Game Theory Basics

Book Draft

Bernhard von Stengel

London School of Economics and Political Science

Department of Mathematics

October 7, 2020

Please send comments to [email protected]

Contents

Preface vii
Aims and Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Mathematical and Scholarly Level . . . . . . . . . . . . . . . . . . . ix
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

1 Nim and Combinatorial Games 1

1.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 1
1.2 Nim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Top-down Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Game Sums and Equivalence of Games . . . . . . . . . . . . . . . . 10
1.5 Nim, Poker Nim, and the Mex Rule . . . . . . . . . . . . . . . . . . 15
1.6 Sums of Nim Heaps . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.7 Finding Nim Values . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.8 A Glimpse of Partizan Games . . . . . . . . . . . . . . . . . . . . . 24
1.9 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.10 Exercises for Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Congestion Games 39
2.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 40
2.2 Introduction: The Pigou Network . . . . . . . . . . . . . . . . . . . 40
2.3 The Braess Paradox . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4 Definition of Congestion Games . . . . . . . . . . . . . . . . . . . . 44
2.5 Existence of Equilibrium in a Congestion Game . . . . . . . . . . . 46
2.6 Atomic and Splittable Flow, Price of Anarchy . . . . . . . . . . . . 48
2.7 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.8 Exercises for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . 50

3 Games in Strategic Form 53

3.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 54
3.2 Games in Strategic Form . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Best Responses and Equilibrium . . . . . . . . . . . . . . . . . . . . 57
3.4 Games with Multiple Equilibria . . . . . . . . . . . . . . . . . . . . 60
3.5 Dominated Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . 62

iii
iv Contents

3.6 The Cournot Duopoly of Quantity Competition . . . . . . . . . . . 65

3.7 Games without an Equilibrium . . . . . . . . . . . . . . . . . . . . 71
3.8 Symmetric Games with Two Strategies per Player . . . . . . . . . . 72
3.9 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.10 Exercises for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . 74

4 Game Trees with Perfect Information 77

4.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 78
4.2 Definition of Game Trees . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3 Backward Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4 Strategies in Game Trees . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5 Reduced Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.6 Subgame Perfect Equilibrium (SPE) . . . . . . . . . . . . . . . . . . 91
4.7 Commitment Games . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.8 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.9 Exercises for Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . 99

5 Expected Utility 103

5.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 105
5.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.3 Decisions Under Risk . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.4 Preferences for Lotteries . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.5 Ordinal Preferences for Decisions Under Certainty . . . . . . . . . 112
5.6 Cardinal Utility Functions and Simple Lotteries . . . . . . . . . . . 115
5.7 Consistency Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.8 Existence of an Expected-Utility Function . . . . . . . . . . . . . . 124
5.9 Risk Aversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.10 Discussion and Further Reading . . . . . . . . . . . . . . . . . . . . 132
5.11 Exercises for Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . 134

6 Mixed Strategy Equilibria 135

6.1 Prerequisites and Learning Objectives . . . . . . . . . . . . . . . . 136
6.2 Compliance Inspections . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.3 Bimatrix Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.4 The Best-Response Condition . . . . . . . . . . . . . . . . . . . . . 145
6.5 Existence of Mixed Equilibria . . . . . . . . . . . . . . . . . . . . . . 149
6.6 Finding Mixed Equilibria in Small Games . . . . . . . . . . . . . . 154
6.7 The Upper-Envelope Method . . . . . . . . . . . . . . . . . . . . . . 158
6.8 Degenerate Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.9 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.10 Exercises for Chapter 6 . . . . . . . . . . . . . . . . . . . . . . . . . 167
Contents v

7 Brouwer’s Fixed-Point Theorem 169

7.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 171
7.2 Labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7.3 Simplices and Triangulations . . . . . . . . . . . . . . . . . . . . . . 173
7.4 Sperner’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.5 The Knaster–Kuratowski–Mazurkiewicz Lemma . . . . . . . . . . 186
7.6 Brouwer’s Fixed-Point Theorem on a General Compact Convex Set 189
7.7 The Freudenthal Triangulation . . . . . . . . . . . . . . . . . . . . . 196
7.8 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.9 Exercises for Chapter 7 . . . . . . . . . . . . . . . . . . . . . . . . . 203

8 Zero-Sum Games 205

8.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 207
8.2 Example: Soccer Penalty . . . . . . . . . . . . . . . . . . . . . . . . 207
8.3 Max-Min and Min-Max Strategies . . . . . . . . . . . . . . . . . . . 211
8.4 A Short Proof of the Minimax Theorem . . . . . . . . . . . . . . . . 216
8.5 Further Notes on Zero-Sum Games . . . . . . . . . . . . . . . . . . 218
8.6 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.7 Exercises for Chapter 8 . . . . . . . . . . . . . . . . . . . . . . . . . 221

9 Geometry of Equilibria in Bimatrix Games 223

9.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 224
9.2 Labeled Best-Response Regions . . . . . . . . . . . . . . . . . . . . 224
9.3 The Lemke–Howson Algorithm . . . . . . . . . . . . . . . . . . . . 230
9.4 Using Best-Response Diagrams . . . . . . . . . . . . . . . . . . . . 235
9.5 Strategic Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . 239
9.6 Best-Response Polyhedra and Polytopes . . . . . . . . . . . . . . . 245
9.7 Complementary Pivoting . . . . . . . . . . . . . . . . . . . . . . . . 250
9.8 Degeneracy Resolution . . . . . . . . . . . . . . . . . . . . . . . . . 256
9.9 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
9.10 Exercises for Chapter 9 . . . . . . . . . . . . . . . . . . . . . . . . . 260

10 Game Trees with Imperfect Information 263

10.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 264
10.2 Information Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
10.3 Extensive Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.4 Strategies for Extensive Games and the Strategic Form . . . . . . . 272
10.5 Reduced Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
10.6 Perfect Recall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
10.7 Behavior Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
10.8 Kuhn’s Theorem: Behavior Strategies Suffice . . . . . . . . . . . . . 283
vi Contents

10.9 Behavior Strategies in the Monty Hall Problem . . . . . . . . . . . 287

10.10 Subgames and Subgame Perfect Equilibria . . . . . . . . . . . . . . 291
10.11 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
10.12 Exercises for Chapter 10 . . . . . . . . . . . . . . . . . . . . . . . . . 296

11 Bargaining 299
11.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 300
11.2 Bargaining Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
11.3 Bargaining Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
11.4 The Nash Bargaining Solution . . . . . . . . . . . . . . . . . . . . . 305
11.5 Splitting a Unit Pie . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
11.6 The Ultimatum Game . . . . . . . . . . . . . . . . . . . . . . . . . . 311
11.7 Alternating Offers Over Two Rounds . . . . . . . . . . . . . . . . . 315
11.8 Alternating Offers Over Several Rounds . . . . . . . . . . . . . . . 319
11.9 Stationary Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
11.10 The Nash Bargaining Solution Via Alternating Offers . . . . . . . . 327
11.11 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
11.12 Exercises for Chapter 11 . . . . . . . . . . . . . . . . . . . . . . . . . 329

12 Correlated Equilibria 333

12.1 Prerequisites and Learning Outcomes . . . . . . . . . . . . . . . . . 333
12.2 Examples of Correlated Equilibria . . . . . . . . . . . . . . . . . . . 334
12.3 Incentive Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 337
12.4 Coarse Correlated Equilibrium . . . . . . . . . . . . . . . . . . . . . 341
12.5 Existence of a Correlated Equilibrium . . . . . . . . . . . . . . . . . 343
12.6 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
12.7 Exercises for Chapter 12 . . . . . . . . . . . . . . . . . . . . . . . . . 349

References 351

Index 358
Preface

This book is an introduction to the mathematics of non-cooperative game theory.

Each concept is explained in detail, starting from a main example, with a slow-
paced proof of each theorem. The book has been designed and tested for self-study,
and as an undergraduate course text with core chapters and optional chapters for
different audiences. It has been developed over 15 years for a one-semester course
on game theory at the London School of Economics and the distance learning
program of the University of London, attended each year by about 200 third-year
students in mathematics, economics, management, and other degrees. After
studying this book, a student who started from first-year mathematics (the basics
of linear algebra, analysis, and probability) will have a solid understanding of the
most important concepts and theorems of non-cooperative game theory.
The intended audience are primarily students in mathematics, computer
science, and mathematical economics. For mathematians, we provide complete
self-contained proofs (in an economics course, these may be used as reliable
background and reference material). For computer scientists, we introduce
important ideas in algorithmic game theory, such as traffic equilibria, the “parity
argument” for the existence and computation of Nash equilibria, and correlated
equilibria. For economists, we use some important economic models as examples,
such as Cournot’s quantity competition, which is the oldest formal definition
of a game. Commitment games are applied to Stackelberg leadership, and in
Chapter 11 to the iterated-offers bargaining model. However, given the many
good introductions to game theory for economists (for example, Gibbons, 1992, or
Osborne, 2004), economic applications are not central to this book.

Aims and Contents

The first aim of this book is to let the student become fluent in game theory. The
student will learn the modeling tools (such as game trees and the strategic form)
and methods how to analyze games (finding their equilibria). This is provided
in the core chapters on non-cooperative games: Chapter 3 on games in strategic
form introduces important games such as the Prisoner’s Dilemma or Matching
Pennies, and the concept of Nash equilibrium (in the book nearly always just called

vii
viii Preface

“equilibrium” for brevity). Chapter 4 treats game trees with perfect information.
Chapter 6 explains mixed strategies and mixed equilibrium. Game trees with
imperfect information are covered in Chapter 10.
A second aim is to provide the conceptual and mathematical foundations of
the theory. Chapter 5 explains how the concept of expected utility represents a
consistent preference for risky outcomes (and not “risk neutrality” for monetary
payoffs, a common confusion). Chapter 7 proves Brouwer’s fixed-point theorem,
used by Nash (1951) to show the existence of an equilibrium point. Chapter 8
gives a self-contained two-page proof of von Neumann’s minimax theorem for
zero-sum games. These chapters form independent “modules” that can be omitted
in a taught course that has less emphasis on mathematical proofs, and can be used
for reference. If only one proof from these chapters is presented, it should be the
short proof of the minimax theorem.
A third aim is to introduce ideas that every game theorist should know but which
are not normally taught, and which make this book special. They are mathematical
highlights, typically less known to economists:
• An accessible introduction to combinatorial games in Chapter 1, pioneered by
Berlekamp, Conway, and Guy (2001–2004), with a focus on impartial games
and the central game of Nim.
• Congestion games, with the famous Braess paradox where increasing network
capacity can worsen congestion, and where an equilibrium is found with the
help of a potential function (Chapter 2).
• An elegant constructive proof of Sperner’s lemma, used to prove Brouwer’s
fixed-point theorem, and the Freudenthal simplicial subdivision that works
naturally in any dimension (Sections 7.4 and 7.7 in Chapter 7).
• The geometry of two-player games (which is my own research specialty) and
the algorithm by Lemke and Howson (1964) for finding a Nash equilibrium,
which implies that generic games have an odd number of equilibria (Chapter 9).
• An introduction to correlated equilibria in Chapter 12, with an elegant existence
proof due to Hart and Schmeidler (1989) that does not require a fixed-point
theorem but only von Neumann’s minimax theorem.
These chapters are optional. I teach the course in variations but always include
Chapter 1 on combinatorial games; they are much enjoyed by students, who would
not encounter them in an introductory economics course on game theory. These
chapters give also a short introduction to algorithmic game theory for computer
science students.
The general emphasis of the book is to teach methods, not philosophy. Game
theorists tend to question and to justify the approaches they take, for example the
concept of Nash equilibrium, and the assumed common knowledge of all players
Preface ix

about the rules of the game. These questions are of course very important. In fact,
these are probably the very issues that require a careful validation in a practical
game-theoretic analysis. However, this problem is not remedied by a lengthy
discussion of why one should play Nash equilibrium.
I think that a student of game theory should first learn the central models of
the strategic and extensive form, and be fluent in analyzing them. That toolbox will
then be useful when comparing different game-theoretic models and the solutions
that they imply.

Mathematical and Scholarly Level

General mathematical prerequisites are the basic notions of linear algebra (vectors
and matrices), probability theory (independence, conditional probability), and
analysis (continuity, closed sets). Rather than putting them in a rarely read
appendix, important concepts are recalled where needed in text boxes labeled as
Background material .

For each chapter, the necessary mathematical background and the required
previous chapters are listed in a first section on prerequisites and learning objectives.
The mathematics of game theory is not technically difficult, but very conceptual,
and requires therefore a certain mathematical maturity. For example, combinatorial
games have a recursive structure, for which a generalization of the mathematical
induction known for natural numbers is appropriate, called “top-down induction”
and explained in Section 1.3.
Game-theoretic concepts have precise and often unfamiliar mathematical
definitions. In this book, each main concept is typically explained by means of a
detailed introductory example. We use the format of definitions, theorems and
proofs in order to be precise and to keep some technical details hidden in proofs.
The proofs are detailed and complete and can be studied line by line. The main
idea of each proof is conveyed by introductory examples, and geometric arguments
are supported by pictures wherever possible.
Great attention is given to details that help avoid unnecessary confusions.
For example, a random event with two real values x and y as outcomes will
be described with a probability p assigned to the second outcome y, so that the
interval [0, 1] for p corresponds naturally to the interval [x, y] of the expectation
(1 − p)x + p y of that event.
This book emphasizes accessibility, not generality. I worked hard (and enjoyed) on
presenting the most direct and overhead-free proof of each result. In studying these
proofs, students may encounter for the first time important ideas from topology,
convex analysis, and linear programming, and thus may become interested in the
more general mathematical theory of these subjects.
x Preface

Bibliographic references, historical discussions, and further extensions are

deferred to a final section in each chapter, to avoid distractions in the main text.
References to further reading may include popular science books that I found
relevant and interesting. This book is primarily directed at undergraduate students
and not at researchers, but I have tried to be historically accurate. For important
works in French or German I have also cited English translations where I could
find them.
References to statements inside this book are in upper case, like “Theorem 2.2”,
and to other works in lower case, like “theorem 13.6 of Roughgarden (2016)”.
I use both genders to refer to players in a game, in particular in two-player
games to distinguish them more easily. Usage of the pronoun “she” has become
natural, and (like “he”) it may stand collectively for “he or she”.
Each chapter concludes with a set of exercises. They test methods (such as how
to find mixed equilibria), or the understanding of mathematical concepts used in
proofs (in particular in Chapters 5 and 7). Complete solutions to the exercises are
available from the publisher for registered lecturers.

Acknowledgments

Some of the material for this book has been adapted and developed from a course
guide originally produced for the distance learning International Programme
offered by the University of London (http://london.ac.uk/). Rahul Savani and
George Zouros gave valuable comments on that first course guide. Many students,
some of them anonymously, have pointed out errors and needs for clarification in
the subsequent versions of lecture notes. I thank Paul Dütting, Aviezri Fraenkel,
Sergiu Hart, Urban Larsson, Richard Nowakowski, David Tranah, and Peyton
Young for further suggestions.
Martin Antonov is the current lead programmer of the Game Theory Explorer
software (http://www.gametheoryexplorer.org), a joint project with Rahul Savani
and (as part of the Gambit software) Theodore Turocy, for solving games in strategic
and extensive form. From Game Theory Explorer, one can export games into
the LaTeX formats used in this book for drawing payoff tables and game trees
(available on request).