
finally))

rVs

#(rV-*)r'*-',
)

v(S) = Max-Min-

K(,

= Min-

Max-^

K(,

).

25.2. of Discussion the

Concept

should be determined tions can be formed. The \"compensations\" essenthe possibilitieswhich existfor each partner in the coalition S tially by for other (i.e. eachk belongingto *S), to forsakeit and to join some coalition a T. (One may have to consider lso the influence of possiblesimultaneous d and concertedesertions sets of several partners in S etc.) I.e.he t by of v(/S) to the playersk belongingto S shouldbe determined))) \"imputation\"

set function. We call it the characteristic function of the game T. As we have repeatedly indicated,we expectto basethe entire theory of the zero-sumn-persongameon this function. It is wellto visualize what this claim involves. We proposeto determine everything that can be said about coalitionsbetween players, compensaor tions betweenpartners in every coalition, ergers fights between coalim function v(*S) alone. Prima in terms of the characteristic facie, tions, etc., this program may seem unreasonable, particularly in view of thesetwo facts: to (a) An altogetherfictitious two-persongame,which is related the was used to define realn-persongameonly by a theoretical construction, v(S). Thus v(S) is based on a hypothetical situation, and not strictly on the n-persongameitself. w (b) v(S) describes hat a given coalition of players (specifically, the set S) can obtain from their opponents (the set AS) but it fails to of h describeow the proceeds the enterpriseare to be divided among the k belonging to S. This division, the \"imputation,\" is indeed partners , r n ), k belongdirectly determinedby the individual functions 3C*(Ti, to S,while v(S) dependson much less.Indeed, (S) is determinedby v ing their partial sum JC(r5, r~5) alone,and even by lessthan that sinceit is 5 the saddle value of the bilinearform K( , y ) based on 5C(r , r~\" s) (cf.the formulae of In 25.2.2. spite of these considerationswe expectto find that the function v(S) determines verything, including the \"impucharacteristic e tation\"(cf. (b) above). The analysisof the zero-sumthree-personamein g Chapter V indicatesthat the direct distribution (i.e.,"imputation\ \ , r n ) is necessarilyoffset by some by means of the 3G*(ri, system of which the players must make to eachotherbefore coali\"compensations\"

The 25.2.1. above function v(S) is defined for all subsets 8 of I and has a numerical real numbers as values. Thus it is, in the senseof 13.1.3.,

25.1.3.).

strated this for the zero-sum three-personamein ChapterV. One of the g main objectives f the theory we are trying to build up is to establish the o same thing for the general -persongame. n

by

241 THE CHARACTERISTICUNCTION F the other v(T) and not by the 3C*(ri, , rn ). We have demon1

this function as a mathematical entity in itself. We know that it is a numerical set function, defined for all subsets S of / = (1,2, , n) and we now proposeto determineits essential roperties. p It will turn out that they arethe following:

w acteristic function v(S) for the generalheory of games, e shall investigate t

25.3.1. we undertake to elucidatethe importanceof the Before

26.3. Fundamental

Properties

char-

(25:3:a) (25:3:b)
fulfills with

(25:3:c)

v(S u T) ^ v(S) + v(T),

v(-S)= -v(S),
if

v(0) = 0,

Sn T =

0.
of every game

(25:3:a)-(25:3:c). out 25.3.2. simplestproof is a conceptual which can becarried The one,

We prove first that the

set characteristic function

v(>S)

one might desirea formalistic proof in terms of the operations Max strictly mathematical, and Min and the appropriate vectorial variables. We emphasizetherefore that our conceptual roof is strictly equivalent to the desired formalistic, p out without mathematical one, and that the translation can be carried any real difficulty. But sincethe conceptual roof makes the essential p ideas clearer, in a briefer and simplerway, while the formalistic proof and would involve a certain amount of cumbersomeotations, we prefer to n w give the former. The reader ho is interested may find it a good exercise one. to construct he formalistic proof by translating our conceptual t 2 Thecoalition so it always has no members, Proof of (25:3:a) t getsthe amount zero,herefore v(0) = 0. from the same(fictitious) Proof of (25:3:b): and v(-S)originate v(*S) zero-sum two-person game, the one played by the coalitionS against

exactmathematical expressionsor v(S) in f

practically no mathematical formulae. However, since we gave

25.1.3.,

25.3.3.

1All

should we are treating even the empty set asa coalition. Thereader the think this over carefully. In spite of its strange appearance, step is harmless and quite in the spirit of general set theory. Indeed,it would be technically quite a nuisance to exclude empty set from consideration. the Ofcoursethis empty coalition has no moves, no variables, no influence, no gains, and no losses. But this is immaterial. The complementary set of , the set of all players 7, will alsobe treated as a possible from the set-theoretical coalition. This too is the convenient procedure point of view. To a lesserextent this coalition alsomay appearto be strange, sinceit has no opponents. Although it has an abundance of members and henceof moves and variables it will (in a zero-sum game) equally have nothing to influence, and no gains or losses. But this

existence. 1Observe that

this

is very

much in the

senseof the remarks

in

on 4.3.3.the

role of

\"

virtual\"

too is immaterial.)))

242

GENERAL THEORY:ZERO-SUMn-PERSONS

the coalition

S. The value of a play of this game for its two composite playersisindeedv(S)and v( S) respectively. Therefore v( S) = v(S). The Proof of (25:3:c): coalition S can obtain from its opponents (by using an appropriatemixedstrategy) the amount v(S) and no more. The the coalition T can obtain similarly the amount v(77) and no more. Hence coalition S u T can obtain from its opponents the amount v(S) + v(T), 1 with S even if the subcoalitions and T fail to cooperate eachother. Since the maximum which the coalition S u T can obtain under any conditionis 2 v(S u T) this impliesv(S u T) ^ v(S) + v(2T). 25.4.Immediate Mathematical Consequences Before conclusionsrom the above 25.4.1. we go further letus draw some f These (25:3:a)-(25:3:c).will be derived in the sensethat they hold for irrespective any numerical set function v(S) which fulfills (25:3:a)-(25:3:c) function of a zero-sumn-person of whether or not it is the characteristic
gameF. (25:4)

= 0. (25:3:b), = v(-0) = v(7) Proof:* By (25:3:a),

v(7)

-v(0) = 0.
+ v(S9)

(25:5)
if

v(5i u

Sp)

v(5i) +

Si,

Sp arepairwisedisjunct subsets of 7.

I applicationof (25:3:c). Proof:mmediatelyby repeated

(25:6)

and even somewhat less can replace (25:3:a)-(25:3:c) equivalently.

+ v(S,) g v(Si) + of if Si, p , Sp are a decomposition 7, i.e. airwisedisjunct subsets of 7 with the sum 7. u Sp = 7, hence v(Si u - - - u Sp) = by Proof:We have Siu T (25:4). herefore (25:6)follows from (25:5). :3 of 25.4.2. (25:4)-(256)areconsequences (25:3 While :a)-(25 :c), they :
(25:A)

Precisely:

areequivalent to the asserThe conditions(25:3:a)-(25:3:c) tion of (25:6) the values p = 2, 3 only; but (25:6)must for then be statedfor p = 2 with an = sign, and for p = 3 with a
sign.

1,

1,

1Observe If S and T had common elements, we that we are now using S n T could not break up the coalition S U T into the subcoalitions S and T. 1This proof is very nearly a repetition of the proof of a + b 4- c ^ in One t couldeven deduce (25:3:c) that relation:Considerhe decomposition of into from our the three disjunct subsets T, (SU T). Treat the three corresponding (hypothetical) S, absolute coalitions as the three players of the zero-sum three-persongame into which this c loc. cit.; ence h transforms T. Then v(), v(T), v(S U T) correspondto the a+b+c means -v(S) v(T) 4- v(S U T) 0; v(S U T) v(/S) + v(T). For a v(S) originating from a game, both (25:3:a) (25:4)are conceptually and contained in the remark of footnote 2 on p. 241.)))

0.

22.3.2.

i.e. -a,-b,

GIVEN CHARACTERISTICUNCTION F (we write

243

Proof:(25:6) for p = 2

for Si (25:6) p = 1with an = sign statesv(/) = (in this case must be this /) which is exactly (25:4). wing to (25:3:b), is exactly the same O as (25:3:a). the above proof of (25:4).) (Cf. for (25:6) p = 3 with an g signstatesv(S) + v(T) + v(-(S u T)) ^ (we write S, T for Si,S2; henceSB is (S u T)), i.e. -v(-OSuT)) vOS) +v(T). this By (25:3:b) becomes u T) ^ v(S) + v(T) which is exactly (25:3:c). v(S So our assertionsareequivalent preciselyto the conjunction of (25:3:a)-

(25:3:b).

S for S\\,

= sign states v(S) + v(~S) = henceSiis S); v( S) = v(S) which is exactly i.e.
with

an

(25:3:c).

a there exists fulfills the conditions (25:3:a)-(25:3:c) function. zero-sumn-person F of which this v(S) is the characteristic game In orderto avoid confusion it is betterto denotethe given numerical set function which fulfills (25:3:a)-(25:3:c) We shall define by v (S). with its help a certainzero-sumn-person game F, and denotethe characteristic function of this F by v(/S). It will then benecessaryto prove that)) function v(S) which

26.Constructionof a Game with a Given Characteristicunction F 26.1. Construction The 26.1.1. now prove the converseof 25.3.:That for any numerical set We 1.

1This game F is essentially a more general analogue of the simple majority game of three persons, efined in d We shall accompany the text which follows with footnotes pointing out the details of this ~ analogy. 2 The w-element set 7 has 2n l subsets S containing k, which we can enumerate by an l index r*(/S) i 2, , 2*~ nIf we now let the player choose,nstead of *, its ~ index r* Tk(Sk) 2, , 2 l , then the game is already in the normalized form of 1 Clearly all fa 8 The are the analogues of the couples in The contents of footnote 3 on rings i p. 222apply accordingly; in particular the rings are the formal conceptn the set of rules of the game which inducesthe coalitions which influence the actual course eachplay. of 4 Verbally: A ring is a set of just players, in which every onehas chosen this set. Theanalogy with the definition of a couplein is clear. Thedifferences aredue to formal convenience: in we made eachplayer designate the other element of the which he desires; we expect now him to indicate the entire ring. A closer nalysis a couple of this divergencewould be easyenough, but it doesnot seemnecessary. 6 Proof:Let S and T be two Sk rings with a common element k\\ then by (26:1) = S and Si, T, and so S

Let therefore a numerical set function v (S) which fulfills (25:3:a)1 be (25:3:c)given. We define the zero-sumn-person F as follows: game a Eachplayer k = 1, , n will, by a personalmove, choose subset , 2 Sk of / which containsk. Eachone makeshis choice independentlyof the choice the otherplayers.2 of After this the payments to be made aredeterminedas follows: Any set S of players, for which Sk = S for every k belongingto S (26:1) 6 is called ring*** Any two rings with a common elementareidentical. a

21. 1.

1,

11.2.3.

1,- 2*' .

A;

21.1.

21.1.

1. 21.

T.)))

244

GENERAL THEORY:ZERO-SUMn-PERSONS

In other words:The totality of all rings (which have actually formed in a play) is a system of pairwisedisjunct subsets of /. Eachplayer who is containedin none of the rings thus defined forms by himself a (one-element) which is called solo set. Thus the totality set a of all ringsand solosets(which have actually formed in a play) is a decomposition of i.e. system of pairwisedisjunct subsets of I with the sum I. a Denotethese setsby Ci, numbers of their , Cp and the respective elements y n\\, , n p. b Consider now a player k. He belongs to precisely one of these sets Ci, , Cp say to Cq . Then player k getsthe amount))

/;

(26:2)

nq

_v

.(C.)n

r-l

(Cr).1

function v

This completes descriptionof the game T. We shall now showthat the this F is a zero-sumn-person game and that it has the desiredcharacteristic

Eachoneof the

amount, stated in playersbelongingto it getsthe same Hence players of Cq togetherget the amount the (26:2).
nq

26.1.2. of the zero-sumcharacter: onsiderone of the C Proof

().

sets Cq .

(26:3)

r-l 1,
,n

In orderto obtain the total amount which all players 1, must sum the expression o over (26:3) all setsCq, i.e.ver all q =
This sum is clearly))

get, we
,

p.

i.e. ero. z 2

Q-l

- r-lv,(C),
r

the function is v (S) Denote characteristic Proof that the characteristic hold for function of T by v(S). Remember that (25:3:a)-(25:3:c) v(S) because is a characteristic it function, and for v (S)by hypothesis. Consealso hold for both v(S) and v (/S). quently (25:4)-(25:6) We prove first that for all subsets 8 of /. (26:4) v(iS) ^ v (S) If S is empty, then both sidesarezero by (25:3:a). we may assumethat So S is not empty. In this casea coalition of all playersk belongingto S can 1Thecourse the play, that is the choices i, of of , Sn or, in the sense footnote 2 S on p. 243,the choices , n , r n determine the Ci, ; Cp, and thus the expression i (26:2).Of course(26:2)s the JC*(n, , r n ) of the general theory.
p 2 Obviously

2^ n q

n.)))

AND ESSENTIALGAMES INESSENTIAL

245

o govern the choices f its Sh so as with certainty to make S a ring. It sufhis fices for every k in S to choose Sk = Whatever the other players , Cp, (in S) do, S will thus be oneof the sets(ringsor solosets)Ci, hencethe entire say Cg Thus each in Cq = S gets the amount (26:2); coalition S getsthe amount (26:3). ow we know that the system)) N

S.

fc

is a decompositionof

/;

henceby
fl

(25:6) ]

1 i (26:3)s ^ v (C ) = VoOS). In other words,the playersbelongexpression for ing to the coalition S can secure themselvesat leastthe amount v (5) irrespectiveof what the playersin S do. This meansthat v(S) ^ v (S);

r-l

(Cr) ^

0.

That is, the

i.e.26:4). (

Now we can establishthe desiredformula))

(26:5) v(S) to this Apply (26:4) Owing to (25:3:b) means -v(S)^ (26:6) v(S) g VoOS). 2 (26:4),(26:6) together(26:5). give

-S.

-v (S), i.e.

26.2. Summary

26.2. sum up:in paragraphs 25.3.-26.1. obtained a comTo we have functions v(S) of plete mathematical characterization of the characteristic all possiblezero-sumn-persongamesF. If the surmisewhich we expressed in 25.2.1. to be true, i.e. we shall be able to base if the entiretheory proves of the game on the globalpropertiesof the coalitionsas expressed v(S), by then our characterization of v(/S) has revealed the exactmathematical of substratum of the theory. Thus the characterization v(S)and the funcareof fundamental importance. tional relations(25:3:a)-(25:3:c) We shall therefore undertake a first mathematical analysisof the meaning and of the immediatepropertiesof theserelations. We call the funcin

tions which satisfy them characteristicfunctions even when they to themselves,without reference any game.

areviewed

27. Strategic and G Equivalence. Inessential Essential ames

TheReducedForm Consider a zero-sum n-person game T with the characteristic function v(S). Let also a system of numbers aj, , 2 be given. We

27.1.1. (),

27.1. Strategic Equivalence.

the expression(26:3), the total amount obtained by the coalition o alone. But we derived for it a lower by the choices f the players in bound v which is determined. 2 Observe that in our discussion of the good strategies of the (fictitious) two-person and (our aboveproof really amounted to that), we game between the coalitions considerednly pure strategies, and no mixed ones. In other words, all these two-person o gameshappened to be strictly determined. This, however, is irrelevant for the end which we are now pursuing.)))

1Observehat t is not determined

i.e.

S,

246

GENERAL THEORY:ZERO-SUMn-PERSONS

with F in all details exceptor now form a new game which agrees f is playedin exactlythe same way as F, but when all is over, player k gets in F' the amount which he would have got in F (after the sameplay),plus (Observe that the , 2 are absolute constants!) Thus if F is with the functions)) brought into the normalized form of

F'

F'

this:

aj,

a.

11.2.3.
with

r 3C*( i>

3C*(ri, - - , r n ) ajj. with F) if and only if n-persongame (along

*
>

then

F' is also in

''

rn)

this normalized form,

the correspondingfunctions Clearly F' will be a zero-sum

(27:1)
which we assume.

fc-1

= 0,

the function of F'by v'(S),then clearly)) Denote characteristic

(27:2)

v'(S)

kinS

of Now it is apparent that the strategic possibilities the two gamesF and F' difference between thesetwo games cononly sists of the fixed payments J after each play. And these payments are absolutelyfixed;nothing that any or all of the players can do will modify them. Onecouldalso say that the positionof eachplayer has beenshifted t possibilities, he inducementsand by a fixed amount, but that the strategic to form coalitionsetc., entirely unaffected. In otherwords: are possibilities functions v(S) and v'(S) arerelated eachotherby to If two characteristic 2 then with the characteristic function v(S)is fully equivaevery game (27:2) , lent from all strategicpoints of view to some game with the characteristic and conversely. I.e. (S) and v'(S) describe strategitwo function v'(S), v v(S)and v'(S) may themcally equivalent families of games. In this sense

are exactly the same. The

selvesbe consideredquivalent. e in Observe that all this is independent of the surmise restated 26.2., to according which all games with the samev(S) have the same strategic

characteristics.

The 27.1.2. transformation (27:2)(we needpay no attention to (27:1), as cf. footnote 2 above) replaces, we have seen,the set functions v(/S) by a

1The truth of this relation becomes apparent if one recallshow v(S), v'(S) were It is alsoeasy to prove (27:2)formalistically defined with the help of the coalition with the help of the3C*(n, , rn , r n ), 3Ci(n, s Under these conditions follows and need not be postulated separately. v(7) v'(7) 0, hence(27:2) gives Indeed,by (25:4)in

S.

25.4.1., klal
2)
<4

(27:1)

).

- 0;

i.e.

- 0.

k-l)))

INESSENTIAL AND ESSENTIALGAMES

247

strategically fully equivalent set-function v'(S). We therefore call this relationshipstrategicequivalence. of We now turn to a mathematical property of this concept strategic functions. equivalenceof characteristic f It is desirableto pick from eachfamily of characteristicunctions v(S) in strategicequivalencea particularly simple representativev(/S). The ideais that, given v(S),this representativev(S)shouldbe easyto determine, and that on the otherhand two v(S)and v'(S) would be in strategic equivalenceif and only if their representatives v(*S) and v'(*S) are identical. we theserepresentativesv(S) in such a fashion Besides, may try to choose that their analysisis simplerthan that of the original v(S). 27.1.3. we started from characteristic functions v(S) and v'(S), When o couldbe based upon (27:2) then the concept f strategic alone; equivalence ensued (cf. footnote 2, p.246). However,we proposeto start now (27:1) a from one characteristic function v(S)alone, nd to survey all possiblev'(S) the which arein strategic with it in order to choose representaequivalence tive v(/S) from among them. Therefore the questionarises which systems the f aj, , aj we may use, i.e.or which of thesesystems(using(27:2)) fact that v(S) is a characteristic function entails the same for v'(S). The answer is immediate, both by what we have said so far, and by direct is verification: The condition (27:1) necessaryand sufficient.1 Thus we have the n indeterminate quantities aj, , a at our an ar disposalin the searchfor a representativev(S); but the S> ' ' S (27:1). o we have n 1 free parameters at subjectto one restriction: our disposal. t 27.1.4. may therefore expecthat we can subjectthe desiredrepreWe the sentative v(S) to n 1 requirements.As such we choose equations))
*
>

(27:3)

v((l)) = v((2)) =

2 . = v((\.

I.e. e requirethat every one-man coalition every player left to himself w should have the same value. a into (27:3) nd statethis together ith (27:1), w We may substitute (27:2) and so formulate all our requirementsconcerning aj, the So ,
we obtain:

a.

(27:1*)
(27:2*)

t-i))

al = 0,

It is easy to verify that theseequationsaresolvedby preciselyonesystemof

'))

1This detailed discussion may seempedantic. We gave it only to make clearthat when we start with two characteristicfunctions v(S)and v'(S)then is superfluous, but when we start with onecharacteristicfunction only, then is needed. 1Observe that theseare n and not n equations.)))

(27:1) (27:1)

248

GENERAL THEORY:ZERO-SUMn-PERSONS))

(27:4) So we can say:

a{
function v(S)reducedif and only if it We call a characteristic satisfies (27:3). hen every characteristicunction v(S) is in T f with precisely one reducedv(S). This strategicequivalence a a v(S)is given by the formulae (27:2) nd (27:4), nd we call it the

(27: A)

beenlooking.

reducedform of v(S). The reducedfunctions will be the representativesfor which we have

27.2.nequalities. The Quantity r I 27.2. us considera reduced characteristic Let function v(S). denotethe joint value of the n terms in (27:3) y 7, i.e. b (27:5)

We

7))

We can

state(27:5)also this way: set (27:5*) v(S) = 7 for every one-element S.

in i transforms (27:5*)nto (25:3:b)25.3.1.

Combinationwith

(27:5**)
We

v(S) = 7

for every (n

- l)-element S. set
i.e.:

of of i a defining 7 just a restatement (27:3), .e. characterization the reduced nature of v(S). sets to Now apply (25:6)in 25.4.1. the one-element Si = (1), , = (n). (Sop = n). Then (27:5)gives -ny ^ 0, <S

i ( re-emphasizethat any one of (27:5),(27:5*), 27:5**)s besides

(27:6)

7^0.
,

elements: = (fti, S s element etsSi = (fci),

Apply

Let p be the number of its Considernext an arbitrary subset S of in k p). Now apply (25:5) 25.4.1. the oneto ,

/.

Sp =

(k p).

Then (27:5) gives))

25.3.1.,
1

this also to

the above inequality now becomes

S which

has n

p elements.Owing to
v(/S)

in (25:3:b)

-*(fl)2S -(n
to al

p)75

i.e.
in

Proof:Denotethe joint value of the n terms becomes -v((fc)) + 0, and so (27:1*)

n0

S (n p)y. a (27:2*)y 0. Then (27:2*)mounts b

n

- % v((*)) - 0;
n

i.e. -

fc-i

*-i)))

AND ESSENTIALGAMES INESSENTIAL

249

Combiningthesetwo inequalitiesgives:

(27:7)
this way:

py

v(/S)

g (n

p)y

for every p-element set

S.

and (27:5*) v(0) =

in can (i.e. 25:3:a) 25.3.1.) also be formulated (

(27:7*)
this

Forp = 0,1we have = in the first relation of (27:7). can a (27:5**) nd v(7) = (i.e.25:4)in 25.4.1.) also be formulated (
way:

(27:7**) Forp = n

1,n we have = in the secondrelation of (27:7).

alternatives. This distinctionis based on (27:6):

perfectly trivial

27.3.nessentiality and Essentiality I In 27.3.1.analyzing theseinequalitiesit is best now to distinguishtwo

for possibilities. Thereis no occasion any strategy of coalitions,no element of struggle or competition: player may play a lone hand, sincethere each is no advantage in any coalition. Indeed, every player can get the amount zero for himself irrespective of what the others are doing. And in no coalition can all its memberstogether more than zero. Thus the value get of a play of this game is zero for every player, in an absolutelyunequivocal way. If a generalcharacteristic function v(S) is in strategicequivalence form is v(S) = then we have the with such a v(S) i.e. its reduced if same conditions,only shifted by a for the player k. A play of a game F with this characteristic function v(S) has unequivocally the value ajj for o the player k:he can get this amount even alone,irrespectivef what the others aredoing. No coalition coulddo betterin toto. We call a game F, the characteristic function v(S) of which has such a
1 form v(S) = 0,inessential. reduced in 27.3.2. case > 0. By a change unit 2 we couldmakey = Second :y

Firstcase:7 = 0. Then (27:7)gives v(S) =0 for all S. This is a case,in which the game is manifestly devoid of further

I.

o This obviously affects none of the strategically significant aspects f the and it is occasionally convenient to do. At this moment, howgame, quite ever, we do not proposeto do this. In the present case, t any rate,the players will have good reasonsto a t want to form coalitions. Any player who is left to himself loseshe amount 1players who he gets cf. (27:5*) (27:7*)), hile any n or w y (i.e.

-7,

0.)))

1 That this coincidesith the meaning given to the word inessential in 23.1.3. w (in the c special aseof a zero-sum three-persongame) will beseenat the end of 27.4.1. 2 Since In a wider senseit might payments are made, we mean the monetary unit. be the unit of utility. Cf. 3 This would not have been in possible the first case,where 7

2.1.1,

250 or

GENERAL THEORY:ZERO-SUMn-PERSONS

the their win cooperate together amount 7 (i.e. coalition gets7, cf. (27:5**)
1 (27:?**)).

We call a gameT essentialwhen 2 form v(S) not = O. reduced

an Hence appropriatestrategy of coalitionsis now of greatimportance. its characteristic function v(S) has a

27.4.Various Criteria.

Non-additive

Utilities

Given 27.4.1. a characteristic function v(S),we wish to have an explicit for form v(S). (Cf. above.) expression the 7 of its reduced o Now 7 is the joint value of the v((fc)), i.e.f the v((fc)) + aj,and this))

is by (27:4)

i V v((j)).
j-i

Hence
7=

(27:8)
Consequentlywe have:

i))

y-i

(27:B)

The game F is inessentialif and only

if

3-1

t v((j)) =

7 (i.e. = 0),

and it is essentialif and only if

S ;-lv((/)) <

7 (i.e. > 0).*

Fora zero-sumthree-person amewe have, with the notations of 23.1., g v((l)) = -a,v((2)) = -6,v((3)) = -c;so 7 = iA. Therefore our conin the to ceptsof essentialand inessentialspecialize those of 23.1.3. case of a zero-sumthree-person ame. Consideringthe interpretation of these g i t conceptsn both cases,his was to be expected.
but

1This is, of course,not the whole story There may be other coalitions of > 1 < n 1 players which are worth aspiring to. (If this is to happen, n 1 must n ^ 4.) This depends exceed1 by more than upon the ^(S)of the setsS with > but < n 1 elements. But only a complete and detailedtheory of games can the appraise role of thesecoalitions correctly. 1 player coalitions (the biggest Our abovecomparison of isolatedplayers and n suffices only for our present purpose:to coalitions which have anybody to oppose!) establish the importance of coalitions in this situation. 1 Cf again footnote 1 on p. 249. 8 So is the of footnote 1 on p. 248.

1, i.e.

-7

4 We have well

seenalready that one or the other must be the case,since 2}v ((/))^ as

as 7

;'-i

0.)))

AND ESSENTIALGAMES INESSENTIAL

251

27.4.2.
(27:C)

We can formulate, some other criteriaof inessentiality: The game F is inessentialif and only if its characteristic function can be given this form: v(S)

v(S)
for a suitable system a?,

tin

2) aj
5 , aJJ.

this I by Proo/: ndeed, expresses (27:2)preciselythat v(S) is in strategic i with v(S) = 0. As this v(S) is reduced,t is then the reduced equivalence
form of v(S)

and this is the meaning of inessentiality.

(27:D)

The game F is inessential if and only if its -characteristic of i.e. function v(S) has always = in (25:3:c)25.3.1.;when if Sn T = V (S u T) = v(S) + v(T)

0.

A Proof: ecessity: v(S) of the N this property. possesses

form given in

(27:C)above obviously

i.e. 25.4.1.;
v(Si u
- - -

Sufficiency: Repeatedapplication of this equation gives

u

= in (25:5)of

Sp) = v(SO +

+ v(Sp) if Si,

Sp arepairwisedisjunct.
Si =
(fci), ,

SP = (k p) give So we have

Consideran arbitrary

S,say S = (fci,

, k p). Then

v(S) = v((*0) + v(S) = 5) a

km 8

+ v((*,)).
k

erally additive were among the most important, but they offered significant difficulties to every theoretical a approach; nd one cannot say I that thesedifficulties have ever beenreally overcome. n this connection one should recallthe discussionsof conceptslike complementarity,total phase value, imputation, etc. We are now gettinginto the corresponding of our theory; and it is significant that we find additivity only in the uninfor

= v((l)), , a = v((n))and so F is inessentialby (27:C). t 27.4.3. criteria(27:C) and (27:D) expresshat the values of all Both 1 coalitions ariseadditively from those of their constituents. It will be remembered role additivity of value, or rather its frequent absence, what the has played in economiciterature. The cases which value is not genin l
with

a?

1Thereaderwill understand the quantity v(S).)))

that we

are using

the word \"value\"

(of the coalition

S)

252))

GENERAL THEORY:ZERO-SUMn-PERSONS))

g c teresting(inessential) ase,while the really significant (essential) ames

have a non-additive
will

t t make this further observation:he additive v(S) i.e.he inessential games are exactly the measurefunctions of /, which give I the total f measure zero. Thus the generalcharacteristicunctions v(S) are a new o generalization of the concept f measure.Theseremarks are in a deeper v w e o senseconnectedith the precedingnesconcerningconomicalue. How2 ever, it would lead too far to pursue this subjectfurther.

function. 1 characteristic w readers ho arefamiliar with the mathematical Those

theory of measure

27.5. he Inequalities T (27:7*),(27:7**). For 7 = clear. Assume therefore that

y

in the

Essential Case trivially

in Let 27.5.1. us return to the inequalitiesof 27.2., particular to (27:7),

> (essential case).))

(inessential ase) everything is c

Figure 50. Abscissa: number of elements of S. Dot at 0, 7, 7, or heavy line: Range of p, v possible alues v(S) for the S with the corresponding p.

s Now (27:7),(27:7*),(27:7**) et a range of possiblevalues for v(S) This range is pictured for each for every number p of elementsin - ,n = 0, 2, n n in Figure 50. 2, p We can add the following remarks: 27.5.2. It will be observedthat in an essentialgame when First: Otherwise the formulae (27:7), (27:7*), 7 > necessarily n ^ their (27:7**) or Figure50,which expresses content lead to a conflict: For n = 1 or 2 an (n l)-element S has or 1 elements, enceits set h

1,

1,

S.

3.

i.e.

1We are, of course,concerned this moment only with a particular aspectof the at of concerted of behavior acts subject:we are considering values of coalitions only and not of economicgoodsor services. Thereaderwill observe, owever, that the speh cialization is not as far reaching as it may seem goods and servicestand really for the s for a concerted of behavior. economic of their exchange act act in *Thetheory of measure reappears another connection. Cf. 41.3.3.)))

i.e.

i.e.

INESSENTIAL AND ESSENTIALGAMES v(S) must on the one hand be 7, and on the other hand or
1 impossible.

253

7, which is

Second: the smallestpossiblenumber of participantsin an essential For f game,i.e.or n = 3,the formulae (27:7),(27:7*),27:7**) or Figure50 ( determineverything:they statethe values of v(S)for 0,1, 1, ~element n n e setsS;and for n = 3 the followingareall possibleelement umbers: ,1, , n 0 2 3. (Cf. also a remark in footnote 1on p.250.) This is in harmony with the fact which we found in 23.1.3., to only according which there exists onetype of essentialzero-sumthree-personames. g Third:Forgreater umbersof participants, i.e.or n ^ 4, the problem n f assumes a new complexion.As formulae (27:7),(27:7*),(27:7**) or Figure50 show, the elementnumber of p of the set S can now have n othervalues than 0,1, 1, . I.e.he interval n t 2gp^ n 2 (27:9)

2 becomes available. It is in this interval that the above formulae no a longer determine unique value of v(S); they set for it only the interval

now

(27:7) -P7 ^ v(S) S (n p)7, the length of which is ny for every p (cf.again Figure50). In 27.6.3. this connection the question may be asked whether really the entireinterval (27:7)is available, i.e. hether it cannot be narrowed w further by some new, more elaborate considerations concerning v(S). The answer is:No. It is actually possibleto define for every n 4 a v single game F p in which, for each p of (27:9), (5) assumesboth values sets p7 and (n p)y for suitablep-element S. It may sufficeto mention the subjectherewithout further elaboration. To sum up:The realramifications of the theory of gamesappear only when n ^ 4 is reached. (Cf. footnote 1 on p. 250, where the same idea was expounded.) 27.6. ector Operationson CharacteristicFunctions V someremarksof a more formal nature In 27.6.1. concludingthis section

seem appropriate.

in 25.3.1., the The conditions (25:3:a)-(25:3:c) which describe characteristic function v(S), have a certainvectorial character: they allow ana1., loguesof the vector operations,defined in 16.2.of scalarmultiplication, and of vector addition. More precisely: funcScalarmultiplication:Given a constant t ^ and a characteristic function. Vector tion v(S), then tv(S) = u(S) is also a characteristic
3 functions v(S),w(S); then G addition: iven two characteristic

sum two-person

1Of course,n a zero-sum one-persongame nothing happens at all, and for the zeroi games we have a theory in which no coalitions appear. Hencethe inessentiality of all these casesis to be expected. s It has n 3 elements;and this number is positive as soon as n ^ 4. 8 Everything here must refer to the same n and to the same set of players / (1,2, - - - ,

n).)))

254
from

GENERAL THEORY:ZERO-SUMn-PERSONS

function. The only difference v(S) + w(S) z(S) is also a characteristic is the corresponding definitions of 16.2.that we had to requiret S> O.1-2

27.6.2. The

two operations defined above allow immediate practical

3 constituent games. 27.6.3. do not proposeto enter upon a systematicinvestigation of We o theseoperations,i.e. f their influence upon the strategicsituations in the gameswhich they affect. It may be useful, however, to makesomeremarks on this subject without attempting in any way to be exhaustive. We observefirst that combinations the operationsof scalar ultiplicaof m tion and vector addition also can now be interpreted directly. Thus the

multiplication by the factor t. Vector addition:This correspondsto the superpositionof the games correspondingto v(S) and to w(S). One would imagine that the same players 2, , n are playing these two games simultaneously,but no move made in one game is supposedto influence independently. the othergame,as far as the rules areconcerned. n this case characthe I teristic function of the combinedgame is clearlythe sum of those of the two

interpretation: t If Scalarmultiplication: t = 0, then this producesu(/S)as 0, i.e.he in So eventlessgame considered 27.3.1. we may assume t > 0. In this caseour operationamounts to a changeof the unit of utility, namely to its

1,

I.e.,

characteristic function

(27:10)
y

z(S) = tv(S) + w(5)

gravity in the senseof

belongsto the game which arisesby superpositionof the gamesof v(S) and w(S)if theirunits of utility arefirst multipliedby t and s respectively. If s = 1 t then (27:10) correspondsto the formation of the centerof

as very involved consequences regards strategy. We observenextthat therearesomecases here our operationshave no w
in consequences strategy. First,the scalar ultiplication by a t m unit, has no such consequences.

in (16:A:c)16.2.1. It will appear from the discussionin 35.3.4. in particular footnote 1 (cf. on p.304 below) that even this seeminglyelementaryoperation can have

> alone,being a merechangein

t < would upset Note that a multiplication of the in original 3C*(n, , r n ) with a t < would be perfectly feasible. It is simplest to considera multiplication by t a change in sign. But a change of sign of the 3C*(Ti, , doesnot at all correspondto a changeof sign of the v(S). This should be clearby common sense,as a reversalof gains and losses modifies all strategic considerations in a very involved way. are (This reversal and someof its consequences familiar to chessplayers.) A formal corroboration of our assertion may be found by inspecting the definitions of * Vector spaceswith this restriction of scalar multiplication are sometimes called positive vector spaces. We do not needto enter upon their systematic theory. 3 This should be intuitively obvious. An exact verification with the help of involves a somewhat cumbersome notation, but no real difficulties.)))

indeed,

(25:3:c) 25.3.1.

r)

1,i.e.

25.1.3.

25.1.3.

S GROUPS, YMMETRY AND FAIRNESS

255

Second and this is of greatersignificance the strategicequivalence discussedin 27.1. a superposition: pass from the gameof v(S) to the is we strategically equivalent gameof v'(S) by superposingon the former an 1 inessentialgame. (Cf. (27:1) (27:2)n 27.1.1. concerning i and inessenand, and this W tiality, 27.3.1. (27:C)in 27.4.2.) e may express in the following way: we know that an inessentialgame is one in which coalitionsplay no role. The superpositionof such a gameon another one does not disturb i strategicequivalence,i.e.t leaves the strategicstructure of that game
unaffected. a Game Let us now considerthe roleof symmetry, or more generally,the n or their numbers in an effects of interchanging the players of gameF. This will naturally bean extension the corresponding n-person for the zero-sumtwo-persongame. study made in This analysisbegins with what is in the main a repetition of the steps taken in for n = 2. But since the interchanges f the symbols o n than for n = 2, it is , n offerfor a general many more possibilities that indicated we should go about it somewhatmore systematically. Considerthe n symbols , n. Form any permutation P of these by , n, into which p symbols. P is described stating for every = = i (also , n), P carriest. So we write:

28.1. Permutations,

28.Groups,Symmetry and Fairness

Their Groups, and Their Effect

28.1.1.

1,

17.11. 17.11.

!,-,
p

on

1,

1,

i 1,

(28:1)

P:i-*i ,

or by way of complete enumeration:))

the permutationssomedeservespecial mention: The identity In which leaves every i(= 1, (28:A:a) changed:
Among

, n) un-

i - iln = i.

(28:A:b)

Given two permutations P, Q, their product PQ, which consistsin carrying out first P and then Q:))

1With

the

characteristicfunction w(S)
v'OSf)

\\i,

fcinS
vGSf)

2}

<**

then

in

our

above notatibns

+w(S)
2* (1 I/
2\\

The identity (cf.below) is /

- (!'' ' ' ' ', *\\ z,

n/)))

256

GENERAL THEORY:ZERO-SUMn-PERSONS
n\\

The number of all possiblepermutationsis the factorial of n r

=1

...

n,

and they form togetherthe symmetric group of permutations 2n subsystemG of 2n which fulfills thesetwo conditions: /n belongsto G, (28:A:a*) PQ belongsto G if P and Q do, (28:A:b*)
1 is a group of permutations.
A

Any

subset S

permutation
p *

P carriesvery subset S of / = (!, e

, n) into another

Perform a permutation P on the symbols , n denotingthe players k = n by k p instead of fc; this transd of F. I.e. enotethe player , of forms the game F into another game F p The replacement F by Fp must in two respects: the influence which eachplayer in make its influence felt i on exercises the courseof the play, i.e.n the index k of the variable r k and of i which eachplayer chooses; in the outcome the play for him, i.e.n this. the index k of the function 3C* which expresses 8 So Tp is again in the normalized form, with functions 3C(ri, , r n ), k = , n. In 3C(ri, , r n) by means of 3C*(ri,p expressing , r n ), we must remember:the player k in F had 3C*; now he is k p in F , so he has3C>. If we form 3C* with the variables TI, the , r n , then we express outcome of the r game Tp when the player whosedesignationin F p is k chooses*. So the T player k in F who is k p in Tp chooses*/>. So the variablesin 3C* must be , r n p. We have therefore: TIP,

these generaland preparatory remarks we now proceed t to apply their conceptso an arbitrary n-persongame F.

After 28.1.2.

1,

1,

1,

the important and extensive theory of groups compare L. C. Mathewson: Elementary Theory of Finite Groups, Boston 1930;W. Burnside: Theory of Groups of Finite Order,2nd Ed. Cambridge A. Speiser: Theorie Gruppen von endlicher der Ordnung, 3rd Edit. Berlin 1937. We shall not needany particular results or concepts f group theory, and mention o the above literature only for the use of the readerwho may want to acquire a deeper insight into that subject. Although we do not wish to tie up our exposition with the intricaciesof group theory, we nevertheless introduced someof its basictermini for this reason a real understanding of the nature and structure of symmetry is not possible without some familiarity with the who may want to (at least)the elements of group theory. We want to prepare reader in proceed this direction, by using the correctterminology. For a fuller exposition of the relationship between symmetry and group theory, cf. H. Weyl: Symmetry, Journ. Washington Acad. of Sciences, ol. XXVIII (1938), p. V p 253ff. *If then p kp (* , (ibf, , 8 Cf the similar situation for n = 2 in footnote 1 on p. 4 The readerwill observe that the superscript for the index k of the functions 3C themselves appears the left-hand side,while the superscript P for the indices k of the on variables r* appearon the right-hand side. This is the correctarrangement; and the to was needed establish it. argument preceding (28:3) Theimportance of getting this point faultless and clearliesin the fact that we could)))

(28:3))) 1For

1911;

*,),

).

109.

S 257 GROUPS, YMMETRY AND FAIRNESS Denote the characteristic functions of T and Tp by v(S) and v p (S) the respectively. Since players,who form in Tp the setSp, arethe sameones
who form in T the

(28:4)
invariant

setS,we have v p(Sp ) = v(5)

with

for every

1 S.

28.1.3. a particularP) T coincidesith If (for w

or symmetric

expressed by
(28:5)
When

t respecto
, r n)

P.

Tp, then we say that T is By virtue of (28:3)this is

, T|| ). -

3Mn,

3C*(nP,

this is the case,hen (28:4) t becomes

(28:6)

v(8p) = v(S)

for every

S.

group of F.

Given any T, we can form the systemGr of all P with respect which T to is symmetric. It is clearrom (28:A:a),28:A:b)bove, that the identity 7n f a ( belongsto Gr, and that if P, Q belongto Gr , then their productPQ doestoo. SoGr is a group by (28:A:a*), a (28:A:b*)bove. We call Gr the invariance

Observethat (28:6)can now be stated in this form:

(28:7) v(S) = v(r)

thereexists P in Gr with Sp = T, a i.e. hich carries into T. w S

if

The size of Gr i.e.he number of its elements gives somesort of a t measure of \"how symmetric \" F is. If every permutation P (otherthan identity 7n ) changes , then Gr consistsof 7n alone, r is totally unsymmetric. T If no permutation P changesT, then Gr contains all P, i.e.t is the symi metric group Sn , F is totally symmetric. Thereare,of course, umerous n intermediate cases between thesetwo extremes, the precisetructure and s of T'ssymmetry (or lack of it) is disclosedby the group Gr . The 28.1.4. conditionafter (28:7)impliesthat S and T have the same number of elements.The converse implication, however, need not be if true if Gr is small enough, i.e. F is unsymmetric enough. It is therefore
not otherwise

order) to

be sure that successivepplications of the superscripts P and Q (in this a give the same result as a (single) application of the superscript PQ to r. in The readermay find the verification of this a good exercise handling the calculus of

r will

permutations. For n

2 and P J , application of P on either sidehad the sameeffect,soit is Cf footnote 1 on p. to not necessary be exhaustive on this point. F 6 In the zero-sum two-person game, 3C 3C B 3C , and similarly 3C t 2 3Cf = 3Cf p (n, n) -3C(ri, ). 3C b Hence this case(cf. above,n 2 and P \\\\ ^ J (28:3) ecomes in n and This is in accordwith the formulae of But this simplification is possible only in the zero-sum two-person game; in all

I'

109.

14.6. 17.11.2.

'j

alone. must rely upon the general formula (28:3) than a computational one, which could The latter, however, would causeno difficulties be basedon the formulae of either, only more extensive notations.)))
other

1This conceptualproof is clearer simpler and

caseswe

25.1.3.

258

GENERAL THEORY:ZERO-SUMn-PERSONS

those groups G = Gr which permit this converse of interest to consider implication, for which the following is true:

i.e.

(28:8)

If S, T have the same number of elements,hen there exists t which carries into T. a P in G with Sp = T, S

i.e.

This condition (28:8)is obviously satisfiedwhen G is the symmetricgroup f Sn , i.e.or the G = Gr = Sn of a totally symmetricF. It is alsosatisfied 1 f than total symmetry. for certainsmaller groups, i.e.or certainF of less

28.2. Symmetry
from

and

Fairness

At 28.2.1.any rate,whenever (28:8)holdsfor G = Gr, we can conclude

(28:7):

(28:9)
That

i v(S)dependsonly upon the number of elementsn

S.
,

is:
v(S) = v p i where p is the number of elementsn
/S, (p

(28:10)

in Consider the conditions (25:3:a)-(25:3:c) which give an of all characteristic functions v(S). It is easy to exhaustive description holds. They become: rewrite them for v p when (28:10)

25.3.1.,

0,1,

ri).

+ q g n. o is (27:3)in 27.1.4. clearly a consequencef (28:10) of (28:9)), (i.e. so that such a v(S) is automatically reduced, with 7 = VL We have i i i.e. therefore,n particular, (27:7),(27:7*), 27:7**)n 27.2., the conditions ( of Figure50.
v p+q

(28:ll:a) (28:ll:b) (28:ll:c)

vp

v v n _p vq

= 0, = -vp,
for

to that of

can Condition (28:ll:c) be rewritten, by a procedure which is parallel

(25: A)

in

25.4.2.

so cf.several recedingreferences); G = 2n is the only possibility of any symmetry. p Considerherefore n 3, and call G set-transitive if it fulfills (28:8).The question, t which G T* Sn are then set-transitive, is of a certain group-theoretical interest, but we ourselveswith it in this work. neednot concern For the readerwho is interested in group theory we neverthelessmention: Thereexists a subgroup of Sn which contains half of its elements (i.e.n!), nown as i k the alternating group O. This group is of great importance in group theory and has been there. For n ^ 3 it is easilyseento be set-transitive too. extensively discussed So the real question is this: for which n 3 do there exist set-transitive groups
It is easyto show that for n = 3, 4 none exist. For n ** 5, 6 such groups do exist. 5 a set-transitive group G with 20 elements exists, while S6, as have 120,60 elements respectively. For n = 6 a set-transitive group G with 120elements exists, while Se,tte have 720, 360elements respectively.) For n 7, 8 rather elaborate grouptheoretical arguments show that no such groups exist. For n 9 the question is still open. It seemsprobable that no such groups exist for any n > 9, but this assertionhas not yet beenestablished for all these
(For n
n.)))

2' (1

2\\
1

* S;a?

<

S GROUPS, YMMETRY AND FAIRNESS

Put r

269

=n

follows:

q\\

then

(28:ll:b) us to permits

state (28:ll:c) as

if (28:ll:c*) v p + v, + v r ^ p + q + r = n. is symmetric with respecto p, g, r;1 hencewe may make t Now (28:ll:c*) w ^ q ^ r by an appropriate permutation. Furthermore, hen p = p r-= n - g), then (28:11 follows from (28:11 (28:11 (hence :c*) :a), :b) (even with =). Thus we may assume p 7*.o. So we need to require only (28:ll:c*) for 1 g p ^ g g r, and therefore the same is true for Observefinally that, as r = n p q, the inequality g ^ r (28:ll:c). this: means p + 2q ^ n. We restate only (28:12) It sufficesto require (28:ll:c) when

^ P^

q,

p + 2g ^ n.2

clear This becomes when we considerit in the simplestpossiblespecialcase:

for n that This means in the terminology of 17.11.2., the game T is fair. We The extendthis concept: n-persongame F is fair when its characteristic i.e. function v(S) fulfills (28:9), when it is a v p of (28:10). Now, as in this notion 17.11.2., of fairness of the game embodieswhat is really essential o in the concept f symmetry. It must be remembered, however, that the

The of function is a conse28.2.2. property (28:10) the characteristic quenceof symmetry, but this property is also important in its own right.

= 2. 8 for vanishes. simply Indeed, n = 2 (28:10) meansthat the v' of 17.8.1.

similarly that of total symmetry of the game t may or may not imply that all individual players can expecthe same fate in an individual play (provided that they play well). For n = 2 this for the former, and implication did hold, but not for n ^ 3 (Cf. footnotes and 2 on 225 for the latter.) or in We observe,finally, that by (27:7), 27:7*), 27:7**) ( ( gamesaresymmetricand hence fair, when n = 3, by Figure50, all reduced in but not when n ^ 4. (Cf.the discussion 27.5. .) Now the unrestricted 2 zero-sum n-person game is brought into its reducedform by the fixed extrapayments i, , a n (to the players , n, respectively),as Thus the unfairness of a zero-sum three-personame in described g what is really effective in its asymmetry is exhaustively expressed by these i, 2, 8 ; that is, by fixed, definite payments. (Cf. also the

conceptof fairness and

28.2.3.

p.

17.11.2.

27.2.,

i.e.
1

27.1.
7

1,

\"basic values/

I a',6',c' of 22.3.4.)n a zero-sumn-person game with

Both in its assertion and in its hypothesis! 1These the inequalities replace original p + q ^ n; they are obviously much As they imply 3p ^ p + 2g ^ n and 1 -f 2q ^ p -f 2g ^ n, we have 1 n n q P^

stronger.

3'

* ^~\"

8 v((2)). For n 2 the only essentialassertion of v((l)) By definition v' is v((l)) v((2)). Dueto the above, this means (28:9)(which is equivalent to (28:10)) that -v',i.e. v' preciselythat v'

0.)))

260

GENERAL THEORY:ZERO-SUMn-PERSONS

n ^ 4, this is no longer always possible,sincethe reducedform need i not be fair. That is, there may exist,n such a game, much more fundamental differences between the strategicpositions of the players, which an , cannot be expressed the by by fixed, definite payments. This will becomemply clearin the courseof Chapter VII. In the same a it connection is also useful to recallootnote 1on p.250. f

!,,

i.e.

We are now prepared for the main undertaking: o formulate T l of the principles the theory of the zero-sum n-persongame. Thecharacteristic function v(S),which we have defined in the precedingections, s provides the necessarytool for this operation. w Our procedure ill be the sameas before: e must select specialcase W a to serve as a basis for further investigation. This shall be one which we and which we neverthelessdeem sufficiently charachave already settled teristicfor the generalcase. By analyzing the (partial) solution found in this special w case, e shall then try to crystallizethe rules which should it and govern the generalcase. After what we said in 4.3.3. in 25.2.2., ought to beplausiblethat the zero-sum three-personame will be the special g

29.1.1.

29.Reconsideration the Zero-sum of Game Three-person 29.1. Qualitative Discussion

argument by which our present solution of the zero-sum three-person game was obtained. Clearly the essential asewill be the one of interest. We know now that we may as c 2 well considert in its reduced i form, and that we may also choose = I. y The characteristic function in this caseis completelydetermined, as discussedin the second caseof 27.5.2.:))

casein question. 29.1.2. us therefore reconsiderhe Let t

(29:1)

v(5)))

-1 1

when

S has))

elements.))
3))

it with

above.
4 In

1Of coursethe general n-person game will still remain, but we shall be able to solve the help of the zero-sum games. Thegreateststepis the present one the passage to the zero-sum n-person games. Cf. 27.1.4. 27.3.2. and 1In the notation of this means a b c The general parts of the discussionsreferred to were those in 23. The above specialization takes us caseof So our considerations of 27.1. actually back to the earlier(more special) (on strategic equivalenceand reduction) have actually this effect in the zero-sum threeperson games:they cany the general caseback into the precedingspecial ne, as stated o

conclusions: Three coalitions may form, and accordingly the three players will finish the play with the following results:

We saw that in this game everything is decided the (two-person) by 4 coalitions which form, and our discussions produced the following main

23.1.1.

22.2., 22.1. 22.3.,

just elaborations

1.

22.2.3.; 22.2.2., but theseare really

of those in

22.1.3.))) 22.1.2.,

RECONSIDERATION))

261))

(1,2))) (1,3))) (2,3)))

-1)) -1))
Figure

51.

This \" solution \" callsfor interpretation,and the followingremarkssuggest 1 themselvesin particular:

29.1.3.

(29:A:a)

The three distributions specifiedabove correspondto all of strategic possibilities the game. None of them can be considered solution in itself; it is the a (29:A:b) of all three and their relationshipto eachother which system really constitute the solution. The three distributions possesstogether,in particular, a (29:A:c)

\"stability\" to which we have referred thus far only very sketchily. Indeed no equilibrium can be found outside of thesethree distributions;and so one should expecthat any t kind of negotiation between the players must always in fine lead to one of thesedistributions. (29:A:d) Again it is conspicuousthat this \"stability\" is only a characteristicf all three distributions viewed together. No o one of them possessesalone; achone,taken by itself, could it e be circumvented if a different coalition pattern should spread to the necessarymajority of players.

heuristic principleswhich lead us to the solutions of Figure 51,always keepingin mind the remarks (29:A:a)-(29:A:d). A more precise statement of the intuitively recognizable \"stability\" of the systemof threedistributionsin Figure51 which shouldbe a concise r summary of the -discussionseferred to in footnote 4 on p. 260 leads us 2 back to a position already taken in the earlier, ualitative discussions. q It can be put as follows:

29.1.4. We

now

to proceed search for

an

exactformulation

of the

(29:B:a)

If any otherscheme distribution should be offered for of considerationto the three players, then it will meet with

4.4. and 4.6.2.))) L

1These remarks take up again the considerations of 4.3.3. connection with In half t may (29;A:d)he second of 4.6.2. alsobe recalled. * These viewpoints permeate all of 4.4.-4.G., they appear more specifically in but

262

GENERAL THEORY:ZERO-SUMn-PERSONS
for rejection the following

players1 prefer, in their own interest,at leastone of the disa o tributions of the solution (i.e. f Figure 51), nd are con2 vinced or ean be convinced of the possibilityof obtaining the advantagesof that distribution. (29:B:b) If, however, one of the distributions of the solution is offered, then no such group of players can be found.
We proceed discuss the merits of this heuristic principlein to exactway.

reason:a

sufficient number of

a more

I.e. betweenthe players 1,2,3.

ft

29.2. Quantitative 29.2.1. that 0i, 2, 0s is Suppose

j8>

Discussion a possible method of distribution

+ + 0s = 0.

distribution with ft

Then,sinceby definition v((i))(= 1) is the amount that player can get o for himself (irrespectivef what all others do), he will certainly block any

< v((z)).

We

assumeaccordinglythat))

We may

so permute the players 1,2,3 that

Pi ^ A g;

it t the higher amount Besides, is clearhat they can getthe advantage of o that distribution (irrespective f what the third player does),sincethe amounts which it assignsto them do not exceed v((2,3)) = together on the other hand, /32 ^ then a fortiori 0i ^ Since 8 ^ If, this is possibleonly when /3i = 2 = 0a = when we have the first distribution of Figure footnote 3 above.) (Cf.

ft. the Now assume 2 < i- Then a fortiori # 3 < . Consequently players where they both get will both prefer the last distribution of Figure51, 2,3 4

i.

i,i

i,

51.

i,

1,i.e.

1. 1,

4.4.3. Our discussion which follows will make it perfectly clear. 3 Since made an we the last distribution unspecified permutation of the players 1,2,3 of Fig. 51 really stands for all three. 4 Observe that eachone of thesetwo players profits by such a change separately and It would not suffice to have only the totality (of thesetwo) profit. Cf., individually. the first distribution of Fig. 51 with the second;he players t asa totality would and neverthelessthe first distribution profit by the changefrom the former to the latter, is just as good a constituent of the solution as any other. In this particular change, player 3 would actually profit (getting instead of and for player 1 the changeis indifferent (getting in both cases). Neverthelessplayer 1 will not act unless further compensations are made and thesecan be disregardedin this connection. For a more careful discussion of this point, cf. the last part of this section.)))

1Of course,n this case,two. i 1What this \"convincing\" in means was discussed

e.g.,

1,3

1),

EXACT FORM OF THE GENERAL DEFINITIONS 263

in immediate: eachof the three distributionsof Figure51there is,to be

This establishes (29:B:a) the end of at

29.1.4. loc. (29:B:b) cit.is

sure,one playerwho is desirousof improving his standing,1but sincethereis only one,he is not able to do so. Neither of his two possiblepartnersgains anything by forsaking his present ally and joining the dissatisfiedplayer: already eachgets and they can getno more in any alternative distribution 2 of Figure51. 29.2.2. point may be clarified further by someheuristicelaboration. This

i,

that the dissatisfied We see player finds no one who desires spontaneously to be his partner, and he can offer no positive inducementto anyone to m join him; certainly none by offering to concede ore than from the proceeds of their future coalition. The reasonfor regardingsuch an offer as ineffective can be expressed two ways: on purely formal groundsthis offer may in be excluded because correspondsto a distribution which is outside the it scheme Figure the real subjectivemotive for which any prospective of a partner would considerit unwise3 to accept coalition undersuchconditions is most likely the fear of subsequentdisadvantage, there may be further negotiationsprecedingthe formation of a coalition, in which he would be found in a particularly vulnerable position. (Cf. the analysis in

51;

Sothereis no way for the dissatisfied playerto overcome the indifference of the two possiblepartners. We stress: is, on the side of the two there no positivemotive against a change into another distribupossiblepartners tion of Figure51, ut just the indifference characteristic certaintypes of b of
stability.4

22.1.3.)

22.1.2.,

30.The Exact Form of the General Definitions The 30.1. Definitions

to the caseof the zero-sumn-person game n. Let the characteristic function of T be v(S). general
We We 30.1.1. return F
with with In accordance the suggestionsof the preceding paragraphs we

to proceed give the decisivedefinitions. a

'thefollowing properties a, ^ v((0) (30:1)

mean by a distribution or imputation

setof n numbers

i,

with

for

i = 1,

- - , n,

(30:2)
1The one who gets 1Thereadermay find

t a good exerciseo repeatthis discussion with a general (not with reduced) (S), i.e. generala, 6, c, and the quantities of 22.3.4.The result is the v same;it cannot be otherwise, sinceour theory of strategic equivalence and reduction is correct. (Cf.footnote 3 on p. 260.) 8 Or unsound, or unethical. 4 At every change from one distribution of Fig. 51to another, one player is definitely for it; and so the remaining player blocks the change by his against, one definitely
indifference.)))

1. it

i-l

a, = 0.

264

GENERAL THEORY:ZERO-SUMn-PERSONS as vectors in the

It may be convenient to view thesesystems L 2.:)) l n-dimensionalinear space n in the senseof 16.1
A

!,-,
ft

tion

a setS (i.e.subset of / = a , if

1,
\"*

, n)

is called effective for the imputa81

(30:3)
An

imputation a. dominates anotherimputation

iinS

* vO )-

, in symbols))

if

a thereexists set S with the following properties: (30:4:a) S is not empty, (30:4:b) S is effective for a ,

(30:4:c)
A

<*

> ft
ft

for all in

i S.

: the setV of imputations is a solutionif it possesses followingproperties

(30:5:a) No in V is dominatedby an a in V, (30:5:b) Every ft not in V is dominatedby some a in V. and can (30:5:a) (30:5:b) be stated as a singlecondition:

(30:5:c)

The elements V are preciselythoseimputations which)) of . 4 ^ o areundominatedby any element f V.))

_ _

of the earlier iscussions 4.4.3. d

when we recallhe considerations the preceding t of paragraphs and also of

30.1.2. meaning of these definitions can, of course,be visualized The

(Cf.footnote 1on p.40.)

Tobegin with, our distributionsor imputationscorrespondo the more t intuitive notionsof the same name in the two placeseferredto. What we r call an effective setis nothing but the playerswho \"areconvincedor canbe

convinced\"of the possibilityof obtaining what they areoffered by a ; cf. a in The in again 4.4.3nd (29:B:a)29.1.4. condition(30:4:c)the definition of domination expresses all these players have a positive motive for that > > preferring a to ft . It is therefore apparent that we have defined domination entirely in the spirit of 4.4.1., of the preference escribedy and d b

The definition of a solutionagrees ompletelywith c as well as with (29:B:a), in (29:B:b)29.1.4.)))

in (29:B:a)29.1.4.

that given in

4.5.3.,

EXACT FORM OF THE GENERAL DEFINITIONS 265

30.2. Discussionnd a
t placeso
which we

Recapitulation

further in 4.6. n a qualitative basis. We can now reformulate o elaborated those ideas in a rigorousway, consideringthe exactcharacter hich the w 1 has discussion now assumed. Theremarkswe wish to make arethese:

referred in detail in the courseof the last paragraph. We shall neverthelessre-emphasize some of their nlain features particuo larly the concept f a solution. We have already seenin 4.6.hat our concept f a solution of a game t o correspondsprecisely to that of a \" standard of behavior\" of everyday which parlance.Our conditions (30:5:a), (30:5:b), correspond to the conditions (4:A:a),(4:A:b)of 4.5.3., expressust the kind of \"innerstaj which is expectedf workablestandards of behavior. This was o bility\"

30.2.1. motivation for all these definitions has been given at The

the

30.2.2.
(30:A:a)
Consider a solution V. We have not excluded an for of imputation ft in V the existence an outside imputation a ' H ft .2 If such an a ' exists, attitude the (not in V) with a I of the players must be imagined like this:f the solution V \" this (i.e. systemof imputations) is \" acceptedby the players1, , n, then it must impressupon their minds the idea that

'

only the imputations

An

a ' not in V with a H ft , although preferable to an effective set of players, will fail to attractthem, because is it of d \"unsound.\" (Cf.our detailediscussion the zero-sum three-

ft

in V

'

are\" sound\"ways of distribution.

as for e persongame, specially to the reason the aversion of each m amount in a coaliplayer to accept ore than the determined and tion. Cf. the end of 29.2. its references.)Theview of the
\"
>

\" unsoundness of a ' may alsobe supportedby the existence

of an

in V

with

>

arguments are, of course, circular in a sense and again of depend on the selection V as a \"standard of behavior,\" i.e. of B as a criterion \"soundness.\" ut this sort of circularity is not unfamiliar in everyday considerations dealing with

a (cf.(30:A:b) elow). All these b

'

\"soundness.\"

4.6.3.
a

presentation

irThe remarks (30:A:a)-(30:A:d) which follow are a more elaborateand precise bears the same relationship to Remark (30:A:e) of the ideas of

4.6.2.

'H

1Indeed,we
>

shall

that seein (31:M)of 31.2.3.

in inessential
games.)))

an imputation

for which never

, exists only

266

GENERAL THEORY:ZERO-SUMn-PERSONS
If the players 1, the , n have accepted solutionV as a 1 standard of behavior/ then the ability to discreditwith the o help of V (i.e.f its elements) ny imputation not in Vi is a f necessaryin orderto maintain their faith in V. Indeed,or a every outside a (not in V) theremust exist n a in V with a H a (This was our postulate (30:5:b).) f Finally there must be no inner contradictionin V> i.e.or
\"

(30:A:b)

(30:A:c) (30:A:d)

'.

'
if

t domination, i.e.he relation H , weretransio tive, then the requirements (30:A:b)and (30:A:c)(i.e. ur and would excludethe rather postulates (30:5:a) (30:5:b)) I delicateituation in (30:A:a). s Specifically: n the situation of
Observethat

(30:5:a).)

a,

ft

in

V, never a

ft

(This was our other postulate

(30:A:e)

(30:A:a),ft belongsto V, a doesnot, and a *- ft . By t an (30:A:b)here exists a in V so that a H a Now if t domination were transitive we could concludehat a H ft , which contradicts30:A:c)ince a , ft both belongto Vs ( t The above considerations makeit even more clearhat only

'

'

'.

its entirety is a solutionand possesses kind of stability any but none of its elements individually. The circularcharin m acterstressed (30:A:a) akes it plausiblealso that several solutionsV may existfor the same game. several stable standards of behavior may exist or the same f factual situation. and Eachof thesewould, of course, e stable consistentin itself, b and but in conflict with all others. (Cf.also the end of
V in

I.e.

the end of 4.7.)

4.6.3.

In many subsequent discussionswe shall seethat this multiplicity of solutionsis, indeed, a very general henomenon. p
more formalistic nature. So far we have paid attention mainly to the w but meaning and motivation of the concepts hich we have introduced, the notion of solution, as defined above, possesses formal features which some

30.3.1.seems ppropriate to insert at this point someremarks of a It a

30.3. Concept f Saturation The o

deserveattention.

The formal logical considerationswhich follow will be of no immediate use,and we shall not dwell upon them extensively,continuing afterwards more in the vein of the precedingreatment. Neverthelesswe deem t that theseremarksareuseful herefor a more complete understandingof the structureof our theory. Furthermore, procedures be used herewill to the have an important technical pplicationin an entirely different connection a
in

51.1.-51.4.)))

EXACT FORM OF THE GENERAL DEFINITIONS 267

validity of (R between two elements , y of D is x the formula x&y.1 (R is defined by a statementspecif ing y unambiguously for which pairsx,y of D, xfay is true,and for which it is not. If xGiy is equivalent to 2/(Rrc, then we say that xfay is symmetric. Forany relation (R we can define a new relation (R 5 by specifying 8y to mean the x(R conjunction of xfay and yfax. Clearly(R 5 is always symmetricand coincides with (R if and only if (R is symmetric. We call (R a the symmetrized form of (R.2 relation x&y

30.3.2. a Consider domain (set)D for theelements ,y of which a certain x

exists. The

expressed by

for all x, y of A. (30:B:b) A subset A of D and an elementy of D are (Si-compatible if and only if x(Ry holds for all x ot A. From theseone concludes immediately: i (30:C:a) A subset A of D is (Si-satisfactoryf and only if this is true: The y which are(R-compatiblewith A form a superset f A. o We define next: (30:C:b) A subset A of D is (Si-saturatedif and only if this is true: They which are(R-compatiblewith A form preciselythe set A

define: if (30:B:a) A subset A of D is (Si-satisfactory and only

We now

if x(Ri/

holds

Thus the requirementwhich must be added to (30:C in order to secure :a)

is this: (30:C:b)

(30:D)

If

thereexists an x in

is not in

A,
A

then it is not (R-compatible with A; such that not x(Ry.

i.e.

Consequently(R-saturation may be equivalently defined by

further, we give some The verification of the assertionsmade in them is easy and will examples. be left to the reader. First:Let D be any set and x&y the relation x = y. Then (R-satisfactorinessof A means that A is eitherempty or a one-element while set, set. (R-saturation of A means that A is a one-element Let Second: D be a set of real numbers and x(Siy the relation x ^ t/. 3 Then (R-satisfactorinessof A means the same thing as above,4 while (R-saturation of A means that A is a one-element consistingof the greatest set, no elementof D. Thus there exists such A if D has no greatestelement 1It is sometimes more convenient to use a formula of the form (R(or, y), but for our purposes x&y is preferable. 2 Someexamples: D consistof ail real numbers. The relations x Let y and x 9* y are symmetric. None of the four relations x y, x ^ y, x < y, x > y symmetric. The symmetrized form of the two former is x y (conjunction of x ^ y and x y),
\\R

(30:D).

and (30:B:a)

Before 30.3.3. we investigate theseconceptsany

65.4.1. 4
Cf

the symmetrized form of the two latter is an absurdity 8 D couldbe any other set in which such a relation

is defined, cf

(conjunction

.the secondexampley).
in

of x

< y and x >

.footnote

1 on

p.

268.)))

268

GENERAL THEORY:ZERO-SUMn-PERSONS
if

that Third:Let D bethe plane and xfay express the points #, y have the same height (ordinate). Then (R-satisfactoriness of A means that all l points of A have the same height, i.e.ie on one parallel to the axis of abscissae.(R-saturation means that A is preciselya line parallelto the axis

f (e.g.or the set of all realnumbers) and A is unique when it is finite). element(e.g.

D has a greatest

same thing as (R a-satisfactoriness. Thus satisfactorinessis a conceptwhich needbe studied only on sym-

Fourth:Let D bethe setof all imputations,and x<S(.y the negation of the with of domination x H y. Then comparison our (30:B:a), (30:D) (30:5:a), or equallyof (30:C:b) (30:5:c) shows:R-saturawith id. ( in (30:5:b) tion of A means that A is a solution. suffices to seethat satisOne 30*3.4. look at the condition (30:B:a) as factorinessfor the relationx<S(y is the same for the relation yfax and so also for their conjunction x(&8y. In other words: (R-satisfactorinessis the

of abscissae.

30.1.1.,

metric relations. :B Thisisdue to the x, y symmetricform of the definitory condition (30 :a) does The equivalent condition (30:C:a) not exhibit this symmetry, but of this course does not invalidate the proof. for Now the definitory condition (30:C:b) (R-saturation is very similar It in structure to (30:C:a). is equally asymmetric. However, while an this possesses equivalent symmetricform (30:B:a), is not the (30:C:a) The casefor (30:C:b). correspondingequivalent form for (30:C:b) as is, and we know, the conjunction of (30:B:a) (30:D) and (30:D) is not at all

So we see: (30:E)
(R

symmetric.

I.e.(30:D) is essentially alteredif x(Rt/

While
(R5,

is replacedy b

y(Rx.

(R-satisfactorinessin unaffected by the replacement of doesnot appear that this is the case (R-saturation. for by Condition (30:B:a) (amounting to (R-satisfactoriness) is the same for Condition(30:D) for (R s is impliedby the same for (R since(R 5 (R and (R 5 implies(R. So we see: s (R -saturationis impliedby (R-saturation. (30:F) The difference betweenthesetwo types of saturation referred to above it o is a realone: is easy to give an explicit examplef a set which is (R s-sat1 urated without being(R-saturated. to Thus the study of saturation cannot berestricted symmetricrelations. For symmetric relations (R the nature of saturation is simple complicationswe assume for this enough. In orderto avoid extraneous 2 that section x(Rx is always true. 1E.g.: first two examplesof 30.3.3.in the relation of (R 5 and (R to eachother The are of (cf.footnote 2 on p. 267);their concepts satisfactorinessare identical, but those of
it

30.3.5.

saturation

* This is clearly the sincenever x H

x.)))

differ.

casefor

our

decisivexample of 30.3.3.:the e x&y

negation

of *

H y

EXACT FORM OF THE GENERAL DEFINITIONS 269

Now we prove:

(30:G)

Let (R be symmetric. Then the (R-saturation of A is equivalent to its being maximal (R-satisfactory. I.e.t is equivalent i to:A is (R-satisfactory,but no proper supersetof A is.

togetherwith condition (30:D). So we need only prove:If A is (R-satisfactory, then (30:D)is equivalent to the non-(R-satisfactorinessof all proper of supersets A.
I Sufficiency of (30:D):f B D A is (R-satisfactory, then any y in B,but not in A, violates (30 1 :D). Considera y which violates (30:D). hen T Necessityof (30:D):

c Proof:(R-saturation means (R-satisfactoriness (i.e. ondition (30:B:a))

B=

A u

(y) D A.

f Now B is (R-satisfactory, i.e.or #', y' in B, always x'<S(y'. Indeed, hen w x',y' are both in A, this follows from the (R-satisfactorinessof A. If x',y' areboth = y, we aremerely assertingy&y. If one of z',y' is in A, and the other = y, then the symmetry of allows us to assume x in A, y' = y. w Now our assertioncoincidesith the negation of (30 :D). t If is not symmetric,we can only assert his:
(R f (R

(30:H)

(R-saturation of A impliesits being maximal (R-satisfactory.

ness, cf. (30:E). As (R is symmetric,this amounts to (R saturation by of (30:G). And this is a consequence (R-saturation by (30:F). a The meaning of the result concerning symmetric (R is the following: as with any (R-satisfactory set, this set must be increased long as Starting u i.e.ntil any further increasewould involve the loss of (R-satispossible, factoriness. In this way in fine a maximal (R-satisfactory set is obtained,
2 n (R-saturated one by (30:G). This argument secures ot only of the existence (R-saturated sets,but it also permits us to infer that every to (R-satisfactory set can be extended an (R-saturated one.
1
<H

M Proof: aximal (R-satisfactorinessis the sameas maximal (R-satisfactori5

i.e.an
8

Note that none of the extra restrictions on has beenused so far. it is over after a finite number of This processof exhaustion is elementary steps when D is finite. However, sincethe set of all imputations is usually infinite, the caseof an infinite D of is important. When D is infinite, it is still heuristically plausible that the process exhaustion referred to can be carried out by making an infinite number of steps. This k process, nown as transfinite induction, has been the objectof extensive set-theoretical studies. It can be performed in a rigorous way which is dependent upon the so-called axiom of choice. Thereaderwho is interested will find literature in F.Hausdorff, footnote on p. Cf.alsoE. ZermelOj Beweisdass jede Menge wohlgeordnet werden kann. Math. Ann.

i.e.

1,

61.

a 59 (1904) . 514ff. nd Math. Ann. Vol. 65 (1908) . 107ff. p p for Thesematters carry far from our subjectand are not strictly necessary purposes. We do not therefore dwell further upon them.))) Vol.

our

270

GENERAL THEORY:ZERO-SUMn-PERSONS

statement is also true. of It 30.3.6.would be very convenient if the existence solutionsin our theory couldbe establishedby such methods. The prima facie evidence, the however, is against this: relation which we must use,a;(Ry negation of the domination x s- y, cf. 30.3.3. clearlyasymmetrical. Hence canwe is m not apply (30:G),but only (30 :H): aximal satisfactorinessis only necesf sary, but may not be sufficient for saturation, i.e.or being a solution. That this difficulty is really deep can seated be seenas follows:If we could the replace above (R by a symmetric one,this could not only be usedto prove the existence solutions, but it would also prove in the same operation of the possibility extending (R-satisfactoryset of imputations to a solution of any a (cf. above). Now it is probable that every game possesses solution,but we shall seethat there existgames in which certain satisfactory setsare 2 subsets of no solutions. Thus the device of replacing (R by something this symmetric cannot work because would be equally instrumental in provthe first assertion, hich is presumablytrue, and the second w ing one,which 8 is certainly not true. The reader ay feel that this discussionis futile, sincethe relation x&y m which we must use (\"not H y\") is defacto asymmetric. From a technical x of view, however, it is conceivable that another relation x$y may be point found with the following properties: is not equivalent to x&y; indeed, x$y is symmetric,while (R is not, but-saturation is equivalent to (R-saturation. In this case(R-saturated sets would have to existbecausethey are the -saturated ones,and the g-satisfactory but not necessarilythe (R-satisto i (R-saturated factory sets would always be extensible g-saturated, .e. 4 ones. This program of attack on the existence of solutionsis not problem as arbitrary as it may seem. Indeed, e shall seelatera similar problem w which is solvedin preciselythis way (cf. 51.4.3.). this is, however, for All the time beingjust a hope and a possibility. In 30.3.7. the last sectionwe considered question whether every the (R-satisfactory set is a subset of an (R-saturated set. We noted that for the relation x&y which we must use (\"notx H y\" asymmetric)the answer is in the negative. A brief comment upon this fact seems be worth while. to If the answerhad beenin the affirmative it would have meant that any setfulfilling (30:B:a) be extended one fulfilling (30:B:a) (30:D); can to and that or,in the notations of 30.1.1., any setof imputationsfulfilling (30:5:a) can be extended onefulfilling (30:5:a) (30:5:b). to and
1

It shouldbe noted that every subset of an (R-saturated setis necessarily (R-satisf ctory l The above assertion means therefore that the converse a

1Clearly property (30:B:a)not lost when passing to a subset. is 1Cf.footnote 2 on p. 285. 8 This is a rather useful Theinapproprinciple of the technical sideof mathematics. priateness of a method can be inferred from the fact that if it were workable at all it would prove too much. 4 The point is that (R- and S-saturation are assumed to be equivalent to eachother, but <H- and S-satisfactoriness not expected be equivalent.))) are to

EXACT FORM OF THE GENERAL DEFINITIONS 271

statement becomes: standard of behavior which is free from inner Any can contradiction be extendedto one which is stable, i.e. ot only free n
from inner

It is instructive

to

restatethis in

the terminology of 4.6.2. hen the T

The observationin 30.3.6., to according which the above is not true in in general,is of some significance: orderthat a set of rules of behavior
1 inner contradictions.

of it.

contradictions,but also ableto upset all imputations outside

should be the nucleus (i.e. subset) of a stablestandard of behavior, it a freedom from s may have to possessdeepertructural propertiesthan mere 30.4.ThreeImmediate Objectives

strictedzero-sumn-person game and can therefore begin the systematic with the investigation of the properties of this concept. In conjunction of this investigation it seems to carry out three early stages appropriate cases special enquiries. Thesedeal with the followingspecial First: hroughout the discussionsof 4.,the idearecurredthat the unsoT in phisticated conceptof a solution would be that of an imputation, our presentterminology, of a one-element V- In 4.4.2. saw specifiwe set w t cally that this would amount to finding a \"first\" element ith respecto domination. We saw in the subsequentparts of 4.,as well as in our exact that discussionsof 30.2., it is mainly the intransitivity of our conceptof domination which defeats this endeavorand forcesus to introduce setsof

30.4.1. have formulated the characteristicsa solutionof an unreWe of :

i.e.

imputations V as solutions. It is, therefore, of interest now that we are in a positionto do it to give an exactanswer to the following question:For which gamesdo oneelementsolutions V exist? What elsecan be said about the solutionsof such games? from were Second: postulates of 30.1.1. extracted our experiences The in its essential ase. It is, therefore, c with the zero-sum g three-personame, in t of interestto reconsiderhis case the light of the present,exacttheory. Of course,we know indeed this was a guiding principle throughout our discussions that the solution which we obtained by the preliminary methods of 22., 3., re solutions in the senseof our presentpostulates 2 a too. Neverthelessit is desirableto verify this explicitly. The real point, to however, is to ascertainwhether the presentpostulates do not ascribe further solutionsas well. (We have already seenthat it is not thosegames s that inconceivable thereshouldexist everalsolutionsfor the same game.) all We shall therefore determine solutions for the essentialzero-sum g three-person ames with results which are rather surprising,but, as we shall see,not unreasonable.
1If the relation S referred to at the end of could be found, then this $ and the which standards of behavior are such nuclei not (R would disclose subsets): S-satisfactory ones. where the corresponding operation is performed Cf. the similar situation in

30.3.6.

(i.e.

51.4.,

successfully.)))

272
We

GENERAL THEORY: ZERO-SUMn-PERSONS

n

that observedin the first remarkof 27.5.2. for n = 1, , thesegamesare 2 c of so inessential; this, togetherwith the inessentialand the essential ases n = 3,takes careof everything in n ^ 3. When this program is fulfilled we are left with the games n ^ 4 and we know already that difficulties of a new kind begin with them (cf. the allusionsof footnote 1,p.250, and the end of 27.5.3.). Third: the 30.4.3. We introduced in 27.1. conceptof strategicequivalence. It appeared plausible that this relationship acts as its name two expresses: games which are linked by it offer the same strategical possibilitiesand inducementsto form coalitions,etc. Now that we have o basis, this heuristic expectation put the concept f a solution on an exact demandsa rigorousproof. Thesethreequestionswill be answeredin (31 in 31.2.3. :P) ; in 32.2. ; and in in (31:Q) 31.3.3., respectively.

These 30.4.2. two itemsexhaustactually all zero-sum gameswith

3.

which surround them, like inescerning solutions,and the other concepts sentiality, essentiality, domination, effectivity. Sincewe have now put all thesenotions on an exact asis, the possibilityas well as the obligation b arises to be absolutelyrigorousin establishingtheir properties. omeof S the deductionswhich follow may seem pedantic,and it may appearoccasionthe ally that a verbal explanationcould have replaced mathematical proof. Suchan approach, however, would be possiblefor only part of the results t of this sectionand, taking everything into account,he best plan seemsto be to go about the whole matter systematicallywith full mathematical rigor. Someprincipleswhich play a considerablepart in finding solutionsare : : w (31 (31B), 31C), 31:F),(31G), (31H), hich for certain coalitions :A), : ( : ( a decide priori that they must always, or never, be taken into consideration. It seemedppropriate to accompanytheseprincipleswith verbal explanaa tions (in the senseindicated above) in addition to thqir formal proofs. Theotherresults possessvarying interestof their own in different directions. Togetherthey give a first orientation of the circumstances which surroundour newly won concepts.Theanswersto the first and third quesare tions in 30.4. given in (31 and (31Q). Another question which :P) : in arose :M). previously is settled (31 Considertwo imputations a , ft and assumethat it has become

31.1. Convexity, Flatness, and SomeCriteria for Domination This 31.1.1. sectionis devoted to proving various auxiliary results con-

31. irst Consequences F

31.1.2.

whether a H necessaryto decide

30.1.1.

ft or not. This amounts to deciding a whether or not there exists set S with the properties (30:4:a)-(30:4:c) in is One of these,(30:4:c)

> ft

for all in

S.)))

FIRSTCONSEQUENCES

273

We call this the main condition. The two others, (30:4:a), are (30:4:b), the preliminary conditions. Now one of the major technical ifficulties in handling this concept f d o in finding solutionsV in the sense domination of is the presence of thesepreliminary conditions. It is highly desirableto beable, o to say, s to short circuit them, to discovercriteria nder which they arecertainly u satisfied,and othersunder which they arecertainly not satisfied. In looko that they should ing for criteriaf the latter type, it is by no meansnecessary involve non-fulfillment of the preliminary conditions all imputations a for

i.e.

30.1.1.

i.e.

it sufficesif they involve it for all those imputations a which

fulfill

the main

condition for someother imputation (Cf.the proofs of (31 or (31F), :A) : where exactly this is utilized.) in We areinterested criteria of this nature in connectionith the quesw tion of determiningwhether a given setV of imputationsis a solutionor not; whether it fulfills the conditions (30:5:a), the (30:5:b) condition

i.e.

aredominatedby elementsof Vwhich disposeof the preliminary conditionssummarily, in the Criteria situation described above, are most desirableif they contain no reference
at all to ^V' 2

in This (30:5:c) 30.1.1. amounts

to determining which imputations

(Cf. (31F), 31 (31H).) But : ( :G), : which involve a may be desirable.(Cf. (31 even criteria :A).) We shall consider even a criterion which deals with S and a by referring to the
*

if i.e. they refer to 8alone.

b behavior of another a (Of course, oth in V. Cf. (31B).) : In order to cover all these possibilities,we introduce the following terminology: We considerproofs which aim at the determinationof all imputations o We /3 , which aredominatedby elements f a given set of imputationsVwith are thus concerned the relations a H ft (a in V), and the question whether a certainsetS meets our preliminary requirementsfor such a rela-

'.

tion.

We call S certainly necessaryif we know (owing to the fulfillment

by

t of some appropriate criterion) that S and a always meet he preliminary if conditions. We call a set S certainly unnecessary, we know (again owing b to the fulfillment by S of some appropriate criterion, ut which may now involve otherthings
1Thepoint

c too, f. above) that

the possibilitythat S and

of a **
ft

a meet

being that in our original definition

the preliminary

conditions

does. referto 5 and to (30:4:b) (but not to ft ). Specifically: 1The hypothetical element of V, which should dominate ft
.)))

274

GENERAL THEORY:ZERO-SUMn-PERSONS

t the preliminary conditionscan be disregarded(becausehis never happens, or for any otherreason. Cf. also the qualifications made above). a but Theseconsiderationsmay seem complicated, they express quite natural technical tandpoint.1 s of We shall now give certaincriteria the certainly necessaryand of the certainly unnecessary characters.After each criterionwe shall give a verbal explanationof its content, hich, it is hoped,will makeour technique w t clearero the reader.

31.1.3. threeelementarycriteria: First,

an t in

(31 :A)

S is certainly unnecessaryfor a given a S with a, = v((i)).

(in V) if

thereexists

A if Explanation: coalition neednever be considered it does not promise to every participant (individually) definitely more than he can get for himself.

I Proof:f a fulfills the main conditionfor someimputation, then on > ft. Since ft is an imputation, so ft ^ v((i)). Hence > v((i)). This cona< tradicts = v((i)). cti

S is certainlyunnecessaryfor a given a (in V) if it is certainly for necessary(andbeingconsidered) anothera (in V), suchthat ; ^ a, for all i in S. (31:1) A if Explanation: coalition neednot be considered another one, which has the sameparticipants and promisesevery one (individually) at least as much, is certainto receive consideration. Let i a, Proof: a and ft fulfill the main condition: > ft for aliin S. Then a and ft fulfill it also, y (311), > ft for all i in S. Since and a' are S b : a(

(31B) :

'

'

For the readerwho is familiar with formal logicwe observe following: the The attributes \"certainly necessary\"and \" certainly unnecessary\" are of a logical nature. They arecharacterizedy our ability to show (by any means whatever) that a b certain logicalomission will invalidate no proof (of a certain kind). Specifically: a Let > > beconcerned ith the domination of a ft by an element a ofV. Assume that this w proof > > domination a H occurring with the help of the set S ( a in V) beunder consideration. Then this proof remains correctif we treat S and a (when they possess attribute in the without our question) as if they always (or never) fulfilled the preliminary conditions, these conditions. In the mathematical proofs which we shall actually investigating will be applied frequently. carry out, this procedure
1

Itcan even happen

that the

same Swill

turn out (by the

use of two

different criteria)

to be both certainly necessary certainly unnecessary (for the same a , e.g. all of and for them). This means merely that neither of the two omissions mentioned abovespoils * any proof. This can happen, for instance, when a fulfills the main condition for no and (31:G)n the casedescribed i (An exampleisobtained by combining (31:F) imputation. in (31 Another is pointed out in footnote 1on p. 310, and in footnote on p. 431.))))

::b).

FIRSTCONSEQUENCES

275

o being considered, thus establishthat ft is dominatedby an element f they to Vi and it is unnecessary considerS and a .

(31:C)

S is certainly unnecessaryif another set TzSis certainly necessary(and is being considered).

A if Explanation: coalition neednot be considered a part of it is already certainto receive consideration. L Proof: et a (in V) and ft fulfill the main conditionfor S\\ then they will obviously fulfill it a fortiori for T S. SinceT and a arebeingconsidered, o they thus establishthat ft is dominatedby an element f V and it is unneces-

the following consideration: For an arbitrary set S = (k\\, Si = (ti), p , S = (fc p). Then
obtains, i.e.

sary to considerS and a . 31.1.4. now introduce some further criteria,and on a somewhat We broader basis than immediately needed.For this purpose we begin with
, k p) i apply (25:5)n
with 25.4.1.,

v(S) ^ v((*0) +

' ' ' + v((fcp))

(31:2)
of the coalition
if vanishes,i.e.

v(S) g; % v((fc)).
kinS

o The excessf the left-hand side of (31 over the right hand side expresses :2) inherentin the formation the total advantage (for all participants together)

S. We

call this the convexity of

vGS)

S. If this

advantage

(31:3)
(31D) : : (31D:a) (31:D :b) (31D:c) : (31E) :

kin

% v((fc)), 3

then we call S flat. The following observationsareimmediate:

Every subset of a flat set. i Any oneof the following assertionss equivalent to the inof the game: essentiality , n) is flat, (31:E:a) / = a Thereexistsn S such that both Sand S areflat, (31 :E:b) (31:E:c) Every 8 is flat.

The following setsarealways flat: The empty set, set, Every one-element

(!,

A Proof: d

is For (31:D:b):thesesets(31:3)obvious.))) (31:D:a),

276
Ad

GENERAL THEORY: ZERO-SUMn-PERSONS

: flat. (31D:c):Assume SsT,T Put R = T S. (31:4) v(5))) k in S (31:5) v(ft) X v((fc)).

tinR

Then by (31 :2)

SinceT is flat, so by (30:3)

(31:6)
As

v(T)

- I v((*)).
*inT
v(T),)) kinT
R

Sn ft = 0, S u ft = T;therefore v(S) + v() ^

* in

* in

Hence :6)implies (31

(31:7)

v(S)+v()3*inS v((*))+

fcinfl

Now comparisonof (31:4), and s (31:5) (31:7)hows that we must have equality in all of them. But equality in (31:4) just expresses the flatness of

S.

The (31:E:c): assertioncoincideswith (27:C):a27.4.2. For (31:E:b): an inessentialgame this is true for any S owing i (31E:c).Conversely,f this is true for (at least one)S,then : v(5) = v((*)), v(-S)= S v((*))
f

Ad Ad Ad

w The (31:E:a):assertioncoincidesith

(27:B)in

27.4.1.

to

kin

k not in

henceby addition (use (25:3:b)25.3.1.), in

/fe-i

t v((*))
f

t i.e.he game is inessentialby (31:E:a)by (27:B)in 27.4.1. or 31.1.6. arenow in a positionto prove: We

(31:F)

S is certainly unnecessaryif it is flat.

A if Explanation: coalition neednot be considered the game allows no total advantage (for all its participantstogether) over what they would get 1 for themselvesas independentplayers.

1Observehat this is relatedto (31A), but not at all identical with it Indeed (31A) t with dealswith the a t , i.e. the promises made to eachparticipant individually. (31:F) with dealswith v(*S)(which determines flatness), i.e. the possibilities of the game for all with what participants together. But both criteria correlatethesewith the v((i)), i.e. eachplayer individually can get for himself.)))

: :

FIRSTCONSEQUENCES
S 5* 2) t
in

277

Proof:If a ^

>

>

. cti>fa
>
t in

ft

with

for all
As

<

v((t)).
<*

S is flat,

i in

the help of this

S and

be effective,

UnS

* v(fi>), which

S'then have: Necessarily we ^ v((i)), hence on > v((i)). So fa this means J) <* > v(/S). But S must i in S is a contradiction.

(31:G)

opposesone of the kind describedin (31F). : > T Proof: he preliminary conditionsarefulfilled for all imputations a .
Ad Ad
n

S is certainly necessaryif S is flat and 8 ^ A i Explanation: coalition must be consideredf it (is not empty and) S^ (30:4:a): Q was postulated. ^ v((t)), so (30:4:b): Always

0.

t not in

^
t not in

v((i)). Since
the))

5) t-l

= 0, the left-hand side is equal to

S

^ en.

Since S is flat,

in (use (25:3:b) right-hand side is equal to v( S), So % ^ -vGS), S is effective. a< g v(S),

t in

a.

From

s p-element et is certainly necessary if p = n 1,and if n certainly unnecessary p = 0, 1, . if Explanation:A coalition must be considered it has only one opponent. i A coalition neednot be considered,f it is empty or consistsof one player o only (!), r if it has no opponents. h p Proof: = n 1: S has only one element, enceit is flat by (31:D) above. The assertionnow follows from (31:G). Immediate (31 and (31F). : :D) by p = 0, 1: = n: In this casenecessarily = / = rendering the p t main condition unfulfillable. Indeed, hat now requiresa< > ft for all

(31:G) (31:F), we obtain by specialization:

A

t in

i.e. i.e.

to 25.3.1.) v(S).

(31:H)

>S

(l,*--,n)
_ _
^ ^

i = 1,

, n,

hence

sides vanish, and this is a contradiction. to Thus thosep for which the necessityof S is in doubt, arerestricted n t i.e.o the interval p ?* 0, 1, 1,n, 2. 2gpgn (31:8) This interval plays a role only when n ^ 4. The situation discussedis and and similar to that at the end of 27.5.2. in 27.5.3., the casen = 3 more as oneof particularsimplicity. appears once

t-i

a >

t-i

ft.

But as a ,

areimputations,both

We 31.2.1. now discussthe structureof the system of all imputations.

The 31.2. System of All Imputations.

One-elementSolutions

(311) :

Foran inessential thereexists reciselyoneimputation:))) p game

278

GENERAL THEORY:ZERO-SUMn-PERSONS

(31:9) a = {i,
them.

Foran essentialgame existinfinitely an (n continuum but l)-dimensional

= -,), therev((i))
<*

for

= 1,

n.

is (31:9) not one of

many

imputations

C Proof: onsideran imputation = {01, - ,/M, and put for i = 1, ft = v((0) + *

n.

c Then the characteristiconditions(30:1), of (30:2) = for i^ , n. (31:10)

(31:11) (31:11) (31:10), amount

imputation.

i-i
to ci =

- - *-i ((*))v

i 1,

become 30.1.1.

If T is inessential, then (27:B)in

27.4.1. gives

= *n = 0, i.e. 31:9) the is (

- t'-iv((i)) = 0; so
n

unique

the If T is essential, (27:B)in

27.4.1. gives

solutions,which form an (n l)-dimensional continuum; 1so the sameis true for the imputations ft . But the a. of (31 :9) = en = now violate (31:11). is not oneof them, because = ei An immediateconsequence: A solution V is never empty. (31J) :
many

(3111) infinitely : possess

> so t-iv((t)) 0, (31:10),

Proof:I.e.the
imputation
ft

is not a solution. Indeed: onsiderany C and no , thereexists at leastoneby (311). ft is not in :

empty
ft

set

in

We 31.2.2. have pointed out before that

has a H

So violates
~2 *-

2 in 1. (30:5:b)30.1.

the simultaneousvalidity of
\"^

(31:12)
(31:K)
S (31:12)

7, 7

**

3 is by no means impossible.However:

conditions for the imputations conthen V without would be a solution. flicted 3 The sets of thesetwo dominations would have to be disjunct. :H) By (31 these must have ^ 2 elements each. Hence(31 can occuronly when n ^ 4. 4 can be excluded;ut for every n 5 b By a more detailed consideration even n is really possible.)))

Never a H 1Thereis only one equation: 1This argument may seempedantic ; but

a.

(i.e.

(31:11).if the
Q
:12)

(31:1)),

FIRSTCONSEQUENCES
in conflict Proof: 30:4:a), ( (30:4:c)30.1.1. for to = ~p.

279

(31:L)

Given an essentialgame and an imputation there exists an imputation ft such that ft H a but not a H ft. 1

a,

P Proof: ut

a =
Considerthe equation

{,,
<*

}.

(31:13)

= V ((0).

Sincethe game is essential, 31:1) t be ( excludeshe propositionthat (31:13) = 1, valid for all i , n. Let (31:13) say forf = i . Sincea is an fail, the me&ns imputation, so a S v((i )), hence failure of (31:13) a o > v((i )),

i.e.

(31:14)
Now define a vector))
by ft

<.

= v((i )) + e,

>0.

ft o =

aio

= v((i )),
for
i

=a

n ft

t'

.
n

t These equationsmakeit clearhat

^ v((i))2 and that

i-l
.

ft

^ a = O. -l

So

is an imputation along with a . a We now prove the two assertionsconcerning ,

ft

ft

a : We have ft > a for all i ^

H

io,

for i.e. all i in the setS =

>

(I'D)

This set has n

1elements it fulfills and

a.

the main condition (for

ft

a ),

hence(31 gives :H)

Not a H ft :Assumethat a H . Then asetS fulfilling the main condition must exist,which is not excluded (31H). So S must have by : g: 2 elements.So an i ^ i'o in S must exist. The former impliesft > a
1Hence a ?* 1For t to, we have

>

actually

0,

v((*' )). For i 5* i , we have

0, > a

v((i)).
t*

X
-)))

-i

ft

\"*

2 t-i

a kecausethe
*

difference

of ft

and

a.is c for one value of i (i

) and

for n

1values of i (all i **

io).))

280

GENERAL THEORY:ZERO-SUMn-PERSONS

(by the constructionof ft ) ; the latterimpliesa > ft (owing to the main condition) and this is a contradiction. We 31.2.3. can draw the conclusionsin which we were primarily inter-

ested: (31:M)

imputation a , for which never 1 only if the gameis inessential.

An

'

a , exists if and

by S Proof: ufficiency: If the game is inessential,then it possesses (31:1) preciselyone imputation a , and this has the desiredpropertyby (31:K). and If Necessity:the game is essential, a. is an imputation, then a ' = of (31 gives a' = ft H a. :L)

(31N) :

a one-element solution2 is necessarily A game which possesses inessential. solutionin questionby V = ( a ). This D Proof: enotethe one-element
in This (30:5:b)30.1.1. meansunder our present circumft

V must satisfy

stances: Every

otherthan a is dominatedby
ft

a. I.e.:
>

j a

>

implies

>

ft

.
ft

condition.

then Now if the game is essential,

An

(31L)providesa :
with

which violates this

(310) :

inessentialgame possesses preciselyonesolution V. This

the

is the one-element V = ( a ) set

A

of

(311). :

there B Proof: y (31:1) exists preciselyone imputation, the a of (31:1). solution V cannot be empty by (31 hencethe only possibility is :J); = ( a ). Now V = ( a ) isindeeda solution,i.e. fulfills (30:5 (30:5:b) it V :a), :the in 30.1.1. former by (31 the latterbecausea is the only imputation :K), by (31:1). We can now answer completelythe first questionof 30.4.1.: A gamepossesses a one-element solution(cf.footnote 2 above) (31P) : if and only if it is inessential; nd then it possesses other a no solutions. T Proof: his is just a combination of the results of (31 and (31:0). :N)
1 Cf. the considerations of (30:A:a)n i and particularly footnote 2 on p. 265. 1 We do not excludethe possibility that this game may possess other solutions as well, which may or may not be one-elementsets. Actually this never happens (under our with present hypothesis), as the combination of the result of (31:N) that of (31 :0) or the result of (31 :P) shows. But the present consideration is independent of all this.)))

30.2.2.,

FIRSTCONSEQUENCES

281

to Corresponds Strategic Equivalence Considertwo games F and F' with the characteristicunctions f v(S) and v'(S) which are strategically equivalent in the senseof 27.1. We proposeto prove that they are really equivalent from the point of view of the concepts defined in This will be done by establishingan isomorphiccorrespondencebetweenthe entitieswhich form the substratum of the definitions of i.e.the imputations. That is, we wish to establish a one-to-one orrespondence, between the imputations of F and c those of F',which is isomorphicwith respecto those concepts, .e. hich t i w carriesffective sets,domination, and solutionsfor F into those for F'. e The considerationsare merely an exactelaboration of the heuristic indicationsof hencethe reader ay find them unnecessary. Howm \" ever, they give quite an instructive instance of an \" isomorphismproof, and, besides, our previous remarks on the relationship of verbal and of exactproofs may be applied again. Let the strategic be , ajin the sense equivalence given by a i, Which

31.3.1.

31.3. Isomorphism The

30.1.1. 30.1.1.,

27.1.1.,

31.3.2.

of

one correspondence

Considerall imputations a = \\a\\ , an ) (27:1), in 27.1.1. (27:2) of F and all imputations a ' = {a' , a^jofF'. We lookfor a one-to9

lf

(31:15)
with

~^<=~^'

the specifiedproperties. (31:15) to be is easilyguessedfrom the motivation at the ought therethe passage from F to F' by adding We described beginning of to the game a fixed payment of a to the player k. Applying this principle to the imputations means = ak + ak * = n.1 for k (31:16) (31:15)the equations(31:16). by Accordingly we define the correspondence We now verify the asserted propertiesof (31:15), (31:16). Theimputationsof F aremappedon the imputationsof This means that (30:2)in by (30:1), - - . n. a ^ v((t)) for (31:17)
What

27.1.1. a'

!,-,
, n,

31.3.3.

30.1.1.,

F':

i - 1,

(31:18)
go over into

'-! = 0,
,

(31:17*) (31:18*)
l

a( ^ v'((i))

t-1
a'
the (fixed)
>

2:-0.
- lJ,

for

i = 1,
,

lf

we introduce

vector a
>

written vectorially of the imputations.)))

a + a

it . I.e. is a translation

i|
(by

then

may (31:16) be
the

a ) in

vector space

282

GENERAL THEORY:ZERO-SUMn-PERSONS

This is so for

because (31:17), (31:17*) v'((t)) = v((t)) + ? (by (27:2)in n becauseJ) a? = (by (27:1) and for (31:18), id.). 27.1.1.), (31:18*)

t-1

Effectivity for F goesover into effectivity for in 30. that

1.1.,

F':This means by (30:3)

2) goesover into

^ v(S)

tin))

2 J s vw.))
:
(27:2).

over into domination for F' Thismeansthe same Domination for F goes is in is for (30:4:a)-(30:4:c) (30:4:a)trivial; (30:4:b)effecthing assertshat a > ft goesover into at > 0J t which we settled: (30:4:c) tivity, which is obvious. The solutionsof F are mapped on the solutionsof in These This means the same for (30:5:a), (or (30:5:b) (30:5:c)) conditionsinvolve only domination, which we settled.

evident by comparisonof (31:16) with This becomes

30.1.1.

30.1.1.

F':

We

restate theseresults:

(31:Q)

If two zero-sumgames F and F' arestrategicallyequivalent, an then there exists isomorphismbetween their imputations a one-to-one apping of those of F on those of F' which m leavesthe concepts efined in d invariant.

i.e.

30.1.1.

games.

32.Determinationof all Solutions of the Essential ero-sum Z Game Three-person $2.1. Formulation of the Mathematical Problem. The Graphical Method We now turn to the secondproblem formulated in The determinationof all solutionsfor the essential ero-sumthree-person z

32.1.1.

30.4.1.:

We know that we may considerthis game in the reduced form and that we can choose = 1 Thecharacteristic function in this case completely is 7 determinedas we have discussedbefore:

I.
=))

2))

(32:1)
An

v(fl)

-1
1))

when

S has))

elements.))

imputation is a vector))

=))

27.1. the and 1

1Cf.the

the beginning of in Cf. the discussion at the beginning of

discussion

remark second

at

27.3.

or t 29.1.,the referenceshere given: the or 20.1.,the secondcaseof 27.5.)))

end of

ALL SOLUTIONS F O

THE THREE-PERSON GAME

283

whosethree componentsmust fulfill (30:1), in (30:2) ditions now becomeconsidering ( (30:1))

30.1.1.* Thesecon-

(32:2) (32:3)
We know, from

ai

, -1, +aa*^+ -1, 0. =

on 3

-1,
a3 form
only a two-

dimensional ontinuum i.e.hat they shouldberepresentable the plane. c t in Indeed,32:3)makes a very simpleplane representationossible.)) ( p

in that (31:1) 31.2.1.,

these ai,

<*

2,

Figure 52.))

a,

-I))

a*

-))

7))
angles each other. For any point of the plane we define on, 2, as by directed perpendicular distancesfrom these three axes. The whole a t arrangement, nd in particular the signs to be ascribedo the <*i, as,
of 60
with
as)))

32.1.2. this purposewe takethree axesin the plane, making For

Figure

53.

284))

GENERAL THEORY:ZERO-SUMn-PERSONS))
that for any point the algebraic thesethree perpendiculardistancesvanishesand that conversely

sum of

aregiven in Figure52. It iseasy to verify

any triplet

a =

point. Sothe planerepresentationf Figure52 expresses o preciselythe condition The remaining condition (32:2)is therefore the equivalent of a (32:3). restriction 52. This imposedupon the point a within the plane of Figure is restriction obviously that it must lie on or within the triangle formed a by the threelines i = 1, 2 = 1, = 1. Figure53 illustratesthis. Thus the shaded area,to be called fundamental triangle, represents the the a which fulfill (32:2), a (32:3) i.e.ll imputations. 32.1.3. next expresshe relationshipof domination in this graphical We t representation.As n = 3,we know from (31:H) (cf. also the discussion
of))

{i,

2,

8}

to for which the sum vanishes,corresponds a

Figure 54.))

Figure

55.))

at that (31:8) the end of 31.1.5.) among the subsets 8 of / = (1,2, 3) those of two elementsare certainly necessary, and all others certainly i the unnecessary. I.e., setswhich we must considern our determination of all solutionsV arepreciselythese
:))

Thus for
domination

(1,3); (1,2); (2,3).

a = {ai,a2, as}, a
H
B))

ft

= {ft,ft,

ft}

means that))

(32:4)
as

> ft.

Either i

> ft, a2 > ft; or ai > fa,

a*

> ft; or

> ft,

1 others, in Figure54.)) 1In

particular,

Diagrammatically: a dominates the points in the shaded areas,and no

no points on the boundary

lines of theseareas.)))

ALL SOLUTIONSOF
>

THE THREE-PERSON GAME

285

Thus the point a dominates three of the six sextants indicated in Figure55 (namely A, C, B). From this one concludeseasily that a is dominated by the three other sextants (namely B, D, F). So the only points which do not dominate a and are not dominatedby it, lieon the threelines (i.e. half-lines) which separatehesesextants. I.e. six : t

(32:5)

If neither of a. , ft dominatesthe other,then the direction > from a. to ft is parallel to one of the sides of the fundamental

triangle.

32.1.4. the systematicsearchfor all solutionscan begin. Now Consider a solution V, i.e. set in the fundamental triangle which a fulfills the conditions(30:5:a), In of (30:5:b)30.1.1. what followswe shall
use theseconditionscurrently,
without

occasion.

referring to them explicitlyon each

Sincethe game is essential, V

must contain at

leasttwo points say

1

a and ft . By (32:5)the directionfrom a to ft is parallel to one of the sidesof the fundamental triangle; nd by a permutation of the numbersof the a w i the players 1,2,3e can arrange this to be the side i = 1, .e. horizontal. * So a , ft lie on a horizontal line I. Now two possibilities riseand we treat a them separately: (a) Every point of V lies on I. (b) Somepoints of V do not lie on L

32.2. Determination
>

of All Solutions
fulfill

32.2.1. consider(b) first. Any point not on I must w We (32:5) ith it t respecto both a and , i.e. must be the third vertex of one of the two equilateraltriangleswith the base a , : one of the two points a ', a of
ft
>

56. Soeither a or a \" belongsto V. Figure

from

>

'

ft

\"

and a 'or a must again fulfill > three points a , ft and a or a \". This, however, is impossible,as an these 56 inspectionof Figure immediately reveals. SoV consistsof precisely o three points, i.e. f the threeverticesof a triangle which is in the position of triangle / or triangle // of Figure57. Comparisonof Figure57 with Figures54 or 55 showsthat the vertices of triangle / leave the interior of 2 this triangle uridominated. Thisrules out I.

a,

\"

ft

'

point of V which differs to but (32:5), now with respect all

Any

1 This is also in directly observable Fig. 54. 2 This provides the example referred to in not dominate eachother, they form a satisfactory nevertheless unsuitable as a subset of a solution.)))

i.e.

The 30.3.6.:three set in

the

of sense loc.cit. They are

verticesof triangle

/ do

286

GENERAL THEORY: ERO-SUMn-PERSONS)) Z

Figure 57.))

Fundamental

Triangle
58.))

danienUl

Triangle
59.)))

Figure

Figure

ALL SOLUTIONS FTHE THREE-PERSON O GAME

287

The same comparison shows that the vertices of triangle // leave undominatedthe dotted areasindicated in Figure58. Hence triangle // in must be placed such a manner in the fundamental triangle that these dotted areasfall entirely outside the fundamental triangle. This means that the threeverticesof II must lieon the threesides of the fundamental triangle,as indicated in Figure 59. Thus these three vertices are the middlepoints of the threesidesof the fundamental triangle. Comparisonof Figure59 with Figure54 or Figure55 shows that this set V is indeed a solution. One verifies immediately that thesethree middlepoints arethe points (vectors) (32:6)
t i.e.hat

{-!,*,*}, 41,It,*,-1|, {*,-!,

is the setof Figure51.))
Fundament*!
Triangle))

this solution V

Figure

60.

all 32.2.2. us now considera) in 32.1.4. this case of V lieson the Let In ( horizontal line L By (32:5)no two points of I dominate eachother, so that every point of / is undominatedby V. Hencevery point of I (in the fundae mental triangle)must belong to V. V is precisely that part of /

I.e.,

which is in the fundamental triangle. of V arecharacterizedy an equation b

So the elementsa =

{<*i, 2,

s}

(32:7)

c.

F Diagrammatically: igure60. Comparisonof Figure60 with Figures54 or 55 shows that the line I t leavesthe dotted areaindicatedon Figure60 undominated. Hencehe line I must be placed such a manner in the fundamental triangle that the in dotted areafalls entirely outside the fundamental triangle. This means that I must lie belowthe middlepoints of thosetwo sidesof the fundamental)))

288

GENERAL THEORY:ZERO-SUMn-PERSONS

1 c On the trianglewhich it intersects. In the terminology of (32:7):< other hand, c ^ 1 is necessary to make I intersecthe fundamental t triangleat all. So we have:

|.

(32:8)

-1S c <i

Comparisonof Figure60 with Figures54 or 55 showsthat under these 2 I conditions the set V i.e. is indeeda solution. But the form (32:7)of this solution was brought about by a suitable Hence have two further solutions, we permutation of the numbers 1,2,3. characterizedy b

(32:7*) and characterizedy b (32:7**)

always with

a, = c,
3

= c,

32.2.3. Summingup:
For every

(32:8)

This is a complete of solutions: list

c which
fulfills

(32:A) (32:B)

(32:7*), (32:7**).
of

The (32:8): three sets (32:7)

The set (32:6).

we had already decided hich solution we wanted; and this solution reapw now as (32 the But we found other solutionsbesides: (32: :B). pears A), which areinfinitely many sets,eachone of them an infinite set of imputations itself. What do thesesupernumerarysolutionsstand for? Consider, .g.,he form (32:7)of (32: This gives a solution for e t A). 2, 3 } which fulfill every c of (32:8) consistingof all imputations a = = c. Besides they must fulfill only the requirements, i.e. (32:7), i this,

33.1.1. result of 32.calls for careful considerationand comment. The determined all solutions of the essentialzero-sum three-person before were game. In 29.1., the rigorous definitions of 30.1. formulated,
We have

33.1. Multiplicity The

33.Conclusions
Solutions.
Discrimination and

Its

Meaning

{i,

1The limiting position of J, going through the middle points themselves, must be excluded. The reason is that in this position the vertex of the dotted area would lie on the fundamental triangle, and this is inadmissable sincethat point too is undominated
by V,

our present position of V, on the other hand, the vertex under consideration to belong to V, This exclusion of the limiting position causesthe < and not the ^ in the inequality which follows. * (32:8), I intersects fundamental the triangle, but below its middle.)))

to V. In

Observe that the corresponding prohibition did not occurin case (b), i.e. the for dotted areasof Figure 58. Their vertices were undominated by V, but they belong too doesnot

i.e. I. by

i.e. I.

i.e.

CONCLUSIONS

289

(30:1), of (30:2) Our solution consistsof all

(33:1)
\"^

30.1.1. (32:2), of 32.1.1. other words: i.e. In (32:3)

= {c,a, -c -a},

-1g a 1- c.
i.e.

The interpretation of this solution consists manifestly of this: ne O of the players(in this case is beingdiscriminatedagainstby the two others 1) (in this case 2,3). They assignto him the amount which he gets,c. This amount is the same for all imputations of the solution, of the accepted standard of behavior. The place societyof player 1is prescribedby the in two other players; he is excluded from all negotiationsthat may lead to coalitions.Suchnegotiationsdo go on, however, between the two other players: the distribution of their share, c, depends entirely upon their s bargainingabilities.The solution, the acceptedtandard of behavior, no restrictions upon the way in which this share is imposesabsolutely dividedbetweenthem, expressed a, c a.1 This is not surprising. by Sincethe excluded player is absolutely\"tabu,\" the threat of the partner's desertionis removed from each participant of the coalition.Thereis no 2'3 way of determiningany definite division of the spoils. It o Incidentally: is quite instructive to seehow our concept f a solution as a set of imputationsis able to take careof this situation also. Thereis more that shouldbe said about this \"discrimination\" against a player. First,it is not done in an entirely arbitrary manner. The c, in which is discriminationfinds its quantitative expression, restrictedy (32:8)in b is clear nough in its meaning, but the c Now 32.2.2. part of (32:8),^ e (cf. significance of the other part c < 4 is considerablymore recondite E b however, below). It all comes ack to this: ven an arbitrary system of discriminationscan be compatiblewith a stable standard of behavior orderof society but it may have to fulfill certainquantitative conditions, that it may not impair that stability. in order t Second, he discriminationneed not be clearly disadvantageousto the his fixed player who is affected. It cannot be clearlyadvantageous, value c cannot be equal to or betterthan the bestthe others may expect. that c^ which is exactly This would mean,by (33:1), c ^ 1 c, forbids. But it would be clearly disadvantageousonly for what (32:8) but c = 1 and this is a possiblevalue for c (by (32:8)), not the only one. ; but to c = 1means that the player is not only excluded, also exploited

i.e.

33.1.2.

1,

i.e.

i.e.

i.e.

i,

1 is, of course,he 1 ^ a ^ 1 c of (33:1). t at the end of 25.2.Observethat the arguments which we adducedthere to motivate the primate of v(S) have ceasedto operatein this particular case and v(S) nevertheless determines the solutions! 8 Observe the amount in the \"spoils\", that due to (32:8) c, can be both positive and negative. 4 And that is excludedin c < J, but not in c ^
2 Cf. the discussions

any

1Exceptthat both outside help.

must

be <z

i.e. what

the player

can get for

himself, without

a, c

&

32.2.2.,

i.e.
1.)))

290 GENERAL THEORY: ERO-SUMn-PERSONS Z 100 cent. The remaining c (of (32:8))with 1< c < i correspondo t per forms of segregation. gradually lessand lessdisadvantageous It that 33.1.3.seems remarkable our conceptof a solutionis able to a (32 a expressll thesenuancesof non-discriminatory :B), nd discriminatory (32: standards of behavior the latterboth in their 100er centinjurious A), p form, c = 1,and in a continuousfamily of lessand lessinjurious ones 1< c < i. It is particularly significant that we did not look for any of such thing the heuristicdiscussions 29.1 erecertainly not in this spirit w

but we werenevertheless forcedto theseconclusions y the rigoroustheory b itself. And thesesituations arose even in the extremelysimpleframework of the zero-sumthree-person ame! g a For n ^ 4 we must expect much greaterwealth of possibilitiesfor all sorts of schemes discrimination,prejudices, of privileges,etc. Besides o i.e. these,we must always\"look out for the analoguesf the solution (32:B), \" the nondiscriminating objective solutions. But we shallsee that the conditions arefar from simple. And we shall also seethat it is preciselythe solutionswhich leads to a investigation of the discriminatory\"inobjective\" proper understanding of the generalnon-zero-sumgames and thenceto applicationto economics.

33.2. and Statics

Dynamics

indeedit was really meant for the phase which our theory has now reached. and In 29.2. in the places the referred to there, we considered negoand fears which precede formation of a coalition the tiations, expectations its and which determine conditions. Thesewere all of the quasi-dynamic in described 4.8.2. he same applies to our discussionin 4.6. nd T a type in 30.2., how various imputations may or may not dominate each of again other dependingon their relationshipto a solution; .e., the conducts i how an establishedstandard of behavior do not conflict with each approvedby other,but can be used to discreditthe non-approvedvarieties. in and The excuse, the necessity,for using such considerations a static were set forth on that occasion.Thus it is not necessaryto repeat theory
them
now.)))

At 33.2. this point it may be advantageousto recallthe discussionsof s 4.8.2. concerningtaticsand dynamics. What we said then appliesnow;

CHAPTER VII ZERO-SUMFOUR-PERSON GAMES

34.1. are now in possessionof a generaltheory of the zero-sum We b n-persongame, ut the stateof our information is still far from satisfactory. Save for the formal statement of the definitions we have penetratedbut little below the surface. The applicationswhich we have made i.e.he t cases in special in which we have succeeded determiningour solutions can be rated only as providing a preliminary orientation. As pointed out in these n 30.4.2., applicationscover all cases ^ 3,but we know from our past discussions ow little this is in comparison h with the general roblem. Thus p we must turn to games with n ^ 4 and it is only herethat the full complexity to of the interplay of coalitionscan be expected emerge.A deepernsight i into the nature of our problemswill beachievedonly when wehave mastered the mechanisms which govern thesephenomena. The present chapteris devoted to zero-sum four-persongames. Our a information about these still presents many lacunae. This compelsn 1 w inexhaustive and chiefly casuistictreatment, ith its obvious shortcomings. But even this imperfect expositionwill disclose various essential ualitative q propertiesof the generalheory which couldnot be encounteredreviously, t p it (for n g 3). Indeed, will be noted that the interpretation of the mathematical results of this phase leads quite naturally to specific \"social\" a concepts nd formulations. 34.2.Formalism of the Essential Zero -sum Four-person Game In 34.2.1.orderto acquire an idea of the nature of the zero-sum fourpersongameswe begin with a purely descriptiveclassification. Lettherefore an arbitrary zero-sum four-persongameF be given, which 2 form: and also let us choose = I. we may as well considerin its reduced y as in Theseassertionscorrespond, we know from (27:7*) nd (27:7**) 27.2., a
to the following statements the concerning

34.Preliminary Survey 34.1. Viewpoints General

characteristic functions:))
elements.))

(34:1)
1

v(5)

=))

-1 1

when

S has))

2 Cf. and that

a emphasis on heuristic devices. E.g., considerable 27.1.4. 27.3.2.The readerwill note the analogy and
of

be said later.

29.1.2. concerning

the zero-sum

three-persongame.

between this discussion About this more will

291)))

292

GAMES ZERO-SUMFOUR-PERSON

then we have

Thus only the v(/S) of the two-elementsets S remain undetermined by o thesenormalizations. We therefore direct ur attention to thesesets. six of all The set I = (1,2,3,4) players possesses two-elementsubsets S: (1,3), (2,3),(2,4),(3,4). (1,2), (1,4), Now the v(S) of these sets cannot be treatedas independentvariables, as because each one of theseS has another one of the same sequence its t and complement.Specifically:he first and the last, the second the fifth, the third and the fourth, are complementsof each other respectively. their Hence v(S) arethe negativesof eachother. It is also to be rememberedthat by the inequality (27:7)in 27.2.with n = 4, p = 2) all these ( if v(S) are ^2, ^ 2. Hence we put)) )) = 2xi, (34:2) v((2,4))= 2x2, 1 v((3,4))= 2x3,

(34:3)
and in addition

= lv((l,2)) -2*3,

v((2,3))= v((l,3)) =
x 2,

-2*!,
-2x2,
x, ^

(34:4)

-1^ xi,
25.3.1.

1.

Conversely:If any three numbers Xi, x 2, x 3 fulfilling (34:4)are given, then we can define a function v(S) (for all subsets S of I = (1,2,3,4)) by function but we (34:l)-(34:3), must show that this v(S)is the characteristic this of a game. By 26.1. meansthat our presentv(S)fulfills the conditions and are of (25:3:a)-(25:3:c) Of these, (25:3:a) (25:3:b) obviously this remains. By 25.4.2. means showing that so only (25:3:c) fulfilled,
of Si,S2, Ss area decomposition /. If (Cf. also (25:6)in 25.4.1.) any of the setsSi,S2, S is empty, the two arecomplementsand so we even have equalityby (25:3:a), others (25:3:b) in 25.3.1. we may assume that none of the setsSi,S2, S is empty. So o a Sincefour elements reavailable altogether, ne of thesesets,say Si = S,

v(Si) + v(S2)

+ v(S3) ^

if

must have two

our inequality

becomes
v(S)

w elements, hile the

two

sets.Thus othersareone-element
v(S) ^

-2^0,

i.e.
2,

2. 2,
:)))

the inequality into

If we express for all two-elementsetsS,then (34:2),(34:3)transform this

-2*! 2, g
which

2*ig 2,

-2*, 2, -2x,g 2, g

2x2 ^

2x3 ^

is equivalent totheassumed(34:4) Thus we have demonstrated .

PRELIMINARY SURVEY))

293))

(34: A)

1 function, and its x\\, z 2, x* is given by the equations(34:l)-(34:3). 34.2.2. above representationf the essentialzero-sumfour-person The o games by triplets of numbers x\\, z 2, x 8 can be illustrated by a simple geometrical picture. We can view the numbers x\\, x$, x 3 as the Cartesian coordinatesf a point.2 In this case inequalities(34:4)describe part)) o the a

The essential ero-sumfour-persongames (in their reduced z with the choice = 1) correspond 7 exactly to the triplets of numbers xi, a?2, z 8 fulfilling the inequalities (34:4). The between such a game, i.e.its characteristic correspondence
form

Figure

61.
=

of space which exactly fills a cube Q. This cube is centeredt the origin a of the coordinates, its edges of length 2 because sixfacesarethe and are its

six planes

Xi

as shown in Figure61. Thus each essentialzero-sum four-person game F is representedy b preciselyone point in the interior or on the surface of this cube,and vice versa. It is quite useful to view thesegamesin this manner and to try to correlate their peculiarities with the geometrical conditions in Q. It will be particularly instructive to identify thosegames which correspondto definite significant points of Q.
1 Thereadermay now compareour result with that of concerning the zero-sum three-persongames. It will be noted how the variety of possibilities has increased. 2 We may also consider these numbers as the components of a vector in L$ in the senseof et seq. This aspectwill sometimes be the more convenient, as in footnote 1 on p.

1,

Z2

1,

X3

1,

29.1.2.

16.1.2.
304.)))

294

ZERO-SUMFOUR-PERSON GAMES

But even before carryitig out this program, we propose to consider certain questions of symmetry. We want to uncover the connections betweenthe permutationsof the players 1,2,3,4, the geometrical transand to formations (motions)of Q. Indeed: 28.1. former correspond the the by the symmetriesof the gameT, while the latterobviously express symmetries of the geometrical object.

i Indeed,n describingthe v(S) of the two-elementsetsS,we had to single out three from among these sets (which are six in number), in order to introducethe coordinates\\ x 2, x 3 We actually did this in (34:2), x (34:3) the by assigning player 4 a particular role and then settingup a correspondencebetween the players 1,2,3nd the quantities Xi, x 2, x 3 respectively a will Thus a permutation of the players 1,2,3 induce the same (cf.(34:2)). of the coordinates i, x 2, x 3 and so far the arrangement is permutation X symmetric. But theseareonly six permutations from among the total of 24 permutationsof the players 1,2,3,4.* a permutation which replaces So the player 4 by another one is not accounted in this way. for 34.3.2. us considersuch a permutation. For reasons which will Let t t appear immediately,considerhe permutation A, which interchangeshe 2 1and 4 with eachother and alsothe players2 and 3. A lookat the players equations (34:2), (34:3)suffices to show that this permutation leaves Xi x 2, x 3 Similarly one verifies: x while it replaces2, x 3 by invariant, The permutation which interchanges and 4, and also 1 and 3, leavesx 2 2 x 3 The permutation C, which invariant and replacesi, x 3 by Xi, X 3 and 4 and also 1and 2, leavesx s invariant and replaces x 2 Xi, interchanges
y

In 34.3.1. evolving the geometricalrepresentationof the essential zero-sum four-person game we had to perform an arbitrary operation, i.e. ne which destroyed part of the symmetry of the original situation. o

34.3.Permutations

of the

Players

5,

Thus eachone of the three permutations A, B, C affects the variables each x2, x8 only as far as their signs are concerned, changing two signs Xi, and leaving the third invariant. Since t they also carry 4 into 1,2,3, respectively, hey produceall permutations of the players 1,2,3,4,combinedwith the six permutations of the if to Now we have seen that the latter correspond the six players 1,2,3. x 3 (without changesin sign). Consequentlythe permutations of Xi, x 2, 24 permutations of 1,2,3,4 correspondto the six permutations of Xi, x 2, x 3, of with no changeof sign or with two changes each one in conjunction
8 sign.
1 Cf. following the definitions 2 With the notations of

by

-Xi,-x2.

28.1.1.,

29.1.:

( (28:A:a), 28:A:b).

1,2,3,4\\

_ B~

/1,2,3,4\\

c))

/1,2,3,4
X
be.)))

3 These 4 possibilities in sign changesare 1 + 3 t operationson Xi, x^x\\ to representhe 24permutations

eachcase,so we have 6 of 1,2,3,4, as it should

24

IN SPECIAL POINTS THECUBE

Q))

295))

also statethis as follows:If we considerall movements carry the cube Q into itself, it is easily verified that they consistof the permutationsof the coordinate xi, x 2, x* in combination axes the with any reflections on the coordinateplanes (i.e. planes x*, x 9 ; x\\ t Xf #1, xj). Mathematicallythesearethe permutationsof x\\ X*, x\\ in combinao tion with any changes f sign among the xi, X2, x 8 Theseare48 possibili1 is ties. Only half of these,the 24 for which the number of sign changes even (i.e. or 2), correspondto the permutationsof the players.))
in

34.3.3. may We

spacewhich

IV))

Figure 62.))

It is easily verified that thesearepreciselythe 24 which not only carry the cubeQ into itself, but also the tetrahedron V, VI, VII,as indicatedin I, s Figure62. One may also characterizeuch a movement by observingthat
it always carries vertex a

of Q into a vertex ; and equallya vertex ointo a 2 but never a into a o. o, We shall now obtain a much more immediate interpretation of these statements y describingdirectly the gameswhich correspondto specific b to points of the cubeQ: the vertices or o,to the center origin in Figure (the and to the main diagonalsof Q. 61), vertex

four corners 7,

Therefore it suffices to by suitable permutations of the players 1,2,3,4. consider neof them,say o

35.1.1. begin by determiningthe gameswhich correspondto the We : V, VI,VII. We have seenthat they arisefrom eachother

35.Discussionf Some Pointsin the Cube Q o Special 35.1. Corner / (and 7, VI, VII) The

/.

1For eachvariable *i, xtt x\\ there are two possibilities: change or no change. This gives altogether 21 8 possibilities. Combination with the six permutations of Xi, t, x* yields 8 X 6 48 operations. * This group of motions is well known in group theory and particularly in crystallogthe point further.))) raphy, but we shall not elaborate

296

ZERO-SUMFOUR-PERSONGAMES

of The point 7 correspondsto the values 1,1,1the coordinatesx\\, x z, Thus the characteristic function v(>S) of this game is:))

-1
2))

1
2
when

(35:1)v(S)))

-2 1

S has))
4))

elements

(and 4 belongsto S)

2 (and 4 doesnot belongto S) 3

(Verification is immediate with the help of (34:1), (34:2),(34:3)in 34.2.1.) l Insteadof applyingthe mathematical theory of Chapter VI to this game,et us first seewhether it does not allow an immediate intuitive interpretation.

Observefirst that a player who is left to himself losesthe amount 1. This is manifestly the worst thing that can ever happen to him sincehe 1 can protect himself against further losses without anybody else'shelp. 1 as completely Thus we may considera player who gets this amount defeated.A coalition of two pla3r ers may be consideredas defeated if it get I.2 3 getsthe amount 2, sincethen eachplayerin it must necessarily In this game the coalition of any two players is defeated in this senseif it does not compriseplayer 4. Let us now pass to the complementary sets.If a coalition is defeated in the above sense,it is reasonable consider the complementaryset to as a winning coalition. Therefore the two-elementsets which contain the player 4 must be rated as winning coalitions. Also since any player who remainsisolated must be rated as defeated, three-personoalitionsalways c win. This is immaterial for those three-element coalitionswhich contain the player 4, sincein these coalitionstwo members are winning already if the player 4 is among them. But it is essentialhat 1,2,3 a winning be t 4 s coalition, inceall its proper subsets aredefeated.
-

of the matter is corroboratedy our results concerning the three-person b and more fundamentally by our definition of the imputation in particularly condition 2 Since neither he nor his partner need acceptlessthan and they have together 2, this is the only way in which they can split. 8 In the Thereis of courseno gain, and this coalition is flat. terminology of therefore no possible motive for two players to form such a coalition. But if it happens that the two other players have combined and show no desireto acquire a third ally, we may treat the remaining two as a coalition even in this case. \" \" 4 We warn the reader that, although we have used the words defeated\" and winTheseconcepts indeed, are, ning\" almost as termini technici, this is not our intention. The \"defeated\" and \"winning\" coalitions very well suited for an exact treatment. those for w in :G)in actually coincide ith the sets considered (31 and in (31 which is flat or is flat, respectively. But we shall consider this question in such a way only in Chap.X. For the moment our considerations are absolutely heuristic and ought to be taken 22. in the same spirit as the heuristic discussions the zero-sum three-persongame in of a Theonly difference is that we shall be considerably briefer now, sinceour experiencend routine have grown substantially in the discussion. As we now possess exact theory of solutions for games already, we are under))) an

game

30.1.1., 32.2.,
in

1This view

23. and

(30:1).

1,

31.1.4.:

:F)

31.1.5.;

21.,

SPECIAL POINTSIN THE CUBE Q

297

So it is plausibleto view this as a strugglefor participation in coalitions: any one of the various possible (1,4),(2,4),(3,4),(1,2,3), where the amounts obtainablefor thesecoalitionsare: v((l,2,3))= (35:3) v((l,4)) = v((2,4))= v((3,4))= 2, (35:2) were:

35.1.2.

1.

Observe that this is very similar to the situation which we found in the essential zero-sum three-persongame, where the winning coalitions

(35:2*)

(1,2), (2,3), (1,3),

and the amounts obtainablefor these coalitions:

v((l,2)) = v((l,3)) = v((2,3))= 1. (35:3*) In the three-persongame we determinedthe distribution of the proceeds35:3*) mong the winners by assuming:A player in a winning coalia ( tion shouldget the same amount no matter which is the winning coalition. by ( Denoting theseamounts for the players1,2,3 a, $, y respectively, 35:3*)
gives

(35:4*)
from which follows

+ = a + 7 =:0+ 7 = l

a = ft = y = (35:5*) Thesewere indeedthe values which thoseconsiderations ielded. y

Let us assume the same principlein our present four-person game. t Denoteby a, 0,7, 5, respectively,he amount that eachplayer 1,2,3,4 gets coalition. Then (35:3) ives in if he succeeds participatingin a winning g

(35:4) (35:5)
All

a + 6 = + 6 = 7 + $ = 2, = =7 =
up:
*

a + + 7 = 1, =

from which follows

the heuristicarguments usedin 1 be repeated.

i22., i 21.,
for the

g three-personame could

35.1.3. Summing

(35:A)

This is a game in which the player 4 is in a speciallyfavored positionto win: any one ally sufficesfor him to form a winning coalition. Without his cooperation,on the other hand, three itself players must combine. This advantage also expresses in

obligation

to follow up this preliminary heuristic analysis by an exactanalysis which is basedrigorously on the mathematical theory. We shall come to this. (Cf.loc.cit. above, and also the beginning of 36.2.3.) 1 Of w course, ithout making this thereby a rigorous discussion on the basisof 30.1.)))

298))

GAMES)) ZERO-SUMFOUR-PERSON

should get when he is which each player 1,2,3,4 among the winners if our above heuristic deduction can be trusted. These amounts are respectively. It is to be noted that the advantage of player 4 refers to the caseof victory only; when defeated,all playersarein the same position

the amounts

i, i,

|,

(i.e. -1). get

The last mentioned circumstance of course,due to our normalizais, tion by reduction. Independentlyof any normalization, however, this game exhibits the following trait: One player's quantitative advantage over another may, when both win, differ from what it is when both lose. This cannot happen in a three-person game,as is apparent from the 2 T formulation which concludes 2.3.4. hus we get a first indication of an when the number of participantsreaches important new factor that emerges four. strategicadvantage consistedin the fact that he neededonly one ally for victory, whereaswithout him a total of three partners was necessary. form Onemight try to passto an even more extreme by constructinga game not containplayer 4 is defeated.It is in which every coalition that does t essentialo visualize that this is not so,or rather that such an advantage of in nature. Indeed such a game)) is no longer a strategic
One 35.1.4. last remark seems appropriate.In this gameplayer 4's

1
2
3))

if

8 has))

a elements nd 4 does not belongto

S,))

hence))

1
if

S has))

2 a elements nd 4 belongsto o
4))

S.

This is not reduced, s a

v((I)) = v((2)) v((3)) = -1,)) v((4)) = 3.)) If we apply the reduction processf 27.1.4. this v(S) we find that its o to reduced form is
0.))

in

In other words:Player 4'sadvantage in this gameis one of a fixed of The payment (i.e. cash),and not one of strategic possibilities. former of course, more definite and tangiblethan the latter,but of less theoretical is, interestsinceit can be removedby our processf reduction.))) o

t i.e.he game is inessential. (This couldhave beenshown directlyby (27:B) 27.4.)Thus this gamehas a uniquely determinedvalue for eachplayer 1,2,3,4: , 1, 1,3,respectively. 1

SPECIAL POINTSIN THECUBE Q

299 the corners

VI, VII differ from J only by permutationsof the players. It is easily role verified that the special of player 4 in / is enjoyedby the players1,2,3, in V, VI, VII, espectively. r Gameand a \"Dummy The 35.2. Corner VIII(and //,///,77). The Three-person
V,
91

35.1.5. observedat the beginning of this sectionthat We

o://,///,IV,VIII. As they arisefrom eachotherby suitablepermutations it sufficesto considerone of them, say VIIL of the players 1,2,3,4, t Thepoint VIII correspondso the values 1, 1, 1of the coordinates
function v(S) of this xi,x 2, x*. Thus the characteristic game is:))

We 35.2.1. nextconsiderthe gameswhich correspondto the four corners

-1 -2))
(35:6) v(S)
=))

1
2
when

S has))
4))

elements

(and 4 belongsto S)

2 (and 4 doesnot belongto 5) 3

i (Verification is immediatewith the help of (34:1), (34:2),(34:3)n 34.2.1.) instead of applying to this game the mathematical theory of Chapter Again, VI, let us first see whether it does not allow an immediate intuitive

interpretation. The important feature of this game is that the inequality

: an becomes equality, i.e. 25.3. v(S u T) = v(S) + v(T) (35:7)

when T

in (25:3:c)

if

1 by expressed (35:7).

a If S represents coalition which doesnot contain That is the player 4, then the addition of 4 to this coalition of no advantage; it does not affect the strategicsituation of this coalition nor of its opponents in any way. This is clearly the meaning of the additivity

= (4).

is:

S n T = 0,

i.e.

This 35.2.2. circumstance suggests the following conclusion, which 2 o is of coursepurely heuristic. Sincethe accessionf player 4 to any
1Note that
the indifference
in acquiring

the

and not by

That is, a player is \"indifferent\" of a coalition but if he brings into the coalition exactly the he is worth outside. than This remark may seemtrivial; but there existsa certain in non-reduced games where v((4)) > 0, particularly

v(S U T) vGS). asa partner, not if his accessionoesnot alter the d

amount which danger i.e. here w

cooperationof 4 is expressedy (35:7), b

value and no more

the (although strategically indifferent!) actually increases value of a coalition. a Observe lsothat the indifference of and T (4) to eachother is a strictly reciprocalrelationship. 1 We shall later undertake exactdiscussion on the basisof At that time it will be found also that all these games are specialcasesof more general classesof some importance. (Cf.Chap.IX, particularly 41.2.))))

of misunderstanding, of the accession 4

30.1.

300

ZERO-SUMFOUR-PERSONGAMES

coalition appears to he a matter of completeindifference to both sides, it seemsplausibleto assumethat player 4 has no part in the transactionsthat constitute the strategy of the game. lieis isolated from the others and the amount which he can get for himself v(/S) = 1 is the actual value on of the game for him. The other players 1,2,3, the other hand, play the game strictly among themselves; hence they are playing a threefunction v(S) person game. The values of the original characteristic which describesthe original three-personame are g
:))

=v((3)) = -1,)) = v((l,3)) = v((2,3))= 2, (35:6*) v((l,2)) v((l,2,3))=

1,))

is /' = (1,2,3)now the set

of all players.))

(Verify this from

At

function the new characteristic

systematically. (Cf. footnote 2 on p. 299.) It is obvious, however, that this condition can be remediedby a slight generalization of the transformation used in 27.1.We modify the game of 1,2,3 assuming that each by for one got the amount in cash in advance, and then compensating this amounts from the v(S) values in (35:6*). Just by deducting equivalent this as in 27.1., cannot affect the strategy of the game,i.e.it producesa l strategicallyequivalent game. After considerationof the compensations mentionedabove2 we obtain

first sight this three-personame represents the oddity that v(/') g is now the set of all players!)is not zero. This, however, is perfectly reasonable: eliminating player 4 we transform the game into one which by is not of zero sum; sincewe assessed the others retain player 4 a value a value We do not yet propose to deal with this situation together

(35:6).)

(/'

1.

1,

v'((l)) = v'((2)) = v'((3)) = *, (35:6**) v'((l,2)) = v'((l,3)) = v'((2,3))= *, v'((l,2,3))= 0. This is the reduced form of the essentialzero-sum three-person amedisg f cussedin 32. exceptor a difference in unit : We have now y = $ instead
1In the terminology we have infringed is
zero-sum game. Even 2^ ai
putting
\"

v'(0)= 0,

of

(27:1): aj = 0.
t

27.1.1.: al a? =

al

This is necessary sincewe started

if we included

|.

The condition

there which with a non-

could

be safeguarded
leave him

player

4 in

our considerations, the

compensation would make v((4)) = 0, with results which are obvious. Onecan sum this up by saying that in the present situation it is not the reduced form of the game which provides the best basis of discussion among all strategically equivalent forms. 2 deduction of as many times J from v(S) as8 has elements.)))

aj =

1. This would

just

as isolated as before, but

necessary

I.e.

SPECIALPOINTS THECUBEQ IN

301

in of the 7 = 1of (32:1) Thus we can apply the heuristicresults l or the exactresults of 32. Let us restrict of ourselves,at any a rate,to the solution which appears in both cases nd which is the simplest of Thisis the setof imputations(32:6) one:32:B) 32.2.3. in which ( we must multiply by the present value of y = ;

23.1.3.,

32.1.1.

i -1.
1,
(35:B)

I (Theplayers are,of course, ,2,3.)n otherwords:Theaim of the strategy 1 of the players 1,2,3 to form any coalition of two; a player who succeeds is who is victorious, getsf , and a player who is defeated gets $. in this, Now eachof the players 1,2,3 our original game gets the extraamount of hence the above amounts f , must be replaced y beyond this, b

ft, -it!, (-*,*,!}, lit, -*}

i.e.:

32.2.1.,

i.e.

36.2.3. Summingup:

inflated by the addition of a \"dummy\" player 4. We shall seelater 1,2,3, that this concepts of a more generalsignificance.(Cf. footnote 2 on i

This is a game in which the player 4 is excluded from all coalitions. The strategicaim of the other players 1,2,3 to is form any coalition of two. Player 4 gets 1at any rate. Any otherplayer 1,2,3etsthe amount 1when he is among the wing ners,and the amount 1when he is defeated.All this is based on heuristicconsiderations. One might say more conciselythat this four-persongameis only an \"inflated\" three-personame: essentialhree-personame the players the g g of t

compare the dummy roleof player 4 in this game in with the exclusion playerundergoes the discriminatorysolution (32: a A) in 32.2.3., Thereis, however, an important differas discussedin 33.1.2. encebetween thesetwo phenomena. In our presentset-up,player 4 has really no contribution to make to any coalition at all;he stands apart by function v(S). Our heuristic considerations virtue of the characteristic in from all coalitions all acceptable indicatethat he should be excluded t solutions. We shall see in 46.9.hat the exacttheory establishesjust this. The excludedplayer of a discriminatorysolution in the senseof 33.1.2. is excluded only in the particularsituationunder consideration.As his far as the characteristic function of that gameis concerned, roleis the same as that of all other players. In otherwords:The \"dummy\" in our facts of the situation presentgameis excluded virtue of the objective by 2 The excluded function v(S)). player in a discrimi(the characteristic natory solution is excludedsolely by the arbitrary (though stable) that \"prejudices\" the particular standard of behavior (solution)expresses.
1

p.299.)

One 36.2.4. might

Of coursethe present discussion is heuristic

ment,

2 This is the

cf. footnote 2 on p. 299. \"

in any

event.

As

to the exacttreat-

physical background,'

in the

senseof 4.6.3.)))

302
We

ZERO-SUMFOUR-PERSONGAMES

observedat the beginning of this sectionthat the corners 777, //, from VIIIonly by permutations of the players. It is easily role verified that the special of player 4 in VIIIis enjoyedby the players in 1,2,3 77,777,77, respectively.

IV differ

of Q to the game which corresponds the centerof Let us now consider of the coordinates } 2, g. This game is clearly to the values 0,0,0 x\\ Q, i.e. unaffected by any permutation of the players 1,2,3,4, it is symmetric. in Q, since total symmetry means Observethat it is the only such game o in variance under all permutationsof x\\, 2, x* and sign changes f any two = x a = x s = 0. hence of them (cf. 34.3.); Xi function v(S) of this game is:)) Thecharacteristic
Concerning the Interior

Some 85.3. Remarks

35.3.1. i.e.

-1
(35:8) v(S)
=))

1 when

S has

2 elements. 3
4))

any (totally) symmetricgame is clearly that any majority of the players(i.e. when two coalitions coalition of three) wins, whereasin caseof a tie (i.e. form, eachconsistingof two players)no paymentsaremade. Thecenterof Q represented only (totally) symmetric game the with to in our set-up: respect all permutations of the players 1,2,3,4. The considerationof another symmetry as well: p geometrical icture suggests to with respect all permutationsof the coordinates x^ x 8 In this way x\\ t we selecthe points of Q with

(34:2),(34:3)in 34.2.1.) (Verification is immediatewith the help of (34:1), The exactsolutionsof this game are numerous;indeed, one must say that variety. It has not beenpossibleyet to they are of a rather bewildering order them and to systematizethem by a consistent pplicationof the exact a theory, to such an extent as one would desire. Neverthelessthe known instructive insight into the ramifications of the theory. give specimens some We shall considerthem in somewhat more detail in 37.and 38. remark:The idea of this At present we make only this (heuristic)

35.3.2.

(?5:9) (35:10)
In 1,2,3. otherwords:
1This

xi = xi = si,
7-center-y777.

which form a main diagonal of Q, the line We saw at the beginning of that the gameis invariant with
shows representation this property.

34.3.1. this symmetry means precisely that

t respecto all permutationsof the players
and uniquely

characterizedy b

oncemore that the game is symmetric, Cf.the analysis of 28.2.1.)))

SPECIAL POINTSIN THECUBE Q

303

The main diagonal (35:9), representsall those gameswhich are (35:10) t i.e. symmetric with respecto the players 1,2,3, where only player 4 may have a specialrole. Q has threeothermain diagonals(//-center- ///-center- /,/F-center F F, F//),and they obviously correspondto those gameswhere another player alone may have a specialrole. respectively) (players1,2,3, Let us return to the main diagonal (35:9), Thethree games (35:10). which we have previously considered(/,F///, Center) on it;indeed in lie 1 all thesegamesonly player 4 had a specialrole. Observethat the entire such category of games is a one-parameter variety. Owing to (35:9), a by game is characterized the value x\\ in

(35:11)

-1g

xi

1. 1,

The threegamesmentioned above correspond the extremealues x\\ = to v 1 and to the middlevalue Xi = 0. In order to get more insight into Xi = the working of the exact heory, it would be desirableto determineexact t solutionsfor all thesevalues of Xi, and then to seehow thesesolutionsshift It as x\\ varies continuously along (35:10). would be particularly instructive to find out how the qualitatively different kinds of solutionsrecognized v for the special alues x\\ = 0, 1 go over into eachother. In 36.3.2. we shall give indicationsabout the information that is now available in this

1,

regard.

Another question of interest is this: onsider a game, C a point in Q, where we can form some intuitive picture of what solutionsto e t expect, .g.he corner F///. Then consider a game in the immediate of o neighborhood F///, i.e. ne with only slightly changedvalues of Xi, s x 3 Now it would be desirableto find exact olutionsfor theseneighboring games,and to see in what details they differ from the solutions of the how a small distortion of x\\, xj, x s distorts the soluoriginal game, 2 and at c of tions. Special ases this problem will be consideredin as well as in 38.2.7. the end of 35.3.4. far we have consideredamesthat arerepresentedy points So b g 8 a of Q in more or less special positions. A more general, nd possiblymore typical problem ariseswhen the representative point X is somewhere in in the interior of Q, in \"general\" osition, i.e. a position with no p particular distinguishing properties. Now one might think that a good heuristiclead for the treatment of the problemin such points is providedby the following consideration.We have someheuristic insight into the conditionsat the corners/-F/// (cf. and 35.2.).ny point X of Q is somehow\"surrounded\" these by A more corners; precisely,it is their centerof gravity, if appropriateweights 1In the centernot even he. 1This procedure familiar in mathematical physics, where it is used in attacking is problems which cannot be solvedin their general form for the time being:it is the analysis

35.3.3.

i.e.

x,

i.e. 37.1.1.,

36.1.2.,

35.1.

of perturbations.
8

t Corners,he center,and

entire main

diagonals.)))

304

GAMES ZERO-SUMFOUR-PERSON

one areused. Hence might suspectthat the strategy of the games,repreo sentedby X, is in some a combination of the strategiesf the (more way of by -VI familiar) strategies the games represented 1 II. One might even be hope that this \" combination \" will in somesense similar to the formation 1 X of the centerof gravity which related to I-VIII.
that and shall seein 36.3.2. in 38.2.5.7. this is true in limited parts of of Q, but certainly not over all of Q. In fact, in certaininterior areas Q different from anything exhibited phenomenaoccurwhich arequalitatively in c by I'VIII. All this goesto show what extreme aremust beexercised a or in making guesses bout them. dealing with notions involving strategy, The mathematical approachis in such an early stageat present that much in will more experience be neededbefore one can feel any self-assurance this respect.
We

these as vectorsin L8 and

gravity tX

The 36.1.1. systematic theory of the four-person game has not yet list advancedso far as to furnish a complete of solutionsfor all the games b representedy all points of Q. We arenot able to specifyeven onesolution for every such game. Investigations thus far have succeedednly in o one, determiningsolutions (sometimes sometimes more)in certainparts of -VI Q. It is only for the eight corners1 II that a demonstrablycomplete A list of solutionshas been established. t the presentthe parts of Q in which solutionsare known at all form a rather haphazard array of linear, plane and spatial areas. They aredistributed all over Q but do not fill it out completely. The exhaustive list of solutionswhich areknown for the corners -VIII I can easily be established with help of the results of Chapters IX and X, where thesegames will be fitted into certainlargerdivisionsof the general theory. At presentwe shall restrictourselvesto the casuistic approach 1 Consider two points X = (t/i, y^ 2/s) in Q. We may view |a?i, z2, xt and Y
\\

of 36.Discussion the Main Diagonals The 36.1. Part Adjacent to the Corner VIII.:Heuristic Discussion

it

is indeed in

this

sensethat

the formation

of a centerof

- - + (1- + (1is understood. (Cf. (16:A:c) 16.2.1.) [xi, z , z| sense (34:l)-(34:3) v(S) w(S) 34.2.1., same a characteristic + (1u(S)s
+ (1
t)Y (toi
Ol/i, tx* in

Oyt,

to,

+ (1

<)y,}

the

Now if X in the and algorithm,

and Y

of

\\yi, y^ 2/3) define the in then tX

/)wOSf).

function

+ (1

characteristic functions
t)Y will give, by

tv(S)

27.6.3.

(It is easy to verify this And this same u(S)was

relationship introduced

by inspection of the formulae which we as centerof gravity of v(S) and by

w()

in (27:10)

quoted.)

Thus the considerations of the text are in harmony with those of 27.6. That we with centers gravity of more than two points (eight: of instead of only t two, is not essential:he former operation can be obtained by iteration of the latter. follows from theseremarks that the difficulties which are pointed out in the text below have a direct bearing on as was indicated there.)))

are dealing

I-VIII)

It

27.6.3.,

which

known.

accountof precise instances which,

it

purpose of this expositionto give a the momentary state of these investigations 1 and it would take up an excessivemount of space. We shall only give some a

OF DISCUSSION THE MAIN DIAGONALS)) 305)) consists in describingparticular solutions in cases where such are

It would scarcelyserve the

36.1.2. considerfirst conditionson the main diagonal /-CenterWe VIII in Q nearits end at VIII, = x 2 = x 3 = 1(cf. 35.3.3.),we shall and
x\\
try))

is hoped, arereasonablyillustrative.

The diagonal
VIII))

Center
Center))

VIII

redrawn

Figure

63.

to extendover the On this diagonal))

x\\

= #2 =

> 1as far as possible.(Cf. Figure63.) 1

2

_^
2*!))

(36:1)))

when

S has))
4))

elements

(and 4 belongsto S)

2 (and 4 doesnot belongto S) 3

for in i for (Observethat this gives (35:1) 35.1.1.xi = 1and (35:6)n 35.2.1. = 1.) We assume that x\\ > 1 but not by too much, just how x\\ is later. Letus first considerthis much excess to be permitted will emerge situation heuristically. t SinceXi is supposedtobe not very far from 1, he discussionof 35.2. A coalition of two players from among)) may still give someguidance.
1This will be done by one of us in
subsequent

mathematical publications.)))

ZERO-SUMFOUR-PERSON 306 GAMES 1,2,3ay even now be the most important strategicaim, but it is no longer m the only one: formula (35:7) 35.2.1. true, but instead the of is not if SnT= v(S u T) > v(S) + v(T) (36:2)

= (4).1 Indeed,t is easily verified from (36:1) this excess i that is always2 2(1 1). Forx\\ = 1this vanishes,but we have x\\ slightly +
when T

be only slightly < 4. Thus the first coalition (between two players, other than player 4), is of a much strongerexturethan any other(whereplayer 4 entersinto the t picture), but the latter cannot be disregarded nevertheless.Sincethe first coalition is the stronger that one,it may be suspected it will form first and that onceit is formed it will act as one player in its dealingswith the to two others. Then some kind of a three-personame may be expected g take place the final crystallization. for 36.1.3. e.g. 1,2)or this \"first\" coalition, the surmisedthree( f Taking, persongame is betweenthe players (1,2), 4 In this game the a, 6, c of 3,4.

so the expression slightly > 0. Observethat for the preceding is > 8 in coalition of two players other than player 4, the excess (36:2) is by = 1 this is 4, and as we have x\\ slightly (36:1) 2(1 Xi). Forx\\ always

1,

> 1,it will

a 23.1.re

a = v((3,4))= 2xl9 b = v((l,2,4)) = c = v((l,2,3))= 6 if Hence, we may apply the results obtained there (all of this is extremely

1,

I.

= 1 x\\, the ^ heuristic!) player (1,2) ets the amount a = g & if successful in joining the last coalition), and a = 2xi if defeated.The ( = x\\ if successful,and b = 1 player 3 getsthe amount = ^ z
if

Since \"first\" coalitions (1,3),2,3) may form, just as well as (1,2), ( there are the sameheuristic reasons in the first discussionof the threeas t person game (in 21. to expecthat the partners of thesecoalitionswill -22.) split even. Thus, when such a coalition is successful (cf. above), its members may be expected get -^ 1 each, and when it is defeated to &
Xi

defeated.The player 4 getsthe c = 1 if defeated. ful, and

amount 7

ft

= x\\

if

success-

each.

follows:

36.1.4. the Summing up:If thesesurmisesprove correct, situation is as

I.e. the presnatural

persons. *
negative;

1 Unless S * in (36:2). in or T, in which casethere is always ent situation S must have one or two elements. 2 By footnote 1 above,Shas oneor two elementsand it doesnot contain 4. 8 now S, T are two one-elementsets,not containing player 4. 4 Onemight say that i (1,2)s a juridical person, while 3,4are, in our picture,

I.e.,

In all the

hence

formulae which follow, remember that Xi Xi is a gain, and x\\ is a loss.)))

is near to

1, i.e. presumably

DISCUSSION THE MAIN DIAGONALS OF

307

If the \"first\" coalition is (1,2), if it is successfuln finding an ally, i and and if the player who joins it in the final coalition is player 3,then the 1_ 1 .T ~ players 1,2,3,4 the amounts 2i S ~ S Xi, 1respectively. If the get i player who joins the final coalition is player 4, then these amounts are 1~ X S is If the \"first\" coalition (1,2) not XL replacedy \"I Xl , b
/*

&

&

-1,

i if successful, .e. the players3,4 combineagainst it, then the playersget the amounts Xi, xi, Xi, x\\ respectively. If the \"first\" coalition is (1,3) (2,3), hen the corresponding t or permutation of the players 1,2,3ust be applied to the above. m
heuristicsuggestionmanifestly correspondsto the following surmise: Let V be the setof theseimputations:
~~>,

36.2. Part Adjacent to the Corner VIII. Exact Discussion : The 36.2.1.is now necessaryto submit all this to an exactcheck. The It

(l - si' 1 -)) xi
2
x

(36:3) a\"

>

=))

J
I
\\

x\

' ~~*

Xl

an d the imputations which originate from theseby perthe)) muting the players, (i.e.

components) 1,2,3.))

that (Cf. footnote 5, p.306.) We expect this V is a solutionin the rigorous senseof 30.1.Xi is near to 1and we must determine if whether this is so> and preciselyin what interval of the x\\. This determination,if carried out, yields the following result: The set V of (36:3) a solution if and only if is (36: A)

-1^

xi

-*.

This then is the answer to the question, how far (from the starting point i the the xi = 1, .e. cornerVIII) above heuristicconsiderationguidesto
1 correctesult. r

significant technical ifficulty. d

Xi) the

The 36.2.2. proof of (36: A)

1 We wish to emphasize that only solution of the game in question. However, attempts with numerous in further solutions for x\\ ^ J (i.e. the range of similarly built setsfailed to disclose (36:A)). For x\\ slightly > (i.e.lightly outside the range of (36:A)), where the s it. V of (36: ) is no longer a solution, the same is true for the solution which replaces A

can be carried rigorouslywithout any out It consistsof a rather mechanical disposal A (36: ) doesnot assertthat V is (in the specified range of

the finite solutions V which are now under consideration. Theseare the arguments which seemto justify our view that in the nature of the solutions occurs at
x\\
G

We do not question, of course, hat other solutions of the \"discriminatory\" type, t as repeatedly discussed efore, lways exist. But they are fundamentally different from a b

in Cf. (36:B)

36.3.1.

somequalitative change

(on the diagonal

/-center- III).))) V

308

GAMES ZERO-SUMFOUR-PERSON

and o of a seriesf special cases, doesnot contributeanything to the clarifica1 tion of any questionof principle. The reader ay therefore omit reading m w it if he feelsso disposed,without losingthe connectionith the main course of the exposition. He shouldrememberonly the statement of the results

T Neverthelesswe give the proof in full for the following reason: he set w (36:3)was found by heuristic considerations,i.e. ithout using the at exacttheory of 30.1. all. The rigorous proof to be given is based on 30.1. and thereby brings us back to the only ultimately satisfactory alone, standpoint, that of the exacttheory. The heuristic considerationswere a only a deviceto guess the solution,for want of any bettertechnique; nd it is a fortunate feature of the exactheory that its solutionscan occasionally t be guessedin this way. But such a guess must afterwards be justified in m by the exact ethod, or rather that method must be used to determine the guesswas admissible. what domain (of the parameters involved) t We give the exact proof in order to enable the readero contrast and tocompareexplicitlythesetwo procedures,the heuristicand the rigorous. 36.2.3. proof is as follows: The a c If x\\ = 1,hen we arein the corner t VIII,nd the V of (36:3) oincides with the set which we introduced heuristically(as a solution) in 35.2.3., and which can easily be justified rigorously(cf.also footnote 2 on p.299). we Therefore disregardthis case now, and assumethat
V of

in

(36:A).

(36:4)

xi >

-1.

are We must first establish which sets (1,2,3,4) certainly since necessary or certainly unnecessaryin the senseof 31.1.2. we are carryingout a proof which is preciselyof the type consideredthere. The following observationsareimmediate:

Ss/=

(36:5)
(36:6)

three-element S arecersets :H) By virtue of (31 in 31.1.5., necessary, two-elementsetsare dubious, and all other 2 setsarecertainly unnecessary. Whenever a two-element set turns out to be certainly necessets sary, we may disregardall those three-element of which it is
tainly

a subset,owing

to

(31C)in 31.1.3. :

Consequentlywe shall now examine the two-elementsets.This of course must be done for all the a in the set V of (36:3).
1Thereadermay contrast this proof with somegiven in connection with the theory of the zero-sum two-person game, the combination of with Such a proof is more transparent, it usually covers more ground, and gives some qualitative elucidation of the subjectand its relation to other parts of mathematics. In somelater parts of this in 46. But much of it is still in the primitive theory such proofs have beenfound, and technically unsatisfactory state of which the considerationswhich follow are typical. * This is due to n

e.g.

16.4. 17.6.

e.g.

4.)))

DISCUSSION THE MAIN DIAGONALS OF 309 Consider first those two-elementsets S which occur in conjunction with a V As aj = 1 we may exclude (31 in 31.1.3.possibility the by :A) that S contains 4. S = (1,2) ould be effective if a{+ a' ^ v((l,2)), w 2 1 i.e. xi ^ -2xi, i ^ which is not the caseby (36:4). = (1,3) x S
>

-1

is effective if a( + the condition

^ v((l,3)), i.e.
Xl

(36:7)
which we assumeto

^-^ ^ -i
46

-2*!,
Xl

-f

Thus

be satisfiedmakesits first appearance. = (2,3)we S do not need,since1and 2 play the same rolein a (cf.footnote 1above).
t passto a \". As a' = 1we now excludehe S which contains3 8 i (cf.above). S = (1,2)s disposedof as before,since a and a \" agree in
We now

these components.S = (1,4) would be effective if a\" + a\" g v((l,4)), i.e. ^ 2xi,xi ^ which, by (36:7), not the case. S = (2,4)is is

i.e.

discardedin the same way. r Finally we take ~Z \". S =

2xj. =

'

'

'

reason: e W

and 3 (cf.footnote

= v((l,2)) n 2xi. S = (1,3) eed not be considered the following for are already consideringS = (1,2) a '\",if we interchange for 2
is (1,2)
effective:
\"

a'/'+ a' 2

-Xi, -XL Our original S = (1,3)for a'\"with the components-Xi,Xi is thus renderedunnecessaryby (31 in 31.1.3., x t ^ xi owing to as :B) (36:7).S = (2,3) is discarded in the same way. S = (1,4) would be \" effective if a'/'+ a' g v((l,4)) i.e. g 2xb xi ^ 0, which, by (36:7),is 4 not the case. S = (2,4)is discardedin the same way. S = (3,4)is effective:ai\" + a'/'= v((3,4)),i.e.xi = 2Xl . 2
Summingup:

1above) this goesover into (1,3), with

the components

(36:8)

Among the two-element sets S the three given below are certainly necessary,and all others arecertainly unnecessary:

31.1.3.may exclude we ' and 3 for a \". Consequentlyonly (1,2,3)left those containing 4 for a is for and (1,2,4) a\". Of these the former is excluded (36:6), for by as it contains the set (1,3) f (36:8). or ^'\"every three-element o F set

t sets Concerninghree-element S:By (31:A)in

for and (1,3) ^',(1,2) (3,4)for T'\".

'

1 Here,and in the entire discussion which follows, we shall make useof the freedom asstatedin (36:3),n order to abbreviate the argumentation. i apply permutations of Hencethe readermust afterwards apply thesepermutations of to our results.)))

1,2,3

to

1,2,3

310

GAMES ZERO-SUMFOUR-PERSON

it or hence contains the set (1,2) the set (3,4)of (36:8); we may exclude by

(36:6).

Summingup :

(36:9)

s the three-element ets <S, the one given below is 1 and all others arecertainly unnecessary: certainlynecessary,
Among

We 36.2.4. now verify

for (1,2,4) 7\". in i.e. (30:5:a)30.1.1.,that no a of V dominates

any
with

1,2,3ermutation of a ' or a \" or a '\"? This requiresfirst p of the existence a component < x\\ (this is the 3 component of a. ') among the 1,2,3omponentsof the imputation in question. Thus a' and a'\" c
this S any but the 3 componentwill do. But now another one of the 1,2,3 components 1 \" - (this is the 1componentof a ') of this imputation a must be < ~ Z
t and this is not the case;he a.
with 2 are areexcluded. Even in a \" the 1,2 (cf. components excluded footnote 2)

^ = a ':By (36:8), we must useS = (1,3).Can a ' dominate (36:9)

of V.

componentsof 1,2

a. \"

areboth =

-1

&

that the 4 componentof the imputation in question be < x\\ (this is the 4 componentof a\-") Thus a \"and a \"' areexcluded.For a 'we must

= a \": y (36:8), we must useS = (1,2,4). a \" dominate Can (36:9) B ' or a \" or a '\"? This requiresfirst this S any 1,2,3 permutation of a

componentsbe requirefurther that two of its 1,2,3

<

-^ (this is the 1

as well as the 2 componentof a \,") and this is not the case; nly one of these o is components 5* ~
*\"

1?'\": (36:8), we must useS = (1,2) then S = (3,4). and (36:9) By :Can above? This requires 5 = (1,2) a'\"dominatewith this as described
\"a*

of the existence two components < zi (this is the

~\"*

1as well as the 2 com-

a componentsof the imputation in question. ponent of a'\") mong the 1,2,3 a This is not the casefor a'\", s only one of thesecomponentsis j& x\\
1 As by every three-element set is certainly necessary (36:5)above,this is another instance of the phenomenon mentioned at the end of footnote 1 on p. 274. 1Indeed rand -x\\ x\\, Xi both by (36:7). x\\ a?i,

5^

i.e.

V)))

i.e.

DISCUSSION OFTHE MAIN DIAGONALS

* *

311

there. Nor is it the casefor a ' or a \", as only one of those componentsis * 1 x\\ 1 ?* s there. S = (3,4):Can a '\" dominate with this S as described t
above? This requires first that the 4 component of the imputation in > '\" questionbe < Xi (this is the 4 componentof a \"'). Thus a \" and a are

excluded.For a ' we must require further

the

of existence a component

< x\\
this

(this is the 3 componentof a '\") among its 1,2,3omponents,and c is not the case; ll thesecomponents ^ Xi (cf.footnote 2 on p.310). a are This completes verification of (30:5:a). the
ft

is undominated by the elementsof V must belongto V. of Consider a ft undominated by the elements V. Assume first that 04 < Xi. If any one of 0i, 2, 03 were < Xi, we couldmake (by permuting H 3 1,2,3) < XL This gives 7'\"H with S = (3,4)of (36:8). ence

36.2.5. verify next (30:5 in 30. i.e. an imputation We 1.1., that :b)

which

7
1

ft i,

02,03 ^ ~

XL

If any two of 0i, 2, 03 were

<

, we couldmake (by permuting 1,2.3)))

with

0i,02 < ~V^' 2

This gives a \" H

S=

of H (1,2,4) (36:9). ence,

Xl

is at most one of 0i, 2, 3 i we 1,2,3, can thus make))

<

~- i.e.wo are ^ t

By permuting))

Clearly

' ponent of a , and sinceboth areimputations 2 it followsthat they coincide: = a ', and so it is in V. Assume next that 4 ^ x\\. If any two of 0i, 2, 0s were < Xi, we * > H with couldmake (by permuting 1,2,3) 2 < XL This gives a\"' 0i, two at of S = (1,2) (36:8). ence, most one of 0i, 2, 0s is < Xi, i.e. are H w ^ XL By permuting 1,2,3e can make
0 01,2 S -Xi. If 0s ^
1And

1. Thus eachcomponent of

comis ^ the corresponding

*i,then

all this impliesthat each componentof

xi -xi,i.e.

is ^ the cor-

2 Consequently

i-^1

-1.
of all components is the same:zero.)))

for both the sum

312

GAMES ZERO-SUMFOUR-PERSON

both areimputations(cf.footnote 2, a o responding component f a '\", nd since a on p.311) follows that they coincide: = a '\", nd so it is in V. it ft Assume therefore that fa < XL If any one of 0i,fa were < ~ \"'
we could make (by permuting 1,2) fa = (1,3) (36:8). with of Hence

<I

^ &

x\\

This gives

* a'H

*
ft

A, A ^ Clearly
8

^J-))
ft

1. Thus eachcomponentof

comis g: the corresponding

2,
it p.311),

ponent of

a\",and sinceboth are imputations (cf. footnote

followsthat they

coincide: = a \", and so it is in V. ft 1 This completes verification of (SOiSib). the So we have establishedthe criterion(36: 2 A).
When 36.3.1.

i.e. passes outside the domain (36 of 36.2.1.,when :A) t then the V of (36:3)id.ceaseso be a x\\ = solution. It is actually possibleto find a solutionwhich is valid for a certain domain in x\\ > (adjoiningx\\ = i),which obtains by adding to the the further imputations V of (36:3)
Xi

36.3. Parts of the Other

Main

Diagonals

it

i at crossests border

i,

(36:10)a lv =

j^-1, xi,

1 -,

xi\\

and permutations as
3 (36:3).

in))

The exact tatementis actually this: s The set V of (36:3)and (36:10)a solutionif and only if is (36:B)

- i < x, <

O.4

1The readerwill observe that

have
\"7\"'
f

in the

beenused for
of (36:3).

dominations,

and

courseof this analysis all setsof (36:8),(36:9) had to be equatedsuccessively all three a a \", to

',

proof. J, this goes this ( (1,3)and with it (2,3))is no longer effective for a Ofcourse wrong: Theset S which was excluded solely because(1,3) rehabilitates the three-element set = (1,2,3) (and (2,3))is contained in it.
a-t

Concerning zi 3 An inspection

of

the

1,cf. the remarks made above shows that

proof

the beginning of this when x\\ becomes>

'.

m .Thus domination by this element of V, a now becomes ore difficult, and it is of in therefore not surprising that an increase the set V must be considered the search for a solution. 4 Observe the discontinuity at x\\ A J which belongs to (36: ) and not to (36:B)l Theexacttheory is quite unambiguous, even in such matters.)))

',

THE CENTER AND ITSENVIRONS

313

The proof of (36:B)is of the same type as that of (36: given above, A) and we do not proposeto discussit here. The domains (36:A) and (36:B) exhaust the part xi ^ of the entire available interval

36.3.2. Solutionsof a nature similar to V described (36:A) of in and in (36:B) of have been found on the other side x\\ > the half Center-/ f the diagonal as well. It turns out that on this half, o qualitative changesoccurof the same sort as in the first half coveredby (36: and (36:B).Actually three such intervals exist, amely: A) n

V/77-Center-/.

1 g x\\ g 1 i.e.he half F/77-Center the diagonal of t

36.3.1.,

36.2.1. i.e.

(36:C)
(36:D)

(36:E)
VIII

(Cf. Figure64, which is to be comparedwith Figure63.)

Omtw

^ xi < i, i < xi g *, 1.

g Xl g

J
4
i

*.

- -i

*))

-*
Figure

o *

64.

1 discussthe solutionspertaining to (36:C),36:D), :E). (36 ( = appears as belonging The readermay however observe this:\\ x to both (neighboring) omains (36:B) and (36:C), nd similarly xi = to a d both domains (36:D) and (36:E). This is so because, a close as inspection in of the corresponding solutionsV showsthat while qualitative changes the a nature of V occurat x\\ = and thesechanges renot discontinuous. The point Xi = on the other hand, belongs to neither neighboring domain (36:C)or (36:D). It turns out that the types of solutionsV which the are valid in thesetwo domainsareboth unusable at x\\ = Indeed, conditionsat this point have not beensufficiently clarified thus far. We shall not

i,

i,

i.

37.TheCenterand ItsEnvirons
Center to The considerationsof the last sectionwere restricted a onedimensionalsubset of the cube Q:The diagonal FJ/7-center-/.By using this as the permutations of the players 1,2,3,4, describedin 34.3., can be made to disposeof all four main diagonalsof Q. By techniquesthat are s similar to thoseof the last section, olutionscan also be found along some other one-dimensionalines in Q. Thus there is quite an extensivenet of l lines in Q on which solutionsare known. We do not proposeto enumerate t them, particularlybecausehe information that is available now corresponds probably to only a transient stateof affairs.
Orientation Concerning the Conditions around the

37.1.1.

$7.1. First

in discussed

1Another

family and C 38.2. f.insolutions, 38.2.7., footnote particular

of

which

alsocoverpart of the same territory, 2 on p.

328.)))

will

be

314

ZERO-SUMFOUR-PERSONGAMES

such This, however, shouldbe said: a searchfor solutionsalong isolated one-dimensional body of the cubeQ lines,when the whole three-dimensional waits for elucidation,cannot be more than a first approachto the problem. If we can find a three-dimensional art of the cube even if it is a small p one for all points of which the same qualitative type of solutionscan be ideaof the conditionswhich areto be expected. used, we shall have some Now such a three-dimensional art exists round the centerof Q. Forthis a p reason we shall discussthe conditionsat the center. of 37.1.2. centercorrespondsto the values 0,0,0 the coordinates The the as x\\, #2, #3 and represents, pointed out in 35.3.1.,only (totally) symfunction of this game is metric game in our set-up. The characteristic
:))

-1))

(37:1)

v(S)

=))

when

S has

2 elements. 3
4))

i casesn 35.1., 36.1., begin 35.2., we id.) As in the corresponding (Cf. (35:8) with a heuristicanalysis. again efforts This game is obviously one in which the purposeof all strategic c is to form a three-personoalition. A player who is left alone is clearlya loser, ny coalition of 3 in the same sensea winner, and if the game should a terminate with two coalitionsof two players eachfacing eachother,then this must obviously be interpreted as a tie. T The qualitative question which ariseshere is this: he aim in this is probablethat in the negotiations game is to form a coalition of three. It the which precede play a coalition of two will be formed first. This coalition with the two remaining players, trying to securethe will then negotiate o cooperationf one of them against the other. In securingthe adherence of this third player, it seems uestionablewhether he will be admitted into q the final coalition on the same conditionsas the two original members. If the answeris affirmative, then the total proceeds the final coalition, 1, of will be divided equally among the three participants: . If it is negative, then the two original members (belongingto the first coalition of two) will probablyboth get the same amount, but more than . Thus 1 2e with an e > 0. will be dividedsomewhatlike this: + c, + c, 37.1.3. first alternative would be similar to the one which we The in encountered the analysis of the point / in 35.1. erethe coalition H if (1,2,3),it forms at all, contains its three participants on equal terms. to Thesecond alternative corresponds the situation in the interval analyzed in 36.1. Here -2. o any two players (neither f them beingplayer4) combined and this coalition then admitted eitherone of the two remaining playfirst,

i,

|,

erson lessfavorable terms.

The 37.1.4. presentsituation is not a perfect analogue of eitherof these.)))

In the first because they absolutelyneeded him:if 3 combinedwith 4, then 1 and 2 would be completelydefeated;and (1,2)could not, as a coalition, combine with 4 against 3,since 4 neededonly one of them to be victorious (cf.the t In descriptionin 35.1.3.).our present game this is not so:he coalition (1,2)can use 3 as well as 4, and even if 3 and 4 combine against it, only a tie results. In the second casethe discrimination against the member who joins the coalition of threeparticipantslast is plausible,sincethe original coaliis of a much strongertexture than the final coalition of three. c as x\\ tends to 1, he latter coalition tends to become t worthless; f. Indeed, In the remarksat the end of 36.1.2. our present game no such qualitative difference can be recognized: first coalition (of two) accountsfor the the difference betweendefeat and tie, while formation of the final coalition (of betweentie and victory. three)decides t We have no satisfactory basis for a decision excepto try both alternatives. Before we do this, however, an important limitation of our considerations deservesattention.
tion of two

THE CENTER AND ITSENVIRONS 315 the case coalition (1,2)could not make stiff terms to player 3

be noted that we assumethat the same one of the two we alternatives above holdsfor all four coalitions of threeplayers. Indeed, s are now looking for symmetric solutions only, i.e. olutionswhich contain, 2, a, QU}, all its permutations. along with an imputation a = Now a symmetry of the game by no means implies in general the symmetry in eachone of its solutions. The discriminatory corresponding make solutions discussedin 33.1.1. this clear already for the three-person instancesof this for the symmetric f game. We shall find in 37.6.urther four-person game now under consideration. for It must be expected, however, that asymmetric solutions a symmetric to be discovered a first heuristic by game are of too reconditea character in survey like the present one. (Cf. the analogous occurrence the threefor referred to above.) This then is our excuse looking at person game, presentonly for symmetric solutions. i One 37.2.2. more thing ought to be said:t is not inconceivable that, while asymmetric solutions exist, general organizational principles,like to thflse corresponding our above two alternatives, are valid either for the of all participantsor not at all. This surmisegains somestrength totality that the number of participantsis still very low, and from the consideration be too low to permit the formation of several groupsof parmay actually ticipants with different principlesof organization. Indeed,we have only four participants, and ample evidencethat three is the minimum number will for any kind of organization. Thesesomewhat vague considerations o c find exact orroboration in at least one specialinstancein (43:L)et seq. f F 43.4.2. orthe presentcase, owever, we are not able to support them by h any rigorousproof.)))

It 37.2.1.will

T 37.2. he Two

Alternatives

and the

Role of

Symmetry

{i,

316

ZERO-SUMFOUR-PERSONGAMES

87.3. he First T

Alternative

at the Center

We take Let us now considerthe two alternatives of them up in reverseorder. Assume first that the two original participants admit the third one under much less favorable conditions. Then the first coalition (of two) must be considereds the core which the final coalition (of three)crystala on to lizes. In this last phase the first coalition must therefore be expected act as one player in its dealings the two others,thus bringing about with somethinglike a three-person ame. If this view is sound, then we may g t considerations of repeathe corresponding Taking, e.g. 1,2), the \"first\" coalition, the surmised three-person ( as gameis between the players (1,2), 3,4. The considerationsreferred to a above therefore apply literally, only with changednumerical values: = 0, 6 = c = 1 and so a = ft = y = O.1 Sincethe \"first\" coalition may consist of any two players, there are heuristicreasonssimilarto those in the discussion the three-personame of g t to (in -22.) expecthat the partners in it will split even:when an ally is found, as well as when a tie results, the amount to be divided being 1or

37.3.1.

37.1.2.

36.1.3.

1,

21.

2 respectively.

as follows:

If the distribution of the players is different, then the corresponding must permutation of the players 1,2,3,4 be applied to the above. It is now necessaryto submit all this to an exactcheck. The heuristic suggestionmanifestly correspondsto the following surmise Let V be the set of thesefollowing imputations

0,0,0,0.

if is not successful,i.e. a tie results, then these amounts are replacedy b

i 37.3.2. the Summing up:f the above surmisesprove correct, situation is If the \"first\" coalition is (1,2)nd if it is successfulin finding an a ally, and if the player who joins it in the final coalition is 3,then the players get 1,2,3,4 the amounts , ,0, 1 respectively. If the \"first\" coalition

C\\7-9\\

^' i> 0> ~M ^\"=(00001

a =

'

We

t expecthat this V

the theseby permuting the players (i.e. 1,2,3,4. components) is a solution.

and the imputations which originate from

1Theessential ifference between this discussion and that referred to, is that player 4 d is no longer excludedfrom the \" first\" coalition. 1The argument in this caseis considerably weaker than in the casereferredto (or in
the corresponding
in two different

sinceevery \"first\" coalition may now wind up application in ways (tie or victory). The only satisfactory decisionas to the value of the argumentation obtains when the exacttheory is applied. Thedesired justification is c indeed, it is the special ase actually contained in the proof of

36,1.3.)
y*
y*

38.2.1. -3.;

y\\

- - - -l
y*

in of (38:D)

38.2.3.)))

THE CENTER AND ITSENVIRONS

A

317

rigorousconsideration,of the same type as that which constitutes this V is indeed a solution in the senseof 30.1. e do W not give it here,particularly because is contained in a more general roof it p which will be given later. (Cf. the reference of footnote 2 on p.316.) shows 36.2., that
Center Assume next that the final coalition of three contains all its 37.4.1. the participantson equalterms. Then if this coalition is,say (1,2,3), players 1respectively. get 1,2,3,4 the amounts It would be rash to concludefrom this that we expecthe set of imput tations V to which this leads, to be a solution; the set of theseimputaat the

37.4.TheSecond Alternative

i,i;i,

tions a =

{i, 2,

i.e.

<*3, a*}>

(37:3) a '\" =

(i,i,i, -1}

and permutations as in

(37:2).

We have made no attempt as yet to understand how this formation of the \" final coalition in \"one piece comesabout, without assumingthe previous

of existence a favored two-personcore.

able. The stratified form of the final coalition is expressed the imputaby tion a and the motive for just this scheme f distributionliesin the threat o ~-* \"-* of a tie,expressed the imputation a \". To put it exactly: a form a the by solution only in conjunction with the a \", and not by themselves. In (37:3)this secondelement is lacking. A directcheckin the sense that of 30.1. discloses the a \"' fulfill condition (30:5 :a)there,but not (30:5 :b).

In 37.4.2. the previous solution of (37:2)such an explanationis discern-

'

'

I.e.hey do not dominate eachother, but they leave certainotherimputat further elements ust be added to V.1 m tions undominated. Hence of This addition can certainly not be the a\" = {0,0,0,0)(37:2)since

that imputation happens to be dominatedby a 2 In other words the of of extension stabilization,in the sense 4.3.3.)a '\" to a solutionmust (i.e. be achievedby entirely different imputations (i.e.hreats) in the caseof t of of (37:2). as the ~^'\"of (37:3) in the case the It seemsery difficult to find a heuristicmotivation for the steps which v are now necessary. Luckily, however, a rigorous procedureis possible from here on, thus renderingfurther heuristicconsiderations unnecessary, 1Toavoid misunderstandings It is by no means generally true that any set of imputations which do not dominate each other can be extended to a solution. Indeed,the problem of recognizing a given set of imputations as being a subset of some (unknown)

'\".

~'

solution

is still unsolved. Cf 30.3.7. In the present casewe are just expressing the hope that such an extension a f possibleor the V of (37:3),nd this hopewill be further justified below.
With

will

prove

(1,2,3).)))

318

GAMES ZERO-SUMFOUR-PERSON

o o Indeed, ne can prove rigorously that there exists ne and only one symmetric extension the V of (37:3)to a solution. This is the addition of of theseimputations a = {ai,a2, as,<**}

(37:4) ~2IV =

(i,i,

- i, - *}

and permutationsas in

(37:2).

this solution, i.e.of its to not constituent a lv of (37:4),s wanted, it must be said that it does seem i t be a tie at all (like the correspondinga \" of (37:2)) rather,it seemso be some kind of compromisebetween a part (two members) of a possible victorious coalition and the othertwo players. However,as stated above, we do not attempt to find a full heuristicinterpretation for the V of (37:3) and (37:4)indeedit may well be that this part of the exactheory is already t ; 1 will s beyond such possibilities.Besides, ome subsequent examples illustrate the peculiaritiesf this solution on a much wider basis. Again o we refrain from giving the exact roof referredto above. p
If 37.4.3. a common-sense interpretation of

\" terming them discriminatory\.") Only one solution in that casewas a the finite set of imputations; that solution alone possessed same symmetry t as the gameitself (i.e. as symmetricwith respecto all players). This w time conditionsarequite different. We have found two solutionswhich are 2 the both finite setsof imputations, and which possess full symmetry of the e The discussionof 37.1.2. that it is difficult to consider ither shows game. i solutionas \" abnormal\"or \"discriminatory\"n any sense;hey aredistint of guishedessentiallyby the way in which the accession the last participant to the coalition of threeis treated, nd therefore seem correspond two to to a normal principlesof social rganization. perfectly o If 37.5.2. anything, the solution (37:3), may seem lessnormal the (37:4) of one. Both in (37:2)and in (37:3),(37:4)the character the solutionwas

37.5. omparison of the Two Central Solutions C 37.6.1. two solutions (37:2)and (37:3), The (37:4),which we found for the game representing center, resenta new instanceof a possible the p multiplicity of solutions. Of coursewe had observed this phenomenon 1 . of g before,namely in the case the essentialhree-personame in 33..1 But t this thereall solutionsbut one werein someway abnormal (we described by

determined by thoseimputations which describedcomplete a decision,a > ^ t \"' \" and a respectively. To thesethe extra stabilizing\"imputations, a \" > > t and a /v , had to beadded. Now in the first solutionthis extraa \" had an
1This is, of course, well a
if they originate

'

known occurrence mathematical-physical theories, even in in heuristic considerations. * An easycount of the imputations given and of their different permutations shows that the solution (37:2) of consists 13elements,and the solution (37:3),(37:4)of
10.)))

THE CENTER AND ITSENVIRONS

obvious heuristic interpretation as a
IV

319

is surroundedby somepeculiarphenomena which can neitherbe explained nor foreseenby the heuristic procedurewhich providedeasy access this to

tie, while in the secondsolution the nature of the extraa appearedto be more complex. A more thorough analysis discloses, however, that the first solution

interpretations of our theory. We shall therefore analyze them in some detail in what follows. We add that a similar expansionof the second solution has not been found up to now.
To 37.6.1.beginwith, thereexistsomefinite but asymmetricalsolutions which arecloselyrelated (37:2)n 37.3.2. to i becausehey contain someof the t 1J.1 One of thesesolutionsis the one which obtains i the center the imputations {i, ,0, V7// from either when we approach along diagonal /-Centerand use there the solutionsreferred to in 36.3. i I.e.:t obtains by side, continuous fit to the domains (36:B) and (36:C) there mentioned. (It = 0,i.e.he center, elongsto both will be remembered the point that t b
Xi

solution. Thesephenomena arequite instructive from a general oint of view too, p becausethey illustrate in a rather striking way some possibilitiesand

37.6. nsymmetrical U

Central Solutions

a sui generisprincipleof social rganization, we shall describe briefly. it o This solutionpossesses samesymmetry as thosewhich belongto the the on VIII, games the diagonal/-Center- as it is actually one of them symmetric 2 t while player 4 occupies special osition. a with respecto players 1,2,3, p it We shall therefore describe in the same way we did the solutions on i Hereonly permutations of the the diagonal, e.g.n (36:3)in a w in the descriptions (37:3) nd (37:4) e are w of players 1,2,3 suppressed, hile suppressedall permutationsof the players 1,2,3,4. with w 37.6.2. the sake of a bettercomparison, e restate this notation For only) the definition of our first fully(i.e. allowing for permutationsof 1,2,3 8 in I symmetricsolution (37:2) 37.3.2. t consistsof theseimputations:

thesedomains,cf. 36.3.2.) this solution can be taken also to express Since

36.2.1.

.
I.e.

= \"?/// = P /v =
ft\"

7' = (i,i,o, -i}

\\\\ all

{i, , 1,0} i (in 1 i> tiiU, l, {0,0,0,0}

exhaust the

and the imputationswhich originate from theseby permuting the players 1,2,3.

is what

1 somebut not * That the position

distinguishes

a \", id. '\" because this a had to be represented the three imputations \", ft by to system of representation makes it necessary state in which one of the three possible the positions of that imputation (i.e. values },0, 1)the player 4 is found.)))
Our

7',7\",7\"' '

of the 12permutations of this imputation. of player 4 in the solution is really different from that this solution from the two symmetric onesmentioned

before.

of the others,

a'of (37:2)in 37.3.2.,_ while

',

_ * is

320

ZERO-SUMFOUR-PERSON GAMES
which we refer

Now the (asymmetric) solution to imputations:

consistsof these

(4,0, Op players 1,2,3. Oncemore we omit giving the proof that (37:5) a solution. Instead is we shall suggestan interpretation of the difference between this solution o and that of (37:2) i.e.f the first (symmetric)solutionin 37.3.2. This 37.6.3. difference consistsin replacing
F

?' (37:5) V

\"a*\"

P ' \"?\"

-*,

QQintt7'9*^ asm^/.z;

natefrom theseby permuting the

and the imputationswhich origi-

by

7'\"= (4,0,-i,i)

7
ft

(4,0,-4,0}.
y

That

getssomewhatlessand the losingplayer among 1,2,3 this arrange(in ment player 3) gets somewhatmore than in ft '\". This difference is preto cisely4, so that player 4 is reduced the tie position0,and player3 moves from the completely defeatedposition 1to an intermediate osition 4 p \" form Thus players 1,2,3 a \" privileged group and no onefrom the outside will be admitted to the \"first\" coalition. But even among the three membersof the privilegedgroup the wrangle for coalition goeson, since
now

is:the imputation '\" in which the player 4 would belongto the \"first\" coalition (cf. 37.3.1.), o the group which wins the maximum i.e. t is removed,and replacedy another imputation . Player 4 amount b
ft

the \"first\" coalition has room for two participantsonly. It is worth noting that a memberof the privilegedgroup may even be completelydefeated,as
in
ft

\" but only by a majority of his \" class who form the \" first \" coalition and who may admit the \"unprivileged\" layer 4 to the third membership p of the \"final\" coalition, to which he is eligible. w The 37.6.4. reader ill notethat this describesperfectly possibleform a of social rganization. This form is discriminatoryto be sure, although o not in the simpleway of the \"discriminatory\"olutionsof thethree-person s a a game. It describes more complexnd a more delicate type of social d inter-relation, ue to the solution rather than to the gameitself.2 One a may think it somewhatarbitrary, but sincewe are considering \"society\" of very smallsize, ll possiblestandards of behavior must beadjusted rather a preciselyand delicatelytothe narrownessof its possibilities. We scarcely need to elaborate the fact that similar discrimination instead of 4) couldbe expressed suitable against any otherplayer (1,2,3 by

\",

but

1This imputation ft v is reminiscent in its arrangement of a it has not beenpossible make anything of that to 1As to this feature, cf. the discussionof.35.2.4.))) analogy.

>

/F in

(37:4)of 37.4.2.,

NEIGHBORHOODF THE CENTER O

321

w solutions, which could then be associatedith the three other diagonals of the cubeQ.

38.A Family of Solutionsfor a Neighborhoodof the Center 38.1. Transformation of the Solution Belonging to the First Alternative at the Center
to appear that it can be subjected a peculiartransformation without losingits character s a solution. a This transformation consistsin multiplying the imputations (37:2)of 37.3.2. a common (positive)numerical factor z. In this way the followby ing set of imputationsobtains:
in

I 37.3.2. t will

38.1.1. continue the analysisof the ramifications of solution (37:2) We

f^K-n

^ =
__

'

|2'2* ^

~Z

7\" = {0,0,0,0} In orderthat thesebe imputations,

t (i.e.he
z

anc^

theseby permuting the players 1,2,3,4.

^e im Putations which originate from

1,

all their componentsmust be ^ 1 common value of the v((i)). As z > this means only that w i.e. e must have

(38:2) other z of

< z ^ 1.
it

I with (37:2) 37.3.2. t would not seem of coincides Forz = 1our (38:1) shouldbe a solution for the same gamefor any likely a priori that (38:1)
if and only if z
A (38:2). nd yet a simplediscussionshowsthat i w > $ i.e. hen (38*:2)s replacedy b f < z g 1. (38:3)

is a solution

wider applicabilityin theseinvestigations. The interpretation of theseresults will be attempted afterwards. We 38.1.2. beginby observingthat considerationof the set V defined in for the above (38:1) the game describedy (37:1) 37.1.2. the (i.e. b by centerof Q), could be replacedy considerationof the original set V of b was o in (37:2)in 37.3.2. another game. Indeed, ur (38:1) obtained from by z. Instead of this we could keep(37:2)and multiplying (37:2)by function (37:1) l/z;this would destroy the by multiply the characteristic normalization 7 = 1which was necessaryfor the geometrical representation t but we proposeto accepthat. by Q (cf.34.2.2.) What we arenow undertakingcan be formulated therefore as follows: with Sofar we have started a given game,and have lookedfor solutions. Now we proposeto reversethis process, starting with a solution and looking)))

The importanceof this family of solutionsis further increased the fact by to that it can be extended a certainthree-dimensional iece p surrounding the centerof the cubeQ. We shall give the necessarydiscussionin full, it because offers an opportunity to demonstratea techniquethat may be of

322

GAMES ZERO-SUMFOUR-PERSON

we for the game. Precisely: start with a given set of imputations Vi and 1 this function v(S) (i.e. ask for which characteristic games) V is a solution. in by a common factor means Multiplicationof the v(S)of (37:1) that we stilldemand

37.1.2.

(38:4) (38:5)

v(/S) =

when S is a two-elementset,

i.e. of character the game (cf.27.1.4.), but beyond this only the reduced
v((l)) = v((2)) v((3)) v((4)).

and is value of (38:5) l/zand therefore (38:4), (38:5) in multipliedby (25:3:a) (25:3:b)25.3.1. that this v(S) is just (37:1) yield is above l/z. Our assertion (38:3) means that the V of (37:2)in 37.3.2.a if solutionfor (38:4), (i.e. l/z) (38:5) and only if the joint value of (38:5) is ^ 1and > f. t Indeed,his joint

S i ( reduction, .e. 38:5). o we demand of v(S) only (38:4),restrictingits the values for two-elementsetsS. We restate final form of our question: a (38: Considerll zero-sumfour-persongameswhere A) (38:6) v(S) =0
for all two-elementsets

Now 38.1.3. we shall go one step further

and drop the requirementof

S.

i a solution? Forwhich among theseis the setV of (37:2)n 37.3.2. It will benoted that sincewe have droppedthe requirementsof normalwith ization and reductionof v(S) all connections the geometrical representation in Q aresevered.A special manipulation will be necessarytherefore at the end, in order put the results which we shall obtain back into the to

framework of Q.

38.2. Discussion Exact 38.2.1. unknowns of the problem (38: areclearlythe values The A) = -2/1, v((2)) = -2/2, v((3)) = -2/3, v((4)) = -y4. (38:7) v((l
We what restrictionshe condition in (38:A) actually t proposeto determine on thesenumbers2/1, 2/*> 2/a, 2/4. places
method

the deflects

1This reversed o is procedure quite characteristic f the elasticity of the mathematical for the kind and degreeof freedom which existsthere. Although initially it
inquiry

into

the strictest mathematical

technical manipulation other way. After our previous exampleswhere the guidance camefrom heuristic considerations, it is quite instructive to study this casewhere no heuristic help is reliedon and solutions are found by a purely mathematical trick, the reversalreferredto above. For the readerwho might be dissatisfied with the useof such devices exclusively technical and non-conceptual ones),we submit that they are freely and legitimately used in mathematical analysis. We have repeatedly found the heuristic procedure asierto handle than the rigorous e one. The present caseoffers an example of the oooosite.)))

which must be considerednnatural from any but u point of view, it is neverthelesseffective; by an appropriate it finally disclosesolutions which have not been found in any s

a direction

(i.e.

NEIGHBORHOODF THE CENTER O

323

1 This gameis no longersymmetric. Hence permutations of the the are players 1,2,3,4 now legitimatenly if accompanied the corresponding o by permutationsof yi, y^ y&, y^ Tobegin with, the smallest componentwith which a given player k is in ever associated the vectorsof (37:2)n 37.3.2., 1. Hencehe vectors i is t will be imputations if and only if 1 *z v((fc)) i.e.
k = 1,2,3,4. for o The characterf V as a set of imputations is thus established; us let now seewhether it is a solution. This investigation is similar to the proof given in 36.2.3-5. The 38.2.2. observations(36:5), of apply again. A two(36:6),

(38:8)

yk

*l

s element et S = (i,j) is effective for a = #2, s, ouj when a + ay ^ In we (cf.(38:A)). Hence have for the a ', a\"of (37:2): a \" every two' :No two-elementsetS which doesnot cons element etS is effective. In a tain the player 4 is effective, while those which do contain him, S = (1,4), S we (2,4),(3,4)clearlyare. However,if we consider = (1,4), may discard 1 the two others; = (2,4)arises from it by interchanging and 2, which S does not affect a 3 S = (3,4)is actually inferior to it after 1 and 3 are since ^ O.4 interchanged,
Summing

{i,

36.2.3.,

up:

'; i

(38:B)

Among

tainly

the two-element sets S, those given below arecernecessary,and all others arecertainly unnecessary:

6 all for 7\". (1,4)for sets: three-element Owing to the above we may exclude by Concerning ' thosewhich contain (1,4)or sets all (36:6) three-element for a \", and for a 6 for (2,4). This leavesonly S = (1,2,3) a

7',

Summingup:

'.

(38:C)

sets S, the one given below is the three-element certainly necessary,and all othersarecertainly unnecessary:
Among

for (1,2,3) 7'.

Unless y\\ y<. y\\ y* 2 But there is nothing objectionable such usesof the permutations of 1,2,3,4 we in as in made in the formulation of (37:2) 8 This and similar oneslater are clearly legitimate devicesin spite of permutation footnote 1 on p. 309and footnote 2 above. above. Observe footnote 4 As i when 1 we coulddiscardall thesesets,ncluding i (1,4), v((4)) But we are under no obligation to do this. We which is a possibility. when 2/4 prefer not to do it, in orderto beableto treat 3/4 1 and j/ 4 > 1 together.
1

37.3.2.

i.e.

1,

1;

6 And

all permutations

of

these 1,2,3,4; modify a ' too.

6Thelatter obtains from the former by interchanging

1and 2, which doesnot

affect

'.)))

324

GAMES ZERO-SUMFOUR-PERSON

We leave to the

necessarysteps.)

36.2.4. Actually 38.2.3. We

which

a 'of V

>

readerhe verification of (30:5 in 30.1.1.,that no t i.e. :a) dominatesany ft of V. (Cf. the corresponding of the proof in part
the proof of
which (30:5:b),

follows, also contains the an imputation

ft

is undominatedby the elements f V must belongto V. o > Considera ft undominatedby the elements f V. If any two of fa, 182, o 0s, ft 4 were < 0, we could make these (by permuting 1,2,3,4) 2 < 0. fa, > \" This gives a H ft with S = (1,2) (38:B). encet most one of fa, 2, of H a 03, 4 is < 0. If none is < 0,then all are ^ 0. Soeachcomponentof ft is
it

next verify

in i.e. (30:5:b) 30.1.1., that

the corresponding componentof a\",and sinceboth areimputations it follows that they coincide,t = and so f (cf. footnote 2 on p.
<

Hence we preciselyone of 0i, 2, 0s, ft 4 is < 0. By permuting 1,2,3,4 can make it 4 If any two of fa, ft*, 3 were < we could make these (by permuting of ft 1,2,3)i, 02 < Besides,4 < 0. So the interchange 3 and 4 gives

is in V.

311),

a\";

so this necessitates 1/4 ^ $ $, i.e./ 4 ^ f . Thus we need i/ 4 < f to this exclude possibility,and as we arepermuting freely 1,2,3,4,even need we

2, Ifnoneis<, then0i,03 ^i.

'

i.

i,

ft

with

S=

of H (1,2,3)(38:C). enceat most one of fa, fa,

Hence4 ^

4 f. But0 ^ v((4))= -y4,

3 is

< i.

(38:9)

yk

<}

for

1,2,3,4.

If this conditionis satisfied,then we can conclude that preciselyoneof we By permuting 1,2,3, can make it 3 fa, 02, 0a is < 1 So0i,02 ^ 03 ^ 0. If 4 ^ then eachcomponentof is ^ the

it

c correspondingomponentof a ', and sinceboth areimputations (cf. footit note 2 on p.311) follows that they coincide: = a' and so it is in V. Hence 4 < 1. Also 3 < So interchangeof 1 and 3 gives ~^' H with S = (1,4)of (38:B). and the This, at last, is a contradiction, thereby completes verification 1. of (30:5 in 30.1. :b) Thecondition(38:9), we needed this proof, is really necessary: which for

i. i,

>

- 1,

i.

is easy to verify that

7'= i*,*,*, -i}

1 If if y 4 then this is certainly v((4)) to assumeit. (Cf.footnote 4 on p. 323.))))

1,i.e.

1,

the

b case; ut we do not

wish

NEIGHBORHOODF THE CENTER O

then gives (38:9). Thus we need precisely(38:8) and

is undominated by our V, and the only way to prevent it from being an < v((4)) = y 4, i.e. < f.1 Permuting 1,2,3,4 imputation is to have 2/4

(38:9). umming up: S The V of (37:2) 37.3.2.a solutionfor a gameof (38:A) in is (38:D) if and only if (with (38:6), (38:7) there) 1^ y < h = 1,2,3,4. for (38:10)
k

325

38.2.4. us now reintroducethe normalization and reduction which Let we abandonedtemporarily,but which arenecessaryin order refer these to results to Q, as pointed out immediately after (38: A). The reduction formulae of 27.1.4. that the shareof the player show must bealtered the amount ajj where)) by
fc

|
2/7

= yk
and))

- i(2/i+

= i(2/i+ 2/2 + 2/3 + 2/0Fora two-element et S = (i,j),v(S)is increasedrom its original value to s f +a = + i(2/i+ 2/2 + 2/3 + 2/4) = i(2/ + 2/y yk yi) (fc, I arethe two playersotherthan i,j). Theabove y is clearly ^ 1 > (by (38:10)), the game is essential. hence Thenormalization is now carriedout by dividing the characteristic function, as well as every player's share,by y. Thus for S = (t, j), v(S) is now
<*? 2A

7=

2/2

+ 2/3 + 2/4)

-- -

modified further

to))

2/i

+ y* + 2/3 + 2/4

This then is the normalized and reducedform of the characteristic for the representationy Q. (34:2)id.gives, b function, as used in the togetherwith the above expression, formulae

34.2.1.

ft

' to V
If

1Observe that
(when
j/

the failure 4

!). Indeed,
y4

it would be imputations,
t/j

etc.
t/j

necessaryto remove a
2/t

ss

form in transform

Thus 37.4.1. further

it gradually
further.)))

of and (37:1) 37.1.2.,it carriesthe

with

attempts to make our V over into

of V to dominate
ft
\"

' dominates
above

this

' could not becorrected adding by

a
\"

- (0,0,0,0)
creating

with

so S (1,2,3),
back to
the

from V, thereby

new undominated

t then

a change of unit
ft

' into the

by

brings our game

IV

started

Theseconnectionsbetween the

(37:2)in 37.3.2.

into

(37:3),(37:4)in

This 37.4.1-2. is noteworthy,

a solution

- J, i,i,-1|(37:3) of
{

would probably

sincewe

two solutions

(37:2)and (37:3),(37:4)should be

investigated

326))

GAMES)) ZERO-SUMFOUR-PERSON

x1 _

(38:11)))

~ 2/2 + + 2/2 + 2/3 + 2/4' 2/1 + 2/4 = -2/i+ 2/2 * + #2 + + 2/4 ; 2/1 _ -2/i-- 2/2 + ys + 2/4; + 2/4 yi + 2/2 +
yi
2/8 2/4^ 2/8 2/8 8
'))

2/8

for the coordinates x*, x\\,

thesesolutions i.e.he solution (37:2)in 37.3.2., transformed as indicated t above can be used. This definition is exhaustive,but implicit.Let us make it explicit. I.e.,iven a point of Q with the coordinates i, x^ 3, z g let us decide hether (38:10) (38:11) then be satisfiedtogether(by w and can appropriate 2/1, 2/2, 2/s, 2/4). We put for the hypotheticalyi, 2/2, 2/8, i/
4))

38.2.5. (38:10) (38:11) define the part of Q in which Thus and together

X*

in

Q.

(38:12)
with

2/i

+ 2/2 + 2/s +
y*

2/4

= -))

z indefinite.

Then the equations (38:11) become))

2/i

, ~ 2/3 + 2/4 = xi
>))

(38:12*)))

, -2/i+ 2/2

2/3))

2/4

=
z))

>

-2/i
~~~

2/2 \"h

2/8

+ 2/4 =
=))

8>))

and can t (38:12) (38:12*) be solvedwith respecto 2/1, 2/2, 2/s, 1+ Xi x z x i/ 2/2 z (38:13))) 1+ Xi + X
\\ I))

-*

+))

2/4

=))

2))

Now (38:11)satisfied, and we must use our freedom in choosing to is z satisfy (38:10). Letw be the greatest nd v the smallestof the four numbers)) a (38:14))) Now that
+))

are These known quantities, sinceZi,

(38:10) means that clearly

$w

1^

x 2, 0:3areassumedto

v/z and that w/z

it < $, i.e. means

be given.

(38:15)
Obviously this conditioncan

<zgv.
(for z) if and only if
v.)))

befulfilled
fu;

^38:16)

<

NEIGHBORHOODOF THE CENTER

And if

327

is allows (38:16)satisfied,then condition (38:15) infinitely many values an entireinterval for z. 38.2.6. we draw any conclusionsfrom (38:15), we give Before (38:16), the explicit formulae which expresswhat has becomef the solution (37:2) o of 37.3.2. owing to our transformations. We must takethe imputations

share),and divide this by y. Thesemanipulations transform the possiblevalues of the componentk and which are 0, 1in (37:2) as follows. We considerfirst k = 1, use the above expressionsor a and y as well as (38:13). f Then:))

a ', a \", loc. it.,dd the amount a* to the component (i.e. the playerfc's k to c a

i,

2 g0

1 e8 into

*5! +
y

=2

4y.
i

- (y. + y. + y, +
+ + 3+
a
4))

_ goesinto Jll+Jf =

If. + If. + = Ogoesinto^** yi (if. yt+ + yt + yt *) = Xl + y

- Xt
.)))

Xt

-4 + 4.-(yt * + * + +
2/1

+ 2/2 + ^8 + 2/4

+ Xi ~ Zj ~ X|.

Xi

1 respectively.

a Fortheotherk = 2,3,4 expressionsre changedonly in so far that their these 2 + Zs, Zi + 2 + z$, X2 zs is replacedy Xi + x 2 z 8, Zi b
Summing up

(and recalling (38:14)):

fc

(38:E)

The component is transformed as follows: goesinto z/2 + w* 1, goesinto u k 1, 1goesinto z + uu 1,

the
Wi, w 2, MS, u 4 of

with

(38:14).

paying due attention to carrying out correctly the permutations 1,2,3,4 which arerequiredthere. It will be noted that for the center i.e.x\\ = x 8 = X* = (38:E) as it of the reproduces formulae (38:1) 38.1.1., should. We 38.2.7. now return to the discussionof (38:15), (38:16). that Condition(38:16) expresses the four numbers1*1,W2, u*, M 4 of (38:14) arenot too far apart that their minimum is more than f of their maximum their t i.e.hat on a relative scale sizesvary by lessthan 2:3.= This is certainly true at the center, where x\\ = x* = X* 0; there Hencein this casev = w = 1,and (38:15) ui, t*2, MS, ^4 are all

We leave it to the

with t readero restate (37:2)

the modification (38:E),

=1.

owing to the form the permutations sidering in Xi, xj, xs as described 34.3.2.))) the influence

1This is immediate,

of

and of the equations (38:13), on of the players 1,2,3,4

equally by conthe coordinates

328

ZERO-SUMFOUR-PERSONGAMES

i becomes < z g 1 proving the assertionsmade earliern this discussion f

interior.

is which (38:16) true by Z. Then even a sufficiently small neighborhoodof the center belongs to Z.1 So Z is a in three-dimensional piece the interior of Q, containing the centerin its own also express relationshipof Z to the diagonalsof Q, say to the VIIL Z contains the following parts of that diagonal. (Use /-CenterFigure64): on one side preciselyC, on the other a little less than half of J3.* We add that these solutions are different from the family of solutions valid in (36:B)and (36:C)which were referredto in 36.3.
We can

in (cf.(38:3) 38.1.1.). Denotethe part of Q in

severalremarkablefeatures.

The 38.3.1. family of solutionswhich we have thus determinedpossesses

38.3. Interpretation

of The Solutions

intersection

We note first that for every gamefor which this family is a solution 1 at all (i.e. every point of Z) it gives infinitely many solutions. And all in we said in of appliesagain:thesesolutionsarefinite sets imputations* 6 and possessthe full symmetry of the game. Thus thereis no \" discrimination \" in any one of thesesolutions. Nor can the differences in \"organizational principles,\" hich we discussed cit., loc. beascribedto them. There w 1 is neverthelessa simple \" organizational principle' that can be stated in a verbal form, to distinguish these solutions. We proceed to qualitative formulate it. 38.3.2. Consider (38:E), hich expresses changesto which (37:2) w the is in 37.3.2. to be subjected.It is clear that the worst possibleoutcome for the player k in this solution is the last expressionsince this corresponds ( z + Uk This expression > or = to 1), is to according whether z is < or = u k Now Wi, w 2, MS, w 4 arethe four numbersof (38:14), of z the smallest which is v. By (38:15) v, always z + w* 1^ 1, = occursonly for the greatestpossiblevalue of z, z = v, and then and only for thosek for which Uk attains its minimum, v. 1If Xi, xs, Xi differ from by < A then eachof the four numbers MI, MJ, MI, u 4of (38:14) is < 1 -f A I and > 1 A 1; hence on a relative sizethey vary by < J t f. So we are still in Z. In other words: Z contains a cubewith the same centeras Q,but with A of Q's(linear) size. Actually Z is somewhat bigger than this, its .volume is about yAo of the volume of Q. * On that diagonal x\\ x xt , so the MI, MI, MI, M 4 are: (three times) 1 x\\ and 1 -f 3xi. So for zi 0, v 1 xb w 1 + 3xi, hence (38:16) becomesx\\ < J. And for si g 0, v 1 + 3xi, w 1 Xi, hence (38:16) becomes> A. So the

37.5.1.

i.e.

1.

1,

i.e.

is this:

Xi

xi

which we found contained four parameters:\\, yi, y t , y* while the y for which they are valid had only three parameters: i, x>, xv & 4 Eachone has in elements, like (37:2) 1In the centerxi xt xt we have y\\ y* y* y 4 (cf. On the diagonal Xi xt x9 we have j/i metry in yf y t (cf. symmetry in 1,2,3.)))

1Thesolution

- - -< is precisely C) >> - i). (B is

t (this

xi

13 1,2,3,4.

- - - -37.3.2.- - - (38:13)), i.e. i.e. (38:13)),

sym-

games

NEIGHBORHOODF THE CENTER O

We

329

this: restate
In this family of solutions,even as the worstpossibleoutcome, a player k faces, in general, omethingthat is definitely better s than what he could get for himself alone, v((fc)) = This advantage disappearsonly when z has its greatest ossible p value, z = v, and then only for thosek for which the correspondattains the minimum in ing number iii, w 2, w 8, w 4 in (38:14)

(38:F)

i.e.

1.

(38:14).

In other words:In thesesolutionsa defeated player is in generalnot \" exploitedcompletely,not reducedto the lowestpossiblelevel the level which he could maintain even alone, i.e. v((fc)) = 1. We observed beforesuch restraint on the part of a victorious coalition,n the \"milder\" i kind of \"discriminatory\" olutionsof the three-person amediscussedin s g 33.1. when c > 1,cf. the end of 33.1.2.). thereonly oneplayer But (i.e. o couldbe the object f this restraint in any one solution,and this phenomenon went with his exclusion from the competitionfor coalitions. ow there is N no discriminationor segregation; instead this restraint appliesto all players in general, nd in the centerof Q ((38:1)38.1.1., z < 1) the solution a in with
11 1 is even symmetric!

losethis advantage, sincein generalhe four numbersui, wj, MS, t are Ui of (38:14) different from eachother and only one is equal to their it minimum v. All four playerswill lose simultaneously only if u\\, w, Wt, u\\ t are all equal to their minimum v i.e.o each other and one look at suffices to show that this happens only when x\\ = x\\ = x\\ = 0, (38:14) i.e.t the center. a a This phenomenon of not \"exploiting\"defeatedplayer completelyis a
player will
very important possible(but by no means necessary)feature of our soluof social rganizations. It is likely to play a greaterrolein the o tions,

Even 38.3.3. when z assumesits maximum

value v, in general nly one o

general heory also. t We conclude y stating that someof the solutionswhich we mentioned, b are also in but failed to describe 36.3.2., possessthis feature. These the differ from the solutions solutionsin C of Figure64. But nevertheless they h which we have considered ere.
Thereis a quantitative difference of some significance as well. Both in our present set-up (four-person game, centerof Q) and in the one referredto (three-persongame in the the senseof 33.1.), besta player can do (in the solutions which we found) is J, and
1

i.e.

the worst

he is not completely c (i.e. with 1 & c

Theupper limit of what he may get in caseof defeat,in those of our \"
exploited,\"

is

1.

is now

(i.e. zwith

\\

< z 1)and it was \\ then

solutions where

< J).

the Sothis zonenow covers interval,

while

fraction

__

1)

**

\"*

9*

i.e. % of the 22|

significant

it then

covered100%.)))

CHAPTER VIII SOME REMARKSCONCERNINGn

5 PARTICIPANTS of in 39.The Number of Parameters Various Classes Games The 39.1. Situation for n = 3,4

We 39.1. know that the essentialgames constituteour real problem form and with 7 = 1. and that they may always be assumedin the reduced In this representation there exists precisely one zero-sum three-person game, while the zero-sum four-persongames form a three-dimensional manifold.1 We have seenfurther that the (unique) zero-sumthree-person game is automatically symmetric, while the three-dimensionalmanifold of all zero-sum four-person games contains preciselyone symmetricgame. this Let us express by stating, for eachone of the above varieties of i i h how many dimensionst possesses, .e.ow many indefinite paramgames, a etersmust be assignedspecific(numerical) values in order to characterize of that class. Thisis best done in the form of a table, iven in Figure g game 2 i 65 in a form extended to all n ^ 3. Our statementsabove reappearn the entriesn = 3,4 of that table.

complete the table by determining the number of zero-sum n-person game,both for the class all these of a games, nd for that of the symmetricones. function is an aggregate as many numbers v(S) of The characteristic as therearesubsetsS in / = (1, of 2n Thesenumbersare , n), t of subjecto the restrictions(25:3:a)-(25:3:c) and to thosedue to the a reducedcharacter nd the normalization 7 = expressedy (27:5)in b fixes 27.2.Of these (25:3:b) v(-S) whenever v(S) is given, hence it 3 so halves the number of parameters: we have 2n~l instead of 2n Next :a) (25:3 fixes oneof the remaining v (S) v ; (27:5)fixesn of the remaining , v((n)); hencethey reducethe number of parameters v(S):v((l)), ~ need by n + 4 So we have 2n l n 1 parameters.Finally (25:3:c) sinceit containsonly inequalities. not be considered, If 39.2.2.the gameis symmetric,then v(S)dependsonly on the number of elements of Thus it is an aggregate as of p v(S) = v p, cf. numbers v p as there are p = 0, i.e.n + These many , n, 1Concerning the general remarks, cf. 27.1.4. 27.3.2.; and concerning the zero-sum g three-personame cf. 29.1.2.; concerning the zero-sum four-person game cf. 34.2.1. * Thereare no essential zero-sum games for n 1,2! 8 S and S are never the same set! 4S Qt (1), , (n) differ from eachother and from eachother'scomplements.
now

39.2.1. We

39.2. Situation The

for All n

of parameters the

i.e. . 25.3.1., 1,

()

I.

S:

28.2.1. 1,

1.

330)))

NUMBER OF PARAMETERS))

331))

numbers are subjectto the restrictions (28:ll:a)-(28:ll:c) the of 28.2.1.; i reducedcharacters automatic, and we demand also Vi = y = 1. (28:11fixes v w_p when v p is given; henceit halves the numbersof those :b) for When n n parameters which p = p l i.e. = 2p, which hap= n/2 (28:11showsthat this v pens only when n is even, and then p :b) must vanish. So we have ^ parametersif n is odd and 5 if n is even, instead of the original n + 1. Next (28:ll:a) one of the remaining fixes v p : Vo) and Vi = 7 = 1 fixes another one of the remaining v p : Vi;

n-p^p.

p))

hencethey reducethe number

or
n))

of

2 parametersby 2: so we have))

2))

need sinceit parameters.Finally (28:ll:c) not be considered ^ contains only inequalities. 39.2.3. collect this information in the table of Figure65. We also We all state explicitly the values arising by specializationto n = 3,4,5,6,7,8, the first two of which were referred to previously.))
Number

~2

of players)
3)

All

games)

Symmetric) games)

0*)
3)

0*) 0*)
1)

4)

5)

10) 25) 56)

6)
7)

1)

2)

8)

119)

2)

..
n)

.)

..
2

.)

Of)

or n

odd

n |-2for))even)

* Denotes game is unique. the

Figure

65.

Essential

and games. (Reduced

1.)

o The rapid increasef the entriesin the left-hand column of Figure65 how serveas anotherindication,if onebe needed, the complexityof a may with increases the number of its participants. It seems oteworthy n game
1Contrast this with footnote 3 on p. 330! 1p 0, differ from each other and from eachother's n of because n

- p.

(The latter

only

3.))))

332

REMARKS CONCERNING n

5 PARTICIPANTS f i.e.or the symmetric

in that thereis an increase the right-hand column too, b games, ut a much slowerone.

formulated

game. The systematic theory is not far enough advanced to allow it; and for a descriptive and casuisticapproach (as used for the zero-sum, four-persongame)the number of its parameters, is ratherforbidding. 10, It is possiblehowever to examine symmetriczero-sumfive-person the games in the lattersense. The number of parameters, is smallbut not 1, zero,and this is a qualitatively new phenomenon deservingconsideration. For n = 3,4 there existed only one symmetric game,so it is for n = 5 that it happens for the first time that the structureof the symmetricgame showsany variety. 40.1.2. symmetriczero-sumfive-persongame is characterizedy the The b v p , p = 0,1,2,3,4,5 of 28.2.1., subjectto the restrictions (28:ll:a)-(28:ll:c) there. (28:ll:a), state (28:ll:b) (with 7 = 1) v = 0, v 4 = 1, v 6 = (40:1) and v 2 = v 3, i.e. v 2 = -ij, v3 = (40:2)

40.1.1. shall not attempt a directattackon the zero-sum five-person We

40.The SymmetricFive -person Game 40.1. Game Formalism of the Symmetric Five-person

vi=-l,

7j

statesv p+g ^ v p + v q for p + q ^ 5 and we can subject id. to the further restrictions (28:12) Therefore p = q = 1,2,* of PJ q and so thesetwo inequalitiesobtain (using (40:1), (40:2)):
Now

(28:ll:c)

1,
i?

i.e.

p=

1,q = 1: -2 ^ -17;
up:
t\\

p=
rj

1,q = 2:

iy;

(40:3)
Summing

-i g

g 2.

(40:A)

The symmetric zero-sumfive-person game is characterized T by one parameterwith the help of (40:1), (40:2). hedomain of variability of y is (40:3).

40.2.The Two Extreme Cases It 40.2.1.may beuseful to give a directpicture of the symmetricgames o a describedbove. Let us first considerthe two extremesf the interval

(40:3):

1= 2, - *.
j

1This is easily verified by inspection of (28:12), by using the inequalities of footor note 2 on p. 250. Thesegive 1 p f, 1 ^ 4 2; henceas p, q are integers, p
q

1,

1,2.)))

This tells the whole story: In the gradual crystallizationof coalitions, the from defeat to victory occurs when the size is point at which the passage increases two to three,and at this point the transition is 100 .2 from %
Summing

THE SYMMETRIC FIVE-PERSON GAME 333 Consider first = 2: In this casev(S) = 2 for every two-element 1 set S;i.e.very coalition of two players is defeated. Thus a coalition of e three (being the set complementaryto the former) is a winning coalition.
17

up:
77

(40:B)

players is to form coalitionsof three players. In this casewe argueas follows:)) 40.2.2. Considernext =
17

= 2 describes gamein a

which the only

objectiveof all

i.

1))

4))

when

S has))

elements.))

3 coalition of four always wins. Now the above formula showsthat a coalition of two is doingjust as to well, pro ratdj as a coalition of four; henceit is reasonable considerthe former just as much winning coalitions as the latter. If we take this v broader iew of what constitutes winning, we may again affirm that the I whole story of the game has beentold:n the formation of coalitions,the is from defeat to victory occurs when the size point at which the passage from increases one to two; at this point the transition is 100 .4 %

Summing

up:
rj

as Oneplayer is just as much defeated two, four are no more victorious than three. a Of course coalition of three has no motive to take in a fourth partner ; it seems (heuristihim only on the worst possible terms. But cally) plausible that if they do they will accept neverthelesssuch a coalition of four wins if viewed as a unit, sincethe remaining isolated player is defeated. 8 In 1 wins, sincean isolated player any zero-sum n-person game any coalition of n is always defeated. Cf.loc.cit. above. 4 One player is defeated,two or four players are victorious. A coalition of three players is a compositecasedeserving some attention v (S)is J for a three-element set Thus a Coalition of St it obtains from the of a two-element set by addition of three is no better than a winning coalition of two (which it contains) plus the remaining isolatedand defeated player separately. This coalition is just a combination of a.wingroup whose situation is entirely unaltered by this operation.))) ning and a defeated

a describesgamein which the only objectiveof all is to form coalitionsof two players. players 40.2.3. the basisof (40:B)and (40:C)it would be quite easy to guess On heuristically solutions for their respectivegames. This, as well as the exactproof that thosesetsof imputations arereally solutions,is easy; but t we shall not considerhis matter further. of let Beforewe pass to the consideration the otherrj of (40:3) us remark of that (40:B) and (40:C) are obviously the simplest instances a general 1 Cf. the discussion in 35.1.1., footnote 4 on p. 296. particularly 2

(40:C)

i.e.

1.

334

REMARKS CONCERNING n

5 PARTICIPANTS

method of defining games. This procedurewhich is more general than that ( exhausof Chapter X, referred to in footnote 4 on p.296) will be considered to tively elsewhere(for asymmetricgamesalso). It is subject somerestrict tions of an arithmetical nature ; thus it is clearhat therecan be no (essential in which every coalition of p is winning symmetriczero-sum)n-persongame couldbe formed and everythen n/p suchcoalitions if p is a divisor of n, since l would win with no losereft. On the other hand the samerequirement body t 1 does not restricthe game at all (cf. footnote 3,p.333). with p = n

40.3.Connection between the

T (40:3). he situation is We have some somewhat similar to that discussedat the end of heuristic insight into the conditionsat the two ends of (40:3)(cf. above). \" \" i Any point ry of (40:3)s somehow surrounded by theseend-points. More 1 precisely,it is their center of gravity if appropriate weights are used.

40.3.1.

Four-person Game Considernow the rj in the interior of

Symmetric Five-person Gameand the 1,2,3-symmetric

35.3.

c a The remarksmade loc. it. pply again:while this constructionrepresents as c (40 ( i all gamesof (40:3) combinationsof the extremeases :B),40:C),t t of is neverthelessnot justified to expecthat the strategies the former can from be obtainedby any direct rocess thoseof the latter. Our experiences p
in the

Thereis, however, another analogy with the four-person game which is gives someheuristicguidance. Thenumber of parametersin our case the (40:3)
while the

caseof the zero-sum four-persongame speak for themselves.

same as for those zero-sumfour-persongames which are symmetric with we t respecto the players 1,2,3; have now the parameter which runs over

-i ^

77

T?

g 2,

x\\ gamesreferred to had the parameter which varies over

(40:4)

-1^ xi

I.2

40.3.2. Considera symmetric five-person gameF with its in (40:3). Let us now modify this game by combining the players 4 and 5 into one the o person,i.e.ne player 4'. Denote new game by F'. It is important to that in F realize that F' is an entirely new game: e have not asserted w f 4 and 5 will necessarilyact together, orm a coalition, etc., that or players there areany generally valid strategical onsiderationswhich would motic 8 vate just this coalition. We have forced 4 and 5 to combine; e did this w modifying the rules of the game and thereby replacingF by F'. by 1Thereadercan easily carry out this composition in the senseof footnote 1 on p. 304, (40:2)in 40.1.2. relying on our equations (40:1), * Cf.35.3.2. the representation in Qused there, x\\ = xt In xt .
77

four-person game is so far entirely formal. There is, 1,2,3-symmetric a deeperignificance behindit. To seethis we proceed follows: as s however,

This analogy betweenthe (totally) symmetricfive-person game and the

tion
cally

This ought to be contrasted with the discussion in of two players was formed under such conditions that
justified.)))

where 36.1.2., this

merger

a similar combinaseemed strategi-

1 four-person game. Given the of T we shall want to determine the of I\" in order to seewhat correspondence (40:3) nd (40:4)hisdefines. t of a xi Afterwards we shall investigate whether therearenot, in spite of what was said above, someconnections betweenthe strategies i.e.he solutions of t T and T'. The characteristicunction v'(S) of r' is immediately expressible in f terms of the characteristic function v(S) of T. Indeed:
>

Now T

THE SYMMETRIC FIVE-PERSON 335 GAME is a symmetricfive-person game,while r' is a 1,2,3-symmetric

v'((l)) = v((l)) = v'((2)) = v((2)) = v'((3)) = v((3)) = v'((4')) = v((4,5))= v'((l,2)) = v((l,2)) = v'((l,3)) = v((l,3)) = -n, = v'((2,3))= v((2,3)) v'((l,4')) v((l,4,5))= = v'((2,4'))= v((2,4,5)) , v'((3,4'))= v((3,4,5)) = = v((l,2,4,5))= 1, v'((l,2,3)) v((l,2,3))= n, v'((l,2,4')) = v((l,3,4,5))= 1, v'((2,3,4'))= v((2,3,4 = 1; v'((l,3,4')) 5)) and of course
>

.-,,

-1, -1, -,,

-1, -i,; - ,, ,;

= 0. v'(0)= v'((l,2,3,4'))

Q of 34.2.2. Let us therefore apply first the normalization formulae of 27.1.4. hey T must show that the share of the player k = 1,2,3,4' be altered by the amount ajj where and
Hence))

T'into that

While F

was normalized and reduced,\" is neither;and we must bring r form sincewe want to computeits x\\, 2, 3, i-e.efer it to the r

al =,

7=

l 5 This 7 is clearly ^ 3 ^~ = ^ > (by (40:3));hence the game is out essential.The normalization is now carried by dividing every player's share by 7. Thus for a two-elementset S = (i, j),v'(S) is replaced by))

7
Consequentlya simplecomputation
yields))

= = \" ~ v\"((l,4'))= v\"((2,4')) v\"((3,4')) 2(3

1 The participants T. The participants

o -t- i?

1} the :4'. original Clearly

in

have 1,2,3 still

in

the

same role, but

and r' areplayers 1,2,3,4,5,the players 1,2,3

r are

who

4' is different.)))

all have the same role in compositeplayer (4,5)

336

REMARKS CONCERNING n

5 PARTICIPANTS

This then is the normalized and reducedform of the characteristic for function, as used in 34.2. the representation Q. (34:2)in 34.2.1 by the gives,togetherwith the above expression, formula))

Takingx\\ = x* = x* for granted,this relation can alsobe written asfollows:

(40:5)

(3

- xO(3+

n)

= 10.

four-persongamesF, class class. comparisonwith the 1,2,3-symmetric by ^ 17 g 2, Usingthe tablein Figure66 we obtain the zonesA-E in 1 x\\ ^ 1. The details which are the images of the zones A-E in appear in Figure67. A detailed analysis of the symmetric five-person games can be carried that out on this basis. It discloses the zonesA, B do indeed play the role b which we expect, ut that the zones C, D, E must be replacedy others, b ^ z* 2 and their inverse imagesA-D' <?',D'. Thesezones A-D'in 1 ^ Xi ^ 1 (again obtained with the help of the table of Figure66) in
t\\

Now it is easy to verify that (40:5)maps the Tj-domain (40:3)on the Xi-domain (40:4). The mapping is obviously monotone. Its details are shown in Figure66 and in the adjoining tableof correspondingvalues of t xi and 17. The curve in this figure representshe relation (40:5)in the x\\ ly-plane. This curve is clearly (an arcof) a hyperbola. 40.3.3. analysis of the 1,2,3-symmetric Our four-person game has The game, i.e.the diagonal culminated in the result stated in in 7-Center-V7/7 Q which representsthem, is divided into five classes A-E, each of whrch is characterized a certain qualitative type of by solutions. The positions of the zones A-E on the diagonal 7-CenterVIII, the interval 1 g x\\ ^ areshown in Figure64. The present results suggesttherefore the consideration of the coro respondingclassesf symmetricfive-person gamesF in the hope that some heuristic leadfor the detectionof their solutions may emergefrom their
9

36.3.2.:

i.e.

1,

areshown in Figure68.

It is remarkablethat the Zi-diagramof Figure68 showsmore symmetry than that of Figure67, although it is the latterwhich is significant for the
four-persongames. 1,2,3-symmetric

The 40.3.4. analysis of symmetric five-person games has also some heuristic value beyond the immediate information it gives. Indeed, y b comparing the symmetric five-person game F and the 1,2,3-symmetric to gameF'which corresponds it, and by studyingthe differences four-person between their solutions, one observes the strategiceffects of the merger of players4 and 5 in one (composite) player 4'. Totheextentto which the in d solutionspresentno essential ifferences (which is the case the zonesA, d above)onemay say that this merger id not affect the really))) B,as indicated

THE SYMMETRICFIVE-PERSONAME G

337))

Corresponding values of Xi

*i:

*:-!and
17:
\\

Figure

66.
}
A B J

-1 -I -i i i
I)<

i !

1
2))

1)

|C| D
t
1

\\

|)

-I)

i
B
\\C\\ D
I

1)

A)

fi)

1))

o t

Figure 67.))

-l))

-i
A

C-)) -t-))

1*1*1))
Figure
68.)))

338

n REMARKS CONCERNING 1

5 PARTICIPANTS

considerations. On the otherhand, when such differsignificant strategic encesarise(this happens in the remaining zones) we face the interesting in situation that even when 4 and 5 happen to cooperate T, their joint 2 is dislocatedby the possibilityof their separation. position o basedon the rigorousconcept f solutions. f Spaceorbidsa fuller discussion
1Of course must expect,n solutions of F, arrangements where the players 4 and 5 i one are ultimately found in opposing coalitions. It is clear that this can have no parallel in a of d in F'. All we mean by the absence essential ifferences is that those imputations to solution of T which indicate a coalition of 4 and 5 should correspond equivalent imputa-

tions in the solution of but Theseideasrequire further elaboration, which is possible, it would lead too far to undertake it now. 1Already in that the our first discussion of the three-person game disclosed w division of proceeds ithin a coalition is determined by the possibilities of eachpartner In our in caseof separation. But this situation which we now visualize is different. present it can happen that even the total share of player 4 plus player 5 is influenced by this \"virtual\" fact. A qualitative idea of such a possibility is best obtained by considering this When a preliminary coalition of 4 and 5 is bargaining with prospective further allies, their bargaining position will be different if their coalition is known to be indissoluble (in than when the opposite known to be a possibility (in r).))) is

r'.

22.2.,

r')

CHAPTER IX

COMPOSITION AND DECOMPOSITION GAMES OF

41.1.1. last two chapters will have conveyeda specificideaof the The a rapidity with which the complexityof our problemincreasess the number n of participants goesto 4,5, etc. In spite of their incompleteness, those considerationstended to be so voluminous that it must seemcom1 to pletelyhopeless push this casuistic approachbeyondfive participants. the of Besides, fragmentary character the results gainedin this manner very i seriouslylimits their usefulnessn informing us about the general ossibilities p of the theory. On the other hand, it is absolutely vital to get some insight into the conditionswhich prevail for the greatervalues of n. Quiteapart from the fact that theseare most important for the hoped for economicnd socioa With every increase of logicalapplications,there is also this to consider: n, qualitatively new phenomena appeared.This was clearfor each of n = 2,3,4 20.1.1., 35.1.3., also the remarks of footnote 2 and (cf. 20.2., on p.221), if we did not observeit for n = 5 this may be due to our and lack of detailed information about this case. It will developlater,(cf. the end of 46.12.), very important qualitative phenomena make their that
first

41.Composition Decomposition and 41.1. for n-personGamesfor Which All Solutions Can Be Determined Search

is imperative that we find some technique for the attackon games with higher n. In the presentstateof things we cannot hope for anything systematic or exhaustive. Consequently the c natural procedures to find some special lassesf games involving many o i 2 that can be dealt with. It is a general xperience decisively e participants that a completenderu in many parts of the exactand natural sciences c but of suitable special ases which are technicallymanageable, standing to which embody the essential rinciples has a good chance be the pacep maker for the advance of the systematicand exhaustivetheory. cases. They We will formulate and discusstwo such families of special can be viewed as extensive generalizationsof two four-persongames so that eachone of these will be the prototype of one of the two families. to Thesetwo four-persongames correspond the 8 cornersof the cube Q, w introduced in 34.2.2.: Indeed, e saw that thosecornerspresentedonly
1As was seen in Chapter VIII, it was already necessary for restrict ourselvesto the symmetric case. 1And in such a manner that eachone plays an essential ole! r
339)))

41.1.2. thesereasonsit For

f appearanceor n = 6.

five participants

to

340

tions to which this chapter and the following one will be devoted.

OF AND COMPOSITION DECOMPOSITION GAMES two strategically different types of games the corners/, Vj VI, VII, discussedin 35.1. the corners//, ///, IV, VIII,discussedin 35.2. and Thus the corners/ and VIIIof Q are the prototypes of those generaliza-

to the other

themselvesand entirely unrelated to the other set. The natural generalization of this is to a game F of n = + I participants, with the following property: The participantsfall into two setsof k and I elements, espectively,which have no dealings with each other. r I.e.he playersof eachsetmay be consideredlaying a separate ame,say t g p A and H respectively,strictly among themselvesand entirely unrelated
A;

A consider cornerVIIIof Q, discussedin 35.2. s the this 35.2.2. gamehas the following conspicuousfeature: a Thefour participants fall into two separateets(oneof threeelements nd s the otherof one)which have no dealingswith eachother. I.e.he players t of eachset may be considereds playing a separate ame,strictly among a g

41.2.1. first We

41.2. First The

Type. Composition

and Decomposition

was brought out in

We will describe relationshipbetween the gamesT, A, H by the this following nomenclature: Composition of A, H produces F, and conversely 2 F can be decomposed into the constituents A, H. 41.2.2. we deal with the above verbal definitions in an exact Before way, some qualitative remarks may be appropriate: and o First,it shouldbe noted that our proceduref composition decomposition is closelyanalogousto one which has been successfullyapplied in 3 many parts of modern mathematics. As these matters are of a highly technicalmathematical nature, we will not say more about them here. Sufficeit to statethat our presentprocedure as partly motivated by those w analogies. The exhaustive but not trivial results, which we shall obtain of of 35.2. second the setconsisted oneisolated player, who was This suggestsan alternative generalization to the aboveone: A game in which the participants fall into two setssuch that those of the first set play a set while those of the second have no influence upon game strictly among themselves etc., are the game, neither regarding their own fate, nor that of the others. (These then the \"dummies.\ This is, however, a special aseof the generalization in the text. It is subsumed in c it by taking the game H of the second setas an inessential one, onewhich has a definite value for eachoneof its participants that cannot be influenced by anybody. (Cf.27.3.1. and the end of 43.4.2. A player in an inessential game could conceivably deteriorate his We ought to excludethis possibility for a position by playing inappropriately. but this point is of little importance.) \"dummy\" Thegeneraldiscussion,which we are going to carry out (both games A and H essen-

1 set.

alsotermed a \" dummy.\"

1In the

original game

i.e.

c a tial) will actually disclose phenomenon which doesnot arisein the special aseto which the corner of the case of \"dummies\" (H inessential). The belongs new phenomenon will be discussed 46.7., 6.8., nd the case of \"dummies\" in where 4 a in 46.9. nothing new happens 2 would seemnatural to extend the of concepts composition and decomposition to more than 2 constituents. This will becarried in 43.2., out 43.3. 3 Cf.G. Birkhoff & A MacLane: Survev of Modern Ahrebra. New York.

VIII 35.2.

i.e.

It

Chapt.XIII.)))

S.

1941.

COMPOSITION AND DECOMPOSITION

341

and alsobe able to use for further interpretationsarea rather encouraging symptomfrom a technical oint of view. p 41.2.3. the readermay feel that the operation of composition Second, is of an entirely formal and fictitious nature. Why should two games, A and H, played by two distinct setsof players and having absolutelyno influence upon eachother,be considered one game F? as Our result will disclose that the completeeparation of the games A s and H, as far as the rules are concerned, does not necessarilyimply the same for their solutions. I.e.: lthough the two sets of players cannot A as influence eachother directly, neverthelesswhen they are regarded one set, one society there may be stable standards of behaviour which 1 establishcorrelations betweenthem. Thesignificanceof this circumstance will be elaborated ore fully when we reachit loc. it. c m 41.2.4. of it Besides, shouldbe noted that this procedure composition is quite customary in the natural sciences well as in economicheory. as t Thus it is perfectly legitimateo considertwo separate echanical systems m t situated, to take an extremease, ay one on Jupiterand one on Uranus c s o as one. It is equally feasibleto consider internal economiesf two the betweenwhich aredisregarded asone. countries the connections separate This is, of course, he preliminary step before introducingthe interacting t in forcesbetweenthosesystems. Thus we could choose our first example as those two systems the two planets Jupiter and Uranus themselves (both in the gravitational fieldof the Sun),and then introduceas interaction the gravitational forces which the planets exerton each other. In OUT secondexample, interaction enters with the considerationof interthe national trade,international capital movements, migrations,etc. We could equally use the decomposable game F as a stepping stone to othergamesin its neighborhood, in their turn, permit no decompowhich,
2 sition. In our present considerations,however, these lattermodifications will i not be considered. ur interestis in the correlationsntroduced by the O solutionsreferred to at the beginning of this paragraph.

to Let us now proceed the strictly mathematical description of the composition and decomposition games. of Let k players , k ) play the , fc', forming the set J = , V) , I\", forjning the set K = (1\", game A; and I players 1\", play the game H. We re-emphasizethat A and H are disjoint sets of players and8 that the gamesA and H arewithout any influence upon each 1 There is sortie analogy between this and the phenomenon noted before (cf. 21.3., that 37.2.1.) a symmetry of the game neednot imply the same symmetry in all solutions. a 1 Cf. 35.3.3., to the neighborhood of corner which according to 35.2. is applied g decomposableame. The remark of footnote 2 on p. 303 on perturbations is also

41.3.1.

Exact 41.3. Definitions

1',

(!',

/.,

pertinent. 8 If the same players entirely different situation

1,

prevails. That is

t n

are playing

two games, then an simultaneously the superposition of games referred to

in)))

OF AND COMPOSITION DECOMPOSITION GAMES functions of these two games by other. Denotethe characteristic 342

and v H (!T)respectively,where S zJ and T K. In forming the compositegame F, it is convenient to use the same 1\", ,/\" for its n = k + I players.1 They symbols 1',
form

v&(S)

Clearlyevery set R

the set / = J u K =

,*',

(41:1)
(41:2)

/ permits a unique representation # = SuT, SsJ, TzK;

S = #n/,
T

1 (!', -,*',\",

V).

the inverseof this formula being

= R n K*

Denotethe characteristic function of the game F by vr(fi) with R /. The intuitive fact that the games A and H combinewithout influencing :The eachotherto F has this quantitative expression value in F of a coalition R I obtains by addition of the value in A of its part S (sJ) in J and of the value in H of its part T (sK) in K. Expressedby a formula: where R, S, T are linked by (41:1), vr(fl) = v (S) + v R (T) (41:3) 8 i.e.41:2). ( the 41.3.2. form (41:3) The vr(R) expressed composite by means of its constituents VA(), v H (T). However, it also contains the answer to the To inverseproblem: expressA(/S), v H (T) by v r (#). V
A

Indeed v (0) = v H (0) = O.4 Hence putting and S = Q in (41:3) gives:

A

alternately T =

(41:4) (41:5)

vA
vH

(5)=vr (S) (r) = v r (r) / J

for for

T <=

SsJ, #.
6

two suitable gamesA (among the elements and H (among the elements J) of K). As stated,this is an implicitproperty of F involving the existence as of the unknown A, H. But it will be expressed an explicit property of F. If Indeed: two such A, H exist,then they cannot be anything but those t described (41:4),(41:5). Hencehe property of F in questionis, that the by

the of We arenow in a positionto express fact of the decomposdbility the t the given game F F with respecto the two sets and game o into (among the elements f = u K) is such that it can be decomposed

K. I.e.:

to formulate it. is feasible without difficulty. a Of course, rigorous deduction on the basisof All of appliesin this case. 4 Note that the empty set is a subset of both and K\\ sinceJ and K are disjoint, it is their only common subset. 6 This is an instance of the technical usefulness of our treating the empty set asa coalition. Of.footnote 2 on p. 241.)))
find it profitable
8

I and 27.6.2. alsoin 35.3.4.ts influences on the strategy are much more complex and describabley generalrules, as was pointed out at the latter loc.cit. b scarcely 1Insteadof the usual 1, , n. 1Theseformulae (41:1), :2)have an immediate verbal meaning. Thereadermay (41

25.3.2.

25.1.3.

COMPOSITION AND DECOMPOSITION

A,

343

H of (41:4), fulfill (41:5) into (41 using (41 to :3),

(41:6)
Or,
if

(41:7)

of proper light, a detailedreconsideration the basic principlesupon which 41.4.-42.5.2. which they rest,is necessary. This will be done in sections follow. However, two remarks concerning the interpretation of these equationscan be made immediately. First: 41:6) that ( expresses a coalition between a set S s J and a set T SK has no attraction that while there may be motives for players within J to combine with eachother, and similarly for players within K, thereareno forces acting acrossthe boundariesof J and K. Second: those readers ho are familiar with the mathematical theory To w of measure,we make this further observation in continuation of that made at the end of 27.4.3.: is exactly Carath6odory'sdefinition of (41:7) i measurability. This concepts quite fundamental for the theory of additive measureand Carath6odory'sapproach to it appears to be the technically l in i superiorone to date. Itsemergence the present context s a remarkable fact which seemsto deservefurther study.

= v r (fi n J) + v r (fl n X) for Bel. 41.3.3.order to seethe role of the equations (41:6), in In (41:7)
vr(fi)

#. we use (41:2)(expressing of S,,rin terms of R) in place (41:1)

v r (S u T)

= v r (S) + v r (T)

:1)

R express in terms of /S, T gives this:

(41:3). Substituting,therefore, (41:4), (41:5)

for

J,

T7 S

the

41.4.1. obtained the criteria (41:6), of Y's decomposability We (41:7) r into by substituting the v A (/S), v H (T) obtained from (41:4), (41:5) the fundamental condition (41 :3). However,this deductioncontainsa lacuna: We did not verify whether it is possibleto find two gamesA, H which producethe v A (S), v H (5P) formally defined by (41:4), (41:5). Thereis no difficulty in formalizing theseextrarequirements.As we know from 25.3.1. mean that v*(S)and vH (T) fulfill the conditions they eod. It (25:3:a)-(25:3:c) must be understood that we assume the given
that vp(fi) Vr(B) to originate from a game F, Hence following questionpresentsitself: the in v r (B) fulfills (41:A) above

41.4. Analysis

of Decomposability

i.e.

fulfills

theseconditions.
the

is not the case,which further postulate must be imposedupon

W i.e. (41:6), (41:7). ill then the v (S) and vH (T) of in 25.3.1.? (41:5) (41:4), also fulfill (25:3:a)-(25:3:c) Or, if this
A

25.3.1. (25:3:a)-(25:3:c) togetherwith

7 and v H (7).

ment for VA(*S) and v H (T) as for v r (#).

1Cf. C. Carathtodwy:
Vorlesungen liber

It is convenient to takethem up in a different order. 41.4.2. (25:3:a):virtue of (41:4), this is the same stateAd (41:5), By
ReelleFunktionen,
Berlin,

of 25.3.1. Todecidehis, we check(25:3:a)-(25:3:c) separatelyfor v (S) t

A

vr(B)?

1918, Chapt.

V.)))

344
Ad Vn(T)

AND DECOMPOSITION GAMES OF COMPOSITION

from Vr(fi)

By (25:3:c):virtue

of

we the Beforediscussing remaining (25:3:b), insert a remark concerning o it is Sincethis is a consequencef (25:3:a)-(25:3:c), (25:4)of 25.4.1. from it and it will be seenthat this for legitimate us to draw conclusions anticipationsimplifies the analysisof (25:3:b). sets From hereon we will have to use promiscuously complementary in 7, J, K. It is, therefore, necessaryto avoid the notation S,and to write instead I S,J S,K S,respectively. : Ad (25:4) Forv (S) and v H (T) the role of the setI is taken over by the setsJ and K, respectively. Hencethis condition becomes:
A

S J and T s K.

it amounts only to a

o (41:4), this carries ver to v (S) and (41:5), restriction from the R c I to
A

v A (J)

= 0,
0.))

Owing to

(41 t (41:4), :5),his means

VrO/)

(41:8) (41:9)
SinceK = I
(41:10) and Thus (41:8) assumedto hold) gives))

vp(K)

=0, =

0.
v r OS) for which it was

/, therefore

(25:3:b) (applied to + Vr(K) = 0.

v r (J)

(41:9) eachotherby virtue of the identity (41:10). imply or we (41:8) (41:9) have actually a new condition, which does not or follow from (41:6) (41:7).
In
Ad (25:3:b): will derive its validity for v A (S) and v H (7) from the We assumedonefor VF(#). By symmetry it sufficesto considerv&(S). The relation to be proven is

(41:11)
By (41:4)this means

v A (S)

+ v&(J S) = 0.

0. vrOS) + v T which we must require anyhow, this may be written Owing to (41:8), (41:12) (41:13)
For thesesets/ becomes
vrOS) + v r (J

(J - S)
R

- S) =

v r (J)

S (Of course, J.) Toprove (41:13), (25:3:b) v r (fl) to R for apply

= SuK and /

i.e.

VrOS)

- VrOSutf) = -v (X),
r

= J S and R = J. = K, respectively. So (41:13)

VrOS u

case and this is the special of (41:6) T = with

K) = vrOS) + v r (X),
K.)))

MODIFICATIONOF THE THEORY paragraph and answeredthe questionsof (41: A).

345

Thus we have filled in the lacuna mentioned at the beginning of this The further postulate which must be imposedupon vr(fi) is this:41:8), (41:9). i.e. (

(41:B)
All

theseput togetheranswerthe questionof 41.3.2. concerning decomThe game F is decomposable ith respect the sets J and K w to if and only if it fulfills theseconditions: 41:6), (cf. 41.3.2.) (

posability:

(41:C)

and i.e. (41:7) (41:8), (41:9).

i.e.

The two conditionswhich we proved equivalent to decomposis ability in (41:C)areof very different character.(41:6) (41:7)) the (i.e. really essentialone,while (41:8)(i.e. (41:9))expresses a rather incidenonly tal circumstance.We will justify this rigorously below,but first a qualitative remarkwill be useful. The prototypeof our concept f decomposition o was the game referred to at the beginning of the game represented but by the corner VIII of 35.2.Now this game fulfilled (41:6), not the latterfrom v(J) = (41:8). former follows from (35:7)in (The v((l,2,3))= 1 7* 0.) We neverthelessconsideredthat game as decomK posable (with J = (1,2,3), = (4)) how is it then possible, that it for violates the condition (41:8)which we found to be necessary the decom-

41.5.1.

41.5. Desirability

of

Modification

41.2.1.: 35.2.1.,

posability?

In conditionthat we derived (41:8)). other words:the constituentsof F are not zero-sum games. This point, of course,was perfectly clearin where 35.2.2., it receiveddue consideration. Consequentlywe must endeavor to get rid of the condition (41:8), recognizing that this may force us to considerother than zero-sum games.

F 41.5.2. answeris simple: or the above game the constituents A The J = (1,2,3))and H (in K = (4)) do not completelysatisfy (25:3:a)(in the consequence To in t (25:3:c) 25.3.1. be precise,hey do not in 25.4.1.: = (#) = is not true (and it was from this (/) (25:4)
fulfill VA VH

42.1. CompleteAbandoning No

42.Modification of the Theory

of the Zero-sum
n

Condition

1 42.1. Completeabandonment of the zero-sum condition for our games it would mean that the functions 3C*(ri, , r ) which characterized t are in the senseof 11.2.3.entirely unrestricted. I.e.hat the requirement

(42:1)

Jb-l

2) JC*(n,

, r n) ,
n.)))

BE

1We again denote the players by

1,

AND OF COMPOSITION DECOMPOSITION GAMES is dropped, with nothing elseto take its place. This of 11.4. 25.1.3. and a would necessitaterevision of considerable significance, sincethe construci.e. and function in 25. ependedupon (25:1), (42:1), tion of the characteristic d would therefore have to be taken up de novo. n Ultimately this revision will becomeecessary(cf. Chapter XI) but not yet at the presentstage. idea In order get a precise of just what is necessarynow, let us make to

346

containedin the auxiliary considerations

42.2.2. 42.2.1., below.

the a?, , aj described 1 to the samefor F'. there. It is evident,that (41:6)for F is equivalent from F to The situation is altogetherdifferent for (41:8). Passage F' changesthe left hand side of (41:8) ajj, hencethe validity of by
kinJ

game r' in the senseof

42.2.Strategic Equivalence. Constant-sum Games Consider a zero-sum game F which may or may not fulfill 42.2.1. and Passfrom F to a strategically equivalent conditions (41:6) (41:8).

27.1.1., with 27.1.2.,

this is true: (42:A)

in (41:8) one caseis by

no means implied by that in the

other. Indeed

a Forevery F it is possibleto choose strategicallyequivalent F'so that the latterfulfills (41:8). game we can choose aj,
,a
with

T Proof: heassertionis1 that

in so that (this is (27:1) 27.1.1.)

] = *iaj

v(J) +
Now this is obviously possibleif

kinJ

aj =
or

J 7*

then v(J) =

given any assignedvalue.

by

This result can be interpreted follows:If we refrain from considering as 2 other than zero-sum games, then condition (41:6) that expresses while the gameF may not be decomposable it is strategicallyequivalent itself, 3 to some decomposable gameF'. 42.2.2. above rigorous result makes it clearwhere the weakness The of our present rrangement lies.Decomposability an important strategic is a property and it is therefore inconvenient that of two strategicallyequivalent one without the other. It is, therefore, games may be termed decomposable
(27:2)loc.cit.
1By (27:2)in

and (25:3:a)25.3.1. (25:4)in 25.4.1.

in

ForJ = Q or J, there is nothing

/, sincethen *m J ]

can be

to prove, as

27.1.1. that Observe

p. 300shows

the v r GS), v r 'OS) f (42:A)are the o

v(S),v'(S)of

* we require this not only for r, but * The treatment of the constituents in

I.e.

alsofor its constituents

tion

of footnote 1 on

35.2.2. amounts

to exactly this, as an inspec-

A,

H.

explicitly.)))

MODIFICATIONOF THE THEORY

347

desirable to widen these conceptsso that decomposabilitybecomesn a invariant under strategic equivalence. In otherwords:We want to modify our concept o that the transformas tion (27:2) 27.1.1., defines strategic of which d equivalence, oesnot interfere with the relationshipbetweena decomposable ame F and its constituents g A and H. This relationshipis expressed (41:3): by
T c K. S c J, (42:2) vr(S u T) = v (S) + vK (T) for Now if we use (27:2) ith the samecf for all threegamesF, A, H then (42:2) w
A k

is manifestly undisturbed. The only trouble is condition (27:1). his states F, A, H that)) T for
* in /
k in

with

the preliminary

= 0,

a
k in K

fc

- 0,
27.1.1.

respectively and while we now assumethe first relation true,the two others
o way out is to discard (27:1)f altogether. widen the domain of games, which we consider,by including all those games which arestrategicallyequivalent to zero-sumonesby virtue of the transformation formula (27:2)alone without demanding (27:1). this amounts to replacingthe functions)) As was seenin
may fail.

I.e.to

Hencethe natural

27.1.1. ' '

\"

\"

>

Tn)

of the

latterby

new functions

(The a?, functions 3C(ri, of functions 3Cfc(ri, (42:3)

Summing

, r n ) which areobtained in this way from the system in areeasy to char, r n ) which fulfill of the property is acterize.Thecharacteristic (in place

- , rn ) + JC*(n, 3Ci(n, , r n) T , aj are no longer subject to (27:1)). he systems of

al

(42:1) 42.1. loc. (42:1) cit.)

EE

up:
We

*i

3C;(n, -

, r n)

.i

(42:B)
At

passing from
18 is an
game from
arbitrary

the domain of gameswhich we consider, by the zero-sum games to the constant-sum games.* o the same time, we widen the concept f strategic equivalence,
constant

arewidening
> <

= 0.

In the transformation

this (27:2)which produces

a zero-sum one, there is obviously))

which we

1This gives a precise meaning to the statement at the beginning of to arenot yet prepared considerall gamesunrestrictedly.)))

42.1. according to

348

OF AND DECOMPOSITION GAMES COMPOSITION introduced in

eod. (27:2)loc. it., ut droppingthe condition(27:1) c b d 42.2.3.is essentialo recognize our above generalizations o not It that t alterour main ideason strategicequivalence.This is best done by consideringthe following two points.

27.1.1.,defining it by

again by transformation

zero-sumgames. Thereasonis this: (42:C)

that First,we stated in 25.2.2. we proposedto understand all quantif tative properties of a gameby means of its characteristicunction alone. One must realize that the reasons this are just as good in our present for domain of constant-sum games as in the original (and narrower) one of Every constant-sum game is strategically equivalent to a zero-sumgame.
a of ,

t Proof:The transformation (27:2)obviously replaceshe

n

(42:3)

*-i n as to make this s + J) Jb-i

above by s +

aj. Now it

is possibleto choosethe aj,

aj so

= 0,i.e.o carry the given constant-sumgame t

into a (strategicallyequivalent) zero-sumgame. Second,our new conceptof strategicequivalencewas only necessary for the sake cf the new (non-zero-sum) that games we introduced. Forthe old (zero-sum) gamesit is equivalent to the old concept.In otherwords If two zero-sumgamesobtain from eachotherby meansof the transformation (27:2)in t then (27:1) automatically fulfilled. Indeed,his is was already observedin footnote 2 on p.246.

27.1.1.,

42.3.The CharacteristicFunction in the New Theory 42.3.1. a constant-sum game F\" (with the 3C(ri, , rn) Given function v'(S) by (42:3)),we could introduce its characteristic fulfilling 1 On repeatingthe definitions of 25.1.3. the other hand, we may follow the procedure 42.2.3.: can We by suggested the argumentation of 42.2.2., obtain F' with the functions 3Cl(ri, , rn ) from a zero-sum game F with the functions JCjb(ri, i.e. , r w ) as in 42.2.2., by))

(42:4)
with

jei(n,
,

rn)

s3C*(r,,

, r n)

+ <tf

characteristic function v'(S) of Tf by means of (27:2) in

(42:5) v'(S) ^ V (5) +
kinS

appropriateotj,

aj (cf.footnote 1on p.246),and then define the

i.e. 27.1.1.,by

aj.
25.1.3.
A

1The whole arrangement of can be repeatedliterally, although T is no longer we must add to the and zero-sum, with two exceptions. In (25:1) (25:2)of extreme right hand term. (This is so, becausewe now have (42:3)in placeof (42.1).) This difference is entirely immaterial.)))

25.1.3.

MODIFICATION THE THEORY OF

349

Now the two procedures reequivalent, i.e.he v'(/S) of (42:4),(42:5) a t w coincidesith the oneobtained by the reapplicationof 26.1.3. an Indeed, shows inspectionof the formulae of 25.1.3. immediately,that the substitul f tion of (42:4) there producesthe result (42:5). 42.3.2. is a characteristicunction of a zero-sumgame,if and only f v(S) if it fulfills the conditions (25:3:a)-(25:3:c) as was pointed out of 25.3.1., there and in 26.2.(The proof was given in 25.3.3. 26.1.) hat do and W theseconditionsbecome the caseof a constant-sumgame? in In order answerthis question,let us remember, (25:3:a)-(25:3:c) to that loc. it.imply (25:4)in 25.4.1. c w Hence, e can add (25:4)to them, and modify (25:3:b) adding v(7) to its right hand side (this is no change by T o owing to (25:4)). hus the characterizationf the v(S) of all zero-sum this: gamesbecomes
*

and

(42:6:a) v() = 0, + v(-S)= v(7), (42:6:b) v(S) (42:6:c) v(5) + v(T) g v(S u T)
v(7)

if

S n T = 0,

(42:6:d)
ing them to

= 0.

Now the v'(S) of all constant-sumgamesobtain from thesev(/S) by subjectthe transformation (42:5) f 42.3.1. o Howdoesthis transformation affect (42:6:a)-(42:6:d)? areentirely unaffected, One verifies immediately,that (42:6:a)-(42:6:c) 3 is completelyobliterated. So we see: while (42:6:d)

(42:D)

if and only

f v(S) is the characteristicunction of a constant-sumgame

if

it

satisfiesthe conditions(42:6:a)-(42:6:c).

(We write from now on v(S) for As mentionedabove,

is (42:6:d)no longer valid.

v(7)

v'GS)).

However,we have

(42:6:d*)

= s.

It of t Indeed,his is clearfrom (42:3),consideringthe procedure 25.1.3. can also be deducedby comparingfootnote 1 on p. 347 and footnote 3 is intuitively clear: above (our v(S) is the v'(S) there). Besides42:6:d*) ( A coalition all players obtains the fixed sum s of the game. of
1The verbal equivalent of this consideration is easily found. 1Had we decided define v'(*S>) by means of (42:2),(42:5) to only, a question of ambican obviously be A guity would have arisen. Indeed: given constant-sum game w obtained from many different zero-sum games r by (42:4), ill then (42:5) always yield the same v ;(5)? It would be easy to prove directly that this is the case. This is unnecessary, however, becausewe have shown that the v'(S) of (42:5)is always equal to that one of and that v'(S)is defined unambiguously, with the help of F' alone.

r'

25.1.3.

1According to (42:5),he t
and this sum

right hand

sideof (42:6:d) over into goes

t in

?>

i-e
t

is completely

^1 ? -

arbitrary.)))

AND OF 350 COMPOSITION DECOMPOSITION GAMES 42.4.Imputations,

Domination, Solutions in the New Theory

functions of any 42.4*1. now on, we areconsideringcharacteristic From f constant-sumgame,i.e.unctions v(S)subjectto (42:6:a)-(42:6:c) only. to Our first task in this wider domain, is naturally that of extending it o the concepts f imputations,dominations,and solutionsas defined in 30.1.1. Let us begin with the distributions or imputations. We can take over from 30.1.1. interpretation vectors their as

a = {ai,
(42:7)

>

Of the conditions(30:1), eod. (30:2) we may conserve

-,). (30:1):

v((t)) 1 unchanged the reasonsreferred to there are just as valid now as then. eod., (30:2) however, must be modified. The constant-sumof the game above),eachimputation shoulddistribute being s (cf.(42:3)and (42:6:d*)
this amount

a,

i.e.t is natural i

to postulate

(42:8)
this By (42:6:d*) is equivalent to

(42:8*)
from which

-ia,

= v(/).

Thedefinitions of effectivity, domination, solutionwe take over unchanged

no led up to thosedefinitions, appear to lose strength by our present generalization. their final corroborations observ42.4.2. considerations These receive by
ing

3 1. 30.1.the supportingarguments brought forward in the discussions

this:

o Forour new concept f strategic of equivalence constant-sum 4 an gamesF, F', there exists isomorphismof their imputations, a i.e. one-to-one apping of those of F on those of F', which m 5 o 1. leaves the concepts f 30.1.invariant. This is an analogue of (31 in 31.3.3. it can be demonstrated in and :Q) the same As there,we define the correspondence way.

(42:E)

(42:9)
they must
with

1a < v((t)) would be unacceptable, e.g.he beginning of cf. t 2 For the special aseof a zero-sum game c c v(7) = so (42:8),(42:8*) oincide as

7^7'

29.2.1.

'I.e.30:3);(30:2) or loc. ( (30:4:a)-(30:4:c); (30:5:b), (30:5:c) cit., espectively. (30:5:a), r 4 As defined at the end of 42.2.2., (27:2) 27. i.e. without eod. by 1.1., (27:1) As redefined in 42.4.1.)))
in

loc.cit.

MODIFICATION THE THEORY OF

between the imputations a =
\\a h

351

an
J

of F and the imputations

\"?' =

a'] of
n

I\"

by

(42:10)
- ,

a' = a +
k k

where the a?, a arethose of (27:2)in Now the proof of (31 in carriesver almost literally. Theone o :Q) difference is that (30:2) of is replacedy our (42:8) since(27:2))) but b

31.3.3. 30.1.1.
n))

27.1.1.

27.1.1. v'(/) = v(7) + <*?,this too takes careof itself.1 The gives t-i readerwho goesover 31.3. will seethat everything elsesaid there again,
in

appliesequally to the present case. 42.5.Essentiality, Inessentiality, and

that 42.2.3. every constant-sumgame is to a zero-sum game. Hence :E) allowsus to strategically equivalent (42 carry over the general resultsof from the zero-sum gamesto the constantsum onesalways passingfrom the latter classto the former one by strategic (42:C)in

42.6.1. know from We

Decomposability in the New Theory

31.

equivalence.

This forces us to define inessentiality for a constant-sum game by strategicequivalence to an inessential zero-sum game. We may state

therefore:
(42:F)

A zero-sum game is inessential and only if it is strategically if or (27:C) equivalent to the game with v(S) = 0. (Cf. in 27.4.2.) y the above, the same is true for a constant-sum B game. (But we must use our new definitions of inessentiality and of strategicequivalence.)

23.1.3.

d Essentialityis, of course, efined as negation of inessentiality. to the Application of the transformation formula (42:5) of criteria 27.4. of shows,that thereareonly minor changes.

42.3.1.

(27:8)in (42:11)))

must 27.4.1.

be replaced by))

sincethe right hand side of this formula is invariant under (42:5)and it the c f goesover into (27:8)loc. it.or v(I) = (i.e. zero-sumcase). of for The substitution of (42:11) (27:8)necessitates replacement the of 27.4.1. on the right hand side of both formulae in the criterion (27:B) by
n

1And
in

this was the only point in the proof

referredto, at which

which 27.1.1.,

we no

longer require)

is used.)))

2, t-i S

(i.e.27:1) (

AND OF 352 COMPOSITION DECOMPOSITION GAMES are (27:D)of 27.4.2. invariant under (42:5), v(7). The criteria(27:C),

or this time no extracondition arises (like w Tosimplify matters, e give this proof in full. Ad We will derive its validity for v A (S) and v H (T) from the assumedone for vrCR). By symmetry it sufficesto consider
but applicable,

42.5.2. can now return to the discussionof compositionand decomWe in the 41.3.-41.4., wider domain of all constant-sumgames. All of 41.3. be repeated can literally. to the When we come 41.4.,, question (41 :A) formulated there again presents itself. In orderto determinewhether any postulates beyond are i.e. of (41:6), (41:7) 41.3.2. now needed,we must investigate (42:6:a)in 25.3.1. in instead (42:6:c) 42.3.2., of (25:3:a)-(25:3:c) (for all three of v r (/Z), v (S),v,(I*)). are (42:6:c) immediately disposed of, exactly as (25:3:a), (42:6:a), in the in is (25:3:c)41.4.As to (42:6:b), proof of (25:3:b)41.4. essentially
positionin
A

and henceunaffected.

loc. it.). c

(41:8) (41:9)

(42:6:b):

The relation to be proven

A

is))

becomes

v (S) + v (J S) = (42:12) this By (41:4) means VrOS) + v r (J (42:12*) S) = v r (J). for To prove (42:12*) (42:6:b) v r (fl) to R = / S and R = J. apply For these I R = SuK and I R = K, respectively. So (42:12*)
A

i.e.

Vr(S) + v r (J)
v r (S u

- Vr(S

K) = v r (7)

v r (X),

K) = Vr(S) + v r (X),

and this is the special of (41:6) T = K. with case Thus we have improved upon the result (41:C)of 41.4. follows: as (42:G)

In the domain of all constant-sum games the game F is the setsJ and K (cf. 41.3.2.) if and only if it fulfills the condition (41:6), (41:7). i.e.
with t decomposable respecto

42.5.3. shows and Comparisonof (41:C)in 41.4. of (42:G)in 42.5.2. that from zero-sumto constant-sumgames the passage rids us of the unwanted for condition (41:8), (41:9) decomposability. is now defined by (41:6), (41:7)alone,and it is Decomposability invariant under strategic equivalence as it should be. know that when a gameF is decomposed two (constituent) into We also A and H (all of them constant-sum only!),we can make all these games for gameszero-sum by strategicequivalence. (Cf. (42:C) in 42.2.3. F, et f and then (42:A) in 42.2.1.sequ. or A, H.))))

i.e.

i.e.

THE DECOMPOSITION PARTITION

363

consideration. In the remainder of this chapter we will continue to consider constant-sum games, unless the opposite is explicitly stated. 43.The Decomposition Partition 43.1. Sets. Constituents Splitting 43.1. defined the decomposabilityof a gameF not per se,but with We
to of respect a decomposition the set/ of all playersinto two complementary K. sets,J, Thereforeit is feasible to take this attitude:Consider the game F as given, and the setsJ', K as variable. SinceJ determines K (indeed K = 7 J), it suffices to treat J as the variable. Then we have this

Thus we can always use one of the two domains of games zero-sum or constant-sum whichever is more convenient for the problemjust under

question:

and (41:4) 41.3.2.) J-constituent of F.1 of is the A splitting set J is thus defined by (41:6), (41:7) 41.3.2., i.e. in where K = 7 J must be substituted. h w The reader ill note that this concept as a very simpleintuitive meanA splittingsetis a self containedgroup of players,who neither influence, ing: nor areinfluenced by, the othersas far as the rulesof the game areconcerned. 43.2.Propertiesof the System of All Splitting Sets 43.2.1. totality of all splitting setsof a given gameis characterized The an aggregate simple properties.Most of these have an intuitive of by immediatemeaning, which may makemathematical proof seemunnecessary. We will neverthelessproceed systematically and give proofs,stating the intuitive interpretations in footnotes. Throughout what follows we write
function of F). v(S)for v r (S)(thecharacteristic

setof players/) for which setsJ s / (and the K = / J) is F decomposable? corresponding We call those J(QI) for which this is the casethe splitting sets of T. The constituent game A which obtains in this decomposition(cf. 41.2.1.
Given a game F (with the

(43: A)

is one.
2

J is a splitting set if

and only if its complementK = /

of T Proof: he decomposability F involves J and K symmetrically. 8 and / aresplitting sets. (43:B) w or Proof: (41:6) (41:7) ith J = , K = / are obviously true, as = 0. v(0)
and same definition the game H (cf. (K / J) of T. 1That a set of players is self-containedin the senseof ment, as that the complement is self-contained. 1That theseare self-containedis tautological.)))

1By the K-constituent

- -

is then 41.2.1. (41:5) 41.3.2.) in

the

is clearly 43.1., the same state-

354 (43:C)

AND COMPOSITION DECOMPOSITION GAMES OF

43.2.2.

1 J\" and J' u J\"aresplitting sets if J', J\"are. A A for Proof: d J' u J\": s J', J\" are splitting sets,we have (41:6)f J, K equal to J', 7 J' and J\", / J\". We wish to prove it for J u J\", I (J'uJ\.") Consider therefore two S J u J\", T 7 (J9 u Let S'be the part of S in J', then S\"= S S'lies in the complementof J', and as S J' u J\", S\" also lies in J\". So S = S'+ S\", S'c J',
n

/'

/\

S\"S /\". Now S'

(43:1)
Also

Next S\" 7 also gives

S'c

j'.-

v(5) = v(S') + v(S\. J' and T s I (J'u J\")s 7 J' so ClearlyS'u (S\"u T) = S u T. Hence(41:6)for J', 7

j',S\"c/ - J' and (41:6)for J', I - J' give

S\"uTsI-J'. J'

J\",7

v(S u T) = v(S') + v(S\"u 7 ). for (J'u J\") 7 J\". Hence (41:6) Finally S\"<=J\" and Ts7 (43:2) (43:3)

- J\"gives

v(S\"u T ) = v(S\")+ v(T). Now substitute (43:3)into (43:2)and then contractthe This by (43:1). gives

right

hand side

v(S u T) = v(5) + v(T),

as is (41:6), desired. J' n J\":Use (43:A) and the above result. As J', J\"are splitting for obtains successively 7 J', I J\",(I J') u (7 sets,the same n 7\") 2, and J' n /\" the last one beingthe desired which is clearly7 (/'
which
Ad

/\

expression.

43.3.Characterization

Sets. TheDecomposition Partition It 43.3.1.may be that thereexistno othersplitting setsthan the trivial w ones , 7 (cf. (43:B)above). In that case, e call the game F indecomposof the System of All Splitting

able* Without studying this question any further, 4 we continue to investigatethe splitting sets of F. setsJ',J\" should
1The intersection

readeras odd, that two self-contained a non-empty intersection at all. This is possible, however, as the reasonis that a self-containedset may well be the J\" shows. The deeper example sum of smaller self-containedsets (proper subsets). (Cf.(43 in 43.3.) Our present :H)

/'

have

J' n J\":It may strike the J',

assertionis that plausible.

/\" have a non-empty intersection if two self-containedsets subset. In this form it will probably then this intersection is such a self-contained

/'

appear

/\",

T Thesum U J\": hat the sum of two self-containedsetswill again be self-contained stands to reason. This may be somewhat obscuredwhen a non-empty intersection above. The proof which n J\" exists, but this case is really harmless as discussed follows is actually primarily an exactaccountof the ramifications of just this case. 1Thecomplement of the intersection is the sum of the complements. 1Actually most games are indecomposable; otherwise the criterion (42 :G)in 42.5.2. :6),(41:7) in requires the restrictive equations (41 4 Yetl Cf. footnote 3 and its references.)))

/'

/'

41.3.2.

when

THEDECOMPOSITION PARTITION 355 Considera splitting set J of F and the /-constituent of F. A (43:D) Then a J' J is a splitting setof A if and only if it is one of F.1 J' Proof: onsidering(41:4), is a splitting set of A by virtue of (41:6) C
for 8 J', T fi J J'. (43:4) v(S u T) = v(S) + v(7) (We write v(S) for v r (/S)). Again by (41:6)J' is a splitting set of F when for S J', J c / J'. (43:5) v(S u T) = v(S) + v(T) A We must prove the equivalence of (43:4)and (43:5). s J J, so is clearly a special caseof (43:5) hencewe needonly prove that (43:4) (43:4)implies(43:5). Assume,therefore, (43:4). We may use (41:6)for F with J, K = / J. f s be the of T ' Consider two Sin J', TJ. 7 SoJ'.=Let uTT\", T cpartT\" c /in J, then T' = T T' lies 7 T T' J and J,
1

(41:6)
(43:6) (S u
T;)

for F

with

J, I J give
T' cJ
7

- -

Next

SzJ'zJ and '=

u

v(r) = v(2\")

77

S u T.

(43:7)

Finally (43:4)gives

Scj'and T'&I-J'and

J, J also gives v(S u T) = v(5 u T') + v(T\.

with

so Hence(41:6)for F

SuT'sJ.Also

+ v(T\.

T\" 7

/ - J.

Clearly

T'cJ, so

T'sJ-J'. Hence
right

(43:8)
by

T (43:6). his gives preciselythe desired (43:5). t 43.3.2. :D) makes it worth while to considerhose splitting setsJ, (43 for which J ^ 0, but no propersubset J Q of J is a splittingset. We
1

v(S u T') = v(S) + v(2\. Now substitute (43:8)into (43:7)and then contractthe

hand side

call such a

setJ, for obvious reasons,a minimal splitting set. Consider our definitions of indecomposabilityand of minimality. (43:D) impliesimmediately: A if The J-component (of F) is indecomposable and only if (43:E) J is a minimal splitting set.
The minimal splitting sets

statement follows:

form an arrangement with very simple and they determine the totality of all splitting sets.The properties,

from the

(43:F) Any two different minimal splitting setsaredisjunct. The sum of all minimal splitting setsis 7. (43:G) 1 To be self-contained within a self-containedset, is the same thing as to be such in the original (total) set. The statement may seemobvious; that it is not so, will appear
proof.)))

AND OF 356 COMPOSITION DECOMPOSITION GAMES (43:H) o By forming all sums of all possibleaggregates f minimal splitting sets,we obtain precisely the totality of all splitting
1 sets.

Proof:Ad (43:F) Let J\"be two minimal splitting setswhich are and a not disjunct. Then n J\" 7* Q is splitting by (43:C), s it is SJ\". So the minimality of and J\" implies that J r n J\" is equal to
both J 1 and J 11 . Hence = J\".
Ad

: /', J'

splitting set.

(43:G):It sufficesto show that every k in / belongsto someminimal

/'

J'

J'

J does.Then J' is among the setsof which J is the intersection. J Hence 1aJ. But as J' J and J' 7* J, this is impossible. Ad (43:H): very sum of minimal splitting setsis splitting by (43 :C), E so we needonly prove the converse. Let K bea splittingset. If J is minimal splitting, then J n K is splitting by (43:C),also J n K J henceeither JnK = QoTjnK= J. In the first case/, K aredisjunct, in the second K. So we see: J (43:1) Every minimal splitting set J is either disjunct with K
that
1 or))

Thereexist plitting setswhich contain the player k (i.e. ; let J be the s /) o intersectionf all of them. J is splittingby (43:C). If J werenot minimal, then there would exista splitting set J' ?* , J, which is J. Now a J\" = J J' = J n (/ J') is also a splitting set by (43:A), (43:C), nd also J\" j* , J. Either J' or J\" = J1 must contain k say clearly

LetK'bethe sum of the former J, and K\" the sum of the latter. K'u K isthe sum of all minimal splitting sets,henceby (43:G) = K'uK\" 7. (43:9)
By their origin K is disjunct with K, and K\" is

11

'

with the sum 7. We call this the form a partition in the senseof decompositionpartition of T, and denoteit by H r Now (43:H) can be

X\"K (43:10) K\" Now (43:9), (43:10) togethernecessitate = K\\ henceIf is a sum of a of minimal sets,as desired. suitable aggregate 43.3.3. :F),(43:G) make it clearthat the minimal splitting sets (43 8.3.1.,
as expressed follows:
(43:H*)
A

#'/-#,

K. I.e.

is characterized the following splitting set by The points of eachelementof II go togetheras r property: far as K is concerned i.e. achelementof II lies completely e r insideor completelyoutside of K.

Ksl

1The intuitive meaning of theseassertionshould be quite clear. They characterize s the structure of the maximum possibilitiesof decomposition of F in a plausible way.)))

THE DECOMPOSITION PARTITION

without

357

Thus n r expresses far the decomposition T in / can be pushed, how of destroyingthoseties which the rulesof T establishbetweenplayers.1 of r By virtue of (43:E) the elements II are also characterizedy the fact b that they decompose F into indecomposable onstituents. c 43.4.Properties the DecompositionPartition of 43.4.1. nature of the decompositionpartition Ilrbeing established, The it is natural to study the effect of the finenessof this partition. We wish to When II is as fine as possible, r analyze only the two extreme ossibilities: p i.e. hen it dissectsdown to the one-element and when Erisas coarse w / sets as possible,i.e. hen it doesnot dissect at all. In otherwords:In the first w I caseHr is the system of all one-element (in 7) in the secondcaseIlr sets consistsof / alone. The meaning of thesetwo extremeasess easily established: c i n r is the system of all one-element (in 7) if and only if sets (43 :J) the gameis inessential.
Hr

(43:H) or (43 :H*) that the stated property of is equivalent of sayingthat all setsJ(s. ) aresplitting. I.e. by 43.1.) ( 7 that for any two complementarysets J and K(= I J) the gameF is This holds in all those cases. This decomposable. means that (41:6) S that the condition imposedby (41:6)on S,T (i.e. Si J, implies, however, T K) meansmerelythat S,T aredisjunct. Thus our statement ecomes b

Proof:It is clearfrom

v(S u T) = v(S) + v(!T)

for

Sn T =

Now this is preciselythe condition of inessentialityby (27:D)in 27.4.2. n r consistsof 7 if and only if the gameF is indecomposable. (43 :K)

I t Proof: t is clearfrom (43:H)(or (43:H*)),hat the statedproperty of Ilr is equivalent to saying that 0, 7 arethe only splitting sets.But this

is exactly the definition of indecomposability the beginningof 43.3. at Theseresult^ show that indecomposabilityand inessentialityare two for oppositeextremes a game. In particular, nessentialitymeans that the i at can decompositionof F, described the end of 43.3., be pushed thrdugh to the individual players,without ever severingany tie that the rules of the 2 t game F establish. The readershould comparehis statementwith our original definition of inessentialityin 27.3.1. 43.4.2. connectionbetween inessentiality, decomposability,and The the number n of playersis as follows: n = 1:This caseis scarcelyof practicalmportance. ucha gameis S i 8 and time inessentialby the first clearly indecomposable, it is at the same remark in 27.5.2. 1 I.e. ithout of w impairing the self-containednessthe resulting sets.
2 that every player is self-containedin this game. 3 As is a one-element set , are its only subsets.)))

I.e.
/

AND 358 COMPOSITION DECOMPOSITION GAMES OF It should then be noted that indecomposabilityand inessentialityare ( by (43:J), 43:K) incompatiblewhen n ^ 2,but not when n = 1. n = 2:Sucha game, oo,s necessarilyinessentialby the first remark oi t i 27.5.2. ence is decomposable. H it n ^ 3:For these gamesdecomposability an exceptional is occurrence, with Indeed, ecomposabilityimplies (41:6) some J ^ Q, I] henceK = d / J T* 0, /. So we can choose in J, k in K. Then (41:6)with S = (j), j T = (k) gives (43:11) v((j, A;)) = v((j)) + v((fc)). Now the only equationswhich the values of v(S) must satisfy, are (25:3:a), or of (25:3:b)25.3.1.zero-sumgames are considered) (42:6:a), (if (42:6:bj of 42.3.2. 43:11)neither of these,sinceonly the sets (j), (fc), (j, k] is ( occurin (43:11) thesearenone of the setsoccurring thoseequations in and i.e. or 7 or complementsas n ^ 3. Thus (43:11) extracondition is an which is not fulfilled in general. By the above an indecomposable cannot have n = 2 henceit has game = 1 or n ^ 3. Combining this with (43:E),we obtain the following n result: peculiar

Every elementof the decompositionpartition Tlr is eithera one-element or elseit has n ^ 3 elements. set, Note that the one-element in II arethe one-element sets r splitting sets' i.e.hey correspond those players who are self-contained, to separatee t from the remainder of the game(from the point of view of the strategy oi and coalitions). They arethe \" dummies\" in the senseof 35.2.3. footnote on p.340. Consequently, ur result (43:L)expresses fact :Those this o playen constituentgame* who arenot \"duYnmies,\" aregroupedin indecomposable of n ^ 3 players each. This appears to be a general rincipleof social rganization. o p of 44. Decomposable Games. FurtherExtension the Theory (43:L)
1

a (Decomposable) Gameand Solutions of Its Constituents 44.1. have completed descriptivepart of our study of composi the We tion and decomposition. us now pass to the centralpart of the problem Let The investigation of the solutionsin a decomposable game. for Considera gameT which is decomposable J and / J = K, wit! as the J- and /^-constituents and H. We use strategicequivalence, A to make all three gameszero-sum. at explained the beginningof 42.5.3., Assume that the solutionsfor A as well as thosefor H areknown; doei this then determinethe solution for T? In other words: How do th< solutionsfor a decomposable obtain from thosefor its constituents? game w Now thereexists surmisein this respect hich appears to be the prime a to and we proceed formulate it. facie plausibleone, 1For n = 2 it is otherwise; (j,k) /, (j) and (A;) are complements. 1Such a splitting setis, of course, utomatically minimal.))) a
of

44.1. Solutions

GAMES DECOMPOSABLE

359

44.2.Composition and Decomposition of Imputations and of Setsof Imputations 44.2.1. us use the notations of 41.3.1. as we write v(S) for But Let v A (S), v H (5). Vr(S) this also replaces (41:4), (41:5), by On the other hand, we must distinguish between imputations for 1 this F, A, H. In expressing distinction, it is betterto indicatethe set of to whom an imputation refers,instead of the gamein which they are players w engaged. I.e. e will affix to them the symbols/, /, K ratherthan F, A, H. In this sensewe denotethe imputations for I (i.e. by T)

{i>, and those for J, K (i.e. H) by A,

(44:1)

aj =

*>,

ai,

r},

7/= {0i', ,0*}, (44:2) 7* = {7i\", ' ,7r). (44:3) If three such imputations arelinked by the relationship
(44.4^ (** A)
then we say that

a,.= 7,\"
>

*' =

&'

for for

*' j\"

=r '' ' = 1\",

>

'

>

k '>

I\",
+

a / obtainsby compositionfor /, 7 *, that /, 7 * obtain by decompositionfrom a / (for J, If), and that /, 7 K are the (/-,K-) constituentsof a /. Sincewe are now dealing with zero-sumgames,all these imputations
must
fulfill

>

o the conditions (30:1), (30:2)f

Now 30.1.1.

one verifies

immediatelyfor
Ad

linked by (44:4). :The of (30:1)30.1.1. validity of this for /, y K is clearlyequivalent

K

a/, ft j, y
a /.

to its validity for Ad (30:2) of

For 30.1.1.: \"/?/,

~y K

this states(using (44:4))

*'

(44:5) (44:6)

X t'-r**
i\"

>

y-i\"
amounts to

a, = 0.
i\"

For a / it
(44:7)
1

It is now

convenient

to re-introducethe

'!'
*'

J)

<*'

;_!/,

,- = 0.
of
for the 41.3.1.
players.)))

notations

AND OF 360 COMPOSITION DECOMPOSITION GAMES

Thus its validity for /3 j, y K impliesthe same for a /, while its validity does imply the for a i does not imply the same for /, y K indeed (44:7) and (44:6), ut it fails to imply the validity of either b equivalenceof (44:5)

one.

So we have:
can be composedto an while an imputation a / can bedecomposed two /, y K if and of
Any

A (44: )

two imputations
fulfills

ft

/,

a/,

only

if

it

i ( (44:5), .e.44:6).

themselves: ompositionis always possible,while decompositionis not. C 1 is occurrence. Decomposability again an exceptional of It ought to be noted, finally, that the concept f composition imputao tionshas a simpleintuitive meaning. It corresponds the same operation to two of \" viewing as one\" separate ccurrences, playedthe correspondwhich o
ing

44.2.2. situation This

We call such an

a / decomposable J, K). (for

is similar to that, which prevails for the games

role for games in

ft

41.2.1., 41.2.4. 41.2.3., Decompositionof an

a/

is possibleif and only if the two self-contained of players sets \" J, K aregiven by the setsof imputations a / preciselytheir just dues\" which arezero. This is the meaning of the condition(44:A) (i.e. (44:5), of
(into

/,

K)

(44:6)).
>

44.2.3. Considera setV/ of imputations / and a setW/c of imputations

Let U/ be the set of thoseimputations a / which obtain by composition
> >

K.

of all

j in V/ with all 7 K in Wx. We then say that U/ obtainsby composition from V/, Wxi that V/, W* obtain by decompositionfrom U/ (for JT, K\\ and that V/, W* arethe (J-, K-) constituentsof U/. Clearly the operation of compositioncan always be carried whatout, ever V/t WK whereasa given U/ need not allow decomposition J, K). (for If U/ can be decomposed, call it decomposable we (for J, K). Note that this decomposability f U/ restricts very strongly;it implies, o it (cf. among other things that all elementsa / of U/ must be decomposable the interpretation at the end of 44.2.2.). In order interpret theseconceptsor the setsof imputations U/f V/, to f Wx more thoroughly, it is convenient to restrictourselves to solutions of the gamesr, A, H.
ft

for in

41.3.2.; (41:9), (41:8),

1Thereare great technical games and for imputations.

differences between the concepts f decomposability o h Observe, owever, the analogy between (41:4), and our (44:4), (41:1(V) in (44:5),(44:6),(44:7).)))

41. 4.2.;

(41:5)

etc.,

DECOMPOSABLE GAMES 44.3.Composition and

Decomposition of Solutions.

361

The Main Possibilitiesnd Surmises a 44.3.1. V/f W* be two solutionsfor the gamesA, H respectively. Let Their compositionyields an imputation set U/ which one might expect of to be a solution for the game F. Indeed, is the expression a standard U/ of behavior which can be formulated as follows. We give the verbal formulation in the text under (44:B:a)-(44:B:c), the mathematical equivastating w lents in footnotes, which, as the reader ill verify, add up preciselyto our definition of composition.

The players of J always obtain togetherexactly their 1 a \"justdues\"(zero), nd the sameis true for the playersof K. w (44:B:b) Thereis no connection hatever betweenthe fate of play-

(44:B:a)

(44:B:c)

ersin the set/ and in

The fate of the playersin J is governed by the standard of behavior Vj> 3 the fate of the players in K is governed by the standard of behavior W*. 4

the

2 setK.

from

If the two constituent gamesareimagined to occurabsolutelyseparate each other, then this is the plausibleway of viewing their separate solutionsVy, W/c as one solution U/ of the compositegameF. However,sincea solution is an exactconcept,this assertion needs a w proof. I.e. e must demonstratethis:

(44:C)

If V/, W* aresolutionsof A, H, then their composition is U/ a solution of F.

the way, is another instanceof the characteristic between common senseand mathematical rigour. Although relationship that U/ is a solution whenever V/, W* are) an assertion(in the present case is requiredby common sense,it has no validity within the theory (in this unless based on the definitions of 30.1.1.) proved mathematically. To case this extentit might seemthat rigour is more important than common sense. This, however, is limited by the further considerationthat if the mathematical proof fails to establish the common senseresult, then there is a the strong casefor rejecting theory altogether. Thus the primate of the on e mathematical procedure xtends only to establishchecks the theories in a way which would not be open to common sensealone.
1

44.3.2. by This,

Every element

a / of U/ is decomposable.
is used in
forming U/ and any
>

* Any

/ which

7 *, which is used in

forming U/, give

by composition an element

* The above mentioned

4 The above mentioned

a / of U/. / are preciselythe elements of V/. y K are preciselythe elements of

W/c.)))

AND OF 362 COMPOSITION DECOMPOSITION GAMES It will be seenthat (44:C)is true, although not trivial. t true, i.e.o demand a proof of this: If U/ is a solution of F, then (44:D)

t One might be tempted to expecthat the converse of (44:C) is also

into it can be decomposed solutions V,, W* of A, H. S This is prima facie quite plausible: inceF is the compositionof what arefor all intents and purposestwo entirely separate -games,how couldany solutionof F fail to exhibit tjiis compositestructure? The surprising fact is, however, that (44:D) is not true in general. The reader ight think that this shouldinduceus toabandon or at least m if we tomodify materially our theory (i.e. 30.1.1.) take the above method\" statement seriously. Yet we will show, that the \" common sense ological o basis for (44:D) is quite questionable.Indeed, ur result, contradicting (44:D) will provide a very plausible interpretation which connectsit with well known phenomena in socialorganizations. successfully The 44.3.3. proper understanding of the failure of (44:D) and of the rather detailed conit, validity of the theory which replaces necessitates siderations. Before we enter upon these,it might be useful to make, in anticipation,someindicationsas to how the failure of (44:D) occurs. It is natural, to split (44:D) into two assertions: If U/ is a solution of F, then it is decomposable J, K). (44:D:a) (for (for (44:D:b) If a solution U/ of F is decomposable K), then its constituentsV/, W/c aresolutionsfor A, H. Now it will appear that (44:D:b)true, and (44:D:a)s false. I.e. is i it can happen that a decomposable an game F possesses indecomposable

/,

obtained.

o However,the decomposability f a solution (or of any setof imputations) in 44.3.1. one or more of thesecondiis expressed (44:B:a)-(44:B:c) So by tions must fail for the indecomposable solution referred to above. Now that it will beseen(cf. 46.11.) the conditionwhich is not satisfiedis (44:B:a). This may seemto be very grave, because is (44:B:a)the primary condition in the sensethat when it fails, the conditions (44:B:b), cannot (44:B:c) even be formulated. The concept of decompositionpossesses certain elasticity. This a 42.2.2. 42.5.2., we succeeded ridding and where in appeared_jn 42.2.1., ourselvesof an inconvenient auxiliary conditionconnected with the decomposability of a gameby modifying that concept. It will be seenthat our difficulties will again be met by this procedure so that (44:D) will be a we replaced a correct nd satisfactory theorem.Hence must aim at by so modifying our arrangements, that the condition(44:B can be discarded. :a) in We will succeed doing this, and then it will appear that conditions make r (44:B:b), (44:B:c) no difficulties and that a completeesult can be
1This is similar to

1 solution.

metric solution.

Cf. 37.2.1.)))

the phenomenon

that

a symmetric game may possessn asyma

GAMES DECOMPOSABLE 44.4.Extension

of the

383

constant-sum. These being understood, consider a game F which is decomposable (for J, K) with J-, -K-constituents A, H. The theory of composabilityand decomposabilityof imputations, as 44.2.2. now be repeated insignificant changes. could with given in 44.2.1., are (44:l)-(44:4)may be taken over literally, while (44:5)-(44:7) only modified in their right hand sides. Since(30:2) f 30.1.1.beenreplaced has o o those now by (42:8*) f 42.4.1. formulae (44:5)-(44:7) become:
k'

It 44.4.1.is now time to discardthe normalization which we introduced That (temporarily)in 44.1.: the games under considerationare zero-sum. We return to the standpoint of 42.2.2. to according which the games are

Theory. OutsideSources

(44:5*) (44:6*)
and

(44:7*)
by

= or = i'-i', + /'-i\" v(7) v(J) + v(K).

in or equally (Thelast equation on the right hand side by (42:6:b)42.3.2., in with (41:6) 41.3.2. S = /, T = K.) The situation is exactly as in 44.2.1., it really arisesfrom that oneby the isomorphismof 42.4.2. indeed, are b Thus a i fulfills (44:7*), ut for its decomposability(44:5*), 44:6*) ( needed and (44:7*)does imply the equivalenceof (44:5*)and (44:6*),

i'-ra,- v(J), /'!\"rv(K)

but it fails to imply the validity of either. is again true, only So the criterion of decomposability :A) in 44.2.1. (44 with our (44:5*), 44:6*)n place of its (44:5), 44:6). And the final coni ( ( of clusionof 44.2.2. be repeated: Decomposition an a / (into ft /, y K ) may is possibleif and only if the two self contained setsof players J, K are given by this imputation a / preciselytheir just dues which arenow v(J),))

of Sincewe know that this limitation of the decomposability imputaof difficulties, we have tions the reasonfor (44:B:a) 44.3.1. a source in is i to remove it. This means removal of the conditions(44:5*), 44:6*), .e. ( of the condition(42:8*)n 42.4.1. which they originate. from i

constant-sumgame F

44.4.2. According to the above, we will

with

(i.e. (42:7)of 42.4.1. on 2 In other words

i of alone, (30:1) 30.1.1.) without (42:8*)n 42.4.1.

attempt to work the theory of a o a new concept f imputations,which is basedon

1Insteadof zero,as loc.cit. * We again denote the players by

1,

n.)))

364
An

AND COMPOSITION DECOMPOSITION GAMES OF extended imputation is a system of numbers ai, , an with this
<* ^

property:

(44:8)
We

v((t))

for
n

i = I,

n.

impose no conditionsupon

-. We view theseextended imputa-

a tions, too, s vectors)) be necessaryto reconsider our definitions which are all of imputation i.e.hose of 30.1.1. 44.2.1. t and But, concepts we do this, it is well to interpret this notion of extended before imputations. of i a Theessence this concepts that it represents distributionof certain between the players, without demandingthat they should total amounts up to the constant sum of the gameF. would be extraneous the picture that the players to Suchan arrangement with eachother. However,we have always conceived of areonly dealing proposedto the totality of all players. imputations as a distributivescheme all S 4.5.; 1 (This ideapervades,e.g. of 4.4., it is quite explicitin 4.4.1.) uch from one of the players, but this is immaterial. We a proposalmay come submit varying imputations to can equally imagine, that outside sources the considerationof the players of F. All this harmonizes with our past sources\" anifested thembut m considerations, in all this, those \"outside selvesonly by making suggestions without contributing to,or withdrawing
now in rooted the

It 44.4.3. will

of from, the proceeds the

game.

44.5.The Excess
Now 44.5.1. our\"present conceptof extended imputationsmay be taken \" that which actually to express the outside sources can make suggestions involve contributions or withdrawals, i.e.transfers. For the extended imputation

a =

{i,

,}
e

the amount of this transfer is

n

(44:9)

-i a,

- v(/)

if no transfer takesplace, for a withdrawal. 1 Who tries to form a coalition. Sincewe considerthe entire imputation as his o proposal,this necessitatesur assuming that he is even making propositions to those players, who will not be included in the coalition. Tothesehe may offer their respective and There minima v((t)) (possibly more, cf. 38.3.2. 38.3.3.). may alsobe players in intermediate positions \"between included and excluded\" (cf. the secondalternative in Of 37.1.3.).course,those lessfavored players may make their dissatisfaction effective, o this leadsto the concept f domination,

the and will be called excess a . Thus of e> for a contribution,

(44:10)

e e

<

etc.)))

GAMES DECOMPOSABLE

365

orderto obtain realisticproblems;and we will take due account of this.

The transfers are part of the suggestionsmade from outside, which are o by acceptedr rejected the players,weighedagainst eachother,according 1 to the principlesof domination, etc. In the course of this process, any

It will be necessaryto subjectthis to certain suitable limitations, in It is important to realize how thesetransfers interact with the game.

dissatisfiedset of players may fall back upon the game F, which is the solecriterion of the effectivity of their preferenceof their situation in one 2 (extended)imputation against another. Thus the game, the physical under consideration, determines the background of the social process stability of all details of the organization but the initiative comesthrough the outside suggestions,circumscribedby the limitations of the excess referredto above. c 44.5.2. simplestform that this \"limitation\" of the excessan take, The consistsin prescribingits value e explicitly. In interpreting this prescrips tion, (44:10) houldbe remembered. The situation which exists when e may at first seem paradoxical. w Thisis particularly true when e < 0,i.e. hen a withdrawal from outside is attempted. Why should the players, who could fall back on a game of h a constant sum v(/) accept n inferior total? I.e. ow can a \"standardof order,\"based on such a principlebe stable? Thereis, behavior/'a \"social neverthelessan answer:The game is only worth v(l) if all players form a a coalition, ct in concert. If they are split into hostile groups, then each more pessimistically and such a may have to estimate its chances group division may stabilizetotals that areinferior to v(/).3 w c The alternative e > 0, i.e. hen the outside interferenceonsistsof a free gift, may seemless difficult to accept. But in this casetoo, it will be

1 This a is, of course, narrow and possibly even somewhat arbitrary description of the socialprocess. It should be remembered, however, that we use it only for a definite and solutions. The concluding remarks T limited purpose: o determine stable equilibria, i.e. should of 4.6.3. make this amply clear. 2 We are,of course, lluding to the definitions of effectivity and domination, cf. 4.4.1. a We and the beginning of 4.4.3. given in exactform in 30.1.1. will extend the exact i definitions to our present conceptsn 44.7.1. 3 For a first quantitative orientation, in the heuristic manner: If the players are

, p , then the total of their own valuagrouped into disjunct sets(coalitions) S\\, in tions is v(Si) + + v(SP This is v(J) by (42:6:c) 42.3.2. in v(7) when p 2 by (42:6:b) 42.3.2. Oddly enough, this sum is actually in this model the disagreements between three or more groups are the effective sources

).

of damage. Clearly
n

i.e.
g>

by

i (42:6:c)n 42.3.2.the above

sums v(8i)

+ v(Sp) are all

n,

t-i S
K

V v ((*')). On the other hand, this latter expression is one of them (put p

(t)). So the

when damage is greatest

eachplayer is isolatedfrom
when

all

others.

The whole

phenomenon

therefore, disappears,

game is inessential.

in (Cf. (42:11) 42.5.1.))))

^ t-iv((i))

v(7) ,

i.e. the when

366
gift

AND DECOMPOSITION GAMES OF COMPOSITION

necessaryto study the gamein order to seehow the distribution of this among the playerscan be governed by stable arrangements.It has to that d be expected, the optimisticappraisal of their own chances, erived from the possibilitiesof the various coalitionsin which they might participatewill determine the players in making their claims. The theory must then provide their adjustment to the available total. 44.6.Limitations of the Excess.
The Non-isolatedCharacter of a Gamein the New Setup e These 44.6.1. considerationsindicate that the excess must be neither a too small (when e < 0), nor too large(when e > 0). In the former case situation would arise where each player would prefer to fall back on the 1 i game,even if the worst should happen, i.e.f he has to play it isolated. it In the lattercase will happen that the \"free gift\" is \"toolarge,\" that no player in any imaginedcoalition can makesuchclaimsas to exhaust the available total. Then the very magnitude of the gift will actas a dissolvent of on the existingmechanisms organizations. We will see in 45. that these qualitative considerationsare correct and we will get from rigorousdeductionsthe details of their operationand a e v the precisealue of the excesst which they becomeffective. In 44.6.2. all these considerationsthe game F can no longerbe consince i as sidered an isolated occurrence, the excesss a contribution or a withdrawal by an outside source. This makes it intelligible that this w whole train of ideasshould come up in connectionith the decomposition of the game F. The constituent games A, H are indeed no longer theory 2 with but entirely isolated, coexistent each other. Thus, there is a good reasonto look at A, H in this way whether the compositegame F should o as betreatedin the old manner (i.e. isolated), r in the new one,may be debatable.We shall see,however, that this ambiguity for F does not influence the result essentially,whereasthe broaderattitude concerning, A and H proves to be absolutelynecessary(cf.46.8.3. also 46.10.). When a game F is consideredin the above sense,as a non-isolated with occurrence, contributionsor withdrawals by an outside source,one T be tempted to do this: reatthis outside sourcealso as a player, might him togetherwith the other players into a largergame F'. The including rules of F' (which includesF) must then be devisedin such a manner as to provide a mechanismfor the desiredtransfers. We shall be able to meet this demand with the help of our final results, but the problem has some t intricacieshat arebetterconsideredonly at that stage.

i.e.

n
1

This happens
We will

when the

total proposed v(/) + e is < this

in this (42:11)42.5.1.)

sion

41.2.4.))) cf. concerned; 41.2.3.,

'

is equal to v(7)

This in spite of the

seein 45.1. that

ny (by

is preciselythe criterion for e being \"too small/' absenceof \"interactions,\" as far as the rules of the game are

-iv((i)). means e <

ny.

As the

last expres-

DECOMPOSABLE GAMES 44.7.Discussion the of

New

367))

Setup

44.7.1. reconsiderationf our old definitions mentioned at the The o is beginningof 44.4.3.a very simplematter.
For the
extended imputations we have the new definitions of

(i ),

The definitions of effectivity and domination we takeover unchangedfrom 1 1. 30.1. the supportingarguments brought forward in the discussionwhich led up to thosedefinitions appear to lose strengthby our present generalno 2 izations. The sameapplies to our definition of solutions eod.with one caution: he definition of a solution referred to makes the conceptof a T solution dependent upon the set of all imputations in which it is formed. Now in our presentsetup of extended imputationswe shall have to consider them notably concerning limitations concerning as their excesses indiTheserestrictions ill determine set of all extended w catedin 44.5.1. the a o imputations to be considerednd thereby the concept f a solution.

44.4.2.

is a given e . The meaning of this restriction that the transfer from in of of outsideis prescribed, the sense the discussion 44.5.2. w where only an upper limit of the excess Second, e shallconsiderthe case T is prescribed. henwe have an inequality
with

First,we shall considerthe casewhere the scribed. Then we have an equation e = eo (44:11)

44.7.2. we Specifically shall considertwo types of limitations.

value of the

excess preis

(44:12)
with

outside is assigneda

a given e

ge
from

is The meaning of this restriction that the transfer

maximum

is that one which we arereally interested the first one, will case prove technicallyuseful for the clarification of 44.5.2. he second T of the first one although its introduction may at first seemartificial. further alternatives because e will be able to w We refrain from considering alone. the indicateddiscussionwith thesetwo cases complete the Denote set of all extended (first imputations fulfilling (44:11) case) as we in ( by E(eo). Considering44:9) 44.5.1.,can write (44:11)
n

receive it). The casein

(from the point of view of the

players who

i.e.

(44:11*)
Q

-i
n

= v(7) + e .

case) the Denote set of all extended (second imputations fulfilling (44:12) a we can write (44:12) s in by F(e). Considering(44:9) 44.5.1.,

(44:12*) i-l g v(7) + e,. loc.cit., I.e.30:3); (30:4:a)-(30:4:c) respectively. ( I.e.30:5:a), or (30:5:c) (30:5:b) (
1

eod.)))

AND OF 368 COMPOSITION DECOMPOSITION GAMES we t of Forthe sakeof completeness, repeathe characterization an extended as well as to (44:12*): imputation which must be added to (44:11*), - , n. = a, v((0), for (44:13)

i 1,

of areinvariant under the isomorphism 42.4.2.

a and Note that the definitionsof (44:9) s well as (44:11*), (44:12*) (44:13)

definition, adjusted to the present conditions. Throughout the definition which follows, E(e^) can be replacedy F(e<>), as indicatedby [ b if A setV E(eo) the [F(ed)]is a solution for E(e ) [F(eo)] it possesses follow-

of that of w Because the central role this concept e restate

Now 44.7.3. the definition of a solution can be taken over from 30.1.1.

].

ing

properties:

(44:E:a)

No

(44:E:b)

inV.

is dominatedby an a in V. 9 )] Every /} of E(e*)[F(e not in V is dominatedby some a

in V

can and (44:E:a) (44:E:b) bestated as a singlecondition: o o (44:E:c) Theelements f V arethoseelements f E(e<>)[F(eo)]which

areundominatedby

o any element f V.

It will be noted that E(0) takes us back to the original 30.1.1. (zero-sum (constant-sumgame). game)and 42.4.1. The 44.7.4. conceptsof composition, decomposition and constituents of extendedimputations can again be defined by (44:l)-(44:4) f 44.2.1. o the As pointedout in 44.4.2. technical purposeof our extending concept the
as of imputation is now fulfilled. Decomposition well as compositioncan
now

we simple; will deal with it as the necessityarises. For the composition,decompositionand constituents of setsof extended can imputationsthe definitions of 44.2.3. now be repeated literally. 45. Limitations of the Excess. Structureof the Extended Theory

out. always becarried o Theconnectionf theseconcepts ith the setsE(e ) and F(eo)is not so w

Excess In 45.1. the setups of 30.1.1. of 42.4.1. and imputations always existed. It is now different: Either set E(e), F(e) may be empty for certaine . or of conflict with Obviously this happens when (44:11*)(44:12*)44.7.2. eodem and this is clearlythe casefor (44:13)
of the
Q

45.1. LowerLimit The

v(7)
in

+ eo <

-iv((i))
ny

both alternatives. As the right hand side is equal to v(7) in 42.5.1., means this (42:11)
e

by

(45:1)

< -ny)))

STRUCTURE THE EXTENDEDTHEORY OF

369

in

If J(e) [F(e<>)]is empty, then the empty setis clearly a solution for it 1 and sinceit is its only subset, it is alsoits only solution. If, on the other hand, E(e ) [F(e )] is not empty, then none of its solutionscan be empty. This follows by literal repetitionof the proof of (31J)in 31.2.1. : The right hand side of the inequality (45:1) determinedby the game is T; we introducethis notation for it (with the oppositesign,and using(42:11)

42.5.1.):

(45:2)

|rK = ny = v(/)

- -iv((t)).

Now we can sum up our observationsas follows: If (45: A)

o

<- |r|,,

then E(e ),
Q

solution. Otherwiseneither eithercan be empty.

tative

F(e) are empty and the empty set is their only E(e<j) nor F(e) nor any solution of

This result gives the first indication, that \"too small\" values of e the exist. (i.e. in the senseof 44.6.1. Actually, it corroborates quantie)

estimatef footnote o

1 on

p.366.

45.2.The Upper Limit of the Excess. Detachedand Fully DetachedImputations 45.2.1. us now turn to those values of e (i.e.), which are \"too Let e When does the disorganizinginfluence of large\" in the senseof 44.6.1. the magnitude of e, which we thereforesaw, manifest itself? As indicated in 44.6.1., critical the T phenomenonis this: he excess may be too largeto be exhausted by the claims which any player in any to imaginedcoalition can possiblymake. We proceed formulate this idea in a quantitative way. It is best to considerhe extendedimputations a themselves,instead t of their excesses Suchan a is past any claimswhich may be made in e. any coalition,if it assigns to the players of each (non-empty)set S I if more than thoseplayers could get by forming a coalition in F, i.e.

(45:3)

2v
in

cti

> v(S)

for every non-emptyset S &

I.

shows Comparingthis with (30:3)in 30.1.1. that our criterion amounts to demandingthat every non-emptyset S be ineffective for a . In our actual deductions it will prove advantageousto widen (45:3) somewhatby includingthe limiting caseof equality. The condition then

becomes
actually

1In spite of

repeatsfootnote 2 on p.

its triviality,

this circumstance should not

278.)))

be overlooked.The text

370 (45:4)

OF AND COMPOSITION DECOMPOSITION GAMES iinS

<xi

v(S)
*

for every

SsL
1

We call the a of (45:3)ully f and those of (45:4)detached. As indicated, the latter concept detached, that will be really neededin our proofs both termini aremeant to express t is detached from the game,i.e.hat it cannot be the extended imputation effectively supported within the game by any coalition. 45.2.2. more remarkis useful: One o The only restriction imposedupon extended imputations is (44:13) f

It is convenient to give these a a name.

44.7.2.:
(45:5)

at ^ v((i))

for

i = 1,

n.

i and hence Now if the requirement (45:4)of detachednesss fulfilled is o a fortiori if the requirement45:3) f full detachedness fulfilled then it is ( unnecessary to postulate the condition (45:5)as well. Indeed,(45:5) for of is the specialcase (45:4) S = (i). This remark will be made use of implicitly in the proofs which follow. those 45.2.3. we can revert to the excesses, characterize which Now i.e. to detached(or fully detached)imputations. This is the formal belong characterization

(45:B)

The game

determinesa number

(45:B:a) (45:B:b)

properties:
A
fully

|r|

2 with with

the following the e excess

if exists and only if

A

detachedextendedimputation
e

> |r|. 2
with

if

and only if

detachedextendedimputation
e^

the

e excess exists

|r|,.
inS

E of Proof: xistence a detacheda 3: Let a be the maximum of all v(S), S Si (so a ^ v(0) = 0). Put ^ = {aj, , aj} = (a, , a }. ;> v(5). This is Then for every non-empty S S I we have a? ^ a
s (45:4), o a is detached.
to unlike (45:3) true when is necessary exclude S , since(45:4) Indeed,then both sidesvanish. J The intuitive meaning of these statements is quite simple:It is plausible that in or a order to produce detached a fully detached imputation, a certain (positive) minimum excessis required. t is this minimum, or rather lower limit. Sincethe notions \"detached\" and \"fully detached\" differ only in a limiting case (the sign in (45:4)), it stands to reasonthat their lower limits be the same. These things find an exact :B). expressionin (45 8 Note that it is to necessary prove this! The evaluation which we give hereis crude, for more precise onescf. (45:F)below.)))

It is no longer

|r|

STRUCTURE THE EXTENDEDTHEORY OF

of Properties the detached a :According to the above, detached a = , an ]

371

{!,

exist,and
a minimum

with

them their

e excesses=

it 8 = 1)all thesee are 2> 0. Hence followsby continuity, that theseehave

t-i

a<

v(7). By (45:4)(with
,

e*.

1 We now put

e*. Choosea detached a * = {af, = |r|, e*.

n

a*}with

this

excess

(45:6)
Proof ot definition e

If V (45:B:a), (45:B:b): = {ai,

detached, {i, , ) i-1 v(7) ^ then (45:3)remains true we subtract a sufficiently small 6 > from each So a' = {on 5, by 8} is detached.Hence defini, a

on

e*.

If

a=

t an is detached,hen by
}

an is fully

if

oti.

tion e

nb

i-i(a<

- - v(7) ^ e*,e > e*.

) ,

Considernow the detached a * = {af,

a*) with

-i
Then (45:4)holds for
$

a*

- v(7) = e*.

a*;hence(45:3)holds if we increaseeacha* by a

is > 0. So a\" = (a* + 5, , a* + 6j is fully detached.Itsexcess n = (a* + 5) v(7) = e* + n5. So every e = e* + n, 6 > 0, i.e. i-l o imputation hencea fortiori every e > e*,is the excessf a fully detached

of a detached t one;and e* is, of course,he

o excessf a detached imputation

a*.

also closely connected with the conceptof domination. The properties involved aregiven in (45:C)and (45:D)below. They form a peculiarantithesisto eachother. Thisis remarkable,since two concepts restrongly our a one f analogousto eachother indeed,the second arisesrom the first oneby the inclusion of its limiting cases.
1This continuity
argument the is valid because = sign

hold Thus all parts of (45:B:a), (45:B:b) for (45:6). 46.2.4. fully detachedand the detachedextendedimputations are The

is included

in

(45:4).)))

372
(45:C)

AND OF COMPOSITION DECOMPOSITION GAMES

extended detached imputation a dominatesno other extended imputation ft .

A fully

I Proof:f a

0,then a must

possessa non-emptyeffective set.

if

(45:D)

extendedimputation a is detachedif and only dominatedby no otherextended imputation ft .

An
}

it

is

Let Proof: Sufficiencyof beingdetached: a = {ai, ,an be detached. Assume a contrario ft H a , with the effective setS. ThenSis not empty; < fti for i in S. So 2} < 2} ft ^ v(5) contradicting45:4). ( t in 8 f in 8 Assumethat a = Necessity of being detached: detached.Let S be a (necessarilynon-empty) set for

i.e.

{i, , a which

n}

is not

oti

< v(S). Thenfor a sufficiently small 6 > 0,even

(<*
tin

(45:4)fails,

+ *) v(S).
,

Put

and 5is effective for

= {0i,

ft n \\ ft

= {ai+ 6,
ft

an + $},hen t g v(5). Thus H a .

always a<

< ft

45.3.Discussion the Two Limits |r|i,rt Their Ratio of | 45.3.1. two numbers and |F|, as defined in (45:2)of 45.1. The 2 and in (45:B) of 45.2.3. both in a way quantitative measures the are of of F. More precisely: essentiality

|r|i

|.

follows:
(45:F)))

|F|i T of Proof: hestatements concerningF|i, hich is = ny by (45:2) 45.1., | w coincide with the definitions of inessentiality and essentiality of 27.3., as reasserted 42.5.1. in The statementsconcerning 2 follow from those concerning by |F| |F|i, means of the inequalitiesof (45:F), hich we can use here. w 45.3.2. quantitative relationshipof |F|i nd |F| is charactenzed The as a 2
Always))

(45:E)

If F is inessential, then JF^ = 0,|F| = 0. 2 If F is essential,hen t > 0,|F| > 0. 2

-^rirai)))

STRUCTURE THE EXTENDEDTHEORY OF

373

As a Proof: we know, |F|ind 8 areinvariant understrategic equivalence, hencewe may assumethe game T to be zero-sum,and even reducedin the senseof 27.1.4. e can now use the notationsand relationsof 27.2. W = Since|T|i ny, we want to prove that

|r|

(45:7)

* jJLj y g |r|, Sfef^ y.

L Proof of the first inequality of (45:7): et a = {ai, , a ) be detached.Then (45:4)gives for the (n l)-element 8 = I (fc), set n 5) ai ctk = 5) on ^ v(S) = 7, i.e.

-1))

(45:8)
Summing

,
(45:8) over k 1)

r
n
r))

=
n

1, :

.
-

n,

gives

] i-l

r.

*-l

a*

i.e.(n

a ^

717,

^a ^

y 7.

Now v(7)

= 0, so e =
r

^ a*.

Thus e ^ n

r 7 for all 1

detachedimputations;hence|F|t ; w

17.

inequality of (45:7): Proof of the second - , = Put a00 = ^-y^7, and

This a

o the number of elements f

i i is detached, .e.t

is p = Q:5 = , (45:4) trivial. = 1: = (t), (45:4)becomesa00 ^ v((i)), S p n i - i i i.e. s 2 y ^ ~~~7 which is obvious.

S. Now we have:

fulfills

00 , a00}. aj = {a , Let (45:4)for all Sfi I. Indeed: p be

{a?,

p^

2:(45:4)becomes 00 ^ pa

v(S), but by (27:7)in 27.2. p)7, vGS) ^ (n

so it suffices to prove pa00 2> (n

p)7 i.e. p

7^

(w

p)7. This

amounts to p 7 ^ ^7> which follows from p ^

2.

Thus a

is is indeeddetached.As v(/) = 0,the excess

eoo na oo = n(n
^\"

2) 7.

Hence

|r|i^

7-)))

374
n

AND DECOMPOSITION GAMES OF COMPOSITION

worth while to

= 1,2,3,4, successively: n = 1,2: thesecases coefficient In the

46.3.3. is It

consider the inequalities of (45:F) for

= 3:In this casethe two coefficients 1 T and ^ s coincide: 2 n I Both areequal to to So the inequalitiesmerge an equation: = ilrK. (45:9)
n

so the contradictionsdisappear.

1 t inequality is greaterhan the coefficient z of the upper bound., This 5 absurd. But sinceF is necessarilyinessentialfor n = 1,2 (cf. may seem = 0,and the first remark in 27.5.2.), have in thesecases = 0, we

j of the lower bound of the

|r|i |r|i
o

i.

|r|,

4:In thesecaseshe coefficient t

^ of the lower boundis definitely

^ o = smallerthan the coefficient JL of the upper bound.2 Sonow the inequalities leave a non-vanishing interval open for |F2
n n

The lower bound |F| = ft i.e. 2 |F|iis precise, there existsfor each f ^ 4 an essential ame for which it is assumed. Therealso exist or each g b r |F|i, ut it is probably not pos^ 4 essential ames with |F| > 2 g

^r
L.

|.

sible to reach the upper bound of our inequality, |F| = n ~ 2 r li- The 2 i v precisealue of the upper boundhas not yet beendetermined. We do not needto discussthesethings hereany further. 8 In 45.3.4. a more qualitative way, we may therefore say that |F|i,|F|j are both quantitative measuresof the essentiality of the game F. They measureit in two different, and to a certainextent, independent ways.
l

eachn ^
We

w 2 Indeed,the ratio |F|/|F|i, hich never occursfor n = 1,2 essential (no and games!), is a constant for n = 3 (its value is ^), is variable with F for

4.

thesetwo quantities actually measure the \" dictated excess ill not \" disorganize the players, w in the senseof 44.6.1. e i Judgingfrom our results, an excess < |F|is \"too small\" antl an excess > |F| is \"toogreat\"in that sense. This e 2
saw in limits, within
view will
which a

45.1., that 45.2.,

-i! __are .--2} 1They ,

for n

in becorroborated a much 1;

s more preciseensein

46.8.

1, for n

2. Note alsothe paradoxicalvalues

<*

and

s 1 < ti means 2 < (n l)(n 2) which is clearly the casefor all n 4. 1For n 4 our inequality is J As mentioned above, e know an w ^ itial game with and alsoone with 8 J |r|i.)))

|r| - |r|i|r|i |r|i |r|i. - i |r|

STRUCTURE THE EXTENDEDTHEORY OF

46.4.Detached Imputations
and Various Solutions.

375

The Theorem Connecting (e ), F(e ) in 45.4.1. and (44:E:c) the definition of a solution in 44.7.3. our result (45:D) in 45.2.4. immediately: give A solution V for E(e^ [F(e)] must contain every detached (45:G) extended imputation of E(e<>)[F(e)].
After what was said at the beginning of 44.7.2. about the roles E(e^) of and F(e), the importance of establishingthe complete inter-relationship between these two caseswill be obvious. I.e.we must determine the b connectionetweenthe solutionsfor E(e ) and F(e). Now the whole difference betweenE(e ) and F(e) and their solutions is not easy to appraise in an intuitive way. It is difficult to seea priori I the \"gift,\" made why thereshouldbe any differenceat all:n the first case to the players from the outside, has the prescribedvalue e , in the second caseit has the prescribedmaximum value e . It is difficult to seehow the w \"outsideource/' hich is willing to contributeup to e can ever be allowed s to contributeless than e in a \" stable\" standard of behavior (i.e. solution). i will However,our past experience caution us againstrash conclusionsn this Thus we saw in 33.1. 38.3. already threeand four-person and that respect. games possesssolutions in which an isolated and defeated player is not \" \" exploited up to the limit of the physicalpossibilities and the present casebears some analogy to that. 45.4.2. :G)permits us to make a more specificstatement: (45 A detached extended imputation a belongsby (45:G)to every solutionfor
Q Q

Theimportanceof this resultis due to its role in the following consideration.

F(e), if it belongsto F(e).

any solutionfor E(e )
Q

if

it

On the otherhand, a clearly cannot belongto does not belongto E(e ). We now define:
Q

(45:10)

D*(e<>) is

the

extended setof all detached imputations a E(e).

in

F(e), but not in

of So we see:ny solution of F(e) contains all elements D*(eo);ny solua A o tion of E(e<>) containsno element f D*(e). ConsequentlyF(e) and E(e ) have certainly no solution in common if D*(e) is not empty. Now the detached a of D*(e) arecharacterized having an excess by
Q

6,

but not e = e

i.e. y b

Fromthis we conclude:
(45:H)
D*(e<>) is empty
if

(45:11)

<e .

and only if

e,

|r|
t
.)))

376
Z>*(e )

AND OF COMPOSITION DECOMPOSITION GAMES

This is further evidencethat

above, the non-emptinessof Proof:Owing to (45:B) and to (45:11) o to is equivalent to the existencef an e with |r|jg e < e i.e. the emptinessof D*(e) amounts to e ^ |r|a. o > |r|j. Hence Thus the solutionsfor F(e) and for E(e$)aresure to differ, when eo>|F|i.
Q Q

>

onebetweenthe solutionsfor E(e*)and for F(e). More precisely: (45:1) The relationship

|r|. 45.4.3.

e<>

\" is \"too large for normal behavior when it is

Now we can prove that the difference indicatedabove is the only

(45:12)
Q

V^W = VuD*(e)

This will

establishes a one-to-one relationship between all solutions V for E(e ) and all solutionsW for F(e). be demonstratedin the next section. 46.5.Proof of the Theorem

45.5.1. beginby proving someauxiliary lemmas. We The first one consists of a perfectly obvious observation,but : applicability
(45:J) Let the two extended imputations y = {71, 5 == {81, , dn ] bear the relationship

of wide

, yn

\\

and

(45:13)

7, ^

5,

for all

i = 1,

n\\

then for every

a, a
>

H y

implies a H 5 .

The meaning of

this

result is, of course,that

some (45:13) expresses

kind of inferiority of 6 to y in spite of the intransitivity of domination. This inferiority is, however, not as completes one might expect. Thus a onecannot makethe plausibleinferenceof y ** ft from 5 H ft , because the

effectivity of a

set S for & may not imply the same for y . (The reader t should recallhe basic definitions of 30.1.1.) It should also be observed,that (45:J)emergesnly because e have w o the o extended concept f imputations. Forour older efinitions (cf. 42.4.1.) d
n

we would have had 2) 7

t-i
>,

=
==

^ i-ii<;
1,
y

hencey< ^
, n,

fl

for all
&

i = 1,

, n

necessitates = 5, for all i %

(45:K)
If
y

45.5.2. four lemmasleadingdirectly to the desiredproof of (45:1). Now

a H ft with a detached in F(e) and ft in/?(e), then and thereexists n a H ft with a ' detached in E(e a and
^

i.e.y = .

'

Q ).)))

STRUCTURE THE EXTENDEDTHEORY OF 377 L in 30.1.1. Proof: et S bethe setof (30:4:a)-(30:4:c) for the domination

ft

. S = / would imply

a{ > ft

for all
ft

, - v(J) >
But as

i = 1,
v(/).))

,n

so

o is in F(e) and ft

in

U(e ), so % o - v(7) g e =
n

n))

ft

v(7),

contradictingthe above.

a' = {a'j,

So S T* 7. Choose,herefore, an t
,

t'

a'} with
n

= 1,

, n, not in

S. Define

e choosing ^

= go. Thus all o{^ ; hence a is detachedand it is clearly in (e ). Again, as a( = on for i 5^ t'o hence for all i in S,so our a H /3 implies a H /3 . of (45:L) Every solution W for F(e) has the form (45:12) (45:1)
~~ <

A so that 2)

~*

V CO

'

'

1 for a unique V E(eo). O Proof: bviously the V in question if it existsat all is the intersection should hold for)) W n E(ev), so it is unique. In order that (45:12)

we needonly that the remainderof W

(45:14)

be equal to Z)*(eo),i.e. E(e<>) = D*(e).

Let us therefore prove (45:14). and o Every element f D*(e<>)is detached in F(e) so it is in W by (45:G). E(e<>). Thus)) Again, it is not in E(e*\\ so it is in W

(45:15)
If also

- E(e)

W-(eo)D*(eo), (45:16) as then (45:15), (45:16) togethergive (45:14),desired.Assume therefore, that (45:16)not true. is
an Accordingly,consider a =
in

{i,

an

in W
n

D*(e). Then a is in F(e), but not in E(e<>),so

1 We do not yet assertthat
this V

-i

a<

- v(J) <

E(eJ)and not
e

As

is a solution

for

f(

t)

that will

comein (45:M).)))

378

OF AND COMPOSITION DECOMPOSITION GAMES

in

a is not

non-empty

set8 with a =

>*(e ),

i this excludests being detached.Hencethereexists a

tin

Now form

'

a < v(S).
,

{&(, a'i

a'

n } with

= cti + c a< = a<

n
<*(
\\

for in S, for not in

i i

S,
in

choosing
a

is in F(e). If it is not
ft

>>>
in W
with
ft

>

so that still
I
in W,

- v(7) g e and
^
, this

then (as W is a solution for F(e)) thereexists

Q

; ^ v(S).
> >

So a '

a'.

As all

This is impossible, inceboth ft s must be in W. Now ; > at for all

implies ft H a by (45:J). a , a belongto (thesolution)W. Hence

'

'

i in S, and

<*<

v (5)-

So

tin

S this

diction. (45:M)

a . But as both a ', a belongto (the solution)W,

is a contra-

of (44:E:b)44.7.3.

The V of (45:L)is a solution for E(e). o a V along Proof: c E(e ) is clear, nd V fulfills (44:E:a)f 44.7.3. with W is a solution for F(e)), since V W. So we need only verify (which
Q

Considera
W|

in

#(<)),but not in V. Then

in W
with

/3

is alsoin F(e) but not in

Q

hencethereexists a an
>
Q

a. H

ft

(W

is a solution forF(e)!).
Q

a belongsto E(e), then it belongsto W n E(e ) = V, i.e. e have w an a in E(e ) with a H ft . If a doesnot belong to E(e$),hen it belongsto W t #(e ) = D*(e), and so it is detached.Thus a H ft , a detached in f(eo). Hencehere and t ' detached in E(e<>). By (45:G)this a exists y (45:K)an a H , a and b
If this

Thus (44:E:b)44.7.3. at any rate. of holds in If V is a solution for JZ(<* ), then the W of (45:12) (45:1) (45:N) is a solutionfor F(e). s of Proof: sF^o) is clear, o we must prove (44:E:a), W (44:E:b)44.7.3.
Ad

belongsto W, (E(eo) F(e), W is a solution for F(e )\\); henceit belongs s = V. So we have an a ' in E(e ) with a *- .] to W n E(e )
Q Q Q

'

'

'

(44

Assume a :E:a):
ft

ft

for two

a,

ft

in W.

8-

^3* and

(45:D)

exclude that

be detached.So
W

is not in D*(e), henceit is in D*(6) = V.)))

STRUCTURE THE EXTENDEDTHEORY OF

379

Hencea

ft

excludeshat a too be in (the solution)V. So a is in t

W

= D*(o).
which

Consequentlya is detached. Now (45:K)produces a ' H an

>

ft

detached,a' belongsby (45:G) to (the solution for E(e )) V. As a ', ft both belongto (the solution) V and a ' H ft , this is a contradiction. Ad (44:E:b): Considera T = {0i, , ftn] in F(c), but not in W. Now form ft (c) = {^(e), , n (e)}= {fr + t, , n + } for every Let increase from until one of thesetwo things occurs the for
^
Q

and is detached in E(e ). Being Q

6^0.

first time:))

(45:17)

(0is in
2 \"/?()is detached.

(45:18)
We

detached. = 0, then ft = ft (0) is in E(e ). As ft is not in V W, there If an exists a H in (the solution for E(e<>)) V. A fortiori a in W. there Assume next *i > 0, and (ci) in V. As ft (ei) is not detached, c / with ^ ft(ei) < v(S). Besides, lways exists a (non-empty) S a
1
Q

distinguishthesetwo possibilities: (45:17)appens first, say for c = ei ^ 0:ft (ci)is in E(e ), but it is not h
Q

>

fti(ci)

> fti. So

ft

(ci) H
*i

Assume,finally,

>

S . And ft (1)is in V, hencea fortiori in W. and ft (*0 not in V. As ft (i) is in E(e<>) there t
tin

existsn a ^ ( l ) in (thesolution for E(e*)) . Since a V always ft i(i)> ft, a a H (ei) impliesa ^ ft by (45:J). And a is in V, hence fortiori in W. or simultaneously with (45:17), for = ej ^ 0: say (45:18) happensfirst, ft (j) is still in F(e), and it is detached. If 7(s) is in J(e), then it is by (45:G)in (the solution for ()) V. If 6 (e2) is not in #(e ), then it is in D*(e). So ft (*2) is at any ratein W.))
1

I.e.he t
%

and the

I.e.5) 0,(e) vOS)forallS7.Each

in

o excessf ft () is c . For * w excessf (e)increases ith , o S

(0)

S)))

is in F(c ),

its i.e. excessb

c.))

e,

w ft (t) increases ith in

AND OF 380 COMPOSITION DECOMPOSITION GAMES c This excludess

==

0,since ft = ft (0) is not in W. So 61 > 0.

s a (e) is not detached, o thereexists non-empty8 / continuity a non-empty))

For < <

with

2 0*W
/
*

< V OS)- Hencethere existsby

i,

ft

even

with

&()

a v(S). Besides, lways ftfo)

> ft,

hence))

(i)H
*

- And

() belongsto W.
promisedproof:

a up:In every casethereexistsn a H ft in W. (This a was is a , ft (ci), a , (ej)above,respectively.) So (44:E:b)fulfilled.

Summing We can now give the

Proof of

Immediate, combining (45:L),(45:M),(45:N). by (45:1):

45.6.Summary and Conclusions 46.6.1. main results,obtainedso far, can be summarizedas follows: Our

(45:0) (45:O:a)

If e
Q

< -|r|i,
the empty

then U(e ),

solution.
If

F(e) are empty and

set is their

only

(45:0:b)

-|r|i e ^ |r|,
f))

then J5(e ), ^(^o) are not empty, both have the same solutions, which areall not empty. If e> (45:0:c)
Q

|r|,,

thenE(eo), (e) arenot empty, they have no solutionin common, F all their solutionsarenot empty.

and 45.1. following (45:H) in 45.4.2. these concerning points:That it is here

I and by A), Proof: mmediate combining (45: (45:1) (45:H). This result makes the critical characterf the points = |F|i,r|* o | at quite clearand it further strengthens the views expressed the end of
e<>

later(in 46.5.). Let W be a (45:P)

eo

\" \"too where 60 becomes small\"or \"toolarge in the senseof 44.6.1. 46.6.2. arealsoable now to prove some We relationswhich will beuseful

-|r|i.

non-empty solution for Then

a F(e), i.e. ssume that

(45:P:a)

Max-. e(O = e iii

*\" iff)))

DETERMINATION OF ALL SOLUTIONS

381

(45:P:b)
Also
**

Min-^ w e(7)

- (.,|r|,).
Min
iff

M . (45:P:c) ax-*in TTe(~a) Min-in e(a)= Max (0, e . f since Proof: 45:Prc)ollows from (45:P:a), ( (45:P:b)
ot

- |r|,).

Min (e ,

2 |T|) = Max (e

We now prove Write W =

e ^
Q

|r|i,

As VuD*(e), V a solution for J0(e ), following (45:1). so Vis not empty (by (45: or (45:0)). we know e( a) = e As A)

(45:P:a), (45:P:b).

- - |r| =
e,e
2)

Max (0, e

- |r|).
2

throughout V and e( a) < e throughout D*(e). Now for e ^ |T|, D*(6) is empty (by (45:H)),so 2

(45:19)
(45:20)
And

Max-*

. e(a)= Max-*m w e(a)= , a. V Min- w e(T) = Min- w e(\"^)= . . . a mV

in W in
>

> |T|, D*(e) is not empty (again by (45:H)),it is the setof all 2 detacheda with e( a ) < e . Hence (45:B:b) in 45.2.3. e( a ) have these by
for e
Q

a minimum,

2 |T|. So we have in

this

case:
V))

(45:19*) Max-. w/ e(2 = Max-.w e(2)= ) in W in

(45:20*)
Min

7^w

e()= Min-. ^^ e()= |r|,.

fa

and (45:19), (45:19*) togethergive our (45:P:a), (45:20), (45:20*) ive g our (45:P:b). together 46.Determinationof All Solutionsin a Decomposable Game of Decompositions 46.1. Elementary Properties

stituents.

of 46.1.1. us now return to the decomposition a game T. Let for Let F be decomposable J, K( I J) with A, H as its J-, IC-conGiven any extended imputation

a = (ai,
i in

, <x} for 7, we form

its

J-, ^-constituents
excesses

ft

(ft

= a, for

J, 7^ = a for i in K), and their

%w

1Our assertion includes the claim that these Max and Min exist. in W in W 1Verbally: The maximum excess the solution W is the maximum excessllowed in a in FMi o. The minimum excessin the solution W is again c , unless e > in which caseit is only the minimum is as nearly e as possible, considering that it must

|r|t. never exceed t . |r| I.e.

The \"width\" of the

|r|t,
if

interval

in of excesses W ia

the

of excess e over IrU,

any.)))

OF 382 COMPOSITION DECOMPOSITION GAMES)) AND

o Excessf (46:1))) a. in
ft

/:e = e(a ) = ]
JRC:

a
~~ v (^)>

t-i))

Excessf o

in

J:/ = /( a ) = J/ a tin/
g

o Excessf 7 in

= g( a) = ]? a
tin X

v(X).1

Since (46:2)

+ v(K) = v(7) with in or equallyby (41:6) 41.3.2. S = J, T = K) (by (42:6 in 42.3.2., :b) therefore e=f + g (46:3)
v(J)
(46:A)
We have

(46:A:a) (46:A:b) (46:A :c) turn. (46:4)

A Proof: d

+ |r|i= |A|i + |H|i, = .

H areboth inessential. to (46:A:a): the definition (45:2)in 45.1. T, A, H ir Apply

F i is inessentialf and only if
A,

|r|
2

|A|,

JHJ 2

|r|i= v(7) |A|i = v(J)

(46:5) (46:6)

tin

v((t)), v((i)),

|H|,= v(K)
y

t in

% v((i)).

and (46:6) with the sum of (46:5) Comparing(46:4) gives (46:A:a), owing t<

detached(in /)

(46:2). Ad (46:A:b): ~2, ~P , Let

if

Then a be as above (before (46:1)).

for all
R

i:

tin R

2) a, S v(R)
\"^

S 7.
C T o /,
r~

i Recalling (4t:6)n

we 41.3.2.may write for this

(4o:7; 2/ a t in S
f7\\

/ A G.

\\^

\\

t in

2/ a
\\^

/C\\ v w)

+ v \\^ )
I

,~/fTi\\

**.-* *or

\" a 11

H^

V 1 r IK.

Again

(46:8) (46:9)
1Up to of a

7 aredetached(in J, X) if for all % a, ^ v(S)

iin

SsJ,
TsX.
g.)))

'sexcessupon . e

this point

tinT it was not necessary give explicit expressionto to a We do this now.for eas well as for/,

J) a, ^ v(T)

for all

the

dependenc

DETERMINATION OF ALL SOLUTIONS

383

Now (46:7)is equivalent to (46:8) (46:9) Indeed (46:7)obtainsby adding , and f a (46:8)and (46:9); (46:7) pecializesor T = to (46:8) nd for S = s to (46:9).

i Thus a is detached,f and only if its (/-, constituentsft , 7 areboth K-) e detached.As their excessesand /, g arecorrelatedy (46:3),this gives b

for their minima (cf. (45:B:b))

i.e.ur formula (46:A:b). o

= |r|,

|A| 2

2 + |H|,

Ad (46:A:c): o w Immediate combining (46:A:a) r (46:A:b) ith (45:E) by as appliedto F, A, H. Thequantities 2 |F| areboth quantitative measuresof the essentiality of the game F, in the sense 45.3.1. above result states of that both are Our of additive for the composition games. 46.1.2. Another lemma which will be useful in our further discussions:

|r|i,
>

(46:B)

If a H ft (for F),then the setS of with S J or S K without can be chosen

>

for 30.1.1.this domination

any

l lossof generality.

for Proof:Considerthe set S of 30.1.1. the domination a H ft. If accidentally S Si J or S K, then there is nothing to prove, so we may assume that neither S J nor S K. ConsequentlyS = Siu Ti, where Si /, Ti S K, and neither Sinor T\\ is empty. for We have a > ft for all i in S,i.e. all i in Si,as well as for all i in T\\. Finally

28 at ^ v(S). The left hand sideis clearlyequal to ^ a + ^ a, while the right hand * Si
tin

in Thus)) side is equal to v(Si) + v(Ti) by (41:6) 41.3.2. i in Si

in

t in TI

henceat leastone
must

of))

in

Sl

t in 2*,
>

be true.

Thus of the three conditions of domination in 30.1.1. a ^ ft ) (for for holds (30:4:c) for both of Si,TI and (30:4:b) at leastone of (30:4:a), them. Hence, may replaceur original S by eitherSi(sJ) or Ti(fi -K\. we o This completes proof. the
1
t I.e.his

extra

restriction

on

S does not

(in this

case!) odify m

the

concept of

domination.)))

384

OF AND COMPOSITION DECOMPOSITION GAMES

46.2.Decompositionand Its Relation to the Solutions:First Results Concerning F(e ) 46.2.1. now directour coursetowards the main objectiveof this We of the theory: Thedeterminationof all solutionsU/ of the decomposable part a gameT. This will be achievedin 46.6., concluding chain of sevenlemmas. We begin with some purely descriptiveobservations. Considera solution U/ for F(e) of T. If U/ is empty, there is nothing more to say. Let us assume, therefore, that U/ is not empty owing to this (45: (or equally to (45:0)) is equivalent to)) A)
in we Usingthe notations of (46:1) 46.1.1.form:

Max-. U/ am (46:10)

./()
i.e.
,

= *>,

= p, . 'Max-.U/ flf() = ft m
Min-* , , /( a ) in U/

)= inU,^ *-'
question

1That all these quantities can be formed, that the maxima and the minima in exist and are assumed, can beascertained a simple continuity consideration. by

Indeed/( a )
functions

'2/J a
in

~~

v (^) and

0r(

a ) = 2^ a
i in X

v(K) are both

continuous and domain

minima is therefore

of a ,

of i.e. its components 01,

a well

known

of consequence the

o Theexistence f their maxima

continuity

properties of the

of a

the set U/. the readerwho is acquainted with the necessary mathematical background we give the precise statement and its proof. (Theunderlying mathematical topology facts are discussed by C. Carathtodory, loc. footnote 1 on p. 343. Cf. there pp.136-140, particularly theorem In order to be sure U/ is a set in the n-dimensional linear spaceLn (Cf. that every continuous function has a maximum and a minimum in U/, we must know that U/ is bounded and closed. Now we prove: is a bounded and closed Q (*) Any solution U for F(e ) [E(e)] of an n-person game set in Ln

For

e.g.

5).

cit.,

30.1.1.).
r

If Proof:Boundedness: a
n

{on,

, <x) belongs to

U, then every

a,

2>

v((i))

and

2/

~ v(7)

=s

o, hencea,

v(7)

+e

2} / ^ V CO+ *o

^ v((t)).

Soeach is restricted to the

fixed interval

and sothese a form a bounded set. Closedness: is equivalent to the openness f the complement of U. That set is, This o > > in the set of all which are dominated by any a of U* (Observe))) by (30:5:c)

30.1.1.,

DETERMINATION OF ALL SOLUTIONS

Given two
>

385
>

unique

a = 7 = {71,

{!,
,
-y n }

an j,

which

= {ft, a , n there exists has the same as J-component a , and

ft
}

the same JC-componentas (4Q.H)

ft

:
for for in

7. = 7 = ft

i K.
V
of

tin/,

46.2.2. now prove: We

(46:C)
if

and only if

If ^T, V belongto U/, then the

(46:11) to U/ belongs

(46:C:a)
Incidentally

/(*)+0(7)S

o.

(46:C:b)

6(7)=/U)+0(7).

F Proo/: ormula (46:C:b): (46:3)in By and clearly/( 7 j( 7 ) = fif(7)-

)=/(),

46.1.1. = /(V)+ 0(V), e(V)

*

Since t Necessity of (46:C:a): U/fif^o),herefore e(y) ^ e is necesthis with (46:C:a). sary and by (46:C:b) coincides y Sufficiency of (46:C:a): is clearly an extendedimputation, along
with

and (46:C:a), that y belongsto F(e).1 (46:C:b) guarantee a Now assumethat y is not in U/. Thenthere exists 8 H 7 in U/. for set /S of 30.1.1. this domination may be chosenby (46:B) with The > Now clearly 5 H 7 implies,when S J that 6 H a

a,

ft

,))

that

we areintroducing For
any

the solution

a denotethe > a of U. the sum of all a Sincethe sum of any number (even of infinitely many) open setsis again open, it this: If ft ^ a , then for every ft of suffices to prove the openness each D-+,i.e. a

-,

o character f U at this point!) H a by D-+. Then the complement of U is set of all

<*

'

which

ination, ft ^ a by And the validity of

is sufficiently > *

near to

ft

, we have also ft

'H a.
->

Now in the definition

of dom-

(30:4:a)-(30:4:c)in30.1.1.,
(30:4:c)
same is not

since(30:4 is a < relation. :c)

(Note that
the

in appears the condition (30:4 only. is clearly not impaired by a sufficiently small change of 0<, in true for a , because a appears (30:4:b) also,and is arbitrary small changes, since(30:4:b) a ^ relation.

:c)

(30:4:b) ight be destroyed by m

this property for ft , and not for But we needed 1This is the only useof (46:C:a).)))

a !)

AND O 386 COMPOSITION DECOMPOSITIONF GAMES and when S K that tives areimpossible.
d

ft

As 5 ,

a,

ft

belongto U/, both alterna-

Hence7
We

must

restate(46:C)in

belongto

U/>

as asserted.

an obviously equivalent form:

(46:D)

If-constituentsof U/.
U/

Let V/ be the set of all /-constituents Wx the set of all and

Then U/ obtains from theseV/ and Wx as follows:

is the set of all those y , which have a /-constituent a in V/ and a /^-constituent ft in W* such that

'

'

(46:12)

e(

O + e(7')^ e .
Q

46.3.Continuation 46.3. Recalling the definition of U/'sdecomposability (for/, in (44:B) with little difficulty, one that it is equivalent to this: in 44.3.1., sees U/ obtains from the V/, Wx of (46:D) as outlined there, but without

/)

the condition
in

Thus (46:12) be interpreted as expressing to what extent U/ may just This is not decomposable. is of someinterestin the light of what was said
in One may even go a step further: The necessityof (46:12) (46:D) is i.e. easy to establish. (Itcorrespondsto (46:C:a), to the very simplefirst two steps in the proof of (46:C)). Hence :D) expresses U/ is no that (46 than unavoidable. further from decomposability, in All this, in conjunction with (44:D:b)44.3.3., suggests trongly that s o V/, WK ought to be solutionsof A, H. With our present extensionsf all which F(/o), F(go) to take; conceptsit is necessary, however, to decide 2 w /o beingthe excess e proposeto use in and go the one in K. It will are t appearhat the p, # of 46.2.1. these/o, g<>.

(46:12).

about there. 44.3.3. (44:D:a)

/,

we Indeed, can prove:

V/ is a solution of A for (46:E:a) W/c is a solutionof H for (46:E:b) It is convenient, however, to derive first another result: 1Note that these a
uents

(46:E)

', ' are not the a , of (46:C) they are their /-,K-constitas well as those of . e( a ;), e( ') are the excesses a ', ' formed in /, of
ft

ft

ft

to

they are equal to /( a ), g( ft ) as well as to /( 7 ), g( y ). (All of this is related (46:C)). * Thereaderwill note that this is something like a question of distributing the given e excess in / between / and

K. But

K.)))

DETERMINATION OF ALL SOLUTIONS

387

(46:F)

(46:F:a) (46:F:b)

p+
$

=e, + }= e .

Note that in (46:E), s well as in (46:F),he parts (a), (b) obtain from t a eachother by interchanging J, A, p, f with Kt H,#, $. Hence sufficesto it prove in eachcase only one of (a), (b) we chose(a).
Proof of
mum

in therefore (46:3)

p . Sincenecessarily a ) ^ e and sinceby definition g( a. ) ^ , e(

Choose (46:F:a): an *a

>

in U/ for which
>

/( a ) assumesits maxi*

46.1.1. gives

(46:13)
Assume now that
further

$+ $ e . is (46:F:a) not true. 9+ < 6 .

Then

would (46:13) imply

(46:14)

also Use the above a in U/ with /( a ) = p, and chose a in U/ for which 0( ) assumes its minimum . Then /( a ) + 0( ) = 9 + ^^00 (by
or Thus (46:13) (46:14)). the V of (46:C) belongs to U/, too, Again

(46:C)together

with

(46:14) ives g

e(y) /(\")+ |/(7)= 9 + * <

o,

i.e. 7i < v(7) + e . ^ t-i

5

Now define

{i,
n

,}
5<

= {71 +

>'>7 + c),
^

choosing
>

>
^

so that

-i

= v(/) + e . Thus
TJ

belongsto F(e).

existin U/. By s (45:J) t? H 7 , which is impossible, ince y , 7 areboth in U/. Hence 8 v(J) > ^ 7 ~ v(J) = ^ a, v(J), belongs to U/. Now i in / J iinJ t in i.e. d ) > /( a ) = ?,contradictingthe definition of p. /( must Consequently(46:F:a) be true and the proof is completed. o :If ' Proof of (46:E:a) a belongsto V/j> then it is the J-constituent f an of U/. Hence footnote 1on p.386)e( a ') =* /( a ) g ?,so that a (cf.
If 6 did not belong to U/, then an

H. 5
>

would

belongsto F($). Thus V/)))

'

AND OF 388 COMPOSITION DECOMPOSITION GAMES of So our task is to prove (44:E:a), (44:E:b)44.7.3. ' H ft ' happenedfor two a ', ' in V/. Ad (44:E:a): Assume,that a
>

Then

a ',

>

>

' are the J-constituents two of

>

7,

>

5 in

U/. But a ' H

>

>

'

clearlyimplies 7 *< 6 , which is impossible. :Consideran a ' in F(<p) but not in V/. Then by definition Ad (44:E:b) > in U/ mentionedin the above proof of (46:F:a), e( a ') g . Use the be the jfiC-constituent of this ft , so that for which g( ) = . Let + ~ft' is in W* and e(T')= gCft) = f Thus 6(T') e(7')^ ^ + = e Form (use (46:F:a)). the y (for /), which has the J-, ^-constituents

'

~ \"^',T' Then e(7 ) = e( a ') + e(]*') e i.e.y belongsto F(e). ' to its y does not belong U/ because /-constituent a does not belong thereexists d H y in (the solution for F(e)) U/. a to V/. Hence, for Let S be the set of 30.1.1.the domination 5 H y . By (46:B)we
Q Q

has the same Jf-constituentsft ' as from d H 7 that 5 H . Sinceboth 5 , ft , we can conclude to belong U/, this is impossible. Denotethe J-constituentof d by 6'; as 6 Consequently ' belongsto U/, therefore 6 belongsto V/. 7 has the J-constituenta Assume first
y

may assumethat

Ss J or S c K. that S K. As

SsJ.

f we Hence canconcluderom

6 H

Thus we have the desired 6

'

7 that 5 from V/ with

' a '. ' a '.

H 5 H

'.

the 46.4.1. :D),(46:E) expressed generalsolution U/ of T in terms (46 of appropriate solutions of V/, Wx of A, H. It is natural, therefore, to this procedure: start with the V/, W* and to obtain U/. To try to reverse It must be remembered, however, that the V/, W* of (46:D) are not t of in entirely arbitrary. If we reconsiderhe definitions (46:10) 46.2.1. t this light of (46:D),hen we seethat they can also bestatedin this form:))

46.4.Continuation

(46:15))))

DETERMINATION OF ALL SOLUTIONS

And

389

a (46:F) expressesrelationshipof thesep, tf>, #, which aredetermined with eachother and with e by V/, W/c 46.4.2. will show that this is the only restraint that must be imposed We the V/, W*. To do this, we start with two arbitrary non-empty upon solutions V/i W/c of A, H (which need not have beenobtained from any solution U/ of T), and assert s follows: a Let V/ be a non-empty solution of A for F($) and Wx a non(46:G) empty solution of H for F(#). Assume that p, ^ fulfill (46:15) above, and also that with the ^>, ^ of (46:15)

(46:16)
>

Forany a ' of V/

+ = ? + = eo.
and any
ft

' of W/c with

(46:17)

(O+e(7') 60,
form the 7 (for /) which

has the 7-, ^-componentsa ',

>

'.

Denotethe set of all these 7 by U/. The U/ which are obtained in this way are precisely all solutionsof F for F(e). A a Proof: ll U/ of the stated character re obtained in this way: Apply a (46:D) to U/ forming its V/, W/c. Then all our assertions recontainedin constructed with the help of V/,
Consideran U/ obtained in this way have the stated character: a W/c as describedbove. We have to prove that this U/ is a solution T for F(e).
All

(46:D), (46:E), (46:F)togetherwith (46:15).

U/
Q

o Let a ', ' be the /-, K-constituentsf 7 and 6 ', 6 the J-, X-cona stituents of from which they obtain as describedbove. Let S be the for set of 30.1.1.the domination y H y . By (46:B)we may assumethat
77

of U/ our (46:17) ives e( y ) = e( a ') + e( ft ') ^ e , so)) g that belongsto F(e<>). Thus U; F( of So our task is to prove (44:E:a), (44:E:b)44.7.3. Ad (44:E:a): Assume that rj H y happened for two i? , y in U/.

For every 7

<=

'

S c J or
which

SSi K.

Now S c J would cause

>

is impossible,since 6 ', a both belong to V/; and S X would both cause if H 7 to imply ' H /3 ' which is impossible,since e ',

'

T;

7 to

>

imply

5 H

'

a ',

'

to belong W*.
Ad

of Assume per absurdum, the existence a 7 (44:E:b):

> 17

>

in

/-,

not in U/, such that thereis no o It-constituents f

7)))

of U/

with

17

7 . Let a ',

F(e) but be the

'

OF 390 COMPOSITION DECOMPOSITION GAMES)) AND

Assume first e( a ') ^ Then belongs to F(p). Consequently either a belongsto V/ or thereexists a 6 in V/ with 5 H a In the

'

<f>.

a'

'

'

lattercase choosen a
Form the
17

6 in Wx

'

'.

for which e( ') assumesits minimum value

with

the

/-, /^-constituents 5 ',

respectively,and as e( d ') + e(*') g + = e , therefore i? ' T belongsto U/. Besides ; H y owing to 6 ' H a (thesebeing their Jy constituents). Thus 17 contradictsour original assumptionconcerning . Hencewe have demonstrated, for the caseunder consideration,that a ' must belongto V/. In otherwords:

'.

As

', e ' belong to

V/,

W*,

(46:18)

Either

~' belongsto V/, or e(7') > ?.

<f>,

Observe that in the first casenecessarilye( a ') ^ in the second casee( a ') > ^ <p. Consequently:

and of course

ratee( a. ') ^ into J ( (46:19) these: Interchanging and K carries 46:18),

(46:19)
(46:20)

At

any

^>.

Either
At

r
~f$

belongsto
f

Wx

or e(V')

>#
gives in con-

(46:21)

any

ratee(~^) ^ ^.
then (46:18), this

Now if we had the second alternative of with (46:21) junction

e(7) = e( O + e(7')> ^ + ^ = c ,
which is impossible, s y belongsF(ed). The secondalternative of (46:20) a is equallyimpossible. and i Thus we have the first alternatives in both (46:18) (46:20),.e.))

a ', ft belongto V/, W*.

'

As y

belongsto F(e), therefore

e

e(~O+ e(7;) = e(7) ^

Consequently 7 must belongto U/

.
Q

contradictingour original assumption.

46.5.The CompleteResult in F(e ) 46.5.1. result (46:G)is, in spite of its completeness, The unsatisfactory in one respect: on The conditions(46:16) (46:17) which it dependsare and

altogetherimplicit. We
much

more transparent conditions.)))

will,

therefore,

them replace

by equivalent, but

DETERMINATION OF ALL SOLUTIONS

391

To do this, we beginwith the numbers<p, $ which we assumeto be given first. Which solutionsV/, W* of A, H for F(p),F(#) can we then use in of the sense (46:G)? First of all, V/, W* must be non-empty;applicationof (45: ) or (45:0) A to A, H (insteadof F) showsthat this means

* -|A|i, # -|H|i. to A, Considernext (46:15). Apply (45:P)of 45.6.1. H (insteadof T). while Then (45:P:a) ecures s the two Max-equationsof (46:15), (45:P:b) transforms the two Min-equations (46:15) of into 2 (46:23) $ = Min (?, |A| 2), * = Min (},|H|). Let us, therefore, define by (46:23). Now we express46:16), i.e. ( +^ = + =e. (46:16) The first equation of (46:16) alsobe written as may
(46:22)
v>,
\\l/ _<,

i.e. y (46:23) b
(46:24)
Max (0, ?

<P

=#

- ^,
<p

= Max (0,# |H|).1 2 Now a 46.5.2. two cases repossible: a Case(a): Both sides of (46:24) re zero. Then in eachMax of (46:24) is the 0-term ^ than the other term, i.e. 2 |H| ^ 0,i.e., |A| 2 ^ 0,#
|A| 2)

^ |H|,. i b holds, then (46:24) ecomes = 0, i.e.t is autoConversely:If (46:25) matically satisfied. Now the definition (46:23)becomes (46:26) ? = v, ^ = ^, (46:25)
|A| 2,

?^

and so the

full

condition

(46:27) a g (46:25) nd (46:27) ive also e g |A| 2 + |H| = 2. 2 (46:28) are Case(b): Both sides of (46:24) not zero. Then in each Max of > is the 0-term < than the other term i.e. |A| 2 > 0,# |H|, 0, ? (46:24)
Q

becomes (46:16) 2 ? +*= e.

|r|

i.e.

(46:29) 1Cf. (45:P:c) and 1

based.
Of which
we

? > |A| 2,
first part

#
to

> |H| 2

.'
which this discussion

its proof. used only the

obtain

o (46:24), n

is

1 Note that the important point is that (46:25), 46:29)xhaust all possibilities e ( d This is, of course, ue to 2, or * > |A|i, that we cannot have * |A| a, $ > vanish or neither. that both sides which forces the equation (46:24), The meaning of this will appearin the lemmas which follow.)))

|H|

|H|j.

i.e.

i i.e.t is not automatically satisfied. ? * |A|i + , (46:30)

OF AND 392 COMPOSITION DECOMPOSITION GAMES b $ h Conversely:If (46:29) olds,then (46:24) ecomes |A|2 = #

We can

express46:24) y writing ( b

-|H|,

= |H| + o>, 2

b and then (46:29) ecomes simply

(46:31)
(46:32)
and so the
full

co

> 0.
2 * = |H|,

b Now the definition (46:23) ecomes

? = |A|2,
|A|2

condition (46:16)*ecomes b

i.e.
(46:33) (46:34)

+ |H|t+ co = 6 ,
e

=
6

|r| +
2

w.

and g (46:31) (46:33) ive also

> |F|. 2

46.6.3. Summingup :
(46:H)
o The conditions(46:16), (46:17) f (46:G)amount to
m One of the two following cases ust

together
and

Case(a): 1) (
with

(2) [(3)
(4)

-|r|i e |r|, -|A|I*S |A|,, -|H|i**S|H|,,

9+ * = .e,> |r|,,
P > |A|i, * > |H|( J
|A|t

hold:

this:

Case(b): (1)
togetherwith
and

(2) (3)
(4) 6

^^

knew all along,that 6 ^ |r|iand ^ |A|i, The (46:28), (46:27) -|H|i.the other conditionscoincide case. (46:25), which contain of this

- |r|, 9 =

= *~

|H|,.
<f>

C Proo/: ase(a):We

with

of w which subsumes(46:31) under (l)-(3)).

complete description Case (b): These conditions coincidewith (46:34),(46:29), (46:30), w (after elimination descriptionof the case (46:33) hich contain the complete

1Cf.footnote 2 on p. 1The readerwill note that while (l)-(3)or (a) and for (b) show a strong analogy, f the final condition (4) is entirely different for (a) and for (b). Nevertheless,all this was obtained by the rigorous discussion of one consistent theory! More will be said about this later.)))

391.

DETERMINATION OF ALL SOLUTIONS

393

46.6.The Complete Result in E(e ) 46.6.46:G) and (46:H) characterizehe solutionsof T for F(e) in a t ( and t complete explicitway. It is now apparent, too,hat the casesa), (b) ( of (46:H) coincide with (45:0:b), in 45.6.1.: Indeed (a), (b) of (45:0:c) a (46:H) are distinguishedby their conditions (1), nd these are precisely

(45:O:b), (45:O:c).
We now

all our information.

T (45:0). his will

combinethe results of (46:G), (46:H) with those of (45:1), give us a comprehensive icture of the situation, utilizing p
If eo <

(46:1) (46:I:a)

(1)

-|r|i,
r, for E(e ) as well as

then the empty set is the only solution of for F(e). 9 If ^e (46:1 :b) (1)

-|r|i

|r|,,

then T has the samesolutionsO/ for E(e ) and for F(e). These O/ arepreciselythose sets,which obtain in the following manner: Choose two p, # so that any
Q

and

(2) (3)
(4)

^ -|A|i p

^ -|H|i * g |H|,
2

|A| 2,

? + # = eo.
two

44.7.4.
If

solutionsV/, W* of A, H for JE(p), E($). Then U/ is the compositionof V/ and WK in the senseof

Chooseny a

(46:I:c)

(1)

>

|r|
2
,))

F(e). TheseO/ and

(2) (3)

then F does not have the samesolutions O/ for E(e^)and U/ for

U/ are preciselythose sets which F in the following manner: orm the two numbers , # with

obtain

which

aredefined by
(4) eo

- |r| = 9 - |A|, = * - |H|.

2 2

9 > |A|i, 2 f > |H|,

Choose two solutionsV/, WK of A, H for E(v),E$). any The Then O/ is the sum of the following sets: compositionof a V/ and of the setof all detached (in K) with e(ft ') = |H|; the of composition the setof all detacheda (in J) with e( a ') =

7'

'

|A|*)))

394

OF AND DECOMPOSITION GAMES COMPOSITION

of and of WA:; the composition the set of all detacheda (in J) with e( a ') = and of the set of all detached (in K) with
<p

'

'

and
U/

e( ft ') = ^, taking all pairs (5) |A| 2 < *, < 9,

(6)

^>,

^ with

|H|i< * < *,

?+ ^ = e . ?+ * g 6 .

obtains by the same process, replacingthe condition (6) only

by

(7)
A Proo/: d Ad

This w (46:I:a): coincidesith (45:O:a). f This is a restatement f case(a) in (46:H) exceptor the o (46:I:b): modifications:

First: he identification of the E and F solutionsfor F, A, H. This is T to ( justified by applying (45:O:b)F, A, H which is legitimate by (1),2),(3) of (46:I:b).
This is justified by observing that (46:17)is the condition (46:17). fulfilled: V/ = Vj #(?), W* = W* E($), hence for automatically a in V/ and ft in W* always e( a. ') = 9, e( ft ') = # and so by (4)

following

The Second: way in which we formed O/ = U/ from V/ = V/, W/c = Wx in which differed from the one described (46:H) insofar as we omitted

'

'

e(

O+
Ad

b consider oth E and F solutions for F (not only F solutionsas in (46:H)),and use only E solutionsfor A, H (not F solutionsas in (46:H)). The way in which the former O/, U/ of F areformed from the latter(Vj of different from the onedescribedin (46:H). A, Wx of H) is accordingly In orderto remove these differences, one has to proceed follows: as and to ( Apply (45:1) (45:O:c) F, A, H which is legitimate (1), 2), (3) by of (46:I:c). Then substitute the defining for the defined in (46:H). If these out then manipulations arecarried on (46:H) (in the presentcase (46:I:c)), l preciselyour above formulation results.
We

modification:

This o (46:I:c): is a restatement f case(b) in

(7') = *o.

f (46:H), xceptor this e

46.7.Graphical

Representation

of a Part

of the Result

o expressionf several simple qualitative principles. The only the precise reasonfor going through the intricaciesof the preceding mathematical t derivation was, of course, hat theseprinciplesare riot at all obvious,and that this is the way to discoverand to prove them. On the otherhand our result can be illustrated by a simplegraphicalrepresentation. 1If the readercarries out, he will seethat this transformation, although somewhat this
cumbersome, presents absolutely
no
difficulty.)))

46.7. he results of (46:1) seem T but may complicated,

they

areactually

DETERMINATION OF ALL SOLUTIONS))

We
A

395))

begin with a more formalistic remark. look at the threecases discloses W (46:I:a)-(46:I:c) this: hile nothing more can be said about (46:I:a), two other cases(46:I:b), the (46:I:c) have some common features. Indeed,n both instances desiredsolui the tions O/, U/ of F areobtained with the help of two numbers , # and certain solutionsV/, W* of A, H. Thequantitative elementsf the o corresponding o representationf O/, U/ arethe numbers , #. As was pointedout in footnote 2 on p.386, represent omethinglike a distribution of the given they s e excess in / betweenJ and K.))

Figure

69.

in and arecharacterized the cases (46:I:b) (46:I:c) their respecby theseconditionsfor (46:I:b) conditions (2)-(4). Let us compare tive and ,#

for

, They have this common feature They force the excesses # to belong of e to the same case A, H as the one to which the excess belongsfor I They differ, however, very essentiallyin this respect:n (46:I:b) they imposeonly one equation upon , # while in (46 :c)they imposetwo equa1 tions. Of course,the inequalities too,may degenerate occasionallyto but s equations (cf. (46:J)in 46.8.3.), the general ituation is as indicated. The connections etween e and p, # are represented b graphically by

(46:I:c).

r.

:1

Fig.69.))

for two equations

1 (2), (3) are inequalities

in

in both

(46:I:c).))

cases, (4) stands for oneequation

in

(46:I:b)

and)))

OF AND 396 COMPOSITION DECOMPOSITION GAMES This figure showsthe p, #-planeand under it the 6o-line. On the latter mark the division into the three zonescorresponding The tocases (46:I:a)-(46:I:c). ^-domainwhich belongsto case(46:I:b) t marked (b) in the p, #-plane; he , ^-domain, coversthe shaded rectangle coversthe line marked (c) in the p, ^-plane. which belongsto case(46:I:c) leads to its e value Given any $,#-point, followingthe line thus 6, V yield a, a',respectively. Given any e -value the reverse process all discloses its 9, ^-points, thus a producesan entireinterval at 6, while 1 a'yieldsthe unique point b'.

the points

-|r|i, |r|i

<?>,

46.8.Interpretation

:The Normal

Zone.

Heredity

of Various

Properties

46.8.1. 69 calls for further comments,which are conduciveto a Figure fuller understandingof (46:1). indications(for the last time in the comFirst: have beenrepeated There that and ment following (45:0)), the cases (46:I:a) (46:I:c),e < r k values of e and 60 > |F|j,espectively arethe \"toosmall\"or \"toolarge\" r e g 2, is in some of in the sense 44.6.1.; that case(46:I:b), ^ e way the normal zone. Now our pictureshows that when the excess of F lies in the normal zone, then the correspondingexcesses, # of A, 2 n H liealso within their respective ormal zones. In otherwords: in is hereditary from Thenormal behavior (positionof the excess (46:I:b)) T to A, H. In t Second: the case(46:I:b)he normal zone $ arenot completely on determined eo, as we repeatedly saw before. In case(46:I:c), the by other hand, they are. This is picturedby the fact that the former domain w is the rectangle in the <p, #-plane, hile the latterdomain is only a line (c). (b) It is worth noting, however, that at the two ends of the case(46:I:b) 8 for e = the interval available for p, # is constricted a point. to to Thus the transition from the variable 9, # of (46:I:b) the fixed onesof

i.e.

i.e.,

|F|i

|r|

-|

<f>,

|T|i, |r|f

Third: ur first remarkstated that normal behavior (i.e., the position O that is of the excessorrespondsto (46:I:b))hereditary from T to A, H. It is c remarkablethat, in general, o such heredity holds for the vanishing of the n not i excess, that eo = 4 does in generalmply p = 0,# = 0. It is preour w cisely the vanishing of the excess hich specializes presenttheory (of 44.7.)to the olderform (of 42.4.1. as we know, is equivalent to the which, t We will examinehe variability of p, # when e = original one of more closelyin the last (sixth)remark. Beforewe do that, however, we give our attention to the connection between our presenttheory and the older form.

is (46:I:c)continuous.

i.e.

30.1.1.).

1 We leaveto the readerthe simple verification that the geometricalarrangements of Fig. 69.express,ndeed,the condition of (46:I:b), i f Se * H#li * S l#k cf. (46:I:b). implies 1This is onecaseof degeneration,alluded to l*k end of 46.7. ftt the 4 Of e liesin the normal case s course,

I.e.-lrU
Q

|r|,

-|A|i

(46:I:c). * (46:I:b): -|r|i

|r|
.)))

DETERMINATION OF ALL SOLUTIONS

must of

397

Fourth: It is now evident that the present, wider form of the theory necessity receiveconsideration,even if our primary interest is with the original form alone. Indeed: order to find the solutionsof a in decomposable game T in the original sense(for e = 0), we need solutions for the constituent gamesA, H in the wider,new sense(for p, # which may
Q

not be zero). This gives the remarks of 44.6.2.more precise eaning:It is now a m apparent how the passagefrom the old theory to the new one specifically becomes necessarywhen the game (A or H) is lookedupon as non-isolated. The exact ormulation of this idea will come 46.10. f in Fifth: 46.8.2. We can now justify the final statements about (44:D) in 44.3.2. (44:D:a), and shows that (44:D) (44:D:b)in 44.3.3. 46:I:b) ( is true in the case(44:I:b), relinquishthe old theory for the new one; if we shows even (44:I:c) that (44:D) is not true in the case(44:I:c) at that to price. Thus the desire securethe validity of the plausibleschemeof (44:D) motivates the passageto the new theory as well as the restriction t to case(44:I:b)he normal case. If we insist interpreting (44:D), 44:D:a), by (44:D:b) the old theory, ( then (44:D),(44:D:a)ail,1 while the conditional statementof (44:D:b) f

remainstrue.2

saw that e = does not in generalimply p = 0, doesthis \"in general\" ean? m # As are subjectto the conditions (2)-(4)of (46:I:b). eo = 0, so , # the and permitsus to express remaining (2),(3)in means that # = (4) this:)) terms of alone. They become

= 0.

46.8.3. We Sixth:
What

<p

Now apply (45:E) to A, H. Then we see: a t If A, H areboth essential,hen the lower limits of (46:35) re < and the upper limits are > 0,so p can really be 7* 0. If eitherA or H is inesseni tial, then (46:35)mpliesp = and hence# = 0.

statethis explicitly: holds e = implies9 = 0,# = 0,i.e.44:D) of 44.3.2. even ( (46:J)

We in the

senseof old theory if and only

if eitherA or H is inessential.

46.9.Dummies We 46.9.1. can now dispose of the narrower type of decomposition, in described footnote 1on p.340 the addition of \"dummies\" \"to the game. Considerthe game A of the players1', ,*'.*\" Inflate it by adding of to it a series \"dummies\" ;i.e.omposeA with an inessentialgame H K c of the players 1\", game is F. , I\". Then the composite
1 As of (44:B:a)

e - 0, ^ 0, # & 0. Then the decomposability is violated, asstatedin 44.3.3. 44.3.1. * Representing the special asee c 0, 0, # - 0. 1
we may have
\"-

requirement

It is now

convenient

to reintroduce

the notations

of

for the 41.3.1.

players.)))

OF AND 398 COMPOSITION DECOMPOSITION GAMES

in We will use the old theory for all these games. By (31:1) 31.2.1. there existsprecisely one imputation for the inessential game H say 1 a H or :P) in 31.2.3.possesses , 7?\"}. By (31:0) (31 Vi = M\", s T unique solution: heoneelement et ( y ). s the Now by (46:J)and (46:I:b) general olution of F obtains by composing the generalsolution of A with the generalsolution of H and the latterone is unique! In otherwords: \" Inflate \" every imputation / = {j8r, A , fly) of J (i.e. ) to an

imputation

s tion of F from the general olutionof A. This result can be summedup by sayingthat the \"inflation\" of a game d by the addition of \"dummies\" oes not affect its solution essentially it is the only necessaryto add to every imputation componentsrepresenting a \"dummies,\" nd the values of these componentsare the plausible ones: What each \"dummy\" would obtain in the inessential game H, which their describes relationshipto eachother. We 46.9.2. conclude adding that (46:J)statesthat if and only if by the compositionis not of the specialype discussedabove, the old theory t to ceases have the simple properties of the new one, and its hereditary character as indicatedin the third remark of 46.8.1. fails, 46.10. Imbedding of a Game In the fourth remark of 46.8.1. reaffirmed the indicationsof we to which the passage from the old theory to the new one 44.6.2., according becomes necessarywhen the game is lookedupon as non-isolated.We will now give this ideaits final and exact xpression. e It is more convenient this time to denotethe game under consideration It ought to be understoodthat this A by A and the setof its playersby is perfectly general no decomposability A is assumed. of We beginby introducingthe concepts hich areneeded treata given to w A as a non-isolated This occurrence: amounts to imbeddingit without game a modifying it, into a wider setup, which it is convenient to describes anothergame F. We define accordingly: is imbeddedinto F, or F is an A 2 imbedding of A, if F is the compositionof A with another game H. In 1 otherwords,A is imbeddedin all thosegamesof which it is a constituent. 1Recallthe notations of 44.2. * Thegame H and the Bet of its players K are that K and J perfectly arbitrary, except must be disjunct. 1Sincea constituent of a constituent is itself a constituent (recallthe appropriate

a / of I (i.e. by composing with y J, i.e.y addingto it the it b T) ' , 7?-: / = [Pi',' ' ' , A*, 7?\", , 7?\"}.Then components7?\", of i.e.f composition producesthe generalsoluo this process \"inflation\"
*

46.10.1.

J.

definitions,

in particular

imbedding.
considering

i an imbedding of an imbedding is again an (43:D)n In other words; Imbedding is a transitive relation. This relieves from us relationships

43.3.1.),
basedupon

any indirect

it.)))

DETERMINATION OF ALL SOLUTIONS

399

Let 46.10.2. us now investigate the solutionsof A viewing A as a nonI isolatedoccurrence.n the light of the above,this amounts to enumerating all solutionsof all imbeddingsF of A, and interpretingthem, as far as A is concerned. The last operation must be the taking of the J-constituent in the senseof 44.7.4. We know from the fifth remark in 46.8.2. this that i is only feasible,f we consider solutionsfrom outsidethe normal zone (b). no One might hesitatewhether the solutionsof F should be taken in the senseof the old or the new theory. The former may appear to be more :The justified on the standpointof 44.6.2. outsideinfluences upon the game for having beenaccounted by the passagefrom A to H, thereis no longer for any excuse going outside the old theory.1 It happens, however, that we need not settlethis point at all, because result for A will be the the i same,rrespectivelyof which theory we use for F. But if we use the new as theory for F, we must restrict ourselvesto the case(46:I:b), discussed above. Thus the questionpresentsitself ultimately in this form: (46:K) Considerall imbeddingsF of A, and all solutionsof theseF: f (a) in the senseof the old theory, i.e.or (0),

(b) in the senseof the new theory

Which

in the normal zone,

i.e.

for))

o arethe J-constituents f the solutions? The 46.10.3. answeris very simple:

(46:L)

arepreciselythe following

The J-constituents the F solutions)referredto in (46:K) (of All sets: solutionsfor A in the normal This f zone, i.e.or any JE(p) of (46:I:b). is true for both (a)and (b) of (46:K).
Q

4 on p.396),hence we need only show that all the sets (a) is narrower than (b). Therefore, obtained from (b) are among the onesdescribed above, and that all these setscan even be obtained with the help of (a). of The first assertionis only that of the hereditarycharacter the normal zone (b). if assertionfollows from (46:I:b),we can do this: iven a $ G Thesecond find a game H and with with t ^ 9 # g |H|, such solutions for E(ifr). Now such an that $ + # = and that H possesses a H exists, nd it can even be chosenas a three-person ame. g Let Indeed: H be the essentialthree-person amewith general7 > 0. g = i Then by (455)in 45.1. = 87and by (45:9)n 45.3.3. = J|H|i %y. |HK and what we know now amounts to We have required # =

(cf. Proof: = belongsto case(46:I:b) footnote

-|A|i

|A|,

-|H|i

|H|,

<p

-87 ^ * ;Sfy.
imbedding

1Besides,he transitivity pointed out in footnote 3 on p. 398,shows t of r can be regarded directly as an imbedding of
A.)))

that any further

400

OF AND DECOMPOSITION GAMES COMPOSITION

This canclearlybemet by choosing sufficiently great. Then we alsoneeda y of solutionof H for E($). Theexistencesucha solution(for 87 ^ # g fy)

will

of I First:f we wanted to handlethe process imbeddingin such a manner T that the old theory remainshereditary, we would have to seeto this: he of T from A and H has to be such that e = implies = composition (and hence# = 0). By (46:T) this means\"that eitherA or H areinessential. The lattermeans (cf. eod),that only dummies\"areadded to A.
Q
<f>

To 46.10.4.this result two more remarksshouldbe added:

beshown in 47.

Summing

up:

(46:M)

The old theory remains hereditary if and only if eitherthe or to original gameA is inessential, the imbeddingis restricted \" the addition of \" dummies to A.

which

theory to the new one,as another player. Our above result (44:L)justifies a slightly modified view: The outside is of source 44.6.2.the game H which is added to A or rather the setK of

to It Second: was suggested already in 44.6.2. treat the outside source, and t createshe excesses paves the way for the transition of our old

its players.

in t Now we have seenthat the game H must beessential, ordero achieve w desiredresult. Furthermore e know that an essentialgamemust have the n ^ 3 participants, and the proof of (44:L)showedthat a suitableH with n = 3 participants doesindeedexist. So we see:

(46:N)

can as The outside sourceof 44.6.2. be regarded a group of t new players but not as one player. Indeed, he minimum effective number of membersof this group is

3.

The 46.10.5. foregoing considerationshave justified our passagefrom the old theory to the new one (within the normal zone (b)) and clarified \" the nature of this transition. We see now that the \" common sense fails surmiseof 44.3. to hold in the old theory, but that it is true in precisely that new domain to which we changed. This rounds out the theory in a

satisfactorymanner. The leading principle of the discussionsof 44.4.3.-46.10.4. was this: under considerationwas originally viewed as an isolated occurThe game but rence, then removed from this isolationand imbedded,without modification, in all possibleways into a greater ame. This orderof ideas is not g alien to the natural sciences, particularly to mechanics.The first standto closedystems,the second point corresponds the analysisof the so-called s to their imbedding, without interaction,into all possible greaterclosed
systems.)))

DETERMINATIONOF ALL SOLUTIONS

401

hasbeenvariously emphaThe methodical importance of this procedure sized in the modern literature on theoretical physics,particularly in the analysisof the structure of Quantum Mechanics. It is remarkablethat it couldbe made use of so essentiallyin our present investigation.

a game T in the normal zone i.e. fortiori for every solution in the sense of the old theory numbers p, #. This and the immediate properties of p, # in connection with the solution, appear to be so fundamental, as to deservea fuller non-mathematical exposition. two We areconsidering gamesA, H playedby two disjunctsets of players and K. The rules of these gamesstipulate absolutelyno physicalconnection between them. We view them neverthelessas one game T but w i o this game, f course,s composite, ith the two isolatedconstituentsA, H. of the comLet us now find all solutionsof the entire arrangement, outsideof T, we it positegame T. Since is not desiredto consideranything 1 and 42.4. Then we have shown adhereto the original theory of 2 with the following propa that any such solution U/ determines number p in erty: For every imputation a of U/ the players of A (i.e. J) obtain the in the amount , and the playersof H (i.e. K) obtain together together Thus the principle of organization embodiedin U/ must amount stipulate (among other things) that the players of H transfer under all conditionsthe amount to the playersof A. o The remainder of the characterization of Uj i.e. f the principle of in or standard of behavior embodied it is this: organization T First: he players of A, in their relationship with eachother,must be regulated by a standard of behavior which is stable, provided that the 3 transfer of 9 from the other group is placedbeyonddispute. The Second: players of H, in their relationshipwith eachother, must be regulated by a standard of behavior which is stable, providedthat the 4 transfer of to the other group is placedbeyonddispute. Third:The octroyed transfer z> must lie between the limits (46:35) of 46.8.3.

46.11. Significance of the Normal Zone defines The 46.11.1. result (46:I:b) for every solution of the composite

i.e.

30.1.1.

1.

-p.

(46:35)
by o any stable social rder of T is based upon payment of one of the two groups to the other. The amount of this tribute is an those which can integral part of the solution. The possibleamounts,
any solution, The 46.11.2. meaning of these rules is clearly that definite tribute i.e. a

i.e.

I.e. - 0. - + # e 0, we do not introduce - -$. 1I.e. the /-constituent V/ of U/ is a solution of A for E($). that I.e.hat the ^-constituent W* of U/ is a solution of H for E(-$).))) t
* Since$

402

A COMPOSITIONND DECOMPOSITIONF GAMES O

a occurin solutions,arestrictly determinedby (46:35) bove. This condition showsin particular: t of T First: he tribute zero, i.e.he absence a tribute is always among the
The Second: tribute zero is the only possibleone if and only if one of the two gamesA, H is inessential(cf.the sixth remark in 46.8.3.). b I Third: n all other cases oth positive and negative tributes arepossible b i.e. oth the playersof A and the playersof H may be the tribute paying

possibilities.

2 under any conditions. outsideacceptance The results (45:E) and (45:F) of 45.3.1. acquire a new signifi-2.now cance: ccording to thesethe two numberscan only vanish together A (when A is inessential) and their ratio always liesbetweendefinite limits.

group has a minimum belowwhich no form of socialorganization can depress and, eachgroup has a maximum, above which it cannot |A|i, raise itself under any form of socialorganization: |H| |A|i, 2 a Thus, for a particularphysicalbackground,i.e. game, say A, the two numbers|A|i, |A| 2 can be interpretedthis way: |A|i is the worst that will be endured under any conditionsand |A|* is the maximum claim which

group. a The limits of (46:35) reset by both games A, H, i.e., the objective by 1 that each physical possibilitiesof both groups. These limits express

it:

|H|i;

may

find

46.12. First
secondand

o Occurrencef the Phenomenon

of

Transfer:n

have seen repeatedly (thus in (46:J)in 46.8.3. in the and remarksin 46.11.2.) the characteristic element that new of the theory of a compositegame T manifests itself only when both constituents A, H areessential. This is the occurrencef eo = 0,but o
third

46.12. We

-6

i.e. non-zero tribute in the senseof 46.11. a

Now we know that in order to be essential a game must have 3 players. If this is to be true for both A, H, then the compositegame T must have ^ 6 players. Six players are actually enough as the following considerationshows: Let A, H both be the essential three-person games with 7 = 1. Then = = = S 9 f, Hence |A|i = |H|i 3,|A|, |H|, f. (Cf.in 46.10.3.). for both p and # = $ lie between 3 and This implies,as will be shown in 47., he existence solutions t of V/, W* of A, H for U($),E($). Theircom-))

? = -** 0,

|.

the objective data, but by the solution, i.e. standard of behavior which happens to be generally accepted. 1It must be recalled that all this takes the value of the coalition of all players of A, we v(/), as zero;i.e., are discussing the losseswhich are purely attributable to lack of co-operationmong the group, and unfavorable general social rganisations and gains, a o which are purely attributable to lack of co-operation outside groups and favorable in general social rganizations.))) o

1But

where the actual amount

? liesbetween

those limits, is not determined

by

those

THREE-PERSON GAME Sincep was only restrictedy f b Thus we have demonstrated:

403

position U/ is then a solution of the compositegame T with the given 9. it g p ^ f , we can choose non-zero.

(46:0)

n = 6 is the smallestnumber of players for which the characteristicew element of our theory of compositegames (the n possibility of e = with $ = -# ^ 0, cf. above) can be
Q

observedin a suitable game.

We have repeatedlyexpressed belief that an increase the number the in of players need not only causea more involved operation of the concepts which occurred smaller numbers,but that it also may originate qualifor w tatively new phenomena. Specifically such occurrences ere observed as the number of playerssuccessively to increased 2,3,4.It is, therefore, of interestthat the samehappensnow as the number of playersreaches ix. s 1

47. TheEssential hree-person in the New Theory Game T

47.1. for Need

This Discussion

47.1.t remains for us to discussthe solutionsof the essentialthreeI t persongame,accordingo the new theory. This is necessary,since we have already made use of the existence of and these solutions in 46.10. 46.12., the discussionpossesses an but also interestof its own. In view of the interpretation which we were induced to put on these solutions in 46.12. also of their central role in the and 2 theory of decomposition,it seemsdesirableto acquirea detailedknowledge of their structure. Furthermore, a familiarity with thesedetails will lead a to otherinterpretationsof somesignificance. (Cf.47.8.nd 47.9.) Finally, we shall find that the principlesused in determining the solutionsin question areof wider applicability. (Cf. 60.3.2., 60.3.3.)
= 3, t = f (cf. Thus T, in the normalization 7 = wish to determine the solutionsof this T for (e ).8 In the referredto above we neededonly the normal zone 3 ^ e ^ fer to discussnow all e
47.2.Preparatory Considerations 47.2.1. considerthe essentialthree-personame,to be denoted by We g

1.

|r|i |r|

We 46.12.).

f but

applications,

we pre-

This discussionwill be carried out with the graphicalmethod, which we used in treating the old theory in 32. We will, therefore, follow the i scheme 32.n severalrespects. of
1For someother qualitatively
players, solution

* This is the only problem of absolutely general

*(- course, Of
before.)))

1 We are writing

-.

cf. 53.2.

new phenomena

which

emergeonly

when

there are six

at present!
the

o character, f which

we have

a complete

r, e although the applications employed the notations A, ? and H, r present r has nothing to do with the decomposable considered

404

OF AND COMPOSITION DECOMPOSITION GAMES The characteristic function is the sameas in 32.1.1.:))

(47:1)
An

v(S)

=))

-1 1
(X

when

S has))

elements.))

(extended)mputation is a vector)) i
l^l> ^2> ^8}
fulfill
>

whosethreecomponentsmust

(47:2)
and this is now
/j/7.O\\ yxt &)

^
the

-1,
d.\\ ~T~

now: in which (44:13)44.7.2., becomes ^ ^ be e , according to

s
c/0*

-1,

-1.

in Besides, E(e )

excess ust m

in (44:11*)44.7.2.

^J_xvJ_xv #2 I ^3

/>!

47.2.2. wish to representhese a by the graphicaldeviceof 32.1.2. We t But that procedure pictures only number triplets of sum zero. Therefore
we define
/ (47:4)
A*7

A\\

a1 = ai
1

- ^,

^0

2 a9 = a2

-^

^0

ft

= as

- ~-

Then (47:2),(47:3) become

(47:2*) a1 (47:3*)

a + a2 + a3 = O.2 Now the representation 32.1.2. becomespplicable, need only we of a 1 2 8 8 by a , a , a . With this reservation, Figure52 can be replace i, 2,
used.

Forthesereasons, e form for every vector a = w 2, 3} of I?(e) not only its componentsin the ordinary sensebut alsoits quasi-components in the sense (47:4): 1, a2, a3; and with the help of the quasi-components, of a o we utilize the graphicalrepresentationf Figure52. So this plane representation xpresses e preciselythe condition (47:3*). The remaining condition (47:2*)is therefore equivalent to a restriction imposedupon the point a , within the plane of Figure52. This restriction
difference

(i,

of with the sole reader should compare (47:l)-(47:3) (32:l)-(32:3) liesin (47:3). * Comparing these (47:2*),(47:3*)with (32:2),(32:3)of it appears that a (47:3*)and (32:3)coincide, nd that (47:2*)and (32:2)differ only by the factor of 1
-f)))

* The

32.1.1. 32.1.1.,

proportionality

THREE-PERSON GAME
obtains in the sameway as the similar one in

406
lie
M
within

the triangle formed by the threelines a1 =

- (l + |M,a = 2

32.1.2.:must a

+ 5lY

a8 =

1+ sr )

This is preciselythe situation of Fig.53., for except the

proportionality factor

1+ o and it |V

is represented Figure70. The on

).))

shaded areato be called fundamental triangle, representshe a which the t fulfill i.e.hose of E(e9 t (47:2*), 47:3*), (

a'-O))

70. We the 47.2.3. express relationship of domination in this graphical of representation.As we areusingthe new theory, the considerations
Figure

31.1.

for t concerninghe set 8 of 30.1.1.a domination a H ft i.e. oncerning c its certainlynecessaryor certainly unnecessarycharacterno longer apply. So we discussS de novo. It is still true,that 8 cannot bea one-or a three-element In the first set. caseS = (i), so by 30.1.1.^ v((t)) = on on > ft, hence ft < so ft by (47:2). In the secondcaseS = (1,2,3),by contradicting ^ 30.1.1. > 181, a > ft, as > ft, hence i + a,+ 8 > ft + ft + ft, ai a contradicting i + 2 + as = Pi + ft + ft = e by (47:3).

-1

-1,

-1,

1Cf If

.footnote

2 on p. 404. HereWe assume,of course,hat 1 -f =r ^ 0, i.e. t

o))

6 then 1+ ^ < 0, i.e. < -3- -|r|i, 5

know from

the

conditions in this

of (47:2*), 47:3)conflict, and (

case.)))

indeed we

(45: A)

that

E(e ) is empty
Q

OF AND COMPOSITION DECOMPOSITION GAMES 1 Thus S must be a two-elementset, S = (i, j). Then domination = 1and > ft, a, > ft, i.e.hat t meansthat a, + a, ^ v((z, j)) 406
*

S1 + ai < 1
_i_

2e
>

and

> /#, ' > 0>.

the By (47:3*) first conditionmay bewritten))

We

this: restate Domination

a
>

H |3

means that

eithera1 >
(47:5)))

1 ,
1

a2 > a8 > a8 >

and
and and

a3 ^
2

or or

a1 >

a2 > )32,

- (l - ^-)) a ^ -U -^ a ^ - - ^M/
1 (1
\\

47.3.The Six Cases the Discussion. Cases(I)-(III) of

thesepreparationswe can proceed discussthe solutions to E(e), for all values of e . It will be found convenient to distinguishsix cases. Of theseCase(I) Cases correspondsto (45:O:a), (II)-(IV)and one point of (V) to (45:0:b) (the normal zone),and Cases and (VI)(without that point) to (45:0:c) (V)
V of T for
Q

After 47.3.1.

Case e 47.3.2. (I): < -3. In this case1+ o < 0,so (47:2*), 47:3*) ( conflict and E(e) is empty (cf.footnote 1on p.405)so V must be empty too. Case (II):e = -3. In this case 1+ ^ = 0, so (47:2*),(47:3*) o 1 2 = ^ = 1, a imply a = a = a = 0,i.e. i = a2 =
Q

(all in

45.6.1.).

o))

as in the proof of

SoE(eo)is a one-element and V must be = E(e<>)by the same set, argument

in Thus the conditionsare very similar (31:0) in to thoseencountered an inessentialgame,cf. loc. it. c 1i, k a permutation of This differs from the corresponding at the end of each
line.)))

31.2.3.

j,

1,2,3.

(32:4)in

32.1.3. by only

the

extra

condition

THREE-PERSON GAME

407

Case(III):

-3< e

0.

Figure 70. Also

1+ ~ ^ 1 -~; so the extraconditions of (47:5)in

In this case 1+ ^

> 0, so we

can use

47.2.3. automatically are

fulfilled throughout the fundamental triangle. So (47:5) coincides (32:4)n 32.1.3.footnote 2 on p.406). Consewith i (cf. quently the entire discussionof 32.1.3.-32.2.3. applies again, if the proe portionality factor 1 + ~ is inserted. o Thus we obtain the solutionsof (e ) in this case simplyby taking those in described 32.2.3., each component by 1+ o and adding multiplying
>

- (to o

pass from

a* to

at )47.4.Case(IV) :First Part this case

47.4.1. (IV): < e < ? In Case

&

< 1 ?O < 1+ ^O
f
<>\\

Consequentlythe lines

ai = ~ (\\ 1
V

- 3)'a - - /i1- 2e a = - (^l - 2e 3)' ( 17

2en \\
2
\\

t (which bound the extraconditionsof (47:5))are situated with respecto the fundamental triangle of Figure70 as indicated on Figure71. They e subdividethe fundamental triangle into seven areas, achof which can be which two-element setsS areeffective in it in the characterizedy stating b senseof (47:5). The list is given below Figure71. Now we can draw of the analogue Figure54, indicating for each point of the fundamental 1 triangle the shaded areas which it dominates. This is done in Figure72 to according (47:5). It is necessaryto treateachoneof the seven areasof Figure71separately,and every shadedareaof Figure72 must be continued t acrosshe entirefundamental triangle. It is clearfrom Figure72, that no point of the area can bedominated 2 of by a point outside that area. Hencethe condition (44:E:c) 44.7.3., f the which characterizes solution V for (e ), i.e.or theentirefundamental triangle,must also hold for the part of V in when taken for (in place i < of the entirefundamental triangle, .e.E(e )). But is a triangle like the f fundamental triangle of Figure53, except or the proportionality factor in Figure72 shows that the 1 3 8 Comparisonof Figure54 with ^-

conditionsof domination arethe same.

1

1Including

Excluding

their boundaries. its boundary.

Note that 1

>

0.)))

408

OF AND COMPOSITION DECOMPOSITION GAMES

1.3. 2.3. 47*4.2.Consequently the entire discussion of 32. -32. applies 2eo to the part of V in , if the proportionalityfactor 1 5- is inserted.))

'))

\\

\\))

Figure

71.))

Area:))

Effective two-element

sets-S:))

(1,2), (1,3), (2,3) (1,2), (1,3) (2,3) (1,2),)) (1,3), (2,3)

(2,3))) (1,3))) (1,2)))

/T^nty

,-.-.
Figure

\\

^..-.... \\
\\
SA.))

^6)

/
72.))

the Hence part of V in must be eitherthe set or the set)) can be in any indicatedin Figure73. (The line

position)))

GAME)) THREE-PERSON

409))
all))

to must be subjected belowthe points .) However i t permutations of 1,2,3, .e.o rotations of the triangle by 60, 120 to produce , all solutions. (Cf. 32.2.3., is (32:B), is (32:A) there.)

0,

Having found the part of V in , we to proceed determine the remainder of V. SinceV is a solution, this remainder must liein the areawhich is undominated by the Comparisonof Figure73 part of V in with Figure 72 showsthat this undominated Figure 73. areais the following one: it consistsof the three ^ trianglesof Fig.74, for the For the set it consistsof the three &, trianglesof Figure75. set It is clearfrom Figure72, that no point in any one of thesetriangles 2 the can be dominatedby a point in another one. Hence condition (44:E:c) for the entire the which characterizes solution V for E(e), of 44.7.3., taken for)) fundamental triangle and which holds for the part of V in

/ /

\\\\))
l))

i))

i.e.

\\
i.e. :

\\))

Figure 75. Figure 74. of E(e9)) too states preCD (in place the entire fundamental triangle, taken)) h ( cisely this: 44:E:c) olds for the part of V in each triangle for that triangle. 47.5.Case(IV) SecondPart Let 47.5.1. us therefore take one of those triangles,denotingit by T. Itspositionin the fundamental triangle, 8 and the shaded areasdominated 1The position of all these triangles are clearly indicated by the drawings, exceptfor the lower triangle in Figure 75. This triangle lies certainly outside the inner triangle cf.also (area ) this is equivalent to the restriction (32:8)in 32.2.2., Figure 60there. definite It may shrink Its position with respectto the outer (fundamental) triangle is less to a point or even disappear altogether. It is not difficult to see that the latter phenomenon is excluded,unless the (linear) sizeof the inner triangle is ^ of the outer one this means 1 ~ * 4 ^* to We do not propose discussthis subjectfurther. e i.e. 1All this refersto Figure 74, or all to Figure 75 but, of course, ever to both in the n

1.

\"*\"

3/

same argument! *Up to a rotation by doesnot lie on the inner alter out discussion.)))

triangle, but below it,

75 0,60,or 120.For the lower triangle of Figure this the apex doesnot (cf. footnote 1 above) but

OF AND 410 COMPOSITION DECOMPOSITION GAMES

by a given point in it (taken over from Fig.72) are shown on Figure76. of o We may now restrict urselvesto this triangle T, and to the concept the domination which is valid in it and determine solution of (44:E:c) with respecto this. We redraw T and the setup in it separately, also t 77.) x introducinga system of coordinates , y in it. (Figure Note that the apexo is undominated by points of T henceit must
to belong V.1'
2))

-LiM*))

Figure
Point
o))

77.))

Line/))

78. 47.5.2. considerwo points of V in T at different heights y. In Now t orderhat the upper one shouldnot dominatethe lower one,the lattermust t not liein the two shaded sextants,belongingto the former, the lower point must be in the middle sextantbelow the upper one, and vice versa. Thus, if a point of V in T is given, then all points of V in T at different heightsy must lie in one of the two sextants indicatedin Figure 78. mi
Figure

i.e.

triangles T) than the lower one of Figure 75, this follows from too: As Figures 74, 75 show, the apexof such a triangle lieson the consideration, borderof the inner triangle (area ) and belongs to what we know to be the part ol V in 1 When the lower triangle t T) of Figure 75 degenerateso one point (cf. footnote 1on p. 409)which is, of course, o then this determines the part of V in another

1For other

J^ (i.e.

^ (i.e.

T.)))

THREE-PERSON GAME))

411))

Now 47.6.3. assumethat a y\\ is the height of more than onepoint of V. Let then p and q be two different pointsof V with this height yi (Figure 79). Now choosea point r in the interior of the triangle Comparison of Figure79 with Figure77 showsthat this r dominatesboth p and q. As theremust exista point * p, q belongto V t r cannot belong to it. Hence in V which dominates r. Now a secondcomparisonof Figure79 with Figure77 showsthat a point which dominatesr must also dominateeither p or q. Sinces, p, q all belongto V this is a contradiction.))

Point*))

-))

-Line/))

Figure

79.))

Point

o))

-Line*))

Figure

80.

the apexo) is the height of no point of V. Thereexistcertainlypoints of a the V with heights y ^ t/i, e.g. apexo is such a point. Choose point p of V with a height y ^ y\\ as low as possible,i.e. ith its y minimum. 1 w this (Figure80.) Denote minimum value with y = i/ 2. Clearly yi < y* the definition of y^ no point of V has a height y with yi ^ y < y* and By by the above p is the only point of V with a height y = y*. henceit is dominated by an
on Now project perpendicularly y p
8 in 1Thisis possible set. Cf (*) of footnote 1 on sinceV is a closed

47.5.4. assumethat Next

y\\

(in triangle T,

i.e. etweenthe baseI and b

= yi, obtainingq. q cannot bein V Hence s cannot liebelow q, i.e. this V.

p.384.)))

AND OF 412 COMPOSITION DECOMPOSITION GAMES))

its height y : y\\. Consequentlyy J j/i. Comparisonof Figure80with * Figure77 showsthat p doesnot dominate q. Hence 7* p, necessitating a y T* yi. Thus y > yi, i.e. lies (definitely) above p. Now a second comparisonof Figure80with Figure77 shows that if a point a above p dominates q, then it must also dominate p. Sinces,p both belongto V, this is a contradiction. 47.5.5. Every y (betweenI and o) is the height of precisely Summing up: w onepoint of V. If y varies, then this point changes ithin the restrictions))
Point
o))

CurveV))

-Line/))

Figure V: The points

81.))

V: The line
and the curvet
,/))

withe curvet /))

Figure

82.))

Figure

83.))

of

w Fig.78.,i.e. ithout

leaving the sextantsHH indicated there, In other))

words:))

(47:6)

of V (in T) is a curve from o to i, the direction which never deviates from the vertical by more than 3001 (cf.Figure81).

to Conversely, if any curve according (47:6)is given then comparison of Figure81and Figure77 makes it clearthat the areasdominated by the points of V sweepout preciselythe complementof V in T. So (47:6) 2 is the exact eterminationof the part of V in T. d We can now obtain the general olution V for E(e) (i.e. the fundas for mental triangle) y insertingcurvesaccording Figure into eachtriangle to 81 b
Q

1Henceit is continuous. 1 It is equally true when T degenerates a point, cf to

.footnote

1 on page

409.)))

of

1 respectively. It will be observedthat thesefigures show still marked similarity with thosepertaining to the solutionsof the essential hree-person amein the t g old theory (cf. 32.2.3., in the inner triangle of Figure73). Thenew shown element onsistsof the curves in the smalltriangles, ll of which aresituated c a in the fringe between the two major triangles Figures and 83. The 82 of width of this fringe, as shown in Figure71,et sequ., s measuredby e .2 i So when e tends to zero, our new solutionstend to the old ones. It is also worth pointing out that the variety of the solutionsis much now E greater than ever before: ntire curves can be chosenfreely (within

413 Figures74 and 75. The results are shown on Figures82 and 83,
THREE-PERSON GAME

the limitations of (47:6)above).

We will see later, that these curves motivate an interpretationwhich is of further significance. (Cf. 47.8.) 47.6.Case (V)

47.6.1. (V): g e < 3. Case

2^n

8 as 5- J < 1 + q\"- Theseinequalitiesexpress, is easily verified, that the orientation of the inner triangle of Figure is inverted,but that it is 71 still situated entirely within the outer (fundamental) triangle,as indicated 84. Thelatteris again subdividedinto sevenareas, achof which e on Figure can be characterizedy stating which two elementsetsare effective in it b in the senseof (47:5)in 47.2.3. he only difference between the present T situationand that onein Case (IV) (i.e. Figure71)is the behavior of area The list is given belowFigure84. 54 Now we can draw the analogue of Figures and 72,indicatingfor each 4 point of the fundamental triangle the shaded areas which it dominates. t This is done in Figure85,accordingo (47:5). It is clearfrom Figure85 that no point of the area 5 is dominatedby 6 V any point. Hence must contain all of 1 The lower to altotriangle of Figure 83 may degenerate a point or even disappear gether, cf. footnote 1 on p. 409. * Thesidesof the outer (fundamental) triangle are given by = f 1 + -^\\ those

|
P

In this case1

-^ g

< 1+ ~ and

of the and

inner triangle by
f
.))

'

- - f 1 - -jfj(Cf.

Fig.

71). Thedifference

of

- ( I -f
*

~-))

~ ) is c 1 * This latter inequality is equivalent to e

4

< 3.

Excluding 6 Including

a in E(CQ). It is easyto show that they are dominated by no a at all :D)in 45.2.4. they are the detached imputations, by (45 The points of the interior of the area dominate no other points either. I.e. they dominate no a in #(e ). Again, it is easyto show that they dominate no a at all :C) they are the fully detached imputations, cf. (45 in 45.2.4. Thesestatements can alsobe verified directly, by using the definitions of 45.2.)))
8

I.e.y any b

'

their boundaries.)) its boundary.

414 COMPOSITION DECOMPOSITION GAMES)) AND OF

--ft1^))
AD))

Figure

84.))

Area:))

Effective two-element

setsS:))

( (1,2), 1,8)
(1,2), (1,3)
(1,2)))

(2, 3) (2, 3)

(1,3)

(2,3)

Figure

85.)))

THREE-PERSON GAME))

415))

47.6.2. to Having found the part of V in , we proceed determinethe remainderof V. SinceV is a solution,this remaindermust lie in the area which is undominated by the already known part of V f i.e.y . Conb siderationof Figure85 showsthat this undominated areaconsistsprecisely of the threetriangles@, , 1 It is clearfrom Figure85 that no point in any of the three triangles can be dominated by a point in another one. Hencethe argument of 47.4.2. shows, that our requirement of V must be precisely this:

0.

taken for that triangle (in place of

the entire fundamental triangle,

of must (44:E:c)44.7.3.

hold for the part of V in eachone of thesetriangles,

i.e. v. Thethe
and

triangle
curves))

0, in

The conditionsin the triangles

the Hence entiredeductionof 47.5.1.47.5.4. be repeated may literally, and

Figures76, 77 for the triangle T.

arethe sameas thosedescribed

0,

the parts of V in are curves as shown on Figure81, haracterized c Figure 86.)) in by (47:6) 47.5.5.)) We can now obtain the general solution V for E(e^) (i.e. the fundafor mental triangle) by insertingsuch curves into , in Figure 85. The result is shown in Figure86. Forfurther remarks concerninghesesolut tions cf. 47.8., 47.9. 47.7.Case(VI)))

0,0,

Y))
o))

0,

47.7.e ^
Q

3. In this case))
t

<0 < 1

+))

and))

+^

-2

asis These inequalities express, easilyverified,

84 that the inner triangle of Figure has still the same orientation, but that a of the outer (fundamental) triangle, nd possibly it reaches boundaries the The only differencebetweenthe present as beyond,* indicatedon Figure 87. of situation and that one in Case(V) (i.e. Figure84) is the disappearance The list is given belowFigure87. the areas@, , The analogue of Figures54, 72 and 85 indicatingthe domination rela-

tions, is containedin Figure88. can The argument of 47.6.1. be repeated literally, proving that V conleaves no part tains all of . Considerationof Figure88 showsthat 4 V of the fundamental triangle undominated. Hence is precisely . For this further remarks concerning solution,cf. 47.9.
1The remainder
belongs to
8

which

* This last inequality

When

of the fundamental

triangle
<

is

dominated

by the boundary

of

4 Theremainder of the fundamental

> 3.

is equivalent to e

3.
is dominated
by the boundary

triangle

of

which

belongsto

.)))

AND 416 COMPOSITION DECOMPOSITION GAMES)) OF

47.8.Interpretation of the Result:

The Curves (OneDimensional Parts) in the Solution 47.8.1. solutionsobtained in the discussionsof 47.2.-47.7. The deserve a brief interpretative analysis. It is quite conspicuousthat the repeated))

v^_

,a'

- -II

\\))

IL))

-))

' - -())
Figure
87.))

Area:))

Effective two-element

setsS:))

(2,))

(1,3))) (1,2)))

Figure

88.
in)))

o appearance f a smallnumber of qualitative features goesfar in characterizing their structures insofar as they deviate from the types familiar

THREE-PERSON GAME

417

the solutionsof the essentialthree-personame of the old theory. These g features are:Thecurves arbitrary within the restriction(47:6) f 47.5.5. o which occuras soonas e > (and as long as e < 3); and the two-dimensional areas,which appear when e > We will now undertake their

interpretation.

f.

Considerfirst Case(IV): < e < f (in the \"normal\" zone). Let us considerthosesolutionsof the presentcase the which extend non-discriminain Such and tory solution of the old theory (cf. 33.1.3. (32:B) 32.2.3.). a solution is pictured on Figure82. This figure showsthe threepoints which form the analogue of a solution in the old theory. Taking, e.g.he lower point , one verifies easily t
Q

that

there))

l \" i

i.e.

- - 1- -57 =
V Oil

\" i . 1

+
1

T
o

, a2 = a3 = 2 1 T V
2

Q,

= i-

Thus thesethree points expressn arrangement where two players have a formed a coalition, obtainedits total proceeds a (amounting to 1),nd divided them evenly but the defeatedplayer has not been reduced his minimum to value because retained beyondthat the total available excessQ. he C Now the curves,starting from thesepoints (in the fringe betweenthe two triangles), xpress situation where the total excess is not left the e e in the indisputedpossession the defeatedplayer. By claiming any part of of the excess,he victorious coalition exactsmore than the amount 1 t which it can actually get in the game i.e.t ceases be effective. (Cf. i to in the Figures 71 and 72.) Therefore the conduct the areas(2), of affairs of this coalition the distribution of the spoils within it is no longer determinedby the realitiesof the game i.e. y the threats between b the partners but by the standard of behavior. This is expressed the by curve, which is part of the solution. The possiblethreats between the t partners still restricthis curve to a certain extent (cf.(47:6)in 47.5.5.),

1,

0,

but

that beyond that it is highly arbitrary. It must be re-emphasized this arbitrarinessis just an expression the multiplicity of stable standards of of behavior but a definite standard of behavior, i.e.olution,meansa definite s r curve, i.e.ule of conductin this situation. 47.8.2. considerationssuggestthe following tentative interpretaThese

tion:
(47:A)

In the presence a positive excesst may happen that a of i coalition can obtain beyond its effective maximum also some fraction of the excess.This possibility is then due entirely to the standard of behavior and not to the physicalpossibilities of the game. The fraction of the excess obtainedmay vary thus from 0% to 100% be left undeterminedby the standard of and behavior. The latter will prescribe, however, uniquely, how)))

AND OF 418 COMPOSITION DECOMPOSITION GAMES

the fraction obtainedis to be distributedbetweenthe members of the coalition. This rule of division will depend on which o* the many possiblestable standards of behavior is chosen, nd a
if

the latteris varied, this rule quite unrestrictedly.

will

vary widely, although not

t interpretation seemso fit them in every case. o betweenthe victorious Theindefiniteness f the distributionof the excess and the defeated player (in a given solution) is an instancehow coalition certainsocial djustments may be left open even within a specifiedsocial a the order. Our curves express further nuance that while such an indefinite is decided distribution upon, some players can be tied to each other by definite conventions. (We will seefurther instances of this in the third 67.3.3. in 62.6.2.) and remark of 67.2.3,

to We have seenalready, that undeterminedcurves according (47:6) in the future. Theabove occurin many solutions,and they will occur again

47.9.Continuation

:The Areas (Two-dimensional Parts)

47.8.2. 83.

in the Solution

t instructive view-points, particularly with respecto the curve in the lower of Fig.83. However,we refrain from elaborating this caseany triangle further. We turn, instead, to the Cases (V) and (VI), specifically when e > J in the \"toolarge\"excesses the senseof 44.6.1., 45.2).These (theseare that their solutions contain the circumstance, casesare characterizedy b areas. Actually, two different situations may arise: two-dimensional < e < A solution V contains the two-dimen(a) Case(V), sionalarea but besidesalsocurvesas discussedin 47.8. Fig.86). (cf. The unique solution V is the two-dimen(b) Case(VI), e ^ sional area , and nothing else(cf. Fig.88). a o The emergencef two-dimensional reaswithin the solution indicates that the standard of behavior fails to contain rules of distribution at least

couldbe tested by applying The interpretation (47:A) in solution of the old theory (cf. o it to the extensionf the discriminatory as pictured on Figure This would bring up some and (32:A) in

47.9.1.

32.2.3.)

33. 1.3.

0,

i.e. i.e.

|
Q

3.

3.

within certainlimits. In the Cases (b) theselimits arespecified. In (a), t the case(a) the curves of 47.8. ppear outside of those limits, i.e.he a standard of behavior still sanctionscertaincoalitions in the Case(b) this is no longerthe case. e 47.9.2. we see that the \"disorganizing\" ffect of a \"too large So manifests itself in two excess i.e.ift from an outside source(cf. 44.6.1.) g I successive stages:n the Case(a) it is present in a certaincentralarea,but doesnot excludecertain conventional coalitions. In the Case(b) the standard of behavior no longerallows coalitionsbut it setscertainlimiting
11

principlesfor the distribution.

.)))

GAME THREE-PERSON
at 60 =
We have

419

Theseconsiderationsseem to be quite instructive in a qualitative way for the possibilities standards of behavior and organizations.It appears of in likely that they will provide useful guidance the further development
of the theory. But the

1,

1 3 respectively.

seenthat.,theseuccessive s are stagesof disorganization reached

reachingconclusionsfrom the quantitative results:They apply to the 2 which is thus shown to be the three-personame with an excess, g simplest model for their operation. But it must have become amply clearby
now that an

readermust be cautioned against drawing

far

increase in

the number of participants will affect conditions

fundamentally.
1Note that 2 Hence alsoto a

|r|i f .

decomposition six-persongame

in the

old theory, cf.46.12.)))

CHAPTER X

SIMPLEGAMES
W 48.Winning and LosingCoalitionsand Games here They Occur 48.1. TheSecondType of 41.1. Decision Coalitions by

theory in the senseof 48.1.2. As was brought Considerthe corner7 of Q, discussedin this game has the followingconspicuous eature: aim of the The out in f eitherof player4 and oneally, playersis to form certaincoalitionsconsisting orof all threeotherplayerstogether. Any one of thesecoalitionsis winning in the full senseof the word. Any coalition which falls short of theseis defeated.I.e.he quantitative element,he paymentsexpressed t t completely as by the characteristic function, can be treated somethingsecondary the in primary aim in this game is to succeed forming certaindecisivecoalitions. This description suggestsstrongly that the number four of players of a and the particular scheme decisivecoalitionsare specialnd accidental from this particular and that a moregeneralprinciple can be extracted

48.1.1. program formulated in 34.1. The provided for far-reaching of of to generalizations the gamescorresponding the 8 corners the cubeQ, introducedin 34.2.2. Thecorner77/7 (alsorepresentativeof 77,777,77) was taken up in 35.2.1. providedthe source a generalization, and leadfor to the theory of composition to and decomposition which all of Chapter ing IX was devoted. We now pass to the corner7 (alsorepresentativef 7, o which we will treat in a similarfashion. 77, 777), i By generalizing the principle, of which a specialnstancecan be discernedin this game,we will arrive at an extensiveclass of games,to be called simple. It will be seenthat a study of this class yields a body of information which is of value for a deeper nderstanding of the general u 34.1. 35.1.1., 35.1.

arrangement. In 48.1.3.carrying out this generalization, following observationis the the useful. In our above example, decisivecoalitions the attainment of aim of the players were these: which is the sole

(48:1)

(1,4), (2,4),(3,4),(1,2,3).
but coalitions, also

Now it is convenient to view not only theseas winning

all their (proper) supersets:

(1,2,4), (1,3,4), (2,3,4),(1,2,3,4). (48:2) The point isthat although the coalitions(48:2)contain participants whose is presence not necessaryin orderto win, the coalitionis neverthelessa
420)))

WINNING AND LOSINGCOALITIONS

winning

421

1 opponents are defeated. These opponents form thosecoalitionswhich arethe complements the seta in (48:1), i.e. of (48:2),

one

i.e.the

the sets
(48 3) ( }

(2,3),(1,3), (4). (1,2), (3), (2), (i), eThus (48:1), a (48:2),contain the winning coalitions, nd (48:3)contains the defeated ones. It is easily verified that every subset of 7 = (1,2,3,4) to prebelongs 2 ciselyoneof thesetwo classes: (48:2), (48:1), or (48:3).
48.2.Winning and Losing Coalitions 48.2.1. us now considera set of n players: 7 = (1, , n). The Let scheme 48.1.3. of to generalizes subdividingthe system of all subsets of I into two classes and L, such that the subsets of W will representthe W coalitionsand the subsetsof L will represent losingones. The the winning o can analogues f the propertiesformed in 48.1.3. be formulated as follows: Denotethe system of all subsets of / by 7.8 The mapping of every subset S of / on its complement(in /) :

(48:4)

S-+-S

is clearlya one-to-one appingof I on itself. Now we have: m :A :a) (48 Every coalition is eitherwinning or defeatedand not both i.e. and L arecomplementarysetsin 7. W a (48:A:b) Complementation(in 7) carries winning coalition into a the losingoneand viceversa i.e. mapping (48:4)maps W and L on eachother. A coalition is winning, W if part of it is winning i.e. con(48:A:c) tains all supersets its elements. of A coalition is losing,if it is part of one which is losing i.e. (48:A :d) L contains all subsets of its elements. 48.2.2. we discuss the conceptsof winning and losing in their Before relationship to the game, we may analyze the structure of conditions somewhatfurther. (48:A:a)-(48:A:d) W f Thefirst conspicuousact is that, although we needboth classes andL d to interpretthe game,theseclassesetermineeachother. Indeed they do c a in two ways: Given one of W or L (48:A:a) s well as (48:A:b) an be this other. I.e. tarting from one of thesesets,the other t used to construct he s oneis obtained in this way: the A to According (48: :a):Take given setas a whole and form its complement (in
1 of the complements are flat in the sense * (1,2,3,4) 24 16subsets, f these8 are in o has
As

I.e.

J).

(48:3). 1

Cf. 31.1.4. the discussion

a (48:1), (48:2), nd

the remaining 8 in

in

35.1.1.

/ has n elements, 7 has 2 elements.)))

422

According to given set separately, it and replace by its complement(in 7).1 the It shouldbe notedalso that if the given set,W or L, possesses propor (48:A:d) espectively,then the other set obtained from r erty (48:A:c) the former by (48:A:a)or by (48:A:b)will possess the other property

SIMPLE GAMES o T (48:A:b): akeeachelement f the

It follows from the above, th&t we can basethe entirestructure now on under consideration eitheroneof the two setsW and L. We must only l a requirethat both transformation (48:A:a) nd (48:A:b)ead from it to the same set (which is then the otherone of W and L) and that it must satisfy a the pertinent one of the two conditions(48:A:c) nd (48:A:d) other (the a is condition of (48:A:c) nd (48:A:d) then automatically taken careof,
Thus we have only two conditionsfor W or L : Firstthe equivalenceof a a o (48:A:a) nd (48:A:b) nd second(48:A:c) r (48:A:d). of The former condition means this: henon-elements the set coincide T o with the complements (in 7) of the elements f the set. In other words:Of two complements 7) S, S,one and only onebelongsto the set. (in
to according
what we have

2 o (48:A:c) r (48:A:d).

seen).

ThesetsTF(
(48:W)

Summing

up:

7)

arecharacterizedy theseproperties: b

(48:W:a)

belongsto L. (48:L:b) L containsall subsets of its elements.

(48:W:b) The setsL( I) arecharacterizedy theseproperties: b (48:L) Of two complements (in 7) S, S, one and only one (48:L:a)

Of two complements(in 7) S, S, one and only one belongsto W. W containsall supersets f its elements. o

1 Thereader will note the remarkable structure of this condition The given set must produce the same result, irrespectively of whether complementation is applied to it as a unit, or to its elements separately. 1This is actually true for (48:A:a) s well as for (48:A:b), nd independent of the a a a question whether (48:A:a)nd (48:A:b) roduce the sameset. Precisely: p then Leta set M possess property (48:A:c)(48:A:d)], both setswhich the (48:B) [ are obtained from it by (48:A:a) nd (48:A:b) we do not assume that they a are identical possess other property (48:A:d) (48:A:c)]. the [

a c Proof:We must show that both transformations (48:A:a)nd (48:A:b)arry (48:A:c) a (48:A:d) nd vice versa. i Clearly (48:A:c)s equivalent to this: If S is in M, and T is not, then SS T is excluded. (48:A:c*) i Again (48:A:d)s equivalent to this: If S is not in M, and T is, then S T is excluded. (48:A:d*) Now the transformation (48:A:a)nterchanges \"being in M\" and \"not being in M \" with i T each other. Hence it interchanges (48:A:c*)and (48:A:d*). he transformation and (this is brought about by individual complementation for i (48:A:b)nterchanges the elements S, T\\ besidesthe symbols S, T must be interchanged). Henceit, too, a interchanges (48:A:c*)nd (48:A:d*).))) into

CHARACTERIZATION THESIMPLE OF GAMES

We

423

restate:

If W [L] fulfills (48:W) [(48:L)],then (48:A:a) nd (48:A:b) ield the a y :A samesetL [W]. W andL fulfill (48:A :a)-(48 :d)and L [W] fulfills (48:L) then [(48:W)]. Conversely,if W, L fulfill (48:A:a)-(48:A:d), they fulfill separately (48:W), (48:L). 49.Characterization of the SimpleGames 49.1. Conceptsf Winning and Losing Coalitions General o We now pass to the considerationof the connection between in winning and losingcoalitions the game itself. Assume, therefore, that an n-persongame F is given. In all the considerationswhich follow, it is advantageousto restrict urselvesto the old o or 42.4.1. theory in the senseof Consequently,as pointed out in we 42.5.3.,may assumeF to be zero or constant-sumas we desire. Forthe F present we prefer to choose as a zero-sum game. a Beyond this F is not restrictednd in particular no normalization is

49.1.1.

30.1.1.

assumed.

1 we may argueas follows: The playert, essentiallywhat was said in 35.1.1., when left to himself, obtains the amount v((i)). This is manifestly the worst thing that can ever happen to him, sincehe can protecthimself w against further losses ithout anyone else's help. Thus we may consider the player i when he getsthis amount v((i)) to be completelydefeated.A coalition S may be consideredas defeated,if it getsthe amount ] v((t)), i in S 2 Thus the criterion sincethen eachplayeri in it must necessarily getv((i)). of defeat is

49.1.2. us first analyze the concept f a losingcoalition. Repeating Let o

v(S) =

v tinS

<)-

In the terminology of 31.1.4., means that the coalition S is flat. (Cf. this also footnote 3 on page 296.) We have obtained a satisfactorydefinition of the systemLr 8 of all losing (defeated)coalitions: Lr is the set of all flat setsS(&/). (49:L) It is now easy to say what a winning coalition It is plausiblyone, is. the opponentsof which arelosing,i.e. system Wr* of all winning coalithe tions is this:
1The difference is that our present F is more general. 1Since player t needever acceptlessthan v((0), nd those in no a together ^/ v ((*')) this is the only way in which they can split.))
the coalition

S have

48.2.2. difference confusion, The between

discussion of and \"losing\" from a specificgame T. The two viewpoints will

1In orderto avoid

we will use the symbols Wr, Lr instead of the W, L of is this and the former is that 48.2.2. a postulational of the properties which appeareddesirable for the concepts \"winning\" (described W, L) while we are now analyzing definite sets obtained by

be merged

in

(49:E)of 49.3.3.)))

equivalence.

SIMPLE GAMES W r is the set of all setsS(z for which -Sis flat. (49:W) /) It should be conceptuallyclear,and is also immediately verified with the help of 27. .1. that the sets Wr, I/r are invariant under strategic 1 -2.,
424
Wr, Lr to fulfill the conditions of 48.2.1. game in its presentgenerality The (48:A:a)-(48:A:d) W, L) (for neednot beof the simpletype referredto,where the only aim of all players is to form certaindecisivecoalitionsand there areno othermotives which 1 require a quantitative description. It will therefore be necessary to restrictin orderto expresshe property we have in mind. Indeed,he t t formulation of this restriction our immediateobjective. is precise Nevertheless,we beginby determininghow much of (48:A:a)-(48:A:d) holdstrue for the T in its present enerality. We give the answerin several g

49.1.3. cannot We

t expect he above

stages.
(49:A)

W r , Lr always

fulfill

:A (48:A :b)-(48 :d)

But (48:A:d) oincidesith (31:D:c) in 31.1.4., w c considering(49:L). Thus the main difference between our presentWr, Lr and the setup of 48.2. in (48:A:a) i.e.n the question whether or not Wr and Lr are lies i this into two parts: We can decompose assertion complements.

in Ad (48:A:c), 48:A:d): we have (48:A:b), e can apply (48:B) w Since ( 4S.2.2. therefore (48:A:c)nd (48:A:d)mply eachother., and a i
8

2 49.1.2.

Immediatey comparing(49:L) and (49:W) Proof:Ad (48:A:b): b

in

(49:1) (49:1 :a) (49:1 :b)

(49:B)

TT r

nLr = 0,4 JF r uLr = /.6

leadsback to familiar concepts: (49:l:a)

(49:B:a) (49:B:b)

holds (49:l:a) if

If T is inessential, then W r

and only

if

T is essential. = L T = /.6

1Actually the conceptof \" winning \" was basedon the conceptof \"losing\" by just this operation of complementation. 1It appearsnow why we separated(48:A:a)rom (48:A:b)n i We have now f b (48:A:b),ut not (48:A:a). 4 It may seemodd that this no coalition can at the same time be both winning and :B) losing must bestated separately. The meaning of this condition will appearin (49 and footnote is definitely winning or This states that every coalition every subset of T t losing. This is, of course,he ideaon the basisof which we wish to specialize . Thus a coalition can at the same time be both winning and losing, when the game is in inessential manifestly because this caseboth statesare irrelevant.)))

1Our discussion of the four-person game has provided many illustrations of such motives, for which the end of provides a good instance. This situation is, indeed, the usual (general) one the classof games at which we are aiming now, is in a certain sensean extreme case,cf. the concluding observation of 49.3.3.

36.1.2.

48.2.1:

6.

i.e.

CHARACTERIZATION THESIMPLE OF GAMES such that both S and

425

(31:E:b)in31.1.4. Ad (49:B :Wr = Z/r = / means that every Sin 7 is flat. :b) to inessentialityby (31:E:c) in 31.1.4.
which

A is The of Proof: d (49:B:a): negation of (49:l:a) the existence an S S are flat. This amounts to inessentialityby

Thisamounts

Before passing to (49:l:b) note that Wr, I/r possessone property we did not occurin (48:A:a)-(48:A:d).

(49:C)

sets. Lr contains the empty setand all one-element 1

w below. Thus our plausiblediscussion (48:A:d); e will verify this in 49.2. a of 48.2. overlooked necessaryfeature of the Wr, Z/r- We must, therefore, make sure that the present conditionscontain everything. I.e.hat the t conditions(48:A:b)-(48:A:d) (49:C),ogether ith the resultsof (49:B) and w t on inessentiality,characterize Wr, LT completely. This will be shown the
in

T with in 31.1.4. Proof: his coincides (31:D:a), (31:D:b) is really a new condition, i.e. is not a consequence (48:A:a)it of (49:C)

below. 49.3.

49.2.The Special ole of One-element ets S R 49.2.1. begin with the example announced above:Two systems We 2 but L which fulfill (48:A:a)-(48:A:d), not (49:C). Actually, we can TF, determineall such pairs. but i (49:D) W, L fulfill (48:A:a)-(48:A:d), not (49:C), f and only if they have the following form :W is the setof all S containing i , L is the setof all S not containing i , where IQ is an arbitrary but definite player. Proof: ufficiency: It is immediately verified that the W, L formed as S indicated fulfill (48:A:a)-(48:A:d). :C) is violated, sincethe one ele(49 ment set (f ) belongsto TF, and not to L. but Assume that W, L fulfill (48:A:a)-(48:A:d), not (49:C). Necessity: 3 Let (to) be a one-element which does not belong to L. Then (t' ) set, to W. belongs Every S containing t' has $2 (i ), henceit belongsto W by (48:A:c). If S does not contain i , then S containsit;hence S belongsto W by the above and S belongsto L by (48:A:b). henceW is preciselythe set of the Finally W, L aredisjunctby (48:A:a), the S containing andL is precisely setof the S not containing t' . 49.2.2. may be worth while to comment briefly upon this result. It
Q t*

1 It is clearly in the spirit of our entire analysis of games that a coalition of one player in a is to beconsideredsdefeated as this player has not succeeded finding partners for a coalition.

2 We mentioned originally (48:A:b) (48:A:d)only, but the above strengthening requires no extra effort. 8 If the Aid) empty setdoesnot belong to L, then no set can belong to L owing to (48:

henceany (to) will

do.)))

426

SIMPLE GAMES

The W, L formed in (49:D) cannot be the Wr, Lr of any game since odd, since(49:D)appearsto convey a they violate (49:C). This may seem d b very clearidea of the kind of \"winning\" and \"losing\" escribedy its the t W, L. Indeed,hey describe situation where a coalition wins if the i not. Why can no game be conplayer io belongsto it, and losesf he does structedto fit this specification? would The reasonis that under the conditionsdescribed, \"winning\" 1 The player is \"victorious\" not bea matter of forming coalitionsat without anybody else's help. Still worse,in our terminology this position 2 of to is no victory it is not the result of any strategicoperation, but a 8 fixed stategiven him by the rules of the game. A game in which coalitions 4 involve no advantage is inessential, even if one player should have a fixed considerable advantage in it. w t The reader ill understand, of course,hat all this is just an additional comment on results which were already rigorously established above (in

all:

(49:D)). (49:0),

the secondsubjectmentioned at the end of Let 49.1.3. two systems W, L( 7) be given, which fulfill the conditions a and a :a). (48:A:b)-(48:A:d) (49:C),nd also (49:1 6 We wish to construct n essentialgame F with Wr = W, Lr = L. In doing this, we normalize F with 7 = 1. The setsS in Lr arecharacterizedy their flatness, i.e. y v(S) = p, b b o 6 where p is the number of elements f S. The setsS in Wr arecharacterized by the fact that S belongsto Lr , i.e. y v( S) = (n p), owing b tothe above. Now v( S) = v(/S), hencewe may write for this v(S) = n p.
turn to

49.3.1. now We

49.3.Characterization

of the Systems W, L of

Actual

Games

Lr = L areequivalent to this: Fora g-element S, (q = 0,1, set ,n 1,n) (49:2) =n q (49:2:a) v(5)
The desiredrelations WT =
W,

Hence have shown: we

if and only if

(49:2:b)

if and only if 1The equivalent consideration was carried out in a special asein c 1We always considerhis to be the same thing as forming appropriate coalitions. t * Cf. our treatment of the basicvalues a', c'in the three-personame, in 22.3.4. g The entire discussion of strategic equivalence,cf. was made in the same spirit: advantages like this one can be removed by strategically equivalent transformations, while those which are really due to forming coalitions, cannot. 4 Hence WT, Lr are not the desired its but in ones,described (49:D), those of (49:B:b). * We require (49:l:a) we because aim primarily at essentialgames (cf.(49:B)). Sub-

v(S) = -q S belongsto L.

S belongsto W,

and

35.1.4.

6',

27.1.1.,

sequently

'

Recallthat

we will make our discussion exhaustive -1.))) aU y

v() -

as will beseenin (49:E).

CHARACTERIZATIONOF THESIMPLE GAMES

427

characteristic function v(S) which fulfills (49:2). 49.3.2. determinesv(S) for the S of W and L, so we needonly (49:2) define it for thoseS which belongto neitherset. We try therethe value 0.
Accordingly we define

Thus our task is to construct game F (normalized and 7 = 1) with a a

: H Sl Sag-element with ff( r . n set \"V o m L ^1 L r q tor J

1 otherwise

= 0,1,

,n-l,n

We first prove that v(/S) is a characteristic t function, i.e.hat it fulfills in 25.3.1. (25:3:a)-(25:3:c) We prove theseconditionsin their equivalent form of (25:A) in 25.4.2.: Case = 1with = :Thisis v( I) =0,immediatesinceI is in W, because p = -7 is in L by (48:A:b),49:C). ( Case = 2 with = :This is v(S\\) + v(S2) when Si,Sjarecomplep ments. If both Si,S2 arenot in W, L, then v(Si) = v(S2) = 0. If one of
==

Si,S2 is in W or L, then the other is in L or W, respectively,by (48:A:b). h Assume,by symmetry, Si in L, S2 in W. Let Si have q elements, ence S2 has n q. Then v(Si) = -g,v(Si) = g.
2 v OSi),v(S2), v(S3) ^ O. If one of S2, Sa is in W, we may assumeby Hence Sa = u S2 is in L by (48:A:b), so and symmetry, that it is S2 arein L by (48:A Let Si have q\\ elements, 2 have g2 elements, S hence S8 has n gi g2 Then v(Si) = v(Si) = 2, v(Si) = q\\ So at any ratev(Si) v(S2) v(S8) Thus v(S)belongsto a game T. We now establishthe remaining

any ratev(Si) + v(Si) = 0. Case = 3 with g :This is v(Si) + v(S2) + v(S8) ^ when Si,St, S, p are pairwisedisjunct with the sum /. If none of Si,S2, Sa is in W, then

So at

$3.

:d).

Si, Si

Si,

v(S) (i.e. is normalized and 7 1:Indeed, (49:C)all v((i)) = 1. T) by and v(-S)= -v(S),he two t v(S) fulfills (49:2): Owing to (48:A:b) S parts of (49:2)go over into eachother if we interchange and S. We considerherefore only the secondhalf. t If S is in L, then clearly v(S) = q. If S is not in L, then v(S) = q would necessitate = #,8 or q = 0. But this means that S is empty, (49 contradicting :C). all So the game F possesses desiredproperties.
==

assertions.

49.3.3.

-cji,

g 0.

-g

+ g.

arenow able to prove the following exhaustivestatement: In orderthat two given systems W L(/) be the TFr, Lr (49:E) of a suitable game T, these requirements are necessary and
We
}

sufficient:
F 1That

T essential: (49:1 (49:0), :a). (48:A:b)-(48:A:d),

W inessential: = L = /.

the two first specifications do not conflict, is due to if is not in W. v(S) 1As n q T* could not be in W', hencev(S)

Clearly

-7,S

(49:l:a).

0.)))

In concludingwe mention another interpretation of (49:2), Recalling the inequalities (27:7)of 27.2. also shown on Figure50), which specify ( limitations for v(S), it appears that Wr is the setof thoseS for which v(S) t reacheshe upper limiting value, and L r the set of thoseS for which vCS) t reacheshe lower limiting value.
49.4.Exact Definition of Simplicity give a rigorous definition of that classof and and to which we alluded in 48.1.2. 48.2.1., which was circumgames Where scribed in more detail at the beginningof 49.1.3.: the only aim of all players is to form certaindecisivecoalitionsand where thereareno other involved motives which require a quantitative description. g By combining the part of (49:E) which refers to essential ames with it o (49:1),appears that the formal expressionf this idea is

SIMPLE GAMES T Immediate (49:B:b). by Proof: inessential: T essential: necessity was establishedin (49: (49:B:a), :C). The A), (49 Thesufficiencyis the contentof the constructionwhich we have carried out.
428

49.4.49:E) permits us to (

:b) (49:1

Wr

uLr = /.

that t Indeed,his condition expresses any given coalition S belongseither to the winning or to the losingcategory without any further qualification. A We define accordingly: n essentialgame which fulfills (49:l:b) is

called simple.

The conceptof simplicity is invariant under strategicequivalence, sincethe setsWr, Lr are. o let concept, us consider ncemore the closingremark of 49.3.In the sense of that remark an essential ame is simple, if v(S) liesfor every S on the g boundary1 of the areaassignedto it by the inequalities(27:7)in 27.2. The variety of all essentialn-persongames (normalized,7 = 1) can beviewed as a geometrical configuration of a certainnumber of dimensions,
given in Figure65. More preciselythe inequalities referred to define a of convex polyhedricdomain Qn in the linearspace the dimensionalityin 2 a question,and the points of this domain represent ll thesegames. = 3 the dimensionalityis zero,and the domain Q 49.5.2. For n E.g. a single point. For n = 4 the dimensionalityis 3, and the domain Q4 the cubeQ of 34.2.2. Now the simple gamesare those for which we are on the boundary t of eachdefining inequality. With respecto the convex polyhedricdomain of T are Qn this means: he simplegames the vertices Qn , n = 3,4. 1The boundary consists two points:the upper limiting value n p and the lower of one p, (y 1). v(S)must be oneof thesetwo, no matter which. Thereaderwho is familiar with n-dimensional linear geometry, will note:SinceQn a is defined by linear inequalities, it is a oolvhedron. The discussion of 27.6.llows to mathematical 49.5.1. we take up the detailed Before

49.6.SomeElementary Propertiesof

Simplicity

discussionof this

'

concludethat

it

is convex.)))

CHARACTERIZATIONOF THESIMPLE GAMES

429

l essential 3-personame is simple. Forn = 4 Q4 is the cubeQ, so the simple g i c I-VIII.* gamesarethe vertices, .e. orners

: E.g. For n = 3

Q3 is a singlepoint, i.e. othing but a vertex, so the n

49.6.Simple Gamesand Their W, L. The Minimal Winning Coalitions : W m 49.6.1. Combining(49:E) with the definition of simplicity,we obtain:
(49:F)

In orderthat two given systems W, L(s/) be the Wr, Lr of a suitable simple game F, these requirementsare necessary and sufficient: (48:A:a)-(48:A:d), (49:C).

That the S referred to in (49:2)exhaust all subsets of 7, is definitory for simplicity. Consequentlyit is for simplegames and for these alone, that knowledgeof Wr, Lr determines v(S), provided that the game is normalized and 7 = without the last proviso, that it determines the gameup to a strategicequivalence.

1. I.e.,

only then, the game F is determined by its Wr, L r up to a strategicequivalence. t Consequently,accordingo (49:F) and (49:G)a theory of simplegames with is coexistensional the theory of those pairs of systems W, L which
fulfill

this restate : In caseof simplicity,and (49:G)

We

(49:C). (48:A:a)-(48:A:d), should In 49.6.2. studying the pairs TF, L describedbove, 48.2.2. be a and particularly (48:W), (48:L) there and (49:2). ccording A recalled,
these,it is sufficient to pair W, L.
name either W or L in

to

orderto determine the

by their complements.
with

are as The conditions (48:A:a)-(48:A:d) then to be replaced follows: If W is used, by (48:W);if L is used, by (48 :L). it As to (49:C), refers to L directly. We can equally well refer it to W, by applying (48:A:b) then the sets mentioned in it must be replaced
we For the sakeof completeness, restate(48:W) forms of (49:C). the corresponding and (48:L),together

The setsW(
(49:W*)

b I) arecharacterizedy theseproperties:

(49:W*:a) (49:W*:b) (49:W*:c)

Of two complements(in /) S, S, one and only one belongsto W. o W contains the supersets f its elements. sets. W contains / and all (n l)-element

*Cf.also (50:A)in 1As far as the corners7, V, VI, VII are concerned,his is no surprise:Our dist cussionstarted with thesein 48.1 and our conceptof simplicity obtained from them by
generalization. in

50.1.1.

as 35.2. the

of The reappearance the corners 777, IV, 77,

easily from

(50:

prototypes of decomposability. However, they A) and the beginning of 51.6.)))

i VIIIs more puzzling:

are simple too, as follows

We treated them

430

GAMES SIMPLE

The setsL (

I) arecharacterizedy theseproperties: b

Of two complements (in /) S, S, one and only one belongsto L. (49:L*:b) L containsthe subsets of its elements. sets. (49:L*:c) L containsthe empty setand all one-element

(49:L*:a)

(49:L*)

As pointedout above, we couldbase the theory on eitherW with (49:W*) or onL with (49:L*). Since 49.6.3. it is more in keepingwith the usual way of thinking about we thesematters to specify the winning rather than the losing coalitions, shalluse the first mentionedprocedure. In this connection observe that a certainsubset of W shares the we S importanceof W. This is the set of thoseelements of W of which no of proper subset belongsto W. We call these S the minimal elements W W (i.e. r ) and their set W m (i.e. F?). T minimal winning The intuitive meaning of this concept clear:These is coalitionsarethe really decisiveones,thosewinning coalitionsin which no that participant can be spared. (It will be remembered our discussionof began with the enumeration of thesecoalitionsfor the game we were then considering.)

48.1.3.

49.7.The Solutions of Simple Games 49.7.1. heuristic considerationswhich led us to the conceptof The simplegames make it plausible,that the discussionof gamesbelongingto this category,may turn out easierhan that of (zero-sum) t n-persongames in general. Fora corroboration this we must examine how solutionsare of the determined in a simple game. Sincewe arenow considering old form 1 of the theory, 30.1.1. be consulted. We begin with the observation must that a considerable from the fact that in a simplification must be expected setis eithercertainly necessaryor certainly unnecessary simplegameevery

sary, and all setsS of Z/ r are certainly unnecessary. I Proof:f S is in Z/r then it is flat, hencecertainlyunnecessaryby (31:F) in 31.1.5. S is in W?, then S is flat (because is in L r ) and S 7* Q If it (because is in Z/r, hencenot in Wr). So S is certainly necessaryby
We can now fulfill our above promiseconcerning simplegames indeed, this can be done in two different ways. 1In the terminology of the new form of the theory as introduced in 44.7.2. sequ. et this means:We are looking for solutions for #(0), .e. excess being restricted to the i the is

(cf.31.1.2.). 49.7.2. orderto establishthis assertionwe prove first: In In any essential ame T all setsS of Wr arecertainly neces(49:H) g

(31 in 31.1.5. :G)

value

Thesigmficance.of this

0.

restriction

will

become clearerin

the third remark

of 51.6.)))

MAJORITY GAMES AND THE MAIN SOLUTION

431

(49:1)

In any simplegame F all sets S of W r arecertainly necessary, and all others arecertainly unnecessary.

Proof:Combine (49:H) with the fact, that for a simple game Lr is o preciselythe complement f Wr.
(49:J)

In any simplegame F all setsS of W arecertainly necessary, l and all others arecertainly unnecessary.

W t we can Proof: e can replacehe Wr of (49:1)by its subset W^, transfer all S of WT W? from the certainly necessary classinto the certainly unnecessary,owing to (31 in :C) Indeed,very in Wr e a possessessubset T in TFp. Of these two criteria(49:1) (49:J),the latter is more useful. and Their importancewill be establishedby actual determination of solutions 2 in simple games. Indeed,this analysis of simple games permits the deepestpenetration yet effected into the theory of games with many

i.e.

31.1.3.

3 participants.

it is appropriateto give someexamples of the pairs W, L of (49:F) in 49.6.1. know from i We of simple games, .e. that 49.6.2. it sufficesto discussthe W as characterized (49:W*) there. by Let us therefore considersomepossible ways of introducing such W o possibledefinitions of a concept f winning. The principle of majority suggests itself as a particularly suitable it Hence is plausibleto define W as the systemof all definition of winning. those Swhich contain a majority of all players. It will be noticed, owever, h ties indeed (49:W*:a) tatesfor this W, that for s that we must exclude every S either S or S must contain the majority of all players, thus excludingthat both may contain exactly half. In other words:The total number of participants must be odd.

50.1.1. going any further, Before

50.The Majority Gamesnd the Main Solution a 50.1. : Examples of Simple Games The Majority Games

i.e.

4 So if n is odd, we may define W as the setof all S with > ^ elements. the The simplegame which obtainsin this way, 5 will be called direct majority

game.
1Comparison
certainly

remark at

necessaryand

1Cf. 50.5.2. 1Cf.

55.2.-55.11.,

that the of Wr W are simultaneously for the certainly unnecessary. This is another illustration the end of footnote 1 on p. 274. and 55.2. and in particular the general remarks of 54. of

a :J) (49:1)nd (49 shows

4 Sincethe smallest integer

j

> ^ is

(n

w odd!), e

may also say:S must have

elements. T Precisely: heclassof strategically

equivalent

ones(of n

participants).)))

432

GAMES SIMPLE
1

The smallestn for which this can be done, is 3. We know that there exists one essentialhree-personame,and that for this W it consists t g only o s preciselyof the 2 and 3-element ets i.e. f the setswith > f elements. So we see: The (unique) essentialthree-person i (50:A) game is simple;t is the direct ajority game of threeparticipants. m
majority game is merely onepossibilityamong many. The 50.1.2. directmajority gameis only available, when n is odd, and yet simplegames existfor even n as well indeed our prototype of simple had 48.1.3.) n = 4. games (cf.48.1.2., to the concept f majority is easilyextended cover the case o of However, even n as well. To this end we introduceweighted majorities in the following L manner: et eachoneof the players 1, , n be given a numerical weight, w n respectively. Define W as the set of all thoseS which , say Wi, contain a majority of total weight. This means:

For the subsequent n

which

are eligible,n = 5,

7,

, the

direct

(60:1)
or equivalently,

i in 8

w. >

i -1
t

Wi,

(50:2)
We must

generality of our presentsetup, it is betterto proceed immediatelyto a discussionof (49:W*). complete Let 50.1.3. us see,therefore, what restriction(49:W*) imposesupon
the MI,
Ad

-8 ties. However,owing to the greater again take careto exclude

iinS
iin

% Wi>

J)

Wi.

Sincewe can express S belongsto that (49:W*:a): S belongsto W when

i in 8

wn

by

so (50:2),

(50:3)

Wi

<

i in

-8

Wit

m So (49:W*:a) eans that always This means clearly,that never

or (50:2) (50:3) but holds,

never both.

(50:4)

Wi

=
t in

i in 8

-S w^
Wi.

or equivalently, (50:5)
1
w I.e. hich
ia

that never
Wi

tinS

i -l

odd and for

which

a game can be essential.)))

MAJORITY GAMES AND THE MAIN SOLUTION

433

Ad (49:W*:b): singthe definition of W in the form (50:1), requirethis U ment is clearlysatisfiedif all Wi ^ O.1 Ad (49:W*:c): sing again (50:1),is clearthat / = (1, it U , n) belongs to W. For the general(n l)-elementset S = / (i' ), the

condition(50:1) statesthat))

Summing

up:

The weights Wi, , w n can be used to define by (50:1) or (50:2)W which satisfies(49:W*) if and only if they fulfill a the following conditions: (50:B:a) Forall t = ,n

(50:B)

1,

;S 0** Wi

<i
^x

n
V^
2_t

~y

Wi.

(50:B:b)

Forall S

I))

tinS

-l
this W 3 will
9 * * *

2 total weight.

the total weight or more; o combination of playershas preciselyhalf of the n The simplegamewhich obtains from
participants with be called weighted the
,
wn

Verbally: A player has always non-negative weight, but never half of

majority game (of n

also designate game by the symbol[t0i, this , w n]. Thus the direct ajority game has the symbol[1, , 1], m It will be noted that the four-person game represented the corner by 7 of Q, discussedin 48.1.2., as 48.1.3. be described a weightedmajority can can t game. Indeed,he principleof winning found in 48.1.3. be expressed while player 4 has the have by saying that players 1,2,3 a common weight, double weight. I.e.his game has the symbol[1,1,1,2]. t

the weights w\\

).

We

will

60.2. Homogeneity
majority games and their explanatory is a step in the directionof a quantitative (numerisymbols[wi, , o cal) classificationand characterizationf simple games. There are good to reasons think that it would be most desirable carry out such a program to terms and fully: Simplicity was defined in combinatorial,set-theoretical would make them it is to be expected that a numerical characterization
w n] we

60.2.1. introduction of The

1This is, of course, perfectly plausible condition; indeed, the surprising thing is that a that to are not forced require u> > i.e. wecan permit a weight to vanish. 1The first requirement obviates the difficulty of 49.2., second t excludesies. the 1Precisely: heclassof strategically equivalent ones.))) T

434

SIMPLE GAMES

numerically, and therefore it seemslikely that a numerical characterization will correspond them more directly than a combinatorial one. to However,this first step is far from carrying out the transition. On the one hand, a simple gamemay possess ore than one symbol m [t0i, , w n] indeed, every simplegame that has oneat all has infinitely many. 1 On the other hand, we do not know whether all simple games 2 possesssuch a symbolat all. We begin by consideringthe first deficiency. Sincethe same simple game may possessseveral symbols[10 1, , w n ], the natural procedure is to single out from among them a particular one by some convenient principleof selection.It is desirableto specifyin this principlesuchrequire- - , wn ments which increase significance and usefulness the w\\ the of First somepreliminary observations. Theconditions(50:1), sug(50:2) gestconsiderationof the difference
9

easiero handle. Sucha characterization t usually facilitates a more exhausin tive, quantitative understandingof the notion considered. esides, our B for present problemwe areultimately searching solutionsthat aredefined

(50:6)
properties:
(50:C)

aa

=2

t in

w<- %*><= %
i 1
\\

i in 8

in

1*.

This d s expresses much the coalition S outweighsits opponents how how much of a weighted majority it possesses. heseare its immediate T
as as as as

-a_s

U Proof: se the last form

of (50:6)for a s
if and only if if and only if

(50:D:a) (50:D:b) (50:D:c)

A Proof: d Ad Ad

> <

is impossible.

S belongsto W. S belongsto L.

that the amount of a s which secures victory be the same for eachwinning coalition.It would be unreasonable, however, to require this actually for T all S of W: If Sbelongsto W, then its propersupersets do too,nd they may a a T > a s z Since T containsparticipants who arenot necessary such a have for winning, it seemsatural to disregardit. I.e. e requirethe constancy w n of a s only for thoseS of W which arenot proper supersets f otherelements o

(50:D:a): Definitory. and Immediate (50:D:a) (50:C). (50:D:b): by since Immediate (50:D:a), (50:D:c): (50:D:b) W, L exhaust all by S. It also coincidesith (50:B:b). w 60.2.2. it is natural to try to arrangethe weightsw\\, Now , w n so

1Obviously, sufficiently small changesof the u> will not disturb is by particularly since(50:5) excluded 1We will seein that certain simple games have none. 1Thus for T for all i not in S, a/ > as unless u><

/a

53.2.

(50:B:b).

the validity

of (60:1),

8.)))

MAJORITY GAMES AND THE MAIN SOLUTION

of W.

435

stant for the

We define accordingly:

In the terminology introducedin 49.6.3.: required to be cona s is minimal elements f W i.e. elements f W m . o the o (50:6)have a common value, to be denoted by a, for all S of
W1 *.

(50 :E)

The weights

w\\ y

wn

are homogeneous,if

the a s of

Whenever (50:E) is valid we shall indicate this by writing [wi, , w n ]h instead of [wi, w n]. , Clearlya > 0. A common positive factor affects none of the significant of propertiesof w\\, , w n , therefore we can use this in the case homofor a final normalization: Makinga = 1. geneity We conclude observing that the games mentioned at the end of by 50.1.3. homogeneousand normalized by a = 1. Theseare the direct are the majority games of an odd number of participants [1, , l],.and corner7 of Q [1,1,1,2] can accordinglybe written [1, which , 1]* and [1,1,1,2]*. the reader ill verify with easethat a 3 = 1 for all w Indeed, S of W m in both instances.
A 50.3. MoreDirectUseof the Conceptof Imputation in Forming Solutions 60.3.1. homogeneouscaseintroduced above is closely connected The

with the ordinary economiconcept f imputation. We proposeto show o c this now. More precisely:We defined in 30.1.1. a general oncept f imputations o c and based on it a concept f solutions. In forming thesewe were led by o the same principlesof judgment which areused in economics, therefore and somerelationshipwith the ordinary economic o concept f imputation must be expected.However, our considerationstook us rather far from that concept.This appliesespeciallyto the constructionswhich were necessary when we found that sets of imputations i.e. solutions and not single imputationsmust be the subjectof our theory. It will now appear that for certainsimplegames the connection the ordinary economic oncept with c of imputation can be establishedsomewhat more directly. One might say that for the special b gamesin questionthe connectionetweenthis primitive conceptand our solutions can be directly established. Actually it will providea simplemethod to find a particular solutionfor eachone of those

The 50.3.2. two conceptsof solution, i.e.the two procedures, support eachother quite effectively. The ordinary economic onceptprovides a c useful surmiseas to the form of a certainsolution. And then our mathematical theory may be used to determinethe solutionsin question and to make the requirements of the ordinary approach complete.(Cf. 50.4. on the one hand, and 50.5. sequ.on the other.) et Theseconsiderations alsoserve another end: They bring out the limitations of the ordinary approach with greatclarity. The ordinary approach)))

games.

436
functions in this and not entirely a disclosell the remarks on this
in

SIMPLE GAMES
a form only for the simplegames, nd even therenot always it not unaidedby our mathematical theory. Besides, does solutions for the games to which it applies. (Further subjectoccurthroughout the discussion,and particularly

theseXi, (50:7)

w In this connection e emphasizeagain that any game is a model of a organization and any solution is a possible possible social or economic stablestandard of behavior in it. And the gamesand solutionsnot covered c b by the method referred to i.e. y the unimproved economic onceptof theory. imputation will prove to be quite vital ones for socialor economic It will be seenthat the simplegames which can be treatedby this special with the homogeneousweighted majority method are closely connected gamesof which they area generalization. 50.4.Discussion This DirectApproach of Considera simplegame F which we assumein the reduced form 60.4.1. = 1,but which we do not yet restrict ny further. Let us try to with 7 a ideas without making use of discussit in the senseof the ordinary economic our systematictheory. aim Clearly,in this game the sole of playersis to form a winning coalition, and oncea minimal coalition of this kind is formed, thereis no motive for its participants to admit additional members.Consequentlyone can assume that the minimal winning coalitions the S of W m are the structures that will form. It is therefore plausible to assume that a player's fate in presents only two significant alternatives:He either succeeds joining one of the desirablecoalitionsor he does not. In the lattercasehe is defeated, hencehe obtains the amount 1. In the former casehe is suct cessfuland accordingo ordinary ideas ought to ascribeo this success one t a value. This value may vary from one player to another;for player i it we denote by 1+ z< so that x is the margin betweendefeat and success l for player i. 60.4.2. now formulate the requirementswhich must be imposedon Letus

50.8.2.)

First: y the very meaning of the z* necessarily B

Xi

xn in

the course a conventional economic iscussion. of d

0.
winning

certainplayer i, then there exists him no alternative to the value for and so we neednot define any z for him. 2
1We assumehere that there is only one way
of winning,

I Second:f

it

happens that no

minimal

coalition contains a

1,

same whichever (minimal winning) coalition the player succeeds joining. This is in in plausible sincethere is only one kind of success a simple game the complete one or every coalition being either fully defeated fully winning. It will appearin 50.7.2. 50.8.2. far this standpoint carries. As far as it and how does,it can be advantageously combined with our systematical theory.

i.e. that

the margin

x, is the

1For the really to some minimal

in 51.7.3.)))

important
winning

simple games such i do not exist coalition. Cf. the first observation

i.e. player every

in

and 51.7.1. (51:0)

belongs

MAJORITY GAMES AND THE MAIN SOLUTION

437

Third:If a minimal winning coalition S becomes effective, then the division betweenthe playerswill be this: ach player not in S obtains E each player t in S obtains + x* The sum of theseamounts must be zero. This means

-1

1,

0=

i.e.
(50:8)
a =

i not in S

= (-1)+ S (-!+*<) -n+i 2)S*,

t in in

In our system of notations

with
f
\\

{!,,}
__ Qt\\

*t

n.))

tin))

this

distributionis describedy the vector b

t

the components
(

1 for 1 -f Xi

not in S, for in

i S. 30.1.1.

Our first condition and the presentone actually statejust that a is an imputation in the senseof 50.4.3. Continuing the usual line of argument, \\ve shall now want to determinethe x\\, , xn by means of the equations and inequalities of the three above remarks. In doing this, one more point must be conW sidered: e have stated in the third remark, that its S must be minimal i.e. elongto W m However it may be asked whether all S of W m b winning, can be used. t Indeed,he presentprocedureis nothing but the usual one to determine the imputation of values to complementary goodsby meansof their alterna1 tive uses. Now these alternative uses may be more numerous than the W than n.2 different goodsunderconsideration i.e. m may have more elements In such a situation onemight expecthat someof the uses areunprofitable t w and neednot be includedin the third remark. Indeed, e already made use of this principleby taking the S of W m only, and not all elements f W, o are because S of W W m (the non-minimal winning coalitions) clearly the as wasteful. Are we now sure that all S of W m must be considered of profitable uses? They areclearlynot wasteful in the crude equivalents senseof the S mentionedabove; no participant of an S inW m can be spared without causing defeat. But unprofitability can arisein less directways show. Thus the questionremains than this, as numerouseconomic examples m areto be used in the third remark. unansweredas to which S of W i It is clear, owever, that if an S of W m is not includedthere,i.e.f h
We by ' 6

denotethis vector

a s.

(50:8)
1In
total

t in

J)

Xi

=n

is this caseit would be more suitable to say, services. Theobjectconsidered the service player i in cooperatingwithin a coalition which he joins. of 1 Cf.The fourth remark in 53.1.)))

438

SIMPLE GAMES

fails to hold for it,then it must be definitely unprofitable. have > in place = in (50:8): of

I.e.we must

(60:9)

tin

5)

Xi

> n.

Thus the questionarises: y what criteria we to determinewhich S are B of W m fall under the third remark h for which must (50:8) old. Denote U. So their set by U(s Then (50:9) must hold for the S of W m W\.") the problemis to determine 1 U.

i.e.

point by going back to our systematictheory. From the statementmade 50.4. carry over this much:Considera system of minimal winning we Form the imputations)) coalitions, a set U W m and the
in

50.5. Connection with the GeneralTheory. Exact Formulation 60.5.1. of attempting a verbal description, let us settlethis Instead

i.e.

z.

as in

50.4.

'

1for i not in S 1+ Xi for i.. S] in

}
\\

when

. . rr S is in U.

as That these a 5, S in J7, areindeed imputations, is expressed, we know, by the conditionsof 50.4.

(50:7)

Xi

(50:8)

x
tin

=n

z 0,
when S is in

U.

whether U and the Xi Formthe set V of the a 5, S in U. We will decide are satisfactory,by determiningwhether this V is a solution in the sense of

be seenthat the result which is obtained in this manner can be from the ordinary economic perfectly reasonable point of view. But it may be questioned whether it could have been T unequivocally establishedby the usual procedures. his may serve as an illustration of how our mathematical theory can serve as a guide even for the purely verbal discussions of the ordinary economic approach,

30.1.1.
II will

statedverbally and is

(cf.50.7.1.).

50.5.2. proceed investigatewhether V is a solution. to We

*
would

Let us determine , /3n } first, when a given imputation ft = {0i, is dominated by a given a T, T in U. Sincethe gameis simple,the set S
is the real objective!)))

(50:9).This would

1It

be utterly mistaken to
not

restrict the Xi,

try

to

*,

define W m

sufficiently

U (and so 17) by means of and their determination

MAJORITY GAMES AND THE MAIN SOLUTION

of
W m,

439

30.1.1.this domination can be assumedto belong to W (or even to for use (49:1) (49:J) in 49.7.2.). every i in S, af > ft ^ -1; or For for every i not in T a = - 1 : henceS T. Now T is in U c jp 5 is in fF, therefore S S T yields S = T. So we see: he set S of 30.1.1. T for this domination must be our T. And T can be used there, sinceit is certainly necessary,as it belongsto U W 5 W, cf. above. Hencethe
9 f 7

domination

aTH

ft

amounts to
ft,

(50:10)

<

-1+ z,
ft

a this:r >
f

ft*

for i
1

in

T,

i.e.

for

i in 7 .
,
n

Denotefor any imputation

= {0i,

(50:11)
by

ft

-1+x.

the

setof all i with

alterna-

72( ft ). Then (50:10) statesthat 72(7)and T aredisjunct. tive way of writing this

is:

An

(50:12)
We

-72(7)2^.
<*

repeat:
r *-

(50:F) (50:G)

is equivalent to

(50:12).

From this we can infer:

Let U* be the set of all 72( I) which possesssome subset belongingto U. Let U+be the set of all 72(s7) for which R doesnot belong
to

U*.

is undominated by any elementof V 7?(7) belongsto U+.

Then
ft

if

and only

if

o T b Proof: hat ft is dominated by someelement f V i.e. y some a r , holds T in U means that (50:12) for some T in U. This is equivalent to

sayingthat

-72(
)

t is in U*,i.e.hat
r

72(

is not in U+.
if

HenceR( ft o element f V.
the

belongs to L +

if

and only

is dominated by no

Before 60.5.3. going any

further we U+

set t/+of (50 :G)

observe four simple properties of

if

(50:H:a)

U*

=W

U=

Wm

Then U* consistsof those setswhich possess a a subset belongingto W i.e. minimal winning subset. HenceU* = W. The operationwhich leads in (50:G)from U* tot/+is the combination of the in Now we noted already))) and (48:A:b) 48.2.1. transformation (48:A:a)

Assume U = W m . Proof: m

440

GAMES SIMPLE

e then, that thesetwo transformations compensateachother,when applied to W Hence7* = W gives t/+ = W (50:H:b) U*is a monotonic and U+ is an antimonotonic operation. I.e.f/i C/2 implies E7? C7? and Uf C7f. t I Proof: t sufficesto recallhe definitions in (50:G),to seethat Ui Uz

=>

implies U*

s U* and this in turn

All

(50:H:c)

our U sW m have *7* c TF c f/+.

Uf 2 C/J.

C and of (with Proo/: ombine(50:H:a) (50:H:b) U, W m in place Ui, C/2). of (50:H:d) Both U*and [7+ contain all supersets their elements. T Proof: his is obvious for U*. The property under considerationis the i same one which was formulated in (48:A:c)n 48.2.1. taking the place (W

of our [/*,J7+.) Now the operationwhich leads in (50 :G)from U* to U+, a i is the combination of the transformations (48:A:a) nd (48:A:b)n 48.2.1. of (48:B)in 48.2.2. thesetwo to the proof of (50:H:a)). Application (Cf. transformations shows that the property in question is conserved when passingfrom U* to U+. Note 50.5.4. that U*, U+ allow a simple verbal interpretation. If we knew only of the coalitionsbelongingto U that they arewinning, of which coalitionscould we then assertthat they are certainly winning, and of

set of the last mentioned ones. Now the meaning of (50:H:a)-(50:H:c) clear:For U = W m , becomes everything is unambiguous:The certainly winning coalitionsare precisely those which are not certainly defeated, and they form the set W. As U from decreases W m , the gap widens. The first setdecreases through subsets of W, the second increases one of W. through supersets Theassertion (50:H:d) of is equally plausible.
the

arenot certainly defeated? for Theformer is the case the coalitions with subsetsin T,i.e.or those f of U*. Thecertainly defeatedonesarethe complements these,i.e.hose of t not in U+. HenceU* is the set of the first mentionedcoalitions, and {7+
which that they

60.6. Reformulation (50:1)

V

of the

Result

50.6.1. of 50.5.2. us to state: allows (50:G)

is a solution if and only if R(ft) belongsto U+ precisely when ft belongsto V. So we must only decide when (50:1) holds. For this purpose we consider > > an R in U+and determine ft for which R(ft) = R. the Considerthe three possibilities :

(50:13)

i not in

= (-1)+ (-!+*<) 0, i
V
in ft
<)))

MAJORITY GAMES AND THE MAIN SOLUTION

441

i.e.
(50:14)
>

Xi

= n.

inB

<

If a

ft

with

12(

then ) = R exists, we have

(50:15)

i 1

- ft

i not in

+ (-1) i % (-1+ *,),

in ft

i.e. in (50:13), So t ^ (50:14). > in (50:13), excludeshe existence (50:14)

of any \"/? with R(~ft) = R. I.e.he setsR in C7+with > in (50:13), t (50:14) neednot be consideredurther. Consideron the other hand, an R in E7+ f with < in (50:13), Then (50:14). thereareinfinitely many ways of choosing ft(

7
)

with

ft
l

and ft ^

(\"I+ I 1

For 5**'***!. /c)

Xi

for t m

all these

it SinceV is necessarily2 ft. Hence belongsto V by (50:H:d). s finite, these ft cannot all belongto V. This is a contradiction. I.e. ets R in U+with < in (50:13), must (50:14) not exist. It 60.6.2.remainsfor us to considerthe setswhich arein U+ with = in (50:13), (50:14). According to the above, these must furnish precisely

the V of V.
If
R ( ft )
ft

to belongs

V,

i.e. = a T, T in
ft

C7, then

we have this situation:

Also

is T plusthe setof thoset for which x> = 0. T belongso U c [7*c [/+ t hence R(ft) belongsto U+. (for the second relation use (50:H:c)),

2_ i

Xi

in R(0)

i in T

2)

Xi

= n-

Hence So we have = in (50:13), (50:14). the ft of V areall takencareof. Addition a (50:14). Conversely:Consider n R in U+with = in (50:13), to of all i for which x> = to R affects neitherthe fact that R belongs [7+ So nor (by (50:H:d)), the equation (50:14). we may assume that R contains all thesei. If now an imputation ft has R( ft ) = R, then ft ^ + x> for i in fi.

-1

Always ft

1. As

ft

= 0,this implies:))
for not in for t in
R.)))

(60:16)

_}))

442

GAMES SIMPLE

implies that ft is an imputation with R( ft ) = ft. Conversely: (50:16) be our Hence requirementin this casemust be that the ft of (50:16) an

the inclusion of all such i into R. Summing up: C V is a solution if and only if this is the case: all an i indiffer(50:J)
ent when z t

i a r , T in U. This means,that T and R differ only in elements for which = 0. And this property is insensibleto our original modification of fi, Xi
= O.
1

Then we have

(50:8*)

in

J)

Xi

=n

for the T of U and, of course, lso for thosewhich differ from a these only by indifferent elements. And we must require

(50:9*)
tin

xt T

>n
first, then verify whether thesex<

for all other T of U+.

attempt to determinethe Xi from fulfill the inequalities

In making use of this result, one may chosethe set U c W m

and (50:8*) finally
Xi

(50:7)
and (50:9*). This 50.5.1. is it:
A

The 60.7.1.

50.7.nterpretation I

of the Result

result (50:J) permits the verbal statement promised in

solutionV is found by choosing arbitrarily the set U of thoseminimal ( coalitions (i.e. 7cWm ), which are to be considered profitable. The Xi must then satisfy the corresponding But equations (50:8*). after this, it must be verified that certain other coalitionsaredefinitely inprofitThis ablein the senseof (50:9*). must be requirednot only for thosecoalitions which areknown to be winning, (i.e. but for all those which cannot W),
winning

1Thesei constitute a slight complication which is further aggravated by the fact we have no example of a game in which they actually occur. It may be that they never exist;an indifferent i characterizes player who belongs to some minimal winning a a coalitions,but never receives share. The excludedplayer in a discriminatory solution of the three-persongame is in this with c situation (cf. (32: ) in A 1). But that solution is an infinite set, whereas our V must be finite. It would be of interest to decidethis existential question. At any rate we must at presentprovide for the indifferent i to avoid lossof generality or rigour.)))
that

32.2.3.

MAJORITY GAMES AND THE MAIN SOLUTION

443

be establishedas definitely defeatedby the coalitionsof U alone (i.e. U+) of t excepting, course,he coalitionsof U itself.1 The readermay now judge whether the concluding remark of 50.5.1. is justified by this formulation. 50.7.2. question of finding the proper U for (50:1)a rather deliThe is cateone. The antimonotony of U+ (cf. (50:H:b) 50.5.3.) itself in makes felt now: Decreasing i.e.he number of equations, increases t [7, [/+, the number of inequalities,and viceversa. In particular, if we choose as largeas possible,i.e. = W m , then the U U with U+ createno difficulties at all. Indeed: inequalities associated U = W\" implies U+ = W by (50:H:a) 50.5.3. T of W certainly in A a possessessubset S which is minimal in W, i.e.elongsto U = W m . Now b if T differs from this S by more than indifferent elements,hen we have t x = n, i.e.50:9*) desired. hence x > as X{ > for somei in T S, (
i in 8 Thus U = W m always yields a solution V, if its equations (50:8*) an c be solvedat all (with (50:7)). But as we pointed out in we have no right to expect priori a that this will always be the case especiallysincetheremay be more equations i elementsn TFm), than variablesXi. The last objection not an absolute one;indeed it is easy to find a is the simple game for which the number of theseequations exceeds number
in

^T

50.4.3.,

(i.e. (50:8*)

2 of variables and the solution nevertheless exists. On the other hand thereexistsimplegamesfor which thoseequations have no solutions. An of example this is somewhat more hidden,8 but the phenomenon is probably o fairly general. When this occurs, ne must investigate whether a solution V cannot be found by appropriate choicesU c W m The difficulty and delicacyof this questionhas beencommentedupon alreadyat the beginning 4 of this section.

50.8. Connection with the Homogeneous Majority Games w 60.8.1. now restrict urselvesto the caseU = W m . I.e. e assume We o
that the
full

system of equations
tinS
Xi

(50:17)
can be solved with

Xi

=n

for all S in W m ,

(50:7)

0.

1And those which differ from them only by indifferent elements. 1This for the first time for n 5, cf.the fifth remark in 1This happens for the first time for n 6, cf the fifth remark in 53.2.5. happens m 4 No instance of a simple game with a solution V derived from U W is known, nor is it established that none exists. The further-going question whether every simple W m is equally open. solutions V of suitable U game possesses The problem seemsto be of some importance. It may be difficult to solveit. It appearsto have somesimilarity with the solved questions mentioned in footnote 1 on so p. 154,but it has not beenpossible, far, to exploit this connection.)))

53.1.

444
We saw that
in

SIMPLE GAMES

the this case setV of all a 5, S in W m , is a solution. In this situation and only then, we call V a main simplesolution of the game. There is a certain similarity between these requirements and those t a which characterize homogeneous weighted majority game. Indeed,he latteraredefined by

(50:18)
tin

*>

&

for all

S
a

S in

where))
Wi;
())

a,

>
Wi

( (combine(50:D) 50:E)of 50.2.)

and

(50:19)
Actually,

0.

1 is follows: The quantity b of (50:18) positive. Multiplicationof all Wi leaveseverything unaffected, and by choosing by a common positive factor b this factor as n/b we can replace in (50:18) n. Now we can simplyput by
Xi

thereis more than similarity. Thus, if a systemof w iy fulfilling (50:7) (50:19) (50:18), is given, a system z< fulfilling (50:17), obtains as

= Wi and (50:18), become50:17), (50:7). ( (50:19)

If conversely a systemof Xi
fulfilling
Wi

extra difficulty.

We may put

(50:17), is given, thereis an (50:7) ss x<.2 Then (50:7) becomes50:19) (

with and (50:17) yields (50:18) 6 = n,

i.e. = 2n a
>

w questionarises hether the last requirementa

is fulfilled

] w. -i

But now the

i.e. hether w

(50:20)
Summingup:

-i

Xi

< 2n.
a possesses

(50:K)

simplesolution. a Conversely,if a (simple)game possessesmain simple solufor the game can be derived from it tion, homogeneous eights w if and only if (50:20)s fulfilled. i
main

Every homogeneous,weighted majority game

This 50.8.2. connectionetweenhomogeneous eightsand main simple b w solutions is significant. But it must be stressed that a homogeneous, t weighted majority gamewill in generalhave other solutions besideshe
and Then (50:6)of 50.2.1. (50:19) a. g gives not the

1Otherwiseall i occurring

in the

S of W m would

have

have no

1The i which
Xi

case.

for the

8of W

w>

by
t

hencea

and (60:18) (50:19).

0, which is

cluded from

contingency

belong to no minimal winning setcausea slight disturbance, sincethey while we require their IP<. (cf. the secondremark in 50.4.), However, the is unimportant (cf.loc.cit.) and we can put theseWi as is easily conthe referencesf footnote 2 on p. 436.))) o

ENUMERATIONOF ALL SIMPLEGAMES

main

445

1 simpleone. And a game with a main simplesolution may not fulfill i.e. (50:20), thereneed not be < in

(50:21)

2 % x{ = 2n.

-i

< >

Beyondall this, finally, we must not lose sight of the main\" limitation\" Whether we take the conceptof ordinary of these considerations: U (i.e. = W m) or in its wider imputation in its narrower form of m it is 50.7.1. C7s W , cf. (50:1) 50.6.2.), cerin (i.e. original form of 50.6., to to tainly restricted simplegames. That it is necessary go beyondthese, and beyond the specialsolutionsdescribedhere,and that this forces us to was pointed out fall back completelyon the systematicaltheory of at the end of 50.3.

50.8.1.

30.1.1.,

61. ethodsfor the Enumeration of All SimpleGames M 51.1. Preliminary Remarks with 50.1.1. we introducedspecific 51.1.1. simplegameswhich Beginning instead of the original permitted characterization by numerical criteria We o set theoretical nes(cf. the beginning of 50.2.1.). saw, however, that

c thesenumerical proceduresouldbe carried out in several ways and that for there was no certainty that all simplegames couldbe accounted with their help. It is therefore desirableto devisecombinatorial (settheoretical) methodsthat producesystematicenumeration of all simplegames.
in This is, indeed,indispensable order to gain an insight into the possibilitiesof simplegamesand particularly to seehow far the above mentioned numerical proceduresarry us. It will appear that the decisiveexamples c obtain only for relatively high numbers of of the non-obvious possibilities 8 players, so that a mere verbal analysiscannot be very effective. that 51.1.2. pointedout at the end of 49.6.3. the enumeration of all We o simplegames is equivalent to the enumeration of their sets W, i.e. f all W setsW which fulfill (49:W*) in 49.6.2. e alsonoted there that it may be the advantageous to replace use of W (all winning coalitions),by W m

Either procedure provides an enumeration of all simple games. The use of W is preferable from the conceptualstandpoint sinceW has the
cf.the end of solution of the essential three-person game and We know from (32:B)of original solution of that other solutions exist. is the cf. the end of The main simple solution of corner 7 of Q We will discuss this game, together with the more general original solution of one , n 2]*(n participants) in 55.and obtain all solutions. All thesereferences ake it clearthat the solutions other than the main simple one m are quite significant, cf. and s occursfor the first time for n 6, cf. the fourth remark of 53.2.4. occurs for > the first time for n 6 or 7, cf.the sixth remark of a Both theseexamples re quite interesting in their own right. 8n 6, 7 cf. 53.2.)))

(all minimal

winning

coalitions).

is 50.2.) the

1The main simple

33.1.

i.e. 29.1.2.,

32.2.3. 32.2.3. ([1,1,1]*,

50.2.)

[1,

1,

35.1.3.

([1,1,1,2]*,

33.1 54.1. -

53.2.6.

446 simpler definition and

Wm

GAMES SIMPLE

and saturation introducedin

was introduced indirectly with the help of W. a enumeration of all simplegames which is our present im Fora practical the use of W m is preferablesinceW m is a smallersetthan W l and therefore more readily described. We will give both proceduresuccessively. It will appear that these s o discussions providea natural applicationof the concepts f satisfactoriness

30.3.

i.e. 51.2.1. sets W are characterizedy (49:W*) in 49.6.2. by the The b conditions(49:W*:a)-(49:W*:c) constitute (49:W*). which Let us for a moment disregard (49:W*:c), nd consider(49:W*:a), a of (49:W*:b). Thesetwo conditionsimply that no two elements W can 2 be disjunct. In other words:Denotethe negation of disjunctness i.e. of Sc\\T7*Q by ScRiT. Then (49:W*:a),(49:W*:b)imply (Hi-satis8 factoriness. A more exhaustivestatementalong theselinesis this:
a a :A) (51 (49:W*:a), (49:W*:b) reequivalent to (Rrsaturation. ( Proof:Ri-saturation of W means this: (51 :1) Sbelongsto W, if and only if Sn T * for all T of W.

51.2. Saturation The

Method Enumeration

by

Meansof

If

doesnot belongto S n T = 0.

(49:W*:a), imply (49:W*:b)

belongsto W, then we know that

W, then T

Let (51:1): W

-S

T of W. If S a to W by (49:W*:a), nd belongs

fulfill

Sn T

7*

Q for all

(49:W*:a), (49:W*:b).

the reverseorder. Ad (49:W*:b):f S meets I then the criterionof (51:1), every superset of S does too. Hence contains the supersets its elements. of W Ad (49:W*:a): wing to the above, S is not in W if and only if no O subset of S is in W. I.e. hen every T of W is not w S,or again,when for every T of W, S n T ^ . By (51:1) means preciselythat S is this
in (49:W*:b) (49:W*:a), in W.

L (51:1) implies (49:W*:a), 49:W*:b): et W (

fulfill

We (51:1).

prove

in

W m varies, but it is always considerably smaller. (Cf fourth remark in 1Proof: et T belong to W, n T L Then T, hence belongsto IF by thus (49:W*:b), violating (49:W*:a). 1Cf. the definitions of 4 It will be remembered that we alsoassumed in the general validity of s<JU?in this caseof Sto\\S. This means soit fails for & exclude form W, hencewe may use as domain D However, (49:W*:a),49:W*:c) ( of in the sense instead of (the system ofall subsets of equally well 7 (the

exactly 2 elements. The number of elements in

W

48.2.1. Togetherthey ~1
S,

Thus, at any rate,preciselyone of /S, S belongsto W. 5.4 Now SfaiT is clearlysymmetric,hencewe can apply (30:G) in 30.3. 1 W, Lare disjunct sets. They have the samenumber of elements owing to (48:A:b)
exhaust

7 which has 2n elements. HenceW as well asL has

the

53.1.)

- 0.

-Sa

-S

30.3.2.

i.e.

S Q

30,3.5.

30.3.2.

system of all non-empty subsetsof 7). This rids us of S

S /)

()

.)))

ENUMERATION OF ALL SIMPLE GAMES

447

51.2.2.ordero discuss(49:W*) on this basis,we must take (49:W*:c) t In also into account. This can be done in two ways. The first way will be useful for a subsequentcomparison.
(51 :B)
W fulfills

tains neither

(49:W*) if and only if it is (Ri-saturated and connor any one-element set.

Proof: (49:W*) is the conjunction of (49:W*:a),(49:W*:b)and (49:W*:c). The first two amount by (51 to (Ri-saturation. Taking :A) for (49:W*:a) granted,(49:W*:c) be stated thus:If S is / oran (n 1)may elementset, then S is not in W. I.e.: either nor any one-element N

set is in

W.
Q

Thesecond is more directly useful. way Let V be the system of all sets of (49:W*:c) i.e.f 7 and all (n o element ubsetsof I. Thenwe have: s

- 1)u VQ

(51 :C)
fills

V is a subset of a is (Ri-satisfactory.

W fulfilling

(49:W*) if and only if V

(Ri-saturated set. Now we know that (30:G) of 30.3.5. applies, and hencethe considera1 tions of the last part of 30.3.5. too. This extensabilityis equivalent apply to the (Ri-satisfactoriness V u VQ. of We rephrase :C)more explicitly: (51

fulfilling (49:W*) amount to this: *V,W fulW (49:W*:a),(49:W*:b) i.e.W is (Ri-saturated by (51 :A) W fulfills W V for (49:W* :c) i.e. 2. Q. In other words:We arelooking an (Ri-saturated W 2. u VQ i.e. e are asking whether V u VQ can be extended an V w to

Proof:Wz. V and W

51.2.3.

(51 :D)

is a subset of a

8 be disjunct.

sessesheseproperties: t No two S, T of V aredisjunct. (51 :D:a) 2 nor V containsneither set. (51 :D:b) any one-element W to according (51 the (Ri-satisfactorinessof :C), Proof: e must express, V u VQ. I.e.hat no two S,T of V or VQ aredisjunct. t w S,T areboth in V: This coincidesith (51:D:a). h S, T are both in VQ: Both have ^ n 1elements, encethey cannot

W fulfilling

(49:W*)

if

and only

if

it

pos-

Of S,T oneis in V and the otherin W We may assumeby symmetry, S as the former and T as the latter. Soan S of V must not bedisjunct with

/ or any (n

set. :D:b). l)-element Thisis precisely(51

1Note that the domain (cf.footnote 4 on p. 446)is finite. 1For this cf.alsofootnote 4 on p. 446. 1We are using that 2(n 1) > n This should have beenstated n ^ n >2 sets but it is a natural assumption, sincesimple games (i.e. explicitly at the beginning with (49:W*)) exist only for n (Cf.49:4,49:5.))))

!>-/-()
i.e.
3.

i.e. 3.

448
which

GAMES SIMPLE

we then have such a W. among the subsetsof the W (with (49:W*)) in In performing this gradual building up process all possibleways, we obtain all W in question. The reader ay try this for n = 3 or n = 4. It will appear that the m i even for small n, although it is rigorousand procedures quite cumbersome exhaustive for all n.

(51 solvesthe question of1enumerating all W: Startingwith any V :D) it fulfills (51:D:a), (51 .-Bib) we may increase gradually as long as When this can be done without violating (51:D:a), (51:D:b). this process cannot be continued any further, than we have a V which is maximal

i.e.

1In principle we may start with the empty set. Thereaderwill note that sion of from V (cf above)doesnot affect the possibility of V 1 V is the only system which can possibly meet theserequirements, W it may

has the desired roperties: p

of 49.6. these We wish to characterize W m directlyand to find somesimpleprocess to construct them all. In what follows, we will derive two different ways both of characterization, being of the saturation type. The first will be by will means of an asymmetricrelation,while the second be by a symmetric one one. Thus it is the second which is suited for constructionpurposes,in analogy with the constructionof W in We give neverthelessboth characterizations because equivalenceis the T s instructive: he first one is in some (technical)espectsimilarto the r quite and and 30.3.7.), therefore the transition definition of a solution (cf. to the equivalent secondform is of interestsinceit points a way to solve how problemsof this type. We have mentionedbefore (in 30.3.7.) desirable the corresponding transition for our concept f solution would be. o Let W be a system which contains all supersets its elements: of W * :b) Then the systemof its minimal elements m deterW fulfilling (49 e.g. i mines W: Indeed,t is clearthat W is the system of the supersets all of elements f W m o if Hence a system V is given, and we arelookingfor a W with (49:W*) such that V = JFm , then this W must necessarilybe the system V' of the of o supersets all elements f V. ConsequentlyV = W m for a W with (49:W*) if and only if thesetwo requirementsare met by W = V. 2 We are now goingto transform this of characterization the V = W m into one of the saturation type. Denote the assertion that neitherS n T = nor S D T, by S&iT. Then we have: V = W for a W with (49:W*) if and only if V is (H saturated (51 :E) and contains neither nor any one-element set. A Proof: ccording to the above,we must only investigatewhether W = V
m

Let t 51.3.1. us considerhe setsW

Reasonsfor Passingfrom 51.3.

to

W.

Difficulties

of Using W~

51.2.

30.3.3.

61.3.2.

: .

I.e.
fail.)))

the but

exclueven

ENUMERATIONOF ALL SIMPLE GAMES

V

449

Let S be a minimal element f this W. Then 8 2.T for some o T of V. HenceT7 is in W, and so the minimality of S excludes D T. So S S = T i.e. belongsto V. S Thus only the converse property must be discussed: Whether every S of V is really minimal in W. Any S ot V clearlybelongsto TF. So the that S D T 2 T, minimality means that S D 7\", T' of W is impossible; T of F is impossible. This impliesthe impossibility SD T, T of 7 and of is impliedby it (put T7 = T). Sowe have this condition:
Wm
7

i.e.

'

:2) (51

Never S D T for S,T in

F.

TF fulfills (49:W*): e must consider(49:W*:a), W (49:W*:c) (49:W*:b), separately. We do this in a different order. Ad (49:W*:b): s Clearly W = F containsall supersetsof its elements o this is automatically fulfilled. Ad (49:W*:c):Take (49:W*:a)for granted. (Cf. below.) Then (49:W*:c) be stated thus:If S is I or a (n l)-element et,then S is may s not in TF. I.e. either nor any one-element is in TF; that is, no subset n set of theseis in F. So we have this condition: set (51:3) Neither nor any one-element is in F.

S', S'cannot both belongto W: I.e. S,T belongto F, then we canT not have S&S', S'. Now the existencef such an S' implies o S n T = and is impliedby it (put S' = S). So we have this condition:
if

Ad

We (49:W*:a): considerthis in

two

parts:

(51 :4)

Never S n T

for S,T in

F.

One of S, S must belongto TF:Assume that neither of S, S belongs the lattermeaning to W. This means that no T of F has T c S or T c n Sn T = I.e. o T of F has T = S or S D T or S n T1 = Or again: S is not in F and no T of F fulfills the negation of S(R2 I.e. is not in F, but S(R8 !Tfor all T of F. S that Now we have to express this is impossible:

-S,
r.

i.e.:

(51:5)

If S(R2 for all T of

F,then S belongsto F.

TofF. I.e.:
(51 :6) Hence and this

we arethe Thus (51:2)-(51:5) criteria want. like this: &zTfor all S, can S Now (51:2) (51:4) be stated together and
S(R2T for all T of

F,if S belongsto F.
(Rj-saturation of

and p togetherexpress reciselythe (51:5) (51:6)

wanted to prove. :B). (51 is of someinterestbecauseit is a perfect analogue of (51 :E) of of Thus thesecharacterizations W and TFm differ only in the replacement
1This is indeed Sz>TorSnT-Q.)))

(51:3)

F.

form the criterion

and this is preciselywhat we

450 by

SIMPLE GAMES S<RiT:not S n T = Q

S(R2

r:neither S

nor S D T.

be used in (51 :A).

But as this replaces symmetric(Ri by the asymmetric(Ra, (51 cannot the :E) or the way in which we used (51:B) rather the underlying

is this: very S of V is a minimal element E belong to W but its proper subsets must not of i.e. belongto W. As W fulfills (49:W*:b), containsthe supersets all its it sufficesto statethis for the maximal proper subsets of S only; elements, w f i.e.or the S (i), i in S. As W fulfills (49:W*:a) e may say instead of = (-AS)u (i) is in W. So S (i) not belongingto W, that (0)

T We now turn to the second procedure. his consistsin analyzwhat does it mean for a W G ing the following question: iven a system (49:W*) that VsWm ? fulfilling

51.4.1.

61.4. Changed Approach :

Enumeration

by

Meansof W m

F,

The meaning of the

of W.

I.e.such an S must

Wm

we see:

-(S-

(51 :F)

cW
V,

S of

belongsto
We now

means precisely this: or every F (W with (49:W*)) u S belongsto W ; and for every i of this S, (-S) (i)
m

W.

prove:
V

(s51:G)

is a subset of the No two S, T of

of

Wm

of a W

(51:G:a) (51:G:b) (51:G:c)

only if it

these possesses properties: No two S,T of V aredisjunct.

fulfilling

(49:W*)

if

and

(51:G:d)

s element et.
Neither
to V.
Vi

For S, T

V have SD T. F, S u T = / implies that S n 7

is a one-

nor any one-element nor 7 must belong set,

that means by (51:F), 7u 7i W. This is possible for someW with t accordingo (51 if V u Vi fulfills (51:D:a), :D), (51:D:b). (49:W*) for Let us therefore formulate (51:D:a), (51:D:b)V u Vi. :S, w Ad (51 :D:a) T7 areboth in V: This coincidesith (51 :G:a). S Tareboth in Vii I.e. = (-S')u(i), = (-T')u(j),8' in V, tin t 7\" S,

L Proof: et

be the set of all

(-5) (i), Sin i in S. Then c

u V, V

JF\" 1

fl',j'inl\".

i i.e. in T'. S o Summingup: 'u T' = 7;i,j two different elements f both S'and T' o i.e.f S'n
I\".)))

T'disjunct, .e. u T' = 7; i S' Thedisjunctnessof S,T meansthese: i i.e. 7* j\\ S',(j) disjunct,i.e. in S';-T', (i) disjunct, j (i), (j) disjunct,
7

-S',-

ENUMERATIONOF ALL SIMPLE GAMES

451

S'

u T' = 7, then Now we must state that this is impossible. I.e.f i r n T cannot possess two different elements.As n T' cannot be empty this set. by (51:G:a), means that it must be a one-element Thus precisely(51 T obtains (S', 'in placeof its S,T). :G:c) Of S, T one is in V and the other in Fr. We may assumeby symmetry, 1 that S is the former and T the latter. So T = ) u (j), T' in V, in The disjunctnessS, ( T') u (j) means: , T1 are disjunct, S S T'\\ S (j) aredisjunct, i.e. not in S. o S Summing up: S T', an element f T'not in S. Now we must state that this is impossible. I.e.not S c Thus obtains. (T7 S in placeof its S, T.) precisely(51:G:b) Ad set Neither nor any one-element must belongto V nor to V\\. The lattermeansthat neither must be a ( S) u (i), S in F, in set Only a one-element couldbe such a ( S) u (i) and this would mean:

S'

S'

T'.

(-T

j i.e.

(51:D:b):

',

T'.

i S.

S -S= 0, i.e. = /.
Summing

to V.

desired. (51 solves the problem of enumerating all W m in perfect analogy :G) to the solutionby (51 of the corresponding :D) problemfor the W: Starting 1 we increase gradually as it with any V which fulfills (51:G:a)-(51:G:d)
When this long as this can be done without violating (51:G:a)-(51:G:d). c processannot be continued any further, we have a V which is maximal we have such a W m among the subsets of the W m of a W with (49:W*) of In performing such a gradual process buildingup in all possibleways, we obtain all W m in question. e Our last remarks show that the practical numeration of all u can :G) and we will, indeed, ndertake it in simplegames be based on (51 out arebettercarried at first. 52. But someotherconsiderations We now proposeto analyze the assertionthat (51 is a conditionof the :G) saturation type a little more closely. Observefirst, that as (51:G:b) to two arbitrary S, T of F, we refers t w can interchangehesein it. I.e. e can replace51:G:b) this: ( by

set, up:Neither nor any one-element nor I must belong w This coincidesith (51 :G:d). Thus we have obtained preciselythe conditions (51:G:a)-(51:G:d) as

i.e.

51.4.2.

:G (51 :b*)
t i.e.hat

No two S,T of F have S D T or S c T.

7

Denotethe assertionthat S, T fulfill (51:G:a), (51:G:b*), (51:G:c) nor S D T , nor S c T, nor S u T = / without neither S n T = set S n T being a one-element by S(RjT. states that F is (Rs-satisfactory, together with Then (51:G) simply Now (51:G:d). let the domain D be the system 7 of those subsets of / i.e. nor a one-element nor I. Then which fulfill (51:G:d) neither set,
the last remarks of subsets of 7.

51.4.1. that the W arethe maximal (Rs-satisfactory show

start with the empty
set.)))

0,

1In principle we may

452
S<R8 T

This gives:

SIMPLE GAMES 1 is clearly symmetric. Hence may apply we

V

(30:G) of

30.3.5.

(51 :H)
(in from

W m for a W with

(49 W*) if and only if V is (R 3-saturated

7).

in :E) :H) Comparing(51 with (51 showsthat we have succeeded passing the asymmetric(Ra to the symmetric (Rs fulfilling the promisemadein

footnote

51.4.3. quite instructive to compare(R* (in 51.3.2.) our It is with

1p.271.

(R 8

nor S a T. S(R2T:neitherS n T = S(R3T:neitherS n T = , nor S => T, nor S T, nor u T = 7 without S n T7 beinga one-element

0,

set.

the One can infer from this how recondite operationsmust be by which out if this proves to be feasible the program of 30.3.7. might be carried at

others.

would Mere symmetrization of (R 2 (cf.30.3.2.) give the threefirst parts of (Rs, but not the last one. This last part is the essential chievement of a in :H) (51 and (51 and not connected any obvious way with the three :G)

all

61.5. Simplicity

and

Decomposition

game Assume,therefore, that F is a decomposable with the constituents H (J, K complementsin /). Then we must answer this question: What does it mean for A, H that F is simple? we t We beginby determiningthe setsW, L. Since must considerhem for all three gamesF, A, H, it is necessary to indicatethis dependence.
A,

betweenthe concept f a simple o 61.5.1. us considerthe connections Let gameand that of decomposition.

We write therefore Wr,

It shouldbe added that

A,

L?\\

ization for the gamesF, all in a zero-sumform. 2

we assumeneitheressentialitynor any normalH. It is convenient, however, to assumethem

W^ LA;

Wu

Z>H.

(51:1)

T fi K) belongs to W? [Lr ] if (R R belongsto TT [LA ] and T to W H [LH A

S=RuT

J,

].

and only if

T by their respective complements(in

of Proof: eplacement S by its complement(in R

/, K).

This transformation inter-

/), / S,3 replaces , R

S S S /

1 And StovS holds in 7 n = occursonly for , S => never, U henceneither of thesehappen for an of only for 2 Thereaderwho recalls the discussionsof may want to know at this point how the question of the excesses (in A, H the e , , ^ loc.cit.) is to be handled. This question will be clarified in the discussion of 8 It is preferableto write the complement in this way, instead of the usual T, sincewe are complementing in different sets.)))

:S S

r,

46.10. 51.6.

S S 7.

-S,-ft,

ENUMERATION OF ALL SIMPLE GAMES

y

453

our changesWr, W Wn with Z/r, I/A, Z/ H Hence statementconcerning the W impliesthat one concerning L and viceversa. We are going to the prove the latter. That S belongsto Z/r is expressed by

(51:7)

v(S) =
t in

v((0)

sinceA, H arethe constituentsof T, we have v(S) = v(ft) we can write (51:7)hus: t

+ v(T). Hence

(51:8)
That
ft

v(fl)

+ v(T) = S t in
vGR)

ft

<) + 27 v 0).
t in

belongsto LA and T to LH is expressed by

(51:9)

(51:10)

v((t)). v(T) iinT

tin
ft

S v((0),

The assertion which we must prove, then, is the equivalenceof (51 and :7)
drawn sincealways

(51:10). (51:9), the Clearly (51:9), (51:10) (51:7); reverseimplication imply

v(fl) ^ X y (W)>
i in ft

can be

v(T) ^ J)
in (cf. (31:2)31. 1.4.).

V (W)

in 71

51.5.2. arenow able to prove: We

A,

(51 :J)

F is simpleif and only if this is true:Of the two constituents H one is simple,and the otheris inessential.

T Simplicityof F means this: Proof: he condition is necessary: (51:11) Forany S / one and only one of thesetwo statementsis true:

(51:ll:a) (51:ll:b)
results: : (5112)

S is in W r . S is in Lr .

Put S = flu T (Rz J, TsK) and apply

is statements true:
R R

Then to (51:1) (51:11).

this

For any

two R

J, TfiK one and only one of thesetwo

.)))

(51 :12:a) :12:b) (51

is in W* and T is in W tt . is in LA and T is in LH

454

GAMES SIMPLE
7

Now put R = , T = K. Then R belongsto LA and'7 belongsto WV for Hence (51 W& :12:b) :12:a) and LA have a common element: and for (51 R, T. By (49:E) in 49.3.3. WH and LH have a common element: (appliedto a A, H instead of its r) the former implies that A is inessential, nd the latter,that H is inessential. So we see: (51:13) If F is simple,then eitherA or H is inessential.

tial.

The condition is sufficient: We assume,by symmetry, that H is inessenThen (49:E)in 49.3.3. of (appliedto H in place its F) showsthat every T K belongs to both WH and Z/H. Hence can now reformulate the we
of the (51:12) simplicityof F.
R

characterization

(51:14)

(51:14:a) (51:14:b)

For any true:

R R

J one and only

one of thesetwo statementsis

is in W*. is in LA .

This is preciselythe statementof the simplicityof A. So we see: then (51:15) If H [A] is inessential, the simplicityof F is equivalent

[H]. the (51:13), togethercomplete proof. (51:15)

to that one of A
Simplicity and Composition.

This is not true for simplicity,which as we know, is the simplest form of :J) essentiality: y (51 a decomposable is not simple if its two conB game stituents are. (51 shows that a simple gameA remains simple under :J) compositionif and only \"if it is combinedwith an inessentialgameH i.e. \" with a set of dummies (cf. footnote 1 on p.340). In this connectionour further remarks areappropriate: f above from the conFirst:If the simple game F obtains as described stituent (simple)gameA by an addition of \"dummies\" of the inessen(i.e. tial game H), then the solutionsof F aredirectly obtainablefrom thoseof A. l in this Indeed, is described detail in 46.9. that Second: stated at the beginningof 49.7. we use the old form of We the theory for simplegames. It is therefore worth noting that the type of compositionto which we were led (cf. the above remark) is preciselythe one for which the old form of the theory is hereditary. (Cf. the end of or 46.9. (46:M)in the first remark in 46.10.4.)
1Thisis, of course, hat common senseleadsone to expectanyhow. The surprising w turns of the theory of decomposition cf. in particular the resum6 in show, p however, that it is unsafe to losesight of the exactresults. In this case46,9. rovidesthe

while to compare :J) with (46:A:c)of 46.1.1. (51 There we found that a decomposable is inessential if and only if game its two constituentsare i.e.nessentialityis hereditary under composition. i

61.6. Inessentiality, It 51.6. is worth

Treatment of the

Excess

46.11.

firm ground.)))

ENUMERATION OF ALL SIMPLE GAMES

455

Third:In this connection becomes it clearerlso why we had to refrain a from considering than otherexcesses zero the new form of the theory in the senseof 44.7. for the theory of simplegames. If we had been able to carry this out successfully,then the

i.e.

results of 46.6. 46.8. of and would enableus to deal with all compositions of simplegames. Now we have seenthat a composition simplegamesis not a simple game. In other words: A theory of simplegames with general excess ould indirectly embraceon-simplegamesas well. It is therefore w n 1 not surprisingthat we could not proceed generality. in Fourth:In the light of the analysisof 46.10. above remarksconcernthe a ing the excessssume the following significance: They show that the con2 cept of simplicity doesnot stand the generaloperation of imbedding. cannot This showsthat the methodicalprincipleconsideredin 46.10.5. be appliedunder all conditions.
A 51.7. Criterion

Indeed:

To when tacklethe converseproblem: decide decomposable.

We now

51.7.1.51.5. discussedwhen a decomposable F is simple. In we game

a simple game is o setof all significant elements f /

of Decomposability

in

Termsof W m

Let a simplegameF be given. It will appear that the following concept is of importance: i of / is significant if and only if it belongsto some S An
of W m .* Denotethe
W by 7 in We now proceed

S in

(51 :K)

severalsuccessive steps: If F is simpleand decomposable, if the simpleconstituent and is A (cf. (51 and the use of notations of 51.5.) F and A then :J) have the same W m .
to

i.e.he sum of all t

T (RzJ, K) of W r and any T of WH- H is inessential (by obtain by taking any R of W& (51 hencethe T of WH aresimply all T K (cf. the proof of (51 :J)). :J)), it belongs to Wf if its Consequentlythis S = R u T is minimal R, T areminimal. This means that R belongsto TFj and that T =

Proof:According

(51 the S = R u :1)

Ts

i.e.

Thus W and TFj coincide, F and A have the same W m . i.e. With the same assumptionsas in (51 :L) (51 :K), necessarilyJ 2 7 . m F Proof: and A have the same W (by (51 hencethe samesignificant :K)), elements therefore thoseof F, which form the set7 , areall among the participants of A, which form the setJ.
1In a certain sensethis may be viewed as an application of the methodical principle referred to in footnote 3 on p. 270. 1Unlessit is merely an addition of \"dummies\" asdiscussedbove. a 1Thus a player i is significant if there exists a minimal winning coalition to which he essential ervice may render. s he belongs; if there exists a conceivable of It will beseenthat the opposite this is a \"dummy \" (cf.the end of All this refers, of course, o simple games.))) t

S = R.

0,i.e.

i.e.

51.7.3.).

456

GAMES SIMPLE
1 Assume only that F is simple. Then 7o is a splitting set, the Jo-constituentA beingsimple,and the (/ 7 )-constituent

(51 :M)

H inessential(cf. (51 :J)).

a C Proof: onsider n S = R u T, R 7 , T I (51 :16) S is in W if and only if R is in W.

Then:

If Indeed: R is in W } then SZL R is too. Conversely:Let S be in W. m Then a minimal TmW with T S exists. So T is in TT , every of T is in 7 HenceT 7 Thus T7 S n 7 = -R, and therefore fl is in W along with T.

s .

(51:17) TisinL. Indeed: ReplaceS by T(sI

As

is in L, (51 We now prove:

for :16) permitsto infer the same T.

o ); this replacesur

72,

T by

Q, T.

(51:18)
v(S) =

v(S)

- v(ft)+v(r).
t in

Considerthe S of L and of TV separately: S is in L:/2, T c 5arealso in L. Hence

t in

^S v((t)) =

t in A

X v((0) +

ST v((t)) = v() + v(D,

i.e. 51:18). (
S in
W:

By (51 (51 f :16), :17)l is in IF and T in L. Hence))

v()
and so

t not in

2; t not in

v((o>
ft

--

t not in

2)

v S

(^) -

i in T

2)

v(T) = 2) v((0), iin T v(S) = v(R)

+ v(T).
statementthat /o is a splitting set. For all

i.e. 51:18) ( is (51:18)precisely the TzI h (51:17) gives

v(D

- 2) v((t)),
in

hencethe 7 7o constituentH is inessential. Consequentlythe 7o-constituent A must be simpleby (51:J). Thus the proof is completed. 'In the senseof 43.1.)))

51.7.2. We
(51 :N)

SIMPLE GAMES FOR SMALLn 457 are now able to describehe decomposibilityof a simple t

game F completely senseof 43.3.

With

i.e. e can name its decompositionpartition Up in the w : i .

the sameassumptionsas in (51 :M) Thedecomposition n r consists of the set 7o and of the one-element sets partition for all in 7 7 (i)

A Proof: ll (t), i in / 7 , belongto n r :By (51 7 7 is a splitting :M) setof F with an inessentialconstituent . Hencevery (i), i in 7 7o, is a H e and splitting set of H (use,e.g.43:J)in 43.4.1.) so of F (use (43:D) in ( 43.3.1.). a one-element (i) is necessarilyminimal. Henceit Being set,

r belongsto II . 7o belongsto
T

No further J belongsto II : Any other J of II must be disjunct with r r and with all (i), i in 7 7 , (use(43:F) in 43.3.2.). the sum of these As sets is 7, this would necessitate but is not an elementof n r J (cf. the beginningof 43.3.2). Thus the proof is completed. 1 51.7.3. with Combinationof (43:K)in 43.4.1. (51 :N)gives: if (51:0) A simplegame F is indecomposableand only if 7 = 7, i.e. if and only if all its participants aresignificant. We conclude y proving: b
7

tollr.

J= .

, then

(51 appliesto J or to 7 :L)

Both exclude J

7 lip:

is a splitting set by (51 :M). If J is a splitting set

c7 .

it Thus 7 is minimal. Hence belongs

J, henceeitherJ 2 7

or 7

J 2 7o,

(51 :P)

A simple gameF possesses preciselyone/-constituentwhich is simpleand indecomposable: with J = 7 . That

T c :M). Proof: he 7 -constituent an be formed and is simpleby (51 Now consider simple/-constituent.Thenit has,by (51 the same a :K), W m and the same as significant elements F itself, hencethe latterform the set 7 . So the indecomposabilityof the /-constituentis by (51 :0) equivalent to / = 7 . A We callthe 7 -constituent of F its kernel. All otherparticipants of 7 7 are \"dummies.\" i.e.those (Cf. (51 :M) or (51 and the :N), in Hence that matters in the gameF takesplace all last part of 43.4.2.). it sufficesto apply the first remark in 51.6. A ; to see its kernel this, f 52.the SimpleGamesor Smalln
o Role. Disposal f n = 3 for of is Our next objective the enumeration all simplegames the to v smalleralues of n. We propose push this casuistic analysisso far as is referred to in 50.2. footnote 2 on (cf. necessaryto producethe examples
No

52.1.

2 52.1. Program:n 1, Play

* Or more directly of

(43:K)in

with 43.4.1. (51:L),(51

:M).)))

468

GAMES SIMPLE

(cf. 2 (cf. p.434), 50.7.2., footnotes 2,3,4 on p.443), 50.8.2. footnotes 1, , on p.445). w Sinceevery simplegame is essential, e needonly consider ameswith g n 3.

is simpleand it has the symbol[1,1,1 1]*. So we can assumefrom now on that n ^ 4. < f Sets 52.2. Procedureor n ^ 4 : The Two-element and Their Rolein Classifying the W m all n ^ 4 be given. We wish to enumerate simplegames 52.2.1. an Let to introducea principleof to with this n. In order do this it is convenient further classification of thesegames which is very effective for the smaller

The Forn = 3 the situation is this: (unique)essentialhree-personame t g

values of n. The enumeration in question amounts to the enumeration of the sets that one of available e.g. W m for which we have various characterizations

Considerthe smallest sets which may belong to W m . Since(51:G:d) sets c the loc. it.excludes empty set and the one-element from W m , this means consideringthe two-elementsets in W m . Thesesets possess the following property: A two-element et belongsto W m if and only if it belongsto s (52: A)
JT.2

(51 in :G)

4.1. 51.

Proof:The forward implication is obvious. Now assume conversely, subsets of S are that the two-elementset S belongs to W. The proper empty or one-element hencenot in W. Therefore S belongsto W m sets, to s We proposeto classify according two-element etsin W m 52.2.2. ConceivablyW m may contain no two-elementsetsat all. We denotethis possibilityby the symbol Co. Thenext alternative is that W m contains preciselyone two-elementset. , n we can make this setto be (1,2). By a permutation of the players

We

by the symbol C* Now 52.2.3. assume that W m doescontain further two-elementsets. Assumefurthermore that not all of them contain 1. Considertherefore a two-element set of W m not containing 1. By loc. a (51:G:a) cit.it must have common elementswith (1,2)nd (1,3) 1 thesemust be2 and 3 so the setmust be(2,3). beingexcluded,
1Cf. (50:A)in
*

loc. they must have a common element. two of these. By (51:G:a) cit. a permutation of the players , n we can make the common By t o elemento be and the two otherelements f thesetwo sets2 and a SeW m contains (1,2)nd (1,3). We denotethe possibilitythat W m contains no further two-elementsets

denotethis possibilityby the symbol Ci. Further, W m may contain two or more two-elementsets.Consider

1,

1,

1,

3.

a I.e. non-minimal

and 50.1.1.the last remark of 50.2.2.

set in

W must

have at least three elements.)))

SIMPLE GAMES FOR SMALLn

Thus (1,2), (2,3)belongto (1,3),
Wm

459

(To this extentwe have perfect symmetry in 1,2,3.) Now considerany othertwo-element set which may belongto W m It cannot contain all threeof by a permutation of theseplayerswe can Now it must have arrangeit so that the setin questionfails to contain with (1,2)nd (1,3) 1 being excluded,hese must be common elements a t 2 and 3 so the set must be (2,3),but we assumedit to be different from (2,3)(among others). Thus W m containsthe two-element ets(1,2), s (2 (1,3), 3), and no others. We denotethis possibilityby the symbolC*. 52.2.4. remaining alternative is that W m contains other twoThe element etsbesides(1,2), but that they all contain s (1,3), By a permutation of the players4, , n we can maketheseplayersto be 4, , k + 1with a k = 3, ,n Thus W m containsthe two-element sets(1,2), (1,4), + (1,3), (1,* 1), and no others. We denotethis possibilityby the symbolC*. It 52.2.5.is convenient to brackethe cases , Ci, C* of 52.2.2. the t C and 1of 52.2.4. We Ck, k = 3, ,n together: then have the cases

1,2,3;

1.

- 1.
,n

1.

Ck, k = 0,1,

- 1.

m InthecaseCjknowTFcontaijasthetwo-elementsets(l,2), + l), ,(l,fc 1 and no others. By an additional permutation of the players 1, , n we can replace thesesetsby (l,n), , (k,n). Itis in this form that we aregoing to usethe case k = 0,1 C*, , n 1. Now Ck containsthe two-element sets(l,n), , (fc,n), and no others. Besides C* the only alternative is C* of 52.2.3. we will not these which

transform.

Cases C*_i, C*_i C*, 52.3.1.all thesealternatives three can be disposedof immediately: Of < C*, Cn -2, ?n-i. We discussthesein a different order. Ad C*: onsideran S 7. If S contains two or more of 1,2,3 (at C say t least) 1,2,hen Sa (1,2). 1,2) elongsto W, henceS does too. If S ( b contains one or fewer of 1,2,3, (at most) 1,then Sfi (2,3).(2,3) say W belongsto W, (2,3)to L, henceS to L too. Sowe see: consistsprecisely W Hence m consistsprecisely of those S which contain two or more of 1,2,3. is the for of the sets(1,2), (2,3). So (1,2,3) 7 of 51.7. this game. (1,3), 2 In other words: The kernel of the game under consideration a threeis its person game with the participants 1,2,3, W m consistingagain of (1,2), t (2,3). (1,3), As mentionedbefore for the last time in 52.1.his game has The the symbol [1,1,1]*. remaining n 3 players, 4, , n are \" \" dummies.
of the

52.3. Decomposability

* Namely
they exhaust

,n \\n, 2, 1Thesewere the two-element

Wm
completely.)))

(!' ?'}'\"'* i)' I/ 1, sets

cf. 28.1.1.
definition but

of Wm by

we have

now shown that

460

SIMPLE GAMES

So we see: T CaseC* is represented preciselyone game: he three-person ame g by of \"dummies.\" with the necessarynumber (n 3) [1,1,1]*,

set then S. If S has no further elements, it is the one-element (n), and then S a (t, n) . n so in L. If Shas further elements, t = 1, , 1, say
to

52.3.2. Cn -i:Consider an Case

Ss/.Assume first, that

i.e.

belongs

i Now this (i, n) belongsto W, hence doestoo. In otherwords:f n is in S, S w then S belongsto W, except hen S = (n). Applying this to S gives: if and only if If n is not in /S, then S belongsto W, when S does not,

1,n - 2]. obtains immediately from IT. It consists precisely of these S: (1,n), 1,n) and (1, , n I).1 It is now easy to verify , (n t that this gameis homogeneous nd normalized by a = 1. I.e.hat as = 1 a n for all the Sof this W . Hence e can write [1, w 2] . , 1, (cf.50.2.) So we see: CaseC -i is represented precisely one game:The n-person game by [1,- - , 1,n 2],. 62.3.3. C -2: Consideran S /. Assume first, that n belongsto S. Ad IfS has no further elements therthan possiblyn 1,hen S (n 1, ). t o n So Now (n 1,n) is not in JP , hencenot in W (by (52:A) in 52.2.1.)other Sis inL along with (n 1, ). UShas further elements, than n 1, n say i = l, , n 2, then Ss (t, n). Now this (i, n) belongsto W w henceS does If too. So we see: n is in S,then S belongsto W, except hen S = (n) or (n 1,n). Applying this to S gives:If n is not in S then S = (n) or (n 1,n), S belongsto W when S does i.e. and only not, - ,n = (1, i.e. S 2). ,n 1) or (1, Hence consistspreciselyof thesesetsS:All setscontaining n, except W (n), and (n 1,n); no set not containing n, except 1, ,n 1) and ( [1,
,
TFm
m

t W Hence consistspreciselyof theseS:All setscontaining n, excepthe smallest one (n); no set not containing n, exceptthe largest one, (1, , n 1). One verifies easily that this IF indeed fulfills the as requirements(49:W*). Also that this game can be described a weighted 1 having a common weight, majority game, all players 1, ,n 2 fold weight. I.e.his gamehas the symbol t while player n has the n

-S= (n), i.e. = (1, S

,n

- 1).

if

if

ment (49:W*).
Wm
f

(1,

, n

2). One verifies easily that

this

indeed fulfills the require-

Thenew fact is that the only further element of W m is (1, , n 1). Note that this last set is not a two-element set only because n ^ 4. of 1Thus the two-element setsin W m are (1,n), , (n 2, n) as it should be by Thenew fact is that the only further element of W m is (1, definition. , n 2). For n 4 this last set is alsoa two-element set, thereby falsifying the classof the C game. (Itbecomes *instead of C_2, Ct .) Hencethis class(C_j)is void, unless n
definition.
f

obtains immediately from W. It consists precisely of these S: (n n (1, ), 2,n), and (1, , n 2). So (1, , n 2,n) is the /o of 51.7. this game. for 1Thus the two-element setsin Wm are (1,n), n , (n 1, ), as it should be by

i.e.

5.)))

SIMPLE GAMES FOR SMALLn

(n

461

In other words: The kernel of the game under considerationis an l)-persongame with the participants 1, , n 2, n, its W m of (1,n), consisting again , (n 2, n), (1, 2). Thus , n
this is the case n -2for n C n players (replacing by n

[1,

So we see:
,

1,n

3]*.

1players the analogue of the caseCn_i for n 1!) discussedabove. Henceit has the symbol 1is a \"dummy.\" The remaining player n
3]* with one dummy.

game[1,

1 CaseCn-2 is represented preciselyone game: The (n by

1,n

l)-person

62.4.The Simple GamesOther than

Ck , k

52.4. results of 52.3. The deserve to be consideredsomewhat further and to be reformulated. We saw that for every I ^ 4 the homogeneous I 2]* can be formed.2 weightedmajority game of I players , I= it is the direct ajority game of three Then We can even form it for m So we will use it for all / ^ 3. participants If n ^ 4 then we can obtain a simplen-person game by forming this / 2]* for any / = 3, , , n and adding to it the necessary number of \" dummies/' was T The result of 52.3. this: his game with I = 3,n, and (for n ^ 5) n 1exhauststhe cases * Cn Cn-2. C The odd thing about this result is that thesevalues of I do not exhaust the full set of its possibilitiesI = 3, , n (cf. above). That is to do this for n = 4, 5, but not for n ^ 6. There remain the say, they 2 for n ^ 6. What is their significance? / = 4, ,n I The answeris this: onsiderthe game C 2]* (I players) , n and n ^ 4. The I \"dummies.\" Assume only I = 3, with n , 8 we W m consistsof (1, ), , 1 I). Hence I) and (1, 1 , (I 4 C* have case when I = 3 and case Cz_iwhen I = 4, , n. from Thuswehave in thesegamesspecimens the cases *,Cs, , Cn C T can Theresult of 52.3. now be formulated like this:he cases *, C_j,Cn C 6 areexhaustedby the pertinent ones among thesegames.

-[1, 0, 1,

1,I - 2]* (with n -3

,

Dummies)

:The Cases

[1,

[1,1,1]*. 1,

3:

[1,

1,

-i,

- 1,

[1,

1,

-i.
-i

We

this restate conclusion:

We wish to

(52:B)

The game

number (n I) of \"dummies\" is a simple n-person game for n. Its caseis C*, C3, all Z = 3, 4, , C*-i, respec-

[1,

enumerateall simple n-person games n ^ 4. 2]* (I players) with the necessary , 1,I

For n 6 it is void for n 4. Cf.footnote 2 on p. 460. 1Cf Case _! above,with I in placeof n. 3 We take , 1 2]*and , I as the participants of the kernel players of , n as \"dummies.\" This differs from the arrangement in case players / -f 52.3.where / n 1 and player n 1 was \"dummy\" by an interchange of players n 1 and n. 4 For / Ci 1)is in this cosea two-element set. 3, C*replaces since(1, 1Hence is void for n ^ 4, sinceit occurs the second but not on the first one. on Ci list, Cf, 52.3.)))
1

1,

1,

[1,

1,

C.i

,/ -

482

GAMES SIMPLE

cases Ci, Co,

i instancesn We 52.6.1.
will

lively.

All

other simple n-person games(if any) are in the

, 1 Cn-s.

n = 6,7. = 4 is easily settled. By (52:B) above, we need only investigate n s W Co,Ci for this n. In thesecases m contains g 1two-element ets.Howof a two-element etis a twos Since ever this is impossible: the complement s element et,theremust be the samenumber of two-element etsin W and in s h L. I.e. alf of the total number,which is 6. So W contains3 two-element setsand the sameis true for W m .2 Thus the only simplegames for n = 4 are thoseof (52:B). We state

52.5. o Disposal f n = 4, 5 discussthe values n = 4, 5 fully and somecharacteristic

this as follows: Disregardinggames which obtain by adding dummies to (52:C) 8 simplegamesof < four persons, thereexistspreciselyone simple four-persongame: 1,1,1,2]*. [ = 5. By (52:B) above we must investigate Consider next n 62.6.2. a concrete I = 0,1,2.n contrast to the n = 4 case, ll of these represent I

possibilities.

and of all setsof ^ threeelements, W m of all setsof threeelements.Hence this is the direct ajority game A. m i Ci:(1,2)s the only two-elementset in W m in W. Passing to the set the others are (3,4,5) complements: is the only three-element in L in W. Thus W consistspreciselyof thesesets:1,2), three-element sets ( all a but (3,4,5), ll four- and five-element sets.It is easy to verify that this fulfills (49:W*) and also that its W m consistsof the following sets: and (a,6,c), (1,2), where a = 1,2, 6, c = any two of 3,4,5. Now one shows without difficulty, that this game has the symbol

i Co: No two-elementset is in W m i.e.n W. So they are all in L and t their complements,he three-element areall in W. Thus W consists sets,

[1,1, 1,1,1]

i.e.

i.e.

[2,2,1,1,1]*. i W Passing d:(1,2), arethe only two-elementsetsin W , i.e.n setsin!/ i.e. (1,3) to the complements: a (3,4,5), 2,4,5) rethe only three-element ( the othersare in W. Thus W consistspreciselyof thesesets:1,2), ( (1,3), all three-element but (2,4,5),(3,4,5),all four- and five-element sets. sets It is easy to verify that this W fulfills (49:W*) and also that its W consists of the following sets: (1,2), (2,3,4),(2,3,5), (1,3), (1,4,5).
m m

Now one showswithout difficulty that this gamehas the symbol[3,2,2,1,1]*. 1All those caseswhich we succeeded exhausting so far were void or contained in preciselyone game. This is, however, not generally true. Of.the first remark in 53.2.1. 1Owing to (52: ) in 52.2.1. will beusedin what follows continuously without This A
further

'I.e.

reference*. to the unique simple

g three-personame [1,1,1]*.)))

NEW

POSSIBILITIESSIMPLE OF GAMES

463

Hence simplegamesfor n = 5 arethesethree,and those of (52:B). the We stateas follows:

(52:D)
Disregardinggameswhich obtain by addingdummiesto sim1 four plegamesof < five persons, thereexist recisely simplefivep persongames: [1,1,1,2,2]*,' [1,1,1,1,3]*. [1,1,1,1,1]*, [1,1,2,2,3]*,'
of 53.The New Possibilities SimpleGamesor n ^ f 53.1. Regularities Observedfor n < 6 The 6

we have obtained so far, possesseda were homogeneousweighted majority , Wn]h, symbol, [wi, verified this for n = 4, 5, the question arises whether it is games. Having true. As stated in footnote 3 on p. 443, this is not so;the first always f comesor n = 6. counter-example So Second: far every class which containedany gameat all, contained C* one. This too fails from n = 6 on. (Cf. the first remark in only Third:One might think a priori, that thereis greatfreedom in choosing the weights for a homogeneous weightedmajority game. Our lists show, are One that the possibilities very limited: eachfor n = 3,4, and however, 8 four for n = 5. We emphasizethat sinceour lists are exhaustive,this is a rigorously established objective fact and not a more or less arbitrary peculiarityof our procedure. Fourth We can verify the statement of footnote 1on p.446 that while in the number of elements W is determinedby n (it is 2n~l)> that one oi m for simplegames of the same n. This phenomenon begins W may vary for n = 5. W For n = 3 W has 4 elements, m in the unique instancehas Foi = 4:W has 8 elements, m in the unique instance4. Forn = Whas W n W 16 respectively. elements, m in the four instances 10,7,5,5, Fifth: We can verify the statement of footnote 2 on p. 443, that the m 50.6.2. equations (50:8)of 50.4.3., (with U = W ) may be more numerous a system ol than their variables, and neverthelesspossessa solution imputations in the ordinary sense. The former means that W m has > r for the weighted majorit} elements, latteris certainly the case homogeneous

lists.

53.1. we go further, letus draw someconclusionsrom the above Before f

First: All simple games which

i.e.they

53.2.1.)

3. 5:

i.e.

:K) games((50 in 50.8.1.). We saw above that for n = 3, 4

for n
1

4 w homogeneous eightedmajority games.

= 5, it

may have
to

a 10or 7 elements s well.

Wm

b necessarilyhas n elements, ul And all thesegames an

increasing arrangement of weights. 3 Disregarding permutations of the players! 4 Thus we have the first counter-examples for n game) and [1,1,1,2,2]*.)))

* We permute the players of these games (belonging to Oi and

t I.e.o [1,1,and IK

[1,1,1,2]*.

C2) in order to have ai

5:[1,1,1,1,1]* majority (the direct

464
in

SIMPLE GAMES

53.2.5.

w Fora simplegame, here thesesolutionsdo not exist, f. the fifth remark c

thesen.

= would be rather voluminous. We forego it for this reason. instancesof simplegamesin n = 6, 7 We will only give somecharacteristic which illustrate certainphenomena which begin as mentionedbefore at
even of n

T 5S.2. he Six Main Counter-examples n 6, 7) (for exhaustionof thesecases, We 53.2.1. now pass to n = 6,7. A complete

6,

that mentioned in the secondremark of 53.1. for n = 6, a it case may contain severalgames. Indeed, is not difficult to verify that Ck the two homogeneous eightedmajority games w

First: e W

[1,1,1,2,2,4],, [1,1,1,3,3,4],,
(cf. footnote 2 on p.463) aredifferent from eachotherand belongboth to C,. that 63.2.2. We mentioned in the first remarkof 53.1. for n = 6 Second: w a simplegame exists hich is not a homogeneous eighted majority game, w o i.e.ne which doesnot possessany symbol[u>i, , i0 ]*. By (50:K) in n this 50.8.1. is necessarilythe casewhen thereexists o main simplesolution; n i.e.o systemof imputations in the ordinary sense. (Cf.the fifth remarkin 53.1.)
n

It is possibleto find one which

it

not is alsopossibleto find one which does even have that property. We begin with the first mentionedalternative. Put n = 6:Define W as the systemof all thosesets8 I = (1, , 6) have ^ 4 elements), r which eithercontain a majority of all players (i.e. o have 3 elements), a majorwhich contain exactly half of all players (i.e. but (i.e. ity of all the players 1,2,3 ^ 2 of these). In otherwords:The players form 1,2,3 a privilegedgroupas againstthe players4,5,6 but their privilege is rather limited: ormally the overall majority wins; only in caseof a tie N doesthe majority of the privilegedgroup decide. It is easy to verify that this W satisfies(49:W*). Thegameis clearlya I weighted majority one: t suffices to give the members of the privileged someexcess eight over those of the others (4,5,6),which w group (1,2,3) must beinsufficient to overridean overall majority. Any symbol
[w, w,
with w,

Sucha game existsindeed,and it is even possibleto differentiate further: is neverthelessa weightedmajority game w i.e. hich possessessymbol[wi, a (without homogeneity!), , w n ], and

1,1,1]

1 1<w < 3 will do.

1w > 1is necessary, for 8 (1,2,4) defeat -S (3,5,6) 2u> + 1 > w + 2). to e.g. (i.e. t < 3 is necessary, for S (3,4,5,6)o defeat -S (1,2) i.e. + 3 > e.g. ( w
2u>).)))

--

--

NEW
Wm

POSSIBILITIESSIMPLE GAMES OF
(1,2,3)))

465

is quicklydetermined; consistsof thesesets: it

(Si):
(fli)))

(S'/):

(a,&,/0

(Si\:")

(o,4,5,6)

where a,b = any two of 1,2,3, h = 4 or 5 or 6 where a = 1or 2 or 3.

l))

The equations (50:8)of 50.4.3., (with U = W m ) 50.6.2. are: mine a main simplesolution in the senseof 50.8.1. /ET/\\. /. I X} -j- X$ = A / /X\\ \\&i). 0,

which

deter-

(Ei):
f

xa

+ x + xh = 6,
b

-f\"

(E'i ) :

xa

+ z 4 + z 5 + c = 6,
b &

= where 0,6 any two of h = 4 or 5 or 6 where a = 1or 2 or 3

1,2,3,

2 = These equations(Ei)cannot besolved. Indeed, with a = 1,6 2 (EC) and /i = 4,5,6 shows that x 4 = x = x ; (EC')with a = 1,2,3 shows that hence(EC)gives 4 + x 4 = 6, x\\ = 2 = x 8 ; now (J) gives 3^i = 6, Xi = 2; x 4 = 2;and then (#\"') gives 2 + 6 = 6 a contradiction. It shouldbe noted that the ordinary economicspect f this occurrence o a would be this:Si') (i.e. showsthat the servicesof players4,5;6 can be (EC)) ( substituted for each other hencethey are of the same value. (Si\ N (i.e. E{\ shows the same for 1,2,3. ow comparisonof (Si) and (Si') (

showsthat one player of the group 1,2,3 be substituted for one player can of the group 4,5,6 and comparison (S'/)and (Si\") howsthat one player of s of the former groupcan be substituted for two playersof the latter. Hence no substitution ratebetweenthesetwo groups can be defined at all. The natural way out would be to declareomeof the setsof W m enumeratedin s to be \"no profitable uses\"of the players' services. In the senseof (Si) and 50.4.3. amounts to choosingU c W m (Cf. also 50.7.1. footnote 4 on this Whether in this game a U c W m can have the requiredproperties p.443). couldbe decidedby a simplebut somewhat lengthy combina(cf. 50.7.1.) o out. The existencef torial discussion,which has not yet been carried it can be shown that it would have such a V is highly improbable,because
if it mathematically unlikely characteristics existed. It This gameis also very peculiarin another respect: is possibleto prove that there existsno solution V which contains only a finite number of the imputations and which possesses full symmetry of the game itself; a invariance under all permutations of the players 1,2,3nd under all permutations of the players 4,5,6.We do not discussthis rather lengthy 1 proof at this place. Thus the type of solution which onewould term the natural onedoes not exist.

i.e.

1Thus W m has 1 + 9 + 3 13elements. 1They are 13equations in 6 variables, but this in itself is not necessarilyan obstacle, as the fifth remark in shows. 8 Whether We suspectthat even any finite solution V exists at all, is not known. this question will be anwered in the negative.)))

53.1.

466

GAMES SIMPLE

This is an indication of how extremelycareful one must be in terming them. extraordinarysolutions\"unnatural/'or in trying to exclude 63.2.3. Let us now considerthe secondalternative referred to Third: in the second remark above: simplegame for n = 6,which is no majority A at all i.e. hich has no symbol [u>i, w game , w*]. This alternative itself can be subdividedfurther:It is possibleto find a game such that it a possessesmain simplesolution (cf. above) and it is also possibleto find one that has no main simplesolution. Considerthe first case: Put n == 6. Define W as the systemof all thosesetsS(fi 7 = ( 1, ,6)) o which eithercontain a majority of all players (i.e. have ^ 4 elements), r which contain exactly half (i.e. but have 3 elements), an even number of the players 1,2,3i.e. ( have or 2 of these). Comparingthis with the example T in the second remark above, this observationmust be made: he players still but 1,2,3 form a group of special significance, it would be misleadingto
from the tying (i.e. call their significance a privilege sincetheir absence set three-element) S is just as advantageousas their strong representation of o (presencef preciselytwo of them),and the presence all of them just as o disastrousas their weak representationpresencef preciselyoneof them). ( in n Theybring about a decisionot by their presence S but by an arithmetical 1 relation:
2 that this W fulfills (49:W*) in 49.6.2. us now determine m . SinceW contains all ^ four-element sets, W Let no ^ five-element setcan be in W m. Considernow a four-element setin W. If the number of players 1,2,3 it is even, remove from it a player 4 or 5 in 1 in or 6. If the number of players 1,2,3 it is odd, remove from it a player 1 4 subset with an even number of or 2 or 3. At any rate a three-element o o 1,2,3btains i.e.ne in W. Sono four-element setcan be in W m . players W Hence m consistsof the three-element in W. Theseare: sets

It is easy to verify

(S\:")

(Si):

(4,5,6)
(<*>>b,h)

where a,6 = any two of 6 h = 4 or 5 or 6.

1,2,3;
S.

1Note alsothat the group 4,5,6has a similar significance:SinceS must have three elements (in order that these criteria becomeoperative), the statement that an even number of is in S is equivalent to the statement that an odd number of 4,5,6is in This lends further emphasis if any beneeded to our frequently made observation forms of socialorganization, and the concerning the great complexity of the possible extreme wealth of attendant phenomena. 1Note in particular that always oneof S and Sbelongsto W: This is evident if one of the two has ^ 4 elements (and so the other 2s 2). Otherwise both S and S have 3 elements. Henceoneof them contains an even number of players 1,2,3and the other an odd one. 1This is possible, s a are only 3 players. 4This is a possible, s 4,5,6are only 3 players. 1Thus W has 1 + 9 elements.)))

1,2,3

1,2,3 - 10

NEW

OF POSSIBILITIESSIMPLE GAMES
, w n] then
9

467

If this game had a symbol [wi,

Wi

therewould be

i in 8
Apply this

>
t not in

Wi

for all S in W.

to the

setsof W m enumeratedin (Si). This gives in particular:

+ Wi + Wj u>i + w Wj + W
Wi u>i

+ W* > Wi + Wi + w > w* + Wi + MS > Wi + wi + Wi > Wi +

W>5

+ Wj, + MS, + toe, + W*.

W*

Adding

theseinequalitiesgives:

2(wi + w 2 + u> 8 + W* a contradiction.

mine a main

+ W* +

WB)

> 2(wi + ti?2 +

+ w 4 + Ws + 6),
ti>

The equations (50:8)of

simplesolution on the other hand are:

x4 xa

50.6.2. 50.4.3., (with

U=

W m}

which

deter-

EQ:
Ey):
They

+ x* + x< = 6, + x + xh = 6,
b

where a,6 = any two of 1,2,3; = 4 or 5 or 6.

In the ordinary economic terminology one would have to say that the structural difference betweenthe groupsof players 1,2,3 4,5,6 cannot be and expressed weights and majorities,and that as far as values areconcerned, by thereis no difference. 63.2.4. Fourth:Note that the above exampleis alsosuited to establish the difference between the homogeneous weighted majority principleand the existencef a main simplesolution, as discussedin 50.8.2. o Indeed, = x = 2 (cf. it is an instance of = in (50:21) cit.: inceZi = loc. S s above), o Fifth: 53.2.5. Now

areobviously solved by x\\ =

1 = x e = 2.

t-i

Xi

= 12= 2n.

consider the second casedescribed in the third remark above:A simple game for n = 6, for which neither a symbol
[Wl,

',

Wn ]

nor a main simplesolution exists. Compared with the two previous examplesgiven in the secondand third remark above this oneis based on less transparent principles. This a is not surprisingsincenow all our simplifying criteria reto be unfulfilled.

This is the example: Put n = 6. Define W as the systemof all thosesets / = (1, S( ,6)) o have ^ 4 elements), r which contain eithera majority of all players (i.e.
1It is easily seenthat
this

is their

only

solution.)))

468

SIMPLE GAMES
or (1,3,4)(1,5,6)* W satisfies(49:W*)in 49.6.2. be determined without serious difficulties. It turns out to
8

which contain exactly half have 3) elements, nd fulfill a further condition: Either containsplayer but it is not

S 2' o or S is (2,3,4) r (2,5,6). It is easy to verify that this

Wm

(i.e.

1,

the following

consistof thesesets:

can

(Si): (S' : z') : (Si\")

(1,2,6) (l,a,6)
(2,p,g)

(SJO:
2)

0,4,5,6)'
>
i in

where 6 = 3 or 4 or 5 or 6 where a = 3 or 4, b = 5 or 6 4 where p = 3,q = 4, or p = 5, q

,
w n ],

4 = 6.

If this gamehad a symbol[w\\,

Wi

then there would be

i in 8
Apply

-8

Wi

for all S in W.

this to the setsof W m , enumeratedin (S3). This gives in particular:))

W2

w*

+ + W 4 > Wi + + Wt, + MS + w 6 > Wi + Ws + W4.

W*
Wf>

Adding

thesefour inequalitiesgives:

2(wi + Wi

+ W* +

WA

+ ws + w 6) > 2(wi + wz + w 3 +

W*

+ w& + 6),
u>

a contradiction.
(E't):

The equations (50:8)of 50.4.3., (with U = 50.0.2. mine a main simplesolution on the otherhand are:

W m)

which

deter-

+ xi + x = 6, Xi + x + x = 6, (E't): : + xp + x = 6, (#i\") (#'/): 0:3+ 0:4+ 0:5+ 0:6 = 6.

x\\
b

X<L

where 6 = 3 or 4 or 5 or 6, where a = 3 or 4, b = 5 or 6, where p = 3,q = 4, or p = 5, = 6,

8 Theseequations (1?) cannot be solved. Indeed 3 (/?')showsthat o:3 = x 4 and xs = x, hence (E'\") gives x 2 + 2x8 = 6, x 2 + 2x6 = 6, therefore 3

1 it is (l,a,b)with a 2, 6 3 or 4 or 5 or or with a 3 or 4, b 5 or 6. 1The complements of the previously excludedsets(1,5,6) (1,3,4). and 1If this last exception concerning (1,3,4), and (1,5,6) (2,3,4),(2,5,6) were omitted, then W would be defined by this principle:Theplayer 1is privileged normally the overall majority wins, but ties aredecided player by this caseis even It is easyto verify that this is simply the game example in the secondremark above simpler, than our in some ways, analogous sincethe privilege existing here has a numerical value in the conventional sense. and Thus the complicating exception concerning (1,3,4), (1,5,6) (2,3,4),(2,5,6) is in decisive bringing forth the real characterof our example. 4 Note that a,6 vary independently of eachother, while p,q do not! elements. Thus PF w +l 1They are equations in 6 variables, cf footnote 2 on p. 465.)))

I.e.

6;

1.

I.e. [2,1,1,1,1,1]*.

10

has4+4+2

- 11

NEW x9

POSSIBILITIESSIMPLE OF GAMES
3

469 ^a = f Finally

= a? 4 = z 6 = * Now (E*v ) &ves *x * = 6 whence (#','), #',\") ieldxi + 3 = 6, z 2 + 3 = 6,i.e.l = x 2 = 3. x ( y 3 (JEJ) becomes+ 3 + f = 6, a contradiction.
B

and so

As to the interpretation of this insolubility, essentiallythe same comments arein order s at the corresponding a point of the secondremarkabove. 63.2.6. We have already referred to the difference between the Sixth: o homogeneousweighted majority principle, and the existencef a main in This simplesolution,as discussed 50.8.2. wasdone in the fourth remark loc. w above, where an examplefor = in (50:21) cit. as given. We will now

give an

whether an exampleof the desired kind exists n = 6 the one which for will be given has n = 7. Put n = 7. Define W as the systemof all thosesetsS( / = (1, ,7)) which contain any one of the 7 following three-element 1 sets:

Sincewe found that for n ^ 5 all simple games were homogeneous w weightedmajority games, e must now assumen <t 6. We do not know

for loc. example > in (50:21) cit.

(S4 (1,2,4), (3,4,6),(4,5,7),(5,6,1), (7,1,3) (6,7,2), (2,3,5), The principleembodiedin this definition can be illustrated in various
ways. This is one:The 7 setsof (SO obtain from the first one (1,2,4) by all cyclicpermutation. I.e. y increasing its elements any one of the by b numbers 0,1,2,3,4,5,6 all three by the sameone provided that the but 2 areidentified numbers8,9,10,11,12,13 with 1,2,3,4,5,6 respectively. In other words: They obtain from the set marked x x x on Figure89, by any one of the 7 rotations which this figure allows. Another illustration:Figure 90 shows the players 1, , 7 in an in arrangement which it is feasibleto mark 7 setsof (SOdirectly. They are indicated by the 6 straight linesand the circle 1 O. The verification, that this W fulfills (49:W*) is not difficult, but we in if prefer to leave it to the reader he is interested this type of combinatorics. W m consistsobviously of the 7 sets of (SOIt is easy to show along the lines given in the third and fifth remarks above that this is not a weightedmajority game. We omit this discussion. 50.6.2. The equations (50:8)of 50.4.3., (with U = W\") which determine a main simplesolution on the otherhand are: r (#0: xa + x + xc = 7, where (o,6,c)uns over the 7 setsof (SO= Xi = 4 They areobviously solvedby zi =
b

):

i.

1Thus Wm has 7 elements. f In the modulo 7. terminology of number theory Reduced 1Thereaderwho is familiar with projective geometry will note that Figure 90is the the so-called point plane geometry. The seven sets in question are its 7 picture straight lines, eachone containing 3 points, and the circleO alsorating as such. Oneshould add that other projective geometriesdo not seemto be suited for our present purpose. 4 is easily seenthat this is their only solution.)))

o!

It

470))

SIMPLE GAMES))
in Indeed:)) > holds in (50:21)50.8.2.

Now we can establishthat

i-lx*

= V- > 14 2n.

t As the games discussedin the second,hird and fifth remarks,this one too correspondsto an organizational principlethat deservesclosertudy. s In this game the setsof W m , the decisivewinning coalitionsarealways minorities (three-element sets).Nevertheless,no playerhas any advantage show that any cyclicpermutaover any other: Figure89 and its discussion o tion of the players 89 , 7 i.e.ny rotation of the circle f Figure a in leaves the structureof the game unaffected. Any player can be carried 1 this manner into any other player's place. Thus the structure of the game))

i.e.

1,

Figure

89.))

Figure

90.))

2 not of is determined by the individual properties the players all are,as we saw,in exactlythe sameposition but by the relation among the players. It is,indeed,the understandingreached among 3 playerswho arecorrelated 8 a (/S4) which decidesbout victory or defeat. by

64.Determinationof All Solutionsn Suitable Games i

Our 64.1.1. discussionof simplegames thus far placed most emphasis k the specialind of solutionsdiscussedin 50.5.1. and particu-50.7.2. upon on the main simplesolution of 50.8.1. the basis of what we have On larly
from the examples 53.2.this of learnedin the previoussections especially t does appearo do justiceto all aspectsf our problem. not o approach

54.1. Reasonsto ConsiderOther Solutions

than the Main Solution in Simple

Games

1The game is nevertheless not fair in the senseof sincee.g.he two threet and act element sets(1,2,4) (1,3,4) differently: Theformer belongsto W, the latter to L. form of the game, with y the v(S) of the former is 4, and that of (Soin the reduced the latter is 1Which the rules of the game might give them. exist 'There in this game no significant relations between any two players:It is t possibleo carry any two given playersinto any two given onesby a suitable permutation (of all players , 7) which leavesthe game invariant.)))

-3.)
1,

- 1,

28.2.1.,

ALL SOLUTIONSN SUITABLEGAMES I

471

Tobegin with, we have seenthat we cannot expectll simplegamesto a have solutionsof the type mentioned. Already for n 6 a wealth of new possibilitiesemerged.This is significant, since 6 is a sizeable number from the point of view of combinatorics, ut a small one when viewed in b the context social rganization. of o But further, even when thesesolutionsexist,ndeed i even for the homogeneous weightedmajority games,hey do not tellthe whole story. Forthe t most primitive specimen of that class,the essentialthree-person game which as we know has the symbol there existmany solutions. And our discussion 33. howedthat they are all essential or our underin f s and standing of the characteristics the implicationsof our theory actually somefundamental interpretationswere first obtained at that point. 64.1.2. Consequentlyit is important to determineall solutions of a simplegame and, as long as we arenot able to do this for all simplegames, to do it for as many simplegames as possible. particular this should In be done for at leastone simplegame at eachvalue of n. Suchresults would provide some information about the structural possibilitiesand principles of classification solutionsfor n* participants. of It is true that this information would be equally welcomeif it couldbe obtained for otherthan simplegames. Howeverthe simple gamespossess a manifest advantage over all others when solutions are to be determined systematically:For simple games the so-calledreliminary conditions of p causeno difficulties (cf. sincethere every set S is certainly necessaryor certainly unnecessary(cf.49.7.). It is equallytrue that the determinationswhich we envisagewould only a provide information concerning few isolated cases. But they would neverthelesscover all n enableus to vary n at will. This is bound to lead to essentialnsights. i

[1,1,1]*,

30.1.1.

31.1.2.),

i.e.

54.2.Enumeration of ThoseGamesfor Which All Solutions Are Known 64.2.1. us take inventory of the casesfor which we already know Let all solutionsof a game. Therearethree: by complemented (31:1) :P) (a) All inessentialgames(cf. (31 in 31.2.3.,
in

in stituents areknown. (Cf. (46:1) 46.6.) we can use the device(c) to combinethe games providedby (a) Clearly 1 and (b) thus obtaining games for which all solutions are known. In \" \" this processf building up (a) furnishes only dummies (cf. the end of o hence we may well dispense with it, since we want structural 43.4.2.),
the

game both in the old theory (excess (b) The essentialthree-person for zero) and in the new one (generalexcess).(Cf. 32.2.3. the former -47.7. and the analysisof 47.2.1. for the latter.) games provided that all solutions of the con(c) All decomposable

31.2.1.).

1This can alsobe expressed the following way: in A given game F is the compositeof its indecomposable constituents, according to a :E)eod. definition of the decomposition partition at the end of 43.3. nd (43
We)))

472
information.

iterated applicationof (c) to (b). In this way we can obtain gameswhich 1 arethe compositeof essentialhree-personames. t g 64.2.2. gives n = 3fc-persongamesfor which we know all solutions. This Sincek is arbitrary, we can make n arbitrarily great. Tothis extentthings are satisfactory. Howeverthe fact remains that such an n-person game
is just a polymer of the essential hree-personame the players form in g t reality setsof 3 which the rules of the game fail to link to eachother. It the is true that our results concerning solutions of decomposable games show that a linkageof thesesets of players is neverthelessprovidedfor in the typical solution i.e. y the typical standard of behavior. But natub rally we want to seehow the ordinary type of linkage,explicitlysetby the t rules of the game,affects the organization of the players i.e.he solutions or standards. And we want this for greatnumbersof players. Consequentlywe must look for further n-persongames for which it is possibleto determineall solutions.

GAMES SIMPLE Thus we are left with those games which are obtained by

n t 64.3.Reasonsto Considerhe Simple Game[1, - , 1, 2]* As 64.3.1.pointed out2 above, we aregoing to look for thesespecimens among the simplegames. Now it turns out that thereis a certainsimple out. game for every n ^ 3, for which this determination can be carried This is the only n-persongame,of a general , for which we succeeded thus n far in the generaldetermination.This obviously gives it a position of interest. We will also seethat it permitsinteresting interpretations special in several respects. The game in question has already occurred 52.3. in (52:B) of in and n 52.4.It is the homogeneous eighted majority game [1, , 1, 2]*, w (n players). that the sets of participants into which this partition subknow from (43 in 43.4.2. :L) divides them, are setsof 1 or ^3 elements. The simplest possibility is therefore that they are all one-elementsets. According this means that the game is inessential it takes us back to the to (43 in :J) case(a) above. Thenext simplest possibility is that they are all one-or three-element sets. These are exactly the games which we can form accordingto (c) from (a) and (b). it is for thesethat we know all solutions. This is satisfactory sinceit shows that -a classification based on the sizesof the o indecomposableconstituents (i.e. f the elements of the decomposition partition, cf. is :L) (43 in 43.4.2.) a natural one:Our progressin obtaining all solutions follows precisely the lines drawn by it. h alsostressesow limited theseresults are: is indeed a very special ccurrence o at when a game is decomposable all. (Remember the denning equations of (41:6) or in The typical n-person accordingto the criterion at the end of (41:7) and by game is indecomposable cannot bereached means of (c). 1By application of strategic equivalence we can assume them all to be in the reduced form. But denoting their y by 71, , 7* respectively, we cannot expectto make them all equal to 1 by a changeof unit (unless 1). Indeed,their ratios 71: yk are unaffected by changesof unit. * For this reasonwe use the old theory, excess zero. Cf the third and fourth

43.4.1.

i.e.

I.e.

It

It

41.3.2.,

42.5.2.!)

A;

remarks in 51.6.)))

i.e.

THESIMPLE GAME [1,

1,n - 2],

473

are theseS:(1, ), n

wins as soonas he finds any ally at all, but if he remainscompletely isolated 1 then he loses. This result invites someremarks: First: hestatementof this rule suggests trongly that player n is in a T s He privilegedposition: needs only one ally to win, while the others need each other without exception. ctually the situation is this: layer n A P needs a coalition of two, the others togetherneedone of n hencea 1> 2, n ^ 4. privilege exists only if n For n = 3 there is, indeed, no difference between the three players: We have then the game the unique essentialthree-person game which is obviously symmetric. Second: privilege of player n is as extensiveas a privilege can be: The We required that n must find at leastone ally in order win and it would to 2 not have beenpossibleto require less. It is impossibleto specifythat n can win without an ally, i.e.o declarehat the one-element (n) to be t t set this is incompatiblewith the essentialityof the game. (This was winning

54.3.2. discussedin 52.3. this game the minimal winning coalitions As in , (n - 1, ) and (1, n , n - 1). I.e. layer n p

i.e. [1,1,1]*,

1,

discussedextensivelyin

49.2.)

The determinationof all solutionsof the gamewhich we discussed above will show that they fall into a complexarray of classes, exhibiting These create an opportunity for the widely varying characteristics. interpretations we have alluded to previously. We will discusssome of along the same line will probably follow in them, while further discussions
subsequentinvestigations. Theexact erivation of this complete of solutionswill be given in the d list This derivation is of not inconsiderwhich follow (55.2.-55.11.). sections able complexity. We are giving it in full for the same reasonsas the the games analogousone concerning solutionsof decomposable in Chapter IX.:Theproof itself is a convenient and natural vehicle for certaininterpretations. It presentsat several stagesan opportunity to bring out verbally the emerging structural features of the organizationsunder consideration. will In fact this circumstance be even more pronouncedin the proofsof this than in thoseof Chapter IX. chapter
The Chief Player. Cases and (II) (I) to we After thesepreliminaries proceed the systematicinvestigation of the game 2]* (n players). Assume that it is in the , n reduced form, normalized by y = 1 As set every one-element must. 1We stated abovethat player n is not at all privileged in this game when n 3 and now we state that he is as privileged as he possibly can be! Yet n 3 is no exception from this statement:Sincethere exists only one essential three-persongame,

66.1.

65.The SimpleGame [1, , 1,n 55.1. Preliminary Remarks

- 2]

55.2. Domination.

66.2.1.

[1,

1,

1.

the)))

474
We

GAMES SIMPLE begin with an immediate observation on domination

:))

(55:A)

For

{i,
a
and
a<

an },
ft

{j9i,

'
=

A.},

if and only if

either

(55:1)
(55:2) n (1, ),

or

> Pn oti > ft

> ft
i

for somei

for all

= 1,
,n

1, , n - 1.

,n

1,

T since w Proo/: his coincidesith (49:J)in 49.7.2., W m consistsof the sets

, (n
n

- 1,
a =

n)
n

and (1,

- 1).

Note that
of

-i

-ift

= permit us to infer from (55:2) validity the

ctn

(55:3)

Hence:
(55:B)

< ftn.
i

a ^

ft

necessitate an ^

ft*.

Proof: y symmetry we needonly considera s-< /3 . We saw that this B or henceat any ratean 5^ ft. implies (55:1) (55:3), These two results, simple as they are, deserve some interpretative
comment.
We 1

discussedin 54.3. the player n has a privilegedrolein this that He is in a situation which is comparableto that of a monopolist, game. with the inescapable limitation (cf. the second remark loc. it.) c that he must find at leastone ally. I.e. generalcoalition of all others against a him but nothing lessthan that can defeat him. We will call him the 2 chief player in this game. 55.2.2. Thesecircumstances brought out clearly in (55:1) (55:2). and are One may say that (55:1)the directorm of domination by the chief player is f and an arbitrary ally (any player i = 1, may be ,n 1) while (55:2) a termed stateof general ooperationgainst him. (55:1), or (55:B) (55:3) c a show that in a domination the chief player is certainly affected: Advan(the tageously in (55:1) directform of domination with the chief player), adverselyin case(55:2) thegeneral ooperationgainst the chief player). ( a c 8 i Any other player can be unaffected, left aside,n a domination.
o the position in which a player finds himself there may as well becalled bestpossible ne sinceit is the only onethere is. 1Exceptthe casen 3, about which more will be said later. 1As to the casen 3, the end of the first remark in 54.3.hould be kept in mind. s 1 I I.e.t may happen
actually for an w only possible hen n

4,

-cf1, .

again

1 that ,n the observations

>

>

H ft and a< concerning n

0<. This is
3.)))

THESIMPLE GAME [1,

1,n

2]*

475

55.2.3. considera solution V of this Now

Max-*
mV
n n

1 game. Form

. a = w, Min~> . a = tf. 2 , in V
Clearly
~~

1 ^ tf ^

w.

The meaning of #, o> is plain:They representthe. worst and the best for possibleoutcome the chief player, within the solution V. We distinguishtwo possibilities: (I)

(II)

= w,

< w.
for all

i.e.that
within

65.3. o Disposal f Case (I) Consider .55.3.1. the case(I). This means that an = J>, (55:4)

in V

the chief player obtains the same amount under all conditions the solution. In other words: (I) expresses the chief player that in is segregated the game in the senseof Consideringthe central roleof the chief player it is not unreasonablethat the first alternative 8 distinctionin our discussionshould proceedlong this line. a

33.1.

Let 55.3.2. us now discussV in case(I).

V is preciselythe set of all a fulfilling (55:4). 1In the senseof the old theory, cf footnote 2 on p. 472. 1That these quantities can be formed, that the maximum and minimum exist and are assumed can beascertained the same way as in footnote 1 on p.384. Cf in particin

(55:C)

i.e.

ular (*) the

this procedure analogous to that one of is essential three-persongame. that the essential hree-person This will appeareven more natural if it is recalled t t caseof the one we consider ow pertaining to n 3. (Cf.e.g.he n game is a special end of the first remark in 54.3.) Closerconsideration of the case n 3 shows, however, that this analogy suffers from a rather unsatisfactory limitation: In this casethe game is really symmetric, and so the chief player. (Cf alsofootnote 2 any one of the three players could have beencalled on p. 474.) In t the segregation in question was indeed applicableo any one of the three players,and now we have arbitrarily restricted it to player n! Yet there is no way so far to apply this to the other players too if we want our discussion to coverall n ^ 3 (and not only n 3) For n 4 the chief player and his role are unique. The only sensein which this situation can be accepted temporarily is that of keeping in mind that case(II)must in fine turn out to bea compositeone. which is analyzed in Thus for n 3 comparison with the classification of shows this: Our case(I)is one of the possibilities of (32:A)there:discrimination the other two possibilities of against player 3. Our case (II),on the other hand, covers

loc.cit. *Thereferenceo 33.1. t that re-emphasizes

33.1.

33.1.

32.2.3.

solution.

(32: discrimination against A): So (II)is really an

This schemewill,

55.12.5.)))

aggregate of 3 possibilities when n indeed, generalize for all n. Cf. (e) in

players

1,2together

with

t (32:B),he

non-discriminatory 3.

the fourth

remark of

476

SIMPLE GAMES
>

ft

ct

Thus V is determined easily enough, but we must now answer the G conversequestion: iven an w ^ 1,is the V defined by (55:4)(i.e.y b of a solution? I.e. oes it fulfill (30:5:a), d (30:5:b)30.1:1.? (55:C)) excludea *- ft for a , ft in V, hence(30:5:a) Now (55:B)and (55:4) of is automatically satisfied. Therefore we need only investigate (30:5:b) I.e. 30.1.1. we must securethis property:

W Proof: e know already that all a of V fulfill (55:4). If, conversely,a t fulfills (55:4), hen every a of V has an = 0n, hence(55:B) excludes H ft . Hence belongsto Vft

(55:5)
upon
w.

If

ftn

w,

then

ft

for some a

with

an = w.

W More explicitly: e must determine what limitations (55:5) imposes

Theftn

9* & of

(55:5)can be classified:
ft n
ftn

(55:6) (55:7)
We show first:

> u, < .

(55:D)

In the case(55:6)condition(55:5)is automatically fulfilled.

,*

A f Proof: ssume ft n > w, i.e.t n = & +

> 0.

Define))

by a< = fti

-=

71

\"\"*

1for t = 1,

,n

1,and a

= ft n

= co. a is

an imputation of the desiredkind with a H ft by (55:2). remains. Concerningthis case have: we Thus only the case(55:7)

(55:E)

Forw =

f Proof:tn ^

The possibilityw

(55:F)

l > 1is somewhat deeper. Assume > 1 and case (55:7). Then ris equivalent to w < n 2
o>

1,(55:7)is impossible. 1,hencenot < = 1.

ft n w

--

condition (55:5)

1 means that the chief player is not only segregated alsodiscriminated but (Cf. way. (by V) in the worst possible 1 gives a solution outright, while w > 1 necessitates more the Thus w This is not surprising: An extreme form of discrimination detailedanalysis of (55 is a more elementary proposition and requires lessdelicateadjustments than an interagainst

1<i

33.1.)

:F).

mediate one.)))

THESIMPLE GAME [1,

1,n

2]>

477

= o>, (55:3) 55.2.1. of is Proof:Assume ft n < w. For an a with i.e. excluded, domination a H ft must operate (and not through (55:1) in (55:2)!)(55: Sincean > n , this condition amounts merely to A). > fa for somei = 1, , n 1. (55:8)
/

Thus (55:5) o requires the existencef an imputation a

with

and

Consider first (55:8)for a fixed i = 1, 1. Then this con, n dition and an = o> can be met by an imputation a if and only if ft and add up with n 2 addends 1 to < 0. I.e.ft + <o (n 2) < 0, 2 W. Consequently(55:8)s unfulfillable for all i = 1, i ft < n ,
n_r-

(55:8).

an =

1,if and only

ft n

if

(55:9)

-2 -

J>

for all

i = 1,
no
ft

,n

- 1.
ft n ft n

that (55:5) xpresses this should happen for e no imputation means that n to
ft

with

> 0. I.e.(n <n

2
n

Combining (55:E),(55:F) and recalling(55:D) and the statements w made concerning55:5)and (55:6),55:7), e can summarize as follows: ( (

-r>
1))

1 addends n

- l)(n - - - 1 >0, - 2
w)

could have (55:9)togetherwith 1^ 2 co and one addend

n

J>

1 must add up > n ? p and so

1))

< o>. I.e. < w. 1 This

as desired.

(55:G)

Let w be any number

with))

n))

Form the

set V

of all

a
n

with

Thesearepreciselyall solutionsV
n

in the

case(I).

2 w gt 1, (55:4)unfulfillable by

1 We are assuming

>0.

that (55:9) ^ implies n Indeed w > n and n imputations:

1.

1 for i This means ,n 1 must be excluded sinceit makes 1 addends 1 would then add up to

1,

1.

33.1.
For n

the Therefore hypotheses of (55:F)imply this: & n 1 In pursuance with the parallelism with the discussion of the specialcasen 3 in to referred to in footnote 3 on p. 4Z5 wenote that this corresponds the cloc.cit.

1.

3 our

the r becomes

J occurring

there.)))

478))

SIMPLE GAMES))
n

The first values of the quantity

n)

2
4)

-- -

r are:))
5) 6)

3)

u n

n-1)

1)

2 -0.5)
n*)

5 3
Figure

1A7) 1 6r)

'

11 '
4)

27*) -6.70

19 IB) 5 3
8))

-'

91.

of This standard of behavior (solution) is based on the exclusion the chief player from the game. This makes the distribution between the other players quite indefinite i.e. ny imputation which gives the chief a amount J> belongsto the solution. The upper limit player the \"assigned\" 2 could also be motivated of the \"assigned\"mount o>, n a but we will not considerthis question. following the lines of
=

65.3.3. interpretation this result is not difficult: The of

--

33.1.2.,

55.4.Case(II) :Determination

of V

the last part of footnote 3 on p.475.)

We 66.4.1. now pass to the considerablymore difficult case(II). (Cf. We then

have))

disjunctsets:

This suggeststhe following decompositionof V into three pairwise

V, setof all a in V with an = w, V, setof all a in V with an = < an < o>. V*, set of all a in V with
a>, o>

V, By the very nature of co, o> (cf. the beginningof 55.2.3) V cannot be V 1 concerning *. empty while we cannot makesuch an assertion We 66.4.2. beginby investigating V.))

(55:H)
t

If

1,

belongsto V and
,n
1.))

ft

to V

n 1. Now Proof:Otherwise ft > c^ for a suitable i which is impossible, n = w, 0n > , so jSn > nj hence ft H a by (55:1), sincea , ft belongto V. 1V*
ia actually

=!,,
<x<

V*, then

for

all))

empty in the

caseconsideredreceding (55:V).))) p

THESIMPLE GAME [1,

Form
cti

- ,

1,n - 2]
- ,n

479

= Min^.

for

i = 1,

I.1

Now (55:H) gives immediately:

(55:1)

If ~0 belongsto V u V*, then = for all g ^ fa

i 1,
0.'))

- I.

We prove further

(66J)

n-l
&

o>

A 7 Proof: ssume that V a> + o? < 0. Then we can choose

i

n-1

>g

for))

=
>

1,

7
V

<v , l7l>

is not empty, choose a

, n

...
, n
\"v
> >

1, 7n =

with

2) i-i?

= 0,

forming the imputation

4 7n|.

all

i = 1,
a>

- 1,hence

from

by

V. Then (55:2)7 H

by
ft

fa (55:1) ^ As

a<

< 7< for

belongsto V,

this excludes7 from V.

Hencethere existsan a in V with a - 7 . If a belongs to V, then an = = 7 n , hence a ** 7 contradicts :B).So a must belong (55 to S/u V*.\" Now by (55:1), ^ a, < 7, for all t = 1, 1. a , n ^ * But both (55:1) (55:2)n (55 require since a H 7 that a, > 7, and i :A) for at leastone i = l, Thus we have a contradiction. ,n

Now the determinationof V can be completed V has preciselyone element:)) (55:K)

l.

1That thesequantities can be formed, can be ascertained the same way as in in

i.e. that

footnote

theseminima exist and are assumed 1 on p. 384. Cf. in particular (*)

loc.cit.

is stated there concerning V, is equally true for V, Jhe intersection of V with the closed set of the a with a* # 1 Note that this cannot be assertedfor the ft of V since0, may exceed minimum the
What

value

however, 1Cf. however, (55:12) below. 4 Note that by their definitions all henceall our 7* & ,n (t

g,. Cf.,

(55:L).

1,

- 1,

g<

-1,(i - 1, 1,
n).)))

- 1)

and

2 -1,

480

SIMPLE GAMES
\\cn,

Proof:Let a =

(55:10)

\"'--'

a_i, <*} be an ' \" 1. f r

,

elementof V. Then

by the definition of these quantities. Now ]?

n-l

-i

ou

and by (55 :J),

-i

oti
\"\"

from + w ^ 0,hence> is excluded

all inequalitiesof

(55:10).

Thus

(55:11)

Ia = y a* = w,
<

for

= *' '

' ' ' n ~ *'

,

i.e.

{ai,

an -i,a} =

{!,

an -i,a>).
,

is not empty, this is its unique element. 55.4.3. that as a = Note , an _i, w} belongs to V it is an imputation. So we can strengthen (55:J)to necessarily

So V can have no elementother than {ai,

an _i, wj. SinceV

f

{i,
a<

(55:12)
We

n-l

-i

+ = 0.

can also strengthen (55:1):

If \"^ belongsto V, then <* ^ fa for all

(55:L)

i = 1,

,n

1.
for (55:1),
ft

in V u V* this has been statedin (55:K) yields even ft = a,-. We concludehis part of the analysisby proving: t

For Proo/: )3

in V

(55:M)
A Proof: ssume w

= o > 1,i.e. >= 1+

-1.
n

,c

> 0.

Define))

by ft = a< + ^ ^
from

for

{=!,-,1,
n

and

^=

1.

is an imputation (cf. (55:12) above).

ft
V.)))

excludes < w, or equally (55:L),

THESIMPLE GAME [1,

for all

- 2k

481

Hence there exists an a in V with a H ft . By (55:L)on g a< < fa i = 1, - , n 1. But both (55:1) (55:2)in (55:A) requireand since a H ft that on > fa for at leastone i = 1, 1. Thus , n

we have a contradiction. Note that now (55:12) becomes

*

(55:N)

i-ia,
1).

= 1.

The essentialresults of this analysisare (55:K), (55:L),(55:M). They 1 can be summarized as follows: The worst possible outcome for the chief player is complete defeat Thereis one and only one arrangement i.e.mputation (value i one (in V) which does this, and for all otherplayersthis is the best possible
This arrangement (in V) is the stateof completeooperationagainst c the chief player.1 The readerwill note that while this verbal formulation is not at all it complicated, could only be establishedby a mathematical, not by a
verbal,
(in

V).

procedure.

55.5. (II) :Determination Case 56.5.1. arenow able to investigate V. We

(55:0)

of V

Consider an imputation for some i = 1, , n

to

>

= {0i,
o>.

n } with
>

&^

a,

V.

I and /3n ^

Then

belongs

in

i 1,

(55:A) must hold. As a is in V n ^ w ^ *, and this excludes 55:1). (55:L)a< ^ a, for all ( By = hencea ^ a ^ fa for at least onet = ,n ,n So we have a contradictionin both cases. and this excludes 55:2). (
f

A Proof: ssume that does not belongto Vor V with a H 0. Hence(55:1) (55:2)of

Then there exists an

1,

1,

1,

i.e.hat t

ai ^ n (55:P) A Proof: ssume that

-2
<

-2 - - 2) +
<n
(n

J>

for

i = 1,
o>

,n

- 1.
1,

for a

suitable i =

,n

- 1,
in the

a<

+ w < 0.
somewhat different

1All this applies, f course, o the case(II)only. t o 1This expressionwas alsoused in a related, but last part of 55.2.)))

sense

482
Then we can choose ft of j) fa
<
r

ife

a,-, ft n

> w with ]

= j-ift 0,forming

GAMES SIMPLE 1(j = 1, , n

n 1,j ^ i i.e.
}

2 values

the imputation

7 = I0i,
This
n
o>

, ft>}.

hence meets requirementsof (55:0), it belongsto V. But this the necessitatesn ^ by the definition of that quantity contradicting

> ^.

Now put

(55:13)
(55:14)

a*

= Min.i

-i

i-

Then (55:P) statesthis: a* ^ n

-2,n

w. 2

Denotethe set of all i(= 1,

1) with

(55:15) by S*. By its nature

(55:Q) (55:R)
(55:S)
V

a = a*

this set must have thesetwo properties: 1), S* is not empty. S+ e (1, ,n

65.5.2. continue: We
a* a* =
>

=n

-2-

a>.

a consists of these elements: where i rijns over all

{a\\,
n , a* _i, a* n

and where

a, = a*
1

for = f , for j = n,

| j

S*,

with

otherwise.

an Proof of (55:R) and (55:S) We beginby considering elementft of \\f. then (55:2)gives a H /3 since If ft < g for all i = ,n

1,

1,

both in V hencethis is impossible. So (55:16) ft ^ g ^ a* for some i = Necessarily

a = {gi,

an-i, w).

As

a belongs to

V by

(55:K), so a
,n

are

1,
,n

1.

(55:17)

ft

-1

for all

j = 1,
ft> with

- 1,; *i,

and since is in V, so

(55:18)
1 This time the minimum is formed 1 Cf. however, (55:R) below.)))

respectto a finite domain!

THESIMPLE GAME [1,

n

1,n - 2]k

483

Now % Pi

excludedfrom

y-i

i belongsto S*. And i.e.

= and by (55:14) (n 2) + a* + ^ 0, hence > is all inequalities of (55:16), Henceg< = a*, i.e. (55:17).
for for

otherwise,

j = i, j = n,

So we see:

ft

= a as defined above.
*

(55:19)
\"-*

Every

ft

of V is necessarilyan a* with

i in S*.

Now V is not empty, hencean

this a* is an imputation,

a i in V (i in S+) exists. Consequently hence 2}aj = 0,i.e. (n 2) + a* + w = 0.

This is equivalent to (55:R). Considerfinally any i of S+. Since(55:R) is true, we have

;-i

- - 2) + a* +
(n
^

a>

= 0.

Hence
y-i
even in

a} = 0,

i.e.a* is an

imputation. But } =

that a belongsto hence(55:0)guarantees

V. I.e.

'

=*,
a>,

= J>, a i is

V-

And

sincea* = n

(55:20)

Every

a i with

i in S* is an imputation and belongsto V.

and (55:19) (55:20)ogether establish (55:S). (55:R) was demont stratedabove. Thus the proof is completed. 55.5.3. Theessential esults of this analysisare (55:R), (55:S)together r with the introduction of the set S*. It is again possibleto give a verbal summary.1 The best possibleoutcome for the chief player assignsto him a certain to value o>. In order achieve this he needspreciselyone ally who can be a selectedt will from a certainsetS+ of players. This setconsistsof those 1 who areleastfavojed in the state of , n among the players 1, a cooperationgainst the chief player, referred to at the end of 55.4. complete 1makebetween Thus the arrangementswhich the players 1, ,n themselves, when they combine to defeat the chief player completely, success. where he achievescomplete his determine conduct in thosecases This \" interaction \" between fundamentally different situations is worth
1

All this

o t applies, f course, o the case(II)only.)))

484
1

SIMPLE GAMES

absolute oppositionagainst him. 2 The concludingremark of 55.4. oncerning contrastof formulation the c and proof applies again.

noting. It isalsoof interestthat the natural alliesof the chief player,when he aims at completeuccess, the leastfavored membersof a potential are s

55.6. (II) :a and Case

S+

in determined 55.4., the two parts V, V of V.1 It is 55.5. therefore time to turn to the last remaining part of V:V*. Let a be the set of all a with a = g, = a* for all i in S+. Then we have: VuV'sa. (55:T)

55.6. We

en only exclude

C Proof: onsideran a in V u V*. We must prove a = g< for all i in S+. Now at g a* for all i = 1, 1 by (55:L). Hencewe need , n

< g, when i is in S*.

*

For i in S+ form the a. of (55:S).It belongsto V, so a\\ = o>; a belongs to V u V*, so an < w. Henceai > an . Now a, < a< means * = (ft ;> hence a'H a by (55:1)nd this is impossible,since a a *, a both belongto V. set VfiG if and only if S+ is a one-element or a* = 1; (55:U) otherwiseV and a aredisjunct.
\"~

>

*i

i>

>

Proof:Consider an a

in V>
*

Then a = a* (from (55:S)),i in S+.

t Comparingthe definitions of a and & makesit clearhat this belongsto Q, i if and only if S+ has a unique element or , = 1.
1In we insisted on the influence exercised the \"virtual\" of existence an by of its belonging to a certain standard of behavior (solution) on all imputation other imputations of the same standard. Almost all solutions of n ^ 3 persongames A specific which we found can be used to illustrate this principle. to reference it was The present instance, however, is made at an early stage of the discussion, in particularly striking. 1Political situations to illustrate this principle are well known and in connection with them its general validity is frequently asserted. It is difficult to deny, however, that the casewhich can be made purely verbally for this principle is no better than that which could be made for a number of other conflicting ones. The point is that for the particular game socialstructure we consider at present, this and no other principle is valid. To establish it a mathematical proof of some complexity was needed. All purely verbal plausibility arguments would have beeninconclusive and ambiguous. The set S* is still unknown, although restricted by (55 The numbers are alsounknown, but restricted by (55:N). They determine a* (their , tin, The determination of theseunknowns minimum), w, are given by (55:M),(55 will be attended to later. Cf. (55 a (55:N')nd (55 Nevertheless,the form of V and of V has beenfound, and the remaining uncertainties are of a lessfundamental character.)))

4.3.3.

i.e.

25.2.2.

i.e.

-i

:Q).

:0')i.e. ( (55:L'),

:R).

:P')).

THESIMPLE GAME [1,

1,n

2]*

485

The verbal meaning of (55 (55:U)is this: ach player of the least E :T), favored group (S*, cf. the end of 55.5.) reacheshis optimum 1 in every i apportionment in which the chief player is not fully successful(i.e.n in V u V*). When the chief player is fully defeated (i.e. V), this is even true for all players 1 (cf.the end of 55.4.). the chief When ,n t player is fully successfulhen this is true for one and only one player, who favored group (S+,cf. the end of may be any memberof the least

1,

55.5.).

65.7. ases(II')and (II\.") Disposal f Case(II') C o 55.7.1. the caseS+ = (1, , n 1),o be calledcase(IP). Consider t In this case = a* for all i = 1, n 1, o (55:N) ives (n l)a* = 1, s g g , = =n 2 i.e., 71 -> and (55:R) gives a* 1 1 If a belongsto Ct 71
o>
~~\"~

then

at = gt = a* = =
Tt

|
u

for
;

i = 1,

-=-

, n

1. Hencea

1,))

i.e.a
all

-t

7 in V

l|- By (55:T)this is equally true for

V*.

This a is clearly the unique elementa of V by (55:K), enceV* is h V empty. Hence = V u Vi and now (55:K), (55:S)give

(55:V)

consistsof theseelements:))

where

i = 1,

7' =
n
TJ 71

,n
(a\\,

1and where
-

with))

1
2) &

1))

j = i, for j = n,
for otherwise.)

in the case(II'). This that this V is eithera solution or in doesnot necessarilyimply, however, i case Indeed,f it failed to meetany one of thesetwo requirements, then we would only have shown although in a rather indirect way that

(55:V) determinesthe only possiblesolutionV

(!!')
1,

player

solution V. For the His individual optimum within the given standard 1)this optimum (maximum) is owing to (55:L) although i( ,n defined as his pessimum (minimum) in the part V of Hi was originally
1
< V*)))

i.e.

SIMPLE GAMES no solution in the case(IF) exists. We will prove, therefore, 1 requirementsare met.
486

that both

66.7.2.
(55 :W) The V of (55 is the unique solution in the case(II'). :V)
W Proof: e need only show that this V is a solution in the then follows from the above, i.e.rom (55:V). the uniqueness f Case(IF) is easilyestablished: Clearly,for this V

case(IF)

^=

1,

o>

=n

2))

S* =

(1,

,n

- 1).

to verify (30:5:c) prove that V is a solution, To this end we must determine the imputations ft which are in undominated by elements f V. o

It remains for us to
> >

i.e.

30.1.1.

For a H (55:1) excluded is sinceoj = 1. So this domination it can operate h ft a through (55:2) lone, ence amounts to aj > ft, i.e. < 71 - r 1 = 1, 1. for t ,n For a*H ft , k = 1,' , n 1, (55:1) excludedwhen i 9* k is and (55:2)is excludedsinceaf = 1 for i ^ k. So this domination can with operate through (55:1) i = k alone,henceit amounts to a* > ft for 2 p ft. < n j = fc, n, i.e.k <

Hence is undominated by elements V if and only if this is true: of ..holds for some i = 1, , n 1 and it holds even for all ;>

711))

thesei in case that

Thus /3n
ft,

-r>
n

- n -))

-))

<n

-<n
n
2))

1. HenceA ft 2)

-i
ft,

=
^
n

-1 = -* . i.e. a yields = for all these^ relations,

tt

r 8\\. 1 necessitates

, |8n _i

y-i

r for one 1 n and ft ^ 1for the othern 2 values of j. Hence *'(=1, , 1) n _ _ i ft ^ ^ ft = again yields = for all these ^ relations, .e. = a *. JJ
On the other hand 2
n n 1Cf. this situation with (55:G), where the case (I)was settled. No such secondary considerationswereneeded w and there, because (65:G) as ab wo necessary sufficient.)))

--

r ft 1 necessitates ^

ft

r- Also

THE SIMPLEGAME [1,

Sothe undominated by V are aand a a *~l ; i.e. recisely p the elements f V, as desired. o 65.7.3. solution is of importancebecauseit is a finite set as we This shall seeit is the only solution with that property. If the general oalition c
against the chief player is formed, the n 1participantsshare in it equally as describedy a . If the chief player finds an ally, he gives him the b same amount as a and retains the remainder as describedby

',,
,

1,n - 2],

487

7',this is ?-i. reasonable and perfectly

,
All

this is not the only possiblesolution we found another one in in w (55:G))and more will emerge the sections hich follow.

1 non-discriminatory. Nevertheless

55.3. (cf.

65.8.1. Consider next case(II\.

55.8. (II\") :a and V. Domination Case the caseS* 9* (1, , n

1),to

be called

Using (55:Q)we may also formulate this as follows: (55:X) S* c (1, ,n 1), S* not empty. We can also say: Cases (II')and (II\")are characterizedespectively r and by the presence discriminationwithin the possible of by the absence c general oalition against the chief player. As we enter upon the discussionof case(II\")the following remark is indicated: b -55.7.was mathematical, ut the (interThe argumentation of 55.4. mediate)results which were obtained thereallowed simpleverbal formulations. I.e.it was possible to work into the mathematical deduction relatively frequent interruptions, giving verbal illustrations of the stages attained successively. This situation changes ow, insofar as a longer mathematical deduction n where is needed carry us to the next point (in 55.12.) a verbal interpretato tion is again appropriate.

to now proceed give this deduction. V V = a n V (thepart of V in a). By (55:T), 55:11)= V u V* ( to or V = V u V* u V = V, according whether the condition of (55:U)is
Write

55.8.2. We

not or is satisfied.

(55:Y)

holds The condition(30:5:c) for V

in ft.

in (30:5:b)30.1.1. by Proof: eplace(30:5:c) the equivalent (30:5:a), R o elements f V cannot dominateeachother Ad (30:5:a): V V, Since is the because same true for V.
discriminatory 35.1.)))

1The specialcasesn 3, 4 of this solution are familiar: For n 3 it is the nonin 4 it was discussed solution of the essential three-persongame; for n

488
Ad
with

SIMPLE GAMES
Let (30:5:b): ft
H ft.
> >

in

Ct not

be in

V.

Then we must

find

an

in

is even not in V. Hence a in V with a H ft an exists. This a must be in V if it is not in S? (cf. the remarks preceding (55 :Y)),and this would establishour statement. Sowe needonly to exclude
To begin with,
ft

from

V. a is in

S+. We have a*H ft. (55:1) excludedwhen i ** k, and (55:2)is excludedsince is ak. = 1 for i j k (i = 1, , n 1). So this domination can only = fc, hence impliesaj > 0*, i.e. < * = a*. with it operate ftk through (55:1) i s However,this is impossible, ince ft belongsto Ct.
f

Assume that

_ V

* i.e.(by 55 :S) a = a*, k

>

in

domination in Ct.

in for (30:5:c) 30.1.1.) Ct.

56.8.3. our task is now to find Thus

For a,
in

all all solutions (i.e. setsfulfilling This necessitates determining the nature of

(55:Z)
a

> fti for someiin(I,

k

ft

Ct,

ft

is equivalent to
,n

this: n > a

ft n

and

1)

S+.

i is when i is in S*,and (S&:2) s Proof:For a H ft (55:1) excluded excluded sincea = 0*(=g* = <O for all k of S*. So this domination can only operatehrough (55:1) i in (1, with , t n And this means an > ft n and a, > ft, as asserted. 1) S*. We have replaced set of all imputations by Ct and the concept of the domination describedin (55: by that onedescribed (55:Z). Otherwise in A) the problemof finding all solutionshas remainedthe same. The progress is that the concept f domination in (55:Z)can be workedmore easilythan o that onein (55: as will be seenin what follows. A)

55.9.1. p be the number of elementsn S+. Let i

Then we have:
X

56.9. (II\") :Determination Case

of V'

(55:A)

1 :g p
(55:X).

- 2.
1))

I Proof: mmediate by
(55:B')))
in

.1)) So)))

Proof: 1 g a* is evident. Next a< = a* for i in 5*,a< > a+ for i (l , n 1) S+, and by (55: neither set is empty. A')
y

THE SIMPLEGAME [1,

n-l))

1,n - 2]

489))

as desired.
An

- . > (n l)a*,and hence(55:N)gives 1> (n l)a*,a* < n

a a, i
= a*), t
in

__

^ 1

a in has p fixed components: (= o the in (1, the variable ones: p , n) fif*. to the conditions
n

These are subject

9

S*;and

(55:21)
n

a,

-1
In

for t in

(1,- -

, n)

- S*

and

] a,- = 0,i.e. -i (1,... ,*)a,

(55:22)
The lower limits in

- -pa,.
\"\"\"

add (55:21) up to less than the sum prescribedin n t i.e. (55:22), (n p) < -pa*. Indeed,his means a* < p P = - 1. p 5 = 1_,and (55:B') 1 And by (55:A') p<n-l, 1 1 P

so--l> 71
\"\"*

71

guaranteesa*

<

^--r
(n p

Sowe see: The domain a is (55:C')

Put
co* =

l)-dimensional.
1

to 65.9.2. now proceed a closernalysisof V and of a. We a

(55:23) By (55:R) we can write (55:24)

(55:D')

n-p-l-pa*.
- (p - !)(* 1). +
(55: A'),
this (55:B'), is immediate

a>* = w if and only if S* is a one-element (i.e. = 1) or set p if is a* = 1i.e. and only if the condition of (55:11)unfulfilled;

otherwisew*
from

< co.

S Proof: incep ^ (55:24). ; ) (55:E

1,a* ^ 1 by

Max-in a an = w*. a

which follow are the analytical *The lemmas equivalent of the The graphical deduction of 47.5.2.-47.5.4. technical background differs, but the analogiesbetween the two proofs are nevertheless very marked the interested readermay follow them up step by step. (55*C')shows that a graphical discussion would have to take placein a (n we p an This is the reasonwhy n use A dimensional space(by (55: ') this is referred to above took placein a plane, it analytical one. (The graphical proof 2 dimensions.)))) required

(55:D')-(55:P')

1,g

- 2).

1)

i.e.

490

SIMPLE GAMES))
This maximum is assumedfor preciselyone
*
with))

(55:F')

in <J:

= {r,)) = a*
(<* 1
a>*

for in S+, for = n,

1

otherwise.

i i

a It is clearfrom the definition of that for Proof of (55:E') nd (55 the a of (i the variable componentan assumesits maximum when the other in n variable components at , 1) S* assume their minima. Theseminima are So for this maximum)) = a* for in S*,

:F): i

1.

(!,,
i
,n

Now

an =

n-1
^/

-1)) (g

foriin(1,

- 1) p

S*.))

pa* + (n

1 p)=n

1 pa*. By

this (55:23) means an = a>*.

This provesall our assertions.

(55:G')

^* belongs V. to
a*

Proof:a * belongsto a,for any a

:S:(= a,*).

of

a (55:E'),55:F') ( give

follows:

So (55:Z) excludesa H a *, and therefore (55:Y) necessitates a* that to belong V. After 55.9.3. these preparations the decisive part of the deduction (55:H')
If

a,

ft

belongto

V, then an = ft n implies a = ft .
with

C Proof: onsidertwo a , ft in V Put 7, = Min (at , 0,)(i = 1,

dn =

, n

5) Ti < 0,say 5) 7. =
Put 6

1,n)

and assume first that

> 0.
n

= {81,

, 6n

i S*, 7 . _1^ (7 + n p for i in (1,

-i,6

where for in

- 1, - S*.
n)

;) 1Comparison of this definition with (55:D shows that this a * is an a , i in it belongsto V if and only if the condition of (55:11) that is fulfilled. Since a * belongsto a this is in agreement with the result of (55:11).)))

i.e.

THESIMPLE GAME [1,

1,n - 2]

491

This 6 is clearly an imputation, and as i in/S* gives 5l = 7i=ai==/3 g< = a# , l= so 5 belongs to ft. We have 6,, > y n = a n = ft*, and for t in

(1,
an
A
ij

, n in

1)
with
rj

S*, 5i > 7 = a
H 5 5n and

or ft, hence 5 H
from

or 6 H

Since a ,

belong to

V, this excludes5

V.

Hencethere exists
,n

Now by (55:Z) n ry fortiori rj n

17

> 5t for an i in (1, > dn > y n = an = ft, > 5; > 7< = a t or ft.
ry

>
,

1) S+.
t?
S-\"

Thus

>

or

As

a,

77

all belongto

V'

this

is a contradiction.

Consequently

^ t-i

<

is impossible, o s
n

(55:25)
n

7. i-1
ft and

^
j8

0.
= 0. Hence(55:25) ields = for y
> >

Now 7i

g ai; 7i ^

2}ca =
=i

^i =

all these ^ relations,i.e. 7

= a = ft. This proves a =

>

as desired.

(55

:!')

The values of the an for all a

interval))

in

V make up preciselythe

F Proof: oran a in V, an ^ 1 is evident, and an g o>*follows from the of we Hence need only exclude existence a yi in (55:E').

-1
V.

tfi

co*,

such that a

7^

2/1

for all

in

Thereexistcertainly elementsa of V with an ^ and to V by (55:G'), a* = co* ^ y l8 Form

y\\\\

Indeeda * belongs

Min... a in V

with

1In this caseit is not necessary form but the procedurewhich to this achieves is somewhat longer than the one used below. That this minimum can be in that formed, i.e. it exists and is assumed, can beascertained the same way as in footnote 1 on p. 384. Cf. in particular (*) loc.cit. What is stated there for V is equally set true for the analogous set V in a and for the intersection of V with the closed of the a with an ^2/i. o Because f this needfor closurewe must use the condition a* ^ ^i and not n > y\\ although we are really aiming at the latter. But the two will be seento be equivalent b in the caseunder consideration. (Cf.(55:26)elow.))))

2/2, an ^ 2/i the exactminimum,

an =

492

GAMES SIMPLE

an and choose a + in V with a+ ^ y\\ for which this minimum is assumed: this a+ = t/ 2. By (55:H') a + is unique. So 2/2 ^ 2/i, and sincenecessarilya+ 5^ y it so t/ 8 ^ 2/1, i.e.

(55:26)

2/1

< 2/2-

It follows from
(55:27)
Now put
t/i

the definition of 2/2 that

with

a+ ^
Ct

-ft,

and %>frforaniin (1, , n 1) S+. ^2- Tn = 2/2 would ; + ), imply 7 = a (by (55:H cf. above). Hence 7 = at^ < ft for the above in (1, , n 1) S+, and not 7, > ft as required. Hence Now 7n

7^7. this means7n> By (55:Z)

> 2/2-

and

= y\\ excludesft

+p
f

g an < 2/2 for no a in V. = 2/2 ~- > and form the imputation

y\\
>

= 2/2
or

t in

'

= {01, , fti-l, fti} = 2/i ft = a+ = a< = a* for i in S*,ft = * (i . . . n _ i) _ 5 Clearly belongs to

* *

from
n

V.

a Hencethere exists

y in

with

> 0n = 2/i necessitates (55:27)y n ^ by

Thus 7n > 2/2= <*n and y t > ft > at^ for the above iin(1, ,n 1) S^. +, and as 7 , a + both belongto V r this is a contradiction. So 7 H a F 56.9.4. (55:1'),:H')we see: orevery y in (55 By

y*

-1^

^\"*
by))

a thereexists unique a

in

with

an = y. Denotethis a

= n (2/) = 2/ and a t (t/) = matterarethe at (t/) for i in (1,

Clearly Combiningthis with

* for

(55:1') gives:

,n

i in S*. Sothe functions which

1)

S*.

(55:J')

V consistsof theseelements:
7(2/)
where y runs over the
interval)))

THESIMPLE GAME [1,

and where a (y) = [ai(y)
9

, ,

1,n - 2]
<x

493

an_i(y), n (y)} with

for in S*, for = n, a suitable function of y (and i) for

y

= a*

(1,

, n

- 1) - S,.
(1,

i i

i in

55.9.5. And
(55:K')
fulfill

to

conclude:
on(y), t in

The functions

(55:K':a)
(55:K':b) (55:K':c) (55:K':d) I.e.

the following conditions: The domain of at (2/) is the interval

,n

- 1) - <S,of (55:J')

i/!g y 2
tin

-1^

co*. 1 at (i/i) ^ a,(yi). i/.2 8

*

implies

a.(-l)= a,. =

(55:K':e)

.(*)

(l,-,n-l)-S*

a,(y) = -pa*

-1. |yi

1 a,(t/) is an antimonotonic function of y. 1From theserelations the continuity of all functions (t/), i in (1, ,n follows. Indeed,we can even prove more, the so-calledipschitz condition: L

1)

(55:28)

\\<x*(yi)

a,(yO| ^

yi|.
y\\

Proof:This relation is symmetric to y\\ and y application of (55:K':e) y

tin

in yi, y*

hence we may assume

give

2/2

and subtraction

y*.

Now

(l,-,n-l)-S*
a(2/j)are ^ 0, hencethey arealso ^ than
the middle term their

all By (55:K':b) theseaddends a(j/0 sum y* Thus y\\.

Theseinequalities make
that the

last term is \\yt

yi|. Hencewe

it

alsoclearthat
have

is

\\a t (y t

a(t/i)|and

as desired. Thereaderwill note that

quite interesting 8 Note that

a<(-l) ~pa# 4- 1, hence ] tm(l,-,n-l)-S* n-1 1 agreeing with (55:N). a* J S a (55:K':c) requires ^/ -l im(l ,n-l)-5*
For
y

- (55:K':c), -1(55:K':e) f

we never assumed any continuity from the technical mathematical point of view. do not conflict with

we proved

it! This is

(55:K':d)
gives

Indeed: (55:K':e).

\"*

\"\"P

\"^\"

< ==

For
(55:23).)))

w*

g (55:K':e) ives (n

(55:K':d) requires

1)

- ~p*-w*,w*-n~p~l-pa*,

J)

a<(w*)

- ~pa# -

*,

hence))

agreeing with

494
A Proof: d Ad

SIMPLE GAMES

Containedin (55 :J'). (55:K':a): Assume the y\\ (55:K':b): - opposite:^ t/ 2 and a(t/i) < a(y 2) (for a suitable i in (1, 1) S*). This excludes\\ = !/2, so y\\ < y 2. , n y

toV.

Then a (7/2) H a (t/i),

which

is impossible since a (y\\), a (y 2) both belong

o This Ad (55:K':c): is a restatement f the fact that a belongsto V, indeed it belongsto V. (Cf. (55:K),(55:M).) o This Ad (55:K':d): is a restatement f the fact that a * belongsto V
Ad

a (y) (55:K':e): is an imputation, hence ^

n
4

(y)

= 0.

By (55:J') this means that

t i.e.hat
tin

a (2/) + Pa* + 2/ = 0>

t/>

(l,-,n-l)-S>

(?/)

pa*

as desired.

o 55.10. Disposal f Case (II\

contain The results obtained in 55.S.-55.9. a completeescripd As tion of the solution V. Indeed: we saw at the beginning of 55.8.2. it V = V u V, although the addend V may be omitted (because is c V) is satisfied. V is describedin (55:S), if and only if the condition of (55:11) makeuse of the parameters V in (55:J')- Thesecharacterizations

55.10.1.

at (t/)(iin (1,
which

: a. (i = 1, ,
n

,n

S, - - 1),S,,-1g 1) or*,
w,

*,
y

g *),

are subjectto the restrictionsstated in (55:N); 55:13), in ( (55:15) i (55:24)n 55.9.2.; (55:K'). (55 55.5.1.;:R);(55:23),
result restatethe complete
i Sincethis material is dispersedover seven sections,t is convenient to
in

one

place:

(55:L')

(55:L':a) (55:L':b)

S* is a set c (1, , n 1), not empty. Let p be the number of elements f S*, o that 1 g p g n o s 2.

gi,
Put
w

gn -i arenumbers ^

-1,

n-l

with

^ at =
,n

1.
that

(55:L':c)
(55:L':d)

Forall i in
w*

S,a> = a^, for all i in (1,

- 1) - S*

= n 2 a*, a>* = /i = (p l)(a* +

1).)))

- -

-p-

~ pa^, so

THESIMPLE GAME [1,

1,n - 2]*
,n

495

(55:L':e)

ai(y) is defined for t in

-1

(1,
*.

1)

S+,

Thesefunctions satisfy the conditions(55:K':a)-(55:K':e) V consistsof theseelements: (a) a (y) where y runs over the interval 1 y ^ w*, and
where))

with
&i y

= a*

the

(1,

av(y)
-

for i in S+, for i = n,

of
f

(55:L':e) n- 1) -S,.
for

i in

(b) a where i runs over all S* and where

*

with

for for

otherwise.
I Remark:f p = 1 (S* a one-element or a* = 1,then eo = w* set) w and the a of (b) coincide ith a (y) of (a) for y = w*. If this is not the case i.e.p *z 2, a* > 1,then > a>* and the a of (b) are disjunct from the a (y) of (a). The readerwill verify with little difficulty that all these statements arenothing but reformulations of the results referred to above.
*

j = i, j = n,

a>

be followed by similar considerationsas (55:V). We must investigate whether all V obtained from (55:L')aresolutionsand in the case(II\.")Thoseof them which meet both these requirements form the complete system of all solutionsin the case(II\.")We will prove
meet theserequirements. that all V of (55:L') (55:M')

65.10.2.:L')must (55

a The V of (55:L') repreciselyall solutionsin the

case(II\.

W Proof: e need only show that every V of (55:L/)is a solution in the case(II\") that theseV are preciselyall such solutionsthen follows from

(55:L').

Case(II\")is easily established: Clearlyfor this V,

1,

and)))

496
(in

GAMES SIMPLE
in

1 in quantities designated (55:L/)by thesesymbols, hence

the senseof their definitions given

are precisely the 55.2.-55.5.)

by

It remains for us to
Ad

(55:L':a).

S,c (1,
V fulfills
ft

,n

- 1)
in

we will

do this by proving that Assume (30:5:a): a

H

prove that V is a solution. In the present case

in (30:5 (30:5:b)30.1.1. :a),
ft

for

a,
ft

cases(a), (b) of (55:L/) a , combinations:

to which

distinguish belong. Thereare four possible

V.

We must

(a):I.e.a = a (2/1), ft = a (1/2) and so a (t/i) H a (t/ 2). is Now (55:1) excludedwhen i is in S* and (55:2)is excludedsince (yO = <*(y) = * for i in S*. So this domination can operate this with through (55:1) i in (1, , n - 1) - S* only. By (55:L':e) means an (t/i) > ctn(yd, y\\ > 2/2, and a(s/i) > a,(t/2) for a suitable i in (55:K':b). (1, , n 1) S*,contradicting

a,

ft

in

- ft

tradiction. a
in

(b): I.e. a = a (i/), = a * (i in S*), and so is since otn (y) = y ^ w* ^ = aj,, (j/) Ha*. Now (55:1) excluded, i s and (55:2)s excluded, ince <(i/) = a\\ = g = a*. So we have a con-

>

in

>

(a),

in

a>

(b), ft in (a):I.e. a = a* (i in S*), ft = a (y), and so a * H a (#). Now a\\ = a(v) = at = a^ and for j j* i, n, a} = 1^ a,(j/), i.e.a} g a,(j/) for all j = 1, , n 1. This excludesboth (55:1), and (55:2), gives a contradiction.

= a * (i, A in (b): I.e.a = a *, = a* = w, thus contradicting :B) Now ai (55 a,

Ad

in

S*), and so a

a *.

is undominated by the elements V. of We wish toprove that this impliesthat ft belongso V which establishes t
ft

Assume (30:5:b): that

in

^ w. If ft < a, = a* for all i = 1, - , n 1, then a( 1) H ft , contradictingour assumption. Henceft ^ g for some t*=l, , n 1. Now the argument used in the proof of (55:R) shows that necessarilyi in S* and a *. Therefore belongsto V
Assumefirst that
ft n
==

(30:5:b).

this case.

l w obtains (55:L':c).)))

from

(b), 31,

, gn

-i from

(a) with

- -1,nd a

then

a*,

>

from

THESIMPLE GAME [1,

Assume next that
n

1,n

2]

497

< o>.
=

If ft

< g = a*
1

for some

i in S*,then i.e.

clearly a H
*

in

S+.
n

our contradicting assumption. So ft ^ g = a* for all

-1^ that Assume

ft>

Now % ft

-i

=
ft

gives

ft,

a)*.

have clearly ft we have even

Put y -ft.. ^ a(y) for all i in (1, , n 1) S*. Then we 2> a(2/) for all i = 1, n. (For i in S* and f = n ,
, cf.

- i-i
n

ft

- - 1p

pa* = *,

above.) Hence

^ -i

ft

t-i i(t/)

= necessitates

= a (y). Therefore/3 that we have = in all these ^ relations.So to too. belongs V in this subcase There remainsthe possibilitythat ft < a<(y) for a suitable in (1, , n o 1) S+. A sufficiently small increase f y (from y = to some 1 For this new y we have y > j8 n ) will not affect this relation ft < <*i(y).

> ft, a,-(y) > ft,

and therefore a (y) s* ft contradictingour assumption. Thus all possibilities accounted for. are

>,

>

55.11. Reformulation

of the CompleteResult

the undetermined parameters ccurringin (55:L/) i.e.in o (55:M'), describingcase (II\") is the set S*. According to (55:L':a) of this is any set c (1, 1) ,n ,n 1) with the exception (1, it is not possibleto find some and This raises question whether the = for cases + = (1, IJandS* ,n S interpretation theseexcluded
Among

Thesethree cases(I), (IF),(II\") into which we subdivided ; our problem have been completely settledby (55 ) :G),(55:W), (55:M of respectively. Let us now seeto what extentthesethreeclasses solutions arerelated eachother. to

55.11.1.

0.

also.

1) the answer is easy. If we use this S* to ( (disregarding55:L':a)this extent),then we obtain (usingall otherparts = g-i = a* = n _ t of (55 gi :L')): = n 1 by (55:l/:a),= p
, n

For S* =

(1,

by

= (55:L':b), (55:L':c), n
, n

since of (55:L':e), as 1) S+ is empty. Inasmuch the interval 1^ y ^ w* it must be noted that it shrinks to the plays a role(in (a) of (55:L':e)), point y = 1 (sincew* = 1). Now comparisonwith (55:V) discloses w that under theseconditions(55:L/)coincidesith (55:V).
to Thereis no occasion introduce the functions a(y)

- - jpz-y
2

<>* =

-1,by (55:L':d).

(1,

1cu (y) is continuous!

Cf.footnote 2 on p.

493.)))

498

SIMPLE GAMES = (1, If we includein (55 ,n 1)(hence = p a then (55:L') enumerates ll solutionsin the cases 1) also, t Case (IF)and (II\:") (IF) correspondso 5* = (1, , n 1) n and case(II\")to S+ * (1, ,

So we have :

(55:N')

:L':a)S

- 1).

the result one might feel inclinedto correlate remainwith the remaining case(I). However,inspection and comparisonwith (55:G)show that this is not possible at leastnot in this direct way. Use Indeed: of (55:L)with S+ = Q (hencep = 0) gives an empty (b), with (a) i.e. is the set of all so a V coinciding V

S* exception = with of[(55:L') S* =

ing

After 55.11.2.

this

suitablefunctions i(t/), , a_i(y). Disregarding 1 other maladjustments we note:In this arrangement the an of an a in V a n -i;while in (55:G)an was constant and i determinests
w*, with

1^ y ^

!,-,

arbitrary

!2

Summing

up:
All

(55:0')

would producea V which is not the solution of (55:G),and choice indeedis no solution at all. We 65.11.3.concludewith the following observations:

w c ( (55:N') CasesIF) and (II\.")(55:N') oincidesith (55:L'), is widenedto includeall S+ (1, (55:L/:a) 1) ,n with t 7* . The exclusion of S+ = Q is necessary; his S*

solutionsV are enumerated by (55:G) Case(I) and

when

(55:F) (55:P':a)

o>* =

In case i S (IF),.e. * =

(1, t 1,i.e.he interval 1 ^

=

,n
y

a point. Also a+

- 1),p = - 1,we have:

n
w*

of (55:L':e) shrinksto

(55:P':b)
w*

In Case(II\,")i.e. * c (1, S
t > 1,i.e.he interval
Also))
w

-1g

shrink to a
1

point.

.
a>*

1),p ,n y <; o>* of

does (55:L':e) not

may
have))

<n

- 1,we have
be

1))

> w, and
For S+
give

Owing

to p so

Max-*

(55:23)now gives
M&\\_ l

no solution at all. It

The V

Min^j

-i of (55 :L')
,n

- - -( ^y

n))

+ 1),hence we

w*, although

(55:L':b), (55:L':c) Min.^j gave

g<

althou gl 1 tn ^ former was the definition with is thus not a set from our list of solutions, would have beeneasyto verify this directly.)))

> a*

n-1gi

it should

8-

; for of a

w!))

they

henceit is

THESIMPLE GAME [1,

1,n

2]*

499

(55:N').
Ad
w*

A Proof: d

We (55:P':a):proved thesestatementsimmediately preceding

We t (55:P':b):saw in the proof of (55:B')hat

a*

< 2-ZJP,hence

+l=n

~p- pa^ > 0,

i :0'), .e.

w*

>

-1.a* <

was stated in (55

:B').

We can now beginto interpret this result. It is hardly possible to do this in an exhaustive way for two reasons.Firstthe final result containedin (55 in (55 ( :G),(55:K'),55:L') is rather involved, hencea precisetatementmust necessarily mathematical and not verbal. s be Any verbal formulation would fail to do justiceto some of the numerous nuancesexpressedy the mathematical result. Secondwe still lack the b and necessaryexperience perspectivefor a really thoroughgoing interpretation of a situation like the presentone. The game which we consider here is a characteristic n-person game in some significant ways, as we set forth in and 54.3.But our successn determiningall of its solutions i is still an isolated occurrence case 54.2.1. of (the notwithstanding). It will take many more discussionslike this one before one can attempt really exhaustiveinterpretations of characteristic-persongames. n It is nevertheless to do a certain amount of interpreting without useful We any claim of completeness. have seen in several previous instances that such interpretations give valuable guidance for the further progress of the theory. Besides, this procedure doesthrow some light on the of our rather complicatedmathematical result. significance Sincewe do not try to be complete, interpretation is best made the in the form of several remarks. in First: hesolution of Case(I) described (55:G)is an infinite T set of imputations. The same is true for the solutionsof (II\,")described in (55 :L')(cf. (55:N')) sincethe y mentioned there varies over an entire On interval which does shrink to a point. (Cf. (55:P':b).) the other not hand the solution of Case(IF) is a finite set of imputationsas was already l observedat the end of 55.7. This solution alsohas the attractive property i of sharingthe full symmetry of the game i.e.nvariance under all permutations of the players ,n Thus it is in severalways the simplestsolution of our game. Heuristic n discussions its special of cases = 3,4 (in 22., respectively)led to this n solution and it is easy to extendthem to the general .2 It takesthe full machineryof our formal theory to find the other solutions. t It will besufficiently clearo the readery now that theseothersolutions b and the can in no way be disregarded. esides, existence the uniqueness B 1 Thereader same effect (55:P':a), (55:P':b). may compare to the *The (heuristic) argument would run as follows: The chief player needsan ally to 2. Thus if he wishes to retain the amount w (this win, with any such ally he obtains n eachally n 2 w. t correspondso the of our exactdeductions) he can concede

55.12.1.

55.12. Interpretation

of the Result

54.1.2.

55.12.2.

1,

1.

35.1.

o>

If)))

500
1 meansgeneral.

SIMPLE GAMES

b of a finite solution is a favorable contingency in the present ame, ut by no g

in the organizational principlethat the chief player is segregated the game that of (55:G) discloses this standard of of in the sense Inspection solution offers absolutelyno principleof division among the behavior t all imputations where the chief player receiveshe preother players to it. This is perfectly reasonable common scribed amount belong by the sense: he chief playerbeing excluded, other playerscan only combine T in with eaol^ other unanimously. All quantitative checks their relationof siding with the chief player) being forbidden, the ships (i.e. possibility of there is no telling what the outcome their bargaining with eachother

remark it possesses in described (55:G). Like that one in the preceding thesetwo the CasesI) and (IF) the full symmetry of the game. Indeed ( 2 arethe only oneswith this symmetry. it On the otherhand this solution is infinite. As we sawin 55.3.expresses

to Second: 66.12.3. Theabove solution corresponded the largestpossible i ,n 1). The other extremes the solution which we associThis atedwith S* = Q (cf. preceding(55:0')). is the solution in Case(I)

S+:(1,

33.1. i.e. i.e.

will

two solutionsdealt with above. and the most involved one of deductions.Their interpretation, oo, s more difficult our mathematical t i the main points only. We and complicated. will indicate in We describedin (55:L') detail how in all imputations of a standard a solution of this categorythe playersof (1, behavior , n 1) S*

in form

a morecomplicatedgroup than the part Indeed,hey took up a considerable t

Third: 66.12.4. The remaining solutionsarethosein Case , described (II\") i.e. , n - 1). They (cf. (55:L') (55:N')), those with S* ^ 0, (1,

be.

i.e.

his n

1 potential allies together can make more

(n

- l)(n - 2 - ) < 1,
are destroyed
(n

than that,

i.e. if
is
the only limit n

of then his chances exactions. Thus

w

finding

an ally by

and this

to

his

is

So w

- -2
n

only n

limited

l)(n 2

w 1,i.e.

2-

Sothe chiefplayer
and

obtains n

2
the other that this the

r if

he succeeds in

forming

a coalition, are it (55:V),

and r

of course 1 ifhe doesnot. For

Case(II').

-1. The readercan

players the is just

corresponding

amounts
in

now verify

the solution

arrived at

i.e.
in

fails is analyzed in 1 Any other solution belongsto Case(II\") and so has an S* ,n , Hencean appropriate permutation of the players 1 will carry an element ,n of S* into oneoutside, thus changing 5* and with it the solution under consideration.)))

uncertainty An 53.2.2. instance where concerning the uniqueness 1

1As to the

existence the end of the secondremark cf.

1,

38.3. . * (1, -

- 1).

THE SIMPLEGAME [1,

1,n - 2]*

501

h are causallylinked to the chief player. I.e. ow the respectiveamounts to which they get areuniquely determined the amount assigned the chief by b player. This connection was expressedy definite functions.1 These

functions couldbe chosenin different ways, thus yielding different standards s of behavior i.e.olutions but a definite standard meant a definite choice of thesefunctions. Thus the uncorrelatedness the players of ,n so prominent in the secondremark, is now gone. Thereis obviously some kind of indefinite bargaining going on betweenthe chief player and those 2 of (1, , n 1) S*, but the relationshipof the latter players to determinedby the standard. eachother is completely It is worth while to emphasizeoncemore this difference betweenthe between in remark and in the presentone situation described the second therewas bargaining between the Cases and (II\.")In the former case (I) t all players excepthe chief player with absolutelyno rules or correlations to cover it, 8 so that the standard of behavior had to make no provision in this respect. Now we have bargaining betweenthe chief player and some of the others,but this time the standard must provide definite correlations and rules for the opponentsof the chief player. Accordingly there is a standards. multiplicity of possible Thequalitative types of indefinitenessarising in the CasesI) and ( as discussedabove, are a more general form of that one which we investia 4 gated in 47.8., 7.9. The remarks made there' bout the 2-dimensional (area) and one-dimensional (curve) parts of those solutions are indeed applicableto our present Cases(I) and (II\,")respectively. While it is possibleto motivate this difference by verbal arguments of some plausibility, they are all far from convincing. The mathematical deductionalone,such as we gave it, gives the real reason and its relative complicationshows how difficult it must be to translate it into ordinary but language. This is another instanceof a result which can beexpressed, scarcelydemonstrated,verbally. 66.12.6. Fourth:The situation of the remaining players those in S+

1,

1,

i.e.

(II\,

has alsoits interestingaspects. of Inspection (55:L/) shows that in every imputation of our solution eitherall theseplayersget the amount a* , or one of them getsa% and the othersthe amounts 1. From this one infers immediately:

set (a) If S* is a one-element then the player in S* gets always the same amount:a* then eachplayer of S* always gets the same amount: (b) If a* =

(c)

-1. -1 If neither of

i :U) (a) or (b) is the case i.e.f the condition of (55 i referred to in (55:D'))s fulfilled then each player in S+ (also

1The a<(y), t in (1, , n 1) 5*. Of.also(55:P':b). 1Thiscorresponds the variability of y in (55:L':e). to 1 Exceptfor the assignment to the chief player who is segregated.)))

502

GAMES SIMPLE
always gets one of the two different amounts a* and 1 neithercan be omitted.

1, and

From thesewe can draw the following interpretative conclusions: (d) In the two cases and (b), but not in (c), the players of S+ are (a) in segregated the senseof set: , (e) The case(a) where S* is a one-element S+ = (i), = n o the expresses segregationf the player alone. The value a* which is then assignedto him, is limited by (55

33.1.

1,

(55:29)

-1Sa, < ^~

i 1, :B'):

of This is a satisfactory complementto the segregation the chief 2 in player, Case (I), described the secondremark. The value w which was then assignedto the chief player was limited by (55 :G):

(55:30)

-lgw<n-2^^
1.

set, (f) If S* is not a one-element then only the possibility(b): a* =

In other words:

thereis within the cases (b) (a),

If more than one player is to be segregated, then their setmust not contain the chief player, nor all other players, and the segregated players must all be assignedthe value:

(55:31)

a* = from (e), (f) that thosesetsof players which can be (g) We conclude a segregated repreciselythe setsof L 3 the defeatedsets. then (e) showsthat he neednot be (h) If only oneplayer is segregated, discriminatedagainstin an absolutelydisadvantageous ay. I.e. e h w more than 1 (55:29) (55:30) statethe upper also , may be assigned I limit of what this assignmentcan be: t is clearly the same amount which this player would get in the finite solution of Case(I), dis4 cussedin the first remark. It is very satisfactorythat this extends from n = 3 to all n. the result of 5 (i) If, on the other hand, more than one player is segregatedthen shows (55:31) that there can be no concessions: must all be They given the absolute minimum

-1.

33.1.2.

1.

in

1 both occurin appropriate imputations of the solution. 1This resolveshe difficulty pointed out in footnote 3 on p. 475. t 8 This is bestverified by recalling the enumeration of the elementsof W and so of L the caseC_i in 52.3.
4n

I.e.

lessthan
*
happen

I.e.he t

(i) and

2. Since n 2 (cf. number p of elements in # is this can p 2^ 2 n 4. This is the reasonwhy the phenomena of only when n in (j) were not observed the discussion of n

2 n i for the theseamounts.

chief player, n

for the

others. Theassignment must be

i.e. :

(55:L:a))

3.)))

THESIMPLE GAME [1,

1,n - 2]*

503

some of which were even functions which could by numerousparameters, be chosenwith considerablereedom. The main classification,however, f 1 was rather simple:It was affected by the set S* (1, , n I). all partitions of 7 = (1, The pairs S*, S* exhaust obviously , n) into two sets.Possibly this is the first indication of a generalprinciple. In a simplegame a partition into two complementsseemso decideveryt e thing, sinceone of them is necessarilywinning and the other necessarily defeated. In generalgames partitions into more sets may be equally case important. At any ratethe role of S* in the presentspecial gives the first idea of what may be a general lassifying principlein all games. c form. We arenot in a position,as yet, to give this surmisea more precise
1We use as in the secondremark S+ preceding (55 notwithstanding.)))

If must be qualified to the following extent: S+ has (j) This assertion a more than one element,the a* of (55:29) re still all possible indeed (55 :L')with (55:B')allows for them explicitly. But the situation of the players in S+ is then described (c), and can no by be longer termed segregation: may join coalitionsand thereby They improve their status. It is clearthat these remarks, particularly (g), (h), (i), invite further comment. However we will now restrictourselves to these indications and return to the subjectat another occasion. 55.12.6. We found a great number of solutions, characterized Fifth:

:0')

to symbolize

the

case (I),the

discussion

CHAPTER XI GENERAL NON-ZERO-SUMGAMES

Our considerations the have reached stageat which it is possible to drop the zero-sum restriction games. We have already relaxed for this condition onceto the extent of consideringconstant-sumgames with a sum different from zero. But this was not a really significant extension of the zero-sum case sincethesegameswere related it by the isomorphism to of strategicequivalence(cf. 42.1. 42.2.). e now proposeto go the and W whole way and abandon all restrictionsconcerning sum. the We pointed out before that the zero-sum restrictionweakensthe con1 nection betweengames and economic roblemsquite considerably. Spep it emphasizesthe problem of apportionmentto the detriment of cifically, problemsof \" productivity*' proper (cf. particularly footnote 2 on 34 ; also5.2. Thisis especially 1 i clearn the case the one-person of .) p. game behavior in this situation is manifestly a matter of productionalone,with no conceivable imputation (apportionment)betweenplayers. And indeed the one-personame offers no problemat all in the zero-sumcase, nd a a g case(cf. perfectly good maximum problemin the non-zero-sum the Accordingly our presentprogram of extending theory to all non-zerosum gamesmust be expected bring us into closer ontactwith questions to c of the familiar economic which follow, the reader type. In the discussions will soonobservea change the trend of the illustrative examples of in and the interpretations: e shall begin to deal with questions of bilateral w monopoly, oligopoly, arkets, etc. m abandonmentof the zero-sumrestrictionfor our games Complete that means, as was pointed out in 42.1., the functions 5Cjk(ri, , r n) which characterized in the senseof it arenow entirely unrestricted. that the requirement))

66.1.1.

56.Extensionof the Theory 56.1. Formulation of the Problem

4.2.1.,

12.2.1.).

66.1.2.

I.e.,

11.2.3.
- , r n)

(56:1)

*-l

K (r ly
k

1It should be noted that zero-sum games not only coverthe type of gamesplayed for but alsothat many of them describe entertainment (cf. quite adequately relationThereaderwho has progressedp to this point and nature. u ships of a definitely social recallsthe interpretations which we have made in numerous caseswill be fully aware of the validity of this statement. Thus the distinction between zero-sum and non-zero-sum games reflectsto a certain extent the distinction between purely socialand social-economic questions. (Thenext in t sentence the text expresseshe sameidea.)

5.2.1.),

04)))

EXTENSION OF THE THEORY

505

is droppedwith nothing to take its place. Accordingly of and o we proceedn this basis from now on. This changenecessitates completeeconsiderationof our theory a r with all the attendant concepts n which it is based. Characteristic o funca tions, domination, solutions, all theseconcepts reno longer defined when is (56:1) dropped.We emphasizethe fact that the problemwhich arises hereis a conceptual and not merely technical s were all thosetreated a one, 1 in Chapters VI-X,on the basis of our theory of the zero-sum games. The prospect of having to start all over again would be very we discouraging: have already spent considerableeffort on theseconcepts w and the theory based on them. Furthermore e face a conceptualproblem and the qualitative principleson which our theory was based do not seem to carry beyond the zero-sumcase. Thus this final generalization the passagefrom the zero-sum to the non-zero-sumcase would seem to nullify all our past efforts. We must find therefore a way to avoid this

11.4. 25.1.3.

56.1.3.

difficulty.
At

42.2.Thereour transition from the zero-sum case the constant-sumcase to

threatened on a narrower scale with similar consequences. hey were T avoidedby an appropriateuse of the isomorphisms f strategic o equivalence,
as effected in 42.3. 42.4. and The usefulnessof this particular device was, however, exhausted by the applicationreferred to:strategicequivalencesextendthe family of all zero-games preciselyto the family of all constant-sumgamesand no further. of 42.2.3. 42.3.1.) or from the considerations 42.2.2., (This should beclear Sowe must find someotherprocedure link the theory of the non-zeroto sum games the establishedtheory of the zero-sumgames. to
Player. The Zero-sum Extension r Beforegoing further we have to clarify a point of terminology. Thegameswhich we shall now considerarethosewhere as stated in condition (56:1) dropped without anything else is taking its place. We talked of theseas non-zero-sum games,but it is important to realize that this expression meant in the neutral sense, that we do not wish to is exclude thosegamesfor which (56:1) happens to be true. It is therefore preferableto use a lessnegative name for these games. Accordingly we shall call the gameswith entirely unrestricted ,T)generalgames.* 3C*(Ti, We have formulated the program of linking the theory of the general gamesin some way to the theory of the zero-sumgames. It will actually a be possibleto do more: ny given general amecan be re-interpreteds a g a The 56.2. Fictitious

this point one might recallhe comparablesituation which arose in t

66.2.1.

56.1.2.

i.e.

zero-sumgame.

1Among thesetechnical problems was one which we preferred to treat by a method involving a certain conceptualgeneralization: the caseof the constant-sum games,which will be referred to further below in the text. 1This is in agreement with 12.1.2.)))

606

GAMES GENERAL NON-ZERO-SUM

This may seem since g paradoxical the general amesform a much more will extensivefamily than the zero-sum games. However,our procedure zero-sum game. beto interpret an n-persongeneral ame as an n + 1 -person g caused the passagefrom general amesto zero-sum Thus the restriction g by f gameswill be compensatedor indeed made possible by the extension 1 in due to the increase the number of participants. 66.2.2. procedurey which a given general -persongameis re-interThe n b zero-sum game is a very simpleand natural one. preted as an n + 1-person It consistsof introducing a fictitious n + 1-st layer who is assumed p to losethe amount which the totality of the other n real players wins i and viceversa. Hemust, of course, ave no directnfluence on the course h this Let us express mathematically : Considerthe general -persongame n the players 1, , n, with the functions 3C*(ri, , r n ) (k = We 1, , n) in the senseof 11.2.3. introducethe fictitiousplayer n + 1 by defining
F of

of the game.

(56:2)

3C (r,, n+1

,r )
w

*i3C*(n

r).
1,

real players , n, their influence on the courseof the game. respectively. They represent Sinceit is intended that the fictitious player have no influence on the 2 w courseof the game,a variable r n+i which he controls, as not introduced. In this way we obtain a zero-sum n + 1-person ame, the zero-sum g extensionof F, to be denoted by T. 66.3.Questions oncerning the Character P of C In stating that we have re-interpreted general -persongame the n F as the zero-sumn + 1-personame T, we imply prima facie that the g entire theory of T has validity for T. This assertion requires,of course,

Thevariablesri,

, r n arecontrolledy the b

56.3.1.

closest scrutiny.
We

that this cannot be a purely mathematical analysis,like the analysesin the chapterswhich were based on a definite theory. We areanalyzpreceding ing oncemorethe foundations of a proposedtheory. Consequentlythe analysismust in the main be in the nature of plausibilityarguments, even if intermixedwith subsidiarymathematical considerations.The situation is exactlythe sameas in those earliernstanceswhere we made our decisions i 1This may serveas a further illustration of the principle stated repeatedly that any in increase the number of participants necessarilyentails a generalization and complication of the structural possibilities of the game. The formalism of 11.2.3. provided a variable r k for every player k. (In order to replaceit for the present casewe must replaceits n by our n H- 1.) Henceone might insist on the appearance f the variable r n +i of the fictitious player n + o This requirement is easyto meet. It suffices to introduce a variable T+I with only one possiblevalue (i.e.o put +i 1,loc. cit.). Actually one could even use any t domain of T*+I (i.e. /3 +i) as long as all 3C*On, any , T, r n+i) areindependent of TH+I, so that they are really functions 3C*(n, , r) as used in the text.)))
B

shall now undertake this investigation.

It must

be understood

1.

OF EXTENSION THE THEORY

507

and go on to 56.12.) The result of our analysis will be that it is not the entiretheory of F as a zero-sumn + 1-person which game,in the senseof 30.1.1. appliesto F, but only a part of it which we shall determine.In other words,we shall find that not the system of all solutions for P, but only a certainsubas system produceswhat will be interpreted the solutionsof F. 66.3.2. fictitious player was introduced as a mathematical device The to make the sum of the amounts obtained by the playersequal to zero. It is therefore absolutelyessentialhat he should have no influence whatever t on the course the game. This principlewas duly observedin the definiof tion of T as given in 56.2.2. must neverthelessput to ourselvesthe We question whether the fictitious player is absolutely excludedfrom all transactions connected with the game. This caveat is not at all superfluous. As soon as T involves three or 1 more persons the game is ruled by coalitions,as we observedat an early stageof our analysis. A participationof the fictitious player in any coalition which is likely to involve the payment of compensations etweenthe b participants would be completelycontrary to the spirit in which he is t introduced. Specifically:he fictitious player is no player at all, but only a formal device for a formal purpose. As long as he takesno part in the o gamein any direct r indirectform, this is permissible. But as soonas he t to interfere, his introduction into the game i.e.he passagefrom begins F to T ceases be legitimate. hat is, T cannot then be regardedas an T to of the , n, equivalent, or a re-interpretation F, since real playersof T, 1, have to provide against dangers or may profit by possibilitieswhich may certainly do not existin F. 56.3.3. might think that this objectionis void due to the way in One t which the fictitious player was introduced. Indeed,he amounts end of the play, do not players 2 he has no moves in the play. on any variable which he controls depend How can he then be a desirablepartner in a coalition? a It may appear t first that this argument has somemerit. The condithat make tions described it seem any coalition of real playersis just as well 1I.e. hen n -f 1 ^ 3 which means n ^ 2. Thus only the general one-personame w g w is free from the objections hich follow. This is in harmony with the fact which we have emphasized repeatedly, that the general n-persongame is a pure maximum problem only
which the real

the concerning theoriesof the zero-sum two-, three-,n-person games. 17.1.-17.9. zero-sum two-persongame;Chap.V for the (Cf. 14.1.-14.5., for the zero-sum three-person 30.2. game;29.,30.1., for the zero-sum n-person game. For the generaln-person game i.e.the relationship betweenthe theories T and T the equivalent sections of begin with 56.2.,

1,

n 3Ci, , 3C , n obtain at the ,

* *

i.e.

when n

1.

2 Nor

doesthe

amount

3C _n n

2^ 3C
fc

which

he obtains.)))

508

GENERAL NON-ZERO-SUMGAMES

specificexample.

Ishe anything but a dummy? If this wereso, he theory of T couldbe appliedwithout any further qualit fications to P. Howeverthis is not the case. It is true that the fictitious player, having no moves to influence the n courseof the game,is not a desirable partner for any coalition. I.e. o for player or groupof playerswill pay a (positive)compensation his cooperation. Howeverhe himself may have an interestin finding allies. The amount which he getsat the end of the play 3C +i(ri, n , depends on the moves of the other players on n, , r n and it may be worth his while to pay one or more among the players a (positive)compensation to with for ceasing cooperate the others. It is important not to misunderA this: s long as F is played, as long as the fictitious player is stand really a formalistic fiction, no such thing will happen;but if the game i really played is T, i.e.f the fictitious player behavesas a real player would t t in hisposition, hen his offerof compensationso the othersmust be expected. As 56.3.4. soon as the fictitious player begins to offer compensations w to otherplayersfor cooperationith him which, as we sawabove,amounts with w to their non-cooperation others he is an influenceto be reckoned ith. He offers to join coalitionsand to pay a price for this privilege and his t i willingnesso pay is fully as goodas a directnfluenceon the game exercised to make significant moves. by ability Thus the fictitious player gets into the gamein spite of his inability toinfluence its coursedirectlyby moves of his own. Indeedit is just this to w impotence hich determineshis policyof offering compensations others, and thus setsthe above mechanism into motion. Fora betterunderstandingof the situation it may be helpful to give a
off without the fictitious playeras with him.

r)

i.e.

if while the 1,2 left to himself can securefor himself only the amount It is easy to specifydefinite rules two togethercansecurethe amount 1 for a game to bring this about. A particularly simple combinatorial

66.4.Limitations of the Useof f 66.4.1. Considera generalwo-persongame in which eachof the players t

1.

1,

2 w arrangement hich does it, is as follows: Each player will, by a personalmove, choose of the numbers 1,2. one uninformed about the choice the otherplayer. of Eachone makeshis choice

amount
just

chosenthe number 1,each gets the amount

After

this the payments will be made as follows:if both players have otherwise each gets the

l.
a

i,

1Thus it will be seenin to that the bilateral monopoly corresponds this. This construction should be compared with the one used in defining the simple with which it has a certain similarity. majority game of three personsin 1With the notations of all 2 2 0,

60.2., 61.3. 61.2.,

21.1., 11.2.3.:

and)))

EXTENSION THE THEORY)) OF

509))

It is easy to verify that this gamepossesses desired roperties. the p Let us now considerhe fictitious player 3 and form the gameas defined t in 56.2.2., its characteristic with function v(S), (1,2,3). According to what we said above))

Ss

-1,))
Obviously))

v(0) 0,)) and by the general roperties the characteristic of function (of a zero-sum p game) v((l,3)) v((2,3))= v((l,2,3)) =
Summingup:))

= *3))= -v((l,2)) -1, = =

-v((2)) 1,

-v((l))= 1,

-v(0)= 0.))

-1
(56:3)))

v(S)

=))

when

S has))

elements.))

This formula (56:3)preciselythe (29:1) 29.1.2.; is theessential is of i.e.P zero-sum three-person game in its reducedform, with 7 = 1. Thus it coincides equallywith the simplemajority gameof threepersons,which was 1 discussedin 2 1. Now we learned of -23. previously from the heuristicdiscussions 21. that this gameis nothing but a competitionof all playersfor coalitions. Indeed, this is immediatelyobvious,considering nature of the simplemajority the of three persons (cf.21.2.1.). a Hence fictitious player will certainly game I show a strong tendency to enter into coalitions. n fact the gameT is, as far as the characteristic function is concerned, completelysymmetricwith to its three players. Consequentlythe two realplayers 1,2 play respect exactly the same roleas the fictitious player 3, and so there is no reason 2 why their ability to enter coalitionsshouldbe at all different from his. to the argument used in the last part of 56.3.3. 56.4.2. canalsorevert We I and apply it to this game:f the fictitious player 3 in T behavesas a realone of to would,he has every reason try to prevent the formation of a couple the and wins is t since players 1,2, he loseshe amount 1if this couple formed,
1Ofcourseall thesegames coincide only as far as their characteristic functions are is basedon the characteristicfunctions alone. but concerned, the entire theory of this: To avoid misunderstandings we re-emphasize The rules of the game f, fully 3C depends t t expressedy the JC,are not at all symmetric with respecto the players b which f on n, rs but not on It is only the characteristicunction v(), (1,2,3), is But we know that v(S)alonematters. (Cf.footnote above.)))) symmetric in

30.1.1.

1,2,3.

n.

1,2,3; SS 1,

510

GAMES GENERAL NON-ZERO-SUM

This compensationcan be determinedby tively, equal to 2 instead of the considerations f 22., and turns out to be $ 2 Thereader ay verify o m 23., l to w this for himself, together ith the fact that this procedureeads the known results concerning simplemajority game of threepersons. the formu56.4.3. examplef 56.4.1. substanceto the objection The o gives latedin 56.3.3. 56.3.4. hus the fictitious player n + 1can influence and T the game T not directly through personalmoves but indirectlyby offering compensationsand thereby modifying the conditions and the outcome this of the competitionfor coalitions. As pointed out at the end of does mean that any such thing happensin T, as long as the fictitious not formalistic fiction. It does happen in T if the theory of player is a mere if the fictitious player is permitted to is applied to it literally, behave (in offering compensations)as if he were a real one. In other of words,thfi considerations the last paragraphsdo not mean that we want to attribute to the fictitious player abilities conflicting with the spirit in which he was introduced. They servedonly to show that an uncompromising application of our original theory to T brings us into such a conflict. a t Hence must concludehat the zero-sumgameF cannot beconsideredn we g unqualified equivalent of the general ameF. to What arewe then to do? In order answerthis question,it is best to o return to the analysisof the specific wherje examplef 56.4.1. the difficulty

1 h the amount 2if it is not formed. Hencee will offer player 1or player 2 a f compensationfor disrupting this couple,i.e.or choosingTI or TJ, respec-

1.

i.e.

56.3.3.,

30.1.1.

i.e.

was expressed fully.

56.5. Two PossibleProcedures The f 56.5.1. might try to escaperom our presentdifficulty by observing One use that it was brought about in 56.4.1.the exclusive of the characteristic by the game T there coincided with the simple majority function. Indeed

closernspection, however, this suggestionappears to be entirely i at leastfor the problemunder consideration. First: bandonment of the characteristic A function v(S) in favor of the 3C* would deprive us of all means to handle the problem.For underlying
At

game of three persons where the mechanism of coalitionformation is beyond doubt only to the extent that they had the samecharacteristic (cf. functions, but not the same3C* in particularfootnotes1and 2 on p.509) Thus a possibleexpedient ight be to abandon the claim that the characm function alone matters, and to base the theory on the 3C*themselves. teristic
without merit

coalition to which

- -K,(r ) - .(,,, r This is the the loss -1,o a t is player 3) i of the 1 2. It also
, r.)
lf

1By

footnote

3 on p. 508and

by

(56:2):
t
)))

from

compensation which brings up the player 1 or 2 (who joined the fictitious

gain
which

1,

players and is indeed what it should

what he would obtain in a couple, in a brings the fictitious player'sgain from 2 down

i.e.

be.)))

EXTENSION OF THE THEORY

zero-sumgames we possessno generaltheory other than that based on v(S) exclusively. Thus the adoptionof this program would make our passage from the generalgame F to the zero-sum game T entirely useless,sinceit would renderus just as incapable of handling zero-sum this gamesas, originally, generalgames. Hence sacrifice of our entire that despite existing theory would bereasonablenly if it werequite certain o its adequacy in all other respectshere was no other avenue of escape. t However,neitherof thesetwo conditionsis fulfilled. Second: retrogression the characteristic from The function to the undernot meet he objectionsf the previousparagraphs. Indeed, o t lying 3C*does at the end of 56.3.2.well as in 56.4.2. did operaten a manner which we as i took the 3Cjt into account. We establishedthe necessityfor the fictitious in player in T to offer compensations a direct manner, a necessitywhich was in no way dependentupon a replacement T by a different gamewith of

511 of 30.1.1.,

Third:It will appearfrom the discussionwhich follows that it is not necessaryto sacrificethe theory based on the characteristic function, but c rather that the objectionsan be met by a simplerestrictionof its scope. 66.6.2. Reconsideration 56.S.2.-56.4.2. that we were not of shows justified in placingthe blame for our presentdifficulties, with regard to the behavior of the fictitious player, entirely upon the theory of 30.1.1. The considerationsof 56.3.2.-S6.3.4. were entirely heuristic. and 56.4.2. 56.4.2. the undesirableresult Thisis particularlyimportant in the case where the instance. Indeed, treatment wasobtainedin a definite way for a specific of 56.4.2. referredto the \" preliminary\" heuristicdiscussionof the essential zero-sumthree-personame in 21. and not to its exactheory in 32. -23., t g as can What happenedthere in 56.4.2. well as in 56.4.1. bedescribed in the terminology of the exacttheory as follows:the general two-person led g gameF of 56.4.1. to a zero-sum three-personame T which coincides with the simple majority game of three persons.The exacttheory of 30.1.1. provided various solutionsfor this game,which wereclassifiedand and of amounted N analyzed in 33.1. ow the considerations 56.4.1. 56.4.2. the to selecting particularonefrom among thesesolutions: non-discriminaa tory solutionof 33.1.3. to W j Consequentlywe must ask ourselves: as it reasonable select ust this the non-discriminatory solution? Is it not possible that another a oneamong the solutions i.e. discriminatoryonein the senseof 33.1.3. w is free from the objections hich hold us up?
The 56.6. Discriminatory
Solutions

the same characteristic function. l

z 66.6.1. had approachedthe essential ero-sumthree-personame If we g i.e.he simplemajority game of three persons from any otherangle,and t

a if it had beennecessaryto select particular onefrom among its solutions, 1We made repeateduse of such replacementsin but not in the subsequent argument of 56.4.2.

56.4.1.,

!)))

512

GENERAL NON-ZERO-SUM GAMES

the standard of behavior which it represents tory one. This solution for to gives all three playersequal possibilities compete coalitions,and of in the absence any definite motive for discrimination one is tempted to 1 treatit as the most \"natural\" solution of this game. However,in our present situation thereis every reasonto discriminate: In the game P, players 1,2 real players,the original participants of r, are while player 3 is, as repeatedly emphasized,just a formalistic fiction. Throughout the discussionof the precedingparagraphs we have stressed that this player shouldnot compete a coalition, and that he shouldnot for be treatedlike the others. In other words,if we expecto be able at all t to apply the theory of to this situation, then there is an absolute to choose ne o necessityof discriminating againstthe fictitious player3, of those solutionswhich were termed discriminatory in the excluded playerbeingthe fictitious player We saw, loc. it.,hat thesediscriminatory solutionsare characterized c t the standard of by the fact that the excludedlayer whom the solution, p from competingfor a coalition is assigneda fixed behavior, disqualifies amount c in all imputations of the solution. It appearedin that this amount neednot be the minimum at which the excluded playercan maintain himself alone, i.e. ot necessarilyc = n Actually c could be chosen from a certaininterval: 1 ^ c < 66.6.2. this point it may be useful to interrupt the discussionfor a At t moment in ordero comment briefly upon the discriminatory solution which excludes fictitious player in the worst possiblesituation, i.e.with the c= this solution consists of precisely those According to and eachone of the two imputationsin which the fictitious player3 gets real playersgets ^ As pointed out loc. it.his means that the solution c t the standard of behavior restrictsn no way the division of the proceeds i betweenthe two real players. The reason given there is now valid in a much more fundamental way: the bargaining of the players 1,2 becomentirely has e n the s unrestricted, ot only because acceptedtandard of behavior excludes theinterferencef player 3 which was the only normative influence in the o that player 3 relationshipof players 1,2 but also for the still betterreason doesnot exist. It is easy to seethat this removes the threat that player 1 or 2 will forsake cooperationwith the other if his \"fair share\" is not concededy his partner and that insteadhe will cooperate player3 and with b obtain a compensation from that source.

therewould have beena strongpresumptionin favor of the non-discrimina-

i.e.

30.1.1.

3.

i.e. 33.1.,

i.e. 33.1.2.

i.

1.

1.

33.1.1.

1.

1,

i.e.

66.6.1.

66.7.Alternative Possibilities 66.7.1. us now continue the discussionwhere it left off at the end of Let
rigorous

1Ofcourse other solutions are just as goodin the the abovestatement is nevertheless reasonablerime facie.))) p

senseof

but the 30.1.1.,

EXTENSION OF THE THEORY

513

It may seem questionablewhether we shouldinsist upon c = or allow the entire variability 1 ^ c < The first alternative is the more c plausibleprima fade. Indeed, > 1meansthat the realplayersdo not i.e. exploitthe fictitious playerto the utmost of their possibilities, that they do not endeavor to gain as much (asa totality) as feasible. One might view such a self-denial as a compensation paid to the fictitious player by virtue s of the acceptedtable standard of behavior. And sincewe have decided to excludeny participationof the fictitious playerin the interplay of coalia thereis somejustification in forbidding this. tions and compensations, It must be conceded, however, that this argument is not altogether cogent. A (positive) compensationpaid by the fictitious player is a qualitatively different thing from one paid to him. The former is a patent absurdity, sincethe fictitious player does not existand will therefore not pay compensations. The latter,on the other hand, is not absurd at all. i a self-denialn exploiting a possible collective advantage, It merely expresses and we have had several instancesshowingthat a stablestandardof behavior 1 can requiresuch conduct. It is not a priori evident that such a self-denial it is out of the questionin the present situation.2 To exclude would mean o that a stablestandard of behavior in the presencef completeinformation necessarilyentails attainment of the maximum collective benefit. The l readerwho is familiar with the existingsociologicaliterature will know of that the discussion this point is far from concluded. in succeed settlingthis questionwithin the frameWe shall nevertheless work of our theory by showingthat c must berestrictedto its minimum value. 3 For 56.7.2. the moment, however, we must developboth alternatives

1,

concurrently. Tothis end wereturn to the general -persongame r, and the correspondn game T. We are now able to formulate the ing zero-sum (n + l)-person relevant conceptsigorously. r the Denote setof all solutionsV of T by 12. (56:A:a) Given a number c, denote the system of thosesolutionsV (56:A:b)
of T for which every imputation e V has a n.fi = c,4 by Q

a =

{i,

an , <*n+i}

of

1This is, of course,ust another way to express possibility of Another the j Still another of instance, in a zero-sum four-person game, is given in (38 obtains for all decomposable (In this last instance the self-denialis games in exercisedy the players of A when < 0, and by those of H when & > 0, cf loc.cit.) b u We emphasize that such self-denial is exercisednder the pressure of the accepted to be standard of behavior, although the players are assumed as always in our theory informed fully about the possibilities of the game. 1However if it occurred, it would be regarded normally as an inefficient though stable form of social rganization. o 8 the self-denial in question doesnot occurand the maximum socialbenefit is we s always obtained. This result is not as sweeping as it may seem, ince are assuming a numerical and unrestrictedly transferable utility, as well as completeinformation. 4 where the fictitious player gets the same amount c in all imputations of the

46.11.

:F)

33.1.2. 38.3.2.

I.e.

I.e.

solution.)))

514
(56:A:c) (56:A:d)
ft\".

GAMES GENERAL NON-ZERO-SUM

Denotethe sum of all setsQc by ft'. Denotethe ft c of c = v((n + 1)) =

1

-v((l,

n)) by

Forsomec the set ft c is empty. Thesec may obviously be omitted when ft' is formed. Thus an+i ^ v((n + 1)) , n)) necessiv((l, otherwiseft c is empty. Again)) tatesc <z v((l, , rc)), =
c necessitates^

w w In connectionith (56:A:c) e notethis:

a. g -))
ft c

fc-i

^ v((fc)), otherwise
,

is empty. So c is subjectto
n

the restriction (56:4)

Actually it

-v((l,

n)) S c S
2

-))

Jb-l

The ft\" of (56:A:d) elongsto the b

is usually even more restricted.

minimum

c of

(56:4).

The 56.8. New

Setup

the solutionsof ft all significant for T. The analysis of restricted those solutions further, but it left the question unansweredwhether the system of all significant solutionsis ft' or 12\". Thus the systems 12'and 0\" correspondto the two alternativesreferred

are certainly not

showed Our 56.8.1. discussionof S6.3.2.-56.4.3. that

56.6.1.

to.

to We now proceed differentiate t Considerhe imputations

between 12'and
,

ft\".

(56:5)

*=

{i,

n,

n+i}

1, , n respectively,are to obtain from this imputation. The last a the a component, +i on the other hand, expresses fictitious operation: amount attributed to the fictitious player n + 1. Further,this component
1 where the fictitious player gets, in all imputations of the solution, only that amount which he could obtain for himself even in opposition to all others. This means as we know that the real players obtain together the maximum collectiveenefit. b 1Thus in the essential ero-sum three-persongame, (56:4) ives z g
while

of the game T. Among the components i, r the , an expressealities: amounts ones,01,

, n , an+i the n first which the real players,

I.e.

we know

from

that 32.2.2. the exactdomain

-1
-1

c c<

2, of c (with
J.)))

non-empty

ft) is

EXTENSION OF THE THEORY

515

an +i is not
zero) (56:6) ents

only fictitious in the interpretation of P, but it is also matheit is determinedif the ai, matically unnecessary, , an areknown. Indeed (sincethe sum of all componentsof the imputation a must be

i.e.

an+i =

- J) a
n

.))

a Consequentlyit may be preferableto express by specifying its componan only, always remembering an+i can be obtained if that i, , T desired from (56:6). hus we shall write
(56:7)
We observethat

7= {{at,

-,.}}.
{

w i.e. e wish to befree to useboth (56:5)and (56:7),whichever may be more

the this notation is not intendedto supersede original one,

w i suitable. Indeed,t is in orderto avoid misunderstandings hich might ensue from this double notation, that we are using the double brackets 1 in (56'.5). }}in (56:7)instead of the simpleones The 56.8.2. imputation a in its form (56:5)was subjectto the zero-sum a restriction, nd also to the restriction a, ^ v((t)) for i = 1, , n, n + 1. (56:8) for We must express56:8) (56:7)(with (56:6)). ( for t = 1, Now , n (56:8)is unaffected by the transition from S but for t = n + 1 we must make use of (56:6). o it to (56:5) (56:7),
{ {

becomes

-ia* ^
1

-v((n+ 1)) = v((l,

all along,

n)).

Of coursewe could have done this games. Herean imputation

for i.e. the

original zero-sum n-person

a
is determined
if

only

its components

, a*) i ?* are given ,-,

i*

(ai,

(for any fixed t

), since))

that the imputain In conformity with this we have observed already in l)-dimensional, and not an tions of the (essential)zero-sum n-person game form an (n n-dimensional, manifold. However, there was no particular advantage to be gained by getting rid of an v which a f should be eliminated, if any. In the graphical and there was no way to decide xero-sum three-person game we actually made an effort to discussion of the essential keepall a, in the picture. (Cf. r The situation now is altogether different, considering the special ole of <x+i. The elimination of <x+i will be essential for our subsequent deductions.)))

(31:1) 31.2.1.

32.1.2.)

516 (56:9) (56:10)

GAMES GENERAL NON-ZERO-SUM

Thus (56:8) over into goes

a<

this:
for
<

v((t))

i = 1,
,

n;

t-i

g v((l,

n)).

can get for himself in oppositionto all others. (56:10), however, appears the for the first time. Its meaning becomes transparent if we consider more closely. quantity v((l, , n)) v((l, , n)) is the value of the game for the compositeplayer comprisingall realplayers , n, and playing against the fictitious n + The amount which this compositeplayer gets at the end player of a play is, course)) of

56.0. Reconsiderationof the CaseWhere T is a Zero-sum Game 56.9.1. us stop for a moment to interpret theserestrictions. Let is not new. It expresses (56:9) again what we had already for the i less zero-sumgames, amely that nobodywould acceptn any case than he n

1.

1,

in

control all moves the fictitious player having no influence on the course of the game. In comparingthis with the zero-sumtwo-personschemedescribed in

a He controls the variables TI, , r n , i.e. ll the variables which occur T this expression. hus in the zero-sum two-persongame the real players

14.1.1. our

to the one variable TI there, while no domain of variability in our present set-upcorrespondsto the variable r 2 there. It is intuitively clearthat the value of such a game (for the first player) t obtains by maximizing with respecto all variables (sincethey are all

2) *-l

3C* correspondsto 5C there, all

our variables TI,

- , rn

controlled him). This by

is))

(56:11)
in our

Maxrj

.....
%

k-l

3C*(n,

, r n)

in set up, the correspondingexpression

the scheme of

14.1.1. being
v8
in)))

(56:12)
in

MaxTi OC(ri,r 2)

(r*

is really absent).

is void. Sothe gameis strictly determinedand has the value

since are 14.4.1.equal to eachotherand to (56:12), the operation MinTj

the Of course systematictheory of

17. 14., gives the same result:Vi,

(56:12)

EXTENSION OF THE THEORY

517

the senseof

14.4.2. 14.5. onsequentlythe generaltheory of and C yieldsnecessarilythe same value.

So we see:))

17.

(56:13) v((l,

- -

,n))

- Max

Ti

.....
%

*-i

X*(n,

, r n ).

this: Consequently(56:10) expresses No imputation should offer all (real) i more than the totality can expectn the most favorable players together and the bestpossiblestrategy.1 2 i.e.ssumingcomplete case, a co-operation
*

Summing

up:

(56:B)

(56:B:a)

No real player must be offered lessthan he can obtain for himself even in oppositionto all otherplayers (cf. (56:9)). (56:B:b) All real playerstogethermust not be offered more than the i totality can expectn the most favorable case, assuming and complete o-operation the best possible strategy (cf. c

The imputations of (56:7) are subject to the : restrictions

following

i.e.

and (56:10) (56:13)).

This formulation makes the common-sense eaning of our restrictions m (i.e. quite (56:B:b)) clear:A violation of (56:9) (56:10) (56:B:a), (56:9), a of means (i.e. (56:B:a)) that one of the (real)playersreceivesn offer which is more unfavorable than what can be enforcedagainsthim. A violation of means (i.e. (56:10) of (56:B:b)) that the totality of all (real)playersreceives t an offer which is more favorable than it could ever expecto achieve. It t r seemseasonableo considertheseas preciselythe conditionsunder which a (an players who actrationally will refuse to consider distributionscheme it imputation) because is manifestly absurd. Before 56.9.2. proceedingny further we must oncemore retraceour a steps and compareour present set up with the previousone,in the cases where both apply. of A Specifically: ssume that we are applying the procedure the past to sections an n-person game T which is already zero-sum. Accordingly we form for this gamethe zero-sumn + 1-person game T as describedin a and then proceeds in 56.8.2. 56.2.2.,
1Note that the concept f a beststrategy for the totality of all real players is clearly o then the totality facesa pure maximum defined: if there is complete co-operation, problem. 1If the game in its original form and beforethe normalization of referredto above is performed contains chance moves, then the \"most favorable case\" of and optimal choice must not be taken to include thesetoo. only co-operation is strategies to be assumed, while the chancemoves must be accounted for by forming from the in i expectationvalues. Indeed,t is in this way that we passed 1 , Tn) 9*(T, Ti, which we are the influence of all chance , moves) to the JC*(n, (TO representing

i.e.

11.2.3. 12.1.1.

I.e.

1.2.3.

r)

using

now.)))

and are Obviously the operationsof 56.2.2. 56.8.2. entirely unnecessaryif of F itself is a zero-sumgame; e possessa theory which disposes this case. w on But if a more generaltheory, valid for all games,is to be constructed old this basis, then we must demand that it agreewith the (morespecial) i theory as far as the lattergoes. I.e.n the domain of the old theory, where 1 the new theory is superfluous,the new must agreewith the old. 56.9.3. F is a zero-sumn-persongame means That
n

518 GENERAL NON-ZERO-SUMGAMES It is important not to misunderstandthe meaning of this operation.

*-l

^(n,

, r n) m

0,

solutions of F and T (in the old theory) correspond to each other, the only difference being that the latter takes careof the dummy (thefictitious z player n + 1) also,assigninghim the amount v((n 4- 1)), i.e. ero. (Cf. or 46.9.1.(46:M)in 46.10.4.) Our proposed new theory would obtain the solutions for F from the t (oldtheory) solutionsof P. Hencehe above considerationroves that all p the new solutionsto be obtainedfor F will be among the old ones. Furtherwe more seethat in this casewe can indeedmust take the entiresystem Q of (56:A:a) 56.7.2. t must be noted,however, that in this caseall in I solutionsof Q automatically assignthe fictitious player n + 1 the amount

: + 1 is added to (or removed from) the set*S. I.e. (56:14) v(S) = v(S u (n + 1)) for 8 (1, , n). The special cases = , (1, S , n) give v((n + 1)) = 0, (56:15) (56:16) , n + 1)) = 0. v((l, with (56:14), togethershow that the gameT is decomposable (56:15) the splitting sets (1, , n) constituent , n) and (n + 1). Its (1, is the original game F, while the fictitious player n + 1 is a dummy.2 F as (For the decompositioncf. the end of 42.5.2. well as 43.1. or the dummiescf. footnote 1 on p.340, and the end of 43.4.2.) : Now we can observe 56.9.4. SinceT obtains from F by the addition of a dummy, the First:
player n

i.e. Cn+iCn, O

, r n) m

0.

Thus v(S) is not affected if the fictitious

v((n + 1)). I.e.here

1This is a well

fl

Oc

with

= v((n + 1)), i.e.fl *

ft\".

(Cf.
in the

known methodological principle of mathematical generalization. * Thereadershould recallthat the fictitious player is not, in general, a dummy for game f This may sound paradoxical,but it was establishedin 56.3. the very

caseof the general two-person games r. Indeed,it is just because rules of the the do not in general assign him the role of a dummy that we must restrict the solution V of F to those which do restrict him to such a role. This is the meaning of the discussions
of

special game f

56.3.2.-56.6.2.
We shall determine
in

orderthat

57.5.3. which

the fictitious player

be a dummy.)))

properties of

r are necessary and

sufficient in

EXTENSION OF THE THEORY

519

a i c (56:A:b) nd (56:A:d),oc. it.) Consequentlyany sets we might define between ft and 0\" in particular,both 12'and 8\" of (56:A:c)nd (56:A:d), a Ioc. it. coincide ith Q and areequallyacceptable our purpose. w for c betweenQ' and 8\" which is still ahead of us In other words:the choice is of no significance in this case. Both alternatives herearein agreement with the old theory;indeed,thereis no need hereto abandonthe old theory
at

The 56.9.5. Second: imputations for a zero-sum n-person game were defined in the old theory in this way:

1 all.

(56:C:a) (56:C:b) (56:C:c)

a =
<*

v((t))

(i, for
n

i = 1,
= 0.

-,};
.

n;

-i

<

1 Our new arrangement of (56:7)in 56.8.differs from this.

\"^ ({ (56:C:a*) and and by (56:9), (56:10) (56:16), i = 1, ^ v((t)) for (56:C:b*) n a, 0. (56:C:c*)
<

-,}};
,

wehave Here

n;

i-l

We already know from the precedingremark that there can be no real 2 difference in the present casebetween the old theory and the new one. It is neverthelessuseful to seedirectly that (from the point of view of the and (56:C:a*)-(56:C:c*) old theory) the two procedures 56:C:a)-(56:C:c) ( contain really no discrepancy. The only difference between these two arrangementslies in (56:C:c) we and (56:C:c*). Recalling the definitions of 44.7.2., seethat the differcan alsobe and (56:C:a*)-(56:C:c*) stated encebetween (56:C:a)-(56:C:c) solutionsfor E(G) the second, in this way: Thefirst amounts to considering ; that solutionsfor F(0). Now we have noted in 46.8.1. lies to considering in in the \"normal\" zone of the game T, and by (45:0:b) E(0) and have the same solutions. Thus we have a perfect agreement. F(0) Thesetwo remarksmade systematic use of the theory of composition of and decomposition Chapter IX, in order to analyze the influence of our This procedure on new procedure the zero-sum games contemplated from T to T, which as we saw amounted to consistedmainly in the passage more specialthan the the addition of a dummy to F. This is considerably

45.6.1.,

r.

generalcompositionsdealt with
sum game

c Ioc. it. The specificresults used could

we have not yet

1The necessity of restricting

1Orrather any new one built between ft' and decision

ft\".)))

r.

ft

was

in deduced 56.5.-S6.6. by considering a non-sero

along the lines contemplated

made the

520

GENERAL NON-ZERO-SUM GAMES

accordinglyhave beenobtained with lesseffort than by the use of the far more generaltheoremsreferred to. We shall not enter into this subject a r further sincethe generalesultsof Chapter IX areavailable in any case, nd more our the because above treatment projects present considerations clearlyupon their proper background.
We 56.10.1. now sum in

66.10. Analysis

of the Conceptof Domination

be possible). The concepts f 30.1.1.all based on that of domination. We thereo are fore beginby expressing meaning of domination as defined loc. it.or c f the of imputations (of T) with the new arrangement (56:7)in 56.8.1. Considerwo imputations t
would not

Certainly all solutionsof T in generalcannot be used to define a satiso factory concept f solutionsfor F. This was establishedby the considerLet ation of a special ase i.e. y casuisticprocedure in S6.5.-56.6. us c b the now approachthis problemsystematically; .e.pply to the game formal i a in definition of a solution as given in 30.1.1. try to determine full and which of its features areunsatisfactory and require modification. generality In doing this we shall use the concept f imputation (of T) in the new o in 56.8.1. important point about this arrangement The arrangement (56:7) is that it stressesb initio the primary importanceof the realplayersin P, d i.e. irectsour attention more to F than to T. This doesnot impair, of d to the t course, he fact that we apply the formal theory of 30.1.1. zero-sum n + 1-person ame T, and not to the generaln-person gameF (which g

56.8.

return to the generaln-person game F, its zeroextension F, and the new treatment of imputations as introduced

7 = {{a,, .a.}}, 7 = {{/Si, -,&)}.

Domination))

means that there exists non-empty set S S (1, a

effective for

n, n

+ 1) which

is

a , i.e.
<

(56:17)
such that

inS

V(

S)'

i 8. We wish to express in terms of the ,,ft with i = 1, this , n alone. It is therefore necessaryto distinguishbetweentwo possibilities: 56.10.2. S doesnot contain n + 1. Then First: S s (1, (66:19) , n), S not empty.)))
(56:18)
at > ft
for all in

EXTENSION OF THE THEORY

521

The conditions(56:17), above (56:18) need not be reformulated sincethey involve only the a, ft with i = 1, S , n. Besides (1, , n) in the v(S) of (56:17). Second: does contain n + 1. Put T = S (n + 1). Then S

(56:20)

(1,-

n),

T may

beempty.

The conditions (56:17), above (56:18)

must be reformulated sincethey involve an +i, fti+i. It is natural to form S in (1, , n, n + 1), as (1, , n, in (1, as (1, n + 1) <S; and T. These , n); , n) two setsareclearly equal, but it is neverthelessuseful to have symbolsfor both. We denotethe first by _LSand the second by T.

-T

i.e.

i.e.

Since

n+l

^ a< = 0,

so))

v(S) = -v(JLS) = Hence becomes (56:17)

iinS

tin

-v(-T).

in

-T

(56:21)

in

This involves only the a, with , n. Besides Ts (1, becomes Next (56:18) in the v( 5P) of (56:21). and

-7 i = 1,
1

,-fcv(-T).
, n)

(56:22)

>&
<*n+l

for all t in T,

> 0n+l'

This last inequality meansthat

(56:23)

f i < t ft-i -i
with

(56:22), also involve only the a, ft (56:23) Summing up:

(56:D)

1,

n.

(56:D:a) (56:D:b)

either H means that there exists and an S with (56:19) (56:17), (56:18); or a a T with (56:20) nd (56:21), (56:23). (56:22),
,

i Note that thesecriterianvolve only setsS,T,

cti,

and to the realplayers 1,

ft

with

i 1,

n.

I.e.they
,
n.)))

n) and the (1, refer only to the original game T

f

-Tfi

522

GAMES GENERAL NON-ZERO-SUM

56.10.3. criterion(56:D) of domination was obtained by a literal The the applicationof the original definition of 30.1.1.,applicationbeing made directly to T and then translated in terms of T. This rigorousoperation the out, let us now examine result from the point of having beencarried i let view of interpretation; .e. us see whether the conditions of (56:D) d producea reasonableefinition of domination for the present case. and According to (56:D) domination holds in two cases(56:D:a)
1 original definition of that It expresses thereexistsa groupof (real)players (thesetSof (56:19)),

(56:D:b).

is (56:D:a)merely a restatementof the

30.1. 1.

in a to that in ft (this is t a and who know that they areableas a group,i.e. s an alliance,o (56:18)), enforce preferenceof theirs (this is (56:17)). this on (56:D:b) the other hand is, when viewed in terms of T and of the real players alone,something entirely new. It requires again that there e exista group of (real)players (the set T of (56:20)) achof whom prefers

eachof whom prefers his individual situation

The his own individual situation in a to that in ft (this is (56:22)). is ( ability of this group to enforce the preferencein question (i.e.56:17)) not required. Instead we have the condition that the real players left out of this group must not be able to block the preferred imputation in 2 question, that is insofar as it affects them (this is (56:21)). Finally there is the peculiarcondition that the totality of all (real) s players i.e.ociety as a whole must be worse off under the (preferred) regimea than under the (rejected) egimeft (this is (56:23)). r 56.10.4. strangealternative (56:D:b) of course,obtained by This was, treating the fictitious player n + 1 as a real entity. If we refrain from
1 Applied, however, to the generalgame F, for which that theory was not 1The (real) players left out, those of 7\\ could block the preferred

i.e.
in

intended imputation

!
if))

if they

could get for themselves separately more than a

assigns

to them together;

i.e.

-T
would not

(Note that we had to excludeequality here,sincethat of this is indeed:

block a .) Thenegation

(56:21)
This may be comparedwith its i.e. enforce preference,

tin

J)

-T

<fcv(-T).
ability

the

expressionof the

of the

original group

T to

(56:17)

iinT

J)

v(T).

Itshould
that possible

be noted that neither of (56:17), (56:21) implies the other: It is perfectly the group T can enforcea as far as it affectsthe members of T, and that

at)))

EXTENSION OF THE THEORY

523

o doing that and try to appraisematters in terms of realitiesi.e. f the real The players then it becomes difficult to interpret (56:D:b). best very one can say of it is that it seemso assume the effective operation of an t influence which is definitely setto injure societyas a whole (i.e. totality the of all real players). Specifically this casedomination is asserted when in all players of a certaingroup (of real players)prefer their individual situation in a. to that in , if the remaining (real)players cannot block this and arrangement, if it is definitely injurious to societyas a whole. In comparingthis domination (56:D:b) the ordinary one (56:D:a) to

the followingdifferencesareparticularly conspicuous: First,that in (56:D:a) the ability to enforce one'spreferenceis essential, hile in (56:D:b) w the t essential point is the ability of the others to block it. Second,hat in the (56:D:a) active group had to be a non-empty set,whereasin (56:D:b) it couldalso be an empty set (cf. (56:19) (56:20)). and Third, the antibut social iewpoint figures in (56:D:b), not at all in (56:D:a). v will is The reader have noticed by now that (56:D:b)of a rather irrabut u tional character, neverthelessnot altogether nfamiliar. It would be is an of upon the imagesand allegories which (56:D:b) exact easy to enlarge formalization. Thereis no need to dwell further upon this subjecthere. in What mattersis that we have every reasonto see the alternative (56:D:b) case for those difficultiesfor which a special was analyzed c the general ause is in 56.5.-56.6. Clearly (56:D:b)not an immediately plausibleapproach is. o to the concept f domination in the sensein which (56:D:a) We shall therefore try to resolveour difficulties by the simpleexpedient of rejecting (56:D:b) altogether. Discussion to We have decided redefine domination by rejecting (56:D:b) o This in and retaining (56:D:a)(56:D) of 56.10.2. new concept f dominato tion can be stated in two ways which both seem deserveconsideration. amounts to A First: s pointed out at the beginning of (56:D:a) The only difference)) definition of of a repetition the corresponding

66.11. Rigorous

56.11.1.

56.10.3., 30.1.1.

T can block a as far as it affectsthe members of T. the same time the group that neither group can enforceor prevent anything. other hand it is alsopossible
n

On the

However, if T is a zero-sum game, and if we require

then

(asin the

old theory)

2/

-i

~ 0,

so

are and (56:17) (56:21) equivalent. Indeed,in this case , n)) v(T) + v(-T) v((l,

-0,

2) iin-r

- - 2)

iinT

v(-T)

- -v(T),

from which the equivalence follows

as asserted.)))

524

GAMES GENERAL NON-ZERO-SUM

is that then F was a zero-sum n-person game,while now it is a general n-persongame! m Thus our presentprocedure eans that we extend to the present case the definition of domination in 30.1.1. unchanged,irrespectiveof the fact 1 that the gameis no longer requiredto be of zero sum. from Let Second: us now view the restriction to (56:D:a) the standyielded point of T rather than of F. Our original discussionin 56.10. and the two cases dependingupon the following disjunc(56:D:a) (56:D:b) the tion. In the senseof 30.1.1.domination in T had to bebased on a setS. obtains when n + 1 does not belong to S, while (56:D:b) Now (56:D:a) amounts to requirt obtains when it does.Hencehe restrictionto (56:D:a)

ing that the set S must not contain n

We repeat:Our new concept f domination means, in terms of T, o we add to the conditions that in the definition of domination in imposedupon the set/S, the further condition that S must not contain a specified element, amely n n

+ 1.

(30:4:a)-(30:4:c)

loc. it.: e regarda set S as effective only if it doesnot contain n -f 1. c W t loc. i (Of course, he original condition(30:3) cit.s also required.) We t 66.11*2. now proceedo study the new conceptof solution for P,
In analyzing it we shall rely upon the game T and the form (56:5) f impuo tations (ratherthan upon the game F and the form (56:7)of imputations) and the definition of domination as formulated in the second remark of

This can alsobe construedas a restriction the concept f effectivity o on

30.1.1. + 1.

f i.e.or F, based upon the new conceptof domination, introducedin 56.11.1.

66.11.1.
(66:E)

We obtain our

result by proving four successive lemmas:

If V is a solution for T in the new sense,then every

an+i) a = of V has aw+1 = v((n + 1)). Proof:Assume the opposite.Necessarily an+i ^ v((n + 1)), hence a therewould exist n a = , a , nfi} in V withan +i > v((n + 1)). Put an +i v((n + 0. Define = |0i, , 0,P*+i\\ with

{i,

-,,
7

!))+,>

(i,

Pi

= cti + for i = 1, , n; ^ n +i = <*+! t = v((n + 1)).

1It may seempeculiar that it took us so long to reachthis simple principle, in fact we needthe further considerations of beforewe acceptit finally. However, the act of taking over the definition of in spite of the without any alternatives, extremely wide generalization which is now performed, requires most careful attention. Thedetailedinductive approach given in theseparagraphs seemed be best suited for to

56.11.2.

30.1.1.

this purpose.)))

EXTENSION THE THEORY OF

n

525
ft

Since

t-i = 1, t

ft

-ft+i = -v((n+ 1)) = v((l,

Vf
ft

n)) and
ft

>

<

for

, n,

a belongs to

l = n) establishes -sothe use of S (1, it. ,Hencethere existsHa a . in As _ cannot belong to V y

with the 7 H ft . Now consider set S which enforcesthis domination. SinceS doesnot contain n + 1,we have S (1, , n). As ft > a< > _ for t = 1, , n, 7 H implies 7 H a . But 7 , a areboth in VJ hencewe have a contradiction.
>

(56:F)

sensehere imply the samewith domination in the old sense. Now domination in the new sense hence impliesdomination in the old sense; is our assertion S concerning30:5:b)immediate. o only (30:5:a) ( requires closernspection. i Assume therefore that (30:5:a)invalid in the old sense,i.e.hat for is t two a. , ft in V, a H ft in the old sense. Let Sbe the setwhich enforces this domination. By (56:E) an+i = ft*+i (= v((n + 1))), h$nce n + 1 to cannotbelong S. Consequently a H ft in the new sense,i.e. 30:5 ( :a) fails in the new sensetoo. This completes proof. the If V is a solutionfor T in the old sense,and if every)) (56:G)
in the new

old sense. W Proof: e must showthat

in the

If V is a solution for T in the new sense,then it is so also

with (30:5:a), of 30.1.1., domination (30:5:b)

sense.
in the W Proof: e

of \\f has an+i = v((n + 1)),then V is also a solutionin the new

must show that
with (30:5:a), of 30.1.1. (30:5:b)

domination

old sensehere imply the samewith domination in the new sense. is Now domination in the old sense impliedby domination in the new sense; is hencethis time our assertion concerning (30:5:a)immediate.So only

closernspection. i requires (30:5:b)

t Considerherefore an a = , an , <*n+i} not in V- As (30:5 :b) _ = {0i, , n , n+i} in V with a t holds in the old sense,here exists ft ft H a in the old sense. Let S bethe setwhich enforcesthis domination. to Necessarilyan +i ^ v((n + 1)), and since ft belongs V by assumption, to S. ft ft n +i = v((n + 1)). Hence n +\\ g an +i and so n + 1 cannot belong holds ( Consequently ft H a in the new sense,i.e. 30:5:b) in the new sense the too. This completes proof.
>

{i,

1This domination as well as all others in

this proof

are in

the new sense.)))

526

GAMES GENERAL NON-ZERO-SUM

i S?is a solutionof T in the new sense,f and only if it belongs i to the system Q\" of (56:A:d)n 56.7.2.

(56:H)

Proof:The forward implication results from (56:E) and (56:F),the inverseimplication from (56:G), that 56.11.3. In interpretingthe result of (56 we must remember the :H) the discussionoriginatedfrom the necessityof restricting system 8 of all that solutionsof T for the purposesof the theory of F. We saw in 56.7. the plausibleresult of this restrictionshould be the set Q' or ft\" (or possibly t some betweenthe two). Thereafter our effort was directedo making set a decisionbetween these two possibilities. Furthermorewe concluded
in might

answer our problem.Now the statementof (56:H) is that this modification of the conceptof domination leads preciselyto the set Q\". r By theseconcurrent esults the decisionis clearly indicated. We accept Q\" as the system of all solutionsfor F.

that a modification of the 56.10.-56.11.1.

conceptof domination

in T

56.12. New The

Definition of

a Solution

t 56.12. reformulate this togetherwith referenceso the main results We on which the decisionwasbased

(56:I:a)

(56:1)

For a generalw-person gameF, a solution is any solution of its zero-sumextension, the (in the original senseof zero-sumn 1-person game T, for which all

30.1.1.)

a =
in V have

{i,

-,,,
<* n
{

+i}

(56:24)

an+1 = v((n + 1)). Thesesolutionsform preciselythe set Q\" (56:I:b) Using the form (56:7),a = {ai;

i of (56:A:d)n 56.7.2. - , an } for these } F and its players rather than T), imputations (i.e. emphasizing transforms (56:24) bove into a
fc

(56:25)

-i

<

= v((l,

n)).

(56:I:c)

This is clearlya strengthenedorm of (56:10)56.8.2. f in In the special in which F is itself a zero-sumgame,our case new concept f solutions (for F) coincides ith the old one, o w i.e.he unmodified applicationof 30.1.1. the first remark t (Cf. in 56.9.4.) hus it is no longer necessary to distinguish T betweenthe old theory and the new one. (Cf.alsofootnote 1
on

p.518.))))

(56:I:d)

obtained by applying the definitions of 30.1.1. (which were intended for zero-sumgames only) to F directly and without o any modification. Theconcept f imputationsfor F must then be used in the form (56:7). (Cf.the first remark in 56.11.1.) means (56:I:e) The validity of (56:I:d) that nothing must be added o to the characterizationf the imputations in the form (56:7) the as given in 56.8.2. However,by (56:I:b) equation (56:25) will then automatically hold in each solution \\7. Hence we i.e. of may, if we wish, add (56:25), strengthen (56:10)

THE CHARACTERISTICUNCTION 527 F For a generaln-person gameF, the solutions can also be

(56:I:f)

Therestriction imposedin (56:I:a) the solutionsof F upon o can also be expressed modifying the concept f domination by for F but then allowing all solutionsin the modifiedsense. This modification consists of imposingupon effective sets (in the the further requirement that they must not senseof remark of contain n + (Cf. the second

to 56.8.2.(56:25).*

30.1.1.) 1.

56.11.1.)

F 57. TheCharacteristicunction and Related Topics The 57.1. CharacteristicFunction :TheExtended and the Restricted Forms of We 57.1. arenow in the possession a theory which appliesto all games

which actually define functions OC*(ri, , n of , r n ), k = the game,do not affect the theory directly, but only through the characteristic function v(S).2 There is, however, a difference between the use of the characteristic a function v(S)for a zero-sumgame, nd for a general ame. Fora zero-sum g function v(/S) is defined for all sets F the characteristic n-person game F S (1, n , n) and for only these. (Cf. 25.1.) ora general -person the F we had to form its zero-sumextension, zero-sum (n + l)-person game function v(S) was actually formed like the game F, and the characteristic function (in the old sense)of F. (This is the v(S) which characteristic

o for and like the theory of 30.1.1.zero-sum games, f which it is an extenfunction. I.e.the sion is based exclusively upon the characteristic

1,

11.2.3.,

1This permissibility

of restricting
n

(56:10)
to

i-i
n

v((l,

,n))

(56.25)

2} t-i 56.9.5.

. - v((l,

, n)) E(Q)and F(0) referred to in ,

is analogous to (but more general than) the equivalenceof remark of the second 1Of course (S)is defined with the help of the 3C*(n, v

r).

Cf.25.1.3.58.1.))) and

528

GENERALNON-ZERO-SUM GAMES

figured in all of our recentdiscussions,particularly throughout Accordingly 56.7.2., 56.8.2., , n, n + 1) and for only these. v(S) is now defined for all setsS (1,We may, however, if we wish, considerv(S) for the setsS (1, , n) only. When this is done,we shall speakof the restrictedcharacteristicfuncall , n, n + 1) tion; while v(S)in its original domain, embracing S S (1, is the extendedcharacteristicfunction. H in From this we conclude the special caseof a zero-sum game: ere function of the old theory is the restrictedharacteristic c the characteristic 1 function of the new one. function Returning to the general ames,we seethat the characteristic g is the basis of our entirepresent theory. Among the equivalent formulafuncin characteristic uses tions of that theory, (56:I:a)56.12. the extended o uses while (56:I:d) the restrictedne. tion, Consequentlyour next objective is necessarilythe determination of to the nature of thesecharacteristic functions, and of their relationship each

56.5.1.,

56.4.1., 56.9.1., 56.9.3.-56.10.3.,56.11.2.-56.12.)

other.

57.2.Fundamental Properties

Considera general n-persongame F, and its two characteristic functions, as defined above:The restrictedne,v(S) defined for all subsets o = (1, , n) and the extendedone,v(S)definedfor all subsetsS of

57.2.1.

Soil
(1, (1,
the

- - , n, n In what follows we must distinguishbetween two possibilitiesin our = as in the secondremark in notations for = as while for S , n, n 1) we can form S in 7, S in as We denote again , n) we can also form the first setby XS and the secondone by We proposeto determinethe essentialpropertiesof both characteristic for and functions of the general -persongame just as we did in n

J = (1,-

+ !).

S,

56.10.2. ForSsJ / i.e. / S, /, i.e. / S*

S.

Consider the extended characteristic function first. Sinceit is the

in the old sensefor the zero-sum (n l)-person it must have the properties formulated in game T, = , n only with 1) in placeof the = , n) there. In this way we obtain:

function of the zero-sum n-persongame. characteristic

25.3. 26., +

f characteristicunction

(1,

(25:3:a)-(25:3:c) / (1,

25.3.1.

(57:1a) :
1

v(0)
that for

- 0,

All

thesedistinctions

lished

in (Cf. (56:I:c) 56.12.)

fact

all zero-sum games the

and definitions cannot and do not affect the rigorously estabtwo theories are equivalent to each other.

are defined.
1Formed SandT.)))

* We denote them by the same letter, v,

for the same that they

sincethey have the samevalue wherever

there we formed them for two different

both

cit. we claimed

S (of course8 /), thesetwo setsare clearly different.

are equal, but

Loc.

sets

(57:1b) : (57:1c) :

THE CHARACTERISTICUNCTION F v(S) = -v(S), 5 n T = 0. v(8 u T) v(5) + v(T) if

OS,TfiJ).

529

Consider next the restricted characteristic function. We obtain conditions for it from (57:l:a)-(57:l:c), ourselvesto subsets of J. by restricting This is immediately feasiblefor (57:l:a), but (57:l:c), it is impossiblefor (57:1b).* In this way we obtain: :

(57:2:a)

(57:2:c) v(S u T)

v(fl)

v(0) = 0, + v(T) if

Note that we cannot replace by (57:l:b) something equivalent for 8. a Indeed,ll we can do with S,is to put T = S in (57:1c). This gives :

(5, Ts7)

Sn T =

0.

(57:2:b)
Even if v(7)

v(-S)g v(7) - v(S). v(-S)^ -v(S),

in (25:3:b)25.3.1.

= 0,which neednot be the case,(57:2:b) becomes merely

(57:2:b*)
but not the equivalent of

v(-S)= -v(S).
as well (57:l:a)-(57:l:c) as (57:2:a), (57:2:c) re, by virtue of their a or derivation, only necessaryproperties of the (extended restricted)charf acteristicunctions. We must now seewhether they aresufficient as well. If 67.2.2. T were an arbitrary zero-sum(n + l)-persongame,then we f that could concluderom the result of 26.2. any v(S) which fulfills (57:l:a)function of a suitable T, i.e.he t is (57:l:c)the (old sense)characteristic function of a suitable generaln-persongame F. extendedcharacteristic are In other words:This would prove that the conditions(57:l:a)-(57:l:c) mathematical necessary and sufficient that they contain a complete functions of all possible general of characterization the characteristic

the However,T is not at all arbitrary. As we saw in 56.2.2., (fictitious) h n + I has no influence on the courseof the game i.e. e has no player , rn , r n +i) do not really depend on his moves; the 3C*(ri, personal i that variable rn+i. Furthermore,t is clearfrom 56.2.2. this is the only If restrictionto which T must be subjected: in a zero-sumn + 1-person gameF the player n + 1has no influence on the courseof the game,then of we can view T as the zero-sum extension a generaln-person game F the remaining players 1, , n.2 playedby
contain

gamesF. n-person

1S,

game

-f 1 as if he were a fictitious player as far_as the rules of the are concerned. We know, of course,that there are solutions V for f which bring in but not in Q\", cf. (56:A:a)-(56:A:d) out the fact that he is a real player (those in recallalso56.3.2., and 56.7.2. (56:1 in 56.12.; 56.3.4.).))) a)

I.e.

1. we can treat n
-f

S cannot be both

/(!,

, n) since one of them

must

necessarily

530

GENERAL NON-ZERO-SUM GAMES

arenecesConsequentlythe following question arises: 57:l:a)-(57:l:c) ( functions in the old sary and sufficient conditions for the characteristic senseof all zero-sumn + 1 -person games. Howmust they be strengthened so as to do the same thing for the (old sense)characteristic functions of all those zero-sum n + 1-personames in which the player n + 1 has no g
of influence on the course the game? m Answering this question would amount to giving a complete athematical characterization for the extendedcharacteristic functions of all g generaln + 1-personames. But then the problem of doing the same c functions would still remain. for the restrictedharacteristic It will be seen that by attacking the last problem first, a somewhat t more advantageousarrangement obtains:he first problem can be solved in a few lines with the help of the latterone. However,our approach will

be dominatedby the above considerations.

D 67.3. etermination

of All CharacteristicFunctions

We to 57.3.1. proceed prove that the necessary conditions (57:2:a), are (57:2:c) also sufficient: For any numerical setfunction v(S) which fulthere a fills (57:2:a), (57:2:c) exists generaln-persongame F of which this c function. 1 v(/S) is the restrictedharacteristic In orderto avoid confusion it is betterto denotethe given numerical set-function which fulfills (57:2:a>, (57:2:c) y v (S). With its help we b shall define a certaingeneraln-person gameF, and denotethe restricted characteristic function of this F by v(/S). It will then be necessaryto prove that v(S) = vo(S).

Let therefore a numerical set-function v (S) which fulfills (57:2:a), 2 We define the general -persongame F as follows: n = 1, Each player k a , n will, by a personal move, choose subset Sk of / which containsfc. Eachonemakeshis choice independentlyof the o choicef the otherplayers. After this the payments aremade as follows: Any set S of players for which be (57:2:c)given.

(57:3)

Sk

=S

for every

A;

belongingto S

a is called ring. Any two rings with a common element reidentical.In a otherwords:Thetotality of all rings (which actually have formed in a play) is a system of pairwisedisjunct subsets of Each player who is contained in none of the rings thus defined forms a set by himself a (one-element) which is called solo set. Thus the totality of all ringsand solosets(which actually have formed in a play) is a decompoa sition of 7, i.e. system of pairwisedisjunct subsets of / with the sum I. numbers of their Denotethese sets by Ci, , Cp and the respective n P. elements y n\\, b ,

/.

1The construction which follows has much in common with that of 1Thereadershould now compare detailswith those in 26.1.2.))) the

26.1

Now

Ci,
(57:4)

THE CHARACTERISTICUNCTION F 631 consider player k. He belongs to preciselyone of these sets a

,

Cp, say to

C,. Then the player k getsthe amount

nq

I v,(C,).
Vo(C

This completes descriptionof the game F. F is clearly a general the a w extension is. We emphaT n-persongame, nd it is clear hat its zero-sum size in particular that in F the fictitious player n + 1getsthe amount

(57:5)

- -i
Q

fl

).'

We arenow going to show that F has the desiredestrictedharacteristic r c function v (/S). 57.3.2. Denotethe restricted characteristic function of F by v(S). h Rememberthat (57:2:a), (57:2:c)old for v(S) becauseit is a restricted characteristic function, and for v (S) by hypothesis. If S is empty, then v(S) = v (S) by (57:2:a). we may assumethat So S is not empty. In this casea coalition of all playersbelongingto S can of govern the choices its Sk so as with certainty to make S a ring. It his Whatever the otherplayers sufficesfor every k in S to choose Sk = Cp (in S) do, S will thus be one of the sets(rings or solo sets)C\\ Cq Eachk in C, = Sgetsthe amount (57:4), ence theentirecoalition h say getsthe amount v (S). Consequently

S.

(57:6)

v(S) ^

(5).

A coalition of all playersk belongt Now considerhe complement of to S can govern the choices its k so as with certainty to make S ing a sum of rings and solo sets. If S is empty, then this is automatically If S is not empty, then it suffices for every true, sincethen 5 = k in S to choosehis Sk = Hence S is a ring, and therefore S sets. is a sum of rings and solo Thus S is the sum of someamong the setsCi, , Cp say

S.

/.

S.

of))

togetherget the amount

r

r' among , p). (!',s' =, i' aresomegetsthe the numbers 1,hencethe n Each in C = amount (57:4), r') players in C
k

q q

(q

(C),and so all players of S togetherget the

CV,

amount

s-l

v (CV)
in

Since the
C9 get
together

CV

are pairwise disjunct sets

p

1The n q

players

the amount

henceall players vo(C) by (57:4);

the amount

1,

, n,

i.e.ll a

players in Ci,

, C , get
p

together

2^ vo(C).

(57:5)ensues.)))

-i

Now

632
with

GENERAL NON-ZERO-SUMGAMES

the sum

5,

repeated applicationof

(57:2:c) gives

I.e.: hatever the players of S do, togetherthey get an amount W

Consequently

-i

(C,')^ ^

v v

(S). (S).

v(S) g vo(S). Now (57:6),57:7)togethergive ( v(S) = voGS), (57:8) (57:7)

as desired.

use will be denotedby v(/S).

of which this v(S) is the extended function. characteristic In ordero avoid confusion, it is again betterto denotethe given numert ical set function which fulfills (57:l:a)-(57:l:c) by Vo(S). The extended characteristic function of the generaln-persongame F which we shall

are necessary. We (57:l:a)-(57:l:c) shall prove that they arealso sufficient: That for any numerical setfunction a there v(S) which fulfills (57:l:a)-(57:l:c)exists generaln-person gameF Herewe

57.3.3. us Let

now considerthe know that the conditions

functions. extended characteristic

1 such that its restricted characteristic function has always v(S) = voOS) = v ($) for S /. and so its extendedcharacteristic function has v(S) if I.e., we revert to the natural domain of theseS,2 then we have: v(S) = vo(S) if n + 1is not in S. (57:9) Now let n + 1 be in S. Then it is not in S. Hence (57:9)gives

Considerit for a moment for thesets5 / = (1, , n) (57:l:c)begiven. our Hence constructionof 57.3.1., only, then it fulfills (57:2:a), (57:2:c). 57.3.2. beappliedto this vo(/S). Soa generaln-persongameF obtains, can

Let therefore a

numerical

set function

(S) which

fulfills

(57:l:a)-

it :c) v(JL/S) = vo(J_S). (57:1a)-(57:lhold for v(S)because is an extended : characteristic function, and for v (/S) by hypothesis. Therefore(57:l:b) gives v(J_/S) = v(/S), v (J_S) = v (/S). All these equations combine to if n + 1is in v(5) = v (fl) Now (57:9), 57:10) ( togethergive (57:11) v(S) = v (S)

(57:10)

S.

unrestrictedly,as desired.

mathematical character67.3.4. sum up:We have obtainedcomplete To functions and izations of both the restricted the extendedcharacteristic of all possiblegeneraln-person F. The former aredescribed games v(S) and by (57:2:a), (57:2:c), the latterby (57:l:a)-(57:l:c).
1The \"always\" in this caserefers,of course, nly to the S o 1 Which in this caseconsists all S fi of
/.)))

S/.

THE CHARACTERISTICUNCTION F

533

We follow therefore the comparableprocedure 26.2., call the of and functions which satisfy theseconditionsrestrictedcharacteristicfunctions or extendedcharacteristic Junctions, respectively even when thev are viewed in themselves,without reference any game. to

67.4.Removable Setsof Players we obtained for extended characteristic functions can also bestated as follows:Every characteristic function (in the old charactersense)of any zero-sum(n + l)-person game is also the extended 1 istic function of a suitable generaln-person game. Rememberingthe discussion of 57.2.2., means:Every characteristic this function of any zero-sum n + 1-person function of a game is also the characteristic suitable zero-sumn + 1-personamein which the player n + 1 has no g of influence on the course the play. i Let us replacen this statementn + I by n, obtaining the equivalent for zero-sumn-person games and the roleof the player n. In orderto formulate this result,it is convenient to define: Let a zero-sum (57: n-persongame PandasetSsI = (1, A) , n) be given. Then we call S removable for F, if it is possibleto find another zero-sumn-person game I\", which has the same function as F but in which no playerbelonging S to characteristic of has an influence upon the course the game. that Using this definition, our assertionbecomes the set S = (n) is removthe able. Given any playerk = , n, we can interchange rolesof the players k and n, hencethe set S = (k) is also removable.So we see:

67.4.1. result which The

!,

set Every one-element >S is removable in every game F. to Now it shouldbe noted that according our theory the entirestrategy i of coalitionsand compensationsn A gamedependsonly on its characteristic function. Consequentlythe two games F and F' of (57:A) are exactly alike from that point of view. of as Hence (57:B)can beinterpreted follows:Therole any one playerin of p any zero-sumn-person game insofar as the strategic ossibilities coalitions and compensationsare concerned can be duplicatedexactly in an of w arrangement hich deprives him of all directinfluence upon the course sense: in the game. Herewe mean his \"role\" the most extended including his relationshipto all otherplayers,and his influence on their relationships to eachother. a mechanismby which In otherwords:We described S6.3.2.-56.3.4. in a player who has no directinfluence upon the courseof the gamecan neverthelessinfluence the negotiations for coalitionsand compensations. t We have now shown in (57:B),hat this mechanismis perfectly adequate of 25.3.1. with (25:3:a)-(25:3:c) coincide 'Indeedthe conditions (57:l:a)-(57:l:c) with 7 (1, , (1, , n, n + 1)in placeof its /
(57:B)

n).)))

534 to

GAMES GENERAL NON-ZERO-SUM

describe influence which any player in any game could have in this the respect. This statement must be taken absolutely literally: Our result t details and nuanceswill be reproduced. guaranteeshat all conceivable 57.4.2. (57:B)every playerk = , n is removable individually By i.e.he one-element S = (k) is but this doesnot mean that all these t set t players areremovable simultaneously i.e.hat the set S = / = (1, , n)

!,

is. Indeedwe have:

(57:C)

The set S = / is removable inessential.

if and only

if

the game T is

no player k = 1, , n has an influence on the course of the game F',means that all functions 3C[(Ti, , r n ) are independent of all their variablesTI, that they areconstants , rn T Proof: hat

i.e.

(57:12)
From this

aci(n,v(S) = %
kin
<*k

,r ) = <**.
n

(57:13)

for all

SzL

i is Conversely,f (57:13) required,it can be secured (57:12). by is Hence function of a game T for which such (57:13)the characteristic a I*exists and (57:13) preciselythe definition of inessentiality. is For n = 1,2 every game F is inessential,hencethere the set S = / 1 and with it every set is removable. For n ^ 3 there existessential a n games, nd therefore S = I is in general ot removable. Therefore this questionarises: Which arethe removable sets for an essentialgame F? (57:D) T setsareremovable, (57:B),57:C)contain a partial answer: heone-element ( the n-element (S = /) is not. Where is the dividing line? set The 57.4.3. upper extremes reachedwhen all (n l)-element i sets and with them all setsexcept are removable. We call such a game / extreme. It is worth while to visualize what this property entails: he T strategicsituation in such a game is equivalent to that where only one of player has an influence on the course the game,and the roleof all others consistsmerely in trying to influence his decisions. hemeansof influencT him is, course, ffering him compensations; motive is to induce of the ing o 1The main result concerning zero-sum two-person games, according to which each game of this type has a definite value for eachplayer (say v, v cf. the discussion of 17.8., means just this: It statesthat the game is equivalent to the fixed payments 17.9.), v, v to the two players and this is an arrangement where neither of them can influence
anything. for coalitions

all

In every essential ame, on the other hand, there exists the interplay of negotiations g and compensations and this excludeshe simultaneous removability of t

players.)))

THE CHARACTERISTICUNCTION F 535 him to makedecisions which arefavorable to the playeror playerswho make
the offer. (57:E) (57:F)
Now we can prove:

For n = 3: The essential ero-sum three-person z game is extreme. Forn = 4 :Thereexist xtremes well as non-extreme e a essential zero-sumfour-persongames.

More in detail:

For the essentialzero-sum three-person game, all twos element etsareremovable. For an essentialzero-sum four-person game all three(57:F*) 1 element etsareremovable, or all but one.-2 s The proofsof thesestatements present no serious difficulties, but we do not
(57:E*)
proposeto give them here. The results (57:B), 57:0), :E),(57:F) show that a generaltheory (57 ( of removable setsand extremeames is not likely to be very simple. It g will be considered systematicallyin a subsequentpublication.

simply of the characteristic function of F. This is also in harmony with our preceding games,sincenow the old terminology for zero-sumn-person o function areconcordant. and the new use of the concept f a characteristic (Cf.the remarks next to the end of 57.1.) t Consideringthese arrangements,he definition of the conceptof a of solutionmust be that describedin (56:I:d) 56.12.mputations arebest I and in defined as described (56:I:b) in the last part of (56:I:e) It this w seems orth while to restate latterdefinition explicitly: An imputation is a vector

57.5. trategic Equivalence. Zero-sum and Constant-sum Games S 57.5.1. have exhaustedthe usefulnessof the zero-sumextensionP We of the general -persongame F, and therefore from now on we shall discuss n the theory of generaln-persongames without referring to that concept. h Consequently, ereafter we shalluse only the gameF itself and its restricted characteristic function, unlessexplicitlystated to the contrary. Forthis \" reasonthe qualification \" restricted will be dropped, and we shall speak

(57:14)
the components

(57:15)

!,,
a,
v((i))

id.

7 - {{!, -,)}
for

being subjectto the conditions

i = 1,

n;

1Every two-element set is a subset of two three-elementsets(remember that n 4), and by the above at least one of these is removable. Henceevery two-element set is removable in any event. 1The parts of the cube Q of 34.2.2. to which correspond thesevarious alternatives

canbe explicitly

determined.)))

536

GENERAL NON-ZERO-SUMGAMES
en

(57:16)
We

- v(/).

and i.e. setup. This will be done exactly as in 42.2. 42.3.1.,in

o can now extendthe concept f strategicequivalenceto the present analogy to

27.1.1.:
(57:17)

and a setof constants aj, tions 3C*(ri, ' ' , rn ) by

*

Given a generaln-person , rn ) game T with the functions 3Cfc(ri, we define a new gameT'with the func, J

3Ci(r!, , rn ) = 3C*(ri, , rn ) + aj. functions From this we conclude, exactly as before,that the characteristic and v'(S) of thesetwo gamesareconnected by))

(57:18)
We call two

v'(S) = v(S) +

kinS

aj.

such games,as well as their characteristic functions, stra-

tegicallyequivalent. Sincewe arefree of all zero-sumrestrictions, constants a\", the , aJJ areunrestricted, as in (42:B)in 42.2.2. just We note that this strategic equivalenceinducesan isomorphismof the of F and I\" exactly as in the two previousinstanceseferredto r imputations and of above. Specificallythe considerationsand conclusionsof 42.4.2. over to the presentcaseunchanged, that it seems nnecesu so carry sary to reformulate them explicitly. 67.5.2. domain of all characteristic The functions (of all general -person n which games)was characterized the conditions (57:2:a), (57:2:c), we by

31.3.3.

restate: (57:2:a)

v(0) = 0, (57:2:c) v(S iTT) ^ v(S) + v(T) for S n T = functions of zero-sum Among thesethe characteristic gamesand of constantsum gamesform two special lasses.The former are characterized c by

0.

condition

of 25.3.1. I (25:3:a)-(25:3:c) (Cf. 26.2.) .e.we must add to our (57:2:a), with (which (57:2:c) coincide the (25:3:a), mentioned)the further (25:3:c)

Thelatterarecharacterizedy (42:6:a)-(42:6:c) (cf.id.).I.e. e in 42.3.2. w b must add to our (57:2:a), w coincideith the (42:6:a), (which (42:6:c) (57:2:c)
1As was pointed
out

(57:19)

v(-S)= -v(S).

loc.cit. we could have used equivalently

n

X i-l
Indeedthis is the
original form

<

* v(7).
However, we prefer (57:16).)))

of this

condition.

THE CHARACTERISTICUNCTION F
mentioned)the further condition

537

assertion.Hence sufficesto showthat (57:19) it impliesv(7) = 0. Indeed, = v(-0)= -v(0)= 0. (57:2:a), give v(I) (57:19) Note that (57:20)s the assertion i that equality holds in (57:2:c) when S u T = I.2 Thus the v(S) of constant-sum games are characterized
of by the property that the merger two distinct coalitionsSand T produces no further profit if togetherthey contain all players. For the v(S) of zero-sum gamesthe further requirement v(7) =0
must

(57:20) v(fi) +v(-S) v(J). Sincethe zero-sumgamesarea special of the constant-sumgames, case m of (57:20) ust be a consequence (57:19), assuming(57:2:a), always (57:2:c). This is indeedso; e can actually prove somewhatmore, amely: w n is (57:G) (57:19) equivalent to the conjunction of (57:20)with v(/) = 0. 1 and Proof: Assuming v(/) = 0, (57:19) (57:20)are clearly the same

To conclude, emphasizethat the extraconditions(57:19) (57:20) or we do not mean that any gamewith such a characteristic function is necessarily a zero-sumor a constant-sumgame. They imply only that such a characteristic function must belong amongothers to at leastone zerosum or constant-sum game. It can happen that a gamewithout being zero-sum (or constant-sum)itself has such a characteristic function, the characteristic function of a zero-sum (or constant-sum) game. In this case will behave from the point of view of the strategy of coalitions, it and the compensationslike a zero-sum (or constant-sum) amewithout g actually being one. 67.6.3. are now in the position to settlea question which was in We the foreground several times in our discussions.The analysis of 56.3.2.56.4.3. concerned was already with the fact that the fictitious player in n of his unreality is not ipsofacto a dummy. I.e. ot onein the sense spite function and the decomposition of the extended characteristic theory of the zero-sumextension 8 This subjectcameup again at the beginningof I\\ where 56.9.3., we noted that he is a dummy for zero-sumgamesF. F The question which we will answer now is accordinglythis: or which T is the fictitious player a dummy? 4 We prove: g general ames The fictitious player is a dummy if and only if T has the (57:H) if function as a constant-sum game samecharacteristic i (57:20)s fulfilled.

beadded.

i.e.

i.e.

'IndeedS U T T ~8 '

1Essentially

this argument was made in and the usual hypothesis

-I

42.3.2.
of

S (57:2:c) n T
then that for

mean that

to
ft\"

We had to exclude him from the game by explicitly restricting

\".

the solutions from

4 The argument

the coincide i.e.

of the

first remark in restriction of the solutions

shows 56.9.4.,

of f is unnecessary.)))

thesegames and

638

GAMES GENERAL NON-ZERO-SUM

a A Proof: s observedat the end of 43.4.2.player is a dummy, if and set) only if he forms (as a one-element a constituent of the game. We must apply this to the fictitious player n + 1 in the zero-sum game T. That (n + 1) is a constituent,means obviously that

+ (57:21) (fl)+v((n + l))-v(Su(n l)) v

Now we have

for all

Sfi(l,

,n).

v((n + 1)) = v(Su(n + l)) =

-v(7),

+ -v(Su(n 1)) = -v(-S).

v(7)

Hence becomes (57:21)

i.e.
(57:22)
And

v(fl)

-v(-S),

v(S) + v(-5)= v(7).

F 58.Interpretationof the Characteristicunction 68.1. Analysis of the Definition

this is preciselythe condition (57:20).

have arrived at a formulation of the theory of the general function n-person game,and found that the conceptof the characteristic is just as fundamental in it as it was in the preceding of the zero-sum theory n-persongame. It is therefore appropriate to survey the meaning of this conceptonce more,putting its mathematical definition into an explicit
form

58.1. We

Consider ccordinglya general -persongameF, described the funca n by tions JC*(ri, The , r n ) (k = 1, , n) in the senseof 11.2.3. value function for a set S 7 = (1, v(S) of the characteristic , n) obtains by forming this quantity for the zero-sumn + 1-person gameT the zerosum extension F.1 Hence can express by means of the definitory of it we
formulae of

and adding someinterpretative remarks.

25.1.3.:

(58:1) v(S) = Max- Min-> K(7,7) = Min-> Max- K(T,7),

where we have:

{ is a vector with

the components T

,))

function. function,

1 We restrict ourselvesto the S 7 (1, The use of all , (1, is contrary to our present standpoint.

/-

of + 1),i.e. the extendedcharacteristic (Cf the

, n)

i.e. the restricted characteristic to

beginning

of 57.5. 1

.
))))

INTERPRETATION
17

539

is a vector

with

the components r -a rj
t?r-

^ 0,

T-a

J) ir- =
k in

1;

TS

is the aggregate the variables r*, of 1 variablesT*, in S; and finally

fc

of S;r~5 is the aggregate the

(58:2)
where

K(7,7)=
3C(r5, r~5)

OC(rV-*)W-<,
rV-a

(58:3)

=
fcinS

JC*(n,

, r w)

.2
a Loss

58.2.TheDesireto Make a Gain vs. That

to Inflict

58.2.1. K(

>

77

v ) is obviously the expectationalue of a play of the game

T for the coalition

S,if
8

the coalition S uses the mixed strategy

>
17

and the

opposingcoalition

S uses the mixed strategy Hence(58:1) defines the value of a play for the coalition S under the assumptionthat the v(S), value K( , ), while the coalition S wants to maximize the expectation t opposingcoalition S wants to minimize it, and they chooseheir respect\\

tive (mixed)strategies , accordingly. i Now this principle is certainly correctn the zero-sum n + 1-person 4 but we are with the general n-person game F really dealing game T, 1 I S denotes S. Sincewe are dealing with f , we should have formed Swhich is I S. (Cf. the beginning of 57.2.1.) owever, this is immaterial, becauseno H variable T I exists. (Cf.the end of 56.2.2.) 1 We use only the original 3C*, = the i , n, i.e. W n +i of (56:2)n 56.2.2.))
T? W+ A;

!,

(58:4)

OC

nr l(T,,

fcl

d doesnot occurhere. This is, of course, ue to the fact that S / = It must be remembered that formula (58:3)above is the first formula loc.cit. gives Thesecond formula of (25:2) 25.1.3.

(!,,
n).
of

(25:2)in

(58:5)

s OC(rV-)

s ~ S 3C*0-i.IS
fcin

r).
S S,

loc.cit., since we are for the (Note that we, must now definitely use it is in _US; hence n dealing with T. Cf.also footnote 1 above.) Since + 1is not in of (58:5) doescontain the JCM i of (58:4), However, (58:4) the sum guarantees, kin as it must, the identity of the right-hand sidesof (58:3)and 3 The observations of footnote 1 above apply again. 4 if we view S as really representing

IS

I.e.

S /

(58:5). 1.5 /
S.)))

540

GENERAL NON-ZERO-SUMGAMES

of P is merely a \"working hypothesis\"!And in T the desire the coalition S to harm its opponent,the coalition S,is by no meansobvious. Indeed, the natural wish of the coalition S should be not so much to decrease

its the expectation value K( , t? ) of the coalition S as to increase own value K'( , ). These two principles would be identical expectation > o if every decrease K( , i? ) were equivalent to an increasef K'( , ). of 1 This is of coursethe casewhen F is a zero-sumgame, but it need not at all be so for a general ame F. g I.e.in a generalgame F the advantage of one group of players need not be synonymouswith the disadvantageof the others. In such a game in moves or rather changes strategy may existwhich are advantageous to both groups. In other words, there may exist an opportunity for o in of genuine increases productivity, simultaneously all sectorsf society. 58.2.2. this is more than a merepossibility the situations to Indeed, which it refers constituteone of the major subjectswith which economic and socialheory must deal. Hencehe questionarises: oes ur approach D o t t Did we not losethis cooperative not disregardthis aspectaltogether? o side of socialelationshipsbecausef the great emphasiswhich we placed r on their opposite, side? antagonistic, We think that this is not so. It is difficult to presenta complete case, sincethe validity of a theory is ultima analysi only establishedby success in the applications and we have made no applicationsin our discussion thus far. We will suggesttherefore only the main points which seem to
i\\ i\\

1This is sobecause when T is zero-sum

then

(58:6) 7 K(7,7) + K'(T, ) This is clearby common sense a formal proof obtains in ; (58:7)
where
a

- 0.
this way

:Clearly

K'(7,T) - 5) rc'(r,'- )*r *iT-*

5 3 3C'(T, T- )
this

(58:8)
(Note that
parison

s
*in

-S

3C (Ti,
fc

.. . ,

Tn

).

is not

the

Jbin _LS

2} 3C*(n,
that

r) which

occursin (58:5)). ow coinN

of (58:2) ith (58:7) hows w s

that

is (58:6) equivalent to 5 + 3C'(r, T~3) a 0,

(58:9)
and

5 3C(r , r~s)
n

(58:5) (58:3), imply

(58:9)amounts to % JC*(n, - - - , r.)

- 0,

i.e. the

zero-sum condition for

T.)))

INTERPRETATION *
definite corroboration.

541

and support our procedure, then refer to the applicationswhich provide a

58.3. Discussion
this connection:
in a

The 58.3.1. following considerationsdeserve particular attention

in

not be directly profitable not necessarilyzero-sum)game,but it is the way to exert general(i.e. on pressure him. He may be induced by such threats to pay a compento adjust his strategy in a desiredway, etc. Hence is not a limine it sation, unreasonablethat this category of strategicpossibilitiesshould be taken in into account;and our procedure forming the characteristic function, as be the proper one to do just that. It must be analyzed above, might admitted, however, that this is not a justification of our procedure it the merely prepares ground for the realjustification which consistsof success
in

First:nflicting losses n the adversary may o I

further considerationpointing in the same directionis this. We have seen that in our theory all solutions correspond attainment to 1 of the maximum collective profit by the totality of all players. When this maximum is reached, further gain of one group of players must be any compensatedy an at leastequal loss of the others. True,there couldbe b overcompensation one group might obtain a gain by inflicting a greater loss on the others. However,we have assumed complete information for a all players,and a perfect interplay of threats,counterthreatsnd compensa2 will one tions among them. Hence may assumethat such possibilities be will actions be obviated effective only as threats,and that the corresponding always by negotiationsand compensations. y this we do not mean that B these threats are \" bluffs \" which are never \"called.\"Sincethere exists information for all players, there can never be any doubt. But complete than the other when an action is threatened which one party gains less by t one loses,hen there exists facto the possibilityof avoiding it by comipso 8 pensationsin a way which is advantageousto both sides. And when this happens it is again true that one side gainsexactly what the otherloses. If this argument is accepted generally valid, then our difficulties as

applications.
A Second:

:i.e.

in

Our entire attitude towards coalitions and compensations was basedon this, already the theory of the zero-sum games. 1We do not proposeto determine here the amount of the compensation the nature of the compromise. This is the task of the exacttheory which we possess already. It will be the main subject in each application. (Cf. the various interpretations in -63.) At this point we want only to show that actions which would leadto a lossfor above.))) the totality of all players, can beavoided by the mechanism described

disappear. Third: 58.3.2. It may be said that the argumentation of the two prenot remarksis too sketchy and that it does justify our theory in the ceding f exactorm in which we proposeto useit. Thisis true,but our very detailed 1Cf. the end of 56.7.1., particularly footnote 3 on p. 513. 1

i.e.

61.

542

GAMES GENERAL NON-ZERO-SUM

4

meets motivation of that theory, as given in 56.2.2.-57.1.the latterrequirein r those sections the light of the two prement. If the reader econsiders remarks,then he will seethat the detailedjustification in the desired ceding sense was their subject. Indeed,the possibility of the objectionnow of was under consideration our reasonfor making the discussion our theory 1 and so detailed, avoiding plausibleshortcuts. Fourth: In spite of all this, the readermay feel that we have overe and emphasizedthe roleof threats,compensations, tc., that this may be a of one-sidedness our approach which is likely to vitiate the results in applications. The best answerto this is, as repeatedlypointed out before, the examination of thoseapplications. We shall therefore consider definite applications which correspondto t familiar economic problems.Their study will disclosehat our theory leadsto results which are,up to a certain point, in satisfactoryagreement views on thesematters. This is the caseas with the usual common-sense long as the two following conditionsare fulfilled: First that the setup is simple enough to allow a purely verbal analysis, not making use of any that those factors which are inseparable mathematical apparatus. Second in from our theory, but often excluded the ordinary,verbal approach coalitions and compensations have not come essentiallyinto play. This situation will be found to existin the application of 61.2.2.-61.4. Indeed, that examplerovidesthe decisivecorroborationof our procedure. p Beyond this point, where the first condition is still satisfied, but not the second,we shall find discrepancies in the direction and to the just extent to which the difference in standpoint justifies it. This will be and 62.6. in particularly clear the applicationsof the as even the first conditionfails, because problemis no longer Finally, elementary, we gradually reach ground where the theoreticalprocedure necessarilytakesover the leading role from the ordinary, purely verbal

61.6.3. 61.5.2.,

2 one.

59.General onsiderations C 59.1. of Discussion the Program

games for small values of n. systematic discussionof all generaln.-person for It will appearthat we can carry this out in absolute completeness the F samen as for the zero-sumgames: orthe n ^ 3. The discussionfor the f greater alues,i.e.or n ^ 4, is necessarilyat leastas difficult as it was for v
1A possible would have beento define the characteristic function as in one to comeout then with a flat generalization of the theory for zero-sum games,

t proceedo the applicationsof our theory of the game. The best way of starting such applicationsis a generaln-person

We 69.1.1. can now

58.1. and

This gradual transfer of the emphasis from corroboration of the theory by the reliablecommon-senseesults in the simple case,to overriding any un theoretical approach r by the theory in the complicatedones, s, of course, uite characteristicin the formation of i q
scientific theories.)))

:d) (56:1 in *

56.12.

i.e. with

GENERAL CONSIDERATIONS kinds.

n

643

the zero-sum gameswhere we couldonly disposeof special cases f various o

less proposeto do considerably in the way of analyzing gameswith We can afford to be considerably briefer now than we were in discussingthe zero-sum games: he detaileddiscussionthere was T necessaryin order to reassureourselvesof the proprietyof our procedure, and of the general deas and methodicalprinciplesunderlying it. At the i stagewhich we have reached s now, the general etup of the theory appears to be justified, and we want only to gain assurance concerning one the out generalizing step carried in this chapter. Forthis purposea lessextensive analysisof applicationsshouldsuffice. Further,it will be possiblealready to connectthe generalgames with n g 3 with some typical economic roblems(bilateral monopoly, duopoly p versus monopoly, etc.)which allow judgment of the appropriatenessof our indicatedbefore. theory in the sense Moredetailedinvestigations of general ameswith n ^ 4 will beunderg taken in subsequentpublications. 69.1.2. systematicapplicationof our new theory is best introduced The by a generaldiscussion,similar to that of 31. It will not be necessary, i w however, to carry out the equivalent considerationsn detail; e must only analyze to what extentthe resultsobtainedtherecarry over to the present situation, or what modifications arerequired. a We need not discussagain the role strategic of equivalence, s expounded in 31.3., this subjecthas already been dealt with satisfactorily in since On 57.5.1. the other hand, we shall take up certainmatters originating reducedforms, inequalitieswhich hold for the charelsewherethan in 31.: -27. f acteristicunction, inessentialityand essentiality (cf. 27.1. 5.);urther, f and the absolute values |T|i,T|2 (cf. 45.3.), finally someremarksconcerning | of the theory of decomposition Chapter IX.
We

^ 4 this time.

Forms. 69.2. TheReduced

The Inequalities

conceptof strategicequivalence,as introduced in 57.5.1. forms for all characteristic can be used to define reduced functions, along

69.2.1. The

the lines of 27.1. function v(S) its general trategicallyequivalent Given a characteristic s in transformation is given by (57:18) by))

i.e. 57.5.1.,
kmS))

(59:1)
unrestricted,while they
n

v'GS) = vOS) +))

in This is precisely(27:2)

but 27.1.1.,the a?, - , <* arenow completely were subject loc.cit.to the condition (27:1):
J, ,

<x

= 0. Hence the

oj are now

independentparameters,)))

644
while they

GENERAL NON-ZERO-SUM GAMES

restrictive possibilitiesof normalization than we found in 27.1.4. Indeed, c t we desired loc. it.o obtain a particular v'(S) to be denoted by v(S) 1conditions(27:3) which fulfills the n

representedonly n 1 independent parameters formerly 1 (cf.27.L3.). It would be erroneousto assume, however, that this leads to more

= v((n)). v((l)) = v((2)) = a functions consideredt that time belongedto zeroYet, the characteristic sum games; hencewe had automatically

(59:2)

(59:3)

v((l,

n)) = 0.
n

In imposingthis as a normalizing requirement,we now have and S (59:2) (59:3). owe obtain (59:4) v(/)

conditions:

v((n)) + (59:5) v((l)) + a? = v((2)) T (59:5) expresses59:2). hese equations cor( (59:4) expresses59:3); ( a loc. it., nd it is easy to verify that they are to (27:1*), respond (27:2*) c solvedby preciselyone system of aj, , a:))

+&==
v((*))

= 0,

a.

(59:6)
So we can say:

a{ = -v((A)) +

v(J)))

(59: A)

we call it the reducedform of v(S). 59.2.2. Another possiblerequirementfor the n parameters a?, , aj| in requiringfor v'(S) to be denoted V(S) the n conditions)) consists by

We call a characteristic function v(S) reducedif and only if it function satisfies (59:2), :3). Then every characteristic (59 3 v(S) is in strategic equivalence with precisely one reduced and and vGS). This v(/S) is given by the formulae (59:1) (59:6),

(59:7)
constant-sum
f

= = *((!)) *2))
joint value of n terms in

= ?((n)) = 0.
similar

1Our present standpoint in this respectis games in 42.2.2.

to

that

which we took for the

aj

- -v((fc)) +
8

Proof:Denotethe
ft

and so (59:4)becomes (7) v

(59:5)by
n

0. Then (59:5)amounts
-f n/3

to

v((fc))

- 0,

i.e.))

This is preciselythe

definition

of 27.1.4.)))

GENERAL CONSIDERATIONS This means

(59:8) v((l)) + a\\ = v((2))+

i.e.
(59:9) So we can say:
(59:B)
if

$...

545

v((n)) +

= 0,

cf k

= -v((fc)).

We call a characteristic function v(S)zeroreducedif and only it satisfies (59:7). Then every characteristicunction v(/S) f is in strategic with precisely one zero-reduced equivalence v(/S). This v(S) is given by the formulae and and we call it the zero-reduced form of v(S).

(59:1) (59:9),

59.2.3. us considerthe reduced characteristicunction v(S). We Let f denote the joint value of the n terms in (59:2) by 7, i.e.
(59:10)
Hence
y

-7 = v((l)) = v((2)) =
= v((fc)) +
and so (59:6) gives
n

= v((n)).

(59:11)
v(/) = v(J) +
n

7=^

jv(/)- Jb-1

If we use the zero-reducedform v(S) of the same v(S) then we have

(59:11)
(59:12)

*-i

a , henceby (59:9)v(/) = v(7)

k

- % v((fc)), i.e.sing u
n

*-i

ny

= ?(/).

Returning to the reducedform v(/S), we seethat someequalitiesand all are inequalitiesof 27.2. still valid. can To begin with, (59:10) be stated as follows:

set 7 for every one-element S. v(S) i l c w w This coincides ith (27:5*) oc. it., hile (27:5**)d.fails, sincewe saw

(59:13)

in

in is now that 57.2.1. the equivalent of (25:3:b)25.3.1. missing,and this t from (27:5*)here. was requiredto derive (27:5**) in to , (n) Repeatedapplication of (57:2:c) 57.2.1.the sets (1), -ny ^ 0,i.e.: gives by (59:13)

(59:14)

7^0.

in w This coincidesith (27:6) 27.2. Considernext an arbitrary subset S of I. Let p be the number of its in S elements: = (*i, applicationof (57:2:c) 57.2.1.))) , k p). Repeated

646 to the

GENERAL NON-ZERO-SUMGAMES

sets(fci),

, (k p) gives by

(59:13) v(S) ^ -P7p elements.Owing to

in (57:2:b)57.2.1.

Apply

this to /S which has n and (59:3), have we

1 v(-S)^ -v(S)

hencethe preceding inequality now becomes v(S) (n p)y. Combiningthesetwo inequalitiesgives : s (59:15) py ^ v(/S) ^ (n p)y for every p-element etS. This coincidesith (27:7) 27.2. in w

in

and (59:13) as follows:

can ( v() = (i.e.57:2:a) 57.2.1.) also be formulated

formulated as follows

Forp = 0,1 we have = in the first relation of (59:15). This coincides in 27.2. v(7) = (i.e. 59:3))can also be with (27:7*) (

(59:16)

relation of (59:15). Forp = n we have = in the second = n 1is missing,for t This coincidesith (27:7**)oc. it., xcepthat p l c e w i the samereasonfor which the equivalent of (27:5**)d.is missing(cf.the

(59:17)

remark following our

(59:13)).

Therearetwo alternatives, based on (59:14): First case:7 = 0. Then (59:15) v(S) = for all S. This is gives with precisely the inessentialcasediscussedin 27.3.1., all the attributes there. Considering(59: the inessentialgamesareprecisely enumerated A), thosewhich areequivalent to the gamewith v(S) = 0, the gamewhich is perfectly \"vacuous.\" case: > 0. By a changeof unit we could make 7 = 1,with Second y A the consequences pointed out in 27.3.2. nd just as there,we refrain as from doing this immediately. Forthe same reasons pointed out there, the strategy of coalitionsis decisivein such a game. We call a gamein this
n

59.3.1. 27.3.1.

59.3.Various Topics Theseinequalitiescan now be treatedin the

sameway as in

caseessential. f or The criteria :B),27:C),27:D)of 27.4. inessentialityand essenti( ( (27

ality

areagain valid:In (27:B)

in our

*-i

b v((fc)) must be replacedy

inequality

in which (25:3:b) 25.3.1.,

1Note that

present application this was used in 27.2.)))

replacesthis

missing equality

GENERAL CONSIDERATIONS

547

:C),

v(7),))

w essentialcase, ith the normalization 7 = 1 to the present situation. 59.3.2. can now pass to the considerationswhich correspondto We those of 31. o Theremarksof 31.1.1.-31.1.3. the structure of the concept f concerning domination and certainly necessaryand certainly unnecessary setscan be without repeated

while (27 i (27:D) are completely unaffected. Indeed,t is easy to that the proofs given therecarry over to the presentcase,heir bases t verify in beingprovided t of We leave it to the readero apply the considerations 27.5. for the

59.2.1.

of any change. The concepts convexity and of flatness The conclusions of arealso can be introducedas in in and (31 in for (31:E:b) as well as unaffected, except :G) Thesearethe only oneswhere (25:3:b) of (31H)id.for p = n :

is just as dubiousas thoseincluded what we said above, the value p = n the p for which the necessityof S is in doubt, are in to the interval to restricted p 7* n,

31.1.4. 31.1.4.-31.1.5. 31.1.4. 31.1.5., - 1. 25.3.1. is used. (cf.57.2.1.) must Finally, the remark at the end of 31.1.5. be modified. Owing to
1 c loc. it. I.e. (31:8) 0,1, i.e. 2gp gn 1. (59:18)

Thus this interval beginsto play a rolewhen n ^ 3, not only, as loc. it., c
when n

T who consults that secConsidernext the results of 31.2. he reader

will

4.1

tion

have no difficulty

in verifying the following:

(31:1),:J),(31K) (31 :

are unaffected. In (31 the constructionof ft with the help of a can :L) t out becarried without any change;he first assertion,ft H a , cannot be
maintained, sinceit uses that part of
which (31H) in 31.1.5. :

is no longer n valid; the secondassertion, ot a H ft , is unaffected. This weakening it of (31 removes (31 :M). (31:N) remainstrue because uses the intact :L) of (31 only. (310), 31 areunaffected. : ( :P) :L) part let To o 59.3.3.conclude, us considersomeof the concepts f Chapter IX. the We defined therethe two numbers|F|i, the former in 45.1., latter |T|i and in 45.2.3., we discussedtheir propertiesin 45.3. 45.2. Both definitions i.e.the pertinent considerations of 45.1., There are, however, essentialchangesin 45.3.: In carry over literally. of the proof is valid, but not the first part, (45:F) only the secondpart in sincethat and that alone makes use of (25:3:b)25.3.1. 57.2.1.). (cf.
n

-f 1-person games, which was prominent

1This is in agreement

with

the connection of general n-person games and zero-sum throughout 56.2-56.12.)))

548

GAMES GENERAL NON-ZERO-SUM

W Specifically: e still have

(59:19)
and so we can evaluate

|r|s?L=_?|r|
t
If

b 2 |F| in terms of |F|i;ut we do not have

(59:20)
nor can weevaluate that

|r|i

(n-))

we i 2 |F|in termsof |F| at all. Indeed, shallseein 60.2.1.

(59:21)

> |F|! 0,

= |r|,

for occurs certaingames. In consequence this, the remarks of 45.3.3.-45.3.4. pointless. become of The same i holds for 45.3.1., ts result (45:E) fails in so far as it concerns i.e. |F|j. It is true for but this is merely a restatementof the definitions. above, (59:21) we seethat (45:E) must be Consideringthis, and (59:19),

|r|i

weakenedas follows: then (59:C) If T is inessential, If T is essential,hen t

0. |r|i= 0,|F|^ =0. |r|i> 0,|r|

2 2

obtainedin ChapterIX, when consideringzero-sumgames.

w o G 60.TheSolutions f All General ames ith 60.1. TheCasen - 1

The theory of compositionand decomposition,which is the main object in of ChapterIX, can be extended its essential arts to our presentset-up. p a The difference betweenthe behavior of |F|i nd |F| discussedabove,neces2 b sitatessomeminor changes, ut these are easily applied. Of coursethe and theory of excesses of solutionsin the setsE(eo)and F(e) (cf there)must to beextended the present case but this, too, entailsno realdifficulties. would lengthen our expositionbeyond A detailednalysisof this subject a the the limits that we setourselvesin Furthermore, interpretative value of the results would not differ materially from what was already

59.1.1.

^3

n ^ as announcedin games This casehas already beenconsidered f Considerirst n = (and, for in In particular, we pointedout in purposes,settled) practical we that in this (and only in this) case dealwith a pure maximum problem. to is nevertheless desirable verify that our generalheory producesin this t 2 result. We apply therefore the c (trivial) special asethe common-sense mathematical rigor. generalheory in complete t
with

60.1. We

to proceed the systematic discussionof all generaln-person

3,

1.

59.1.1.

12.2.

12.2.1.

It

1 (59:20) nd a

* Thisbrings us backto the fourth remark in 58.3.2.)))

the (59:19) express two

:F), parts of (45 respectively.

SOLUTIONS OR n F
A

S3

549

This general ame F with n = 1is necessarilyinessential: is clear g by the function v(/S) of its reducedform, since considering characteristic = 0,i.e. = 0. then (59:16) (59:17)59.2.3. (for p = and in 7 give We may alsouse without reducing any one criterion(27:B), :C), (27 c i (27:D) of 27.4.cf.59.3.1.). (27:C) loc. it.s clearlysatisfied,with ( E.g. in ai = v((l)). Note that this is v(7), i.e. y (56:13) 56,9.1. the (with b

l=n)-7

notation of

Max 12.2.1.) T3C(r). We restatethis:

SinceT is inessential, can apply (31:0) (31 in 31.2.3. 59.3.2.). we or (cf. :P) This gives: T possesses set (60: A) preciselyone solution, the one-element (a)
where))

(60:1)

ai = v((l)) = v(7) = MaxT3C(r).

the ai of (60:1). \" This is obviously the \" common-sense result of

with

as 12.2.1. it shouldbe.

60.2. TheCasen - 2 60.2.1. Considernextn = 2. Themain fact is that a general amewith g = 2 neednot be inessential thus differing from the zero-sumgames n with n = 2. (Thelatterareinessentialby the first remark in 27.5.2.) : IndeedThecharacteristic form is completely function v(S)of its reduced determinedby (59:16) (59:17) 59.2.3. t is and in I (60:2)
v(S) =
<

I -7 I

when S has

desired.
1

the possibilitiesin (60:2). Thus 7 > 0, i.e. ssentiality, is among the possibilities,as asserted. e In the caseof essentiality we may further normalize 7 = 1,thereby comThus there exists only one type of essential pletely determining (60:2). generaltwo-persongames. = 2?may thus be > 0,thereis always (for n = 2) Notethat while = 0. It sufficesto prove this for the reduced f form, i.e.or (60:2). we and Indeed: Recallingthe definitions of 45.2.1. 45.2.3. see that a a* {{ai, j}) is detachedwhen a\\ a ^ 7, a\\ + 2 ^ 0,and that the a e minimum of the corresponding = a\\ + a2 is O.1 Hence|F|j= as

fulfills the conditions Now one verifies immediatelythat a v(S) of (60:2) function of a that it is the characteristic of (57:2:a), (57:2:c) if and suitable T (cf.57.3.4.), only if 7 ^ 0. This is preciselythe condition are in in t So (59:14) 59.2.3. we see:he 7 ^ of (59:14)59.2.3. precisely

i.e. 57.2.1.,

|r|i

|r|i

It is assumed e.g.or ai f

a*

0.)))

550

GENERAL NON-ZERO-SUMGAMES

a Summingup:Forn = 2 a zero-sumgamemust be inessential, general = gameneed not be. Accordingly the former must have |F|i 0;the latter > too. But both have always |F| = 0. 2 may have |F|i t We leave it to the readero interpretthis result in the light of previous a discussions, nd particularly of 45.3.4. The 60.2.2. solutionsfor a generalgame F with n = 2 areeasily deter-

mined.

in but

simply defined by the property that no imputation can be outside of it. t I.e.hereexists t preciselyone solution:he set of all imputations. as The generalmputation is given in this case a = (ai,a2 J },subject i in 57.5.1., now become:)) which to the conditions(57:15), (57:16)
{

we sincen = 2,theseexhaust all subsets. Hence may determine the solutions of T as if domination never held. Consequentlya solution is

a all 59.3.2.)setsS / with 0,1or n elements recertainly unnecessary

By the valid part of

(cf. (31H) in 31.1.5. the :

pertinent observations

(60:3)
(60:4)
We

a!

ai +

= v((l,2)) = v(/).

only if

the restate result: F possesses preciselyonesolution,the set of all imputations. (60:B) Thesearethe)) a = {{ai,2 |} a a with the ai, 2 of (60:3), (60:4). a a Note that (60:3), (60:4)determine unique pair i, 2 (i.e. ) if and

(60:5)

v((l)) + v((2)) = v((l,2)).

precisely the inessentiality of F. By the criteriaof 27.4.this expresses This result is, s it shouldbe,in harmony with (31 in 31.2.3. 59.3.2.). a (cf. :P) Otherwise

(60:6)
and thereexistnfinitely i
ity for

v((l)) + v((2))< v((l,2)),

many i,
2,

of The interpretation theseresults will be given in

F.

i.e.a .

This is the case essentialof

61.2.-61.4.

TheCasen = 3 60.3. 60.3.1. Considerfinally n = 3. These games include the essential zero-sumthree-person > 2 gamefor which |F|i and |F| > (cf. 45.3.3.).

Forn = 3 a zero-sumgameas well as a generalgamemay be essential, and both |F|i and |F| > ate possibilities.))) > 2

: So we see

SOLUTIONSFOR n

3
F

551
or in (31:O) (31:P)

31.2.3. 59.3.2.). assumetherefore that We (cf.

can

The casewhere T is inessentialis taken careof by

Use the reducedform of F in the normalization 7 = 1. Then we its describe characteristicunction v(S) with the help of (59:16) f and in as (59:17) 59.2.3. follows:))

is essential.

(60:7)
and

v(S) =

-1

when

S has v((l,2)) = a 3 S has

(60:8) v((2,3))= en,

And it

v((l,3)) = a 2,

when

2 elements.

is verified immediately that a v(*S) of (60:7),(60:8) fulfills the conditions (57:2:a), of i.e. function (57:2:c) 57.2.1.,that it is the characteristic of a suitable F (cf.57.3.4.), only if if and

(60:9)

-2 g ai, a , a ^ 1.
2 3

Note that this F can be chosen zero-sum,i.e.hat t holds if and only if

of (25:3:b)25.3.1.

(60:10)
our

ai

= a 2 = <i8 =

1.

In other words: The domain (60:9)represents ll generalgames, hile its a w the represents (unique) zero-sum game of upper boundary point (60:10)

60.3.2. us now Let

case.

g three-person ame. a2, 8 }), The generalimputation is given in this caseas a = which in (57:16) 57.5.1., now become: subjectto the conditions(57:15), ^ S i (60:11)

determine tffie solutions of this (essential) eneral g

{{i,

for Theseconditionsare preciselythose of 32.1.1. i, a2, s (cf. (32:2), t (32:3)there),i.e.hose used in the theory of the essentialzero-sum threeperson game. They agree also, apart from the factor 1+ -$'with the o w for conditionsof 47.2.2. a , a2, a3 (cf. (47:2*), 47:3*)here),i.e.ith those t ( used in the theory of the essentialzero-sum three-personame with excess. g in d Consequentlywe can use the graphicalrepresentation escribed 32.1.2., in particular in Figure52. We obtain the domain of the a as the funda53. It is also similar to that in 47.2.2. in Figure mental triangle in 32.1.2. in Figure70.
1

(60:12)

-1, + a +-1,0. = ai
2 3

-1,

the express relationshipof domination in this graphicalrepresentafor the tion. Concerning setS of 30.1.1.a domination a H ft , the follow-)))
We
> >

552

GENERAL NON-ZERO-SUMGAMES

(cf. ing can besaid. By the valid part of (31 in 31.1.5. the pertinent :H) observationsin 59.3.2.) setsS / with 0, 1or n elements certainly all are our unnecessary but sincen = 3, this restricts analysis to two-element

setsS.

Put therefore S =
on

(t,

j).1

Thendomination means that

k

+ ai ^ the By (60:12) first

We

v((t, j))

=a

and

a, > ft,
ctk

a, > ft.
ak

restate : Domination this

condition may be written a H

means that

i the against Figures71,84, 87 (which all describe sameoperation,n the successive ases C (IV),(V), (VI))is only this: The six lines))

a S #3; and > fa, 2 > ftz and 2 ^ ai > 0i, a3 > j;2 and ai. on ^ 2 > ft, 3 > 03 Thecircumstances can describedin (60:13) now beadded to the picture of the fundamental triangle. Thesimilarity is now more with 47. than with 32. The operationcorrespondsto the transition from Figure70 to Figures the 71,72, or to Figures84, 85,or to Figures87,88. Indeed, difference as
f

either

(60:13) j or 1or

ft*

(60:14)
which form

the configuration there,arenow replacedy the six lines b

(60:15))) the respectively. Hence secondtriangle (formed by the three last lines) which appears in the fundamental triangle(formed by the three first lines) t neednot be placed symmetrically with respecto the latter,as it is in the
11, k a permutation of 1This is quite similar to (47:5) in

threefigures mentioned.

j,

1,2,3.
except 47.2.3., that
we have

there 1

r- in place
60

on

of all three 01,az, at. Thereis alsothe change of scaleby the factor -f -r referred to o after (60:12). cf. The relation to (32:4)in is the sameas for (47:5)in 47.2.3., footnote 2

(60:11),

p. 406.)))

32.1.3.

SOLUTIONSFOR n
the

553

It 60.3.3.is convenient to distinguishtwo cases, ccordingto whether a (60:16)

ai ^

-di,

a2 ^

-a ,
2

c* 8

S -as

sidesof the threelast lines of (60:15) (wherethe threedomination relations

i o are of (60:13) valid) intersectn a common area, r not. Owingto (60:12) the former meansthat ai + a 2 + a 8 > 0, (60:17:a) while the latter means that ai + a 2 + a3 ^ 0. (60:17:b) We call thesecases and (b), respectively. (a) t Case(a): We have the conditionsof Figures71,72, excepthat the to inner triangle need not be placedsymmetrically with respect the fundamental triangle, as it is there. If this is borne in mind, then the discussion can of Case (IV),as given in 47.4.-47.S. be repeatedliterally. The solutions aretherefore, with the same qualification, those depictedin Figures 83. 82, sides of the inner then the corresponding We note that if an a< = and and the fundamental triangle coincide(cf. (60:15)), the corresponding 1 curve disappears. Case(b): We have essentially the conditionsof Figures84, 85 of which those of Figures 88 arebut a variant with the same proviso for 87, as in Case above. (a) asymmetry We redraw the arrangement of Figure84, the fundamental triangle being marked by / and the inner triangle by \\: Figure92. The arrangethe ment has several variants, because inner triangle can stick out from 2 the fundamental triangle in various ways. Figures92-95epict these d

1,

variants.8 -4

of If thesecircumstances rebornein mind, then the discussion Case a (V) 5 The solutionsare c as given in 47.6. an be repeated therefore, literally. 1Thus in the zero-sum case,where ai =03= a none of these curves occur 1,

accordwith the result of 32. 2 By (60:9) 2 ^ a, ^ This means, as the readermay easily verify for himself, that eachsideof the \" inner\" triangle must passbetween the corresponding sideof the exhaust all possibilities fundamental triangle and its oppositevertex. Our Figures 92-95
in

1.

within

* The only ones which can occur in a zero-sum game, for ai those which can be symmetric: Figures 92, 95. Of these, Figure

this restriction.

i.e.

Figure 84, and Figure 95 correspondsto Figure 87. 4 The Figures 92-95differ from each other by the successive isappearance the of d o Besides ne or more of the areas and , , may degenerateto a areas , , It is sometimes not quite easyto distinguish between linear interval or even to a point. A rule which allows mentioned above, and this \"degeneration.\" the \" disappearance\" differentiation between the four cases corresponding to the Figures 92-95and in which this difficulty doesnot present itself, is this: Figures 92-95correspondrespectively to the caseswhere the \"inner\" triangle meets 0, 2, 3 sidesof the fundamental triangle. (Meeting a vertex counts as meeting both sidesto which it belongs.) * The discussion of Case(VI) in 47.7, may alsobe consideredas such a repetition under much simpler conditions.)))

a* at 1,are 92 correspondsto

1,

554))
with

GAMES)) GENERAL NON-ZERO-SUM

the necessaryqualifications of asymmetryand the possibledisappearo and anceor degenerationf some areas (cf. Figures92-95 footnote in 4 on p.553),those depicted Figure86.))

Figure

92.))

Figure

93.))

Figure

94.))

Figure

95.))

60.4.Comparison with the Zero-sum Games 60.4.1. have determinedall solutionsof the generaln-person We games with n = 3 in a rigorousway, but we have not yet made an attempt to analyze the meaning of our results. We therefore pass now to this analysis. Let us begin with someremarks of a rather formal nature. We have i seenthat the smallestn for which a general ame can be essentials n = 2, g while for the zero-sumgames the corresponding number was n = 3. We have also seen that there exists (assuming reductionand normalization 7 = 1) precisely one essential generalgame for n = 2, whereas for the zero-sumgames the same thing was true for n = 3. Again the essential generalgames for n = 3 (under the sameassumptions as above) form a three parameter manifold, while for the zero-sumgames this was true for n = 4. All this indicates analogy between general -persongames and an n zero-sumn + 1-person T games. Of course,we know the reason: he zeroo sum extensions f the generaln-persongames are zero-sumn + 1-person)))

ECONOMIC INTERPRETATION FOR n =

1,2

555

games and we saw that every zero-sum n + 1-personame can be obtained g in this way. 1 It 60.4.2. must be remembered,however, that while the zero-sum n + 1-personames are exhausted by this procedure,their solutions are g not the solutionsof a general -persongame form only a subset of those n of its zero-sumextension e.g.56:I:a) 56.12.). in (cf. ( Thus our determinationof all solutionsof \"all general three-personames g means only that we know some, ut not all solutionsof all zero-sum fourb t person games. Indeed,he voluminous and yet incompletediscussionof Chapter VIIshowsthat determiningall solutionsof all zero-sum four-person games is a task of considerablygreatersize. Our results concerning the games imply, however, this much:Thereexistsolugeneralthree-person tions for every zero-sumfour-persongame. (The casuisticdiscussionof ChapterVII did not reveal this.)

61. conomicInterpretationof the Results for n = 1,2 E 61.1. n - 1 TheCase We to 61.1. now come the main objectiveof our presentanalysis:The of our results for n = 1,2,3. interpretation Considerfirst n = 1: What matters in this casewas already stated or

O referredto in 60.1. ur result was,as it had to be,a repetitionof the simple t maximum principle which characterizeshis case and this caseonly, the o which therefore describes \" Robinson Crusoe\" r completelyplanned communistic economy.
M 61.2. TheCasen = 2. The Two-person arket

Thereexists preciselyone solution. It consistsof all those imputations where each player gets individually at leastthat amount which he can f secureor himself, while the two gettogether reciselythe maximum amount p which they can secure together. Herethe \" amount which a player can get for himself\" must be understood to be the amount which he can get for himself, irrespectiveof what e his opponentdoes, ven assumingthat his opponentis guidedby the desire 2 to inflict a loss rather than to achieve a gain. In examining the solution we find the opportunity to fulfill the promise We contained in the fourth remark in 58.3.2. must seewhether our above definition of the \" amount which a player can get for himself\" based on a of hypothetical desire the opponent to inflict a loss rather than to achieve

o 61.2.1. nextn = 2:Our result for this case, btainedin 60.2.2. Consider can be stated verbally as follows:

beginning
8

1Precisely It is strategically of 57.4.1.)

equivalent

to one which is so obtainable. (Cf

. the
the

player k

and Cf.the detaileddiscussion at the end of 58.2.1.in cansecurefor himself is, of course, ((fc)).))) v

58.3.Theamount

which

556

GENERAL NON-ZERO-SUM GAMES

1 the r gain leads to common-senseesults. In order to compare result it is of our theory in this way with \"common-sense/' desirableto present to the generalwo-persongame in a form which is easilyaccessible ordinary t somefundamental intuition. Sucha form is readily found by considering b economic relationshipswhich can exist etweentwo persons. 61.2.2. Accordingly we considerthe situation of two personsin a market, a a seller nd a buyer. We wish to analyze one transaction only and it will appear that this is equivalent to the generaltwo-person game. It is economic obviously also equivalent to the simplest form of the classical problemof bilateral monopoly. the seller1 and the buyer 2. The The two participants are i of transactionwhich we considers the sale one unit A of a certaincommodthe value of the possession A to 1by u and for 2 of ity by 1to 2. Denote while v u representshe best alternative use of A for the seller, t by v. is the value to the buyer, after the sale. In order that such a transaction have any sense,the value of A for the that w buyer must exceed onefor the seller. I.e. e must have

1,2:

I.e.,

(61:1)

u<v.

result is that the pricep will have some value The common-sense betweenthe limits setby the alternative valuations of the two participants,
1Thereaderwill understand that we do not ascribethis desireto the opponent. It is only that our theory can be formulated as if he had this desire. What matters is not this possible formulation, but the results of the theory. Indeed,this \"malevolent\" behavior of the opponent determines only some, but not all features of the solution It gives the lower limit of what eachplayer must obtain but what both get together can only bedescribedy the opposite b individually, hypothesis of perfectcooperation. (Cf.above.) c This is just a special aseof the general fact, that only the entire, rigorous theory is a reliable guide under all conditions, while the verbal illustrations of its parts are of limited applicability and may conflict with eachother. All this can be brought out evenbetter by the detailed discussion of 58.3. 1 We are purposely disregarding the possibility of describing a saleas an exchangeof which we have stated repeatedly, to goodsfor goods. Our theory forcesus, for reasons in use an unrestrictedly transferable numerical utility, which we may as well describe terms of money. We shall deviate from this standpoint only in XII. 1We leaveit to the readerto formulate this in Chapter our original combinatorial terms of definition of games.)))

3 amounts u, O.

It is convenient to use the stateof the buyer when no saleoccurs i.e. his original financial position as the zero of his utility. 2 this Let us now describe as a game. In doing this it is best to omit A and to deal instead with the value connected from the picture altogether with its transfer or its alternative uses. We may then formulate the rules of the game as follows. or 1 offers 2 a \"price\", which 2 may \"accept\" \"decline.\"In the p In the second they get the case first case1, get the amounts p, v 2, p.

ECONOMIC INTERPRETATION FOR

1,2

557

e.that
51:2)
Vhere u

gp ^

v.

actually be between the limits of (61:2) dependson factors this description.Indeed,his rule of the game t rovides for one bid only, which must be accepted declined this is or learly the final bid of the transaction. It may have been precededy b egotiating,bargaining,higgling, contractingand recontracting,about rhich we said nothing. Consequentlya satisfactory theory of this highly the entireinterval (61:2) available for p. implified modelshoulcHeave

ot taken into accountin

will

F Market and Its Characteristic unction Two-person Before going any further we add two remarks concerning this of the game,which is our model for the economic under escription set-up

61.3.1.

61.3. Discussion the of

onsideration.

First: t would be possible to use more elaborate I models allowing for reater(but limited)numbers of alternative bids, etc. Thereis a prima facie evidence consideringuch variants, sinceall for s a r markets regovernedby more or lesselaborateules for successive xisting ids by all participants, hich appear be essential or the understanding w f to f their character. Besides, e did investigate in detail the game of w i 'okern 19. This gameis based on the interplay of the bids of all partici-

o c t and ants and we saw loc. it.hat the sequence arrangement f thesebids for its structureand theory. (Cf.in particular as of decisive importance in 11. 14. -19. t le descriptive 19.1. 3.,he variants discussed 19. -19. and part of 19.16.) leconcludingummary s A closer inspectionshows, however, that in our present setup these d etailsdo not become ecisive.The situation is altogetherdifferent from okerwhich is a zero-sumgameand where any loss of one player is a gain 1 t >r the other one. Specifically, he reader may discuss any more compliEtted market way as we shall (but with only two participants!) in the same He characteristic it for our simple version in 61.3.3. will find the same the mctionas we obtain (61 (61 of 61.3.3. Indeed, deductionsgiven :5), :6) The tiere apply mutatis mutandis in any market (of two participants!): out ^ader who carries this comparisonwill observethat all that matters 2 1 those proofs is that the seller(or the buyer) may, if he wishes,insist o bsolutely on the particular pricementioned there,irrespectivef the 8 b ounter-offers he may get and the number of successiveids required. lead Theseelaborations essentially to the sameresults as our simple

them. from considering refrain therefore 1 This in appliesdirectly to Poker as a two-persongame, as considered 19.If more tan two persons participate, then our treatment by means of coalitionsbrings about the ime situation. 2 The significant one is the proof of (61:5) 61.3.3. in 3 Returning to our previous remarks concerning Poker:.Thereader may verify for to imself how a corresponding simple overall policy would not work there-due

lodel.We

the)))

558 further.

GENERAL NON-ZERO-SUMGAMES

between (co-operating) t Indeed,he mechanism of compensations is players,which we assumedin all parts of our theories perfectly adequate bids of prices. I.e.it is not necessary to introduceoffering, to replace as acceptingor declining of prices part of the rules of the game. The is mechanismof compensations fully able to takecareof this, includingthe a bargaining,higgling, contractingnd recontracting. preliminary negotiating, as Sucha simplified game couldbe described follows:Both players 1,2
n o then to choose exchanger not. If eitheronechoosesot to exchange, then they get the u, 0. If both choseto exchange amounts w', u\" where u' u\" are two arbitrary but fixed quantities with
may

Second: 61.3.2. On the other hand our model could also be simplified

get 1,2 the amounts

t;. 1
\" price p = u'

the sum
\"

In other words:The rules of the game

(then
v

they

will

may provide for an arbitrary p u\,") which the players cannot influence nevertheless they desirey appropriate b bring about any otherprice

c is neitherthe Thus it appears that the arrangement hosen in 61.2.2. one. We areusingit because seemso be it t nor the most complete simplest t best suited to bring out the essentialraits of the situation without unnecessary details. \" \" amounts in the terminolresult The 61.3.3. common-sense of 61.2.2. of imputationsto this :Thereexists reciselyonesolution and this is the p ogy

compensations.

set of all imputations))

with

(61 :3) (61:4)

ai ^ u, 2 ^ 0, ai + a2 = v.
<*

Comparingthis with the application of our theory in o coincide with that agreement btains when (61:3), (61:4) there. This meansthat we must have))

we 60.2.2., see

(60:3), (60:4)

(61:5) (61:6)

v((l)) =u, v((2)) =0, v((l,2))

=.

we completeness do this
penalties which

It is easily verified that (61:5),61 areindeed true. Forthe sakeof ( :6)

for both arrangements of

61.2.2. 61.3.1., and

the rules of that game inflict upon any prohibitive, excessive, in any or of other simple way uniform scheme bidding. One could, of course,incorporate similar provisions into the rules governing a market. Indeed,there are certain traditional forms of transactions which are possibly of this type, such as options. But it doesnot seem advisable to include them in this first, elementary survey of the problem. 1Thecharacteristicfunctions of both arrangements (that of and the one above) and they will be found to be identical.))) will be determined in

61.3.3.

61.2.2.

ECONOMIC INTERPRETATION FOR n

the 61.3.2.,first being dealt with

- 1,2

559

in the text,and the variants requiredfor alternative in brackets [ the second Ad (61:5): Player 1 can make sure to obtain u by offering the price not p = u [by choosing to exchange].Player 2 can makesure that player 1 obtains u by decliningevery price [by choosingnot to exchange].Hence))

].

u by p = v [the same conduct of both players] yields in the same way that v((2)) = 0. Ad (616): he two playerstogether : T get eitheru or v the latterarising from p + (v v is p) [from u' + u\"}. By (61:1) preferable;hence

of Replacement p

v((l,2)) = v.

the u, 0,v as observedin may appear fairly trivial. Thereis, one significant point about It wasobtainedwith our definition however, of the characteristicunction to which the criticisms of 58.3. f and I.e.t is dependentupon eachplayerascribingto his opponent in i apply. a certainpart of the theory but not in all of it the desire inflict a loss to rather than to achieve a gain. It is important to realize that this dependences really significant, i that modification of this assumptionwould alterthe result, and therefore falsify it, sincethe result was seento be correct. Thisis best donewith the of arrangement 61.2.2. a Indeed, ssume that player 2 would under certain conditions prefer to make a profit for himself rather than to inflict a loss upon player Assume that theseconditionsexist, .g. hen player 1offers a certainprice e w and In this caseplayer 2 obtains v p if he accepts, PO > u but < v. he if he declines.Hence gains by accepting. On the other hand player a 1obtains if player 2 acceptsnd u if he declines.Hence layer 2 inflicts p our a loss (uponplayer 1) by declining. Consequently presentassumption the concerning intentions of player 2 meansthat he will accept. Thus under these conditionsplayer 1 can count upon obtaining the to amount p This conflicts with our previous result according which the a that interval (61:2) shouldbe permissible, nd we saw in 61.2.2. entireprice the natural one. it is the latterresult that must be considered of t Summing up:The discussion the general wo-persongame which we has shown that the generaltwo-persongame is carried in 61.2.-61.4. out function crucialwith regard to the decision whether the characteristic shouldbe formed as used in our theory. The setup was simpleenough as to allow a ''common-sense\" predictionof the result and any changein f the characteristicunction would have altered the procedure forming of In this way we have obtained by the the theoreticalesult significantly. r applicationof the theory, a corroborationin the senseof the fourth remark
with

61.4. coincidence the values of the characteristic The of function v(S)

61.4. Justification

of the Standpoint

of

68.

61.3.3. it:

61.2.

i.e.

1.

p<>

in 58.3.)))

560

GAMES GENERAL NON-ZERO-SUM

but it neverthelesssufficed for the task of corroboration which we had setfor ourselves. Besides, interpretingoneessential eneralwo-person g t by game,all were interpreted,sinceall of them are strategically equivalent to onereduced form (which couldbe normalized to y = 1). So far everything is satisfactory. But it is still desirableto verify that our theory can do equal justiceto somewhatlesstrivial economicetups. s Forthis purposewe will first extendthe descriptionof the two-personmarket somewhat. It will be seenthat this yields nothing really new. Then we shall turn to the general hree-personames. Therewe will find genuinely t g new corroborations opportunitiesfor more fundamental interpretations. and in the 61.5.2. us return to the situation described 61.2.2: seller1 Let and the buyer 2 in a market. We allow now for transactions involving and mutually substitutable) units AI, , At any or all of s (indivisible 1 of of a commodity. Denote value of the possession tf( = 0,1, the , s) of theseunits for 1by u t and for 2 by v t . Thus the quantities

referred 61.5.1. discussionof 61.2.-61.4. to a very \" The elementarycase \"

61.6. Divisible Goods. The \"Marginal

Pairs\"

(61:7)

UQ
VQ

(61:8)
describehe t

= 0, = 0,

t;i,---

MI,

,.,
, M.,

we 61.2.2.use for the buyer his original position as the zero of his utility. Thereis no need to repeatthe considerationsof 61.2.2., 61.3.2. 61.3.1., the concerning rules of the game which modelsthis setup.

variable utilities of these units to each participant. As in

It is easy to see,what its characteristic function must 2 that player can blockall sales,it followsas in 61.3.3. (61:9)

be. Sinceeach

v((l)) = 11., v((2)) =0. the Sincethe two players togethercan determine number of units to be with a transfer of t units they obtain together t + v t , transferredand since u,therefore

+ v((l,2)) = Max,..o,i. (*.-, v>). This v(S) is a characteristic function, henceit must fulfill the inequalities (57:2:a), of 57.2.1. the (57:2:c) Considering(61:9), (61:10), only one

(61:10)

....

which is not immediately obvious is

(61:11)

v((l,2))

v((l)).
v<>

This obtains, by observing that the left-hand side is ^ u 9 + (use (61:10) t = 0), and the right-hand sideis = u t .
1We could also allow for continuous divisibility, but this would make difference. 21 by offering an inacceptibly high price,2 by declining every price.)))

= u, by

no material

ECONOMIC INTERPRETATION FOR n say

Q

1,2))

561))

needonly be stated for the t ^ fa and we can stateit for the t ^ We may write theseinequalitiesas follows:

61.5.3. now the t for which the maximum in (61 is assumed, Consider :10) t = t . It is characterizedy w._t + ^ u,-. + for all t. This b o
v
to

vt

fa

separately.

for u.-t u._t ^ vt v u.-t u..t v t v t for to w Specialize (61:12)t = + 1 (except hen
9 t9

(61 :12) (61 :13)

fl

t t

> fa, <

fa.))

u,- - u.-*^> v t+l - vt , t i to (61:14) / and (61:13) t = t to 1 (exceptwhen to = in

\\

is vacuous):

fa

fa

s in which case(61:12)
* 7
Q

IB

-t-i

vacuous)

ti._i0^ i - i/._! < v t - vt-i. (61:15) + Note that (0:12), t (0:13) (without the specialization can be written as follows)) (6:14), (6:15))

which

case(61:13) is

00

fa

1 that

led to

(61 :16)))
(61:17)))

+>))

(, - ';-)))

> (,,))
for))

that is, that the utility the utility ably introduce assumptionof decreasing for as of each additional unit decreases, the total holding increases, both 1,2. As a formula)) participants

is In generalwe can say that (61:14), (61:15)necessary only, while (61:16), is necessaryand sufficient. However,we may now profit(61:17)

(61:18)
(61:19)))
This implies
r))

U\\
t'l))

>
>))

Mi
V\\

>
>))

> V,

?'-!))

for
V))

>))

(61:20)))

W.-,)

(/O))

for I

<))

encenow

Consequently(61:14),))) (01:17). (61:15) (61:14), imply (61:16),

562

GENERAL NON-ZERO-SUM GAMES

too (61:15) (61:15) arenecessaryand sufficient. Combining(61:14), with we part of (61:18), (61:19)may also write:

Eachoneof

(61:21)
units actually

1 '-'*~ \"\"/r' 7 \"'- , is greaterthan eachone

U
1 Vt
>

of))

2 definition of the \" marginal pairs.\" So we see:

ideas,the maximizing t = to is the number of transferred. We have shown that it is characterized by and w is preciselyBohm-Bawerk's (61:21), the reader ill verify that (61:21)
According to the usual

(61 :A)

to of units transis determined in accord criterion with Bohm-Bawerk's ferred,

t The size of the transaction, i.e.he number

\"

of the

marginal pairs.\"

To this extentwe may say that the ordinary common-senseesult has r our theory. beenreproducedy b It may be noted, to conclude, the case this game is inessential that when has a simplemeaning. Inessentialitymeans here))

this i.e.y (61:9) (61:10) means b Considering(61:9), equality in (61:11). = =

that the maximum we see:
in

the

latteris assumedat

t 0,i.e.hat

/o

0.

So

: (61B)
in

it

Our game is inessentialif and only 3 i e.when t = O.

Q

if

no transferstake place

m we In order to provide an interpretationin this respect must consider ore a the (unique)solution of our game, s providedby the considerations closely of 60.2.2.

61.6.1. us now pass to the determinationof the price in this set-up. Let

61.6. ThePrice.

Discussion

Mathematicallythe present set-up is no more generalthan the earlier : both one analyzed in 61.2.-61.4. representessentialgeneraltwo-person a only games, nd we know that there exists one such game. Nevertheless, to that set-upwas only a special caseof our present one: orresponding C 8 = 1. This difference will be felt as we now pass to the interpretation.
line is of the first line with the second term of the second second first is (61 and :15). omparing first and first is an C from (61:18); and second, ne from (61:19). second o inequality 1E.von Bohm-Bawerk: Positive Theorie Kapitals, 4th Edit. Jena 1921, des p. 266ff. 8 Note that in our earlier we forced the occurrence a transfer of arrangement of 61.2.2. Our both possibilities open.))) by requiring (61:1). present set-upleaves

(61:14); comparing

1Comparing

the first term

similarly

ECONOMIC INTERPRETATION FOR n =

1,2

563

with in Comparisonof (61:5), in (61.3.3.) (61:9), (61:6) (61:10)61.5.2. shows that the mathematical identity of these two setups rests upon t substituting the u, v of the former accordingo t (61:22) u = u. v = Max,_,i (u,-. + v t ). The (unique)solution consists,therefore, of all imputations
y

. .....
2

a =
fulfilling

In (61:4) (61:3), in 61.3.3. terms of ^

2

a {{!,}}
^
v

(61:23)

- u.

2 this

means

Let us now formulate this in terms of the ordinary concept f prices o instead of the imputationswhich arethe meansof expression our theory.2 of t units Since,as we concludedin 61.5.3., will have been transferred to the must be buyer 2,there
Q

(61:24)
if

t9

Up

2,

the price paid was p per unit. that p,

Consequently(61:23) means,in terms of

(61:25)
<o

T(U.-u.-) ^ p ^}\\* to to

This can also be written as

(61:26)

V f ^-/ (u.^.1
to

61.6.2. the limits in (6126)arenot at all those which the BohmNow : Bawerk theory provides. According to that theory, the pricemust lie of 61.5.3., betweenthe utilities of the two marginal pairs named in (61:21)
i i.e.n the interval

-l

w,_ -) ^ p g 1 V fe - ^0.
t
^o

'o

J-l

^W

K^-^oUp^(*<-<o--<o - ^-i)) This can also be written as M (61:28) ax (w,-e - v ,+i - ^ p

(61:27)
+i w._
vt v

f|)

^
this In orderto compare interval a further interval
with

Min (ti,_

<o

tt._ o_i,

v <o

t;

<o

_i)

it is (61:26), convenient* to form

w

(61:29)
suited

u8

to+ i

tt._i

^ p^

-i.

1We could baseour discussion equally well on on, but the present procedureis better to be repeatedin the caseof the three-personmarket. 1It may be worth re-emphasizing: This is interpretation, and not the theory itself! 3 Note that by (61:22) w,, v = w< + u t

t><

.)))

564

GAMES GENERAL NON-ZERO-SUM

in The two last inequalitiesof (61:20) 61.5.3. t = 0) yield that the (with and lower limit of (61 is ^ that of (61 :26), that the upper limit of (6129) :29) : the is ^ that of (6126).Hence interval (61 is containedin the interval :29) : ( : : (6126).Again, (6129)obviously containsthe interval (6127),i.e.6128). : : The intervals (61:26), (61:29), contain eachother,in (61:28) Summing up:

this order.

So we see:
while

(61C) :

(61:28).

The price p per unit is limited to the interval (61:26) only, it Bohm-Bawerk's theory restricts to the narrower interval

The two results (61 and (61 give a preciseicture of the :C) :A) p relation of our theory, in the present application,to the ordinary common sense standpoint.1 They show that there is complete greementcona what will happen in fact i.e.he number of units transferred but t cerning a divergenceas to the conditionsunder which it will take place the o priceper unit. Specifically, ur theory provideda wider interval for that than price the ordinary viewpoint. at That the divergenceshould come this point and in this direction is readily understandable. Our theory is essentiallydependent upon assumm among ing (among other things) a complete echanism of compensations the players. This amounts to possiblepayments of varying premiumsor i rebatesn connection with the various units transferred. Now the narrow interval of the ordinary standpoint (defined by Bohm-Bawerk's price of is \"marginal pairs\") notoriously dependentupon the existence a unique valid for all transfers which occur. Sincewe are actually price equally a allowing premiums and rebates, s indicated above, the unique priceis obliterated.Our pricep per unit is merely an average price indeed it and it is therefore quite natural in was defined as such by (61:24) that we obtained a wider interval than the one defined by \" marginal

61.6.3.

i.e.

61.6.1.

pairs.\" we Toconclude, observethat suchabnormalitiesin the formation of the w structureare also quite in agreement ith the fact that the market price

one. under considerationis a bilaterallymonopolistic

Case 62.EconomicInterpretationof the Resultsfor n = 3 : Special Market TheCasen 3, SpecialCase. The Three-person 62.1.
We proposeto obtain an interpretation This as samesense was outlined in 61.2.1. will be done by extending the model of 61.2.2. dealing with two persons in a market to one dealing n in the
with

Consider 62.1.1. finally

= 3.

three persons.

the views

as

1 We took Bohm-Bawerk's of most other writers

treatment as representative for that standpoint. on this subjectsinceCarl Menger are essentially

the

Indeed,
same

his.)))

ECONOMICINTERPRETATION FOR n

- 3:SPECIAL

565

As we have pointed out before,the first mentioned discussion couldnot fail to be exhaustive,sincethereexists nly one essentialgeneralwo-person o t game. On the other hand we know that the essentialgeneral three-person in family and their detailed discussion games form a 3 parameter 1 forced us to distinguish numerous alternatives. Accordingly several

60.3.2.

models would be required to accountfor all possibilitiesof the essential o g generalthree-person ame. We shall restrict urselvesto the discussion

would be somewhat lengthy of one typical class. An exhaustive discussion and would not contributeproportionally to our understandingof the theory but it would not present any additionaldifficulties. in We consider accordingly the situation of three-persons a a a market, one seller nd two buyers. The discussionof two sellersnd one buyer would lead to the same mathematical setup and to corresponding conclusions.For the sakeof definitenesswe discussthe first form of the to problemand leave it to the reader carry out the parallel discussionof form. the second Thethree participants are 1,2,3the seller the (prospective)buyers a 2,3. We shall consider successivelythe special rrangement of 61.2.2. In contrast to what we found there,the and the more general neof o latterwill now provide a real generalization of the former. The transaction which we conLet us begin with the setup of unit A of a certain commodity by 1to sider is the saleof one (indivisible) of the Denote value of the possession A for 1 by u, for 2 by v, either2 or and for 3 by w. that thesetransactionsshouldmake sensefor all participants, In order that the value of A for eachbuyer must exceed for the seller. Also, unless the two buyers 2,3 happen to be in exactly equal positions,one of them able to derive a greaterutility from than the other must be stronger of the the possession A. We may assumethat in this case strongerbuyer is Theseassumptionsmean that we have

62.1.2.

1,

61.5.2. 61.2.2.:

3.

i.e.

3.

u < v ^ w. (62:1) we use As in 61.2.2. 61.5.2. for each and buyer his original positionas the zero As in 61.5., there is no need to repeatthe considerationsof 61.2.2., the 61.3. concerning rules of the gamewhich modelsthis setup.

of his utility.

f Sinceeach its characteristicunction must be: a can block sales him, and the seller s well as both buyers together to buyer that it follows as in can block all sales (cf.

It is easy to seewhat

61.5.2.),

61.3.3.

(62:2)

(62:3)
(62:4)

v((2)) = v((3)) = 0, v((l)) = u, = v, v((l,3)) = w, v((2,3))= 0, v((l,2)) = w* v((l,2,3))

being subdivided into the four

1The two main cases(a) and (b), the latter represented the Figures 92-95. by 1Ofcourse makes use of w < v 5* this
w.)))

subcases

566

GENERAL NON-ZERO-SUMGAMES

it This v(S)is a characteristic function, hence must fulfill the inequalities in The out of (57:2:a), (57:2:c) 57.2.1. verification can be carried with little trouble, and is left to the reader. By the nature of things the gameto which v(S) belongsis not constant sum,1 henceit is a fortiori essential.

62.2. Preliminary
We 62.2.

Discussion

can now apply the results obtained in 60.3. the concerning essential eneralthree-person ame,to obtain all solutionsfor our present g g the problem.We shall again compare mathematical result with what the applicationof ordinary common sensemethodsgives. w The agreement ill turn out to bebetterthan in 61.5.2.-61.6.3. up to a the certainpoint specifically limits to be derived for the pricewill be the to same with both methods. This is probably ascribable the fact that we When we pass to * aredealing now with oneunit only, just as in 61.2.2. will reappear. the complications 61.5.2.-61.6.3. of units, in 63.1.-63.6., Beyond the point referred to, however, there will be a qualitative discrepancybetween our theory and the ordinary view point. It will be
T seenthat this is due to the possibilityof forming coalitions. his possibility a becomes reality for the first time for three participants, and it must be that expected our theory will do it full justice while the ordinaryapproach o it. Thus the divergence f the two procedures ill also w usually neglects onefrom the point of view of our theory. turn out to be a legitimate
t 60.3.2. We 62.3.1. proceedo the application of 60.3.1., to the v(S) of The 62.3. Solutions :First

Subcase

above. (62:2)-(62:4)

The imputations in this setuparethe))

with

at 0, a, 0, ai ^ w, (62:5) a2 + as = w. ai + (62:6) it In orderto apply 60.3.1., 60.3.2.,is necessary to bring this to its
reducedform, and then to normalize y =
the

to of The first operationcorresponds the replacement our

1.

aj,aj,a

of

ai,aj,

by

(62:7)
aj obtain as

ai = a* + a
in

in and and ismentionedin 57.5.1. discussed 31.3.2. in 42.4.2. he a?, aj, T in described the discussionwhich leadsto (59: A) 1Proof:(57:20)n 57.5.2. i is violated, e.g.y b

59.2.1.

v((2,3

<w

v()))

ECONOMIC INTERPRETATION FOR n Specifically,

- 3:SPECIAL
o,

567

' , (62:8)a{ =
AO Q\\

-w

+ 2u
g

>

a, = a2

-w

u
*

-w

o Thecorresponding in changes n v(S)aregiven by (59:1) 59.2.1. carry ; they into (62:2)-(62:4) (62:9)))

v (62:10) '((l,2)) =

-*-* ' i, '

3
v v

(( 3)))) \\\\*t**JJ

>

v'((2,3))

(62:11)
Thus 7 =
o
>

= v'((l,2,3)) 0.

- - ?l^p),

and so the secondoperation consistsof dividing everything by this quantity. Insteadof doingthis, we prefer to apply 60.3.1., 60.3.2. directly, inserting everywhere (where 7 = 1 was assumed) the proportionalityfactor o -1 shows Comparisonwith (60:8)in 60.3.1. that
fll

2(w 1

u) ,

az

ag

3v

2w _

The six lines of

**1

now:)) we derived our solutions,become

in which the (60:15) 60.3.2., describe triangle from

w
u)

which

(62:12)
v
;))

a,i

2(w

')

, a2

w w

o
3)

u)
j

a
At')

n?

2) ')

3)

3
3t>

u))
;)

a,2

u)

,) 3)

2w
3)))

u)

62.3.2. We
Clearly

can now discussthis configuration in the senseof

i

60.3.3.

hencewe have

which one of its four subcases,represented Figremains to decide by 1The procedure analogous to that used in the discussion of the essential ero-sum z is and in ( three-persongame with excess 47.,in particular in 47.2.2. 47.3.2.Case(III)), 47.4.2. certain phaseof Case(IV)). (a * The 1 in (60:15) loc.cit. stands for 7, so we must multiply it by the proportionality
8

+ 2 + 3 = v w rg 0, a w loc. (60:17:b) cit. i.e. e have the Case(b) id., nd it

factor

The a i,
u
r))

r * w factor

mentioned

ai,

at

in

above. w loc. (60:15) cit., hich reappearhere, include already the

568))

GAMES)) GENERAL NON-ZERO-SUM

from here on by graphical is ures 92-95, present. Therefore we proceed representation. w Forthis representation e use,as before,the planeof Figure52. Repre-

in were as sentingthe sixlinesof (62:12)those of (60:15)60.3.2. represented))

u)))

, .<

u 3-

Figure 96.))
V: The line and the curve))

Figure

97.))

Figure

98.

we by Figures92-95, obtain Figure96. The qualitative features of this figure follow from the following considerations:)) (62:A:a)))
first ajj- and

of The secondaj-linegoes through the intersection the Indeed:)) c^-lines. 2(w

u)))

u))

u))

= 0.))
first

(62:A:b)
(62:A:c)))

The two a-lines areidentical. Theseconda^-line is to the left of the It has a greater 3'-value, since a
3v

one. Indeed:

2w

.w
-,

~))

shows 92-95 that it is a (rotated Comparisonof this figure with Figures form of Figure94 1 The area is degenerated a point to and) degenerate vertex of the fundamental triangleA), the areas , also (the upper 1For this and the remarks which follow, cf. footnote 4 on p.
553.)))

ECONOMICINTERPRETATION FOR n

3:SPECIAL

569

but degenerated, to two linear intervals (the upper and the lower part of the left side of the fundamental triangle A), while the areas , are still undegenerated trapezon and the smallertriangle, into which the (the
fundamental triangle A is divided on our figure). This dispositionof of the five areas Figure94 is shown in Figure97. The general olution V s now obtains,as stated at the end of 60.3.3., 86 by fitting the pictureof Figure 1 into the situation describedby Figure97. Figure98 showsthe result.

62.4.The Solutions : GeneralForm Before we go any 62.4.

validity, assuming
further we note that u

Figure97 is of general

(62:13) (62:14)
When

<v ^
v

w,

but the picture it gives refers qualitatively to

< w.
= w,

(62:15)

t then the area in Figure97 i.e.he upper interval on the left side of the in t fundamental triangle degenerateso a point. (Cf. (62:A:c) 62.3.2.) in o Hence this caseFigure98 assumesthe appearancef Figure99.))
V: Theline \\ and the curve

V: The curve))

Figure

99.

Figure

100.

t This discussioncan be rendered quite symmetric with respecto the the two buyers by: players 2,3 we or (62:13) the weaker by Assuming (62:14) (62:15), may replace condition

(62:16)

< v, w.

Let us therefore assume(62:16) and not (62:13) (62:14), with only (62:15). Thismeansthat eachbuyer derivesa higher utility from the possession A of than the seller,ut it does not place the buyers with respect eachother. b to (Cf. the discussionin the first part of 62.1.2.)
1The curve in Figure 98 is like those in there: (47:6)in 47.5.5.)))
Figure

86, subject to the

restriction

stated

570
Now

GENERAL NON-ZERO-SUMGAMES

and leaves (62:14), (62:16) threepossibilitiesopen: (62:15), v > w. (62:17) were The solutions of (62:14), (62:15) given by Figures98,99. (62:17) the obtains from (62:14) interchanging two players 2,3 the two buyers by and w. This means that Figure98 must be reflectedon its vertical middleline (after interchanging, w). This is shown in Figure100. v Summingup:
t>,

(62:B)

1 as follows:

The 62.5.1. result expressedy Figure98 can be statedalgebraically b

for 98, 9,100 v <, =, > w, respectively. 9 62.6. Algebraical Form of the Result

Assuming

the (62:16), generalsolutionV is given by Figures

The solution V consistsof the upper part of the left side of the fundaa mental triangle, nd the curve The first part of V is characterizedy b

~.

'

~~

3v

~\"

2w

u
~~

~~

__ w

in Owing to (62:8) in Now (62:6)

this 62.3.1., means that

2

= 0,
a\\

0.

62.3.1. gives

=w 3, hencethe above conditioncan be written as a8 = w v g ai ^ w, a* = 0, (62:18)

The secondpart of
V

- 01.
>

(the curve) extendsfrom the smallesta{above

of

to the absolute minimum

Its geometrical V a{ ( shape (cf. can by (47:6)in 47.5.5.) be characterized stating that along it 2 <*s are functions of a{. We may again pass from both monotonic decreasing
i in

monotone decreasing functions of

u (62:19) g i

in to i, 2, as by (62:8) 62.3.1. a\\ varies from its minimum Then above (62:18) (v) to its absoluteminimum (u), and 2, s areagain both
2>

i.

So we have:

2 aremonotonic decreasingunctions of ai. 8 f and Thus the general olutionV is the sum of the two setsgiven by (62:18) s

g v,

c* , 2

1Note that it holds whenever v ^ it>, (62 notwithstanding. :B) 2 They must, of course,ulfill (62:5),(62:6) in f 8 As w Figure 98shows, the lowest point on the line coincidesith the highest point on the point a.\\ *= v of and of is the same. the curve. Hencewe couldexclude ai v from either (but not from bothl) of (62:19).)))

62.3.1.

I.e.

(62:18)

(62:19)

(62:18),

ECONOMIC INTERPRETATION FOR n

- 3:SPECIAL 571

It are (62:19). will be noted that the functions mentioned in (62:19) a arbitrary (within certainlimits),but that a definite solution (i.e. definite standard of behavior) corresponds a definite choiceof these functions. to This situation is entirely similar to those analyzed in (47:A) of 47.8.2. and in 55.124. 62.5.2. (62:18), can be used whenever v ^ w (cf.footnote 1on (62:19) to p.570). Forv = w (62:18) simplifies (62:20)
We
==

a\\

= v,

= 0.

shall therefore use (62:18), only (62:19) when v < w, and (62:20), when (62:19) v w. 1 If v > w, then we can utilize (62:18), (62:19) interchanging the by players 2,3 the two buyers and v, w. Then (62:18), become (62:19) w ^ ai g v, a2 = v a3 = O.2 (62:21) i, 1 functions of ai. (62:23) u g a\\ g w a*, 3 aremonotonic decreasing Summingup:

(62:C)

the (62:16), generalsolution is given by (62:18), (62:19); (62:23) (62:20), (62:19); (62:21), for v <, = , >w
Assuming

respectively.

62.6. Discussion the ordinary, common-sense analysis to the apply market of one seller nd two buyers and one indivisibleunit of a good, in a its orderto compare result with the mathematical onestatedin (62:C). a The lines of this common-sense procedure reclearlylaid down* we are of cases the theory of actually dealingherewith one of the simplestspecial \"marginal pairs.\" The argument runs as follows: offers only one indivisibleunit of the good under considerThe seller ation and there are two buyers. Hence will be includedin the transone action,and one will be excluded.Clearly the strongerbuyer will be in w the first position except hen the two buyershappen to be equally strong, eitheris eligible.Accordingly the priceat which the transin which case will actiontakes place liebetweenthe limits of the includedand the excluded must beprecisely buyer and if they happen to beequallystrong,the price w their common limit. The limit of the seller, hich must be assumedto be

62.6.1. us now Let

l The observation of footnote 3 on p. 570 concerning (62:18), (62:19) appliesalsoto H ( (62:20), 62:19). encewe could omit (62:20) ltogether, but it is more convenient to a keepit, for the sakeof the interpretation in 62.6. * Note that, owing to the aboveinterchange, (62:4) 62.1.2. becomes in (62:22) v((l,2,3)) , becomes and so (62:6)in 62.3.1. ai + a* -f as = v. (62:6*) 8 Theobservation of footnote 3 on ) (62:19) to (62:21, applies p. 570concerning (62:18), a (62:23) lso. Of coursewe must replaceits v by

w.)))

572

GENERAL NON-ZERO-SUM GAMES

lower than that of either buyer in orderto have a genuine three-person in into play. market, comes no case In our mathematical formulation the limits of the seller nd of the two a buyers were w, v, w. The above remarkmeans

(62:16)

< v, w.

the amount to Thestatements concerning price v for v < w, (62:24) p w for v = w, (62:25) p=v w gp gv for v > w. (62:26) in A buyer who is excluded, finisheswhere he started i.e. our normalization
of utility at

Consequentlyour present statements correspondexactly to (62:18), as providedfor by (62:C). (62:20), (62:21), results agree. But the Sofar the mathematical and the common-sense i limit of this agreements also in evidence: (62:C) providedfor the further and thereis no trace thesein the ordinary of (62:19), of imputations (62:23), above. as presented treatment, D a What then is the meaning of (62:19), (62:23)? o they express conflict betweenour theory and the common-sense standpoint? It is easy to answerthesequestions,and to seethat thereexists real no but that (62:19), a perfectly proper extension conflict, (62:23) represent of the common-sense standpoint. i The 62.6.2. amount obtained by the sellern a given imputation, i, is when that imputation is offered. In (62:19), clearlythe price envisaged p to a\\ (62:23), varies from u to v or w (according which is smaller) i.e. limit to the weaker buyer's limit. There varies from the seller's the price b is also a definite (monotonic) functional connectionetweenthe (variable) amounts obtained by the two buyers.1
the Thesetwo facts strongly suggestgiving (62:19), (62:23) following b The verbal interpretation: two buyers have formed a coalition, ased on a definite rule of division for any profit obtained, and arebargainingwith the seller. The rule of division is embodiedin the monotonic functions that u No occurin (62:19), (62:23). bargainingcan depressthe seller nder his 2 above the limit of the weakerbuyer own limit. On the otherhand a price him from any possibilityof exerting influence. would exclude and The specificrules contained in (62:19), (62:23), the rolesof all verbal treatparticipants in thesesituations may be given more extended ment. We shall not do this here,sincethe above shouldsufficeto establish our main point: the one hand (62:18), On (62:21) t (62:20), (i.e.he upper to correspond the competition of the two parts of V in Figures98-100) i buyers for the transaction in which the strongerplayer, if one exists,s 1 All these \"amounts\" refer to the utility in which we reckon of the goodsunder
consideration

zero.

there exists only one, indivisible, unit. 1Theseller's is his bestalternative use (instead limit

of

a sale)for

A.)))

ECONOMICINTERPRETATION FOR n = 3:GENERAL 573 sure to win. On the otherhand, (62:19), ( the (62:23) i.e. lower part of V in Figures the 98-100, curves)correspondto a coalition of the two buyers, against the seller. Thus it appears that the classical argument at leastin the form used in 62.6.1. gives the first possibility only, disregardingcoalitions. Our from the beginning, is theory, to which the coalitionscontributeddecisively I both i necessarilydifferent in this respect: t embraces possibilities, ndeed it gives them weldedtogether, s a unit, in the solutionswhich it produces. a The separation, ccording schemes ith and without coalitions,appears to w a only as a verbal comment on the relatively simple three-personame g there is no reasonto believe that it can becarried for all games,while out the mathematical theory appliesrigorously in all situations.

was in the same way that the two-personsetup of 61.2.2. extended

63.Economicnterpretation of the Resultsfor n = 3 : General Case I 63.1. Divisible Goods It in 63.1.1.remainsfor us to extend the three-personetup of 62.1.2. s

61.5.3.

61.5.2.,

1 in the return to the situation described 62.1.2. seller Let us accordingly and the (prospective)buyers 2, 3 in a market. We allow now for transactions involving any or all of s (indivisibleand mutually substitutable) units Ai, - - - , A, of a particular good. (Cf. also footnote 1 on p.560.) Denotethe value of t (= 0, , s) of theseunits by u t for by v t Thus the quantities w t for for 2,and by

3.

1,

1,

(63:1) (63:3)
(63:2)

t/o
v
MO

= 0, = 0, = 0,

MI,

vi,

wi, -

, u,, , v., , w,,

the describe variable utilities of theseunits for eachparticipant.

As before, we use for eachbuyer his original positionas the zero of his utility. and we neednot repeat he considerations t As in the rules of the game which models of concerning

this setup. f It is easy to seewhat its characteristicunction must be:Sinceeach a to canblock sales him, and the seller s well as both buyers together buyer it followsas in 61.3.3. that and can blockall sales 61.5.2.62.1.2.), (cf.

61.5.3. 62.1.2., 61.5.2., 61.3.1., 61.2.2., 61.3.2.

(63:4) (63:5)
3 by
t,

v((l)) = u., v((2)) = v((3)) v((2,3))= 0.

- 0,

1 the Denoting number of units transferredfrom the seller to the buyers 2,

w r, respectively,it is easy to express hat the remaining coalitions t i.e.he sellerwith either one or both buyers can))) (1,3), (1,2), (1,2,3)

574

GENERALNON-ZERO-SUM GAMES

achieve. The familiar argumentsgive)) )) = Max,_ ,i ) = Maxr _ = Max,._0>l r (63:7) v((l,2,3))

0>1

.(_. ..... ...... ......

(u._<

+ v t), + ,), (u._,_r + v,

+))

This v(S)is a characteristic function. We leave it to the readero verify t the inequalitieswhich areimpliedthereby. The discussion to when this game is essentialcan be carriedout as in as

i.e.

when oneof the two buyers2,3 becomesdummy in the sense our theory a of of decomposition. e shall not considerthis either;the result is not W difficult to obtain, and though not surprisingit is not uninteresting. r 63*1.2. Restricting in the maximum of (63:7)to the value converts this into the first maximum of (63:6). estrictingt there to the value R convertsit into u, By eachone of theseoperationsthe value becomes , g we have))

2 61.5.3. 61.5.2., and is left to the reader. It is also possibleto determine

(63:8)
If we do the same to r, t in the reverseorder, e obtain similarly w

(63:9)

v((l))

v((l,3))

Considerthe first inequality in that the first maximum in (63:6)is assumed for t = 0. According to the usual ideas on the subjectthis meansthat the seller nd buyer 2, in the a of absence buyer 3, would effect no transfers. that buyer 2, in the of buyer 3, is unable to make the market function. absence T Consider the secondinequality in (63:8). o have equality there meansthat the maximum in (63:7)s assumedfor r = 0. According to the i usual ideas on the subject,this means that the seller nd buyer 3,in the a that buyer 3, in presenceof buyer 2, would effect no transfers. of the presence buyer 2,is unable to participate in the market. Summing these up, togetherwith the correspondingstatements for (63:9)which obtain by interchanging buyers 2, 3, we have:

v((l,2,3)). T (63:8). o have equality there means

I.e.,

I.e.,

(63: A)

to influence the market. 1The extra condition t -f r ^ under this Max expresses that units sold cannot exceed number of units originally possessed the by
discussion of the relationship with in when 8 can alsobe carried out easily. Thediscussion and footnote 3 on p. 562should be remembered.)))

(63:9) Equality in any one of the four inequalitiesof (63:8), is a sign of someweaknessof one of the buyers. In the first inequality of (63:8), it [(63:9)] means that 2 [3], in the absence buyer 3 [2], is unable to make the of buyer market function. In the second it [(63:9)] inequality of (63:8), that buyer 3 [2], in the presence f buyer 2 [3], is unable o means,
in or with 62.1.2. (62:16) 62.4., of (61 at the end of 61.5.3. :B)

'The

- 1,

(62:1)

the number t the seller.

-f r of

ECONOMICINTERPRETATION FOR n

- 3:GENERAL 575

The really interestingcase arises, obviously, when all theseweaknesses are excluded.It is therefore reasonableto make the hypotheseswhich this: express

(63:B:a) (63:B:b)

We have

(63:9). (63:9).

< in

the first inequalitiesof both (63:8)and

We have

< in the secondinequalitiesof both (63:8)and

of the Inequalities

63.2.Analysis

Thismeansthat one of the two playersis absolutelystrongerthan the other. More precisely:That he is at least as strong as the other player,even when from the market. t he tries to excludehe other player completely i we Hence may expectn this casea result which is similar to that u when only obtainedin 62.1.2.-62.5.2., one (indivisible)nit A was available. I.e.the divisibility of the supply into units Ai, , A, which we have effective. shouldnow become here, Thisis indeedthe case. To prove it, introducethe quantitiesu, v, w by

but 63.2.1. for a moment (63:B:a), Assume

the negation of

(63:B:b).

(63:10) v((l)) = u,

v((l,2)) = v

v((l,3)) = w.

Then the second inequality of (63:8)and of (63:9)and the negation of give (63:B:b)

(63:11)
(63:12)
with
v

v((l,2,3))= Max (v, w)

and and (63:8) of (63:9)
u

while the first inequality of

(63:B:a) give

<v,w.

of the coincides Now we have precisely conditions (63:12) in while (63:4), 63:10) (62:2), in 62.4., and give ( (62:3) (62:16) v ^ w) or (62:22) in 62.5.2. in 62.1.2. (when (when gives (63:11) (62:4)

62.1.2.-62.5.2.: 62.1.2.,

and are the Consequently results of 62.4. 62.5.2. valid, with the u, v, w solution obtains as described, in (62:B) in The of (63:10). general e.g. to according Figures98-100. 62.4., are From 63.2.2. now on we assumethat (63:B:a), (63:B:b) both valid. We introducethe quantities u, v, w, z by
v, (63:13) v((l)) = u, v((l,2)) = = v((l,3)) = w, v((l,2,3)) z. (63:14) state and Then (63:8), (63:B:b) that)) (63:9) (63:B:a),

g w).

(63:15)

u<)))

576
worth while to

GENERAL NON-ZERO-SUMGAMES
them compare
in

This arrangement iffers from that of d

id. loc. it.,nd (63:4),(63:13), correspondto (62:2)-(62:4) (63:14) c a It is now convenient to introduce again the assumption of decreasing In 61.5.3. fact we need it now at a already utilized in 61.5.2., utility, somewhatearlier tagethan we did then : It is now (at leastin part) useful s
in

detail:(63:15) correspondsto (62:1)

but 62.1.2., it

is nevertheless

the mathematical part of the theory,1 while we neededit there only in the interpretative part. We statethe decrease utility for all three participants 1,2,3,: of

(63:16) (63:17) (63:18)

can be written as follows:

Max, , i
(w.-<

> u w,_i, > > > v, v._i, > w, w,_i. Wi wi > In the immediateapplicationonly (63:16) be required. This is it: will v + w > z + u* (63:19) the O (63:7) (63:14), assertion(63:19) Proof: wing to (63:6), and (63:13),
u\\

vi

Uo t;

WQ

> > 2 > wt

M2
t>

ui

Vi

r + v t) + Maxr.0,i (u,- + w r) > MaX,, r_ , r (Ut-t- + t + Wr) + U.. t+ri

1 V

t Considerhe t t r for which the maximum on the right-hand sideis assumed. i.e. Sincewe have (63:B:b), < in the secondinequalitiesof (63:8), (63:9), f that we canconcluderom the argumentation of 63.1.2. these/, r are 7* 0. our We denotethem by to, r . Hence assertionis this

Max,., i

(u 9

-+
t
V

v t)

+ Maxr_ , i

r , (u,- + w r)

> W.-l -r + + Wr + U,.

Vt

we T I.e., claim: hereexisttwo J, r with

W,_l r + t + U -. + Wr > W._
9
<o

_r

+ t\\ + Wr + U,.

for Now this is actually the case then be written as may

=/,r =r

The above inequality

statesthat
1But not somewhat. * In

w._ _ro > ti. w.w._r o (63:20) v It should be conceptuallyclearthat this follows from our assumption of in utilities. Formally it obtains from (63:16)this way: (63:20) decreasing
<o

case

of the necessary; absence this property would only complicatethe discussion this 62.1.2. was trivially true. Indeedusing (63:13), we obtain in that (63:14)
u

<v ^

and this gives (63:19) immediately.)))

ECONOMIC INTERPRETATION FOR n

(63:21)

- 3:GENERAL 577 - > 5) (tt.-*fi - U.-i). *-l(tt.-r.-tfi W.-r.-4) i-1

<e

(63:16) implies
tt,' W,'_i

>U

Ut-\\

whenever

s' < s\", hencein particular

tt,_r

-i+i ~ M-r-< > Wt _,-+i

U._,-,

and from this

Discussion 63.3. now apply We 60.3.2. the present setup. This will to for prove to be quite similar to the applicationcarriedout in 62.3. the setup of Theexpositionwhich follows will therefore be more concise, and is bestread parallelwith the corresponding parts of 62.3. As to the comparisonof the mathematical result with that of the ordinary, common sense approach, the remarks of 62.2. apply again. We indicated already there what complicationsthe present setup produces. We shall consider situation only briefly, although it is a rather important the one. The generalviewpoints were sufficiently illustrated by our earlier, and d simpler examples, the specific, etailed interpretative analysis of this o setup and other,even more general nes will be taken up auo jure in a

follows. (63:21) 63.3. Preliminary

60.3.1.,

62.1.2.

subsequentpublication. 63.4.The Solutions 63.4.1. imputations in the presentsetup arethe)) The

with

^ 0, a* ai ^ u, 0, (63:22) i + 2 + as = z. (63:28) form. It is again necessaryto introducethe reduced

<* 8

This amounts to a

transformation

(63:24)

discussedthere determinethe J, in as described 62.3.The processes now becomes)) aj,so that (63:24)

(63:25)i

- Bl

-, !-,v(/S)

a' = a* + ak
k

aj,

i= a.

on The correspondingchanges

(63:14) (63:13), they carry (63:4),

are again

given by

in (59:1) 59.2.1.;

into)))

578

GAMES GENERAL NON-ZERO-SUM

(63:26)

v'((l)) = v'((2)) = v'((3)) =

(63:27) v'(( l,2)) =

*L^J>,

^^

- -jA - 2. _8

-))

(63:28)
Thus 7 = -s
>

= v'((l,2,3))0.
and we again refrain from passing to the normalization))

60.3.2. 60.3.1., as described in 62.3.This proportionality factor

u

Hencewe

must again

insert a proportionality factor when applying is now

now

~3~\"
Comparisonwith ai==
2(z

_,
u)

in shows (60:8) 60.3.1. that

a2 =

3w

-_ 2z

a8 =

- 2z ___ - u
3v

Thesixlinesof (60:15)60.3.2. describe trianglesfrom in which the

now:)) derived our solutions,become

which we

!-

(63:29)))

,= z u g-> 2(g - ^) 'o

2
/

,= ~ z

-3-, - 2z _ _
3w

, a, = - z u -3-,
, = 1
3v

2
Q))

63.4.2. Applying

the criterium of 60.3.3.,find that we

fli
8

+ 02 + a = v + w 2z g 0. Hence e have again (60:17 loc. it. i.e.he Case(b) id.,nd it remains w t a :b) c w tobe decided hich oneof its four subcases, represented Figures92-95, by
is present.
of as Following the same procedure graphical representation in 62.3., we obtain Figure The qualitative features of this figure follow from the following considerations:

101.

(63:C:a)

a'- and aa-lines. Indeed: 2

2(z 3
u) z u

Thesecond o a[-line oesthrough the intersectionf the first g

~3~ \"3

n U

'

(63:C :b)

(63:C:c)

Thesecond [a{-] is to the right [left] of the first one. aj- line

Indeed: has a greater j-[a,-]alue, since))) It a v

ECONOMICINTERPRETATION FOR n
3w
3v

- 3:GENERAL
> 0,
>0.))

579))

2z 3 2z

u
u))

z
*

w
v

(63:C:d)
g
\"\"

The first a{-lineies below the intersectionf the secondaj l o and aj-lines. Indeed: u 3w 2g u _3v 22 w_ , n 3 3 3

z+u-v-w
__ __

in by (63:19)63.2.2. showsthat it is again a Comparisonof this figure with Figures92-95 form of Figure 94 (cf. footnote 1 on p. 568), (rotated and) degenerate than the correspondingFigure96 in 62.3.: The)) although less degenerate

Figure 101.)) V: Thearea

and the curve

Figure 102.))

I))

Figure

103.

area is again degenerated a point (theupper vertex of the fundamental to are still undegeneratedthe four ( triangle A), but the areas , , , areasinto which the fundamental triangle is divided in our figure). This dispositionof the five areasof Figure94 is shown in Figure 102.The

by generalsolution now obtains as stated at the end of 60,3.3.,fitting the of Figure into the Situation described Figure102.Figure103 86 picture by showsthe result (cf.footnote 1 on p.569).)))

580

GAMES GENERAL NON-ZERO-SUM

Assuming (63:B

Summingup:

(63:D)
V

is given by Figure103.

the s ( :a),63:B:b)and (63:16), general olution

shows Comparisonof this figure with thoseof 62.S.-4. that Figure103 and is a form intermediate etweenthoseof Figures98-100, thosefigures b

arein

turn

f degenerateorms of Figure103.

63.6. Algebraic
in

Form of the Result

the same as wasdone for Figure98 in 62.5.1. way In Figure103 solution V consistsof the area m and the curve ~ . the first part of V is characterizedy The b
3w

can The 63.5. result expressed Figure103 be stated algebraically, by

2z 3

^ a, =

'

3-'
=Z
2

3v

2z 3

s a,' = ~ z

u
3\"\"'

Owing to

in this (63:25)63.4.1., means that z v ^ z w ^ a? ^ 0, in Now (63:23)63.4.1. gives

1

0.

3,

and so the exact angeof i is r

v

+w
u by

+.w .z ai z, Q a* z- w, Q a* z v, ai + 2 + <*s = z. o Therangesin the first line of (63:30) the precisenesfor i, 2, s. are : Thesecond of V (the curve) can be discussedliterally as in 62.5.1. part ai varies from its minimum in (63:30)above (v + w z) to its absolute
v

in We (Recallthat v + w z> (63:19) 63.2.2.) state all these than conditionstogether,the result being somewhatmore complicated its in It analogue (62:18)62.5.1. is this:

z^ai^z.

minimum

we have:

functions of (u), and a2, as are monotonic decreasing

i

ai. So

u (63:31) g

if

+w

z,

aa , a8 aremonotonic decreasing functions

w Ofi.

Thus the general olutionV is the sum of the two setsgiven by (63:30) s and (63:31). will be noted that the roleof the functions in (63:31) It is the same that discussedat the end of 62.5.1. as
the lowest point in the area coincidesith the highest point w on the curve. the point ai v -f w zof (63:30)nd of is the same. a Hencewe could exclude v + w z from either one (but not from both!) of

1They must, of course,ulfill (63:22), 63:23)n i f (

* As Figure

I.e.

103shows,

63.4.1.

(63:31)

<*\\

( (63:30), 63:31).)))

ECONOMICINTERPRETATION FOR n
Summing

- 3:GENERAL 581

up:
tion V is given by
Assuming

(63:E)

common-sense nalysis to the market of one seller two buyers and and a indivisibleunits of a particulargood,in order to compareits result with the mathematical one stated in (63:E). Actually the interpretation which ought to be carried out now must combinethe ideas of 61.5.2.-61.6.3. of 62.6.: former apply with those the because have divisibility into s units; the latterbecause market is we the oneof threepersons. As indicated in 63.3., do not proposeto go fully we into all details on this occasion. The two parts (63:30), of (63:31),which our present solution consists arecloselysimilarto the two parts (62:18), (62:19) (62:20), or (or (62:19), with (62:21), obtained in 62.5.(Cf. also (63:E) in 63.5. (62:C) (62:23)) in 62.5.2.) appears most reasonable, It therefore, to interpret them in the same way as we did in the correspondingsituation in 62.6.2.: (63:30) describes situation where the two buyers compete the s units in the the for describes situation where they have the seller's possession,while (63:31)

and the (63:B:a), (63:B:b) (63:16), general solu(63:30), (63:31). 68.6. Discussion 63.6.1. now perform the equivalent of 62.6. applythe ordinary Letus and

united. The readerwill have no formed a coalition and face the seller in amplifying the details, in parallelto 62.6.2. difficulty the Thesebeing accepted, is nothing new to besaidabout (63:31), there situation in which the buyers have combinedand do not compete.(63:30) their however, which describes competition,still deservessomeattention. to and Let us considerthe imputations belonging (63:30), letus formulate their contents in terms of the ordinary conceptof prices. This is the 61.6.2. point in equivalent of what we did at the corresponding We introduceagain the t, r for which the maximum in (63:7)
r v((l,2,3))= Max,, t+r

J is assumed:, r

.
9

61.6.1.,

,i

+ (u.-*-, v t + w r)
i

Sinceour imputations
2

a =

actually distribute the amount v((l,2,3)),these < numbers of units actually transferred by the sellerto the buyers
Q

{i, ,

s}}

with

+ 2+

s = v((l,2,3)) the , r must represent

2,3,

respectively. 61.5.3. The analysis of 61.5.2., leading up to (61 could now be :A), mutatis mutandis. It would show that numbers of units transrepeated criterion of with Bohm-Bawerk's in ferred, f , r , can be described accord \" \" number c f as was done loc. it.or the corresponding marginal pairs just this discussionwould bring up nothing new, we shall of transfers to. Since
not dwell upon the point any
further.)))

582

GAMES GENERAL NON-ZERO-SUM

to the questionof prices. Thebuyers2,3 received,

63.6.2. now turn We

as we saw, J , r units, respectively. Theimputation a on the otherhand, ascribes the amounts at, s. Thesetwo descriptionscan be harmothem
nized only by establishingthe equations

(63:32) (63:33)

vt,
u? fg

- - at| - =
Up
r*q
8

and interpretingp, q as the pricespaid per unit by the buyers 2,3 respecbut in are tively. (63:32), (63:33) the equivalents of (61:24) 61.6.1., it for the two must be emphasizedthat we obtained two different prices buyers! Now (63:30) an bestated in terms of p, q l as follows: c

(63:34)

(63:35)

We in Theseinequalitiesare the analoguesof (61:25) 61.6.1. could treat them in a way similar to that there,and compare them with thoselimits which result from the applicationof Bohm-Bawerk's theory. We shall not

rK-*+ uOS P 1 .- + g
10

r
J<>

\\,
u>

(w r

v)

63.3.A few remarksmay neverthelessbe appropriate. The intervals (63:34) nd (63:35) reagain wider than those of Bohma a numerical examples Bawerk'sheory just as in 61.6. (61 id.).Some t (cf. :C) show, however, that the difference tends to be smaller. It is therefore possible although nothing has beenproven in this respect that a further t in increase the number of buyers may tend to obliteratehis discrepancy in that part of the solution which corresponds no coalition betweenthe to with the greatest buyers. This surmise must, however, be considered sincewe know too well how rapidly the complicationof solutions caution, with increases the number of participants and how difficult the interpretation of different parts of the solution may then become. It will beobservedalsothat we had to introducetwo (possibly)different for prices the two buyers, and this in spite of our, still valid, assumption of complete information. This is perfectly in harmony with the interpretations of 61.6.3.:saw therethat what we calledprices erereally only We w of several different transactions, hat the sellerand the t average prices buyers must have beenoperatingwith premiumsand rebates and all this is necessarily conducive to a differentiation betweenthe two buyers. All Finally, we may statethe equivalent of the last remark of 61.6.3. theseabnormalitiesin the formation of the price structurearealso quite in
carry this out in detail for the reasonsstated in
into statements on

( concerning a t , as can be translated by means of (63:32), 63:33) p, q. The statement of (63:30) oncerning a\\ is merely a consequence those on at, at of c z. Henceit neednot be considered.))) using ai -f at -f- aj

I.e.ts statements i

THE GENERALMARKET
monopoly versusduopoly.

683

w agreement ith the fact that the market under considerationis one of a

markets which we have consideredso far were very or of three participants. We shall now a step further and considera more general market, which consistsof go I + m participants: I sellersnd m buyers. This is, of course, till not the s a

restricted: consistedof two They

64.1.1. The

64.The General Market Formulation of the Problem 64.1.

T most generalarrangement: hat would have to provide among other whether he will that eachparticipant can choose for the possibility things f or or sell; again, that he may be selleror one classof goods,and buyer buy for another. In this study, however, we shall content ourselveswith the above case. Further,we proposeto considerone kind of goodsonly, of which units
A\\
f

It is convenient to denotethe sellers 1, by 1

- - - , A,

areavailable.

the buyers by
and the

*,,
m*,

L=

(l, '

,0;

, I, their

setby

their set by

M= (1*,
- -

,m*);

setof all participantsby / = LuM = (1,

,1,1*,

1 m*).

Denotethe number of units of the goodsunder consideration, riginally o

in the

i of possession the seller by

s. Then clearly
*= .
,

(64:1)
the Denote utility of t( = 0,1,
u\\

t-i
utility

and the

v}*.

Thus the quantities

it<

of t( = 0,

1,

, s) units of the ,
f

s) units

of the goodsto the seller by goodsto the buyerj*by

, 0, (i = 1, m*), (j = t>r the describe variable utilities of theseunits to eachparticipant. As before,we use for eachbuyer his original positionas the zero of his

(64:2) (64:3)

= 0,u\\, 4* = 0,i^*,

<

!*,,
, I, I +

utility. As in

we 61.5.3., and 61.5.2., 62.1.2. 63.2.1.,need not repeatthe conthe siderationsof 61.2.2., 61.3.2. 61.3.1., concerning rulesof the game which

modelsthis setup.
1We use this

notation

instead

of the

conventional

1,

1,

,I+

m.)))

584

GAMES GENERAL NON-ZERO-SUM

native

function v(S) of this The 64.1.2. determination of the characteristic gameis easy: Clearly S / = L u M. We now consider successivelythree alter-

no transactions among themselves.One merelystatestheir original position:

First: &L. In this caseS consistsof sellersnly, who can carry out S o
sees immediately
that v(S)
v(fl)

possibilities.

(64:4)

tinS

2<

S Second: s M. In this case consistsof buyers only, who areequally unable to carry out any transactions among themselves. One seesagain
that v(S) merely statestheir original position:

v(S) = 0. in Third:Neither S cL nor S c M i.e. has elements common both S a with L and with M. In this case containssellerss well as buyers, hence S transactions between theseare definitely possible.On the basis of these, the following formula obtains: (64:5) (64:6) v(5) = Max^-o,i. - , ..(.-inani) in ,(;* Sfl M) f-,,-0,1, S , + S v \" S
in

( :in % \\ S
<

u\\ t

' + j*

t
in

in S fl L SH If in In this expression n Z/ is the set of all sellers S,S n M the set of all S i buyers in S,J, the number of units transferred from the seller (in S n L), r,.the number of units transferredto the buyerj* (in S n Af). 1 Thereader

S fl L

j* in

will

now have no difficulty

in verifying

the formula

(64:6).

64.2.SomeSpecial roperties. Monopoly and Monopsony P 64.2.1. arefar from being able to discussthe theory of this game We the market of I sellers nd m buyers exhaustively. We have at present a a cases nd beyondthis only a only some fragmentary information on special wider areas. The problems which arisein this few surmises concerning connection seemto be of definite mathematical interest,aside from their I economicmportance. t would seem i h premature, owever, to discussthis beforethe investigation has penetrated subject deeper. Instead shall draw someimmediateconclusionsrom the two simpler we f of our equations: (64:4) (64:5) They areas follows: . , All setsS s L and all setsS s M areflat. (64:A)
which

1Thereis no need to state here which selleris transferring each particular unit to buyer: The resulting utilities which alone enter into v(S) are not affected by this. All negotiations between individuals, coalitions, compensation, etc.,must be taken careof by the application of our theory.))) automatically

THE GENERAL MARKET

T Proof: his meansthat v(5) = v((*))
JfcinS

585

for

ScL

and for

8 s M.
is inessential.

which follows immediately from

(64:4),(64:5).
if and only
if

(64:B)

The game is constant sum

it

Proof: ufficiency:Inessentialityclearly impliesconstant-sum. S Assume that the gameis constant-sum. Necessity: As L, M arecomplementary sets (64:7) v(7) = v(L) + v(M). Now by (64: (with S = L, M) A) (64:8) v(L) = % v((fc)) v(M) = % v((*)).
f

kinL

kinAf

Combining(64:7)and (64:8)we obtain

(64:9)
in our

v(7)

=
*in/

v((*)).

to Now the modification of (27:B) in 27.4. hich applies according w of as case, ivesjust (64:9) a criterion inessentiality. g It may beworth noting that the criterion inessentiality(64:9) ecomes, of b T this: he maximum in when statedexplicitlyby means of (64:4)-(64:6), m Now this is the value of the expression aximized (64:6)is equal to u\\ that (64:6)when t* = s, r, = 0. So the statementbecomes, the maxir ; = 0,i.e. hen no transactions w mum in (64:6)is assumed when U =
in
t in L

59.3.1.

..

takeplace. Hence :B) can alsobe formulated as follows: (64

(64:B*)

that the at no transactions take place all is r, == (64:6) is assumed when J, = or t to these:hat the gameis constant-sum; equivaequivalent t lently (in this case!)hat it is inessential. a Thesalientpoint of this result is that our game,representing market, can be constant-sum only at the priceof the market being absolutely this problem belongs quite intrinsically to games of ineffective. Hence

are such that

maximum

a Thefact that the individual utilitiesof the sellersnd buyers

in

s,

i.e.

a nonconst nt-sum.
(64:C)

We 64.2.2. now continue in a somewhatdifferent direction.

t Considerwo imputations,

a =

{{ai,

off,

i,

,am'!K)))

586

GAMES GENERAL NON-ZERO-SUM

Assume that
H

0,
Then neither

S n L nor Sn M can be empty.1

for S being the set of 30.1.1. this domination.

S O Proof: therwisewe should have S M or S L. Hence is flat by (64:A)and therefore certainly unnecessary(cf.59.3.2.). We concluderom (64:C)that in this case f for at leastone i in L, a > ft (64:10) for at leastone j* in M. (64:11) a/. > ft*
have and Theseformulae (64:10) (64:11) a roleof some interest,when set: eitherL or M is a one-element I = 1or m = 1. This meansthat there that o exists reciselyoneseller r preciselyonebuyer, i.e. we have monopoly p

is uniquely determined: or In thesecaseshe i of (64:10)the j*of (64:11) t = i = 1or,;* 1*. Se we have: (64:D)

or monopsony.

>

>

(64:12) (64:13)

implies

ai* > 0i*

*

ai > 0i

when when
*

1=
m

1, 1.

The remarkable thing is that both

tradiction in

and are (64:12) (64:13) transitive

H ft is not. Thereis, of course,no conis or (64:13)merely a necessarycondition for this, (64:12) But it is neverthelessthe first time that the domination concept a H in an actualgameis so closelylinked to a transitive relation. t s f This connectioneemso be a quite essentialeature of the monopolistic 2 situations. It will play a roleof some importance in (or monopsonistic)

relations,while domination a

65.9.1.
I.e.

1 S must contain both sellersand buyers. 1The verbal interpretation of (64:12), is (64:13) simple and plausible:No effective without the monopolist (or monopsonist).))) domination is possible

CHAPTER XII EXTENSIONSOF THE CONCEPTSOF DOMINATION AND SOLUTION

with

domination and solution, which were then unambiguously established. Neverthelessin the subsequentdevelopmentof the theory thereoccurred underwent variations. These repeatedly instances where these concepts w instances ere of threekinds: I First:t happenedin the courseof our mathematical deductions,based on the original definitions, that concepts ose importancewhich r to strictly were obviously analogousto the original ones (of imputation, domination, it solution) but not exactlyidenticalwith them. In this case was conventhem by those names,necessarilyremembering differthe ient to designate ences. Examplesof this areto be found in the investigation of theessential in where the discussionof the game with excess 47.S.-47.7. three-person fundamental triangle is reducedto that of one of the various smallertriangles in it. Another exampleis offered by the investigation of a special simple 2. of gamein 55. -55. where the discussion the original domain is n-person to reduced that oneof V in (cf. the analysisof I Second: n the courseof our considerationson decomposabilityin the o ChapterIX, we explicitlyre-defined (generalized) concepts f imputaThis to and solution in 44.4.2.-44.7.4. corresponded an tion, domination o extensionf the theory from zero-sumto constant-sumgames. Throughout what followed we emphasizedthat we were investigating a new theory, analogousto,but not identicalwith, the original oneof are Actually these two types of variations of our concepts not fundatype can be subsumedunder the first one. mentally different: Thesecond t Indeed,he new theory was introducedin orderto handle the problemof decompositionof the original one more effectively. This motive was w stressed throughout the heuristicconsiderations hich led to this generalization. In the analysis of imbeddingin 46.10., particularly in (46:K) and there, we establishedrigorously that the new theory can be sub(46:L) ordinated to the original one preciselyin this sense. o T Third: he concepts f imputation, domination and solutionswereagain 56.12. in re-defined(generalized) Chapter XI, specifically in 56.8.,

Our 65.1.1. mathematical considerations f the n-persongamebeginning o the definitions of 30.1.1. use of the concepts imputation, made of

65.TheExtension. Special Cases 66.1. Formulation of the Problem

11., a

55.8.2., 55.8.3.).

30.1.1.

56.11.,

587)))

588

EXTENSIONS THE CONCEPTS OF

to of This corresponded the final extension the theory to generalgames. that from thereon we were investigating a new theory We again emphasized ones. analogousto,but not identicalwith, the preceding This generalization was, however, fundamentally different from the two o It a widening of the theory and preceding nes: represented realconceptual not the changesreferred to above it was in evidence that while the conceptsof imputation, domination and solution varied some connectionmong them remained a (particularlyregardingextension), invariant. In orderto acquirea generalinsight into thesechanges and other analogousoneswhich may follow it is necessaryto find a precise formulation of this invariant connection. When this is done we can permit a generalityin all respectsnd reformulate the theory on that basis. complete it will appearhat this the e t By recalling instances numeratedin is invariant connection the processy which the conceptof a solution is b derived from thoseof imputation and domination. This is the condition in and Hence (or (30:5:c) the equivalent ones (30:5:a) (30:5:b)) the we reach perfect generality if we release notions of imputation and domination from all restrictions,but define the solutions in the way

66.1.2. Throughout

technical onvenience. a mere c

65.1.1.,

30.1.1.

Instead of imputations we considerthe elementsof an arbitrary but fixed domain (set)D. Insteadof domination we consideran arbitrary but fixed relation S betweenthe elements , y of D.l x Now a solution (in D for S) is a set V D which fulfills the following condition:

indicated. a with In accordance this programwe proceeds follows:

(65:1)

x which xSy holdsfor no element of V.2

of The elements

are preciselythoseelementsy

of D for

General 66.2. Remarks

It shouldbe noted that our presentconceptof solutionbears the same and relation to that oneof saturation analyzedin 30.3. particularlyin shouldbe comIn particular our (65:1) of as the original concept our presentS corresponding the in to with the fourth example pared of the (R there. It is especiallysignificant that in the searchfor negation with solutionsall difficulties connected the lackof symmetryof the relation and the remarks made in 30.3.6. 30.3.7. ariseagain. to considered, this effect apply oncemore.

senseindicated.

66.2. definitions providethe basisfor a more generalheory in the These t

30.1.1. 30.3.3.,

30.3.5.,

I.e.,

1x$y expresseshat this relation holds between the specific elements x and y. The t readershould recallthe discussionsat the beginning of 30.3.2. 1Thisis the equivalent of (30:5:c)30. in as promised.)))

11., .

THE EXTENSION. SPECIAL CASES 689 We shall seesubsequentlyhow thesedifficulties can be resolvedat least 1 in some specificcases. In orderto acquirea betterunderstandingof the entiresituation, we must consider some specializationsof the relation x&y. Indeed,n our i and t presentexpositionS is entirely unrestricted we cannotexpecto find any particularly deepresult while 8 remains in this generality. On the otherhand, the original concept f a solution,as defined in 30.1.1., o remains the most important applicationof $ and it seems ery difficult to discover v

any simpledistinguishingpropertiesof this particular relation. Therefore thereis no apparent way to introducespecialization, owever desirablethis h would be. We will nevertheless discussthreefrequently used schemesf specializao tion for relationsx&y and finally find a fourth onewhich possesses a certain limited applicabilityto our problem proper. In orderto carry this out, we need a few mathematical preparationswhich follow. the domain D) which and \" smaller.\"This f sharethe essentialeatures of the concepts greater in order ideashas receiveddetailedand careful considerations the matheof t matical literatureand there exists today rather generalagreemento the list effect that a complete of thesepropertiesruns as follows:
\" \"

65.3.1. first consideruch relationsxSy (with We s

66.3. Orderings, Transitivity,

Acyclicity

(65:A:a) (65:A:b)

ing relationsholds:)) x

Forany two x, y

of D oneand only one of the threefollow-

= t/,

2 imply zSz. xSy,ySz together We call a relation S with thesepropertiesa completeordering of D. orderingsareeasy to give and conform to ordinary Examplesof complete \" o intuition:The usual concept f \" greater for the set of all realnumbersor \" 8 o for any part of it. The concept f smaller\" under the same conditions. Even the points of the plane possesscomplete x e.g. orderings, this one:Sy o meansthat x must have a greater rdinate than y or the same but then one, than a x must have a greater bscissa y. 4 of complete 66.3.2. concept The orderingcan be weakenedconsiderably attention h still remains. This, too, as received so that a significant concept 5 and is of in the theory of utilities. in mathematical literature importance It obtains by weakening(65:A:a) bove,but retaining (65:A:b) nchanged. u a 1Cf.the results of 65.4.65.5., the less onesof 65.6.-6S.7. and t superficial 'The reader who substitutes the ordinary \"greater \" relation x > y for xSy in
verify that theseare indeed the basicproperties of \"greater.\" integers, or any interval, Without this last proviso, our S would fall under the next section. Cf. Birkhoff: LatticeTheory, loc.cit. Chapt. In this book orderings, partial Extenin orderings and similar topicsarediscussed the spirit of modern mathematics. to sive references literature are given there.))) will ( (65:A:a), 65:A:b), 1

t E.g.he 4

etc.

0.

I.

590

EXTENSIONS THE CONCEPTS OF

I.e.:
(65:B:a) (65:B:b)
relationsholds:))
x&y, y&z

For any

two

x, y
x

of D at most one of the three following

= y, xSy,y$x.

togetherimply xBz. 1 a We calla relation S with thesepropertiespartial ordering of D. Two in holds of D for which noneof the threerelationsenumerated (65:B:a) x,y is the ordering partial, this is a possibility)arecalled (since incomparable t (with respecto S). of Examples partial orderingsareeasy to give : The points of the plane, t that the ordinate of x is greaterhan that of y (cf.footnote 4 x&y meaning on p.589).We may also define that xSy means that the ordinateand the f o t abscissaf xareboth greaterhan their counterpartsor y. 2 Another good obtains in the domain of positive integers, $y meaning that x is x example
i of 65.3.3. two preceding The concepts orderingmaintained (65:A:b)n This the same while (65:A:a) as modified (weakened)to (65:B:a). w form, of t emphasizedthe importance (65:A:b),he property of transitivity.* We and will now undertake to weaken the combination of (65:B:a) (65:A:b) so that (65:A:b)s essentiallyaffected, too. i further, Notefirst that (65:B:a) is equivalent to thesetwo conditions:

divisibleby y excludingequality.

(65:C:a) (65:C:b)
%
'i

excludesthese three combinations:x = y, xSy; Indeed (65:B:a) x x&y> y& x- Now the first and the second merely two ways are y> y&
while (65:C:a), the third

Never xSx. Never xSy,ySx together.

of writing

We now

(65:D)

prove: Considerhe assertion: t

Never xiS^o,x&xi,
XQ,

is precisely(65:C:b).
m , x mSx

(A m )

Then we have: :D (65:B is equivalent to (A i), (A 2) together. (65 :a) :a) (65: (65:D:b) (65:B:a), Altogether all (A,),(A 2), (A 8), imply Ad (65:D:a): is (65:C:a) (4,) (65:C:b). and is Clearly (Ai) Proof: the relationsof (A m) in thereverse and applying (65:A:b) Writing order,
m

xi,

Xm~i

to belong D.

-i, where

x<>

= xm and

1Note that the word partial is used in the neutral sense,i.e.,completeordering is a a c special aseof the partial ones,since(65:A:a)mplies (65:B:a). i * Note that this is close a plausible type of partially ordered to utilities in the senseof the last remark of 3.7.2. ach imagined event may be affected with two numerical E both in characteristics, of which must be increased orderto produce a clearand reproducible preference. 1Some other important relations, not at all in the nature of an ordering, alsopossess
this

(65:B:a).

1 times gives x &xo.

m

As

xm = x , this means

XO&EO,

contradicting

property: E.g. quality, x e

y.)))

THE EXTENSION. PECIALCASES S

591

This result suggestsconsideringthe total aggregate all conditions of . They are impliedby (65:B:a), i b (Ai), (A 2), (A s), (65:A:b), .e. y a as a partial ordering, nd represent, will appear, further weakeningof this property.
We define accordingly:

(65:D:c)

relation S is acyclic if it
-

fulfills

all conditions(Ai), (Aj),

(A m )

(A.),

The readerwill understand why we call this acyclicity: If any should fail, therewould bea chain of relations))

which is a cycle,sinceits last element,m , coincidesith its first one,x x w We have already remarkedthat acyclicityis impliedby partial ordering a t (this of course,he contentof ; and hence fortiori by complete remains to show that it is actually a broader oncepthan t ordering. c that a relation can be acyclical without being an partial ordering,

is,

It

(65:D:b))

ordering(partial or complete).

i.e.,

Theseare examples the latterphenomenon: D be the set of all of Let and x$y the relation of immediate succession, .e., i positive integers, x = y + 1. Or, let D bethe setof all realnumbers,and xSt/ the relation ol than, but not by too much say by no more than 1 i.e., beinggreater the relationy + 1 ^ x > y. of anc this We conclude section observingthat our examples complete by
of partial orderings and of acyclical relationscould easily be multiplied to forbidsus to go into this here,but it may be suggested the readei Space t as a useful exercise.The referenceso the literaturein footnote 1on pag< 62 and footnote 5 on page589 can also beconsultedto advantage.

65.4.The Solutions :For a Symmetric Relation. For a CompleteOrdering Let of 65.4.1. us now discussthe schemes specializationreferredto at th( end of 65.2. In S First: is symmetricin the senseof 30.3.2. this caseit is expedien w to to go back the connectionith saturation, ointed out at the beginningo p 65.2.Owing to the symmetry of S it will provide all information abou solutionswhich we desire. S we Second: is a complete ordering. In this case define as usual:x is i convenient to indi maximum of D if no y with yBx exists. It is sometimes with a complete rderingby calling it an absolut cate the connection o in maximum of D. (Cf. this with the corresponding place the nextremark. 1 D has eitherno maximum or preciselyone. Clearly
Now we have:

(65:E)

V is a solutionif and only if it is a one-element consistin set, of the maximum of D. 1Proof: If X, y are both maxima of D, then ySx and x&y being excluded,(65: a A x necessitates
y.)))

592

EXTENSIONS THE CONCEPTS OF

a y in V. If zSy, then x cannot be in V, hence u in V with exists. The transitivity gives u&y which is impossible,sinceu, y are both in V. Sono x (in D!) xistswith xSy, 1and y must be a maximum of D. e So D has a maximum which must be unique (cf.above). Hence is a V one-element consistingof it. set, Sufficiency:Let a?o be the maximum of D, V = (XQ). Given a y (of D!), the validity of xy for no x of V amounts simply to the negation of x&y. Sincey&xo is excluded,his negation is equivalent to y = t So these y form the set V. Hence is a solution. V 66.4.2. there existsno solution y if D has no maximum, while a Thus solutionexists nd is unique if D has a maximum. a If D is finite, then the latteris certainly the case. This is intuitively and quite plausibleand also easy to prove. Forthe sake of completeness alsoto makethe parallelismwith the corresponding parts of the next remark a Consider
uBx
XQ.

empty either.

Let Proof: ecessity: V be a solution. SinceD is not empty, V is not N

more evident, we neverthelessgive the proof in full:

(65:F)
any
Xi

If D is finite, then it has a maximum. maximum. Choose

Theseresults show that both the existence uniquenessof V parallel and those of the maximum of D.
The 66.5. Solutions :For a Partial Ordering 66.5.1. S is a partial ordering. In this casewe take over literally Third: remark. It is somethe definition of a maximum of D from the preceding

in Z), then an # 2 with x&x\\, then an x$ with XsSx* etc. By x x the Zi, x 2, 8 , (65:A:b) m xn for m n, henceby (65:A:a) m T xn areall distinct from eachother,and so D is infinite.

t Proof:Assume the opposite, i.e.,hat D has no

>

. I.e.,

etc.,

times convenient to indicatethe connection with a partial orderingby in place calling it a relative maximum of D. (Cf. this with the corresponding the preceding remark. This contrast is quite useful, footnote 2 belownotwithstanding.) D may have no maximum, it may have one and it may have 2 several. Thus relative maxima are not necessarilyunique, while the 3 absoluteones are.
1A similar
8

in situation was already discussed 4.6.2. on w Theargument of footnote 1 on p. 591fails, sinceit depends (65:A:a) hich is now weakened to (65:B:a). take by E.g., for D the unit square in the plane and define in it a partial ordering the Then in either one of the two processes the two first examplesat the end of 65.3.2. maxima of D form its entire upper edge, or the upper and the right edgestogether, respectively. 1Thereaderis warned against mixing up our notion of a relative maximum with that onewhich occursin the theory of functions: Therea localmaximum is frequently called a relative one. Sincethe quantities involved there are numerical, hencecompletely ordered,this has nothing to do with our present considerations.)))

THE EXTENSION. SPECIALCASES 593 The questionof existence plays a different rolefor relative maxima also than for absoluteones. It will appearthat the decisive propertynow isthis : If y in D is not a maximum, then a maximum x with x$y (65:G) exists. For absolute maxima i.e., S is a completerdering (65:G) expresses if o l of preciselythe existence one. For relative maxima this need not be the i f of case, .e.or a partial orderingthe mere existence some(relative) maxima neednot imply (65:G). Examplesof this areeasy to give, but we will not pursue this matter further. Suffice it to say, that (65:G)will prove to be the proper extension of the existencef an absolute maximum (cf. the o of precedingemark) to the case relative maxima (cf.below). r
Now we have:

(65:H)

V is a solutionif and only if (65:G)is fulfilled (by D and S!) and V is the set of all (relative) maxima. Let Proof: ecessity: V be a solution. N If y is not in V, then an a; in V with xSy exists, hence is not a maximum. y So all maxima belongto V. If y is in Vi then the argument given in the proof of (65:E)in the preceding remark can be repeated literally, showing that y is a maximum. So V is preciselythe set of all maxima. a maximum, not in V, then an x in V, If y is not a maximum, with xSy exists, (65:G)is fulfilled (by D and S). so Sufficiency: Assume that (65:G) is fulfilled, and let V be the set of all maxima. s Forx, y in V, x&y is impossible, incey is a maximum. If y is not in V f in V, not a maximum, then by (65:G)an x which is a maximum, with x&y exists. So V is a solution by (65:1). to The reader hould verify how this result (65:H) specializes (65:E) of s remark when the orderingis complete. the preceding n Our result (65:H)showsthat thereexists o solutionV if D and Sdo not while a solution exists and is unique if this fulfill the condition (65:G), conditionis fulfilled. If 65.5.2.D is finite then the latteris certainly the case. We give the proof in full: (65:1) If D is finite, then it fulfills the condition (65:G). not A Proof: ssume the opposite, that D does fulfill (65:G). Call a y exceptional, if it is not a maximum and xSy holds for no maximum x. The failure of (65:G)means that exceptional exist. y Consideran exceptional Sinceit is not a maximum, an x with xy y. this exists. Sincey is exceptional, x is not a maximum. If a maximum u 1Proof:SinceD is not empty (65:G)implies the existence f a maximum. o Let be the maximum of D. Then for every y not a maximum, i.e., Conversely: of (65:A:a) (completeordering!)give y j x , the exclusion of j/8x and the validity

i.e.,

i.e.,

i.e.,

i.e.,

i.e.,

x<>

XoSy.)))

594
with

EXTENSIONS THE CONCEPTS OF

uSx existed, would give by (65:B:b) contradicting excepthe this uSy o characterf y. Hence such u exists, .e., too is exceptional. no i x

tional

I.e.:

If y is exceptional thereexists n exceptional with x&y. x a then an Now choose exceptional and exceptional with xaSxi, an excepx* x\\ tional #3 with x&xi etc., tc. By (65:B:b) xn for m > n, henceby xm e xm areall distinct from eachotherand (65:B:a) 5^ xn xi,#2, #3, so D is infinite. (Cf the last part of this argument with the proof of (65:F)in the precedits ing remark. Observe, that we could replace (65:A:a) the weaker by

(65:J)

. I.e.,

resultsshow that the existence a solution now doesnot correThese of of spond to the existence a maximum, but to the condition(65 :G). This is remark the quite remarkable considering concludingpart of the preceding in 65.4.2. corroborates earlier bservationthat in the presentcase It our o of partial ordering (65:G) is the proper substitute for the existencef a o
maximum.

(65:B:a).)

I The uniquenessof the solutionis even more remarkable. n the light of the last part of our preceding n remark, it would have seemedatural for this uniquenessto be connectedith that one of the maximum. But we see w now that the solution is unique, while the (relative) aximum neednot be, m
1 as was already mentioned.

In thosetwo cases e determinedhe necessaryand sufficient conditions t w for the existence a solutionand we alsofound that when they aresatisfied of it the solutionis unique. (Cf. (65:E)and (65:H).) Furthermore, was seen that when D is finite these conditionsare certainly satisfied. (Cf. (65:F)

65.6.1. S is acyclic.We know that this casecomprisesthe two Fourth: t that preceding nes,i.e., it is more generalhan both. o

66.6. Acyclicity

and Strict Acyclicity

with the (unique) set of all maxima. 1Since used the qualification \"absolute\" in the second, nd \"relative\" in the third a we unnecessary, however, remark, we should now employ another still weaker one. It seems to bring in such a terminological innovation at this occasion.)))

maximum, but

and (65:1).) In the acycliccasewe will find conditionswhich aresimilar to thesein w i many ways and in some respects e will gain deepernsights than before. It will benecessary,however, to vary our standpoint somewhatin the course of our discussion and our results will be subjectto certainlimitations. The in case a finite D will again besettled an exhaustiveand satisfactoryway. of o the It is again convenient to introduce concept f maxima, 2 and not only for D itself but alsofor its subsets. Sowe define :x is a maximum of E(s D) if x belongsto E and if no y in E with y&x exists. We denotethe set of all maxima of E by Em (z E). 1 (65 shows that the solution V is not connected with any particular (non-unique) :H)

THE EXTENSION. SPECIALCASES

$

595

w Our discussions ill show that it is of decisive importance whether D and possessthis property: E j* (for E D) impliesEm 7* Q. (65:K) m I.e.: non-empty subset of D possesses axima.1 Prima (65:K) facie Every does not appear to be related in any way to acyclicity, but there exists connection.Beforewe attack our proper objective, actually a very close w the role of solutionsin the presentcase, e investigate this connection. For 65.6.2. this purposewe drop all restrictionsconcerningD and S,even that of acyclicity. It is convenient to introducea propertywhich is a variation of the (A m)

them: (AJ

of (65:D) in

and 65.3.3., which

will turn

w out intrinsically connected ith

* * *

2 Never x X belongto D. , where x , Xi, x 2, , x2Sxi, sSx 2, We define, for reasonswhich shall appearsoon: A relation Sis strictly acyclicif it fulfills the condition (AJ. o We now clarify the relationshipof strict acyclicity i.e. f (A*) both to (65:K)and to acyclicity,by proving the five lemmas which follow. The are a r essential esults are (65:0)nd (65:P);65:L)-(65:N) preparatoryfor (

ir

(65:0).
(65:L)

Strictacyclicity impliesacyclicity. x Assumethat Sis not acyclic. Then thereexist , Xi, ' ' xm _i Proof: xmSx -i. Now extendthis m and xm = x in Z>, such that XiSx , x 2Sxi, , x by putting , xm _i to an infinite one x , Xi, x 2, sequence , Xi,
>

X Xl Xm_i

= Xm = X 2m = = Xm+i = X 2m +l = = X 2m
__i

'' ^ Xsm-1 = '

*

* *

, , -

Then clearly X&XQ, x&x^rr 8Sx , 2 etc., etc., and sostrict acyelicityfails. Acyclicity without strict acyclicity impliesthis: (65:M) * in D with this x Thereexists a sequence , Xi, x 2, (J5*)

x,

property:
if Sis a completeordering, this property (65:K) is of great importance in set Thosereaderswho are familiar with that theory will observe, that (65:K) is theory. o preciselythe fundamental concept f well ordering. (In this caseS must be interpreted as \"before\" instead of \"greater.\") For literature cf. A. Fraenkel, loc.cit. p. 195ff, and 299ff, and F.Hausdorff, loc.cit. p. 55ff , both in footnote 1 on p. 61; also E. Zermelo loc. cit. in footnote 2 on p. 269. It is remarkable that the same property plays a role in of connection with our concept solution for arbitrary relations. Themajor part of the with this property and considerations which make up the remainder of this chapter deals its consequences. considerablefurther this subject and its ramifications appearto deserve Actually study from the mathematical point of view. 1Thesequence; , *i, 3j, - should be infinite in the sensethat the indicesmust go a on ad infinitum, but the s themselves neednot all be different from eachother. 1Cf this with footnote 2 above,and the last part of this lemma.)))

1Even

596

OF EXTENSIONS THECONCEPTS

1 is necessary. are pairwisedifferent (B*)implies that the x , Xi, x 2, from eachother and therefore D must be infinite in this case. x in D, such S Proof: inceSis not strictlyacyclic,thereexist , Xi, x 2, that XiSxo, x 2Sxi,x 3Sx , ' ' ' Hence = g + 1is sufficient for xp x q . 2 p Now assume that xpSxg . We wish to prove the necessity of p > q. Assume the opposite:^ q. Now x p+iSx, Xp+jSXp+i, p p , x Sxg_i, 2 = q p + 1:It suffices (A XpSXg and theserelations contradict m ) with m to replace x , Xi, its xm and xm = x by our xp, xp+i, , , x and x p. This conflicts with the acyclicityof S. Thus all parts of (B*)areestablished. Now the consequences (J8*) : If the x , Xi, #2, ' ' were not pairwise of then x p = x q would occurfor some p > q. By (#*) x q+i&x q , hence distinct, Xg+iSXpj by (5*) this implies q + 1 > p, i.e. ^ p, but q < p. So the g x 2, arepairwisedistinctand therefore D must be infinite. Xo, Xi, (65:N) impliesthis: Non-acyclicity Forsomem(= 1, , 2 ) we have: There existXQ, Xi, (B*) , x m _i and xm = XQ in D with this property: ForXpSx g , p = q + 1is necessaryand sufficient.8 S Proof: inceS is not acyclic,thereexistx , Xi, , #m _i and x m = x in D such that XI&XQ, x&Xi, xm Sx _i. Choose such a system with its m , m(= 1,2, ) as small as possible. Clearlyp = q + 1 is sufficient for x pSxq . We wish to prove that it is necessarytoo. Assume therefore xpSxq but p ^ q + 1. Now a cyclicalrearrangement the Xo, Xi, of , xw_i, x m = x does not affect their properties we can apply this so as to make xp the last and t t element i.e.,o carry p into m. I.e.,here is no loss of generality in assumingp = m. Now p 7* q + 1, .e., 7* m 1. We can also assume i q that q 7* m, since q = m could be replaced q = 0. So q g m 2. by After these preparations we can replace Xi, XQ, , xm _i, x m = x by 4 x , Xi, This replaces , xg , xm = XQ without affecting their properties. m by g + 1,which is < m, and this contradicts the assumed minimum property of m. Thus all parts of (J5*) areestablished. 65.6.3. Summingup:

For x pSx7 , p = q + 1 is

sufficient and p

>q

fl

-i

fl

(65:0) (65:0:a)

Acyclicity is equivalent to the negation of all (B*),(fij),

1In connection with this result cf.also 1Theseare preciselyq p relations, hence tl^ey do not appearif p 1Observe of the interrelatedness of the that the characterization completein (BJi), but not in (B*). This will be of importance below.
4

65.8.3.

q.
XD,

x\\,

x^

is

I.e.,

omit)))

THE EXTENSION. SPECIAL CASES

597

(65:O:b) (65:O:c)

(*?),

lent to it for the finite

Strict acyclicity is equivalent to the negation of all (fif), and.of(B*). Strictacyclicity impliesacyclicity for all D but it is equiva-

D.

The (65:O:a): condition is necessarysince(5*) contradicts hence acyclicity. Thecondition is sufficient by (65:N). (-A), contradicts since The Ad (65:0:b): condition is necessary non-acyclicity strict acyclicity by (65:L), and (J5*) contradicts (A ) hencestrict acyclity. The condition is sufficient sincethe negation of strict acyclicity of permits the applicationof (65:M)in case acyclicity,and the application
A Proof: d
00

of

finite the

(65:L). If D is reverseimplication and hencethe equivalence results from the last remark in (65:M). w Finally we establishthe connectionith (65:K): (65:P)

above (65:0:a) in caseof non-acyclicity. The Ad (65:0:c): forward implication was stated in

#2i,

Strictacyclicityis the exact quivalent of the property (65:K), Sowe see: e t we expecto be fundamental. Acyclicity and strictacyclicity are which to closelyrelated eachother. The particular role of the finite D begins make itself felt: For finite D the two above conceptsare already to
equivalent.
T 65.7. he Solutions
We 65.7.1. now
turn

obtainsand XI&XQ,

(65:K)is equivalent to strictacyclicity. A Proof:Necessity: ssume that S is not strictly acyclic. Choose Then#= (so,Zi, z 2, z 2, 2 ) x&x\\, x 8Sx , inDwithziSzo, XQ, Xi, no maxima.So (65:K)fails. is D and 7* , and it possesses clearly Sufficiency: Assume that (65:K) fails. Choosea nonempty EzD an maxima. l Choose XQ in E. XQ is not a maximum in E,so choose without a an Xi in E with XI&XQ. Xi is not a maximum in E, so choosen x* in E with in E, hencein D, etc. In this way a sequenceo, Xi, #2, Z etc.,

' '

X&KI, zs&r*,

This contradictsstrictacyclicity.

:For an Acyclic Relation

The objective: investigation
of the

D for S. It is at this point that it will appear, hy we attribute w (65:K)will turn out to the property (65:K)sucha fundamental importance: of w to be quite intimately connectedith the existence preciselyone solution. that thereexists reciselyonesolution (in D for S) We beginby showing p ourselvesto finite if (65:K) is fulfilled. In proving this we will restrict the solution can even be obtained by an explicit sets D, in which case construction. This constructionis effected by finite induction. The finiteness of D is not really necessary,but for an infinite setD the constructionin 2 questionwould be more complicated. 1Thereadershould comparethis proof with that one of (65 in 65.4.2. :F) solutionsin
* It would

to our main

to be necessary make use of more advanced set-theoretical concepts

(cf.)))

598

OFTHE CONCEPTS EXTENSIONS

Sincewe must assume (65:K), this means by (65:P)that D must be from indistinguishable strictly acyclic.SinceD is finite, this is by (65:O:c) for . ordinaryacyclicity Soit doesnot matter the moment, whether we state that we requireacyclicity or strict acyclicity of D. It is nevertheless that s appropriate to remember we are using (65:K), i.e.,trictacyclicity, the and that the assumptionof finiteness, which obliterates distinction in question, couldbe removed. F We repeat: orthe remainderof this paragraphfiniteness of D is assumed strictacyclicity. and the property (65:K) i.e., acyclicity,i.e., Let us now carry out the inductive constructionreferred to. This will will bedone first and the announcedproperties beestablishedafterwards. three sets B, C (all D) as We define for every i = 1,2, 3, follows:A i = Z>. If for an i (= 1, ,3, - ) A* is already known, then 2 J5 Bi, d and A{+\\ obtain in this way: Bi = A? i.e., is the setof thosey in A< for which x&y for no x in Ai. C is the setof thosey in A* for which x&y for x some in B. Finally A+i = A Bi C.

4,

Now we prove:

(65:Q) (65:R)

Bi, d aredisjunct. I by Proof: mmediate their definitions.

Ai^O implies A^ c Ai. 1 2 Proof:Ai^Q implies Bi = A? 9* Q by (65:K), hence

Ai+i

= Ai
with

- -d c
Bi
Ai

A>.

(65:S)

an Thereexists t

= 0.
A*\"=>

the finiteness of D. Let i be the smallesti of (65:S),then (65:T)

D=
AI
=>

Proof:Otherwise by (65:R) D = A\\^>

At

A*^>

, contradicting

4,o_i 3 A, o = . o R Proo/: estatement f (65:R) and (65:S).

^ Az ^
=>

(65:U)

Bi t
,
JB

''

the sum D.

Bi9-i>

C\\,

C, _i, are disjunct sets, with

BI Ci,
u

B Proof: y the

definition of Ai+\\ we have JS u C< = A, A+i. _i u C o_i, arepairwisedisjunct and their sum is
A,

Hence

A^

=D

= D.
B\\
y

Combining this
Thesematters

with

(65:Q) shows that

C\\ 9

, J3 t

-i, C _i, i.e.,

to

o the referencesf footnote 2 on p. 269and footnote induction or some equivalent technique.

will

1 on

i p. 595),n particular of transfinite

1The point is that we have and not merely 1Thisis the only but decisive! usewe make of (65 :K).)))

be consideredlsewhere. e

THE EXTENSION. SPECIAL CASES BI, , B,_i, Ci, , C f _i, are pairwisedisjunct, and o sum is also D.

599 that their

65.7.2. now put We (65:2) (65:3)

Vo

= B!U

uB,-i. t
f

Then (65:U)gives

D-Vo= C!U- - u C -i.

If V is a solution (in D for S),then V
V

Now we prove:

(65:V)

Assumethe opposite and considerhe smallest i for which Bi V fails t to be true. Let z be an element f this B not in V. Then ySz for some o Considerthe smallest y y in V. z is a maximum in A t hence is not in A,. k for which y is not in A*. Then g i and as y is in D = AI, so k j& 1. Put j = t 1,hen 1 j < i. y is in A; but not in A,-+i = A k, henceit is in B, u C, = Aj A/+I. 2; is in Bi A, c A,-. So if y were in B,, ySz would imply that z is in C,. This is not so, incez is in B. Hence is in C,. s y Now necessarily x in B, with x&y exists. Since is in V, this excludes an y x from V. Thus Bj V cannot hold. As j < i, this contradictsthe assumedminimum property of i.
fc fc

W Proof: e beginby showing that Bi

= Vo. for all i = 1,

,*

-i.

So we see:

(65:4)

B<

for all

i = 1,

to

- 1.
V by

If y is in then an x in Bi with xSy cannot be in V. (65:4), y So we see:

C,

exists. Sincethis x is in
,t

(65:5)

d s -V
Vo

for all

i = 1,

- 1.
V must

w (65:5) ith Comparing(65:4), w coincide ith Vo, as asserted.

(65:2), above showsthat (65:3)

(65:W)

is a solution (in D for S).

in))

W Proof: e prove this in two steps: Assumethe opposite: to If x, y belong Vo then xSj/ is excluded: x, y

x, y belongto
in

sincey is in B,. hencexSyis impossible. in Thus we have a contradiction any

B,

A,

A.

f say x to Bi and y to B. If i j, then y is is in B, so xSy impliesthat y is in d. This is not so, If t > j, then x is in B< S A c A/, y is a maximum in A,,

Vo f

event.)))

600

OF EXTENSIONS THE CONCEPTS

in If y is not in Vo> then xBy for somex in Vo y is in Voi hence some in Hence&y for an x in 5<, and this x is in consequence Vo. x This completes proof. the :V) Combining(65 and (65:W) we can state: Thereexists ne and only one solution (in D for S),the Vo of o (65:X)

'

d.

t Let us reconsiderhe last three remarks, still retaining for a t moment the assumptionof finiteness, in ordero avoid further complications. It is conspicuousthat they all yielded the sameresult, although under o varying assumptions. In each casewe proved the existencef a unique t but the hypothesiswas first complete ordering,hen partial ordersolution, it was weakened at ing, and finally (ordinary or strict) acyclicity every step. with the last i This beingso,t is natural to ask whether we have reached remark the limit of this weakening or whether acyclicitycouldbereplaced of without impairing the existence a uniquesolution. by even less It must be admitted, that this line of investigation takes us away from in of the theory of games.Indeed, that theory the existence solutionswas but that therecouldbe no question of primary importance, we have learned of uniqueness. with Nevertheless,sincewe now have someresults on existence uniquet we will continue to study this case. We will seelater, hat it has even ness, indirectlya certainbearingon the theory of games. (Cf. 67.) W of In the senseoutlined we shouldask therefore this: hich properties that the relation S arenecessaryand sufficient in order thereexista unique solution? It is easy to see,however, that this questionis not likelyto have the a simpleand satisfactoryanswer. Indeed, solution(in D for S) discloses with S). The acyclicalcase little about the structure of D (together only but o suited to judgethis, sinceit is somewhatcomplicated, the cases f is less or partial orderingmake the point quite clear. Therethe solution complete to a is only related the maxima of D and it does not expresst all what the o of the otherelements f D are. properties It is not difficult to eliminate this objection.Consider a set E . D instead of D. The relationS in D is also a relationin E and if it was a or orderingor a partial ordering (ordinarily or strictly)acyclicin D, complete 1 the samein E. Hence result (65:X) implies that in our then it will be a there exists unique solution (for S). Now thesesolutions, every ESiD when formed for all E Z), tellmuch more about the structureof D. It is o o o bestto restrict urselvesagain to the cases f (completer partial) ordering. Clearly the knowledgeof the maxima of E for all setsE D gives a very information about thestructureof D (together ith S). w detailed 1I.e., leastthe same it can happen that a partial ordering in D is completein E at relation in D is an ordering in or that an acyclic

65.8.1.

above. (65:2) 66.8. Uniqueness of the

Solutions,

Acyclicity

and Strict Acyclicity

i.e.,

E.)))

THE EXTENSION. SPECIAL CASES

601

65.8.2. we arrive at the following question:Which propertiesof Thus the relation S arenecessaryand sufficient in order f that thereexist or each E D a unique solution (in E for S)? We can show that hereacyclicity and strict acyclicityarethe significant concepts, i although the subject s not completelyexhausted. Thetwo lemmas which follow contain what we can assert n this matter. o
(65:Y)
be replacedy acyclicity,owing to (65:0:c). b o Forinfinite D this is dependentupon the extensionf (65:X) to infinite sets(cf.the beginningof 65.7.1.) I Proof:f D is (ordinarily or strictly) acyclic,then the sameis true of all E S D (cf. above). Now all assertionsof our lemma becomebvious. o In orderthat there existfor each E&D a unique solution (65:Z) (in E for S) acyclicityis necessary.
may

In orderthat there existfor each E&D a unique solution (in E for S),strictacyclicity is sufficient. For finite D this follows from (65:X), and strictacyclicity

= 1,2, put E = (XQ,

m

yields the validity of a (B*), Proof; If D is not acyclic,then (65:O:a) , z m _i and x m = XQ and , in (65:N). Form its X , z , S and (B*)describesin E Then EsiD , xm Xi, x 2, completely. Let us considerthe solution V in E (for S). Considersuch a solution V. If Xi is in V, then Xi+i is not, since a^+iSx,-. If xt is not in V then thereexists a y in V with y = X, with z,&r. x This means = + I,1 so y = Xi+\\, and hence t+\\ is in V- Sowe see:

-i).

j i
Xi

?/,i.e.,

(65:6)
(65:7)
XQ

is in V

if

and only

if Xi+i

is not.

of Iteration (65:6)gives: If k is even, then XQ is in V if and only if Xk is. If k is odd, then XQ is in V if and only if Xk is not.

= ar m , (65:7)involves a contradictionif m is odd. Hencethere As exists o solutionin E (for S) if m is odd. If m is even, then (65:7)implies n that V is eitherthe setof all x with an even or the setof all Xk with an odd k. And it is easy to verify that both these setsare indeed solutions in E (for S).
k
fc

So we have: (65:8) The number of solutions in E = (XQ, x\\, (with the ZQ, Xi, , Zm-i from (B*))is 2 or
whether m is even or odd.

, a;m

to according

-i) for S

a Consequentlythereis in no case unique solutionin this E( D). of The and (65:Z)we see: existence a unique solution Combining(65:Y) For for characterised finite sets: these E for S) for all E D is completely (in
1If
f

m, then

replaceit

by

0.)))

602
it

OF EXTENSIONS THECONCEPTS

necessaryand strictacyclicityis sufficient. i The 66.8.3. gap which existsn this casecan only bebridgedby a study of the acyclic,but not strictly acyclic(infinite) setsD and their subsets E. we (65:0:b) see that such a D satisfies (B*). By comparing (65:0:a), and put D* = (X , Xi #2, Formits x , XL x* ). This is also we of acyclicbut not strictly acyclic,hence may study it in place D. this: Thus the questionhas become fulfills (B*). Will then (65:9) Assume that D* = (x , Si, every E D* possessa unique solution (in E for S)? c The answer to (65:9)annot be given immediately, because(B*) x the describes relation xSyin D* i.e.pSxq only incompletely. The cor) was answeredin the proof 2, respondingquestion for (J3*) (m = 1, of (65:Z)in the negative, but (B*)describedthe relation xSy in its set i.e. XpSx q completely. Thus the answer to (65:9) requires an exhaustive analysis of all possibleforms of the relation x p&xq which fulfill (B*). The 1 problemappears to be one of considerable difficulty.
Q

the same thing. For infinite sets D we can only say that acyclicity is

to is equivalent to acyclicity,i.e., strictacyclicity,which in this caseis

*,)

pointed out before,no direct earing on the theory of games. b As regardsstrictacyclicity,it suffices to emphasizeits equivalenceto that (65:K) (by (65:P)), and to remember in the theory of games even D no itself (the set of all imputations) possesses maxima (i.e., undominated

Our 65.9.1. above results on acyclicityand on strict acyclicityhave, as

66.9. Application

to Games Discretenessnd a

Continuity

t Ordinary acyclicity too is violated, e.g., already in the essential hree 3 person game. Neverthelessthere weresituations that arose during the mathematical of o w discussion certaingames, here the concept f acyclicitycouldhave been in applied. Thesesituationsareto beregarded the spirit of the first remark and the examples of 65.1.1. they arespecifically referred to there. among Thus in the trianglesT discussed 47.5. we have an acyclicalconcept f in 1. o 4 as the inspectionof figures 76,77 shows. Furtherin the setCt domination, in described 55.8.2. is an acyclicconcept f domination as the criterion there o 6 makes apparent. (55:Z) Finally, in the market discussedin 64. there is an acyclic conceptof of domination in the case monopoly or monoposony, s the discussion the at a
1It lieson the boundary line of combinatorics and set theory, and seemsto deserve further attention. * This holds for all essential ames. Cf (31 :M)in g 1Thereaderis invited to verify this, on the diagram of Figure 54. It is easyto ascertainthat (BJ)holds (and (A m) fails) for all m 4 Heredomination ordin ate. a 1Here domination implies having a greater n-component, and from this acyclicity implies having greater obviously follows.)))

2 elements).

e.g.,

31.2.3. 3.

GENERALIZATIONOF THECONCEPT UTILITY OF end of and 64.2.2. in

603

1 there particular (64:12), (64:13) shows. We may the accentuate concludingremarkmadethereby observingthat an intrinsic i connections to be surmisedbetweenthe monopolisticituationsin the ecos nomic sphereand the mathematical concept f the acyclicityof domination. o It is very remarkable,therefore, that in all thesecases articularly extenp n sive families of solutionswere found to exist. Indeed, ot only numerical but parameters, even highly undeterminedcurves or functions enteredinto and in a thosesolutions. Forthis cf. 47.5.5. Fig.81. the first instance, nd the in the second one. In the third instance can only we fifth remark in 55.12. c The refer to the mathematical discussionof a special ase: three-person market monopoly versus duopoly which was analyzed in 62.3., and 62.4.

The 65.9.2. great number of solutionsin the acyclicsituationsreferred to above may seem natural, if the infiniteness of these D (the set of the is imputationsunder consideration) emphasized.After all, it was only for finite sets D that acyclicity implied uniqueness of the solution, for the the infinite onesstrictacyclicitybecame crucial concept. (Cf the last part And in n of 65.8., particular 65.8.2.) all theseexamples of course, ot are, strictly acyclic,as can be verified with ease. The situation is nevertheless paradoxical, for the following reason: in of o Modifications the concept f utility, which will be considered 67.1.2. can be applied in such a manner as to make the sets in question finite. mentionedwill have unique solutions. Now these Then the acyclicalgames finite modifications can be made to resemble arbitrarily closelyto the origithe nal, unmodified games. Hence original acyclicgameswith many solucan tions (infinite D!) be approximatedarbitrarily closelyby the modified acyclicalgames with unique solutions (finite D!). How can the unique \" solutionsbe \" arbitrarily close approximationsof the non-unique ones?

63.4.

in s This paradoxical ituation will bedescribed detail in 67. Theanalysis are going to give there will clarify this lackof continuity and a present n opportunityfor someinterpretationsof a certaininterest. which we

66.Generalizationf the Conceptof Utility o 66.1. Generalization. The

of The Two Phases the Theoretical reatment T

In the past sections have generalized concept f a soluo we the tion based on a relation S, which takesthe role of domination in a most extensiveway. These generalizationsshould be used in our theory as o follows Our concepts f imputation, domination and solutionsrestupon the more fundamental one of utility. Now if we desire vary the formalism to

66.1.1. :

1Herethe domination implies having acyclicity obviously follows. If neither monopoly nor monopsony then apply instead of this caseacyclicity doesnot prevail.)))

a greater 1- 1*-) component, (or

exist,
if i.e.,with

and from this

(64:11) (64:10),

eod. (64:13) (64:12),

the notations loc.cit.I, m > is easy to verify that in

It

1,

604

OF EXTENSIONS THECONCEPTS

difficulties must beovercome in order carry out this program. to Our theory of games divides clearly into two distinct phases: Thefirst onecomprising treatment of the zero-sumtwo-persongame and the one leadingto the definition of its value, the second dealingwith the zerosum n-persongame, asedon the characteristic b function, as defined with the help of the values of the two-persongames. We pointed out above, how eachof these two phases makes use of specificproperties of the utility if modiconcept. Therefore, any of thesepropertiesareto be generalized, fied or abandoned,we must study the effect of such a change eachphase. in It is therefore indicatedto analyze thesetwo phasesseparately.

used to describe latter,we can try to renderthesevariations adequately the of by appropriategeneralizations the former concepts. Of course, e do not wish to carry out generalizationsor their own sake, w f but there are certain modifications which would make our theory more realistic.Specifically: e have treatedthe conceptof utility in a rather W narrow and dogmaticway. We have not only assumedthat it is numerical for which a tolerablygoodcase be made (cf.3.3. 3.5.)but also and can that it is substitutableand unrestrictedlytransferablebetweenthe various The in r We players (cf. 2.1.1.). proceeded this way for technical easons: numerical utilities were needed the theory of the zero-sumtwo-person for of values had to play game particularly because the rolethat expectation in it. Thesubstitutabilityand transf erability werenecessary the theory for of the zero-sumn-person gamein orderto produceimputations that are vectors with numerical components and characteristic functions with numerical values. All these necessities themselvesimplicitly in present o every subsequentconstructionbuilt upon the preceding nes and so in fine in our theory of the general -persongame. n Thus a modification of our conceptof utility in the nature of a generalization appears desirable, at the same time it is clear that definite but

66.1.2.

66.2. Discussion the of

First

Phase

utilities.

I Specifically: t is difficult to seehow a definite value can be assignedto a unlessit is possiblefor eachplayer to decide all cases hich of the in w game, various situations that may arise preferablefrom his point of view. This is means that individual preference ust define a completerdering of the m o

66.2.1. difficulties of generalizing the first phase are very serious. The The theory of the zero-sumtwo-persongame as expoundedin ChapterIII makesull useof the numerical character utility. f of

Next the operationof combining utilities with numerical probabilities cannotbe dispensedwith either. We have seenthat the rules of the game i may explicitlyrequiresuch operations,f they provide for chancemoves. But even when this is not the case,he theory of ChapterIIIleads general t in totheuseof mixedstrategies the same with effect. (Cf.17.))))

GENERALIZATIONOF THE CONCEPT UTILITY OF

605

Now it is well known that the completelyordered character utilities of the numerical one. But we have seenin that complete ordering in conjunction with the possibility of combining utilities with numerical probabilitiesimpliesthe numerical character utility. of Thus we have at presentno way to adscribea zero-sum two-person game a value unlessnumerical utilities areavailable. In the n-person game the characteristicunction is defined with the f help of the value in various (auxiliary) zero-sumtwo-persongames. Our reduction of the general -persongamesto the zero-sum ones,in addition, n made use of the transferability of utilities from one player to another.

does imply not

3.5.

can ]? 3C* in 56.2.2. hardly begiven fc-i function in any other meaning. Thus the definition of the characteristic an n-persongame is technicallytied up with the numerical nature of utility in a way from which we cannot at present escape. The values v(S) of the characteristic function of such a game are the values of the correspondingsets coalitions of players Henceour conclusioncan also be stated in this way: Our general ethod to adscribea m value to every possiblecoalition of players is essentiallydependent upon n the numerical nature of utility and we areat present ot able to remedythis. We have pointed out before that the hypothesisof the numerical nature of utility is not as special it is generally believedto be. (Cf the discusas d sion of 3.) Besides, can avoid all conceptual ifficulties by referring we it to our considerations a strictly monetary economy. Nevertheless would be more satisfactory,if we could free our theory of theselimitations and that of it must be conceded the possibility doingthis has not beenestablished thus far. In 66.2.2. spite of this inadequacy in general,there are many games function is never serious. where the difficulty of defining the characteristic Thus the examples of and of 57.3. were such that the characteristic function could be determineddirectly, without realneed for the elaborate of considerations the theory of the zero-sumtwo-persongames. It is true were examples that these synthesized in orderto obtain a known, prefunction hencethe easewith which they can be assigned characteristic

c Indeed, onstructionslike 3C i = n+

S.

26.1.

function causesno difficulty whatever throughout the the characteristic theory of simple gamesof Chapter X. Again the various markets con1

i handled in this respects scarcelysurprising. However,there existother Thus of instances the same phenomenon which areof a certain significance:

all had in functions which were easily and sidered 61.2.-64.2. characteristic

directly obtained.

i In thesecasest would beeasy to replace numerical utilitiesby more the We proposeto takethem up on another occasion. c general oncepts. 1Thesegames were defined
by stating which are the winning of the characteristic function.))) coalitions and this

implied an implicit determination

606

EXTENSIONS THE CONCEPTS OF

SecondPhase If the characteristic function is taken for granted,we can pasi
of 66.3. Discussion the

treatmentof the secondphase. To begin with, when we renounce transferability of utilities,as wel the their numerical character, like as when we renounce concepts zero-sumo] it constant-sumgamesarenot immediatelydefined. Hence is bestto dea directly with generalgames. a th< Letus therefore consider general -person n game. Sincewe possess theory of ChapterXI, we may forget its origin in the theory of zero-sun of ( gamesand try to extendit directly to the case more general non-numeri utilities. cal,non-transferable) The imputations

Herethe necessityfor a numerical utility can be entirely circumvented do not propose describehis in complete etail,sincethe entin to t d to not subjectdoes seem be mature yet for a final mathematical formaliza tion. Indeed,the first phase is obstructedby unsolved difficulties ai t described above. Besides, appearso besome there justification for believ unified form of the theory, of which we can at presentse< ing that a more only the outlines,might lead us to the desired goal. We shall therefore give only some generalindications relative to th<
We

to the second phase.

66.3.1.

a =

>

will still be vectors, but their components i, a , an may not be num bers. It must be noted, that if we give up the numerical character o that eachparticipant i(= 1, utility, it is best to concede , n) has *

{{a!,

an }}

the domain of individual utilities % of his own. I.e., Od, , <u n will ii a be different. In this setupthe component must belongto if*. general It mustbenoted that even if all utilitiesarenumerical i.e.,if<Vi, , u, w coincideith eachother and with the set of all realnumbers we may stil omit the assumption of transferability. Also we may considerthe cas< b where transferability exists, ut subjectto certain restrictions.Indeed o in an examplef this will be discussed detailin 67. Now o a 66.3.2. the restrictions n thesecomponents must be considered are of two kinds:Firstthe domain of all imputations was definedii They 56.8.2. by
<

(66:1)
(66:2)

a,

v((t))
a<

for

=
,

1,- - - , n,
n)).
1

in 56.12.)))

*~i we defined domination with the help of a concept f effectivit} o Second on based 1We prefer to use (56:10) instead of the alternatively possible here o (56:25) f (56:I:b

v((l, -

GENERALIZATIONOF THECONCEPT UTILITY 607 OF

(66:3)
which
All

ai
tinS

v(S),

is (30:3) 30.1.1. of theseinequalitiesbelongto a common type: A certainset T is given T T (T = (t) in (66:1),= (1, - - , n) = / in (66:2), = S in (66:3))and *. the imputation a is requiredto place set coalition T into a position the which is at leastas good (in (66:1)) at most as good (in (66:2) in or and (66:3))as that one statedby v(!T). The position of the coalition T i.e., compositeposition of all its the in participants is expressed all theseinequalitiesby the sum of their components: ] ak . Fornon-numerical utilitiesthe domainsa/!, l^may
Jbin

be different from each other, and besides, there may existno addition in them thus rendering formations like % ak senseless. ut even if the B
kin

utilities are numerical, the use of ^ ak in the above contextis clearly

kin

t equivalent to assumingunrestricted transferability. Indeed,he position of a coalition can be described the sum of the amounts given to its memby to bers without any reference the individual amounts themselves only when thosemembersare able to distribute that sum among themselvesin any way in which they all agree, if thereareno physicalobstructionsto In general,herefore, we shall have to forego the useof ]? a . Instead, t fcin T we must introduce the domain of utilities for the composite person, conthis of all membersof a given coalition T. Denote domain by ^(T). sisting Clearly,^((k)) is the same thing as ^ . ^(T) must be obtainableby some o processf synthesis from the ^ of all k in T. It is not at all difficult to but devisethe proper mathematical procedureequiredfor this process, we r
k k k
7

transfers.

i.e.,

proposeto discussit on another occasion. of The aggregate the ak) k in T , as well as the value v(T) of the characo function must be elements f this system. The inequalities(66:1), teristic in refer then to preferences that systemof utilities. (66:2), (66:3) 66.4.Desirability of Unifying the Two Phases w I 66.4.n the hope that the reader ill not find the analysisof 66.3. too unification of the two we now indicatein which way the desired sketchy, for. Our theory of the zero-sumtwo-persongame phases may be looked was really based on the same general rinciplesas the subsequentstructure p of imputations,domination and solutionsfor zero-sumn-persongamesand t even for generaln-persongames. Specifically, he decisivediscussionof in of the inter-relatedness various strategies a zero-sumtwo-persongame the carried in 14.5., 17.9.i.e., analysisof the conceptof a good))) out 17.8.,

608

OF EXTENSIONS THECONCEPTS

with W To apply this specificallyto our experience games: e had to t dividethe difficultiesin ordero overcome them and to considersuccessively zero-sum twozero-sum two-persongames with strict determinateness, s zero-sumn-persongames, with games general trict determinateness, person generaln-person games. However, all these steps but two were finally mergedinto the generaltheory: Only the zero-sumtwo-persongameand o the general -persongame remained. Our insistence n the characteristic n function amounts to insisting that for the zero-sumtwo-persongamean intermediaryresult beobtainedwhich is much sharper than that onewhich 1 as we accepted satisfactory for the n-person game. Of course,we were ableto fulfill this requirement in the caseof a numerical, nrestrictedly u transferable utility. However,this may be different when these assumpr tions concerningtility arediscarded.And it seemsather plausiblethat u to our difficulty with the n-person gamemay be ascribed our continued on insistence this special etup for the zero-sumtwo-persongame. Our s forces us to insist in this respect, ut this b present technicalprocedure neverthelessbe misplaced. insistence ay m

extraneousifficulty. d

strategy is in many ways analogous to our use of dominations of * imputations. that the weaknessof our presenttheory liesin the Now it would seem T in necessityto proceed two stages: o producea solution of the zero-sum two-persongame first and then, by using this solution, to define a characfunction in orderto be able to producea solution of the general teristic function. Generalexperience n-pergongame,basedon the characteristic indicates such a two stage that in mathematicsand in the physicalsciences procedurewith an intermediary halt representedin our caseby the function has two essential spects. In the early stagesof characteristic a the investigation it may be advantageous,sinceit divides the difficulties. it In the laterstages,however, when full conceptual enerality is desired, g can bea handicap. The requirementof producinga sharply defined quanthe function tity in the middleof our procedure in our case characteristic might be an unnecessarytechnicality,saddling the main problem with an

for thesedifficulties.

treatmentfor theentiretheory of the n-person game without artificial halt at the zero-sumtwo-persongameand it now appears) the (as function may therefore in fine prove to be the remedy the characteristic
A unified

67. Discussionf an Example o

67.1. Descriptionof the

Example

1For the zero-sum two-person game we obtained a unique value imputation. For the generaln-persongame (as well as for the zero-sum one)we had only a usually not unique solution, and eventhe individual solution is a setof imputations!)))

modifications do not represent a and transferability are modified. These

in We o 67.1.1. shall now discussan example which the concepts f utility

i.e.,

DISCUSSION AN EXAMPLE OF

609

t particularly significant broadeningof our standpoint with respecto those i concepts.The interest of our examples rather that it permits of an applicationof our results concerningcyclicityand thereby yields conclua sionswhich throw somenew light on the subject iscussed the end of 65.9. d at i o Specifically,t is hoped that procedures f this kind will provide a more mathematical approachto the phenomenon of bargaining. adequate 67.1.2. modification to be consideredis this: e assume that The W or its monetary equivalent is made up of indivisibleunits. I.e. utility we do not questionits numerical character require but that its value be in restricted appropriateunits an integer. Thus transferstoo arenecessarily to integers, ut we do not restricthem further. We proposeto use the b t characteristic function as before, but also with integervalues. The conof cepts domination and solution,after this, areunaltered. If this standpoint is applied to generalone and two-persongames,no significant changes occur; everything remainsessentiallyas in our old It is therefore unnecessaryto enterupon the detailsof thesecases. theory. The three-person ame,on the other hand, offers some new features, even g in its old zero-sumform. It gives rise to some quite peculiar difficulties which appear be of considerableinterest,but are not yet sufficiently to analyzed. We therefore prefer to postpone this discussionfor a later

i.e.,

occasion.

a consistingof one seller nd two buyers. 67.1.3. obtained in our previousanalysis of this casevarious soluWe tions, dependingon whether we assumedthat only one (individual) transor actioncouldtake place several,also dependingon the relative strength and in of the two buyers. Thesesolutionswere described (62:C)of 62.5.2. s i that the general olution in (63:E) of 63.5.In all thesecasest appeared was made up of two parts:(62:18) (62:20), (63:30)) (or (62:21), and (62:19) Our (or (62:23), (63:31)). discussionthere showedthat the parts of the to correspond the situation where the two buyersarecompeting type (62:18) with eachother,while the parts of the type (62:19) correspondto the situation wherethey have formed a coalition againstthe seller. Thetype (62 :18) with a part was uniquely determinedand in essential greement the ordinary, common senseeconomic ideas on the subject. The type (62:19) on part, the other hand, was defined with the help of some highly arbitrary functhe as tional connections.Theseexpressed, we saw in 62.6.2., various possibilitiesto set up a rule of division between the allied buyers for any profit obtained. I.e.they constituted their standard of behavior within additional their coalition.Our presentdiscussionis going to providesome mechanism. information concerning functioning of this part of the social the to In orderto do this effectively, it is reasonable eliminatefrom our w to all thoseelements hich do not contribute this aspect. I.e. problem
we)))

a of This excludesn exhaustive discussion the general hree-personame g t the new setup. We shall, however, analyze a special asewhich bears c m directly upon the nature of bargaining. This is the three-person arket,
in

610
wish to

EXTENSIONS THE CONCEPTS OF get rid of the type

omitted altogether footnote 1on p.571) when v w in the notations (cf. This meansthat only one (indivisible)ransaction can take place t and that the two buyersareof exactlyequal strength. Thesolution is then a of given by (62:20) nd (62:19)62.5., ((62:20) eingsuperfluous, cf. above), b or equivalently by Figure99. So we assume v = w in the schemeof 62.1.2. can simplify the We situation further, without any significant loss,by putting the \"alternative usefor the seller\" = 0. In this way the (62:2)-(62:4) u of defining the characteristic function, simplify to

62.6.1. 62.5.2., that

(62:18) of the solution. We know from part this part is of the smallestsize and indeedcouldbe

c loc. it.

62.1.2.,

(67:1)

v((l)) = v((2)) = v((3)) = 0, v((l,2)) = v((l,3)) = w, v((2,3))= 0, = w. I v((l,2,3))

f

The imputations arenow defined by))

a2, aa
with

(67:2:a) (67:2:b)
given w

ttl ttl

Szl

now assume all these quantities to be integers i.e.he t and all permissible i, 2, 3 of (67:2:a), (67:2:b). We define domination as before, i.e.ollowing 56.11.1. which means f that we repeathe definitions of 30.1.1. t literally. It is therefore necessary to determine the characterof the sets = (1,2,3) respecto their rolein defining a domination. It is with t to show that the sets easy

67.1.4. We

+ 2+

0,

a,^ 0,a,^ 0, 3 ^ w.
<* 1

S = (1,2), (1,3)
are certainly necessary,and
1Note that
we 2 all others certainly unnecessary. Thus we amounts to using former condition is the original one (cf., and the equivalence of the two, madeuse of in fails in the setup to be used now. It will be seenin the first remark of that the ^ and the in (67:2:b) must produce different results, but that this divergence nevertheless fits into the general t instead of in (67:2:b) ould lead to results which picture. Besides,he use of w differ only in detailsof secondaryimportance from thosewhich we are going to obtain. J The conditions for and certainly unnecessary setswere derived certainly necessary in and reconsidered in Since standpoint has changed again (cf.above, our and particularly footnote it would be necessary reconsider to thesethings oncemore. It seems simpler to take them up de now; in Owing to with above, and the condition is certainly unnecessary. This disposesof (30:3) (2), (3), every Againv(S) (2,3). above give ai + a2 w (67:2:b) hence ai + at w are certainly necessary. And since(31 in is clearly still valid, :C) this renders certainly unnecessary.)))

taken in the discussion of

are using (67:2:b) with and not with This is the standpoint in In in (66:2) 66.3.2. the terminology of (56:I:b) 56.12., it and (56:10) not (56:25). The reason for this procedure that the is

-.

56.12.,

e.g., 6.8.2.), 5

67.2.3.

31.1.,

1),

59.3.2.

(67:2:a)

(67:2), 8 - (1,2), (1,3) 8 - (1,2,3)

- -

S (1), v((l,2)),

30.1.1.,S (67:1), S - v((l,3)), 31.1.3.

DISCUSSION AN EXAMPLE OF canusethe definition of domination with S = (1,2), (1,3).I.e.: a ^

means that
>

611

ft

(67:3:a)
and

ai > fa a2 > fa
or

(67:3:b)

a3 > fa-

Thus domination implies (67:3:a), therefore it is clearly acyclical. and the correspondingdiscussion of 65.9.)Furthermore, he domain t (Cf. the (67:2:a),1 of the a is finite, because componentsa\\, a2, as must (67:2:b) be integers. Now we can apply (65:X) of 65.7.2.: There existsone and only one id. solutionVo which is characterized the formulae (65:2), (65:3), by

67.2. he Solution T In 67.2.1. orderto

must

and

Its Interpretation
we (65:2), of 65.7.2., (65:3)

do this for J5i, Ci. BIis the set of those a which cannot bedominated. To dominate a in we must increase i and a2 or a8 without violating (67:2:a), (67:2:b) a w Theseincreases reby 1at least, hile the otherone of a2, as may 67.1.3. bedecreased far as to 0. Hencea can be dominated,if either as

the Let determine sets5,,C, defined at the beginningof 65.7.1. us

apply the formulae

(ai + 1) + (*s + 1) g w
So BIis defined by (67:4)

or

(ai + 1) + (a,+ 1)
8

w.

(a!+ 1) + (ai + l)>w, (a!+ 1) + (a + 1) > w.

a < 2, a2 < 2,i.e. 2, as = 0, 1. Now the (67:2:a), (67:2:b), following ai = w,
w

this (67:2:b) impliesas By (67:2:a), in conjunction with (67:4) gives,

possibilities:
(67:A)
(67:B)))

a2 = as = 0,
or

1;

(67:C)
by

a2 = , as = a2 = as = 1,

ai = w

- 2.
y

of Ci is the set of those a which are dominated by elements B\\ i.e. thosein (67:A)-(67:C).It is easy to verify that thesearecharacterized
1This was, of course, ot n
the

casein

the original continuum

setup.)))

612
by

EXTENSIONS THECONCEPTS OF

(67:D)

which is better suited to this particular case. This usean inductive process processoesas follows: g t Considerhe a with

Now 67.2.2. it is betterto deviate from the scheme (65:2), of of (65:3) 65.7.2. is, not to continue by determiningJ52, Cz, J5a, Cs, ; that , but to

|
2

(,
(3

= <M

or

}, = 0j

a, S w - 2.

(67:E)

or

= 0.

( They make up exactly (67:A), (67:B),67:D). We know that among these Vo containspreciselythe (67: (67 :B). The remaining a arethosewith A), (67:F)

a,,a

1;

henceundominatedby (67: (67:B). So we form Vo by taking (67: A), A), a o the (67:B) outside of (67:F), nd repeating processf finding a solution in
(67:F).

is that
w

in The ( (67:2:b)67.1.3. only difference Compare67:F) with (67:2:a), w b 2, s areincreasedy 1. Hence must be treatedas if it were

2.

Thus Vo now contains further

(67:G) (67:H)
t of and we must repeathe process finding a solutionin

(67:1)

a2, a8 ^ 2.

of a Repetition this procedure ssigns i = tu 2 = a = 2, 4, w 5; (67:J) = 3,aa = 2 ( 2 or 1, ai = 10 5; (67:K) { a2 = 2, a8 = 3 ] ( t o us to Vo> and requires to repeathe processf finding a solutionin 2, 2 ^ 3, (67:L)
\\

e etc., tc.
(67:M)

as This set can becharacterized follows:

ThusVoConsistsof(67:A),(67:B),(67:G),(67:H),(67:J),(6
ai

- 0,1,

10

;))

DISCUSSION AN EXAMPLE OF
//3T (67:N)))
XT\\

613))

[ a*

=
W

1 OLl)

if

a\\

is even;))

1
2
or)

(67:0)))

- 1

ai)

2 2

or)

'-)

if w

ai is odd.))

+1
2))

01)

The 67.2.3. results (67:M)-(67:O) t suggestheseremarks: First: he values of a\\ + a2 + as in this solution are w and w 1. T Thus we cannot replace ^ in (67:2:b)67.1.3.=, the resultstated the of by in (56:I:b) 56:12no longer true. The maximum social enefit is not b is of

o necessarilyobtained and this appears as the directconsequencef the existence an indivisibleunit of utility. 1 o This Second: \"discrete\"tility scaleconvergestoward our usual, conu oo . (Cf. the corresponding tinuous one,if w considerationsconcerning The and discrete continuous \"hands\"in Poker,in 19.12.) difference of more and w, mentioned above, is at most 1. So it becomes i + 2+ s more insignificant as w - oo , i.e.his aspect f the situation tends to t o and what it was in the continuous case. Third: 2, a differ from eachother by at most 1. Sothis differencetoo when we approach the continuous tends to insignificance as w > oo. I.e., casethe solutiontends to looklike this:)) (67:P) (67 :Q)))
As
with 3
=))

^
w))

w,

this pointedout in the first part of 67.1.3.,solutionmust be compared solutions case areindeed similar,but our solution coversonly one special of (62:19): functions of a\\ mentionedtherecoincide with The monotonic decreasing
in using (62:19)62.5.1., the values u = 0,v = w. The two

eachotherand with ^ as the Thosefunctions describe, discussedin 62.6.2., rule of division upon which the two buyers agreedwhen forming their coalition (which is In by expressed (62:19)). the continuouscasethis rule was highly arbi-

c w trary. But now, in the discrete ase, e find that it is completelydetermined the two buyers must betreated exactly alike! What is the meaning of this symmetry? Are the other distribution o the rules -i.e. otherchoices f the functions in (62:19)eally impossible r
in the
1

\"

discrete\" case?))
with

Cf. this

footnote

3 on p. 513.)))

614

OF EXTENSIONS THECONCEPTS
Utility

the symmetry (betweenthe two buyers), but conserve,the \"discreteness.\" This will be done by altering the setup of 67.1. so far that we assign in the indivisibleunit of utility for the buyer 2 a value different from that one for the buyer 3. Specifically: et us prescribe the values of i, a2 L that must beintegers, hile thoseof as must be even integers. Apart from this, w remainsunaltered. everything in 67.1. of We now carry out the equivalent of the considerations 67.2. Accordingly, we beginby determiningthe

G 67.3. eneralization :Different Discrete Scales In 67.3.1.orderto answerthe above questions,we shall try

to destroy

setsB\\, Ci of 65.7.

which cannot be dominated. To dominate a in we must increase and a2 or a8 without violating a\\ Theseincreases are (for i, a2) or 2 (for a8) at least,while the

Biis the setof those a

>

>

67.1.3.

(67:2:b) (67:2:a),
>

other one of a2, a8 may be decreaseds far as to 0. Hencea can be a dominated if either (on + 1) + (a2 + 1) w or (ai + 1) + (a* + 2) g w. So BIis defined by (67:5) (ai + 1) + (at + 1) > w,

(i + 1) + (as + 2) > w.
ai = w, ai = w ai = w

possibilities:
(67:R) (67:S) (67:T)

a this (67:2:b) impliesas < 2, a2 < 3,i.e. 2 = 0, 1,2, a8 = 0. By (67:2:a), the Now (67:5)gives, in conjunction with (67:2:a), (67:2:b), following a2 = 0, a2 = 1, a2 = 2,
as = 0, as = 0, as = 0,

w-1; - 1, - 2; - 2.
w

o Ci is the setof those a , which aredominatedby elements f BI i.e.y b It is easy to verify that thesearecharacterized thosein (67:R)-(67:T). by

a2 = 0, ai g w 2; a2 = 1, ai w 3. Insteadof determining Now 67.3.2. we repeatthe variant of 67.2.2.: we use a different inductive process. C2, B8, Cs, , 5s, t Considerhe a with a2 = 0,1. (67:W)
(67:U) (67:V)
4 ( They make up exactly (67:R), (67:8),67:U),(67:V). We know that :S). The remaining a among these,Vo contains preciselythe (67:R), (67

arethosewith
(67:X)

a2 ^
s 1, inceit must

2;
be even.)))

1Note that a* cannot be

DISCUSSION AN EXAMPLE OF

615

henceundominated by (67:R), (67:S). So we form Vo by taking (67:R), the of (67:S)outside of (67:X), and repeating process finding a solution in (67 :X). in The Compare(67:X) with (67:2:a), (67:2:b) 67.1.3. only difference 1 is that a2 is increased 2. Hence must be treated if it werew 2. w as by
Thus Vo now containsfurther
(67:Y)

(67:Z)

a2 = 2, 2 = 3,

<* 8

a8 = 0, = 0,

ai = w ai = w

- 2, - 3; - 3, - 4;
w w

and we must repeathe process finding a solution in of t 2 ^ 4. (67:A') Repetition of this proceduressigns a
(67:B') (67 :C')
2 = 4, a2 = 5,

a3 = 0, a, = w 4, w 5; as = 0, ai = w 5, w 6; t to Vo and requiresus to repeathe process finding a solution in of

(67:D')
2

- - -

^ 6,

etc. etc.,
(67 :E') (67:F') (67 :G')

Thus Vo consists of (67:R), (67:8),67:Y), (67:Z),(67:B'),67:C'), ( ( a . This setcan be characterizeds follows:

ai = 0,1, = w ai,w 2

the one (excluding second when

- 1- ai
w\\

a8 = 0.

ct\\

= w)\\

identically a = 0. Approaching continuity, tends to looklike this:

The 67.3.3. results (67:E')-(67:G') theseremarks: suggest the First and second: Concerning sum ai + c*2 + and its relation to w we may repeatiterally the corresponding parts of 67.2.3. l Third:Herethings are altogetherdifferent from 67.2.3. e have W
a

f i.e.or w

* oo

the solution

(67:H')
(67 J )

(67:1') ;

a2 = w as = 0.

^ -

w,

i,

in as made in the cor(62:19) T we seethat the situation is now this: hemonorespondingpart of 67.2.3., the rule of division betweenthe which describe tonic functions of (62:19), two allied buyers (cf.loc. it.) reagain completelydetermined but this c a in the time we find (instead of the equal treatmentthey received 67.2.3.) entireadvantage going to buyer

Repeatingthe comparisonto

62.5.1.,

2!

1Note the (67:F)there.)))

difference

between this

and

the

corresponding

following step in 67.2.2.

616
We must now

OF EXTENSIONS THE CONCEPTS

this compare result with

and interpretthe entirephenomenon.

the corresponding in one

67.2.3.

67.4.Conclusions Concerning Bargaining 67.3.3. T 67.4. he conclusionfrom the results of 67.2.3., is evident. In the the former case two buyers had exactlyequal powersof discernment i.e.qual units of utility and the rule of distributionwas found to treat e them equally. In the lattercase buyer 2 had a betterpowerof discernment 2 unit of utility was half of 3's and in the rule of divithan buyer 3 i.e. 's sion the advantage went in its entirety to buyer 2. Clearly,if their abilities had beenreversed,the result would have been also. We may also say: The advantage in the rule of division betweenalliedbuyers is equallydivided if and they have equally fine utility scales, goesentirely to the one with the
finer
utility

how the ability of discernment of a the fineness of his subjectiveutility scale has a deterplayer specifically 2 mining influence on his positionin bargaining with an ally. It is therefore that to be expected problemsof this type can only be settled completely c when the psychological onditionsreferred to are properly and systematiof cally taken into account. The considerations the last paragraphmay be a first indicationof the appropriatemathematical approach. 1It is possible consider more subtle arrangements: We can assign to 2 and to on to ranges of varying density. In this case we have still a unique solution for the same as reasons before. The correlation of 2, when plotted in the 2, s plane will be a above:Symmetric in 2, s, parallel to the combination of the three types described bisectrixof the two coordinateaxes;parallel to the a-axis; parallel to the a 2-axis. to It is actually possible bring about any desired combination of theseelements by Any desired choosing the ranges of 2 and a a appropriately. shapeof the curve can be approximated arbitrarily well in this manner. In this way the original generality of the continuous caseis recovered. to We do not propose considerthis matter, and various related ones, in detail here. 1This occurs, f course, nly when the theory with continuous utilities allows several o o different rules of division between allies which is plainly the casewhere bargaining plays a
<**

1 scale otherwise. case This is true in the discrete where each participant has a definite the and scale the rule of division (i.e. solution)is uniquely determined. utility of In the continuous casethe \" fineness\" the utility scale undefined and is in the rule of division can be chosen many different ways, as we have seen.

So we observe for the first time

i.e.

role.)))

APPENDIX. THE AXIOMATIC TREATMENT OF UTILITY

A.I.Formulation of the Problem

t prove in this Appendix, hat the axiomsof utility enumera number up to a linear transformation. 1 More We of precisely: will prove that those axiomsimply the existence at least one mapping (actually, of course,of infinitely many) of the utilities on with the properties :a), (3:1 and we numbersin the senseof (3:1 :b); will also prove that any two such mappingsare linear transforms of each other, connected a relation (3:6). by Before we undertake this analysis of the axioms (3:A)-(3:C) of them may be useful in dispellingpossible two further remarks concerning

ated in

W A.I.I. e will

make 3.6.1. utility

3.5.1.,

i.e.

3.6.1.,

misunderstandings. A.I. The first remark is 2.

do not axiomatize the relation =, but interpret it as true to The alternative procedure, axiomatize = also,would be matheidentity. t matically perfectly sound,but sois our procedureoo. The two procedures are trivially equivalent and represent nly variants in taste. The practice in o the relevant mathematical and logical literature is not uniform and we have therefore adhered to the simplerprocedure. A The secondremark is this: s pointed out at the beginningof we are using the symbol > both for the \"natural\" relation u > v affecting utilities u, v and for the numerical relation p > a affecting numbers p, <r; both for the \" natural\" alsowe are usingthe symbola -f (1 a) operation an + (1 a)v affecting utilities u, v and for the numerical operation ap + (1 a)cr affecting numbers p, a (a is a number in either t c case). One might objecthat this practicean lead to misunderstandings and to confusion; however, it does not, providedthat one keeps always in whether the quantities involved are utilities (u, v, w) or numbers evidence (a, 0,7, , p, <r). This identification of the designationsfor relations a and operations in the two cases (\"natural\" nd numerical) has a certain simplicity and facilitates keeping track of the \"natural\" and numerical in pairs of analogs. Forthese reasonsit is fairly generallyaccepted similar a situations in the mathematical literature, nd we proposeto makeuse of it. The deductionswhich follow in A.2. are rather lengthy and may be somewhat tiring for the mathematically untrained reader. From the that viewpoint thereis the further objection, purely technical-mathematical d they cannot be considered eep the ideas that underly the deductionsare
We

this: heseaxioms, specifically the group T o t characterizehe concept f completeordering, based on the relations (3:A),

>, <.

3.5.1.,

A.1.3.
I.e.ithout w

fixing

a iero or a unit of utility.

617)))

618

THE AXIOMATICTREATMENT OF UTILITY

e had quite simple,but unfortunately the technical xecution to be somewhat to voluminous in order be complete.Possiblya shorterexpositionmight be found later. At any rate,we are now forced to use the esthetically not quite satisfactory mode of expositionwhich follows in A. 2. A.2. Derivation from the Axioms
The of (3:A)-(3:C) 3.6.1. deductionwill be broken up into severalsuccessive steps and it will be carried out in this sectionand the four next ones. The
final

A.

W 2.1. e

now

to proceed

carry out our deductions from the axioms

result will be stated in (A:V), (A:W).

If u

(A:A)

to

:a) By (3 :C this can be (1- p)u +

Thus (1

< v, then a < implies (1 a)u + av < (1 P)u + PV. (applied to u, v, Proof:Clearly a = 7/3 with < 7 < 1. By (3:B:a) in place of u v, a) u < (1 flu + pv, and hence (3:B:b) by (applied (1 flu + PV, u, 7 in place of u, v, a) -(1 flu + PV > 7(0 P)u + pv) + (1 y)u.

ft

ft

written
pv

> y(pv

+ (1 - flu) + (1 - y)u. VQ UQ VQ,

Now by (3:C:b) (appliedto v, u, 7, 0,a = yp in placeof u, v, a, 0,7 = a/3) the right hand side is av + (1 <*)u, hence by (3:C:a) (I a)u + av.

a)u + av

(A:B)

< (1 flu + pv, as desired. Given two fixed UQ, with < considerthe mapping a w = (1 a)u + av .
>
Q Q

next.

This is a one-to-onend monotone mapping of the interval a < a < 1on part of the interval u < w < v .1 Proof:The mapping is on part of the interval UQ < w < UQ < w coincides ith (3:B:a) w (appliedto u , v , 1 a in place of u, v, a), w < v coincidesith (3:B:b) w (appliedto , wo, in place of u, v, a). One-to-oneharacter: c Followsfrom the monotony, which we establish
VQ\\
t>

Monotonecharacter: Coincideswith (A:A).

The mapping of (A:B)actually maps the on all the w of UQ < w < V
Q

. A < < i that Proof: ssume that this were not so, .e. some with were omitted. Then for all a in < a < 1 (1 a)u + av 7* , i.e. (1 a)u + av ^ w . According to whether we have < or >, let a
WQ UQ WQ VQ
Q Q

(A:C)

a of < a < I
u>

Itwill

appearin (A:C),that

this

part is actually the whole

interval

<w <

t>o-)))

DERIVATIONFROM THE AXIOMS

619

I I belongto class or II. Thus the classes,II, which areclearly mutually the exclusive,exhaust together interval < a < 1. Now we observe:
MO,

First: lass I is not empty. This is immediate by (3:B:c) C (appliedto of 1 -~ a i*1 place u, w, v, a). Class Second: II is not empty. This is immediateby (3:B:d) (applied of to Wo, u , a in place u, w, v, a). Third:If a is in I and is II, then a < 0. Indeed,sinceI and II are disjunct, necessarilya ^ ft. Hencethe only alternative would be a > ft. But then the monotony of the mapping of (A:B)would imply, that since a is in I, ft toomust be in I but is in II. Hence a < ft is possible. only thesethree propertiesof I,II, there must existan a with Considering < o < 1 which separateshem, i.e.uch that all a of I have a g c* , s t and all a of II have a ^ ao. Now a itself must belongto I or to II. We distinguishaccordingly: First:a Then (1 ao)uo 4- a v < w ^ Alsow < Applying inl. y in placeof u, w, v, y) we obtain a y (3:B:c) (1 a )wo + o^o, (with with < y < 1 and y((l o)^o+ ao^o)+ (1 7)^0 < w , i.e.y (3:C:b) b (with t*o, 7, 1 ao, 1 a = 7(1 a ) in place of u, v, a, 0,7 = aft) a Hence = 1 7(1 a ) belongsto I. However (1 a)w + av < 1 (1 a ) = a , although we shouldhave a ^ a . a> Second:a in II. Then (1 a )wo + > W . Also < a )u + aoVo, Wo, 7 in placeof u, w, v, a) (with (1 Applying (3:B:d) we obtain a 7 with < 7 < 1and 7((1 a )w + vo) + (1 7)^0 > w , i.e. (3:C:a) v + (1 o)w ) + (1 7)^0 > w , henceby (3:C:b) by y(a (with = 7o in place u, v, a, ft, y = aft) av + (1 a)u > of Wo, 7, o, i.e.by (3:C:a)1 a)uo + avo > Hence a = belongs to II. ( Howevera < ao, although we shouldhave a ^ ao.
Wo,
t>o,

Vo,

ft

ft

VQ.

MO, tfo,

Vo,

WQ.

<*ot>o

UQ

WQ.

MO,

<*

Vo,

WQ,

WQ.

ya<>

Thus we obtain a contradictionin each case. Therefore the original a assumptionis impossible, nd the desiredproperty is established. It A.2.2. is worth while to stop for a moment at this point. (A:B)and (A:C)have effected a one-to-one mappingof the utility interval w < w < VQ (w , v fixed with U < v , otherwisearbitrary!) on the numerical interval < a < 1. This is clearlythe first step towardsestablishinga numerical o representationf utilities. However,the result is still significantly incomto pletein several respects.Theseseem be the major limitations: w First: he numerical representation as obtained for a utility interval T WQ < w < v only, not for all utilities w simultaneously. Nor is it clear, how the mappingswhich go with different pairs WQ, #o fit together. o Second: numerical representationf (A:B),(A:C)has not yet been The is N correlatedith our requirements (3:l:a), w (3:l:b). ow (3:l:a) clearly
Q Q

class

a perfectly rigorous inference. effecting the introduction of irrational numbers, the theorem concerning the Dedekind cut. Detailscan be found in C.Carathtodory texts on real function theory or on the foundations of analysis. Cf loc.cit. footnote 1 on p. 343. Cf.there p. Axiom VII. Our class should be substituted for the set (a ) mentioned there. Theset ( A ] mentioned there then contains our
t w Indeed,it coincides ith one of the classicalheorems
fairly

1This is intuitively

plausible. It is, furthermore,

11,

.e.g. I

II.)))

620

THE AXIOMATICTREATMENT OF UTILITY

the satisfied:It is just another way of expressing monotony that is secured remainsto be established. by (A:B). Howeverthe validity of (3:l:b) all these requirements jointly. The procedurewill We will fulfill b by suggested the first remark,ut in the process primarily follow a course the requirements of the secondremark and the appropriate uniqueness results will alsobe established. We beginby proving a group of lemmata which is more in the spirit of the secondremark and of the uniquenessinquiry; however it is basic in o t to order make progressowardsthe objectives f the first remark too. Let u , V be as above:U , v fixed, UQ < V . For all w in (A:D)
Q

/(w) = /* ,v ,(w) as follows:

the interval

UQ

<w <

t>

define the numerical function

(i) /(wo) = (ii) /(t>o) = w for UQ (iii) f(w) for w ?* w , fo, to a which corresponds w in the

0.

1.

< <1

i.e.

< < y , is the numbera in

(A:E)

The mapping
w

senseof (A:B),(A:C).

>/(w)
u

has the following properties: (i') It is monotone.

(ii') For

/((I - 0)Uo+ = (iii') For < < and ^ /((I - 0> + = 10W)

< < 1 and w ^

1
w

0f(w).

VQ

0ii>)

(A:F)

of this interval are mapped on numbers > 0, < on numbers > than the map of UQ and < than the map of v Hencewe have monotony throughout u<> g w g v Ad Forw = t/ The statementis /((I 0)u + 0t; ) = 0,and this coincidesith the definition in (A:B)(with in place a). w of
All w

and either(ii') or (iii'),is identical with the mappingof (A:D) ( Proof: A:D)is a definition; we must prove (A:E)and (A:F). Ad (A:E):Ad (i'): or < w < the mappingis monotone by (A:B). F the bers,which possesses properties (ii) (i),
UQ VQ

mapping of all

w with

UQ

+ 0/(uOg w g v on any set of num-

(ii'):
7*
t>

. :

1,i.e.

Forw

u i.e. < w < :Put/(w) = a, i.e.y (A:B) b

v

= (1 a)u + at/o.
Q

0,a. in place of u, v a, 0,and using (3:C:a)))) Hence (A:B)/((I 0)w + 0^) = 0a = 0/(^), as desired. by Ad (iii'): w = u : The statementis /((I 0)t> + 0^o) = 1 0, For and this coincidesith the definition in (A:B)(with 1 w of in place a and

Then by (3:C:b) (with

,u

Qy

using (3:C:a)).)))

DERIVATION FROM THE AXIOMS

621 :C :a))

Forw

7* w ,

i.e. < w < :Putf(w) = a, i.e.y (A:B) b

UQ

vQ

Then by (3 :C (with :b)

. ,by ,AW hence (A:B)

as desired.
Ad

(1-

ft)v

+ ftw

= (1

/((I

0)t>

+ ftw)

+ ft((l + = 1 - 0(1- a) = 1 t>

, ,

0,1- a

= (1

avg. a)w in place w, v, a, ft, and using (3 of

ft)v<>

o)w

avo)

ft

+ 0a = 1 = 0(1

a)w))

+ 0/(i,),

Considera mapping (A:F):

(A:l) w-^Mw) with (i), (ii) and either(ii') or (iii').The mapping (A:2) w->f(w) w on is a one-to-one a I, henceit can be mapping of inverted: a->^(). (A:3) i w Now combine(A:l) with (A:3), .e. ith the inverse of (A:2):
UQ VQ

(A:4)

a^A^(a))=

,().

Sinceboth (A:l) and (A:2) fulfill (i), (ii),we obtain for (A:4) ^(0) = 0, ^(1) = 1. (A:5) If (A:l) fulfills (ii') or (iii'),then, as (A:2)fulfills both (ii') and (iii'), we
obtain for (A:4) or

(A:6) (A:7) ^(1

v(fta)

= M),

j8

+ j8a) = 1

j8

+ j8^(a).
<p(l

Now putting a = 1in (A:6) and using (A:5)gives ^(j3) = j8, (A:8) and putting a = in (A:7) and using (A:5) gives
it < ft < 1. However (A:5) extends

to ft = 0, 1 too, to all ft with the t |S Considering definition of v?(ex) by (A:3),(A:4), he general the of (A:8)expresses identity of (A:l) and (A:2),which is prevalidity ciselywhat we wanted to prove. Let w , Vo be as above:u , v fixed, u < V Let also two (A:G) < fto be given. Forall w in the interval fixed a , /?o with w g VQ define the numerical function g(w) = fif\";;S|(^) UQ ^

ft Replacing by 1 ft gives again (A:8). Thus (A:8)is valid at any rate, (ii'), (iii') restrictit to the

0) =

ft.

1.

i.e. .

ft with

as follows:

g(w)

o)/(w) + o, to ) according (A:D)).)))

= (j8o

622

THE AXIOMATICTREATMENT OF UTILITY

We

note:
This mapping
w

(i)
(ii)

(A:H)

->g(w)

has the following properties: (i') It is monotone.

(ii') For g((l (iii') For g((l

A

< < I and w 7* ft)u + ftw) = (1 0) + ftg(w). < ft < 1and w ^ + ftw) = (1 /3)0o + 00(w).
ft
Q

ft)v<>

UQ

VQ

(A:I)

mappingof all w with w ^ t^ ^ VQ on any setof numbers the which possesses properties (ii) and either(ii') or (iii'), (i), w is identical ith the mappingof (A:G). betweenfunctions U Proof: singthe correspondence
gi(w)

or equivalently))

= (0o a )/i(t0)+ <*o,

of (for fi(w), g\\(w), and also for f(w), g(w)\\ the statements (A:G)-(A:I) o g of over into the statements (A:D)-(A:F). Hence(A:G)-(A:I)ollow from f

(A:D)-(A:F). (A:J)
(u in
Q

Assuming

0v) = (1 P)g(u)+ pg(v) g u < v g t; ) with u = U , v ^ U is equivalent to (ii') (A:I), and with u 7* , v = v it is equivalent to (iii') in
Q Q

g((l - ftu +

(i) ; (ii) in

(A:G),the equation

VQ

form, with the necessaryuniquenessproperties.We can now begin to fit the various mappings
w

(A:I). A of Proof: d (ii'): Put u , w, ft in place u, v, ft. Ad (iii'): Put w, v , 1 ft in place u, v, ft. of A.2.3. (A:G)-(A:J)he mapping of a utility interval ^ w g In t on a numerical interval a ^ a ^ /3o has beengiven its technicallyadequate
UQ VQ

g(w)

together.
(A:K)
To = i

= gl'tyw)
with

Considergf^;Jj and a WQ

gW

v<>.

Put))

Then g^\\(w) coincides ith gj,'ij(w)in the latter'sdomain w coincides))) ^ w ^ WQ (if 1^0 T* u , i.e. < WQ\\ and 0jj(u>)
UQ
Q

UQ

DERIVATIONFROM THE AXIOMS

with

623
v

i.e. e < W
MO,

0ij5j(w>) in
t><>).

the

latter'sdomain W

< w <

(if

WQ

7*

t>

A the Proof: d 0\"j;i|(w): gl\\(w) possesses properties(i), (ii') (of (A:G), t0 , because for a , To, coincidewith those for a , Ho, w , v they (A:H)) (sincethey involve only the lower end a , MO). It also possesses (of (ii) * 70. Henceit follows from (A:G))for a , 70, w, u>o, because ff'fywo)

*w
d****

(A:I) that
Ad

g\"j;Jj

fulfills

within

u%

5w

t0

a unique characterization of

To, Ho, u>o, VQ

properties (ii),(iii') (of (A:G),(A:H))for because they coincidewith those for , Ho, ^ , (since they involve only the upper end Ho Vg). It also possesses (of (A:G))for (i) it To, , Wo, y o, because g*$(wg) TO- Hence followsfrom (A:I)that g*fy of fulfills within W & w v* & unique characterization *',!*. Consider a gCj^J and two u\\ v\\ with u ^ u\\ < vi ^ v . (A:L) Then ^(w) coincides Put en ^;;J;(wO, Hi
gl**,\\\\
t>o
Q

the 0C;;J;possesses

with

g\"

lt l

(w) in

the

' lattersdomain ui

- d$(i).
}

v\\.

( with A Proof: pply first (A:K)to g?***and gjjjj i.e. w , o, o, /3o, t>i, /3i in placeof u , v , a , /8o, ^ , %; note that /8i = jC$(i)) this shows that w ^ v\\. Apply domain w ^*j;Jj(w;) coincidesith 0Cj;J|(w>)in the latter's
UQ

with next (A:K)to g\"j;Jj and g^\\ (i.e.

= a , Ho, wo, TO;note that = gl*(ui) fif^f;(wO) this showsthat gl*!\\(w), and hence also gl**\\(w), coincideswith gl^\\(w) in the latter's domain
U\\

, i, o, Hi,

t^i,

i in

place of u , t; ,

(A:L) has to be combinedwith a secondline of reasoning. At this w point we also assume that two w*, t>* with u* < v* have been chosen; e will considerhem as fixed from now on until we get to (A:V) and (A:W). t
We now

g Vi.

(A:M)

prove If
(i)

wo

u*

< v* ^

i>

, then

one there exists and

only one

03\",Jj(t0)

rf(u) - 0, (ii) rf(O * Idenotethis g*'(w) y /i^ (w). b

Vf

such that

We

Forvariable a , Ho (A:G) gives ^(w) = (Ho a )/(w) + o- Hencethe = 1, + o)/(w*) 4- o above (i), (ii) mean that (Ho o)/(u*) o = 0,(Ho
1 and these two equations determine ao, Ho uniquely. Hencethe desired

Form Proo/: the/(w) = /^.(w)of (A:D).

As u*

< v*, BO/(U*)< /(*).

\"

~)))

/(**)

/(*)

'))

624

THE AXIOMATICTREATMENT OF UTILITY

and exists is unique.)) If
with
u\\
(M>)

latter'sdomain u\\ g w g VL Proof:Put <*! = A r ,t (wi), 0i = ft.,,>i).Then, by (A:L), A ti ,t (iiO ' w coincidesith flC ;? ^) in the latters domain u\\ w ^ v\\. Applying this = u* and to w = v* gives fl|;J|(M*)= A , (M*) = and ^j;Jj(v*)= to w = *,.,(*>*)1- Hence by (A:M), gll*\\(w) = A , (w). Consequently domain HI ^ w ^ v\\. w A , (w) coincidesith A , (w) in the latter's We can now establish the decisivefact: The functions h (w) all fit
AWI(VI

(A:N)

MO

u*

< v* g vi g MI,
*

then

h Vo ,V9 (w)

coincides

in the

UO VO

Mi Vi

Wo>t

Ui Vi

U9tVt

togetherto one function. Specifically: (A:0)

M Given any w, it is possibleto choose , VG so that M ^ M* For all such choices f MO, *>o, o VQ and UQ^ w ^ ^o. A Ati ,v e (w) has the same value. I.e. , (w) depends on w only. We denote it therefore by h(w).
O O

<v*^

Woft o

o Proof:Existencef M , : M = Min (M*, the desired properties. obviously possess

O

VQ

M;)

and

= Max
O

(v*, w)

Mj, vj: and M ^ M* < v* ^ vj, M ^ w ^ v'. Put MI = Max (M , M ), Vi = Min (t> , ^o)- Then?/ ^ MI ^ M* < y* ^ t^i ^ t; , g vi, and Mj ^ MI ^ M* < v* ^ Vi g v , ?/i ^ w ^ VL Now two MI ^ applicationsof (A:N) (first with M , VQ, MI, wi, i^, then with wj, vj, MI, Vi, t^) give A (M;) = h U} , (w} and ft u ,;(w) = AU,,,^^). Hence)) h Ui>V9 (w) w
VQ MO

M*

< v* ^

depends on
v

only: Choose two such pairs M ,

and

UQ

VQ

M?

Uot1>f

Vl

;
f

A.2.4. he function A(n;) of (A:O) is defined for all utilities and it has T all numerical values. We can now show with little trouble that it possesses the properties we need. that This is most easily done with the help of two auxiliary lemmata.

as desired.

(A:P)
MO

Given any two

M*

< V* g

M, v with

MO

< thereexisttwo < g !>o.

M tf, V

MO

with

P Proof: ut

MO

= Min (M*, M), v = Max (v*, v).

(A:Q)

Given any two M, v with M < v, put A(M) = a, A(t>) = /3. w Then a < ft, and A(ti?) coincides ith g\"'%(w) in the latter's domain u w v.

as indicated in (A:P). By (A:M) h UitV9 (w) is a w two suitable a , j9o- By (A:0) A(M?) coincides ith A^^w), QW\\( W ) w i.e. ith gfy(w), in the latter'sdomain u w v . Applying this w

Proof:ChooseM

VQ

to)))

DERIVATIONFROM THE AXIOMS

w

625))

gives g^\\(u) = h(u) = a and gl**\\(v) = A(t>) = 0. Since 0j;Jj(w)is monotone, this implies a < /3. Next by (A:L) (with w , t>o, coincides o, 0o, w, v, a, in place of w , v , o, 0o, w,, vi, ai, &i)gl'\\(w) in the latter'sdomain u g w ^ v. Consequently with the sameis *(w) true for h(w). After these preparations we establish the relevant properties of h(w)

= u and

to

=v

g'
t

(A:R)

The mapping
w
+

h(w)

of all w; on a
(i) (ii) (iii) h(w) (iv) For

MO 0. MO = 1. is monotone, < y < 1and u < v MU y)u + yv) = (1

- setof numbershas the -

following

properties:

- y)h(u) +

yh(v)

(A:S)
the

(A:R).

mappingof all w on any set of numbers,which possesses (i), properties (ii) and (iv) is identicalwith the mappingof

A by Proof: d (A:R):Ad (i), (ii):Immediate (A:0) and (A:M).

: (A:P)and then a, and g*'(w)accordof (A:Q). Now by (A:H),(ii') (with u, v, v, y in place u , v , w, y) = (1 y)g%(u) + yg*Z(v). Henceby (A:Q) 0:;5(U 7)w + yv)
ing to

Ad (iii):Containedin (A:Q). to Ad (iv) Chooseu, v according

T)U

+ yv) = (1
w

- y)h(u) +

yh(v)))

as desired.
Ad

(A:S):Considera mapping
>

hi(w)

of all utilities w on numbers, which fulfills (i), (ii) and (iv). Choosetwo = ^i(^*), == ^i(v *)- Then, Uo, v with u g u* < v* ^ v , and put in w by (A:I), /ii(tu) coincides ith ^uJ'^C^) the latter'sdomain w g i^ g VQ. Putting
tn(v*)

w By (A:0) this means that hi(w) hwiM,i.e. ith h(w), in w ^ t^ ^ and h(w) are altogetherdentical. i A.2.5. (A:R),(A:S)give a mapping of all utilities on numbers, which
t>o.

= u* and 10 = v* we get sC$(u*) = hi(u*) = 0, ^J(w*) = = 1. Henceby (A:M) gift is fc,^. Thus hi(w) coincides ith w
w

and b possesses plausibleproperties is uniquely characterizedy them, and therefore we might let the matter restthere. However,we are not yet)))

an and it introduces arbitrary normalization in (i), (ii) (by means of the arbitrary u*, v*). In what follows,we will eliminate thesemaladjustments. This will prove fairly easy.
We first

THE AXIOMATICTREATMENT OF UTILITY The in quite satisfied,for the followingreasons: characterization (A:R) does not coincidewith that one by (3:l:a), (3:l:b)A:R) goeslessfar in (iv) ( for in (this is asserted (3:l:b) all u, v, in (iv) only for those with u < v)',
626

extend(iv)

in

(A:R).
y)u

(A:T)
or

Always

(1

+ yu = u.

F Proof: oru (1 y)u + yu say that y belongsto classI (uppercase) II (lower case). If y is in class I or II and if < < 1,then u $ (1 ft)u + ft((l y)u + yu} $ (1 7)14 + yu and (For by (3:B:a) (3:B:b). y in classI or II, respectively:First,u, or + yu, 1 in placeof u, v, a in (3:B:a) (3:B:b). Second, (1 y)u or + yu, u, ft in placeof u, v, a in (3:B:b) (3:B:a).) By(3:C:a) (1 y)u of and (3:C:b) (with u, u, ft, y in place u, v, a, ft) (1 ft)u + ft((l y)u + yu) (1 fty)u + ftyu. u Hence $ (1 7)\"+ ftyu (1 ft)u + ftu. Put * * 187. Since is t < 7 < 1, free in < ft < 1,herefore 6 is free in < 3 < 7. Assuming < 6 < 1,we have therefore: I If 7 is in classI or II,then every 5 < 7 is in the sameclass (A :9) or II.

ft

ft

- -

ft

(A:10)

Under the conditionsof (A:9)))

(1 t)u + &u $ (1 J)u + yu,

respectively.
The expression (1 7)1* + yu is unchangedif we replace by 1 7. 7 5 is equivalent to 7 > d, we can put 1 7, 1 6 in place y <I t becomehis: of 7, a in (A:9). Then (A:9)and (A:10)
As

(A:ll)
(A:12)

or II.

If 7

is in class or II,then every 6 I

> 7 is the same classI

Under the condition of (A:ll)

(1

)u

+ 81* (1

- y)u +

yu,

respectively. Now (A:9)and (A:ll)how,that if 7 is class or II, then every 5(<7 I s or = 7 or > a) is in the same class or II. I.e.f eitherclassI or II is not I i < 6 < 1. Assume this to be the case empty, then it contains all 6 with two 7, d with 7 < 5. Then by (A:10) (for classI or II), and consider of (1 t)u + tu $ (I y)u + yu, and by (A:12)(with d, y in place 7,

*))))

DERIVATIONFROM THE AXIOMS

in

827

(1 5)u + $w ^ (1 y)u + yu. Henceat any rateboth < and > hold (1 6)w + &u (1 T)W + yu. This is a contradiction.Therefore I both classes and II must be empty. never u ^ (1 y)u + yu, i.e. Consequently always (1 y)u + yu w,
as desired. (A:U)

< 7 < 1,any u, ). Proo/:For u < v this is (A:R),(iv). For u > v it obtains from the former by putting v, u, 1 y in place of u, v, y. Foru = v it follows from
(0

*((!- 7)u + yv) * (1- 7 )*(u)+ 7*W

Always

(A:T).
to and At c form, i.e.orresponding (3:l:a) (3:l:b).
We

can now prove the

this point we also drop of the assumedfixed choice w*, v*, which was introducedbefore (A:M). a Thereexists mapping (A:V)
w
>

existence uniquenesstheorem in the desired and

v(w)

of all
(i)

u?

on a setof numberspossessing following the

properties:

(ii) For < 7 < 1and any v((l 7) u + yv) = (1

Monotony.

u,

T)V(U)

+ yv(v).

(A:W)

the Forany two mappingsv(w) and v'(w) possessing propties(i), (ii),we have v'(w) = o)ov(iy) + coi,
with

two suitable but fixed w , wi and

1

> 0.

different utilities, u* ^ v*. u* then interchange and v*. Thus at any rate u* < v*. for (A:L)-(A:U).We now Use theseu*, v* for the constructionof h(w), prove Ad (A:V)-:The mapping If u*

L Proof: et u*, v* be two

> v*,

i.e.

w fulfills

>

h(w)

Ad

(i) by (A:R),(iii),and (ii) by (A:U). v (A:W):Consider (w) first. By (i) v(u*)

hi(w)
=))

< v(v*).

Put

v(*) ~ v(u*)))

v(u;)

v(u*)))

1Strictly speaking, the axioms permit that there should be no two different utilities. of. If there are no two This possibility is hardly interesting, but it is easily disposed different utilities, then their number is zeroor one. In the first caseour assertions are case:Thereexists one and only one Assume therefore the second vacuously fulfilled. A function is just a constant v(u> ) o- Any such function fulfills (i), (ii) utility u> 1 and wi ai in (A:V). In (A:W), with v(w) o, v'(u>) aj, choose

--

o>

- -

.)))

628 Then hi(w)

THE AXIOMATICTREATMENT OF UTILITY

in

fulfills (i),(ii)in (A:R) automatically, and (iii),(iv) (i), (ii) above. Henceby (A:S)hi(w) = h(w), i.e.

(A:R)by

(A:13)
where ao, i are fixed Similarlyfor v'(w):))

= <x*h(w) + lf numbers: = v(v*)

v(w)

v(w*)

> 0,

= v(u*).

where a[ are fixed numbers: J Now (A:13) and (A:14) ive together)) g

a,

= v'(t>*)

v'(w*)

(A:15)
where
wo,
<*>i

v'(w)
are fixed numbers: o w

the desiredresult.

= ~ > 0, wi = ao

-^

l
<*o

> 0, = v(i/*).
<x{

This is

A.3. Concluding Remarks

are theorems called for in 3.5.1. Consequentlythe assertionsof 3.5.-S.6. established their entirety. in the analysisof the concept At this point the readers advised to reread i as and of utility and of its numerical interpretation, given in 3.3. 3.8. There o are two points, both of which have beenconsideredr at leastreferred to and the concept of complementarity. Simply additive formulae, like of would seem indicate that we are assumingabsence any form to of complementarity betweenthe things the utilities of which we are combinthat ing. It is important to realize, we aredoingthis solelyin a situation wheretherecan indeedbe no complementarity. As pointed out in the first our u, v are the utilities not of definite and possiblycoexistpart of in ent goods or services,but of imagined events. The w, t; of events w, y, of which only one can particular refer toalternatively conceived and will becomeeal. I.e.(3:1 deals with either having u (with the r :b) probability a) or v (with the remaining probability 1 a) but since the two arein no case conceivedas taking place t together, hey can never complementeachother in the ordinary sense. It shouldbe noted that the theory of gamesdoesoffer an adequate way of dealing with complementaritywhen this conceptis legitimatelyappliI cable: n calculating the value v(S) of a coalition S (in an n-persongame), as described 25., possibleforms of complementaritybetweengoodsor in all betweenservices, hich may intervene, must be taken into account. Furw that thermore,the formula (25:3:c) expresses the coalition S u T may be worth more than the sum of the values of its two constituent coalitions S the it betweenthe services))) T, and hence expresses possiblecomplementarity
worth reemphasizing now. loc. it., ut which seem c b A.3.2. first one deals with the relationshipbetweenour procedure The

A.3.1. (A:V)

and (A:W)

are clearly the

and existence uniqueness

(3:l:b),

3.3.2.,

(3:l:b)

CONCLUDINGREMARKS

629

of the membersof the coalition S and those of the membersof the coalition

T.

the question,whether our approach forces one to value a loss exactly as much as a (monetarily) equal gain, whether it permitsto attach a utility or a disutility to gambling (even when values balance), the expectation etc. We have alreadytouchedupon these (cf. questions in the last part of 3.7.1. also the footnotes 2 and 3 eod.). remarks may be useful. However,someadditional and more specific Considerthe following example: DanielBernoulli proposed(cf.footnote 2 on p.28),that the utility of a monetary gain dx should not only be proportional to the gain dx, but also (assumingthe gain to be infinitesimal that is, asymptoticallyfor very small gainsdx) inversely proportionalto the amount x of the owner's total possessions, xpressed money. Hence in e (using a suitable unit of numerical utility), the utility of this gain is x The The

(Cf. also 27.4.3.) remarkdealswith The A.3.3. second

excesstility u excess utility

X

of owning

= In over owning x 2, is then / #2 xtj X of gaining the (finite) amount over losingthe same
x\\,

amount is In

In

the gains and losses latter are more strongly felt than the former. A 50%-50% gamblewith equal risks, is definitely disadvantageous. Bernoulli's our axiomsand obeysour results: Nevertheless utility satisfies the utility of possessing units of money is proportionalto In x, x However,

_ XT]= In ( 1
\\

77

X2

\\

This is

o < 0,i.e. f equal

u and not to x! Thus a suitable definition of utility (which in such a situation is esseneliminatesin this case specific the tially uniquely determinedby our axioms) to or disutility of gambling, which prima facie appeared exist. utility not because think that it is parwe B We have stressed ernoulli's utility, or much nearerto reality than many other more or less ticularly significant, that similar constructions.The purposewas solelyto demonstrate, the use involve assumingthat 50%-50% not of numerical utilities does necessarily gambleswith equal monetary risks must be treatedas indifferent, and the
8 like.

1That the utility of x units of money may be measurable, was pointed out in footnote 3 on p.

terms of In

gambling has under all conditions a definite utility or disutility, where cannot of numerical utilitiesfulfilling the calculus mathematical expectations In such a systemsomeof our direct or indirect. be defined by any process, 1 The a discussedboveinvolved equal risks in terms of x, but not in 50%-50% gamble

It constitutesa much deeper roblemto formulate a system, in which p

x.

8 As stated in remark we (1)in 3.7.3., are disregarding transfers of utilities between several persons. The stricter standpoint used elsewherein this book, as outlined in of utilities between persons, doesforceone to specifically, the free transferability assume proportionality between utility and monetary measures. However, this is not relevant at the present stageof the discussion.)))

18.

but not proportional

to x,

2.1.1.,

THE AXIOMATICTREATMENT OF UTILITY

which axiom or groupof axiomsismost likely to undergosucha modification. A.S.4.Thereare neverthelesssome observationswhich suggestthemselvesin this respect.

axiomsmust be necessarilyinvalid.

It is difficult

to foreseeat this time,

relationshipof incomparability, denoted by u v, which meansthat neither = v nor u > v nor u < v. It shouldbe noted that the current method of indifferencecurves does properlycorrespond this possibility. Indeed, not to in that case the conjunction of \" neither u > v nor u < v,\" corresponding to the disjunctionof \" either u = VOTU\\\\V,\" and to bedenotedby u v, can be treated a mere as o broadeningof the concept f equality (of utilities, cf. also the remark concerning identity in A.I. 2.). Thus if u u',v v', then u', v' can replace, v in any relationship,e.g. u in this caseu < v implies u' < v'. Hencein particular u u' and v = v' have this consequence, u = u' and v v' have this consequence.I.e., and writing v, w, u for w, t>, u' and u, v, w for u, v, v', respectively:
\\\\

First: he axiom (3:A) or, more specifically, (3:A:a) expresses T the t of of completeness the ordering of all utilities, i.e.he completeness the individual'ssystemof preferences.It is very dubious,whether the idealization of reality which treatsthis postulate as a valid one,is appropriate or o even convenient. I.e. ne might want to allow for two utilities u, v the
u

|| ||

\\\\

\\\\

(A:16) (A:17)

u u

< v and

||
v

and

v v

< w imply
\\\\

imply u

< w. < w.

However, for the really interesting casesof partially ordered systems neither (A:16) (A:17)s true. (Cf. e.g. secondexampleat the end nor i the of which is also dealt with in footnote 2 on page 590, where the connection the concept f utility is pointed out. This is the ordering with o of a plane so that u > v means that u has a greaterordinate than v as well as a greater bscissa v.) a than In Second: the group (3:B)the axioms (3:B:a) (3:B:b) and a express property of monotony which it would be hard to abandon. The axioms on and what (3:B:c) (3:B:d),the other hand express is known in geometrical axiomatics the Archimedean property:No matter how much the utility as v exceeds (or is exceeded the utility w, and no matter how little the by) w exceedsor is exceeded ( utility by) the utility u, if v is admixed to u with a sufficiently small numerical probability, the difference that this admixturemakesfrom u will be lessthan the difference of w from u. It is probably desirableto require this property under all conditions,sinceits abandonmentwould betantamount to introducing infinite utility differences. 1For a statement of the Archimedean property in an axiom at ixation of geometry, where it originated, cf. e.g. . Hilbert, loc.cit. footnote 1 on page74. Cf. there Axiom D V.I. The Archimedean property has sincebeen widely used in axiomatizations of number systems and of algebras. Thereis a slight difference between our treatment of the Archimedean property and its treatment in most of the literature we are referring to. We are making freeuseof the conceptof the real number, while this is usually avoided in the literature in question. Therefore conventional approachis to \"majorise\" the \"larger\" quantity by successive))) the

65.3.2.,

CONCLUDINGREMARKS L tion: et any completelyorderedsystemof utilities

it In this connection is also worth while to

681

make the following observa<u be given, which does not allow the combination of events with probabilities,and where the utilities are not numerically interpreted. (E.g. system based on the a familiar ordering by indifference curves. Completeness this ordering of obtains, as indicatedin the first remark above, by extendingthe concept f o u b equality i.e. y treating the concept v, that we introduced there, as u t equality. In this case * v means,of course,hat u and v lie on the same indifference curve.) Now introduce events affected with probabilities. Thismeansthat one introducescombinations of, say, n ( = 1, , 2 ) events
with

respectiveprobabilitieson,

an

(i,

,
c

This requires the introduction of the corresponding(symbolic) utility combinationsot\\ui + + anu n (HI, , u n in u). It is possibleto order thesectiUi + + anu n (any n = 1,2, and any a\\ , aw and ui, a , u w subjectto the above conditions)completely, nd without making them numerical if the orderingis allowed to be non-Archimedean. Indeed,comparing, say, mui + + anu n and 0it>i + + ft mvm we may assumethat n = m and that the ui, u n and the v\\, , , vm
9

0, % a,= 1). -i

+ + + + Ov and Ot/i + + /Mm for a^i+ + Oun + P\\VI + + On u n and 0it>i + + n an , 0, $ v , and then replace + m; MI, , u n , Vi, - - - , v ai, , , 0; 0, , 0, ft, , ft. by n; u,, , u\\ h , an ; 0i, + a w n and fti^ -f , n ). Then we compare a^i+ + 0i/. Next make, by an appropriate permutation of 1, , n u\\ > - - > u n . After thesepreparationsdefine a \\u\\ + + nu n > ftiUi + as meaning that for the smallest i( = 1, + , n) for which a, 7* fa, say i to, there is ai > . It is clearthat these utilities are non-numerical.Their non-Archi<*nW Ot>i

coincide (write aiUi +

m m

m \\

<x

<x

ffnUn

ft i%

small

medean character ecomes b clearif one visualizes that here an arbitrary excess robability att & affecting u will outweigh any potential p of the remaining u t , = t'o + 1, oppositeexcess , n, probabilities i.e. f utilities < u, f . (This then excludeshe applicationof criteria like o t that one in footnote 1 on page 18.) Obviously, they violate our axioms
i%

t*

Sucha non-Archimedean orderingis clearly in conflict with our normal ideas concerning nature of utility and of preference.If, on the other the
while we \"minorise\" Hilbert'a procedureoc. l entity (the utility discrepancybetween w and v in our case)by a suitable small numerical multiple (the a-fold in our case) of the \"larger\" entity (the utility between9 and u in our case). discrepancy This difference in treatment is purely technical and doesnot affect the conceptual o situation. The readerwill alsonote that we are talking of entities like \"the excessf v over u\" or the \"excess u over v\" or (to combine the two former) the \"discrepancy of of u and v\" (u, , being utilities) merely to facilitate the verbal discussion they are not part of our rigorous, axiomatic system.)))

and (3:B:c) (3:B:d).

the

addition! f the \"smaller\" one(cf o

\"smaller\"

.e.g.

cit.),

including system, satisfying our axioms (3:A)-(3:C) and hencepossessing the Archimedean property then the utilities would have to be numerical, sinceour deductionof A. 2.applies. Third:It seems probable, that the really critical roup of axiomsis g This (3:C)or, more specifically,the axiom (3:C:b). axiom expresses the combination rule for multiple chancealternatives, and it is plausible, that a specificutility or disutility of gambling can only existif this simple combinationrule is abandoned. a Some changeof the system (3:A)-(3:C), t any rate involving the abandonmentor at leasta radicalmodification of (3:C:b), perhapslead may and to a mathematically complete satisfactory calculusof utilities, which allowsfor the possibilityof a specific utility or disutility of gambling. It is hoped that a way will be found to achieve this, but the mathematical Of t to difficulties seem be considerable. course, his makes the fulfillment verbal means appear even of the hope of a successfulapproach by purely more remote. f It will be clearrom the above remarks,that the current method of using indifference curves offers no help in the attempt to overcome these difficulties. It merely broadens the conceptof equality (c.f.the first remark above),but it gives no useful indications and a fortiori no specificinstructions as to how one should treat situations that involve probabilities, with utilities.))) which areinevitably associated expected

THE AXIOMATICTREATMENT OF UTILITY to hand, one desires define utilities (and their ordering)for the probability632

INDEX OF FIGURES
Figure 1

2 3 4 5 6
7

10 11 12 13
15
14

8 9

20 21 22 23 24 25 26 27 28a 28b 29 30

16 17 18 19

Page 62 63 63 64 64 65 65 65 65 78 93 94 94 94 99

Figure

35 36 37 38 39 40

Page

184

Figure

192 192 192

203 203

70

71
72 73 74 75 76
77

197

41

Page 405 408 408 409 409 409

131 132 133 133 133 133

135 135

42 43 44 45 46 47 48 49 50

212 212 212 216 216

217 217
230 252 283 283 284 284 286 286 286 286 287 293 295 305

78 79 80

81
82 83 84 85 86 87 88 89 90

410 410 411 411 412 412 412

414 414

410

51
52 53 54 55 56 57 58 59 60

261

415
470 470 478 554 554 554 554 568 568 568 569

416 416

91
92 93 94 95 93 97 Q8
Qg

136

31
32 33 34

137 137 169 175 176 177 179

61
62 63 64 65 66 67 68 69

180 181 181 182

313 331
337 337 337 395

JJ

JJJ

102 103

579))

579
579))

633)))

INDEX OF NAMES))
Bohr,N., 148 Bonessen, 128 T.,
6 Archimedes, 3O 8 Bernoulli,D.,28, 3, 629 Birkhoff, G., 2, 3, 64, 66, 340, 589 6 E6 582 5 B6hm-Bawerk, .von, 9, 562, 64, 581,

Brouwer,L. E.J.,154 Burnside, ., 256 W 3 Caratheodory,C., 43,384, 619 Chevalley,C.,vi D'Abro, A., 148

219 E Borel, .,154,186,

Kakutani, S., Kaplanski, I.,vi 4 Kepler, Kdnig, D.,

6O))

128 5 Frankel, ., 61, 95 A 2 595 Hausdorff, F.,61, 69, W., Heisenberg, 148 Hilbert, .,74, 76, 63O D v Hurwicz,L., i 154

Euclid,23 Fenchel, ., W

Dirac,P.A. M.,148 1 Doyle, C.,176, 78

Dedekind, 29, 1 619

Lipschitz,493 v Loomis, L. H., i 3 MacLane,S., 40 Marschak,J.,vi 2 Mathewson,L. C., 56 5 Menger,C., 64 Menger,K.,28, 176 Mohs,22 1 Morgenstern,O.,176, 78 Morse,M., 95 Neumann, J. von, 1,154,186 Tarski, A., 62 2 Tintner, G., 8 Tycho de Brahe, 4 7 Vebien, O., 6
Ville, 154, Wald, A., v, vi A Speiser, ., 256

129 Kronecker,

2 Pareto, ., 18, 3, 29 V

Newton, 4, 5, 6, 33

198 186, Weierstrass,129 256 76, 128, Weyl, H., Young, J. W., 76 595)) Zermelo,E.,269, J,,

634)))

INDEX OF SUBJECTS))
Acyclicity,

589,591,594,595, 596,598, 600,601,602,603,609;strict, 594,595, 597, 598,600,601,602,603 of value, 251,628 Additivity \" found out,\" 105 Adversary Agreements, 221,224;sanctity of, 223 Ally, 221 Alternatives, 55;number of, 69 Anteriority, 51, 2, 77, 78, 112, 124 5 117, 35,41,504 Apportionment, Archimedean property, 630,631
Assignment, actual, 75; pattern of, 75 Asymmetric, 270, 448 Austrian School, 9 Axiomatization, 68, 74, 76 Axioms, 25, 26, 28, 73; independenceof, 76; logistic discussion of the, 76

Backgammon,
Bargaining,

Barter exchange,7
Behavior, Bid, 557

616

52,58, 79, 124,125,164 338,501,512,557, 558,572,

34;expected, 46;standards of, 1 seeStandards of behavior Bestway of playing, 100

Bidding, alternate, Bilateral monopoly,

Bilinear form,
Bluffing,

154,156,157,166,233 54, 164,168,186,188,204,205, f 206,208,218,541;ine structure of, 209 Boolean algebra, 62 Bound, 59, 60 Bounded, 384 Bounds, lower, 100; upper, 100 Bridge, 49, 52, 53, 58, 59, 79, 86, 224;
Duplicate,

556

211 1,6, 35,504,508,543,

530, 532; interpretation, 538; in the 348;normalized form of the, 325;reduced,248,325,543,544,545; restricted, 528, 529, 531, 532, 533; strategically equivalent, 536; vector 5 operations on, 253;zero reduced, 45 Characteristic set function, 241 Chess,49, 52,58, 59, 113, 124,125,164 good, 125 Chief player, seePlayer, chief Choice, 51, 9,69,222,508;actual, 75; 49, 5 anterior, 72; pattern of, 75;umpire's, 80, 81, 183 82, a Choice,xiom of, 269 Circularity, 40,42, 56 Closed 384 set, Coalition, 15,35, 47, 221,222,224,225, 229,234,237, 240,260,276, 289,418, 420,507, 509,510,531,533,539,566, 572, 573, 605; absolute, 231,238; certainly defeated, 40;certainly winning, 4 2 440;decisive, 20;defeated, 96;final, 4 f 315,317;irst, 306,307, 315,316,320; losing, 420,421,423; interplay of, 291; minimal winning, 429, 430, 436, 438, 445; profitable minimal winning, 442; 437; weighted majority, unprofitable, 434; winning, 296, 297, 333,420,421, 423,436,445,470 Coalitions, competition for, 329;of differnew theory,

Closed systems, 400

ent strengths,

227

Buyer,

14, 556, 557, 565,569, 572, 574, 581,583,585,609,610,613 Compatible, 267 ff. 6 Compensations, 36,44, 47, 225,227, 233, Calculus, 3, 5, Calculusof variations, 11, 95 234,235,237, 240,507, 508,510, 511, 513,533,541,558 Calling off, 179,180,541 1 15,249,509 Competition, 1, 3, Cartels,15,47 Complement, 62 Categoricity, 76 Centerof gravity, 21, 31, Complementarity, 18,27, 251,437, 628 1 303 Complementation, 422 Chance, 9,52, 87 3 Characteristicunction, 238 ff., 240,245, Completeordering, seeOrdering, complete f 2 509,510, 511, 529,530,533,535, Completely defeated, 96 527, extended, 528, Composition, 340,359,360,454,548; of 557, 574, 584,605,610; simple games, 455)) 529,532,533;game with a given, 243,))
635)))

113 113; Tournament,

Column of a matrix, 93, 141 Combinatorics, 45 Commodity, 10,13,560,565 Communications, 86;imperfect,

Commutativity, points

91ff.,94. SeealsoSaddle

86

636)) Conjunction, Constituent,

INDEX OF SUBJECTS))
66 340, 353, 359, 360,

Curves, undetermined, 418 Cutting the deck,185,186

\"Cyclical\"

Economic equilibrium, 4; fluctuations, 5; indecomposable,457, 471; inessential, models, 12,58; statics, 8 -453,457;simple, 453,455,457 Economics, mathematical, 154 Contribution, 364,366 Economies, internal, 341 Conventions, 224 Convex bodies,128 Economy, planned communistic, 555; Robinson Crusoe,9; SocialExchange, Convexity, 128ff., 275, 547 Cooperation, 221,402,474, 481,508,517; 9 ft. Effectivity, 272, 282,350,367, 524 complete,483 Energy, 21 Couple, 22,226,243,509 2 Entrepreneur, 8 Crusoe, , 15,31, 555 87, 9 197 CubeQ,293,295;centerand its environs, Equidistributed, center 227, 313; of, 316,317,321; corner,303, Equilibrium, 4, 34,45,see 365 304,307, 340,429;interior of, 302,303, Equivalence strategic, Strategic equiv304;main diagonal of, 302,304,305,312; alence neighborhood of the center, 321; special Essentiality, 249,272, 351,452 Exceptional, 593 points in, 295 ; three-dimensional part Excess, 64,367, 417,418,454,455,548; 3 of, 314
ff

518; Ecarte*, 59

E(e ), Solutions

for

r in,

393ff.

dominations,

39

342, 357, 360; analysis of, 343 Decomposition,242, 292, 340, 359,360, 452,537, 548;elementary properties of, 381 its relation to the solutions, 384 ;
Decomposability,

Defeated,296; certainly, 440; fully, 436.

Decomposition Partition, decomposition

see

Partition,

10,28,29,32,33,83, 87, 117, 18, 1 126, 149,156,157;moral, 28,83; values, 183 Exploitation, 30, 329,375
Extensive form,

of the, 418;limitation of distribution the, 365,366; lower limit of, 368; too great, 374,380, 19;oo small, 374, 380; 4 t upper limit of, 369 Exchange economy, 9, 31 Exchange, indeterminateness of, 14 Excluded player, seePlayer, excluded Expectation, 12,28,83,539; mathematical,

F(<? ), Solutions for T in, 384ff. 164, Defensive, 205 Determinateness,general strict, 150,155, Fairness, 166,167, 225,255,258,259,470 Fictitious player, seePlayer, fictitious 158ff.; special trict, 150,155;strict, s out\" the other player, 148 \"Finding 106ff., ff., 165,178, 179 \"Finds out 11 his adversary, 106, 148 Diagonals,separation of the, 173 First element, 38, 271 Differential equations, 6, 45 Fixed payments, 246,281,298,534 Directsignaling, 54 Fixed Point Theorem, 154 Discrimination, 30,288,289, 28,475,476, 3 Flatness, 276, 547 512.Seealso

Seealso Players; Coalitions

119 112,

Ill

110,

Disjunction,

2 Distance, 0
Distribution,

66

Solution, discriminatory

Frame of reference, 29 1
Function,

Found out,

148

552, 299,301,340,358,397, 398,400, 553, 569, 570, 587; area, 579, 580; 454,455, 457, 460,461,508,518, 537, curves in, 412, 570, 580; undominated 538 area, 409 543,603 Duopoly, 1,13, Dynamic equilibria, 45 Gain, 33, 128,145,539,556,559,629 Dynamics, 44,45, 189,290)) Gambling, 27, 28, 87, 630,
Dummy,
631)))

intransitive nation of, 37 Double-blind Chess, 8, 72, 79 5 Duality, 104

89,90,128,157 Domination, 38, 264, 272, 350, 367, 371, 415, 474, 520 ff., 522, 523, 524, 587; c acyclical onceptof, 602;asymmetrical, 270; extension of the conceptof, 587;
Domain,

364,437

37, 87, 226,261,263,350,

see Imputation, Fully detached, fully

detached

Functional, 157 Functional Calculus, 88, 154 Functional operations, 88,91 Fundamental triangle, 284, 405 ff.,

88, 128;arithmetical, 89; characteristic,238 ff.; continuity of, 493; measure, 252;numerical, 89; numerical set, 240, 243,530, 532; of functions, 157;set, 89

INDEX OF SUBJECTS))
Game and socialorganizations,
metric,

637))

43; asym334;auxiliary, 101 axiomatic ff.; definition of a, 73;chance component of the, 80; classification of, 46; complete o concept f, 55;completesystem of rules of, 83; composition, 339 ff.; constantsum, 346ff., 347,350,351,504,505,535, 536,537, 585;decomposable, 454,471, solution of, 358, 81 3 ; 518; decomposable, decomposition, 339 ff.; direct majority, 431,433;elements of the, 49; essential, 231, 232, 245, 250, 331, 534, 546; essential three-person, 220 ff., 260 ff., 471, 473; everyday concept of, 32; extensive form of the, 85,105,186,234; extreme, 534,535; fictitious, 240; final simplification of the description of a, general, 504ff., 505,538;general 79, 81;
description description

291ff.; zero-sum n-person, 48, 85, 238 ff.; zero-sum three-person, 220 ff., 260 ff.; zero-sum three-person, solution of essential, 282 ff.; zero-sum 169 two-person, 48,85 ff., 116, ff., 176 Geometry, 20, 74, 76; linear, 428;plane, 7-point, 469;projective, 469 Goodway (strategy), 103 Goodway to play, 108,159 Goods,omplementary, 437, 628;divisible, c 560,573 Group, 22, 76, 255 ff.; alternating, 258; invariance, 257; of permutations, 256; set-transitive, 258; symmetric, 256; theory, 256, 58, 95; 2 2 totally symmetric, 257;totally unsymmetric, 257 Group (of players), privileged, 320
four-person,

formal general 130,137,139 Half-space, 46-84;general n-person, Hand, 53,186,187,190,197,613 530,606;general n-pereon, Hands, discrete, 08 48, 85, 112, 2 application of theory, 542 ; imbedding Heat,3, 17, 21 of a, 398; indecomposable,354; ines- Hereditary, 454 sential, 231, 232, 245, 249, 251, 471; Heredity, 396,400 of a, 398; invariant, 257; Heuristic, 7, 25, 33, 120,238,263,291, \"inflation\" kernel of, 457, 459; length of the, 296,298,301,302,307, 316,318,322, 75; main simple solution of the, 444; 333,499,509,511,587 100, 102, 103, 119,49; Heuristic argument, 147,181, 227 majorant, 1 182, majority and the main solutions, 431; Higgling, 557, 558 minimum length of, 123; minorant, 100, 119, 101, 149; non-isolated character Homo occonomicus, 228 464 of a, 366;non-strictly determined, 110;Homogeneity, 433, 130, 137, 139; supporting, Hyperplane, non-zero-sum, 47. Seealso Game,gen134ff. normalized eral;normal zoneof the, 519; 183,234,239, form, 85, 100,105,119, 255,617 322,325,452,473;of chance,87, 185; Identity, Imbedding, 398,399,400,455,587 8 one-person, 5, 548; partitions which Imputation, 34, 37, 39, 240,251,263,264, describe a, 67; perfect information, 577, 112 plan of the, 98; plays of, 49; 350,376, 437, 517,520,527, 566,359; ff.; composition of, 587, 606, 610; reduced,248,259,473, 543ff.; rules of decomposition of, 359;detached,369, 147, the, 32, 49, 59, 80, 113, 224,226, ff., 413; set-theoretical 370, 375 detacheddetachedextended, 227, 334, 426, 472; extended, fully, 370, see 370, 375; of a, 60, 67; simple, description e f 372;detached,ully, 369,413; conomic Simple game; simplified conceptof a, conceptof, 435;extended, 364 ff., 367, ; 48;strategies in the extensive form, 111 368,369,372;single, 34,36,37, 39,40; strictly determined, 98 ff., 106,124,150, unique, 608 strictly determined, 165,172,174,516; generally, 150;strictly determined, Imputations, extended, sets of, 368;finite set of, 465,499;infinite set of, 288,499; specially, 150;struggle in, 125;superisomorphism between, 282; set of, 34, position of, 254,255; surprise in, 125; 44,608;system of, 36,277, 464 symmetric, 165,167, 192,195,334,362; 630 symmetric five-person, 332,334; sym- Incomparable, 590,357 [630, 31,632] 6 total of, 259; three-person, 35, Indecomposability, metry, 220ff., 282ff., 403 ff., 457, 550;three- Indifference curves, 9, 16, 19,20, 27, 29, person, simple majority of, 222 ff.; Indifference of player, 300 Indirect proof, 147 totally symmetric, 257; totally unsymvacuous, 116,Individual planning, 15 metric, 257;unique, 331; 123; 116; 546; value of the, 102,103,170,516; Induction, 112, complete, 113, finite, 597; transfinite, 598 weighted majority, homogeneous, 444; zero-sum extension of, 529; zero-sum)) Inessential games,

of, 57;

of,

ff

44)))

638))
Inessentiality, Information,

INDEX OF SUBJECTS))

249,272, 351,357, 454 143,144,164,166, Matching Pennies, 47, 51, 4,55,56,58, 67, 71, 5 g 169,176, 178, 185; eneralized forms, 175ff. 109, absurd, 67;actual, 67, 79; chance, complete, 30, 541, 582; Mathematical Method, difficulties of, 2; in Economics, imperfect, 30, incomplete, 30, 86; pattern of, 67, 69;perfect,51, 23,124, Mathematical physics, 303 1 164,233;perfect,verbal discussion, 153;diagonal of, 173, Matrix, arbitrary, player'sactual, 75; player'spattern of, negative 174; elements, 93, 138, 75; sets of, 77; umpire's actual, 75; transposed, 141,142;rectangular, 93, umpire's pattern of, 75, 77; umpire's scheme,diagonal of the, 138,140,

111,

183; 112, 182; 182; 115

1-8

126;

141;

Initiative, 189,190 Inner triangle, 409,413,553 Interaction, 341,366,400,483 International trade, 7, 341 Intransitivity, 39, 52 Inverted signaling, 54 Irrational, 128,523

state of,

Interval, linear,

131

173;skew symmetric, 142,143,167 Max operations, 89 ff. Maximum, 88,89,591,593,594;absolute, benefit, 513, 514, 591, 593; collective 13,42,86,87, 541,613; problem, 10,11, 220,504,517,555;relative, 592; satis-

141;

281 Isomorphic correspondence, Isomorphism, 149,350,504 Isomorphism proof, 281

Just dues,360,361 Kernel, 457 Killing the variable, 91
\"Kriegsspiel,\"

Mechanics,4
Min

faction, Measurability, 16,343 Measure, additive, 343; mathematical theory of, 252,343 Measurement, principles of, 16,20

10

Method, mathematical, 322 Method, saturation, 446

58

Laisserfaire, 225

\"Lausanne\" Theory, 15 Linear interpolation, 157 Linear transformation, 22,23 Linearity, 128ff. Lipschitz condition,

88,89 Mistake, 162,164,205 Mixing the deck, 185

Minimum,

Min-Max problem, 154 Minimal elements of W,

operations, 89 ff.

430,448 5 12,8;

6 Logic, 2,66, 74, 274 6 Logistic, 6 Losing, 421,426 Loss,33,128,145,163,167,168,205,539, 5 118, Move, chance, 0,69, 75, 80,83, 112, 555,559,629 122,124,126,183,185,190,517,604; removing of, 183;dummy, 127; imposMain condition, 273, 274, 279 sible, 72; in a game, 49,55, 58, 59, 72, Main simple solution, see Solution, main of the first kind, 50; of 98, 109, simple the secondkind, 50; personal, 50, 55, Main theorem, 153 70, 75, 112, 122,126,183,185,190,223, Majorities, weighted, 432 508,510 Majority, principle of, 431 Majority game, homogeneous, 443;homogeneous weighted, 444, 463, 469; Negation, 66 Negotiations, 263,534,541 weighted, 433,464

493

Money, 8, 10 Monopolist, 474 Monopoly, 13,474, 543,584,586,602,603 Monopsony, 584,586,602,603 23 Monotone transformations, Motivation, 43

Models, axiomatic, 74; economic, mathematical, 21 ff., 32,43, 74

111;

Majority principle, homogeneous weighted, Majorization of rows or columns, 174,180 ff. Mapping, 22, 581 Marginal pairs, 560,562,563,564,571, 7 Marginal utility school,

New

46r, 469

618

Market, 47,
21))

504,556,557, 560,563,581, 605; general, 583; three-person, 564; two-person,555 Mass,20,

in the, 351;olutions s the, 350 Normal zone, 396,399,401,417 un essentiality

domi348;decomposability in the, 351; nations in the, 350; essential threeperson game in the, 403 ff.; essentiality in the, 351;mputations in the, 350; i
in

Theory, characteristicfunction

in

the,

Numerical utilities axiomatic treatment 24. Seealso Utility, numerical)))

of,

INDEX OF SUBJECTS))
Numerical utility, numerical Numerical weight,

639))

17. See aho

432

Utility,

Players, interchanging

Offensive,
Oligopoly, Opposition Optimality, Strategy

164,205 1,13,47, 504

of interest,
permanent,

the, 104,109,122, 165,255;permutation of the, 294,463; privileged group of, 320,464;removable sets of, 533;strategies of the, 49. See

also Strategy. 167

Plays, set of all, 75 Poker, 52, 54, 56, 58, 59, 164,168,186, Optimum, 38 187 ff., 557, bids, 209;Draw, 187; Optimum Behavior, 34 general forms of, 207; good strategy, Orderof society,41,43. Seealso Organi196;interpretation of the solutions, zation, Standard of behavior mathematical description of all soluOrdering, 19,37, 38,589;complete,19,26, tions, 216 ff.; overbidding, 188, 28,589,591,593,595,600,617;artial, p passing, 188,190, 199;seeing, 188, 590,591,600;well, 595 190, 199, solution, 199,202, 95, Oreographical, 97 204;strategies, Stud, 187

11, 205. 220,484

162,

Playing

appropriately,

102,103,107,159,

Seealso

613;

Organization, socialand

224, 328, 366, 401, 419; economic, 1, 43, 225,318, 4 319,329,358,362,365,402,436,471; social,complexity of, 466 Origin (point in space),129 Outsidesource,363,364,366,375, 419 S Overbid, 186. ee also Poker

191, 191, 218; 191;

21

218; 190;

Position,

Positive octant, 133 Positive quadrant, 133

Preferences, 17, 18,23,522,590,607, 15, 630; completeness of, 29, 630, 631; transitivity of, 27. Seealso Utility Preliminarity, 51, 2,77, 78, 112, 124 5 117, Parallelism of interests, 11, 221 220, Preliminary condition, 273, 471 Partial ordering, 590. SeealsoOrdering Premiums, 582 33 31, Participants, Price,556,559,562,563,564,571, 572, Partition, 60,63,64,66,67, 69,84, 114 ff.; 582,585;average, 564,582;unique, 564 decomposition, 353,354,356,357, 457, Privilege, 464 471;logistic interpretation, 66 Probabilities, choice 145 of, Pass,95,97 17, Probability, 11, 19,39,81, 128, 46, 87, 1 Passing, seePoker 197;geometrical, 197;numerical, 14,19, Patience,86 27, 28,69,75, 80,113, 147,156,183, 145, Per absurdum proof, 147,148 604;of losing, 144;of winning, 144 Permutation, 255, 262, 294, 309, 319;Production, 5, 13, 504 cyclic,230,470 Productivity, 33,34,504,540 303,341 Perturbations, Profit, 33,47, 572 43 Physical sciences, 1, 2 23,32,401 Proper subset, 61 Physics, 2, 3, 4, 45,76, 148, Propersuperset, 61 mathematical Planning, 86 phenomena, Psychological Plateau,rectangular, 97 treatment, 28,77, 169
Plausibility

Postulates, 27.

Seealso Axioms

considerations,7 Play, 49,59, 71; ctual, 82; courseof the, Quantum mechanics, 3, 33,148,401 a 68,70, 84;individual identity of the, 71; of outcomeof the, 82;sequence choices, 49; value of a, 104,105,124,127,150, Randomness, 146 33, b Rational, 9, 517; ehavior, 163,165,178,238 127,150,160,224,225;playing, 54 Player, chief, 473, 474 ff., 481,483,500, 502; chief, segregated,500, 502; com- Rationality, 99, 128

8-15, 31,

4 defeated, 18; Rebates,582 posite,231,232,239,516; 2 discriminated, 476, 502;excluded, 89, Recontracting, 557, 558 form, 248,544 301, 512;ictitious, 505,506, 507, 508, Reduced f 513, 509,511, 514,516,518,537;found Reduction, 322,325 3 i out, 146;ndifferent, 299;isolated, 75; Relativity Theory, 23, 148 privileged, 473; segregated,476, 502, Removable set, seeSet,removable Ring, 243,530,531 353, 357; splitting 503;self-contained, Risk, 163 the, 86; unprivileged, 320; victorious, s Robinson Crusoe, eeCrusoe))) 426))

640))
Rolling dice,166 Roulette, 87 Row of a matrix, 93,

INDEX OF SUBJECTS))

Social roduct, 46 p Socialtructure, 484 s 141. eealso Matrix Society,42,320,341,523,540 S Solitaire,86 153; Soloset, 244,530,531 Saddle,95, 97; points, 93, 95, 110, Solution, 102, 64,350,367, 368,417,478, 2 value, 88, 95, 107 526, 527, 587, 588; areas (two-dimenSt. Petersburg Problem, 28,83 sional parts) in the, 418;asymmetric, Satisfaction, maximum, 8, 10,15 concept 315,362; composition of, 361; Satisfactoriness,267 ff., 446 ff.; maximal, o of a, 36;concept f imputation in form269 ing, 435;curves (one-dimensional parts) Saturation, 266,267 ff., 446ff., 448,591 in the, 417; ecomposable, decom362; d Scalarmultiplication, 129,253,254 definition, 39, 264; disposition, 361; seePoker Seeing, 301, 307, 318, 320, 329, criminatory, Segregation, 290 essential zero-sum three512; Seller,556, 557, 565,569, 572, 574, 581, 442,511, 282; existenceof, 42; person game, 583,585,609,610 of the concept of, 587; extension numbers, 173 Separated families of, 329, 603; finite, 307, 500; 1 ff.; Set, 60, 61, 14 certainly necessary, finite sets of imputations, 328; for an 273, 274, 277, 308,309,323,405,430, acyclic relation, 597; for a complete 471, 547; certainly unnecessary, 273, for ordering, 591; r in E(e ), 393 ff.; 274, 276, 308,309,323,405,430,471, for T in F(eo), 384 ff.; for a partial 547; completely ordered, 19; onvex, c ordering, 592;for a symmetric relation, 131, elementsspanned, 131; 133;convex effective, 591;eneral games with n 3, of all, g 38,264; of, 61; 241, empty, 61, 548; general three-person game, 551; 380;finite, 61; 275, 276, 423,424; flat, indecomposable,362; inobjective (disminuend, 62; one-element,61; partially 290; main simple, 444, criminatory), ordered,19; removable, 533,534,535; 464,467, 469;multiplicity of, 266,288; set; solo,seeSolo splitting, 353,457, 518; natural, 465; new definition of, 526; splitting minimal, 355, 356; splitting, objec290,475, 511; non-discriminatory, system, of all, 354; subtrahend, 62; tive (non-discriminatory), 290; one60 ff. theory, 45, element, 277, 280;set of all, 44;superSets,difference of, 62; disjunct, 62; interunique, numerary, 288;symmetric, 315; sectionof, 62;logistic interpretation, 66; 594, 600, 601, 603; unsyrametrical of imputations, composition of, 359;of central, 319 imputations, decomposition of, 359; Soundness, 265 pairwise disjunct, 62;product of, 62,66; half, 137; 1 self-contained,354;sum of, 62,66;sys- Space,Euclidean, 21, 28,129; linear, 128; linear, 157; n-dimensional tems or aggregatesof, 61 positive vector, 254 5 Signaling, 51, 3; direct, 54; inverted, 54 56 f 420 ff., 454, 605; adding Specialorm of dependence, Simple game, dummies to, 462; and decomposition, Splitting the personality, 53 452; characteristic function, 427; char- Stability, 36,261,263,266,365;inner, 42, 43,265 acterization of the, 423; complementa4 tion in, 422 ff.; enumeration of all, Standard of Behavior, 31,0, 41,42, 44, 265,266,271,289, 361,365,401,418, 445 ff.; for n ^ 4, 461;for n ^ 6, 463; discriminatory, 472, 478, 501,512,513; for small n, 457; indecomposable,457; one-elementsetsin the, 425 ; six main 290;multiplicity (of stable,or accepted), 42,44, 417;non-discriminatory, 290 counter-examples, 464;solution of, 430; strategic equivalence, 428; systems W, Statics,44,45,147,189,290 Statistics, 10,12,14,144 L, of, 426ff.; with dummies, 461 143,144,164, Stone, Paper,Scissors, Simplicity, 433, 452, 454; elementary 185 properties of, 428; exact definition of, 428 Stop rule, 59,60 166 Skew-symmetry, Strategic equivalence, 245,247, 248,272, Social xchangeeconomy, 9 ff. e 281,346,348, 373, 426,429, 472, 505, Social rder, 365 o 535,543;isomorphism of, 504,505 Social organizations, see Organizations, 117, 44,50,79, 80,84, 101, 119; Strategies,
Q ff

111,

social))

combination

of,

159)))

INDEX OF SUBJECTS))
Strategy,

641))

as move, 84; asymptotic, 210;Understandings, 223,224,237 c best, 124,517; hoiceof, 82, 145,147; Utilities, comparability of, 29; complete conceptof a, 79; found out, 151, 153, ordering of the, 19,26,29,604,617ff. ; domain of, 23,, differences of, 18, 631; 158,160,168;good, 108,146,160 ff., 161, 162,164,170,172,178, 179,183, 607; nonadditive, 250; non-numerical, 196,205, 206; higher order of a, 84; 16, 606, 607; numerical, 17 ff., 157, mixed, 143ff., 146,148,149,155,157, 605,606,617 ff.; numerical, substitut168,174,183,192,232,539,604; ability, 604;partially ordered,19,590; 161, optimal, 127,517; permanently optimal, 8, 604, system of, 26; transferability, 163,164,165;pure, 146,148,155,157, 606,608,629;variable, 560 168, 161, 181, statistical, 144,146; Utility, 8, 15,23,33,47, 83,156,556,563, 182; strict, 146; structure, continuous, 197; 565,569,572, 573, 583,585,603, 608, structure, fine, 197.; tructure, granular, s 616,617ff.; axiomatic treatment, 26 ff., 197 617 ff.; decreasing, 561,576; discrete, Strict Determinateness, see Determinate613; 60330;indivisible of the expected, generalization ness ff ; units, 609, concept f, o Struggle, 249 s 613, 614;marginal, 29, 30, 31; cale, fineness of, 616; Subpartition, 63,64,69 total, 34, 35 Subset, 61 Substitution rate, 465 Value, economic,252,467, 556,565;of a Superiority, intransitive notion of, 37 function, 88;of a play, seePlay, value of Superposition, 64 Variables, 12, 13, 88; aggregate of, 239; 61 \" Superset, alien,\" partial sets of, 12ff. Symmetry, 104,109,165,166,190,224, Vector, 129,140;addition, 130,253,254; 255,256,258,267, 315,446ff., 500,591. components, 129,404; coordinate, 129, Seealso Group 157; distance, 134; length, 134;oper129; quasi-components, ations, 404; 40 Tautological, 8, spaces,254;zero, 129 21 Temperature, 17, 296. See also Player; CoaliVictorious, Theory, extended, structure of the, 368 tion; Winning; Losing Theory, new, 526, 528. See also New Virtual existence, 36,45,338,484 theory 23 Thermodynamics, Wants, 10 Thermometry, 22 Wave mechanics, 148 Tie,125,315 Weights, 433,434,463;homogeneous, 435, Topology,154;384 444 Totalvalue, 251 Who finds out whom, 110 Transfer, 30,364,365,401,402 Winning, 296, 421, 426; certainly, 440; of utility, 8, 608 Transferability fully, 436 Transfinite induction, 269 Withdrawal, 364,366 23 Transformation, 22, 38,39,51, 89,590 Transitivity, 5 Zero-reducedform of characteristic funcTrees,66,67 Tribute, 30,402 tion, 545 Zero-sum condition, 345 Tug-of-war, 100 Zero-sum extension of r, 505, 506, 527, Umpire, 69,72, 84 531,538 Zero-sum restriction, 84, Uncertainty,

11;

35))

504)))