INTRODUCTION TO

COMBINATORICS
SECOND EDITION
 DISCRETE
MATHEMATICS
ITS APPLICATIONS
R. B. J. T. Allenby and Alan Slomson, How to Count: An Introduction to Combinatorics,
Third Edition
Craig P. Bauer, Secret History: The Story of Cryptology
Jürgen Bierbrauer, Introduction to Coding Theory, Second Edition
Katalin Bimbó, Combinatory Logic: Pure, Applied and Typed
Katalin Bimbó, Proof Theory: Sequent Calculi and Related Formalisms
Donald Bindner and Martin Erickson, A Student’s Guide to the Study, Practice, and Tools of
Modern Mathematics
Francine Blanchet-Sadri, Algorithmic Combinatorics on Partial Words
Miklós Bóna, Combinatorics of Permutations, Second Edition
Miklós Bóna, Handbook of Enumerative Combinatorics
Miklós Bóna, Introduction to Enumerative and Analytic Combinatorics, Second Edition
Jason I. Brown, Discrete Structures and Their Interactions
Richard A. Brualdi and Dragos̆ Cvetković, A Combinatorial Approach to Matrix Theory and Its
Applications
Kun-Mao Chao and Bang Ye Wu, Spanning Trees and Optimization Problems
Charalambos A. Charalambides, Enumerative Combinatorics
Gary Chartrand and Ping Zhang, Chromatic Graph Theory
Henri Cohen, Gerhard Frey, et al., Handbook of Elliptic and Hyperelliptic Curve Cryptography
Charles J. Colbourn and Jeffrey H. Dinitz, Handbook of Combinatorial Designs, Second Edition
Abhijit Das, Computational Number Theory
Matthias Dehmer and Frank Emmert-Streib, Quantitative Graph Theory:
Mathematical Foundations and Applications
Martin Erickson, Pearls of Discrete Mathematics
Martin Erickson and Anthony Vazzana, Introduction to Number Theory
 Titles (continued)

Steven Furino, Ying Miao, and Jianxing Yin, Frames and Resolvable Designs: Uses,
Constructions, and Existence
Mark S. Gockenbach, Finite-Dimensional Linear Algebra
Randy Goldberg and Lance Riek, A Practical Handbook of Speech Coders
Jacob E. Goodman and Joseph O’Rourke, Handbook of Discrete and Computational Geometry,
Second Edition
Jonathan L. Gross, Combinatorial Methods with Computer Applications
Jonathan L. Gross and Jay Yellen, Graph Theory and Its Applications, Second Edition
Jonathan L. Gross, Jay Yellen, and Ping Zhang Handbook of Graph Theory, Second Edition
David S. Gunderson, Handbook of Mathematical Induction: Theory and Applications
Richard Hammack, Wilfried Imrich, and Sandi Klavžar, Handbook of Product Graphs,
Second Edition
Darrel R. Hankerson, Greg A. Harris, and Peter D. Johnson, Introduction to Information Theory
and Data Compression, Second Edition
Darel W. Hardy, Fred Richman, and Carol L. Walker, Applied Algebra: Codes, Ciphers, and
Discrete Algorithms, Second Edition
Daryl D. Harms, Miroslav Kraetzl, Charles J. Colbourn, and John S. Devitt, Network Reliability:
Experiments with a Symbolic Algebra Environment
Silvia Heubach and Toufik Mansour, Combinatorics of Compositions and Words
Leslie Hogben, Handbook of Linear Algebra, Second Edition
Derek F. Holt with Bettina Eick and Eamonn A. O’Brien, Handbook of Computational Group Theory
David M. Jackson and Terry I. Visentin, An Atlas of Smaller Maps in Orientable and
Nonorientable Surfaces
Richard E. Klima, Neil P. Sigmon, and Ernest L. Stitzinger, Applications of Abstract Algebra
with Maple™ and MATLAB®, Second Edition
Richard E. Klima and Neil P. Sigmon, Cryptology: Classical and Modern with Maplets
Patrick Knupp and Kambiz Salari, Verification of Computer Codes in Computational Science
and Engineering
William L. Kocay and Donald L. Kreher, Graphs, Algorithms, and Optimization, Second Edition
Donald L. Kreher and Douglas R. Stinson, Combinatorial Algorithms: Generation Enumeration
and Search
Hang T. Lau, A Java Library of Graph Algorithms and Optimization
C. C. Lindner and C. A. Rodger, Design Theory, Second Edition
San Ling, Huaxiong Wang, and Chaoping Xing, Algebraic Curves in Cryptography
Nicholas A. Loehr, Bijective Combinatorics
Toufik Mansour, Combinatorics of Set Partitions
 Titles (continued)

Toufik Mansour and Matthias Schork, Commutation Relations, Normal Ordering, and Stirling
Numbers
Alasdair McAndrew, Introduction to Cryptography with Open-Source Software
Elliott Mendelson, Introduction to Mathematical Logic, Fifth Edition
Alfred J. Menezes, Paul C. van Oorschot, and Scott A. Vanstone, Handbook of Applied
Cryptography
Stig F. Mjølsnes, A Multidisciplinary Introduction to Information Security
Jason J. Molitierno, Applications of Combinatorial Matrix Theory to Laplacian Matrices of Graphs
Richard A. Mollin, Advanced Number Theory with Applications
Richard A. Mollin, Algebraic Number Theory, Second Edition
Richard A. Mollin, Codes: The Guide to Secrecy from Ancient to Modern Times
Richard A. Mollin, Fundamental Number Theory with Applications, Second Edition
Richard A. Mollin, An Introduction to Cryptography, Second Edition
Richard A. Mollin, Quadratics
Richard A. Mollin, RSA and Public-Key Cryptography
Carlos J. Moreno and Samuel S. Wagstaff, Jr., Sums of Squares of Integers
Gary L. Mullen and Daniel Panario, Handbook of Finite Fields
Goutam Paul and Subhamoy Maitra, RC4 Stream Cipher and Its Variants
Dingyi Pei, Authentication Codes and Combinatorial Designs
Kenneth H. Rosen, Handbook of Discrete and Combinatorial Mathematics
Yongtang Shi, Matthias Dehmer, Xueliang Li, and Ivan Gutman, Graph Polynomials
Douglas R. Shier and K.T. Wallenius, Applied Mathematical Modeling: A Multidisciplinary
Approach
Alexander Stanoyevitch, Introduction to Cryptography with Mathematical Foundations and
Computer Implementations
Jörn Steuding, Diophantine Analysis
Douglas R. Stinson, Cryptography: Theory and Practice, Third Edition
Roberto Tamassia, Handbook of Graph Drawing and Visualization
Roberto Togneri and Christopher J. deSilva, Fundamentals of Information Theory and Coding
Design
W. D. Wallis, Introduction to Combinatorial Designs, Second Edition
W. D. Wallis and J. C. George, Introduction to Combinatorics, Second Edition
Jiacun Wang, Handbook of Finite State Based Models and Applications
Lawrence C. Washington, Elliptic Curves: Number Theory and Cryptography, Second Edition
 DISCRETE MATHEMATICS AND ITS APPLICATIONS

INTRODUCTION TO
COMBINATORICS
SECOND EDITION

Walter D. Wallis
southern illinois university
CarbonDale, usa

John C. GeorGe
GorDon ColleGe, barnesville
GeorGia, usa
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2017 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

Version Date: 20161110

International Standard Book Number-13: 978-1-4987-7760-5 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access
www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc.
(CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization
that provides licenses and registration for a variety of users. For organizations that have been granted
a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data

Names: Wallis, W. D.. author. | George, J. C. (John Clay), 1959- author.

Title: Introduction to combinatorics / Walter D. Wallis and John C. George.
Description: Second edition. | Boca Raton : Taylor & Francis, 2017.
Identifiers: LCCN 2016030318 | ISBN 9781498777605
Subjects: LCSH: Combinatorial analysis–Textbooks.
Classification: LCC QA164 .W35 2017 | DDC 511/.6–dc23
LC record available at https://lccn.loc.gov/2016030318

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
For Our Families
Contents

List of Figures xv

Preface xvii

1 Introduction 1
1.1 Some Combinatorial Examples . . . . . . . . . . . . . . . . . 1
1.2 Sets, Relations, and Proof Techniques . . . . . . . . . . . . . 13
1.3 Two Principles of Enumeration . . . . . . . . . . . . . . . . . 15
1.4 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Systems of Distinct Representatives . . . . . . . . . . . . . . 20
Exercises 1A . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Exercises 1B . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2 Fundamentals of Enumeration 27
2.1 Permutations and Combinations  . . . . . . . . . . . . . . . . 27
n
2.2 Applications of P (n, k) and . . . . . . . . . . . . . . . . 29
k
2.3 Permutations and Combinations of Multisets . . . . . . . . . 32
2.4 Applications and Subtle Errors . . . . . . . . . . . . . . . . . 36
2.5 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Exercises 2A . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Exercises 2B . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3 Probability 45
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Some Definitions and Easy Examples . . . . . . . . . . . . . 45
3.3 Events and Probabilities . . . . . . . . . . . . . . . . . . . . 48
3.4 Three Interesting Examples . . . . . . . . . . . . . . . . . . . 52
3.5 Probability Models . . . . . . . . . . . . . . . . . . . . . . . 54
3.6 Bernoulli Trials . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.7 The Probabilities in Poker . . . . . . . . . . . . . . . . . . . 56
3.8 The Wild Card Poker Paradox . . . . . . . . . . . . . . . . . 57
Exercises 3A . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Exercises 3B . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

ix
x Contents

4 The Pigeonhole Principle and Ramsey’s Theorem 65

4.1 The Pigeonhole Principle . . . . . . . . . . . . . . . . . . . . 65
4.2 Applications of the Pigeonhole Principle . . . . . . . . . . . 66
4.3 Ramsey’s Theorem—The Graphical Case . . . . . . . . . . . 69
4.4 Ramsey Multiplicity . . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Sum-Free Sets . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.6 Bounds on Ramsey Numbers . . . . . . . . . . . . . . . . . . 77
4.7 The General Form of Ramsey’s Theorem . . . . . . . . . . . 81
Exercises 4A . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Exercises 4B . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5 The Principle of Inclusion and Exclusion 87

5.1 Unions of Events . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2 The Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.3 Combinations with Limited Repetitions . . . . . . . . . . . . 93
5.4 Derangements . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Exercises 5A . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Exercises 5B . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6 Generating Functions and Recurrence Relations 103

6.1 Generating Functions . . . . . . . . . . . . . . . . . . . . . . 103
6.2 Recurrence Relations . . . . . . . . . . . . . . . . . . . . . . 107
6.3 From Generating Function to Recurrence . . . . . . . . . . . 112
6.4 Exponential Generating Functions . . . . . . . . . . . . . . . 113
Exercises 6A . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Exercises 6B . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

7 Catalan, Bell, and Stirling Numbers 121

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.2 Catalan Numbers . . . . . . . . . . . . . . . . . . . . . . . . 122
7.3 Stirling Numbers of the Second Kind . . . . . . . . . . . . . 126
7.4 Bell Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.5 Stirling Numbers of the First Kind . . . . . . . . . . . . . . . 132
7.6 Computer Algebra and Other Electronic Systems . . . . . . 135
Exercises 7A . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Exercises 7B . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
 Contents xi

8 Symmetries and the Pólya–Redfield Method 143

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
8.2 Basics of Groups . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.3 Permutations and Colorings . . . . . . . . . . . . . . . . . . 150
8.4 An Important Counting Theorem . . . . . . . . . . . . . . . 151
8.5 The Pólya and Redfield Theorem . . . . . . . . . . . . . . . 154
Exercises 8A . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Exercises 8B . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

9 Partially Ordered Sets 163

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9.2 Examples and Definitions . . . . . . . . . . . . . . . . . . . . 164
9.3 Bounds and Lattices . . . . . . . . . . . . . . . . . . . . . . . 166
9.4 Isomorphism and Cartesian Products . . . . . . . . . . . . . 168
9.5 Extremal Set Theory: Sperner’s and Dilworth’s Theorems . 171
Exercises 9A . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
Exercises 9B . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

10 Introduction to Graph Theory 177

10.1 Degrees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
10.2 Paths and Cycles in Graphs . . . . . . . . . . . . . . . . . . 180
10.3 Maps and Graph Coloring . . . . . . . . . . . . . . . . . . . 183
Exercises 10A . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Exercises 10B . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

11 Further Graph Theory 193

11.1 Euler Walks and Circuits . . . . . . . . . . . . . . . . . . . . 193
11.2 Application of Euler Circuits to Mazes . . . . . . . . . . . . 198
11.3 Hamilton Cycles . . . . . . . . . . . . . . . . . . . . . . . . . 200
11.4 Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
11.5 Spanning Trees . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Exercises 11A . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Exercises 11B . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

12 Coding Theory 223

12.1 Errors and Noise . . . . . . . . . . . . . . . . . . . . . . . . . 223
12.2 The Venn Diagram Code . . . . . . . . . . . . . . . . . . . . 224
12.3 Binary Codes, Weight, and Distance . . . . . . . . . . . . . . 226
12.4 Linear Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
12.5 Hamming Codes . . . . . . . . . . . . . . . . . . . . . . . . . 232
12.6 Codes and the Hat Problem . . . . . . . . . . . . . . . . . . 234
xii Contents

12.7 Variable-Length Codes and Data Compression . . . . . . . . 235

Exercises 12A . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Exercises 12B . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

13 Latin Squares 241

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
13.2 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . 245
13.3 Idempotent Latin Squares . . . . . . . . . . . . . . . . . . . . 251
13.4 Partial Latin Squares and Subsquares . . . . . . . . . . . . . 253
13.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
Exercises 13A . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Exercises 13B . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

14 Balanced Incomplete Block Designs 265

14.1 Design Parameters . . . . . . . . . . . . . . . . . . . . . . . . 266
14.2 Fisher’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . 270
14.3 Symmetric Balanced Incomplete Block Designs . . . . . . . . 272
14.4 New Designs from Old . . . . . . . . . . . . . . . . . . . . . 274
14.5 Difference Methods . . . . . . . . . . . . . . . . . . . . . . . 276
Exercises 14A . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Exercises 14B . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282

15 Linear Algebra Methods in Combinatorics 285

15.1 Recurrences Revisited . . . . . . . . . . . . . . . . . . . . . . 285
15.2 State Graphs and the Transfer Matrix Method . . . . . . . . 287
15.3 Kasteleyn’s Permanent Method . . . . . . . . . . . . . . . . 294
Exercises 15A . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Exercises 15B . . . . . . . . . . . . . . . . . . . . . . . . . . 301
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302

Appendix A: Sets; Proof Techniques 305

AA.1 Sets and Basic Set Operations . . . . . . . . . . . . . . . . . 305
AA.2 The Principle of Mathematical Induction . . . . . . . . . . . 313
AA.3 Some Applications of Induction . . . . . . . . . . . . . . . . 315
AA.4 Binary Relations on Sets . . . . . . . . . . . . . . . . . . . . 317
Exercises A . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Exercises B . . . . . . . . . . . . . . . . . . . . . . . . . . . 320

Appendix B: Matrices and Vectors 325

B.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
B.2 Vector and Matrix Products . . . . . . . . . . . . . . . . . . . 328
B.3 Inverses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
 Contents xiii

B.4 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . 332

Exercises A . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Exercises B . . . . . . . . . . . . . . . . . . . . . . . . . . . 334

Appendix C: Some Combinatorial People 339

Solutions to Set A Exercises 347

Hints for Problems 377

Solutions to Problems 383

References 407

Index 417
List of Figures

1.1 Pancake with three slices. . . . . . . . . . . . . . . . . . . . . 3

1.2 A Sudoku puzzle. . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 The Tower of Hanoi puzzle. . . . . . . . . . . . . . . . . . . . 6
1.4 Königsberg and its bridges. . . . . . . . . . . . . . . . . . . . 11
1.5 Some Venn diagrams . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 K6 and K3,4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 A graph and its complement. . . . . . . . . . . . . . . . . . . 19

2.1 How many ways to walk? . . . . . . . . . . . . . . . . . . . . 31

3.1 Tree diagrams for coin tossing. . . . . . . . . . . . . . . . . . 49

3.2 Tree diagrams for Examples 3.6 and 3.7. . . . . . . . . . . . . 50

4.1 K5 in two colors. . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.2 Decomposition of K8 proving R(3, 4) ≥ 9. . . . . . . . . . . . 71

5.1 Enumerating X ∪ Y using (5.1). . . . . . . . . . . . . . . . . 87

5.2 Illustration of (5.2) (left) and (5.3) (right). . . . . . . . . . . . 89

7.1 Triangulating the hexagon. . . . . . . . . . . . . . . . . . . . 122

7.2 Rooted binary trees with n internal points. . . . . . . . . . . 125

8.1 Square with numbered corners. . . . . . . . . . . . . . . . . . 145

8.2 Three hexagons with labeled vertices and sides. . . . . . . . . 149
8.3 Two figures for coloring. . . . . . . . . . . . . . . . . . . . . . 153
8.4 Coloring corners or edges. . . . . . . . . . . . . . . . . . . . . 155
8.5 Square with eight points to color. . . . . . . . . . . . . . . . . 162

9.1 Poset of subsets of a set of three elements. . . . . . . . . . . . 164

9.2 Poset of subgroups of Z12 , ordered by containment. . . . . . . 165
9.3 Positive divisors of 30 ordered by divisibility. . . . . . . . . . 165
9.4 Poset with no unique minimum or maximum element. . . . . 167
9.5 A lattice that is not distributive. . . . . . . . . . . . . . . . . 169
9.6 Two posets with almost the same Hasse diagram. . . . . . . . 170
9.7 The Hasse diagram for B1 × B1 and for B2 . . . . . . . . . . . 170
9.8 Poset for exercises and problems. . . . . . . . . . . . . . . . . 175

xv
xvi List of Figures

10.1 Testing degree sequences. . . . . . . . . . . . . . . . . . . . . 178

10.2 Tabular form of Dijkstra’s algorithm. . . . . . . . . . . . . . . 184
10.3 Two representations of K4 . . . . . . . . . . . . . . . . . . . . 185

11.1 The original Königsberg bridges. . . . . . . . . . . . . . . . . 193

11.2 The Königsberg bridges today. . . . . . . . . . . . . . . . . . 197
11.3 Find an efficient eulerization. . . . . . . . . . . . . . . . . . . 198
11.4 Graphical representation of a maze. . . . . . . . . . . . . . . . 199
11.5 Two graphs from the map. . . . . . . . . . . . . . . . . . . . . 200
11.6 A labeling of K4 . . . . . . . . . . . . . . . . . . . . . . . . . . 203
11.7 All Hamilton cycles of the K4 . . . . . . . . . . . . . . . . . . 204
11.8 Some trees. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
11.9 Illustration of Lemma 11.3. . . . . . . . . . . . . . . . . . . . 205
11.10Graphs for counting trees. . . . . . . . . . . . . . . . . . . . . 210
11.11Counting trees. . . . . . . . . . . . . . . . . . . . . . . . . . . 212

12.1 Venn diagram encoding. . . . . . . . . . . . . . . . . . . . . . 225

12.2 Venn diagram decoding. . . . . . . . . . . . . . . . . . . . . . 226
12.3 The codewords in Venn diagram encoding. . . . . . . . . . . . 227
12.4 The Morse code . . . . . . . . . . . . . . . . . . . . . . . . . . 236

13.1 A golf design for seven clubs. . . . . . . . . . . . . . . . . . . 258

13.2 Array to generate a golf design for 17 teams. . . . . . . . . . 259

14.1 Residual and derived designs. . . . . . . . . . . . . . . . . . . 276

14.2 Some difference sets. . . . . . . . . . . . . . . . . . . . . . . . 278
14.3 Some supplementary difference sets. . . . . . . . . . . . . . . 279

15.1 Gear shift graph. . . . . . . . . . . . . . . . . . . . . . . . . . 287

15.2 State graph and adjacency matrix. . . . . . . . . . . . . . . . 288
15.3 A diagram and two adjacency matrices. . . . . . . . . . . . . 289
15.4 Digraph and adjacency matrix for the string problem. . . . . 291
15.5 Digraph and adjacency matrix. . . . . . . . . . . . . . . . . . 292
15.6 Digraph and adjacency matrix for domino tiling. . . . . . . . 294
A.1 R ∩ (S ∪ T ) = (R ∩ S) ∪ (R ∩ T ). . . . . . . . . . . . . . . . . 310
Preface

This book is a text for introductory courses on combinatorics, usually offered

around the junior year of college. The audience for such a course consists pri-
marily of mathematics majors, although the course is often taken by computer
science students and occasionally by electrical engineering students.
The preparation of the students for these courses is very mixed, and one
cannot assume a strong background. In some cases this will be the first
course where the students see several real proofs, while others will have a good
background in linear algebra, will have completed the calculus stream, and will
have started abstract algebra. For this reason we have included two chapters
on background material—one on sets, induction and proof techniques, and one
on vectors and matrices. We have made them appendices so that students can
get straight to the meat of the course; the better-prepared students can ignore
the appendices entirely, while others can refer to them as needed.
Some textbooks contain biographical material on the mathematicians who
invented the concepts that are studied, but students are often irritated by
the interruption. Others omit this; the reader misses out on some interesting
information, and a different set of students is unhappy. In an attempt to get
around this problem, we have included some brief sketches, but have made
them into a separate section of the book– a third appendix. We refer readers
to the biographies when relevant to the topic being studied.
Not all books discuss the increasing role played by technology (computer
algebra systems, among many other examples); we have added a discussion
of Maple TM and Mathematica TM where appropriate and commented on other
technological tools (e.g., spreadsheets and calculators). These are not empha-
sized, so that students or instructors who prefer not to have them do not need
access to any of these tools. Similarly we have used more modern examples
for some of the traditional problems and exercises (e.g., digital music tracks
in MP3 players, rather than books on a shelf).

What Is Combinatorics Anyway?

Broadly speaking, combinatorics is the branch of mathematics that deals
with different ways of selecting objects from a set or arranging objects. It
tries to answer two major kinds of questions, namely, counting questions
(how many ways can a selection or arrangement be chosen with a particu-
lar set of properties?) and structural questions (does there exist a selection
or arrangement of objects with a particular set of properties?).

xvii
xviii Preface

The Book: Problems and Exercises

Exercises are collected at the end of the chapters. They are divided into
three parts. Exercise sets A and B are fairly straightforward, and the only
difference is that answers and partial or complete solutions are provided for
the A exercises. Many textbooks provide answers for every second exercise or
for what appears to be a random selection, but we have found that the “two
sets” model works well in the classroom. There is also a set of “Problems.”
These contain some more difficult or more sophisticated questions, and also a
number of exercises where the student is asked to provide a formal proof when
a result has been treated informally in the text. We have provided solutions
for some Problems, and hints for some others. There are also Exercises, but
not Problems, in the Appendices on background material.

The Second Edition

We have introduced two new chapters, one on probability and the other on
posets. We have also attempted to correct errors from the first edition, and
would like to thank several of our readers for pointing these out. In addition,
there are numerous new illustrations, exercises, and problems.

The Book: Outline of the Chapters

We start by briefly discussing several examples of typical combinatorial
problems to give the reader a better idea of what the subject covers. Chapter
1 also contains some basic information on sets, proof techniques, enumeration,
and graph theory, topics that will reappear frequently in the book. The next
few chapters explore enumerative ideas, including the pigeonhole principle
and inclusion/exclusion, and also probability.
In Chapters 6 through 8 we explore enumerative functions and the relations
between them. There are chapters on generating functions and recurrences,
on some important families of functions, or numbers (Catalan, Bell, and Stir-
ling numbers), and the theorems of Pólya and Redfield. These chapters also
contain brief introductions to computer algebra and group theory.
The next six chapters study structures of particular interest in combina-
torics: posets, graphs, codes, Latin squares, and experimental designs. The
last chapter contains further discussion of the interaction between linear al-
gebra and combinatorics.
We conclude with the appendices mentioned earlier, solutions and hints,
and references.

Acknowledgments
This book has benefited significantly from the comments of Ken Rosen,
and those of anonymous reviewers. We would also like to thank Robert Ross,
 Preface xix

David Grubbs, and Bob Stern of Taylor & Francis Group for their assistance.
Every edition of any textbook will, unavoidably, contain errors. Each of
us wishes to state categorically that all such errors are the fault of the other
author.
Chapter 1
Introduction

Broadly speaking, combinatorics is the branch of mathematics that deals with

different ways of selecting objects from a set or arranging objects. It tries to
answer two major kinds of questions, namely the existence question (Does
there exist a selection or arrangement of objects with a particular set of prop-
erties?) and the enumerative question (How many ways can a selection or
arrangement be chosen with a particular set of properties?). But you may be
surprised by the depth of problems that arise in combinatorics.
The main point to remember is that it really doesn’t matter what sort
of objects are being discussed. For example, we shall often assume that we
are talking about sets of numbers, and sometimes use their arithmetical or
algebraic properties. But these methods are used to prove results that we
then apply to all sorts of objects.
In the first section we shall show you a few examples of combinatorial
problems. In Section 1.2 we briefly summarize some ideas of and notations
of set theory (if you need to review this material, see Appendix A). The
remainder of the chapter introduces some basic combinatorial ideas.

1.1 Some Combinatorial Examples

Some of these problems have a recreational flavor, puzzles and so on, be-
cause they will be more familiar; but all these ideas have very serious appli-
cations. We address many of them in more detail in subsequent chapters.

Passwords
We start with an enumerative problem. Enumeration (the theory of count-
ing) is an important part of combinatorics.
Most computer systems require passwords, in order to protect your informa-
tion and your privacy. For example, the social networking program YoufaceTM
requires everybody to have an eight-character password made up of letters and
digits. The passwords are case-sensitive, but the letter O is not allowed (to
avoid confusion with zero), so there are 60 symbols available. How many
different passwords could you choose?

1
2 Introduction to Combinatorics

There are 60 choices for the first character. For each of those there are
60 possible second characters, for 3600 possible 2-character starts, and so on.
In all, there are 608 , or about 168 trillion, passwords. This calculation is an
example of the multiplication principle, which we’ll discuss further in Section
1.3, later in this chapter.
Suppose a hacker wants to break into your Youface account. She has a
program that can generate and test a thousand passwords per second. Is she
dangerous?
Well, if she tests every possibility, it will take her over 5,000 years. So no,
your password is pretty secure.
However, you need to be careful about passwords. Dr. John Walters, a
computer science professor that we shall meet again in this volume, always
uses the login jwalters. Having a poor memory, he always starts his password
with his initials, jw, and ends with the last two digits of the year. The hacker
found this out, so in 2009 she worked out that his password had the form
jw****09. There are 604 = 12, 360, 000 passwords of this form. That sounds
like a lot, but in 3.6 hours she could run every possible password. Even if
she is very unlucky, and her computer does not find the actual password until
very late in the run, she could still hack into his account in an afternoon.
In order to protect people like Dr. Walters, Youface introduced some further
requirements. Your password cannot begin with the same two symbols as your
username, and the last two digits cannot be the last two digits of the year. In
2010, when he rebuilt his account, Dr. Walters could not choose any password
of the form jw****** or ******10.
How many possible passwords remained? We start with 608 possibilities,
and subtract the number of passwords that were banned. There were 608
passwords originally. There are 606 passwords of type jw******, and 606 of
type ******10. Subtracting these numbers leaves 608 − 2 × 606 . However, we
have taken away some passwords twice: those of form jw****10. There are
604 passwords like this, so we have “oversubtracted” 604 from the total. So
the final result is 608 − 2 × 606 + 604 .
This method of counting—subtract all objects in one class, do the same
with those in another class, then add back those that were common to both
classes—is the first case of the Principle of Inclusion and Exclusion. We shall
deal with this further in Chapter 5.

The Pancake Problem

Another good example of a combinatorial problem is this: Into how many
regions can the plane be divided by n straight lines, given that no lines are
parallel and at most two lines intersect at any point? These conditions ensure
the maximum number of regions. This is sometimes called the Pancake Prob-
lem because we may draw a large enough circle (the “pancake”) surrounding
all points of intersection of the lines, so that the problem may be described
 Introduction 3

as: What is the maximum number of pieces remaining after cutting a pancake
with n cuts (none parallel and no three concurrent)?

2
1
3
5
4
6
7

FIGURE 1.1: Pancake with three slices.

The first thing we do is define a notation and try a few examples. This
is almost always a good start to any kind of problem. We let Pn be the
maximum number of pieces. Then it is easy to see that P0 = 1 (with no cuts,
the pancake is still one piece) and P1 = 2. When we notice that P2 = 4,
we see the progression 1, 2, 4 and wonder whether we might have powers of
two; perhaps, we think, Pn = 2n ? This would make sense if each cut passes
through every region, so that each region is cut into two. However, this does
not happen, and we find P3 = 7, as we can see from Figure 1.1. A few more
tries will give us P4 = 11 and P5 = 16. Clearly, another approach is needed.
How many new regions do we get by adding a new cut? If we have n − 1
cuts in the pancake, and add another cut, the rules require that the new cut
must intersect each of the n − 1 old cuts. If we start the cut on the edge of
the pancake in the middle of a region, we cut along the pancake and cut off a
new region when we reach the first of the n − 1 old cuts. Then as we continue
cutting in a straight line, we intersect each of the other old cuts, cutting off a
new region when we do. Then at the end, we reach the edge of the pancake,
and we cut off a final piece. This means that we have added n pieces; one
for each of the original n − 1 cuts, and one when we reach the edge of the
pancake. This tells us that Pn = Pn−1 + n. This is a formula for Pn , but not
a nice one because we cannot easily calculate, for example, P100 without a
great deal of effort. We will work a little harder and see whether we can find
a more efficient formula.
We start by repeating the formula for larger values of n. We get P1 = P0 +1,
and P2 = P1 +2 = P0 +1+2, then P3 = P0 +1+2+3, P4 = P0 +1+2+3+4, and
so on. The pattern becomes clear, and we write Pn = P0 +(1+2+· · ·+n). Now
we nearly have our answer. The reader may be familiar with the formula for
1 + 2 + · · · + n or of the first n terms of any arithmetic progression; the story
goes that the eminent mathematician Carl Friedrich Gauss (see Appendix
C) found the sum of the first one hundred integers almost immediately by
realizing how the formula worked (for more details, see [16], p. 509). The
4 Introduction to Combinatorics

“two-rows proof” is as follows:

1 2 3 ... n−1 n
n n−1 n−2 ... 2 1

Each of these rows clearly sums to the same value, S. If we add the columns,
we find that each column adds to n+1, and there are n columns; thus the sum
of all elements in both rows is 2S = n(n+1). Thus 1+2+· · ·+n = n(n+1)/2.
This gives us, finally, our formula for Pn , the maximum number of pieces
after n cuts.
n(n + 1)
Pn = 1 +
2

Sudoku
Over the last few years we have seen a great surge of interest in the Sudoku
puzzle. For those few readers who have been vacationing on another planet,
the puzzle consists of a 9 × 9 array, partitioned into 3 × 3 subarrays (sub-
squares), in which some cells contain entries chosen from {1, 2, 3, 4, 5, 6, 7, 8, 9}.
The object is to complete the array so that every row, every column and every
subarray contains each symbol precisely once.
Sudoku was apparently invented by an American architect, Howard Garns
(see Appendix C), in 1979, but did not achieve great popularity until 1986,
when it started appearing in Japanese publications. Worldwide circulation
dates to about 2004. A nice history can be found in [137].
We’ll use the phrase Sudoku square for a completed 9 × 9 square that could
be the solution to a Sudoku puzzle. In a well-posed puzzle, there is one and
only one Sudoku square that completes the original partial array.
Figure 1.2 shows a typical Sudoku puzzle in the top left-hand position. It
needs a 1 in column 3. There are already 1s in the top left and bottom left
subsquares, so there is only one possible space in the column; the top right
array shows the 1 in place.
Now look at the top right subsquare. It needs a 2, 4, 5, and 7. The only
possible place for a 4 in this subsquare is the middle top. At first, it looks
as though the 5 can go in any of the other three vacant cells. However, the
second row is not available (the 2 must go there), and the 7 must go in the
last column (there is already a 7 in the seventh column, and the 4 filled up
the eight column). So the 5 only has one possible home. The completion of
the top right subsquare is shown in the lower left copy.
Continuing in this way, we obtain the Sudoku square shown at the lower
right of the figure.
The methods used in solving Sudoku are combinatorial in nature. However,
not many of the millions who play Sudoku realize that a Sudoku square is an
example of a combinatorial object, called a Latin square, that is widely used
in statistical and geometric applications. We shall study Latin squares in
Chapter 13.
 Introduction 5

9 6 2 8 1 9 6 2 8
1
1 8 3 1 8 3
2 1 6 9 2 1 6 9
7 8 4 3 7 8 4 3
5 9 4 6 5 9 4 6
6 3 7 8 6 1 3 7 8
5 4 7 8 5 4 7 8
1 9 5 1 9 5
3 2 5 9 4 3 2 5 9 4

9 6 2 8 5 4 1 9 7 3 6 2 8 5 4 1
1 2 8 3 6 1 5 9 4 7 2 8 3
2 1 6 9 7 8 4 2 5 3 1 6 9 7
7 8 4 3 7 2 8 1 6 4 9 3 5
5 9 4 6 5 3 9 8 7 2 4 1 6
6 1 3 7 8 4 6 1 3 9 5 7 2 8
5 4 7 8 2 5 4 7 1 3 8 6 9
1 9 5 1 9 7 4 8 6 3 5 2
3 2 5 9 4 3 8 6 2 5 9 1 7 4
FIGURE 1.2: A Sudoku puzzle.

The Tower of Hanoi

Virtually every student in combinatorics and computer science runs into
the Tower of Hanoi problem. Because it is such a basic example of combina-
torial reasoning and so quickly yields to straightforward methods of attack,
and because it may be used to illustrate such ideas as recurrence relations and
recursive programming and has ties to such topics as Gray codes, it finds its
way into textbooks in programming, algorithms, and discrete and combinato-
rial mathematics. The puzzle itself predates the modern study of these topics;
it was first described in 1883 by Edouard Lucas (see Appendix C), who also
publicized the Fibonacci numbers (as discussed in Chapter 6). Lucas intro-
duced the puzzle under the name M. Claus. One year later, Henri de Parville
published the following fanciful legend concerning the puzzle [26]:
In the great temple at Benares, beneath the dome which marks
the center of the world, rests a brass plate in which are fixed three
diamond needles, each a cubit high and as thick as the body of a
bee. On one of these needles, at the creation, God placed sixty-
four discs of pure gold, the largest disk resting on the brass plate,
and the others getting smaller and smaller up to the top one. This
is the Tower of Bramah. Day and night unceasingly the priests
transfer the discs from one diamond needle to another according
to the fixed and immutable laws of Bramah, which require that
the priest on duty must not move more than one disc at a time
6 Introduction to Combinatorics

and that he must place this disc on a needle so that there is no

smaller disc below it. When the sixty-four discs shall have been
transferred from the needle on which at the creation God placed
them to one of the other needles, tower, temple, and Brahmins
alike will crumble into dust, and with a thunderclap the world will
vanish.
(translation by W. W. Rouse Ball [92]). The legend, oddly, places the towers
in a temple to the god Brahma in Benares (modern Varanasi, India), which
renders the reference to Hanoi (in Vietnam, over 1000 miles away) puzzling.
In essence, the puzzle is this; we have three “needles” or “spindles” and a
set of n differently sized disks with holes in the middle to allow the disks to
be placed over the needles. The disks are arranged on the leftmost needle in
order of diminishing size (smallest on the top). We are to move the entire
column to a different needle (say, the rightmost) according to the rules:
• At each play, precisely one disk is moved.
• No disk may be placed over a spindle that already has a smaller disk.
An illustration for the case n = 3 is shown below. Commercial versions of the
puzzle typically include from five to eight disks.
The problem is twofold; first, what is the most efficient way of moving the
tower? Second, how many plays does this most efficient way take for a tower
of n disks? One might add a third (or more properly, first ) question: Can it
be done at all? A careful analysis will answer all three questions.
First of all, observe that the problem is trivial in the cases of n = 1 (simply
move the disk from the starting needle to the ending needle) and n = 2 (move
the smallest disk to the middle needle; then the larger disk to the ending
needle; then place the smallest over the largest). Either of these may form the
basis case for a proof by mathematical induction. (If you are unfamiliar with
induction methods, or need to revise this material, see Section 1.2, below.)
Suppose that it is known that we can solve the puzzle for n disks, and we
have n+1 disks before us. We may use the solution for n disks to move the top
n disks to the middle needle, ignoring for the moment the largest disk. Now
we may freely move the largest disk from the leftmost needle to the rightmost
needle. At this point, we use the solution for n disks to move the n disks

FIGURE 1.3: The Tower of Hanoi puzzle.

 Introduction 7

from the middle needle to the rightmost needle. This will cause no problems
since the disk on the rightmost needle is the largest, so that any disk may be
placed on it without violating a rule.
It is at least intuitively clear that this technique gives us the most efficient
way to move a stack of n disks. Clearly the basis case uses the fewest possible
moves (one move for one disk, three moves for two disks). Also, it seems clear
that there is no shorter way to move n + 1 disks than by moving the stack
of n disks first. Without going through a rigorous proof, it is at least very
plausible that our approach uses the fewest moves possible.
Now the hard part: How many plays does this method take for n disks?
We denote this (still unknown) quantity by Hn . Clearly, we have H1 = 1 and
H2 = 3. For any n > 2, we can compute Hn = 2 · Hn−1 + 1. This follows
because we use Hn−1 plays to move the n − 1 disks to the middle needle, one
play to move the largest disk to the rightmost needle, and another Hn−1 plays
to move the n − 1 disks from the middle to the rightmost needle.
This kind of formula for Hn , like the formula for Pn of the previous example,
is called a recurrence because the expression we are solving for “recurs” in
the equation. This does not give us a simple formula for Hn , but it gives us
a way to calculate it. We will run through a few values.

n 1 2 3 4 5
Hn 1 3 7 15 31

At this point, we might be tempted to conjecture, based on the numbers

shown, that Hn is always one less than a power of two, or more precisely,
Hn = 2n − 1. How might we prove this? Again, proof by induction is the
easiest approach with which we are familiar (another method will be presented
in Chapter 6). The basis is n = 1; suppose that Hn = 2n − 1 and use the
recurrence to find Hn+1 , which our guess says should be 2n+1 − 1. The
recurrence tells us that Hn+1 = 2 · Hn + 1. The induction hypothesis allows
us to replace Hn by 2n − 1, yielding Hn+1 = 2 · (2n − 1) + 1. We simplify to
obtain Hn+1 = 2n+1 − 1, as required.

The Seven Wargamers

A group of friends (Adam, Beth, . . . ) meet frequently to play the wargame
Vampire Cat Wars, a game for four individual players (no partnerships) in
which the players are vampires who have to beat off attacks by the others.
They decide to conduct a tournament over several sessions. In each session
four of the players will take part in one game. They would like it if each
competitor played against each of the others equally—in this case let’s call
the competition equitable. Various questions come up: for example, what is
the minimum number of games in such a tournament?
When they first decided to organize a tournament, there were seven players,
initials A, B, C, D, E, F, G. At first, they thought maybe everyone could play
8 Introduction to Combinatorics

everyone else once. If Adam were going to play every other competitor exactly
once, he would play in two matchups; writing the initials of each player to
represent a game, the schedule could include ABCD and AEFG. But whom
will Beth play? One matchup is already given, ABCD. She cannot meet A,
C or D again, so the only possibility is BEFG. But a schedule cannot include
both AEFG and BEFG: pairs like EF would meet twice.
So they decided to try a schedule where everyone plays everyone else twice.
Everybody will have four games (each person must meet six opponents twice
each—12 appearances—and each game accounts for three). We may as well
assume Adam plays in ABCD. He must play against Beth again, and the two
matches AB... could have a total of four, three or two players in common.
If we try four players—that is, use the matchup ABCD again—Adam’s other
two games must be AEFG twice. But Beth has been scheduled against A, C
and D twice, so her other two matchups must each be BEFG, and EF (as well
as other combinations) has occurred too many times. Similarly, if there are
two games with three players in common, say ABCD and ABCE, we strike
problems: Adam plays twice more, and must meet both F and G twice. The
only possibility is ADFG, AEFG. But the same argument shows that we must
have BDFG and BEFG. Now F and G have played each other too often.
Okay, this argument shows that the two AB... matchups have no other
member in common, and the same must apply to all other pairs. So no
two games can have three players in common. Try to start with ABCD and
ABEF. Adam’s other two matchups have the form A...G, in order for him to
meet Greg twice. They cannot use B again, and ACDG would have three in
common with ABCD, so they try ACEG and ADFG (or ACFG and ADEG,
which would be equivalent, just exchange two names). The same argument,
avoiding three in common with the first two games, leads to either BCEG or
BCFG, and BCEG clashes with ACEG, so they chose BCFG and BDEG. The
final matchup is CDEF, and the schedule is

ABCD, ABEF, ACEG, ADFG, BCFG, BDEG, CDEF.

The players are very happy—and then Helga joins their group. What now?
And what if other players join up?
Let’s write v for the number of players, and say each pair plays together λ
times. Then each player has λ(v − 1) instances of an opponent. Each matchup
accounts for three of these. So each player takes part in λ(v − 1)/3 games.
You cannot schedule partial games, so λ(v − 1) must be a multiple of 3. So,
for v = 8, λ must be 3 or 6 or . . . . For the 8-player version, the smallest
possible case is λ = 3, in which case there will be 14 games and each player
is in three of them.
The wargamers didn’t feel up to calculating a schedule for 14 games, so
they started playing the 7-player version again. Each week they scheduled
one game. Each week the other three players came to watch, and played a
game with Helga. And, at the end of seven weeks, they realized they had
 Introduction 9

played an equitable 8-player tournament

ABCD, ABEF, ACEG, ADFG, BCFG, BDEG, CDEF,
EFGH, CDGH, BDFH, BCEH, ADEH, ACFH, ABGH.
The problem here is, given a set with v elements (the players), to choose a
collection of subsets of equal size (4 in the example, more generally written
k) such that every pair of elements occurs in the same number of subsets (our
λ). This is called a balanced incomplete block design, and we’ll study them
in Chapter 14. The word balanced refers to fact that pairs occur together
equally often, incomplete to the fact that k is smaller than v (or else there
would be no problem), and the equal-size subsets are called blocks. It is called
a design because the main application is in the design and statistical analysis
of experiments. The numbers v, k and λ are the parameters of the design
(together with b, the number of blocks, and r, the number of blocks in which a
given object occurs; but b and r can be calculated from the other parameters).
It is common to refer to these designs by specifying their parameters, for
example, one says “a (v, b, r, k, λ)-design”; the wargamers’ first example is a
(7, 7, 4, 4, 2)-design.
We have already seen that some sets of parameters are impossible, because
r and b have to be integers. But there are other cases, for example v = 15, k =
5, λ = 2 (for which r = 7 and b = 21, both whole numbers) for which there
is no design; and there is an infinitude of parameter-sets for which we do not
know whether a design exists.
The second example—the 8-player schedule—is an (8, 14, 7, 4, 3)-design in
which the blocks can be partitioned into sets, where every object belongs
to exactly one member of each set. This special case is called a resolvable
balanced incomplete block design. These designs are of special interest for
statisticians and also arise in finite models of Euclidean geometry.

The Hat Game

Here is the outline of a new television game show. There is a team of three
contestants. Just before the show begins, the players are taken into separate
rooms. Each is blindfolded and a hat—either red (R) or black (B)—is put
onto his or her head. The blindfolds are removed, but the player cannot see
which color hat they have been given. The players are then taken into the
studio and line up behind three desks; they can see each others’ hats, but still
not their own. No communication between them is allowed.
The M.C. says, “Players, do you think your hat is red or black? Please write
down your answer. If you don’t know, you can pass—don’t write anything.”
Then the players’ “votes” are revealed. If any of them has made the wrong
guess, the team loses. If they all pass, they lose. If at least one player gets the
right answer, and there are no wrong answers, they win three million dollars.
As you would expect, the players meet beforehand to discuss their options
and choose a strategy. The organizers say they decide on the hat colors by
10 Introduction to Combinatorics

tossing a coin for each player, so that the allocations are independent and
each player has a 50-50 chance of red or black. We’ll assume they are telling
the truth, and we’ll also assume the players don’t try to concoct any type of
cheating scheme. What is the team’s best plan?
One obvious strategy would be for the team to nominate one of their number
who will always say “red” while the other two players always pass. This
would give them a 50% chance of winning the prize. Can they do better?
Most people, and in fact most mathematicians, initially think not. Since
each person’s hat color is independent of the other players’ colors and no
communication is allowed, it seems impossible for them to learn anything just
by looking at one another; all the players can do, it seems, is guess.
But in three out of four cases, we expect two of the players will have hats of
the same color and the third player’s hat will be the opposite color. The group
can win every time this happens if the players follow the following strategy:
Once the game starts, look at the other two players’ hats. If you see two hats
of different colors, pass. If they are the same color, guess that your hat is
the opposite color. Every time the hat colors are distributed two and one,
one player will guess correctly and the others will pass, and the team will win
the game. When all the hats are the same color, all three players will guess
incorrectly and the group will lose; but on average this will happen in only a
quarter of cases.
There are eight ways the hats can be distributed: RRR, RRB, RBR, RBB,
BRR, BRB, BBR and BBB (from left to right). Suppose the game was played
eight times, and suppose it just so happened that every possible combination
occurred once. In the first and last cases, all three players make the wrong
guess; in the other six there is just one correct guess. So, of the twelve guesses,
half are right and half are wrong, as you would expect.
How about more players? With four team members, the easiest strategy is
to tell one player to stay silent throughout. The other three don’t even look at
that player! They play as they would if member number four was not there:
each looks at the remaining two players and proceeds as if it was a team of
size three. They still have a 75% chance of winning.
The Hat Game problem was introduced by Todd Ebert in his dissertation
[28], and is in fact related to error-correcting codes. It has been discussed in
the press [90], and a recent survey paper is [17]. We’ll examine the problem
further, and explain the connection to error-correcting codes in Chapter 12.

The Bridges of Königsberg

Leonhard Euler (see Appendix C), a pioneering Swiss mathematician and
physicist, is seen by most as the leading mathematician of the eighteenth
century. Euler (pronounced OY-ler) founded the area of combinatorics called
graph theory (see Section 1.4, Chapter 10 and Chapter 11) and invented the
concepts of Latin squares and orthogonality (see Section 13.1). He also made
 Introduction 11

equally significant contributions to analytic geometry, trigonometry, geometry,

calculus and number theory.
In 1735, Euler spoke to the St. Petersburg Academy on a problem in recre-
ational mathematics. The Prussian city of Königsberg was set on both sides
of the Pregel River. It included a large island called the Kneiphof, and the
river branched to the east of it. So there were four main land masses—let’s
call them the four parts of the city, and label them A, B, C, D—connected to
each other by seven bridges, as shown in the rough map in Figure 1.4:

B
D

FIGURE 1.4: Königsberg and its bridges.

The problem was to find a walk through the city that would cross each
bridge once and only once, and visit all four parts. The only way to go
between the different pieces of land was to cross the bridges.
So we have a combinatorial problem. We have a set with seven elements—
the seven bridges—and we need to find out whether they can be arranged in a
certain way, specifically so that the corresponding bridges could be traversed
in the given order.
The solution is very easy. Look for example at the north bank of the river,
part A. Whenever you walk over a bridge that leads to A, your next bridge
must take you back out of A. So walking into A and out again uses up two
bridges (we say “uses them up,” because you cannot cross a bridge twice).
There are three bridges; the walk must take you onto A and off again once,
and there is one bridge left over. The only possibility is that the walk must
start or finish in A (but not both).
The same argument can be applied to C and D (with three bridges each),
and to B (five bridges, so you walk through B twice). Each part of town is
either the start or finish. But a walk can have only one start, and only one
finish, and there are four parts of town! This is clearly impossible, so there is
no such walk.
Of course, Euler proved that no walk is possible. But he did much more.
As we said, he essentially invented graph theory and showed how it could be
used to represent any set of islands and bridges, or indeed any set of towns and
roads joining them, and concocted an algorithm for traversability problems.
We shall look at his methods in Section 11.1.
12 Introduction to Combinatorics

Six Degrees of Separation

We have all experienced the following phenomenon: meeting a perfect
stranger, and subsequently finding out that we have a common acquaintance.
While this is very rare, a little further investigation may well show that the
new person is a “friend of a friend” or a “friend of a friend of a friend.”
Guglielmo Marconi, the pioneer of radio, thought that these phenomena were
becoming far more common because of the growth of communication. In his
1909 speech accepting the Nobel Prize, he suggested that the average two peo-
ple could be connected through a chain of at most five acquaintances (that is,
six links).
This idea has been explored by many people since. In particular, Harvard
psychologist Stanley Milgram [75, 117] studied the problem experimentally.
The following description of his experiment is based on [136].

1. Milgram typically chose individuals in Omaha, Nebraska and

Wichita, Kansas to be the starting points and Boston, Massachusetts
to be the end point of a chain of correspondence.
2. Information packets were sent to randomly selected individuals
in Omaha or Wichita. They included letters, which detailed the
study’s purpose, and basic information about a target contact per-
son in Boston, a roster on which they could write their own name,
and business reply cards that were pre-addressed to Harvard.
3. Upon receiving the invitation to participate, the recipient was
asked whether he or she personally knew the contact person in
Boston. If so, the person was to forward the letter directly to that
person. (Knowing someone “personally” was defined as knowing
them on a first-name basis.)
4. If the person did not personally know the target, they were to
think of a friend or relative they knew personally who they thought
more likely to know the target. They then signed the roster and
forwarded the packet to the friend or relative. A postcard was also
sent to Harvard.
5. When and if the package eventually reached the contact person
in Boston, the researchers could examine the roster to count the
number of times it had been forwarded from person to person.
Additionally, for packages that never reached the destination, the
incoming postcards helped identify the break point in the chain.

This experiment has a number of obvious flaws. Many people refused to

participate. In one case, 232 of the 296 letters never reached the destination.
However, completed chains seemed to run to six or fewer links. Subsequent
experiments have shown that this number is about right, and the phrase “six
degrees of separation” has become fashionable.
We can look at the world as a huge communication network. Smaller com-
munication networks include social networks like Facebook and MySpace, the
 Introduction 13

internal communications (or intranet ) of a large company, the physical net-

work of wires connecting components of a computer system, and so on. The
average length of communication links could be very small (for example, if
all pairs are directly linked) or very large (some computer networks consist
of a ring in which each component is linked only to the pair of components
adjacent in the ring). Networks with a small average length are called small
world networks (as in the phrase, “it’s a small world”). The analysis of these
networks—for example, finding out whether a particular type of network ex-
hibits “small world” properties—is basically a combinatorial problem, and
again graph theory comes into play.

1.2 Sets, Relations, and Proof Techniques

A set is a collection of objects with a well-defined rule, called the mem-
bership law, for determining whether a given object belongs to the set. The
individual objects in the set are called its elements or members and are said
to belong to the set. If S is a set and s is one of its elements, we denote this
fact by writing
s ∈ S.
We say a set S is a subset of a set T if every member of T is also a member
of T .
While nearly all readers will be familiar with basic set theory, you may wish
to review the standard set ideas and notations in Appendix A. In particular,
you need to know about the standard operations on sets: union, intersection,
complementation; you need to know de Morgan’s Laws; and you should be
familiar with the standard number sets, the set of integers, denoted Z, the
rational numbers Q, the real numbers R and the complex numbers C.
If n is a positive integer, the residue classes modulo n are the n sets
[0], [1], . . . , [n − 1], where [i] is the set of all integers that leave remainder
i on division by n. Members of the same residue class are called congruent
modulo n and we write y ≡ x(mod n). When no confusion arises we simply
write x instead of [x].
If you have done a little abstract algebra, you will know that the set Zn of
residue classes modulo n is an example of a finite group. We discuss groups
further in Chapter 8.
We often illustrate sets and operations on sets by diagrams. A set R is
represented by a circle, and the elements of R correspond to points inside
the circle. If we need to show a universal set (for example, if complements
are involved) then this universal set is shown as a rectangle enclosing all the
other sets. These illustrations are called Venn diagrams, because they were
popularized by John Venn (see Appendix C) in 1880 [118, 119], although
14 Introduction to Combinatorics

similar diagrams were studied earlier by Leibniz and Euler, and related ideas
were studied by Lull as far back as the thirteenth century (see, for example,
[5]). It has recently become fashionable to refer to “Euler diagrams,” but
Venn was primarily responsible for their modern usage.
For example, the Venn diagrams or R ∪ S, R ∩ S, R, R\S and R ∩ S ∩ T
are shown in Figure 1.5.

R S R S

R ∪S R∩S
R R S

_
R R\S
R

S T

R∩S∩T

FIGURE 1.5: Some Venn diagrams

We define the Cartesian product (or cross product ) S × T of sets S and T

to be the set of all ordered pairs (s, t) where s ∈ S and t ∈ T :

S × T = {(s, t) : s ∈ S, t ∈ T }.

There is no requirement that S and T be disjoint; in fact, it is often useful to

consider S × S.
A (binary) relation ρ from a set S to a set T is a rule that stipulates,
given any element s of S and any element t of T , whether s bears a certain
relationship to t (written s ρ t) or not (written s \ρ t). Alternatively, we can
define a binary relation ρ from the set S to the set T as a set ρ of ordered
pairs (s, t), where s belongs to S and t belongs to T , with the notation that
s ρ t when (s, t) ∈ ρ and s \ρ t otherwise. This means that, formally, a binary
relation from S to T can be defined as a subset of the Cartesian product S ×T .
A relation α on a set A is called an equivalence relation on A if and only if
it is reflexive, symmetric, and transitive.
 Introduction 15

The obvious equivalence relation is equality, on any set. In sets other than
number sets, equal objects are often called “equal in all respects.” More
generally, an equivalence relation can be considered as a statement that two
objects are “equal in some (specified) respects.” One example, on the integers,
is the relation α, where a α b is true if and only if a = ±b—a and b have the
same absolute value. Another, is congruence, on the set of all plane triangles.
Equivalence relations give us an alternative way to discuss multiple ele-
ments. You can view a multiset as an object based on a set of distinct ele-
ments on which an equivalence relation has been defined, so that the elements
of a multiset are the equivalence classes of the underlying set.
One of the most important proof techniques for working with finite sets or
with sets of integers is the method of mathematical induction, the application
of the principle of mathematical induction:
Suppose the proposition P (n) satisfies
(i) P (1) is true; and
(ii) for every positive integer n, whenever P (n) is true, then P (n + 1) is
true.
Then P (n) is true for all positive integers n.
If you are not familiar with induction, or need to review the topic, we treat
it in detail in Appendix A.

1.3 Two Principles of Enumeration

In this section we introduce two easy rules for enumerating the numbers of
elements of certain sets or certain types of arrangements.
The addition principle states that if we are building an arrangement of
objects starting with a set of mutually exclusive beginning states, then the
number of arrangements is the sum of the number of arrangements starting
from each beginning state. This rather complicated-sounding sentence is far
more trivial than it might appear. For instance, suppose that a young woman
wants to go on a job interview, and has only one clean dress skirt and one
clean pair of dress slacks. With the skirt, she can create three presentable
outfits; with the pants, five. It follows that she has 3 + 5 = 8 ways to dress
appropriately.
The multiplication principle states that if we are building an arrangement
of objects in stages, and the number of choices at each stage does not depend
on the choice made at any earlier stage, then the number of arrangements is
the product of the number of choices at each stage. Again, this principle is
simpler than it appears; if a restaurant, for example, offers two kinds of salad,
16 Introduction to Combinatorics

nine entrees, and seven desserts, the number of meals consisting of one salad
and one entree and one dessert is 2 × 9 × 7 = 126.
Although these principles seem painfully obvious, each may be proved by
results from set theory; for instance, the multiplication principle follows from
the fact that the cardinality of a Cartesian product of sets is the product of
the cardinalities of the underlying sets. However, we prefer to emphasize the
uses rather than the proofs of these principles.
Example 1.1: We wish to have lunch at a fast-food place, and there
are three within walking distance. One place offers eight soft drinks, seven
sandwiches, and two choices of a “side dish” (French fries or onion rings);
another offers only six drinks, but 10 sandwiches and two sides; and the last
(a coffee shop) offers 16 drinks, five sandwiches, and only one side. How many
different lunches are possible?
The addition principle says that (because we can eat lunch at only one
place) the answer is the sum of the number of ways to eat lunch at each place.
The multiplication principle says that each place has a number of options
equal to the product of the options at each stage: drink, sandwich, and side.
So the first restaurant offers us 8 × 7 × 2 = 112 possibilities; the second,
6 × 10 × 2 = 120; and the third, 16 × 5 × 1 = 80. Altogether, then, there are
112 + 120 + 80 = 312 possibilities for lunch. 
The next few examples will introduce some mathematical concepts that
may already be familiar to the reader. These concepts arise frequently in
enumeration problems. The reader should already be familiar with n! to
denote the product 1 · 2 · · · · · n. By convention, we assume 0! = 1! = 1.
Example 1.2: A set has n distinct elements. How many subsets does it
have?
We line the elements of the set up in some order; beneath each, we will
place a mark if the element is in our subset, and no mark if the element is
not. We will make n decisions as we go through the list of elements, and each
decision has two alternatives, mark or no mark. Since no decision will affect
any other decision, we see that there are 2n subsets of a set of n elements. 

Example 1.3: Suppose we have five tunes stored on our portable digital
music player, and we wish to listen to each of them one after another. How
many ways are there to arrange the playlist?
We have five tunes, so there are five possibilities for the first tune. Once
we have chosen this tune, then regardless of which tune we choose to play
first there are four tunes left (because we don’t wish to repeat a tune). After
we choose the second tune, there are three possibilities left, and so on. The
result is that we may choose a playlist in 5! = 5 × 4 × 3 × 2 × 1 ways, by
the multiplication principle. In more general terms, we find that there are n!
 Introduction 17

ways to order n distinct objects; so we can arrange seven books on a shelf in

7! ways, place 12 people into a line in 12! ways, and so forth. 

Example 1.4: A license plate consists of three letters of the alphabet and
three digits from 0 to 9, with the letters preceding the digits. How many
license plates are possible? How many are possible if we do not use any letter
twice? How many are possible if we may use letters twice but not digits? How
many if we do not use either twice?
We imagine six boxes in a row as shown.

For the first box, we may choose any of the 26 letters, so we have 26 choices.
For the first question, where we may repeat letters and digits, we have 26
possible choices for each of the next two boxes as well. The last three boxes
each have 10 possibilities. It follows from the multiplication principle that
there are 263 · 103 possible license plates. If we may not repeat a letter, then
we must work a little harder; the second box has only 25 possible letters, and
the third has only 24. We then get 26 · 25 · 24 · 103 possible plates. In the
same way, there are 263 · 10 · 9 · 8 plates where letters may be repeated but
digits may not; and 26 · 25 · 24 · 10 · 9 · 8 possible plates where neither may be
repeated. 

Example 1.5: How many n-digit numbers in base k are there if we do not
allow a leading digit of 0? How many numbers without leading zeros in base
k are there with at most n digits?
We recall that a number in base k has possible digits 0, 1, . . . k − 1, so that
each of the n digits may be assigned any of the k symbols, except for the
first. We apply the multiplication principle to see that there are (k − 1) · k n−1
possible numbers. To find how many numbers of at most n digits in base k
there are, we may simply permit leading zeros; so the number 1 (in any base)
might be considered a “five-digit number” 00001. We find that there are k n
such numbers. 

Example 1.6: A large corporation gives employees ID codes consisting of

two or three letters and two digits. How many codes are possible if we may
use any of the 26 upper-case letters and any of the digits from 0 to 9, and
letters precede digits?
The multiplication principle tells us that there are 262 · 102 possibilities for
the two-letter codes, and 263 · 102 for the three-letter codes. The addition
principle tells us that there are
262 · 102 + 263 · 102 = 67, 600 + 1, 757, 600 = 1, 825, 200
codes. 
18 Introduction to Combinatorics

1.4 Graphs
A great deal of mathematics has been done concerning collections of subsets
of two elements of a larger set. In that case the members of the universal set
are often represented graphically, as points in a diagram, and the set {x, y} is
represented by a line drawn joining x and y. Provided the universal set is not
empty, such a structure is called a graph. The sets are called edges (or lines),
and the elements of the universal set are vertices (or points). The universal
set is the vertex-set of the graph. For any graph G, we write V (G) and E(G)
for the sets of vertices and of edges of G.
The edge {x, y} is simply written xy, when no confusion can arise; x and
y are called its endpoints. When x and y are endpoints of some edge, we say
they are adjacent and write x ∼ y for short; the vertices adjacent to x are
called its neighbors. The set of all neighbors of x is its (open) neighborhood,
N (x). If x and y are not adjacent we write x 6∼ y.
Two vertices either constitute a set or not, so a graph can never contain two
edges with the same pair of vertices. However, there are some applications
where two edges joining the same vertices might make sense. For this reason
we sometimes talk about networks or multigraphs in which there can be several
edges joining the same pair of vertices; those edges are called multiple edges.
Another generalization is to allow loops, edges of the form xx. There is no
very good term for a graph-type structure in which loops are allowed, and we
will usually call one of these a looped graph or looped network or network with
loops although strictly speaking it is not a graph or network at all. Moreover,
when no confusion arises, the word graph can be used for any generalized
graph.
Any binary relation can be represented by a diagram. If ρ is a binary
relation on the set S, the elements of S are shown as vertices, and if xρy is
true, then an edge is shown from x to y, with its direction indicated by an
arrow. Provided the set S is finite, all information about any binary relation
on S can be shown in this way. Such a diagram is called a directed graph
or digraph, and the edge together with its arrow is called an arc. If ρ is
symmetric, the arrows may be dropped and the result is a graph (possibly
with loops).
Several families of graphs have been studied. Given a set S of v vertices,
the graph formed by joining each pair of vertices in S is called the complete
graph on S and denoted KS . Kv denotes any complete graph with v vertices.
As you would expect, we often call K3 a triangle. The complete bipartite
graph on V1 and V2 has two disjoint sets of vertices, V1 and V2 ; two vertices
are adjacent if and only if they lie in different sets. We write Km,n to mean a
complete bipartite graph with m vertices in one set and n in the other. K1,n
in particular is called an n-star . Figure 1.6 shows copies of K6 and K3,4 .
 Introduction 19

FIGURE 1.6: K6 and K3,4 .

Suppose H is a graph all of whose vertices and edges are vertices and edges
of some graph G—that is, V (H) ⊆ V (G) and E(H) ⊆ E(G). Then H is a
subgraph of G; we write H ≤ G. Every graph G has itself as a subgraph;
if H is a subgraph of G but H 6= G, H is a proper subgraph of G, and we
write H < G. In particular, if S is some set of vertices of G then hSiG is
the subgraph consisting of all edges of G with both endpoints in S. If G
is a complete graph whose vertex-set contains S then the subscript “G ” is
dropped, and hSi is the complete subgraph based on S. Any subgraph of a
complete bipartite graph is itself called bipartite.
Instead of saying Figure 1.6 shows two graphs, we could say it is a single
graph that consists of two separate subgraphs, with no edges joining one part
to the other. We call such a graph disconnected ; a graph that is all in one piece
is called connected. The separate connected parts of a disconnected graph are
called its components.
The graph G is trivially a subgraph of the complete graph KV (G) . The set
of all edges of KV (G) that are not edges of G will form a graph with V (G) as
its vertex set; this new graph is called the complement of G, and written G.
More generally, if G is a subgraph of H, then the graph formed by deleting
all edges of G from H is called the complement of G in H, denoted H − G.
The complement K S of the complete graph KS on vertex set S is called a null
graph; we also write K v as a general notation for a null graph with v vertices.
Figure 1.7 shows a graph and its complement.

A A

B C B C

D E D E
G G
FIGURE 1.7: A graph and its complement.
20 Introduction to Combinatorics

We define the degree or valency d(x) of the vertex x to be the number of

edges that have x as an endpoint. If d(x) = 0, then x is called an isolated
vertex. A graph is called regular if all its vertices have the same degree. If
the common degree is r, it is called r-regular. In particular, a 3-regular graph
is called cubic.

THEOREM 1.1 (Sum of Degrees)

In any graph or multigraph, the sum of the degrees of the vertices equals
twice the number of edges (and consequently the sum of the degrees is an even
integer).

Proof. Suppose the graph or multigraph has e edges; label the edges, say
y1 , y2 , . . . , ye . Consider a list in which each edge appears twice, once for each
endpoint. For example, if y1 has endpoints x4 and x7 , you might make entries
y1 : x4 and y1 : x7 . Vertex x will appear in precisely d(x) entries, so the total
number of entries equals the sum of the degrees of the vertices. On the other
hand, each edge appears twice, so the total number of entries equals twice the
number of edges. 

COROLLARY 1.1.1
In any graph or multigraph, the number of vertices of odd degree is even. In
particular, a regular graph of odd degree has an even number of vertices.
Suppose you encounter two graphs with exactly the same structure but
different vertex-sets. For example, look at Figure 1.7 again; the two subgraphs
with vertex-sets {B, D, E} and {C, D, E} are both triangles. There is no
real difference between them until you need to consider the meaning of the
vertices. Formally, we say two graphs G and H are isomorphic if there is
a one-to-one correspondence between their vertex-sets V (G) and V (H) such
that two vertices of G are adjacent if and only if the corresponding vertices
of H are adjacent. For example, any two complete graphs with v vertices are
isomorphic.
Not all graphs with the same number of vertices are isomorphic; for ex-
ample, the graph G of Figure 1.7 and its complement are quite different, but
both have five vertices.

1.5 Systems of Distinct Representatives

An important topic when discussing collections of subsets of size greater
than 2 is the existence of systems of distinct representatives. If D = {B1 , B2 ,
. . . , Bk } are any k sets, we define a system of distinct representatives, or
 Introduction 21

SDR for D to be a way of selecting a member xi from each set Bi such that
x1 , x2 , . . . are all different.
As an example, consider the sets

{124, 124, 134, 235, 246, 1256}.

One system of distinct representatives for them is

1, 2, 3, 5, 4, 6

(where the representatives are listed in the same order as the sets). There are
several others. On the other hand, the sets

{124, 124, 134, 23, 24, 1256}

have no SDR.
If the collection of sets D is to have an SDR, it is clearly necessary that
∪B∈D B have at least as many elements as there are sets in D. The example
above shows that this is not sufficient.

THEOREM 1.2 (Philip Hall’s Theorem) [52]

A collection D of sets has a system of distinct representatives if and only if
it never occurs that some n sets contain between them fewer than n elements.

Proof. We proceed by induction on the number of sets. If D consists of one

set, the result is obvious. Assume the theorem to be true for all collections
of fewer than b sets. Suppose D has k sets B1 , B2 , . . . , Bk and between them
they have v elements {1, 2, . . . , v}; and suppose D satisfies the hypothesis that
the union of any n sets has size at least n, for 1 ≤ n ≤ k. By induction, any n
of the sets will have an SDR, provided that n < k. We distinguish two cases.
(i) Suppose no n sets contain between them fewer than n + 1 elements,
for n < k. Select any element x1 ∈ B1 , and write Bi∗ = Bi \{xi }, for i ∈
{1, 2, . . . , k}. Then the union of any n of the Bi∗ has at least n elements, for
n = 1, 2, . . . , k−1. By the induction hypothesis, there is an SDR x2 , x3 , . . . , xk
for B2 , B3 , . . . , Bk , so x1 , x2 , x3 , . . . , xk is an SDR for the original design D.
(ii) Suppose there is a collection of n of the sets whose union has precisely
n elements, for some n less than k. Without loss of generality, take these
sets as B1 , B2 , . . . , Bn . For i > n, write Bi∗ to mean Bi with all members
of B1 ∪ B2 ∪ . . . ∪ Bn deleted. It is easy to see that the design with sets
∗ ∗
Bn+1 , Bn+2 , . . . , Bk∗ satisfies the conditions of the theorem (if Bn+i ∗
1
∗
, Bn+i2
,
∗
. . . , Bn+it were t sets whose union has fewer than t elements, then B1 , B2 ,
∗ ∗ ∗
. . . , Bn , Bn+i 1
, Bn+i 2
, . . . , Bn+it
would be n + t sets of D whose union has
fewer than n+t elements, which is impossible). From the induction hypothesis
both sets have SDRs, and clearly they are disjoint, so together they comprise
an SDR for D. 
22 Introduction to Combinatorics

Exercises 1A
1. In each case, represent the set in a Venn diagram.

(i) R ∪ S ∪ T (ii) R ∪ S ∪ T

2. Show that the Hat Game team has at least a 75% chance of winning,
for any number of players greater than three.

3. Find a formula for the sum of the first n odd positive integers,

1 + 3 + . . . + (2n − 3) + (2n − 1)

(i) using Gauss’s formula for the sum of the first n positive integers;
(ii) using induction.

4. The seven wargamers decide to play a new game, where players compete
in groups of three. Show that they can find an equitable schedule of
seven games.

5. Prove by induction that

n
X
i(i + 1) = 13 n(n + 1)(n + 2).
i=1

6. Define the relation τ on the integers by m τ n whenever m + 2n is

divisible by 3. Determine whether τ is an equivalence relation, and if
not which of the properties (reflexive, transitive, symmetric) it lacks.

7. A phone store has four models of Samsung phones, three kinds of HTC
phones, and two models of Apple phones. In how many ways can we
purchase a phone from this store?

8. A computer store has four brands of desktops, three sizes of monitor,

and six models of printer. In how many ways can we take home a system
consisting of one tower, one monitor, and one printer?

9. In order to gain a high-paying career in the exciting field of basketweav-

ing, we wish to take a class in Underwater Basketweaving and a class
in Mountaintop Basketweaving. The local junior college offers three
sections of each course. The nearby university offers six sections of Un-
derwater and four of Mountaintop. A somewhat reputable online school
has ten sections of Underwater and two of Mountaintop. Assuming that
we will take both courses from the same school and there are no other
conflicts, how many ways can we do this?
 Introduction 23

10. We have recorded eight movies and we want to watch three of them
tonight. How many ordered sequences of three of the eight movies are
possible?
11. A state has license plates consisting of two letters and four digits. How
many license plates are possible?
12. Suppose a set X has 2n + 1 elements. How many subsets of X have n
or fewer elements?
13. What are the degrees of the vertices in these graphs?

(i) (ii) (iii)

14. How many edges does the star K1,n have? What are the degrees of its
vertices?
15. A graph has 48 edges. There are no isolated vertices. What are the
minimum and maximum numbers of vertices the graph can have?
16. Eight people attend a meeting. At the beginning various people shake
hands. In total there are eighteen handshakes.
(i) Show that there is at least one person who shook hands at least
five times.
(ii) Is it possible there were exactly three people who shook hands an
odd number of times?
17. Do there exist SDRs for the following sets?
(i) 12, 145, 12, 123
(ii) 123, 145, 12, 13, 23.

Exercises 1B
1. The seven wargamers have become more ambitious. They wish to play
a tournament as follows: each week two groups of three (with no one in
common) play; the seventh member acts as host. Find a schedule for
them to do this in seven weeks.
2. In each case, represent the set in a Venn diagram.
24 Introduction to Combinatorics

(i) R ∪ (S ∩ T ) (ii) (R ∩ S) ∪ T

3. Use induction to prove that, when r 6= 1, a + ar + ar2 . . . + arn−1 =

a(1 − rn )/(1 − r).

4. A ball is dropped from a height of 10 feet. Each time it hits the ground,
it bounces back to half the previous height. How far will it have traveled
when it hits the floor for the tenth time?

5. A furniture store has in stock four square dining tables, six expandable
rectangular dining tables, and two circular dining tables. In how many
ways can we purchase a dining table from this store?

6. A furniture store has three kinds of dining tables, six different sets of
dining room chairs, and three kinds of sideboard. In how many ways
can we purchase a dining room set consisting of a table, a set of chairs,
and a sideboard?

7. To fulfill graduation requirements, we need a history course and an En-

glish course. At the local junior college, there are three courses that
satisfy the history requirement and two that satisfy the English require-
ment. At the nearby university, there are four history and four English
courses that satisfy requirements. An online institution offers two of
each. Assuming that the credits transfer and we only wish to enroll
in one institution, in how many ways can we complete the graduation
requirements for these two courses?

8. Define the relation φ on the integers by m φ n whenever m + 3n is

divisible by 4. Determine whether φ is an equivalence relation, and if
not which of the properties (reflexive, transitive, symmetric) it lacks.

9. There are nine new videos we want to see on YouTube, but we only
have time to watch four of them right now. How many ways are there
to watch four of the nine videos in some particular order?

10. A state has a license plate that consists of two letters, three digits, and
one letter. How many license plates are possible?

11. What are the degrees of the vertices in these graphs?

(i) (ii) (iii)

 Introduction 25

12. The n-wheel Wn has n + 1 vertices {x0 , x1 , . . . , xn }; x0 is joined to every

other vertex and the other edges are
x1 x2 , x2 x3 , . . . , xn−1 xn , xn x1 .
How many edges does Wn have? What are the degrees of its vertices?
13. A graph has 9 vertices and each vertex has degree at least 4. What are
the minimum and maximum numbers of edges the graph can have?
14. Draw a graph with nine edges and:
(i) five vertices;
(ii) six vertices.
15. Do there exist SDRs for the following sets?
(i) 123, 124, 134, 1234, 234
(ii) 123, 145, 167, 13, 23, 24.
(iii) 123, 145, 167, 13, 23, 12.

Problems

Problem 1: A relation ρ on a set S is said to be circular if and only if

for every x, y, z ∈ S, x ρ y and y ρ z together imply that z ρ x. Show that a
reflexive circular relation is an equivalence relation.
Problem 2: Find the error in the following “proof.”
Theorem. A symmetric and transitive relation must be reflexive.
Proof. Let ρ be a symmetric and transitive relation on a set A. If a ρ b, then
b ρ a because a is symmetric. But a ρ b and b ρ a together imply a ρ a, since ρ
is transitive. Therefore ρ is reflexive.
Problem 3: Show by induction that 2n ≥ n2 for integer n ≥ 4.
Problem 4: Prove by induction: if n people stand in line at a counter, and
if the person at the front is a woman and the person at the back is a man,
then somewhere in the line there is a man standing directly behind a woman.
Problem 5: Assume that the sum of the angles of a triangle is π radians.
Prove by induction that the sum of the angles of a convex polygon with n
sides is (n − 2)π radians when n ≥ 3.
Problem 6: Prove that no graph has all its vertices of different degrees.
Problem 7: Five married couples attend a dinner party. During the evening
some of the people shake hands; no one shakes hands with his or her spouse.
26 Introduction to Combinatorics

At the end of the party, the hostess asks each person (other than herself, of
course), “with how many people did you shake hands?” It turns out that no
two of them shook hands with the same number.
(i) With how many people did the hostess shake hands?
(ii) With how many people did her husband shake hands?
(iii) Generalize this to the case of n couples.
Problem 8: We have five student organizations O1 , O2 , . . . O5 , each of which
wishes to send a student ambassador to the National Congress of Useless
Student Organizations. National Congress policy decrees that no student
may represent two organizations, so we require an SDR of the memberships
of the organizations. Of all students in these organizations, only five (A, B,
C, D, and E) are interested in being ambassadors. Now, O1 contains A and
C; O2 contains A, B, C, and D; O3 contains A and C; O4 contains B, D,
and E; and O5 contains A, C, and D.
(i) Find an SDR from among the interested students.
(ii) Find a student and organization such that if the student resigns from
the organization, no SDR is possible.
(iii) Find a different student and organization such that if the student resigns
then no SDR is possible.
(iv) Is it possible that if one of these five students were to join one other
organization, part (ii) would have no solution? Prove your answer.
Chapter 2
Fundamentals of Enumeration

In this chapter we explore some applications of the two principles that were
introduced in Section 1.3.

2.1 Permutations and Combinations

In practice, two particular applications of the Multiplication Principle arise
many times. Because of this, they have their own notations.
Example 2.1: An antiques dealer has twelve different pieces of iridescent
glass and wishes to display five pieces in a window display. In how many ways
can she arrange five of the twelve pieces in some order in the window?
Since there are twelve pieces to choose from, she has 12 ways to put a piece
in the first spot. There are 11 ways to put a piece in the second spot (because
after the first piece has been placed, only 11 remain); 10 for the third; and so
forth. The Multiplication Principle tells us that the total number of ways is
12 × 11 × 10 × 9 × 8 = 95, 040. 
This argument may be generalized nicely. We assume that the reader is
aware that a permutation is an ordered subset, and a combination is an un-
ordered subset, of a set.

THEOREM 2.1 (Counting Permutations)

The number of ways to arrange k objects in order from a set of n objects is
n!/(n − k)!, commonly denoted P (n, k) or [n]k .

Proof. There are n possible choices for the first object, (n−1) for the second,
and so forth, until we find that there are n − k + 1 possible choices for the kth
object. The Multiplication Principle tells us that there are n(n−1) . . . (n−k+
1) arrangements altogether. When we divide n! by (n − k)!, this is precisely
the result. 

Example 2.2: How many ways are there for an awards committee to
award three distinct scholarships among 12 students entered?

27
28 Introduction to Combinatorics

The formula gives us at once 12!/9! = 1, 320. 

We wish to ask what happens if we do not wish to order the k chosen
objects. Thus, if we have three (identical) scholarships and 12 students, how
many ways can we award the scholarships? The answer will be much smaller
than the 1, 320 that we got for the previous example, because we do not
distinguish among scholarship winners. Thus, if students A, B, and C are
selected to win a scholarship, that one case corresponds to six cases in the
last example; for we might have A first, then B and C last, or A, then C, then
B, and so on. Because there are three winners, there are 3! ways to order
them. We can employ the Multiplication Principle by asking how many ways
there are to order the three

students. Let the number of ways to choose 3 of 12
students be denoted 12 3 or C(12, 3). Both notations are commonly used; we
say “12 choose 3.” Then there are C(12, 3) ways to choose the students, and
3! ways to order the chosen students; this says that C(12, 3) × 3! = P (12, 3).
It follows that C(12, 3) is just P (12, 3)/3! = 220. This example also may be
generalized.

THEOREM 2.2 (Counting Combinations)

The

number of ways to choose k objects from a set of n distinct objects is
n
k = n!/(k!(n − k)!), also commonly denoted C(n, k).

Proof. By Theorem 2.1, there are P (n, k) ways to choose k elements in a

particular order,

and there are k! ways of ordering the given k elements. It
follows that nk × k! = P (n, k), and dividing by k! completes the proof. 

Example 2.3: An automobile manufacturer produces eight models of

sports car. In how many ways can we choose three models to be blown up
during the exciting climax of our next action movie?
The solution is 83 = 8!/(3!·5!) = 56.




Example 2.4: A school decides to offer a new bioinformatics program,

and decides to appoint a committee consisting of two mathematicians, three
biologists, and three computer scientists to plan the implementation. If there
are six mathematicians, ten biologists, and five computer scientists available
to serve on this committee, in how many ways can the committee be formed?

There are six mathematicians who could serve and we must choose two
of them; this may be done in C(6, 2) = 15 ways. We select the biologists
in one of C(10, 3) = 120 ways. The computer scientists are chosen in one
of C(5, 3) = 10 ways. By the Multiplication Principle, we have a total of
15 × 120 × 10 = 18, 000 possible committees. 

Example 2.5: For the new bioinformatics program of the previous exam-
ple, we find that Dr. John Walters of computer science and Dr. Walter Johns
 Fundamentals of Enumeration 29

of mathematics cannot serve together due to their tendency to become dis-

tracted and make dreadful puns during committee deliberations. How many
committees are there that contain at most one of these two individuals?
We could calculate the answer by doing cases, but it is easier to subtract the
possible committees that contain both these professors. The committees that
contain both involve a choice of one of the five mathematicians in addition
to Dr. Walter Johns, two of the four computer scientists in addition to Dr.
John Walters, and three biologists. There are C(5, 1)·C(10, 3)·C(4, 2) = 3, 600
such committees. We subtract these from the 18, 000, so we have 18, 000 −
3, 600 = 14, 400 possible committees. You should observe that we get the same
result from calculating that there are C(5, 2)·C(10, 3)·C(4, 3) committees that
contain neither professor, C(5, 2) · C(10, 3) · C(4, 2) committees that contain
John Walters but not Walter Johns, and C(5, 1)·C(10, 3)·C(4, 3) that contain
Walter Johns but not John Walters, and then adding the results. 

 
n
2.2 Applications of P (n, k) and
k
Here we discuss some problems whose solutions involve counting permu-
tations and combinations, as well as a generalization of the concepts. In
particular, the student is probably aware that the symbol C(n, k) is called
a binomial coefficient . A “subscripting technique” proves the theorem that
explains this name.

THEOREM 2.3 (The Binomial Theorem)

n  
n
X n k n−k
(x + y) = x y
k
k=0

Proof. We write the product as (x1 + y1 )(x2 + y2 ) . . . (xn + yn ). Then each

term of the product will be of the form a1 a2 . . . an−1 an , where each a is either
an x or a y. To find a term that will contribute to the coefficient of xk y n−k ,

k of the as to be x and the other n − k of them to be y. There are
we need for
clearly nk ways to pick k of the as to be x. Removing
the subscripts means
that we can collect all such terms together to get nk xk y n−k . This works for
each value of k from 0 to n. 

COROLLARY 2.3.1    
n n
=
k n−k
30 Introduction to Combinatorics

Proof. Clearly (x + y)n = (y + x)n . 

COROLLARY 2.3.2
n  
X n
= 2n
k
k=0

Proof. Set x = y = 1 in Theorem 2.3. 

COROLLARY 2.3.3
n  
X
k n
(−1) =0
k
k=0

Proof. Set x = 1 and y = −1 in Theorem 2.3. 

There are many other identities and formulas related to binomial coeffi-
cients. Some of these are explored in the exercises and problems. We mention
one last identity as it is important.

THEOREM 2.4 (Pascal’s Identity)


n−1
 
n−1
  
n
+ =
k−1 k k

Combinatorial identities often have two or more kinds of proofs. One kind
of proof for a theorem such as this is the algebraic proof, where we simply
write both sides of the equation, expand in factorials, and transform one side
into the other by means of elementary (and reversible) algebraic operations.
Another approach is to use a combinatorial proof, which involves showing
that each side of the equation counts the same set, or counts sets between
which a bijection exists. The combinatorial proof is often preferred because
it provides greater insight into the meaning of the identity.
Proof. Consider a set of n distinct objects, one of which we distinguish with
a label of some kind from the others. We
count the number of ways to select
k of these objects. Clearly, there are nk ways to do this. Now, if we wish to
separate these subsets of size k according to whether or not they contain the
distinguished object, how many of each kind are there? The sets containing
the distinguished object may be formed by taking the object
and choosing
k − 1 of the remaining objects of the n − 1; so there are n−1
k−1 subsets of size
k that contain our distinguished element. Similarly, if we do not include the
 Fundamentals of Enumeration 31

distinguished object, there are n−1

k ways of choosing k objects from the n−1
not-distinguished objects. Since every subset either does or does not contain
the distinguished object, the identity follows. 
This method of proving an identity is sometimes called the distinguished
element method. We will see many other identities proved in this way.
Binomial coefficients are usually presented in a triangular array, called Pas-
cal’s Triangle (although it certainly predates Pascal; see [16] or [24], which
specify earlier Chinese, Indian, and European sources). In the figure below,
the entry in row n and column k is nk .

n\k 0 1 2 3 4 5 6
0 1
1 1 1
2 1 2 1
3 1 3 3 1
4 1 4 6 4 1
5 1 5 10 10 5 1
6 1 6 15 20 15 6 1

Example 2.6: Suppose we live at the lower-left corner of the diagram of

Figure 2.1, and we need to walk to the upper-right corner along the streets
represented by the vertical and horizontal lines. In how many ways can we
make this walk? We assume that we never take shortcuts through parking
lots or other open areas, and each block is walked either north or east (so we
never retrace our steps or detour south or west).

All
All
All

FIGURE 2.1: How many ways to walk?

We may describe our path by a string of 12 characters, each of which is

either N or E to represent the direction we take. So if we start our string
with “NNEE,” this would indicate that we first walk two blocks north and
then walk two blocks east. The string “EEEEEEENNNNN” indicates that
we walk 7 blocks east along the southernmost street, then turn left and walk
5 blocks north to reach our goal. Thus, our path is a string of 12 characters,
32 Introduction to Combinatorics

of which 7 must be “E” and 5 must be “N.” How many such strings are there?
Clearly C(12, 5) = 792. 

Example 2.7: In how many ways can we make this walk if we determine
to walk along the block marked b1 in Figure 2.1? How many ways if the block
marked b1 is blocked off and we must avoid it?

The first question is simple enough; to guarantee we walk along block b1,
we must walk from our starting point (denoted (0, 0)) to the westmost point
of the block (at (2, 2)); then walk from the end of that block (at (3, 2)) to
our destination at (7, 5). The Multiplication Principle gives us the number of
ways to do this; it is C(4, 2)·C(7, 3) = 210.
The second question requires us to take the 792 total ways to walk from
one corner to the other and subtract the 120 ways that entail walking down
block b1; this gives us 582 ways to walk without going down block b1. 

We will see other problems involving counting lattice paths like this else-
where in the book. You should also see the exercises and problems at the end
of the chapter that make use of the blocks labeled b2 and b3.

2.3 Permutations and Combinations of Multisets

Theorem 2.2 tells us how to select some k items from a set of n items. It
is also possible that we wish to select more than one “copy” of some items.

Example 2.8: Consider the two questions: How many ways can we rear-
range the letters of the word KEPT ? How many distinct rearrangements are
there of the word KEEP ?

The first question is simple. Because there are four distinct letters, we may
arrange them in 4! different ways, using Theorem 2.1. The second question,
however, requires some effort. We begin by using subscripts, as we did in
the proof of Theorem 2.3. The word KE1 E2 P now consists of four distinct
symbols, and there are accordingly 4! distinct rearrangements. However, when
we remove the subscripts, there are 2! ways of permuting the two Es. Thus, we
have 4!/2! = 12 rearrangements. This is a small enough number we can check
by listing them explicitly. We find EEKP, EEPK, EKEP, EKPE, EPEK,
EPKE, KEEP, KEPE, KPEE, PEEK, PEKE, and PKEE. Each of these turns
into two permutations if we leave the subscripts on the E s. 

Example 2.9: In how many distinct ways can we rearrange the letters of
MADAM IM ADAM ?
 Fundamentals of Enumeration 33

There are 11 letters, of which four are M, 4 are A, two are D, and one is I.
Thus, we have 11!/(4!·4!·2!·1!) = 34, 650. We divide by 4! for the Ms, 4! for
the As, and 2! for the D. 

Example 2.10: In almost every combinatorics textbook, the reader is

shown the number of distinct ways to rearrange the letters of the word MIS-
SISSIPPI. How many ways are there?
Since there is one M, four each of I and S, and two P s, we get 11!/(1!4!4!2!)
or 34, 650. 

THEOREM 2.5 (Permutations of Multisets)

The number of distinct permutations of n objects of r types, with k1 objects
of the first type, k2 of the second, and so on, where two objects of the same
n!
type are considered identical, is .
k1 !k2 ! . . . kr !

Proof. As in the preceding example, we apply subscripts to all objects of a

given type, so that we have n distinct objects. This gives us n! ways to arrange
the distinct objects. Now, we divide by k1 ! to enumerate the ways in which
the objects of the first type might be rearranged, and by k2 ! to rearrange
objects of the second type, and so on. The result follows. 
This theorem may be described as a theorem about permutations of mul-
tisets; the reader will recall that in Chapter 1, we described a multiset as
being able to contain more than one “copy” of a given element. Thus, PEEK
contains one P, one K, and two copies of E. Theorem 2.5 counts the number
of permutations of a multiset.
The binomial theorem may be generalized to powers of more involved sums.
We define n!/(k1 !k2
! . . . kr !) where k1 + k1 + . . . kr = n to be the multinomial
coefficient k1 k2n...kr .

THEOREM 2.6 (The Multinomial Theorem)

 
X n
n
(x1 + x2 + · · · + xr ) = xk1 xk2 . . . xkr r
k1 k2 . . . kr 1 2
k1 +k1 +···+kr =n

where the sum is over all ways to write n as a sum of non-negative integers
k1 + k2 + · · · + kr .

The proof is similar to that of Theorem 2.3 and is omitted. 

Here we look at a problem whose solution involves nk . Ordinarily, nk

counts the number of ways of selecting k of n distinct elements. We wish to

divide up n identical objects into k distinct classes.
34 Introduction to Combinatorics

Example 2.11: If we are given a dozen identical chocolate eggs and wish
to arrange them into four distinct Easter baskets, in how many ways can we
do this? In how many ways can we do this if each basket is to receive at least
one egg?
We are not concerned with fairness; we might give all twelve eggs to one
basket, for instance, and do not need to put three eggs in each. Because the
baskets are distinct, a solution in which all 12 eggs go into the first basket is
different from one in which all 12 go into the second, third, or fourth baskets.
We set this problem up by lining up the twelve eggs, and inserting three
dividers between eggs to indicate which eggs go into which basket. Thus a
diagram like this indicates three eggs in each basket:

OOO|OOO|OOO|OOO

If all twelve eggs were to go into the first basket, the diagram would be this:

OOOOOOOOOOOO|||

Each vertical divider separates the eggs that go into one basket from the
eggs that go into the next. A little thought will reveal that each diagram
corresponds to a distinct way to arrange the eggs, and each arrangement
of eggs is represented by exactly one diagram. Thus we have a one-to-one
correspondence between diagrams and ways to arrange the eggs, and we may
count the diagrams to solve the problem.
How many diagrams are there? With 12 Os and three |s, we have
15 symbols
altogether; and three of these symbols must be |. There are 15 3 = 455 ways
to arrange the eggs. Now, if each basket must contain at least one egg, we
merely place one of our twelve eggs into each basket. This leaves us 8 eggs
and four distinct baskets (each with an egg in
it), and we repeat the problem
by taking 8 eggs and three dividers, to get 11
3 = 165 ways. 
We use the reasoning of the foregoing examples to establish the next theo-
rem.

THEOREM 2.7 (Sorting Identical Objects into Distinct
Classes)

We can sort n identical objects into k distinct classes in n+k−1
k−1 ways.

Proof. We create a string of length n+k −1 that will consist of n objects and
k − 1 dividers. The ith divider will separate the objects that go into class i
from the objects that go into class i + 1. Because there are C(n + k − 1, k − 1)
such strings, there are C(n + k − 1, k − 1) ways to separate the objects. 

Example 2.12: We are organizing a party for small children and need to
buy 18 party hats; the store has 5 different designs. In how many ways can
we select 18 hats, assuming any hats with the same design are identical and
the store has an unlimited number of hats of each design?
 Fundamentals of Enumeration 35

At first glance, this problem does not appear to be solvable using the the-
orem. However, with a little thought we can see that it may be stretched to
apply. Imagine that we have 18 identical hat boxes, and we wish to place the
hat boxes into one of five categories depending on the design of hat that it will
hold. Then by Theorem 2.7 there are C(18 + 5 − 1, 5 − 1) = C(22, 4) = 7, 315
possibilities. 
This example illustrates an idea that arises often enough in different con-
texts that it deserves its own theorem. Just as Theorem 2.5 dealt with the
permutations of a multiset, this deals with combinations of a multiset.

THEOREM 2.8 (Combinations of Multisets)

The number of ways of selecting n (not necessarily distinct) items from a
set of k distinct items where there is no limit on the number of repetitions is
C(n + k − 1, k − 1).

Proof. We imagine n boxes. Each box is to hold one item, and each is to be
classified into one of k classes, depending on which kind of item the box is to
hold. Again, we use Theorem 2.7 to show that there are C(n+k−1, k−1) ways
in which to classify the boxes. We fill each box with a copy of the object of
that type, and after discarding the boxes, we have our desired combination.
There is clearly a bijection between the ways to classify the boxes and the
ways to select the items from the multiset. 
Another common counting problem is that of cyclic permutations. Suppose
we wish to seat n people at a round table with identical seats. At first glance,
this laudably Arthurian arrangement might be accomplished by lining up
our people in one of the n! possible ways, and simply seating them in order.
However, this actually overcounts the number of ways to seat people. Suppose
we moved each person one or more places to the left. This would be the same
seating arrangement, because the same people would be seated next to each
other. Since there are n ways to choose the seat where a selected person sits,
and any of the n seating arrangements we obtain by rotating people left or
right are equivalent, we see that there are actually n!/n = (n − 1)! distinct
cyclic seating arrangements.
Example 2.13: Suppose that we are to seat 10 children at a circular table.
First of all, in how many distinct ways can this be done? In how many if two
of the children are siblings and always sit next to one another? In how many,
if two of the children are mischievous siblings who on no account may sit next
to one another if sanity is to be preserved?
The first question is a simple matter of applying the formula; there are
10!/10 = 9! = 362, 880 distinct seatings. The second question takes some
more thought. Suppose we “glue” the siblings c0 and c1 together into a single
unit. We may do this in two distinct ways; we may glue them so that c0
36 Introduction to Combinatorics

is on c1 ’s left, or on c1 ’s right. Having done this, we have the two siblings

as one unit, so there are 9 “units” to be seated in a circle. We arrive at
the figure 2 · 8! = 80, 640 by the Multiplication Principle. Finally, to count
the arrangements in which the children are kept separate, we subtract from
the total number of arrangements the arrangements in which the children are
adjacent; this leaves 282, 240 arrangements. 

There are very many works that discuss the uses of binomial coefficients,
identities involving them, and interesting uses. Among others, the reader may
benefit from [41] and [63].

2.4 Applications and Subtle Errors

We introduce a number of counting problems by use of a standard deck
of playing cards. There are four suits (clubs, diamonds, hearts, and spades);
each suit consists of thirteen ranks, the ace (or one), the deuce (or two), the
trey (three), the four through ten, the jack, the queen, and the king. Many
card games involve sets (hands) of varying numbers of cards; for instance,
bridge consists of four hands of thirteen cards each and draw poker consists of
a variable number of hands of five cards each. We will consider some subtleties
of counting using cards and similar devices.

Example 2.14: For instance, one poker hand is known as two pair ; this
means that of the five cards, we have two cards of one rank (say, the eight
of clubs and the eight of diamonds); two cards of a different rank (such as
deuces of clubs and spades); and a fifth card of a different rank (say, jack of
clubs). How many such hands are there?

An unsuspecting student might reason as follows: We have thirteen ways

to select the rank of the first pair; that is 13


1 or 13. Once we have chosen
that rank, we choose the suit of the first card of that rank; there are four
suits. Then we choose the suit of the second card of that rank, with three
suits left. Then there are twelve ways to select the rank of the second pair,
four choices for the suit of the first card, and three for the suit of the second
card. Finally, there are 11 ranks left for the fifth card, and four possible
suits for that card. Applying the Multiplication Principle, we appear to have
13 × 4 × 3 × 12 × 4 × 3 × 11 × 4 or 988, 416. This is tragically wrong!
To see why this does not work, let us see how the sample hand given above
might be selected. We might choose an 8 of the thirteen ranks, then choose
a club for the first suit, then a diamond for the second. Then we might
choose a deuce from the 12 remaining ranks, and choose club for our first suit
(from four possibilities) and spade for our second (from the three remaining).
 Fundamentals of Enumeration 37

Finally, of the 11 remaining ranks, we would choose the jack and choose clubs
from the four possible suits.
Suppose instead we chose a deuce first, and chose a spade for the first suit
and a club for the second. Then we chose an 8, and chose a diamond for the
first suit and a club for the second. Then we again chose the jack and clubs for
the fifth card. Notice that, even though we made distinctly different choices,
we still obtain the same hand! This means that each possible hand is being
counted more than once.
How might we choose this hand correctly? The problem arose because the
poker hand does not care whether the eight or the deuce was dealt first; and
it makes no difference for the eight whether we choose diamonds or clubs first.
Thus, we must choose both pairs at once. Now consider that we want two
ranks for our two pairs,
and we wish to choose them from the thirteen ranks
available; that is 13 or 78. Now we choose two suits for one of the pairs; we

2
4
can do this in 2 or 6 ways. Another 42 selects the two suits for the second

pair. Now, as before, we have 11 ranks remaining and four suits for each rank.
Now our total is 78 × 6 × 6 × 11 × 4 or 123, 552. This is exactly one-eighth of
the total we had before (and is correct). 

Example 2.15: A flush is a poker hand in which each of the five cards is
of the same suit; a straight flush is a flush in which the cards are consecutive,
as the 4, 5, 6, 7, and 8 of clubs. A royal flush is a straight flush consisting of
the 10 through ace. How many of each of these hands are there?
Clearly, there are only four royal flushes; one for each suit. (The royal flush
is the strongest hand in poker because it is the rarest.) How many straight
flushes are possible if we do not count royal flushes? We count according to
the lowest card. The ace may be the lowest or the highest card, so a straight
flush may have as its lowest card ace, deuce, trey, and on through 9. Since a
straight flush is entirely determined by its lowest card, and since there are 9
possible lowest cards in each of 4 suits, we have 4 × 9 = 36 straight flushes
that are not royal flushes. Finally, to count flushes, we choose a suit in 4
ways; we choose five cards from that suit in 13


5 ways; and we subtract the
straight and royal flushes from this total. This gives us 5, 108 possible flushes
that are not straight flushes or royal flushes. 

Example 2.16: A hand of bridge consists of 13 cards; each player has

a partner. How many possible ways are there for a player and that player’s
partner to have two hands? Assume it makes no difference if the players
exchange hands.
There are 52


13 possible bridge hands for the player, and the player’s partner
39


can then have 13 possible hands from the remaining cards. We count the
configuration as the same if the two players switch

39hands, so the correct answer
is half the product of these two numbers, 21 52 2.6 × 1021 . 


13 13 , or about
38 Introduction to Combinatorics

Closely tied to the concept of set enumeration is the concept of probability.

We shall discuss probability in the next Chapter.

2.5 Algorithms
Because permutations have such a broad range of applications, we present
two algorithms that may be used to generate permutations. Easy modifica-
tions of these algorithms may be used to produce particular kinds of permu-
tations, or to produce combinations. The first algorithm will produce a list
of all n! permutations on n letters.
Permutation-Generating algorithm: If n = 1, then return the list
consisting only of the letter a1 . If n > 1, then use the algorithm to generate
the list of all permutations on n − 1 letters, using the symbols a1 , . . . an−1 .
Make n copies of each permutation on this list. For each permutation, there
are n places where the new symbol an may be placed (before the first symbol,
between the first and the second, and so on until it is placed at the end). For
the ith copy of this permutation, place an in the ith position, and repeat until
all permutations have been treated this way.
This will become clearer with an example. Consider the case n = 2; we
use the algorithm to generate all permutations consisting of one letter, and we
make n copies of this permutation. This gives us two copies of the permutation
a1 .
The first copy gets an a2 in front; the second copy gets an a2 behind it. This
gives us the two permutations a2 a1 and a1 a2 , which are all the permutations
on two letters. Now we will work through the case of n = 3 in more detail.
The algorithm has given us both permutations on two letters, and we make
three copies of each, which we display in the diagram below.
a1 a2 a2 a1
a1 a2 a2 a1
a1 a2 a2 a1
We now interleave the a3 (boldfaced for clarity) within each to generate all
six permutations on three letters.
a 3 a1 a2 a 3 a2 a1
a1 a 3 a2 a2 a 3 a1
a1 a2 a 3 a2 a1 a 3
The result is the set of six permutations on three letters. This algorithm is
fairly well known, and is discussed in, among other sources, [43] and [13]; our
approach follows that of Brualdi[13].
Often, what is desired is not the set of all permutations on n letters, but
just one, perhaps chosen at random. While this could easily be accomplished
with the foregoing algorithm (merely generate all n! permutations on n letters
 Fundamentals of Enumeration 39

and select one at random), the time and space required for even moderate-
sized values of n is prohibitive. Accordingly, we present an algorithm to select
a permutation on n letters at random.
Random permutation algorithm: Create a list of all n letters in
any order. Each letter is unmarked. Now, for each i between 1 and n, choose
an unmarked letter at random. Now, place that letter in position i of the
permutation and mark the letter in the list. Continue until all letters have
been chosen. No letter may be used twice because at each step the selected
letter is marked so that it cannot be chosen at a later step.

Exercises 2A
1. A program for generating random computer passwords gives one lower-
case letter, one upper-case letter, two digits (0 through 9), and three
letters that may be upper-case or lower-case. How many possible pass-
words can this program produce?

2. At one time, area codes were three-digit numbers with the second digit
0 or 1 and the first digit greater than 1. How many area codes were
possible under this system? How many are possible if we don’t allow 1s
in both the second and third digit (so as not to conflict with 911, 411,
etc.)?

3. A bookstore has a large box of very old books from an estate sale, and
the owner wishes to display one book from the box in each of five window
displays. If there are 24 books in the box, how many ways can this be
done?

4. A number is called a palindrome if its digits read the same forwards as

backwards.
(i) Find the number of palindromes on seven digits.
(ii) Find the number of seven-digit palindromes that have no digit ap-
pearing more than twice.

5. How many of the first million positive integers contain exactly one 3,
one 5, and one 7?

6. A student decides that time is available to join no more than one social
networking site. Of four sites recommended, the student has three ac-
quaintances on the first, five on the second, six on the third, and four on
the fourth. Once joined to any site, the student has for each acquain-
tance the option to befriend or not to befriend that acquaintance. How
40 Introduction to Combinatorics

many possible outcomes are there of the process of choosing a site and
befriending or not befriending each acquaintance on the site?

7. A student decides that time is available to join exactly two social net-
working sites. Of four sites recommended, each has four of the student’s
acquaintances. Once joined to a site, the student has for each acquain-
tance the option to befriend or not to befriend that acquaintance. How
many possible outcomes are there of the process of choosing two sites
and befriending or not befriending each acquaintance on the site? As-
sume that the acquaintances on each site are different.

8. A student decides that time is available to join exactly two social net-
working sites. Of four sites recommended, two have four of the student’s
acquaintances, and two have five. Once joined to a site, the student has
for each acquaintance the option to befriend or not to befriend that
acquaintance. How many possible outcomes are there of the process of
choosing two sites and befriending or not befriending each acquaintance
on the site? Assume that the acquaintances on each site are different.

9. A student wishes to celebrate with a good friend. She has four friends
from which to choose, and one of these friends is under the age of 21.
The restaurants that she may choose from include three that do not
serve anyone under 21 due to state liquor laws, and four with no such
restriction. How many possible ways can the student celebrate assuming
she takes only one friend out? How many ways if she can afford to take
two of her friends out?

10. There are six tutors who are available to handle the 2:00 PM shift at
the Tutoring Center. In how many ways can we select three to assign
to that shift?

11. Compute P (6, 2), P (7, 2), and P (8, 2).

12. Find the row of Pascal’s triangle corresponding to n = 7.

13. Use Pascal’s triangle to expand the expression (x2 + x−1 )5 .

Pn
14. Show that k=1 (−1)k−1 nk = 1 when n is a positive integer.

15. Refer to Figure 2.1. Compute the number of ways to walk from the
southwest corner to the northeast corner if we require that we pass
along block b2. Use this result to count the number of ways if block b2
is closed off and cannot be entered.

16. In how many distinct ways can we arrange the letters in the phrase
SLEEPING STEEPLES ?

17. Use the multinomial theorem to expand (3x + 5y + 7z)3 .

 Fundamentals of Enumeration 41

18. We wish to distribute 10 identical memory cards among three teen-agers

for use in their tablet computers. In how many ways can we do this? In
how many ways if each of the teenagers is to get at least one memory
card?

19. A popular brand of mobile phone cover comes in eight different colors.
We wish to buy three covers. In how many ways can we do this if any
two covers of the same color are considered identical?

20. Using the permutation-generating algorithm to generate all permuta-

tions of {1, 2, 3, 4, 5, 6}, which permutation precedes 356142? Which
one follows it? Which follows 415326?

Exercises 2B
1. A program for generating computer passwords produces two lower-case
letters, one upper-case letter, three digits, and two letters that may
be either upper-case or lower-case, in that order. How many possible
passwords can this program generate?

2. A game store has 10 new games; the manager wants to devote each of
three distinct displays, one to each of three of the new games. How many
ways can the manager select three games, one for each of the displays,
in order?

3. At one time, seven-digit telephone numbers never had a zero or a one

in the first three digits. How many telephone numbers (without area
code) are possible in this system? How many are possible given that we
cannot have any beginning with 555 or with the second and third digits
being 11 (so as not to conflict, for instance, with 911 or 411)?

4. Find the number of 5-digit numbers with no repeated digits.

5. Palindromes were defined in Exercise 2A.4. Find the number of palin-

dromes on 10 digits.

6. A binary sequence is a sequence of 0s and 1s. Find the number of binary

sequences of length n that have an even number of elements 0.

7. There are eight tutors available to handle the 4:00 PM shift at the
Tutoring Center. In how many ways can we choose three of them to
assign to that shift?

8. Compute P (5, 3), P (6, 3), and P (7, 3).

42 Introduction to Combinatorics

9. Find the row of Pascal’s triangle corresponding to n = 8.

10. Use Pascal’s triangle to expand the expression (2x2 + 1)4 .

11. Refer to Figure 2.1. Compute the number of ways to walk from the
southwest corner to the northeast corner if we require that we pass
along block b3. Use this result to count the number of ways if block b3
is closed off and cannot be entered.

12. How many distinct rearrangements of the letters SNIP TIN NIT PINS
are there?

13. Use the multinomial theorem to expand (x + 2y + 3z)3 .

14. There are seven identical memory cards that we wish to distribute
among four teenagers for use in their MP3 music players. In how many
ways can we do this? In how many ways if each of the teenagers is to
get at least one memory card?

15. A popular brand of pocket calculator comes in five different colors. We

wish to buy four calculators. In how many ways can we do this if any
two of the same color are considered identical?

16. Using the permutation-generating algorithm to generate all permuta-

tions of {1, 2, 3, 4, 5, 6}, which permutation follows 2436135? Which one
precedes it? Which precedes 641253?

Problems

Problem 1: Find the total number of integers with distinct digits.

Problem 2: Palindromes were defined in Exercise 2A.4. Prove that any
palindrome of even length is divisible by 11.
Problem 3: Prove the Vandermonde Convolution:
  k   
n+m X n m
=
k i=0
i k−i

Problem 4: Find a combinatorial proof of the identity:

   
n n−1
k =n
k k−1
 Fundamentals of Enumeration 43

Problem 5: Find an algebraic proof of the identity:

   
n n−1
k =n
k k−1

Problem 6: Find a four-letter and a five-letter word where the five-letter

word has fewer distinct rearrangements than the four-letter word. (It is even
possible to find two such words with similar meanings!)
Problem 7: Prove the identity:
X n X n
=
k k
k odd k even

Problem 8: Show that the number of odd-order subsets of a set of n elements

is 2n−1 . Use this result to find the number of even-order subsets.
Problem 9: Using Figure 2.1, find the number of ways to walk from the
southwest to the northeast corner if we must pass through block b1 but cannot
pass through b2.
Problem 10: How many distinct ways are there to rearrange the letters of
the phrase:
(i) BANDS BAN BANDIT BRAND BANDANAS ?
(ii) THE SIXTH SHEIK’S SHEEP’S SICK ?
Problem 11: We have four Republicans and four Democrats who wish to
sit at a round conference table to discuss a bill. In how many ways can they
be seated? In how many ways can they be seated if we require that no two
compatriots (Republicans or Democrats) sit next to each other? How many
ways can we seat four Republicans, four Democrats, and an Independent if
no members of the same party sit next to each other?
Problem 12: We have n pairs of twins. In how many ways can we choose a
committee of n − 1 people from this set of 2n people so that no pair of twins
both serve on the committee?
Problem 13: This problem is for those of you who have been wondering at
the improbably small numbers in problems and exercises up to now. “What
kind of person,” you ask, “would only have five tunes on a digital music player
that can hold 3000 tunes? Who on Earth only knows four people with whom
they would go to a restaurant?” Very well, then, try this.
A student wishes to take two of her 10 friends to a restaurant to celebrate;
three of these friends are vegetarian, and two insist on eating meat. Of the 10
restaurants available, three are vegetarian only and two do not offer enough
vegetarian entrees to interest her vegetarian friends. How many ways can
44 Introduction to Combinatorics

she take two of her friends out to celebrate without violating any of these
restrictions?
Problem 14: Modify the algorithm that produces a random permutation on
n letters to obtain an algorithm to produce a random permutation of k of the
n letters, for any given k ≤ n.
Problem 15: Produce an algorithm to choose a random subset of order k of
a set of n letters.
Chapter 3
Probability

3.1 Introduction
In this chapter we explore some aspects of probability. We talk about
chance in various ways: “there is a good chance of rain today,” “they have no
chance of winning,” “there is about one chance in three,” and so on. In some
cases the meaning is very vague, but sometimes there is a precise numerical
meaning. Probability is the term for the precise meaning.

3.2 Some Definitions and Easy Examples

If we have a set S, we may ask about the consequences of choosing an
element of the set blindly in such a way that any element of the set has the
same likelihood of being chosen as any other. When discussing such choices,
we often call the process of choosing an experiment, and we usually call S the
sample space. The elements of S are often called the outcomes. We define an
event to be a subset E of the sample space. The probability of E is then just
|E| / |S|.
One way to think about the probability that an event will happen. Suppose
the same circumstances were to occur a great many times. In what fraction
of cases would the event occur? This fraction is the probability that the event
occurs. So probabilities will lie between 0 and 1; 0 represents impossibility, 1
represents absolute certainty. Of course, a probability can also be expressed
as a percentage.
For example, consider the question: What is the chance it will rain tomor-
row? We could ask, if the exact circumstances (current weather, time of year,
worldwide wind patterns, etc.) were reproduced in a million cases, in what
fraction would it rain the next day? And while we cannot actually make these
circumstances occur, we can try to get a good estimate using weather records.
Many problems involve a standard deck of playing cards, which we defined
in Section 2.4. Recall that such a deck contains 52 cards and is divided into
four suits: diamonds and hearts are red, clubs and spades are black. (There

45
46 Introduction to Combinatorics

are also decks containing jokers.) Each suit contains an ace, king, queen, jack,
10, 9, 8, 7, 6, 5, 4, 3 and 2, and these are called the 13 denominations. The
cards are ordered as shown, with the ace highest; for some purposes the ace
also counts as the low card (the “one”). When a problem says “a card is dealt
from a standard deck” it is assumed that all possible cards are equally likely,
1
so the probability of any given card being dealt is 52 . As there are four cards
of each denomination, the answer to a question like “what is the probability
4 1
of dealing a queen,” or any other fixed denomination, is 52 , or 13 .
Example 3.1: Suppose we have a game in which each of two players draws
a card from a standard deck of cards, and whoever draws the card of higher
rank wins; in the case that both cards have the same rank, we draw again. If
we draw first, and get (say) a jack, what is the probability that we win the
game?
We will win the game if the other player draws any of the cards below a
jack; we will tie or lose otherwise. In order to satisfy the assumption that
all elements of the sample space (all cards) have the same likelihood of being
drawn, we assume that the cards were well-shuffled. There are 40 cards below
the jack if we count the ace as low, but that is not usual; it is more common
to let the ace be the highest rank, so that there are nine ranks (deuce through
10) below the jack. Because there are four of each rank, that leaves 36 cards.
Since after we have drawn our jack there are 51 cards remaining in the deck,
the probability of a win is 36/51 or (reducing) 12/17. 

Example 3.2: In the above game, suppose we see that our opponent,
having drawn first, draws a nine. What is the probability that we will not
lose?
We consider the question in two ways. First, we may take the probability
that we win and add it to the probability that we have a tie. In this case, we
may win with any rank from among10, jack, queen, king or ace; there are 20
such cards. We may tie with any other nine, and there are three nines besides
the one our opponent has drawn. This gives us 23 “non-losing” cards, so that
our probability of not losing is 23/51 (where the 51 is, as with the previous
example, the number of cards remaining after our opponent’s nine is removed
from the deck).
Another approach works as follows; there are 51 cards remaining in the
deck, and we will lose if we draw any of deuce through eight (seven ranks);
this gives us a probability of losing of 28/51. If we subtract these 28 losing
cards from the 51 cards remaining, we have (51 − 28)/51 = 23/51 as our
probability of not losing. 
The last example demonstrates a standard technique in probability, the
technique of complementary events. Two events are complementary if they
form a partition of the sample space S, or if one is the complement (in the set
 Probability 47

theory sense) of the other. To make the statement more compact, we introduce
the probability function P (E) that assigns to each event E its probability.

THEOREM 3.1 (Probabilities of Complements)

If E1 and E2 are complementary events, then P (E1 ) + P (E2 ) = 1.

Proof. If the sample space S contains n possible outcomes, we observe that

because E1 and E2 are complementary it follows that |E1 | + |E2 | = n. Thus
P (E1 ) + P (E2 ) = (|E1 | + |E2 |)/n = 1. 

Example 3.3: The Florida LottoTM is a game in which players choose six
different numbers between 1 and 53 inclusive. If the player gets the same six
numbers as were randomly drawn by the lottery officials, he or she wins the
jackpot. If the player only gets five of the six numbers, it is a win of a few
thousand dollars; if only four numbers, less than a hundred dollars; and five
dollars for three numbers. Find the probability of each outcome.
The probability of getting the jackpot is quite small; the sample space
consists of all possible C(53, 6) ways to select six numbers from 1 to 53.
This is 22, 957, 480. Since only one of these will win the jackpot, the prob-
ability is therefore just over one in twenty-three million. The probability
of getting five of the six is somewhat more encouraging, though still small;
we choose five of the six winning numbers to go on a ticket, and then we
choose one of the 47 non-winning numbers to go with them. This gives us
C(6, 5)·C(47, 1) or 282 ways to get five of the six winning numbers. Accord-
ingly, our probability is 282/22957480 = 0.000012 or just over a thousandth of
a percent. Four of six gives us more leeway (as well as smaller prize amounts);
we have C(6, 4)·C(47, 2) possible tickets, or 16, 215 ways to get four of the six
winning numbers. The probability here is 16215/22957480 or 0.0007, about
seven percent of one percent. Finally, to get three of the six may be done in
C(6, 3)·C(47, 3) = 324, 300 ways, giving a probability of about 1.4%. 
There are also problems that involve ordinary dice, as used for example in
games like Monopoly. These have six faces, with the numbers 1 through 6
on them. Dice can be biased, so that one face is more likely to show than
another. If we roll an ordinary, unbiased die, what is the probability of rolling
a 5? The six possibilities are equally likely, so the answer is 61 . If the die were
biased, you might try rolling a few hundred times and keeping records.
There are also other dice, with 4, 8, 12 and 20 sides. Usually the die is a
regular polyhedron, and each side contains one number; an n-sided die has
each of 1, 2, . . . , n once.
We say an event is random if you cannot predict its outcome precisely.
This does not mean the chances of different outcomes are equal, although
sometimes people use the word that way in everyday English. For example, if
a die is painted black on five sides, white on one, then the chance of black is 65
48 Introduction to Combinatorics

and the chance of white is 16 . This is random, although the two probabilities
are not equal.
When we talk about the outcomes of a random phenomenon, we mean the
distinct possible results; in other words, no more than one of them can occur,
and one must occur. The set of all possible outcomes is called the sample
space. Each different outcome will have a probability. These probabilities
follow the following rules:
1. One and only one of the outcomes will occur.
2. Outcome X has a probability, P (X), and 0 ≤ P (X) ≤ 1.
3. The sum of the P (X), for all outcomes X, is 1.

3.3 Events and Probabilities

Much of the language of probability developed from the study of experi-
ments. An experiment is defined to be any activity with well-defined, observ-
able outcomes or results. For example, a coin flip has the outcomes “head”
and “tail.” The act of looking out the window to check on the weather has the
possible results “it is sunny,” “it is cloudy,” “it is raining,” “it is snowing,”
and so on. Both of these activities fit the definition of an “experiment.”
The set of all possible outcomes of an experiment is called the sample space.
Each subset of the sample space is called an event. For our purposes, an event
is simply a set of outcomes. For example, when we talk about the weather
we might say there are four possible outcomes: sunny, cloudy, rain or snow.
So we could say the sample space is the set {sunny, cloudy, rain, snow}. The
event “today is fine” consists of the outcomes sunny and cloudy. The event
“It is not snowing” consists of sunny, cloudy and rain. The different events
can occur at the same time (for example, on a sunny day), so we say the
events are not mutually exclusive.
When we roll a die, we are primarily interested in seeing which number is
uppermost, so the sample space is the set S = {1, 2, 3, 4, 5, 6}. The event “an
odd number is rolled,” for example, is the subset {1, 3, 5}, and this event has
three outcomes. A six-element set has 26 subsets, so there are 26 events that
describe the number that is rolled.
Remember, probability 0 means the particular event cannot occur—impossibility.
And probability 1 means the event must occur—certainty.
The probability of an event equals the sum of the probabilities of its out-
comes.
Example 3.4: Consider a die that is weighted so that 1 is rolled one time
in 5 and all other rolls are equally likely. What is the probability of rolling
an even number?
 Probability 49

P (1) = 0.2, so the probabilities of the other five rolls must add to 0.8. They
are equal, so P (2) = P (3) = P (4) = P (5) = P (6) = 0.16. So the probability
of an even roll is 0.16 + 0.16 + 0.16 = 0.48.
Example 3.5: Walter and John roll a die. If the result is a 1 or 6, John
wins $2 from Walter; otherwise Walter wins $1 from John. What is the event
“John wins”? What is the event “Walter wins”?
We defined the event in a die roll, above to be the set of results that produce
the event. So the event “John wins” is {1, 6}, and the event “Walter wins” is
{2, 3, 4, 5}.
The elementary enumeration methods from Chapter 2 will come in handy
when computing probabilities. For example, if event 1 has x possible out-
comes, event 2 has y possible outcomes, and the two events are not related,
then the compound event “event 1 followed by event 2” has xy possible out-
comes. For example, “roll a die, look at score, roll again” has 6 × 6 = 36
outcomes. If an act has n possible outcomes, then carrying it out k times in
succession, independently, has nk possible outcomes.
One convenient way to represent this sort of situation is a tree diagram.

H
H H T
T T H
T

(a) (b)
FIGURE 3.1: Tree diagrams for coin tossing.

Consider the experiment where a coin is tossed; the possible outcomes are
“heads” and “tails” (H and T ). The possible outcomes can be shown in a
diagram like the one in Figure 3.1(a). A special point (called a vertex ) is drawn
to represent the start of the experiment, and lines are drawn from it to further
vertices representing the outcomes, which are called the first generation. If
the experiment has two or more stages, the second stage is drawn onto the
outcome vertex of the first stage, and all these form the second generation,
and so on. The result is called a tree diagram. Figure 3.1(b) shows the tree
diagram for an experiment where a coin is tossed twice.
Example 3.6: A coin is flipped three times; each time the result is written
down. What is the sample space? Draw a tree diagram for the experiment.
The sample space is
{HHH, HHT, HT H, HT T, T HH, T HT, T T H, T T T }.
50 Introduction to Combinatorics

The tree diagram is shown in Figure 3.2(a).

The situation is more complicated where the outcome of the first event can
influence the outcome of the second event, and so on. Again, tree diagrams
are useful. We modify the previous example slightly.
Example 3.7: A coin is flipped up to three times; you stop flipping as
soon as a head is obtained. What is the sample space? Draw a tree diagram
for the experiment.
The sample space is {H, T H, T T H, T T T }. The tree diagram is shown in
Figure 3.2(b).

H
H T
H
T
H T H
T H T H
H
T T H
T H T
T
(a) (b)
FIGURE 3.2: Tree diagrams for Examples 3.6 and 3.7.

Example 3.8: An experiment consists of flipping three identical coins

simultaneously and recording the results. What is the sample space?
The sample space is {HHH, HHT, HT T, T T T }. Notice that the order is
not relevant here, so that for example, the events HHT , HT H and T HH of
Example 3.6 are all the same event in this case.
One common application of tree diagrams is the family tree, which records
the descendants of an individual. The start is the person, the first generation
represents his or her children, the second the children’s children, and so on.
(This is why the word “generation” is used when talking about tree diagrams.)
Since events are sets, we can use the language of set theory in describing
them. We define the union and intersection of two events to be the events
whose sets are the union and intersection of the sets corresponding to the two
events. For example, in Example 3.5, the event “either John wins or the roll is
odd” is the set {1, 6} ∪ {1, 3, 5} = {1, 3, 5, 6}. (For events, just as for sets, “or”
carries the understood meaning, “or both.”) The complement of an event is
defined to have associated with it the complement of the original set in the
sample space; the usual interpretation of the complement E of the event E
is “E doesn’t occur.” Venn diagrams can represent events, just as they can
represent sets.
 Probability 51

The language of events is different from the language of sets in a few cases.
If S is the sample space, then the events S and ∅ are called “certain” and
“impossible,” respectively. If U and V have empty intersection, they are
disjoint sets, but we call them mutually exclusive events (to be consistent
with ordinary English usage).
Example 3.9: A die is thrown. Represent the following events in a Venn
diagram:
A: An odd number is thrown;
B: A number less than 5 is thrown;
C: A number divisible by 3 is thrown.

A
5

1
3
2
6
4
B C

Example 3.10: A coin is tossed three times. Events A, B and C are

defined as follows:
A: The number of heads is even;
B: The number of tails is even;
C: The first toss comes down heads.
Write down the outcomes in A, B, C, A ∪ C, B ∩ C and C. Are any of the
sets mutually exclusive? Write a brief description, in words, of B ∩ C.

A = {HHT, HT H, T HH, T T T },
B = {HHH, HT T, T HT, T T H},
C = {HHH, HHT, HT H, HT T },
A∪C = {HHH, HHT, HT H, HT T, T HH, T T T },
B ∩C = {HHH, HT T },
C = {T HH, T HT, T T H, T T T }.
A and B are mutually exclusive, as are A and B ∩ C. B ∩ C consists of all
outcomes with a head first and the other two results equal.
Suppose all the different outcomes are equally likely. Say S is the sample
space, and E is the set of all outcomes which we can describe as saying a
certain event happens.
For example, rolling a die, say the event is “score 5 or more.” Then

S = {1, 2, 3, 4, 5, 6}
E = {5, 6}
52 Introduction to Combinatorics

and
|E|
P (event happens) = .
|S|
So, in our example, the probability equals 2/6 or 1/3.

3.4 Three Interesting Examples

Probabilities can sometimes be surprising. We shall discuss three examples.

Birthday Probabilities
Suppose there are 30 people in a room. What is the probability that two
of them share the same birthday (day and month)? (To keep the arithmetic
simple, we shall ignore leap years. The probabilities are very similar.) Most
people would say at first that the probability is small, but this is wrong.
First, list the birthdays of the 30 people in the order of their names (al-
phabetical order). If there is no restriction, each person has 365 possible
birthdays. So there are 36530 possibilities.
Suppose no two of them have the same birthday. There are 365 choices for
the first person’s birthday, 364 for the second, 363 for the third, and so on.
With no repeats the total number of possible lists of birthdays is

365 × 364 × . . . × 336.

Therefore the probability of the event “no two have the same birthday” is
365 × 364 × . . . × 336
36530
which is approximately .294. So the probability of the “birthday coincidence”
is about
1 − .294 = .706,
or approximately 70%.
The Game Show
This example was made popular by columnist Marilyn vos Savant.
Consider a TV game show: the contestant chooses between three doors
marked doors 1, 2, 3. Behind one door is a new car; behind each of the other
doors is something almost worthless (a week’s supply of Kleenex tissues; two
movie tickets; a pet goat); call it a booby prize. The contestant chooses a
door, and she gets the prize behind it.
After the choice is announced but before the door is opened, the host opens
one of the other two doors (not the one the contestant chose). Behind the
 Probability 53

door we see a goat. The host then asks, “Do you want to stay with your
original choice? Or would you rather switch to the third door?”
Well, should the contestant stay or switch? Or doesn’t it matter? Most
people would say it doesn’t matter.
To analyze the problem, we need to agree that the host always acts the
same way; he always opens a door to show a goat, then offers the switch. And
on any given night the chance that the car is behind any particular door is
one in three. Moreover, we assume that when the contestant’s first choice is
the door with the car, there is an equal chance that the host will open either
of the other two doors; he doesn’t open the lower-numbered one more often,
or anything like that.
Suppose the doors are numbered 1, 2, and 3. The contestant chooses door
1. We shall write C1 to mean “the car is behind door 1,” C2 for “car behind
door 2” and C3 for “car behind door 3.” Similarly, H2 means “host opens
door 2,” and H3 means “host opens door 3”. (He will not ever open door 1.)
Now suppose the game is played 300 times. We expect the car to be behind
each door in 100 cases. If the car is behind door 1, then we expect the host
will open door 2 in 50 cases and open door 3 in 50 cases. In the 100 cases
where the car is behind door 2, he always opens door 3; in the 100 cases where
the car is behind door 3, he always opens door 2. So we can represent the
data by
C1 C2 C3
H2 50 0 100 .
H3 50 100 0
Now suppose the host opens door 2. This tells us that tonight is one of the
150 cases represented in the row H2. Of those 150 cases, the car is behind
door 1 in 50 cases and behind door 3 in 100 cases. So the odds in favor of
switching are 2 to 1. The same reasoning applies when he opens door 3. So
it is always best to switch.
This problem is based on the game “ Let’s Make a Deal,” with host Monty
Hall, so it is often called the Monty Hall problem. Mr Hall has pointed out
that in the real world, the conditions 1 through 4 do not always apply.
Another Game Show
This example was also published by Marilyn vos Savant. We shall call it
the “three marble problem.”
In this game there are three contestants, Alice, Bruce, and Cathy. The host
produces a bag containing three marbles, colored red, blue and yellow. The
host draws one, then puts it back in the bag. Then the contestants each draw
one at random: Alice first, then Bruce, then Cathy. If the marble drawn is
the same color as one that has already been drawn, that contestant wins, and
the contest is over. Otherwise the marble is replaced and the next candidate
draws. The question is, who has the best chance of winning?
The easiest way to analyze this is to ask what would happen if the game is
played N times. Obviously, Alice has probability 31 of drawing the same color
54 Introduction to Combinatorics

as the host, so she will expect to win in 13 N games. In the remaining games,
Bruce draws. There are now two colors that have been drawn, and if Bruce
duplicates either, he wins. He expects to draw in 32 N games, and the chance
of him winning in those cases is 23 , so the expected number of wins is 23 × 23 N .
So his probability is 49 . If he loses, then all three colors have been drawn, so
Cathy wins in all the remaining cases, and her probability is 1 − 13 − 49 = 92 .
Therefore, Bruce has the best chance of winning.

3.5 Probability Models

A probability model consists of a sample space together with the list of all
probabilities of the different outcomes.
One example is a fair die, whose outcomes 1, 2, 3, 4, 5, 6 have probabilities

1 1
P (1) = 6 P (2) = 6
1 1
P (3) = 6 P (4) = 6 .
1 1
P (5) = 6 P (6) = 6

This is a perfectly good probability model.

Another is a biased die, whose outcomes 1, 2, 3, 4, 5, 6 have probabilities

P (1) = 0.15 P (2) = 0.19

P (3) = 0.18 P (4) = 0.16 .
P (5) = 0.14 P (6) = 0.18

This is still a good probability model, because the probabilities still add to
1.
In some cases there is no numerical value associated with the outcomes.
However, in many probability models, each outcome is a number, or has a
numerical value. For example, this is true of rolling dice.
For example, suppose you are playing a game with three outcomes, call
them A, B, C. If the outcome is A, you win $10. If it is B, you win $2. And
if it is C, you lose $15. The game is completely random, and observations
show that A and C each occur 40% of the time, while B is the result of 20%
of plays.
If the game were played 100 times, your best guess would be that A came
up 40 times, B 20 and C 40. So someone who played 100 times might expect
to win $(10 × 40 + 2 × 20), or $440, and lose $15 × 40 = $600. The net loss is
$160, or $1.60 per play. You would say this is the expected cost of a play.
In general, suppose an occurrence has a numerical value associated with
outcome, and a probability also. The expected value of the occurrence is
found by multiplying each value by the associated probability, and adding.
 Probability 55

This is also called the mean value, or mean of the occurrence. The mean is
most commonly denoted m.
Example 3.11: You are playing a game where you draw cards from a
standard deck. If you draw a 2, 3, . . . , or 10, you score 1 point; a jack, queen
or king is worth 3; and an ace is 5 points. How many points would you get
for a typical draw?
The probability of a draw worth 1 point is 36/52, or 9/13, since 36 of the
52 cards are worth 1 point, Similarly, the probabilities of values 3 and 5 are
3/13 and 1/13 respectively. So the expected value is
9 3 1
×1+ ×3+ × 5 = 23/13.
13 13 13
Similar calculations can be applied to any probability model where the
outcomes have values associated with them.

3.6 Bernoulli Trials

An experiment with exactly two outcomes is called a Bernoulli trial . In-
dependent repetitions of an experiment with exactly two possible outcomes
are also called Bernoulli trials. These are named in honor of the seventeenth
century Swiss mathematician Jacob Bernoulli. Sometimes the two outcomes
are labeled “success” and “failure,” but these are merely labels and should
not be interpreted literally. For example, in Example 3.3 above, event A (the
number of heads is even) could be called either a success or a failure.
Suppose the probability of a success is p and the probability of a failure
is q. Obviously, p + q = 1. These numbers are often expressed as odds: the
odds in favor are p:q, and the odds against are q:p. These odds can also be
expressed as fractions: for example, the odds are p/q in favor. Often, the
odds are simplified to be the ratio of integers. For example, if the probability
of success is .6 and that of failure is .4, then the odds are .6/.4 which equals
3/2, and we say “the odds are 3 to 2 in favor”; if the odds in favor are .4/.6,
it is common to say “the odds are 3 to 2 against.”
A Bernoulli process is a sequence of experiments, each of which is a Bernoulli
trial with the same probability of success. The associated random variable is
the number of successes. In a Bernoulli process consisting of n trials, with
probabilities of success and failure p and q, the probability of k successes is
   
n k n−k n k
P (k) = p q = p (1 − p)n−k ,
k k
and because of the binomial coefficient, this set of probabilities is often called
a binomial distribution.
56 Introduction to Combinatorics

Example 3.12: Suppose a fair coin is tossed four times. What is the
probability that a head occurs exactly twice?
Assume that a head counts as a success. From the formula, with p = q = .5,
n = 4 and k = 2, the probability is
   
n k n−k 4 1 3
P (k) = p q = 0.52 × 0.52 = 6 × 4 = .
k 2 2 8

3.7 The Probabilities in Poker

In regular poker, a standard deck of cards is used. Each person receives a
set of five cards (the hand ), and the players bet against each other; the player
with the most valuable hand wins. There are various methods—in some cases
the players may discard some cards and replace them, in others more than
five cards are dealt and the player chooses which excess cards to delete—but
we will concentrate on the case where five cards are dealt to each player, what
is called straight draw poker. In the other cases the end result depends on the
final five-card hand. The number of possible final hands is
 
52
= 2, 598, 960.
5

A hand consisting of five consecutive cards in the same suit is called a

straight flush. There are 10 such hands in each suit (the ace can be either
high or low) so there are 40 possible straight flushes. These hands are ranked
by their highest cards; in the {5, 4, 3, 2, ace} case the five is the high card.
If the cards are consecutive in value but not in the same suit, the hand is a
straight ; if they are the same suit but not consecutive, it is a flush. Again,
these hands are evaluated by their high cards. Given two hands of the same
kind, the higher card is better.
A pair consists of two cards with the same numerical value; three of a kind
and four of a kind are defined similarly. A hand containing three of a kind of
one denomination and a pair of another is called a full house; two full houses
are ranked according to the three of a kind. In the other cases, the hand with
the higher significant card (the four of a kind denomination in four of a kind,
and so on) is better. A hand containing two pairs of different denominations
is called two pairs and is ranked according to the higher pair. A hand with
no flush or straight and no two cards with the same number is simply called
high card.
It is not hard to calculate the number of hands of the different kinds, and
the probability of receiving such a hand in straight draw poker:
 Probability 57

Poker hand Number Prob

Straight flush 40 0.0015%
Four of a kind 624 0.024%
Full house 3, 744 0.14%
Flush 5, 108 0.20%
Straight 10, 200 0.39%
Three of a kind 54, 912 2.11%
Two pairs 123, 552 4.75%
One pair 1, 098, 240 42.26%
High card 1, 302, 540 50.12%

In poker, the hands on the above list always beat the hands lower than them.
This same ranking is used in all variations of poker.

3.8 The Wild Card Poker Paradox

Another form of poker involves wild cards. The simplest model is where
one additional card, a joker, is added to the deck. A player who receives the
joker can say it is any card she wishes. A card whose meaning can be decided
by the player is called a wild card. There are other models—for example, all
the twos are wild cards, or there are two or more jokers. But we will restrict
ourselves to the one-joker case.

With a 53-card deck, there are 2,869,685 possible hands. We assume that,
if a hand has a joker, the joker is assigned the value that makes the hand
the strongest. For example, given three kings, a four and a joker, the joker
would be a king, and the hand would be a four of a kind. The suit only needs
to be specified when a flush or straight flush is involved. There is one new
type of hand possible, a five of a kind, and it outranks all the other hands.
The numbers of hands of the various types, and their probabilities, are easily
calculated:
58 Introduction to Combinatorics

Poker hand Number Prob

Five of a kind 13 0.0045%
Straight flush 184 0.0064%
Four of a kind 3, 120 0.11%
Full house 6, 552 0.23%
Flush 7, 804 0.27%
Straight 20, 532 0.72%
Three of a kind 137, 280 4.78%
Two pairs 123, 552 4.31%
One pair 1, 268, 088 44.19%
High card 1, 302, 540 45.39%
Note the anomaly: The probability of three of a kind is significantly higher
that the probability of two pairs. This is not surprising. Consider a hand
such as 8, 6, 6, 3, joker. The player could make the joker a six, so that the
hand becomes three of a kind, or an eight, making it two pairs. (It could also
be made a three, but that would be an inferior two pair hand.) The three of a
kind is the better hand, so it is the obvious choice. In all the hands consisting
of one pair, a joker, and two non-matching cards, the result will always be
three of a kind. So the stronger hand, the three of a kind, is more common.
At first sight, one may think that the solution to this is to decide that
two pairs beats three of a kind in wild card poker. But if this is done, the
player would convert the hands in question to two pairs. Then the number of
hands evaluated as two pairs would be 205,920 (probability 7.18%), while the
number of three of a kind hands would be only 54,912 (probability 1.91%).
This unsolvable problem, sometimes called the wild card poker paradox, was
studied in 1996 by Emert and Umbach [30], and also by Gadbois [38], who
showed that the problem is more marked if two jokers are used.

Exercises 3A
1. If the probability that it will rain today is 0.74, what is the probability
that it will not rain?

2. A box contains eight marbles: five red, two blue and one white. One is
selected at random. What is the probability that:
(i) The marble is blue;
(ii) The marble is not blue?

3. Two dice are rolled and the numbers showing are added. What is the
probability that the sum is greater than nine?
 Probability 59

4. An experimenter tosses four coins—two dimes and two nickels—and

records the number of heads. For example, two heads on the nickels and
one on the dimes is recorded (1, 2).
(i) Write down all members of the sample space.
(ii) Write down all members of the following events:
A : There are more heads on the dimes than on the nickels.
B : There are exactly two heads in total.
C : The number of heads is even.
(iii) What are the probabilities of A, B and C?

5. A coin is tossed. If the first toss is a head, the experiment stops. Oth-
erwise, the coin is tossed a second time.
(i) Draw a tree diagram for this experiment.
(ii) List the members of the sample space.
(iii) Assume the coin is fair, so that there is a probability 12 of tossing
a head. What is the probability of the event, “the coin is tossed
twice”?

6. A fair coin is tossed four times. What is the probability that it lands
heads exactly once?

7. We roll two fair six-sided dice and wish to roll doubles, that is, to have
the same number appear on each die. What is the probability of this in
a given roll?

8. The word “SLEEP” is printed on five cards, one letter on each card. The
cards are shuffled and laid out in random order. What is the probability
that the rearrangement will contain the word “EEL” in order from left
to right with no letters intervening (for example, PEELS or SPEEL)?
What is the probability that “LEP” occurs in the same way?

9. Suppose a four-sided die and a six-sided die are rolled together, and the
total is calculated. What are the possible values of the total? Assuming
the dice are fair, what are the probabilities of the different totals?

10. A sample space contains four outcomes: s1 , s2 , s3 , s4 . In each case, do

the probabilities shown form a probability model? If not, why not?
(i) P (s1 ) = .2, P (s2 ) = .5, P (s3 ) = .2, P (s4 ) = .1.
(ii) P (s1 ) = .2, P (s2 ) = .5, P (s3 ) = .6, P (s4 ) = .2.
(iii) P (s1 ) = .1, P (s2 ) = .2, P (s3 ) = .5, P (s4 ) = .2.
(iv) P (s1 ) = .6, P (s2 ) = .2, P (s3 ) = .5, P (s4 ) = −.3.
(v) P (s1 ) = .1, P (s2 ) = .3, P (s3 ) = .3, P (s4 ) = 1.2.

11. A regular deck of cards is shuffled and one card is dealt.

(i) What is the probability that the card is red?
60 Introduction to Combinatorics

(ii) What is the probability that the card is an ace?

(iii) What is the probability that the card is a red ace?
(iv) What is the probability that the card is either red or an ace?

12. Electronic components are being inspected. Initially one is selected from
a batch and tested. If it fails, the batch is rejected. Otherwise a second
is selected and tested. If the second fails, the batch is rejected; otherwise
a third is selected. If all three pass the test, the batch is accepted.
(i) Draw a tree diagram for this experiment.
(ii) How many outcomes are there in the sample space?
(iii) What is the event, “fewer than three components are tested”?

Exercises 3B
1. Exactly one of three contestants will win a game show. The probability
that Joseph wins is 0.25 and the probability that Alice wins is 0.4. What
is the probability that Barbara wins?

2. A student is taking a four-question true/false test. If the student chooses

answers at random, what is the probability of getting all questions cor-
rect?

3. Two fair dice are rolled. What are the probabilities of the following
events?
(i) The total is 7;
(ii) The total is even;
(iii) Both numbers shown are even.

4. The word “TOAST” is printed on five cards, one letter on each card.
The cards are shuffled and laid out in a line, in random order. What is
the probability that the rearrangement will contain the word “TOT” in
order from left to right with no letters intervening (for example, TOTAS
or STOTA)? What is the probability that “ATO” occurs in the same
way?

5. An examination has two questions. Of 100 students 75 answer Question

1 without errors and 70 answer Question 2 without errors; 65 answer
both questions without errors.
(i) Represent these data in a Venn diagram.
(ii) A student’s answer book is chosen at random. What is the proba-
bility that:
 Probability 61

(i) Question 1 contains an error;

(ii) Exactly one question contains an error;
(iii) At least one question contains an error?

6. A bag contains two red, two green and three blue balls. In an experi-
ment, one ball is drawn from the bag and its color is noted, and then a
second ball is drawn and its color noted.
(i) Draw a tree diagram for this experiment.
(ii) How many outcomes are there in the sample space?
(iii) What are the outcomes in the event, “two balls of the same color
are selected”?

7. A die is rolled twice, and the results are recorded as an ordered pair.
(i) How many outcomes are there in the sample space?
(ii) Consider the events:
E : The sum of the throws is 4;
F : Both throws are even;
G : The first throw was 3.
(a) List the members of events E, F and G.
(b) Are any two of the events E, F, G mutually exclusive?
(iii) Draw a tree diagram for this experiment.
(iv) Represent the outcomes of this experiment in a Venn diagram.
Show the events E, F and G in the diagram.
(v) Assuming the die is unbiased, what are the probabilities of E, F
and G?

8. A card is drawn from a standard deck of cards, and then replaced. Then
another card is drawn.
(i) What is the probability that both are red?
(ii) What is the probability that neither is red?

9. An experiment consists of studying families with three children. B rep-

resents “boy,” G represents “girl,” and for example BGG will represent
a family where the oldest child is a boy and the two younger children
are both girls.
(i) What is the sample space for this experiment?
(ii) We define the following events:
E : The oldest child is a boy;
F : There are exactly two boys.
(a) What are the members of E and F ?
(b) Describe in words the event E ∩ F.
(iii) Draw a Venn diagram for this experiment, and show the events E
and F on it.
62 Introduction to Combinatorics

10. At a local supermarket, the number of people in checkout lines varies.

The probability model for the number of people in a randomly chosen
line is

Number in line 0 1 2 3 4 5
Probability 0.08 0.15 0.20 0.22 0.15 0.20

What is the mean number of people in a line?

Problems

Problem 1: In Example 3.1 what is the probability of a roll of 4 or less with

this die?

Problem 2: An experiment consists of flipping a quarter and noting the

result, then flipping two pennies and noting the number of heads. What is
the sample space? Draw a tree diagram.
Problem 3: Represent the following events in Venn diagrams:
A: A regular, unbiased die is thrown and the result is an even number;
B: John wins (in the game of Example 3.3);
C: Walter wins in that game.

Problem 4: In the three marble problem, suppose four different marbles are
used instead of three. Which player has the best chance of winning? Assuming
that each player draws a marble only once, what is the probability that there
is no winner?
Problem 5: Suppose three regular, six-sided dice are rolled, and the two
highest numbers shown are added. What is the probability that this sum is
9? What is the expected value of the sum?
Problem 6: Suppose we roll two fair six-sided dice and wish to roll doubles,
as in Exercise 3A.6.
(i) What is the probability of not getting doubles in one roll? In two
consecutive rolls? In three consecutive rolls?
(ii) What is the probability of getting doubles two or more times in six
attempts?

Problem 7: Verify the calculations of the numbers of hands and probabilities

for regular poker, given in Section 3.7.
 Probability 63

Problem 8: Show that if event E1 has expected value v1 , and E2 has v2 , then
the random event consisting of adding the values of E1 and E2 has expected
value v1 + v2 . Use this result to find the expected value obtained by rolling n
fair k-sided dice when each die has faces numbered 1 through k, and adding
the numbers showing on the upper faces.
Problem 9: Consider all the subsets with four elements of the set {0, 1, . . . , 9}.
The sum of such a set is the sum of the four elements. Without calculating
all the sums or listing all the subsets, show that there must be at least 10
subsets with the same sum.
Chapter 4
The Pigeonhole Principle and
Ramsey’s Theorem

In this chapter we introduce a basic combinatorial principle called the pigeon-

hole principle. It is used in proving the existence of objects that are difficult
to construct. We then go on to the generalization of the principle.

4.1 The Pigeonhole Principle

Our basic idea can be expressed as follows: If more than n pigeons are
roosting in n pigeonholes, then at least one pigeonhole contains more than
one pigeon. We present this principle in a more formal version, which sounds
more mathematical and less ornithological than the above.

THEOREM 4.1 (The Pigeonhole Principle)

(i) If an s-set is partitioned into n disjoint subsets, where s > n, then at least
one of the subsets contains more than one element. P
(ii) Suppose a1 , a2 , . . . , an are real numbers such that nI=1 ai ≥ b. Then for
at least one value of i, i = 1, 2, . . . , n, ai ≥ b/n.

Proof. (i) Suppose the statement is false. Let

S = S1 ∪ S2 ∪ . . . ∪ Sn (4.1)
be a partition of S, where |S| = s > n, Si ∩ Sj = ∅ whenever i 6= j, Si 6= ∅
and |Si | ≤ 1, for i = 1, 2, . . . , n. But then |Si | = 1 for i = 1, 2, . . . , n, and by
(4.1) s = |S| = |S1 | + |S2 | + . . . + |S1 | = 1 + 1 + . . . + 1 = n. Since s > n, this
is a contradiction.
(ii) Again suppose the statement is false, so that we have ai < b/n for every
i, i = 1, 2, . . . , n. Then
n
X
ai < b/n + b/n + . . . + b/n = b.
i=1
Pn
But i=1 ai > b and again we have a contradiction. 

65
66 Introduction to Combinatorics

We can also see that form (ii) of the principle can be used to deduce (i).
Let ai = |Si |. Then by (4.1) we have
n
X
|S| = ai > n + 1;
i=1

by (ii) this implies that for at least one value of i, we have

n+1
ai ≥ > 1.
n

4.2 Applications of the Pigeonhole Principle

Most applications of the pigeonhole principle involve the appropriate parti-
tioning of a set into subsets. Sometimes we must also use some special feature
of the problem and sometimes we need repeated applications of the pigeonhole
principle.
Example 4.1: Show that in a set of eight people there are at least two
whose birthdays fall on the same day of the week this year.
We partition the set of people into seven subsets, the first subset being
those whose birthdays fall on a Monday this year, the second those whose
birthdays fall on a Tuesday, and so on. Since there are more people in our set
than there are days in the week, Theorem 3.1 tells us that at least one subset
contains at least two people; that is, at least two people have birthdays on
the same day of the week. 

Example 4.2: [45] Given N objects to be distributed between B boxes,

what is a necessary and sufficient condition on N and B that in every distri-
bution at least two boxes must contain the same number of objects?
Suppose no two boxes contain the same number of objects; what is the
smallest number of objects we would need? If the first box contained no
objects, the second box one object, and in general the ith box (i − 1) objects,
then we have N = 0+1+. . .+(i−1)+. . .+(B −1) = 21 B(B −1). Similarly, for
any larger value of N , we can arrange the objects so that no two boxes contain
the same number of objects. Hence, a necessary and sufficient condition for
at least two boxes to contain the same number of objects is

N < B(B − 1). 

Example 4.3: [62] Choose an integer n. Show that there exists an integer
multiple of n that contains only the digits 0 and 1 (in the usual decimal
notation).
 The Pigeonhole Principle and Ramsey’s Theorem 67
i
Consider the integers mi = 10 9−1 , for 1 ≤ i ≤ n. For every i, mi is
represented in the decimal scale by a string of i ones. We look at their
residues modulo n: if some mi is divisible by n, the problem is solved; if not,
then the n integers m1 , m2 , . . . , mn must belong to the (n − 1) residue classes
congruent to 1, 2, . . . , n − 1(mod n), so for some i and j, 1 ≤ i < j ≤ n,

mi ≡ mj (mod n).

Now mi j − mi is divisible by n, and it is represented in the decimal scale by

(j − i) ones followed by i zeros. 

Example 4.4: [77] Let A = {a1 , a2 , . . . , a7 } be a set of seven distinct

positive integers none of which exceeds 24. Show that the sums of the non-
empty subsets of A cannot all be distinct.
Since A is a 7-set, it has 71 + 72 + 73 + 74 = 98 non-empty subsets

with at most four elements. The lower bound on the sums of the non-empty
subsets is obviously 1; the upper bound on the sums of the k-sets, for k ≤ 4,
is 21 + 22 + 23 + 24 = 90. Since we have 98 sums that can take at most 90
values, at least two sums must be equal. 

Example 4.5: Suppose 70 students are studying 11 different subjects, and

any subject is studied by at most 15 students. Show that there are at least
three subjects that are studied by at least five students each.
As 70 students take 11 subjects, there must be at least one subject with
at least seven students; call it subject A. At most 15 students take A, so at
least 55 students study the remaining 10 subjects. Hence at least one of these
subjects has at least six students; call it subject B. At most 15 students take
B, so at least 40 students study the remaining nine subjects. Hence at least
one of these subjects has at least five students; call it subject C. Then A, B, C
are the required subjects. 

Example 4.6: [101] Let a1 , a2 , . . . , ak , k > mn, be a sequence of distinct

real numbers. Show that if the number of terms of every decreasing subse-
quence is at most m then there exists an increasing sequence of more than n
terms.
To each term ai of the sequence we assign an ordered pair of integers (m, n)
in the following way: starting with ai , we choose the longest possible decreas-
ing subsequence and we let mi be its number of terms; similarly n is the
largest number of terms of an increasing subsequence beginning with ai . We
claim that for distinct terms ai and aj , i < j, the corresponding pairs (mi , ni )
and (mj , nj ) are distinct. Suppose ai < aj . We know we have an increasing
subsequence of nj terms beginning with aj . Since ai < aj , we can adjoin
ai at the beginning of this subsequence, so that we now have an increasing
68 Introduction to Combinatorics

subsequence of n + 1 terms, beginning with ai . Hence ni ≥ nj + l. Similarly,

if ai > aj , then mi ≥ mj + 1.
So we have k > mn distinct ordered pairs (mi , ni ) and since we assumed
that the number of terms of every decreasing subsequence is at most m we
have 1 ≤ mi ≤ m for all i. If we now have 1 ≤ ni ≤ n for all i, then we can
have at most mn distinct ordered pairs. Since we have more than mn pairs
that are distinct, we must have ni > n for at least one i. But this means we
have at least one increasing subsequence of more than n terms.
In fact, this result is the best possible. For in the sequence m, m−1, . . . , 1, 2m,
2m − 1, . . . , m + 1, 3m, 3m − 1, . . . , 2m + 1, . . . , nm, nm − 1, . . . , (n − 1)m + 1,
we have mn terms; the longest decreasing subsequence has m terms and the
longest increasing subsequence has n terms. 

Example 4.7: [76] A student solves at least one problem every day. Since
she is anxious not to overwork she makes it a rule never to solve more than
730 problems per year. Show that she solves exactly n problems in some set
of consecutive whole days for any given positive integer n.
Consider first the case where n = 1; this gives us a clue to the solution.
Suppose that the student solves exactly two problems per day, throughout any
year with 365 days. She thus solves 730 problems in the whole year and on no
one day has she solved exactly one problem. But during a leap year, solving
exactly two problems per day leads her to exceed her quota: there are at least
two days of a leap year on each of which she must solve exactly one problem.
Since leap years include all those with numbers divisible by 4 but not by 25
(as well as those with numbers divisible by 400), any eight consecutive years
include at least one leap year and contain at least (8 × 365) + 1 = 2921 days.
We start from some fixed date and let Si be the number of problems solved
in the first i days from the starting date. Consider the sets of integers
S = {Si : 1, ≤ i ≤ N = 2921n}
and
T = {n + Si : 1, ≤ i ≤ N = 2921n}.
Notice that Si 6= Sj when i 6= j, since the student solves at least one problem
per day. Similarly n + Si 6= n + Sj when i 6= j. Now
|S| + |T | = 2N = 5842n;
all the elements of S and T are positive integers, the largest of which is n + S,
and
n + Sn ≤ n + (8n × 730) = 5841n.
Hence, there exists at least one positive integer j belonging to both S and T .
So for some h and k,
j = Sh = n + Sk ,
which means that on the days k + 1, k + 2, . . . , h, the student solved exactly
n problems. 
 The Pigeonhole Principle and Ramsey’s Theorem 69

4.3 Ramsey’s Theorem—The Graphical Case

In discussing partitions of the edge-set of a graph it is common to refer to
coloring the edges. A set of labels called colors is used. The set of all edges
that receive a given color is called a color class.
Suppose the edges of a graph G are colored in k colors. We say a subgraph
H of G is monochromatic if all its edges receive the same color; when no
confusion can arise, the k subgraphs formed by all the edges of one color are
called the monochromatic subgraphs for that coloring. A k-coloring of G is
called proper with respect to H if G contains no monochromatic subgraph
isomorphic to H in that coloring. If no subgraph is specified, “proper” will
mean proper with respect to triangles (graphs isomorphic to K3 ). We begin
with an example for the case k = 2.

Example 4.8: Suppose G is a complete graph and its vertices represent

people at a party. An edge xy is colored red if x and y are acquaintances,
and blue if they are strangers. An old puzzle asks: given any six people at a
party, prove that they contain either a set of three mutual acquaintances or
a set of three mutual strangers.
To observe that the result is not true for fewer than six people, consider
the complete graph K5 . It is easy to see that K5 has a proper 2-coloring: in
the example shown in Figure 4.1, color all the heavy edges red and all the
other edges blue.

FIGURE 4.1: K5 in two colors.

On the other hand, there is no proper 2-coloring of K6 . To see this, select

a vertex x in any 2-coloring of K6 . There are five edges touching x, so there
must be at least three of them that receive the same color, say red. Suppose
xa, xb and xc are red edges. Now consider the triangle abc. If ab is red, then
xab is a red triangle. Similarly, if ac or bc is red, there will be a red triangle.
But if none is red, then all are blue, and abc is a blue triangle.
This proves that any 2-coloring of Kv must contain a monochromatic trian-
gle whenever v ≥ 6: if v > 6, simply delete all but six vertices. The resulting
2-colored K6 must contain a monochromatic triangle, and that triangle will
also be a monochromatic triangle in Kv . 
70 Introduction to Combinatorics

LEMMA 4.2
There exists a number R(p, q) such that any coloring of KR(p,q) in two colors
c1 and c2 must contain either a Kp with all its edges in color c1 or a Kq with
all its edges in c2 .

Proof. We proceed by induction on p + q. The lemma is clearly true when

p + q = 2, since the only possible case is p = q = 1 and obviously R(1, 1) = 1.
Suppose it is true whenever p + q < N , for some integer N . Consider any
two positive integers P and Q that add to N . Then P + Q − 1 < N , so both
R(P − 1, Q) and R(P, Q − 1) exist.
Consider any coloring of the edges of Kv in two colors c1 and c2 , where
v ≥ R(P − 1, Q) + R(P, Q − 1), and select any vertex x of Kv . Then x must
either lie on R(P − 1, Q) edges of color c1 or on R(P, Q − 1) edges of color
c2 . In the former case, consider the KR(P −1,Q) whose vertices are the vertices
joined to x by edges of color c1 . Either this graph contains a Kp−1 with all
edges of color c1 , in which case this Kp−1 together with x forms a Kp with
all edges in c1 , or it contains a KQ with all edges in c2 . In the latter case,
the Kv again contains one of the required monochromatic complete graphs.
So R(P, Q) exists, and in fact R(P, Q) ≤ R(P, Q − 1) + R(P − 1, Q). 
(In the example of the six people at the party, we proved that R(3, 3) = 6.)

The same argument can be used when there are more than two colors. The
general result is the graphical version of a result called Ramsey’s Theorem.

THEOREM 4.3 (Ramsey’s Theorem, Graphical Case)

Suppose p1 , p2 , . . . , pn are positive integers. Then there exists an integer

R(p1 , p2 , . . . , pn )

such that if v ≥ R(p1 , p2 , . . . , pn ), then any coloring of the edges of Kv with

n colors will contain a subgraph Kpi with all its edges of color i, for at least
one i.

Proof. We proceed by induction on n. From Lemma 4.2, the result is true

for n = 2. Assume it is true for n = t − 1. Then R(p2 , p3 , . . . , pt ) exists. Write
p0 = R(p2 , p3 , . . . , pt ). Select v ≥ R(p1 , p0 ) and color Kv in colors 1, 2, . . . , t.
Then recolor all edges of color 2, 3, . . . , or t in the new color 0. Either there is
a Kp1 in color 1 or a Kp0 in color 0. In the latter case, revert to the original
colors. Since p0 = R(p2 , p3 , . . . , pt ), the induction hypothesis ensures that the
Kp0 (and therefore the original graph) must contain a monochromatic Kpi in
color i, for some i, 2 ≤ i ≤ t. So the theorem is true for n = t. 
In fact, we have shown that

R(p1 , p2 , . . . , pn ) ≤ R(p1 , R(p2 , p3 , . . . , pn )).

 The Pigeonhole Principle and Ramsey’s Theorem 71

The numbers R(p1 , p2 , . . . , pn ) are called Ramsey numbers. In the particular

case where all n constants pi are equal, to p say, we shall use the notation
Rn (p).
In discussing individual small Ramsey numbers, it is often useful to consider
the monochromatic subgraphs of a given coloring of a complete graph. We
illustrate this with an example.
Example 4.9: Evaluate R(3, 4).
Suppose Kv has been colored in red and blue so that neither a red K3 nor
a blue K4 exists. Select any vertex x. Define Rx to be the set of vertices
connected to x by red edges—that is, Rx is the neighborhood of x in the red
monochromatic subgraph, and similarly define Bx in the blue monochromatic
subgraph.
If |Rx | ≥ 4, then either hRx i contains a red edge yz, whence xyz is a red
triangle, or else all of its edges are blue, and there is a blue K4 . So |Rx | ≤ 3
for all x.
Next suppose |Bx | ≥ 6. Then hBx i is a complete graph on six or more
vertices, so it contains a monochromatic triangle. If this triangle is red, it is
a red triangle in K9 . If it is blue, then it and x form a blue K4 in K9 .
It follows that every vertex x has |Rx | ≤ 3 and |Bx | ≤ 5, so v ≤ 9. But v = 9
is impossible. If v = 9, then |Rx | = 3 for every x, and the red monochromatic
subgraph has nine vertices each of (odd) degree 3, in contradiction of Corollary
1.1.1.

G G

FIGURE 4.2: Decomposition of K8 proving R(3, 4) ≥ 9.

On the other hand, K8 can be colored with no red K3 or blue K4 . The graph
G of Figure 4.2 has no triangle, and can be taken as the red monochromatic
subgraph, while its complement G is the blue graph. (The construction of
this graph will be discussed in Section 4.5, below.) So R(3, 4) = 9. 
72 Introduction to Combinatorics

4.4 Ramsey Multiplicity

We know that any 2-coloring of K6 must contain a monochromatic triangle.
However, it is not possible to find a coloring with exactly one triangle. (This is
easily checked by exhaustion—see the Exercises—and follows from Theorem
4.4 below.) More generally, one can ask: what is the minimum number of
monochromatic triangles in a k-coloring of Kv ?
Such questions are called Ramsey multiplicity problems. The k-Ramsey
multiplicity Nk,v (n) of H in Kv is defined to be the minimum number of
monochromatic subgraphs isomorphic to Kn in any k-coloring of Kv . Clearly
Nk,v (n) = 0 if and only if v < Rk (n).
The 2-Ramsey multiplicity of K3 was investigated by Goodman [47], who
proved the following theorem. Our proof follows that given by Schwenk [100].

THEOREM 4.4 (Goodman’s Theorem)

    2  
n n n−1
N2,n (3) = − .
3 2 2

Proof. Suppose Kv is colored in two colors, red and blue. Write R for the
number of red triangles, B for the number of blue triangles, and P for the
number of partial triangles—triangles with at least one edge of each color.
There are n3 triangles in Kv , so

 
v
R+B+P = .
3

Since N2,v (3) equals the minimum possible value of R + B,

 
v
N2,v (3) = − max(P ).
3

Suppose the vertices of Kv are x1 , x2 , . . ., xv , and xi is incident with ri red

edges. Then it is adjacent to v − 1 − ri blue edges. Therefore the Kv contains
ri (v − 1 − ri ) paths of length 2 in which one edge is red and the other blue.
Let us call these mixed paths. The total number of mixed paths in the Kv is
v
X
ri (v − 1 − ri ).
i=1

The triangle xyz can be considered as the union of the three paths xyz,
yzx and zxy. Moreover, the paths corresponding to different triangles will all
be different. If the triangle is monochromatic, no path is mixed, but a partial
 The Pigeonhole Principle and Ramsey’s Theorem 73

triangle gives rise to two mixed paths. So there are 2P mixed paths in the
Kv , and
v
1X
P = ri (v − 1 − ri ).
2 i=1
If v is odd, the maximum value of ri (v − 1 − ri ) is (v − 1)2 /4, attained when
ri = (v − 1)/2. If v is even, the maximum of v(v − 2)/4 is given by ri = v/2
or (v − 2)/2. In either case, the maximum is
 2 
v−1
,
2
so
 2 
1
Pv v−1
P ≤ 2 L=1 2
  
v v−1
≤ 2 2 12 ,

and since P is an integer,

  2  
v v−1
P ≤ .
2 2
So     2  
v v v−1
N2,v (3) ≥ − .
3 2 2
It remains to show that equality can be attained.
If v is even, say v = 2t, then partition the vertices of Kv into the two sets
{x1 , x2 , . . . , xt } and {xt+1 , xt+2 , . . . , x2t } of size t, and color an edge red if it
has one endpoint in each set, blue if it joins two members of the same set.
(The red edges form a copy of Kt,t .) Each ri equals v/2. If v = 2t + 1, carry
out the same construction for 2t vertices, except color edge xi xt+i blue for
1 ≤ i ≤ ⌊ 2t ⌋. Then add a final vertex x2t+1 . The edges xi x2t+1 and xi+t x2t+1
are red when 1 ≤ i ≤ ⌊ 2t ⌋ and blue otherwise. In both cases it is easy to check
that the number of triangles equals the required minimum. 
Substituting into the formula gives N2,v (3) = 0 when v ≤ 5, N2,6 (3) = 2,
N2,7 (3) = 4, and so on.
The 3-Ramsey multiplicity of K3 has not been fully investigated. We know
that R3 (3) = 17, so the number N3,17 (3) is of special interest. It has been
shown that N3,17 (3) = 5; the argument involves discussion of many special
cases.
The following theorem provides a recursive bound on Nk,v (3).

THEOREM 4.5 (A Ramsey Multiplicity Bound)

   
v−1 3
Nk,v+1 (3) ≤ + 1+ Nk,v (3) .
k v
74 Introduction to Combinatorics

Proof. Suppose F is a k-coloring of Kv that contains Nk,v (3) monochro-

matic triangles. Select a vertex x of F that lies in the minimum number of
monochromatic triangles. Since there are v vertices, one can assume that x
lies on at most ⌊ v3 Nk,v (3)⌋ monochromatic triangles. Since x has degree v − 1
in Kv , there will be a color—say R—such that x lies on at most ⌊ v−1 k ⌋ edges
of color R.
Construct a k-coloring of Kv+1 from F by adjoining a new vertex y. If
z is any vertex other than x, then yz receives the same color as xz, and xy
receives color R. Then xy lies in ⌊ v−1
k ⌋ or fewer monochromatic triangles, all
in color R. The original Kv contained Nk,v (3) monochromatic triangles, so
this is the number not containing x. Finally, the number of monochromatic
triangles with y as a vertex but not x is at most ⌊ v3 Nk,v (3)⌋. So the maximum
number of monochromatic triangles in the Kv+1 is
   
n−1 3

+ 1 + Nk,v (3) . 
k v

This theorem provides the upper bounds 2 and 5 for N2,6 (3) and N3,17 (3),
both of which can be met [47, 96, 120].

4.5 Sum-Free Sets

For this section you need to remember that Zn is the set of integers modulo
n. The residue class of x, the set of all integers leaving remainder x on division
by n, is denoted by [x]. Addition is defined on Z8 by the rule: [x] + [y] =
[x + y]. Two important properties of Zn is that it has an identity element
[0]—[x] + [0] = [x] for any x—and each element [x] has an inverse [−x]—
[x] + [−x] always equals [0]. For example, in Z8 , the inverse of [5] is [3]. For
convenience we shall usually just write x instead of [x]. The set of non-zero
elements of Zn is written Z∗n .
As an introduction we derive the construction of the red graph of Figure
4.2. The vertices of the graph are labeled with the elements of Z8 . The set
Z∗8 is partitioned into two sets:

Z∗8 = {3, 4, 5} ∪ {1, 2, 6, 7}.

Call the two sets R and B, respectively. Then x ∼ y in G if and only if

x − y ∈ R. It follows that two vertices are joined in G if and only if their
difference is in B. Observe that both R and B contain the additive inverses
of all their elements; this is important because the differences x − y and y − x
both correspond to the same edge xy. (This property might be relaxed for
some applications to directed graphs.)
 The Pigeonhole Principle and Ramsey’s Theorem 75

Notice that R contains no solution to the equation

a + b = c;

no element of R equals the sum of two elements of R. We say R is a sum-free

set. By contrast, B is not sum-free; not only is 1 + 6 = 7, but also 1 + 1 = 2
(a, b and c need not be distinct). If xyz were a triangle in G, then x − y, y − z
and x − z would all be members of R; but

x − y + y − z = x − z,

so R would not be sum-free.

In general, a non-empty subset S of Zn is a sum-free set if there never exist
elements a, b of S such that a + b ∈ S. (This means that 0 cannot belong to S,
since 0 + 0 = 0.) S is symmetric if −x ∈ S whenever x is in S. A symmetric
sum-free partition of G is a partition of Z∗n into symmetric sum-free sets. As
examples,

Z∗5 = {1, 4} ∪ {2, 3}

Z∗9 = {3, 4, 5} ∪ {1, 7} ∪ {2, 6}

are symmetric sum-free partitions.

If S is a symmetric sum-free set in Zn , the graph of S is the graph with
vertex-set G, where x and y are adjacent if and only if x − y ∈ S. From our
earlier discussion, it follows that

THEOREM 4.6 (Graphs of Sum-Free Sets)

If S is a symmetric sum-free set of order s in Zn , then the graph of S is a
triangle-free regular graph of degree s on n vertices.

If there is a symmetric sum-free partition S1 ∪ S2 ∪ · · · ∪ Sk of Z∗n , then

one obtains a k-coloring of Kn that contains no monochromatic triangle, by
applying ci to all the edges of the graph of Si for i = 1, 2, . . . , k. So

COROLLARY 4.6.1
If there exists a sum-free partition of an n-element group into k parts, then
Rk (3) > n.
For example, the partition

Z∗5 = {1, 4} ∪ {2, 3}

provides the well-known partition of Z5 into two 5-cycles that is used in prov-
ing that R2 (3) = 6. The partition

Z∗8 = {3, 4, 5} ∪ {1, 7} ∪ {2, 6}

76 Introduction to Combinatorics

yields a (not very interesting) triangle-free three-coloring of K8 .

Students who have studied a little abstract algebra will realize that all of
this can be applied to abelian groups in general. (If you don’t know what this
means, skip to the end of this section.) There are two abelian groups of order
16 that have symmetric sum-free partitions. The group Z4 × Z4 is the set of
all ordered pairs xy where both x and y come from {0, 1, 2, 3} and

xy + zt = (x + z)(y + t)

(additions modulo 4). Then

(Z4 × Z4 )∗ = R ∪ B ∪ G

where

R = {02, 10, 30, 11, 33},

B = {20, 01, 03, 13, 31}, (4.2)
G = {22, 21, 23, 12, 32}.

(Z2 × Z2 × Z2 × Z2 )∗ has a similar partition

R = {1000, 1100, 1010, 1111, 0001},

B = {0010, 0011, 1011, 0111, 1101}, (4.3)
G = {0100, 0110, 0101, 1110, 1001}.

The existence of these partitions proves of course that R3 (3) ≥ 17. To

see that R3 (3) = 17, we use the following argument. Suppose K17 could be
colored in three colors. Select any vertex x. Since there are 16 edges incident
with x, there must be at least six in one color, red say. Consider the subgraph
generated by the other endpoints of those edges. If it has a red edge, then
there is a red triangle; if not, the subgraph is a K6 colored in the two remaining
colors, and it must contain a monochromatic triangle.
The above argument can be used to show that R4 (3) ≤ 66, but there is
no sum-free partition of a 65-element group into four parts. In fact, we know
that 51 ≤ R4 (3) ≤ 65 ([36, 125]). The lower bound was proven by exhibiting
a triangle-free coloring of K50 , while the upper bound comes from a lengthy
argument proving that if a triangle-free four-coloring of K65 existed, then the
adjacency matrices of the monochromatic subgraphs would have eigenvalues
of irrational multiplicities.
The method of sum-free sets can be generalized to avoid larger complete
subgraphs. For example, consider the subset B = {1, 2, 6, 7} of K8 that arose
in discussing R(3, 4). This set is not sum-free, and its graph will contain
triangles. However, suppose there were a K4 in the graph, with vertices a, b, c
and d. Then B would contain a solution to the following system of three
 The Pigeonhole Principle and Ramsey’s Theorem 77

simultaneous equations in six unknowns:

xab + xbc = xac

xac + xcd = xad
xbc + xcd = xbd

(in each case, xij will be either i − j or j − i). But a complete search shows
that B contains no solution to these equations. So the graph contains no K4 .
(The graph is G in Figure 4.2.)

4.6 Bounds on Ramsey Numbers

Very few Ramsey numbers are known. Consequently, much effort has gone
into proving upper and lower bounds.

LEMMA 4.7
If p and q are integers greater than 2, then

R(p, q) ≤ R(p − 1, q) + R(p, q − 1).

Proof. Write m = R(p − 1, q) + R(p, q − 1). Suppose the edges of Km are

colored in red and blue. We shall prove that Km contains either a red Kp or
a blue Kq . Two cases arise.
(i) Suppose that one of the vertices x of Km has at least s = R(p − 1, q)
red edges incident with it, connecting it to vertices x1 , x2 , . . . , xs . Consider
the Ks on these vertices. Since its edges are colored red or blue, it contains
either a blue Kq , in which case the lemma is proved, or a red Kp−1 . Let
the set of vertices of the red Kp−1 be {y1 , y2 , . . . , yp−1 }. Then the vertices
x, y1 , . . . , yp−1 are those of a red Kp and again the lemma holds.
(ii) Suppose that no vertex of Km has R(p − 1, q) red edges incident with
it. Then every vertex must be incident with at least m − 1 − [R(p − 1, q) − 1] =
R(p, q − 1) blue edges. The argument is then analogous to that of part (i). 

THEOREM 4.8 (Upper Bound)

For all integers, p, q ≥ 2,
 
p+q−2
R(p, q) ≤ .
p−1

Proof. Write n = p
+ q. The proof proceeds by induction on n. Clearly,

R(2, 2) = 2 = 2+2−2
2−1 . Since p, q ≥ 2, we can have n = 4 only if p = q = 2.
78 Introduction to Combinatorics

Hence the given bound is valid for n = 4. Also for any value of q, R(2,

q) =
q = 2+q−2 p+2−2


2−1 , and similarly for any value of p, R(p, 2) = p = p−1 , so the
bound is valid if p = 2 or q = 2.
Without loss of generality assume that p ≥ 3, q ≥ 3 and that
 ′
p + q′ − 2

′ ′
R(p , q ) ≤
p′ − 1

for all integers p′ , q ′ and n satisfying p′ ≥ 2, q ′ ≥ 2, p′ + q ′ < n and n > 4.

Suppose the integers p and q satisfy p + q = n.
We apply the induction hypothesis to the case p′ = p − 1, q ′ = q, obtaining
 
p+q−3
R(p − 1, q) ≤ ,
p−2

and to p′ = p, q ′ = q − 1, obtaining
 
p+q−3
R(p, q − 1) ≤ .
p−1
But by the properties of binomial coefficients,
     
p+q−3 p+q−3 p+q−2
+ = ,
p−2 p−1 p−1
and from Lemma 4.7

R(p, q) ≤ R(p − 1, q) + R(p, q − 1),

so
     
p+q−3 p+q−3 p+q−2
R(p, q) ≤ + = . 
p−2 p−1 p−1

If p = 2 or q = 2
or if p
= q = 3, this bound is exact. But suppose p = 3,

q = 4. Then p+q−2
p−1 = 52 = 10, and the exact value of R(3, 4) is 9. Again if
p = 3, q = 5, then p−1 = 62 = 15, whereas the exact value of R(3, 5) is
p+q−2



14. In general, Theorem 4.8 shows that

q2 + q
 
q + 1 q(q + 1)
R(3, q) ≤ = .
2 2 2
But for the case p = 3, this result can be improved. It may be shown that
for every integer q ≥ 2,
q2 + 3
R(3, q) ≤ .
2
The following lower bound for Rn (k) was proved by Abbott [1].
 The Pigeonhole Principle and Ramsey’s Theorem 79

THEOREM 4.9 (Lower Bound)

For integers s, t ≥ 2,

Rn (st − s − t + 2) ≥ (Rn (s) − 1)(Rn (t) − 1) + 1.

Proof. Write p = Rn (s) − 1 and q = Rn (t) − 1. Consider a Kp on vertices

x1 , x2 , . . . , xp and a Kq on vertices y1 , y2 , . . . , yq . Color the edges of Kp and
Kq in n colors c1 , c2 , . . . , cn in such a way that Kp contains no monochromatic
Ks and Kq contains no monochromatic Kt (such colorings must be possible
by the definitions of p and q).
Now let Kpq be the complete graph on the vertices wij , where i ∈ 1, 2, . . . , p
and j ∈ 1, 2, . . . , q. Color the edges of Kpq as follows:

(i) edge wgj wgh is given the color yj yh received in Kq ;

(ii) if i 6= g, wij wgh is given the color yj yh received in Kp .

Now write r = st − s − t + 2 and let G be any copy of Kr contained in Kpq .

Suppose G is monochromatic, with all its edges colored c1 . Two cases arise:

(i) There are s distinct values of i for which the vertex wij belongs to G.
Then from the coloring scheme Kp contains a monochromatic Ks , which
is a contradiction;

(ii) There are at most s − 1 distinct values of i for which wij belongs to G.
Suppose there are at most t− 1 distinct values of j such that wij belongs
to G. Then G has at most (s − 1)(t − 1) = st − s − t + 1 = r − 1 vertices,
which is a contradiction. So there is at least one value of i such that at
least t of the vertices wij belong to G. Applying the argument of case
(i), Kq contains a monochromatic Kt , which is again a contradiction.

Thus, Kpq contains no monochromatic Kr . 

In order to develop the ideas of sum-free sets and obtain some bounds for
Rn (3), we define the Schur function, f (n), to be the largest integer such that
the set
1, 2, . . . , f (n)
can be partitioned into n mutually disjoint non-empty sets S1 , S2 , . . . , Sn , each
of which is sum-free. Obviously f (1) = 1, and f (2) = 4 where {1, 2, 3, 4} =
{1, 4} ∪ {2, 3} is the appropriate partition. Computations have shown that
f (3) = 13 with

{1, 2, . . . , 13} = {3, 2, 12, 11} ∪ {6, 5, 9, 8} ∪ {1, 4, 7, 10, 13}

as one possible partition, that f (4) = 44 and f (5) ≥ 138.

80 Introduction to Combinatorics

LEMMA 4.10
For any positive integer n
f (n + 1) ≥ 3f (n) + 1,
and since f (1) = 1,
3n − 1
f (n) ≥ .
2

Proof. Suppose that the set S = {1, 2, . . . , f (n)} can be partitioned into
the n sum-free sets S1 = {x11 , x12 , . . . , x1ℓ1 }, . . . , Sn = {xn1 , xn2 , . . . , xnℓn }.
Then the sets
T1 = {x11 , 3x11 − 1, 3x12 , 3x12 − 1, . . . , 3x1ℓ1 , 3x1ℓ1 − 1},
T2 = {x21 , 3x21 − 1, 3x22 , 3x22 − 1, . . . , 3x2ℓ1 , 3x2ℓ1 − 1},
...
Tn = {3xn1 , 3xn1 − 1, 3xn2 , 3xn2 − 1, . . . , 3xnℓn , 3xnℓn − 1},
Tn+1 = {1, 4, 7, . . . , 3f (n) + 1}
form a partition of {1, 2, . . . , 3f (n) + 1} into n + 1 sum-free sets. So
f (n + 1) ≥ 3f (n) + 1.
Now f (1) = 1, so the above equation implies that
3n − 1
f (n) ≥ 1 + 3 + 32 + · · · + 3n−1 = . 
2

THEOREM 4.11 (Bounds for Triangles)

For any positive integer
3n + 3
≤ Rn (3) ≤ n(Rn−1 (3) − 1) + 2.
2

Proof. (i) The proof of the upper bound is a generalization of the method
used to establish R3 (3) ≤ 17.
(ii) Let Kf (n)+1 be the complete graph on the f (n) + 1 vertices x0 , x1 , . . . ,
xf (n) . Color the edges of Kf (n)+1 in n colors by coloring xi xj in the kth color
if and only if |i − j| ∈ Sk .
Suppose the graph contains a monochromatic triangle. This must have
vertices xa , xb , xc with a.b.c, such that a − b, b − c, a − c ∈ Sk . But now
(a − b) + (b − c) = a − c, contradicting the fact that Sk is sum-free. Hence
f (n) + 1 ≤ Rn (3) − 1,
so that
3n − 1 3n − 3
+2= ≤ Rn (3),
2 2
which proves the lower bound. 
 The Pigeonhole Principle and Ramsey’s Theorem 81

4.7 The General Form of Ramsey’s Theorem

THEOREM 4.12 (Ramsey’s Theorem)

Suppose S is an s-element set. Write Πr (S) for the collection of all r-element
subsets of S, r ≥ 1. Suppose the partition

Πr (S) = A1 ∪ A2 ∪ · · · ∪ An

is such that each r-subset of S belongs to exactly one of the Ai , and no Ai

is empty. If the integers p1 , p2 , . . . , pn satisfy r ≤ pi ≤ s, for i = 1, 2, . . . , n,
then there exists an integer

R(p1 , p2 , . . . , pn ; r),

depending only on n, p1 , p2 , . . . , pn and r, such that if s ≥ R(p1 , p2 , . . . , pn ; r),

then for at least one i, 1 ≤ i ≤ n, there exists a pi -element subset T of S, all
of whose r-subsets belong to Ai .

The proof is by induction.

Obviously the pigeonhole principle is case r = 1 of Ramsey’s Theorem.
The case r = 2 is the graphical case that we have been discussing; instead of
R(p1 , p2 , . . . , pn , 2) we just write R(p1 , p2 , . . . , pn ).

Exercises 4A
1. How many times must we throw a die in order to be sure that we get
the same score at least twice?

2. Let a1 , a2 , . . . , an be n (not necessarily distinct) integers. Show that

there exist integers s and t such that 1 ≤ s ≤ t ≤ n and as + as+1 . . . +
at ≡ 0(mod n).

3. A night club’s patron parking lot contains 25 cars, and there are 101
patrons in the night club. Assume all of the patrons arrived by car; find
the greatest number k such that some car must have contained at least
k people.

4. An accountant returns from vacation with exactly 10 calendar weeks to

go before the April 15th deadline to file her clients’ income tax state-
ments. She decides to complete at least one tax return per day during
82 Introduction to Combinatorics

the five-day work week, but no more than eight tax returns in any calen-
dar week. Show that there is some set of consecutive work days during
which she completes exactly 19 tax returns.
5. A student in an English class is expected to turn in journal entries.
The class meets four times a week, and the student wants to turn in
at least one journal entry per class day, but no more than six journal
entries during a calendar week for the entire 15-week semester. Show
that there exists some sequence of consecutive days during which the
student turns in exactly 29 journal entries.
6. Let S = {1, 2, . . . , n}, let r be an integer not belonging to S and let
T = S ∪ {r}. Show that there exists x ∈ S such that the sum of all the
elements of T \{x} is divisible by n.
7. Let p be a prime, p 6= 2, p 6= 5. Show that there exists a positive power
of p which, written in the decimal system, ends . . . 01.
8. Let x be any positive real number and let n be any positive integer.
Show that there exist positive integers h and k such that
1 1
h− ≤ kx ≤ h + .
n n
(That is, show that some integral multiple of x is arbitrarily close to an
integer.)
9. What is the relation between R(p1 , p2 , p3 , . . . , pn ; r) and R(p2 , p1 , p3 , . . . ,
pn ; r)?
10. Prove that R(3, 5) ≤ 14.
11. Prove that R(3, 5) > 13 by choosing R = {4, 6, 7, 9}, B = {1, 2, 3, 5, 8, 10,
11, 12} in Z13 . (Note: Together with the preceding exercise, this proves
that R(3, 5) = 14.)
12. Verify that the partition in (4.2) is in fact a symmetric sum-free parti-
tion.
13. Is it possible to 2-color the edges of K35 so that no red K4 or blue K5
occurs?

Exercises 4B
1. How many times must we throw two dice in order to be sure that we
get the same total score at least six times?
 The Pigeonhole Principle and Ramsey’s Theorem 83

2. Let n and B be given positive integers. Show that there exists an integer
divisible by n that contains only the digits 0 and 1 when it is written in
base B.

3. Let A = {al , a2 , . . . , a5 } be a set of five distinct positive integers, none

of which exceeds 9. Show that the sums of the non-empty subsets of A
cannot all be distinct.

4. Let a1 , a2 , . . . , at , t > n2 , be a sequence of distinct real numbers. Show

that this sequence contains a monotonic subsequence of more than n
terms.

5. Suppose we have a 6 × 6 chessboard and 18 dominoes, such that each

domino can just cover two squares of the chessboard with an edge in
common. The 10 lines (five horizontal, five vertical) that divide the
board into squares are called rulings. Show that in any arrangement
of dominoes that precisely covers the chessboard, there is at least one
ruling crossed by no domino.

6. In September, October and November, a student solves at least one

problem per day but not more than 13 problems per week. Show that
she solves exactly 12 problems on some set of consecutive whole days.

7. Let a1 , a2 , . . . , an+1 be n + 1 distinct positive integers. Show that there

exist two of them, ai and aj , such that n divides ai − aj .

8. Verify that, for any p ≥ 2 and any q ≥ 2,

R(2, q) = q, R(p, 2) = p.

9. Prove by construction (not using Theorem 4.4) that any edge-coloring

of K6 in two colors must contain at least two monochromatic triangles.
(Hint: Look at the monochromatic subgraphs.)

10. Prove that any edge-coloring of K7 in two colors must contain at least
three monochromatic triangles.

11. Prove that R(4, 4) ≤ 18.

12. Suppose there exists a painting of K17 in the three colors red, blue and
green that contains two or fewer monochromatic triangles. If v is any
vertex, write R(v), B(v) and G(v) for the sets of vertices joined to v by
red, blue and green edges respectively, and write r(v) = |R(v)|, and so
on.
(i) Select a vertex x that lies in no monochromatic triangle. Prove
that one of {r(x), b(x), g(x)} is equal to 6 and the other two each
equal 5.
84 Introduction to Combinatorics

(ii) Without loss of generality, say r(x) = 6. Let S be the set of all
vertices of K17 that lie in monochromatic triangles. Prove that
S ⊆ R(x).
(iii) If y is any member of B(x), it is clear that S lies completely within
R(y), B(y) or G(y). Prove that, in fact, S ⊆ R(y).
(iv) Prove that there must exist two vertices y1 and y2 in B(x) such
that y1 y2 is red.
(v) Use the fact that S ⊆ R(y1 ) ∩ R(y2 ) to prove that K17 contains
more than two red triangles. So N3,17 (3) ≥ 3.
13. Verify that the partition in (4.3) is in fact a symmetric sum-free parti-
tion.
14. Show that R(4, 4) > 17, by choosing

R = {1, 2, 4, 8, 9, 13, 15, 16}, B = {3, 5, 6, 7, 10, 11, 12, 14}

in Z17 . (Note: Together with Exercise 11, this proves that R(4, 4) = 18.)
15. Is it possible to 2-color the edges of K25 so that no monochromatic K5
occurs?

Problems

Problem 1:
(i) Let S be an s-set, where s > nm. Show that if S is partitioned into n
subsets, then at least one of the subsets contains more than m elements.
Pn
(ii) Let a1 , a2 , . . . , an be real numbers such that i=1 ai > b. Show that for
at least one value of i, i = 1, . . . , n, ai ≥ b/n.
Pn
(iii) Let a1 , a2 , . . . , an be real numbers such that i=1 ai ≤ b. Show that for
at least one value of i, i = 1, . . . , n, ai ≤ b/n.
Pn
(iv) Let a1 , a2 , . . . , an be real numbers such that i=1 ai < b. Show that for
at least one value of i, i = 1, . . . , n, ai < b/n.
Problem 2: Suppose N objects are distributed between B boxes in such a
way that no box is empty. How many objects must we choose in order to be
sure that we have chosen the entire contents of at least one box?
Problem 3: There are 12 signs of the (Western) Zodiac, corresponding
roughly to which month a person is born in. Similarly, there are twelve ani-
mal signs of the so-called “Chinese Zodiac,” (Rat, Ox, Tiger, Rabbit, Dragon,
Snake, Horse, Goat, Monkey, Rooster, Dog, and Boar) corresponding roughly
 The Pigeonhole Principle and Ramsey’s Theorem 85

to which year a person is born in. Suppose there are 145 people in a room.
Show that:
(i) There must be 13 people who share the same sign of the Western Zodiac;
(ii) There must be 13 people who share the same sign of the Chinese Zodiac;
(iii) There must be at least two people who have the same sign in both
Zodiacs.
Problem 4: Let p be a given prime and let R0 , R1 , . . . , Rp−1 be the residue
classes modulo p. Let A = {a1 , a2 , . . . , an } be a set of integers, and define a
partition of A into sets A0 , A1 , . . . , Ap−1 by
Ai = {ai1 , ai2 , . . . , aiki } = A ∩ Ri , 0 ≤ i ≤ p − 1.

(Obviously, ki = |Ai |.) What is a necessary and sufficient condition on n

and p that ki = kj for at least one pair of distinct integers i and j, where
0 ≤ i, j ≤ p − 1?
Problem 5: (For those who have studied a little abstract algebra.) Let
a1 , a2 , . . . , an be n (not necessarily distinct) elements of a group of order n
written multiplicatively. Show that there exist integers s and t such that
1 ≤ s ≤ t ≤ n and as as+1 . . . at = 1.
Problem 6: Choose n2 + 1 closed intervals on the real line. Prove that at
least one of the following is true:
(i) n + 1 of the intervals have a common point;
(ii) n + 1 of the intervals are pairwise disjoint.
Problem 7: Find a three-coloring of the edges of K2R(m,n)−2 so that there
is no K3 of color c1 , no Km of color c2 , and no Kn of color c3 ; find the bound
on R(3, m, n) obtained from this coloring.
Problem 8: Theorem 4.4 uses the quantity
$ $ 2 %%
n n−1
.
2 2
j k
n(n−1)2
For which n does this not simply equal 8 ?

Problem 9: Prove Theorem 4.12.

Chapter 5
The Principle of Inclusion and
Exclusion

5.1 Unions of Events

The addition principle could be restated as follows: if X and Y are two
disjoint sets then
|X ∪ Y | = |X| + |Y |.
But suppose X and Y are any two sets, not necessarily disjoint, and you need
to list all members of X ∪ Y . If you list all members of X, then list all the
members of Y , you certainly cover all of X ∪ Y , but the members of X ∩ Y
are listed twice. To fix this, you could count all members of both lists, then
subtract the number of duplicates:

|X ∪ Y | = |X| + |Y | − |X ∩ Y |. (5.1)
A Venn diagram illustrating this equation is shown in Figure 5.1. To enu-
merate X ∪ Y , first count all elements in the shaded areas and then subtract
the number of elements in the heavily shaded area.

X Y

FIGURE 5.1: Enumerating X ∪ Y using (5.1).

If X is the collection of all the objects of type I in some universal set and
Y consists of all objects of type II, then (5.1) expresses the way to count the
objects that are either of type I or type II:
(i) count the objects of type I;
(ii) count the objects of type II;

87
88 Introduction to Combinatorics

(iii) count the objects that are of both types;

(iv) add answers (i) and (ii) and then subtract answer (iii).
Example 5.1: The makers of Red Hat licorice put coupons in packets of
their product. Each coupon is an anagram of REDHAT, and if your anagram
contains either of the words RED or HAT embedded, you win a free pack
of red licorice hats. For example, HREDTA is a winner; RHEADT is not.
The number of different coupons possible is 6!, or 720. How many winning
anagrams are there?
To calculate the number of winners, we count two sets, R and H. The
collection R of anagrams containing RED can be enumerated as follows: they
are equivalent to ways of organizing the four symbols H, A, T and RED (in
other words, treat “RED” as one symbol). So |R| = 4! = 24. Similarly, H,
the set of those containing “HAT,” has 24 elements. The intersection, R ∩ H,
has only two elements, REDHAT and HATRED. (Unfortunate, that second
one.) From (5.1),
|R ∩ H| = |R| + |H| − |R ∩ H| = 24 + 24 − 2 = 46.
So there are 46 winning anagrams. The percentage of winning anagrams is
about 6.4%. 
Another rule, sometimes called the rule of sum, says that the number of
objects of type I equals the number that are both of type I and of type II,
plus the number that are of type I but not of type II; in terms of the sets X
and Y , this is
|X| = |X ∩ Y | + |X\Y |. (5.2)
For example, suppose Dr. John Walters teaches a course on C ++ and a
course on operating systems. There are 26 students who take both courses,
and 20 who take the C ++ course but not the other. Obviously there are 46
students taking C ++ . This illustrates Equation (5.2): if X is the C ++ class
list and Y is the OS class list, then
46 = |X| = |X ∩ Y | + |X\Y |
= 26 + 20.
The two rules (5.1) and (5.2) can be combined to give
|X ∪ Y | = |Y | + |X\Y |. (5.3)
Figure 5.2 illustrates the last two rules. In the left-hand diagram, X\Y is
dark and X ∩ Y light, and (5.2) tells us to enumerate T by adding those two
parts. In the right-hand diagram, Y is dark and X\Y light, and from (5.3)
we count the contents of the shaded set X ∪ Y by adding those two parts.
We now extend the above work to three sets. In Equation (5.1), replace Y
by a union, Y ∪ Z. Then
|X ∪ Y ∪ Z| = |X| + |Y ∪ Z| − |X ∩ (Y ∪ Z)|.
 The Principle of Inclusion and Exclusion 89

X Y X Y

FIGURE 5.2: Illustration of (5.2) (left) and (5.3) (right).

The term |Y ∪ Z| can be replaced by

|Y | + |Z| − |Y ∩ Z|;

as a consequence of the distributive laws for intersections and unions of sets,

X ∩ (Y ∪ Z) = (X ∩ Y ) ∪ (X ∩ Z), and the last term can be written as

|X ∩ Y | + |X ∩ Z| − |X ∩ Y ∩ Z|.

So the whole expression is

|X ∪ Y ∪ Z| = |X| + |Y | + |Z|
−|X ∩ Y | − |Y ∩ X| − |X ∩ Z|
+|X ∩ Y ∩ Z|.

5.2 The Principle

This formula can be generalized:

THEOREM 5.1 (Principle of Inclusion and Exclusion)

The number of elements in X1 ∪ X2 ∪ · · · ∪ Xn is

|X1 ∪ X2 ∪ · · · ∪ Xn |
Xn n
X Xn
= |Xi1 | − |Xi1 ∩ Xi2 | + · · ·
i1 =1 i1 =1 i2 =i1 +1
n
X X n n
X (5.4)
k−1
+(−1) ··· |Xi1 ∩ Xi2 ∩ · · · ∩ Xik |
i1 =1 i2 =i1 +1 ik =ik−1 +1
+···+ (−1)n−1 |Xi1 ∩ Xi2 ∩ · · · ∩ Xin |.
90 Introduction to Combinatorics

Proof. We can think of the process as: P

1. Count the elements of all the sets (with repetitions), |Xi1 |.
2. Subtract all elements that were counted at PP least twice
(that is, count intersections of two sets), |Xi1 ∩ Xi2 |.
3. Add in all elements that were counted at least P Pthree
P times
(that is, count intersections of three sets), |Xi1 ∩ Xi2 ∩ Xi3 |.
···
Now suppose an object belongs to precisely r of the sets, P say sets Xj1 , Xj2 , . . . ,
Xjr . The element is counted r times in calculating |Xi1 |, and contributes
r to that part of the sum (5.4). It appears 2r times in subsets of the form

Xi1 ∩ Xi2 , since it appears in precisely
those for which {il , i2 } is a 2-set of
{jl , j2 , . . . , jr }; so it contributes − 2r to that part of (5.4). Continuing, we
obtain total contribution
r  
X r
(−1)k−1 , (5.5)
k
k=1

which equals 1 by Corollary 2.3.3 (see Exercise 2A.14). So (5.4) counts 1 for
every member of X1 ∪ X2 ∪ · · · ∪ Xn ; therefore its total is |X1 ∪ X2 ∪ · · · ∪ Xn |.


Example 5.2: The Red Hat licorice makers, following the success of their
first contest, introduce “Red Hat 21,” a casino-type game played with licorice
hats. Again, they run a promotion; this time, the tickets are anagrams of
REDHAT21, and the winning tickets must either contain RED or HAT or 21.
For example, RH21EDTA and HA2RED1A are winners. How many winners
are there this time? What percentage of tickets are winners?

We proceed as we did in Example 5.1, but this time we count three sets, R,
H and T (for “twenty-one”). The anagrams containing RED are equivalent
to ways of organizing the six symbols H, A, T, 2, 1 and RED, so |R| = 6!.
Similarly H, the set of those containing “HAT,” has 6! elements and T has
7! elements. The intersections are counted similarly: R ∩ H, has 4! elements
(RED, HAT, 2 and 1) while |R ∩ T | = |H ∩ T | = 5!. Finally, there are 3!
elements in R ∩ H ∩ T , corresponding to the 3! ways of ordering the three
words RED, HAT and 21. From (5.4), the number of elements of R ∪ H ∪ T
is

R∪H ∪T = (|R| + |H| + |T |)

− (|R ∩ H| + |R ∩ T | + |H ∩ T |)
+ (|R ∩ H ∩ T |)
= (6! + 6! + 7!)
− (4! + 5! + 5!)
+ (3!)
= (720 + 720 + 5040) − (24 + 120 + 120) − 6
= 6480 − 264 + 6 = 6222.
 The Principle of Inclusion and Exclusion 91

So there are 6,222 winners. There are 8! = 40320 tickets, so about 15.4% are
winners. 

Example 5.3: The following diagram represents a set of streets. How

many ways can we walk from the southwest corner to the northeast corner of
the following if all three of the streets labeled b1, b2, and b3 are blocked due
to construction?

All
All
All

Let X be the set of all walks from southwest to northeast. As in Example

2.2, we see that there are |X| = C(12, 5) = 792 total walks. If we let Xi be the
walks that pass through block bi for i = 1, 2, 3,
we
can compute |X1 | (there
are 210 of these, as discussed before), |X2 | ( 63 · 52 = 200 walks), and |X3 |
(C(8, 4) · C(3, 1) = 210 of these). Now, we count the intersections; there are
C(4, 2)·C(5, 2) walks that pass through both b1 and b2, or |X1 ∩X2 | = 60 such
walks. We similarly compute |X1 ∩X3 | = C(4, 2)·C(3, 2)·C(3, 1) walks, or 54.
This comes from counting walks from the southwest corner to the southwest
endpoint of b1, from the northeast endpoint of b1 to the southwest endpoint
of b3, and from the northeast endpoint of b3 to the northeast corner. Next,
|X2 ∩ X3 | = C(6, 3) · C(3, 1) or 60 walks that pass through b2 and b3.
Finally, we ask how many walks include all three forbidden blocks; there
are C(4, 2) · C(3, 1) or 18, because we only count walks to b1 and from b3.
Then the number of walks that include none of these is 792 − (210 + 200 +
210) + (60 + 54 + 60) − 18 = 328 distinct walks. 
Suppose all the sets Xi are subsets of some set S (that is, S is a universal
set for the discussion). Using the general form of de Morgan’s Laws, we get
the following alternative form of (5.4):
|X1 ∩ X2 ∩ · · · ∩ Xn |
= |S| − |X1 ∪ X2 ∪ · · · ∪ Xn |
Xn Xn Xn
= |S| − |Xi1 | + |Xi1 ∩ Xi2 | + · · ·
i1 =1 i1 =1 i2 =i1 +1 (5.6)
n
X n
X X n
+(−1)k ··· |Xi1 ∩ Xi2 ∩ · · · ∩ Xik |
i1 =1 i2 =i1 +1 ik =ik−1 +1
+···+ (−1)n |Xi1 ∩ Xi2 ∩ · · · ∩ Xin |.
92 Introduction to Combinatorics

Suppose we want to know how many of the positive integers smaller than
n are perfect squares, perfect cubes, or perfect higher powers. As an example
we shall find out how many of the positive integers less than 200 are perfect
powers higher than the first. Trivially, 1 is a perfect kth power for every k,
so we omit 1 from our calculations, and add 1 to the answer. In this case we
need only consider up to seventh powers: 28 = 256 > 200, so every eighth or
higher power of x is greater than 200 whenever x ≥ 2. For i = 2, 3, . . . , 6 let
us write Pk for the set of all kth powers of integers that lie between 2 and 199
(inclusive). Then, from (5.4), the number of integers that belong to at least
one of these sets is
|P2 | + |P3 | + |P4 | + |P5 | + |P6 | + |P7 |
− (|P2 ∩ P3 | + |P2 ∩ P4 | + . . . + |P5 ∩ P7 | + |P6 ∩ P7 |)
+ (|P2 ∩ P3 ∩ P4 | + |P2 ∩ P3 ∩ P5 | + . . . + |P5 ∩ P6 ∩ P7 |)
− (|P2 ∩ P3 ∩ P4 ∩ P5 | + . . . + |P4 ∩ P5 ∩ P6 |)
+ (|P2 ∩ P3 ∩ P4 ∩ P5 ∩ P6 | + . . . + |P3 ∩ P4 ∩ P5 ∩ P6 ∩ P7 |)
− |P2 ∩ P3 ∩ P4 ∩ P5 ∩ P6 ∩ P7 |.
All of these sets and their sizes are easy to calculate. The largest square up
to 199 is 142 , so P2 = {2, 3, 4, . . . , 14} and |P2 | = 13, P3 = {2, 3, 4, 5} so
|P3 | = 4, |P4 | = 3, |P5 | = |P6 | = |P7 | = 1. If k > 7 then Pk is empty. In
general Pi ∩ Pj = Pl , where l is the least common multiple of i and j, and
similarly for intersections of three or more, so P2 ∩ P3 = P2 ∩ P6 = P3 ∩ P6 =
P2 ∩ P3 ∩ P4 = P2 ∩ P3 ∩ P6 = P6 , P2 ∩ P4 = P4 , and all the others are empty.
Therefore the total is
|P2 | + |P3 | + |P4 | + |P5 | + |P6 | + |P7 |
− |P2 ∩ P3 | − |P2 ∩ P4 | − |P2 ∩ P6 | − |P3 ∩ P6 |
+ |P2 ∩ P3 ∩ P4 | + |P2 ∩ P3 ∩ P6 |
which equals 13 + 4 + 3 + 1 + 1 + 1 − 1 − 3 − 1 − 1 + 1 + 1 = 19. So the answer
to the original question is 20 (adding 1 for the integer 1).
In some applications of Inclusion/Exclusion, all the sets Xi are the same
size, as are all the intersections of two sets (like Xi1 ∩Xi2 ), all the intersections
of three sets, and so on. If the intersection of any k of the sets Xi always has
sk elements, then (5.4) becomes
   
n n
ns1 − s2 + s3 − · · · + (−1)n−1 sn . (5.7)
2 3
This is called the uniform form of the Principle of Inclusion and Exclusion.
Example 5.4: A bank routing number is a string of nine digits. For
the purposes of this example, let us assume that any digit can be used in
any position, so that there are 109 possible routing numbers (the number of
sequences length 9 on 10 symbols). How many routing numbers contain all
the odd digits?
 The Principle of Inclusion and Exclusion 93

To calculate how many routing numbers contain all the odd digits, we find
out how many do not contain all the odd digits, then subtract that number
from 109 . Write Rx for the set of all routing numbers that do not contain x.
Then we want to calculate |R1 ∪ R3 ∪ R5 ∪ R7 ∪ R9 |. For any x, the number of
routing numbers not containing x equals the number of sequences of length 9
on 9 symbols, so |Rx | = 99 ; similarly, if x 6= y then |Rx ∩ Ry | = 89 ; if x, y, z
are all different then |Rx ∩ Ry ∩ Rz | = 79 , and so on. So, from (5.7), the sum
is      
9 5 9 5 9 5
5×9 − ×8 + ×7 − × 69 + 59
2 3 4
and the answer we require is

109 − 5 × 99 + 10 × 89 − 10 × 79 + 5 × 69 − 59 . 

5.3 Combinations with Limited Repetitions

Suppose we have a set S with n elements. Theorem 2.2 tells us that, if no
repetitions are allowed, the number of ways to choose a collection of size k is
 
n n!
= . (5.8)
k k!(n − k)!

Theorem 2.8 tells us that, if any object can be repeated as often as we wish,
the number of ways to choose a collection of size k is
   
n+k−1 n+k−1 (n + k − 1)!
= = . (5.9)
k n−1 k!(n − 1)!

Instead of repetitions, we could discuss distinguishable and indistinguishable

elements. Suppose we have a set with infinitely many elements of type A and
infinitely many elements of type B. We do not distinguish between elements
of the same type. Then we would say there are three distinguishable subsets
of size 2: AA (that is, a set containing two elements, both of type A), AB
and BB. This is equivalent to saying there are three different ways to choose
a 2-subset of {A, B}, with repetitions allowed. Then (5.9) tells us the number
of distinguishable subsets of size k that can be chosen from a set with n types
of element, and infinitely many of each type.
We now discuss the question, what if there are some indistinguishable ele-
ments, but only finitely many of some or all types?
Suppose a set contains a1 elements of type A1 , a2 elements of type A2 , . . . ,
an elements of type An . We shall denote such a set

{a1 ◦A1 , a2 ◦A2 , . . . , an ◦An }.

94 Introduction to Combinatorics

If there are infinitely many elements of type Ai , we write ∞◦Ai . It is clear

that, in determining the number of subsets of size k, there is no difference
between an element that occurs k times and one that appears infinitely often.
P to count all subsets of size k of the set {a1 ◦A1 , a2 ◦A2 , . . . ,
Suppose you want
an ◦An }, where ai > k. First, consider the collection X of
all k-subsets of
S = {∞◦A1 , ∞◦A2 , . . . , ∞◦An }. From (5.9) there are n+k−1
n−1 such sets. Now
write X1 for the collection of such sets with more than a1 elements A1 , X2 for
the collection of such sets with more than a2 elements A2 , and so on. Then
you want to count all sets belonging to none of the Xi , so the answer is

|X1 ∩ X2 ∩ · · · ∩ Xn |.

From (5.6), this equals

n
X n
X n
X
|X| − |Xi1 | + |Xi1 ∩ Xi2 | + · · ·
i1 =1 i1 =1 i2 =i1 +1
Xn n
X X n
k
+(−1) ··· |Xi1 ∩ Xi2 ∩ · · · ∩ Xik |
i1 =1 i2 =i1 +1 ik =ik−1 +1
+···+ (−1)n |Xi1 ∩ Xi2 ∩ · · · ∩ Xin |.

From (5.9), |X| = n+k−1

n−1 . The set X1 consists of all k-subsets of S with
at least a1 + 1 members of type A1 ; these are equivalent to subsets of size
(k − a1 − 1) (simply add (a1 + 1) more A1 s to each set), so they number
n+(k−a1 −1)−1
= n+k−a 1 −2



n−1 n−1 . The sets X2 , . . . , Xn are enumerated in the
same way. Similarly, X1 ∩ X2 consists of all the subsets with at least (a1 + 1)
A1 s and (a2 + 1) A2 s, and these are derived by adding a1 A1 s
and a2 A2 s to
subsets of size (k − a1 − a2 − 2), so |X1 ∩ X2 | = n+k−a 1 −a2 −3
n−1 , and so on.
Example 5.5: How many subsets of size 10 can be selected from the set

{4◦A1 , 3◦A2 , 6◦A3 }?

Following the preceding notation, |X| = 3+10−1 = 12

2 2 = 66, |X1 | =
7 8 5 3





2 = 21, |X 2 | = 2 = 28, |X 3 | = 2 = 10, |X 1 ∩ X 2 | = 2 = 3, and
|X1 ∩ X3 | = |X2 ∩ X3 | = |X1 ∩ X2 ∩ X3 | = 0. So the answer is 66 − (21 + 28 +
10) + 3 − 0 = 10. 
We chose a small example so that the answer could be checked by hand. If
pqr denotes the subset {p◦A1 , q◦A2 , r◦A3 }, the solutions are

433, 424, 415, 406, 334, 325, 316, 235, 226, 136.

The same technique applies when there are infinitely many elements of some
or all types.
 The Principle of Inclusion and Exclusion 95

Example 5.6: Solve Example 4.5 when you also have unlimited elements
of type A4 .

Now n = 4, so |X| = 13 8 9




3 = 286, |X1 | = 3 = 56, |X2 | = 3 = 84, |X3 | =
6 4



3 = 20, |X1 ∩X2 | = 3 = 4, and |X1 ∩X3 | = |X2 ∩X3 | = |X1 ∩X2 ∩X3 | = 0.
(As type A4 is unrestricted, there is no set X4 .) So the answer to this new
question is 286 − (56 + 84 + 20) + 4 − 0 = 122. Even in this small case, hand
checking is already tedious. 

Just as in the restricted case, there is an application to integer equations.

Example 5.7: How many integer solutions are there to the equation x +
y + z = 14, subject to 3 ≤ x ≤ 7, 1 ≤ y ≤ 4, 0 ≤ z ≤ 6?

Define p = x − 3, q = y − 1, r = z. Then the conditions become p + q + r =

10, 0 ≤ p ≤ 4, 0 ≤ q ≤ 3, 0 ≤ r ≤ 6, and the problem is equivalent to
Example 4.5; the answer is 10.

Example 5.8: Enumerate the solutions to x + y + z ≤ 14 under the same

restrictions.

With the same substitutions, the equation becomes p + q + r ≤ 10, or

equivalently p + q + r + s = 10, where s is any non-negative integer. The
answer, 122, follows from Example 4.6. 

5.4 Derangements
A derangement of an ordered set of objects is a way of rearranging the
objects so that none appears in its original position. A derangement of
(1, 2, . . . , n), for example, is an arrangement (x1 , x2 , . . . , xn ) of the first n
integers in which xi = i never occurs. We write Dn for the number of possible
derangements of n objects; clearly Dn does not depend on the type of objects
in the set.
For consistency, we define D0 = 1. The first few values of Dn are easy
to calculate. For example, D1 = 0 (how can you put the members of a 1-
element set in a different order?) and D2 = 1. The six permutations of 123
are 123, 132, 213, 231, 312, 321, and the only derangements are 231 and 312,
so D3 = 2. But there are nine derangements of four objects, and the number
grows rapidly.
One expression for Dn is derived from the Principle of Inclusion and Ex-
clusion.
96 Introduction to Combinatorics

THEOREM 5.2 (The Derangement Series)

For every positive integer n,

1 1 1 1
Dn = n![1 − + − + · · · + (−1)n ]. (5.10)
1! 2! 3! n!

Proof. We write Xi for the set of all arrangements (x1 ,2 , . . . ,n ) of (1, 2, . . . , n)

in which αi = i. In order to use Inclusion/Exclusion we enumerate Xi for any
i, then Xi ∩ XJ , and so on.
Each arrangement with x1 = 1, x2 = 2, . . . , and xi = i can be derived from
an arrangement of (i + 1‘, i + 2, . . . , n) by putting (1, 2, . . . . , i) in front of it,
and each arrangement of (i + 1‘, i + 2, . . . , n) gives precisely one member of
X1 in this way, so
|X1 ∩ X2 ∩ · · · ∩ Xi | = (n − i)!.
Obviously the intersection of any i sets has the same size, and there are nj

ways of choosing j sets. So

 
XX X n n!
... |Xj1 ∩ Xj2 ∩ · · · ∩ Xji | = (n − i)! = .
i i!

The theorem follows upon substituting this formula into (5.6). 

The series in (5.10) is the expansion of e−1 , truncated after n + 1 terms.
So, by choosing n to be sufficiently large, we can make Dn /n! as close as we
wish to e−1 . In fact, Dn /n! differs from e−1 by less than 1% provided n is at
least 6.
While Theorem 5.2 is useful in theory, it would be very tedious to use it to
calculate large derangement numbers. The following recursive result is much
easier to use.

THEOREM 5.3 (Recursive Derangement Formula)

Dn = (n − 1)(Dn−1 + Dn−2 ).

Proof. The Dn derangements of {1, 2, . . . , n} can be partitioned into n − 1

classes X1 , X − 2, . . . , Xn−1 , where Xk consists of all the derangements with
last element k. (No derangement will have last element n.) To prove the
theorem, it is sufficient to show that each set Xk has Dn−1 + Dn−2 elements.
The elements of Xk can be classified further. Write Yk for the subset of Xk
consisting of all the derangements in Xk with kth element n and Zk for the
set of those whose kth element is not n.
Of the members of Yk , if entries n and k are deleted, the remaining elements
must be a derangement of the (n− 2)-set T = {1, 2, . . . , k − 1, k + 1, . . . , n− 1},
and different arrangements give rise to different members of Yk . So the number
of elements of Yk equals the number of derangements of T , or Dn−2 . The
 The Principle of Inclusion and Exclusion 97

members of Zk are just the derangements of {1, 2, . . . , k−1, n, k+1, . . . , n−1},

so |Zk | = Dn−1 .
Pn−1
So |Xk | = |Yk | + |Zk | = Dn−2 + Dn−1 , and Dn = i=1 |Sk |, giving the
result. 

Example 5.9: Find D5 .

We know D3 = 2 and D2 = 1. So D4 = 3(D3 + D2 ) = 3(2 + 1) = 9. Then
D5 = 4(D4 + D3 ) = 4(9 + 2) = 44. 

Example 5.10: Five people attend a party and all check their coats. When
they leave, they are given the wrong coats; no one gets their own coat. If the
coats were returned at random, what is the chance of this happening?
The coats could have been returned in 5! ways. The chance of a derange-
ment is D5 /5! = 44/120 = 11/30, or approximately 37%. 

Example 5.11: Five couples, man and woman, attend a party and all
check their coats. When they leave, they are given the wrong coats; each man
is given a man’s coat, but no one gets their own coat. In how many ways can
this happen?
Label the couples 1, 2, 3, 4, 5. Say man i gets the coat belonging to man xi .
Then x1 x2 x3 x4 x5 is a derangement of 12345, so there are D5 ways in which
this can happen. As the men’s coats are all accounted for, each woman gets
a woman’s coat, but the wrong one. So this can also happen in D5 ways. By
the multiplication principle, there are D52 = 442 = 1936 ways. 
If we subtract nDn−1 from each side of the result of Theorem 5.3, we obtain
Dn − nDn−1 = −[Dn−1 − (n − 1)Dn−2 ],
provided n ≥ 3. The expression on the right-hand side is derived from the
left-hand side by subtracting 1 from each expression of the variable, and then
negating. Repeating this process, we get
Dn − nDn−1 = (−1)[Dn−1 − (n − 1)Dn−2 ]
= (−1)2 [Dn−2 − (n − 2)Dn−3 ]
= (−1)3 [Dn−3 − (n − 3)Dn−4 ]
= ...
= (−1)n−2 [D2 − 2D1 ].
We know D2 = 1 and D1 = 0. So
Dn − nDn−1 = (−1)n−2 (1 − 0),
or more simply
Dn = nDn−1 + (−1)n . (5.11)
98 Introduction to Combinatorics

Exercises 5A
1. You need to choose a set of four members from {a, b, c, d, e, f, g, h} that
contains either a or b.
(i) In how many ways can this be done?
(ii) In how many ways can it be done if you may not include both a
and b?

2. Find the number of positive integers that are smaller than 200 and are
relatively prime to 21.

3. Find the number of integers between 1 and 5,000 inclusive that are not
divisible by 4, 5 or 6.

4. How many elements are there in the set A ∩ B, if |A| = 28, |B| = 34,
and:
(i) A and B are disjoint?
(ii) |A ∪ B| = 47?
(iii) A ⊆ B?

5. 2,200 voters were surveyed about a new federal building. Of them, 703
were in favor of financing it with a new sales tax, 525 favored a new
property tax, while 336 said they would vote in favor of either measure.
(i) How many favor property tax but not sales tax?
(ii) How many favor sales tax but not property tax?
(iii) How many would vote against either form of funding?

6. Eighty schoolchildren were surveyed about their reading habits. It was

found that 52 had read most or all Harry Potter books, 25 had read
Lemony Snicket and 20 had read Artemis Fowl. Nine read both Harry
Potter and Artemis Fowl, 11 read both Lemony Snicket and Harry Pot-
ter, and eight read both Lemony Snicket and Artemis Fowl. Eight of
the children do not read any of the series.
(i) How many have read all three series?
(ii) How many have read exactly one of the three?

7. Inspired by the makers of Red Hat licorice, the distributors of Red Hat
Linux decide to start a contest. Each contestant receives an anagram
of REDHATLINUX, and is a winner if none of the words RED, HAT,
LINUX appears in its correct order. For example, HLXREDATINU
is not allowed, because of RED. Of the 11! anagrams, how many are
winners?
 The Principle of Inclusion and Exclusion 99

8. Find the number of distinguishable 12-subsets of

{5◦A, 6◦B, 6◦C}.

9. Find the number of distinguishable 12-subsets of

{5◦A, 7◦B, 2◦C, ∞◦D}.

10. Determine the number of integer solutions to

x + y + z + t = 18

that satisfy 2 ≤ x ≤ 4, 1 ≤ y ≤ 6, 3 ≤ z ≤ 6, 4 ≤ t ≤ 7.
11. Find the number of integers between 1 and 6,000 inclusive that are
neither perfect squares nor perfect cubes.
12. How many elements are there in the union of the four sets X1 , X2 , X3 , X4
if each of the sets has precisely 16 elements, each intersection Xi ∩ Xj
has precisely eight elements, each triple of sets has exactly one element
in common, and there is no element common to all four sets?
13. Each of the sets X1 , X2 , X3 , X4 has 24 elements. X1 and X3 are disjoint;
every other intersection of two of the sets has 12 elements. |X1 ∩ X2 ∩
X4 | = 7, |X2 ∩ X3 ∩ X4 | = 1. What is the size of X1 ∪ X2 ∪ X3 ∪ X4 ?
14. Calculate D6 , D7 and D8 .
15. Six women at a party check their coats. In how many ways can the
following occur?
(i) No one gets her own coat back.
(ii) Exactly one woman gets her own coat back.
(iii) At least one woman gets her own coat back.
(iv) Every woman but one gets her own coat back.
16. What is the number of permutations of {1, 2, 3, 4, 5, 6, 7} in which ex-
actly three of the integers are in their original position?

Exercises 5B
1. A survey was carried out. It was found that 20 of the people surveyed
followed football on television, while 15 followed baseball; 12 watched
both football and baseball, but 14 watched neither. Use the formula
(5.1) to find out how many people were interviewed.
100 Introduction to Combinatorics

2. Find the number of positive integers that are smaller than 100 and are
relatively prime to 10.
3. Find the number of integers that are not divisible by 4, 5 or 6:
(i) between 1 and 10,000 inclusive;
(ii) between 5,001 and 10,000 inclusive.
4. How many elements are there in the set S ∪ T , if |S| = 14, |T | = 17,
and:
(i) S and T are disjoint?
(ii) |S ∩ T | = 11?
(iii) S ∩ T is empty?

5. A survey was carried out. It was found that 185 of the people surveyed
drank tea regularly, while 344 were coffee drinkers; 126 regularly drank
both beverages, while 84 drank neither. Use formula (5.1) to find out
how many people were interviewed.
6. Find the number of distinguishable 10-subsets of

{3◦A, 5◦B, 6◦C}.

7. Find the number of distinguishable 9-subsets of

{4◦A, 5◦B, 6◦C, ∞◦D}.

8. Determine the number of integer solutions to

x + y + z + t = 24

that satisfy 0 ≤ x ≤ 4, 2 ≤ y ≤ 8, 5 ≤ x ≤ 10, 2 ≤ x ≤ 5.

9. Find the number of integers between 1 and 10,000 inclusive that are not
perfect squares, perfect cubes or perfect fifth powers.

10. A shopping precinct contains Staples, Office Depot and Best Buy stores.
In a survey of 150 shoppers, it is found that
(a) 87 shopped at Office Depot;
(b) 101 shopped at Staples;
(c) 92 shopped at Best Buy;
(d) 72 shopped at both Office Depot and Staples;
(e) 75 shopped at both Staples and Best Buy;
(f) 62 shopped at both Office Depot and Best Buy;
(g) 52 shopped at all three stores.
(i) How many people shopped only at Office Depot?
 The Principle of Inclusion and Exclusion 101

(ii) How many people shopped at Office Depot and Staples, but not at
Best Buy?
(iii) How many people shopped at exactly one of the stores?
(iv) How many people shopped at none of the three stores?
11. You have five sets, X1 , X2 , X3 , X4 , X5 , each of size 12. X1 , X3 and X5
are pairwise disjoint; |X − 2 ∩ X4 | = 6; every other intersection of two
of the sets has three elements. What is the size of the following sets:
X1 ∪ X2 ∪ X3 ∪ X4 , X2 ∪ X4 ∪ X5 , X1 ∪ X2 ∪ X3 ∪ X4 ∪ X5 ?
12. Eight job applicants are to be interviewed by eight members of a hiring
committee. Each will be interviewed twice. To avoid bias, each candi-
date is assigned their first interviewer at random; the second interviewer
is also chosen at random, but no candidate is to get the same interviewer
twice.
(i) How many different ways are there to schedule the first round of
interviews?
(ii) How many different ways are there to schedule the whole process?

Problems

Problem 1: Derive the following formula, for subsets A, B, C of any universal

set:

|A ∪ C| = |A| + |B ∩ C| + |B ∩ C| − |A ∩ B ∩ C| − |A ∩ B ∩ C|.

Problem 2: Prove that Dn+1 is even if and only if Dn is odd.

Problem 3: At a party, n people are seated in a room with n + 1 chairs.
They go to another room for supper; when they return and sit down again,
it is found that no one occupies the same chair as before. Show that this can
occur in precisely Dn + Dn+1 ways.
Problem 4: Prove
n  
X n
n! = Dk .
k
k=0

(Remember, D0 is defined to be 1.)

Problem 5: Use Equation (5.11) to prove Equation (5.10).
Chapter 6
Generating Functions and
Recurrence Relations

6.1 Generating Functions

Some kinds of combinatorial calculations are modeled by the behavior of in-
finite power series or even finite power series (i.e., polynomials). For instance,
we considered in Chapter 2 how to determine the number of ways of selecting
k items from a set of n; for simplicity, we will consider n − 4, k = 2. Suppose
we wish to multiply (1 + x1 )(1 + x2 )(1 + x3 )(1 + x4 ). In the
product, we
ask how many terms of the form xi xj there will be; clearly, 42 = 6 of them,
since that is how many ways we can choose two of the factors to contribute
an xi . Now, if we set x1 = x2 = x3 = x4 = x, the product is the polynomial
(1 + x)4 . The coefficient of x2 in this product is 42 = 6. (The reader will

recall that this subscripting technique is how we proved Theorem 2.3, the Bi-
nomial Theorem.) In the same way, some numbers of combinatorial interest
(such as a sequence of integers) have properties that become clear or easier
to prove when they are used as coefficients of a polynomial or an infinite se-
ries. These considerations motivate us to define the P generating function of a
sequence a0 , a1 , . . . to be the formal power series i≥0 ai xi .
The expression formal power seriesindexformal power series is understood
to mean that we do not evaluate the power series at any point, or concern
ourselves with the question of convergence necessarily. Indeed, some uses of
generating functions do not require infinite series at all; in this case we shall
assume that an = 0 for all n greater than some specific integer N .

Example 6.1: Suppose we wish to buy a dozen donuts from a shop with
varieties v1 , v2 , v3 , and v4 . We want between three and five donuts of variety
v1 (because the shop has only 5 available, and at least three people will want
one); similarly, we require 1 ≤ |v2 | ≤ 10, 4 ≤ |v3 | ≤ 6, and 0 ≤ |v4 | ≤ 7.

In Chapter 5 we introduced the Principle of Inclusion and Exclusion, and

it could be used to eliminate the cases that violate the given constraints, but
in this case the process would be enormously difficult. Instead, consider the

103
104 Introduction to Combinatorics

product

(x3 + x4 + x5 ) × (x + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 ) ×
(x4 + x5 + x6 ) × (1 + x + x2 + x3 + x4 + x5 + x6 + x7 ) (6.1)

and ask where terms involving x12 originate. If we multiply out the given
polynomials and do not collect like powers of x, a given term of x12 might
come from the x4 in the first factor, the x2 in the second factor, the x6 in
the third, and the 1 in the final factor. This would correspond to a dozen
donuts consisting of 4 of type v1 , 2 of type v2 , 6 of type v3 , and none of type
v4 . Similarly, any other way of making up a dozen donuts satisfying the given
constraints must correspond to exactly one term x12 of the product. Thus we
see that in general the coefficient of xn in the product above determines the
number of ways of making up a box of n donuts satisfying the constraints.
The product is

x8 + 4x9 + 10x10 + 18x11 + 27x12 + 36x13 + 45x14 + 54x15 +

62x16 + 68x17 + 70x18 + 68x19 + 62x20 + 54x21 + 45x22 + 36x23 +
27x24 + 18x25 + 10x26 + 4x27 + x28 (6.2)


Now, multiplying out the product (6.1) above may seem only slightly easier
than enumerating possible solutions by hand; but there are reasons why the
product might be preferable. For instance, we might be faced with the same
problem for 18 donuts, or 10. The polynomial will give us the answers for
those problems with no additional work. The relative prevalence of computer
algebra systems, from Mathematica and Maple on down to some sophisticated
pocket calculators, means that multiplication need no longer be terribly time-
consuming. And finally, algebraic identities can in some cases reduce the effort
of calculating the product.
Example 6.2: Assume that we again want a dozen donuts from the same
shop, but now the shop has an unlimited supply (i.e., at least 12) of each
variety; the lower bound constraints still apply.
Rather than limiting each factor to x12 , we allow each factor to represent
an infinite series. We have

(x3 + x4 + . . . )(x + x2 + . . . )(x4 + x5 + . . . )(1 + x + x2 + . . . ),

and since each factor is a geometric series we can use the familiar formula for
the sum of such a series to see that it is equivalent to the expression

x8 1
= x8
(1 − x)4 1 − 4x + 6x2 − 4x3 + x4 .
 Generating Functions and Recurrence Relations 105

Now, polynomial long division (done with the coefficients in increasing or-
der) will allow us to compute a Maclaurin polynomial, and the coefficient of
x12 will be the solution to our problem. We get

x8 +4 x9 +10 x10 +20 x11 +35 x12 +56 x13 +84 x14 +120 x15+165 x16 +220 x17 +. . .

These figures conform well to our intuition that the numbers should be larger,
as there are fewer constraints (so there are more ways to make up a box of
donuts). 
Just as we can calculate the values for the donut problem above, we can
use infinite series in a formal way to describe any infinite sequence of integers.
Consider the most basic example of this sort, the sequence
P∞ αn = 1. As we saw
in Example 5.2, it will be useful to recall fα (x) = i=1 1xi = x/(1 − x). In
calculus, we were concerned with the values of x for which the series converged;
in combinatorics, we ignore the question.
Another sequence of combinatorial import has a generating function P∞ that
may be simply computed. Suppose βn = n, n ≥ 0. Then fβ (x) = 1 ai xi =
x/(1 − x)2 . This formula is easily derived by differentiating the ordinary in-
finite geometric series, and multiplying by x. Note that since a0 = 0 we
begin the summation at i = 1; if we begin at i = 0, we get the same series.
Differentiating
P∞ 2 iagain, with another multiplication by x, gives us another for-
3
mula: 1 i x = x(1 + x)/(1 − x) . With these formulas, we are able to
find the generating function for any integer sequence produced by a quadratic
polynomial.
Example 6.3: Find the generating function for an = n2 .

We may write an = 21 n2 − 12 n, so that the generating function will be

P∞ 2 P∞
1
f (x) = 21


2 n − 1 n . We sum from 2 because 2 = 0; we could as
easily sum from 0 or 1. Using the values of these sums given above, we see
f (x) = x2 /(1 − x)3 . 
To the elementary manipulations shown above may be added a few others;
for instance, consider the triangular numbers Tn = 1 + 2 + · · · + n n ≥ 1.
The well-known formula is just Tn = n+1


2 . We may re-index the series given
in the foregoing example by letting j = i + 1. So:
∞ ∞ ∞  
X X X j j−1
fT (x) = Ti xi = Tj−1 xj−1 = x .
1 2 2
2
Multiplying the right-hand side by x mustP yield, by the previous example,
∞
x2 /(1 − x)3 ; so the desired result is fT (x) = 1 Ti xi = x/(1 − x)3 .
Now, we pause to point out a pattern P in some of the foregoing results.
n
The sequence αn is related to βn by βn P = i=1 αi . We also notice fα (x) =
n
(1 − x)fβ (x). Upon noting that Tn = i=1 βi , we look at the generating
functions; fT (x) = (1 − x)fβ (x). The pattern holds, and we conjecture and
prove a result.
106 Introduction to Combinatorics

THEOREM 6.1 (Partial Sums) P

Given sequences an and bn with bn = i≤n ai , their generating functions are
related by fa (x) = (1 − x)fb (x).

The proof is left as Problem 6.1.

Other series manipulations will be convenient from time to time; we sum-
marize a few of these in theorem form.

THEOREM 6.2 P∞(Incomplete Sums) P∞

Given fa (x) = i=0 ai xi , we have i
i=k ai x = fa (x) − a0 − a1 x − · · · −
k−1
ak−1 x .

THEOREM 6.3 (Series for Derivative) P

For an and fa (x) as above, we have xfa′ (x) = i≤∞ iai xi .

ai xi and multiply by x.
P
Proof. Differentiate both sides of fa (x) = 
We used the next theorem as part of doing Example 5.3, so we ought to at
least state it.

THEOREM 6.4 (Series for Addition)

If the sequences an and bn have generating functions fa and fb , respectively,
then the sequence c1 an + c2 bn has generating function c1 fa + c2 fb .

We leave the proof to Exercise 6B.9.

There are other identities involving generating functions. We refer the
interested reader to the Problems at the end of this chapter, or to [139].
P∞
Example 6.4: Let an = 3n − 2n , for n ≥ 0. Then fa (x) = i=0 ai xi .
Evaluate this sum.
This may be summed using the geometric series:
∞ ∞ ∞
X X X 1 1 x
(3i − 2i )xi = (3x)i − (2x)i = − =
i=0 i=0 i=0
1 − 3x 1 − 2x (1 − 3x)(1 − 2x)

x
so fa (x) = (1−3x)(1−2x) . 
More generally, it is true that if an is any sequence defined as a linear com-
bination of nth powers, the generating function will be rational. Furthermore,
its denominator will have roots equal to the reciprocals of the bases of the
nth powers. In the last example we had a linear combination of 3n and 2n ,
and the denominator of the generating function has roots 1/3 and 1/2. We
will explore other properties of the denominator later.
 Generating Functions and Recurrence Relations 107

6.2 Recurrence Relations

Frequently in studying a sequence of numbers, we discover that although
we may not have a simple, “closed-form” expression for each number, we may
have some idea of a relationship by which a given term depends upon those
terms that have gone before. We have seen this sort of formula in Chapter 2.
A recurrence relation is said to be linear if the value for an is given as a
finite sum, each summand of which is a multiple of the first power of some
previous term, or a constant or function of n; so, for example, an = nan−1 +2n
is linear, and an = (an−1 )2 − 2n is nonlinear.
Similarly, we say that the recurrence relation is homogeneous if the value
for an is given as a finite sum, each summand of which is a multiple of some
previous terms. So, an = an−1 + 2n + 1 is nonhomogeneous (but linear)
and an = (an−1 )2 is homogeneous (but nonlinear). The coefficients may be
non-constant, so that an = nan−1 is both linear and homogeneous.
The degree of a recurrence is the largest i such that the recurrence depends
upon an−i . We give an = nan−3 −1 as an example of a linear nonhomogeneous
recurrence relation of degree 3. The degree is important because the higher
the degree, the more initial values we need to define a sequence precisely.
If we wish to write down the first several values of an given by an =
(an−1 )2 − an−2 , we must know two initial values, say a0 = 1 and a1 = 2.
From these, we may compute a2 = a21 − a0 = 3, a3 = a22 − a1 = 7, and so on.
These initial values are called the initial conditions. The reader who has taken
a course in differential equations will notice the similarity of the terminology
with that subject. Any sequence of integers may satisfy many recurrence
relations, including linear and nonlinear, homogeneous and nonhomogeneous,
of many different degrees.
It will occasionally be useful to prove that two sequences that appear to
be the same really are the same. One way to do this is to show that the two
sequences satisfy the same recurrence relation with the same initial conditions.
This is a short form of what more formally could be a proof by induction of
the equality of the two sequences.
Example 6.5: Consider the formula an = an−1 + 2an−2 , for n ≥ 2, a0 = 0,
and a1 = 3. Clearly, we can compute any an with this formula by repeatedly
applying the equation to the initial conditions. So a2 = a1 + 2a0 = 3, and
a3 = a2 + 2a1 = 9. Can we find a closed-form formula for an ?
The generating
P function is one of several ways to solve this problem. Con-
sider fa (x) = i≥0 ai xi . We can use the recurrence to evaluate this function.
Multiply both sides of the equation by xn and sum over n to obtain:
∞
X ∞
X ∞
X
an xn = an−1 xn + 2 an−2 xn .
n=2 n=2 n=2
108 Introduction to Combinatorics

Each summation may be interpreted in terms of fa (x), using Theorem 6.2.

We get fa (x)−a0 −a1 x = x(fa (x)−a0 )+2x2 fa (x). Now we can solve for fa (x).
We get fa (x)(1 − x − 2x2 ) = a0 + a1 x − a0 x or f (x) = 3x/(1 − x − 2x2 ). Now,
in Example 5.4, we saw that the reciprocals of the roots of the denominator
of f (x) were used as bases of exponentials in the formula of an . That is, an
involved 2n and 3n , and the denominator had roots 1/2 and 1/3. In this case,
the denominator is (1 − 2x)(1 + x). We simplify using partial fractions.

3x A B
= + ⇒ A = 1, B = −1.
(1 − 2x)(1 + x) 1 − 2x 1 + x
1 1
This yields fa (x) = 1−2x − 1+x ; we can use the geometric series formula
1 P i i 1 P i i
here. We know 1−2x = i≥0 2 x and 1+x = i≥0 (−1) x , so fa (x) =
i i i i n n
P P
i≥0 (2 − (−1) )x = i≥0 ai x . It follows that an = 2 − (−1) . 
We could compute this formula in another way. Suppose we were to as-
sume that the sequence an had a representation as a linear combination of
exponential functions. Then an = rn would give us, via the recurrence, an
equation rn+1 = rn + 2rn−1 . Dividing by rn−1 yields a quadratic equation,
r2 − r − 2 = 0. This is called the characteristic equation of the recurrence.
Finding the roots gives us r = 2 or r = −1, the bases of the exponential
functions. We are now able to surmise that an = c1 2n + c2 (−1)n . Using the
given values for a0 and a1 we are able to solve for c1 and c2 to get the same
formula as above.
That method works for distinct roots. If there are repeated roots, we must
work a little harder.
Example 6.6: Let us consider an+1 = 2an − an−1 , with a0 = 1, a1 = 2.
It is easy to find a closed expression for an using a guess and a proof by
induction, but we will try generating functions. As with our first example, we
multiply by xn+1 and sum over n to arrive at:

fa (x) − a0 − a1 x = 2xfa (x) − 2a0 x − x2 fa (x) or fa (x) = (x − 1)−2 .

Partial fractions will not assist us in this case. Instead, we turn for inspiration
to an earlier example, the sequence βn , which had as its generating function
fβ (x) = x(x − 1)−2 . We write this sequence as βn = (n + 0)(1)n and ask
whether a formula of the type an = (c1 n + c2 )1n will work. A little work
reveals an = n + 1. 

THEOREM 6.5 (Repeated Roots)

If the characteristic equation has a repeated root (x − r)k , this corresponds to
a term of the form (ck−1 nk−1 + · · · + c1 n + c0 )rn in the formula for an .

We omit the proof, but see Problem 6.3.

 Generating Functions and Recurrence Relations 109

Example 6.7: Solve the recurrence an+1 = 4an − 4an−1 , where a0 = −2,
a1 = 2.
The characteristic equation is x2 − 4x + 4 = 0, so the solution is an =
(c1 n + c2 )2n for some c1 and c2 . Using the given values of a0 and a1 we get
−2 = c2 and 2 = 2c1 + 2c2 . Solving for c1 gives an = (3n − 2)2n . 
As a concrete example, we introduce the reader to a sequence of numbers
involved with a problem that is over 800 years old, about rabbits.
The Fibonacci numbers f1 , f2 , f3 , . . . are defined as follows. f1 = f2 = 1,
and if n is any integer greater than 2, fn = fn−1 +fn−2 . This famous sequence
is the solution to a problem posed by Leonardo of Pisa, or Leonardo Fibonacci
(Fibonacci means son of Bonacci) in 1202:

A newly born pair of rabbits of opposite sexes is placed in an

enclosure at the beginning of the year. Beginning with the second
month, the female gives birth to a pair of rabbits of opposite sexes
each month. Each new pair also gives birth to a pair of rabbits of
opposite sexes each month, beginning with their second month.

The number of pairs of rabbits in the enclosure at the beginning of month n

is fn . An English translation of Fibonacci’s work Liber Abaci is available (see
[102]) for those interested.
Some interesting properties of the Fibonacci numbers involve the idea of
congruence modulo a positive integer. We say a is congruent to b modulo n,
written “a ≡ b(mod n),” if and only if a and b leave the same remainder on
division by n. In other words n is a divisor of a − b, or in symbols n | (a − b).
Example 6.8: Prove by induction that the Fibonacci number fn is even
if and only if n is divisible by 3.
Assume n is at least 4. fn = fn−1 + fn−2 = (fn−2 + fn−3 ) + fn−2 =
fn−3 + 2fn−2 , so fn ≡ fn−3 (mod 2).
We first prove that, for k > 0, f3k is even. Call this proposition P (k).
Then P (1) is true because f3 = 2. Now suppose k is any positive integer,
and P (k) is true: f3k ≡ 0(mod 2). Then (putting n = 3k + 3) f3(k+1) ≡
f3k ( mod 2) ≡ 0( mod 2) by the induction hypothesis. So P (k + 1) is true; the
result follows by induction. To prove that, for k > 0, f3k−1 is odd—call this
proposition Q(k)—we note that Q(1) is true because f1 = 1 is odd, and if
Q(k) is true, then f3k−1 is odd, and f3(k+1)−1 ≡ f3k−2 (mod 2) ≡ 1(mod 2).
We have Q(k + 1) and again the result follows by induction. The proof for
k ≡ 1(mod 3) is similar. 
This sequence is one of the most-studied sequences of integers in all of
mathematics. Generating functions make it possible for us to understand
many properties of the sequence. Suppose f (x) is the generating function for
the Fibonacci numbers. We multiply both sides of the recurrence by xi+1 and
110 Introduction to Combinatorics

sum as in Example 5.5.

∞
X ∞
X ∞
X
i+1 i+1
fi+1 x = fi x + fi−1 xi+1
i=2 i=2 i=2

Careful use of the formula for f (x) will help us to represent this equation
in more understandable
P∞ terms. So, using Theorem 6.2, the first term in the
equation is i=2 fi+1 xi+1 = (f (x) − f1 x − f2 x2 ). We substitute similarly for
the other two terms to get
f (x) − f1 x − f2 x2 = xf (x) − f1 x2 + x2 f (x)
which we may solve for f (x), to get
x = f (x)(1 − x − x2 ) or f (x) = x/(1 − x − x2 )
We may confirm that polynomial long division, as in the earlier example,
produces the Fibonacci numbers (see Problem 6.4). But there is a more
useful and fascinating consequence of our calculation of f (x). Recall that for
the earlier case of the sequence an = 3n − 2n in Example 6.4, the generating
function was also a rational function, and that the bases of the exponents were
reciprocals of the roots of the denominator. Alternatively, as in Example 6.5,
we may use partial fractions or the characteristic equation.
First, we assume that the Fibonacci numbers may be represented by an
expression of the form fn = rn , r ≥ 1. Then the recurrence gives us rn+1 =
rn + rn−1 , and dividing by rn−1 gives us 2
√ a quadratic r − r −√ 1 with two real
roots, traditionally denoted φ1 = (1 + 5)/2 and φ2 = (1 − 5)/2. To find
fn we simply assume fn = c1 φn1 + c2 φn2 and use the initial values to solve for
c1 and c2 .
√ √ √ √
1+ 5 1− 5 6+2 5 6−2 5
f 1 = 1 = c1 + c2 , f 2 = 1 = c1 + c2
2 2 4 4
√ √
This gives us c1 = 1/ 5 and c2 = −1/ 5. This formula for the Fibonacci
numbers is sometimes called Binet’s formula, although Binet was not its first
discoverer. Different starting values yield a different sequence, but it is not
hard to see that the values f1 = 0, f2 = 1 will give the same numbers with
different subscripts, and likewise f1 = 1, f2 = 2. So, the “first” (in some sense)
set of initial values that gives us a different sequence is f1 = 1, f2 = 3. This
sequence has also been studied extensively; it is more commonly called the
sequence of Lucas numbers, after the mathematician Edouard Lucas. There
is a formula for the nth Lucas number Ln similar to the Binet formula for
Fibonacci numbers; see Problem 6.5.
The Fibonacci numbers represent the number of ways of writing an integer
as an ordered sum of 1s and 2s. More precisely, fn is the number of ways of
writing n − 1 as such a sum, for n ≥ 2. The proof of this statement comes
from asking how the sum begins. There are fn ways to write n − 1, and we
place a 1 at the beginning of such sequences to create some of the ways of
 Generating Functions and Recurrence Relations 111

writing n as such a sum. Then there are fn−1 ways to write n − 2; we may
place a 2 at the beginning of any of these sequences to obtain ways of writing
n. It follows that fn+1 = fn + fn−1 . Now, f1 = 1 (which counts the empty
sum only) and f2 = 1 since we can only write 1 as a sum of a single 1 and no
2s. Thus we have the same recurrence and initial conditions as the Fibonacci
sequence; it follows by mathematical induction that the sequences are the
same. The Fibonacci numbers arise in other contexts as well; see Problem 6.7
for an example.
Our next example will concern a construction by the Greek mathemati-
cian Theon of Smyrna. At the time when Theon lived, through the early
second century AD, it was already known that the square root of two was
not rational. However, there was work done on finding a rational sequence
of approximations to the value, and it was Theon who found (see [44]) the
process that has become known as Theon’s ladder. We begin with the first
“rung” of the ladder by setting a0 = b0 = 1. Each successive rung is created
by setting an = an−1 + 2bn−1 and bn = an−1 + bn−1 . Then the rational num-
ber rn = an /bn is our nth approximation.
√ This gives us a1 = 3 and b1 = 2,
so that our second approximation to 2 is r1 = 1.5. Continuing, we will find
a2 = 7, b2 = 5, and r2 = 7/5 = 1.4.
Example 6.9: Find a closed form for the quantities an and√ bn and use
this closed form to show that the ratio rn indeed approaches 2.
We clearly cannot deal with these two intertwined recurrences as they stand,
so we must try to eliminate one of the sequences. We observe that the second
recurrence may be rewritten as an−1 = bn −bn−1 and therefore an = bn+1 −bn .
We substitute these values into the first recurrence to get
bn+1 − bn = bn − bn−1 + 2bn−1 or bn+1 = 2bn + bn−1.
Similar calculations will give us an+1 = 2an + an−1 , the same recurrence.
Using the characteristic equation and the values of a1 , a2 , b1 , and b2 as given
above, we arrive at the formulas for an and bn .
√ √ √ √
(1 + 2)n+1 + (1 − 2)n+1 (1 + 2)n+1 − (1 − 2)n+1
an = bn = √
2 2 2
Now we look at√the ratio rn = an /bn and take the limit as n goes to ∞.
We note that (1 − 2)n+1 is less than 1 in absolute value and goes rapidly to
0 as n increases.
√ √
√ (1 + 2)n+1 + (1 − 2)n+1
rn = 2 √ √
(1 + 2)n+1 − (1 − 2)n+1
√
When we replace the (1 − 2)n+1 with 0, we get the result. 
112 Introduction to Combinatorics

6.3 From Generating Function to Recurrence

We have seen that we can use a recurrence for a sequence to develop a
generating function. This process works both ways. Recall that for a sequence
defined by a linear combination of exponential functions we discovered that
the bases were the roots of the characteristic function, and were the reciprocals
of the roots of the polynomial in the denominator of the rational function.
This means that the denominator of the generating function is related to the
recurrence relation.

THEOREM 6.6 (Rational Generating Function)

If a sequence an has a rational generating function with denominator q(x) =
1+c1 x+· · ·+cn xn then it satisfies the recurrence an +c1 an−1 +· · ·+cn a0 = 0.

Proof. We will prove explicitly the case in which the denominator is a

quadratic with distinct non-zero roots, say 1/r1 and 1/r2 . This implies the
denominator is (1 − r1 x)(1 − r2 x), and an = d1 r1n + d2 r2n for some coefficients
d1 and d2 . Since the denominator looks like 1 − (r1 + r2 )x + r1 r2 x2 , the
recurrence we are to satisfy is an − (r1 + r2 )an−1 + r1 r2 an−2 = 0. We easily
confirm this by replacing each ai in the recurrence by d1 r1i + d2 r2i . Other cases
are done similarly. 

Example 6.10: Find a recurrence satisfied by the numbers of boxes of

donuts from Example 5.1.
Since the denominator is 1 − 4x + 6x2 − 4x3 + x4 , we see that the number of
ways an of filling a box of n donuts with at least three of variety v1 , at least
one of variety v2 and at least four of variety v3 satisfies an − 4an−1 + 6an−2 −
4an−3 + an−4 = 0, or an = 4an−1 − 6an−2 + 4an−3 − an−4 . 

Example 6.11: In the last section, we discussed the Fibonacci numbers,

and noted that they enumerate ways to write n as an ordered sum of 1s and
2s. Now, let us ask about unordered sums. That is, in how many ways can we
write an integer n as a sum of 1s and 2s where we consider sums equivalent if
they contain the same numbers of 1s and 2s?
Let this quantity be an . It is not too hard to see that this is merely the
n 2 4 2
coefficient
P 2i P ofi x in the product (1 n+ x + x + . . . )(1 + x + x + . . . ) =
( x )( x ). We get a term of x for each solution to the equation n =
2m+k, which means for each possible factor x2m xk , where the x2m comes from
the first sum and the xk from the second. The generating function, found via
−1
geometric series, is just (1 − x2 )(1 − x) , with denominator 1 − x− x2 + x3 .


This says that our recurrence is an = an−1 + an−2 − an−3 . We can compute
the value of an for small n; doing so for several values of n will give us initial
 Generating Functions and Recurrence Relations 113

conditions for the recurrence, and allow us to check our results.

n 0 1 2 3 4 5 6 7 8 9 10
an 1 1 2 2 3 3 4 4 5 5 6
The recurrence works nicely, and we suspect that the formula for an is just
an = ⌈ n+1
2 ⌉. In fact, this formula is clear when we think about it for a
moment; to write n as a sum of 1s and 2s, we may write it as a sum of n 1s
or n − 2 1s and a single 2, or (up to) ⌊n/2⌋ 2s and a 1 or not depending on
the parity of n. 
We finish with an example that does not use real numbers. Just as Binet’s
formula for the Fibonacci numbers always produces integers despite irrational
quantities, in the same way some recurrences always produce integers despite
complex quantities.
Example 6.12: Solve the recurrence an+1 = 2an − an−1 + 2an−2 with
initial values a0 = a1 = 6, a2 = 4.
We get a characteristic equation of x3 − 2x2 + x − 2 = 0, which factors
nicely to get x = ±i and x = 2. We conclude that a general solution is of the
form c1 in + c2 (−i)n + c3 · 2n . We get three equations in three unknowns using
the given initial values; so that c1 + c2 + c3 = a0 = 6, and (c1 − c2 )i + 2c3 = 6,
and −c1 − c2 + 4c3 = 4. This yields c1 = 2 − i, c2 = 2 + i, and c3 = 2. Our
final formula is an = (2 − i)in + (2 + i)(−i)n + 2n+1 . 

6.4 Exponential Generating Functions

Suppose we are dealing with sequences that grow very quickly, such as
an = n! or bn = 2(n − 1)!. One difficulty with the kinds of generating func-
tions discussed in the earlier sections is that these sequences will both give us
the function that is 0 at x = 0 and is undefined for all other values of x. For
sequences that grow faster than exponentials, another kind of generating func-
tion is desirable. Accordingly, we define the exponential generating function
i
(EGF) of a sequence a0 , a1 , . . . to be the formal power series i≥0 ai xi! .
P

Example 6.13: Find the exponential generating function of an = 1.

From the definition, we have fa (x) = n≥0 xi /i! = ex . Similarly, if bn = rn
P
for some constant r, we have fb (x) = erx . 
There is a formula comparable to that of Theorem 6.2.

THEOREM 6.7 (Incomplete Sums—EGFs)

P∞ i P∞ i (k)
Given fa (x) = i=0 ai xi! , we have i=0 ai+k xi! = fa (x).
114 Introduction to Combinatorics

Proof. Given fa (x) = a0 + a1 x/1! + a2 x2 /2! + a3 x3 /3! + . . . , we differentiate

both sides. This gives us f ′ (x) = a1 + a2 x/1! + a3 x2 /2! + . . . . This is the case
k = 1 of the theorem; induction on k completes the proof. 
Roughly speaking, exponential generating functions enumerate some kinds
of problems in which order is important, where the other, “ordinary” gener-
ating functions enumerate problems in which order is not important.
Example 6.14: We wish to eat k donuts, and there are an unlimited
supply of donuts of each of three varieties v1 , v2 , and v3 . Order is important;
so in the case k = 4 we distinguish between, for instance, eating two of variety
v1 first and then two of variety v2 , as opposed to eating the two v2 donuts
first and then the two v1 types. In how many ways can we do this? In how
many ways can we do this if we require at least two donuts of variety v1 ?
The solution to the first problem is easy enough; each possible way of eating
four donuts corresponds to a string of length k consisting of the symbols 1, 2,
or 3 (denoting variety). The answer is just 3k or for k = 4, 81. Alternatively,
consider the product
x2 x3 x2 x3 x2 x3
     
1+x+ + + ... × 1+x+ + +... × 1 + x + + + ... .
2! 3! 2! 3! 2! 3!

The coefficient of xk /k! in this product will be the solution. Suppose we

eat four donuts, including two of variety 1 and one of each other variety. The
number of ways of doing this is 4!/(2!1!1!), as we saw  in Chapter 2. This
2 x4 4!
corresponds to the term x /2! × x/1! × x/1! = . The coefficient
4! 2!1!1!
of x4 /4! is thus 4!/(2!1!1!), corresponding to the number of distinct ways of
ordering the donuts. The factor of x2 /2! comes from the first factor of the
product, corresponding to donuts of variety v1 , and the factors of x/1! come
from the second and third factors of the product, corresponding to varieties
v2 and v3 . This turns out to be just (ex )3 = e3x . The coefficient of xk /k! is
clearly 3k .
Now, how do we restrict ourselves to the case of at least two donuts of type
v1 ? In the first factor of the product above, the terms 1 + x/1! correspond
to 0 or 1 donut of that variety. So we replace the first factor with (x2 /2! +
x3 /3! + . . . .) This gives us the product (ex − 1 − x)(e2x ). Similarly, we may
find the solution for any other restrictions on donut types. 

Example 6.15: How many k-letter words can be made from a, b, e, or g

if we must use at least one of a and e, and no more than 3 of b or g?
We form the product (x + x2 /2! + x3 /3! + . . . )2 (1 + x/1! + x2 /2! + x3 /3!)2 .
The first factor corresponds to a and e, and the second to b and g.
So, if we want the solution for k = 6 (for example) we form the product
(ex − 1)2 (1 + x + x2 /2 + x3 /6)2 = (e2x − 2ex + 1)(x6 /36 + x5 /6 + 7x4 /12 +
 Generating Functions and Recurrence Relations 115

4x3 /3 + 2x2 + 2x + 1). Expanding the first factor out in series as far as x6 ,
we get

x2 + 3x3 + 55x4 /12 + 19x5 /4 + 1321x6 /360 + . . .

This makes it clear that the coefficient of x6 /6! is 2642, the solution to the
given problem. 
Computer algebra systems make these computations much more convenient.
The student with access to Mathematica may enter a problem as follows:

Series[(E^x-1)^2*(1+x+x^2/2+x^3/6)^2, {x, 0, 9}]

Following this with Shift+Enter (to evaluate the expression) produces:

55x4 19x5 1321x6 89x7 7369x8 259x9

x2 + 3x3 + + + + + + + O[x]10 .
12 4 360 40 6720 576
In Maple, essentially the same result comes from this input:

series((exp(x)-1)^2*(1+x+(1/2)*x^2+(1/6)*x^3)^2, x = 0, 10);

Exercises 6A
1. Find the generating function for the sequences:
(i) an = n3 .
(ii) bn = 3n2 − 2n − 1.
(iii) cn = n3 .

(iv) dn = 3n − 23 2n .
2. A bakery sells plain, chocolate chip, and banana-walnut muffins; on a
particular day, the bakery has three of each kind.
(i) How many ways are there to purchase a selection of n muffins?
(ii) How many ways are there if we wish at least one of each kind?
(iii) How many ways are there if we wish at least two banana-walnut
muffins, but have no restrictions on the numbers of the other kinds
of muffins?
(iv) How many ways are there to purchase n muffins if the bakery has
an unlimited supply of each kind? Find a recurrence relation for
this sequence.
116 Introduction to Combinatorics

3. Find the solution for the recurrence with given initial values:
(i) an+1 = an + 12an−1 , a0 = 1, a1 = 4.
(ii) an+1 = 4an − 4an−1 , a0 = −2, a1 = −2.
(iii) an+1 = an + 12an−1 , a0 = 1, a1 = 3.
(iv) an+1 = 4an − 4an−1 , a0 = 1, a1 = 2.
(v) an+1 = an + 12an−1 , a0 = 0, a1 = 1.
(vi) an+1 = 4an − 4an−1 , a0 = 0, a1 = 1.
4. Compute the first 12 Fibonacci numbers.
5. Find a generating function for an , the number of ways to represent the
integer n as an unordered sum of 1s and 3s. Use this generating function
to find a recurrence relation satisfied by an . Can you find a closed form
for this quantity?
6. Find the generating function for an = 2n − n; use this function to find
a recurrence that an satisfies.
7. Find exponential generating functions for the following sequences:
(i) an = n2 .

(ii) bn = 3n − 2n−1 .
(iii) cn = 2.
(iv) dn = 2n + 2.

8. Find a sequence whose exponential generating function is cosh(x) =

(ex + e−x )/2.
9. To celebrate a holiday, a child is permitted (under supervision in a safe
place) to play with sparklers that when ignited emit showers of sparks
in one of four colors. The child is permitted only one sparkler at a time,
and may have no more than k sparklers total.
(i) Find the exponential generating function for the total number of
ways (counting order) for the child to play with the sparklers.
(ii) As above, find the exponential generating function if the child is
to use at least one of each color (so k ≥ 4).
(iii) Assume that the child may use between one and three of each color
of sparkler, and find the exponential generating function.
 Generating Functions and Recurrence Relations 117

Exercises 6B
1. Find the generating function for the sequences:
(i) an = n4 .
(ii) bn = 6n4 − 4n2 .
(iii) cn = n4 .

(iv) dn = 2n + 2(−1)n .
2. A supply cabinet contains four red pens, three blue pens, and three
black pens.
(i) How many ways are there for a staff member to take n pens?
(ii) How many ways if we wish one of each color?
(iii) How many ways are there if we want at least two red pens but have
no other restrictions?
(iv) How many ways are there to take n pens if the supply cabinet has
an unlimited quantity of each color? Find a recurrence relation for
this sequence.
3. Find the solution for the recurrence with given initial values:
(i) an+1 = an + 2an−1 , a0 = 1, a1 = 2.
(ii) an+1 = 2an − an−1 , a0 = 0, a1 = 1.
(iii) an+1 = an + 2an−1 , a0 = 1, a1 = −1.
(iv) an+1 = 2an − an−1 , a0 = 2, a1 = 1.
(v) an+1 = an + 2an−1 , a0 = 5, a1 = 4.
(vi) an+1 = 2an − an−1 , a0 = 1, a1 = 1.
4. Prove the given result about the Fibonacci numbers, for all positive
integers n.
(i) fn is divisible by 3 if and only if n is divisible by 4.
(ii) fn is divisible by 4 if and only if n is divisible by 6.
(iii) fn is divisible by 5 if and only if n is divisible by 5.
(iv) f1 + f2 + · · · + fn = fn+2 − 1.
(v) f1 + f3 + · · · + f2n−1 = f2n .
(vi) fn2 + fn+1
2
= f2n+1 .
5. Find the generating function for an , the number of ways to represent
the integer n as an unordered sum of 1s and 4s. Use this generating
function to find a recurrence relation satisfied by an . Can you find a
closed form for this quantity?
6. Find the generating function for the sequence an = 3n − 2n; use this
function to find a recurrence relation that an satisfies.
118 Introduction to Combinatorics

7. Find exponential generating functions for the following sequences:

(i) an = n2 + n.
(ii) bn = 4n − 3n−1 .
(iii) cn = −1.
(iv) dn = (−1)n + 3.

8. Find a sequence whose exponential generating function is sinh(x) =

(ex − e−x )/2.

9. To celebrate a holiday, a child is permitted (under supervision in a safe

place) to play with sparklers that when ignited emit showers of sparks
in one of three colors. The child is permitted only one sparkler at a
time, and may have no more than k sparklers total.
(i) Find the exponential generating function for the total number of
ways (counting order) for the child to play with the sparklers.
(ii) As above, find the exponential generating function if the child is
to use at least one of each color (so k ≥ 3).
(iii) Assume that the child may use between one and four of each color
of sparkler, and find the exponential generating function.

10. Prove Theorem 6.4.

Problems

Problem 1: Prove Theorem 6.1.

n
Problem 2: Let an = (cP 1 n + c0 )r , where c1 6= 0. Without using Theorem
i
6.5, prove that fa (x) = i≥0 ai x implies that fa (x) is a rational function
with denominator (1 − rx)2 .
Problem 3: Compute the quotient 1/(1−x−x2) by polynomial long division
and confirm that the Fibonacci numbers are the result.
Problem 4: Find a Binet-type formula for the Lucas numbers using the same
method as was used with the Binet formula for the Fibonacci numbers.
Problem 5: Find the number of ways an to cover a 2 × n checkerboard with
n rectangles (2 × 1 or 1 × 2, also known as dominos).
Problem 6: How many binary strings of length n are there that contain no
two consecutive 1s?
 Generating Functions and Recurrence Relations 119

n
P
Problem P 7: Prove the convolution identity: if fa (x) = n≥0 an x and
n
fb (x) = n≥0 bn x , then fa (x)fb (x) is the ordinary generating function for
the sequence cn = a0 bn + a1 bn−1 + · · · + ai bn−i + · · · + an b0 .
2
Problem 8: Suppose that P∞h(n) = an + bn + c for some real numbers a, b, c;
the generating function n=0 h(n)x is a rational function p(x)/(1 − x)3 for
n

some polynomial p(x) of degree 2 or less. Show that the coefficient of x2 in

p(x) is h(−1); show that p(1) = 2a.
Problem 9: Prove that if fa (x) = n≥0 an xn /n! and fb (x) = n≥0 bn xn /n!
P P
then

fa (x)fnb (x)

is the exponential
generating function for the sequence cn =
n n n


0 a 0 b n + 1 a 1 b n−1 + · · · + i a i b n−i + · · · + n a n b 0 .
Problem 10: We saw the derangement numbers Dn = n!(1/0! − 1/1! +
· · · + (−1)i /i! + · · · + (−1)n /n!) in Chapter 5. Use the recurrence Dn =
nDn−1 + (−1)n , n ≥ 2, to find the exponential generating function for Dn .
Chapter 7
Catalan, Bell, and Stirling Numbers

7.1 Introduction
One of the basic purposes of combinatorics is enumeration, and we have seen
many ways to enumerate arrangements of objects. The binomial coefficients
enumerate ways of choosing a subset of distinct objects from a set; factorials
and permutation numbers enumerate ways of choosing an ordered subset; and
so on. In general, we distinguish several classes of problems involving enumer-
ation of arrangements of objects. The objects may be distinct or identical;
the arrangements may be considered distinct when order is changed (order
important ) or not (order unimportant ); and we may distinguish among the
classes in which they are arranged, or not. Also, we may ask whether repeti-
tion is allowed, or not (and if so, whether the repetition may be unlimited);
whether all objects must be arranged; and whether all the classes into which
they are to be arranged must receive at least one object.
For example, consider the number of ways of putting Halloween candies
into two bags. If the candies are all identical, there are clearly fewer ways
of arranging them than if all the candies are distinct. Again, if the bags are
considered identical, there are fewer ways than if we have already assigned
one bag to a particular child. Of course, if the candies are distinct it is not
likely that the child will be terribly concerned about whether the chocolate
bar went into the bag before or after the lollipop; if this makes a difference,
then clearly there are more ways to arrange the candies. If we decide that
repetition is allowed, then placing the chocolate bar into one bag does not
mean that the other bag cannot also get a chocolate bar; we might have
a limited number of chocolate bars available, or not. Are we allowed not
to distribute (say) the lemon drop, or must we use all the available candy?
Is it fair to leave one of the bags empty? These questions also affect the
number of ways in which the distribution might be done. We have already
seen ways to enumerate arrangements for some of these cases; the purpose
of this chapter is to discuss enumeration in some cases that we have not
before. This discussion will include quantities for which (unlike the binomial
coefficients, for instance) no easy formula exists. These quantities are given
the names of the mathematicians with whom they are chiefly associated.

121
122 Introduction to Combinatorics

7.2 Catalan Numbers

In the late eighteenth century, the extraordinarily prolific Swiss mathemati-
cian Leonhard Euler (about whom more is written in Appendix 3) investigated
the question: How many ways are there of dividing up a regular polygon into
triangles using non-intersecting diagonals? He produced, after great effort,
a sequence that today is denoted Cn , representing the number of ways of so
dividing an (n + 2)-gon. By convention, we will define C0 = C1 = 1. This
sequence turned out to arise in a remarkable number of different places, and
(for reasons we will see later) was named eventually after the Belgian mathe-
matician Eugène Catalan. We provide a table below of the Catalan numbers.

n 0 1 2 3 4 5 6 7
Cn 1 1 2 5 14 42 132 429

To investigate the Catalan numbers, we begin by considering an (n+2)-gon

and we choose one side of this polygon. We can make this chosen side the
side of a triangle in some triangulation of the polygon in exactly n ways; just
choose one of the n corners that do not touch this side and draw lines from
this corner to the two corners at either end of the chosen side. Each such
triangle is the start of a triangulation, so our expression for Cn will involve a
sum over n terms. Now suppose that the triangle we have drawn that includes
our chosen edge leaves a (k + 2)-gon on one side and an (n − k + 1)-gon on
the other side. Notice that this is valid for 0 ≤ k ≤ n − 1; a “2-gon” is
one side of the polygon, and this means that our chosen triangle shares two
sides with the original polygon. Now to finish the triangulation, there are
Ck × Cn−k−1 ways, because we can triangulate the (k + 2)-gon in Ck ways,
and the (n − k + 1)-gon
P in Cn−k−1 ways. Summing, we arrive at the Catalan
recurrence: Cn = 0≤k≤n−1 Ck × Cn−k−1 . We illustrate with Figure 7.1,
using a hexagon (so n + 2 = 6). It should be clear that C0 , the number
of ways of triangulating a “two-cycle,” is 1 by convention just to make the
recurrence work (some authors simply say that C0 is undefined).

k=0 k=1 k=2 k=3

n-k+1=5 n-k+1=4 n-k+1=3 n-k+1=2

FIGURE 7.1: Triangulating the hexagon.
 Catalan, Bell, and Stirling Numbers 123

It was not until the 1830s that Catalan investigated the question of how
many ways one can correctly arrange n pairs of parentheses. Here “cor-
rectly” means that the expression makes sense, i.e., that each right paren-
thesis is matched by a preceding unique corresponding left parenthesis and
vice versa. Thus, for n = 0 and n = 1 we get 1; for n = 2 we get 2,
enumerating the set {(()), ()()}; and for n = 3 we find five ways, namely
{((())), (())(), ()(()), (()()), ()()()}. Because the sequence begins 1, 1, 2, 5, . . .
we naturally wonder whether it is the same sequence as Euler’s.
If we let Cn denote the number of ways to correctly arrange n pairs of
parentheses, we can find a recurrence that relates each value to the preceding
values. Take any correct arrangement of n pairs of parentheses, n ≥ 1, and
look at the leftmost parenthesis. It is matched by a right parenthesis elsewhere
in the expression, either immediately following or at the end or somewhere in
the middle of the expression. Because the expression is a correct arrangement,
there must be a correct arrangement of 0 or more pairs of parentheses between
this first parenthesis and the corresponding right parenthesis, and a correct
arrangement of 0 or more pairs following the corresponding right parenthesis.
Suppose there are k pairs of parentheses between it and its “mate.” Then
there are n − k − 1 pairs of parentheses following the mate. How many ways
can this be accomplished for a given value ofP k? Clearly, there are Ck ×Cn−k−1
ways. Summing over k, we see that Cn = 0≤k≤n−1 Ck × Cn−k−1 . Because
this sequence has the same initial values and satisfies the same recurrence as
the sequence of polygon triangulations, they are identical.

THEOREM 7.1 (Values of Catalan Numbers)

 
2n
Cn = /(n + 1).
n

Proof. In how many ways may we arrange n pairs of parentheses in any

order whatsoever? A string of n pairs of parentheses has length 2n; we select

n of these positions to be the left parentheses. Thus there are exactly 2n n
ways to do so. Write Cn = 2n


n − In , where In denotes the number of
incorrect arrangements of parentheses. We shall count In and the theorem
will follow as a result. Consider an incorrect sequence of parentheses, denoted
a1 a2 . . . a2n , where each ai is either a left or a right parenthesis, and there are n
of each. Because this sequence is incorrect, there must be a smallest subscript
k such that ak is a right parenthesis that is not matched by a preceding left
parenthesis. Because k is the least such subscript, ak is preceded by matching
pairs of left and right parentheses, so k must be odd.
Now, for the trick: we reverse each of the first k parentheses and leave
the others unchanged. This gives us a string of
n + 1 left parentheses and
2n
n − 1 right parentheses. Clearly, there are n+1 such strings. This process
is reversible. Given such a string, we can find the first point k at which the
124 Introduction to Combinatorics

number of left parentheses exceeds the number of right parentheses, and by

reversing the parentheses up to that point, recover precisely the same incorrect
expression we started with. This process provides a one-to-one and onto
correspondence between the incorrect strings of n left and n right parentheses
and the strings of n
+ 1 left and n − 1 right parentheses.
This 2n
establishes the
2n
. We now calculate Cn = 2n 2n



equation In = n+1 n − n+1 = n /(n + 1). 
The Catalan numbers occur in numerous places throughout all of mathe-
matics, and we will provide examples here, in the exercises, and in the prob-
lems.
Example 7.1: We wish to walk to a point that is n blocks north and n
blocks east of our current position, but we never want to be more blocks north
of our current position than we are east. In how many ways can we do such
a walk if we stick to the streets?
We model this problem by starting at the origin (0, 0) and take a total of
2n steps, each of which consists of adding 1 to either the first or the second
coordinate (but not both). The requirement that we may not go further
north than east means that the allowable points (x, y) at each step must
satisfy x ≤ y. We will remember from Chapter 2 that the total number of
paths without this restriction is 2n


n . We might try to count paths for which
y becomes greater than x and subtract, but there is an easier way.
Observe that a walk from (0, 0) to (n, n) may be described by a string of
directions, where each direction is N for north or E for east. The condition
equates to saying that at any point we have no more N s than Es. There-
fore, take any walk, and change N to a right parenthesis, “)” and E to a left
parenthesis, “(”. We see that we have an expression involving nested paren-
theses, and this must be a correct arrangement because as we move from left
to right we never will have encountered more right parentheses symbols than
left parentheses. It is not hard to see that this will provide a bijection be-
tween walks of the sort we are asked for and correct arrangements of n pairs
of parentheses. It follows that there are exactly Cn walks from (0, 0) to (n, n)
that meet the condition that we are never further north than east. 

Example 7.2: A rooted binary tree is a kind of graph (as discussed in

Section 1.5); a collection of points (also called nodes or vertices) where one
distinguished vertex, called the root, is placed at the top of the diagram. Each
vertex has either zero or two edges descending down to any further vertices.
A vertex is said to be internal if it has two edges descending. How many
rooted binary trees with n internal vertices are there?
Figure 7.2 shows the cases for n = 0, 1, and 2.
If we denote by an the number of rooted binary trees with n internal points,
we see from the diagram that a0 = a1 = 1 and a2 = 2. To prove generally
the equality an = Cn , we derive a recurrence. Observe that for n ≥ 1,
removal of the root leaves us with two rooted binary trees. Since the root
 Catalan, Bell, and Stirling Numbers 125

n=0 n=1 n=2

FIGURE 7.2: Rooted binary trees with n internal points.

is an internal point of the original rooted binary tree, the two trees we get
after removing it have k and n − k − 1 internal points, for some value k with
0 ≤ k ≤ n − 1. Now,
Pn−1as with the problem of triangulations of the polygon, we
may write an = k=0 ak an−k−1 . The sequence an has the same initial terms
and satisfies the same recurrence as the Catalan numbers so it is the same
sequence. 
We spent all of Chapter 6 discussing generating functions. The Catalan
numbers will provide us with a challenge.

THEOREM 7.2 (The Catalan Number Generating Function)

∞ p
X
n 1 − 1 − 4(x)(1)
C(x) = Cn x =
n=0
2x

Proof. We will use the convolution identity from Problem 6.8. Let C(x) =
xn = 1 + x + 2x2 + 5x3 + . . . . This identity gives us:
P
i≥0 Cn
X
C 2 (x) = (C0 Ci + C1 Ci−1 + · · · + Ci C0 ) xi .
i≥0

The Catalan recurrence with which we began this section tells us that the
parenthesized
P expression on the right side of this equation is just Ci+1 , so
C 2 (x) = i≥0 Ci+1 xi or C 2 (x) = 1 + 2x + 5x2 + 14x3 + . . . . (The reader
should confirm this by squaring the polynomial 1 + x + 2x2 + 5x3 + 14x4 ,
neglecting terms past x4 .) Now we multiply both sides by x and add 1:
xC 2 (x) + 1 = 1 + x + 2x2 + 5x3 + 14x4 + · · · = C(x). Now what are we to make
of the equation xC 2 (x) − C(x) + 1 = 0? This is just a quadratic equation in
C, for any particular value of x. We may apply the quadratic formula with
a = x, b = −1, and c = 1. We arrive at two possibilities:
√
1 ± 1 − 4x
C(x) = .
2x
To determine whether to use the plus or the minus, look at C(0); we get
C(0) = 1 + 1 · 0 + 2 · 02 + · · · = 1. Now, setting x = 0 in the formula above
126 Introduction to Combinatorics

means division by 0 regardless of whether we use the plus or the minus. A

closer look is in order.
If we choose the plus sign, then as x approaches 0 we arrive at the expression
2/0, which clearly is undefined. If we choose the minus, we arrive at the
indeterminate form 0/0, which is susceptible to L’Hôpital’s rule. This yields
limx→0 C(x) = 1, as required. If there is any doubt that we have found
the correct generating function, we may use Newton’s binomial theorem or
Maclaurin series to expand the radical and confirm that Catalan numbers are
the result. 
We have only touched very lightly upon the subject of Catalan numbers,
a subject about which a great deal is known. For more information, the
interested reader is encouraged to see ([40], ch. 20), reprinted in [43] for
a very readable exposition; much more technical detail is available in [66].
Another source of good information on the Catalan numbers is [108] and the
online addendum [109].

7.3 Stirling Numbers of the Second Kind

Consider the number of ways of distributing n distinct candies to k distinct
children in such a way that each candy is distributed and no child fails to get
a candy (n ≥ k). Our first impulse might be to attach to each of the candies
a number from 1 to k indicating which child gets that particular candy; by
putting the candies in line in some order, we have an n-digit number in base
k, and can easily enumerate the possibilities. However, this does not work,
for we have not assured ourselves that each child gets at least one candy. The
k n possible n-digit numbers in base k include many possibilities that omit one
or more base-k digits.
Likewise, we cannot say (as we might if the candies were identical) that
we will simply distribute one candy to each child in the beginning, and then
arrange any remaining candies afterwards; it makes a difference which candy
is distributed to which child, and a distribution in which (for instance) child
1 gets candies A and B might be counted as two distributions, depending on
whether candy A was handed out first (we are assuming that order is not
important here). Instead we approach this problem from a different point of
view.
We still associate with each candy a number denoting which child receives
it; regard this association as a function from an n-set (the candies) to a k-set
(the children) and we require that it be onto. This means we have rephrased
our original problem as follows. How many surjections are there from an n-set
onto a k-set? (Recall surjection just means onto function.)
There is an ancient mathematical tradition that says that, in order to solve
 Catalan, Bell, and Stirling Numbers 127

a problem, you merely invent a symbol, define the symbol to be the solution
to the problem, and then announce that you’ve solved the problem, giving the
symbol as the solution. In this tradition, we define the Stirling number of the
second kind, nk , to be the number of ways to divide n distinct objects into k

identical classes. Because the children are distinct, we must multiply by k! to

take into account the number of possible ways of ordering the children. Thus
the number of surjections from an n-set onto a k-set is just k! nk . We have


now solved the problem; ifwe are given nine candies and six children, we may
distribute the candies in 6! 96 ways. Other notations for the Stirling numbers
(k)
of the second kind include Sn and S(n, k); our notation was popularized by
Knuth ([48], [64]),who attributes it to Karamata ([55]). Knuth also suggests
that the notation nk be read as “n subset k.”
The irritated demands of those unmoved by the tradition discussed in the
previous  paragraph suggest we look more deeply into the properties of the
numbers nk in order to produce numerical values. To this end, we establish

a few conventions. For instance, we will define 00 = 1 for much the same


reason that we define 0! = 1; it is convenient for formulas that we will develop

soon. If n < k or if n and k are not both
 non-negative integers, then we will
say nk = 0. It is easy to see that n1 = 1, since the only way to put n
objects into one class is for the class to have all the objects. Similarly, if n = k
we clearly have n! waysto arrange the classes so that each gets exactly one
distinct object, so that nn = 1.
We now seek a recurrence relation that will let us calculate nk from


smaller Stirling numbers of the second kind.

THEOREM 7.3 (A Stirling Number Recurrence)

   
nno n−1 n−1
= +k
k k−1 k

Proof. Suppose we assign objects {o1 , o2 , . . . on } to k classes. When we come

to the
n lasto object on , we may place it in a class by itself; if we do this, there
are n−1
k−1 ways in which we might have arranged the n − 1 preceding objects
into k − 1 classes. Otherwise, we already have k non-empty classes, and we
must place the new
 object
into one of the pre-existing classes; there are k ways
to do this, and n−1 k ways in which the n − 1 earlier objects were placed.
The result follows. 
4
Example 7.3: Calculate the value 2 by applying the recurrence above.

             
4 3 3 2 2 2 2
= +2 = +1 +2 +2
2 1 2 0 1 1 2
128 Introduction to Combinatorics

We use S(2, 0) = 0 and S(2, 1) = S(2, 2) = 1 to simplify further.

       
4 2 2 2
=0+ +2 + 4(1) = 3 +4
2 1 1 1
So, 42 = 7.


It is somewhat tedious to find the Stirling numbers of the second kind by
use of the recurrence. One approach is to create a table similar to Pascal’s
triangle and refer to the table, shown below.
To extend the table or verify any en- n\k 0 1 2 3 4 5 6
try in it, choose an element. Multi- 0 1 0 0 0 0 0 0
ply that element by the number at 1 0 1 0 0 0 0 0
the top of its column, and add to it 2 0 1 1 0 0 0 0
the element on its left (or 0 if there 3 0 1 3 1 0 0 0
is no element to the left). The re- 4 0 1 7 6 1 0 0
sult should be the number below the 5 0 1 15 25 10 1 0
chosen element. 6 0 1 31 90 65 15 1

Example 7.4: How many ways are there to factor the number 2 · 3 · 5 · 11 ·
23 = 7590 into a product of two factors each greater than 1? How many for
three such factors?
It is clear that each of these factors needs to use some of the distinct primes
in the unique factorization of 7590. Accordingly, we take two boxes and put
prime factors into them. As there are five prime factors and two boxes, we
find 52 = 15. Note that we count the factorization 2 × 3795 the same as

 5 × 2, for instance, because the boxes are identical. In the same way,
3795
3 = 25, so we have 25 ways of writing 7590 as a product of three factors,
each bigger than 1. 

Example 7.5: Angered by all the candies being given to children at the
beginning of this section, a group of indignant parents take five distinct pam-
phlets on the dangers of childhood obesity to give to two textbook authors. In
how many ways may this be done if the two authors are considered distinct?
In how many ways if the authors are considered identical? In either case, no
author escapes being given at least one pamphlet, and there is only one copy
of each of the five pamphlets.
Here n =  5 refers to the pamphlets, and k = 2 refers to the authors. From
the table, 52 = 15, so if the authors are distinct the answer is 2! × 15 or 30,
and if the authors are identical we get 15. 
.
Example 7.6: Properly chastened, the authors meet with three indignant
parents and offer them a collection of six distinct healthful snacks. (The
authors never offer healthy snacks, as by definition healthy implies that the
 Catalan, Bell, and Stirling Numbers 129

snacks are still alive and potentially able to fight back. Ethical, dietary, and
legal considerations forbid.) Assuming the three parents are distinct, in how
many ways can this be done if each parent is to receive at least one snack? In
how many ways if a parent may refuse a snack (all snacks are distributed)? In
how many ways if we do not require that all snacks be distributed, and any
parent may refuse a snack?

The first problem is just a matter of looking at the table; n = 6, k = 3, so

we get 3! × 63 = 6 × 90 = 540.

The second problem must be done in pieces. If all three parents accept a
snack, 540 is still the answer; if one parent refuses, there are three ways to
choose a parent to refuse and 2! × 62 = 62 ways to distribute the snacks to


the other two parents, so we get another 186 possibilities. Then if two parents
refuse, we get another three possibilities

 (three choices of the parent to accept
all six snacks, or if you like 31 × 61 = 3). Adding these, we find a total of
729 possible distributions of snacks.
The third problem is the easiest or most difficult, and may be solved by
considering the trash can as a fourth potential  snack recipient. If all four
possible recipients get a snack, we have 4! × 64 = 1560. If one recipient

(parent or trash can) gets no snack, we count 41 × 3! × 63 = 2160. Two



snackless recipients yields 42 × 2! × 62 = 372. Finally, if all the snacks go to



one recipient, there are four possibilities. Adding these numbers yields 4096
ways. Realizing that this answer is just 46 , we see that we could have saved
ourselves much trouble by saying: Associate with each snack a number from
{0, 1, 2, 3} according to which recipient receives it. Placing these numbers in
order gives us a six-digit base-4 number, of which there are 46 . 

This example leads us to seek a generalization. A classic theorem concern-

ing Stirling numbers of the second kind gives a formula that has been used as
the definition. We use the notation [r]k = r!/(r − k)! = r(r − 1) . . . (r − k + 1),
the falling factorial notation.

THEOREM 7.4 (A Stirling Number Formula)

r n o
X n
rn = [r]k
k
k=0

Proof. Suppose we have r recipients of n objects, and, although all the

objects must be distributed, some recipients may get no object. If all r re-
cipients get an object, we have r! × nr , which is the k = r summand of


the formula. Forn the ocase in which one recipient gets no object, we count
r n


1 × (r − 1)! × r−1 , which is equivalent to the k = r − 1 summand. In
n o n o
general, if i recipients get no objects, we have ri ×(r−i)!× r−i
n
= r! n


i! r−i
130 Introduction to Combinatorics

which is the r − ith summand. On the other hand, the quantity we are count-
ing is the number of n-digit base-r integers, of which there are rn . 

Example 7.7: Write the polynomial x3 as a linear combination of the

polynomials 1, x, x(x − 1), and x(x − 1)(x − 2).
In the preceding theorem, substitute 3 for n and x for r; this gives us
3  
3
X 3 x!
x =
k (x − k)!
k=0

and we simplify this to get x3 = 0 · 1 + 1 · x + 3 · x(x − 1) + 1 · x(x − 1)(x − 2),

or x3 = x + 3x(x − 1) + x(x − 1)(x − 2). Even though the recurrence requires
that x be an integer, the identity holds for any value of x. 
A number of identities exist for particular classes of Stirling numbers of the
second kind; for instance, in [48] it is shown that
nno
= 2n−1 − 1.
2
This, as we have seen with other combinatorial identities, may be proved
in a variety of ways. The simplest but least enlightening is the use of math-
ematical induction using the recurrence of Theorem 7.3. We adapt the proof
from [48], which is a combinatorial proof giving more insight into the formula.
Consider a set of n objects o1 , o2 , . . . on and two identical empty boxes. We
toss object on into one of the boxes; they are no longer identical, as one is
not empty. Now, for the remaining n − 1 objects, we have a choice; toss the
object into the box with on , or into the other box. There are 2n−1 ways to
do this. One of the resulting distributions consists of all n objects in one box,
and the other box is empty. Because we are counting only distributions in
which neither box is empty, we subtract 1 to arriveat 2n−1 − 1 = n2 .


This formula is one of a number of identities for nk for specific k; see the
problems for more examples. Each may be proved by induction  or combina-
torial techniques. There is another recurrence relation for nk that may be
useful in computing the entries in a table such as the one given above.

THEOREM 7.5 (Another Stirling Number Recurrence)

n n o n−1
X n − 1 i 
= for n > 0
k i=0
i k−1

 first of all that many of these terms will be 0 (when i < k − 1)

Proof. Notice
and recall 00 = 1. Assume we have chosen a way of arranging all n objects
into k boxes, and look at the box that contains the nth object on . That box
will contain n − i total objects, where 0 ≤ i ≤ n − 1. If we eliminate this box,
we have a way of arranging i objects into k − 1 boxes, and there are clearly
 Catalan, Bell, and Stirling Numbers 131
n o
i
ways of doing this. There are also n−1


k−1 i ways of selecting the objects
that
 n do not go into the box with on . Summing over all values of i gives all
k ways of arranging the objects. 

Example 7.8: Prove the identity n2 = 2n−1 − 1 using Theorem 7.5.



Pn−1
Substituting 2 for k, we get n2 = i=0 n−1

 i i
 n Pn−1 n−1
i 1 ; since 1 = 1 for
n−1
i > 0, we have 2 = i=1 i = 2 − 1 by the binomial theorem. 

7.4 Bell Numbers

How many ways can we assign n students to work in groups? We do not
specify the number of students in any group (except that a group must be
non-empty), and do not specify the number of groups (except that there must
be at least one and no more than n groups). We easily see this to be the sum
of Stirling numbers of the second kind. We define the Bell number Bn to be
the number of ways to partition a set of n distinct objects. We may also think
of it as the number of ways to place a set of n elements intoP identical
 boxes,
where the number of boxes is not specified, and write Bn = ni=0 ni .
Example 7.9: We have three distinct snacks and wish to put them into
bags. How many ways can this be done?
There is exactly one way to do this with three bags (each snack into its own
bag); three ways to do this with two bags (depending on which snack goes
into a bag by itself); and one way with one bag. Thus B3 = 5. 

Example 7.10: How many ways are there of writing 2 × 3 × 5 × 7 = 210

as a product of distinct integers?
Because 210 is (as shown) a product of distinct primes, we may think of a
factor as a subset of the set {2, 3, 5, 7}. Thus the number of ways of writing
210 as a product of integers is the number of ways of partitioning that set,
or B4 = 15. Here, the product 1 × 210 represents the partition {2 3 5 7}; the
product 6 × 35 represents the partition {2 3}, {5 7}, and so forth. Notice that
writing 6 × 35 is considered the same as 35 × 6. 
Summing a row of the Stirling triangle will work for smaller Bell numbers,
but it would be worthwhile to find a better way to compute, say, B25 that
would not involve so much work. Such a formula is given in the following
theorem.
132 Introduction to Combinatorics

THEOREM 7.6 (Bell Number Recurrence)

n  
X n
Bn+1 = Bi
i=0
i

The proof is left to the reader (see Problem 6.7). Note that B0 is 1.
Example 7.11: Find B6 .
The recurrence gives B6 = B0 + 5B1 + 10B2 + 10B3 + 5B4 + B5 . We may
compute these by summing rows in the triangle, or by recursively applying
Theorem 7.6 to get B1 = 1, B2 = 2, B3 = 5, B4 = 15, and B5 = 52. This
gives us B6 = 1 + 5 + 20 + 50 + 75 + 52 = 203. This corresponds to the sum
of the sixth row of the triangle in Section 7.3. 
Another approach is called the Bell triangle. Just as binomial coefficients
may be computed easily by means of Pascal’s triangle, the Bell numbers may
be computed by a triangular array. We create the array by means of the
following rules:
1. The first row consists of the number 1 by itself.
2. Each row is one entry longer than the row before it.
3. The first item in each row is equal to the last item in the preceding row.
4. Each entry after the first is the sum of the entry to its left and the entry
directly above that entry.
Applying these rules, we get the following table.
1
1 2
2 3 5
5 7 10 15
15 20 27 37 52
52 67 87 114 151 203
The rightmost entries give us a list of Bell numbers; B0 = 1, B1 = 1,
B2 = 2, B3 = 5, B4 = 15, B5 = 52, and B6 = 203. This is only a small
selection of the properties of the Bell numbers; we refer the interested reader
to ([42]), also found in the CD-ROM collection ([43]) for a very readable
discussion. More technical details and identities may be found in [48], [63],
and [139].

7.5 Stirling Numbers of the First Kind

Recall that Theorem 7.4 tells us that the Stirling numbers of the second kind
may be used to write monomials xn as linear combinations of falling factorials,
 Catalan, Bell, and Stirling Numbers 133

so we may write x3 = 30 + 31 x+ 32 x(x−1)+ 33 x(x−1)(x−2). Indeed,

   

the original definition of nk was as the coefficient in that equation. It is easier

to write the falling factorial in terms of sums of monomials; so we can figure
that [n]3 = n(n−1)(n−2) = n3 −3n2 +2n. These coefficients are referred to as
the Stirling numbers of the first kind, or sometimes as signed Stirling numbers
Pk h i
of the first kind, and we denote them nk . Thus we have [x]k = i=0 ki xi .
 

We refer to the numbers  nk  as unsigned Stirling numbers of the first kind.

 

The unsigned Stirling numbers of the first kind may also be viewed as the
coefficients of the polynomial [x]k = x(x + 1) . . . (x + k − 1) (the “rising
factorial”). For example, where [x]3 = x(x − 1)(x − 2) = x3 − 3x2 + 2x,
we have [x]3 = x(x + 1)(x + 2) = x3 + 3x2 + 2x.
These numbers are more easily calculated than Stirling numbers of the
second kind; we simply multiply out the polynomial [x]k = x(x − 2) . . . (x −
k + 1). It is also easy to find a recurrence for signed or unsigned Stirling
numbers of the first kind.

THEOREM 7.7 (Another Stirling Number Recurrence)

hmi m − 1 
m−1

= − (m − 1)
n n−1 n

Proof. We begin by using the definition and applying a bit of algebra.

m  
X m j
x = x(x − 1) . . . (x − m + 1) = [x]m = (x − m + 1)[x]m−1
j=0
j

Replacing the falling factorial on the right with the equivalent sum gives us:
m   m−1
X m − 1
X m j
x = (x − m + 1) xj
j=0
j j=0
j

and with the distributive law we obtain

m   m−1
X m − 1 m−1
X m − 1
X m j
x =x xj − (m − 1) xj .
j=0
j j=0
j j=0
j
n
m
What is the coefficient
h i of x on eachside of the equation above? We get n
on the left and m−1
n−1 −(m−1)
m−1
n on the right. Since the two polynomials
are equal, the coefficients are equal for each n, establishing the theorem. 
A similar result holds for the unsigned Stirling numbers of the first kind.

THEOREM 7.8 (Unsigned Stirling Number Recurrence)

h m i  m − 1   
 m−1 
= + (m − 1) 
     
n n−1  n
 
134 Introduction to Combinatorics

The proof is essentially the same as that of Theorem 7.7.

0
By convention,
n we take [x] 0 = 1, so 0 = 1. It is not hard to see that for
n > 0 we have 0 = 0 because [x]m is a multiple of x and so has no x0 term.
In the
 same way, since the leading coefficient of the polynomial [x]m is 1, we
see m n
 
m = 1. Also, we set m = 0 whenever m > n. It is not quite as easy
to find other coefficients; some examples are given in the problems.
The unsigned Stirling numbers of the first kind have a combinatorial inter-
pretation like that of the Stirling numbers of the second kind. Suppose we
have m people and wish to seat them at n identical circular tables, leaving no
table empty. In how many ways can we do this, assuming that the arrange-
ment around the table makes a difference? In particular, if there are one or
two people at a table, there is only one way to seat them; with three people at
a table, there are two ways; with four people, there are six ways. In general, if
there are k people at a circular table, then there are (k − 1)! ways to arrange
them; we can line them up in any of k! ways, and then we divide by k because
we do not distinguish between two seatings that differ only by which of the k
chairs the first person sat in. But this does not help us compute the number
of ways of seating the m people, because different seatings may use different
values of k. Let us denote the number of ways to seat m people at the n
tables as s(m, n), where clearly n ≤ m. It should be clear, for example, that
s(m, m) = 1; each table has one person seated at it, and as the tables are
identical this can be done in just one way. Also, s(m, 0) = 0 for m > 0 and
s(0, 0) = 1 by convention.

THEOREM 7.9 (A Combinatorial Interpretation)

h m i
s(m, n) = 
 
n


Proof. We first establish the recurrence relation of Theorem 7.8. Suppose

we have m people labeled 1 through m. We may seat person m at a table
alone and seat the other people in s(m − 1, n − 1) ways, or we may seat the
other m − 1 people at n tables, and squeeze person m into a table in m − 1
ways, according to which of the other m − 1 people is seated to person m’s
right. There are (m − 1)s(m − 1, n) to do this. Adding these two gives us
s(m,n)
 =
 s(m−1, n−1)+(m−1)s(m−1, n). Now, since the numbers s(m, n)
and  mn satisfy

  the same recurrence relation and (as we have already seen)
s(m, 0) =  m0
, it follows that s(m, n) =  m  for any m and n.
n 

The next result requires a definition from the theory of permutation groups.
A permutation of a set of n elements is a bijection from that set to itself. A
cycle of a permutation is a subset of the elements with the property that
each member of the subset is mapped to any other member of the subset by
repeated applications of the permutation. For example, consider the permu-
tation σ on the set {1, 2, 3, 4, 5, 6} given by the table below.
 Catalan, Bell, and Stirling Numbers 135

n 1 2 3 4 5 6
σ(n) 3 4 5 2 3 6
By applying σ repeatedly to the element 1, we see that σ(1) = 3, σ 2 (1) =
σ(3) = 5, and σ 3 (1) = σ(5) = 1; so the set {1, 3, 5} is a cycle. The other
cycles are {2, 4} and {6}. Groups of permutations are discussed in Chapter 8.

COROLLARY  n  7.9.1
There are  m  permutations on n letters with m cycles.
Proof. We establish a one-to-one correspondence between ways to seat n
people at m tables and the specified permutations. Given a seating, let σ be
the permutation that maps i to j if and only if person i is sitting immediately
to the left of person j. Conversely, given a permutation with m cycles, take
the people whose labels correspond to the elements of a cycle, and seat them
at a table in the order they appear in the cycle; repeat this procedure until
all m cycles have been turned into seats at all m tables. This process gives
us a bijection, so the two sets have the same number of elements. 
As with Stirling numbers of the second kind, we may present them in the
form of a table and refer to it, shown below. To extend the table or verify
any entry in it, choose an element. Multiply that element by the number n
at the left of its row, and subtract it from the element on its left (or 0 if there
is no element to the left). The result should be the number below the chosen
element.

n\k 0 1 2 3 4 5 6
0 1 0 0 0 0 0 0
1 0 1 0 0 0 0 0
2 0 −1 1 0 0 0 0
3 0 2 −3 1 0 0 0
4 0 −6 11 −6 1 0 0
5 0 24 −50 35 −10 1 0
6 0 −120 274 −225 85 −15 1

7.6 Computer Algebra and Other Electronic Systems

Although it is important to be able to manipulate the identities and re-
currences involved in the Catalan, Bell, and Stirling numbers, we can quickly
find the values of these quantities with a variety of software packages. Each
of the major packages has some ability to calculate these quantities rapidly.
Even a spreadsheet can produce the Catalan numbers quickly and effi-
ciently. As an example, consider Microsoft Excel. We use the function COM-
136 Introduction to Combinatorics

BIN, which produces binomial coefficients. To generate the Catalan numbers,

open Excel, and place the focus in cell A1. Enter 1 and go to A2. In A2, place
the cursor in the formula bar above the cells and type =A1+1 and press Enter.
At this point, cell A2 should read 2. We will arrange for the cells in column
A to display the number of the row as follows: Select all of the cells A1 down
to A25 (say), by placing the cursor in A1, holding down the Shift key, and
pressing the down arrow until all the cells you wish are highlighted. Then use
the Fill command and Fill Down. (The location of the Fill command may
vary according to the version you are using.)
To compute the Catalan numbers, we place the cursor in cell B1 and type
= COMBIN(2*A1, A1)/(A1+1). This command should insert the formula for
the first Catalan number C1 in that cell. Now select cells B1 to B25 and again
select Fill Down. In less than half a minute you have calculated the values of
all the Catalan numbers up to C25 , and will probably find that the computer
must use exponential notation for those past about C22 .
Both Mathematica and Maple also have the capability of producing Catalan
numbers. For Mathematica, we first type <<DiscreteMath‘Combinatorica‘
to load the package of combinatorial functions. The backwards apostrophe
before and after the name Combinatorica is found to the left of the 1 in the
top row of most keyboards. Now the name in Mathematica of the function
we want is CatalanNumber, with capital letters as shown. This instruction:
Table[CatalanNumber[i], {i, 1, 25}]
produces a list of the first 25 Catalan numbers more quickly than the similar
exercise with the Excel spreadsheet. In addition, we get the larger numbers
such as C25 = 4861946401452 without exponential notation and truncation
as with Excel.
The Combinatorica package is not required for the Stirling numbers of the
first and second kinds, which are available by default. To generate the triangle
of Stirling numbers of the second kind, for instance, we may enter this:
Table[StirlingS2[i,j], {i,0,6},{j,0,6}]//TableForm
The first part of the command, the Table command, produces a two-
dimensional array indexed by the variables i and j. The StirlingS2 function
produces a Stirling number of the second kind (just as the StirlingS1 func-
tion produces those of the first kind). The two sections in braces indicate the
ranges of the indices, and the TableForm command arranges the results in a
square grid.
There is no built-in function to calculate the Bell numbers in Mathematica,
but that causes no difficulty.
Sum[StirlingS2[6, i], {i, 0, 6}]
If we wish to generate Bell numbers frequently, it is better to define a
function to do this for us. The syntax is the same as for other Mathematica
functions.
bell[n ]:= Sum[StirlingS2[n, i], {i, 0, n}]
After entering that line, for example, bell[10] yields 115975.
 Catalan, Bell, and Stirling Numbers 137

The Maple package, like Mathematica, requires a command to load the

combinatorial functions. In Maple, the command with(combinat); loads
this package and enables one to calculate Catalan and the various Stirling
numbers. The Catalan numbers are not built-in, but are easily accessed by
means of the binomial coefficients. These are found using the numbcomb(n,m)
n


function that calculates m . Entering the definition requires the use of the
right-arrow key.
seq(numbcomb(i)/i+1→, i=0..25);
Entering the character / tells Maple to begin the denominator of a fraction;
the right arrow key means that subsequent entries will not appear in the
denominator. It is important to distinguish between this and the right-arrow
symbol in Maple that is generated by the two keys - (hyphen) and > (greater-
than) and which indicates function definition.
This command produces the first 25 Catalan numbers. As with Mathemat-
ica, no exponential notation or truncation occurs.
Stirling numbers are produced in Maple using the commands stirling1
and stirling2. Thus, stirling1(5, 3); produces 35. Again, Maple’s seq
command will produce a sequence, but we can get a two-dimensional array
with Matrix(). Thus the table of Stirling numbers of the first kind could be
produced by this command:
Matrix[7, (m,n)-> stirling1(m − 1, n − 1));
In this line, the combination of the hyphen and the greater-than sign pro-
duce an arrow
h that
i indicates a function taking the pair (m, n) as input and
producing m−1
n−1 as output. The numeral 7 indicates a 7 × 7 matrix.
Maple also has the function bell that produces the Bell numbers; after
loading the combinatorial functions as above, we may type bell(10); to get
115975.

Exercises 7A
1. A ticket-taker sells tickets to a tacky tatting show. Each tatting ticket
is ten dollars. If there are 2n customers lined up to take tickets to the
tatting show, and n of them have $20 bills and the other half have $10
bills, in how many ways can they line up so that the ticket-taker will
always have correct change (she begins with no change)? It may help
to represent the line as “(” for a customer with a $20 bill and “)” for a
customer with a $10 bill.

2. Draw each of the distinct ways of triangulating a regular hexagon.

3. Place 2n dots in a row along the x-axis, and imagine that we connect
pairs of dots with arcs that curve up into the first quadrant. We are to
138 Introduction to Combinatorics

do this so that each dot is connected with exactly one other dot, and
with no arcs crossing one another; show that Cn is the number of ways
this can be done.

4. Use the Catalan recurrence to verify the table value for C5 .

5. Use Theorem 7.1 to calculate C9 .

n o
i
6. Make a 4 × 4 matrix whose (i, j)-entry is j . Calculate the inverse
of this matrix (by hand, with a calculator, or using a computer algebra
system such as Mathematica or Maple). Compare the entries in this
inverse with the Stirling numbers of the first kind.

7. Use the recurrence to calculate k7 for k = 2, 3, 4.



8. How many ways are there to factor the number 2 × 3 × 7 × 19 = 798

into a product of two factors each greater than 1?

9. How many ways are there to distribute six distinct chocolate rabbits
into three identical Easter baskets, where each basket is to receive at
least one chocolate rabbit? In how many ways can this be done if the
baskets are distinct?

10. Repeat the previous exercise if we do not have to distribute all six rab-
bits.

11. Five dishonest lobbyists set out to bribe three corrupt politicians. If each
politician receives at least one bribe, and each lobbyist offers a bribe to
exactly one politician, in how many ways can we assign lobbyists to
politicians, assuming the politicians are distinct?

12. Five dishonest lobbyists are caught engaging in bribery by federal au-
thorities, and thrown into two jail cells. If each jail cell is to hold at
least one lobbyist, in how many ways can we arrange them? Assume
the cells are identical.

13. Six investors wish to invest their savings into four mutual funds; they
have eight funds from which to choose, and each investor will put his
or her savings into just one fund. In how many ways can four funds
be chosen and the money invested? Assume the funds are distinct and
each of the four chosen funds receives some money.

14. How many ways are there to write 154 = 2 × 7 × 11 as a product of

distinct integers? Write out each product.

15. How many partitions of a set of eight elements are there?

16. Use the recurrence of Theorem 7.7 to find 74 , 75 , and 76 .

     
 Catalan, Bell, and Stirling Numbers 139

17. We have six people who will be seated at two circular tables. How many
ways can we do this if the tables are identical? How many ways if one
table is distinguished?

Exercises 7B
1. An election takes place in which each of two candidates A and B will
receive n votes; there are 2n voters. In how many ways can the voters
line up so that candidate B will never be in the lead? It may help to
represent the line with a “)” for a voter for B and “(” for a voter for A.
2. Show each of the distinct ways of properly pairing up four pairs of
parentheses.
3. Use the Catalan recurrence to verify the table value of C6 .
4. Use Theorem 7.1 to calculate C8 .
n o
i
5. Make a 5 × 5 matrix whose (i, j)-entry is j . Calculate the inverse
of this matrix (by hand, with a calculator, or using a computer algebra
system such as Mathematica or Maple). Compare the entries in this
inverse with the Stirling numbers of the first kind.
6. Use the recurrence to calculate k7 for k = 5, 6, 7.


7. How many ways are there to factor the number 2×3×5×11×23 = 7, 590
into a product of two factors each greater than 1?
8. In how many ways can we distribute our last seven (distinct) candy
bars on Halloween night to two identical trick-or-treaters if we ensure
each child gets at least one candy bar? How many ways are there if the
trick-or-treaters are distinct (different costumes)?
9. Repeat the previous exercise in the case where we do not need to dis-
tribute all the candy bars.
10. Seven federal agents set out to investigate three politicians suspected of
being corrupt. In how many ways can this be done, assuming that each
politician is investigated by at least one agent, and the politicians are
considered distinct?
11. Three corrupt politicians are sent to two luxurious minimum-security
penal institutions. If each institution receives at least one politician,
and the institutions are identical, in how many ways can we do this?
140 Introduction to Combinatorics

12. Seven investors wish to invest their savings into four mutual funds; they
have six funds from which to choose, and each investor will put his or her
savings into just one fund. In how many ways can four funds be chosen
and the money invested? Assume the funds are distinct and each of the
four chosen funds receives some money.
13. How many ways are there to write 385 = 5 × 7 × 11 as a product of
distinct integers? Write out each product.
14. How many partitions of a set of seven elements are there?
15. Use the recurrence of Theorem 7.7 to find 71 , 72 , and 73 .
     

16. We have six people who will be seated at four circular tables. In how
many ways can we do this if the tables are identical? In how many ways
if one table is distinct from the other two?

Problems

Problem 1: A group of 2n people are seated around a circular table. In how

many ways can they shake hands simultaneously so that every participant
shakes hands with another and no handshakes cross other handshakes?
Problem 2: Show that i≤n  ni  = n!.
P  

√
1− 1−4(x)
Problem 3: Find the reciprocal of the generating function C(x) = 2x
and rationalize the denominator. Now use Newton’s binomial theorem, an in-
finite series from calculus, or a computer algebra system such as Mathematica
or Maple to expand 1/C(x) in an infinite series. What sequence does this
reciprocal function generate?
Problem 4:
2(2n + 1)
(i) Use the formula of Theorem 7.1 to show Cn+1 = Cn .
n+2
n
Y 2(2i + 1)
(ii) Show Cn+1 = .
i=0
i+2
Cn+1
(iii) Find lim .
n→∞ Cn
n o
p
Problem 5: Show that if p is an odd prime, then p−1 is a multiple of p.
n
Problem 6: Show that if n is odd then 2 is divisible by 3.
 Catalan, Bell, and Stirling Numbers 141

Problem 7: Prove Theorem 7.6.

h i
n
= − n2 .


Problem 8: Prove that n−1
h i
n
Problem 9: Confirm that n−2 = n(n + 1)(n + 2)(3n + 5)/24 for 2 ≤ n ≤ 6.
Chapter 8
Symmetries and the Pólya–Redfield
Method

We have seen how to enumerate circular permutations in Chapter 2; the idea

there is that two configurations are considered identical if one may be trans-
formed into another by means of a rotation. There are other possible ways for
two configurations to be equivalent, and in this chapter we look at the ways to
enumerate configurations that are distinct with respect to some collection of
symmetries. We will refer to these configurations as “colorings,” even though
what is meant by “color” might be nothing that an artist or decorator would
recognize. For instance, we may seat politicians at a circular table, and only
care which of (say) three political leanings (conservative, moderate, or liberal)
the politician has. Thus, we might label seats with the “colors” C, M , and L.
If we are concerned with gender, we might use “colors” M and F ; and with
religion or ethnicity, there would be many possible labels. For class-scheduling
purposes, a classroom might get a label indicating that it can accommodate
large classes or that it is equipped with a computer projector.

8.1 Introduction
In this chapter we discuss how to find the number of ways of coloring
objects (such as seats at a circular table, or charms on a charm bracelet) that
are distinct with respect to certain operations. Thus, we say two colorings
of seats at a circular table are the same (or equivalent, in the sense of an
equivalence relation) if one coloring may be transformed into another by a
rotation. In the same way, colorings of beads on a necklace or charm bracelet
may be considered equivalent under a rotation, or when the necklace is flipped
over.
Suppose we wish to color each corner of a square one of two colors. In how
many ways can this be accomplished? We can start off by naı̈vely dividing
the question into cases. There are two ways (one for each color) if we are to
use exactly one of the colors, so that the square is all one color. If we are to
use both colors, there are more possibilities. So we might do three corners
in color c1 , and the other in color c2 ; or we might do two corners in color

143
144 Introduction to Combinatorics

c1 and the other two in color c2 . In the second case, we realize that there
is a difference if adjacent corners are to get the same color, or not. At this
point, the problem appears to be tedious and simply a matter of eliminating
possibilities by hand. For a large problem (as, say, a dozen seats and four
colors), this approach is wholly inadequate. A more effective approach uses
some basic facts of symmetry groups; we begin by discussing groups.

8.2 Basics of Groups

For our purposes, a group is a collection of bijective (one-to-one and onto)
functions (called permutations) from a (usually finite) set S to itself that
satisfies the group axioms, which follow.

• The identity, commonly denoted e, is the function that maps an element

x ∈ S to itself; e(x) = x ∀x ∈ S. Every group must contain the
identity.
• The group is closed under composition of functions; for any σ, τ in the
group G, the function σ ◦ τ defined by σ ◦ τ (x) = σ(τ (x)) is also in G.
• The group contains inverses; that is, for any σ ∈ G there is an element
σ −1 in G with the property that σ ◦ τ = τ ◦ σ = e.

The reader should be familiar with the set of integers modulo n, denoted Zn ,
under addition, which we saw in Chapter 4. For instance, Z4 = {[0], [1], [2],
[3]}; the group operation + behaves like ordinary addition for small numbers,
and gives us the remainder upon division by 4, so for example [3] + [2] = [1]
in the group. The group axioms are satisfied; the identity element is [0],
each element has an inverse (so [1]−1 = [3], for instance), and the operation
is clearly closed since the only remainders possible upon division by 4 are
0 through 3. We may think of these elements as being functions from the
set {0, 1, 2, 3} to itself as well. So we may consider [1] to be the function
f (n) = n + 1(mod 8); this means that [1] is the function that takes 1 to 2,
2 to 3, and 3 to 0. It is also common to represent these functions as acting
on the set {1, 2, 3, 4}, and we will use this convention when we are looking at
symmetries of objects.
For the most part, we will consider the elements of a group to be functions
acting on the set S = {1, 2, . . . n} but may also use, for example, a set of differ-
ent ways to color an object. More abstract definitions of the group include the
axiom that the operation ◦ is associative, meaning that τ ◦ (σ ◦ π) = (τ ◦ σ) ◦ π.
This clearly holds for composition of functions, and for addition modulo n.
The student should notice that our notation differs significantly from the no-
tation used in many abstract or modern algebra texts. In many texts, the
 Symmetries and the Pólya–Redfield Method 145

notation τ π is used to mean π(τ (x)); we write multiplication “backwards” by

this standard.
For any n > 0, there is at least one group of order n (that is, containing
n elements). The only group of order 1 is the group consisting only of the
identity. We will describe the elements of a group using the two-line notation,
or cycle notation. (We looked at cycles of permutations in Section 7.5.) So
consider the function that maps each corner of a square to the next corner
clockwise from the center, corresponding to a 90◦ rotation clockwise, as in
Figure 8.1.

1 2

4 3
FIGURE 8.1: Square with numbered corners.

This function ρ will rotate the circle one turn clockwise; in two-line form,
this is denoted  
1 2 3 4
.
2 3 4 1
The same permutation, written in cycle notation, is (1234). In the two-line
notation, we evaluate ρ(1), for instance, by looking for the column in which
1 is in the top row; the second row entry in that column is ρ(1). In cycle
notation, we calculate ρ(1) by looking for the first number in the brackets
with 1 directly to the right of 1. To find ρ(4) using cycle notation, we see that
nothing is to the right of 4 within that set of brackets, so we return to the
first number, which is 1; thus ρ(4) = 1. The name is a reference to applying a
permutation σ to an element of the set S repeatedly; it causes that element to
“cycle” through the values in that cycle of the permutation. This permutation
is exactly equivalent to the element [1] of Z4 , as we saw above.
Example 8.1: Find the cycle notation for the permutation
 
1 2 3 4 5
.
3 5 1 2 4

We begin by asking to which element 1 is mapped; this is 3. This means

our permutation begins with (13 . . . ). Now, the third column of the two-line
form shows that 3 is sent to 1. That “closes the cycle,” so we get (13) for the
first cycle. Now, we look at the next element that we haven’t included; that
is 2. The two-line form indicates that 2 goes to 5, and 5 goes to 4, and 4 goes
146 Introduction to Combinatorics

back to 2. Thus the cycle notation is (13)(254). This permutation may also
be written (542)(31), for instance, so the cycle representation is not unique in
general. 

Example 8.2: Find the two-line notation for the permutation whose cycle
form is (12)(3)(45). (It is more common to omit elements that are sent to
themselves, or fixed points, as they are called; so the permutation might be
written (12)(45).)
To find the two-line notation, we begin by listing the top line, so we get:
 
1 2 3 4 5
.

Now inspecting the cycle form, we see that 1 is mapped to 2, 2 is sent to 1, 3

stays at 3, and 4 and 5 exchange places. We get:
 
1 2 3 4 5
2 1 3 5 4

and this is our two-line form. 

Given two groups G (permuting the elements of a set S) and H (permuting
elements of a set T ), we may form a new group G×H, the Cartesian product of
G and H, as follows. Given a permutation σ of G and a permutation τ of H, we
define a permutation (σ, τ ) on the set S × T in the obvious way: (σ, τ )(s, t) =
(σ(s), τ (t)). The group operation will work within each coordinate of the
ordered pair; so, (σ1 , τ1 )·(σ2 , τ2 ) = (σ1 σ2 , τ1 τ2 ). This is clearly well-defined
as a function from S × T to itself, and it satisfies the group axioms; the proof
is left to Problem 8.2.
Example 8.3: Find the Cartesian product of C2 × C2 and give its multi-
plication table.
This group acts on the set {1, 2} × {1, 2} = {(1, 1), (1, 2), (2, 1), (2, 2)}. If
we denote the elements of C2 as e and ρ, then C2 × C2 consists of (e, e), (e, ρ),
(ρ, e), and (ρ, ρ). For brevity, we will denote these as ee, eρ, and so forth; and
the set elements as 11, 12, and so on.
Now it is easy to see that the identity of the group C2 × C2 is ee, and
that each element is its own inverse. So, for instance, ρρ · ρρ = ee. Our
multiplication table is thus

ee eρ ρe ρρ
ee ee eρ ρe ρρ
eρ eρ ee ρρ ρe
ρe ρe ρρ ee eρ
ρρ ρρ ρe eρ ee
The reader should check the multiplication in each case. 
 Symmetries and the Pólya–Redfield Method 147

We will employ both cycle notation and two-line notation to represent per-
mutations as we continue with our discussion of groups. In particular, the
lengths of cycles and the numbers and kinds of fixed points will be important.
Example 8.4: Find the group of permutations that describe the symme-
tries of the square shown in Figure 8.1.
We have already mentioned the rotation ρ that describes rotation by 90◦ ;
the rotation ρ2 that describes rotation by 180◦ , the rotation ρ3 (270◦ ) and e
are the remaining planar motions that take the square into itself. The other
motions are four reflections that take the square out of the plane. We may
reflect the square about the line passing through the points 1 and 3, or 2 and
4, or through the vertical line bisecting the square, or the horizontal line that
bisects the square.

ρ = (1234); ρ2 = (13)(24); ρ3 = (1432); ρ4 = e = (1)(2)(3)(4)

We will use α, β, γ, and δ to represent the listed reflections.

α = (24); β = (13); γ = (12)(34); δ = (14)(23)

We have described eight symmetries, including the identity. Is this all? A
careful study of the square would seem to indicate that no other motions can
carry the square into itself. Because a group must be closed under composi-
tion of functions and inverses, we should check that all compositions and all
inverses are accounted for. The multiplication table for the group is below.

◦ e ρ ρ2 ρ3 α β γ δ
e e ρ ρ2 ρ3 α β γ δ
ρ ρ ρ2 ρ3 e γ δ β α
ρ2 ρ2 ρ3 e ρ β α δ γ
ρ3 ρ3 e ρ ρ2 δ γ α β
α α δ β γ e ρ2 ρ3 ρ
β β γ α δ ρ2 e ρ ρ3
γ γ α δ β ρ ρ3 e ρ2
δ δ β γ α ρ3 ρ ρ2 e
Since we have found no new symmetries, we have a group (and in fact have
found all the symmetries of the square). The reader will notice that each row
and column contains the identity e as well as each other element of the group.
This group is variously called the dihedral group on four letters or the octic
group, and is denoted D4 . 
A number of points should be made here. First of all, notice that ρ ◦ α = γ,
but α ◦ ρ = δ. Unlike many common binary operations, then, function com-
position is not commutative in general. Groups in which the binary operation
is commutative are called abelian. All the groups Zn are abelian, because ad-
dition (and addition modulo n) is commutative. Also, the Cartesian product
148 Introduction to Combinatorics

of two abelian groups is an abelian group; the proof of this assertion is left to
Problem 8.2.
Next, if we ignore the reflections α, β, γ, and δ, we get a set of elements
{e, ρ, ρ2 , ρ3 }. This set forms a group as well; it is closed under composition
and each element has its inverse (the inverse of ρ is ρ3 , and ρ2 is its own
inverse). This is the group of rotations, and is referred to as the cyclic group
on four letters, denoted C4 . In general, any regular n-sided polygon has a
group Dn of 2n symmetries that contains the n rotations of the group Cn and
n reflections about n axes of symmetry. For any n, Cn is abelian, but Dn is
not, for n ≥ 3. It is worth noticing that the group Zn is essentially the same
as the group Cn ; that is, if we take the element [1] of Zn and write it as ρ,
and write [2] and ρ2 , and so forth, we get the same multiplication table. If
we have two groups that, under a suitable renaming of the elements, have the
same multiplication table, we say the groups are isomorphic, and the function
that maps an element of one group to the corresponding element of the other
is called an isomorphism.
We define a subgroup of a group G to be a subset of G that forms a group;
that is, H is a subgroup of G if H ⊆ G, H includes the identity e, and H is
closed under the group operation and under inverse. Thus, we have seen that
Cn is a subgroup of Dn . When we define Sn to be the set of all permutations
of a set of n objects, we call Sn the (full) symmetric group on n letters, and Dn
is a subgroup of Sn . Another subgroup of D4 is {e, α, β, ρ2 }; we will denote
this group by K. Confirmation that this is a subgroup is left to Problem 8.3.
We observe that the function α of D4 sends 1 to 1, and 3 to 3; 1 and 3
are called fixed points of the permutation α. Clearly, every point is a fixed
point of the identity e; also, non-identity elements of Cn have no fixed points.
We will see that fixed points are a useful idea when we count distinct ways to
color an object. The notation F P (σ) will mean the set of fixed points of the
permutation σ, so that F P (α) = {1, 3}.
A group partitions the elements of the set S into orbits; the orbit of an
element of S is the set of all elements of S to which it may be sent by one
of the functions of the group. The group is said to be transitive on S if
there is only one orbit. Equivalently, G is transitive if and only if, given any
two elements a and b of S, there is an element σab with the property that
σab (a) = b. Other characterizations of transitivity exist; see Problems 8.4
and 8.7.
Example 8.5: Find the orbits of the group K.
We observe that β and ρ2 exchange 1 and 3; no element of K takes 1 to 2
or 4, so one orbit is the set {1, 3}. Now, α and ρ2 both take 2 to 4, but no
element of K takes either of 2 or 4 to either of 1 or 3. So there are two orbits
of K, and S = {1, 3} ∪ {2, 4}. 
A symmetry of the square may be thought of as a permutation, not just of
the corners of the square, but of the sides of the square as well. For example,
 Symmetries and the Pólya–Redfield Method 149

if we label the sides of the square shown in Figure 8.1 by the numbers of the
corners, we may look at ρ as the permutation that sends side 12 to side 23,
side 23 to side 34, and so on. We sometimes say that this permutation of the
sides is induced by ρ, and the group of permutations of the corners induces
a group of permutations of the sides. It is not hard to see that the group of
permutations of the corners is in fact isomorphic to the group of permutations
of the sides. We also say that the group acts on the sides.
It is worth noticing that the permutations of the sides induced by the
permutations of the corners will not have the same numbers of fixed points in
general as the permutations that induce them. For instance, α = (24) fixes
two corners, but the permutation of the sides induced by α exchanges side 12
with 14, and side 23 with 34. On the other hand, γ = (12)(34) does not fix
any corner, but the induced permutation fixes the sides 12 and 34.

1 a 2 1 a 2 1 a 2
f b f b f b
6 3 6 3 6 3
e c e c e c
5 d 4 5 d 4 5 d 4

FIGURE 8.2: Three hexagons with labeled vertices and sides.

Example 8.6: Find the permutations in the dihedral group D6 represent-

ing the symmetries of the first of the three hexagons in Figure 8.2. Find the
actions of the permutations of this group on the sides of the hexagon. Do the
permutations induce an action on the triangles within the second hexagon
in this figure? Do these permutations induce an action on the quadrilaterals
within the third hexagon? Explain why or why not.

We have the rotation ρ = (123456) representing a 60◦ clockwise rotation

of the figure, and the powers ρ2 = (135)(246), ρ3 = (14)(25)(36), ρ4 =
(153)(264), ρ4 = (165432), and ρ6 = e. Now we list the six reflections;
they are α = (12)(36)(45), β = (13)(46), γ = (23)(14)(56), δ = (15)(24),
ǫ = (16)(25)(34), and φ = (26)(35).
We can see that ρ will move the side a to the side b, which goes to side c,
and so on; so the action of ρ on the sides is the cycle (abcdef ). In the same
way, ρ2 is (ace)(bdf ), ρ3 = (ad)(bc)(ef ), ρ4 = (aec)(bf d), ρ5 = (af edcb),
ρ6 = e, α = (bf )(ce), β = (ab)(cf )(de), γ = (ac)(df ), δ = (ad)(bc)(ef ),
ǫ = (ae)(bd), and φ = (af )(be)(cd). Notice that a permutation that fixes
some vertices, such as β, may not fix any edge, and a permutation that leaves
an edge unchanged (such as ǫ that sends edge c to itself) leaves no vertex
unchanged.
150 Introduction to Combinatorics

Now, each permutation of D6 acts on the interior triangles of the middle

hexagon. Thus, ρ takes the triangle that includes the side a to the triangle
that includes b, and so forth. Each triangle is sent to exactly one other triangle
by any particular permutation in D6 , so this group acts on the triangles. If
we were to label each triangle A (for the triangle that includes side a) and
B (for that containing b) and so on, we could rewrite each permutation as a
permutation on the letters A through F . Notice that the permutations we get
on A through F are the same permutations as those on the edges, just with
uppercase letters.
Finally, we look at the result of these permutations on the quadrilaterals
of the rightmost hexagon. It is clear that, for instance, the permutation ρ
does not send the quadrilateral with vertices 1, 2, 3, and 6 to either of the
two quadrilaterals. Thus, ρ in particular does not induce an action on the
quadrilaterals. However, some of the  permutations of D6 act on these two
quadrilaterals; in particular, the set e, α, δ, ρ3 is a subgroup of D6 that acts

on the two quadrilaterals. Both the elements e and α leave the quadrilaterals
unchanged, and δ and ρ3 exchange the two quadrilaterals. 

8.3 Permutations and Colorings

In Section 8.2, we saw that a permutation on the corners of the square could
be regarded as a permutation on the sides. In a similar way, we can look at
the permutation of the colorings of the square induced by the permutation
of the corners. If we color the corners of the square with k colors, there are
clearly k 4 distinct colorings, many of which are equivalent to one another
under symmetries of the square.
Consider for concreteness the coloring C1 in which corner 1 is given color
c1 and the other corners are given color c2 . If we apply ρ, we send corner 1
to corner 2, so that we have a new coloring in which corner 2 receives color
c1 , and all the other corners receive color c2 . If we denote this coloring by
C2 , then we can say ρ(C1 ) = C2 , where ρ is the permutation on the set of
colorings induced by the permutation ρ of the set of corners. We will denote
both permutations by ρ when no confusion will result.
Example 8.7: Color the corners of the square of Figure 8.1 with two colors
(black and white) in such a way that the coloring is fixed by the permutation
induced by ρ2 .
Without loss of generality, we will color corner 1 with black. Now, since ρ2
exchanges corner 1 and corner 3, we must also color corner 3 with black. In
the same way, if we color corner 2 white, we must also color corner 4 white.
(We could find other ways to color the corners that would have the same
 Symmetries and the Pólya–Redfield Method 151

result.) In general, all the corners that are in the same cycle as corner 1 must
get the same color as corner 1, and the same for each other corner. 

Example 8.8: Find all colorings of the square with four colors that are
fixed by the permutation ρ.
Since ρ has only one cycle, all four corners must get the same color. Since
we have four colors to choose from, we have exactly four colorings; C1 , in
which all four corners get color c1 , C2 with all color c2 , C3 with all color c3 ,
and C4 with all color c4 . 

THEOREM 8.1 (Enumerating Equivalence Classes)

The number of colorings with k colors that are fixed by a permutation σ with
n cycles is k n .

Proof. We may choose one of k colors to use for each of the points in the
first cycle; then all the points of that cycle must receive the same color. We
choose one of the k colors independently for a point of the next cycle (if the
permutation has more than one cycle). When we perform this process for each
of the n cycles, the multiplication principle tells us that there are k n ways to
color the figure. 

8.4 An Important Counting Theorem

The main result of this section was known to Cauchy, proven by Frobenius,
and published by Burnside, so that it is often called Burnside’s lemma, or
the Cauchy–Frobenius–Burnside theorem. The student is warned that the
demonstration given here is necessarily somewhat incomplete. We have, how-
ever, given an idea of the proof that, while glossing over some group-theoretic
detail, will give the student a good idea of why the result is true (and how a
more complete proof could be made).
Given a group G (such as the group of symmetries of a square) acting on
a set S (such as the set of ways to color the corners of this square), we wish
to know how many distinct colorings (i.e., equivalence classes or orbits of
colorings) there are. Let this number be N . Here F P (σ) denotes the number
of colorings fixed by σ, not the number of corners or sides; we view σ as a
permutation of the colorings. Given a coloring C, we let Stab(C), the stabilizer
of C, be the set of permutations σ of G for which σ(C) = C. The stabilizer is
a subgroup of G; see Problem 8.6.
How many colorings are equivalent to C under the action of permutations in
G? We shall see that the answer is |G| / |Stab(C)|. For, suppose the colorings
152 Introduction to Combinatorics

C1 and C2 are equivalent (so that there is some σ in G for which σ(C1 ) = C2 ).
Then for any τ in the stabilizer of C1 (so τ (C1 ) = C1 ), we have σ ◦ τ (C1 ) = C2 .
It follows that each equivalent coloring C2 of C1 is the image of C1 for exactly
|Stab(C1 )| permutations in G. That means there are |G| / |Stab(C1 )| such
equivalent colorings. Let |[C]| denote the number of colorings equivalent to
the coloring C (we use square brackets to denote an equivalence class). Then

|Stab(C1 )| = |G| / |[C1 ]| for any C1 ∈ S. (8.1)

THEOREM 8.2 (Cauchy–Frobenius–Burnside Theorem)

For G, S, and N as above, we have
1 X
N= |F P (σ)| .
|G|
σ∈G

Proof. We shall count the set {(σ, C) : C ∈ S, σ ∈ Stab(C)}, in two different

ways. Equating the two expressions will give us the result. The first thing we
do is to observe that given a σ ∈ G,
P there are |F P (σ)| colorings that it fixes;
so the quantity in question is just σ∈G |F P (σ)|. Alternatively, we may sum
P all colorings C in S the number of permutations
over P that leave C fixed; this
is C∈S |Stab(C)|. Equation (8.1) turns this into C∈S |G| / |[C]|. Factoring
out |G|, we have
X X
|F P (σ)| = |G| 1/ |[C]| .
σ∈G C∈S

Now, N is the number of equivalence classes that we are trying to enu-

merate. The sum on the right-hand side above may be grouped; the various
colorings in a given equivalence class [C] will each contribute 1/ |[C]| to the
sum. So all these colorings will contribute a 1 to the sum. It follows that this
sum is just N , the number of equivalence classes. 

Example 8.9: How many ways can we color a length of board with up
to three colors if we place one color at each end and one color in the middle?
Two colorings are equivalent if one may be obtained from the other by flipping
the board end-for-end (corresponding to the group C2 ).
We see that there are 27 ways to color the board in all (without considering
equivalence classes); there are three colors for the left end, three for the middle,
and three for the right end, giving 33 . Now, the temptation to simply divide
by two (since each coloring would appear to be equivalent to exactly one other,
obtained by rotating the board by 180◦ ) might be great, but it would lead us
to conclude that there are 27/2 or 13 21 colorings, surely an absurd result. We
will count the equivalence classes of colorings twice, once by combinatorial
reasoning and once using Theorem 8.2. First, consider that the only colorings
that are different but equivalent are the colorings that are not fixed by ρ,
 Symmetries and the Pólya–Redfield Method 153

the permutation that exchanges the colors on the left and right ends of the
board. We see that ρ has just two orbits; one orbit consists of the left and
right ends of the board, and the other consists of the middle of the board.
By Theorem 8.1, ρ fixes 9 colorings, so there are 18 it does not fix. These
18 colorings give us nine equivalence classes. The nine colorings fixed by ρ
are equivalent only to themselves; each is its own equivalence class. Hence we
have 18 different equivalence classes (ways to color the board).
Now, using Theorem 8.2, we use |G| = 2, and |F P (e)| = 27 because every
possible coloring is fixed by the identity permutation. Since |F P (ρ)| = 9 by
the foregoing reasoning, N = (27 + 9) /2 = 18. 

Example 8.10: We wish to seat n people at a circular table; how many

ways are there to do this? We let the group be Cn and observe that there are
a total of n! colorings (i.e., ways to order the people).
In Chapter 2, we observed that the number of ways to accomplish this
is (n − 1)!; we now use Theorem 8.2 to get the same result. Because we
never repeat any of the “colors” (the n people occupy only one seat each),
no permutation fixes any coloring except for the identity, which fixes all the
colorings. Thus the number of equivalence classes is just (1/ |Cn |)(n!) =
(n − 1)!. 

5
5
6 4
4 3
1 3
1 2
2

FIGURE 8.3: Two figures for coloring.

Example 8.11: In how many ways can we color the corners of the pentagon
on the left of Figure 8.3 with k colors if we assume colorings are equivalent
under reflection about the vertical axis? This reflection corresponds to the
permutation (12)(34); the figure’s symmetry group thus has two elements
(this permutation and the identity). Clearly, every coloring is fixed by the
identity permutation; to compute the number of colorings fixed by the other
permutation, we observe that corners 1 and 2 must receive the same color,
and corners 3 and 4 must receive the same color.
154 Introduction to Combinatorics

There are k 5 total ways to color the corners with k colors (multiplication
principle); all k 5 of them are fixed by the identity permutation. The other
permutation fixes only k 3 colorings, because we may choose any of k colors
for corner 1, any of k colors for corner 2, and any of k colors for corner 3; the
colors for the other corners are determined. The Cauchy–Frobenius–Burnside
theorem says that there are 21 (k 5 + k 3 ) non-equivalent colorings. 

8.5 The Pólya and Redfield Theorem

What we have done so far is effective for counting the equivalence classes of
colorings of structures with small groups. If |G| is not too great, we can simply
enumerate the colorings that are equivalent under the various permutations
in G and find ourselves with the number of non-equivalent colorings after only
a little work. However, what are we to do if we wish to color a complex object
in which |G| is very large? Just as in Chapter 6, we will find that generating
functions can take much of the work and automate it.
In particular, each permutation is associated with a cycle enumerator,
which is a product of variables that describe its cycles. First, we take a
permutation and write it as a product of disjoint cycles, which may be done
in essentially one way; for example, consider the permutation (12)(456). It
has one cycle of length 1, one of length 2, and one of length 3. We describe it
by the monomial z11 z21 z31 . More generally, we put a factor of zij to indicate j
disjoint cycles, each of length i, in the permutation.

Example 8.12: Find the cycle enumerators of the eight permutations in

D4 as it acts on Figure 8.1.

The permutations, as outlined earlier, are e, ρ = (1234), ρ2 = (13)(24),

3
ρ = (1432), α = (24), β = (13), γ = (12)(34), and δ = (14)(23). Since e
consists of four one-cycles, it has cycle enumerator z14 . Similarly, ρ and ρ3 are
both z41 . We find ρ2 , γ, and δ are all z22 , and α and β are both z12 z21 . 

Now, to enumerate the non-equivalent ways to color a figure with k colors,

we form a polynomial C(z1 , z2 , . . . zn ) by adding the cycle enumerators and
dividing by |G|. This polynomial is called the cycle index of G (more properly,
of the action of G on the objects being colored). For the previous example,
this gives us 81 (z14 + 2z41 + 3z22 + 2z12 z21 ). What good does this polynomial
do us? By substituting k for each of the zi s, we find the number of ways of
coloring with k colors!

Example 8.13: How many ways are there to color the pentagon of Fig-
ure 8.3 with k colors?
 Symmetries and the Pólya–Redfield Method 155

The permutations consist of the identity e and the reflection (12)(34). These
have cycle enumerators of z15 for e and z11 z22 for the reflection. Our cycle index
is thus C(z1 , z2 ) = 12 (z15 + z1 z22 ). Inserting k for each z, we get (k 5 + k 3 )/2,
just as with the example of the last section. 

THEOREM 8.3
Given a figure with a symmetry group G having cycle index C(z1 , . . . zn ),
the number of colorings not equivalent under action of G using k colors is
C(k, . . . k).

The detailed proof is omitted. Roughly speaking, however, each cycle enu-
merator evaluated at k will give the number of k-colorings that are fixed
under the action of the corresponding permutation. Thus, the cycle index
is the right-hand-side of the equation of Theorem 8.2 when evaluated at k,
giving us the number of equivalence classes of colorings.
Example 8.14: Count the colorings of the hexagon of Figure 8.3.
For this hexagon, two colorings are equivalent under vertical reflection,
horizontal reflection, or rotation by 180◦ . The symmetry group here thus
(with the identity) has four elements; they are α = (13)(46), β = (16)(25)(34),
ρ = (14)(25)(36), and e. The cycle enumerators are z12 z22 for α, z23 for β and
ρ, and z16 for e. Thus, the total number of non-equivalent colorings with k
colors is C(k) = 14 (k 6 + k 4 + 2k 3 ). 

a 2 b 4 c

1 3 5

d 7 e 6 f

FIGURE 8.4: Coloring corners or edges.

It is important to make sure that the cycle index is appropriate to the ob-
jects being counted. Consider, for instance, the rectangle shown in Figure 8.4.
We will consider coloring the corners, which are labeled with letters a through
f , and the sides, labeled with the numbers 1 through 7. In each case, we con-
sider two colorings equivalent if one is moved into the other by reflection
across the horizontal axis, reflection across the vertical axis, or rotation by
180◦ . Notice that this is the group D2 (two rotations and two reflections, if we
count the identity as a 360◦ rotation), even though it is acting on six corners
and seven sides.
156 Introduction to Combinatorics

Example 8.15: How many non-equivalent ways can we color the corners
of Figure 8.4 with k colors?
The four permutations are: the identity, fixing each of the six corners;
the horizontal reflection, corresponding to the permutation (ad)(be)(cf ); the
vertical reflection, corresponding to (ac)(df ); and the rotation, (af )(be)(cd).
The cycle enumerators are, respectively, z16 , z23 , z12 z22 , and z23 . This gives
us a cycle index of 14 (z16 + 2z23 + z12 z22 ). This gives us a total of C(k) =
(k 6 + k 4 + 2k 3 )/4 ways to color the corners with k colors. 

Example 8.16: How many non-equivalent ways can we color the sides
with k colors?
The same four symmetries now have different cycle enumerators. The hor-
izontal reflection is (27)(46); the vertical reflection is (15)(24)(67); and the
rotation is (15)(26)(47). Each of these fixes at least one edge. Our cycle
index is 14 (z17 + z13 z22 + 2z11 z23 ). This gives us C(k) = (k 7 + k 5 + 2k 4 )/4 ways
to color the edges with k colors. Notice that this answer is just k 2 times the
result for the previous example. 
We can now easily find the number of non-equivalent ways to color (using k
colors, for any positive integral value of k) a figure with its symmetry group.
This does not allow us to count the number of non-equivalent colorings subject
to specific restrictions, however. In particular, we might ask how many ways
there are if each of the k colors must be used at least once; how many ways
there are if we must use more of one color than another; and so forth. These
questions require a little more effort, but can still be addressed through a
slight modification of the technique.
Suppose we are going to color with k colors; we introduce variables d1 , d2 ,
. . . dk . If we take the cycle index C(z1 , . . . zn ) and replace each variable zi
with the expression di1 + di2 + · · · + dik , the result is a new polynomial in the
variables d1 , . . . dk . We denote this function by Ĉ.

Ĉ(d1 , . . . dk ) = C(d1 + · · · + dk , d21 + · · · + d2k , . . . , dn1 + · · · + dnk )

THEOREM 8.4 (The Pólya–Redfield Theorem)

Given a figure with symmetry group G having cycle index C(z1 , . . . zn ), the
number of inequivalent colorings with k colors, in which color ci is used mi
times, is the coefficient of dm 1 m2 mk
1 d2 . . . dk in Ĉ(d1 , . . . dk ) (where m1 + m1 +
· · · + mk = n is the number of points being colored).

Example 8.17: Color the corners of the square of Figure 8.1 with two
colors, red and blue (using variables r and b).
We will use D4 for the group; that is, two colorings are equivalent if there
is a permutation in D4 that maps one to the other. As we calculated earlier,
 Symmetries and the Pólya–Redfield Method 157

our cycle index is C(z1 , z2 , z3 , z4 ) = (z14 + 2z4 + 3z22 + 2z12 z2 )/8. Substituting
zi by ri + bi gives us
Ĉ(r, b) = (r + b)4 + 2(r4 + b4 ) + 3(r2 + b2 )2 + 2(r + b)2 (r2 + b2 ) /8.

This reduces, after sufficient algebra, to

Ĉ(r, b) = r4 + r3 b + 2r2 b2 + rb3 + b4 .
Each coefficient may be easily verified by combinatorial reasoning; for in-
stance, to color the corners with two red and two blue, the two red corners
may be next to each other or on opposite sides, giving the two equivalence
classes of colorings. 

Example 8.18: How many non-equivalent colorings of the corners of the

square of Figure 8.1 are there using four colors, if each color must be used
precisely once?
Here we use variables d1 , d2 , d3 , and d4 . Rather than using the multino-
mial theorem of Chapter 2 to calculate (d1 + d2 + d3 + d4 )4 , we will employ
a computer algebra system to do the hard work for us. For instance, the
Mathematica sequence that follows will work.

chr[z1_, z2_, z3_, z4_] := (z1^4 + 2*z4 + 3*z2^2 + 2*z1^2*z2)/8

chr[d1 + d2 + d3 + d4, d1^2 + d2^2 + d3^2 + d4^2,

d1^3 + d2^3 + d3^3 + d4^3, d1^4 + d2^4 + d3^4 + d4^4]

Together[Expand[%]]

The first line defines the cycle index as chr; at the end of the line, using
Shift+Enter causes Mathematica to evaluate the expression and it will recog-
nize the function chr[z1, z2, z3, z4] as defined. So, for instance, entering
chr[3,3,3,3] and Shift+Enter will give the number of non-equivalent ways
to color the corners with three colors.
The second line outputs the expression for Ĉ as a function of d1 through
d4 , but not in an easily readable form. The third line causes Mathematica to
expand the powers and collect like terms so that we may read the result. The
output (again, we end each line with Shift+Enter) is:

(d1^4 + d1^3d2 + 2d1^2d2^2 + d1d2^3 + d2^4 + d1^3d3 +

2d1^2d2d3 + 2d1d2^2d3 + d2^3d3 + 2d1^2d3^2 +
2d1d2d3^2 + 2d2^2d3^2 + d1d3^3 + d2d3^3 + d3^4 +
d1^3d4 + 2d1^2d2d4 + 2d1d2^2d4 + d2^3d4 +
2d1^2d3d4 + 3d1d2d3d4 + 2d2^2d3d4 +
2d1d3^2d4 + 2d2d3^2d4 + d3^3d4 + 2d1^2d4^2 +
2d1d2d4^2 + 2d2^2d4^2 + 2d1d3d4^2 +
2d2d3d4^2 + 2d3^2d4^2 + d1d4^3 + d2d4^3 + d3d4^3 +
d4^4)
158 Introduction to Combinatorics

The coefficient of d1 d2 d3 d4 is 3, as we see in the middle of the fifth

line of output; there are therefore exactly three non-equivalent colorings of
the corners with four colors in which each color is used exactly once. If we
had not used the reflections (used the group C4 instead of D4 ) we would have
gotten 6, corresponding to the six cyclic permutations of four distinct objects.
A similar approach will work for Maple. In this case, we use the character
sequence - > to produce the right arrow, and we use the right arrow key →
to delimit some operations. For instance, if we wished to type x^2 + y^2, we
would need to use the right arrow key to get out of the exponent. That is, we
would type z^2→+y^2→ because all keystrokes after the ^ are interpreted as
part of the exponent otherwise. Here is the Maple code with right-arrow key
noted:

chr := (z1,z2,z3,z4)->(z1^4→+2z4+3z^2→+2z1^2→z2)/8→

chr(d1+d2+d3+d4, d1^2+d2^2+d3^2+d4^2, d1^3+d2^3+d3^3+d4^3,

d1^4+d2^4+d3^4+d4^4)/8

expand(%)

The result will be the same as for the Mathematica code.

Exercises 8A
1. Find the multiplication table of C5 , given ρ = (12345).

2. Find the two-row form of the permutation on five letters whose cycle
form is:
(i) (15)(234)
(ii) (1245)

3. Find the cycle form of the permutation whose two-row form is:
 
1 2 3 4 5
(i)
3 5 1 2 4
 
1 2 3 4 5
(ii)
2 5 4 1 3

4. Find the group of all symmetries of a 2 × 1 rectangle.

5. Find the Cartesian product C2 × C3 and its multiplication table.

 Symmetries and the Pólya–Redfield Method 159

6. Use the permutations of D3 , the symmetries of an equilateral triangle,

to find the cycle index of the triangle. Now replace each variable zi
by (ri + bi ) and expand (by hand or using a computer algebra system
such as Maple or Mathematica) to find all the non-equivalent ways of
two-coloring the corners of the triangle with the various distributions of
colors.

7. A four-foot pole is to be colored in bands of one foot in length. Two

colorings are equivalent under flipping the pole end-for-end. Label each
one-foot band, find the cycle index, and replace each zi by (ri + bi ). Use
the resulting expression to find the number of ways to two-color the pole
so that each color is used at least once.

8. Consider the problem of coloring the corners of a regular pentagon,

where colorings are equivalent under the cyclic group C5 . Find an ex-
pression for the number of distinct ways to k-color this figure; then
find an expression for two-coloring with colors r and b using the Pólya–
Redfield theorem.

9. Consider the problem of coloring the corners of a regular heptagon

(seven-sided polygon), where colorings are equivalent under the dihe-
dral group D7 . Find an expression for the number of distinct ways to
k-color this figure; then find an expression for two-coloring with colors
r and b using the Pólya–Redfield theorem.

Exercises 8B
1. Find the multiplication table of D3 , the symmetries of an equilateral
triangle with vertices 1, 2, and 3.

2. Find the two-row form of the permutation on six letters whose cycle
form is:
(i) (123)(456);
(ii) (23)(45).

3. Find the cycle form of the permutation on six letters whose two-row
form is: 
1 2 3 4 5 6
(i) ;
4 5 6 3 2 1
 
1 2 3 4 5 6
(ii)
6 5 4 3 2 1
160 Introduction to Combinatorics

4. Draw a diagonal from corner 1 to corner 3 of the square of Figure 8.1.

Find the subgroup of D4 that leaves this diagonal fixed. Hint. : This
group has four elements.

5. Find the Cartesian product C2 × C4 and its multiplication table.

6. Use the permutations of D2 , the symmetries of a non-square rectangle,

to find the cycle index of the rectangle (assume we are coloring the
corners). Now replace each variable zi by (ri + bi ) and expand (by hand
or using a computer algebra system such as Maple or Mathematica)
to find all the non-equivalent ways of two-coloring the corners of the
rectangle with the various distributions of colors.

7. A five-foot pole is to be colored in bands of one foot in length. Two

colorings are equivalent under flipping the pole end-for-end. Label each
one-foot band, find the cycle index, and replace each zi by (ri + bi ). Use
the resulting expression to find the number of ways to two-color the pole
so that each color is used at least once.

8. Consider the problem of coloring the corners of a regular pentagon,

where colorings are equivalent under the dihedral group D5 . Find an
expression for the number of distinct ways to k-color this figure; then
find an expression for two-coloring with colors r and b using the Pólya–
Redfield theorem.

9. Consider the problem of coloring the corners of a regular heptagon

(seven-sided polygon), where colorings are equivalent under the cyclic
group C7 . Find an expression for the number of distinct ways to k-color
this figure; then find an expression for two-coloring with colors r and b
using the Pólya–Redfield theorem.

Problems

Problem 1: Show that if G is a subgroup of H, and G is transitive on S,

then H is transitive on S. Find a counterexample to show that the converse
is false.
Problem 2: Prove that if G and H are abelian groups, then so is G × H.
Problem 3: Show that the set K contained in D4 is a subgroup.
Problem 4: Show that a group G is transitive on a set S = {1, 2, . . . n} if
and only if for any i ∈ S there is a permutation σi ∈ G where σi (1) = i.
 Symmetries and the Pólya–Redfield Method 161

Problem 5: Find an isomorphism between the group C2 × C2 and the group

K.

Problem 6: Show that the stabilizer StabG (s) of any element s ∈ S is a

subgroup of G.

Problem 7: A student wrote |G/[c1 ]| instead of |G| / |[c1 ]|, based on the
identity |a| / |b| = |a/b| taught in algebra. The algebra book notwithstanding,
the student’s teacher took off many points! Explain.

Problem 8: We are to color a square with a color in each corner and a color
in the center; two colorings are equivalent if one may be moved into another
by rotations of 90◦ (corresponding to the group C4 ). How many distinct
colorings are there with k colors? Notice that even though this group looks
like C4 , it acts on five points. How many colorings are there if we only color
the corners? Can you find a relationship between the solutions to these two
problems?

Problem 9: We wish to color a 3×1 rectangle so that each of its 1×1 squares
is given a color. Colorings are equivalent under a rotation of the rectangle
by 180◦. Determine the number ck of ways to do this with k colors. Now
use the methods of Chapter 6 to find the generating function of ck . Use the
denominator of this generating function to find a recurrence relation for ck .

Problem 10: Find all 16 permutations in the symmetry group D8 of sym-

metries of the regular octagon. Now group these into sets according to how
many cycles they have, and find the cycle index of the regular octagon. In
how many ways can the corners be colored with k colors?

Problem 11: A regular tetrahedron is a solid with four faces, each an equi-
lateral triangle. It has four corners which we label 1, 2, 3, and 4, and six
edges. The group of symmetries of the tetrahedron has 12 elements. These
include α = (234), β = (134), γ = (124), and δ = (123); the squares of these
elements; three other elements produced by products of these; and of course
the identity. Find all twelve symmetries as permutations of the corners; find
the cycle index of the tetrahedron. Now find an expression for the number of
ways to color the corners of the tetrahedron with k colors.

Problem 12: As in the previous problem, we consider a regular tetrahedron.

Find the twelve symmetries as permutations of the edges of the tetrahedron,
and find the cycle index. How many ways can the edges of the tetrahedron
be colored with k colors?

Problem 13: We wish to color the circles on the square shown in Figure 8.5.
Find the cycle index using the group C4 (rotations but no reflections). If you
have access to a computer algebra program, find Ĉ(r, b) for the two-colorings
of the figure.
162 Introduction to Combinatorics

1 2 3

4 5

6 7 8

FIGURE 8.5: Square with eight points to color.

Problem 14: Find the cycle index for Figure 8.5 using the group D4 (all
rotations and reflections). If you have access to a computer algebra program,
find Ĉ(r, b).
Problem 15: Generalize the exercises to find the number of ways to k-color
the corners of the regular polygon with p sides (where p is an odd prime)
where colorings are equivalent under the group Cp . Repeat using the group
Dp to allow reflections.
Problem 16: Find the number of ways to k-color the corners of a regular
polygon with 2p sides, where p is an odd prime, and colorings are equivalent
under the cyclic group C2p . Repeat using the group D2p to allow reflections.
Chapter 9
Partially Ordered Sets

9.1 Introduction
The first objects that we encounter in mathematics are ordered; that is,
given (for instance) two different integers, we know that one of them must
be larger than the other. Later in mathematics, we find objects that are
not ordered in the same way; for instance, there are two different groups of
order four, and it makes no sense to say that one of them is “bigger than” or
“comes before” the other. Between these two extremes, there are collections
for which it is possible to make a sensible statement that some members of the
collection are bigger or smaller than others, but for other pairs of members,
no comparison is possible. Such a collection of objects, together with the
appropriate definition of “bigger than,” is called a partially ordered set, or
poset for short.
More rigorously, a partially ordered set is a set S together with a relation
≺ that is irreflexive, antisymmetric, and transitive (see Appendix A to review
these terms). We may also use the relation , which is reflexive, antisymmet-
ric, and transitive; it is understood that that a  b means “Either a ≺ b or
a = b.” Either way, the relation is called a partial order on S. If the poset
has the property that any two elements x and y are related or comparable
(that is, either x ≺ y or y ≺ x), we say it is a total order , and refer to it as
“an ordered set.” A poset that is not a total order has at least one pair of
elements that are incomparable, i.e., there are x and y in the poset for which
neither x ≺ y nor y ≺ x is true. We write xky to indicate that x and y are
incomparable. Since the usual sets of numbers N, Q, R and Z are ordered sets,
we see that a partially ordered set is merely a generalization of these familiar
objects.
A poset is said to be locally finite if, for any comparable pair x ≺ y there
are only finitely many z satisfying x ≺ z ≺ y. Thus, N and Z are locally
finite; Q and R are not. If x ≺ y but there is no z for which x ≺ z ≺ y, we
say that y covers x. If the poset contains one or more elements x with the
property that x 6≺ z for any z, this element x is a maximum element of the
poset; similarly, if there is no z with z ≺ x, then x is a minimum element of
the poset.

163
164 Introduction to Combinatorics

9.2 Examples and Definitions

Consider a set X of n elements x1 , x2 , . . . xn , and let S be the collection of
all subsets of X. The partial order will be the relation ⊂, so that S1 ⊂ S2
means that S1 is a subset of S2 . We sometimes say this is the power set
of X ordered by containment. If n ≥ 3, we might have S1 = {x1 , x2 } and
S2 = {x2 , x3 }. In this case neither of the statements S1 ⊂ S2 nor S2 ⊂ S1
is true so that S1 and S2 are incomparable. If S3 = {x1 , x2 , x3 } then clearly
S1 ⊂ S3 and S2 ⊂ S3 ; we sometimes might write S3 ⊃ S1 to indicate S1 ⊂ S3 .
This poset is sufficiently well known that it has a number of names; we shall
call it Bn , referring to Boolean algebra. It is also possible to look at the poset
of multisubsets of a multiset; thus, X might consist of more than one “copy”
of some elements. So, for example, if X = {0, 0, 1}, its multisubsets include
{0}, {0, 0},{1}, {0, 1}, and the empty set ∅, as well as X. This concept will
be more thoroughly explored in the exercises and problems.
If x ≺ y in a poset P , and there is no z in P with x ≺ z ≺ y, we say y covers
x. Thus, the set {1, 2} covers both {1} and {2} in the poset B3 . However,
even though {1} ≺ {1, 2, 3} in the partial order, we do not say that {1, 2, 3}
covers {1} because there are other sets (in particular, {1, 2} and {1, 3}) that
lie between them.
It is convenient to diagram a finite poset in such a way as to show the partial
ordering; we form a graph whose vertices are the elements of the poset, and
an edge goes down from x to y if x covers y in the partial order. Such a graph
is called the Hasse diagram of the poset. We illustrate with the poset B3 in
Figure 9.1.

All All

All All All

All All All

All

FIGURE 9.1: Poset of subsets of a set of three elements.

 Partially Ordered Sets 165

Another common family of posets comes from group theory. Given a group
G, we can consider its collection of subgroups ordered by containment. That
is, if G1 and G2 are subgroups of G, we say G2 ≺ G1 if G2 is a subgroup of G1 .
We illustrate with the Hasse diagram of the subgroups of Z12 in Figure 9.2.

All

All All All

All All All

AllAll All

All

FIGURE 9.2: Poset of subgroups of Z12 , ordered by containment.

Another family of posets is found in number theory; consider an integer n

and all of its divisors. We will say k ≺ n provided that k is a divisor of n.
This is really an infinite poset on the set of all integers, but we may restrict
ourselves to the positive integer divisors of a specified integer. In Figure 9.3,
we see the divisors of 30. We denote the poset of the divisors of n by Dn .

All

All All All

All All All

All

FIGURE 9.3: Positive divisors of 30 ordered by divisibility.

166 Introduction to Combinatorics

Yet another family of posets consists of the posets of all partitions of a

set of n elements ordered by refinement. Recall that a partition of a set is a
collection of disjoint subsets whose union is that set. Suppose Π1 and Π2 are
partitions of the same set of n elements; we say that Π2 is a refinement of
Π1 provided that each subset in Π2 is entirely contained in one of the subsets
of Π1 . (We discussed partitions of finite sets in Section 7.2.) Consider, for
example, the set {1, 2, 3, 4}. There are B4 = 15 partitions of this set. The
partition {1, 2}{3}{4} is a refinement of the partition {1, 2}{3, 4}, but it is
not a refinement of {1}{2, 3}{4}.
An element x of a poset is said to be maximal provided that there is no
y with x < y, and minimal if there is no y with y < x. It is important not
to confuse these two definitions with two similar-looking ones; we say x is a
maximum element provided that y ≤ x for all y in the poset. In this case,
every element is comparable to the maximum element. In the same way, a
minimum element x is one for which every other element is comparable, and
greater. Not every poset has a maximum or a minimum, although every poset
we have seen so far has both. Clearly, if a poset has a maximum element, that
is the only maximal element, and similarly a minimum element is the unique
minimal element.

THEOREM 9.1 (Extreme Elements)

A finite poset possesses at least one maximal and at least one minimal ele-
ment.

Proof. We will prove there exists a maximal element, and leave the proof
of a minimal element to Problem 1. Choose an element x1 ; if there is no
element greater than it, it is maximal and we are done. Otherwise, find x2
with x1 ≺ x2 ; if there is no element greater than x2 , we are done. Continue
this process; it must terminate, as there are only finitely many elements in
the poset. When it terminates, we will have found a maximal element. 
The theorem is easily seen to be false for infinite posets; consider, for in-
stance, the set of all integers with the usual order. A poset without a maxi-
mum or a minimum element is shown in Figure 9.4.

9.3 Bounds and Lattices

If a poset containing elements a and b also contains an element x with x  a
and x  b, we say that x is a lower bound of a and b; similarly an upper bound
of a and b is an element x with a  x and b  x. There may be elements of a
poset that have no upper or lower bound; consider, for example, the elements
c and e of Figure 9.4, which have neither. We say that x is the least upper
 Partially Ordered Sets 167

All All

All All All

All All

FIGURE 9.4: Poset with no unique minimum or maximum element.

bound of a and b if x is an upper bound of a and b, and for any y that is also
an upper bound of a and b we must have x  y. There is a corresponding
definition for greatest lower bound ; x is the greatest lower bound of a and b
provided that x is a lower bound of a and b, and whenever y is a lower bound
of a and b then y  x. (Notice that this implies that the greatest lower bound
and least upper bound of two elements is unique when it is defined.) When
they exist, the greatest lower bound and least upper bound of two elements
a and b are denoted by a ∧ b (also called the meet of a and b) for the greatest
lower bound, and by a ∨ b (the join) for the least upper bound.
Not every pair of elements in a poset has a meet or a join; again, consider
the poset of Figure 9.4. The elements c and e have no upper or lower bounds
at all, so certainly no meet or join. However, in some posets every pair of
elements has a meet and a join; such a poset is called a lattice. Many of the
posets we have discussed are lattices.
The meet and join operations take familiar forms in some kinds of posets;
for instance, in the poset Bn the meet of two sets S ∧ T is the intersection
S ∩ T , and the join S ∨ T is the union S ∪ T . Similarly, in Dn it is not hard to
see that a ∧ b is the greatest common divisor of a and b, whereas a ∨ b is the
least common multiple. The terms meet and join, as well as the shapes of the
operators, help us remember which is which; meet is akin to set intersection,
and is smaller. Join is akin to union, and is the larger.

THEOREM 9.2 (Properties of Meet and Join)

In any lattice, for any three elements a, b, and c, the meet and join operations
satisfy:

1. Associative law: a ∧ (b ∧ c) = (a ∧ b) ∧ c and a ∨ (b ∨ c) = (a ∨ b) ∨ c).

2. Commutative law: a ∧ b = b ∧ a and a ∨ b = b ∨ a.
3. Idempotent law: a ∧ a = a and a ∨ a = a.
4. Absorption law: a ∧ (a ∨ b) = a and a ∨ (a ∧ b) = a.
5. a ∧ b = a if and only if a ∨ b = b if and only if a  b.
168 Introduction to Combinatorics

Proof. We will prove the associative law for the join and leave the others
to exercises. Because we are dealing with a lattice, we know that any two
elements have a meet and a join, so there is no question that both sides
of the equation are defined; we merely need show that they are equal. Let
x = a ∧ (b ∧ c) and y = (a ∧ b) ∧ c. By definition, we see x  a and x  (b ∧ c),
so x  b and x  c. Now a ∧ b is the least element w for which w  a and
w  b, so x  (a ∧ b). (It is not possible for x and w to be incomparable, for if
this were so, a and b would not have a unique least upper bound.) It follows
that x  (a ∧ b) ∧ c = y. An exactly similar proof shows that y  x as well;
because the partial order relation is irreflexive, we must have x = y. 

THEOREM 9.3 (Unique Maximum and Minimum)

Every finite lattice has a unique maximum element 1̄ and a unique minimum
element 0̄.

Proof. Given the finite lattice, we know from Theorem 9.1 that there is at
least one minimal element and one maximal element. Take such a minimal
element, which we will call 0̄. If x is any other element of the lattice, find
0̄ ∧ x = z. By definition, z  0̄, which means (since 0̄ is minimal) that z = 0̄.
Since 0̄ ∧ x = 0̄, by Theorem 9.2(5), we have 0  x. Thus 0̄ is the unique
minimum element. The existence of the unique maximum element is similar.

A lattice is said to be distributive provided that meet and join each dis-
tribute over the other; that is, that

a ∧ b ∨ c = (a ∧ b) ∨ (a ∧ c) and a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c).

We know from set theory (see Appendix A) that union and intersection
of sets distribute in this way, so that every subset lattice is a distributive
lattice. Not every lattice is a distributive lattice; see, for instance, the lattice
in Figure 9.5.

9.4 Isomorphism and Cartesian Products

The reader has probably noticed that the Hasse diagrams of Figure 9.1
and of Figure 9.3 look very much alike; the posets have the same number of
elements, and their Hasse diagrams are arranged in the same pattern. Ac-
cordingly, we define two posets to be isomorphic if there is a bijection φ
between their elements that is order-preserving; that is, x  y if and only
if φ(x)  φ(y). Thus we can see that D30 is isomorphic to B3 ; for instance,
φ({1}) = 2, φ({2} = 3, and φ({3}) = 5 gives rise to an isomorphism between
 Partially Ordered Sets 169

All

All All All

All

FIGURE 9.5: A lattice that is not distributive.

the two. Clearly, any two total orders on n elements are isomorphic; other
isomorphisms are explored in the exercises.
Example 9.1: Find an isomorphism between the poset D12 and the poset
of submultisets of the multiset {a, a, b}.
We see from the Hasse diagrams that these appear to be isomorphic. Create
an isomorphism by mapping each occurrence of a in a submultiset to a factor
of 2 in a divisor of 12, and each b to a factor of 3. Thus {a, b} corresponds to
6, {a, a} to 4, and so forth. We can easily confirm that this is an isomorphism.

Clearly, if two posets have the same Hasse diagram (apart from the labels on
the vertices), they are isomorphic; we must be very careful, however, with the
converse. For instance, consider the two posets of the diagram in Figure 9.6.
The two graphs are isomorphic as graphs, but considered as directed graphs
with edges oriented downwards, they are different, and represent different
posets.
Given two posets X and Y , there are a few ways of constructing other
posets from them. The simplest way is perhaps the direct sum. This is
denoted X + Y and is defined on the set X ∪ Y ; we say a ≺ b in X ∪ Y if a ≺ b
in X, or a ≺ b in Y , and any element of X is incomparable to any element in
Y . The Hasse diagram of X + Y is the disjoint union of the Hasse diagrams
of X and of Y , respectively.
The Cartesian product of X and Y is the poset on the set of ordered pairs
{(x, y) : x ∈ X, y ∈ Y } in which (x1 , y1 ) ≺ (x2 , y2 ) whenever x1 ≺ x2 in X
and y1 ≺ y2 in Y .
Example 9.2: Find the Cartesian product of the poset B1 × B1 .
Recall that B1 is the poset whose Hasse diagram consists of two vertices
and a single edge; we may think of the elements of B1 as {1} and {} = ∅.
170 Introduction to Combinatorics

FIGURE 9.6: Two posets with almost the same Hasse diagram.
Then the product has four elements; they are ({1}, {1}), ({1}, ∅), (∅, {1}),
and (∅, ∅). We show the Hasse diagram of this poset below, next to the poset
of B2 for comparison purposes. 

AllAll All

AllAll All All All

All All

FIGURE 9.7: The Hasse diagram for B1 × B1 and for B2 .

The similarity is not a coincidence. For a more general statement, see

Problem 4.
It turns out that every finite poset is isomorphic to some subset of Bn , for
a suitable n. We shall set out to prove this result. Define a down-set of the
poset X to be a set Y ⊆ X with the property that, for any y in Y , if z ≺ y
then z ∈ Y ; that is, if Y contains any elements smaller than that element.
(Note that Y may be empty; thus every poset has at least two down-sets,
namely itself and the empty set.)
Example 9.3: Determine whether the set {10, 6, 3, 2, 1} is a down-set of
the poset D30 whose Hasse diagram is shown in Figure 9.3.
We see that the element 5 precedes 10 in the partial order, but 5 is not
in the given set, so it is not a down-set. When we omit 10, the resulting set
{6, 3, 2, 1} is a down-set. 
 Partially Ordered Sets 171

THEOREM 9.4 (Representation of Posets)

Any finite partially ordered set is isomorphic to a poset of sets ordered by
inclusion.

Proof. Given a finite poset X, for each element x of X denote by Lx the

down-set {y ∈ X : y  x}. (It is not in general true that every down-set of a
poset is of the form Lx for some x; for instance, in the poset D30 of Figure 9.3,
the set {5, 2, 1} is a down-set but is not of the form Lx .) We create a new poset
Y whose elements are {Lx : x ∈ X} and Lx ≺ Ly whenever Lx ⊂ Ly . Define
φ : X → Y by φ(x) = Lx . We demonstrate that this is an isomorphism. First
suppose φ(x) = φ(y); by definition, x ∈ Lx , and Lx = Ly . Now since x ∈ Ly
we must have x  y. Similarly, we conclude y  x; thus x = y. Also, φ is
clearly onto.
Now suppose x ≺ y in X; we demonstrate that Lx ⊂ Ly . But this is just
a restatement of the transitive property; any element z of Lx satisfies z  x,
and x ≺ y; so z ≺ y, and z ∈ Ly . 
In fact, more is true; for instance, if X is any finite distributive lattice, then
there is some poset for which X is isomorphic to its poset of down-sets. (This
refers to all down-sets, not just those of the form Lx .) For a deeper discussion
of this and related ideas, we refer the reader to a source such as [107].

9.5 Extremal Set Theory: Sperner’s and Dilworth’s The-

orems
A chain is a totally ordered subset of a poset; that is, C ⊆ P is a chain of P
if for any two distinct x, y in C we have either x < y or y < x. The height of a
poset is the length of any maximum-length chain of that poset. An antichain
is a subset of a poset in which no two elements are comparable; the width of
the poset is the number of elements in a maximum-sized antichain. In the
case of the poset B3 of the figure above, we see that its height is 4, because
the chain consisting of ∅, {1}, {1, 2}, and {1, 2, 3} is a maximum-length chain.
Its width is 3, as illustrated by the antichain {1}, {2}, and {3}. A question
of combinatorial interest is finding the size of the largest antichain in a poset;
this question was entirely settled for Bn in 1928.
A collection of sets is called a Sperner family if none of them is a proper
subset of another. It is clear that if we look at subsets of a set of cardinality
n, the subsets that all have k elements for a given k form a Sperner family,
as the proper subsets of each must have fewer than k elements. Recall that
⌊n/2⌋ is the greatest integer less than or equal to n/2, and ⌈n/2⌉ is the least
integer greater than or equal to n/2. Clearly, then, we can find an antichain
in Bn of C(n, ⌊n/2⌋) elements by taking the subsets of cardinality ⌊n/2⌋. Can
172 Introduction to Combinatorics

we do better? Sperner’s theorem says that we cannot.

THEOREM 9.5 (Sperner’s Theorem)

Let S be a Sperner family of subsets of an n-set. Then |S| ≤ C(n, ⌊n/2⌋),
and if equality holds then either S is the collection of all subsets of cardinality
⌊n/2⌋ or the collection of subsets of cardinality ⌈n/2⌉.

Proof. Given X = {x1 , x2 , . . . , xn } take any permutation σ of these elements;

we find a chain ∅ ⊂ {σ(x1 )} ⊂ {σ(x1 ), σ(x2 )} ⊂ · · · ⊂ X. As there are n! such
permutations, there are clearly n! chains of length n (maximal chains) in Bn .
What proportion of those chains contain a particular set A of k elements (say
x1 , x2 , . . . xk )? There are C(n, k) such sets A; by symmetry, each is contained
in an equal number of maximal chains, and that number is n!/C(n, k) =
k!(n − k)!. Now take our Sperner family S. By definition, no chain contains
more than one member of S. If any maximal chain contains no member of
S, we can add any set from this chain to S to make a larger Sperner family.
Assume that we have done this and now have a maximal Sperner family so
that S contains exactly one set from every maximal chain. We now partition
the set of maximal chains according to which set of S they contain; if Sk is
a member of S of cardinality k, it is in n!/C(n, k) maximal chains. Thus we
have
X X 1
n!/C(n, |S|) = n! n

= n!
S∈S S∈S |S|
P
This is equivalent to S∈S 1/C(n, |S|) ≤ 1; to make |S| as large as possible,
we want to make 1/C(n, |S|) as small as possible, so we want the largest
possible value of C(n, |S|). This clearly occurs when |S| = ⌊n/2⌋ or |S| =
⌈n/2⌉.
Now if n is even, we are done; each set in S must have exactly n/2 elements
to make the sum as small as possible. If n is odd, set n = 2r + 1. We could
make a maximum antichain with sets of cardinality r or of cardinality r + 1;
we need to show that we cannot make a maximum antichain with a mixture.
Suppose that S contains an r-set S0 . If T0 is any r + 1-set containing S,
clearly T0 ∈/ S. Now, remove an element of T0 to get a new set S1 6= S0 of
cardinality r. Notice that T0 and S1 are part of a chain; exactly one element
of each chain must be in S, and since T0 is not, S1 must be in S. We now
add a new element to S1 to obtain an r + 1-set T2 , different from T0 or T1 ; as
before T2 ∈/ S since S1 ⊂ T2 . We now remove some element of T2 to obtain a
new set S2 ; do so in such a way as to guarantee that S2 is different from S0
or S1 , and the same reasoning suggests that S2 must be in S. It seems clear
that we can continue this process until all of the sets of cardinality r + 1 have
been shown not to be elements of S, and all of the sets of cardinality r have
been shown to be in S. It is intuitively clear that we can do this; indeed, if S
and S ′ are any two sets of cardinality R, we may travel between the two by
 Partially Ordered Sets 173

alternately adding to S an element of S ′ that it does not contain to get a new

set T , and deleting from T any element of S that is not in S ′ . It follows that
if S contains any r-set, it contains every r-set and no r + 1-sets. If S contains
a set of cardinality r + 1, on the other hand, the same process shows that it
contains all of those sets, and none of cardinality r. 
The last result from extremal set theory that we will prove is Dilworth’s
theorem. This result deals with partitioning a poset into chains; what is the
smallest number of chains that will suffice for each element of the poset to be
in exactly one chain? Clearly, if the poset has width a, we need at least a
chains, since no chain can contain two or more elements of an antichain. The
surprising result of Dilworth is that we can always cover the poset in exactly
a chains.

THEOREM 9.6 (Dilworth’s Theorem)

If the poset X has width a, then there is a set of a disjoint chains whose
union is X.

There are several ways to prove this theorem, some more involved than
others. Our proof below is adapted from that of Marshall Hall ([50]).
Proof. We use induction on the number n of elements of X; if n = 0 or
n = 1, the result is trivial. We therefore take n to be greater than 1 and
assume the truth of the Theorem for any poset with fewer points. Suppose
we find a maximal chain C of our poset (where by maximal we mean that the
addition of any element to C makes it no longer a chain). Consider the poset
X ′ obtained from X by removing the elements of C; if X ′ has width no more
than a − 1, we apply the induction hypothesis to X ′ to cover it in a − 1 chains,
replace the chain C, and we are done. Therefore, we must assume that C does
not include any point from some antichain Y = {y1 , y2 , . . . ya }. Define the
down-set generated by Y to be Y − = {x ∈ X : x  yi for some 1 ≤ i ≤ a};
the set of all elements of X that are “smaller” than some element of Y .
Similarly, Y + = {x ∈ X : yi  x for some 1 ≤ i ≤ a} is the up-set generated
by Y . Notice that if there is any element of X not contained in Y − ∪ Y + , we
could add this element to Y to make a larger antichain; as Y was a maximal
antichain, this is impossible, so Y − ∪ Y + = X, and Y − ∩ Y + = Y .
At this point, the idea is to say that Y − is a poset that is smaller than X,
and has width a, so by the induction hypothesis it can be partitioned into a
chains; and Y + is a poset that is smaller than X, so it may be partitioned
similarly. Now, we “glue” together the chains in Y + to those in Y − by the
element of Y that they have in common; so the chain in Y + for which yi is the
least element will join to the chain of Y − for which yi is its greatest element,
for every 1 ≤ i ≤ a. How can this fail?
It can fail rather easily, in fact. For instance, Y + might be the entire poset,
because Y could be the set of all minimal elements, or of all maximal elements.
In this case the induction hypothesis would not apply. (The largest antichain
174 Introduction to Combinatorics

consisting of minimal elements must consist of all the minimal elements, and
similarly for the maximal elements; see the Exercises.) However, if Y consists
neither of all maximal elements nor all minimal elements, the proof is finished.
So assume that Y consists entirely of maximal elements (if Y consists of
minimal elements, we will turn the poset upside-down, partition it into chains,
and then turn it right side up again). Find a minimal element x1 with y1 ≺ x1 ;
that is, follow any chain from y1 downward as far as possible and give the
minimal element of that chain (which must be minimal in the poset) the label
x1 . Now we have a chain C1 ; look at the poset X \ C1 . What is the width of
this poset?
We claim now that X \ C1 has width at most a − 1. We have already
established that every largest antichain is either the set of all maximal or
of all minimal elements; C1 contains one of each, so we have removed one
element from each antichain of size a. Now the induction hypothesis applies,
and we may partition X \ C1 into a − 1 chains. Throw the chain C1 back in,
to find a partition of X into a chains, and we are done. 
Dilworth’s theorem is useful for a number of purposes, not least of which
is that it provides a fairly short proof of Philip Hall’s theorem (Theorem 1.2)
which we give here, along with the statement of the theorem:
A collection D of sets has a system of distinct representative if and only if
it never occurs that some n sets contain between them fewer than n elements.
Proof. Let D = {B1 , B2 , . . . , Bk } containing among them v elements {1, 2,
. . . , v} and suppose that any union of n of these sets contains at least n
elements. We form our poset P on the elements B1 , B2 , . . . , Bk , 1, 2, . . . , v
and we rule that i ≺ Bj if and only if i ∈ Bj ; D is an antichain, as is
the set {1, 2, . . . v}. However, these may not be the maximum antichains.
An antichain of largest cardinality might consist of some of the elements
and some of the sets. By renumbering if necessary, let the largest antichain
be 1, 2, . . . ℓ, B1 , B2 , . . . Bf ; here we note that ℓ + f ≥ v ≥ k because this
largest antichain must be at least as big as the antichain D and the antichain
{1, 2, . . . , v}. Because this is an antichain, no element below ℓ + 1 may belong
to any Bi for i ≤ k, so we have B1 ∪ B2 ∪ . . . ∪ Bf ⊆ {ℓ + 1, ℓ + 2, . . . , v}. By
our hypothesis, this union contains at least f elements, so we have f ≤ v − ℓ.
Use Dilworth’s theorem to partition our poset into ℓ + f chains. We can
again renumber our sets and elements so that the chains look like

{B1 , 1}, {B2, 2}, . . . {Br , r}, r + 1, . . . , v, Br+1 , . . . Bk .

Now we see that v + k − r = ℓ + f , because we have v + k total objects

in the poset; there are ℓ + f chains, of which r are length-2 chains, v − r are
single elements, and k − r are sets from D. Thus r = k + (v − ℓ − f ). Since
f ≤ v − ℓ, we get v − ℓ + f ≥ 0, and so r ≥ k. But r (the number of sets Bi
used in length-two chains) cannot be greater than k (the total number of sets
 Partially Ordered Sets 175

Bi ); thus v = k. That means that the set {1, 2, . . . , k} is a system of distinct

representatives for D. 

Exercises 9A
1. Find the Hasse diagram for the poset D36 .
2. Find the Hasse diagram for the poset of the multisubsets of the multiset
{a, a, b, b}.
3. Prove that meet and join commute (Theorem 9.2(2)).
4. Prove the idempotent laws for meet and join (Theorem 9.2(3)).
5. Prove that a ∧ b = a if and only if a  b.
6. Illustrate Theorem 9.4 by assigning sets to each element of the poset of
Figure 9.4 so that it is ordered by inclusion.
7. Show that the largest antichain consisting only of minimal elements
must consist of all the minimal elements.
8. Find all minimal elements of the poset of Figure 9.8.
9. Find the width of the poset of Figure 9.8.

All All

All All All All

All All All

All All

FIGURE 9.8: Poset for exercises and problems.

176 Introduction to Combinatorics

Exercises 9B
1. Find the Hasse diagram for the poset D60 .
2. Find the Hasse diagram for the poset of the multisubsets of the multiset
{a, a, b, c}.
3. Prove the associative law for the join (Theorem 9.2(1)).
4. Prove the absorption laws for meet and join (Theorem 9.2(4)).
5. Prove that a ∨ b = b if and only if a  b.
6. Illustrate Theorem 9.4 by assigning sets to each element of the poset of
Figure 9.5 so that it is ordered by inclusion.
7. Show that the largest antichain consisting only of maximal elements
must consist of all the maximal elements.
8. Find all maximal elements of the poset of Figure 9.8.
9. Find the height of the poset of Figure 9.8.

Problems

Problem 1: Show that a finite poset must have a minimal element.

Problem 2: Show that no poset has a Hasse diagram that contains a K3 .
Problem 3: Let n be the product of four distinct prime numbers; find an
explicit isomorphism between Dn and B4 .
Problem 4: Find an isomorphism between Bn × Bm and Bn+m ; derive as a
corollary the result that each subset lattice Bn is isomorphic to a product of
copies of B1 .
Problem 5: Prove that the union of two down-sets is a down-set; prove that
the intersection of down-sets is a down-set.
Problem 6: Find all down-sets of the poset of Figure 9.5 and find the poset
of these sets ordered by inclusion.
Problem 7: Find all largest antichains of the poset of Figure 9.8.
Chapter 10
Introduction to Graph Theory

10.1 Degrees
Recall the definitions of graphs from Section 1.1.4. The degree of the vertex
x was defined to be the number of edges that have x as an endpoint. We write
degG (x) for the degree of x in graph G, or simply deg(x) if only one graph is
being discussed. A graph is regular if all vertices have the same degree, and in
particular a regular graph of degree 3 is called cubic. We denote the smallest
of all degrees of vertices of G by δ(G), and the largest by ∆(G).
The degree sequence of a graph consists of all the degrees of the vertices.
For example, degree sequence (4, 2, 2, 1, 1, 0) corresponds to the graph
0 2 1

2 4 1
(Degrees are shown next to the vertices.) We normally write such a sequence
in non-ascending order.
It is interesting to ask whether a given sequence is the degree sequence of
any graph. If so, the sequence is called graphical ; a graphical sequence is called
valid if there is a connected graph with those degrees. Some sequences are
easily seen not to be graphical. Theorem 1.1 says that the sum of degrees in
any graph must be even, so the sum of the elements of any graphical sequence
must be even. Moreover, if a graph has d + 1 vertices with degree greater
than zero (non-isolated vertices), no vertex can have degree greater than d:
(3, 1, 0, 0) is obviously not graphical. (This last argument does not apply to
multigraphs.)
Example 10.1: Which of the following sequences are graphical: (3, 3, 1, 1, 0),
(3, 3, 2, 1, 1), (3, 3, 2, 1, 0)?
Suppose a graph has five vertices and two of them have degree 3. If those
two vertices are not adjacent, each must be joined to all three of the remaining
vertices (See the left-hand graph in Figure 10.1) so these all have degree at

177
178 Introduction to Combinatorics

FIGURE 10.1: Testing degree sequences.

least 2. Therefore the two vertices of degree 3 are adjacent. There are only
two possibilities: either both are adjacent to the same two vertices out of the
remaining three, or not. In the former case, both of those vertices have degree
at least 2 (it may be greater, if they are adjacent to each other or to the final
vertex), and this is not possible for any of the degree sequences given. So the
graph must look like the right-hand graph in the figure, possibly with other
edges between the lower vertices. This rules out the first and third sequences,
so they are not graphical; however, the middle sequence is graphical, and is
realized by the right-hand graph. 
Of course, we know that the third sequence is not graphical, because the
sum of its elements is odd.
One way to construct a new graph from an old one is to add one vertex
and join it to some of the other vertices. The following theorem of Havel [53]
and Hakimi [49] essentially shows that this process can be reversed.

THEOREM 10.1 (The Havel–Hakimi Theorem)

If v ≥ 2 and S = (d0 , d1 , . . . , dv−1 ) is a sequence of integers satisfying

d0 ≥ d1 ≥ . . . ≥ dk ≥ dk+1 ≥ . . . ≥ dv ≥ 0,

then S is graphical if and only if T is graphical,

T = (d1 − 1, d2 − 1, . . . , dd0 − 1, dd0 +1 , . . . , dv−1 ).

Proof. Suppose you have a graph whose degree sequence is T . You can
construct a graph with degree sequence S by simply adding a new vertex and
joining it to the d0 vertices with degrees d1 − 1 to dd0 − 1. So the “if” part is
easy.
So let us assume S is graphical. Suppose there is a graph G whose vertices
x0 , x2 , . . . , xv−1 satisfy deg(xi ) = di for every i. For convenience we will call
x1 , x2 , . . . , xd0 the earlier vertices and xd0 +1 , . . . the later vertices.
Suppose the vertices of G can be ordered so that x0 is adjacent to each
of the earlier vertices. (Ordering would be relevant if there is more than one
vertex with degree d0 or more than one vertex with degree dd0 .) In that case
we shall say G is type I ; otherwise it is type II. If G is type I, simply delete
x0 and all the edges joining it, and you have a graph with degree-sequence T .
Now we assume G is type II. Then there must be some earlier vertices
xj that are not adjacent to x0 . Suppose there are t such vertices (t > 0),
 Introduction to Graph Theory 179

xj1 , xj2 , . . . , xjt . Let’s write

A = {xj1 , xj2 , . . . , xjt }.

So there must also be exactly t later vertices xk that are adjacent to x0 . We

write
B = {xk1 , xk2 , . . . , xkt }
for this set of vertices. Finally, we define n(G) to be the sum of degrees of
vertices adjacent to x0 .
If all the vertices in A and B had the same degree, we could simply reorder
the vertices by exchanging xji and xki for every i, and obtain an ordering
where x0 ’s neighbors are precisely the earlier vertices; but that would mean G
is type I. So there exist xj in A and xk in B such that dj > dk . There must be
a vertex that is adjacent to xj but not to xk ; say xm is one such. Construct
a new graph, G1 say, by deleting edges xm xj and x0 xk and replacing them
by edges x0 xj and xm xk . G1 has the same degree sequence as G. Moreover,
n(G1 ) = n(G) + dj − dk , which is greater than n(G).
If G1 is type I we are finished. But suppose G1 is type II. Then we could
go through the whole process again, and produce a graph G2 , and so on.
If none of the graphs G1 , G2 , . . . is type I, then the sequence n(G), n(G1 ),
n(G2 ), . . . is a strictly increasing infinite sequence of integers, so eventually
it will be greater than the sum of all degrees of vertices of G. But this is
obviously impossible. So eventually we must construct a type I graph with
the same degree sequence as G, and we can use it to construct the desired
graph. 
Given a non-decreasing sequence S = (d0 , d1 , . . . , dv−1 ) of non-negative
integers, Theorem 10.1 enables us either to construct a graph with S as its
degree sequence or to show that S is not graphical. Construct the degree
sequence T as given in the theorem. (We’ll call this reducing S.) If T is
obviously graphical, then form a graph with degree sequence T and adjoin a
new vertex to the first d0 vertices. If T is obviously not graphical, then S
is not graphical. Otherwise, reduce T , and repeat until you find a sequence
for which the answer is obvious. Some obviously graphical sequences include
an all-zero sequence, corresponding to the null or empty graph with no edges
and a sequence of 2n 1s, for a graph consisting of n disjoint edges. At any
stage, 0s correspond to isolated vertices, and do not change the answer.
Example 10.2: Is the sequence (4, 4, 2, 2, 1, 1) graphical?
When you reduce (4, 4, 2, 2, 1, 1) you get (3, 1, 1, 0, 1), or in standard form
(3, 1, 1, 1, 0). This in turn reduces to (0, 0, 0, 0), the degree sequence of a null
graph with 4 vertices. So the original is graphical. The calculation could be
represented as

(4, 4, 2, 2, 1, 1) ⇒ (3, 1, 1, 0, 1) = (3, 1, 1, 1, 0) ⇒ (0, 0, 0, 0).

The graph is constructed as follows:

180 Introduction to Combinatorics

⇒ = ⇒

Example 10.3: Is the sequence (6, 5, 4, 2, 2, 2, 1) graphical?

(6, 5, 4, 2, 2, 2, 1) ⇒ (4, 3, 1, 1, 1, 0) ⇒ (2, 0, 0, 0, 0)

which is not graphical (you cannot have one vertex of degree 2 and all the
rest isolates). 
To illustrate the remark about added 0s, you can immediately conclude
from the example that (4, 4, 2, 2, 1, 1, 0, 0) and (4, 4, 0, 2, 2, 1, 0, 1, 0) are graph-
ical but (6, 5, 4, 2, 2, 2, 1, 0) is not.

10.2 Paths and Cycles in Graphs

A lot of the terminology of graph theory comes from the application of
networks to modeling road systems. For example, the multigraph

a b c

p q

d e

could represent five camps denoted a, b, c, d, e and the walking tracks between
them; there are two tracks joining c to e, and they are labeled p and q,
and other tracks are denoted by their endpoints. (These representations are
ordered.) A sequence of vertices and edges of the form (. . . , x, xy, y, yz, z, . . .)
will be called a walk, because on the diagram it represents a trail that one
could walk among the camps.
Consider the following walks:

W1 = (a, ab, b, bd, d, da, a, ad, d, de, e);

W2 = (a, ad, d, de, e, p, c, cb, b);
W3 = (c, p, e, p, c, cb, b, bd, d, de, e, q, c).
 Introduction to Graph Theory 181

Each edge of W1 is uniquely determined by its endpoints—it does not contain

p or q—so we could simply write it as (a, b, d, a, d, e), omitting the edges. This
will always be true of walks in graphs. The walk W2 has no repeated edges.
A walk with this property is called simple. W3 returns to its starting point;
a walk that does this is called closed . We shall make the following formal
definitions:

Definition. A walk of order n (or of length n − 1) in a network is a finite

sequence
(x1 , y1 , x1 , y2 , x2 , y3 , . . . , yn−1 , xn )
of vertices x1 , x2 , . . . and edges y1 , y2 , . . . where the endpoints of yi are xi and
xi+1 . x1 and xn are the start and finish, respectively, of the walk, and we
refer to a walk from x1 to xn . The following are special types of walk:
Simple walk : A walk in which no edge is repeated;
Path: A simple walk in which no vertex is repeated;
Circuit or closed walk : A simple walk in which the start and finish are the
same;
Cycle: A simple closed walk in which x1 , x2 , . . . , xn−1 are all different. No
vertex is repeated, except that the start and finish are equal.
A cycle might also be called a simple circuit. Many mathematicians also
use the phrase “closed path,” although strictly speaking this is not possible,
as the start and finish make up a repetition.
Given a path in a graph or multigraph G, you could define a subgraph made
up of the vertices and edges in the path. This subgraph is called a path in G. If
we think of it as a graph by itself, an n-vertex path is denoted Pn and is a graph
with n vertices x1 , x2 , . . . , xn and n − 1 edges x1 x2 , x2 x3 , . . . , xn−1 xn . Simi-
larly an n-vertex cycle or n-cycle Cn is a graph with n vertices x1 , x2 , . . . , xn
and n edges x1 x2 , x2 x3 , . . . , xn−1 xn , xn x1 , and we can think about cycles as
subgraphs in other graphs. A 3-cycle is often called a triangle.
Suppose a walk W from x1 to xn , where x1 6= xn , passes through a point
more than once; say xi is the first occurrence of the point and xj the last. Then
the walk x1 , x2 , . . . , xi , xj+1 , . . . , xn is part of W —we could call it a subwalk —
and contains fewer repeats. If we continue this process, we eventually finish
with no repeats, a path. That is, any x1 -xn walk contains an x1 -xn path. So:

THEOREM 10.2 (Walks Contain Paths)

If a graph contains a walk from x1 to xn , x1 6= xn , it contains a path from
x1 to xn .

In the same way, one could argue that any closed walk contains a cycle, but
there is one exception to be made: it is possible to walk from x1 to xn and
then walk back by retracing your steps. The graph of such a walk will not
contain a cycle, unless the x1 -xn walk contained one. But any circuit must
contain a cycle.
182 Introduction to Combinatorics

We say two vertices are connected if there is a walk from one to the other.
(In view of Theorem 10.2, we could replace “walk” by “path.”) This gives us a
formal definition of connectivity: a graph is connected if and only if all its pairs
of vertices are connected. We can now define a component as the subgraph
formed by taking a maximal set of vertices with every pair connected, plus all
the edges touching them.
The distance between two vertices is the minimum length among the paths
joining them (or the length of the shortest path). The distance from x to y is
written d(x, y). For consistency we write d(x, x) = 0.
Example 10.4: Find the distance from x to each other vertex in the
following graph.

x a d

b c

There is a path of length 1 (a single edge) from x to a, and also to b. The

distance between two distinct vertices cannot be smaller than 1, so d(x, a) =
d(x, b) = 1. To find the shortest path from x to d we look at all the possible
paths; they are
(x, a, d) of length 2
(x, a, b, c, d) of length 4
(x, b, a, d) of length 3
(x, b, c, d) of length 3
and the shortest length is d(x, d) = 2. Similarly, d(x, c) = 2. 
Listing all paths is fine in small cases, but very wasteful of time in a large
graph. In 1959, the Dutch computer scientist Edsger Dijkstra [27] devised
an efficient algorithm to find the distance from a given vertex to every other
vertex in a graph.
Dijkstra’s algorithm works as follows. Suppose you want to find the distance
from vertex x to every other vertex in your graph. Trivially d(x, x) = 0, and
d(x, y) = 1 for all neighbors of x.
We now use the following argument. Suppose you want to find the distance
from vertex x to every other vertex in your graph. Say you know d(x, z) = n.
Then there is a path P = (x, x1 , x2 , . . . , xn−1 , z) from x to z. If t is any
neighbor of z, then there will be a walk of length n + 1 from x to t—simply
append t to P . This might not be a path, because t might be xn−1 (which
is one of z’s neighbors). But in that case we know a shorter path to t. So
d(x, y) is at most n + 1.
So we proceed as follows. Trivially d(x, x) = 0; label x as x0 . We know
d(x, y) = 1 for all neighbors of x. Label these neighbors x1 , x2 , . . . , xp in some
 Introduction to Graph Theory 183

order (where p = deg(x)). We say x has been processed. Now we process x1 .

We consider all the neighbors of x1 and discard all that are already labeled.
Label the remaining vertices xp+1 , xp+2 , . . .. Each is distance 2 from x.
Continue in this way. When you finish processing xi , go on to xi+1 . Discard
any neighbor of xi+1 that is already labeled, and label the remainder with the
smallest unused labels. Each of them is distance d(x, xi+1 ) from x. Go on
until all vertices are labeled.
Example 10.5: Apply the above algorithm to

a c
e
x f

b d

We start by labeling x as x0 and note that d(x, x0 ) = 0. The neighbors

of x0 are a and b. Neither is yet labeled. Label a as x1 and b as x2 . Set
d(x, a) = d(x, x1 ) = d(x, b) = d(x, x2 ) = 1. Processing x0 is completed.
Now we process x1 . Its neighbors are x and c. We skip x as it is already
labeled. We label c as x3 and set d(x, c) = d(x, x3 ) = 1 + d(x, x1 ) = 2. Next
we process x2 . The only unlabeled neighbor is d. So d is labeled x4 , and
d(x, d) = d(x, x4 ) = 1 + d(x, x2 ) = 2.
The only unlabeled neighbor of x3 is e. So e is x5 and d(x, e) = d(x, x5 ) =
1 + d(x, x3 ) = 3. x4 has no unlabeled neighbors. Finally, x5 has one unlabeled
neighbor, f . So f is x6 and d(x, f ) = d(x, x6 ) = 1 + d(x, x5 ) = 4. All vertices
are now labeled.
This could be written in a table, as shown in Figure 10.2. 
Of course, this example is more easily handled by finding all paths. But
the algorithm can be applied to much larger graphs. Writing the table is
tedious, but it illustrates how all the data could be stored on a computer.
Several implementations of this algorithm are available on the Internet. One
example is [111], but there are many others; just type the phrase “Dijkstra’s
algorithm” into your favorite search engine.

10.3 Maps and Graph Coloring

By a map we mean the kind of map you would usually find in an atlas,
a map of a continent (showing countries) or of a country (showing states).
However, we need to make one restriction. We require that you can get from
one point in a country (or state) to another without crossing a border; that
is, each country or state is connected. We’ll call such a map simple. For
example, a map of North America is not simple, because the United States
184 Introduction to Combinatorics

Vertex Neighbors
x = x0 a, b a = x1 d(x, a) = 1
d(x, x) = 0 b = x2 d(x, b) = 1
a = x1 x, c x already labeled
d(x, a) = 1 x c = x3 d(x, c) = 2
b = x2 x, c, d x already labeled
d(x, b) = 1 c already labeled
d = x4 d(x, d) = 2
c = x3 a, b, e a already labeled
d(x, c) = 2 b already labeled
e = x5 d(x, e) = 3
d = x4 b, e b already labeled
d(x, d) = 2 e already labeled
e = x5 c, d, f c already labeled
d(x, c) = 2 d already labeled
f = x6 d(x, f ) = 4

FIGURE 10.2: Tabular form of Dijkstra’s algorithm.

is not connected – to get from Alaska to the other states you need to cross
either Canada or the ocean. We say two countries or states are neighbors if
they have a border—a single common point does not count.
Given a map, we construct a graph as follows: the vertices correspond to
the countries (or states), and two vertices are adjacent when the two countries
share a border. It is always possible to draw this graph in such a way that
no two of the edges cross—the easiest way is to draw the graph on top of the
map, with each vertex inside the corresponding country. A graph that can be
drawn without edge crossings is called planar . So every map corresponds to
a planar graph. It is easy to draw a map from any given planar graph, so the
correspondence goes both ways.
Not all graphs are planar: try as you might, you cannot draw K5 without
crossings. On the other hand, some graphs are planar even though they do
not look that way. Figure 10.3 shows two representations of the same graph,
K4 . One has crossings, the other does not. Remember, we say the graph
is planar if it can be drawn in the plane without crossings, even though the
representation we are using has some crossings.
Suppose you want to print a map in an atlas. It is usual to use different
colors for the various countries. Some colors may be repeated, but the map
is easier to read if you avoid using the same color on two countries with a
common border. The corresponding problem in a graph is to label the vertices
in such a way that no two adjacent vertices receive the same label. The labels
are called colors, and a labeling is called a coloring of the graph, even though
the ideas can be applied to non-planar graphs, where the relationship to maps
does not exist. A coloring in which no two adjacent vertices receive the same
 Introduction to Graph Theory 185

x y x

z t z t

FIGURE 10.3: Two representations of K4 .

color is called proper .

Map and graph coloring has been studied since 1852 when Francis Guthrie
asked whether it was true that any map can be colored using four colors. The
first printed reference is due to Cayley [18]. A year later an incorrect ‘proof’
by Kempe appeared; its incorrectness was pointed out by Heawood 11 years
later. Another failed proof is due to Tait in 1880; a gap in the argument was
pointed out by Petersen in 1891. It was not until 1976 that Appel and Haken
[4] proved the four color theorem:

THEOREM 10.3 (Four Color Theorem)

Any planar graph can be colored with four colors.

Their proof was long, complicated and required extensive use of computers.
Later proofs include [88] and [46].
In a graph coloring, the set of all vertices receiving the same color is called
a color class. There are a number of applications of color classes outside
map printing (in fact, there is not much practical application of the four color
theorem to map printing.) For example, coloring is used in conflict resolution.
Example 10.6: A habitat zoo consists of several large habitats, each con-
taining several species. In such a zoo, you cannot put natural enemies to-
gether. Use graph coloring to design the zoo.
We form a graph whose vertices are the species to be housed in the zoo.
An edge means the two species are “enemies,” and cannot be housed in the
same habitat. Now color this graph. Two species can be put into the same
habitat if they receive the same color. 
In this example we would like the number of habitats to be as small as
possible. So we want the number of colors to be as small as possible. We
define the chromatic number χ(G) of a graph G to be the smallest number of
colors possible for a proper coloring.
Another application is school timetabling. In this case the classes are rep-
resented by vertices, and an edge means that some student will be taking both
classes. Vertices in the same color class can be scheduled simultaneously.
186 Introduction to Combinatorics

The complete graph Kn has n vertices, and every pair is joined, so χ(Kn ) =
n. The cycle Cn has chromatic number 2 when n is even and chromatic number
3 when n is odd. The complete bipartite graph Km,n , which contains no odd
cycles, has chromatic number 2.
Obviously, if a graph is properly colored, every subgraph will also be prop-
erly colored. So, if G contains a subgraph Kn , its chromatic number will be
at least n. This gives us a tool to find chromatic numbers of small graphs:
find the largest complete graph that is contained in your graph. Color all
its vertices differently. Then try to color the remaining vertices using those
colors. Very often this is easy to do. If there is no complete graph, is there
an odd-length cycle? If so, at least three colors will be needed.
Another tool is:

THEOREM 10.4 (Brooks’ Theorem) [12]

Suppose the graph G is not a complete graph, nor a cycle of odd length. Then
χ(G) ≤ ∆(G).

The proof is a little difficult, and can be found in graph theory texts (for
example, [122, p. 980]).
We now present an easy algorithm that will always color a graph G in no
more than ∆(G)+1 colors—not the minimum, but close. Suppose the vertices
of G are x1 , x2 , . . . and the colors are c1 , c2 , . . . , c∆(G)+1 . Apply color c1 to
x1 . Then go through the vertices in order; when xi is to be colored, use cj ,
where j is the smallest positive integer such that cj has not yet been applied
to any vertex adjacent to xi .
Example 10.7: Apply the above algorithm to
x2 x3
x6
x1 x4
x5
The maximum degree is 3. We apply c1 to x1 . As x2 ∼ x1 we cannot use
c1 , and x2 receives c2 . The neighbors of x3 are x2 and x4 , so color c2 is the
only one barred (vertex x4 has not yet received any color); c1 is still free, so
x3 gets c1 . Similarly x4 gets c2 . Now x5 is adjacent to x1 (color c1 ) and x4
(color c2 ) so it receives c3 . Finally x6 is adjacent to all three colors, so it gets
c4 . 
We could represent this coloring by
{x1 , x3 } ← c1 , {x2 , x4 } ← c2 , {x5 } ← c3 , {x6 } ← c4 .
As we said, this method does not necessarily return a minimal coloring; in
fact, the given graph has chromatic number 3, realized by
{x1 , x4 } ← c1 , {x2 , x5 } ← c2 , {x3 , x6 } ← c3 .
 Introduction to Graph Theory 187

Finally, there is an interesting example in the April 2009 issue of Math

Horizons. The cover shows a four-colored map of all United States counties,
illustrating an article by Mark McClure and Stan Wagon [73].

Exercises 10A
1. Prove that in any graph there must be two vertices with the same degree.
Show that this is not true for multigraphs, by exhibiting a multigraph
with degree sequence (3, 2, 1, 0).
2. Which of the following sequences are graphical? If the sequence is graph-
ical, draw a graph that realizes it.
(i) (4, 2, 2, 2, 1, 1) (ii) (5, 4, 3, 2, 2, 2)
(iii) (3, 3, 3, 3, 2, 2) (iv) (5, 3, 3, 3, 1, 1, 1, 0)
(v) (6, 5, 2, 2, 1, 1, 1, 0) (vi) (6, 5, 4, 3, 3, 3, 1, 1, 0)

3. Find all paths from x to f in the graph

x a b c

d e f

4. Find the distance from x to each other vertex in the graph of the pre-
ceding exercise.
5. Find the distance from b to each other vertex in the following graphs.
b
(i) a b c d e (ii) a c d
k
w x y
v w x y z z

6. The graph G contains two paths from vertex x to vertex y, namely

(x, a, . . . , y) and (x, b, . . . , y). Assuming vertices a and b are distinct,
prove that G contains a cycle that passes through x. Show that the
assumption “a and b are distinct” is necessary.
7. G consists of two subgraphs, G1 and G2 , that share one common vertex.
What is the relation between χ(G), χ(G1 ) and χ(G2 )?
8. Find the chromatic numbers of the following graphs.
188 Introduction to Combinatorics

(i) (ii)

(iii) (iv)

(v) (vi)

9. By subdividing an edge in a graph we mean taking an edge xy and

replacing it with a new vertex u and new edges xu and uy. Suppose G
is a bipartite graph, and suppose H is produced from G by subdividing
one or more edges.
(i) Show that H must have chromatic number no greater than 3.
(ii) Either prove that H must have degree 3, or produce a counterex-
ample.
10. Suppose x is a vertex of odd degree in a graph G. Show that G must
contain a walk that starts at x and ends at some other vertex of odd
degree.
 Introduction to Graph Theory 189

Exercises 10B
1. Show that no cubic graph has an odd number of vertices.
2. A bipartite graph has 2n vertices. What is the maximum number of
edges it can have?
3. Which of the following sequences are graphical? If the sequence is graph-
ical, draw a graph that realizes it.
(i) (4, 4, 3, 3, 3, 1, 1, 0) (iv) (4, 3, 3, 2, 2, 2, 2)
(ii) (3, 3, 2, 2, 1, 1) (v) (7, 4, 2, 2, 2, 2, 1, 1, 1, 0)
(iii) (6, 5, 4, 2, 2, 2, 1, 0) (vi) (8, 5, 4, 3, 3, 3, 1, 1, 1, 1, 0, 0)

4. Show that there is no graph with degree sequence (4, 4, 4, 2, 2), but there
is a multigraph with this degree sequence.
5. Find all paths from a to d in the graph
a b c d

e f g h

6. Find the distance from c to each other vertex in the graph of the pre-
ceding exercise.
7. Find the distance from b to each other vertex in the following graphs.

(i) d (ii) b e
g
b
a c f h
a e h
c
f i
d g

8. The square G2 of a graph G has the same vertex-set as G; x and y are

adjacent if and only if d(x, y) = 1 or 2 in G. Find the squares of:
(i) K1,n ; (ii) P5 ; (iii) C5 .

9. Suppose x, y and z are three vertices in a connected graph. Show that

d(x, z) ≤ d(x, y) + d(y, z).

190 Introduction to Combinatorics

10. The graph G contains at least two vertices. One vertex has degree 1,
and every other vertex has degree greater than 1. Show that G contains
a cycle.

11. Prove that any graph with v vertices and at least 1 + 21 (v − 1)(v − 2)
edges must be connected.

12. Find the chromatic numbers of the following graphs.

(i) (ii)

(iii) (iv)

(v) (vi)

Problems

Problem 1: Show that it is possible to find a graph with degree sequence

(5, 5, 5, 3, 2, 2, 1, 1), but it is not possible to find such a graph in which the two
vertices of degree 1 are adjacent.
Problem 2: Graph G has v vertices. What is the smallest number of edges
that G can have, as a function of v and δ(G)?
Problem 3: A graph that is isomorphic to its complement is called self-
complementary.
(i) Prove that the path P4 and the cycle C5 are self-complementary.
(ii) Show that a self-complementary graph can contain no isolated vertices.
(iii) Find all self-complementary graphs on five vertices.
 Introduction to Graph Theory 191

(iv) Prove that the number of vertices of a self-complementary graph must

be congruent to 0 or 1 modulo 4.
Problem 4: Show how to construct a cubic graph with v vertices whenever
v is even, v ≥ 4.
Problem 5: Is Theorem 10.1 true for multigraphs?
Problem 6: [31] Graph G has v vertices and degree sequence (d1 , d2 , . . . , dv ),
written in non-ascending order. Prove that, if 0 < r < v,
r
X v
X
di ≤ r(r − 1) + min{r, di }.
i=1 i=r+1

Problem 7: The square of a graph was defined in Exercise 10B.8. A graph

is called a perfect square if it is the square of some graph. Find all perfect
squares on four vertices.
Problem 8: If x is a vertex in a connected graph G, the eccentricity of
x, ǫ(x), is defined to be the largest value of d(x, y), where y ranges over all
the vertices in G. The radius R of G is the smallest value of ǫ(x), and the
diameter D is the largest value of ǫ(x), for all vertices x. Prove that

R ≤ D ≤ 2R.

Problem 9: For what values of n is Kn a subgraph of Kn,n ?

Chapter 11
Further Graph Theory

11.1 Euler Walks and Circuits

It is reasonable to say that graph theory was founded by the Swiss math-
ematician and physicist Leonhard Euler (see Appendix C). As we mentioned
in Section 1.1.1, Euler addressed the St. Petersburg Academy in 1735 on the
problem of the Königsberg bridges. The Prussian city of Königsberg was
set on both sides of the Pregel River. It included a large island called The
Kneiphof, and the river branched to the east of it. So there were four main
land masses—we shall call them the four parts of the city, and label them
A, B, C, D—connected to each other by seven bridges. For convenience we
repeat our rough map of the city, on the left in Figure 11.1:

A A

B
B D
D

C
C

FIGURE 11.1: The original Königsberg bridges.

The problem was to find a walk through the city that would cross each
bridge once and only once, and visit all four parts. The only way to go
between the different pieces of land was to cross the bridges. In modern
terminology (terminology that would not have existed without Euler’s work!)
he observed that the city could be represented by a multigraph, with the parts
represented by vertices and the bridges represented by edges, as shown in the
second part of Figure 11.1. A solution to the problem would be a simple walk
in the multigraph that included every edge; even better would be a closed
simple walk, so that a person could start and finish from home.
Euler proved that no such walk is possible. But he did much more. He
essentially invented graph theory and showed how it could be used to represent

193
194 Introduction to Combinatorics

any set of islands and bridges, or indeed any set of towns and roads joining
them, and concocted an algorithm for traversability problems.
In graph terminology, we define an Euler walk to be a simple walk that
contains every edge of the network, and an Euler circuit to be a closed Euler
walk. A network that has an Euler circuit is called eulerian.
Suppose G has an Euler walk, and suppose X is a vertex of G. (So there
are d(X) bridges leading to the area X in the corresponding map.) Every
time the walk passes through X, you traverse two of these bridges. By the
end of the walk, this accounts for an even number of edges. The only way X
can have odd degree is if it is either the start or the finish of the walk. A walk
can have at most one start and one finish, so a network cannot have an Euler
walk if it has more than two odd vertices. Similarly, an eulerian graph must
have all its vertices even.
Euler showed [32] that these conditions are sufficient, if we add the obvious
proviso that the graph must be connected:

THEOREM 11.1 (Euler Circuit Characterization)

If a connected network has no odd vertices, then it has an Euler circuit. If a
connected network has two odd vertices, then it has an Euler walk whose start
and finish are those odd vertices.

Proof. We prove the theorem by describing an algorithm to construct an

Euler walk, and then show that it always works.
Suppose a network has no odd vertex. Select any vertex x, and select any
edge incident with x. Go along this edge to its other endpoint, say y. Delete
the edge from the network. Then choose any other edge incident with y and
go along this edge to its other endpoint; delete the edge you just travelled.
Continue in this way: in general, on arriving at a vertex, delete the edge just
used, select one of the remaining edges incident with it that has not yet been
used, and go along the edge to its other endpoint. At the moment when this
walk has led into the vertex z, where z is not x, an odd number of edges
touching z have been used up and then deleted (the last edge to be followed,
and an even number previously—two each time z was traversed). Since z is
an even vertex (as are all the vertices), there is at least one edge incident
with it that is still available. Continue the walk—eventually you will run out
of edges (remember, we assume that all networks are finite)—until you can
go no further. The only possible way this can happen is that the last vertex
you reach must be x—that is, you have constructed a closed walk. It will
necessarily be a simple walk and it must contain every edge incident with x.
If all the edges have been used, we have a closed circuit. Otherwise, the
remaining edges—those you haven’t yet deleted—make up a multigraph with
every vertex even. Moreover, it must have at least one vertex in common
with the walk, or else the original network would not have been connected,
and that case was not allowed. Select such a vertex b, and find a closed simple
 Further Graph Theory 195

walk starting from b in the “remainder,” just as you did before. Then unite
the two walks as follows: at one place where the original walk contained c,
insert the new walk. For example, if the two walks are

x, y, . . . , a, b, c, . . . , x

and
b, s, . . . , t, b,
then the resulting walk will be

x, y, . . . , a, b, s, . . . , t, b, c, . . . , x.

(There may be more than one possible answer, if b occurred more than once
in the first walk. Any of the possibilities may be chosen.) The new walk is
a closed simple walk in the original network. Repeat the process of deletion,
this time deleting the newly formed walk. Continue in this way. Each walk
contains more edges than the preceding one, so the process cannot go on
indefinitely. It must stop: this will only happen when one of the walks contains
all edges of the original network, and that walk is an Euler walk.
The case where there are two odd vertices, say p and q, and every other
vertex is even, now follows very easily from the “all vertices even” case. Form
a new network by adding another edge pq to the original. The new network
has every vertex even, so you can find a closed Euler walk in it. Moreover, it
follows from the proof that you can choose p as the first vertex and the new
edge pq as the first edge. Then delete this first edge; the result is an Euler
walk from q to p. 

Example 11.1: Find an Euler circuit in the graph

A B C D

E F G H

I J K L

First we check that the graph is connected and every degree is even. Then
select any starting vertex–A say–and trace a circuit at random, with the rules:
• You cannot use an edge twice;
• If you reach a vertex, you must continue if there is any edge touching it
that has not been used.
In the example, say we choose to go

ABF CGKF EA.

196 Introduction to Combinatorics

(We omit the parentheses, and commas between the vertices, when no con-
fusion will arise.) We then have to stop because both edges touching A have
been used. Let’s call this the first circuit. Deleting its edges, we get
A B C D

E F G H

I J K L

We now start again from any one of the vertices that were touched in the
first circuit but still have edges in the above diagram. A good choice is F .
One possible circuit is
F JIEGDHGF.
Again you can go no further, E’s edges are used up. This is the second circuit.
Now combine the first and second circuits as follows: go along the first
circuit until you come to the first occurrence of F ; then go round the whole
of the second circuit; then finish the first circuit. In other words, replace the
first F in ABF CGKF EA by F JIEGDHGF : the result is
AB F JIEGDHGF CGKF EA

(where the space shows where we changed circuits; there isn’t really a gap
there).
All the edges in the second circuit are deleted, leaving
A B C D

E F G H

I J K L

and you look for another circuit in the remainder. In this case, starting at B,
you find the circuit (actually the whole thing)
BCGJKLCB.

Feed this third circuit into the big circuit as before:

A BCGJKLCB F JIEGDHGF CGKF EA.
 Further Graph Theory 197

We have finished; we have an Euler circuit. 

By the way, Königsberg is now called Kaliningrad, and is in Russia. Two

of the seven original bridges were destroyed by bombs during the Second
World War. Two others were later demolished but were replaced by a modern
highway incorporating two new bridges. One other was rebuilt, but two of
the original bridges remain from Euler’s time. The new current configuration
and its network are shown in Figure 11.2. So the town now has an Euler walk!
(But not a closed one.)

A A

B
B D
D

C
C

FIGURE 11.2: The Königsberg bridges today.

Euler walks and their generalizations are used in what are called route
inspection problems. For example, suppose a highway inspector needs to drive
along and check all the main roads in an area. The road map can be translated
into a network G by interpreting all the intersections as vertices and the roads
as edges. Since the inspector does not want to cover the roads more than once
on each trip, and since he must start and finish from the city’s vehicle facility,
the most efficient route will correspond to an Euler circuit in G.
What if there is no such circuit? Then some of the roads must be traversed
twice. This is not allowed in an Euler circuit, so we add roads: for example, if
you need to travel the road from A to B twice, you represent that road by two
edges in the corresponding network. This process of adding duplicate edges
to make the network eulerian is called eulerizing the network. The minimum
number of edges needed to eulerize G is called the eulerization number of G,
and an efficient eulerization is one that uses that number.

Example 11.2: Find an efficient eulerization of the graph in Figure 11.3.

There are two odd vertices, namely u and y. You would like to add an
edge from u to y, but this is not allowed: you must duplicate an edge that
is already present. So you try to find a solution that duplicates two edges.
Obviously one edge would need to touch u, and the other to touch y. There
are three possible solutions; one is to duplicate uv and vy. 

As we said, you would prefer to add an edge uy. There are some situations
198 Introduction to Combinatorics

u w y

FIGURE 11.3: Find an efficient eulerization.

where this would be possible. But for many examples—like snow-plowing—

you have to use the existing roads, so duplication of edges is required.
In many cases there is a cost associated with an edge. For example, the cost
of plowing a street might be proportional to the length of the street or to the
time it takes to traverse it. In that case you could assign a cost to each edge,
and define the cost of an eulerization to be the sum of the costs of the new
edges added. The most desirable eulerization would then be the cheapest one
possible. This is the form in which the problem was first studied, by Kwan
Mei-Ko [74]. Several algorithms were described in [29]; see also [68]. Ko
wrote in terms of mail delivery. The author was Chinese, the original paper
was written in Chinese, it appeared in a Chinese journal and the translation
appeared in the journal Chinese Mathematics, so Alan J. Goldman of the
U.S. National Bureau of Standards suggested the name “Chinese Postman
problem,” and the name stuck.
As a final note, we observe that the whole study of Euler circuits can also
be carried out for directed networks. This will make sense when divided roads
or one-way streets need to be taken into account.

11.2 Application of Euler Circuits to Mazes

Apart from such obvious applications as mail delivery and highway inspec-
tion, there are several other applications of Euler circuits and eulerization. As
an example we shall explore one of them.
A maze can be modeled by a network: each intersection of paths, where
you have a choice of ways to go, corresponds to a vertex, as does the end of
every “blind alley,” and edges represent the direct paths between these points.
Figure 11.4 shows a small section of a maze and the corresponding network.
The network of a maze contains one or two special vertices, representing
the entrance and the exit (which may be in the same place or not). If the
maze can be solved, there must be a walk from start to finish. Finding such a
walk is easy enough if you have a map of the maze, but these are not always
available. In fact, your aim may be to construct such a map—for example,
 Further Graph Theory 199

FIGURE 11.4: Graphical representation of a maze.

you may be exploring a newly discovered system of caves. So we shall present

an algorithm for solving an unknown maze. We assume that you have some
way of marking parts of the maze, such as leaving markers or drawing chalk
arrows. Every edge will be marked immediately after it is traversed, and every
vertex will be marked just after you first leave it; we’ll refer to the vertices
and edges as marked after this is done and unmarked otherwise.
Proceed as follows. Start at the entrance. Select any available edge, call it
the ingress, and take it. Do not take the ingress again while any other edge is
available. Apart from that, when you arrive at a vertex, select the next edge
using the following rules:

(i) From a new vertex, any edge may be taken;

(ii) From an old vertex, if the edge just traversed was new, then turn around
and retake it;

(iii) Never use any edge three times.

This walk can be explained as follows: it is an Euler circuit in the network

formed by doubling each edge in the original. It will visit every edge. If your
aim is to get through the maze, you leave the first time you encounter the
exit. If you want to construct a map, complete the whole circuit.

Example 11.3: In the following diagram, is it possible to draw an un-

broken line that goes through each line segment exactly once (you are not
allowed to pass through the corners)?

If we treat this as a map with five countries, we get the map shown on
the left in Figure 11.5. However, we need to look at the outside boundaries,
corresponding to the ocean coast. This is handled by adding a sixth “country”
(or ocean) that encloses the other five. So we put in another vertex, connected
to all five, and look at the corresponding graph, shown on the right in the
figure. The question now asks, is there an Euler walk in this new graph? The
graph has two odd vertices, so the answer is “yes.” 
200 Introduction to Combinatorics

FIGURE 11.5: Two graphs from the map.

11.3 Hamilton Cycles

In 1857 Thomas Kirkman posed the following problem in a paper that
he submitted to the Royal Society: Given a graph, does there exist a cycle
passing through every vertex? Two years later Sir William Rowan Hamilton,
who was Astronomer Royal of Ireland, invented a puzzle called the icosian
game. The board is a regular graph with 20 vertices and 30 edges. (The
graph is a representation of a dodecahedron, with the vertices representing
the corners.) The object of the game was to visit every vertex using each
edge at most once. This was essentially an example of Kirkman’s problem,
because the path generated in visiting the vertices would be the sort of cycle
Kirkman was discussing, but historically a cycle that contains every vertex is
called a Hamilton cycle. (Kirkman’s name is better known in the history of
block designs.) The name Hamilton path is used for the case of a path that
contains every vertex but is not closed.
A graph is called hamiltonian if it has a Hamilton cycle. The first question
is whether or not a given graph is hamiltonian. Some cases are easy: any
complete graph larger than K2 has a Hamilton cycle, and in fact any permu-
tation of the vertices gives rise to one of them. In any cycle in Km,n , alternate
vertices must be in different parts of the vertex-set, so Km,n is hamiltonian if
and only if m = n.
Example 11.4: Find Hamilton cycles in the following graphs, if possible.

(i) u v (ii) u v

w x y z w x y z

In (i), the cycle wxyzvuw is obviously Hamilton. In (ii), any Hamilton

path would necessarily include edge xy. If it is traversed from x to y, then
the first vertex in the path must be one of {u, w, x} and the last must be one
of {v, y, z}—you cannot get back across xy. The edge xy is called a bridge
(we’ll define this idea formally in Section 11.4, below), and its existence shows
that no Hamilton cycle exists. 
The following necessary condition was found by Ore in 1960:
 Further Graph Theory 201

THEOREM 11.2 (Ore’s Condition) [83]

Suppose G is a graph with v vertices, v ≥ 3, and for every pair of nonadjacent
vertices x and y,
deg(x) + deg(y) ≥ v
then G is hamiltonian.

Proof. Suppose the theorem is not true. We can assume that all pairs of
nonadjacent vertices satisfy the given degree condition, and that if p and q
are nonadjacent vertices then the graph formed by adding edge pq, denoted
G+ pq, will be hamiltonian (if not, then join pq and use the new graph instead
of G). We would say G is maximal for the condition.
As v ≥ 3, Kv is hamiltonian, so G is not complete. Select two vertices, x and
y, that are not adjacent. Then G is not hamiltonian, but G∗ = G+xy is, so xy
must lie in every Hamilton cycle of G∗ . Trace such a cycle of the form (yx . . . y)
and delete the first edge; the result is a Hamilton path (x = x1 x2 . . . xv = y)
in G. If x ∼ xi for some i with 2 < i < p, then y 6∼ xi−1 , or else
(xxi xi+1 . . . yxi−1 xi−2 . . . x2 x)
would be a Hamilton cycle in G, which is impossible.
Say the neighbors of x are x2 , xj2 , xj3 , . . . , xjd , where 2 < j2 < j3 < . . . < jd
and d = deg(x). Then y is not adjacent to any of xj2 −1 , xj3 −1 , . . . , xjd −1 , or to
x. So there are d vertices not adjacent to y, and therefore deg(y) ≤ (v−1)−d =
v − d − 1. Therefore
deg(x) + deg(y)d + deg(y) ≤ d + v − d − 1 < v,
a contradiction. 
This gives a sufficient condition, but it is not necessary; many hamiltonian
graphs do not satisfy the theorem. In fact, we know of no simple way to tell
whether a graph is hamiltonian. At first sight, the Euler circuit and Hamilton
cycle problems look similar, but there is a simple test to see whether a graph
is eulerian, and a straightforward algorithm to construct an Euler cycle when
it exists.
Often there are costs associated with the edges, such as airfares or travel
times. In those cases, we define the cost of a Hamilton cycle to be the sum of
the costs of the edges. These ideas were first studied in the 1930s by various
mathematicians. Whitney pointed out that the problem of finding the cheap-
est Hamilton circuit could be motivated as follows: A traveling salesman needs
to stop at several cities (ending back at home); find the shortest/cheapest
route that can be used. He referred to it as the Traveling Salesman Problem
or TSP. (For more history, see [98].) It is often sufficient to talk about the
TSP for complete graphs.
The obvious approach to the TSP is to list all possible Hamilton cycles and
work out the cost of each. However, this can be an extremely long computa-
tion. If there are n cities, there are n! different ways to order them. Clearly
202 Introduction to Combinatorics

(a, b, c, d) costs the same as (b, c, d, a), so you only need to check one in n
cycles, and the cycle gives the same result if you travel it in reverse order,
but this leaves (n − 1)!/2 essentially different cycles. Computer time will be
prohibitive, even for small cases. For example, if you process one hundred
routes per second, it will take more than 55 hours to examine every Hamilton
cycle in K12 . If you consider a salesman who needs to visit 30 or 40 cities,
the problem is far too large.
So we consider some algorithms that give an approximate solution. We’ll
look at two, the nearest neighbor algorithm and the sorted edges algorithm.
We usually assume our graph is a copy of Kn and there is a distance or cost
c(x, y) associated with each edge xy (the airfare, or distance, or whatever).
The algorithms can be applied to incomplete graphs, but there is no guarantee
that any cycle will be produced.
To explain the ideas behind the algorithms, suppose an explorer wants to
visit all the villages in a newly discovered area, and wants to do it as efficiently
as possible. The “costs” are the distances between settlements.
The sorted edges algorithm models the way she might proceed if she has a
map of the whole area, possibly produced by an aerial survey, with distances
marked. She might hope to find an efficient route by taking the shortest
feasible connections.
Sorted edges algorithm: Initially sort the edges from shortest (cheap-
est) to longest and list them in order. Select the shortest edge. At each stage
delete from the list the edge you just used, and any edges that would form a
circuit or if added to those already chosen, or would be the third edge cho-
sen that was adjacent to some vertex (we’ll call this a threeway), then select
the shortest remaining edge. In this case the result does not depend on the
starting point.
In the nearest neighbor algorithm, the explorer starts at one of the villages.
She asks the locals, “what is the nearest village to here?” If possible she goes
to that village next.
Nearest neighbor algorithm: First choose any vertex x1 as a starting
point. Find the vertex x2 for which c(x1 , x2 ) is smallest. Continue as follows:
at each stage go to the nearest neighbor, except that you never close off a cycle
(until the end) or backtrack. In other words, after xi is chosen (i < n), xi+1
will be the vertex not in {x1 , x2 , . . . , xi } for which c(xi , xi+1 ) is minimum.
Finally, xn is joined to x1 . Notice that the result of this algorithm depends
on the starting point.
In either example, it may be that two candidate edges have the same cost,
and either may be chosen. However, if the algorithms are to be implemented
on a computer, some sort of choice must be made. For your convenience in
working problems, let’s assume we always take the edge that occurs earlier in
alphabetical order (if edges bc and bd have the same length, choose bc).
Example 11.5: Figure 11.6 shows a copy of K4 with costs on the edges.
Apply the two algorithms to this graph.
 Further Graph Theory 203

55
75 B 35
60 30
C 45 D

FIGURE 11.6: A labeling of K4 .

In the nearest neighbor algorithm starting at A, the first stop is D, because

c(AD) < c(AB) < c(AC) (that is, 35 < 55 < 75). Next is B (c(DB) < c(DC),
AD is not allowed). Finally, we go to C. The cycle is ADBCA, or equivalently
ACBDA, cost 200. Nearest neighbor from B yields BDACB (the same cycle),
from C, CDBAC (cost 205), and from D, DBACD (same cycle as from C).
In sorted edges we first make the list

edge cost
BD 30
AD 35
CD 45
AB 55
BC 60
AC 75

Then we proceed: select BD; AD is usable, so select it; CD is not acceptable

(threeway at D); AB is not acceptable (triangle ABD); BC and AC are
acceptable. The cycle is ACBDA, cost 200. 

Figure 11.7 shows all three Hamilton cycles in the copy of K4 , with their
costs. Note that the cheapest route, (A, B, C, D), cost 195, was not found by
either method. These algorithms are not guaranteed to give a perfect result.
But they give a good result in a reasonable time. They are more effective
for larger graphs. The examples in this book are all relatively small, so that
you can follow them, but in the real world you would not apply the nearest
neighbor or sorted edges algorithms to graphs like these.
As we said, either algorithm can be applied to graphs that are not complete,
but neither is guaranteed to give a result in that case. If the graph is close to
complete, the algorithms usually work, but not always.

Example 11.6: Find the routes generated by the sorted edges algorithm
and by the the nearest neighbor algorithm starting at each vertex, if possible,
in
204 Introduction to Combinatorics
A A A

55 55 55
75 B 35 75 B 35 75 B 35
60 30 60 30 60 30
C 45 D C 45 D C 45 D
ABCD (195) ACBD (200) ABDC (205)

FIGURE 11.7: All Hamilton cycles of the K4 .

FIGURE 11.8: Some trees.

u v w
4 9
8 7
5 9 8
9 9
4 6
x y z

The sorted edges algorithm gives the edges (in order) uv, xy, ux, yz, wz,
vw, forming a cycle uvwzyx of total cost 36 (which happens to be the cheapest
cycle possible). Now for nearest neighbor: Starting from u, we choose uv (the
cheapest edge from u), then vz (vu would be a repeat), then zy, then yx.
No further edge is available: xu makes a 5-cycle and xv makes a 4-cycle.
Similarly, starting from v we get the path vuxyz, and from x we get xyzvu,
neither of which can be completed. If we start at w, we get wzyxuvw, which
is the same cycle as we found by sorted edges, and from initial point z we get
the same answer. If we start from y, we get yxuvzw, of total cost 37. 
On the other hand, if there are a number of pairs of nonadjacent vertices
(we say the graph is not very dense), a complete search may be feasible. We
give a technique for doing this at the end of the next section.

11.4 Trees
A tree is a connected graph that contains no cycles. Figure 11.8 shows
three examples of trees. Paths Pn and stars K1,n are other examples.
 Further Graph Theory 205

A tree is a minimal connected graph in the following sense. If any edge

is deleted, the resulting graph is not connected. An edge is called a bridge
if its deletion disconnects the graph (or, in the case of a disconnected graph,
disconnects the component in which the edge lies). So we have just said that
every edge in a tree is a bridge. In fact we can prove a stronger result. First
we show:

LEMMA 11.3
An edge xy in a graph is a bridge if and only if it belongs to no cycle in the
graph.

Proof. Write G for the component of the graph that contains x and y. Delete
edge xy from G; call the resulting graph H.
(i) First assume xy is a bridge. Then H is not connected. So there are
two vertices in G, w and z say, that are not joined by a walk in H. But they
are joined by at least one walk in G. The only possibility is that xy was an
edge in every one of those walks. In particular, there is at least one walk of
the form wW xyZz in H, where W and Z are some strings of vertices. (If x
appears before y in every walk, we can swap labels w and z.)
Suppose xy is in a cycle in G, say xx1 x2 . . . yx. Even after the edge xy
is deleted, the path xx1 x2 . . . y is still in H, and there is a w-z walk in H,
namely
wW xx1 x2 . . . yZz.
But we said there was no such walk. This is a contradiction, so the cycle
could not have existed.
This proof is rather wordy and pedantic, but it becomes much clearer if
you look at Figure 11.9. In the left-hand diagram the heavy lines indicate the
walk that includes the edge xy, with the rest of the cycle shown in light dots.

w x y z w x y z

G H
FIGURE 11.9: Illustration of Lemma 11.3.

(ii) Suppose xy is a not bridge, so H is connected. So there is an x-y path

in H, and it does not contain edge xy. Add in that edge, and you get a cycle
in G that contains xy. 
From this we deduce a simple characterization of trees:

THEOREM 11.4 (Bridge Characterization of Trees)

A connected graph is a tree if and only if every edge is a bridge.
206 Introduction to Combinatorics

(Writing out a formal proof of this is left as Exercise 9A.15.)

There is a similar characterization involving vertices. A vertex is called a
cutpoint if its deletion disconnects the graph (or, in the case of a disconnected
graph, disconnects the component in which it lies). If a connected graph is a
tree then every vertex of degree greater than 1 is a cutpoint.
Another important characterization involves the numbers of vertices and
edges.

THEOREM 11.5 (Order Characterization of Trees)

A connected graph G with v vertices is a tree if and only if it has exactly v − 1
edges.

Proof. (i) Suppose G is a tree with v vertices. We prove by induction on v

that G has v − 1 edges.
The result is true for v = 1–the only graph on 1 vertex, K1 , is a tree.
Suppose it is true for v < n, and suppose G is a tree with n vertices. Select
an edge xy and delete it. The result is a union of two components, say Gx
(containing x) and Gy . Gx is a tree, because it contains no bridge, and it has
fewer than n vertices (at least one, y, is gone from G). Say it has m vertices.
By the induction hypothesis it has m − 1 edges. Then Gy is also a tree, with
n − m vertices, and (n − m) − 1 edges. G is the union of three edge-disjoint
parts, namely Gx , Gy and xy, so it has (m − 1) + (n − m − 1) + 1 = n − 1
edges.
(ii) Suppose G has v vertices but is not a tree. It has an edge that is not
a bridge. Delete that edge. If the resulting edge is not a tree, repeat the
process. Continue until only bridges remain. The result is a tree, and has
v − 1 edges. Since at least one edge has been deleted, G had more than v − 1
edges. 

From part (ii) of the above proof we see that any graph with v vertices and
fewer than v − 1 edges cannot be connected.

COROLLARY 11.5.1
Any tree has at least two vertices of degree 1.

Proof. Say a tree has v vertices, of which u have degree 1. Every other vertex
has degree at least 2, so the sum of the degrees is at least u+2(v −u) = 2v −u.
There are v − 1 edges, so the sum of the degrees is 2v − 2. So 2v − u ≤ 2v − 2,
and u ≥ 2. 

Trees can be used to enumerate things. One example is the use of trees to
find all Hamilton cycles in a graph.
Suppose we want to find all Hamilton cycles in a graph G. We use G to
construct a new graph, a tree, which we shall call T (G) or simply T .
 Further Graph Theory 207

First we select any vertex of G as a starting point; call it x. We construct

the first vertex of the tree and label it with x. This vertex of the tree is called
the root. For every edge of G that contains x, draw an edge of the tree. The
set of these new vertices is called the first level of vertices. Each vertex in the
first level is labeled with the corresponding vertex of G.
At level i + 1, we process every vertex y as follows. Look at the string
of labels on the vertices (called a branch leading from the root to the new
vertex). If this string is x, x1 , . . . , xi , y, then P = xx1 . . . xi y is a path in G,
and we add an edge of the tree corresponding to every edge of G that could be
an edge of the Hamilton cycle starting with P (that is, it would not duplicate
an edge of G or complete a small cycle in G), and label the new endpoint with
the name of the corresponding vertex.
As we go on, some of the branches cannot be continued. At level v, the
only possible label is x, and each branch that reaches level v corresponds to a
Hamilton cycle in G. Each Hamilton cycle will be represented precisely twice,
because the cycle is the same if traversed forwards or backwards.
Example 11.7: Find all Hamilton cycles in the graph
x y v

z u

First we select x as the starting point. There are three vertices joined to
x in G, namely y, z and u. So we start three branches, and label them as
shown.
x

y z u
A branch will go from y to vertices that we label z and v, but not to one
labeled x (because edge yx would be a duplication) or u (there is no edge yu
in G). Similarly there are two branches from each of the first level vertices
labeled z and u.
x

y z u

z v y u z v
We continue in this way. For example, the branch x, y, v can extend only to
u, because y would correspond to a repeated edge and v is not connected to x
or z. The branch x, z, u can extend only to v, because z would be a repeated
edge and x would form a small cycle.
208 Introduction to Combinatorics

After x, y, z, u, v there is no possible continuation, because neither vx nor

vz is an edge of G, vy would form a triangle and vu would be a duplication.
Others, like x, z, y, v, u, x can be completed: the branch contains every vertex
of G and finishes back at x—not a small cycle, because all the vertices are
there. The final tree is
x

y z u

z v y u z v

u u v v y y

v z u y v z

x x x x
So the example has two Hamilton cycles: xyvuzx and xzyvux. Each is rep-
resented twice. 

11.5 Spanning Trees

Suppose you have to connect a number of computers in an office build-
ing. For speed and reliability, the management wishes to have a hard-wired
connection between every pair of computers. The connection need not be di-
rect; it is possible to relay a message through an intermediate computer. To
model this, the computers are represented as vertices and the possible direct
connections are represented as edges. The set of edges that are chosen must
form a connected subgraph. Moreover this subgraph must contain all the
vertices—this is called a spanning subgraph.
There will be a positive cost associated with making a direct link between
a pair of computers. So, instead of making direct connections between every
pair of computers for which it is feasible, management might plan to avoid
any cycles. The resulting graph would still need to span the system, so they
want to find a connected subgraph that contains all the vertices, a spanning
tree in the original graph.
Every connected graph contains a spanning tree. To see this, consider any
connected graph. If there is no cycle, it is a tree. So suppose the graph
contains a cycle, such as abcdea. The pairs of vertices ab, bc, cd, de and
 Further Graph Theory 209

ea are all joined by edges. If you need a path from a to e (for example, to
transmit a message between computers), you could use the single edge, or the
longer path a − b − c − d − e. So remove one edge from the cycle. The resulting
spanning subgraph is still connected. If it still contains a cycle, delete an edge
from it. Continue in this way: graphs are finite, so the process must stop
eventually, and you have a spanning tree.
Spanning trees are important for a number of reasons. A spanning tree
is a subgraph with a relatively small number of edges—and therefore easy
to examine—but it contains a significant amount of information about the
original graph. They have proven very useful in designing routing algorithms
and in a number of areas of network design. We shall touch on some of this
here, but for a more complete discussion please consult books on trees such
as [141].
Sometimes it is useful to know how many different spanning trees there are
in a given graph or multigraph. We’ll write τ (G) for the number of spanning
trees in G. Some cases are very easy: if T is a tree, then τ (T ) = 1; if G
is derived from a tree T by replacing one edge of T by a multiple edge of
multiplicity k, then τ (G) = k; the cycle Cn has τ (Cn ) = n.
Example 11.8: Count the number of trees in the following (multi)graphs:

a b a d

c
d c b e
G H

A 4-cycle abcd has four spanning trees, and three of them include edge ab.
Each of those will give rise to two trees in G, one for each of the edges from a
to b. There is one other (the path (bcda)). So τ (g) = 7. In H, each spanning
tree will consist of a spanning tree in cycle (abc) and a spanning tree in cycle
(cde). Each has three, so the combined graph has τ (H) = 3 × 3 = 9. 
It is easy to work out τ (G) in small cases, but an algorithm is needed for
large G. If G is a graph or multigraph and A = xy is any edge of G, Gxy
(or GA ) is formed by deleting both x and y, inserting a new vertex A, and
replacing every edge zx and every edge zy by an edge from z to A. Gxy is
called the contraction of G by xy. We also define a (multi)graph G − xy (or
G − A), formed from G by deleting xy. Examples of these graphs are shown
in Figure 11.10.

THEOREM 11.6 (Deletion-Contraction Counting)

τ (G) = τ (G − A) + τ (GA ).
210 Introduction to Combinatorics

x y xy

z t
G G – xy G xy

FIGURE 11.10: Graphs for counting trees.

Proof. Suppose A is an edge of G. We count the spanning trees of G that

contain A and those that do not. A spanning tree that does not contain A is
a spanning tree of G − A; conversely, the spanning trees of G − A are spanning
trees of G and do not contain A. So there are τ (G − A) spanning trees of
G that do not contain A. Similarly, it is easy to see that each spanning tree
of G that contains A can be formed by taking a spanning tree of G − A and
adding a to it, and the spanning tree of G − A is uniquely determined. So the
number of spanning trees of G that contain A is τ (GA ). 

Example 11.9: Calculate τ (G − xy) and τ (Gxy ) for the graph G of Figure
11.10, and use this to find τ (G).
Every spanning tree of G − xy must contain edge yt, or else y will be
omitted. If neither edge xt is included, the other two edges must both be
used, and there is one such tree. Otherwise, there are two choices for the edge
from xt, and two choices for the edge to contain z (either xz or zt, but not
both). So there are four spanning trees with an xt edge, and five in all.
Similarly, for each edge from xy to t in Gxy there will be two spanning
trees, and there is one other, a total of seven.
So τ (G − xy) = 5, and τ (Gxy ) = 7, and τ (G) = 5 + 7 = 12. 
Here are two multigraphs that are useful in counting spanning trees. An
n-fold path is formed from a path by replacing each edge with a multiple edge
of multiplicity n. An n-fold cycle is defined similarly. The number of spanning
trees in an n-fold path is
τ (nPv ) = nv−1
and the number of spanning trees in an n-fold cycle is

τ (nCv ) = vnv−1 .

Verification is left as an exercise (Exercise 9A.14).

Another important example is the complete graph. Cayley discovered the
following result in 1889:

THEOREM 11.7 (Cayley’s Theorem) [19]

The complete graph Kn has nn−2 spanning trees.
 Further Graph Theory 211

The proof is left to Problems 11.10 and 11.11. Note that we are counting
all spanning trees, not just the number of isomorphism classes. For example,
K3 has three spanning trees, as the theorem says, but they are all isomorphic.
Example 11.10: Calculate τ (G), where G is the following graph. Check
your calculation by deleting and contracting a different edge.

The method of decomposing the relevant graphs is indicated in Figure 11.11.

From (i) we see that τ (G) = τ (G1 ) + τ (G2 ) = τ (G1 ) + τ (G3 ) + τ (G4 ). As
G1 is a 5-cycle, τ (G1 ) = 5; τ (G3 ) = 2 (it is a tree with one double edge); and
G4 is a 2-fold P3 , so τ (G3 ) = 23−1 = 4. So τ (G) = 5 + 2 + 4 = 11. From
(ii) we have τ (G) = τ (G5 ) + τ (G6 ) = τ (G7 ) + τ (G3 ) + τ (G8 ) + τ (G4 ). Now
τ (G7 ) = 1, τ (G8 ) = 4; so τ (G) = (1 + 2) + (4 + 4) = 11. 
Let’s return to the idea of connecting computers in an office building, and
consider the graph whose vertices are the computers and whose edges represent
hard-wired direct links. As we said, it is cheapest if the connections form a
tree. But which tree should be chosen?
We shall assume that the costs are additive. So the costs can be interpreted
as weights on the edges, and the cost of installing the communication system
will equal the sum of the costs of the installed edges. We want to find a
spanning tree of minimal cost, or minimal spanning tree.
A finite graph can contain only finitely many spanning trees, so in theory
you could list all spanning trees and their weights, and find a minimal spanning
tree by choosing one with minimum weight. But this process could take a very
long time, since τ (G) can be very large. So efficient algorithms that find a
minimal spanning tree are useful.
We shall outline an algorithm due to Prim [85] for finding a minimal span-
ning tree in a connected graph G with v vertices.. It finds a sequence of
trees, T1 , T2 , . . .. To start, choose a vertex x. Trivially, the cheapest tree with
vertex-set {x}—the only tree with vertex-set {x}—is the K1 with vertex x0 .
Call this T1 . Then find the cheapest tree with two vertices, one of which is x;
that is, find the cheapest tree that can be formed by adding one edge to T1 .
Call this tree T2 . Continue in this way: if you have constructed Ti , then Ti+1
is the cheapest tree that can be formed by adding one edge to it. The process
can always be carried out provided i < v—the graph is connected, so it must
always be possible to find a new vertex that is joined to one of the vertices of
Ti , so there is at least one candidate tree for the choice of Ti+1 , and finally
Tv is the desired minimal spanning tree.

THEOREM 11.8 (Minimality of Prim’s Algorithm)

The spanning tree constructed by Prim’s algorithm is minimal.
212 Introduction to Combinatorics

(i)
=> ;

G G1 G2

=> ;

G2 G3 G4

(ii)
=> ;

G G5 G6

=> ;

G5 G7 G3

=> ;

G6 G8 G4

FIGURE 11.11: Counting trees.

Proof. We continue the notation we used in describing the algorithm. For

convenience, write yi for the edge that was added in going from Ti−1 to Ti ,
and Vi for the vertex-set of Ti . The cost of edge y is denoted c(y).
Suppose T 1 is a minimal spanning tree in G. If T 1 = Tv then we are done.
Otherwise, find the smallest i such that yi is not an edge of T 1 . We could say
that yi is the first edge used in the algorithm that is not in T 1 . One endpoint
of yi is in Vi−1 and the other is not.
Since T 1 is a spanning tree of G, there is a path in T 1 joining the two
endpoints of yi . As one travels along the path, one must encounter an edge z
joining a vertex in Vi−1 to one that is not in Vi−1 . Now, at the iteration when
yi was added to T , z could also have been added, and it would have been
added instead of yi if its cost was less than that of yi . So c(z) ≥ c(yi ). Let
 Further Graph Theory 213

T 2 be the graph obtained by removing z from T 1 and adding yi . It is easy to

show that T 2 is connected, has the same number of edges as T 1 , and the total
cost of its edges is not larger than that of T 1 , therefore it is also a minimum
spanning tree of G and it contains y1 and all the edges added before it during
the construction of Tv . Repeat this process; each T j differs from Tv in fewer
and fewer edges, and eventually you achieve Tv . So Tv is a minimal spanning
tree. 
The new edge used to construct Ti+1 may not be uniquely defined in this
algorithm, and indeed Tv may not be uniquely defined. This is to be expected:
after all, there may be more than one minimal spanning tree in G. You need
to decide on a “tiebreaking” strategy before you begin. Let us assume, unless
we state otherwise, always to take the edge that comes earlier in alphabetical
order.
Example 11.11: Apply Prim’s algorithm to the graph

a 5 b
7 4
c 6 d
5 3
5 4
e g
6 f 4

(1) We start at a. The two possible edges are ab and ac. As ab is cheaper,
select y1 = ab.
(2) The only possible choices for the next edge are ac and bd, so the cheaper
one, bd, is chosen. T3 is the path abd.
(3) There are now four possible edges: ac (cost 7), cd (cost 6), df (cost 3) and
dg (cost 4), so the choice is df .
(4) Now there is a choice. The possible edges are ac (cost 7), cd (cost 6), cf
(cost 5), dg (cost 4), ef (cost 6) and f g (cost 4). Either dg or f g could be
used. Using alphabetical ordering, we choose dg.
(5) We have ac (cost 7), cd (cost 6), cf (cost 5) and ef (cost 6) (f g is not
allowed). The choice is cf .
(6) Finally, from ce (cost 5) and ef (cost 6) we choose ce. 
The sequence of trees is

Prim’s algorithm was a refinement of an earlier algorithm due to Kruskal

[67]. In his algorithm, you start by listing all edges in order of increasing cost.
214 Introduction to Combinatorics

The first approximation is the K2 consisting of the edge of smallest cost. The
second approximation is formed by appending the next edge in the ordering.
At each stage the next approximation is formed by adding on the smallest
edge that has not been used, provided only that it does not form a cycle with
the edges already chosen. In this case the successive approximations are not
necessarily connected, until the last one.
In large graphs, the initial sorting stage of Kruskal’s algorithm can be very
time-consuming.

Exercises 11A
1. Are the following networks eulerian? If so, find an Euler circuit. If not,
what is the eulerization number?

(i) (ii)

(iii) (iv)

(v) (vi)

2. Are the graphs in Chapter 10, Exercise A.8 eulerian? If so, find an Euler
circuit. If not, what is the eulerization number?
 Further Graph Theory 215

3. Draw the network of the following maze, and use it to solve the maze.

4. Find Hamilton cycles in the following graphs.

(i) (ii)

(iii) (iv)

5. Explain why the following graph has no Hamilton path.

c d
b e
a f

g h
216 Introduction to Combinatorics

6. The following diagram is a plan of the maze at the Hampton Court

palace, near London. You enter at the bottom; the object is to reach
the center. Produce a network corresponding to the maze and solve it.

7. Find a connected cubic graph with 10 vertices that is not hamiltonian.

8. The vertices of K5 are labeled 1, 2, 3, 4, 5, and edge xy has associated

cost x + y.
(i) Find the Hamilton cycle obtained by applying the sorted edges
algorithm to this graph, and find the cycles obtained using the
nearest neighbor algorithm starting from each vertex.
(ii) Show that every Hamilton cycle on this graph has the same length.

9. The complete graph K5 has vertices a, b, c, d, e. In each part we list

the set of costs associated with the edges. Find the costs of the routes
generated by the nearest neighbor algorithm starting at a, by the nearest
neighbor algorithm starting at c, and by the sorted edges algorithm.
(i) ab = 22 ac = 19 ad = 20 ae = 21 bc = 19
bd = 24 be = 22 cd = 19 ce = 18 de = 24.
(ii) ab = 53 ac = 70 ad = 55 ae = 58 bc = 54
bd = 54 be = 46 cd = 72 ce = 63 de = 77.
(iii) ab = 24 ac = 23 ad = 25 ae = 24 bc = 29
bd = 25 be = 27 cd = 22 ce = 26 de = 26.
(iv) ab = 26 ac = 27 ad = 32 ae = 25 bc = 21
bd = 24 be = 28 cd = 29 ce = 23 de = 24.

10. In each part we show a graph labeled with the costs associated with
the edges. Find the costs of the routes generated by the sorted edges
algorithm and by the nearest neighbor algorithm starting at each vertex,
if possible.
 Further Graph Theory 217

(i) u 5 v (ii) w 5 x
5 6 2 3
w x 6 7 8

6 8 7 4
y z
9
y 9 z
11. Show that the following graph has a Hamilton circuit, but that the
sorted edges algorithm fails and the nearest neighbor algorithm fails for
every possible starting vertex.
s t u
5 2
4 3
6 6 4
7 7
4 2
v 4 6
w 5 5
x
y z
12. Show that there are exactly six non-isomorphic trees on six vertices.
13. Suppose a tree T has a vertex of degree n. Show that T has at least n
vertices of degree 1.
14. Verify the formulas for:
(i) the number of spanning trees in an n-fold path;
(ii) the number of spanning trees in an n-fold cycle.
15. Find minimal spanning trees in the following graphs.
(i) a b c (ii) a b
9 c
4
6 3 7 3 8 6 7
6
d 6 e f 6 d e
5 4
9 8 5 3
3
g h i f g h
5
(iii) a b c (iv) a b
6 5 3 5 4 5 8
3 6
c d e f
d e
3
9 3 8 4 3 6 5 8
6 4
5
f 4 g g h i j
7
4 6 3 4 2
h j k l
4 i 4
218 Introduction to Combinatorics

Exercises 11B
1. Are the following networks eulerian? If so, find an Euler circuit. If not,
what is the eulerization number?

(i) (ii)

(iii) (iv)

(v) (vi)

(vii)

2. Are the graphs in Chapter 10, Exercise 10B.12 eulerian? If so, find an
Euler circuit. If not, what is the eulerization number?

3. The graph G has precisely two vertices x and y of odd degree. A new
graph or multigraph H is formed from G by adding an edge xy. If H is
connected, prove that G is connected.

4. Draw the network of the following maze, and use it to solve the maze.
 Further Graph Theory 219

5. For what values of m and n does Km,n have a Hamilton path?

6. Find Hamilton cycles in the following graphs.
(i) (ii)

(iii) (iv)

7. Show that the following graph contains a Hamilton path but no Hamil-
ton cycle.

8. The complete graph K5 has vertices a, b, c, d, e. In each part we list

the set of costs associated with the edges. Find the costs of the routes
generated by the nearest neighbor algorithm starting at a, by the nearest
neighbor algorithm starting at c, and by the sorted edges algorithm.
(i) ab = 44 ac = 39 ad = 30 ae = 51 bc = 49
bd = 46 be = 42 cd = 37 ce = 38 de = 44.
220 Introduction to Combinatorics

(ii) ab = 23 ac = 30 ad = 27 ae = 28 bc = 34
bd = 29 be = 41 cd = 32 ce = 42 de = 37.
(iii) ab = 8 ac = 5 ad = 7 ae = 11 bc = 9
bd = 7 be = 8 cd = 11 ce = 9 de = 11.
(iv) ab = 34 ac = 37 ad = 32 ae = 45 bc = 29
bd = 34 be = 29 cd = 42 ce = 43 de = 39.
9. In each part we show a graph labeled with the costs associated with
the edges. Find the costs of the routes generated by the sorted edges
algorithm and by the nearest neighbor algorithm starting at each vertex,
if possible.

(i) u 9 v (ii) s t u
4 4
7 8 6 3
7 6
w x 5 2 3
9 1
9 5 4 7
4 6
v 3 w 5 x
y 9 z 7 4
y z
10. Verify the results of Example 11.3.
11. Show that it is possible to decompose K9 into four Hamilton cycles, no
two of which share an edge.
12. Show that a graph on v vertices that has no cycle is connected if and
only if it has precisely v − 1 edges.
13. Prove that a graph is a tree if and only if: it has no cycles, but if you
join any two nonadjacent vertices the new graph has exactly one cycle.
14. Find minimal spanning trees in the following graphs.
(i) a b 4 c (ii) a
7 b 12 c
9 5
6 5 7
9 8 13
d e f
3 16 11 8 f
7 6 d e
4 3 9 7
g h i g
5 h
12

(iii) b
a 7 5 c
3 6 4
d 8 e
9 8 6
f 4 g
8 6 5
h i 3 j
 Further Graph Theory 221

(iv) 3
a b
5
2 7
c d e
5 4 5 f
4 8 6 9
h i
g 5 7 4 j
7 3
8 9
k l

15. Prove that if a connected graph is a tree then every vertex of degree
greater than 1 is a cutpoint, but the converse does not hold.

Problems

Problem 1: We did not mention loops in our discussion of Euler walks. How
would you treat them? What sort of road would translate as a loop?
Problem 2: The girth of a graph G is defined to be the length of the shortest
cycle in G. (The girth of a tree is either defined to be ∞ or else not defined.)
(i) Prove that if a regular graph of degree r has girth 4, then it has at least
2r vertices.
(ii) Prove that if a regular graph of degree r has girth 5, then it has at least
r2 + 1 vertices.
Problem 3: Draw a graph that has:
(i) An Euler circuit but no Hamilton cycle;
(ii) A Hamilton cycle but no Euler circuit;
(iii) Both an Euler circuit and a Hamilton cycle;
(iv) Neither an Euler circuit nor a Hamilton cycle.
Problem 4: The vertices of Kn are labeled 1, 2, 3, . . . , n, and edge xy has
associated cost x + y. Prove that the cost of every Hamilton cycle in this
graph is the same.
Problem 5: The graph G has 2n vertices, and every vertex has degree at
least n. Show that G is hamiltonian.
Problem 6: The graph Ka,b,c is defined to have three sets of vertices, of sizes
a, b and c, and two vertices are adjacent if and only if they are in different
sets.
222 Introduction to Combinatorics

(i) Prove that Kn,2n,3n is hamiltonian for every positive integer n.

(ii) Prove that Kn,2n,3n+1 is not hamiltonian for any positive integer n.
Problem 7: Perfect squares were defined in Problem 10.10.3. Show that no
tree with three or more vertices is a perfect square.
Problem 8: See Problem 10.10.3. The center of a graph consists of all the
vertices x such that ǫ(x) = R. Prove that the center of a tree consists of either
one vertex or two adjacent vertices.
Problem 9: T is a tree with precisely four vertices of degree 1 and precisely
one vertex of degree 4. Find the degrees of the remaining vertices of T , and
show that T is the union of two edge-disjoint paths.
Problem 10: Suppose d1 , d2 , . . . , dv are positive integers whose sum is 2v −2.
Write T for the set of all trees on v vertices x1 , x2 , . . . , xv such that xi has
degree di for every i. Show that the number of elements of T is

(v − 2)!
.
(d1 − 1)!(d2 − 1)! . . . (dv − 1)!

Problem 11: Show that there are precisely v v−2 different spanning trees in
Kv .
Chapter 12
Coding Theory

Suppose you want to send an e-mail message to a friend. Your message may
be an ordinary string of English words, but your computer can only send
electronic messages. So a program is used to convert your words into a string
of instances of two states, “power on” and “power off.” For our convenience
let’s write 1 for “power on” and 0 for “power off”; your message no longer
looks like “Let’s meet for lunch,” but something like

100110110011000011011100010110011011100010110011011001001011000.

Such a sequence of 0s and 1s is called a binary string.

The substitution process is called encoding, and the corresponding process
of going back from the string to the original message is called decoding. Coding
theory has been widely studied in mathematics, and there are a number of
good books on the subject. One example is [91].
Coding has also been studied from the viewpoint of secrecy and security;
this area is usually called cryptography. But we shall restrict our attention to
avoiding errors in the encoding and decoding process.

12.1 Errors and Noise

Sometimes there will be errors in a message. In electronic transmissions
they are often due to noise. If your message was sent a long distance, it may
be bounced off a satellite, and photographs of the Moon and Venus are trans-
mitted as radio messages in space; random electrical discharges, sunspots, etc.
can introduce errors. The medium through which the message is transmitted
is called a channel, and some channels are noisier than others.
To see a way around the noise problem, consider what happens in everyday
English, when there is a misprint. If you are reading a book and you come
across the word “ bive”, you are fairly sure this is not a real word. Assuming
one letter is wrong, there are many possibilities: dive, five, live, bile, bite
are only a few. If this were all the information you had, you could not tell
which was intended. But the context—the other words nearby—can often
help. For example, if the sentence is “. . . the next bive years . . . ” then the

223
224 Introduction to Combinatorics

author probably meant “five.” If it says “. . . a review of a bive performance

. . . ” then “live” is most probable. The phrase “. . . teeth used in the bive . . . ”
suggests “bite”, and “. . . let me bive you an example . . . ” probably should be
“give.” The context gives you extra information.
If the encoding results in a binary string, there is no information from
context. You don’t even have a context until after the string is decoded. To
get around this, we add some more symbols. One very simple method is to
break the message up into substrings or words, say of length 4, and after each
subsequence add another symbol (a check digit ), 0 or 1, chosen so that the
subsequence has an even number of 1s. For example, suppose you want to
send
10011011001100001101.
First break it up into the form

1001 1011 0011 0000 1101

and then insert check digits:

10010 10111 00110 00000 11011.

This method is useful when there is a way of checking with the sender: for
example, if the message was received as

10110 10111 00110 00000 11011

you could say “I got the second through fifth words OK; would you repeat
the first?” But this is not useful for electronic communications, where a large
number of symbols are sent every second. In the next section we’ll look at a
way around this.
Another problem is that the method is useless if more than one error occurs
in a word. The same occurs in everyday English: if “ bive” contains two errors,
there are many more possibilities: “bone,” “bane,”, “brie,” and so on. There
is no solution to this problem. There are encoding methods that can detect
(and correct!) many errors, but it is always possible that more errors occur.
For example, in the photos sent back from space, there will be the occasional
small area where the image cannot be decoded.

12.2 The Venn Diagram Code

We assume you are sending a message composed of 0s and 1s. If it is an
ordinary English sentence, maybe you represent a as 000001, b as 000010, . . . ,
A as 011011, and so on. These 64 (= 26 ) strings will enable you to repre-
sent the 26 capitals and 26 lowercase numbers, together with 12 punctuation
 Coding Theory 225

marks. This type of representation is widely used in computers—the ASCII

(American Standard Code for Information Interchange) code, which was de-
veloped for computers, uses seven binary symbols for each letter, giving a
wide range of punctuation and special symbols. Extended ASCII uses eight
binary symbols for each.
Suppose you are happy to assume that no more than one error will occur in
every seven transmitted symbols: the cases where more errors occur will be
sufficiently few that you can recover the message. Proceed as follows. First,
break the message up into substrings, or message words, of four symbols.
Then replace each message word by a codeword as follows. If the message is
ABCD (each of A, B, C and D represents a 1 or a 0), define three more binary
digits, E, F and G, chosen as follows so that each of the sums A + B + C + E,
A + C + D + F and B + C + D + G is even.
This whole process is called encoding or encryption.
Example 12.1: How do you encrypt 110011010110?
First break the message up as

1100 1101 0110.

If ABCD is 1100, then A + B + C + E = 1 + 1 + 0 + E; for this to be

even, E = 0. Similarly A + C + D + F = 1 + 0 + 0 + F, F = 1, and
B + C + D + G = 1 + 0 + 0 + G, G = 1. The first codeword is 1100011. The
other words are 1101000 and 0110010. 
The easiest way to represent this process is as follows. Draw a diagram
(essentially a three-set Venn diagram) of three overlapping circles, and write
the numbers A, B, C and D in the positions shown in the left-hand part
of Figure 12.1. Then the other numbers are chosen so that the sum in each
circle is even. These are E, F and G, as we defined them, and shown in the
right-hand part.

A E A F
C C
B D B D
G

FIGURE 12.1: Venn diagram encoding.

After the codeword is sent, the receiver, (decoder or decrypter ) must re-
construct the original message. Suppose a string of seven binary symbols is
received. The decoder assumes that at most one error has been made in the
encryption. If more than one error occurs in seven symbols, the wrong mes-
sage will be found. We have to assume that this will occur very infrequently.
(If errors are more common, other methods can be used.)
226 Introduction to Combinatorics

Example 12.2: Suppose you receive 1111010. How is it decoded?

A + B + C + E = 3, so one of these symbols must be wrong. (Since there

is assumed to be at most one error, there cannot be three wrong symbols.)
A + C + D + F = 3, so one of these symbols must be wrong.
B + C + D + G = 2, so these four symbols are all correct.
The first and third conditions tell us that exactly one of A and E is incor-
rect; the second and third conditions tell us that exactly one of A and F is
incorrect. So A is wrong; it should have been 0, not 1, the correct codeword
is 0110010 and the original message was 0110.

3 3
0 1 1 0 0 1
1 1
1 0 1 0
0 2 0

FIGURE 12.2: Venn diagram decoding.

This decoding process is very simple when the Venn diagram is used. The
decoder fills in the diagram and puts the number of 1s next to each circle, as
shown on the left-hand side of Figure 12.2. The two top circles are odd; the
wrong symbol must be the one in the unique cell common to both of those
circles. 

This process will always work. Provided there is at most one error, the
Venn diagram method will yield the correct codeword. Moreover, every binary
string of length 7 can be decoded by this method. We’ll prove this in the next
section.

12.3 Binary Codes, Weight, and Distance

There are 16 possible message words in Venn diagram encoding, so there
are 16 codewords, as shown in Figure 12.3.
The complete set of 16 words is called a binary code, or simply code. In
general, a binary code means any set of binary codewords. Obviously you
could encode messages using more than two symbols, but we will concen-
trate on the binary case because of its widespread applications in electronic
communication.
In most cases, we require all words in a code should be same length; such
a code is called a block code, and the common length is called the length of
the code. (There are a few important exceptions; see Section 12.7.)
 Coding Theory 227

Message Codeword Message Codeword

0000 0000000 1000 1000110
0001 0001011 1001 1001101
0010 0010111 1010 1010001
0011 0011100 1011 1011010
0100 0100101 1100 1100011
0101 0101110 1101 1101000
0110 0110010 1110 1110100
0111 0111001 1111 1111111

FIGURE 12.3: The codewords in Venn diagram encoding.

Given two binary strings, we define the distance between them to be the
number of positions in which they differ. The minimum distance between
any two codewords in a code (that is, the smallest of all differences between
two codewords) is called the distance (or minimum distance) of the code.
The weight of a codeword is the number of 1s it contains. That is, it is the
distance from 000 . . . 00. The weight of a code is the minimum weight of all
its non-zero words (if 000 . . . is in the code, you don’t include it in calculating
the minimum weight). We use the notations d(x, y) for the distance between
binary strings x and y and w(x) for the weight of x. The weight and distance
of a code C are denoted w(C) and d(C). (The use of the same notation for
distance between strings and between vertices of a graph is quite intentional;
see Problem 12.7.)
Example 12.3: The (n+1, n) parity-check code was introduced informally
in Section 12.1. It has 2n codewords of length n + 1. The messages are the
possible binary strings of length n, with another symbol added at the end to
make the number of 1s even. What are the words of the (3, 2) parity-check
code? What is the minimum distance of this code?
The possible messages are 1 1, 1 0, 0 1 and 0 0, so the codewords are 1 1 0,
1 0 1, 0 1 1 and 0 0 0. By inspection, the code has minimum distance 2. 
We’ll look further at parity-check codes in Example 12.3 and Exercise
12A.5.
Example 12.4: Here is a code with five words:

A = 1 0 1 0 1 0 1
B = 0 1 0 1 0 1 0
C = 1 1 1 0 0 0 0
D = 0 0 0 1 1 1 1
E = 0 0 0 0 0 0 0
228 Introduction to Combinatorics

What are the length, distance and weight of this code?

The code is length 7. The distances between the words are
A to B: 7 B to D: 3
A to C: 3 B to E: 3
A to D: 4 C to D: 7
A to E: 4 C to E: 3
B to C: 4 D to E: 4.
So the minimum distance—the distance of the code—is 3.
The weights of A, B, C and D are 4, 3, 3 and 4. As only non-zero words
are to be checked, the code has weight 3. 
The distance between binary strings obeys the usual rule for metrics:

THEOREM 12.1 (Triangle Inequality)

If x, y and z are binary strings, then d(x, z) ≤ d(x, y) + d(y, z). 
The proof is left as an exercise.

The 16 words that can arise from Venn diagram encoding form a code of
length 7. If you check the table, you will see that the code has weight 3, and
in fact the minimum distance between words is 3. So this is a weight 3 code.
(At the moment, the easiest way to see this is to check all the pairs of words.
We’ll see an easier method in Section 12.4.)
Suppose a codeword x from this code is to be transmitted, and exactly
one error occurs. Let’s say the actual string received is y. Then the triangle
inequality tells us that the distance from y to any other codeword (other than
x) must be at least 2. So, if it is assumed that the number of errors is at most
1, then the correct message will be reconstructed. We express this by saying
that the Venn diagram code will correct any single error. It is called a single
error-correcting code.
In nearly all cases, decoding follows the rule that, if the received string is
not a word of the code, it is decoded on the assumption that the number of
errors was smallest possible. This is called nearest neighbor decoding.
Suppose a code has minimum distance d. If d or more errors are made when
codeword x is transmitted, it is possible that a different codeword could be
received, and the receiver would mistakenly assume that no error occurred.
However, this is impossible when the number of errors is smaller than d. In
that case the error would certainly be detected. If the number of errors is
smaller than d/2, nearest neighbor decoding will lead to the correct message.
These two facts are usually expressed as

THEOREM 12.2 (Distance Criteria)

A binary code of distance d will detect any d − 1 errors and correct any
(d − 1)/2 errors. 
 Coding Theory 229

(Of course, when d is even, this last number is rounded down to d/2 − 1,
because a fractional number of errors is meaningless.)
It follows from Theorem 12.2 that the Venn diagram method will correct
any single error in a word.
Example 12.5: The (n + 1, n) parity-check code was defined in Example
12.3. How many errors can it detect or correct?
The code has distance 2. Any single error can be detected, but no errors
can be corrected. (For example, if you receive 11100, you cannot tell whether
the original was 01100, 10100, 11100, 11110 or 11101.)
In constructing a block code with distance d, we must select a number of
codewords (binary strings) of length n such that every two words differ in at
least d places. We would like to find as many codewords as possible. What is
the maximum number? This problem is not completely solved, but there are
some useful results.
Example 12.6: What is the maximum number of words in a block code
with length 3 and distance 2?
We know that the (3, 2) parity-check code has length 3 and distance 2, and
this code has four words. Are five words possible?
Suppose we have such a set S of five words. Write S = S0 ∪ S1 , where Si
consists of the words starting with i. One of these must have at least three
elements; say it is S1 , and suppose S1 contains the distinct words 1 a b, 1 c d
and 1 e f . Then a b, c d and e f are distinct binary words each of distance 2.
But this would mean, for example, that a, c and e are all different, and this
is impossible since we only have two symbols to choose from. 

12.4 Linear Codes

The set of integers modulo 2 is denoted Z2 , and consists of the two elements
0 and 1 whose sum is defined by

0+0 = 0 0+1 = 1
1+0 = 1 1+1 = 0

and whose product is

0×0 = 0 0×1 = 0
1×0 = 0 1 × 1 = 1.

The join or bit sum or sum of two binary sequences is defined as follows:
the sum of individual entries is the sum in Z2 , as defined above, and to add
230 Introduction to Combinatorics

sequences, you add terms separately, with no carry; for example,

1 0 10 1 01 1
+ 1 1 1 0 0 0 1 0.
0 1 00 1 00 1

In other words, sequences are added as vectors whose elements, instead of

real numbers, are members of Z2 .
Given this definition the distance between two codewords is the weight of
their sum.
The set of all binary sequences of length n form an abelian group, under the
sum (or join) operation. The identity element is 000 . . . 0, and each element is
its own inverse. We shall denote this set as Vn (2). It can also be interpreted
as a vector space over the field Z2 .
A linear code is one in which the binary sum of any two of the codewords
is also a codeword. In the binary case, this is the same as saying that the
code is a group under the sum operation. Two important properties of linear
codes are that the zero element 000 . . . 0 is always one of the codewords, and
the minimum distance of code equals the weight of code.
Example 12.7: Show that the Venn diagram code is linear.
We have described Venn diagram encoding as follows: “Given a message
word A1 A2 A3 A4 , with elements from {0, 1}, the corresponding codeword is
A1 A2 A3 A4 A5 A6 A7 , where A1 + A2 + A3 + A5 , A1 + A3 + A4 + A6 and
A2 + A3 + A4 + A7 are even.” An alternative formulation is, “Given a message
word A1 A2 A3 A4 , with elements from Z2 , the corresponding codeword is
A1 A2 A3 A4 (A1 + A2 + A3 ) (A1 + A3 + A4 ) (A2 + A3 + A4 ).” Then the binary
sum of A1 A2 A3 A4 A5 A6 A7 and B1 A4 B3 B4 B5 B6 B7 is
(A1 + B1 )(A2 + B2 ) (A3 + B3 ) (A4 + B4 ) (A5 + B5 ) (A6 + B6 ) (A7 + B7 )
= (A1 + B1 )(A2 + B2 ) (A3 + B3 ) (A4 + B4 ) ([A1 + A2 + A3 ] + [B1 + B2 + B3 ])
([A1 + A3 + A4 ] + [B1 + B3 + B4 ]) ([A2 + A3 + A4 ] + [B2 + B3 + B4 ])
= (A1 + B1 ) (A2 + B2 ) (A3 + B3 ) (A4 + B4 ) ([A1 + B1 ] + [A2 + B2 ] + [A3 + B3 ])
([A1 + B1 ] + [A3 + B3 ] + [A4 + B4 ]) ([A2 + B2 ] + [A3 + B3 ] + [A4 + B4 ])

which is a codeword—the codeword corresponding to the message

(A1 + B1 ) (A2 + B2 ) (A3 + B3 ) (A4 + B4 ).

So the binary sum of two codewords is a codeword: the code is linear. 

We can generalize this as follows. Take all binary strings of length n as
messages. Add check digits in a linear fashion: if the message is A1 A2 . . . An
and the check digits are An+1 , . . . , An+j , . . . , then each An+j equals a linear
form in A1 A2 . . . An (over Z2 ). (The alternative formulation of the Venn
diagram code is an example of this.) Then possibly apply a fixed permutation
to every codeword.
Suppose a code can correct e errors. It is quite possible that a binary string
could differ from every possible codeword in more than e places. A very simple
 Coding Theory 231

example is the code C = {0000, 1111}. If two errors are made, for example if
1010 is received, there is no reason to decode as 0000 or as 1111. We would
prefer a code that gives a unique solution under nearest neighbor decoding.

THEOREM 12.3 (Hamming Bound)

Suppose C is an e-error-correcting code with n codewords of length t. Then
2n
n ≤ Pe t .


i=0 i

Proof. Write S for the set of all binary vectors of length n. For each codeword
a in in C, define S(a) to be the set of all members of S that are distance e
or less from a. Suppose x is a common member of S(a) and S(b). If message
x is received, we cannot tell whether the intended message was a or b. So in
this case the code cannot correct a transmission with e (or fewer) errors—a

contradiction. Therefore the sets C(a), for ainC, are all disjoint. There are ti
Pe
vectors distance precisely i from a, so |C(a)| = i=0 ti . There are n words


Pe
in C, so there are at least n i=0 ti binary vectors of length n. But there are


Pe n
precisely 2n such vectors. So 2n ≥ n i=0 ti , and n ≤ Pe2 t .



i=0 (i)

The inequality in Theorem 12.3 is called the Hamming bound after Richard
Hamming (See Appendix C).
A code C of N binary codewords of length t, of width d(C) = d = 2e + 1, is
called perfect if it meets the Hamming bound, that is
2n
n ≤ Pe t .

(12.1)
i=0 i

Then for every possible binary word w0 of length n, there is a codeword w

in C whose distance from w0 is at most e. This word w is unique—C is e-
error correcting, so there cannot be another codeword w′ with d(w0 , w′ ) ≤ e.
So a perfect code can transmit the maximum possible number of different
messages.
The Venn diagram code is a perfect code: it has e = 1, n = 16 and t = 7.
Then
e      
X t 7 7
= + = 1 + 7 = 8 = 128/16 = (27 )/16 = (2t )/n,
i=0
i 0 1

and (12.1) is satisfied.

232 Introduction to Combinatorics

12.5 Hamming Codes

We could describe Venn diagram encoding as follows: “Given a message
word A1 A2 A3 A4 , with elements from Z2 , the corresponding codeword is
A1 A2 A3 A4 (A1 + A2 + A3 ) (A1 + A3 + A4 ) (A2 + A3 + A4 ).” We can use
matrices to prescribe this encoding: if the message is A1 A2 A3 A4 , the three
check digits are the elements of the vector
 
  A1
1 1 1 0 
 1 0 1 1   A2  .

 A3 
0 1 1 1
A4

There is also a matrix method of testing whether a seven-symbol string

A1 A2 A3 A4 A5 A6 A7 is a codeword. Write H for the 3 × 7 matrix
 
1 1 1 0 1 0 0
H =1 0 1 1 0 1 0
0 1 1 1 0 0 1

and U for the vector

U = (A1 A2 A3 A4 A5 A6 A7 ).

That is, V is the string A1 A2 A3 A4 A5 A6 A7 written as a vector over Z2 .

Then the string is a codeword if and only if

HU T = (0 0 0)T .

Now suppose an error has been made, and the string W has been sent instead
of the codeword U . We call the difference, V = W − U (and because we are
working over zz2 , we could just as easily say V = W + U ), the error pattern.
When we test W we get

HW T = HU T + HV T = (0 0 0)T + HV T = HV T .

The useful property of the Venn diagram code is that, if we assume there is
at most one error, HV T tells us exactly which symbol was wrong.
In order to generalize this, we look at the columns of H. If we translate the
columns into numbers by making column (a b c)T correspond to the integer
4a+ 2b + c, each of 1, 2, 3, 4, 5, 6, 7 occurs precisely once. We say that (a b c)T
is the binary column representation of a + 2b + c, the representation in base 2
written as a column (with leading zeroes when necessary). Moreover, the last
three columns—the columns that correspond to the check digits—represent
1, 2 and 4, the first three powers of 2.
 Coding Theory 233

Instead of three rows, let’s take n rows. We construct an n × (2n − 1)

matrix H whose jth column is the binary column representation of j. The
code we construct has as its codewords binary vectors of length 2n − 1. Each
message is a string of 2n − 1 − n binary symbols. We shall refer to the code
as a (2n − 1, 2n − 1 − n) code. The message is converted into a codeword
by adding n check digits; the check digits will be in positions 1, 2, 4, . . . , 2n−1
(that is, the corresponding columns represent the n powers of 2 starting with
20 ), and the message symbols are placed in order in the remaining positions.
For example, in the case n = 3 (a (7, 4) code), we have 2n − 1 − n = 4; the
message A B C D (the letters representing some string of 0s and 1s) would give
rise to a codeword ∗ ∗ A ∗ B C D where the asterisks are the check digits.
How would the check digits be selected? If U = (u1 , u2 , . . .) is to be a
codeword, we require HU T = 0 (the zero vector of appropriate length). This
is a set of n linear equations. The columns of H corresponding to the check
digits will each contain only one 1 and every other symbol 0, so each check
digit occurs in only one equation; and, in fact, the first non-zero element is
the check digit. So we can read off the values very easily. For this reason, H
is called the encoding matrix .
Example 12.8: Show how to find the check digits in the case n = 3.
In this case the matrix H is
 
0 0 0 1 1 1 1
H =0 1 1 0 0 1 1
1 0 1 0 1 0 1

and HU T = 0 is
u4 + u5 + u6 + u7 = 0,
u2 + u3 + u6 + u7 = 0,
u1 + u3 + u5 + u7 = 0.

So we see immediately that the check digits are

u4 = u5 + u6 + u7 ,
u2 = u3 + u6 + u7 ,
u1 = u3 + u5 + u7 .

(Remember that in Z2 , addition and subtraction are essentially the same.)


Suppose it is intended to transmit the codeword U , but an error occurs
and the string U + V is sent—V is the pattern. Assume that at most one
error has occurred; V contains at most one 1. Now calculate H(U + V )T . If
this is zero, we assume the received word was correct, and we reconstruct the
message by deleting the check digits. Otherwise, H(U + V )T = HV T will
equal a column of H, namely the column that corresponds to the error. If
234 Introduction to Combinatorics

the column is the binary column representation of j, then the error was in
position j of the received word, and can be corrected immediately.
The code we have constructed is called a (2n − 1, 2n − 1 − n)-Hamming
code. These codes were first discovered by Richard Hamming. The Hamming
codes are perfect single-error-correcting linear codes. So there is a perfect
single-error-correcting code of length 2n − 1 for positive integer n ≥ 2. (When
n = 1, the message has length 0, so it has no content.)
Example 12.9: Describe the (3, 1)-Hamming code.
This is the case n = 2. The encoding matrix is
 
01 1
H=
10 1

and the two possible messages are 0 and 1. 0 encodes as 0 0 0 and 1 encodes
as 1 1 1. 
The Venn diagram code is essentially the (7, 4)-Hamming code, with the po-
sitions in the codeword permuted to make encoding and decoding by diagram
easier.

12.6 Codes and the Hat Problem

Suppose a code has minimum distance 3. If you are presented with a string
U that is not a codeword, there is at most one codeword distance 1 from U ,
because if V and W are codewords such that d(U, V ) = 1 and d(U, W ) = 1
then d(V, W ) ≤ 2; the only possibility is that V = W . As a consequence, if
you are given a word that is not a codeword, there is at most one place where
changing the symbol will result in a codeword. If the code is perfect, there
will be exactly one such place.
We now go back to the Hat Game that we discussed in Section 1.1. If we
look at the three hat colors in some order (for example, order the wearers’
names alphabetically), the contestants could interpret what they are wearing
as a binary string, like a codeword, with red and black replacing the binary
symbols 0 and 1.
Maybe the three contestants are obsessed with Hamming codes. Maybe
they learned coding theory from Dr. Walter Johns, and they hated his teach-
ing style so much that they never want to see a code again. They say, “Why
don’t we assume that, whatever happens, the string we are wearing is not a
word in the (3, 1)-Hamming code!”
The code consists of the two words 000 and 111, and every possible string
is a codeword or else differs from a codeword in exactly one place. Suppose
the string that occurs is not a codeword. There is exactly one place where
 Coding Theory 235

a change will produce a codeword. If you were the contestant wearing the
hat that corresponded to that position, you would know the color of your hat:
your reasoning would be something like, “If my hat were red, our string would
be a codeword, and we decided that isn’t going to happen.” In that case you
would know that your hat was black. Of course, your assumption (that the
string is not a codeword) might be wrong, and you would lose. But only two
of the eight possible strings are codewords, so you would win the game in two
out of every eight cases, or 75% of the time.
This same sort of reasoning applies whenever there is a perfect code avail-
able. So, if n is any positive integer greater than 1, the (2n − 1, 2n − 1 − n)-
n
Hamming code can be used. There are 2[2 −1−n] codewords, and since they
n [2n −1]
are of length 2 − 1 there are 2 possible binary strings. If the hat colors
are chosen at random (as is true of the Hat Game), the chance that the string
representing them will be a codeword is
n
2[2 −1−n] 1
= n,
2[2n −1] 2
so the probability that the strategy we described will be succesful approaches
1 as the number of players goes to infinity.

12.7 Variable-Length Codes and Data Compression

So far we have looked at block codes or fixed-length codes, where every word
is the same length. Many codes are like this. But some codes use different
length words. They are called variable-length codes.
Perhaps the best-known example is the Morse code. This was designed for
transmission where there can be a break between codewords. Beginning in
1836, Samuel F. B. Morse and Alfred Vail developed the electric telegraph,
and the code was developed by Morse to transmit numbers and expanded by
Vail to an alphabetic code in 1844. The code uses short and long transmissions
(“dots” and “dashes”) to represent letters. A dot is formed by transmitting
continuously for one time period and a dash by three periods. There is a
codeword for each letter in the English alphabet. The break between two
letters is denoted by seven periods with no transmission. For example, to
transmit “dot dash” followed by a new letter, one might depress a key for
one-tenth of a second, let it go then immediately hold it down for three-
tenths of a second, let the key go for seven-tenths of a second, then start on
the next letter.
The Morse code is shown in Figure 12.4. For example the “dot dash” we
mentioned above would translate as the letter A. Notice that letters commonly
used in the English language were assigned short sequences; the most common
236 Introduction to Combinatorics

A J R
B K S
C L T
D M U
E N V
F O W
G P X
H Q Y
I Z

FIGURE 12.4: The Morse code

vowel, E, is represented by a single dot and the most common consonant, T ,

is a single dash. Encoding so that the most common letters (or words or
phrases) get the fewest symbols is called data compression.
Morse code was designed for manual transmission, and the break between
codewords is provided by the person doing the transmitting. Electronically,
variable-length codes can either use the “end-of-line” symbol for a break or
else the codewords can include information that signals the end of a codeword.
For example, (00 at end).
The genetic code provides an interesting example. Genes are strings of four
nucleotides, adenine, thymine, guanine and cytosine, with repetitions possible.
The names are usually abbreviated A, T, G and C, so a gene can be specified
by a sequence of these four symbols.
Suppose we want to use a binary code (that is, 0s and 1s) to represent a
genetic string. One possibility is to use codewords
00 for A
01 for C
10 for T
11 for G.
For example, the string AACACAGTATACA would be represented by
000001000100110010000100,
which uses 26 symbols.
It happens that A is most common in genetic strings, then C, then T and
G. If we use the following (data compression) encoding:
0 for A
10 for C
110 for T
111 for G
then AACACAGTATACA is represented by 0010010011111001100100, which
only takes 22 symbols. When very large collections of very large strings are
to be transmitted, the shorter version (the compressed genetic code) is useful.
In order to decode the symbol for a string of nucleotides, notice that:
 Coding Theory 237

• Every time a 0 occurs, it is the end of a codeword;

• Every time 111 occurs, it’s the codeword for G;

• The only way a codeword can end is with 0 or 111.

Example 12.10: Decode 001100101110110110001000.

First put commas at the ends of words: 001100101110110110000010 be-
comes 0, 0, 110, 0, 10, 111, 0, 110, 110, 0, 0, 10, 0, 0. Now decode symbol by sym-
bol, to obtain AATACGATTAAAACAA.

Exercises 12A
1. What is the distance between the following strings?
(i) 101110111 and 101100110
(ii) 10101010 and 01010101
(iii) 000010000 and 001010110
(iv) 0101101011 and 0011100110

2. Prove (Theorem 12.1) the triangle inequality for distance between binary
strings.

3. Consider the five-word code given in Example 12.3. How would you
interpret the following received messages, using nearest neighbor decod-
ing?

(i) 1011101 (ii) 0001010

(iii) 1101010 (iv) 0000110

4. Consider a code consisting of the seven words 0000000, 0010111, 0101110,

1011101, 1110010, 1100101, 1001011. What are the weight and distance
of this code? What are its error-correcting and -detecting properties? If
the word 0111001 were added, what changes are made to these param-
eters?

5. The parity-check code of length 6 contains words of six binary symbols.

The first five characters represent the message, and the sixth is chosen
such that the sum of the symbols is even.
(i) How many messages are possible?
(ii) How many errors will this code detect? How many will it correct?
238 Introduction to Combinatorics

6. Show that there is a set of four binary strings of length 5 with minimum
distance 3. Is there such a set with five members?
7. Encode the following messages using Venn diagram coding.

(i) 1010 (ii) 0011

(iii) 1001 (iv) 0010

8. Decode the following messages using Venn diagram coding.

(i) 1001011 (ii) 0011111

(iii) 1000101 (iv) 1101000
(v) 0011010 (vi) 1110011

9. To transmit the eight possible messages of three binary symbols, a code

appends three further symbols: if the message is ABC, the code trans-
mits A B C (A + B) (A + C) (B + C). Write down the encoding matrix
for this code. How many errors can it detect and correct?
10. Use the compressed genetic code to encode the following:
(i) AAACAATAACTAGACGAA
(ii) CACAAAACAATACCAAAA
11. Use the compressed genetic code to decode the following:
(i) 010001010000011100011000
(ii) 00010001100011101001000010

Exercises 12B
1. What is the distance between the following strings?
(i) 101110101 and 101100110
(ii) 10101010 and 01100010
(iii) 000011100 and 001010100
(iv) 010110100101 and 101011100110
2. Consider a code consisting of eight words 00000000, 00101110, 01011100,
10111010, 11100101, 11001011, 01110011, 10010111. What are the
weight and distance of this code?
3. Encode the following messages using Venn diagram coding.
 Coding Theory 239

(i) 1111 (ii) 0110

(iii) 1100 (iv) 1101

4. Decode the following messages using Venn diagram coding.

(i) 1001101 (ii) 0010111

(iii) 1011001 (iv) 1100011
(v) 0001101 (vi) 0101010

5. Write down the encoding matrix for the (15,11)-Hamming code.

(i) How does this code encode the message 11001011111?
(ii) How does this code encode the message 00001110000?
(iii) How does it decode 110010101100111?
(iv) How does it decode 011110001010001?
(Assume the check digits are in positions 1, 2, 4 and 8.)

6. The binary repetition code of length n consists of the two codewords

00 . . . 0 and 11 . . . 1 of length n.
(i) Show that the binary repetition code of length 5 is 2-error-correcting
(that is, given any error in at most two bits the nearest neighbor
decoding provides the correct codeword).
(ii) Generalize to the binary repetition code of length 2n + 1.

7. Use the compressed genetic code to encode the following:

(i) AAGAACAACATAAACCACTAA
(ii) ACAAAACAAGAAACTAACAA

8. Use the compressed genetic code to decode the following:

(i) 00010001110100010000111000110010
(ii) 00000011011010010110001010

Problems

Problem 1: The n-dimensional hypercube is a graph with 2n vertices cor-

responding to the (0, 1)-vectors of length n. Vertices are adjacent if and only
if their vectors differ in exactly one position. Explain how block codes corre-
spond to subgraphs of hypercubes.
Problem 2: Write out a formal proof of Theorem 12.2.
240 Introduction to Combinatorics

Problem 3: Suppose you construct a code from the Venn diagram code
by adding an eighth symbol to each codeword so as to make the sum of the
symbols even. That is, if the code had an odd number of 1s, the last symbol
is 1; otherwise it is 0. How many errors can this code detect or correct?
Problem 4: You need to send messages in the five symbols A, B, C, D, E.
Among the possible messages, A occurs most often, B second most, then C,
then D, and E is least frequent. Devise a code that will make the average
length of messages as short as possible.
Problem 5: Prove that a code can correct n errors if and only if the minimum
distance between codewords is at least 2n + 1.
Chapter 13
Latin Squares

13.1 Introduction
“There are thirty-six military officers, six each from six different regiments.
The officers from any one regiment include one of each of six ranks (a colonel,
a lieutenant-colonel, a major, a captain, a lieutenant, and a sub-lieutenant).
They wish to parade in a square formation, so that every row and every
column contains one member of each regiment and one officer of each rank.
Is this possible?”
This problem fascinated Euler in the eighteenth century [33], but he could
not solve it. In fact, it was not until 1900 that Tarry [114] proved, by an
exhaustive search, that no solution was possible. (See Appendix C for biogra-
phies of Euler and Tarry.)
Suppose we are given an arrangement of the officers in which every row
and every column contains one representative of each regiment. It can be
represented as a 6×6 array; if the officer in row i, column j belongs to regiment
k, then the array has (i, j) entry k. For example, one possible arrangement of
thirty-six officers from regiments 1, 2, 3, 4, 5, 6 could be represented by

1 3 6 2 4 5
5 1 3 6 2 4
4 5 1 3 6 2
.
3 6 2 4 5 1
6 2 4 5 1 3
2 4 5 1 3 6

Euler discussed these arrays at length in [33], and gave (essentially) the
following definition:
Definition: A Latin square of side (or order ) n is an n × n array based on
some set S of n symbols (treatments), with the property that every row and
every column contains every symbol exactly once. In other words, every row
and every column is a permutation of S.
The arithmetical properties of the symbols in a Latin square are not relevant
to the definition, so their nature is often immaterial; unless otherwise specified,

241
242 Introduction to Combinatorics

we assume a Latin square of order n to be based on {1, 2, . . . , n}. (Other

symbol sets will be used when appropriate.)
Here are some small Latin squares:

1 2 3
1 2
1 2 3 1 .
2 1
3 1 2
There are Latin squares of side 3 that look quite different from the one just
given, for example
1 2 3
3 1 2 .
2 3 1
but this can be converted into the given 3 × 3 square by exchanging rows 2
and 3.
As we said, the arithmetical properties of the symbols are not considered in
the definition of a Latin square. However, many examples arise from simple
arithmetic and algebra.
Example 13.1: Use modular arithmetic to construct a Latin square of
order 5.
Consider addition modulo 5. If we write Z5 = {0, 1, 2, 3, 4} for the set of
integers modulo 5, the addition table looks like

0 1 2 3 4
0 0 1 2 3 4
1 1 2 3 4 0
.
2 2 3 4 0 1
3 3 4 0 1 2
4 4 0 1 2 3

The core of this,

0 1 2 3 4
1 2 3 4 0
2 3 4 0 1 ,
3 4 0 1 2
4 0 1 2 3
is a Latin square of order 5. 
This same method works if 5 is replaced by any positive integer n, so:

THEOREM 13.1 (Existence of Latin Squares)

There exists a Latin square of every positive integer order.
 Latin Squares 243

Multiplication in modular arithmetic can sometimes give rise to a Latin

square. In this case the square is of order one less than the modulus, because
the zero element must be deleted, or else there would be a row and column of
all zeroes.

Example 13.2: Use multiplication modulo 5 to construct a Latin square

of order 4.

The multiplication table is

0 1 2 3 4
0 0 0 0 0 0
1 0 1 2 3 4
2 0 2 4 1 3
3 0 3 1 4 2
4 0 4 3 2 1

and if we delete the labels and the row and column corresponding to 0, we
obtain
1 2 3 4
2 4 1 3
3 1 4 2
4 3 2 1

which is called the core of the table. It is a Latin square, and symbol 0 never
occurs. 

This construction is not available for all positive integers; for example,
replacing 5 by 4, the 0 occurs in an inappropriate place:

0 1 2 3
0 0 0 0 0 1 2 3
1 0 1 2 3 ⇒ 2 0 2 .
2 0 2 0 2 3 2 1
3 0 3 2 1

If we replace 5 by n, the “multiplication table” construction yields a Latin

square if and only if n is prime.
These considerations lead us to treat Latin squares as the tables of binary
operations.
Definition: A quasigroup (Q, ∗) consists of a set Q together with a binary
operation ∗ such that, for each pair x and y of elements of Q there exist unique
elements u and v of Q for which

x ∗ u = y = v ∗ x.
244 Introduction to Combinatorics

THEOREM 13.2 (Latin Squares and Quasigroups)

Suppose Q = {x1 , x2 , . . . , xn } is an n-set and ∗ is a binary operation on Q
such that (Q, ∗) is a quasigroup. Define an n × n array A by setting aij = k
when xi ∗ xj = xk . Then A is a Latin square.

Proof. Suppose some row, say row i of A, contains a repetition. Then

aij = aih = k for some h, j and k. Then xi ∗ xj = xk = xi ∗ xh . Therefore
xj is the unique element such that xi ∗ xj = xk ; and so is xh . So xj = xh .
This implies that j = h, so k only appears once in row i. Since i and k
were arbitrary, this means each element appears exactly once in each row.
Similarly, no column can contain a repeated element. As every cell of A is
occupied by an element of {1, 2, . . . , n}, each element occurs precisely once,
and A is a Latin square, as required. 
It is clear that a Latin square on symbol-set S will serve as the operation
table for a quasigroup based on S, so the two concepts are equivalent. Theo-
rem 13.1 is a restatement of the fact that the set Zn of integers modulo n form
a quasigroup under addition for any positive integer n. Other quasigroups in-
clude the integers Z, under addition or under subtraction, the reals R under
addition, subtraction or multiplication, and the non-zero reals (often denoted
R∗ ) under multiplication or division. The set Z∗n , the integers modulo n with
zero omitted, form a quasigroup under multiplication (or under division) if
and only if n is prime.
A quasigroup containing an identity element for ∗, an element e such that
x ∗ e = x = e ∗ x for every element x, is called a loop. If the underlying set
is ordered so that e is the first element, then the corresponding Latin square
has its first row and first column identical.
Two special features of quasigroups can be interpreted as properties of
Latin squares. If the operation ∗ on the quasigroup is commutative in the
ordinary algebraic sense then the quasigroup is called Abelian and the Latin
square is symmetric (when considered as a matrix). If the quasigroup is
idempotent (x ∗ x = x for all x ∈ Q) then the entries on the main diagonal
of the Latin square are all different. A set of elements of a Latin square is
called a transversal if it contains exactly one member of each row, one member
of each column, and exactly one copy of each symbol. So the Latin square
from a quasigroup contains a rather special transversal, one that lies on the
main diagonal of the square. We shall discuss transversals further in the next
section.
It is also significant if the operation of a quasigroup is associative (x ∗ (y ∗
z) = (x ∗ y) ∗ z for all x, y, z). An associative quasigroup is a semigroup and
an associative loop is a group. Groups are widely studied objects, and most
readers will be familiar with group theory. However, this significance is more
algebraic than combinatorial.
 Latin Squares 245

13.2 Orthogonality
We now return to Euler’s problem of the 36 officers. As we said, the problem
cannot be solved. However, if we replace 36 by 9, and have three regiments
a, b, c with three ranks α, β, γ represented in each, there is a solution. An
example is
aα bβ cγ
bγ cα aβ .
cβ aγ bα
This array was constructed by superposing two 3 × 3 Latin squares, namely

a b c α β γ
b c a and γ α β .
c a b β γ α

The two squares have the following property: The positions occupied by a
fixed symbol in the first square form a transversal in the second square. (For
example, the positions containing a in rows 1, 2 and 3 contain α, β, and γ,
respectively in the second.) We say the second Latin square is orthogonal to
the first.
Definition: Suppose A and B are Latin squares of side n based on symbol
sets SA and SB , respectively. B is orthogonal to A (written B ⊥ A) if, for
every x ∈ SA , the set of n positions in A occupied by x contain every member
of SB .
It is clear that each member of SB will occur precisely once in the cells
occupied by a fixed element in A.
Equivalently, suppose we construct from A and B an n × n array (A, B) of
ordered pairs, where (x, y) occurs in position (i, j) if and only if x occurs in
position (i, j) of A and y occurs in position (i, j) of B, then B is orthogonal to
A if every possible ordered pair with first element in SA and second element
in SB occurs in the new array.

THEOREM 13.3 (Symmetry of Orthogonality)

Orthogonality is symmetric; that is, B is orthogonal to A implies A is or-
thogonal to B.

Proof. Suppose A and B are Latin squares of side n based on symbol sets SA
and SB , respectively, and B is orthogonal to A. Write SA = {x1 , x2 , . . . , xn ,
and select y ∈ SB . There is exactly one position in which A contains x1 and
B contains y (put x = x1 in the definition). There is also one position where
A contains x2 and B contains y (put x = x2 ). In this way we see that every
246 Introduction to Combinatorics

member of SA occurs once with y. This proof can be applied to every member
y of SB . So A is orthogonal to B. 
The array formed by superposing two orthogonal Latin squares is sometimes
called a Graeco-Latin square. (In fact, the name “Latin square” comes from
Euler’s habit of using the Roman (that is, Latin) alphabet for the symbols
of a square; and when representing a pair of orthogonal squares, he used the
Greek alphabet for the second square, just as we did.) But it is more common
to speak of “a pair of orthogonal Latin squares,” and to use the symbols
{1, 2, . . . , n} for each. We have exhibited a pair of orthogonal Latin squares
of order 3, and stated that there is no pair of orthogonal Latin squares of
order 6. One can ask: for what orders do there exist a pair of orthogonal
Latin squares? Can we find three or more Latin squares that are mutually
orthogonal —each one is orthogonal to all the others?
Before we discuss this problem, we define a standardized Latin square to
be a Latin square whose symbols are the first n positive integers, and whose
first row is
1 2 3 ... n . (13.1)
If A is any Latin square, we can convert it to standardized form by renam-
ing the symbols. If A and B are orthogonal, we can assume that both are
standardized; we rename the symbols in each separately.

THEOREM 13.4 (Maximum Number of MOLS)

If there are k mutually orthogonal Latin squares (MOLS) of side n, n > 1,
then k < n.

Proof. Suppose A1 , A2 , . . . , Ak are Latin squares of side n, n > 1, each of

which is orthogonal to each other one. Without loss of generality, assume that
each has been standardized, so that each has first row (13.1). Write ai for
the (2, 1) entry in Ai . No ai can equal 1 (since the first columns can contain
no repetition), and the ai must be different (if ai = aj , then the n cells that
contain ai in Ai must contain a repetition in Aj ; both the (1, ai ) and (2, 1)
cells of Ai and Aj contain ai ). So {a1 , a2 , . . . , ak } contains k distinct elements
of {2, 3, . . . , n}, and k < n. 
Let us write N (n) for the number of squares in the largest possible set of
mutually orthogonal Latin squares of side n. In this notation, we have just
shown
N (n) ≤ n − 1 if n > 1.
(If n = 1, we can take A1 = A2 = . . . and it makes sense to write “N (1) = ∞”
in some situations. Whenever a theorem requires the existence of at least k
mutually orthogonal Latin squares of side n, the conditions are satisfied by
n = 1 for any k. In other applications it is more appropriate to say N (1) = 1.
To avoid this problem, many authors only define orthogonality for side greater
than 1.)
 Latin Squares 247

For example, N (4) ≤ 3. In fact, N (4) = 3; one set of three mutually

orthogonal Latin squares of side 4 is

1 2 3 4 1 2 3 4 1 2 3 4
2 1 4 3 3 4 1 2 4 3 2 1
.
3 4 1 2 4 3 2 1 2 1 4 3
4 3 2 1 2 1 4 3 3 4 1 2

On the other hand, the theorem tells us only that N (6) ≤ 5, but we know
from Tarry’s result that N (6) = 1.

THEOREM 13.5 (Number of MOLS, Prime Power Order)

N (n) = n − 1 whenever n is a prime power.

Proof. Suppose n is prime. We define a set of n − 1 mutually orthogonal

Latin squares A1 , A2 , . . . , An−1 of order n. Let us write Ah = (ahij ). We define

ahij = i + hj,

where i, h and j are integers modulo n (remainder 0 is written as n, not 0).

Then Ah is easily seen to be a Latin square, and the n − 1 Latin squares are
orthogonal.
The case where n is a prime power is similar, but uses the existence of a
finite field with n elements. 
The Cartesian product of two Latin squares A and B is defined as follows.
If a is any symbol in A, write (a, B) for the array derived from B by replacing
each entry x by the ordered pair (a, x). Then replace every occurrence of a
in A by (a, B).
For example, suppose A and B are Latin squares of order 2:

1 2 x y
A= B=
2 1 y x

Then the Cartesian product A × B is

(1, x) (1, y) (2, x) (2, y)

(1, y) (1, x) (2, y) (2, x)
A×B = .
(2, x) (2, y) (1, x) (1, y)
(2, y) (2, x) (1, y) (1, x)

In general, if A is a Latin square of order n and B is a Latin square of order r

then A× is a Latin square of order nr; if A is based on symbol-set S and B is
based on symbol-set T , then the symbol-set for A×B is the Cartesian product
S × T , as defined in Section 1.2 (which is why the name Cartesian product
248 Introduction to Combinatorics

is used for these Latin squares). A similar idea—the Cartesian product of

matrices—is defined in Appendix B.
It is easy to see that the Cartesian product of Latin squares is a Latin
square, and moreover the taking of Cartesian products preserves orthogonal-
ity: if A1 is orthogonal to A2 and B1 is orthogonal to B2 , then A1 × B1 is
orthogonal to A2 × B2 . Therefore

N (nr) ≥ min{N (n), N (r)}. (13.2)

So we have the following result:

THEOREM 13.6 (MacNeish’s Theorem) [70]

Suppose n = pa1 1 pa2 2 . . . par r , where the pi are distinct primes and each ai ≥ 1.
Then
n − 1 ≥ N (n) ≥ min(pai i ).

We know that there cannot be a pair of orthogonal Latin squares of order

6, and order 2 is also impossible. Are there any further values n for which
N (n) = 1? Euler thought so; he conjectured that N (n) = 1 whenever n ≡ 2
modulo 4. However, the opposite is true:

THEOREM 13.7 (Falsity of Euler’s Conjecture) [11]

N (n) ≥ 2 for every integer n > 6.

The proof can be found in most books on design theory (such as [123]).
The history of this problem and its solution can be found in [39].
Further results are known. For example, many authors have contributed to
the proof that:

THEOREM 13.8 (Three MOLS)

There exist three mutually orthogonal Latin squares of every side except 2, 3,
6, and possibly 10.

A proof appears in [123]. In general, for any given constant k there exists
a lower bound nk such that N (n) ≥ k whenever n > nk , but the only known
value is n2 = 6; we do not know N (10), so n3 may be either 10 or 6. For
the best known lower bounds on nk , as well as a table of the largest known
numbers of Latin squares of orders less than 10,000, see [22].
A Latin square with no orthogonal mate is called a bachelor square. We
already know of bachelor squares of sides 2 and 6. (In fact, any Latin square
of these orders must be a bachelor.) At side 3, there is no bachelor.
Norton [78] (see also [95]) listed the isomorphism classes of Latin squares
of side 7; there are 147 in all. There is one set of six mutually orthogonal
 Latin Squares 249

squares (all six squares are isomorphic); no other set of three or more mutually
orthogonal square exists, and only five of the remaining classes contain squares
with mates. So there are 141 isomorphism classes of bachelor squares. This
led to speculation that bachelor squares might be very common, but empirical
evidence suggests that for orders larger than 7, almost all Latin squares have
mates. However, there are bachelor squares of all orders other than 3 and 7.
The proof for odd orders may be found in [72] and [124].
To discuss even orders we define a function

∆A (i, j) ≡ i + j − aij (mod n).

LEMMA 13.9
If T is a transversal in A, the sum of ∆A (i, j) over the entries aij of T is
1
2 n if n is even and 0 if n is odd.

Proof. Suppose cells (1, j1 ), (2, j2 ), . . . , (n, jn ) form a transversal. As they

form a transversal, their entries are {1, 2, . . . , n} in some order. So
n
X
∆A (i, ji ) = (1 + 2 + . . . + n) + (1 + 2 + . . . + n) − (1 + 2 + . . . + n)
i=1
= (1 + 2 + . . . + n)
1
= 2 n(n + 1). 

THEOREM 13.10 (Latin Squares Without Transversals)

Let A be the Latin square of side 2t defined by

aij ≡ i + j (mod 2t), 1 ≤ aij ≤ 2t.

Then A contains no transversal.

Proof. Suppose cells (1, j1 ), (2, j2 ), . . . , (2t, j2t ) form a transversal. From
Lemma 13.9 X
∆A (i, ji ) = t.
However, aiji ≡ i + ji , so
X
∆A (i, ji ) ≡ 2(1 + 2 + . . . + 2t) ≡ 0 (mod 2t),

a contradiction. 
Obviously these squares have no orthogonal mates.
We conclude this section by exhibiting some small examples of orthogonal
Latin squares whose orders are not prime powers.
250 Introduction to Combinatorics

The transpose of a Latin square is a Latin square of the same order, so it

is possible to ask whether the two squares are orthogonal. A Latin square
that is orthogonal to its transpose is called self-orthogonal . It is in fact not
difficult to construct self-orthogonal Latin squares.
We have not yet seen a pair of orthogonal Latin squares of order 10, so we
now exhibit a self-orthogonal square of that order:

1 2 3 4 5 6 7 8 9 X .
2 X 4 8 6 3 5 9 7 1
3 6 7 1 8 9 2 5 X 4
4 7 X 5 2 1 9 6 8 3
5 3 8 X 9 4 6 1 2 7
6 9 1 3 7 2 8 X 4 5
7 8 6 9 3 X 1 4 5 2
8 5 9 2 X 7 4 3 1 6
9 4 2 6 1 5 X 7 3 8
X 1 5 7 4 8 3 2 6 9

Another interesting example of orthogonal Latin squares is the following:

THEOREM 13.11 (Latin Squares of Side 12) [54]

There is a set of five mutually orthogonal Latin squares of side 12.

Proof. One of the Latin squares is

1 2 3 4 5 6 7 8 9 X E D
2 3 4 5 6 1 8 9 X E D 7
3 4 5 6 1 2 9 X E D 7 8
4 5 6 1 2 3 X E D 7 8 9
5 6 1 2 3 4 E D 7 8 9 X
6 1 2 3 4 5 D 7 8 9 X E
.
7 8 9 X E D 1 2 3 4 5 6
8 9 X E D 7 2 3 4 5 6 1
9 X E D 7 8 3 4 5 6 1 2
X E D 7 8 9 4 5 6 1 2 3
E D 7 8 9 X 5 6 1 2 3 4
D 7 8 9 X E 6 1 2 3 4 5

Each of the other squares is obtained from this by column permutation; the
four permutations give the following four first rows:

[1 7 9 3 8 2 X D 5 E 6 4 ],
[1 4 7 2 X D 3 9 6 5 8 E ],
[1 9 2 D 6 X 4 E 3 8 7 5 ],
[1 5 D E 3 8 9 7 X 2 4 6 ]. 
 Latin Squares 251

THEOREM 13.12 (Latin Squares of Side 15) [97]

N (15) ≥ 4.

Proof. The first rows of the squares are

[1 15 2 14 3 13 4 12 5 11 6 10 7 9 8 ],
[1 14 3 11 6 9 8 7 10 4 13 12 5 15 2 ],
[1 10 7 13 4 2 15 6 11 9 8 3 14 12 5 ],
[1 6 11 10 7 15 2 5 12 14 3 9 8 4 13 ].

In each square the later rows are formed by developing the first row modulo
15. 

13.3 Idempotent Latin Squares

An n × n Latin square is called a transversal square if its diagonal entries
form a transversal—that is, they are all different. In particular a transversal
square whose main diagonal is

(1, 2, . . . , n) (13.3)

is called an idempotent Latin square.

LEMMA 13.13
There exists a set of N (n) − 1 mutually orthogonal idempotent Latin squares
of side n.

Proof. Suppose A1 , A2 , . . . , AN (n) are mutually orthogonal Latin squares

of side n based on 1, 2, . . . , n. By permuting the columns, transform AN (n)
so that the element 1 appears in all the diagonal cells. Carry out the same
column permutation on the other squares. The diagonals of the first N (n) − 1
new squares will be transversals. Now, in each square, permute the names
of the elements so as to produce the diagonal (13.3). We have the required
orthogonal idempotent Latin squares. 
The above proof embodies a simple but useful fact. If you need an idempo-
tent Latin square, it is sufficient to construct a transversal square. The names
of entries can then be permuted to make it idempotent.
We shall write I(n) for the maximal cardinality of a set of mutually orthog-
onal idempotent Latin squares of side n. Using Lemma 13.13, we see that for
every n,
N (n) − 1 ≤ I(n).
252 Introduction to Combinatorics

On the other hand, I(n) cannot possibly be larger than N (n). So

N (n) − 1 ≤ I(n) ≤ N (n). (13.4)

As easy generalizations of earlier results, we have:

THEOREM 13.14 (Orthogonal Idempotent Latin Squares)

If n > 1 then I(n) ≤ n − 2, with equality when n is a prime power.

The proof that I(n) ≤ n − 2 is similar to that of Theorem 13.4, and is

left as an exercise. The other part follows because all but one of the squares
constructed in Theorem 13.5 are in fact idempotent.
The only known cases where I(n) = N (n) are the trivial case n = 1 (some
authors would omit this case, saying that neither N (1) nor I(1) is defined)
and
I(6) = N (6) = 1.
(We know that N (6) = 1, so I(6) = 1 or 0. Proving the existence of an
idempotent Latin square of side 6 is again an exercise.)

THEOREM 13.15 (Orthogonal Idempotent Latin Squares, Com-

posite Order)
For every m and n, I(mn) ≥ min(I(m), I(n)). If n has prime power decom-
position
n = pa1 1 pa2 2 . . . pakk ,
then I(n) ≥ min(pai i − 1) − 1.

Proof. If A and B are idempotent Latin squares, then the Cartesian product
of A and B is a transversal square. The proof now follows that of Theorem
13.6. 
As a consequence of Theorem 13.8, we have:

THEOREM 13.16 (Orthogonal Idempotent Latin Squares, Lower

Bound)
There is a pair of orthogonal idempotent Latin squares of every order except
2, 3 and 6.

Proof. The only case to be investigated is order 10. But the square of order
10 given in the preceding section is a transversal square and is orthogonal to
its transpose, so these two squares are a pair orthogonal idempotent Latin
squares of that order. 
 Latin Squares 253

13.4 Partial Latin Squares and Subsquares

A partial Latin square of side n is an n × n array in which some cells are
filled with members of some n-set while others are empty, such that no row
or column contains a repeated element. In 1960, Evans [34] conjectured that
any partial Latin square of side n that has n − 1 or fewer cells occupied can be
completed, that is, the empty cells can be filled in so that the result is a Latin
square. This turned out to be a difficult problem, but in 1981 Smetaniuk [105]
showed that the Evans conjecture was true. This is the best possible result.
It is easy to find an n-element example that cannot be completed.
Probably the partial Latin square most familiar to our readers is the Sudoku
puzzle, which we mentioned in Chapter 1. This consists of a 9 × 9 array,
partitioned into 3 × 3 subarrays, in which some cells contain entries chosen
from {1, 2, 3, 4, 5, 6, 7, 8, 9}. The array is in fact a partial Latin square, and the
object is to complete the Latin square. However, there is a further condition:
each subarray must contain the nine numbers precisely once each.
We defined a Sudoku square to be a Latin square that could be the solution
to a Sudoku puzzle: a 9 × 9 Latin square partitioned into 3 × 3 subarrays,
where each subarray contains each of {1, 2, . . . , 9} precisely once. In a well-
posed Sudoku puzzle, there is one and only one Sudoku square that contains
the original partial Latin square.
There is no reason to restrict Sudoku to the 9 × 9 case. We define a (p, q)-
square to be a Latin square of side pq partitioned into p × q subarrays, such
that each subarray contains all integers from 1 to pq once each. For example,
here is a very simple (2, 3)-square:

1 2 3 4 5 6
4 5 6 1 2 3
2 3 1 5 6 4
5 6 4 2 3 1
3 1 2 5 6 4
5 6 4 3 1 2
Among partial Latin squares, the Latin rectangle is of special interest. A
Latin rectangle of size k × n is a k × n array with entries from {1, 2, . . . , n}
such that every row is a permutation and the columns contain no repetitions.
For example, the first k rows of an n × n Latin square form a Latin rectangle.
Clearly, k can be no larger than n.
The following important theorem is a consequence of the results on systems
of distinct representatives (SDRs), which we discussed in Section 1.1.5.

THEOREM 13.17 (Completing Idempotent Latin Rectangles)

If A is a k × n Latin rectangle, then one can append (n − k) further rows to
254 Introduction to Combinatorics

A so that the resulting array is a Latin square.

Proof. If k = n, the result is trivial. So assume that k < n. Write S for

{1, 2, . . . , n} and define Sj to be the set of members of S not occurring in
column j of A. Each Sj has k elements. Now consider any element i of S.
The k occurrences of i in A are in distinct columns, so i belongs to exactly
(n − k) of the sets Sj .
Suppose there were r of the sets Sj whose union contained fewer than r
members, for some r. If we wrote a list of all members of each of these r
sets, we would write down r(n − k) symbols, since each Sj is an (n − k)-set.
On the other hand, we would write at most r − 1 different symbols, and each
symbol occurs in exactly n− k sets, so there are at most (r − 1)(n− k) symbols
written. Since k < n, (r − 1)(n − k) < r(n − k), a contradiction. So there is
no set of r of the Sj that contain between them fewer than r symbols. By
Theorem 14.7, there is a system of distinct representatives for the Si . Say the
representative of Sj is ij . Then (i1 , i2 , . . . , in ) is a permutation of S. If we
append this to A as row k + 1, we have a (k + 1) × n Latin rectangle. The
process may be repeated. After (n − k) iterations we have a Latin square with
A as its first k rows. 
A subsquare of side s in a Latin square of side n is an s × s subarray that is
itself a Latin square of side s. A necessary and sufficient condition for an s × s
subarray to be a subsquare is that it should contain only s different symbols.

COROLLARY 13.17.1
If L is a Latin square of side s and n ≥ 2s, then there is a Latin square of
side n with L as a subsquare.
Proof. Say n = s + t. Define M to be the s × t array
 
s+1 s+2 ... s+t
 s+2 s +3 ... s+1 
M = . .. ..
 
 ..

. . 
s+s s+s+1 ... s+s−1

with entries reduced modulo t, where necessary, to ensure that they lie in the
range (s + 1, s + 2, . . . , s + t). Write A for the array constructed by laying row
i of M on the right of row i of L, for 1 ≤ i ≤ s:

A= L M .

Then A is an s × n Latin rectangle; if we embed it in an n × n Latin square,

that square has L as a subsquare. 
This is the best possible result (see the exercises), provided one ignores the
trivial case s = n.
 Latin Squares 255

One can also ask whether or not a given array can be embedded in a Latin
square.
Example 13.3: In each case, can the given array be embedded in a 7 × 7
Latin square? If so, give an example.

1 2 5 6 7 1 2 5 6 7
5 1 2 7 3 5 1 2 7 3
A= 3 7 1 4 2 B= 3 7 1 4 2
2 5 6 1 4 2 5 6 1 4
6 3 4 2 1 7 3 4 2 1

We can assume the given array will appear in the top left corner of the
Latin square, if one exists. Symbols 1 and 2 appear five times each in A.
That means there can be only two more occurrences of each; the only way
will be for them to appear in the bottom right corner of the square, forming
a 2 × 2 subsquare. One example is

1 2 5 6 7 3 4
5 1 2 7 3 4 6
3 7 1 4 2 6 5
A= 2 5 6 1 4 7 3 .
6 3 4 2 1 5 7
4 6 7 3 5 1 2
7 4 3 5 6 2 1

For B there is no solution: 6 occurs only twice in B, so it would have to

appear five more times. But the maximum is four (2 rows, 2 columns). 

13.5 Applications
Designing Experiments Using Latin Squares
Latin squares provide a model for experiments where there are several dif-
ferent factors. As an example, consider a motor oil manufacturer who claims
that using its product improves gasoline mileage in cars. Suppose you wish
to test this claim by comparing the effect of four different oils on gas mileage.
However, because the car model and the drivers’ habits greatly influence the
mileage obtained, the effect of an oil may vary from car to car and from driver
to driver. To obtain results of general interest, you must compare the oils in
several different cars and with several different drivers. If you choose four car
models and four drivers, there are 16 car-driver combinations. The type of
car and driving habits are so influential that you must consider each of these
256 Introduction to Combinatorics

16 combinations as a separate experimental unit. If each of the four oils is

used in each unit, 4 × 16, or 64, test drives are needed.
In this case there are three factors; the oil to be used in a test run is one,
and so are the driver and the type of car.
An experiment is designed using a 4 × 4 Latin square. Suppose the cars are
denoted X1 , X2 , X3 , X4 , and the drivers are Y1 , Y2 , Y3 , Y4 . Call the oils Z1 ,
Z2 , Z3 , Z4 . We call these the four levels of the factors. We assign one oil to
each of the 16 car-driver combinations in the following arrangement:

Y1 Y2 Y3 Y4
X1 Z1 Z2 Z3 Z4
X2 Z2 Z1 Z4 Z3 .
X3 Z3 Z4 Z1 Z2
X4 Z4 Z3 Z2 Z1

For example, the cell in the row marked X2 and the column marked Y3
contains entry Z4 . This means, in the test drive when car X2 is driven by Y3 ,
oil Z4 is to be used. The diagram is a plan for 16 test drives, one for each
combination of car and driver. It tells us which oil to use in each case.
The array underlying the experiment is a Latin square, so each oil appears
four times, exactly once in each row (for each car) and also exactly once in
each column (for each driver). This setup can show how each oil performs with
each car and with each driver, while still requiring only 16 test drives, rather
than 64. If 64 test drives are available, one technique is to use a Latin square
to schedule 16 tests, and then running each test four times. This enables the
experimenters to make a more reliable estimate of the underlying variability
due to uncontrollable factors (sometimes called “error”), which improves the
statistical reliability of the experiment.
This technique can be used with any size Latin square. Suppose there are
three types of treatment, X, Y and Z, and each is to be applied at n levels
(X at levels X1 , X2 , . . . , Xn , and so on). An n × n Latin square provides n3
comparisons with n2 tests; if the (i, j) element is k, then treatment levels Xi ,
Yj and Zk are applied in one test.
Further savings can be achieved by using orthogonal Latin squares. In
the motor oil example, one might wonder whether automobile performance
depends on the type of roads. So four different test routes could be chosen:
one primarily urban, one about two-thirds urban and one-third interstate,
one two-thirds interstate and one all interstate. If these routes are denoted
T1 , T2 , T3 and T4 , then the following design allows us to test all pairs of levels
of the four factors in 16, rather than 256, tests:
 Latin Squares 257

Y1 Y2 Y3 Y4
X1 Z1 T1 Z2 T2 Z3 T3 Z4 T4
X2 Z2 T3 Z1 T4 Z4 T1 Z3 T2 .
X3 Z3 T4 Z4 T3 Z1 T2 Z2 T1
X4 Z4 T2 Z3 T1 Z2 T4 Z1 T3

The design was constructed from a pair of orthogonal Latin squares.

In experiments where interaction between the different factors is unlikely,
orthogonal Latin squares allow a large amount of information to be obtained
with relatively little effort. This idea has been applied in a number of areas
other than experimentation. One interesting application is the use of orthog-
onal Latin squares in compiler testing; for details, see [71].

Golf Designs
We conclude with an application to tournament scheduling. The teams
representing 2n + 1 golf associations wish to play a round robin tournament.
There are 2n+1 rounds; each team serves as “host” for the one round in which
they do not play, and the n matches in that round are played (simultaneously)
on the course of the host team.
The teams wish to play 2n − 1 complete tournaments in such a way that
every pair of teams play exactly once on each of the 2n − 1 available courses.
(We shall call such an arrangement a golf design.) When is this possible?
This question arose in planning an actual golf tournament in New Zealand.
It was reported by Robinson [89], who pointed out that the problem of con-
structing one round robin is equivalent to finding an idempotent symmetric
Latin square; the square has (i, j) entry k, when i 6= j, if and only if i plays j
at k. So the whole problem is equivalent to finding 2n − 1 symmetric idempo-
tent Latin squares such that the 2n − 1 (i, j) entries cover {1, 2, . . . , 2n + 1}
with i and j deleted, for all i 6= j. He proved that case n = 5 has no solution
and exhibited a solution for case n = 7 (see Figure 13.1).
We can use a Steiner triple system of order v to construct a v × v Latin
square L = (lij ). We set lii = i for all i. If i 6= j, find the unique triple that
contains both i and j. If it is {i, j, k}, then set lij = k. The resultant square
L is idempotent and symmetric.
Tierlinck [115, 116] proved that if a golf design exists for order 11, then it
exists for all odd orders other than 5 and possibly 41. Subsequently, Colbourn
and Nonay [23] found an example of order 11, and recently Chang [20] has
constructed one of order 41. So the problem is completely solved.
To solve particular cases, one could try to construct symmetric idempotent
Latin squares cyclically. If the first row of such a square is 0, a1 , a2 , . . . , a2n
258 Introduction to Combinatorics

0 2 4 5 1 6 3 0 3 6 1 5 4 2 0 4 5 6 2 3 1
2 1 6 4 5 3 0 3 1 5 0 6 2 4 4 1 3 2 0 6 5
4 6 2 0 3 1 5 6 5 2 4 1 0 3 5 3 2 1 6 4 0
5 4 0 3 6 2 1 1 0 4 3 2 6 5 6 2 1 3 5 0 4
1 5 3 6 4 0 2 5 6 1 2 4 3 0 2 0 6 5 4 1 3
6 3 1 2 0 5 4 4 2 0 6 3 5 1 3 6 4 0 1 5 2
3 0 5 1 2 4 6 2 4 3 5 0 1 6 1 5 0 4 3 2 6

0 5 3 2 6 1 4 0 6 1 4 3 2 5
5 1 4 6 3 0 2 6 1 0 5 2 4 3
3 4 2 5 0 6 1 1 0 2 6 5 3 4
2 6 5 3 1 4 0 4 5 6 3 0 1 2
6 3 0 1 4 2 5 3 2 5 0 4 6 1
1 0 6 4 2 5 3 2 4 3 1 6 5 0
4 2 1 0 5 3 6 5 3 4 2 1 0 6

FIGURE 13.1: A golf design for seven clubs.

then the square would be

0 a1 a2 a3 ... a2n
a2n + 1 1 a1 + 1 a2 + 1 ... a2n−1 + 1
.
a2n−1 + 2 a2n + 2 2 a1 + 2 ... a2n−2 + 2
... ...

For symmetry we would require a2n+1−j = aj − j (mod 2n + 1). In order

that the rows be Latin we need 0, a1 , a2 , . . . , an , an −n, an−1 −(n−1), . . . , a1 −1
to be distinct modulo 2n + 1; a similar condition on columns exists. For a golf
design we need 2n − 1 such sequences with different entries in all nondiagonal
positions. This property can be derived from the starter rows. We have the
following:

THEOREM 13.18 (Circulant Golf Arrays) [121]

Suppose there exists a (2n−1)×(2n+1) array A with the following properties:
(a) when 1 ≤ k ≤ n, ai,n+k = ai,n+1−k + n + k;
(b) ai0 = 0;
(c) aij + k 6= aik + j when j 6= k;
(d) each row of A is a permutation of {0, 1, . . . , 2n};
(e) column j of A is a permutation of {1, 2, . . . , 2n}\{j}.
(All arithmetic is carried out modulo 2n + 1.) Then a golf design for 2n + 1
teams exists.
 Latin Squares 259

2 5 9 14 16 13 15 12
3 1 10 12 11 15 4 13
4 7 12 2 13 16 1 14
5 8 2 13 15 1 14 11
6 9 15 3 1 14 11 10
7 16 13 8 3 11 2 9
8 14 16 9 6 5 10 2
9 4 14 16 10 7 3 6
10 13 4 7 2 12 5 16
11 15 7 5 14 8 6 3
12 10 5 1 4 9 13 15
13 6 8 15 7 3 16 1
14 11 1 6 8 10 12 7
15 3 11 10 12 5 9 4
16 12 6 4 9 2 8 5

FIGURE 13.2: Array to generate a golf design for 17 teams.

The 2n − 1 Latin squares are derived by developing the rows of A modulo

2n + 1.
The point of this theorem is that it yields a relatively easy computer con-
struction. To construct row i of A it is only necessary to select the n entries
ai1 , ai2 , . . . , ain . Then the row is

0, ai1 , ai2 , . . . , ain , (ain + n + 1), . . . , (ai2 + 2n − 1), (ai1 + 2n)

(reduced mod 2n + 1). Moreover, you may as well fix ai1 to be i + 2. Then
properties (c) through (e) restrict the search considerably.
No solutions exist for 2n + 1 = 5, 7 or 11. It is found that there is a unique
solution for 2n + 1 = 9, and exactly four solutions exist for 2n + 1 = 13.
A solution for 2n + 1 = 17 is illustrated in Figure 13.2. Only columns 1
through 8 of A are shown. Column 0 consists of all zeros; columns 9 through
16 are derived from columns 8 through 1 by adding (9, 10, 11, 12, 13, 14, 15, 16)
modulo 17.

Exercises 13A
1. Write down the Latin square derived from the addition of integers mod-
ulo 6.

2. Verify that the multiplication table of integers modulo 6 does not pro-
duce a 5 × 5 Latin square.
260 Introduction to Combinatorics

3. A reduced Latin square of side n is one in which the first row and first
column are (1, 2, . . . , n) in that order.
(i) Prove that every Latin square is equivalent to a reduced Latin
square.
(ii) Say there are rn different reduced Latin squares of side n. Prove
that the number of Latin squares of side n is

n!(n − 1)!rn .

(iii) Find rn for n = 3 and 4.

(iv) Exhibit all reduced Latin squares of order 4.

4. Prove that the Cartesian product of Latin squares is a Latin square,

and that the Cartesian product of transversal squares is a transversal
square.

5. Recall that a Latin square is called self-orthogonal if it is orthogonal to

its own transpose.
(i) Prove that the diagonal elements of a self-orthogonal Latin square
of side n must be 1, 2, . . . , n in some order.
(ii) Prove that there is no self-orthogonal Latin square of side 3.
(iii) Find self-orthogonal Latin squares of sides 4 and 5.

6. Walter Johns and John Walters wish to invite 25 colleagues to an inter-

disciplinary conference. These colleagues will include five people from
each of five universities A, B, C, D, and E; each school will send one
person from each of five disciplines, d1 through d5 . For the introductory
talks, they wish to seat these people in a five-by-five grid so that no
row or column contains two people from the same discipline or from the
same college.
(i) Find a seating plan that accomplishes this.
(ii) The football teams from College A and College B are blood rivals.
Can you come up with a seating plan in which no member of College
A sits next to a member of College B (to avoid fights)?
(iii) Can you find a seating plan in which each member of discipline d1
is sitting next to a member of discipline d2 ?
(iv) Can you find a seating plan as required in part (iii) if there are
only four universities and four disciplines involved? Can you find
one that also satisfies the condition of part (ii)?
(In this question “next to” refers only to rows, not columns.)

7. Prove that if a Latin square of side n has a subsquare of side s, where

s < n, then n ≥ 2s.
 Latin Squares 261

8. Without using the result of Theorem 13.17, prove directly that any
(n − 1) × n Latin rectangle can be completed to give an n × n Latin
square.
9. Use SDRs to complete the following Latin square.
1 2 3 4 5
5 1 2 3 4
4 5 1 2 3

10. In each case, can the given array be embedded in a 6 × 6 Latin square?
If so, give an example.
1 5 3 4 1 2 3 4
3 2 1 5 2 1 6 5
(i) (ii)
2 3 6 1 3 6 4 1
6 1 4 2 4 3 2 6
11. In each case, is there a value for x such that the given array can be
embedded in a 6 × 6 Latin square? If so, what is that value? Give an
example of an embedding.
1 2 3 4 1 2 3 4
5 6 1 2 4 6 1 2
(i) (ii)
3 4 5 1 3 1 6 5
4 1 2 x 6 4 2 x

12. Find a completion of the partial Latin square

2 3 4
3 2 4
4 2 3
.
2 3 1 4
1 4 2 3
4 3 2

13. Consider golf designs of order 5:

(i) Show that there is a unique symmetric Latin square of side 5, up
to isomorphism.
(ii) Prove that there is no golf design of order 5.
14. Suppose A is a Latin square of side n, and is symmetric (that is, aij = aji
for all i and j)
(i) Say n is odd. Show that the diagonal entries are {1, 2, . . . , n} in
some order.
(ii) Show that the above statement is false when n is even.
262 Introduction to Combinatorics

Exercises 13B
1. Write down the Latin square derived from the multiplication of integers
modulo 7.

2. Prove that Z∗n is not a quasigroup under multiplication when n is com-

posite.

3. (For those with sufficient algebraic background.) Prove Theorem 13.5

in the case where n is a prime power but not a prime.

4. Prove that I(n) ≤ n − 2 for every integer n > 1.

5. Show that the array

1
2

3
can be completed to a Latin square.

6. Construct a (2, 2)-square.

7. Use SDRs to complete the following Latin square.

1 2 3 4 5 6
6 1 4 3 2 5
4 5 1 6 3 2

8. Consider the problem in Exercise 11A.6. Suppose Drs. Johns and Wal-
ters wish to invite n2 colleagues from n universities, each of which sends
one person from each of n disciplines. They wish to seat these people
in a square grid so that no row or column contains two people from the
same discipline or from the same university.
(i) For what n is there a seating plan that accomplishes this?
(ii) For what n is there a plan in which no member of College A sits
next to a member of College B?
(iii) For what n is there a plan in which each member of discipline d1
is sitting next to a member of discipline d2 ?

9. Verify that the array in Figure 13.2 yields a golf design for 17 teams.
 Latin Squares 263

Problems

Problem 1: Prove that there is no idempotent Latin square of side 2.

Problem 2: Construct an idempotent Latin square of side 6.
Problem 3: Prove that there exist at least
n!(n − 1)! . . . (n − k + 1)!
k × n Latin rectangles and therefore at least
n!(n − 1)! . . . 2!1!
Latin squares of side n.
Problem 4: Find a 4 × 4 array containing four entries, one each of 1, 2, 3, 4,
that cannot be completed to a Latin square. Generalize your result to show
that the Evans conjecture is the best possible result.
Problem 5: Verify that the construction in Theorem 13.5 produces Latin
squares, and that the squares are orthogonal. (If you have not studied finite
fields, restrict yourself to the case of prime order.)
Problem 6: Suppose we have t orthogonal Latin squares of side n; the kth
square has (i, j) entry akij . Form a new array of order (t + 2) × n2 as follows:
column (i − 1)n + j is the transpose of
.
 
i, j a1ij , a2ij , .., atij .

This array is called a (t + 2) × n2 orthogonal array.

(i) Construct a 4 × 9 orthogonal array.
(ii) Is there a 4 × 36 orthogonal array?
(iii) Construct a set of 16 codewords of length 5 based on the alphabet
{1, 2, 3, 4} such that, given any two words, there is exactly one position
in which they have the same symbol.
Problem 7: (i) Show that the partial Latin square

1 3 5
1 3
4 5 2
3 4 1
4 5

has a unique completion.

264 Introduction to Combinatorics

(ii) Find all completions of the partial Latin square

1 3 5
1 3
1 .
4
5
Chapter 14
Balanced Incomplete Block Designs

Suppose you want to compare six varieties of corn, varieties x1 , x2 , x3 , x4 , x5

and x6 , in order to decide which variety yields the most grain in your area.
You could find a large field, divide it into six smaller areas (plots) and plant
one variety in each plot. Eventually you could weigh the different yields and
make your comparisons.
However, suppose one part of the field is more fertile than another. The
variety planted there would seem better than the other varieties, but you are
only observing the quality of the soil, not the corn. The way round this is
to partition the field into four midsized areas—blocks—so that the fertility of
each plot is close to uniform. You could then plant six plots, one for each
variety, in each block, for a total of 24.
But now there is another problem. Not all seeds germinate equally, and
not all plants grow equally well, even in identical environments. Living things
are not uniform, and do not develop uniformly. If the plots are too small, this
random variation could affect the yields enough to outweigh the differences
you are trying to study. Maybe 24 plots are too many, the individual plots
are too small.
The solution is to make the plots larger; not every variety will appear in
each block. For example, you might restrict each block to three varieties. If
the first block contains x1 , x2 and x3 , it makes sense to compare the yields
for x1 and x2 , in that block, and in every block where both appear.
For conciseness, let us call the blocks B1 , B2 , . . . , B4 . We identify the block
with the set of varieties, so B1 = {x1 , x2 , x3 }. One possible selection of four
blocks is

B1 = {x1 , x2 , x3 }
B2 = {x4 , x5 , x6 }
B3 = {x1 , x2 , x5 }
B4 = {x3 , x4 , x6 }.

We could compare the yield for varieties x1 and x2 twice (using blocks B1 and
B3 ).
Such an arrangement of varieties into blocks is called an experimental de-
sign, because they were first studied as tools for conducting experiments, such
as the agricultural experiment we have outlined, or designs for short. In the

265
266 Introduction to Combinatorics

particular case when all the blocks are the same size (like our example) the
term block design is used.
In the 24-plot example that we mentioned first, each block is a complete
set of all the varieties. This is called a complete design. Designs like our
second example, whose blocks are proper subsets of the varieties, are called
incomplete.
Formally, we define a block design with parameters v, b, k to be a way of
choosing b subsets of size k (called blocks) from a set V of v objects called
varieties (or treatments). We say the design is based on V . A block design
is regular if every variety occurs in the same number of blocks. This is often
a desirable property when the data is analyzed, because we have the same
amount of information about each variety. If a block design is regular, it
is also called a 1-design; the number of blocks in which a treatment occurs
is called the replication number or frequency of the design, and is usually
denoted r.
The word incidence is used to describe the relationship between blocks and
varieties in a design. We say block B is incident with treatment t, or treatment
t is incident with block B, to mean that t is a member of B. When we study
designs, we shall primarily be involved with incidence and other structural
properties of a collection of subsets of the varieties. It makes no difference
whether the varieties are labels representing types of experimental material
(such as corn), or positive integers, or whatever. We are purely interested
in the combinatorics of the situation. For this reason, the objects we are
discussing are often called combinatorial designs.

14.1 Design Parameters

The four design parameters we have defined are not independent:

THEOREM 14.1 (First Parameter Relation)

In a regular block design,
bk = vr. (14.1)

Proof. We count, in two different ways, all the ordered pairs (x, y) such that
variety x belongs to block y. Since every variety belongs to r blocks, there
are r ordered pairs for each variety, so the number is vr. Similarly, each block
contributes k ordered pairs, so the summation yields bk. Therefore bk = vr.

If x and y are any two different varieties in a block design, we refer to the
number of blocks that contain both x and y as the covalency of x and y, and
write it as λxy . Many important properties of block designs are concerned
 Balanced Incomplete Block Designs 267

with this covalency function. The one that has most frequently been studied
is the property of balance: a balanced incomplete block design, or BIBD, is
a regular incomplete design in which λxy is a constant, independent of the
choice of x and y. This constant covalency is called the index of the design.
Balanced incomplete block designs were defined (essentially as a puzzle) by
Woolhouse [140], in the annual Lady’s and Gentleman’s Diary, a “Collection
of mathematical puzzles and aenigmas” that he edited. The Reverend T. P.
Kirkman [57, 58, 59, 60] studied the case of block size 3; in particular he
introduced his famous “schoolgirl problem”[59]:
A schoolmistress has 15 girl pupils and she wishes to take them
on a daily walk. The girls are to walk in five rows of three girls
each. It is required that no two girls should walk in the same row
more than once per week. Can this be done?
The solution is a balanced incomplete block design with v = 15, k = 3, λ = 1
with additional properties. Yates [142] first studied balanced incomplete block
designs from a statistical viewpoint.
A balanced design with λ = 0, or a null design, is often called trivial,
and so is a complete design. We shall demand that a balanced incomplete
block design be not trivial; since completeness is already outlawed, the added
restriction is that λ > 0.
It is usual to write λ for the index of a balanced incomplete block design. We
often refer to a balanced incomplete block design by using the five parameters
(v, b, r, k, λ), and call it a (v, b, r, k, λ)-design or (v, b, r, k, λ)-BIBD.
As an example, here is a (7, 7, 4, 4, 2)-BIBD:

B1 = {x1 , x2 , x3 , x4 }
B2 = {x1 , x4 , x5 , x6 }
B3 = {x1 , x2 , x5 , x7 }
B4 = {x1 , x3 , x6 , x7 }
B5 = {x3 , x4 , x5 , x7 }
B6 = {x2 , x3 , x5 , x6 }
B7 = {x2 , x4 , x6 , x7 }.

There is a further relation between the parameters of a balanced incomplete

block design:

THEOREM 14.2 (Second Parameter Relation)

In a (v, b, r, k, λ)-BIBD,
r(k − 1) = λ(v − 1). (14.2)

Proof. Consider the blocks of the design that contain a given variety, x say.
There are r such blocks. Because of the balance property, every variety other
268 Introduction to Combinatorics

than x must occur in λ of them. So if we list all entries of the blocks, variety
x is listed r times and we list every other treatment λ times.
The list contains rk entries. So:

rk = r + λ(v − 1),
r(k − 1) = λ(v − 1). 

Relations (14.1) and (14.2) allow us to find all parameters of a BIBD when
only three are given. For example, if we know v, k and λ, we can calculate r
from (14.2) and then deduce b using (14.1). As an example, if v = 13, k = 3
and λ = 1, we have r = λ(v−1)/(k−1) = 12/2 = 6 and b = vr/k = 13×6/3 =
26, and a (13, 26, 6, 3, 1)-design exists. But not all sets {v, k, λ} give a design.
First, all the parameters must be integers; for example, if v = 11, k = 3
and λ = 1, we would get r = 5 and b = 55/3; you cannot have a fractional
number of blocks, so this is impossible. And even when the parameters are
all integers, there might be no corresponding design; for example, there is
no (22, 22, 7, 7, 2)-design (see Section 14.3), even though both equations are
satisfied. In fact, whether there exists a design corresponding to a given set
of parameters is not known in general.
There is no requirement that all the blocks in a design should be different.
If two blocks have the same set of elements, we say there is a “repeated block.”
A design that has no repeated block is called simple.
If a design has b blocks B1 , B2 , . . . , Bb and v varieties t1 , t2 , . . . , tv , we define
a v × b matrix A with (i, j) entry aij as follows:

1 if ti ∈ Bj ;
aij =
0 otherwise.

This matrix A is called the incidence matrix of the design. The definition
means that each block corresponds to a column of the incidence matrix, and
each variety corresponds to a row.

Example 14.1: Exhibit the incidence matrix of the regular design with
varieties t1 , t2 , t3 , t4 and blocks B1 = {t1 , t2 }, B2 = {t3 , t4 }, B3 = {t1 , t4 },
B4 = {t2 , t3 }.
 
1 0 1 0
 1 0 0 1 
This design has incidence matrix A =   0 1 0 1 .
 
0 1 1 0
When we define designs, the order of the varieties does not matter. Sim-
ilarly, the order of the set of blocks is not important. So we do not usually
distinguish between two designs if the incidence matrix of one can be obtained
by reordering the rows and/or columns of the incidence matrix of the other.
Two such designs are called isomorphic.
 Balanced Incomplete Block Designs 269

Suppose we interchange blocks B2 and B3 in the preceding example, and

also reorder the treatments t1 , t4 , t2 , t3 . The new design has incidence matrix
 
1 1 0 0
 0 1 1 0 
B=  1 0 0 1 .


0 0 1 1

These matrices are related by:

     
1 1 0 0 1 0 0 0 1 0 1 0 1 0 0 0
 0 1 1 0   0 0 0 1   1 0 0 1   0 0 1 0 
 =   .
 1 0 0 1   0 1 0 0  0 1 0 1  0 1 0 0 
0 0 1 1 0 0 1 0 0 1 1 0 0 0 0 1

The above equation could be written

B = P AQ,

where    
1 0 0 0 1 0 0 0
 0 0 0 1   0 0 1 0 
P =
 0
,Q= .
1 0 0   0 1 0 0 
0 0 1 0 0 0 0 1
The permutation matrix P represents the reordering of the varieties, and
could be obtained by performing the same reordering on the rows of the 4 × 4
identity matrix. Matrix Q represents the block reordering.
As usual, we write Jmn for an m × n matrix with every entry 1. (If m = n,
we simply write Jn ; and if the order is obvious from the context, we omit
subscripts altogether.) Also, In (or simply I) is an n × n identity matrix.

THEOREM 14.3 (Incidence Matrix Equations)

If A is the incidence matrix of a balanced incomplete block design with pa-
rameters (v, b, r, k, λ), then
Jv A = kJvb (14.3)
and
AAT = (r − λ)Iv + λJv . (14.4)
Conversely, if a matrix A, with every entry 0 or 1, satisfies (14.3) and (14.4)
with k < v and λ > 0, then A is the incidence matrix of a (v, b, r, k, λ)-BIBD
and
r(k − 1) vr
v= + 1, b = . (14.5)
λ k

Proof. First, assume A is the incidence matrix of a (v, b, r, k, λ)-BIBD.

270 Introduction to Combinatorics

For any i, the (i, j) entry of Jv A counts the number of 1s in column j of

A, which equals the number of varieties in block j. So it equals k. Since Jv A
is the product of a v × v and a v × b matrix, it is of size v × b, so (14.3) holds.
The (i, i) entry of AAT is

b
X b
X
aip aip = a2ip ;
p=1 p=1

a2ip is 1 when variety i belongs to block p, which happens for r values of p

(variety i is in r blocks), and 0 otherwise. So the sum is r. When i 6= j, the
(i, j) entry of AAT is
Xb
aip ajp ;
p=1

aip ajp is 1 when varieties i and j both belong to block p, which happens λ
times for given i and j, and 0 otherwise. So the sum is λ. Therefore AAT has
diagonal entries r and off-diagonal entries λ; that is,

AAT = rIv + λ(Iv − Jv ) = (r − λ)Iv + λJv .

Conversely, suppose A is a v × b (0, 1)-matrix that satisfies the equations.

We define a block design with varieties x1 , x2 , . . . , xv and blocks Y1 , Y2 , . . . , Yb
by the rule
xi ∈ Yj if and only if aij = 1.

Equation (14.3) implies that every block has size k. From (14.4) the design
has constant replication number r (because of the diagonal entries of AAT )
and constant covalency λ (from the off-diagonal elements). So the design is a
(v, b, r, k, λ)-BIBD with incidence matrix A. The equalities (14.5) now follow
from (14.1) and (14.2). 

Another useful equation for balanced incomplete block designs,

AJv = rJvb , (14.6)

follows from the fact that every variety belongs to r blocks, so each row of A
contains r 1’s.

14.2 Fisher’s Inequality

In this section we need to know the determinant of the matrix AAT .
 Balanced Incomplete Block Designs 271

LEMMA 14.4
The determinant of
 
r λ λ ... λ

 λ r λ ... λ 

M =
 λ λ r ... λ 

 ... ... 
λ λ λ ... r
is (r − 1)v−1 [r + λ(v − 1)].

Proof. If we subtract the first column of M from every other column we

obtain  
r λ − r λ− r ... λ − r
 λ r−λ 0 ... 0 
 
 λ
 0 r − λ . . . 0 .

 ... ... 
λ 0 0 ... r − λ
Now add row 2 to row 1, then add row 3 to row 1, and so on. The resulting
matrix,  
r + (v − 1)λ 0 0 ... 0

 λ r−λ 0 ... 0 


 λ 0 r − λ ... 0 .

 ... ... 
λ 0 0 ... r − λ
This matrix has the same determinant as M ; and, as all entries above its
diagonal are zero, the determinant is the product of the diagonal elements,
(r − λ)v−1 [r + λ(v − 1)]. 

THEOREM 14.5 (Fisher’s Inequality) [35]

In any balanced incomplete block design, b ≥ v.

Proof. Suppose A is the incidence matrix of a (v, b, r, k, λ)-BIBD. Then

det(AAT ) = (r − 1)v−1 [r + λ(v − 1)]. From (14.2) we can replace λ(v − 1) by
r(k − 1), so
det(AAT ) = (r − 1)v−1 [r + r(k − 1)] = (r − 1)v−1 rk.
On the other hand, (14.2) can be rewritten as
v−1
r= λ;
k−1
by incompleteness, k < v, so r > λ, and (r − λ) is non-zero. Therefore
det(AAT ) is non-zero, so the v × v matrix AAT has rank v. That is
v = rank(AAT ) ≤ rank(A) ≤ min{v, b},
272 Introduction to Combinatorics

because A is v × b. Therefore v ≤ b. 

14.3 Symmetric Balanced Incomplete Block Designs

Designs that only just satisfy Fisher’s inequality, those with b = v, are
called symmetric. From (14.1), such a design also satisfies r = k, so only
three parameters need to be specified; the common phrase is “(v, k, λ)-design.”
Another abbreviation is SBIBD. For these designs, Equation (14.2) takes on
the simpler form
λ(v − 1) = k(k − 1). (14.7)

In the following theorem, the matrices are all v × v, so we simply write I

and J instead of Iv and Jv .

THEOREM 14.6 (Incidence Matrix Equations, Symmetric De-

signs)
If A is the incidence matrix of a (v, k, λ)-design, then

λJAT = kJ (14.8)

and
λAT A = (k − λ)I + λJ. (14.9)

So AT is also the incidence matrix of a (v, k, λ)-design.

Proof. For a symmetric design, (14.3) and (14.4) become

JA = kJ, (14.10)

AJ = kJ, (14.11)

and
AAT = (k − λ)I + λJ. (14.12)

J T = J, so (14.8) is just the transpose of (14.11).

T
p Since A is square, det(A) and det(A ) are defined and are equal; det(A) =
T
det(AA ) is non-zero. So A has an inverse. Multiplying (14.11) on the left
by A−1 , A−1 AJ = kA−1 J, so A−1 J = k −1 J. Now multiply Equation (14.12)
 Balanced Incomplete Block Designs 273

on the left by A−1 and on the right by A:

A−1 AAT A = (k − λ)A−1 A + λA−1 JA

= (k − λ)I + λA−1 kJ
= (k − λ)I + λkA−1 J
= (k − λ)I + λkk −1 J
= (k − λ)I + λJ

as required. It follows from Theorem 14.3 that AT is the incidence matrix of

a (v, k, λ)-design. 

COROLLARY 14.6.1
In a symmetric balanced incomplete block design, any two blocks intersect in
λ blocks.
A block design whose blocks have constant-size intersection is called linked .
So the corollary tells us that a symmetric balanced incomplete block design
is also a linked design.
In general, if a design D has incidence matrix A, the design with incidence
matrix AT is called the dual of D, and is often denoted D∗ . The dual of
any balanced incomplete block design is linked, but the only incomplete block
designs that are both symmetric and linked are SBIBDs.
The following important theorem is called the Bruck–Chowla–Ryser Theo-
rem.

THEOREM 14.7 (Bruck–Chowla–Ryser Theorem) [15, 21, 99]

If there exists a symmetric balanced incomplete block design with parameters
(v, k, λ), then:
(i) if v is even, k − λ must be a perfect square;
(ii) if v is odd, there must exist integers x, y and z, not all zero, such that

x2 = (k − λ)y 2 + (−1)(v−1)/2 λz 2 . (14.13)

Proof. Part (i) only: Suppose v is even, and suppose there exists a (v, k, λ)-
design with incidence matrix A. Since A is square it has a determinant,
denoted det(A), where det(AT ) = det(A), so
q
det(A) = det(AAT ).

But from Lemma 14.4,

det(AAT ) = k 2 (k − λ)v−1 ,
274 Introduction to Combinatorics

so
1
det(A) = ±k(k − λ) 2 (v−1) .
Now A is an integer matrix, so√det(A) is an integer. Since k and λ are integers,
and v is even, it follows that k − λ is an integer also.
The proof of part (ii) requires more mathematical background than we are
assuming. See, for example, Chapter 7 of [123]. 
For example, consider the case k = 7, λ = 2. From (14.7), v must equal 22;
but this is even, so no (22, 7, 2)-design exists.
Some applications require some harder results from number theory, but we
shall give one example where all you need to know is that 2 is not a square in
the arithmetic modulo 3.
Example 14.2: Show that no (141, 21, 3)-design exists.
Suppose a (141, 21, 3)-design exists. Then there are integers x, y and z, not
all zero, such that
x2 = 18y 2 + 3z 2 .
Without loss of generality we can assume that x, y, z have no mutual common
factor. Clearly, 3 must divide x2 , so 3 divides x: say x = 3e. Then

9e2 = 18y 2 + 3z 2

and z must also be divisible by 3; say z = 3f . We have

9e2 = 18y 2 + 27f 2 ,

e2 = 2y 2 + 3f 2 .

Reducing modulo 3, we find

e2 ≡ 2y 2 (mod 3).

If y is not divisible by 3, 2y 2 is congruent to 2, which is not a perfect square in

the arithmetic modulo 3, a contradiction. So 3 divides y, and x, y and z have
common factor 3, another contradiction. Therefore no (141, 21, 3)-design can
exist. 

14.4 New Designs from Old

If B0 is any block of a balanced incomplete block design with index λ, then
any two varieties that do not belong to B0 must occur together in λ of the
remaining blocks, while any two members of B0 must be together in λ − 1 of
the remaining blocks. It follows that the blocks B\B0 form a balanced design
 Balanced Incomplete Block Designs 275

of index λ when B ranges through the remaining blocks, and the blocks B ∩B0
form a balanced design of index λ − 1. We shall refer to these as the residual
and derived designs of the original design with respect to the block B0 .
In general the replication number of a residual or derived design is not
constant, and the block sizes follow no particular pattern. If each variety in
the original design occurs r times, its residual and derived designs also have
constant replication numbers, r and r − 1 respectively.
The block pattern can be predicted if the original design is linked, that is
the size of B ∩ B0 is constant; the blocks of the derived design will be of that
constant size, and the blocks of the residual design will be |B ∩ B0 | smaller
than the blocks of the original. In particular, if we start with a symmetric
balanced incomplete block design, we obtain constant replication numbers
and block sizes.

THEOREM 14.8 (Derived and Residual Design Parameters)

The residual design of a (v, k, λ)-SBIBD is a balanced incomplete block design
with parameters
(v − k, v − 1, k, k − λ, λ), (14.14)
provided λ 6= k − 1. The derived design of a (v, k, λ)-SBIBD is a balanced
incomplete block design with parameters

(k, v − 1, k − 1, λ, λ − 1), (14.15)

provided λ 6= 1.

The two exceptions are made so as to exclude “trivial” designs.

Example 14.3: Exhibit derived and residual designs of a (11, 6, 3)-design.

Figure 14.1 shows the designs. The first column shows the blocks of a
(11, 6, 3)-design B. The other columns show the blocks of the residual design R
and the derived design D with respect to the first block (block {1, 2, 3, 4, 5, 6}).
Designs R and D have parameters (5, 10, 6, 3, 3) and (6, 10, 5, 3, 2), respec-
tively. 
Suppose a balanced incomplete block design D has parameters (v − k, v −
1, k, k − λ, λ) for some integers v, k and λ. One could ask whether there is a
(v, k, λ)-design for which D is the residual. The answer is “not necessarily,”
but in most cases there must be such a symmetric design. The exact situa-
tion is given by the following theorem, which is proven in various texts (for
example, [50]).

THEOREM 14.9 (Existence of Residual Designs) [51, 103, 104]

Suppose D is a balanced incomplete block design with parameters (14.14).
276 Introduction to Combinatorics

B R D
1 2 3 4 5 6
2 5 6 7 10 11 7 10 11 2 5 6
1 4 6 7 8 10 7 8 10 1 4 6
2 4 5 7 8 9 7 8 9 2 4 5
3 5 6 8 9 10 8 9 10 3 5 6
3 4 6 7 9 11 7 9 11 3 4 6
1 3 5 7 8 11 7 8 11 1 3 5
1 2 6 8 9 11 8 9 11 1 2 6
1 2 3 7 9 10 7 9 10 1 2 3
2 3 4 8 10 11 8 10 11 2 3 4
1 4 5 9 10 11 9 10 11 1 4 5
FIGURE 14.1: Residual and derived designs.

(i) If λ = 1 or λ = 2, there is a (v, k, λ)-design of which D is the residual.

(ii) There is a number f (λ), depending only on λ, such that if k ≥ f (λ),
there is a (v, k, λ)-design of which D is the residual.

A design with parameters (14.14) is called quasi-residual ; thus Theorem

14.9(ii) says that “all sufficiently large quasi-residual designs are residual.”
For example, f (3) is at most 90.
The first example of a quasi-residual design that is not residual was given
by Bhattacharya [9] in 1947. It had parameters (16, 24, 9, 6, 3).

14.5 Difference Methods

Suppose you decide a (7, 3, 1)-design is what you need for an experiment,
so you try to construct one. If you took a very simple-minded approach, you

7
would observe that there are 73 = 35 possible blocks, so there are 73 , more


than 64 billion, ways to select seven blocks. With a little thought, you will
realize that (up to isomorphism) you can choose 123 to be a block; moreover,
the design

will be simple, so the remaining six blocks can be chosen in at
most 34 6 , or 12.7 million, ways. This number can be reduced significantly by
using more and more facts about block designs, and finally you will find that
there is only one solution, up to isomorphism. But the point is that there are
a large number of possibilities to be eliminated. It comes as no surprise that
the computer searches involved in looking for block designs can be very long.
 Balanced Incomplete Block Designs 277

In view of this, it is very useful to have techniques for constructing block

designs quickly, even if success is not guaranteed. In this section we look at a
method that has produced a large number of BIBDs.
Consider the following blocks based on varieties {0, 1, 2, 3, 4, 5, 6}, the in-
tegers modulo 7. The blocks are 124, 235, 346, 450, 561, 602, 013. The second
block is formed by adding 1 to every member of the first, the third by adding
1 to every member of the second, and so on. The defining properties of a
block design can be deduced from this additive structure: clearly every block
has the same size, and every variety appears equally often—once in the first
position in the blocks, once in the second, and so on.
There are seven pairs of varieties whose labels differ by ±1, namely 01,
12, 23, 34, 45, 56, 60. The first block contains one of these pairs (12), and
the additive property means that each of the others appears exactly once.
Similarly, there is one pair with each of the other two possible differences, ±2
and ±3 (pairs 24 and 14, respectively). This is enough to prove balance. The
properties of a balanced incomplete block design can all be deduced from the
choice of the first block and the additive structure.
The important property we used was that the first block contained exactly
one pair of entries with difference ±1, one with difference ±2 and one with
difference ±3. If every difference had occurred twice, we could have generated
a design with λ = 2, and so on. This leads us to the following definition.
Definition: A (v, k, λ)-difference set B is a k-element set of integers mod-
ulo v with the property that given any integer d, 0 < d < v, there are precisely
λ ordered pairs of elements of B whose difference is d.
The definition is written in terms of ordered pairs. When we looked at the
introductory example, we talked about unordered pairs and differences like
±1. Using ordered pairs gives the same result (after all, if we say elements x
and y have difference ±d, it is the same as saying the two ordered pairs (x, y)
and (y, x) have differences d and −d), but avoids any confusion in the case
where v is odd, and the specific difference 21 v is being discussed.
Given a set B, we write B + i for the set obtained by adding i to every
element of B. Then we have the following theorem.

THEOREM 14.10 (Difference Sets and Designs)

If B is a (v, k, λ)-difference set then the sets B + i, for 0 ≤ i < v, are the
blocks of a (v, k, λ)-design.

For example, in the case v = 7, we saw that {1, 2, 4} is a (7, 3, 1)-difference

set, and the corresponding design is

124 235 346 450 561 602 013.

Notice that instead of {1, 2, 4} we could have used {0, 1, 3} or {2, 3, 5} or any
of the seven blocks. (These other blocks are called shifts of the original block.)
278 Introduction to Combinatorics

v k λ Difference set
15 7 3 1, 2, 3, 5, 6, 9, 11
21 5 1 1, 2, 7, 9, 19
73 9 1 1, 2, 4, 8, 16, 32, 37, 55, 64
37 9 2 1, 7, 9, 10, 12, 16, 26, 33, 34
FIGURE 14.2: Some difference sets.

COROLLARY 14.10.1
If there is a (v, k, λ)-difference set, then λ(v − 1) = k(k − 1).
Example 14.4: Construct a (7, 4, 2)-difference set D.
There will be exactly two pairs of elements whose difference is 1; without
loss of generality we might as well take one of them as 1 and 2 (if we took
values t and t+1, we would obtain a shift of a block containing 0 and 1). So the
set is {1, 2, x, y} for some x and y. Now we need another pair with difference
1; the possibilities are that D contains {0, 1} (D = {0, 1, 2, x} for some x), D
contains {2, 3} (D = {1, 2, 3, x}), or y = x + 1 (D = {1, 2, x, x + 1}). In the
first case x 6= 3 or 6, because these both give a third pair with difference 1.
So the possibilities are {0, 1, 2, 4} and {0, 1, 2, 5}. To test the first one, we see
that the differences of the elements are

{±(1 − 0), ±(2 − 0), ±(4 − 0), ±(2 − 1), ±(4 − 1), ±(4 − 2)}
= {±1, ±2, ±4, ±1, ±3, ±2},

so it is a difference set (remember that ±4 = ±3 modulo 7). Similarly,

the second set works. The second two cases give solutions {1, 2, 3, 5} and
{1, 2, 3, 6}, which are shifts of the first two. In the third case, the possibilities
are {1, 2, 4, 5} and {1, 2, 5, 6}. Set {1, 2, 4, 5} yields differences

{±(2 − 1), ±(4 − 1), ±(5 − 1), ±(4 − 2), ±(5 − 2), ±(5 − 4)}
= {±1, ±3, ±4, ±2, ±3, ±1},

and there is only one difference ±2, so this is not a difference set. The other
example does not work either. So there are 14 candidates for D—the two
examples that work and their shifts. 
Figure 14.2 shows some further examples of difference sets.
The process of starting with one block, adding 1 to each member to form
a second block, then continuing, is called developing the block. We are only
dealing with cases where the elements are integers modulo v (called developing
mod v), but the same process can be carried out in other arithmetic structures,
such as groups. It is also possible to develop more than one block. For
example, consider the blocks {1, 2, 3, 5} and {1, 2, 5, 7} modulo 9. The first
 Balanced Incomplete Block Designs 279

block contains differences ±1 twice, ±2 twice, ±3 once and ±4 once. So

developing that block yields nine blocks in which every pair with difference
±1—pairs 01, 12, 23, . . . , 78, 80—occur twice. Block {1, 2, 5, 7} contains
differences ±1 once, ±2 once, ±3 twice and ±4 twice; pairs 01, and so on,
will appear once in the nine blocks developed from it. Similarly, every other
pair will appear exactly three times in the 18 blocks obtained by developing
both initial blocks. A quick check shows that the blocks together form a
(9, 18, 8, 4, 3)-BIBD. Blocks like this are called supplementary difference sets.
Further examples are given in Figure 14.3.

v b r k λ Initial blocks
41 82 10 5 1 {1, 10, 16, 18, 37}; {5, 8, 9, 21, 39}
13 26 12 6 5 {1, 2, 4, 7, 8, 12}; {1, 2, 3, 4, 8, 12}
16 80 15 3 2 {1, 2, 4}; {1, 2, 8}; {1, 3, 13}; {1, 4, 9}; {1, 5, 10}
22 44 14 7 4 {1, 7, 12, 16, 19, 21, 22}; {1, 6, 8, 9, 10, 14, 20}
FIGURE 14.3: Some supplementary difference sets.

Exercises 14A
1. You need to construct a block design with five blocks of size 2, based
on variety-set {1, 2, 3}, in which every possible pair of varieties occurs
at least once. Prove that there are exactly two non-isomorphic ways to
do this. Give examples.

2. Suppose v is any positive integer greater than 2. Prove that a (v, v, v −

1, v − 1, v − 2)-design exists.

3. Prove that the following blocks form a (6, 10, 5, 3, 2)-design:

123, 124, 135, 146, 156,

236, 245, 256, 345, 346.

4. Write down the blocks of a (7, 14, 6, 3, 2)-design.

5. Prove that, in any balanced incomplete block design, λ < r.

6. Suppose S is a set with v elements, v > 2. Write 2(S) for the set of all
unordered pairs of elements of S. Show that if we take the members of S
as varieties and the members of 2(S) as blocks, the result is a balanced
incomplete block design. What are its parameters?
280 Introduction to Combinatorics

7. In each row of the following table, fill in the blanks so that the parame-
ters are possible parameters for a balanced incomplete block design, or
else show that this is impossible.

v b r k λ
35 3 1
21 28 3
11 5 5
14 7 4
12 6 3
49 7 1

8. Show that there do not exist symmetric balanced incomplete block de-
signs with the following parameters: (46, 10, 2), (52, 18, 6).
9. Suppose D is a balanced incomplete block design with parameters
(v, b, r, k, λ). The design tD is defined to have tb blocks: each block
B of D gives rise to precisely t blocks, each with the same set of vari-
eties as B. Prove that tD is a balanced incomplete block design. What
are its parameters? (The design tD is called the t-multiple of D.)
10. The balanced incomplete block design D has parameters (6, 10, 5, 3, 2)
and blocks
B0 = {0, 1, 2} B5 = {1, 2, 5}
B1 = {0, 1, 3} B6 = {1, 3, 4}
B2 = {0, 2, 4} B7 = {1, 4, 5}
B3 = {0, 3, 5} B8 = {2, 3, 4}
B4 = {0, 4, 5} B9 = {2, 3, 5}.

Construct the blocks of the dual of D, and verify that it is not balanced.
11. For the following parameter sets, what are the residual and derived
parameters?
(i) (23, 11, 5)
(ii) (15, 7, 3)
(iii) (45, 12, 3)
(iv) (13, 9, 6)
(v) (69, 17, 4)
12. Find an (11, 5, 2)-difference set.
13. Find two initial blocks whose development modulo 13 yields a (13, 26, 6,
3, 1)-design.
 Balanced Incomplete Block Designs 281

Exercises 14B
1. A combinatorial design has six varieties, {1, 2, 3, 4, 5, 6}, and nine blocks
of size 2. Every variety occurs in three blocks, and the design is simple
(no two blocks are identical). Prove that there are exactly two non-
isomorphic ways to do this.
2. Prove that any two (4, 4, 3, 3, 2)-designs must be isomorphic.
3. In each row of the following table, fill in the blanks so that the parame-
ters are possible parameters for a balanced incomplete block design, or
else show that this is impossible.

v b r k λ
7 14 3
6 4 2
13 6 1
17 8 5
30 7
33 44 6

4. Show that, in a symmetric balanced incomplete block design, λ cannot

be odd when v is even.
5. Suppose D is a (v, b, r, k, λ)-BIBD with variety-set V . Write T for the
set of all subsets of V of the form V \B, where B is a block of D. Prove
that the members of T , as blocks, form a balanced incomplete block
design based on V , unless b − 2r + λ is zero. (This design is called the
complement of D.) What are its parameters?
6. The complement of a design was defined in the previous exercise. What
are the parameters of the complement of a design with the parameters
shown?

(i) (7, 7, 3, 3, 1) (iii) (16, 24, 9, 6, 3)

(ii) (13, 26, 6, 3, 1) (iv) (12, 44, 33, 9, 24)

7. Show that a balanced incomplete block design for which b − 2r + λ is

zero must have parameters (v, v, v − 1, v − 1, v − 2), provided r ≤ v.
8. Prove that the dual of any balanced incomplete block design is a linked
design.
282 Introduction to Combinatorics

9. For the following parameter sets, what are the residual and derived
parameters?
(i) (16, 6, 2)
(ii) (22, 7, 2)
(iii) (15, 8, 4)
(iv) (31, 10, 3)
(v) (41, 16, 6)

10. Find a (13, 4, 1)-difference set.

11. Verify that the examples given in Figure 14.2 are, in fact, difference sets
with the stated parameters.

12. Suppose D is a (v, k, λ)-difference set. Write E for the set of all integers
modulo v that do not belong to D. Prove that E is a (v, v−k, v−2k+λ)-
difference set.

13. Verify that {0, 1, 2, 5} is a (7, 4, 2)-difference set, but {1, 2, 5, 6} is not.

Problems

Problem 1: Prove that any two (7, 7, 4, 4, 2)-designs must be isomorphic.

Problem 2: Can you find a simple (7, 14, 6, 3, 2)-design, that is, one with no
repeated blocks? (See Exercise 12A.4.)
Problem 3: Suppose there were a symmetric balanced incomplete block
design with λ = 3 and k = 21 + 27t for some positive integer t.
(i) According to (14.1) and (14.2), what is the corresponding value of v?
(ii) Prove that no such design exists.
Problem 4: Suppose B is a balanced incomplete block design with v < b
(that is, B is not symmetric). Show that B is not a linked design.
Problem 5: (Generalization of Exercise 12A.6.) Suppose v and k are positive
integers, with v > k ≥ 2, and S is a set with v elements. Write k(S) for the
set of all unordered k-sets of elements of S. Show that if we take the members
of S as varieties and the members of k(S) as blocks, the result is a balanced
incomplete block design. What are its parameters?
Problem 6: See the definition of the complement of a design in Exercise
14B.5. Suppose D is a symmetric balanced incomplete block design with
support set S, and suppose E is its complement. Select any block B of D.
 Balanced Incomplete Block Designs 283

Prove that the residual design of E with respect to S\B is the complement of
the derived design of D with respect to B. Select any block B of D. Prove
that the residual design of E with respect to S\B is the complement of the
derived design of D with respect to B.
Problem 7: D is the (15, 15, 7, 7, 3)-design with blocks

B0 = {0, 1, 2, 3, 4, 5, 6} B1 = {0, 3, 4, 9, 10, 13, 14}

B2 = {0, 1, 2, 7, 8, 9, 10} B3 = {0, 5, 6, 7, 8, 13, 14}
B4 = {0, 1, 2, 11, 12, 13, 14} B5 = {0, 5, 6, 9, 10, 11, 12}
B6 = {0, 3, 4, 7, 8, 11, 12} B7 = {1, 3, 5, 7, 9, 11, 13}
B8 = {1, 3, 6, 7, 10, 12, 14} B9 = {2, 3, 6, 8, 9, 11, 14}
B10 = {1, 4, 5, 8, 10, 11, 14} B11 = {2, 4, 5, 7, 9, 12, 14}
B12 = {1, 4, 6, 8, 9, 12, 13} B13 = {2, 4, 6, 7, 10, 11, 13}
B14 = {2, 3, 5, 8, 10, 12, 13}

(i) Write down the blocks of the dual design D∗ of D.

(ii) A triplet is a set of three blocks whose mutual intersection has size 3.
(For example, {B0 , B1 , B6 } is a triplet in D.) By considering triplets,
prove that D and D∗ are not isomorphic.

Problem 8: Consider an SBIBD with parameters (n2 + n + 1, n + 1, 1). (This

is called a finite projective plane.)
(i) What are the parameters of the complement of this design (as defined
in Exercise 12A.5)?
(ii) What are the parameters of the residual of this complementary design?
(iii) Prove that a design with these residual parameters always exists.
Problem 9: A finite binary plane consists of a finite, non-empty set of points
(the varieties) and a set of lines. Each line is a subset of the points, and two
lines are different if and only if they have different point-sets. Moreover, the
following axioms are satisfied:
B1 any two lines have precisely one common point;
B2 any point lies in precisely two lines.
(i) Show that every line must contain the same number of points;
(ii) Construct all finite binary planes.

Problem 10: Prove that no (v, k, λ)-difference set satisfies v = 2k.

Problem 11: Suppose D is a (v, k, λ)-difference set and m is a positive
integer such that m and v are coprime (that is, they have no common factor
greater than 1). By mD we mean the {mx : x ∈ D}, where all numbers are
treated as integers modulo v. Prove that mD is a (v, k, λ)-difference set. Is
this true if m and v are not coprime?
284 Introduction to Combinatorics

Problem 12: Fourteen riders take part in a motorcycle sprint championship.

The qualifying consists of fourteen heats, of four riders each. Each competitor
races in four heats, and no two meet in more than one heat.
(i) Find a schedule for the competition.
(ii) Is there a schedule where the qualifying is divided into seven rounds of
two heats each, and no rider appears in both heats of a round?
Chapter 15
Linear Algebra Methods in
Combinatorics

Every mathematical discipline is enriched by and enriches other mathematical

disciplines. The reader may be familiar with such areas as analytic number
theory, for instance, in which the techniques of complex analysis are applied
to questions from number theory. Combinatorics, as we have seen in earlier
chapters, uses calculus and group theory (among many other disciplines) for
generating functions and Pólya counting. In this chapter, we take a deeper
look at some topics within combinatorics that rely especially upon results and
ideas from linear algebra.

15.1 Recurrences Revisited

We saw in Chapter 6 that there are two ways (generating functions and the
characteristic equation) to solve a linear homogeneous recurrence relation.
A third way involving linear algebra provides a different perspective. We
illustrate with an example.
   
0 1 n 1
Example 15.1: Let A denote the matrix . Find A .
1 1 1
Let the row matrix [xn , yn ] be the result of the matrix multiplication for a
given n. Then because matrix multiplication is associative, we have

[xn+1 , yn+1 ] = A × An × [x, y]T = A × [xn , yn ]T = [yn , xn + yn ].

Given [x0 , y0 ] = [1, 1] = [F1 , F2 ] (the first two Fibonacci numbers, as outlined
in Chapter 6), it follows that [x1 , y1 ] = [1, 2] = [F2 , F3 ], so we see [xn , yn ] =
[Fn+1 , Fn+2 ]. 
This gives us a faster method of computing a given Fibonacci number than
using the recurrence. To find F100 , we would take the matrix A in the example
and square it, then square the square to get A4 , and continue to A64 . Then
calculate A64 × A32 × A2 × [1, 1]T = [x97 , x98 ] = [F99 , F100 ]. This would be
nine matrix multiplications, as opposed to 98 iterations of the recurrence.

285
286 Introduction to Combinatorics

Considering this process, we see  that the matrix

 An consists only of Fi-
Fn−1 Fn
bonacci numbers. In fact, An = . We use this observation to
Fn Fn+1
obtain a Fibonacci identity. Since A × A = Ak+m , we look at the (2, 2)-
k m

entry in the product to see Fk Fm + Fk+1 Fm+1 = Fk+m+1 . In this way, the
matrix identity allows us to see relationships among the Fibonacci numbers
that would be more tedious to prove without matrix methods. Other Fi-
bonacci identities are given easy matrix proofs in [106], and in the problems.
Other two-term recurrences may be treated similarly. (In fact, the differ-
ence between a “method” and a “trick” in mathematics is often said to be that
a trick only works in one case, where a method works in many; the matrix
technique with this convention is a “method,” because it works for many recur-
rences.) Suppose we look at the recurrence an+1 = can +dan−1 , with specified
initial values a1 , a2 . We seek a matrix A for which An [a1 , a2 ]T = [an+1 , an+2 ].
It is not hard to see that the first row will be [0, 1] and the second row (where
the recurrence “lives,” so to speak) will be [d, c] (note the reversal of the
coefficients!).
Although we have seen two ways in Chapter 6 to find an explicit formula
for an in a two-term linear homogeneous recurrence, we will explore a third
way that uses diagonalization. Recall that some square matrices have the
property that there exists a matrix P where P −1 AP = D is a diagonal matrix
(all entries aij are 0 unless i = j). If A has this property, we say A is
diagonalizable, and D’s non-zero entries are the eigenvalues of A and the
columns of P are the corresponding eigenvectors.
Example 15.2: Solve the recurrence an+1 = an + 6an−1 with initial con-
ditions a0 = 2, a1 = 1.
 
0 1
We want eigenvalues and eigenvectors of the matrix A = . The
6 1
characteristic equation det(λI − A) is λ(λ − 1) − 6 = λ2 − λ − 6. The roots are
λ1 = 3 and λ2 = −2. (The characteristic equation of the matrix is the same as
the characteristic equation of the recurrence that we discussed in Chapter 6.)
The eigenvector corresponding to λ1 is the solution to A~v = 3~v . We write
~v = [v1 , v2 ]T . There are infinitely many eigenvectors, so we arbitrarily let
v1 = 1; then A~v = 3~v gives us v2 = 3v1 , or ~v = [1, 3]T .
Similarly, the eigenvector [1, −2] has eigenvalue −2. We conclude that
     
1 1 .6 −.2 −2 0
P = ; P −1 = ; D = P −1 AP = .
−2 3 .4 .2 0 3

To solve the recurrence, we use the fact that a diagonal matrix may be
easily raised to any power.
     
n a0 −1 n 2 n −1 2
[an , an+1 ] = A = (P DP ) = PD P
a1 1 1
 Linear Algebra Methods in Combinatorics 287

The resulting matrix multiplication yields

(−2)n 0 3n + (−2)n
      
1 1 .6 −.2 2
= .
−2 3 0 3n .4 .2 1 3n+1 + (−2)n+1

Thus, we arrive at the solution an = 3n + (−2)n . 

15.2 State Graphs and the Transfer Matrix Method

Occasionally we look at systems that can be in any of a finite number of
states, and can under certain circumstances change from one state to another.
For a simple example, consider an automobile with a standard transmission.
It can be in several states (gears); first, second, third, fourth, stopped or
reverse. (We omit neutral for the sake of simplicity.)
We cannot shift gears entirely at will; for instance, it would be a bad idea
to shift directly from fourth gear to reverse. For simplicity, we agree that we
may shift a car from a particular gear to another that is only one higher or
lower.
We will use a graph to designate the possible changes of state. Our graph
will have a vertex for each legal state and an edge between two vertices indi-
cates that we may safely shift between the corresponding states. This graph
is shown in Figure 15.1.

 
0 1 0 0 0 0

 1 0 1 0 0 0 

 0 1 0 1 0 0 
Rev Stop 1st 2nd 3rd 4th A= 

 0 0 1 0 1 0 

 0 0 0 1 0 1 
0 0 0 0 1 0
FIGURE 15.1: Gear shift graph.

Now consider the question: We have shifted gears five times since we began
with the car in the initial Stop state. How many ways can we be in 3rd gear
now? Of course, this problem is easily solved by simply going through all
possible sequences of changes that might end in 3rd gear. But it becomes
significantly more difficult if we specify 50 or 200 (or n) changes in state. We
could also observe that to get to 3rd gear in n steps we must get to 2nd gear or
4th gear in n − 1 steps, and develop a recurrence, although this method would
fail if our graph were more complex, with more edges. Instead we consider a
6 × 6 matrix, the adjacency matrix A of the graph, which was introduced in
288 Introduction to Combinatorics

Chapter 10. This matrix has a 1 in the (i, j) position if there is an edge from
vertex i to vertex j, and a 0 otherwise. There is a theorem concerning the
adjacency matrix that is useful to us.

THEOREM 15.1 (Counting Walks)

If A is the adjacency matrix of a graph G, then the (i, j) entry of An is the
number of ways to walk from i to j along edges in exactly n steps.

The proof is left to Problem 15.2.

The application of the theorem is clear; we compute the matrix A for the
given graph (also shown in Figure 15.1). The calculation of A5 may be left
to a sophisticated calculator or computer algebra system. The (2, 5)-entry
(corresponding to changes from state 2, or stop, to state 5, or 3rd gear), is 5.
Some changes in state are irreversible. Consider the change of state of
an egg from whole (raw, unbroken) to broken (but uncooked) to scrambled.
In such a case, we use directed edges and the adjacency matrix will not be
symmetric. However, Theorem 15.1 still applies.
Example 15.3: How many ways are there to move from the vertex “Start”
to the vertex “Stop” of the graph in Figure 15.2 below?

 
0 1 1 0 0 0
Start a Stop  0 0
 1 1 0 0 

 0 0 0 0 1 1 
A=  0 0

0 0 1 0 
b d 
 0 0

0 0 0 1 
c 0 0 0 0 0 0
FIGURE 15.2: State graph and adjacency matrix.

In this case, we see a 0 in row 1, column 6, indicating that there is no edge

directly from Start (the vertex of row 1) to Stop (the vertex of column 6).
The two 1s in row 2 (representing vertex a) indicate the edges from a to b
(column 3) and to c (column 4), and the two 1s in column 5 represent the two
edges from b and c to d.
To solve the given problem, we need the various powers of A. Since A2 has
a 1 in the (1, 6) position, we know that there is one path of length 2 from
Start to Stop. The 2 in the (1, 6) position of A3 indicates two paths of length
3, namely Start-a-b-Stop and Start-b-d-Stop. Because the graph has only six
vertices, we don’t need powers beyond A6 , and in fact we may stop at A4 .
We find that there are only 5 ways to walk from Start to Stop. 
To calculate the matrix A + A2 + A3 + A4 is somewhat time-consuming by
hand, but that is what computers are for. Still, there is an easier method.
 Linear Algebra Methods in Combinatorics 289

THEOREM 15.2 (Sum of Powers)

I + A + A2 + · · · + An = (An+1 − I)(A − I)−1 if the inverse exists. Here I
denotes the identity matrix with the same dimensions as A.

Proof. Multiply the left-hand side by A − I; see Problem 15.3 for details. 
This theorem is the matrix analogue of the well-known formula for the sum
of a finite geometric progression. There is a similar analog for the sum of
an infinite geometric series, and that will apply in the case of the foregoing
example.

COROLLARY
P∞ i 15.2.1
−1
i=0 A = (I − A) , where A0 = I, and limn→∞ An = 0.
Proof. The proof of this corollary is a simple generalization of the proof from
calculus of the sum of an ordinary geometric series, applied to Theorem 15.2.

It is worth noting that the hypothesis that An → 0 is equivalent to the
assertion that all the eigenvalues of A are less than 1 in absolute value (or
modulus, if the eigenvalues are complex). The proof is beyond the scope of
this text.
Now applying the corollary to the matrix of our example, we find (I − A)−1
has a (1, 6) entry of 5, so there are a total of five paths from Start to Stop.

1 2 3 4 5 6 1 2 3 4 5 6
1 0 1 0 0 0 0 1 0 1 1 0 0 0
2 1 0 1 1 0 0 2 1 0 1 1 1 0
3 0 1 0 0 1 0 3 1 1 0 1 1 1
4 0 1 0 0 1 0 4 0 1 1 0 1 0
5 0 0 1 1 0 1 5 0 1 1 1 0 1
6 0 0 0 0 1 0 6 0 0 1 0 1 0
FIGURE 15.3: A diagram and two adjacency matrices.

Example 15.4: A marker is placed in the square labeled 1 in Figure 15.3.

In how many ways can we move the marker to square 6 in n steps if a step is
a move from a square to another square with which it shares an edge? In how
many ways if a step is a move to any square sharing an edge or a corner?
The first of the two matrices shown is the matrix that defines two squares
to be adjacent if they have a side in common. The answer is clearly the (1, 6)
entry of the nth power of this matrix, which we denote A, and that may be
easily computed for a specific n. If we choose n = 6, we find that the (1, 6)
entry is 12, so that there are exactly 12 paths of six steps in length from square
290 Introduction to Combinatorics

1 to square 6. If we wish to compute the number of ways in at most n steps,

we find that there is no shortcut; Theorem 15.2 will not help. This is because
the matrix I − A is singular, so we must sum the matrices individually. We do
this using a programmable calculator or computer algebra system. It is also
possible, but tedious, to use Microsoft Excel or another spreadsheet; Excel has
the built-in function MMULT to multiply two matrices, but does not have a
function to raise a matrix to a power. We find, for instance, I + A + · · · + A6
has a (1, 6) entry of 14, indicating that there are 14 ways to walk from square
1 to square 6 in six or fewer steps.
The second matrix is the “side or corner” matrix; so that, for example,
the (1, 3) entry is 1 because square 1 shares a corner (but not a side) with
square 3. Call this matrix B; we find the (1, 6) entry of B 6 to be 162, which
reflects the greater number of permissible steps. It happens that in this case
the matrix I − B is invertible; we may compute the sum I + B + · · · + B 6 or
the product (I − B 7 )(I − B)−1 . The two expressions give us exactly the same
result; the (1, 6) entry is 222.
This works even though the series ∞ i
P
i=0 B does not converge. How could
we tell that the series does not converge? Given any n, there is a walk from
square 1 to square 6 of length n whenever there is such a walk of length n − 2.
To find such a walk, we simply retrace a step. For instance, if our walk begins
by going from square 1 to square 2, we could simply insert two more steps in
our walk by stepping from square 2 back to square 1, and then from square 1
to square 2, and follow the rest of the walk. As a result, the number of walks
of length n from square 1 to square 6 never decreases as n goes to infinity so
the sum can have no finite (1, 6) entry. 

Example 15.5: Suppose we move a marker along the squares of Fig-

ure 15.3 subject to the constraint that the marker may only move from a
lower-number square to a higher-number square. In this case, we replace each
non-zero entry below the main diagonal of either matrix with a zero to get
two new matrices.

1 2 3 4 5 6 1 2 3 4 5 6
1 0 1 0 0 0 0 1 0 1 1 0 0 0
2 0 0 1 1 0 0 2 0 0 1 1 1 0
 = 3 0 0 0 0 1 0 B̂ = 3 0 0 0 1 1 1
4 0 0 0 0 1 0 4 0 0 0 0 1 0
5 0 0 0 0 0 1 5 0 0 0 0 0 1
6 0 0 0 0 0 0 6 0 0 0 0 0 0
Now, however, if we raise these matrices to the sixth power, we get a matrix
of all zeros, because clearly we can move at most five steps before we come to
a halt. We can raise either  or B̂ to any lower power than 6; so if we find,
for example, B̂ 3 , its (1, 6) entry is 3, indicating three ways to go from square
1 to square 6 in three steps by crossing sides or corners and never moving to
a lower-numbered square. If we compute (I − B̂)−1 , we find its (1, 6) entry is
 Linear Algebra Methods in Combinatorics 291

8, indicating that there areP∞ a total of 8 ways to walk from square 1 to square
6. How do we know that i=0 B̂ i converges? Simply because any power of B̂
past the fifth power is a matrix of all zeros! Another way to obtain the same
result would be to place a 1 in the (6, 6) entry of B̂, and raise the resulting
matrix to any power greater than 4. This would work because, for instance,
a walk from square 1 to square 6 in three steps would give us a unique walk
from square 1 to square 6 in four steps in which the last step was from square
6 to itself. The result is this matrix.

1 2 3 4 5 6 1 2 3 4 5 6
1 0 1 1 0 0 0 1 0 0 0 0 0 8
2 0 0 1 1 1 0 2 0 0 0 0 0 5
B̃ = 3 0 0 0 1 1 1 B̃ 5 = 3 0 0 0 0 0 3
4 0 0 0 0 1 0 4 0 0 0 0 0 1
5 0 0 0 0 0 1 5 0 0 0 0 0 1
6 0 0 0 0 0 1 6 0 0 0 0 0 1

Notice that any power of B̃ greater than 4 gives us the same matrix. We
get a matrix whose only non-zero entries are in column 6, and the (1, 6) entry
is, as expected, 8. Similar remarks apply to the matrix Â. 

Example 15.6: In how many ways can we form a string of n = 3 letters

from the set {a, b, c, d, e} subject to the restrictions that e is the only letter
that may appear twice in succession, the only letters that may follow b or c
or d are a and e (that is, each consonant must be followed by a vowel), and c
may not come directly after e?
This problem is similar to problems that we have done by means of inclusion-
exclusion (Chapter 5) and by means of generating functions (Chapter 6), but
more complicated. We form a directed multigraph whose vertices correspond
to the letters, and a directed edge will go from one vertex to the next only
when the corresponding letters may appear in succession in a legal string.
The graph and its adjacency matrix appear in Figure 15.4.

d 
0 1 1 1 1

a c
 1 0 0 0 1 
 
A=
 1 0 0 0 1 

b  1 0 0 0 1 
1 1 0 1 1
e
FIGURE 15.4: Digraph and adjacency matrix for the string problem.

Notice that the (5, 3) entry is 0, but the (3, 5) entry is 1. This represents
the existence of an edge from c to e but not from e to c, because we forbid
292 Introduction to Combinatorics

strings containing ec. Now any legal string of length n corresponds to a walk
along n − 1 edges of the graph and because arbitrarily long walks are possible,
there is no power n such that An = 0. An attempt to apply Corollary 15.2.1
will not work because the matrix (I − A)−1 does not have integer entries.
To enumerate walks of length 2, we find A2 , which has entries from 1 to 4,
indicating the number of walks with the corresponding first and last vertices.
Since the (5, 4) entry is 2, there are two walks of length from d to e. We can
see these walks from the diagram; they are d-e-e and d-a-e. The sum of all
the matrix entries is 46, so there are 46 strings of three characters that satisfy
the given constraints. 

Example 15.7: Find the number of binary strings of length n that:

(a) Do not contain the string 11;
(b) Do not contain the string 111.
A solution to part (a) was done in Section 6.7. To use the transfer matrix,
we take a graph with two vertices, V = {0, 1} with an edge between the two
vertices and a loop at 0 but not at 1. In this case the various non-zero powers
of A form an infinite sequence, and we borrow aPconcept from Chapter 6 on
generating functions. Look at the expression i≥0 Ai xi . Similarly to the
result of Corollary 15.2.1, we may write the sum formally as (I − xA)−1 .
Because this is a formal sum, the question of convergence is unimportant.
Now, a string of length n + 1 corresponds to a walk of length n on the graph.
Each entry in the matrix (I − xA)−1 is the generating function of the number
of length-(n + 1) binary strings that begins at the vertex that corresponds to
the row and ends at the vertex that corresponds to the column.
   −1  
−1 1 0 1 1 1 1 x
(I − xA) = −x =
0 1 1 0 1 − x − x2 x 1 − x

Thus, the number of walks beginning at 0 and ending at 0 is generated by

1/(1 − x − x2 ), and so on. To find the total, we add all entries, and discover
the generating function (2 + x)/(1 − x − x2 ). We can then prove rigorously if
we like that the number of strings is the Fibonacci number fn+1 .
Part (b) will be a little trickier, and requires us to use a bigger graph. Here
the vertices will be {00, 01, 10, 11} and we will have a directed edge from ab
to bc only if abc is a legitimate substring. The digraph that results is shown
in Figure 15.5.

 
00 10 1 1 0 0
0 0 1 1 
A=
1

1 0 0 
01 11 0 0 1 0
FIGURE 15.5: Digraph and adjacency matrix.
 Linear Algebra Methods in Combinatorics 293

In this case, a string of length n corresponds to n−2 steps along the digraph,
because one step (say from 01 to 11) corresponds to the string 011. Thus the
coefficient of xn in our generating function will be the number of strings of
length n + 2. Now we find the matrix (I − xA)−1 and add the entries. This
is more easily done with a symbolic algebra program than by hand, and gives
us the result f (x) = (−2x2 − 3x − 4)/(x3 + x2 + x − 1). Long division gives us
4 + 7x + 13x2 + 24x3 + 44x4 + 81x5 + . . . . For n = 2 there are clearly four such
strings, so we check for n = 3 and find seven strings. For n = 4 there are 13
strings, as we may check by enumeration, and we should be content that we
have not made any mistakes since the coefficients agree. Checking the cases
of n = 5 and n = 6 is a pleasant exercise in inclusion-exclusion. We may also
deduce, using the methods of Chapter 6, that the sequence an we are looking
at satisfies the recurrence an = an−1 + an−2 + an−3 . 
We conclude with a look at another application of the transfer-matrix
method.
Example 15.8: In how many different ways may we cover all the squares
of a 4 × n chessboard with nonoverlapping dominos?
To solve this problem we represent it as a problem enumerating walks on
an appropriate graph. Let each row of a 4 × n chessboard be represented by
a string of four bits, where a 1 represents a domino in the vertical position
extending into the next row, and a 0 represents a domino in a horizontal
position or extending from the row below. It follows that the topmost row
must be represented by 0000. Of the 16 possible bit strings, we find only six
may represent rows of a properly covered 4×n chessboard. Thus, for example,
the string 0000 may represent a row containing two horizontal dominos, or
one horizontal domino and two vertical dominos that rise from the lower row,
or four vertical dominos from below. A little thought reveals that no binary
string with an odd number of 1s can work since it can never “fit” with the
topmost 0000 row. Similarly, a row represented by 1010 must be followed by
a row of 0101, and the two alternate, never leading to any way to create a
0000 row. The other strings are 0011, 0110, 1001, 1100, and 1111. Now we
make a graph of these six strings and make a directed edge from the first to
the next if a row of the chessboard corresponding to the first can fit below
one corresponding to the second. The graph and the matrix that corresponds
to it are shown in Figure 15.6.
For any specific n, the (2, 2)-entry of An will give the number of proper
coverings of a 4 × n chessboard by dominos. Thus, a walk of n = 3 steps, say
0000-1100-0011-0000, corresponds to the domino tiling of the 4 × 3 board in
which two vertical dominos occupy the lower-left four squares and two other
vertical dominos occupy the upper-right four squares; and horizontal dominos
occupy the upper-left and lower-right corners.
This matrix is large enough that to find (I − xA)−1 we use a computer
algebra system. In Mathematica, for instance, we define A and compute
(I − xA)−1 as follows.
294 Introduction to Combinatorics

1111 0000 1100 0110 0011 1001

1111 0 1 0 0 0 0 0011 1111 1001
0000 1 1 1 0 1 1
A = 1100 0 1 0 0 1 0
0110 0 1 0 0 0 1
0011 0 1 1 0 0 0 1100 0000 0110
1001 0 1 0 1 0 0
FIGURE 15.6: Digraph and adjacency matrix for domino tiling.

A = {{0, 1, 0, 0, 0, 0}, {1, 1, 1, 0, 1, 1}, {0, 1, 0, 0, 1, 0},

{0, 1, 0, 0, 0, 1}, {0, 1, 1, 0, 0, 0}, {0, 1, 0, 1, 0, 0}};

B = Inverse[IdentityMatrix[6] - x*A];

B[[2,2]]

The semicolons at the end of each line suppress display of the results; the last
line produces the (2, 2)-entry of B, namely,

1 − 2x2 + x4
.
1 − x − 6x2 − x3 + 6x4 + 2x5 − x6
We can expand this in a power series:

Series[%,{x,0,10}]

and we get 1 + x + 5x2 + 12x3 + 38x4 + 107x5 + 316x6 + 915x7 + 2671x8 +

7771x9 + 22640x10 + O[x]11 .
The constant term of 1 indicates that there is one way of covering a 4 × 0
board with dominos (which is to do nothing). The x term says that there
is one way to cover a 4 × 1 board with dominos, which we easily see is to
place two dominos horizontally on the four squares of the board. That the
coefficient of x2 is 5 follows from the fact that a 4 × 2 board is also a 2 × 4
board, and the result from Chapter 6 about the number of domino tilings of
a 2 × n board. 

15.3 Kasteleyn’s Permanent Method

The permanent of a matrix may be described, loosely, as “the
 determinant

a b
without the minus signs.” More precisely, the matrix A = has
c d
permanent Per(A) = ad + bc. For a 3 × 3 or larger matrix, the permanent
 Linear Algebra Methods in Combinatorics 295

may be computed by Laplace expansion, without the power of −1 that causes

terms in the summation to alternate in sign. To put it another way, if the
reader recalls that the determinant of a matrix, Det(A), may be written as

(−1)sgn(σ) a1,σ(1) a2,σ(2) . . . an,σ(n)

X
Det(A) =
σ
then the permanent may be written as
X
Per(A) = a1,σ(1) a2,σ(2) . . . an,σ(n). (15.1)
σ

In each case the summation is over all permutations on n letters.

Why, the reader will wonder, this temporary digression on the permanent?
The answer is that it has applications in enumeration. To see why, let’s con-
sider a balanced bipartite graph such as C4 . We may recall from Chapter 10
that there may exist a perfect matching, or one-factor. We may also see from
the adjacency matrix that all the information about the graph’s structure is
preserved in only one quarter of the matrix, because the matrix of a balanced
bipartite graph on 2n vertices is comprised of two n × n matrices of all zeroes
together with two copies of an n × n matrix that describes the edges of the
bipartite graph. This n × n matrix is sometimes called the bipartite adjacency
matrix , and we shall denote it B(G). More formally, B(G) is the n × n ma-
trix whose rows correspond to vertices in one part of the bipartite graph, and
columns correspond to vertices in the other part of the bipartite graph, and
the (i, j) position is 1 if i is adjacent to j in the bipartite graph, 0 otherwise.

THEOREM 15.3 (Matchings in Bipartites)

The balanced bipartite graph B has exactly Per(B(G)) perfect matchings.

Proof. We establish a bijection between matchings of G and permutations of

n letters that contribute a 1 to the sum of Equation 15.1. Choose any perfect
matching of the balanced bipartite graph G with parts X and Y , and form
A = B(G) by letting the rows of A correspond to vertices of the part X, and
the columns of A correspond to the vertices of Y . Now let the permutation σ
satisfy σ(i) = j exactly when vertex xi of X has an edge in the matching to
vertex yj of Y . Since each vertex of X has by definition exactly one neighbor
in the matching, this defines a permutation on n letters, and corresponds to
one summand of the sum of Equation 15.1. This summand must have the
value 1, because every matrix entry of the form ai,σ(i) is a 1 that corresponds
to an edge of the matching. In the same way, every non-zero summand of 15.1
corresponds to a permutation that arises from some perfect matching. It is
easy to see that distinct perfect matchings give rise to distinct permutations
and vice-versa, so the theorem is proven. 
The foregoing discussion of the uses of the permanent may be generalized.
Suppose we have a family of k subsets D = {Bi : 1 ≤ i ≤ k} of an n-set U. We
296 Introduction to Combinatorics

will recall from Chapter 1 that a system of distinct representatives or SDR is

a subset R = {xi : 1 ≤ i ≤ k} of the n-set with the property that xi ∈ Bi
for each i, and of course the elements are distinct (because R is a set, not a
multiset). Theorem 1.2 gave a necessary and sufficient condition for an SDR
to exist. However, this result, while it establishes the existence of an SDR,
gives us no help in determining how many SDRs may exist.
We can modify the definition of the permanent so as to count the number
of SDRs, or equivalently to count the number of matchings of an unbalanced
bipartite graph that use all the vertices of the smaller part. The graph will
have k vertices corresponding to the sets Bi in one part, and n vertices corre-
sponding to the elements of the n-set U in the other part. We assume k ≤ n
and look at the k × n bipartite adjacency matrix A of the graph; let V be the
set of all k-permutations of the n-set U. We define the permanent of A to be

X
Per(A) = a1,σ(1) a2,σ(2) . . . an,σ(n). (15.2)
σ∈V

Then the permanent of A again gives the number of SDRs of the associated
family of subsets; the proof is exactly analogous to the proof of Theorem 15.3.
Equation 15.1 is unsatisfactory as a way to compute the permanent; we
mentioned above that the equivalent of the familiar Laplace expansion for
the determinant will work. Thus, the permanent of any 2 × 2 matrix was
given at the beginning of this section as ad + bc; we can use this to compute
the permanent of a larger matrix. So if Ai is the result P of deleting row 1
and column i from the matrix A, we may write Per(A) = ni=1 a1,i Per(Ai ).
However, caution is required when using this technique. In particular, the
traditional shortcut to computing the determinant with Laplace expansion
is to perform some elementary row operations to create a row or column
with many zero entries, and expand about that row or column. This will
work because the determinant is invariant under elementary row or column
operations; however, the permanent is not. Thus, this shortcut will yield an
incorrect value for the permanent. It is permissible to expand the permanent
about a different row of the matrix, as this merely corresponds to relabeling
the vertices of the graph, or reordering the sets Bi . Also, because Per(AT ) =
Per(A), we may expand about any column when A is square.

Example 15.9: Three committees consist of five people altogether. Com-

mittee A is made up of 1, 2, and 3. Committee B contains 1, 3, and 5, and
Committee C consists of 2, 3, 4, and 5. In how many ways can I choose an
SDR consisting of one member of each committee?

The stated problem gives us a 3 × 5 matrix A; we will find its permanent.

 Linear Algebra Methods in Combinatorics 297

 
1 2 3 4 5
A 1 1 1 0 0
PerA = Per 
B 1 0 1 0 1


C 0 1 1 1 1
     
0 1 0 1 1 1 0 1 1 0 0 1
= Per + Per + Per
1 1 1 1 0 1 1 1 0 1 1 1

This equation came from expanding about the first row, corresponding to
the set A. We continue the process; for instance,
 
0 1 0 1
Per = Per[1 1 1] + Per[1 1 1].
1 1 1 1
Now, to find the permanent of a 1 × 3 matrix, we simply add the entries. In
this case, each of these permanents is 3. We get 18 SDRs in all from these
sets. 

Example 15.10: Use the permanent method to calculate the number of

matchings in K3,3 .
 
1 1 1  
1 1
A =  1 1 1  , so Per(A) = 3 · Per = 3·2 =6
1 1
1 1 1
Or, as we knew before, there are 6 permutations on three letters. 

Example 15.11: The 4-rung ladder L4 and B(L4 ) are shown below. How
many matchings does this graph have?
2 4 6 8
1 1 0 0 1
3 1 1 0 1
5 0 1 1 0
7 0 1 1 1
We expand about the first row, as it has only two 1s; we obtain
   
1 0 1 1 1 0
Per(A) = Per  1 1 0  + Per  0 1 1  .
1 1 1 0 1 1
The first permanent on the right-hand side may be expanded about the first
row; the second, however, will most efficiently be expanded about the first
column (which has only a single non-zero element). We get
     
1 0 1 1 1 1
Per(A) = Per + Per + Per = 5.
1 1 1 1 1 1
The graph has five perfect matchings, verifiable by listing them. 
298 Introduction to Combinatorics

By itself, the permanent is of little use in computing matchings for individ-

ual bipartite graphs; its real utility comes in working with families of graphs,
using theorems about the permanent. We illustrate with two such theorems,
provided here without proof.
For a specific m × n matrix A = [aij ], with m ≤ n, let Ar denote one of the
matrices formed by deleting r of the columns of A, where n − m ≤ r ≤ n − 1.
We will add the elements of each row of Ar , and then multiply the
resulting
row sums together to get a quantity we denote ΠAr . There are nr ways to
form A
r by choosing which columns to delete; let ΣΠAr denote the sum over
all nr possible Ar of the quantity ΠAr .

THEOREM 15.4 (Ryser’s Formula)

 
n−m+1
Per(A) = ΣΠAn−m − ΣΠAn−m+1 +
1
 
n−1
· · · + (−1)m−1 ΣΠAn−1
m−1

COROLLARY 15.4.1 Pn−1

For a square matrix A, we have Per(A) = i=0 (−1)i ΣΠAi .
The proof of Theorem 15.4 may be found in [94]; Corollary 15.4.1 follows
by setting m = n. For one use of this result, consider the matrix Jn , an
n × n matrix with 1 in each entry. Clearly, Per(J) = n! since J is B(Kn,n ).
Now, consider Jn − In ; this is B of Kn,n with a perfect matching removed,
and its permanent is the number Dn of derangements on n letters. Now,
applying the corollary, we note that the various matrices Ar all have row
sums n − r or n − r − 1. A little thought lets
us see that a particular ΠAr will
be (n − r)r (n − r − 1)n−r , and there are nr such matrices Ar . We get
 
X n
Dn = (−1)r (n − r)r (n − r − 1)n−r .
r

This formula is different from the usual formula for derangements given in
Chapter 5.

THEOREM 15.5 (van der Waerden’s Inequality)

Let A be an n × n matrix with non-negative entries where the entries in each
row sum to k. Then Per(A) ≥ n!(k/n)n .

Although Bartel Leendert van der Waerden conjectured that this inequality
was true back in 1926, it was not until 1981 that it was first established by
 Linear Algebra Methods in Combinatorics 299

G. P. Egorychev and, independently, by D. I. Falikman (see [69]). This result

tells us more than just that a regular bipartite graph has a matching; it tells us
that it must have many of them. Since the permanent must be an integer, we
will round our answers up using the notation ⌈x⌉ for the least integer greater
than or equal to x.
Example 15.12: How many matchings are in a 3-regular bipartite graph
on 12 vertices?
This graph will have a 6×6 bipartite adjacency matrix; because it is regular
of degree 3, each row will have three 1s and three 0s, so k = 3. Theorem 15.5
tells us that the graph must have (using n = 6) at least ⌈6!/26 ⌉ = 12 match-
ings, no matter how the edges are arranged. 
For more on this subject, as well as for other applications of the permanent,
the interested reader is referred to ([69], Chapter 8).

Exercises 15A
1. Explain how to find, for some square matrix A, the matrix A82 using
only 8 matrix multiplications.

2. Solve each recurrence using the methods of this chapter:

(i) an+1 = 2an + 3an−1 , a0 = 2, a1 = 2.
(ii) an+1 = 2an + 3an−1 , a0 = 1, a1 = 7.
(iii) bn+1 = 2bn + 8bn−1 , b0 = −1, b1 = 8.
(iv) bn+1 = 2bn + 8bn−1 , b0 = 2, b1 = 2.

3. Suppose a machine has five states A, B, C, D, and Halt. From state A

it may transfer to either state C or D; from B, it may go to D or Halt;
from C it may transfer to B or D; from D it may go to A or Halt; and
it cannot transfer out of state Halt.
(i) Find the state digraph that represents this machine.
(ii) Find the adjacency matrix M of the digraph of part (i).
(iii) Compute M 4 ; in how many ways can the machine go from state A
to Halt in exactly four steps?

4. Let M be the matrix from the preceding exercise, and suppose M̂ is

formed from M by changing the (5, 5) entry from 0 to 1 (that is, allowing
the machine to go from state Halt to Halt). Explain why this matrix
would be useful in determining the number of ways that the machine
can go from state A to state Halt in at most four steps. How will this
compare to calculating the matrix (I5 − M 5 )(I5 − M )−1 ?
300 Introduction to Combinatorics

5. A marker starts in square 1 and may move from a square to any square
sharing a side with that square. For each of the four diagrams shown,
find the adjacency matrix A of the graph. Compute (I − A)−1 if it
exists, and find the number of walks of six or fewer steps from square 1
to square 6.

All All
All All All All
All All All
All All
All All All All
AllAll
All All
6. For the same diagrams as the previous exercise, a marker may move to
a square with which its current square shares a side or a corner. As
above, find the adjacency matrix A, the matrix (I − A)−1 if it exists,
and the number of walks from square 1 to square 6 in six or fewer steps.
7. For the same diagrams as Exercise 5, a marker may move only to a
higher-numbered square with which its current square shares a side or
a corner. Find the adjacency matrix A by replacing all non-zero entries
below the main diagonal of the matrices from Exercise 6, find the matrix
(I − A)−1 if it exists, and compute the number of walks from square 1
to square 6 in six or fewer steps.
8. For the same diagrams, allow the marker to move as in Exercise 7, but
also to move from square 6 to square 6. Find (I − A)−1 , 6i=1 Ai , and
P
the (1, 6) entry of A6 . What is the number of ways to move from square
1 to square 6 in six or fewer moves?
9. Suppose that the set A consists of the odd numbers between 1 and 6
inclusive; B consists of the even numbers in that range; and C consists
of the numbers divisible by 3 in that range. Construct a matrix A and
use this matrix to find the number of possible SDRs for this collection
of sets.
10. Find the number of perfect matchings for the bipartite graph shown.

11. Use van der Waerden’s inequality to determine the minimum number
of perfect matchings of a 4-regular bipartite graph on 16 vertices (8 in
each part).
 Linear Algebra Methods in Combinatorics 301

Exercises 15B
1. Explain how to find, for some square matrix A, the matrix A61 , using
at most nine matrix multiplications.
2. Solve each recurrence using the methods of this chapter:
(i) an+1 = 2an + 3an−1 , a0 = 2, a1 = 2.
(ii) an+1 = 2an + 3an−1 , a0 = 1, a1 = 5.
(iii) bn+1 = 2bn + 15bn−1 , b0 = 1, b1 = 1.
(iv) bn+1 = 2bn + 15bn−1 , b0 = 1, b1 = 5.
3. Suppose a machine has six states, A, B, C, D, E, and Halt. From state
A we may go to B or C; from B, we may go to C, D, or Halt ; from C
we may only go to D; from D we may go to E or Halt ; and from E we
may only go to C.
(i) Find the state digraph that represents this machine.
(ii) Find the adjacency matrix M of the digraph of part (i).
(iii) Compute M 6 ; in how many ways can the machine go from state A
to Halt in exactly six steps?
4. Let M be the matrix from the preceding exercise, and suppose M̂ is
formed from M by changing the (6, 6) entry from 0 to 1 (that is, allowing
the machine to go from state Halt to Halt ). Explain why this matrix
would be useful in determining the number of ways that the machine
can go from state A to state Halt in at most six steps. How will this
compare to calculating the matrix (I6 − M 6 )(I6 − M )−1 ?
5. A marker starts in square 1 and may move from a square to any square
sharing a side with that square. For each of the four diagrams shown,
find the adjacency matrix A of the graph. Compute (I − A)−1 if it
exists, and find the number of walks of six or fewer steps from square 1
to square 6.
All
All
All All All All All
All All All All All
All All
All AllAll All
All All
All All
6. For the same diagrams as the previous exercise, a marker may move to
a square with which its current square shares a side or a corner. As
above, find the adjacency matrix A, the matrix (I − A)−1 if it exists,
and the number of walks from square 1 to square 6 in six or fewer steps.
302 Introduction to Combinatorics

7. For the same diagrams as the previous exercise, a marker may move
only to a higher-numbered square with which its current square shares a
side or a corner. Find the adjacency matrix A by replacing all non-zero
entries below the main diagonal of the matrices from Exercise 6, find
the matrix (I − A)−1 if it exists, and compute the number of walks from
square 1 to square 6 in six or fewer steps.

8. For the same diagrams, allow the marker to move as in Exercise

P6 7, but
also to move from square 6 to square 6. Find (I − A)−1 , i=1 Ai , and
the (1, 6) entry of A6 . What is the number of ways to move from square
1 to square 6 in six or fewer moves?

9. For the set {1, 2, 3, 4, 5, 6, 7, 8}, there are three subsets A = {3, 4, 5, 6},
B = {2, 4, 7}, and C = {1, 4, 7, 8}. How many SDRs are possible?

10. Find the number of perfect matchings for the bipartite graph shown.

11. Use van der Waerden’s inequality to determine the minimum number
of perfect matchings of a 3-regular bipartite graph on 20 vertices (10 in
each part).

Problems

Problem 1: We will modify the approach of Section 15.1 to deal with a

three-term recurrence and solve the recurrence an+1 = 2an + an−1 − 2an−2
with a0 = 1, a1 = 0, and a2 = 4.
   
an−2 an−1
(i) Find a matrix A such that A  an−1  =  an .
an an+1
(ii) Find the eigenvalues of the matrix A and corresponding eigenvectors
(you should have three distinct eigenvalues).
(iii) Form the matrix P whose columns are the eigenvectors and find P −1 .
(iv) Find the diagonal matrix D and use it to solve the recurrence.
 Linear Algebra Methods in Combinatorics 303

Problem 2: Use induction to prove Theorem 15.1.

Problem 3: Use the distributive law to find the matrix product (I + A +
A2 + · · · + An )(I − A).
Problem 4: Dr. John Walters and Dr. Walter Johns assist the biology de-
partment in a mathematical analysis of the life cycle of an obscure creature,
the nerdus academicus. The creature goes through five different phases of its
life cycle, colloquially referred to as grad, lecturer, assistant, associate, and
full. However, the progression through the different phases (which may occur
every three to seven years) is not regular. Thus, the grad stage may only
develop into the lecturer stage, from which the creature may remain in the
lecturer state or advance to assistant. From assistant three possible outcomes
exist; the creature may revert to lecturer, remain as assistant, or progress to
associate. It is virtually unheard-of for the creature to revert to the assistant
phase from associate, so the possibilities from there are to remain associate
or to progress to the final full stage. From that stage, no development occurs,
so the creature remains in that position for the remainder of its life, after
which its body (or oeuvre, to use the specialized term) will be used to feed the
next generation. Draw a state graph representing the life cycle of this bizarre
creature, and find its adjacency matrix. In how many ways can the creature
progress from grad to full in six steps?
Problem 5: Prove or find a counterexample: If a matrix has a permanent of
0 it must also have a determinant of 0.
√
Problem 6: Use Stirling’s approximation that says n! ≃ nn e−n 2πn and
van der Waerden’s inequality to get an approximation to a lower bound for
the number of perfect matchings of a bipartite graph that is regular of degree
k = 2. Then use L’Hopital’s rule to find the limit of this expression as n goes
to infinity.
Problem 7: Construct a family of 2-regular bipartite graphs on 2n vertices
for which the number of matchings increases exponentially as n increases.
Appendix A: Sets and Proof
Techniques

AA.1 Sets and Basic Set Operations

A set is a collection of objects with a well-defined rule, called the mem-
bership law , for determining whether a given object belongs to the set. The
individual objects in the set are called its elements or members and are said
to belong to the set. If S is a set and s is one of its elements, we denote this
fact by writing
s∈S
which is read as “s belongs to S” or “s is an element of S.”
A set may be defined by listing all the elements, usually between braces;
if S is the set consisting of the numbers 3 and 5, then S = {3, 5}. We write
{1, 2, . . . , 10} to mean the set of all integers from 1 to 10. Another method
is the use of the membership law of the set: for example, since 3 and 5 are
precisely the solutions of the equation x2 + 8x + 15 = 0, we could write

S = {x : x2 + 8x + 15 = 0}

(“the set of all x such that x2 + 8x + 15 = 0”). The set of integers from 1 to
10 is
{x : x integral, 1 ≤ x ≤ 10}.
Sometimes a vertical line is used instead of a colon, so we might instead write

{x|x integral, 1 ≤ x ≤ 10}.

The set of all integers is denoted Z, so we could also say

{1, 2, . . . , 1} = {x : x ∈ Z, 1 ≤ x ≤ 10}.

The positive integers or natural numbers are denoted Z+ or N. Z0 denotes

the non-negative integers. The rational numbers, real numbers and complex
numbers are denoted Q, R and C respectively. These sets are infinite (as
opposed to finite sets like {0, 1, 3}).
Example AA.1: Find three ways to describe the set consisting of the
three numbers 0, 1 and 2.

305
306 Introduction to Combinatorics

One way is to list the elements: {0, 1, 2}. The listing could also be expressed
as {x : x integral, 0 ≤ x ≤ 2} (or {x : x ∈ Z, 0 ≤ x ≤ 2}). Or we could
note that 1, 2 and 3 are the solutions of x3 + 3x2 + 2x = 0, and write {x :
x3 + 3x2 + 2x = 0}. 
The number of elements in a set is called its order or size. We shall denote
the order of S by |S| (another common notation is #(S)). For finite sets,
the order will be a non-negative integer; for infinite sets we usually use the
infinity symbol ∞.
Suppose n is a positive integer. The residue classes modulo n are the n sets
[0], [1], . . . , [n − 1], where [i] is the set of all integers that leave remainder i on
division by n:

[n] = {. . . , i − kn, . . . , i − n, i, i + n, . . . , i + kn, . . .}.

The set of residue classes modulo n is denoted Zn . Members x and y of

the same residue classes are called congruent modulo n (see Section 15) and
we write y ≡ x(mod n). The operations of addition and multiplication are
defined on Zn in the obvious way:

[x] + [y] = [x + y] ; [x] × [y] = [x × y].

When no confusion arises we simply write x instead of [x].

It is sometimes useful to discuss number systems with the number 0 omitted
from them, especially when division is involved. We denote this with an
asterisk: for example, Z∗ is the set of non-zero integers.
The notation S ⊆ T means that every member of S is a member of T :

x ∈ S ⇒ x ∈ T.

Then S is called a subset of T . We also say that T contains S or T is a

superset of S, and write T ⊇ S. Sets S and T are equal, S = T , if and only if
S ⊆ T and T ⊆ S are both true. We can represent the situation where S is a
subset of T but S is not equal to T —there is at least one member of T that
is not a member of S—by writing S ⊂ T .
The number systems satisfy Z+ ⊆ Z ⊆ Q ⊆ R. Rational numbers that are
not integers are called proper fractions, and real numbers that are not rational
are called irrational numbers.
The definition of a set does not allow for ordering of its elements, or for
repetition of its elements. Thus {1, 2, 3}, {1, 3, 2} and {1, 2, 3, 1} all represent
the same set (which could be written {x | x ∈ Z∗ and x ≤ 3}, or {x ∈ Z∗ |
x ≤ 3}). To handle problems that involve ordering, we define a sequence to
be an ordered set. Sequences can be denoted by parentheses; (1, 3, 2) is the
sequence with first element 1, second element 3 and third element 2, and is
different from (1, 2, 3). Sequences may contain repetitions, and (1, 2, 1, 3) is
quite different from (1, 2, 3); the two occurrences of object 1 are distinguished
by the fact that they lie in different positions in the ordering.
 Sets; Proof Techniques 307

Another way to deal with multiple elements is to use multisets. This is

particularly useful when the objects are distinct but we choose to ignore dis-
tinctions between them. Thus, we might say that a set of items from the
store consists of twelve eggs, three tomatoes, and four onions, and not de-
scribe the set as consisting of 19 elements, but as a set of three elements
in which each element is repeated. This multiset might be described as
{egg × 12, tomato × 3, onion × 4}. The number of times that an element
may be repeated in a multiset is variously referred to as its multiplicity or its
repetition number ; we shall use either as seems appropriate. A multiset S is
said to be a multisubset of a multiset T if the multiplicity of each element of
S is less than or equal to the multiplicity of that element in T . Equivalently,
S ⊆ T provided that each element has at least as large a repetition number
in T as in S.
Example AA.2: Which of these are not multisubsets of the multiset
S = {a × 3, b × 2, x × 1, y × 2, z × 3}?
A = {a × 1, b × 2}.
B = {a × 2, b × 2, x × 2, y × 2, z × 2}.
C = {a × 1, b × 1, c × 1}.
Clearly, A is a multisubset of S because the elements of A do not appear
more times in A than in S. B and C are not multisubsets of S because each
has an element that appears more times than in S. B has two copies of b
where S has one; C has a copy of c where S has none. 
An important concept is the empty set, or null set, which has no elements.
This set, denoted by ∅, is unique and is a subset of every other set.
In all the discussions of sets in this book, we shall assume (usually without
bothering to mention the fact) that all the sets we are dealing with are subsets
of some given universal set U . U may be chosen to be as large as necessary
in any problem we deal with; in most of our discussion so far we could have
chosen U = Z or U = R. U can often be chosen to be a finite set.
The power set of any set S consists of all the subsets of S (including S itself
and ∅), and is denoted by P(S):
P(S) = {T : T ⊆ S}. (1)
The power set is a set whose elements are themselves sets.
Given sets S and T , the union of S and T is the set
S ∪ T = {x : x ∈ S or x ∈ T (or both) },
the intersection of S and T is the set
S ∩ T = {x : x ∈ S and x ∈ T },
and the relative complement of T with respect to S (or alternatively the set-
theoretic difference or relative difference between S and T ) is the set
S\T = {x : x ∈ S and x 6∈ T }.
308 Introduction to Combinatorics

So S ⊆ T if and only if S\T = ∅.

The relative complement of T with respect to the universal set is written
T and called the complement of T . With this notation S\T = S ∩ T , since
each of these sets consists of the elements belonging to S but not to T .
If two sets, S and T , have no common element, then S and T are called
disjoint . The sets S\T and T are necessarily disjoint, and in particular, S
and S are always disjoint. A partition of a set S is a collection of non-empty
subsets S1 , S2 , . . . , Sn of S such that each pair of them are disjoint; we say
that these sets are mutually or pairwise disjoint.
Both union and intersection are commutative; in other words
S∪T =T ∪S
and
S ∩ T = T ∩ S,
for any sets S and T . Similarly, the associative laws
R ∪ (S ∪ T ) = (R ∪ S) ∪ T
and
R ∩ (S ∩ T ) = (R ∩ S) ∩ T
hold. The associative law means that we can omit brackets in a string of unions
(or a string of intersections); expressions like (A∪B)∪(C∪D), ((A∪B)∪C)∪D
and (A ∪ (B ∪ C)) ∪ D, are all equal, and we usually omit all the parentheses
and simply write A ∪ B ∪ C ∪ D. Combining the commutative and associative
laws, we see that any string of unions can be rewritten in any order: for
example,
(D ∪ B) ∪ (C ∪ A) = C ∪ (B ∪ (A ∪ D)) = (A ∪ B ∪ C ∪ D).
The following distributive laws hold:
R ∪ (S ∩ T ) = (R ∪ S) ∩ (R ∪ T ); (2)
R ∩ (S ∪ T ) = (R ∩ S) ∪ (R ∩ T ); (3)
(R ∪ S)\T = (R\T ) ∪ (S\T ). (4)
We have also the equation
R\(S ∪ T ) = (R\S) ∩ (R\T ), (5)
and the analogous
R\(S ∩ T ) = (R\S) ∪ (R\T ). (6)
When we take the particular case where R is the universal set in (5) and
(6), those two equations become de Morgan’s laws:
S ∪ T = S ∩ T, (7)
S ∩ T = S ∪ T. (8)
 Sets; Proof Techniques 309

These laws extend to three or more sets in the obvious way: for example,

R ∪ S ∪ T = R ∩ S ∩ T.

Another property is that both union and intersection are idempotent . That
is,
S∪S =S∩S =S
for any set S.
We often illustrate sets and operations on sets by diagrams. A set R is
represented by a circle, and the elements of R correspond to points inside the
circle. If we need to show a universal set (for example, if complements are
involved), it is shown as a rectangle enclosing all the other sets. These illus-
trations are called Venn diagrams, because they were popularized by George
Venn [118, 119].
Here are Venn diagrams representing R ∪ S, R ∩ S, R, R\S and R ∩ S ∩ T ;
the set represented is shown by the shaded area, and a universal set is also
shown.

R S R S

R ∪S R∩S
R R S

_
R R\S
R

S T

R∩S∩T

Two sets are equal if and only if they have the same Venn diagram. In
order to illustrate this, we again consider the distributive law

R ∩ (S ∪ T ) = (R ∩ S) ∪ (R ∩ T ). (2)

The Venn diagram for R ∩ (S ∪ T ) is constructed in the upper half of Figure

A.1, and that for (R ∩ S) ∪ (R ∩ T ) is constructed in the lower half. The two
are obviously identical.
310 Introduction to Combinatorics

R S R S R S

T T T
R S T R (S T)

R S R S R S

T T T
R S R T (R S) (R T)

FIGURE A.1: R ∩ (S ∪ T ) = (R ∩ S) ∪ (R ∩ T ).

Example AA.3: Consider the set-theoretic identity

(R ∪ S) ∩ C ⊆ (R ∪ T ) ∩ (S ∪ T ).

This can be established using Venn diagrams.

The diagram is:

R S R S

T T

(R » S) « T (R » T) « (S » T)
Sometimes we draw a Venn diagram in order to represent some properties
of sets. For example, if A and B are disjoint sets, the diagram can be drawn
with A and B shown as disjoint circles. If A ⊆ B, the circle for A is entirely
inside the circle for B.

Example AA.4: Consider the argument:

All planets directly circle stars; our Moon does not directly circle
the relevant star (the Sun); therefore the Moon is not a planet.
 Sets; Proof Techniques 311
To examine this, we take the universal set to
be the set of all heavenly bodies, A to be the
set of all those bodies that directly circle stars, A m•
and B to be the set of all planets; m represents B
the Moon. We know B ⊆ A, so the sets look
like the diagram. As m ∈ / A, it must be in the
outer region, so it is certainly not in B. So the
argument is valid. 
Some arguments that look logical at first sight turn out not to be valid.
For example, consider:
All planets directly circle stars; Halley’s Comet directly circles the
relevant star (the Sun); therefore Halley’s Comet is a planet.

We label the sets as in the example. In the

diagram shown, Halley’s Comet could be rep- A
B
resented either by x or by y, so we can draw y•
no conclusion. The argument is not valid. (In x•
fact, x is correct.)

The symmetric difference of two sets consists of all elements in one or other
but not both. That is, the symmetric difference of R and S is

R + S = {x : x ∈ R or x ∈ S but x 6∈ R ∩ S}. (9)

This definition could be stated as

R+S = (R ∪ S)\(R ∩ S)
= (R ∪ S) ∩ (R ∩ S)
= (R ∪ S) ∩ (R ∪ S), (10)

using (7). We could also consider A + B to be the union of the difference

between A and B with the difference between B and A. This implies that

A + B = (A\B) ∪ (B\A),

and hence that

A + B = (A ∩ B) ∪ (A ∩ B). (11)
It is easy to see that symmetric difference satisfies the commutative and as-
sociative laws.
We define the Cartesian product (or cross product ) S × T of sets S and T
to be the set of all ordered pairs (s, t) where s ∈ S and t ∈ T :

S × T = {(s, t) : s ∈ S, t ∈ T }.

There is no requirement that S and T be disjoint; in fact, it is often useful to

consider S × S.
312 Introduction to Combinatorics

The number of elements of S × T is |S| × |T |. (This is one reason why

the symbol × was chosen for Cartesian product.) As an example, consider
S = {0, 1} and T = {1, 2}. Then

S × T = {(0, 1), (0, 2), (1, 1), (1, 2)},

the set of all four of the possible ordered pairs; 4 = 2 × 2.

The sets (R × S) × T and R × (S × T ) are not equal; one consists of an
ordered pair whose first element is itself an ordered pair, and the other of
pairs in which the second is an ordered pair. So there is no associative law,
and no natural meaning for R × S × T . On the other hand, it is sometimes
natural to talk about ordered triples of elements, so we slightly abuse notation
and write
R × S × T = {(r, s, t) : r ∈ R, s ∈ S, t ∈ T }.
This notation can be extended to ordered sets of any length.
There are several distributive laws involving the Cartesian product:

THEOREM AA.1 (Distributive Laws)

If R, S and T are any sets, then
(i) R × (S ∪ T ) = (R × S) ∪ (R × T );
(ii) R × (S ∩ T ) = (R × S) ∩ (R × T ).

Proof. (i) We prove that every element of R × (S ∪ T ) is a member of

(R × S) ∪ (R × T ), and conversely.
First observe that
R × (S ∪ T ) = {(r, s) | r ∈ R and s ∈ S ∪ T }
= {(r, s) | r ∈ R and (s ∈ S or s ∈ T )}.
On the other hand,
(R × S) ∪ (R × T ) = {(r, s) | (r, s) ∈ R × S or (r, s) ∈ R × T }
= {(r, s) | (r ∈ R and s ∈ S) or (r ∈ R and s ∈ T )}
= {(r, s) | r ∈ R and (s ∈ S or s ∈ T )}.
It is now clear that (r, s) ∈ R × (S ∪ T ) and (r, s) ∈ (R × S) ∪ (R × T ) are
equivalent.
The proof of part (ii) is left as an exercise. 
 Sets; Proof Techniques 313

AA.2 The Principle of Mathematical Induction

In working with finite sets or with sets of integers, we shall repeatedly use
a technique of proof known as the method of mathematical induction. The
general idea is as follows: suppose we want to prove that every positive integer
n has a property P (n). We first prove P (1) to be true. Then we prove that,
for any n, the truth of P (n) implies that of P (n + 1); in symbols:

P (n) true ⇒ P (n + 1) true. (12)

Intuitively, we would like to say:

P (1) true
P (1) true ⇒ P (2) true , by 12,
∴ P (2) true;
P (2) true
P (2) true ⇒ P (3) true , by 12,
∴ P (3) true;

and so on. There is a difficulty, however: given any positive integer k, we

can select an integer n such that the proof of P (n) requires at least k steps,
so the proof can be arbitrarily long. As “unbounded” proofs present logical
difficulties in mathematics—who could ever finish writing one down?—we
need an axiom or theorem that states that induction is a valid procedure.
This is the principle of mathematical induction, and may be stated as follows.

Principle of mathematical induction: Suppose the proposition P (n)

satisfies
(i) P (1) is true; and
(ii) For every positive integer n, whenever P (n) is true, then P (n + 1) is
true.
Then P (n) is true for all positive integers n.
This principle is sometimes called weak induction. Another form is:
Strong induction: Suppose the proposition P (n) satisfies
(i) P (1) is true; and
(ii) For every positive integer n, if P (k) is true whenever 1 ≤ k < n, then
P (n) is true.
Then P (n) is true for all positive integers n.
At first sight, the second statement looks as though we have assumed more
than in the first statement. However these two forms are equivalent.
314 Introduction to Combinatorics

Induction can also be stated in strictly set-theoretic form. Suppose S is the

set of integers n such that P (n) is true. Then the principle of mathematical
induction (weak induction form) is:
Let S be a subset of Z+ such that:
(i) 1 ∈ S; and
(ii) whenever n ∈ S, then n + 1 ∈ S.
Then S = Z+ .
One could equally well state the principle in terms of non-negative integers,
instead of positive integers, by changing the case P (0) to P (1), and converting
references from “positive integers” to “non-negative integers.” It can in fact
be stated in terms of any starting point: if S is a set of integers for which
(i) t ∈ S,
where t is any integer, and
(ii) for every integer n ≥ t, if n ∈ S, then n + 1 ∈ S,
then the principle can be used to prove that S contains all integers equal to
or greater than t. This form is sometimes called “induction from t.”

The Well-Ordering Principle

Another principle that is very useful in proving results about sets of positive
integers is:
The well-ordering principle: Suppose S is any non-empty set of positive
integers. Then S has a smallest member.
Any set with the property that every non-empty subset has a least member
is called being well-ordered ; so the principle says “the positive integers are
well-ordered.” Many number-sets, such as the real numbers, are not well-
ordered.
At first the result seems obvious. If S is not empty, then it must contain
some member, x say. In order to find the smallest member of S, one need
only check to see whether or not x − 1, x − 2, . . . , 1 are members of S, and
this requires only a finite number of steps. However, this “proof” contains
the same problem as the “proof” of the principle of mathematical induction:
the starting value, x, can be arbitrarily large.
In fact, the induction principle can be proved from the well-ordering princi-
ple. To see this, let us assume the well-ordering principle is true and suppose
P is any proposition about positive integers such that P (1) is true, and for
every positive integer n, whenever P (n) is true, then P (n + 1) is true. (These
are the requirements for induction.) Write S for the set of positive integers n
such that P (n) is not true. In order to prove that induction works, we need
to show that S is empty.
If S is not empty, then by well-ordering S has a smallest member, x say.
So P (x) is false. We know that P (1) is true, so x 6= 1. But x is a positive
 Sets; Proof Techniques 315

integer. So x − 1 is a positive integer, and P (x) must be true—otherwise x − 1

would be a member of S, and smaller than x. But this means P ((x − 1) + 1) is
true: that is, P (x) is both true and false! This cannot happen, so our original
assumption, that S is not empty, must have been wrong. So induction is
proved.
We can also work the other way. If you assume induction is true, you can
show that the positive integers are well-ordered.
However, there is no absolute proof here. It is necessary to assume that
either induction or well-ordering is a property of the positive integers. So we
assume these as axioms about numbers.

AA.3 Some Applications of Induction

THEOREM AA.2 (Power Set Order)

Let A be a set with |A| = n. Then |P(A)| = 2n .

Proof. To apply induction, we can rephrase the statement as: Let P (n) be
the statement “|P(A)| = 2n for any n-element set A”; then P (n) is true for
all positive integers n. Since the elements of A do not matter, we may as well
assume A = {a − 1, a2 , . . . , an }.
First we consider the case n = 1, so A = {a1 }. Then P(A) = {∅, A}, so
|P(A)| = 21 and in this case the theorem is true. Now suppose the result
has been proved for n = k − 1 and assume n = k; we might as well say
A = A′ ∪ {ak } where A′ = {a − 1, a2 , . . . , ak−1 }. By the induction hypothesis
|P(A′ )| = 2k−1 . Any subset of A is either a subset of A′ or a subset of A′
with the element sk adjoined to it, so to each subset of A′ there correspond
two subsets of A. Hence |P(A)| = 2k−1 · 2 and the theorem is proved. 

THEOREM AA.3 (Order of Cartesian Product)

If |S| = m and |T | = n, then |S × T | = mn.

Proof. We proceed by induction on m. If m = 1, then S = {s1 } and an

ordered pair with s1 as its first element may be constructed in n ways, giving
(s1 , t1 ), (s1 , t2 ), . . . , (s1 , tn ), so the theorem is true for m = 1. Now suppose
the statement is true for m = k − 1, and consider the case m = k.
Let S ′ = {s1 , s2 , . . . , sk−1 }, so |S ′ × T | = (k − 1)n. Also, |{sk } × T | = n.
But S ×T = (S ′ ×T )∪({sk }×T ) and since the two products on the right-hand
side of this equation are disjoint, we know that |SxT | = (k − 1)n + n, proving
the theorem. 
316 Introduction to Combinatorics

Mathematical induction is very often used in proving general algebraic for-

mulas.
Example AA.5: Prove by induction that the sum of the first n positive
integers is
1 + 2 + · · · + n = 12 n(n + 1).
The case n = 1 is 21 · 1(1 + 1) = 1, which is obviously true, so the formula
gives the correct answer when n = 1. Suppose it is true when n = k − 1;
therefore
1 + 2 + · · · + (k − 1) = 21 (k − 1)k.
Then
1
1 + 2 + · · · + (k − 1) + k = 2 (k − 1)k + k
1 2
= 2 (k − k + 2k)
1
= 2 k(k + 1),

and the formula is proved correct when n = k. So, by induction, we have the
required result. 
The following example looks geometrical, but it also yields to induction.
Example AA.6: Suppose n straight lines are drawn in two-dimensional
space, in such a way that no three lines have a common point and no two lines
are parallel. Prove that the lines divide the plane into 12 (n2 + n + 2) regions.
We proceed by induction. If n = 1, the formula yields 21 (1 + 1 + 2) = 2, and
one line does indeed partition the plane into two regions. Now assume that
the formula works for n = k − 1. Consider k lines drawn in the plane. Delete
one line. By induction, the plane is divided into 12 ((k − 1)2 + (k − 1) + 2) =
1 2
2 (k − k + 2) regions.
Now reinsert the (deleted) kth line. It must cross every other line exactly
once, so it crosses (k − 1) lines, and lies in k regions. It divides each of these
regions into two parts, so the k regions are replaced by 2k new regions; the
total is
1 2 1 2
2 (k − k + 2) − k + 2k = 2 (k − k + 2 − 2k + 4k)
2
= (k + k + 2),

and the result is true by induction. 

Here is an example that uses induction from the starting point 4, rather
than 0 or 1.
Example AA.7: Prove that n! ≥ 2n whenever n ≥ 4.
Suppose the proposition P (n) means n! ≥ 2n . Then P (4) means 4! ≥ 24 ,
or 24 ≥ 16, which is true. Now suppose k is an integer greater than or equal
 Sets; Proof Techniques 317

to 4, and P (k) is true: k! ≥ 2k . Multiplying by k + 1, we have (k + 1)1 ≥

(2k (k + 1) ≥ 2k 2 = 2k+1 , so P (k) implies P (k + 1), and the result follows by
induction. 

Example AA.8: Prove by induction that 5n −2n is divisible by 3 whenever

n is a positive integer.
Suppose P (n) means 3 divides 5n − 2n . Then P (1) is true because 51 − 21 =
3. Now suppose k is any positive integer, and P (k) is true: say 5k − 2k = 3x,
where x is an integer. Then 5k+1 − 2k+1 = 5 ·5k − 2 ·2k = 3 ·5k + 2 ·5k − 2 ·2k =
3 · 5k + 2 · 3x, which is divisible by 3. So the result follows by induction. 

AA.4 Binary Relations on Sets

A (binary) relation ρ from a set S to a set T is a rule that stipulates,
given any element s of S and any element t of T , whether s bears a certain
relationship to t (written s ρ t) or not (written s \ρ t). For example, if S is the
set of living males and T is the set of living females, the relation ρ might be
“is the son of”; if s denotes a certain man and t denotes a certain woman, we
write s ρ t if s is the son of t, and s \ρ t otherwise.
Alternatively, we can define a binary relation ρ from the set S to the set
T as a set ρ of ordered pairs (s, t), where s belongs to S and t belongs to T ,
with the notation that s ρ t when (s, t) ∈ ρ and s \ρ t otherwise. This means
that, formally, a binary relation from S to T can be defined as a subset of the
Cartesian product S × T .
For example, suppose S = {1, 2, 3} and T = {1, 2, 4}. The sets correspond-
ing to s < t and s2 = t are L and R, respectively, where

L = {(1, 2), (1, 4), (2, 4), (3, 4)}

and
R = {(1, 1), (2, 4)}.
If ρ and σ are binary relations from S to T , we define the union ρ ∪ σ and
the intersection ρ ∩ σ from S to T in the obvious way: (s, t) ∈ ρ ∪ σ if and only
if (s, t) ∈ ρ or (s, t) ∈ σ or both, and (s, t) ∈ ρ ∩ σ if and only if (s, t) ∈ ρ and
(s, t) ∈ σ. The notation agrees exactly with the usual set-theoretic notation.
If ρ is a binary relation from R to S, and σ a binary relation from S to T ,
then the composition ρσ is a binary relation from R to T , where r(ρσ)t if and
only if there exists at least one element s ∈ S such that rρs and sσt.
If ρ is a binary relation from R to S, and ν a binary relation from P to T ,
then the composition ρν is not defined unless P = S. But when it is defined,
composition of binary relations is associative.
318 Introduction to Combinatorics

Every binary relation has an inverse relation: if ρ is a binary relation from

S to T , so that ρ ⊆ S × T , then ρ−1 is a binary relation from T to S, or
ρ−1 ⊆ T × S, defined by

ρ−1 = {(t, s) : (s, t) ∈ ρ}.

In other words, t ρ−1 s if and only if s ρ t.

The inverse of a composition may be expressed as a composition of inverses:

(ρν)−1 = ν −1 ρ−1 .

Suppose ρ is a binary relation from a set A to A itself or, in other words,

ρ ⊆ A×A. Then we say that ρ is a binary relation on A. One obvious relation
on a set is the identity relation: ιA is the identity relation on A if a ιA b is
true precisely when b = a, that is, ιA = {(a, a) : a ∈ A}.
Suppose ρ is a relation on A.
(i) ρ is called reflexive if and only if a ρ a for every a ∈ A. ρ is irreflexive if
and only if a ρ a is never true for any element a of A.
(ii) ρ is called symmetric if and only if a ρ b implies b ρ a or, in other words,
if (a, b) ∈ ρ implies (b, a) ∈ ρ. Since, by the definition of inverse relation, we
know that a ρ b if and only if b ρ−1 a, we could say that ρ is symmetric if and
only if ρ = ρ−1 . ρ is called antisymmetric if and only if a ρ b and b ρ a together
imply a = b or, in other words, ρ ∩ ρ−1 ⊆ ιA . (Clearly, a reflexive relation will
trivially be antisymmetric.) If a ρ b and b ρ a can never both be true, then ρ is
called asymmetric, or sometimes skew . (If a ρ b and b ρ a are never both true,
then (a ρ b ∧ b ρ a) → a ρ a is vacuously true, so all asymmetric relations are
antisymmetric, but this gives us no information.)
(iii) ρ is called transitive if and only if a ρ b and b ρ c together imply a ρ c,
and atransitive if and only if, whenever a ρ b and bρc, a \ρ c, or in other words
if and only if ρ ∩ ρρ = ∅. Clearly, ρ is transitive if and only if ρρ = ρ.
As an example, the relation of adjacency on the set of integers, defined by
i ρ j if and only if |i − j| = 1 for i and j positive integers, is irreflexive (since
|i − i| = 0), symmetric (since |i − j| = |j − i|) and not transitive (since 1 ρ 2
and 2 ρ 3 but 1 \ ρ 3). It is in fact atransitive, for i ρ j and j ρ k together imply
i = j ± 1 and k = j ± 1, so i = k or i = k ± 2, but i cannot equal c ± 1. So
a\ρ c.
A relation α on a set A is called an equivalence relation on A if and only if
it is reflexive, symmetric, and transitive.
The obvious equivalence relation is equality, on any set. In sets other than
number sets, equal objects are often called “equal in all respects.” More
generally, an equivalence relation can be considered as a statement that two
objects are “equal in some (specified) respects.” One example, on the integers,
is the relation α, where a α b is true if and only if a = ±b—a and b have the
same absolute value. Another is congruence, on the set of all plane triangles.
Equivalence relations give us an alternative way to discuss multiple ele-
ments. You can view a multiset as an object that is based on a set of distinct
 Sets; Proof Techniques 319

elements on which an equivalence relation has been defined, so that the ele-
ments of a multiset are the equivalence classes of the underlying set.
The word ‘congruence is also used as follows: the relation “congruence
modulo n” for a positive integer n is defined on the set Z of integers by
specifying that a is congruent to b modulo n, written “a ≡ b(mod n),” if and
only if n | (a − b). In other words, two integers are congruent modulo n if and
only if they belong to the same residue class modulo n. It is easy to show
that congruence modulo n is an equivalence relation: since n | (a − a), the
relation is reflexive; if n | (a − b) then n | (b − a), so the relation is symmetric;
if n | (a − b) and n | (b − c) then we have a − b = kn, and b − c = ln, for
some k, l ∈ Z, hence a − c = (k + l)n and n | (a − c), which shows that the
relation is transitive. So congruence modulo n is an equivalence relation on
the integers.

Appendix 1 Exercises A
1. Suppose A = {a, b, c, d, e}, B = {a, c, e, g, i}, C = {c, f, i, e, o}. Write
down the elements of

(i) A ∪ B. (ii) A ∩ C.
(iii) A\B. (iv) A ∪ (B\C).

2. Consider the sets

S1 = {2, 5},
S2 = {1, 2, 4},
S3 = {1, 2, 4, 5, 10},
S4 = {x ∈ Z+ : x is a divisor of 20},
S5 = {x ∈ Z+ : x is a power of 2 and a divisor of 20}.

For which i and j, if any, is Si ⊆ Sj ? For which i and j, if any, is

Si = Sj ?

3. If S ∪ T = U and S ∩ T = ∅, prove that T = S.

4. Consider the data: All mathematicians are loners; all surgeons are rich;
no loners are rich. Which of the following conclusions can be drawn?
(i) No mathematicians are rich.
(ii) All surgeons are loners.
(iii) No one can be both a mathematician and a surgeon.
320 Introduction to Combinatorics

5. Show that
R+S =R+S
for any sets R and S.
6. Show that the following three statements are equivalent: S ⊆ T, S ∪T =
T, S ∩ T = S.
7. Prove the given proposition by induction.
(i) The sum of the first n odd positive integers is n2 .
Xn
(ii) r2 = 16 n(n + 1)(2n + 1).
r=1

8. Prove that the sum of the cubes of any three consecutive integers is a
multiple of 9.
9. The numbers x1 , x2 , . . . are defined as follows. x1 = 1, x2 = 1, and if
n ≥ 2 then xn+1 = xn + 2xn−1 . Prove that xn is divisible by 3 if and
only if n is divisible by 3.
10. Suppose ρ and σ are relations defined on S = {1, 2, 3, 4, 5} by the sets
ρ = {(2, 2), (3, 2), (3, 4), (4, 1), (4, 3), (5, 5)};
σ = {(2, 3), (1, 2), (3, 1), (3, 2), (4, 4), (4, 5), (5, 4), (5, 5)}.
(i) Find the set form of the composition ρσ.
(ii) What is the set form of the relation σρ?
(iii) What are the set forms of the relations ρρ and σσ?
11. Suppose A is the set Z × Z. Prove that the relation ρ on A, defined by

(x, y) ρ (z, t) if and only if x + t = y + z,

is an equivalence relation.

Appendix 1 Exercises B
1. Suppose

A = {2, 3, 5, 6, 8, 9} B = {1, 2, 3, 4, 5} C = {5, 6, 7, 8, 9}.

Write down the elements of

 Sets; Proof Techniques 321

(i) A ∩ B. (ii) A ∪ C.
(iii) A\(B ∩ C). (iv) (A ∪ B)\C.

2. Suppose

A = {a, b, c, d, e, f } B = {a, b, g, h} C = {a, c, e, f, h}.

Write down the elements of

(i) A ∩ B. (ii) A ∪ C.
(iii) A\(B ∩ C). (iv) (A ∪ B)\C.

3. Are the given statements true or false?

(i) 2 ∈ {2, 3, 4, 5}. (ii) 3 ∈

/ {2, 3, 4, 5}. (iii) 3 ∈ {2, 4, 5, 6}.
(iv) {1, 2} = {2, 1}. (v) {1, 3} ∈ {1, 2, 3}. (vi) {1, 3} = {1, 2, 3}.

4. In each case, are the sets S and T disjoint? If not, what is their inter-
section?
(i) S is the set of perfect squares 1, 4, 9, . . . ; T is the set of cubes
1, 8, 27, . . . of positive integers.
(ii) S is the set of perfect squares 1, 4, 9, . . . ; T is the set Π of primes.
5. R, S and T are any sets, and U is the universal set.
(i) Prove: if R ⊆ S and R ⊆ T , then R ⊆ (S ∩ T ).
(ii) Prove: if R ⊆ S, then R ⊆ (S ∪ T ).

6. Find a simpler expression for S ∪ ((R ∪ S) ∩ R).

7. For any sets R and S, prove R ∩ (R ∪ S) = R.
8. In each case, represent the set in a Venn diagram.

(i) (R\S)\T (ii) R\(S\T )

(iii) (R ∩ S)\(S ∩ T ) (iv) (R ∪ T )\(S ∪ T )

9. Consider the data: the city zoo keeps all species of brown bears; all large
bears are dangerous; the city zoo does not keep any dangerous bears.
Which of the following conclusions can be drawn?
(i) No large bears are brown.
(ii) All species of large bears are kept in the city zoo.
(iii) Some large bears are kept in the city zoo.
(iv) No large bears are kept in the city zoo.
(v) No dangerous bears are kept in the city zoo.
322 Introduction to Combinatorics

10. Draw Venn diagrams for use in the following circumstances:

(i) All members of this class are math majors;
(ii) No members of this class are math majors.

11. Prove the two distributive laws

A ∩ (B + C) = (A ∩ B) + (A ∩ C);
(A + B) ∩ C = (A ∩ C) + (B ∩ C).

12. Show that if S ∪ T = ∅, then S = T = ∅ and that if S ∩ T = U , then

S = T = U.

13. Prove that R\(S\T ) contains all members of R ∩ T , and hence prove
that
(R\S)\T = R\(S\T )
is not a general law (in other words, relative difference is not associa-
tive).
(i) First, prove this without using Venn diagrams.
(ii) Second, prove the result using Venn diagrams.

14. Prove that union is not distributive over relative difference: in other
words, prove that the following statement is not always true:

(R\S) ∪ T = (R ∪ T )\(S ∪ T ).

(i) First, prove this without using Venn diagrams. (Hint: Use the fact
(R\S) ∪ S = R ∪ S.)
(ii) Now prove the result using Venn diagrams.

15. Prove the associative law for symmetric difference.

16. Prove that A + B + C consists of precisely those elements that belong

to an odd number of the sets A, B and C.

17. Prove the given proposition by induction.

n
" n #2
X X
3
(i) k = k = 41 n2 (n + 1)2 .
k=1 k=1
(ii) 1 + 4 + 7 + · · · + (3n − 2) = 12 n(3n − 1).

18. The numbers a0 , a1 , a2 , . . . are defined by a0 = 3 and

an+1 = 2an − a2n when n ≥ 0.

n
Prove that an = 1 − 22 when n > 0, although this formula does not
apply when n = 0.
 Sets; Proof Techniques 323

19. Prove the given proposition by induction.

n
X
(i) 2 + 6 + 12 + · · · + n(n + 1) = k(k + 1) = 13 n(n + 1)(n + 2).
k=1
1 1 1 1 n
(ii) + + + ···+ = .
1·3 3·5 5·7 (2n − 1)(2n + 1) 2n + 1
(iii) 1 + 3 + 32 + · · · + 3n = 12 (3n+1 − 1).
(iv) n2 ≥ 2n + 1 whenever n ≥ 3.
1
20. The numbers a0 , a1 , a2 , . . . are defined by a0 = 4 and

an+1 = 2an (1 − an ) when n > 0.

Prove that  
1 1
an = 1 − 2n .
2 2

21. Prove that the following divisibility results hold for all positive integers
n.
(i) 2 divides 3n − 1.
(ii) 6 divides n3 − n.
(iii) 5 divides 22n−1 + 32n−1 .
(iv) 24 divides n4 − 6n3 + 23n2 − 18n.
22. In each case a binary relation from S = {1, 2, 3, 4, 5, 6, 7, 8, 9} to itself
is defined. For each of these relations, what is the corresponding subset
of S × S? For each relation, what is the inverse relation?
(i) x ρ y means y = x2 .
(ii) x β y means xy = 6.
(iii) x δ y means 3x < 2y.
23. The relation ρ is defined on {1, 2, 3, 4, 5} by

ρ = {(1, 1), (2, 2), (2, 3), (3, 3), (3, 4), (4, 3), (5, 5)}.

Prove that ρ has none of the properties: reflexive, transitive, symmetric,

antisymmetric.
24. Define the relations ρ, σ and τ on the set {1, 2, 3, 4} by the sets
ρ = {(1, 1), (2, 2), (3, 3), (4, 4)};
σ = {(1, 1), (1, 3), (2, 2), (3, 3), (3, 1), (4, 4)};
τ = {(2, 3), (1, 2), (1, 3), (3, 2)}.
Which of these relations are
324 Introduction to Combinatorics

(i) reflexive; (iv) antisymmetric;

(ii) irreflexive; (v) asymmetric;
(iii) symmetric; (vi) transitive?
Are any of them equivalence relations?
25. A group of people, denoted by P, Q, . . ., are at a lecture. Consider the
following relations: Is each reflexive, symmetric, antisymmetric, asym-
metric or transitive? Is it an equivalence relation?
(i) P µ Q means P and Q come from the same city.
(ii) P ν Q means P is the father of Q.
(iii) P ρ Q means P and Q spoke to each other at the lecture.
(iv) P σ Q means P and Q met for the first time at the lecture.
(v) P τ Q means P spoke to Q at the lecture.
26. Verify: If ρ is a binary relation from R to S, and ν a binary relation
from P to T , then the composition ρν is not defined unless ρ(R) ⊆ P .
But when it is defined, composition of binary relations is associative.
27. Let α and β be equivalence relations on a set A.
(i) Show that αβ is an equivalence relation if and only if αβ = βα.
(ii) Show that α ∩ β is an equivalence relation.
28. Prove that the inverse of a composition may be expressed as a compo-
sition of inverses:
(ρν)−1 = ν −1 ρ−1 .

29. Suppose p is a prime. The relation ρ on Z is defined by a ρ b if and only

if a2 ≡ b2 (mod p). Show that ρ is an equivalence relation.
Appendix B: Matrices and Vectors

B.1 Definitions
A matrix is a rectangular array whose elements—the entries of the matrix—
are members of an algebraic structure with the operations of addition and
multiplication, where multiplication is commutative. Most commonly the en-
tries will be chosen from the real numbers, but one could restrict them to
integers, or to the two numbers 0 and 1, or allow members of some integral
domain or a more general structure. (We refer to a real matrix, a zero-one ma-
trix , and so on.) The possible entries are called the scalars for the particular
class of matrices.
The horizontal layers in a matrix are called rows and the vertical ones are
columns. A matrix with m rows and n columns is called an m × n matrix; we
refer to m × n as the shape or size of the matrix, and the two numbers m and
n are its dimensions.
Matrices are usually denoted by single uppercase letters. As an example,
consider the matrix
 
1 0 1 −2
M =  −1 1 −4 1 .
1 −3 0 2

The first row of M is

1 0 1 −2

and the second column is

0
1 .
1

We refer to the entry in the ith row and jth column of a matrix as the
(i, j)-element (or entry). The (2, 3) element of M is −4. When no confusion
is possible, we write aij to denote the (i, j)-element of a matrix A, using the
lowercase letter corresponding to the (uppercase) name of the matrix, with
the row and column numbers as subscripts. A common shorthand is to write
[aij ] for A. We use the ordinary equality sign between matrices; A = B means
that A and B are the same shape and aij = bij for every i and j.

325
326 Introduction to Combinatorics

In general if M and N are two matrices with the same shape, the sum of
two matrices M = [mij ] and N = [tij ] is the matrix S = [sij ] defined by

sij = mij + tij .

There is no matrix M + N if M and N are of different shapes. For example,

if  
1 −3 1 0  
1 −1 1 1
N=  −2 1 −1 2  , R= ,
−1 2 −2 1
1 3 2 −3
then
   
1 0 1 −2 1 −3 1 0
M +N =  −1 1 −4 1  +  −2 1 −1 2 
1 −3 0 2 1 3 2 −3
 
2 −3 2 −2
=  −3 2 −5 3 
2 0 2 −1
but M + R is not defined.
It is easy to see that this addition satisfies the commutative and associative
laws: if M , N and P are any matrices of the same shape, then

M + N = N M and (M + N ) + P = M + (N + P ).

Because of the associative law, we usually omit the brackets and just write
M + N + P.
We can take the sum of a matrix with itself, and we write 2M for M + M ,
3M for M + M + M , and so on. This can be extended to multipliers other
than positive integers: if a is any scalar, aM will mean the matrix derived
from M by multiplying every entry by a. That is,

aM is the matrix with (i, j) entry amij .

This matrix aM is the scalar product (or simply product ) of a with M . It has
the same shape as M . The scalar product obeys the laws

(a + b)M = aM + bM, a(bM ) = (ab)M, a(M + N ) = aM + aN

for any matrices M and N of the same shape and any scalars a and b. Notice
that if 1s is the multiplicative identity element of the scalars (for example, for
real matrices, 1s represents 1), then 1s M = M .
Omn denotes a matrix of shape m × n with every entry zero. Usually we
do not bother to write the subscripts m and n, but simply assume that the
matrix is the correct shape for our computations. O is called a zero matrix ,
and works for matrix addition like the number zero under ordinary addition:
if M is any matrix, then
M +O =M
 Matrices and Vectors 327

provided O has the same shape as M .

Obviously 0M = O for any matrix M (where 0 is the scalar zero and O is
the zero matrix). From the first law for scalar multiplication, above,

M + (−1)M = 1M + (−1)M = (1 + (−1))M = 0M = O,

so (−1)M acts like a negation of M . We shall simply write −M instead of

(−1)M , and call −M the negative of M . We can then define subtraction of
matrices by
M − N = M + (−N ),
just as you would expect. Again, subtraction only works if the matrices are the
same shape. The usual conventions are followed: −M + N means (−M ) + N ,
and to show the negative of M + N you have to write −(M + N ). To evaluate
M −N +P you first subtract N from M , then add P ; because of the associative
law, no special brackets are needed.
If A is an m × n matrix, then we can form an n × m matrix whose (i, j)
entry equals the (j, i) entry of A. This new matrix is called the transpose of
A, and written AT . A matrix A is called symmetric if A = AT .
A matrix with one of its dimensions equal to 1 is called a vector. An m × 1
matrix is a column vector of length m, while a 1 × n matrix is a row vector
of length n. The individual rows and columns of a matrix are vectors, which
we call the row vectors and column vectors of the matrix.
We shall write vectors with boldface lowercase letters, to distinguish them
from matrices and scalars. (We treat vectors separately from matrices be-
cause, in many cases, it is not necessary to distinguish between row and
column vectors.) In some books a vector is denoted by a lowercase letter with
an arrow over it, ~v , rather than a boldface letter v. The ith entry of a vector
is denoted by a subscript i, so the vector v has entries v1 , v2 , . . .; we usually
write v = (v1 , v2 , . . .).
The two standard operations on vectors follow directly from the matrix
operations. One may multiply by a number , and one may add vectors. If k is
any number, and v = (v1 , v2 , . . . , vn ), then kv = (kv1 , kv2 , . . . , kvn ). If u =
(u1 , u2 , . . . , un ), and v = (v1 , v2 , . . . , vn ), then u+v = ((u1 +v1 ), (u2 +v2 ), . . . ,
(un + vn )). If u and v are vectors of different lengths, then u + v is not
defined. We again write −v for (−1)v, so −v = (−v1 , −v2 , . . . , −vn ), and
u − v = u + (−v). We define a zero vector 0 = (0, 0, . . . , 0) (in fact, a family
of zero vectors, one for each possible dimension), and v + (−v) = 0.
Sometimes there is no important difference between the vector v of length
n, the 1×n matrix (row vector) whose entries are the entries of v, and the n×1
matrix (column vector) whose entries are the entries of v. But in the next
section we shall sometimes need to know whether a vector has been written
as a row or a column. If this is important, we shall write row(v) for the row
vector form of v, and col(v) for the column vector form. If we simply write v,
you usually can tell from the context whether row(v) or col(v) is intended.
328 Introduction to Combinatorics

B.2 Vector and Matrix Products

We define the scalar product (also called the dot product ) of two vectors
u = (u1 , u2 , . . . , un ) and v = (v1 , v2 , . . . , vn ) to be
n
X
u · v = (u1 v1 + u2 v2 + · · · + un vn ) = uk vk .
k=1

Example B.1: Suppose t = (1, 2, 4), u = (−1, 3, 0) and v = (1, −2, 1).
Calculate t · u, (t − u) · v and (u · t).

t·u = 1.(−1) + 2.3 + 3.0 = (−1) + 6 + 0 = 5,

(t − u) · v = (2, −1, 4) · (1, −2, 1) = 2 + 2 + 4 = 8,
u·t = (−1).1 + 3.2 + 0.3 = (−1) + 6 + 0 = 5. 

Clearly the dot product is commutative. There is no need to discuss the

associative law, because dot products involving three vectors are not defined.
For example, consider t · (u · v). Since (u · v) is a scalar, not a vector, we
cannot calculate its dot product with anything.
The product of two matrices is a generalization of the scalar product of
vectors. Suppose the rows of the matrix A are a1 , a2 , . . ., and the columns of
the matrix B are b1 , b2 , . . .. Then AB is the matrix with (i, j) entry ai · bj .
These entries will only exist if the number of columns of A equals the number
of rows of A, so this is a necessary condition for the product AB to exist.
Example B.2: Suppose
   
1 2 −1 1 3
A= ,B = .
1 −1 2 −2 0
What are AB and BA?
   
−1 + 4 1−4 3+0 3 −3 3
AB = = .
−1 − 2 1+2 3+0 −3 3 3
However, BA does not exist. 
As another example consider the matrix M of the preceding section. Sup-
pose Q is any m × n matrix. Since M is 3 × 4, the product M Q will be a
3 × n matrix if m = 4, but it is not defined when n 6= 4. In general, we have

THEOREM B.1 (Conformability)

Suppose A is an m × n matrix and B is an r × s matrix. If n = r, then AB
exists and is an m × s matrix. If n 6= r, then AB does not exist.
 Matrices and Vectors 329

If the dimensions are such that the product AB exists, the matrices A and
B are called conformable.
Matrix multiplication satisfies the associative law: provided the relevant
products all exist,
A(BC) = (AB)C
and the product is usually written ABC. However, the commutative law is a
different story. If A is 2 × 3 and B is 3 × 4, then AB is a 2 × 4 matrix but
BA does not exist. It is also possible that AB and BA might both exist but
might be of different shapes; for example, if A and B have shapes 2 × 3 and
3 × 2, respectively, then AB is 2 × 2 and BA is 3 × 3. And even when AB and
BA both exist and are the same shape, they need not be equal. For example,
     
1 2 1 1 5 1
= ;
1 3 2 0 7 1
     
1 1 1 2 2 5
= .
2 0 1 3 2 4

If AB = BA we say the matrices A and B commute.

If AA is to exist, then A must have the same number of rows as columns;
such matrices are called square. The common dimension is called the order
of the matrix.
If A is square, we can evaluate the product AA. We call this A squared,
and write it as A2 , just as with powers of numbers. We define other positive
integer powers similarly: A3 = AAA = AA2 , and in general An+1 = AAn . As
with numbers, indices apply only to the symbol nearest: AB n means A(B n ),
not (AB)n . (Similarly, AB T means A(B T ), not (AB)T .)
The zero matrix behaves under multiplication the way you would expect:
provided zero matrices of appropriate size are used,

OA = O and AO = O.

This is not just one rule, but an infinite set of rules. If we write in the
subscripts, then the full statement is
If A is any r × s matrix, then Om,r A = Om,s for any positive integer m, and
AOs,n = Or,n for any positive integer n.
There are also matrices that act like the number 1: multiplicative identity
elements. We define In to be the n × n matrix with its (1, 1), (2, 2), . . . , (n, n)
entries 1 and all other entries 0. For example,
 
1 0 0
I3 =  0 1 0 .
0 0 1

If A is any r × s matrix, then Ir A = A = AIs . We call In an identity matrix

of order n.
330 Introduction to Combinatorics

Another way of combining matrices is to form the Cartesian product of two

matrices. If A is a p × q matrix and B is an r × s matrix, their cartesian
product A × B (or A ⊗ B) is formed as follows. In A, replace entry aij by the
scalar product aij B. That is, each cell is transformed into an r × s block of
cells. Therefore A × B is a pr × qs matrix.

B.3 Inverses
If the matrices A and B satisfy AB = BA = I, we say that B is an inverse
of A.
Which matrices have inverses? In the real numbers, everything but 0 has
an inverse. In the integers, only 1 and −1 have (integer) inverses, but we can
obtain inverses of other non-zero integers by going to the rational numbers.
The situation is more complicated for matrices.
Suppose A is an r × c matrix. Then AB has r rows and BA has c columns.
If the two are to be equal, then AB is an r × c matrix; and if this is to equal
an identity matrix, it must be square. So r = c, and A is also a square matrix.
So only a square matrix can have an inverse.
Moreover, there are non-zero square matrices without inverses. One obvious
example is a matrix with one row or one column all zeroes. But there are less
trivial examples, even if we restrict our attention to the 2 × 2 case.
Example B.3: Show that the matrix
 
2 1
A=
1 1
has an inverse, by construction.
If  
x y
B=
z t
is an inverse, then AB = I, so
      
2 1 x y 2x + z 2y + t 1 0
= = .
1 1 z t x+z 2y + t 0 1
Comparing the (1, 1) entries of the two matrices, so 2x + z = 1; from the (2, 1)
entries, x + z = 0. So x = 1 and z = −1. Similarly, y = −1, t = 2. So
 
1 −1
B= .
−1 2
On the other hand,  
3 1
C=
3 1
 Matrices and Vectors 331

has no inverse: if  
x y
D=
z t
were an inverse, then
      
2 1 x y 3x + z 3y + t 1 0
= = .
2 1 z t 3x + z 3y + t 0 1

Comparing (1, 1) entries again, 3x+z = 1, but the (2, 1) entries give 3x+z = 0.
This is impossible.
A matrix that has an inverse will be called invertible or non-singular ; a
square matrix without an inverse is called singular .
We said “an inverse” above; however, if a matrix has an inverse, it is unique.
We say “the inverse” (if one exists) and write A−1 .

THEOREM B.2 (Uniqueness of Inverse)

If matrices A, B, C satisfy AB = BA = I and AC = CA = I, then B = C.

Proof. Suppose A, B and C satisfy the given equations. Then

C = CI = C(AB) = (CA)B = IB = B

so B and C are equal. 

In fact, either of the conditions AB = I or BA = I implies the other, and
therefore implies that B is the inverse of A.
As we said, the usual notation for the inverse of A (if it exists) is A−1 . If
we define A0 = I whenever A is square, then powers of matrices satisfy the
usual index laws
Am An = Am+n , (Am )n = Amn
for all non-negative integers m and n, and for negative values also provided
that A−1 exists. If x and y are non-zero reals, then (xy)−1 = x−1 y −1 . The
fact that matrices do not necessarily commute means that we have to be a
little more careful, and prove the following theorem.

THEOREM B.3 (Invertibility of Product)

If A and B are invertible matrices of the same order, then AB is invertible,
and

(AB)−1 = B −1 A−1 .

Proof. We need to show that both (B −1 A−1 )(AB) and (AB)(B −1 A−1 ) equal
the identity. But (B −1 A−1 )(AB) = B −1 (A−1 A)B = B −1 IB = B −1 B = I =
AA−1 = AIA−1 = A(BB −1 )A−1 = (AB)(B −1 A−1 ). 
332 Introduction to Combinatorics

There are two cancellation laws for matrix multiplication. If A is an in-

vertible r × r matrix and B and C are r × s matrices such that AB = AC,
then

AB = AC ⇒ A−1 (AB) = A−1 (AC)

⇒ (A−1 A)B = (A−1 A)C ⇒ IB = IC ⇒ B = C,

so B = C. Similarly, if A is an invertible s × s matrix and B and C are r × s

matrices such that BA = CA, then B = C.
The requirement that A be invertible is necessary. We can find matrices
A, B and C such that AB and AC are the same size, A is non-zero and
AB = AC, but B and C are different. One very easy example is
     
1 0 1 1 1 0 1 1
= .
0 0 2 1 0 0 1 2

Some other examples are given in the exercises.

Moreover we can only cancel on one side of an equation; we cannot mix the
two sides. Even if A is invertible it is possible that AB = CA but B 6= C (see
the exercises for examples of this, also).

B.4 Determinants

THEOREM B.4 (Invertibility of 2 by 2 Matrices)

The matrix  
a b
A=
c d
is singular if ad − bc = 0. Otherwise it is invertible, with inverse
 
1 d −b
B= .
(ad − bc) −c a

Proof. If ad− bc 6= 0, it is easy to see that AB = I. So we assume ad− bc = 0

and prove that A has no inverse.
If a = 0, then either b or c must be zero, and A will have a zero row or
column (and consequently no inverse). So we need only consider cases where
a 6= 0, and we can write d = a−1 bc. Similar considerations show that we can
also assume c is non-zero. Now suppose
    
a b x y 1 0
= .
c a−1 bc z t 0 1
 Matrices and Vectors 333

Comparing (1, 1) entries, ax + bz = 1, so x = a−1 − a−1 bz. On the other

hand, the (2, 1) entries give cx + a−1bcz = 0, or x = −a−1 bz. So a−1 = 0, an
impossibility. 
The quantity ad − bc is called the determinant of the matrix A, written
det(A). Determinants are defined recursively for square matrices of any order:
if A is the n × n matrix [aij ], where n > 2, write Aij for the matrix derived
from A by deleting row i and column j, and define
n
X
det(A) = (−1)i+j det(Aij ).
j=1

This is called the expansion along the ith row ; it may be shown that the value
does not depend on the choice of i, and there is a similar formula for expansion
along columns. For further discussion of determinants, see standard texts on
matrix theory and linear algebra, such as [3]; among other things, a square
matrix is non-singular if and only if its determinant is non-zero.

Appendix B Exercises A
1. Carry out the following matrix computations.
 
1 1 −1
(i) 3
−2 0 8
   
2 −1 2 −1
(ii) +
−1 0 −1 2
   
2 −1 2 3
(iii) 3  4 −2  + 2  2 2 
−2 1 −1 0
   
x − 2 3 2z y 2z 3
2. Find x, y and z so that = .
y x y 6z y + 2 6z
3. Carry out the following vector computations.
(i) 3(2, 4, 1) (ii) 4(3, 1, −1)
(iii) (1, 3) + (2, 4) (iv) 2(2, −1, 2, 3) − 3(1, 3, −2, −3)

4. A is a 2 × 3 matrix; B is 2 × 3; C is 1 × 4; D is 3 × 2; E is 3 × 4; F
is 4 × 3; G is 3 × 3. Say whether the indicated matrix exists. If it does
exist, what is its shape?
334 Introduction to Combinatorics

(i) A + B (ii) CE T (iii) D(A + B)

(iv) F T (v) 2F C (vi) AD + DA

(vii) F 2 (viii) DA + 3G (ix) G2

5. Carry out the following matrix computations.

    
  3 −1 3 −2 3 −2
1 2 1  (ii)
(i) 2 −2  3 0 −4 4
2 −1 1
−1 1
   
1 −1   2 −1  
2 2 1 1 0
(iii)  1 −1  (iv)  1 0 
1 −1 1 −1 −1
−1 1 2 3
  
2 −1 4 1
(v)  0 1 2   −2 
1 1 0 1

6. Find a matrix A such that

   
12 −1 1
A = .
23 41

7. The matrix A is given. Find A2 and A3 .

   
1 −1 1 2 1
(i)
−1 0 (ii)  0 −3 1 
0 0 −2

8. Find the determinant of the matrix, and use it to invert the matrix or
show that it is singular.
   
56 21
(i) (ii)
23 42

9. In each case show that AB = AC. What is the (common) value?

     
21 01 14
(i) A = , B= , C=
21 37 11
     
111 212 421
(ii) A =  1 0 1  , B =  2 1 0  , C =  2 1 0 
121 222 013
 Matrices and Vectors 335

Appendix B Exercises B
1. Carry out the following matrix computations.
   
8 −4 1 1
(i) 2  1 3  − 5  1 −2 
−2 1 −2 0
   
3 −1 2 1
(ii) 3  2 2  − 2  4 −2 
−2 3 −1 0
   
4 −1 1 −1
(iii) 3 −3
2 2 −1 3
   
2 −1 −1 1 −1 2
(iv) 2 +3
2 3 1 1 1 −2
   
x −1 y + 2 −1
2. Suppose = .
−1 3 −1 x
What are the values of x and y?

3. Carry out the following vector computations.

(i) −(1, −1) (ii) 3(1, −2)

(iii) 2(2, 1) − 2(1, −1) (iv) 2(1, 2, 1) − 2(1, −1, −1)

(v) 3(−2, 2, 3) + 3(2, 1, −1) (vi) 2(1, −2, 2) + 3(2, 3, −1)

(vii) (1, 0, 3) + 2(2, 1, 3) (viii) 3(1, 6, 1, 0) − 3(2, 0, 4, −1)

4. Suppose
     
1 −1 0 −1 3
A= , B= , C= ,
−1 1 2 2 2
     
1 3 1 −2 3
D= , E= , F = .
−1 0 1 0 −1

Calculate the following, or say why they do not exist.

(i) 3A + B (ii) C + D − E (iii) A − D − E
(iv) 2C − 4F (v) C − 3F (vi) 3A − 3A
336 Introduction to Combinatorics

5. A is a 2 × 5 matrix; B is 2 × 5; C is 1 × 3; D is 5 × 2; E is 3 × 5; F
is 5 × 3; G is 5 × 5. Say whether the indicated matrix exists. If it does
exist, what is its shape?
(i) 2A − B (ii) AD (iii) CF
(iv) CF T (v) DA (vi) AGF
(vii) A2 (viii) D2 (ix) G2
6. In this exercise,
     
1 −1 3 0 −1 6
A= , B= , C= ,
−2 3 −1 4 4 2
     
−1 1 −1 1 −1 2
D= 1 3 3 , E =  1 0 , F =  −1  ,
−2 2 0 2 2 2
     
G= 2 1 −1 , H= 2 2 , K= −1 2 .

Carry out the matrix computations, or explain why they are impossible:
(i) BF (ii) AC (iii) CF
(iv) BG (v) D2 (vi) EK + KB

7. Suppose the square matrices A and B satisfy AB = I. Prove that

BA = I.
8. In each case, find the products AB and BA. Do the two matrices
commute?
   
1 −1 2 1
(i) A = , B= .
1 1 −1 2
   
1 2 5 3
(ii) A = , B= .
3 5 2 1
   
3 1 −2 1 0 1
(iii) A =  1 1 −1  , B =  −1 2 0 .
−1 −1 2 0 1 2
   
0 1 1 3
(iv) A = , B= .
2 1 −1 1
   
2 1 −2 4
(v) A = , B= .
1 −1 3 −2
 Matrices and Vectors 337
 
1 3
9. Consider the matrix A= .
5 3
(i) Find A2 and A3 .
(ii) Evaluate A3 − 2A − I.
(iii) Show that A2 − 4A − 12I = O.

10. The matrix A is given. Find A2 and A3 .

   
1 2 1 −1
(i) (ii)
−1 1 −1 1
   
−1 2 1 2 0
(iii)
−1 −1 (iv)  0 1 3 
1 0 −1
   
2 0 3 3 0 −1
(v)  0 −2 −1  (vi)  1 1 3 
−1 1 2 −3 0 −1

11. Suppose A is a square matrix with one row all zeroes. Prove that A
has no inverse. Prove the corresponding result when one column is all
zeroes.
12. Find a matrix A such that
   
3 2 1 3
A = .
2 1 2 −1

13. Find the determinant of the matrix, and use it to invert the matrix or
show that it is singular.
   
−2 2 3 6
(i) (ii)
−2 1 2 5
   
3 2 4 6
(iii) (iv)
−1 1 5 7
   
3 2 4 3
(v) (vi)
6 4 1 1

14. A and B are any two matrices such that AB exists. Prove that B T AT
exists, and that
B T AT = (AB)T .
338 Introduction to Combinatorics

15. For the following matrices, show that A−1 exists, but AB = CA, even
though B 6= C.
     
2 −1 4 0 2 0
A= , B= , C= .
1 1 4 2 2 4
16. Suppose Mx denotes the 2 × 2 matrix
 
1 x
,
0 1

where x may be any real number.

(i) Compute Mx My , and show that the matrices Mx and My commute
for any real numbers x and y.
(ii) Find Mx 2 , Mx 3 and Mx 4 .
(iii) Find a formula for Mx n , where n is any positive integer.
(iv) What is Mx −1 ?
Appendix C: Some Combinatorial
People

This appendix contains brief biographical sketches of some of the people who
have contributed to the development of combinatorics. If you are interested
in further information, we have cited relevant articles, in Wikipedia and else-
where. Several good books on the history of mathematics have concentrated
on biographical sketches; in particular, we recommend [6] and [16]. And nowa-
days, if you want further information about any mathematician, simply type
their name into your favorite web browser.

Leonhard Euler [133]

Leonhard Paul Euler (April 15, 1707–September 18, 1783) was a Swiss
mathematician who spent much of his life in Berlin and St. Petersburg and
produced more mathematical papers than any other mathematician in history.
He developed the modern notion of a function and did much pioneering work
in infinite series, among many other disciplines. Among many other formulas,
he is known for “Euler’s identity” (eiπ + 1 = 0), and
√ introduced the notation
e for the base of the common logarithm and i for −1. He also became very
well known for a popularization of mathematics that consisted of a collection
of letters written at the request of Frederic of Prussia to the Princess of
Anhalt-Dessau, generally known as the Letters to a German Princess.
His contributions to combinatorics include the invention of graph theory
with his solution of the Königsberg Bridges problem in 1736. The term Latin
square,as well as Graeco-Latin square are due to his notation in writing pairs
of orthogonal Latin squares. His related problem of the 36 military officers led
to significant research in the subject, although the idea itself predated Euler,
having first appeared for the 4 × 4 case as a problem of playing cards. He also
developed the Gamma function, which is a continuous analog of the factorial
function, and developed a formula relating the numbers of edges, faces, and
corners of polyhedra that is a mainstay of topological graph theory.
One story of Euler is fairly well-known; while living in Russia, be became
unhappy with the political situation there, and took the opportunity to move
to the Academy in Berlin. Early in his stay there, it was pointed out to him
that he seemed very quiet and untalkative, and was asked why this might be.
According to the story, he replied, “Madam, I come from a country where,
if we speak, we are hanged.” After twenty-five years in Berlin, he returned
to St. Petersburg in 1766. This was possible in part because of a change in

339
340 Introduction to Combinatorics

Russia’s leadership (as Catherine the Great took power), and Euler felt that
he was no longer welcome in Berlin.
Another fairly famous anecdote of Euler comes from his time in St. Pe-
tersburg. The French philosopher Denis Diderot was visiting the court of
Catherine the Great, and allegedly making a stir with his outspoken atheism.
The story goes that Euler was asked to embarrass Diderot, and a joke was ar-
ranged. Diderot was told that the famous mathematician had a mathematical
proof of the existence of God. Euler, according to this story, gave his “proof”
with the statement that “Sir, (a + bn )/z = x, therefore God exists!” Suppos-
edly, Diderot was so ignorant of mathematics that he did not realize that the
statement was nonsense. Since Diderot was an accomplished mathematician
himself, this story is widely assumed to be completely untrue; however, Bur-
ton ([16]) suggests that the incident may have occurred, in which case Diderot
would have realized at once that it was merely a practical joke.
He lost all sight in his right eye in 1738 (likely due to an infection), and lost
all sight in the other eye in 1766 due to a cataract. Remarkably, his blindness
did not impair his mathematical productivity in any way, and indeed some
of his most productive years occurred after his blindness. In part this was
due to an ability to perform mental calculations, and in part it was due to
an amazing memory; he was able to recite Virgil’s Aeneid from memory, for
example. His complete works are being published; there are 76 volumes, with
a few more to appear, consisting of his personal letters.

Évariste Galois [129]

Évariste Galois (October 25, 1811–May 31, 1832) was a French mathemati-
cian who was the first to find necessary and sufficient conditions to solve a
polynomial in radicals. He developed a considerable part of what today we
call group theory, including the original use of the word group as an algebraic
object. He also pioneered the study of finite fields.
He was distracted from mathematics by his political interests, as he was
expelled from his university and arrested twice for various activities. For
reasons that will probably never be fully understood, a month after being
released from jail the last time, Galois became involved in a duel to the death
over a woman. He died of his wounds, but (having expected to die) spent
considerable time before the duel writing up results that he had polished
while imprisoned. These are preserved in a letter to a colleague who had
helped him publish his work previously.

Howard Garns [130]

Howard Garns (March 2, 1905–October 6, 1989) was an American architect
who gained fame only after his death as the creator of Number Place, the
number puzzle that became a worldwide phenomenon under the name Sudoku.
 Some Combinatorial People 341

Garns developed the idea while working at the Daggett architecture firm in
Indianapolis around 1960, but it was not published until 1979, in Dell Pencil
Puzzles and Word Games.
Garns was born in Connersville, Indiana, and by his teens had moved to
Indianapolis with his father, W. H. Garns, an architect. He attended the
University of Illinois, and received a Bachelor of Science in architectural engi-
neering in 1926. He worked for his father’s firm until the Second World War,
when he became a captain in the US Army Corps of Engineers. He joined the
Daggett architecture firm after the war.

William Rowan Hamilton [138]

William Rowan Hamilton (August 4, 1805–September 2, 1865) was an Irish
physicist and mathematician whose work included the development of quater-
nions (which may be thought of as a generalization of the complex number
system) as well as the solvability of a polynomial of degree 5 by radicals.
In his youth he displayed a remarkable gift for languages, having learned a
large number of them including Hebrew, Arabic, Sanskrit, Malay, and many
European languages. His schooling at Trinity College in Dublin was cut short
prior to graduation when Hamilton was offered the position of Professor of
Astronomy at the University of Dublin. Although he was not primarily an as-
tronomer, the expectation was that he would continue to contribute to science
as a whole, which he certainly did as he contributed significantly to optics and
mechanics.
His name appears in combinatorics only in relation to the Hamilton circuit,
based on his study of cycles on a dodecahedron. This was inspired by and
connected to his work on non-commutative algebra, of which quaternions were
the first example.

Richard Hamming [134]

Richard Wesley Hamming (February 11, 1915–January 7, 1998) was an
American mathematician. He was a professor at the University of Louisville
during World War II, and joined the Manhattan Project in 1945, where he
programmed one of the earliest electronic digital computers. The program
was designed to discover if the detonation of an atomic bomb would ignite the
atmosphere. The United States did not use the bomb until the computations
showed that atmospheric disaster would not occur. (It was then tested in New
Mexico and subsequently used twice against Japan.)
Hamming worked at the Bell Telephone Laboratories, where he collaborated
with Claude E. Shannon, from 1946 to 1976. He then moved to the Naval
Postgraduate School, where he worked as an Adjunct Professor until 1997,
then became Professor Emeritus.
342 Introduction to Combinatorics

He was a founder and president of the Association for Computing Machin-

ery.

Thomas Kirkman [10, 61, 81, 84]

Thomas Penyngton Kirkman (March 31, 1806–February 4, 1895) was an
English clergyman and mathematician.
He was born in Bolton, near Manchester, England, and attended school
there until age fourteen, when he was forced to leave school to work in his
father’s office. He returned to University nine years later, attending Trinity
College, Dublin. He entered the Church of England and became a curate in
Bury and in Lymm. In 1839 he became vicar in the Parish of Southworth in
Lancashire, and stayed in that post until 1891.
Kirkman’s first mathematical paper [57], published when he was age 40,
answered a problem posed by Woolhouse in the Lady’s and Gentleman’s Diary
in 1844 [140]. He proved the existence of balanced incomplete block designs
with k = 3 and λ = 1. Steiner did not know of Kirkman’s work, and in 1853
he published an article [110] that asked whether such systems existed; the
solution was published by Reiss [86] in 1859. Kirkman subsequently wrote
“. . . how did the Cambridge and Dublin Mathematical Journal Vol II p. 191,
contrive to steal so much from a later paper in Crelle’s Journal Vol LVI p.
326 on exactly the same problem in combinatorics?” [81] Despite this, these
designs came to be known as “Steiner triple systems.” In 1918, Cummings
[25] said that Kirkman’s paper “appears to have been overlooked by all writers
on this subject,” and even after Cummings’ paper was published, Kirkman’s
contribution was ignored. (For example, in [37], the author says that Steiner
“initially posed the problem” in [110], says that Reiss solved it, and neither
Woolhouse nor Kirkman is mentioned.)
Kirkman is best known for the schoolgirl problem that we discussed in
Section 14.1. He later published in many other areas of mathematics; he
published extensively on the enumeration of polyhedra, then moved into group
theory—finding a recursive method for compiling lists of transitive groups and
giving a complete list of transitive groups of degree ≤ 10—and finally studied
the theory of knots. Overall, his contributions to mathematics—not just to
combinatorics—are very significant.

Joseph Kruskal [132]

Joseph Bernard Kruskal, Jr. (January 29, 1928–September 19, 2010) was
an American mathematician and statistician whose work has influenced the
disciplines of computer science, combinatorics, graph theory, and linguistics.
He was born in New York City, the youngest of five brothers (two others
of whom also went on to become noted mathematicians). He attended the
University of Chicago, and earned his Ph.D. from Princeton University in
 Some Combinatorial People 343

1954. In addition to the minimal spanning tree algorithm from Chapter 11,
his name is attached to the Kruskal tree theorem and the Kruskal–Katona
theorem.

Édouard Lucas [128]

François Édouard Anatole Lucas (April 4, 1842–October 3, 1891) was a
French mathematician who produced a formula for finding the nth term of
the Fibonacci sequence, and did a great deal of other work on sequences and
related problems. He devised methods for testing the primality of numbers.
When he was fifteen years old, Lucas began testing the primality of 2127 − 1
by hand, using Lucas Sequences. Nineteen years later he finally proved that
it was prime; it was then the largest known Mersenne prime.
Lucas was also interested in recreational mathematics, among other things
inventing the Tower of Hanoi puzzle which we discussed in Chapter 1.
Lucas worked in the Paris observatory, served in the military and finally
became a professor of mathematics in Paris.

Samuel Morse [7]

Samuel Finley Breese Morse (April 27, 1791–April 2, 1872) was an Ameri-
can painter of historic scenes.
In 1825, the city of New York commissioned Morse to paint a portrait of
Lafayette, in Washington. While he was there, a horse messenger delivered a
letter from his father that read one line, ”Your wife is dead.” Morse imme-
diately left Washington for his home at New Haven, leaving the portrait of
Lafayette unfinished. But when he arrived home she had already been buried.
In his distress at the fact that for days he was unaware of his wife’s failing
health and her lonely death, he moved on from painting to pursue a means of
rapid long distance communication.
In 1832, Morse met Charles Thomas Jackson of Boston, an expert on elec-
tromagnetism. Morse used these ideas to develop the concept of a single-wire
telegraph. The original Morse telegraph is in the National Museum of Ameri-
can History at the Smithsonian Institution. He then developed the preliminary
version of the code that bears his name, for use with his device. He originally
planned to send only numerals. A version very like the present Morse code
was developed by Alfred Vail in the early 1840s.
Morse was a leader in the anti-Catholic and anti-immigration movements of
the mid-nineteenh century. In 1836, he ran unsuccessfully for mayor of New
York. He worked to unite Protestants against Catholic institutions (including
schools), wanted to forbid Catholics from holding public office, and promoted
changing immigration laws to limit immigration from Catholic countries.

Isaac Newton [87, 135]

344 Introduction to Combinatorics

Isaac Newton (January 4, 1642–March 31, 1727) is possibly the best-known

mathematician, physicist, philosopher, alchemist, and astronomer in history.
Attending Cambridge University beginning in 1661, he developed many of
the methods of calculus and applied them to the broad range of scientific
disciplines in which he made his mark. In combinatorics, he is known almost
solely for the generalized binomial theorem, which is the source of many gen-
erating functions as described in Chapter 6. This result, a generalization of
the well-known Pascal’s triangle method of raising a binomial to an integer
power, gives an infinite series for an expression of the form (a + b)r for any
real exponent r.
The bulk of his creative mathematical work was done during his years at
Cambridge, although much of it remained unpublished until years later. In
particular, Newton himself cites the years of the bubonic plague (1665 through
1667) as the time in which most of his mathematical work was accomplished.
During this time, Cambridge was closed and he returned to the family farm
at Woolsthorpe. He returned to Cambridge in 1667 and became Lucasian
Professor of Mathematics in 1669. His work was not entirely on mathematics;
at this time, he developed his theories of optics, and invented the reflecting
telescope, as well as continuing to work on his mathematical model of gravity.
He also continued to study alchemy and wrote religious tracts.
He took up a position with the Royal Mint in 1696, which required him to
relocate to London, and in 1701 gave up his position at Cambridge in order
to concentrate on his work there. He was, by some accounts, quite active,
personally aiding in locating and shutting down counterfeiters. This aspect
of his life is the subject of Philip Kerr’s mystery novel Dark Matter [56].
Newton was made president of the Royal Society in 1703, and knighted by
Queen Anne in 1705.

Blaise Pascal [16, 126]

Blaise Pascal (June 19, 1623–August 19, 1662) was a French mathematician
with a strong theological bent. A child prodigy, when he was 11 he wrote a
work on the sounds produced by vibrating objects. His father forbade him
from further study of science and mathematics so as not to interfere with
Pascal’s study of Latin and Greek, a ban that was lifted when Pascal (at 12)
was caught proving a theorem in geometry (that the angles in a triangle add
to 180◦) independently.
At 16, he produced but never published an essay on conic sections that was
received with much praise, and shortly thereafter invented a mechanical device
for addition and subtraction capable of adding a column of eight figures. This
device was at first intended merely to assist his father Etienne with his work
dealing with tax records, but Pascal produced many others, some few of which
remain today. His hopes of making a profit from selling these devices were
never realized, however.
 Some Combinatorial People 345

In 1654 an acquaintance sent Pascal a letter asking certain technical ques-

tions concerning probabilities involved in games of chance. The acquaintance,
the Chevalier de Méré, was friends with many eminent mathematicians (prob-
ably because he made his living by gambling) and addressed a question to Pas-
cal concerning the proper way to divide the stakes between two players if a
game is interrupted. Pascal found questions of this sort easy, and his research
led him to re-discover the triangle of binomial coefficients that today bears his
name. He published Traité du Triangle Arithmétique around 1654, although
it was not available until much later. In this work, he develops many of the
relations and patterns involving binomial coefficients that we know and use
today. This work also contains the first published statement of the principle
of mathematical

n−r ninduction, which Pascal used to prove a formula equivalent
n


to r+1 = r+1 r .

Emanuel Sperner []
Emanuel Sperner (December 9, 1905 - January 31, 1980) was a German
mathematician. He was born in Waltdorf, in what is today Poland. He
obtained his doctorate in 1928 from the University of Hamburg for a disser-
tation that contained one of the two results for which he is known today; this
is Sperner’s Lemma, a result concerning the colorings of a triangulation of a
simplex. While geometrical in nature, this result is used to prove powerful
results in topology as well. In 1928, he also published what is now called
Sperner’s Theorem, a major result in the combinatorial discipline of extremal
set theory. His later work served to make him a major figure in modern
geometry as well.
He later worked in Beijing for a time before returning to Germany, where
he took a position at the University of Königsberg until the second World
War. He then worked for the German Navy for a time before resuming his
academic career at Oberwolfach, where he is regarded as one of the founding
members of the Oberwolfach Mathematical Institute. Other appointments
followed at Bonn and then at Hamburg, where he retired in 1974. He took
a number of visiting positions thereafter to institutions in the United States,
South Africa, and Brazil, and spent his final years in Sulzburg-Laufen, where
he died in 1980.

Gaston Tarry [80]

Gaston Tarry (September 27, 1843–June 21, 1913) was a French amateur
mathematician. He was born in Villefranche-de-Rouergue, in south central
France. He joined the French Financial Administration and became part of
the French administration in Algeria, until his retirement in 1902. He was
never employed as a mathematician, but did a great deal of mathematics,
nearly all combinatorial in nature.
346 Introduction to Combinatorics

In 1886, Tarry published a general method for finding the number of Euler
circuits in a graph or multigraph [112]. Then, in 1895, he published a sys-
tematic method so solve a maze [113], using what is essentially a depth-first
search algorithm. In 1900 and 1901 [114] he solved Euler’s 36 Officer Problem,
showing that there is no pair of orthogonal Latin squares of order 6.
Tarry then went on to work on magic squares, producing a substantial
number of results.

John Venn [93]

John Venn (August 4, 1834–April 4, 1923) was an English logician and
mathematician. In 1857 he was named a Fellow of Caius College of Cam-
bridge University; he was ordained a priest 1859 and for a year was curate at
Mortlake. He then began lecturing at Caius College. He introduced what are
now called Venn diagrams in two papers in 1880 [118, 119]. He was elected a
Fellow of the Royal Society in 1883.
The use of diagrams in formal logic certainly originated well before his time.
However, Venn comprehensively surveyed and formalized their usage, and was
the first to generalize them.
Solutions to Set A Exercises

Chapter 1
1. The sets are shown in grey.
(i) R S (ii) R S

T T

2. Select three “active” players. They proceed as if they were the only
players, in a team of three. All other players pass.
3. Let’s write Sn for the required sum, S = 1 + 3 + . . . + (2n − 3) + (2n − 1).
(i) Partition the first 2n integers into the sets of odd and even integers:
1 + 2 + 3 + . . . + (2n − 1) + 2n
= [1 + 3 + . . . + (2n − 3) + (2n − 1)] + [2 + 4 + . . . + (2n − 2) + 2n] ;
so Sn = [1 + 2 + 3 + . . . + (2n − 1) + 2n]−[2 + 4 + . . . + (2n − 2) + 2n].
Now Gauss’s formula tells us that
1 + 2 + 3 + . . . + (2n − 1) + 2n = 2n(2n + 1)/2 = n(2n + 1),
and obviously
2 + 4 + . . . + (2n − 2) + 2n = 2 [1 + 2 + . . . + (n − 1) + n]
= 2 [n(n + 1)/2] = n(n + 1).
So Sn = n(2n + 1) − n(n + 1) = n2 .
(ii) Clearly S1 = 1 = 12 . Assume Sn = n2 for n = 1, 2, . . . , r.
Then Sr+1 = Sr + (2r + 1) = r2 + (2r + 1) = (r + 1)2 .
4. ABC, ADE, AFG, BDF, BEG, CDG, CEF.
Case n = 1 is easy. Assume ri=1 i(i + 1) = 31 r(r + 1)(r + 2). Then
P
5.
r+1
X
1
i(i + 1) = 3 r(r + 1)(r + 2) + (r + 1)(r + 2)
i=1
r  1
= 3 + 1 (r + 1)(r + 2) = 3 (r + 1)(r + 2)(r + 3).

347
348 Introduction to Combinatorics

6. We see that τ is reflexive, since x + 2x is divisible by 3 for integer

x. Similarly, if x, y, and z are integers with x τ y and y τ z, we write
x + 2y = 3m, y + 2z = 3n and add these equations to get x + 3y + 2z =
3(m + n). Thus x + 2z = 3(m + n − y), so the relation is transitive.
Finally, suppose x + 2y = 3n, and consider 3x + 3y = (x + 2y) + (y + 2x)
to see that τ is symmetric.
7. 4 + 3 + 2 = 9, by the addition principle.
8. 4 × 3 × 6 = 72, by the multiplication principle.
9. 3 × 3 + 6 × 4 + 10 × 2 = 53 ways.
10. 8 × 7 × 6 = 336 permutations of three items from a set of 8.
11. 262 × 104 = 6, 760, 000.
12. Write Y for the set of all subsets with n or fewer elements and Z for those
with more. Clearly Y and Z are disjoint, and equal in size (Z consists
of all the complements in X of elements of Y ). Y and Z have 22n+1
elements between them (see Example 1.2), so each has 22n elements.
13. (i) 2, 1, 0, 3, 4, 3, 2, 5, 2 (ii) 2,2,2,2,2,2,2 (iii) 2, 4, 3, 2, 5, 2
14. There are n edges; one vertex of degree n, and n of degree 1.
15. Minimum 11 (the complete graph K10 on 10 vertices has only 45 edges,
so the minimum is at least 11; K11 has 55 edges, and if we delete any
seven of them the resulting graph will have no isolated vertex, so 11 can
be realized.) Maximum 96.
16. Consider a (multi)graph whose vertices are the people, with an edge
for each handshake. (i) If no one shook hands five or more times, the
sum of the degrees would be at most 32, so there could be at most 16
handshakes. (ii) No, because if it were true the graph or multigraph
would have an odd number of odd vertices.
17. (i) Yes, for example (1,4,2,3). (ii) No, because the four sets 123, 12, 13, 23
have only three elements between them.

Chapter 2
1. Since there are 52 letters possible for the last three, the answer is 26 ×
26 × 102 × 523 , or 9, 505, 100, 800.
 Solutions to Set A Exercises 349

2. Solution: Eight choices for first digit, two for second digit and 10 for
last digit gives us 160 area codes possible. If we remove the “x11”
possibilities, there are eight that we disallow; 211, 311, 411, 511, 611,
711, 811, and 911. Subtracting leaves us 152 area codes.
3. Solution: There are 24 choices for the first window, 23 for the second,
and so on. The result is 24!/19! = 5, 100, 480.
4. This is equivalent to finding (i) the total number of 4-digit numbers, (ii)
the number of 4-digit numbers with no repeated digits. The answer to
(i) is 9000 (all the numbers from 1000 to 9999). For (ii), there are 9
choices (any of 1 to 9) for the first digit (x say), 9 for the second (0 is
now available but x is not), 8 for the third and 7 for the fourth, so the
answer is 9 × 9 × 8 × 7 = 4536.
5. Consider the strings of 6 digits, from 000000 to 999999. (000000 repre-
sents 1,000,000; otherwise abcdef represents abc, def with leading zeroes
deleted.) There are six places to put 3; for each choice there are five
places to put 5; for each of these there are four places to put 7; there are
seven choices for each other number. The total is 6 × 5 × 4 × 73 = 41160.
6. Since the sites are mutually exclusive (the student will join only one
site), the addition principle applies; we add the numbers for each site.
For a given site, befriending one acquaintance does not affect whether or
not the student befriends another, so that the multiplication principle
applies. Thus, the first site provides the student with 23 or 8 possibil-
ities; the second, 25 = 32; the third, 64; and the fourth, 16. We add
these to find 120 possibilities.
7. There are C(4, 2) = 6 ways to choose the two sites. If the numbers of
acquaintances were different, we would have to add up the possibilities
for each of the six choices. Because the numbers are the same, each
chosen site gives us 24 or 16 possible ways to befriend or not befriend
the acquaintances on that site. We therefore have 6·16·16 = 1536 possible
outcomes.
8. This must be broken into cases, unfortunately. There is one way to
choose the two sites with four acquaintances; each site admits 16 possi-
bilities, so we have 256 possibilities for this case. There are four ways to
choose one site with four acquaintances and one with five; there are 16
possibilities for the former and 32 for the latter, for 2048 possibilities.
Then there is one way to choose both of the five-acquaintance sites, and
each of those sites admits 32 possibilities, giving 1024 possible outcomes
in this case. Since the three cases are mutually exclusive, we add the
results to obtain 3328 possibilities altogether.
9. If she neglects the underage friend, she has three friends to choose from
and seven restaurants, for 21 possibilities. If she chooses the underage
350 Introduction to Combinatorics

friend, she has one friend to choose from and four restaurants. Since
she is taking only one friend, the two are mutually exclusive, so she
has 25 possibilities. If she can afford to take two friends, then she has
C(3, 2) = 3 ways if she leaves the underage friend behind, and seven
restaurants. If she takes the underage friend, she has C(3, 1) = 3 ways
to choose the second friend, and four restaurants. We get 21 + 12 = 33
possibilities.
Solution: There are 63 ways to select three tutors from a set of 6; this


10.
gives us 20 ways to assign three tutors to the 2:00 PM shift.
11. 6·5 = 30, 7·6 = 42, 8·7 = 56.
12. 1, 7, 21, 35, 35, 21, 7, 1.
13. 1·x10 + 5·x7 + 10·x4 + 10·x + 5·x−2 + 1·x−5 .
14. We can rewrite Corollary 2.3.3 as
Pn
1 + k=1 (−1)k nk = 0.

Pn
So k=1 (−1)k kr = −1. Divide each term by −1.

15. C(6, 3) ways to get to the west corner of block b2 times C(5, 2) ways to
go from the east end of b2 to the northeast corner gives us 200 ways.
Since there are C(12, 5) = 792 total possible paths, we subtract to get
592 paths that do not pass through block b2.
16. There are 16 letters, of which 3 are S, 2 are L, 5 are E, 2 are P, and there
are one each of I, N, G, and T. This gives us 16!/(3!·2!·5!·2!·1!·1!·1!·1!)
or 7, 264, 857, 600.
17. We can write 3 as a sum in 10 ways: 1+1+1, 1+2+0, 1+0+2, 0+1+2,
2 + 1 + 0, 2 + 0 + 1, 0 + 2 + 1, 3 + 0 + 0, 0 + 3 + 0, and 0 + 0 + 3. Computing
3!/(1!1!1!) = 6, 3!/(2!1!0!) = 3, and 3!/(3!0!0!) = 1, we obtain

6·3x·5y ·7z
+ 3 (3x)2 ·5y + (3x)2 ·7z +
(5y)2 ·3x + (5y)2 ·7z + (7z)2 ·3x + (7z)2 ·5y

+ 1 (3x)3 + (5y)3 + (7z)3 .

18. We visualize 10 cards and 2 dividers to separate the allotment of the

teenagers; this is 12 symbols of which we wish to choose 2 to be dividers,
or C(12, 2) = 66. If each teenager is to receive at least one, we distribute
one card to each and have seven cards and two dividers; this gives us
C(9, 2) = 36 ways to distribute the cards.
19. Apply Theorem 2.8; there are n = 3 covers to be purchased, and k =
8 distinct items (colors) from which to choose them. This gives us
C(3 + 8 − 1, 8 − 1) = 120 ways.
 Solutions to Set A Exercises 351

20. 365142; 351642; 641532

Chapter 3
1. 0.26.

2. (i) 1/4 (ii) 3/4.

3. 1/6.

4. (i) (2, 0), (2, 1), (2, 2), (1, 0), (1, 1), (1, 2), (0, 0), (0, 1), (0, 2).
(ii) A = (2, 0), (2, 1), (1, 0), B = (2, 0), (1, 1), (0, 2),
C = (2, 0), (2, 2), (1, 1), (0, 0), (0, 2).
(iii) The probabilities of the different results are P (2, 0) = P (2, 2) =
P (0, 0) = P (0, 2) = 1/16, P (2, 1) = P (1, 0) = P (1, 2) = P (0, 1) = 1/8
and P (1, 1) = 1/4. So P (A) = 5/16, P (B) = 3/8, P (C) = 1/2.
H TH
H H
5. (i) (i) T T TT (ii) {H, T H, T T } (iii) 1/4.

6. 41 × 0.51 × 0.53 = 4 × 1/24 = 1/4.

7. Call the dice die 1 and die 2. There are six possible rolls on each die, or
36 possibilities, and 6 are doubles. So the probability is 6/36 = 1/6.

8. 1/10.

9. 2, 3, 4, 5, 6, 7, 8, 9, 10; P (2) = P (10) = 1/24; P (3) = P (8) = 1/12; P (3) =

P (10) = 1/8; P (5) = P (6) = P (4) = 1/6.

10. (i) Yes (ii) No; add to 1.4, not 1 (iii) Yes (iv) No, contains a negative
(v) No, contains a “probability” greater than 1 (and adds to more than
1).

11. (i) 1/2 (ii) 1/13 (iii) 1/26 (iv) 7/13.

F PF PPF
F F F
12. (i) Solid dots are outcomes: P P P PPP

(ii) 4 (iii) {F, P F }.

352 Introduction to Combinatorics

Chapter 4
1. 7. (This is a pretty trivial exercise, but notice: if we associate pigeon-
holes with the six possible rolls, then seven rolls must result in at least
one pigeonhole with two or more occupants.)
2. Write A1 = a1 , A − 2 = a1 + a2 , . . . , An = a1 + a2 + . . . + an . If n
divides ai , put s = 1, t = i; we are done. If n never divides ai for any i,
write Sk for the set of all ai congruent to k modulo n. There are n sums
and they belong to n − 1 sets (since S0 is empty). So one set contains
at least two elements.
3. By pigeonhole principle, k = 5; for if each car carried only four or fewer
people, there would be at most 84 patrons in the club. We cannot
conclude k = 6 since if all the cars carried at most five people, we might
have had as many as 105 patrons.
4. Let Si be the number of returns completed when i days have passed.
Clearly S1 ≥ 1 and S50 ≤ 80. Also, the numbers {S1 + 19, . . . S50 + 19}
are between 20 and 99. So the set of 100 integers {Si : 1 ≤ i ≤ 50} ∪
{Si + 19 : 1 ≤ i ≤ 50} is a set of 100 integers between 1 and 99, and so
at least two of them must be equal. It follows that there are h, k such
that Sh = Sk + 19, as required.
5. Let Si be the total number of journal entries turned in by the end of the
ith class day. Clearly S1 ≥ 1 and S60 ≤ 90. Each of the 120 numbers
S1 , . . . S60 , S1 + 29, . . . S60 + 29 lies between 1 and 90 + 29 = 119, so there
must be some h and k with Sh = Sk + 29, as required.
P
6. Suppose t∈T t = k. Then k ≡ ℓ(mod n) for some ℓ, 0 ≤ ℓ ≤ n − 1. If
ℓ 6= 0, choose x = ℓ; if ℓ = 0, choose x = n.
7. “Ends in 01” is equivalent to “equals 1 mod 100.” Write p to represent
p mod 100. There are only finitely many different integers mod 100
(100 of them, to be precise!), so there must be repetitions among the
powers of p. Say pi = pj is such a repetition, where i > j. Since p is not
divisible by 2 or 5, p has an inverse mod 100. Then pi−j = pi p−j ≡ 1
mod 100, and i − j is positive.
8. Given n, define Si = {y : ni ≤ y < i+1 n }. The sets S0 , S1 , . . . , Sn−1
partition the real numbers from 0 to 1 (including 0 but not 1), so the
remainder on dividing kx by n lies in one of them, for any k. Write
xk for this remainder. If one of x1 , x2 , . . . lies in S0 , we are finished.
If not, consider x1 , x2 , . . . , xn . There are n numbers, and each lies in
one of the sets S1 , S2 , . . . , Sn−1 . So two of them—say xa and xb —lie in
the same set, Si say. Say b > a. Then xb − xa is the difference of two
 Solutions to Set A Exercises 353

positive numbers, each greater than or equal to ni and less than i+1n . So
|xb − xa | < i+1
n − i
n < 1
n ; but |xb−a | = |xb − xa |, and we are finished.

9. The two are equal (simply exchange the labels A1 and A2 in the defini-
tion, in Theorem 4.12).

10. Suppose the edges of a complete graph are colored in red and blue so
that there is no red triangle or blue K5 . Select a vertex x. If there
are five red edges at that vertex, no two of the other endpoints of those
edges can be joined by a red edge, or else we would have a red triangle.
But the edges joining those five other endpoints cannot all be blue, or
we would have a blue K5 . So there are at most four red edges at x.
Now suppose there were nine blue edges at x. Consider the K9 formed
by the other nine endpoints and the edges joining them. By Example
3.9, this must contain either a red triangle or a blue K4 . The former
is not allowed; the latter together with x would form a blue K5 in the
original graph.
So there are at most four red and eight blue edges touching x. Therefore
the complete graph has at most 13 vertices (4 + 8 + 1—don’t forget to
count x).

11. R and B form a symmetric sum-free partition.

12. Clearly the sets partition (Z4 × Z4 )∗ . To check symmetry, write −R

for the list of negatives of elements of R: R = {02, 10, 30, 11, 33}, so
(retaining the order) −R = {02, 30, 10, 33, 11} = R, and similarly for B
and G, so we have symmetry. To check whether R is sum-free, observe
that the addition table
+ 02 10 30 11 33
02 00 12 32 13 31
10 12 20 00 21 03
30 32 00 20 01 23
11 13 21 01 22 00
33 31 03 23 00 22
contains no element of R among the sums; similarly for B and G.

13. Theorem 3.8 says R(4, 5) ≤ 35, so the answer is “No.”

Chapter 5
1. 199 − 66 − 18 + 6 = 121.
354 Introduction to Combinatorics

2. 5000 − 1250 − 1000 − 833 + 250 + 416 + 166 − 83 = 2666.

3. (i) 35 + 35 − 15 = 55; (ii) 35 + 35 − 2 × 15 = 40.
4. (i) 0; (ii) 31 + 36 − 47 = 20; (iii) 31.
5. (i) 189; (ii) 367; (iii) 1308.
6. (i) 3; (ii) 50.
7. 11! − 9! − 9! − 7! + 7! + 5! + 5! − 3! = 39, 191, 274.
8. 91 − 28 − 21 − 21 + 0 + 0 + 0 − 0 = 21.
9. 455 − 84 − 35 − 220 + 0 + 20 + 4 − 0 = 140.
10. 165 − 56 − 10 − 35 − 35 + 0 + 4 + 4 + 0 + 0 + 1 − 0 − 0 − 1 − 0 + 0 = 37.
11. There are 77 squares less than 6, 000 and 18 cubes less than 6, 000; there
are four sixth powers less than 6, 000. We get 6, 000−77−18+4 = 5, 909.
12. 4 × 20 − 6 × 9 + 4 × 2 − 0 = 34.
13. 4 × 24 − 5 × 12 + 7 + 1 = 44.
14. D6 = 6 × 44 + 1 = 265; D7 = 7 × 265 − 1 = 1, 854; D8 = 8 × 1, 854 + 1 =
14, 833.
(i) D6 = 265. (ii) 61 × D5 = 264. (iii) 6! − 265 = 455. (iv) 0 because


15.
when five women get their own coats back the last woman has no other
coat but her own to get back.
Choose the integers to be fixed in 73 ways; then choose the derange-


16.
ments of the others in D4 ways. We get 35 × 9 = 315.

Chapter 6
1. (i) (x3 + 4x2 + x)/(1 − x)4 . (ii) (4x2 + 3x − 1)/(1 − x)3. (iii) x/(1 − x)4 .
(iv) 1/(1 − 3x) − 3/(2 − 4x).
2. (i) We expand (1 + x + x2 + x3 )3 and the coefficient of xn is the result.
This is 1 + 3x + 6x2 + 10x3 + 12x4 + 12x5 + 10x6 + 6x7 + 3x8 + x9 .
(ii) Similarly, (x + x2 + x3 )3 = x3 + 3x4 + 6x5 + 7x6 + 6x7 + 3x8 + x9 .
(iii) For at least two banana-walnut muffins, we expand (x2 + x3 )(1 +
x + x2 + x3 )2 , to get: x2 + 3x3 + 5x4 + 7x5 + 7x6 + 5x7 + 3x8 + x9 .
 Solutions to Set A Exercises 355

(iv) For an unlimited supply, (1 + x + x2 + . . . )3 = (1 − x)−3 . We may

use Theorem 2.7 to see that we can do this in C(n + 3 − 1, 3 − 1) = (n2 +
3n + 2)/2 ways. To find the recurrence, expand the denominator; this is
1 − 3x + 3x2 − x3 . This gives the recurrence an = 3an−1 − 3an−2 + an−3 .

3. (i) an = 4n . (ii) an = (n − 2)2n . (iii) an = (−3)n . (iv) an = an = 2n .

(v) an = 4n − (−3)n . (vi) an = n2n−1 .

4. 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144.

5. We get (1 + x3 + x6 + · · · + x3k + . . . )(1 + x + x2 + · · · + xk + . . . ) = [(1 −

x3 )(1 − x)]−1 . Our recurrence relation is thus an = an−1 + an−3 − an−4 .
A closed form is ⌈(n + 1)/3)⌉.

We get n≥0 (2n − n)xn = 1/(1 − 2x) + x/(1 − x)2 . This simplifies to
P
6.
fa (x) = (1 − x − x2 )/(1 − 4x + 5x2 − 2x3 ). This gives us the recurrence
an = 4an−1 − 5an−2 + 2an−3 .

7. (i) fa (x) = x2 ex /2. (ii) fb (x) = e3x − e2x /2. (iii) fc (x) = 2ex . (iv)
fd (x) = e2x + 2ex .

8. This is the sequence {1, 0, 1, 0, . . .} or an = (1 + (−1)n )/2.

9. (i) fa (x) = e4x . (ii) fb (x) = (ex − 1)4 . (iii) fc (x) = (x + x2 /2! + x3/3!)4 .

Chapter 7
1. A proper way of lining up customers ensures that we never have more
people offering us a twenty dollar bill than we have had people who pay
with ten dollar bills. This corresponds exactly to a sequence of left and
right parentheses where the number of right parentheses never exceeds
the number of left parentheses that have occurred. This is just Cn .

2. There are six ways to place three chords at one corner; six ways to draw
chords in the shape of the (possibly reversed) letter N; and two ways to
put chords in the form of an equilateral triangle.

3. Each dot is touched by exactly one arc; write a “(” below a dot whose
arc connects it to a dot to its right, and “)” below a dot whose arc
connects it to a dot on its left. Then a set of arcs corresponds exactly
to a proper sequence of parentheses.

4. C5 = C0 C4 + C1 C3 + C22 + C3 C1 + C4 C0 = 14 + 5 + 4 + 5 + 14 = 42.
356 Introduction to Combinatorics

C9 = 18 2


5. 9 /10. This is 18!/[10(9!) ] = (18 × 17 × 16 × 15 × 14 × 13 ×
12 × 11)/(2 × 3 × 4 × 5 × 6 × 7 × 8 × 9). Careful cancellation makes this
problem not too difficult; (17 × 16 × 14 × 13 × 12 × 11)/(8 × 7 × 6 × 4) =
(17 × 13 × 2 × 11) = 4, 862.
−1 
6.
 
1 0 0 0 1 0 0 0
 =  −1 1 0 0 .
1 1 0 0  

1 3 1 0  2 −3 1 0 
1 7 6 1 −6 11 −6 1
7 7 7
7. 2 = 63; 3 = 301; 4 = 350.
4
8. 2 = 7.
6 6
9. 3 = 90; 3 × 3! = 540.

6

 3 6

 4 6

 5 6

 6
10. 3 3 + 2 3 + 1 3 + 0 3 = 350; 350 × 3! = 2100.
5
11. 3 × 3! = 150.
5
12. 2 = 15.

8

 6
13. 4 4 × 4! = 109, 200.

14. B3 = 5; there are five ways, namely 1×154 = 2×77 = 7×22 = 11×14 =
2 × 7 × 11.

15. B8 = 4140.
7 7 7
16. 4 = −735, 5 = 175, and 6 = −21.
 6 
17.   = 274; if one table is distinguished, we have 2 × 274 = 548.
2

Chapter 8

1. e ρ ρ2 ρ3 ρ4
e e ρ ρ2 ρ3 ρ4
ρ ρ ρ2 ρ3 ρ4 e
ρ2 ρ2 ρ3 ρ4 e ρ
ρ3 ρ3 ρ4 e ρ ρ2
ρ4 ρ4 e ρ ρ2 ρ3
 Solutions to Set A Exercises 357
 
1 2 3 4 5
2. (i)
5 3 4 2 1
 
1 2 3 4 5
(ii)
2 4 3 5 1
3. (i) (13)(254); (ii) (12534).
4. For simplicity, we number the corners 1, 2, 3, 4 starting at the upper
left and moving clockwise. We may form a reflection across the horizon-
tal axis, corresponding to the permutation (14)(23); a reflection across
the vertical axis, corresponding to (12)(34), and a rotation by 180◦ ,
corresponding to (13)(24). Together with the identity e, this forms a
group.
Denote C2 = {e, α} and C3 = e, ρ, ρ2 . The group elements are

5.

(e, e), (e, ρ), (e, ρ2 ), (α, e), (α, ρ), (α, ρ2 )


and its multiplication table follows.

(e, e) (e, ρ) (e, ρ2 ) (α, e) (α, ρ) (α, ρ2 )
(e, e) (e, e) (e, ρ) (e, ρ2 ) (α, e) (α, ρ) (α, ρ2 )
(e, ρ) (e, ρ) (e, ρ2 ) (e, e) (α, ρ) (α, ρ2 ) (α, e)
(e, ρ2 ) (e, ρ2 ) (e, e) (e, ρ) (α, ρ2 ) (α, e) (α, ρ)
(α, e) (α, e) (α, ρ) (α, ρ2 ) (e, e) (e, ρ) (e, ρ2 )
(α, ρ) (α, ρ) (α, ρ2 ) (α, e) (e, ρ) (e, ρ2 ) (e, e)
(α, ρ2 ) (α, ρ2 ) (α, e) (α, ρ) (e, ρ2 ) (e, e) (e, ρ)
6. The cycle index is C(z1 , z2 , z3 ) = (z13 + 3z1 z2 + 2z3 )/6. Replacing with
r and b gives us simply r3 + r2 b + rb2 + b3 . Each color distribution gives
us only one equivalence class of colorings.
7. If the bands are labeled 1, 2, 3, 4 from left to right, the group consists of
two elements, e and (14)(23). The cycle index is thus C(z1 , z2 ) = (z14 +
z22 )/2. Expanding in r and b gives us (2r4 + 4r3 b + 8r2 b2 + 4rb3 + 2b4)/2.
Dropping the r4 and b4 terms leaves 2r3 b + 4r2 b2 + 2rb3 , so there are
eight ways to color the pole if each color is used at least once.
8. Numbering the corners 1 through 5 clockwise, we find ρ = (12345),
ρ2 = (13524), ρ3 = (14253), and ρ4 = (15432). This gives us C(z1 , z5 ) =
(z15 + 4z5 )/5. Substituting k gives 5

us C(k) = (k + 4k)/5. To two-color,
5 5 5
Ĉ(r, b) = (r + b) + 4(r + b ) /5.
9. Numbering the corners 1 through 7 clockwise, we find ρ = (1234567),
and all six powers of ρ are cycles of length seven. The various reflections
all look like α = (1)(27)(36)(45). This will give us C(z1 , z2 , z7 ) = (z17 +
6z7 + 7z1 z23 )/14. Substituting k gives us C(k) = (k 7 + 7k 4 + 6k)/14. To
two-color, Ĉ(r, b) = (r + b)7 + 6(r7 + b7 ) + 7(r + b)(r2 + b2 )3 /14.


358 Introduction to Combinatorics

Chapter 9
1. See the diagram on the left in the figure below.

All All

All All

All All All All All

All All

All

2. See the diagram on the right in the figure above.

3. Let c = a ∧ b; then d = b ∧ a. Since c is the greatest lower bound of a
and b, we must have c  d; but d is the greatest lower bound of b and
a, so d  c. It follows that c = d, so a ∧ b = b ∧ a. The proof for a ∨ b is
similar.
4. We show a∧a = a. Notice that a  a, so that a is indeed a lower bound.
Also, if b is any other lower bound, then by definition b ≺ a, so that a
is the greatest lower bound. The proof for a ∨ a is similar.
5. If a ∧ b = a, then a  b follows immediately; if a  b then (since a  a)
clearly a is a lower bound; if c is any lower bound with a  c then c  a
(because it is a lower bound); thus a = c. Thus a is the greatest lower
bound.
6. Associate with vertex a the set {a, c, d, f }; b with {b, d, e, g}; c with
{c, f }; d with {d, f, g}; e with {e, g}; f with {f }; and g with {g}.
7. Suppose M is an antichain consisting of only minimal elements, and let
m be a minimal element not contained in M . We need to show that m
is not comparable to any element of M . But this is immediate; since
m is minimal, there is no x with x ≺ m, and if m ≺ x then x is not
minimal. Thus M ∪ {m} is a larger antichain containing M .
8. The minimal elements are g, h, j, and k.
9. The width is 5.
 Solutions to Set A Exercises 359

Chapter 10
1. Consider a graph with v vertices. All degrees are less than v. If all are
different, the degree sequence is (v − 1, v − 2, . . . , 1, 0). The vertex of
degree v − 1 is adjacent to every other vertex, but it cannot be adjacent
to the vertex of degree 0—a contradiction.
The multigraph example is

2. (i) (4, 2, 2, 2, 1, 1) ⇒ (1, 1, 1, 0, 1) = (1, 1, 1, 1, 0) ⇒ (0, 1, 1, 0) Yes.

(ii) (5, 4, 3, 2, 2, 2) ⇒ (3, 2, 1, 1, 1) ⇒ (1, 0, 0, 1) Yes.
(iii) (3, 3, 3, 3, 2, 2) ⇒ (2, 2, 2, 2, 2) ⇒ (1, 1, 2, 2) = (2, 2, 1, 1) ⇒ (1, 0, 1)
Yes.
(iv) No (sum is odd).
(v) (6, 5, 2, 2, 1, 1, 1, 0) ⇒ (4, 1, 1, 0, 0, 0, 0) No.
(vi) (6, 5, 4, 3, 3, 3, 1, 1, 0) ⇒ (4, 3, 2, 2, 2, 0, 1, 0) ⇒ (2, 1, 1, 1, 0, 1, 0) ⇒
(0, 0, 1, 0, 1, 0) Yes.
Graphs:
(i) (ii) (iii) (vi)

3. xabcf , xabef , xadef , xdabcf , xdabef , xdef , xadbecf , xdebcf .

4. a, 1 b, 2 c, 3 d, 1 e, 2 f, 3.

5. (i) 1 to a, c, w, x, 2 to d, v, y, 3 to e, z. (ii) 1 to a, c, k, 2 to all others.

6. Go along the path x, a, . . . until the first time you meet a vertex (other
than x) that appears in both paths. Say that vertex is c. Then follow
the second path back from c to y. The result is a closed walk containing
x. The walk will be a cycle unless it contains some repetition, and this
will only happen if a = b. In the latter case, the walk generated is not
a cycle, it goes xax, a repeated edge.

7. χ(G) = max (χ(G1 ), χ(G2 )).

8. (i) 4, (ii) 4, (iii) 3, (iv) 3, (v) 3, (vi) 3.

9. (i) Color G with two colors and apply a new third color to all new
vertices. (ii) No: for example, suppose G is a tree.

10. Beginning at vertex x, select some vertex v1 adjacent to x and mark

the edge xv1 “used.” If possible, choose v1 to be of odd degree; then we
are done. If not, v1 has even degree, so there must be at least one edge
incident to v1 that is not yet used. Now choose a neighbor of v1 other
than x and call it v2 . If v2 has odd degree, we are done. Continue this
360 Introduction to Combinatorics

process. Each time we select a new vertex we use one edge, so if the
vertex has even degree there must be another edge that has not been
used, and we can go on. But the process must end, because there are
only finitely many edges. So we must eventually reach an odd vertex.

Chapter 11
1. (i) No; 1 (ii) Yes (iii) No; 3 (iv) Yes (v) No; 3 (vi) 3.
Solutions to (ii) and (iv) are found by the algorithm. Here are sample
solutions for the eulerizations.
(i) (iii) (v) (vi)

2. None is Eulerian. Numbers are 4, 1, 2, 5, 2, 2.

3. Part I shows the maze with a graph imposed on it, part II shows the
solution:

I II

4. These are not the only solutions:

(i) (ii) (iii) (iv)

5. If a is to be on the path, edge ab is necessary. For c, bc is needed.

Similarly g requires bg. So the “path” has a threeway at b.
 Solutions to Set A Exercises 361

6.

7. One well-known example is:

8. (i) Sorted edges: (31245). Nearest neighbors: (12345), (21345), (31245),

(41235), (51234). (ii) Say the cycle is (abcde). The length of this cycle
is
(a + b) + (b + c) + (c + d) + (d+) + (e + a) = 2(a + b + c + d + e).
But {a, b, c, d, e} = {1, 2, 3, 4, 5} (in some order), so the sum is always
30.
9. (i) sorted: cebda (103); a: acebd (103); c: cadbe (103).
(ii) sorted: dabec (289); a: abecd (289); c: cbead (290).
(iii) sorted: bacde (122); a: acdbe (121); c: cdabe (124).
(iv) sorted: dbcea (125); a: aecbd (125); c: cbdea (121).
10. (i) Sorted edges: xvywuz, 26; others: uvyw, fails; vywuzx, 26; wuvyzx,
26; xvywuz, 26; ywuvxz, 32; zxvywu, 26.
(ii) Sorted edges: wxyz, 28; others: wxy, fails; xwzy, 28; ywx, fails;
zwxy, 28.
11. In order to reach z, edges wz and zx must be included, so wx cannot
occur. Sorted edges starts with tu, uw, wx. Every nearest neighbor
includes wx. To find a Hamilton cycle, first assume the path vywzx; to
complete, go xutsv.
12. The trees are

13. Say there are v vertices. There are v − 1 edges, so the degrees add to
2v − 2. Each degree is at least 1. Assume one vertex has degree n. Say
there are x of degree 1 and y of degree 2 or greater. Then x + y = v − 1;
the sum of degrees is at least n + x + 2y = n + x + 2(v − 1 − x) =
2v − 2 + n − x. So n − x ≤ 0, x ≥ n.
362 Introduction to Combinatorics

14. (i) For the multiple path, one has n choices of edge for each edge of the
underlying path, giving nv−1 paths in all.
(ii) For the multiple cycle, each spanning tree is a path; there are v
choices for the pair of adjacent vertices that will not be adjacent in the
spanning tree, and for each choice there are again nv−1 paths.

15. The following solutions were found using Prim’s algorithm. Even so,
they are not unique: For example, in (i), edge ad could have been chosen
instead of de.
(i) a b c (ii) a b 9 c
4
6 3 7 3 8 6 7
6
d 6 e f 6 d e
5 4
9 8 5 3
3
g h i f g h
5
(iii) a b c (iv) a b
6 5 3 5 4 5 8
3 6
c d
e f
d e
3
9 3 8 4 36 6 5 8 4
5
f 4 g g h i j
7
4 6 3 4 2
h j k l
4 i 4

Chapter 12
1. (i) 2. (ii) 8. (iii) 3. (iv) 5.

2. Suppose A = a1 a2 . . . an , B = b1 b2 . . . bn and C = c1 c2 . . . cn are three

codewords. We need to prove that d(A, C) ≤ d(A, B) + d(B, C).
We can partition {1, 2, . . . , n} into four sets: P = {i : ai = bi = ci }; Q =
{i : ai = bi 6= ci }; R = {i : ai 6= bi = ci }; S = {i : ai 6= bi 6= ci }. Notice
that—since there are only two possible values for entries—members of
S all satisfy ai = ci . Write SAB for the set of places where A and B
differ, and so on. Then SAB = R ∪ S, SBC = Q ∪ S, SAC = Q ∪ R. Now
|S| ≥ 0, so
 Solutions to Set A Exercises 363

d(A, C) = |SAC | = |Q| + |R|

≤ |Q| + |R| + 2|S|
= (|R| + |S|) + (|Q| + |S|)
= |SAB | + |SBC | = d(A, B) + d(B, C).

3. Test the received word to see its distance from A, B, C, D, E. If the

distance is 1 or 0, that is definitely the intended codeword. Otherwise,
consider further. (i) A (distance 1). (ii) B (distance 1). (iii) C (dis-
tance is 2, but every other codeword is distance greater than 2 from the
candidate.) (iv) The word is distance 2 from D and also distance 2 from
E, so we cannot decode it.

4. Minimum distance is 4. So the code can detect 3 errors and always

correct 1.

5. (i) 32. (ii) 1; 0.

6. For four sequences, use {00000, 11100, 00111, 11011}.

If we have a set of five there must be three that begin with the same
symbol: say 0abcd, 0ef gh, 0ijkl. Without loss of generality we can
assume that the last of these is 00000 (if, for example, i = 1 then we
can change the second entry in all five codewords, 0 becoming 1 and 1
becoming 0, so as to achieve a set of five where Z = 00..., and similarly
for the other positions). Then at least three of abcd equal 1, as do at
least three of ef gh. Then abcd and ef gh must have at least two common
entries, and 0abcd and 0ef gh are distance at most 2.

7. (i) 1010001. (ii) 0011100. (iii) 1001101. (iv) 0010111.

8. (i) 1001. (ii) 0010. (iii) 1001. (iv) 1101. (v) 1011. (vi) 1100.
 
1 0 0 1 1 0
9.  0 1 0 1 0 1  ; 1; 0.
0 0 1 0 1 1
(For example, you cannot decide whether 100000 was meant to be
000000 (decodes as 000) or 100100 (decodes as 100).)

10. (i) 00010001100010110011101011100 (ii) 1001000001000110010100000.

11. (i) ACAACCAAAAGAAATAA (ii)AAACAATAAGACACAAAC.

364 Introduction to Combinatorics

Chapter 13
1.
0 1 2 3 4 5
1 2 3 4 5 0
2 3 4 5 0 1
.
3 4 5 0 1 2
4 5 0 1 2 3
5 0 1 2 3 4

2. Assume the symbols are 1, 2, 3, 4, 5. Then the (2, 1) and (2, 4) entries
are equal (there are many other problems).

3. (i) Suppose a Latin square L has first row (a1 , a2 , . . . , an ). Carry out
the permutation 1 7→ a1 , 2 7→ a2 , . . . , n 7→ an on the columns of L.
Permute the rows in the corresponding fashion.
(ii) Each reduced square gives rise to n! different squares by column
permutation. Once a column permutation is chosen, there are (n − 1)!
different row permutations available (leaving row 1 as the first row).
The n!(n − 1)! resulting squares are all different (any two will differ
somewhere in the first row or column).
(iii) r3 = 1, r4 = 4.
(iv) The reduced squares are:

1 2 3 4 1 2 3 4
2 1 4 3 2 1 4 3
3 4 1 2 3 4 2 1
4 3 2 1 4 3 1 2

1 2 3 4 1 2 3 4
2 3 4 1 2 4 1 3
.
3 4 1 2 3 1 4 2
4 1 2 3 4 3 2 1

4. Suppose A is a Latin square of side n based on {a1 , a2 , . . . , an }, and B

is a Latin square of side r based on {b1 , b2 , . . . , br }. Rows (s − 1)r +
1, (s − 1)r + 2, . . . , sr of A × B form the array
[(as,1 , B), (as,2 , B), . . . , (as,n , B)] .

Now the elements in each row of (as,t , B) are precisely the ordered pairs
(as,t , bj ), 1 ≤ j ≤ r, once each. As t ranges from 1 to n, as,t covers
each of a1 , a2 , . . . , an precisely once, so the (as,t , bj ) go through all the
(ai , bj ) once each, and each row of A × B contains each pair precisely
 Solutions to Set A Exercises 365

once. This is the row property of a Latin square. The column property
is similar. If A and B are transversal squares, the transversal property
for A × B follows from the same argument.

5. (i) Where will the entry (1,1) occur in (A, AT )? If it is in position ij,
it also appears in position ji. To avoid two occurrences, we need i = j,
a diagonal position. So 1 appears on the diagonal somewhere. The
same argument applies to every symbol. So each of 1, 2, . . . , n is on the
diagonal. There is just enough room for each symbol to appear exactly
once.
(ii) Say A is a self-orthogonal Latin square of side 3 whose diagonal is
(1, 2, 3). The only possible (1, 2) entry is 3, and it is also the only possible
(2, 1) entry. Therefore (3, 3) appears too many times in (A, AT ).
(iii)
1 3 2 5 4
1 3 4 2
4 2 5 1 3
4 2 1 3
5 4 3 2 1 .
2 4 3 1
3 5 1 4 2
3 1 2 4
2 1 4 3 5

6. A seating plan is obviously equivalent to a pair of orthogonal Latin

squares of order 5; one based on A, B, C, D, E (to specify the university),
the other on 1, 2, 3, 4, 5 (for example, 3 means “discipline d3 ”).
(i) Use any two orthogonal Latin squares of side 5.
(ii) One example is

A C B D E 1 3 5 2 4
E A C B D 5 2 4 1 3
D E A C B 4 1 3 5 2.
B D E A C 3 5 2 4 1
C B D E A 2 4 1 3 5

(The second square could be anything orthogonal to the first.)

(iii) This is impossible. Without loss of generality, assume the second
square has all 1s on its diagonal. Then the (1, 2) entry must be 2, as this
is the only cell next to (1, 1); similarly the (5, 4) cell is 2. Either the (2, 1)
or the (2, 3) cell must contain 2; it must be the (2, 1), because otherwise
there will be no 2 in column 1, and similarly the (4, 5) cell contains 2.
Where does the fifth 2 go? The only row and column without a 2 are
row and column 3, and the (3, 3) cell is occupied.
(iv) Yes and yes:
A C B D 1 2 3 4
B D A C 2 1 4 3
D B C A 3 4 1 2
C A D B 4 3 2 1
366 Introduction to Combinatorics

7. Without loss of generality, suppose the subsquare occupies rows and

columns 1 to s. As s < n, there is some symbol (x say) that is greater
than s; it will not appear in the subsquare. It will appear in each of
rows 1, 2, . . . , s, in s different columns, each numbered greater than s.
At the very smallest possible, these will be in columns s + 1, s + 2, . . . , 2s
(and maybe greater numbers will arise). So n ≥ 2s.

8. Each column contains n − 1 different elements; complete the column to

length n by inserting the unique missing element. In the final array,
every symbol appears once in each column, so each occurs n times in
total. In the original Latin rectangle, every row was a permutation, so
every symbol appeared once per row, or n − 1 times in total. So every
symbol must have occurred precisely once in the nth row.
9. We find an SDR for the sets {23}, {34}, {45}, {15}, {12}. Try 2 for the
first set. The remaining members must represent {34}, {45}, {15}, {1}.
So the fifth is 1. Now we need an SDR for {34}, {45}, {5}. Representa-
tive of set {5} is of course 5. In this way we get (2, 3, 4, 5, 1). Now the
last row is determined: the square is

1 2 3 4 5
5 1 2 3 4
4 5 1 2 3.
2 3 4 5 1
3 4 5 1 2

(Note: We did this solution in some detail because the same thinking
will apply to harder problems.)
10. (i) One example:
1 5 3 4 2 6
3 2 1 5 6 4
2 3 6 1 4 5
.
6 1 4 2 5 3
4 6 5 3 1 2
5 4 2 6 3 1
(ii) Impossible because 5 only appears once.
11. (i) x = 6 because there is only one 6 in the array. One solution is

1 2 3 4 6 5
5 6 1 2 3 4
3 4 5 1 2 6
.
4 1 2 6 5 3
6 3 4 5 1 2
2 5 6 3 4 1
 Solutions to Set A Exercises 367

(ii) Impossible. There is only one 5 in the array, but x cannot equal 5
or there would be two 5s in column 4.
12. One example is
5 2 3 6 4 1
3 5 2 4 1 6
1 4 6 5 2 3
.
2 3 5 1 6 4
6 1 4 2 3 5
4 6 1 3 5 2
13. (i) Assume the diagonal is (1, 2, 3, 4, 5). The (1, 2) entry could be 3, 4 or
5; the case of 5 will be isomorphic to the case of 4 (apply permutation
(45) to rows, columns and symbols).
In the former case, the (1, 3) entry cannot be 2 (you would have a 3 × 3
subsquare; see Exercise 12A.7) so it might as well be 4 (the permutation
(45) could be used if the entry were 5). Then you get
1 3 4 5 2
3 2 1
4 1 3
5 4
2 5
and there is no possible (2, 4) entry.
In the latter case, the (1, 3) entry can be 2 or 5 (the permutation (45)
could be used if the entry were 5). If it is 2 there is no completion (the
first row must be (1, 4, 2, 5, 3) and the second row cannot be completed;
if it is 5, the unique solution is
1 4 5 3 2
4 2 1 5 3
5 1 3 2 4.
3 5 2 4 1
2 3 4 1 5

(ii) We would need one of the squares to have 3 in the (1, 2) position.
14. (i) Say there are t copies of 1 above the diagonal. Then there will also
be t copies of 1 below the diagonal, so there are n − 2t copies of 1 on the
diagonal. This number must be non-negative; since n is odd, it cannot
be zero. So it is positive. Similarly for all other entries; so there is at
least one of each symbol on the diagonal. Since there are only n diagonal
positions, each integer from 1 to n must appear there once.
(ii) Try a 2 × 2 square.
368 Introduction to Combinatorics

Chapter 14
1. Three blocks are 12, 23 and 13. The other two must be either two
copies of the same block (e.g., {12, 23, 13, 12, 12}) or different blocks
(e.g., {12, 23, 13, 12, 13}). Any example can be derived from one of these
two by permuting the names of varieties.

2. Write S for the set comprising the first v positive integers, and write
Ss for the set formed by deleting i from S. Then blocks S1 , S2 , . . . , Sv
form the required design.

3. Clearly v, b and k are correct. To see that r = 5, check each element:

for example, element 1 is in blocks 1, 2, 3, 4, 5, element 2 is in blocks 1,
2, 6, 7, 8, and so on. Similarly check to see that λ = 2. The constancy
of k and r show that this is a block design, and the constancy of λ shows
that the design is balanced.

4. The easiest example is to take the seven blocks of a (7, 7, 3, 3, 1)-design,

twice each. For example,
123 145 167 246 157 347 356
.
123 145 167 246 157 347 356
(Note: This example is not simple. See Problem 14.5.)

5. Clearly λ ≤ r. If λr then every block must contain every element, and

the design is complete.

6. (v, 12 v(v − 1), v − 1, 2, 1).

7.
v b r k λ
15 35 7 3 1
21 28 3 r(k − 1) = 8; λ(v − 1) = 20λ > 8
11 11 5 5 2
14 7 4 v>b
12 6 3 11λ = 12; impossible
49 56 8 7 1

8. In each case v is even but k − λ is not a perfect square.

9. (v, tb, tr, k, tλ).

10. Write Ci = {j : i ∈ Bj }. Then the dual has blocks

 Solutions to Set A Exercises 369

C0 = {0, 1, 2, 3, 4} C3 = {1, 3, 6, 8, 9}
C1 = {0, 1, 5, 6, 7} C4 = {2, 4, 6, 7, 8}
C2 = {0, 2, 5, 8, 9} C5 = {3, 4, 5, 7, 9}.

Since {0, 1} lies in two blocks and {0, 2} lies in one block, the design is
not balanced.
11. (i) (12, 22, 11, 6, 5), (11, 22, 10, 5, 4);
(ii) (8, 14, 7, 4, 3), (7, 14, 6, 3, 2);
(iii) (33, 44, 12, 9, 3), (12, 44, 11, 3, 2);
(iv) (4, 12, 9, 3, 6), (9, 12, 8, 6, 5);
(v) (52, 68, 17, 13, 4), (17, 68, 16, 4, 3).
12. One solution: {1, 3, 4, 5, 9}.
13. One solution: {1, 2, 5}, {1, 6, 8}.

Chapter 15
1. Square A; square the result; square again, and repeat until you get A64
(six multiplications). Then A64 × A16 × A2 gives A82 in eight multipli-
cations.
 
0 1
2. (i) Let A = . Its eigenvalues are 3 and −1. We find the cor-
3 2  
1 1
responding eigenvectors and make them the columns of P = .
3 −1
   
3 0 2
The matrix D becomes . We find P −1 and P Dn P −1 to
0 −1 2
get an = 3n + (−1)n .
(ii) We use
 the
 same A, D, and P as for the previous exercise. Now
n −1 1
PD P gives us an = 3(3n ) − (−1)n .
7
     
0 1 1 1 4 0
(iii) Let A = . We get P = , D = , and
8 2 4 −2 0 −2
 1 1 
P −1 = 23 −1 6 . This gives us bn = 4n − 2(−2)n .
3 6
(iv) We use the same A, D, and P as for the previous part. We get
bn = 4n + (−2)n .
370 Introduction to Combinatorics

3.
A D

Halt

C B

A B C D Ha A B C D Ha
A 0 0 1 1 0 A 2 1 1 2 2
B 0 0 0 1 1 B 1 1 0 1 1
C 0 1 0 1 0 C 1 1 1 2 1
D 1 0 0 0 1 D 1 0 1 2 2
Ha 0 0 0 0 0 Ha 0 0 0 0 0
4.
A C

D B

Halt

A B C D Ha A B C D Ha
A 0 0 1 1 0 A 2 1 1 2 5
B 0 0 0 1 1 B 1 1 0 1 3
C 0 1 0 1 0 C 1 1 1 2 4
D 1 0 0 0 1 D 1 0 1 2 4
Ha 0 0 0 0 1 Ha 0 0 0 0 1
The matrix M̂ 4 has a (1, 5) entry equal to the number of ways in which
the machine can go from A to Halt in four or fewer steps. ThePproduct
∞
(I5 − M 5 )(I5 − M )−1 will not yield the correct answer because i=0 M i
does not converge. This is because we may make paths of arbitrary
length by shifting from A to D and back before ending at Halt.

5. (a)   (I − A)−1 does not exist; but

1 2 3 4 5 6 P6 i
1 i=1 A has a (1, 6) entry of 14.
 0 1 0 0 0 0 

2 1 0 1 1 0 0 
 
3 0 1 0 0 1 0 .
 
4 0 1 0 0 1 0 
 
5 0 0 1 1 0 1 
6 0 0 0 0 1 0
 Solutions to Set A Exercises 371

(b)   (I−A)−1 exists with a (1, 6) entry

1 2 3 4 5 6
of 1; and 6i=1 Ai has a (1, 6) en-
P
1 0 1 0 0 0 0 try of 1. Although (I −A)−1 gives

 
2 1 0 1 0 0 0 

3

. the correct answer, this is an ac-
 0 1 0 1 0 0  cident; other entries in (I − A)−1
4 0 0 1 0 1 0 

5
 are negative or fractional. Either
0 0 0 1 0 1 
is clearly impossible.
6 0 0 0 0 1 0

(c)   (I − A)−1 exists with a (1, 6) en-

1 2 3 4 5 6 P6
1 try of 1; but i=1 Ai has a (1, 6)
 0 1 1 0 0 0 
 entry of 6.
2 1 0 0 0 0 0 
 
3 1 0 0 0 0 0 .
 
4 0 0 1 0 1 0 
 
5 0 0 0 1 0 1 
6 0 0 0 0 1 0

(d)   (I − A)−1 exists with a (1, 6) en-

1 2 3 4 5 6 P6
1 try of 1; and i=1 Ai has a (1, 6)
 0 1 0 0 0 0 
 entry of 1. Again, (I − A)−1 acci-
2 1 0 1 0 0 0 

3

. dentally gives the correct answer.
 0 1 0 1 0 0 
4 0 0 1 0 1 0 
 
5 0 0 0 1 0 1 
6 0 0 0 0 1 0

6. (a)   (I − A)−1 does not exist; but

1 2 3 4 5 6 P6 i
1 i=1 A has a (1, 6) entry of 184.
 0 1 1 0 0 0 

2 1 0 1 1 1 0 
 
3 1 1 0 1 1 0 .
 
4 0 1 1 0 1 1 
 
5 0 1 1 1 0 1 
6 0 0 0 1 1 0

(b)   (I − A)−1 exists with a (1, 6) en-

1 2 3 4 5 6 P6
1 try of 1; but i=1 Ai has a (1, 6)
 0 1 0 0 0 0 
 entry of 29.
2 1 0 1 1 0 0 
 
3 0 1 0 1 0 0 .
 
4 0 1 1 0 1 1 
 
5 0 0 0 1 0 1 
6 0 0 0 1 1 0
372 Introduction to Combinatorics

(c)   (I−A)−1 exists with a (1, 6) entry

1 2 3 4 5 6
of 1/4; but 6i=1 Ai has a (1, 6)
P
1 0 1 1 0 0 0 

2
 entry of 29.
 1 0 1 0 0 0 

3 1 1 0 1 0 0 .
 
4 0 0 1 0 1 1 
 
5 0 0 0 1 0 1 
6 0 0 0 1 1 0

(d)   (I−A)−1 exists with a (1, 6) entry

1 2 3 4 5 6
of 11/9; but 6i=1 Ai has a (1, 6)
P
1 0 1 1 0 0 0 

2
 entry of 116.
 1 0 1 1 0 0 

3 1 1 0 1 1 0 .
 
4 0 1 1 0 1 1 
 
5 0 0 1 1 0 1 
6 0 0 0 1 1 0

7. (a)   (I − A)−1 has a (1, 6) entry of 9,

1 2 3 4 5 6
as does 6i=1 Ai . In fact, An is
P
1 0 1 1 0 0 0 

2
 all zero for n ≥ 6.
 0 0 1 1 1 0 

3 0 0 0 1 1 0 .
 
4 0 0 0 0 1 1 
 
5 0 0 0 0 0 1 
6 0 0 0 0 0 0

(b)   (I − A)−1 has a (1, 6) entry of 4

1 2 3 4 5 6
as does 6i=1 Ai .
P
1 0 1 0 0 0 0 
 
2 0 0 1 1 0 0 
 
3 0 0 0 1 0 0 .
 
4 0 0 0 0 1 1 
 
5 0 0 0 0 0 1 
6 0 0 0 0 0 0

(c)   (I − A)−1 has a (1, 6) entry of 4

1 2 3 4 5 6 P6
1 as does i=1 Ai .
 0 1 1 0 0 0 

2 0 0 1 0 0 0 
 
3 0 0 0 1 0 0 .
 
4 0 0 0 0 1 1 
 
5 0 0 0 0 0 1 
6 0 0 0 0 0 0

(d)   (I − A)−1 has a (1, 6) entry of 4

1 2 3 4 5 6 P8
1 as does i=1 Ai .
 0 1 0 0 0 0 

2 0 0 1 0 0 0 
 
3 0 0 0 1 0 0 .
 
4 0 0 0 0 1 0 
 
5 0 0 0 0 0 1 
6 0 0 0 0 0 0
 Solutions to Set A Exercises 373

8. In each case, the matrix A is just like the matrix for the last exercise,
except with a 1 for the (6, 6) entry. The matrix (I − A)−1 does not exist
in each case, because the geometric series will not converge.
a) 6i=1 Ai has a (1, 6) entry of 30, which is far too large. But A6 has
P
a (1, 6) entry of 9, which is the correct number of ways to move from
square 1 to square 6.
b) 6i=1 Ai has a (1, 6) entry of 12, which is far too large. But A6 has
P
a (1, 6) entry of 4, which is the correct number of ways to move from
square 1 to square 6.
c) 6i=1 Ai has a (1, 6) entry of 12, which is far too large. But A6 has
P
a (1, 6) entry of 4, which is the correct number of ways to move from
square 1 to square 6.
P6
d) i=1 Ai has a (1, 6) entry of 26, which is far too large. But A6 has
a (1, 6) entry of 8, which is the correct number of ways to move from
square 1 to square 6.
 
9. 1 0 1 0 1 0
A =  0 1 0 1 0 1 .
0 0 1 0 0 1
It is simplest to expand about the last row; in any event, we get four
1 × 4 matrices each containing three 1s, to give a permanent of 12.
 
10. 1 1 0 0 0
1 0 1 0 0
 
 1 1 1 1 0  = 5.
Per  
0 0 1 0 1
0 0 1 1 1

11. We use n = 8 and k = 4 to get Per(A) = ⌈8! (4/8)8 ⌉ = ⌈315/2⌉ = 315.

Appendix A
1. (i) {a, b, c, d, e, g, i}, (ii) {c, e}, (iii) {g, i}, (iv) {a, b, c, d, e, g}.

2. S1 ⊆ S3 , S4 ; S2 ⊆ S3 , S4 , S5 ; S5 ⊆ S2 , S3 , S4 ; S2 = S5 .

3. S ∪ T = U ⇒ S ⊆ T ; S ∩ T = ∅ ⇒ T ⊆ S.

4. (i) and (iii).

374 Introduction to Combinatorics

5. S + T consists of all the elements that are in S or T but not both. So

S + T = (S ∩ T ) ∪ (T ∩ S) and this is a disjoint union. On the other
hand, (S + T ) = (S ∩ T ) ∪ (T ∩ S) = (S ∩ T ) ∪ (T ∩ S).
S ⊆ T, S ∪ T = T, S ∩ T = S.

6. (i) Assume S ⊆ T . If x ∈ S∪T then x ∈ S or x ∈ T (or both). But, since

S ⊆ T , x ∈ S ⇒ x ∈ T , so in either case x ∈ T . So (assuming S ⊆ T ),
S ∪ T ⊆ T . But T ⊆ S ∪ T is always true. So S ⊆ T ⇒ S ∪ T = T .
(ii) S ∪ T consists of T together with all members of S that are not in
T . If S ∪ T = T then there are no members of S that are not in T , so
S ∩ T = S.
(iii) Assume S ∩ T = S. Consider any element of S. If it were not in
T , then it would be a member of S that is not in S ∩ T . Then the two
would not be equal. So no such member of S exists. So S ⊆ T .
We have proven S ⊆ T ⇒ S ∪ T = T ⇒ S ∩ T = S ⇒ S ⊆ T . So the
three are equivalent.

7. (i) Obviously true for n = 1: 1 = 12 . Say true for n = r: 1 + 3 + . . . +

(2r − 1) = r2 . Then

(1 + 3 + . . . + (2r − 1)) + (2r + 1) = r2 + (2r + 1) = (r + 1)2 .

(ii) Case n = 1 is true. Suppose it’s true for n = k. That is,

k
X
r2 = 61 k(k + 1)(2k + 1).
r=1

We need to prove case n = k, that is,

k+1
X
r2 = 61 (k + 1)(k + 2)(2k + 3).
r=1

Now, (k + 1)(k + 2)(2k + 3) = 2k 3 + 9k 2 + 13k + 6. But, using the

assumption,
Pk+1 2 P 
k 2
r=1 r = r=1 r + (k + 1)2
= 16 k(k + 1)(2k +
1) + (k + 1)2

= 16 2k 3 + 3k 2 + k
+ (k + 1)2
= 16 2k 3 + 3k 2 + k + k 2 + 2k + 1

= 16 2k 3 + 3k 2 + k + 6k 2
+ 12k + 6
= 61 2k 3 + 9k 2 + 13k + 6 .
 Solutions to Set A Exercises 375

8.
(n − 1)3 + n3 + (n + 1)3
= (n3 − 3n2 + 3n − 1) + n3 + (n3 + 3n2 + 3n + 1)
.
= 3n3 + 6n
2
= 3n(n + 2).

We need to show that 3 divides n(n2 + 2). If 3 divides n, we are finished.

If not, n ≡ 1 or 2 (mod 3), so n2 ≡ 1 (mod 3), and 3 divides (n2 + 2).

9. Rewriting slightly, xn = xn−1 + 2xn−2 = (xn−2 + 2xn−3 ) + 2xn−2 =

3xn−2 + xn−3 . So xn ≡ xn−3 (mod 3). Now x1 = x2 = 1, so xn
is not divisible by 3 when n ≡ 1 or 2 (mod 3). On the other hand,
x3 = x2 + 2x1 = 1 + 2 = 3, so xn is divisible by 3 when n ≡ 0(mod3).

10. (i) ρσ = {(2, 3), (3, 3), (3, 4), (3, 5), (4, 1), (4, 2), (5, 4), (5, 5)}.
(ii) σρ = {(1, 2), (2, 2), (2, 4), (3, 2), (4, 1), (4, 3), (4, 5), (5, 1), (5, 3), (5, 5)}.
(iii) ρρ = {(2, 1), (3, 1), (3, 2), (3, 3), (4, 2), (4, 4), (5, 5)},
σσ = {(1, 3), (2, 1), (2, 2), (3, 2), (3, 3), (4, 4), (4, 5), (5, 4), (5, 5)}.

11. ρ is reflexive if and only if

x + y = y + x for all x, y ∈ Z,

but that is just the commutative law for integers. Symmetry means
that, for all x, y, z, t,

x + t = y + z ⇒ z + x = t + y,

which follows from the commutative law for integers and the symmetric
law for equality. For transitivity, assume

x + t = y + zz + b = t + a.

Adding these two equations,

(x + t) + (z + b) = (y + z) + (t + a).

Subtracting z and t from both sides,

x + b = y + a,

which is equivalent to (x, y) ρ (a, b).

376 Introduction to Combinatorics

Appendix B
 
    10 3
3 3 −3 4 −2
1. (i) . (ii) . (iii)  16 −2  .
−6 0 24 −2 2
−8 3

2. x = 11; y = 9; z = 23 .
3. (i) (6, 12, 3). (ii) 4(12, 4, −4). (iii) (3, 7). (iv) 2(1, −11, 10, 15).
4. (i) no; (ii) 1 × 3; (iii) no; (iv) 4 × 3; (v) no; (vi) no; (vii) no; (viii) 3 × 3;
(ix) 3 × 3.
 
    1 3 0
6 −4 17 −14
5. (i) . (ii) . (iii)  1 3 0 .
3 1 9 −6
−1 −3 0
   
3 1 8
(iv)  1 0 . (v)  0 .
−1 −3 −1
 
5 −3
6. A= .
−10 7
   
2 −1 3 −2
7. (i) , .
−1 1 −2 1
   
1 −4 1 1 14 −5
(ii)  0 9 −5  ,  0 −27 19 .
0 0 4 0 0 −8
 
1 −2
8. (i) 3; . (ii) 0; no inverse.
− 32 53
 
  6 4 4
3 9
9. (i) AB = AC = . (ii) AB = AC =  4 3 4 .
3 9
8 5 4
Hints for Problems

Chapter 1
2. What if there is an element a such that aρb is never true for any b?
5. Think of chords.
7. Construct a graph whose vertices are the attendees (other than the
hostess); there is an edge if two people shook hands. The degrees must
be 0, 1, 2, 3, 4, 5, 6, 7, 8. Consider the person that shook hands eight
times; the only person he/she missed must be his/her spouse. What
does this tell you about the other vertices’ degrees?
8. Since E only belongs to O4 , clearly if E resigns no SDR is possible as
there are only four willing students and five organizations. However, if
D were to resign from O5 , then O1 , O2 , O3 would each consist of only
A and C among the willing students. Then, if student E were to join
either O1 or O3 , no one resignation could prevent an SDR.
Checking this is tedious but straightforward; clearly we only have to
check the possibility of a resignation of a student from the organization
that he or she represents in the SDR we found for part (i).

Chapter 2
2. 1001 = 11 × 91; 100001 = 11 × 9091. Generalize this.
4. A combinatorial proof could involve selecting a committee of k individ-
uals from a larger group of m men and n women; the right-hand side of
the equation sorts the possible committees according to gender makeup.
5. You may select a committee of k people, with a chair (or facilitator). You
might choose the chair first, and then choose the other k − 1 committee
members, or you might choose the committee as a whole first and then
select the chair from the chosen committee members.

377
378 Introduction to Combinatorics

7. Use Corollary 2.3.3.

Chapter 3
5. Let the three dice be distinct. Then consider the number of ways to get
a 9 if the low roll is a one; for instance, there are six rolls with values
{1, 3, 6}. Repeat for other values of the low roll.
9. The sum of a subset must lie between 6 and 30. There are 210 subsets,
and 25 sums. So the average number of subsets with a given sum is at
least 8. Try to improve this.

Chapter 4
2. We are virtually using the pigeonhole principle in a reverse form; that
is, if fewer than n pigeons are roosting in n pigeonholes, then at least
one pigeonhole is unoccupied.
3. Look at Theorem 4.1(ii).
4. This is similar to Example 4.2.
7. A bipartite graph will contain no K3 ; what is the bipartite subgraph of
K2R(m,n)−2 with the most edges?
9. Use induction.

Chapter 5
1. Look at a Venn diagram.
2. One way is to prove that Dn+1 is odd if and only if n is even.
3. There are two cases: the previously vacant chair remains vacant or not.
5. Try induction.
 Hints for Problems 379

Chapter 6
P∞
1. Write fb (x) = k=0 (a0 + a1 + · · · + ak )xk , multiply both sides by 1 − x,
and collect like terms.

5. The two leftmost squares of the 2 × n rectangle are either covered by a

single 2 × 1 domino, in which case there are an−1 ways to complete the
process, or they are covered by two 1 × 2 dominos, in which case there
are an−2 ways to finish.

6. Let an be the number of strings of length n with this property. The

leftmost bit of the string is either a 0, in which case there are an−1
possibilities, or a 1, in which case the string begins with 10 and there
are an−2 possibilities.

Chapter 7
n
6. Use the explicit value for 2 and the fact that xk − y k is divisible by
x − y.

7. A distinguished element approach works here; take element n + 1, and

choose n − i others to go in its part of the partition.

Chapter 8
4. Consider for any i, j in S the permutation σj ◦ σi−1 .

7. For any i, j in S, consider σj ◦ σi−1 .

15. The eight permutations of D4 are the four rotations that were found
in the previous problem, together with the four reflections (13)(45)(68),
(16)(27)(38), (18)(25)(47), and (24)(36)(57).
380 Introduction to Combinatorics

Chapter 9
1. Simply modify the proof for the existence of a maximal element.

2. The Hasse diagram has an edge between x and y provided that there
is no element between them. What sort of relationships would three
mutually-adjacent vertices require?

3. Consider B4 where the set of four elements is the set of four primes
whose product is n.

4. It may help to think of the m-set as disjoint from the n-set.

6. Not every down-set is the down-set Lx generated by a single element.

7. Not all largest antichains consist only of maximal elements.

Chapter 10
3. (iv) Remember that the sum of the degrees is even.

4. Start with a cycle of length v.

5. See Exercise 10.B.2.

6. If S is a set of r vertices of G, how many edges can have at most one

endpoint in S?

Chapter 11
2. (i) Consider any two adjacent vertices. Use the fact that they can have
no common neighbor.
(ii) Consider any two adjacent vertices x and y and form a new graph
by identifying them (two vertices in the new graph are adjacent if and
only if they were adjacent in the old graph; all vertices adjacent to x or
y are adjacent to the new vertex).
 Hints for Problems 381

3. (i) Try to include a cutpoint.

(ii) Try to include odd vertices.

6. (i) First, solve the problem for n = 1. To solve the general case, treat
Kn,2n,3n as n copies of K1,2,3 together with further edges.

8. Show that any two vertices in the center of a tree are adjacent.

10. Use induction on v.

11. Use the result of the preceding problem.

Chapter 12
2. See the part of the text preceding the theorem.

5. Suppose a code has minimum distance 2n. What can go wrong?

Chapter 13
1. What is the (1, 2) entry?

3. Use induction on k.

4. If you take the set all integers modulo a prime (or more generally the set
of all members of a finite field) and multiply by one non-zero member,
you get a permutation of the same set.

6. (iii) Use three orthogonal Latin squares of side 4.

382 Introduction to Combinatorics

Chapter 14
1. Any two blocks have intersection size 2. Assume the treatments are
1, 2, 3, 4, 5, 6, 7. Consider the blocks that contain 1. Without loss
of generality, start with 1234. The second containing both 1 and 2
may as well be 1256 (if, for example, it is 1257, the permutation (67)
is an isomorphism). Show that the other blocks containing 1 are now
determined up to isomorphism (what are they?). Continue from there.
2. The proof of part (ii) is a generalization of Example 13.2.
3. The dual of a linked design must be balanced.
7. Use Exercise 14A.2 and Exercise 14A.9.
11. Divide the riders into two groups of 7. In the first seven heats, riders 1
through 7 meet each other once each.

Chapter 15
1. Use the matrix A found in part (i): show first that A×[an−1 an an+1 ]T =
[an an+1 an+2 ]T using the recurrence.
Solutions to Problems

Chapter 1
1. Suppose ρ is reflexive and circular.
xρy and yρz ⇒ zρx.

(i) Say xρy is true. We know yρy is true. Using the above equation
with z = y,
xρy ⇒ xρy and yρy ⇒ yρx.
So ρ is symmetric.
(ii) Apply symmetry to the equation:
xρy and yρz ⇒ zρx ⇒ xρz.
So ρ is transitive.

3. We use the fact that (n − 1)2 ≥ 2 when n ≥ 3. (In fact (n − 1)2 > 2 for
these values.) Therefore n2 − 2n + 1 ≥ 2 and n2 ≥ 2n − 1 + 2 = 2n + 1.
We proceed by induction on n. The result 2n ≥ n2 is true for n = 4
(both sides = 16). Assume n ≥ 4 and 2n ≥ n2 . Then

2n+1 = 2 × 2n ≥ 2 × n2 = n2 + n2 ≥ n2 + (2n + 1) = (n + 1)2 .

So the result is true for all n ≥ 4 by induction.

4. There must be at least two people. The result is true for 2 (the second
person, the man, is directly behind the first person, the woman). As-
sume it is true for n people. Consider any line on n + 1 people with
a woman in front and a man at the back. If the nth person is a man,
then by the induction assumption the required configuration (man di-
rectly behind woman) occurs somewhere in the first n places. If not,
the positions n and n + 1 provide the configuration.

5. The result is true (given) for n = 3. Assume it is true for n = k.

Consider a polygon with vertices A1 , A2 , . . . , An+1 . The sum of the
angles equals the sum of the angles of the polygon A1 , A2 , . . . , An plus
those of the triangle A1 , An , An+1 . The first equals (n − 2)π from the
induction hypothesis, the second is π, so the sum is ((n + 1) − 2)π.

383
384 Introduction to Combinatorics

6. Say there are v vertices. No vertex has degree greater than v − 1 or less
than 0. So, if they are all different, there must be one vertex of each
degree 0, 1, 2, . . . , (v − 1). The vertex of degree (v − 1) must be adjacent
to every other vertex, but it cannot be adjacent to the one of degree 0.

7. (i) 4; (ii) 4.

8. (i) The SDR is easily found; for instance, O1 = A, O2 = B, O3 = C, O4 =

E, O5 = D works.

Chapter 2
1. Write an for the number of positive n-digit integers with all digits dis-
tinct. If n > 10, there are no such integers. Otherwise, there are nine
choices for the first digit (you cannot start with 0), 9 for the second, 8
for the third, . . . , and (11 − a) for the ath place. So an = 9 × P (9, n − 1).
So the number of positive integers with all digits distinct is
a1 + a2 + . . . + a10
= 9[P (9, 0) + P (9, 1) + . . . + P (9, 9)]
= 9[1 + 9 + 72 + 504 + 3, 024 + 15, 120 + 60, 480 + 181, 440
+362, 880 + 362, 880]
= 9 × 986, 410
= 8, 877, 690.
There are just as many negatives, and do not forget 0. So the answer is
17,755,381.

6. There are several examples; pappa and pops are both colloquially used
to refer to one’s father, and there are only 10 distinct rearrangements
of pappa compared to 12 for pops. Now that you have looked up the
answer, try to find some words (not necessarily so similar) on your own.

8. Write On and En for the numbers of subsets of an n-set of odd and even
orders, respectively. Then
n   n  
X n X n
On = , En = .
k k
k=0,k odd k=0,k even

Pn
From Corollary 2.3.3, k=0 nk (−1)k = 0. The terms in this sum are

positive if k is even and negative if k is odd. So

 Solutions to Problems 385

n   n  
X n X n
= ,
k k
k=0,k odd k=0,k even

Pn
or On = En . From Corollary 2.3.2, k=0 nk = 2n , that is On + En =

2n . So On = 2n−1 (and we have shown that En = 2n−1 also).

10. (i) The 27 letters consist of 7 As, 6 N s, 5 Bs, 4 D’, 2 Ss, and one each
of I, T, R. So the number of rearrangements is

27!
− 1 = 520, 951, 478, 183, 135, 999.
7!6!5!4!2!
(We subtract 1 because rearrangements were called for, so the original
ordering is not included.)
(ii) Similarly we get

24!
− 1 = 62, 336, 074, 312, 511, 999.
6!4!4!3!2!2!

11. In the first case, the eight people can be seated in 7! ways. In the second
case, we seat one group first (in 3! ways), and then intersperse the others
between them in 4! ways, so there are 3!·4! = 144 arrangements. In the
last case, we seat the diplomat last, in one of thetwo endpoints or one of
the seven positions between two of the others; this gives us 9·144 = 1296
seating arrangements.

12. Clearly we must choose one person from each of n − 1 pairs, and no one
from the last pair. There are C(n, 1) = n ways to choose the pair that
will be neglected, and there will then be 2n−1 ways to choose one of
the two members of the n − 1 other pairs. So we have n·2n−1 possible
committees.

13. We take the cases one at a time. Practically speaking, there are three
cases; no vegetarian restaurants, no meat-only restaurants, and both.
The case “both” means we have one vegetarian and one meat-eater,
so we must stick to one of the five restaurants with omnivorous offer-
ings. This means that we can choose the friends in C(3, 1)·C(2, 1) = 6
ways, and any of the five restaurants will work, for 30 possibilities.
The “no-meat-only” case means that we exclude the meat-eaters but
not the vegetarians; there are C(8, 2) = 28 choices of two friends and
eight restaurants, so there are 224 possibilities. The “no-veg-only” case
means that we exclude the vegetarians but not the meat-eaters; there
are C(7, 2) = 21 ways to pick the friends, and seven restaurants accept-
able to these friends, for another 147 possibilities. Altogether, then,
there are 401 possibilities.
386 Introduction to Combinatorics

The student will realize that this is an oversimplification of the compli-

cations that occur in real life (for instance, two of the friends cannot
eat highly spicy foods as are served at one of the vegetarian and two of
the omnivore restaurants, and two of the friends have recently broken up
so cannot go anywhere together without causing a scene, and one of the
restaurants does not allow children while two of the friends cannot get a
babysitter so would have to bring the baby, etc.), but we trust that this
gives something of the flavor of the more complicated sorts of counting
problems.

Chapter 3
1. As we saw, P (1) = 0.2, P (2) = P (3) = P (4) = P (5) = P (6) = 0.16.
So the probability of a roll of 4 or less is P (1) + P (2) + P (3) + P (4) =
0.2 + 0.16 + 0.16 + 0.16 = 0.68.
2. We shall write the outcomes in the form An, where A is H or T , the
result of flipping the quarter, and n = 0, 1 or 2, the number of heads on
the pennies. The sample space is {H0, H1, H2, T 0, T 1, T 2}. The tree
diagram is

0
1
H 2
T 0
1
2

3. The diagram is
A

6 24

1 3 5
B C

4. Suppose the game is played N times. Alice has probability 14 of drawing

the same color as the host, so she will expect to win in 14 N games.
In the remaining 34 N games, Bruce draws. There are now two colors
that have been drawn, and if Bruce duplicates either, he wins. If he
 Solutions to Problems 387

draws, the chance of him winning is 12 , so the expected number of wins

is 34 × 12 N = 83 N . So his probability is 38 . If he loses, then three colors
have been drawn, so the chance of Cathy winning is 43 . There have been
1 3 5
4 N + 8 N = 8 N successes, so Cathy draws in f rac58N cases, and she
33 9 9
wins in 4 8 N = 32 N cases, so her probability is 32 . Therefore Bruce
still has the best chance of winning, but now Cathy’s chance is better
than Alice’s. The probability of no winner is 1 − 41 − 83 − 32 9
= 323
.
5. The possible ways to get a total of 9 from two dice are {3, 6} and {4, 5}.
Suppose the lowest roll is a 1; there are six rolls with values {1, 3, 6}, and
six with {1, 4, 5}. If the lowest roll is a 2, there are six rolls with values
{2, 3, 6}, and six with {2, 4, 5}. If the lowest is a 3, there are three rolls
with values {3, 3, 6}, and six with {3, 4, 5}. If the lowest is a 4, the only
possibility for the two highest is {4, 5}, and there are three rolls with
values {4, 4, 5}. So there are (6+6)+(6+6)+(3+6)+3 = 36 possibilities,
out of the 63 = 216 rolls. So the probability is 36/216 = 1/6.
25 125
6. (i) 56 ; 36 ; 216 .
(ii) For exactly k times, there are k6 ways to select the k times, and

the probability for those k times is probability is ( 16 )k × ( 65 )n−k , so the

probability of exactly k doubles is k6 × ( 16 )k × ( 56 )n−k . So the required

probability is found by summing this from k = 2 to 6. However, an easier

way is to calculate the probability of 1 or no doubles, and subtract this
from 1. That probability is
5 5 1 34375
( )6 + 6 × ( )5 × ( ) = ,
6 6 6 46656
12281
or about 73.7%. So the answer is 46656 , or about 26.3%.
8. Suppose E1 has n outcomes with values x1 , x2 , . . . , xn , and the proba-
bility of the outcome xi is pi . Suppose E2 has m outcomes with val-
ues y1P, y2 , . . . , ym , and thePprobability of the outcome yj is qj . Then
n m
v1 = i=1 pi xi and v2 = j=1 qj yj . If E1 has outcome xi and yi and
E2 has outcome yj , then the sum of the outcomes is xi + yj , and the
probability of this occurrence is pi qj . The expected value of the sum of
the two outcomes is hP i
Pn Pm Pn m Pm
i=1 j=1 pi qj (xi + yj ) = i=1 pi j=1 qj xi + j=1 qj yj
Pn Pn Pn
= i=1 pi [xi + v2 ] = i=1 pi xi + v2 i=1 pi
= v1 + v2 .

9. Let N (S) denote the sum of the elements of a
4-set S, and let T be the
set of all 4-subsets of {0, 1, . . . , 9}. T has 10
4 = 210 elements. Write
Tn for the set of all members S of T with N (S) = n. The smallest
possible value of n is 6, and the largest is 30. there is only one 4-set
with sum 6, namely {0, 1, 2, 3}, one with sum 7 ({0, 1, 2, 4}), one with
388 Introduction to Combinatorics

sum 29 ({5, 7, 8, 9}), and one with sum 30 ({6, 7, 8, 9}). So the remaining
206 4-subsets have sums from 8 to 28. If the 21 sets T8 , T9 , . . . , T28 each
contains nine or fewer elements, they would total at most 189 members
between them, less than 206, which is impossible. So some Tn has 10 or
more members.

Chapter 4
Sn
1. (i) Say S = i=1 Ai . If each Ai satisfies |Ai | ≤ m then |S| ≤ nm, a
contradiction. So there is at least one i such that |Ai | > m.
(ii) If ai ≤ nb for all i then ni=1 ai ≤ b, a contradiction.
P
Pn
(iii) If ai > nb for all i then i=1 ai > b, a contradiction.
(iv) If ai ≥ nb for all i then ni=1 ai ≥ b, a contradiction.
P

2. The worst case is when all boxes but one contain 1 object and the
remaining box contains N − (B − 1) = N − B + 1. So at least N − B + 1
must be chosen.
3. For (i) and (ii), suppose the statement is false. Then there are at most 12
people belonging to each of only 12 signs, so there are only 144 people.
But we know that there are 145 people; this contradiction establishes
the results. For (iii), there are 122 = 144 possible pairs of signs, and
145 people, so there are at least two people who have the same pair of
signs.
4. Suppose no two sets contain the same number of entries. At the very
least, one will be empty, one will have order 1, one 2, . . . and the total
number of entries is
p(p − 1)
n = 0 + 1 + . . . + (p − 1) = .
2
So n ≥ p(p−1)
2 is a necessary condition for all the Ai to be of different
size. (If n is larger, simply add more members to the largest set.) On
the other hand, if n = p(p−1)
2 + t, such a set A exists—for example,

A1 = {1}, A2 = {2, p + 2}, . . . , Ai = {i, p + i, . . . , (i − 1)p + i}, . . . , .

But finally
Ap = {p − 1, 2p − 1, . . . , (p + t)p − 1}.
p(p−1)
So n ≥ 2 is necessary and sufficient.
 Solutions to Problems 389

5. Consider the n products a1 , a1 a2 , . . . , a1 a2 . . . ai , . . . , a1 a2 . . . an . There

are two possibilities: if, say, a1 . . . aj = 1 then we choose s = 1, t = j and
the problem is solved; if none of these n products equals the identity,
then they take at most (n − 1) distinct values and hence at least two
of them must be equal, so a1 a2 . . . aj = a1 a2 . . . aj+k for some j and k,
with 1 ≤ j < j + k 6= n. Put a = j + 1, t = j + k and the problem is
again solved.
6. Suppose both (i) and (ii) are false. Partition the intervals into sets
S1 , S2 , . . . , where all members of a set are pairwise disjoint. As (ii) is
false, each set contains at most n intervals. Clearly no set of intervals
with a common element can contain two members of the same Ai ; as
(i) is false, there can be at most n sets Ai . So there are at most n2
intervals—a contradiction.
7. Divide the 2R(m, n) − 2 vertices into two sets X and Y of R(m, n) − 1
vertices each. Color the edges that connect a vertex in X to a vertex
in Y in color c1 . Color the edges that connect vertices in X using only
colors c2 and c3 in such a way that the subgraph induced by X contains
no Km in c2 or Kn in c3 (this coloring must exist since X has fewer
than R(m, n) vertices). Do the same with the vertices of Y . Because
the edges of color c1 form a bipartite graph, there is no K3 in c1 , and
by construction there is no Km in c2 or Kn in c3 . This shows that
R(3, m, n) ≥ 2R(m, n) − 1.
8. First, suppose n is odd, say n = 2t + 1. Then
$ $ 2 %%
n n−1
= (t + 21 )t2 = t3 + 12 t2
   
2 2

and
n(n − 1)2 (2t + 1)4t2
   
== t3 + 12 t2 .
 
=
8 8
So the two are equal.
When n is even, say n = 2t,
$ $ 2 %%
n n−1
= t(t2 − t) = t3 − t2 ,
 
2 2

but
n(n − 1)2 2t(2t − 1)2 8t3 − t2 + 2t
       
3 2 t
= = =t −t + ,
8 8 8 4
with equality if and only if ⌊ 4t ⌋ = 0, that is t < 4. So the two expressions
are equal unless n is even and n ≥ 8.
390 Introduction to Combinatorics

Chapter 5
1. (B ∩ C) ∪ (B ∩ C) = C, and the two sets on the left are disjoint, so
|B ∩ C| + |B ∩ C| = |C|. Similarly (A ∩ B ∩ C) ∪ (A ∩ B ∩ C) = A ∩ C
and |A ∩ B ∩ C| + |A ∩ B ∩ C| = |A ∩ C|. So

|A ∪ C|
= |A| + |C| − |A ∩ C|
= |A| + (|B ∩ C| + |B ∩ C|) − (|A ∩ B ∩ C| + |A ∩ B ∩ C|).

2. We prove by induction that Dn is odd if and only if n is even, or equiv-

alently Dn+1 ≡ n (mod 2). The result is true for n = 2 (and also for
n = 1, D1 = 0, but as we explained some people say D1 is not defined).
Say it is true for n < k. Then Dk and Dk−1 are of opposite parity, so
Dk − Dk−1 ≡ 1 (mod 2), so Dk+1 = k[Dk − Dk−1 ] ≡ k × 1 ≡ k (mod 2).

3. Suppose the people are numbered 1 to n and the chairs are numbered
0 to n; say person x occupies seat x initially and seat x′ after supper.
Suppose the previously vacant chair remains vacant. Then (1′, 2′, . . . , n′)
is a derangement of (1, 2, . . . , n). This can occur in Dn ways. To handle
the other case, 0′ is the number of the chair that is unoccupied after
supper; (0′, 1′, 2′, . . . , n′) is a derangement of (0, 1, 2, . . . , n). This can
occur in Dn+1 ways.

4. Write N for {1, 2, . . . , n}. Select any permutation (d1 d2 . . . dn ) of N

and write S = {i : di 6= i}. The number of different permutations in
which all symbols are in their natural position except for precisely the
members of S is equal to the number of derangements of S, or D|S| .
To get the number of permutations of N with exactly k numbers
out
of position, sum this over all nk subsets S of size k, obtaining nk Dk .


Now sum over k to count all permutations.

Chapter 6
an xn = n n
P P P
2. n≥0 n≥0 c1 n(rx) + n≥0 c0 (rx) . The second sum is
clearly c0 /(1 − rx); the first requires differentiation. We use
d P n n d
dx n≥0 c1 r x = dx c1 /(1 − rx).
 Solutions to Problems 391

This gives us n≥0 c1 rn nxn−1 = rc1 /(1 − rx)2 . Multiplying both sides
P

c1 nrn xn = c1 rx/(1 − rx)2 . Then we easily see that

P
by x gives n≥0 P
n n
P
c
n≥0 1 n(rx) + n≥0 c0 (rx) is a rational function with denominator
2
(1 − rx) , as required.
7. fa (x)fb (x) = (a0 + a1 x + a2 x2 + . . . ) × (b0 + b1 x + b2 x2 + . . . ) = c0 +
c1 x + c2 x2 + . . . . The coefficient cn will be the sum of coefficients of all
products of terms of degree n. So, c0 = a0 b0 because every other term
has a power of x greater than 0. Distributing the product gives us terms
of the form ai bj xi+j , which means cn = a0 bn + a1 bn−1 + · · · + ai bn−i +
· · · + an b 0 .
9. The coefficient of xn is
n n n  
! 
X ai bn−i X ai bn−i n! X n
= = ai bn−i .
i=0
i!(n − i)! i=0
n! i!(n − i)! i=0
i
P∞ n
10. Write E(x) = n=2 Dn xn! , the EGF. (We could take the lower bound
as n = 0, 1 or 2, since D0 = D1 = 0.) Substituting,
∞ ∞
X xn X xn
E(x) = nDn−1 + (−1)n .
n=2
n! n=2 n!

Now
∞ ∞ ∞
X xn X xn X xn+1
nDn−1 = Dn−1 = Dn = xE(x)
n=2
n! n=2
(n − 1)! n=1 n!

and
∞ n ∞
X
nx
X (−x)n
(−1) = = e−x − (1 − x)
n=2
n! n=2
n!
(it is the expansion of e−x with the first two terms omitted). So
e−x
E(x) = xE(x) + e−x + x − 1, E(x) = − 1.
1−x

Chapter 7
n o
2k+1
6. Let n = 2k + 1; then we know 2 = 22k − 1 = 4k − 1k . Now, the
polynomial xk − y knis divisible
o by x − y; in this case, using x = 4 and
y = 1, we see that 2k+1
2 is divisible by 4 − 1 = 3 as required.
392 Introduction to Combinatorics

7. Separate the n + 1st element. If it is alone in its partition, then there

are Bn ways to partition the remaining n elements. If there is one

n
other element in its partition, we may choose that element in n = n−1
ways, and there are Bn−1 ways to partition the remaining elements. In

n
general, we may choose i “companions” for the n + 1st element in n−i
ways, and the remaining n− i elements can be partitioned in Bn−i ways.
Summing over i gives the result.

Chapter 8
2. Say a, b ∈ G and x, y ∈ H. Then ab = ba and xy = yx. So

(a, x)(b, y) = (ab, xy) = (ba, yx) = (b, y)(a, x).

3. K contains the identity e; K also contains α, β, and ρ2 . Since α ◦ α =

β ◦ β = ρ2 ◦ ρ2 , K clearly contains inverses. Now we observe that
α ◦ β = β ◦ α = ρ2 , and α ◦ ρ2 = ρ2 ◦ α = β, and β ◦ ρ2 = ρ2 ◦ β = α, we
see that the set is closed under composition. This establishes that K is
a subgroup of D4 .
4. If G is transitive, then the required permutations σi exist by definition.
Given the permutations σi , we wish to show that for any i, j ∈ S there
exists σij where σij (i) = j. The permutation σj ◦ σi−1 will work.
8. We may modify Figure 8.1 by placing a 5 in the center. Now the rotation
ρ = (1234)(5) gives us z11 z41 , and the powers of ρ are modified similarly;
so ρ3 = (1432)(5) gives z11 z41 , ρ2 gives z22 z11 , and e gives z15 . With k
colors, there are thus 41 (k 5 + k 3 + 2k 2 ) non-equivalent colorings. If we
do not color the center spot, each permutation “loses” a z11 , with the
result that we divide the colorings by k; the number is 14 (k 4 + k 2 + 2k).
9. We easily see that ck = 12 (k 3 + k 2 ), since there are k 2 colorings fixed
by the 180◦ rotation. We use the generating functions for the sequences
{k 2 } and {k 3 } to find C(x) = (2x2 + x)(1 − x)−4 . Expanding the
denominator, we arrive at 1 − 4x + 6x2 − 4x3 + x4 . Using Theorem 6.6,
we get the recurrence relation ck = 4ck−1 − 6ck−2 + 4ck−3 − ck−4 . We
require the initial values c0 = 0, c1 = 1, c2 = 6, c3 = 18, and c4 = 40.
You should check that c5 = 75, c6 = 126, and perhaps c7 = 196 will
follow from the recurrence.
11. Each permutation consists of exactly two cycles, except of course the
identity. As a result, the cycle index will consist of monomials z112 , z1 z3 ,
 Solutions to Problems 393

and z22 . We have then C(z1 , z2 , z3 , z4 ) = (z14 + 8z1 z3 + 3z22 )/12 for the
cycle index. Substituting k for each of the zi s gives us (k 4 + 11k 2 )/12.

Chapter 9
2. Take the poset X with at least three different elements x, y, and z. If
any two of these are incomparable, we cannot have the three mutually
adjacent elements required to form a K3 , so without loss of generality
assume x ≻ y. (If not, rename x and y.) Similarly, assume x ≻ z
(renaming if necessary). Now, either z ≺ y or y ≺ z. If z ≺ y, then we
have z ≺ y ≺ x, so that x does not cover z; thus there is no line from x
to z in the Hasse diagram. If y ≺ z, then there is no line from x to y.

3. Suppose n = p1 p2 p3 p4 , and consider the power set of the set {1, 2, 3, 4}.
The isomorphism σ will map the set A of this power set to the divisor
d of n precisely when d is the product of the primes whose subscripts
correspond to the elements of A. Thus, σ({2, 4}) = p2 p4 , and so forth.
Consider two subsets A1 and A2 where σ(A1 ) = d1 and σ(A2 ) = d2 ;
if A1 ⊂ A2 , then each prime factor of d1 is also a factor of d2 , so that
d1 ≺ d2 in Dn . In the same way, if we have any two divisors d1 and d2
of n, where d1 ≺ d2 , then σ −1 (d1 ) ⊂ σ −1 (d2 ) since each prime factor of
d1 is a factor of d2 .

4. Suppose that Bn is the poset of subsets of S = {1, 2, . . . , n}, and Bm is

all subsets of T = {n + 1, n + 2, . . . , n + m}. Now, for A ⊂ S and B ⊂ T
let σ ((A, B)) = A ∪ B. It is clearly one-to-one; a different pair (A′ , B ′ )
will be sent to a different union. Similarly, it is onto since any subset
C of S ∪ T = {1, 2, . . . n + m} is clearly the union of C ∩ S and C ∩ T .
Finally, if A ⊆ A′ and B ⊆ B ′ , then A ∪ B ⊆ A′ ∪ B ′ , so σ preserves the
order.

7. We can find antichains of five elements, but not of six. We see that
{a, b, d, e, f } (the antichain of all maximal elements) works; as there are
only four minimal elements, the antichain of minimal elements is not a
largest-sized antichain. But we can take {a, b, d, j, k} and {a, b, h, j, k}.
394 Introduction to Combinatorics

Chapter 10
1. Reducing the degree sequence, we get

(5, 5, 5, 3, 2, 2, 1, 1) ⇒ (4, 4, 2, 1, 1, 1, 1)
⇒ (3, 1, 0, 0, 1, 1) = (3, 1, 1, 1, 0, 0)
⇒ (0, 0, 0, 0, 0)

which is graphical (the empty graph). However, if the two vertices of

degree 1 are adjacent, the remaining vertices form a component with
degree sequence (5, 5, 5, 3, 2, 2), and

(5, 5, 5, 3, 2, 2) ⇒ (4, 4, 2, 1, 1) ⇒ (3, 1, 0, 0)

which is not valid (one vertex has degree 3, but there is only one other
non-isolate).
2. ⌈ 12 vδ(G)⌉
3. (i) Here are P4 and C5 and their complements. The complements have
been relabeled to exhibit the isomorphism.
A A
E B C D
B
A B C D B D A C
D C E B
P4 P4 C5 C5
(ii) Suppose vertex x is an isolate. In the complement, x is adjacent
to every other vertex, so there are no isolates.
(iii) The degree sequence consists of five numbers, between 1 and 3
inclusive, and the number of 3s equals the number of 1s. The
possibilities are (3, 3, 2, 1, 1), (3, 2, 2, 2, 1) and (2, 2, 2, 2, 2). A short
exhaustive search shows there is exactly one graph with the first
and third sequences, namely G (below) and C5 ; this uniqueness
means that they must be self-complementary. The other sequence
is realized by two non-isomorphic graphs, which are complements
of each other (H and H, see below). So G and C5 are the only
solutions.

G H H
 Solutions to Problems 395

(iv) Say G has v vertices. Then G ∪ G = Kv , which has v(v − 1)/2

edges. If G is self-complementary, then G and G each have |E(G)|
edges. As they are edge-disjoint, their union has 2|E(G)| edges. So
v(v − 1)/2 is even, so v ≡ 0 or 1 mod 4.
4. Start with a cycle Cv ; draw an edge from vertex i to v/2 + i, for 1 ≤
i ≤ v/2.
5. No.
6. Denote the vertex-set of the graph by V . Write xi for the vertex with
degree di , Ni for the set of neighbors of xi , and Vi = {x1 , x2 , . . . , xi }.
Then Ni ∩ Vr is the set of neighbors of xi that lie in Vr , and |Ni ∩ Vr | is
the number of edges joining members of Vr to xi . Summing over i, we
get
X r Xv r
X Xv
di = |Ni ∩ Vr | = |Ni ∩ Vr | + |Ni ∩ Vr |.
i=1 i=1 i=1 i=r+1

If 1 ≤ i ≤ r, then xi is adjacent to at most the other r − 1 members of

Vr , so |Ni ∩ Vr | ≤ r − 1, and
r
X r
X
|Ni ∩ Vr | ≤ r − 1 = r(r − 1).
i=1 i=1

If i > r, then the number of neighbors of vi that lie in Vr certainly

cannot exceed either r or di , so |Ni ∩ Vr | ≤ min(r, di ), and
v
X v
X
|Ni ∩ Vr | ≤ min(r, di ).
i=r+1 i=r+1

Plugging these two results into the equation we get the result.
7.

8. D(G) is the largest distance between any two vertices in G; say D(x, y) =
D(G). R(G) is the smallest value of ε; let z be a vertex for which
ε(z) = R(G). As R is a distance between some two vertices, R ≤ D.
But by definition D(z, t) ≤ ε(z) = R(G) for every vertex t, so D(G) =
D(x, y) ≤ D(x, z) + D(z, y) ≤ 2R(G).
396 Introduction to Combinatorics

9. n = 1 and 2 only (otherwise, Kn contains an odd cycle).

Chapter 11
1. You would treat a loop as adding 2 to the degree of its vertex. They do
not need special treatment in Euler walks—you can traverse the loop
when you pass through the vertex. As roads, a loop might represent a
crescent. Or a snowplow might need to clear a large circular driveway
on city property.
3. Examples for the first two parts:

(i) (ii)

4. The cost is n(n + 1).

5. Suppose the result is false. Select a graph G on 2n vertices, each of
degree at least n, that is not Hamiltonian, but has the maximum pos-
sible number of edges among such graphs. Select two vertices, x and y,
that are not adjacent. (This must be possible: the complete graph is
Hamiltonian, so G is not complete.) Add edge xy to G; the resulting
graph G + xy must be Hamiltonian. Now select a Hamilton cycle in
G + xy. It must contain xy, or else it would be a Hamilton cycle in G.
Say it is C = (x, x1 , x2 , . . . , x2n−2 , y).
Suppose xi is adjacent to x. Then xi−1 cannot be adjacent to y, or else
(x, x1 , x2 , . . . , xi−1 , y, x2n−2 , x2n−3 , . . . , xi ) would be Hamiltonian in G.
So each of the n or more vertices adjacent in G to x must be preceded
in C by a vertex not adjacent to y. Write S for the set of those vertices.
There are at least n vertices in S, and none is adjacent to y. But there
are also n vertices adjacent to y. This accounts for all 2n vertices.
But what about y itself? It must be a member of S; but that is impos-
sible, because y cannot appear in the list x1 , x2 , . . . , x2n−2 . So we have
a contradiction.
6. (i) The left-hand diagram shows a Hamilton cycle in K1,2,3 . To handle
Kn,2n,3n , treat it as n copies of K1,2,3 together with further edges; the
solution is shown on the right (with unnecessary edges not shown, to
avoid clutter).
 Solutions to Problems 397

Alternatively, write A, B, C for the sets of size n, 2n, 3n, respectively.

Label the vertices of A with integers 1, 2, . . . , n, those of B with integers
n + 1, n + 2, . . . , 3n, and those of C with integers 3n + 1, 3n + 2, . . . , 6n.
One Hamilton cycle is (1, 3n + 1, 2, 3n + 2, . . . , 3n, 6n).
(ii) Write C for the set of 3n + 1 vertices. If the graph is to have a
Hamilton cycle, no two members of C can be adjacent in the cycle, so
every second vertex must be a non-member of C, so there must be at
least 3n + 1 further vertices.
7. First, observe that any tree with three or more vertices has an edge, and
at least one endpoint of that edge is joined to another vertex. So any
tree with 3 or more vertices contains a path with three vertices.
Now suppose T is a tree on three or more vertices, and suppose there
is a graph G such that T = G2 . Select a path abc in T . If both ab
and bc are edges of G, then T will contain the 3-cycle (abc), which is
impossible. So let’s assume ab is not an edge of G. There must be some
vertex x that is adjacent to both a and b in G. Then ax and bx are
edges of T , and again T contains a cycle, in this case (abx).
8. Consider a tree T of radius R. Any two vertices of a tree are joined by a
path; write Px,y for the path joining x to y. Say a and b are both vertices
in the center of T . If they are not adjacent, there must be another vertex
c that lies in Pa,b , and a vertex p such that d(c, p) ≥ R. Then Pc,p must
pass through a (or else d(a, p) > d(c, p) ≥ R, so d(a, p) > ǫ(a)), and
similarly Pc,p must pass through b. But this is impossible if c is on Pab .
So all central vertices in a tree are adjacent. A tree cannot have three or
more mutually adjacent vertices, so the center has one or two vertices,
and if 2 then they are adjacent.
9. Suppose c is the vertex of degree 4 and x1 , x2 , x3 , x4 are its neighbors.
We can think of the tree as consisting of four subtrees T1 , T2 , T3 , T4 ,
with common vertex c: the vertices of Ti are all the vertices that can be
reached from xi without passing through c. The tree looks like

where the shaded sections denote the Ti . Then each Ti is a tree. Ti may
just be the vertex xi , in which case xi is a vertex of degree 1 in T . If
Ti is not a single vertex, its vertices of degree 1 (other than xi ) will all
398 Introduction to Combinatorics

be vertices of degree 1 in T . Each Ti has at least two vertices of degree

1, unless it is the 1-vertex tree xi ; so each Ti contributes at least one
vertex of degree 1 to T . So none can contribute more than one such
vertex. Therefore each Ti is either a 1-vertex tree or a path. So T has
one vertex of degree 4 and four of degree 1, and all others have degree
2. Therefore T1 ∪ T2 and T3 ∪ T4 are paths, they are edge-disjoint, and
their union is T .
10. We proceed by induction. The result is trivial if v = 1 or 2 (there is
one tree in each case: K1 and K2 , respectively). AssumePn >≥ 3, and
assume the result is true for v < n. Say v = n. Since di < 2n, at
least one degree is 1. Let’s assume dn = 1.
Write Sj for the set of all trees with vertices x1 , x2 , . . . , xv−1 where xi
has degree d(xi ) = di when i 6= j and d(xj ) = dj − 1. The members
of Sj correspond to trees in T with xj xn as an edge; if dj = 0 then
Sj is empty. Since xn must have been connected to every tree in T ,
Pn−1
|T | = j=1 |Sj |. If dj > 1 then (by induction)
(n − 3)!
|Sj | =
(d1 − 1)!(d2 − 1)! . . . (dj − 2)! . . . (dn−1 − 1)!
(n − 3)!(dj − 1)
= ;
(d1 − 1)!(d2 − 1)! . . . (dj − 1)! . . . (dn−1 − 1)!

and if dj = 1 then |Sj | = 0 which fits the same formula. So

n−1
X
|T | = |Sj |
j=1
n−1
X (n − 3)!(dj − 1)
=
j=1
(d1 − 1)!(d2 − 1)! . . . (dn−1 − 1)!
n−1
(n − 3)! X
= (dj − 1).
(d1 − 1)!(d2 − 1)! . . . (dn−1 − 1)! j=1
Pn Pn−1
But dn = 1 (so (dn − 1)! = 1) and j=1 dj = 2n − 2, so j=1 (dj − 1) =
n − 2, and
(n − 3)!
|T | = (n − 2)
(d1 − 1)!(d2 − 1)! . . . (dn − 1)!
(n − 2)!
= .
(d1 − 1)!(d2 − 1)! . . . (dn − 1)!

11. If a tree has v vertices, the sum of the degrees is 2v−2. So Problem 11.10
told us the number of trees on v vertices with a given set of degrees. To
find the total number of trees, we sum over all possible degree sets. So
the sum is
 Solutions to Problems 399

X (v − 2)!
(d1 − 1)!(d2 − 1)! . . . (dv − 1)!
d1 +d2 ...+dv =2v−2

and writing ki = di − 1, this becomes

X (v − 2)!
k1 !k2 ! . . . kv !
k1 +k2 ...+kv =2v−2

which equals v v−2 by the multinomial theorem (Theorem 2.6).

Chapter 12
1. The codewords correspond to the vertices of the subgraph, but there
is no obvious relationship to the edges. However, the distance between
two words equals the distance between the corresponding vertices in the
underlying hypercube.

3. Minimum distance is 4. So the code can detect three errors and always
correct one.

4. One method is to encode A, B, C, D, E as 0, 10, 110, 1110, 1111 respec-

tively.

5. If a code has minimum distance 2n or smaller, one can select two words
A and B that differ in at most 2n positions. Say S is the set of positions
where they differ. Consider a received word that is the same as A (and
B) is those positions where the two are the same, is the same as A in n
of the positions where they differ, and is the same as B in the remaining
positions. There is no way to know whether to decode this word to A
or to B. (Note: If the distance is smaller than 2n, it is tempting to
say, “decode to the codeword that gives fewer errors,” but there is no
guarantee that this will be correct.)
400 Introduction to Combinatorics

Chapter 13
2. One example is
1 4 6 5 3 2
4 2 5 6 1 3
2 5 3 1 6 4
6 3 1 4 2 5
3 6 4 2 5 1
5 1 2 3 4 6

5. One solution is
1 2 3
4

The generalization is obvious.

1 1 1 2 2 2 3 3 3
1 2 3 1 2 3 1 2 3
6. (i)
1 2 3 3 1 2 2 3 1
1 3 2 3 2 1 2 1 3
(ii) No. You would need orthogonal Latin squares of order 6.
(iii) Use 3 MOLS of order 4 to construct a 5 × 16 orthogonal array, and
use the columns as the codewords. Let’s say two codewords are properly
separated if they have exactly one entry in common.
Suppose the entries in position (i, j) in the three squares are a, b, c,
respectively. So one codeword is (i, j, a, b, c). There will be three other
columns corresponding to cells in row i, and none will have a j in position
2 (a j there would mean the corresponding entry was also in the (i, j)
position, and there is only one entry per cell), or an a in position 3 (that
would mean two entries a in row i of the first square). Similarly, none
of them has a b in position 4, or a c in position 5. So our codeword
is properly separated from the three other words with i in position 1.
In the same way we see that our codeword is properly separated from
the three other words with j in position 2, the three other words with a
in position 3, the three other words with b in position 4, and the three
other words with c in position 5. Moreover our reasoning shows that all
15 of those codewords are different. Since there are only 16 codewords,
we have shown that (i, j, a, b, c) is properly separated from all the other
codewords.
But (i, j, a, b, c) is completely general. So every pair of codewords is
properly separated.
 Solutions to Problems 401

7. (i) The (1, 4) entry must be 4 (in order to complete column 4). So the
first row is 12345. So the (5, 4) entry must be 3. Now consider the
completion of row 5; a 2 and a 5 are needed and the 5 cannot be in
column 5, so row 4 is 34512. Similarly row 3 must be 45123. The array
looks like
1 2 3 4 5
1 3
4 5 1 2 3
3 4 5 1 2
4 5
The 5 in row 2 must be in column 1, and to complete column 3 the (2, 3)
entry must be 2. So row 2 is 51234. Now the array can be completed:

1 2 3 4 5
5 1 2 3 4
4 5 1 2 3
3 4 5 1 2
2 3 4 5 1

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
5 1 2 3 4 5 1 4 3 2 4 1 5 3 2
(ii) 4 5 1 2 3 3 5 1 2 4 3 5 1 2 4
3 4 5 1 2 2 4 5 1 3 5 4 2 1 3
2 3 4 5 1 4 3 2 5 1 2 3 4 5 1

Chapter 14
1. (i) From the hint, there are blocks 1234 and 1256. There is one further
block containing 13. Since the intersection size is 2, 4 is not included,
and not both of 5 and 6 can occur, so there must be a final 7 (no other
symbols are available). The block is either 1357 or 1367. We’ll choose
1357 (otherwise, use the isomorphism (56)). The fourth block containing
1 can only be 1467.
There are two further blocks containing 2. Since 2 and 7 have not
occurred together, 7 must be in both blocks. One must contain 3 and
the other contain 4 (234 cannot occur together). We either have 2357
and 2367. But 2357 is impossible (357 has already been used) so 2367
is block, and so is 2457 (in order for 2 to occur twice with every other
symbol). The final block is 3456. So we have blocks

1234 1256 1357 1467 2367 2457 3456.

402 Introduction to Combinatorics

Any other choice is isomorphic to this (make a different choice of the

first block, use (56) or (67), or some combination).
2. Yes: one example is
123 145 167 246 157 347 356
124 235 346 457 156 267 137
3. (i) v = 9t + 8.
(ii) If v is even (that is, t is even) then k−λ = 18+27t is a perfect square.
So 2 + 3t is a perfect square. But all perfect squares are congruent to 0
or 1 (mod 3). So assume t is odd. There is a solution to
x2 + (2 + 3t)(3y)2 = ±3z 2
with x, y, z not all zero. Without loss of generality we can assume x, y
and z have no common factor (if there were a common factor t, we
could divide all three variables by t). So, modulo 3, x2 + 18y 2 ≡ 0. So
3 divides x. So the largest power of 3 that divides x2 + (2 + 3t)(3y)2
must be even. But the largest power of 3 that divides 3z 2 is odd.
4. The dual does not satisfy Fisher’s inequality.
v, kv , v−1 v−2




5. k−1 , k, k−2

6. Write B, B1 , B2 , . . . , Bv−1 for the blocks of D. The blocks of E are

S\B, S\B1 , S\B2 , . . . , S\Bv−1 ,
and the blocks in the residual of E with regard to S\B are the v − 1
blocks
(S\B1 )\(S\B), (S\B2 )\(S\B), . . . , (S\Bv−1 )\(S\B).
On the other hand, the blocks of the derived design of D with respect
to B are
B\B1 , B\B2 , . . . , B\Bv−1 .
Now
(S\Bi )\(S\B) = B\Bi
(check with a Venn diagram if necessary). So we have the result.
7. Write C0 , C1 , . . . , C14 for the blocks of D∗ .
(i) C0 = {0, 1, 2, 3, 4, 5, 6} C1 = {0, 2, 4, 7, 8, 10, 12}
C2 = {0, 2, 4, 9, 11, 13, 14} C3 = {0, 1, 6, 7, 8, 9, 14}
C4 = {0, 1, 6, 10, 11, 12, 13} C5 = {0, 3, 5, 7, 10, 11, 14}
C6 = {0, 3, 5, 8, 9, 12, 13} C7 = {2, 3, 6, 7, 8, 11, 13}
C8 = {2, 3, 6, 9, 10, 12, 14} C9 = {1, 2, 5, 7, 9, 11, 12}
C10 = {1, 2, 5, 8, 10, 13, 14} C11 = {4, 5, 6, 7, 9, 10, 13}
C12 = {4, 5, 6, 8, 11, 12, 14} C13 = {1, 3, 4, 7, 12, 13, 14}
C14 = {1, 3, 4, 8, 9, 10, 11};
 Solutions to Problems 403

(ii) D has precisely seven triplets: (012), (034), (056), (078), (0910),
(01112), (0134). Da st also has seven triplets: (016), (024), (035), (125),
(134), (236), (456). Any isomorphism would have to carry one set of
triples into the other; this is clearly impossible.

8. (i) (n2 + n + 1, n2 , n2 − n); (ii) (n + 1, n2 + n, n2 , n, n2 − n); (iii) an

(n + 1, n + 1, n, n, n − 1) design exists (see Exercise 14A.2). Take the
n-multiple of it (see Exercise 14A.9).

9. (i) Say one line has k points. From the axioms, every other line contains
precisely one of these, so there are k + 1 lines in total. If some other line
had h points, there would be h + 1 lines in total. So h = k.
(ii) There are k points per line, and k + 1 lines; each point is on two
lines, so there are v = 21 k(k + 1) points in all. Call them x1 , x2 , . . . , xv .
Say the first line is
x1 , x2 , x3 , . . . , xk ;

the second line has one point in common with this, so (up to permutation
of names of points) it is

x1 , xk+1 , xk+2 , . . . , x2k−1 .

Proceeding in this way we find that the unique (up to isomorphism)

solution for given k is the following set of lines:

x1 x2 x3 ... xk
x1 xk+1 xk+2 ... x2k−1
x2 xk+1 x2k ... x3k−2
x3 xk+2 x2k ... x4k−3
... ...
xk x2k−1 x3k−2 ... xv .

For example, with k = 5, the solution is

1 2 3 4 5
1 6 7 8 9
2 6 10 11 12
3 7 10 13 14
4 8 11 13 15
5 9 12 14 15

10. You would have λ(2k − 1) = k(k − 1), so λ = k(k − 1)/(2k − 1). Now
2 2k−1
clearly 2k > 2k − 1 > 2k − 2, so k−1 > k(k−1) > k2 , and k−1 2k−1
2 < k(k−1) <
k k−1 k
2 . So λ would lie between 2 and 2 , and there is no integer in that
range.
404 Introduction to Combinatorics

11. Say m and v are coprime. Then the integers m, 2m, . . . , (v − 1)m are all
different modulo v, and are congruent to 1, 2, . . . , v − 1 in some order.
Moreover, of the pairs with difference d (mod v) are {x1 , y1 }, {x2 , y2 },
. . . , {xλ , yλ }, then there are precisely λ pairs in mD with difference md,
namely, {mx1 , my1 }, {mx2 , my2 }, . . . , {mxλ , myλ }. On the other hand,
if m and v are not coprime, there may be x and y in D such that mx
and my are congruent mod v, or mx could be zero.

12. (i) Call the competitors {1, 2, . . . , 7, A, B, . . . , G}. Using a symmetric

BIBD with parameters (7, 3, 1), one could try for a solution whose first
seven heats are the columns of

1 2 3 4 5 6 7
2 3 4 5 6 7 1
4 5 6 7 1 2 3
A B C D E F G

and the other seven heats might be

A B C D E F G
B C D E F G A
.
D E F G A B C
∗ ∗ ∗ ∗ ∗ ∗ ∗

The fourth member of heat 8 cannot be one of {1, 2, 4} (pairs 1A, 2A

and 4A were in heat 1), {2, 3, 5} or {4, 5, 7}. The unique answer is 6. If
the fourth members of the subsequent heats are generated by adding 1
(mod 7) we obtain the solution

1 2 3 4 5 6 7 A B C D E F G
2 3 4 5 6 7 1 B C D E F G A
.
4 5 6 7 1 2 3 D E F G A B C
A B C D E F G 6 7 1 2 3 4 5

There may well be other solutions.

(ii) There is only one way to pair up the heats in our solution to satisfy
the conditions of this part: heats 1 and 9 are paired; then 2,10; 3,11;
. . . ; 6,14; 7,8.
 Solutions to Problems 405

Chapter 15
 
0 1 0
1. (i): A =  0 0 1 .
−2 1 2
 
1 1 1
(iii): P =  1 −1 2 .
1 1 4

2. The induction step: Assume An−1 (i, j) is the number of ways to walk
from i to j in n − 1 steps for any i and j. Then let i, j be given and
determine An (i, j). For each neighbor k of vertex j, An−1 (i, k) is the
number of ways to walk from i to k in n − 1 steps. The (i, j) entry
of An , by the definition of matrix multiplication, is the sum over all k
adjacent to j of An−1 (i, k); this adds up all walks of length n − 1 from
i to a neighbor of j, which correspond to walks of length n from i to j.
5. This is true if the matrix has only non-negative entries (as an adjacency
 
1 1
matrix does). However, if we allow negative entries, the matrix
−1 1
is a counterexample.
√
6. The approximation is (2/e)n 2πn which goes to 0 as n → ∞.
406
References

[1] H. L. Abbott, Lower bounds for some Ramsey numbers, Discrete Math.
2 (1972), 289–293.
[2] W. S. Anglin and J. Lambek, The Heritage of Thales, 5th ed. (Springer-
Verlag, New York, 1995).
[3] H. Anton, Elementary Linear Algebra, 9th ed. (Wiley, New York, 2005).
[4] K. Appel and W. Haken, (October 1977), Solution of the Four Color
Map Problem, Scientific American 237 (1977), 108–121.
[5] M. E. Baron, A note on the historical development of logic diagrams,
Math. Gaz. 53 (1969), 113–125.
[6] E. T. Bell, Men of Mathematics, 5th ed. (Simon & Schuster, New York,
1937).
[7] M. Bellis, Timeline: Biography of Samuel Morse 1791–1872,
http://inventors.about.com/od/mstartinventors/a/samuel morse.htm

[8] J. Bernoulli, Ars Conjectandi, Basileae, Impensis Thurnisiorum,

Fratrum, 1713.
[9] K. N. Bhattacharya, A new balanced incomplete block design, Science
and Culture 11 (1944), 508.
[10] N. L. Biggs,T P Kirkman, Mathematician, Bull. London Math. Soc. 13
(1981), 97–120.
[11] R. C. Bose, S. S. Shrikhande, and E. T. Parker, Further results on the
construction of mutually orthogonal Latin squares and the falsity of
Euler’s conjecture, Canad. J. Math. 12 (1960), 189–203.
[12] R. L. Brooks, On colouring the nodes of a network, Proc. Camb. Phil.
Soc. 37 (1941), 194–197.
[13] R. A. Brualdi, Introductory Combinatorics, 4th ed. (Pearson Prentice
Hall, New Jersey, 2004).
[14] R. A. Brualdi and H. J. Ryser, Combinatorial Matrix Theory (Encyclo-
pedia of Mathematics and its Applications #39) (Cambridge UP, New
York, 1991).

407
408 Introduction to Combinatorics

[15] R. H. Bruck and H. J. Ryser, The non-existence of certain finite pro-

jective planes, Canad. J. Math. 1 (1949), 88–93.
[16] D. M. Burton, The History of Mathematics: An Introduction, 5th ed.
(McGraw-Hill, Boston, 2003).
[17] S. Butler, M. T. Hajiaghayi, R. D. Kleinberg, and T. Leighton, Hat
guessing games, SIAM J. Discrete Math. 22 (2008), 592–605.
[18] A. Cayley, On the colourings of maps, Proc. Roy. Geog. Soc. 1 (1879),
259–261.
[19] A. Cayley, A theorem on trees, Quart. J. Math. 23 (1889), 376–378.
[20] Y. Chang, The existence spectrum of golf designs, J. Combin. Des. 15
(2005), 84–89.
[21] S. Chowla and H. J. Ryser, Combinatorial problems, Canad. J. Math.
2 (1950), 93–99.
[22] C. J. Colbourn and J. H. Dinitz, Handbook of Combinatorial Designs,
2nd ed. (Chapman & Hall/CRC, Boca Raton, 2007).
[23] C. J. Colbourn and G. Nonay, A golf design of order 11, J. Statist.
Plann. Inf. 58 (1997), 29–31.
[24] R. Cooke, The History of Mathematics: A Brief Course, 2nd ed. (John
Wiley & Sons, Hoboken, NJ, 2005).
[25] L. D. Cummings, An undervalued Kirkman paper, Bull. Amer. Math.
Soc. 24 (1918), 336–339.
[26] H. de Parville, Recreations mathematiques: La tour d’Hanoi et la ques-
tion du Tonkin, La Nature, part I (1884), 285–286.
[27] E. W. Dijkstra, A note on two problems in connexion with graphs,
Numerische Math. 1 (1959), 269–271.
[28] T. Ebert, Applications of Recursive Operators to Randomness and Com-
plexity, Ph.D. Dissertation (University of California, Santa Barbara,
1998).
[29] J. Edmonds and E. L. Johnson, Matching, Euler tours, and the Chinese
postman, Math. Prog. 5 (1973), 88–124.
[30] J. Emert and D. Umbach, Inconsistencies of ‘wild card’ poker, Chance
9 (1996), 17–22.
[31] P. Erdös and T. Gallai, Graphs with prescribed degrees of vertices [Hun-
garian] Mat. Lapok. 11 (1960), 264–274.
[32] L. Euler, Solutio Problematis ad geometriam situs pertinentis, Comm.
Acad. Sci. Imp. Petropolitanae 8 (1736), 128–140.
 References 409

[33] L. Euler, Recherches sur une nouvelle espece de quarrées magiques,

Verhand. Zeeuwsch Gen. Wet. Vlissingen 9 (1782), 85–239.
[34] A. Evans, Embedding incomplete Latin squares, Amer. Math. Monthly
67 (1960), 958–961.
[35] R. A. Fisher, An examination of the different possible solutions of a
problem in incomplete blocks, Ann. Eugenics 10 (1940), 52–75.
[36] J. Folkman, Notes on the Ramsey Number N (3, 3, 3, 3), Manuscript,
Rand Corporation, Santa Monica, CA, 1967.
[37] W. J. Frascella, The construction of a Steiner triple system on sets of
the power of the continuum without the axiom of choice, Notre Dame
J. Formal Logic 7 (1966), 196–202.
[38] S. Gadbois, Poker with wild cards–A paradox? Math. Mag. 69 (1996),
283–285.
[39] M. Gardner, Euler’s spoilers, Martin Gardner’s New Mathematical Di-
versions from Scientific American (Allen & Unwin, London, 1969), 162–
172.
[40] M. Gardner, Time Travel and Other Mathematical Bewilderments
(Freeman, New York, 1988).
[41] M. Gardner, Mathematical Carnival (MAA, Washington DC, 1989).
[42] M. Gardner, Fractal Music, Hypercards and More... (Freeman, New
York, 1992).
[43] M. Gardner, Martin Gardner’s Mathematical Games (MAA, Washing-
ton DC, 2005).
[44] S. Giberson and T. J. Osler, Extending Theon’s ladder to any square
root, College Math. J. 35 (2004), 222.
[45] S. W. Golomb, Problem for solution, Amer. Math. Monthly 60 (1953),
114, 551–552.
[46] G. Gonthier, Formal proof—The four-color theorem, Notices Amer.
Math. Soc. 55 (2008), 1382–1393.
[47] A. W. Goodman, On sets of acquaintances and strangers at any party,
Amer. Math. Monthly 66 (1959), 778–783.
[48] R. Graham, D. Knuth and O. Patashnik, Concrete Mathematics, 2nd
ed. (Addison-Wesley, New York, 1995).
[49] S. Hakimi, On the realizability of a set of integers as degrees of the
vertices of a graph, SIAM J. Appl. Math. 10 (1962), 496–506.
[50] M. Hall, Combinatorial Theory, 2nd ed. (Wiley, New York, 1986).
410 Introduction to Combinatorics

[51] M. Hall and W. S. Connor, An embedding theorem for balanced incom-

plete block designs, Canad. J. Math. 6 (1954), 35–41.
[52] P. Hall, On representations of subsets, J. London Math. Soc. 10 (1935),
26–30.
[53] V. Havel, A remark on the existence of finite graphs [Czech], Ĉasopis
Pest. Mat. 80 (1955), 477–480.
[54] D. M. Johnson, A. L. Dulmage, and N. S. Mendelsohn, Orthomorphisms
of groups and orthogonal Latin squares I, Canad. J. Math. 13 (1961),
356–372.
[55] J. Karamata, Théorèmes sur la sommabilité exponentielle et d’autres
sommabilités rattachant, Mathematica (Cluj) 9 (1935), 164–178.
[56] P. Kerr, Dark Matter: The Private Life of Sir Isaac Newton (Crown
Press, New York, 2002).
[57] T. P. Kirkman, On a problem in combinations, Cambridge and Dublin
Math. J. 2 (1847), 191–204.
[58] T. P. Kirkman, Note on an unanswered prize question, Cambridge and
Dublin Math. J. 5 (1850), 255–262.
[59] T. P. Kirkman, Query VI, Lady’s and Gentleman’s Diary (1850), 48.
[60] T. P. Kirkman, Solution to Query VI, Lady’s and Gentleman’s Diary
(1851), 48.
[61] W. W. Kirkman, Thomas Penyngton Kirkman, Mem. Proc. Manchester
Lit. Phil. Soc. 9 (1895), 238–243.
[62] N. S. Knebelman, Problem for solution, Amer. Math. Monthly 55
(1948), 100; 56 (1949), 426.
[63] D. Knuth, Fundamental Algorithms: The Art of Computer Program-
ming, vol. 1, 2nd ed. (Addison-Wesley, Reading, MA, 1973).
[64] D. Knuth, Two notes on notation, Amer. Math. Monthly 99 (1992),
403–422.
[65] T. Koshy, Fibonacci and Lucas Numbers with Applications (Wiley, New
York, 2001).
[66] T. Koshy, Catalan Numbers with Applications (Oxford University Press,
New York, NY, 2009).
[67] J. B. Kruskal Jnr., On the shortest spanning subtree and the traveling
salesman problem, Proc. Amer. Math. Soc. 7 (1956), 48–50.
[68] P. A. Laplante, Dictionary of Computer Science, Engineering and Tech-
nology (CRC Press, Boca Raton, FL, 2000).
 References 411

[69] L. Lovász and M. D. Plummer, Matching Theory (North-Holland Math-

ematics Studies #121) (Elsevier, New York, 1986).
[70] H. L. MacNeish, Euler squares, Ann. Math. 23 (1922), 221–227.
[71] R. Mandl, Orthogonal Latin squares: An application of experiment
design to compiler testing, Comm. ACM 28 (1985), 1054–1058.
[72] H. B. Mann, The construction of orthogonal Latin squares, Ann. Math.
Statist. 13 (1942), 418–423.
[73] M. McClure and S. Wagon, Four-coloring the US counties, Math Hori-
zons 16(4) (2009), 20–21.
[74] K. Mei-Ko, Graphic programming using odd or even points, Chinese
Math. 1 (1962), 273–277.
[75] S. Milgram, The small world problem, Psychology Today 2 (1967), 60–
67.
[76] L. Moser, Problem for solution, Amer. Math. Monthly 56 (1949) 403;
57 (1950), 117.
[77] L. Moser, Problem for solution, Amer. Math. Monthly 60 (1953) 262;
713–714.
[78] H. W. Norton, The 7 × 7 squares, Ann. Eugenics 9 (1939), 269–307.
[79] J. J. O’Connor and E. F. Robertson, Emanuel Sperner,
http://www-history.mcs.st-andrews.ac.uk/history/Biographies
/Sperner.html

[80] J. J. O’Connor and E. F. Robertson, Gaston Tarry,

http://www-history.mcs.st-andrews.ac.uk/Biographies/Tarry.html

[81] J. J. O’Connor and E. F. Robertson, Thomas Penyngton Kirkman,

http://www-history.mcs.st-andrews.ac.uk/Biographies/Kirkman.html

[82] C. D. Olds, Continued Fractions (New Mathematical Library 9) (MAA,

Washington, DC, 1963).
[83] O. Ore, Note on Hamilton circuits, Amer. Math. Monthly 67 (1960),
55.
[84] H. Perfect, The Revd Thomas Penyngton Kirkman FRS (1806–1895):
Schoolgirl parades—but much more, Math. Spectrum 28 (1995/96), 1–6.
[85] R. C. Prim, Shortest connection networks and some generalizations,
Bell Syst. Tech. J. 36 (1957), 1389–1401.
[86] M. Reiss, Über eine Steinersche kombinatorische Aufgabe, J. Reine
AngṀath. 56 (1859), 326–344.
412 Introduction to Combinatorics

[87] V. F. Rickey, Isaac Newton: Man, myth, and mathematics, College

Math. J. 18 (1987), 362–389.
[88] N. Robertson, D. Sander, P. Seymour, and R. Thomas, The four-colour
theorem, J. Combin. Theory 70B (1997), 2–44.
[89] D. F. Robinson, Constructing an annual round-robin tournament played
on neutral grounds, Math. Chronicle 10 (1981), 73–82.
[90] S. Robinson, Why mathematicians now care about their hat color, New
York Times (April 10, 2001), D5.
[91] S. Roman, Introduction to Coding and Information Theory (Springer-
Verlag, New York, 1996).
[92] W. W. Rouse Ball, Mathematical Recreations and Essays (MacMillan,
London, 1892).
[93] DS5: F. Ruskey and M. Weston, Venn Diagrams (Dynamic Survey 5),
Elec. J. Combin. (2005).
[94] H. J. Ryser, Combinatorial Mathematics (Mathematical Monograph
#14) (MAA, Washington DC, 1963).
[95] A. Sade, An omission in Norton’s list of 7 × 7 squares, Ann. Math.
Statist. 22 (1951), 306–307.
[96] S. S. Sane and W. D. Wallis, Monochromatic triangles in three colours,
J. Austral. Math. Soc. 37B (1988), 197–212.
[97] P. J. Schellenberg, G. H. J. van Rees, and S. A. Vanstone, Four pairwise
orthogonal Latin squares of order 15, Ars Combin. 6 (1978), 141–150.
[98] A. Schrijver, On the history of combinatorial optimization (till 1960),
in Handbook of Discrete Optimization (K. Aardal, G. L. Nemhauser, R.
Weismantel, eds.) (Elsevier, Amsterdam, 2005), 1–68.
[99] M. P. Schutzenberger, A non-existence theorem for an infinite family of
symmetrical block designs, Ann. Eugenics 14 (1949), 286–287.
[100] A. J. Schwenk, Acquaintance graph party problem, Amer. Math.
Monthly 79 (1972), 1113–1117.
[101] A. Seidenberg, A simple proof of a theorem of Erdös and Szekeres, J.
Lond. Math. Soc. 34 (1959), 352.
[102] L. E. Sigler, Fibonacci’s Liber Abaci: Leonardo Pisano’s Book of Cal-
culation (Springer-Verlag, New York, 2002).
[103] N. M. Singhi and S. S. Shrikhande, Embedding of quasi-residual designs
with λ = 3, Utilitas Math. 4 (1973), 35–53.
[104] N. M. Singhi and S. S. Shrikhande, Embedding of quasi-residual designs,
Geom. Ded. 2 (1974), 509–517.
 References 413

[105] B. Smetaniuk, A new construction on Latin squares I: A proof of the

Evans conjecture, Ars Combin. 11 (1960), 155–172.
[106] M. Z. Spivey, Fibonacci identities via the determinant sum property,
College Math. J. 37 (2006), 286.
[107] R. P. Stanley, Enumerative Combinatorics, Volume 1 (Cambridge Stud-
ies in Advanced Mathematics #49)(Cambridge University Press, Cam-
bridge UK, 1997).
[108] R. P. Stanley, Enumerative Combinatorics, Volume 2 (Cambridge Stud-
ies in Advanced Mathematics #62)(Cambridge University Press, Cam-
bridge UK, 2001).
[109] R. P. Stanley, Catalan Addendum
http://www-math.mit.edu/~rstan/ec/catadd.pdf

[110] J. Steiner, Combinatorische Aufgabe, J. Reine Angew. Math. 45 (1853),

181–182.
[111] J. Stewart, DijkstraApplet.html,
http://www.dgp.toronto.edu/ people/JamesStewart/270/9798s/
Laffra/DijkstraApplet.html

[112] G. Tarry, Géométrie de situation: Nombre de manieres distinctes de

parcourir en une seule course toutes les all’ees d’un labyrinthe rentrant,
en ne passant qu’une seule fois par chacune des all’ees, Comptes Rend.
Assoc. Fr. Avance. Sci. 15 (1886), 49–53.
[113] G. Tarry, La problem des labyrinths, Nouv. Ann. Math. 14 (1895), 187–
190.
[114] G. Tarry, Le problème des 36 officiers, Comptes Rend. Assoc. Fr. 1
(1900), 122–123; 2 (1901), 170–203.
[115] L. Teirlinck, On the use of pairwise balanced designs and closure spaces
in the construction of structures of degree at least 3, Le Matematiche
65 (1990), 197–218.
[116] L. Teirlinck, Large sets of disjoint designs and related structures, Con-
temporary Design Theory: A Collection of Surveys (J. H. Dinitz and
D. R. Stinson, editors), (Wiley, 1992), 561–592.
[117] J. Travers and S. Milgram, An experimental study of the small world
problem, Sociometry 32 (1969), 425–443.
[118] J. Venn, On the diagrammatic and mechanical representation of propo-
sitions and reasonings, Dublin Phil. Mag. J. Sci. 9 (1880), 1–18.
[119] J. Venn, On the employment of geometrical diagrams for the sensible
representation of logical propositions, Proc. Camb. Phil. Soc. 4 (1880),
47–59.
414 Introduction to Combinatorics

[120] W. D. Wallis, The number of monochromatic triangles in edge-

colourings of a complete graph, J. Comb. Inf. Syst. Sci. 1 (1976), 17–20.
[121] W. D. Wallis, The problem of the hospitable golfers, Ars Combin. 15
(1983), 149–152.
[122] W. D. Wallis, A Beginner’s Guide to Graph Theory, 2nd ed.
(Birkhäuser, Boston, 2007).
[123] W. D. Wallis, Introduction to Combinatorial Designs, 2nd ed. (Chap-
man & Hall/CRC, Boca Raton, 2007).
[124] I. M. Wanless and B. S. Webb, The existence of Latin squares without
orthogonal mates, Des. Codes. Crypt. 40 (2006), 131–135.
[125] E. G. Whitehead Jr., Algebraic structure of chromatic graphs associated
with the Ramsey number N (3, 3, 3; 2), Discrete Math. 1 (1971), 113–
114.
[126] Wikipedia, Blaise Pascal,
http://en.wikipedia.org/wiki/Blaise Pascal
[127] Wikipedia, Carl Friedrich Gauss,
http://en.wikipedia.org/wiki/Carl Friedrich Gauss
[128] Wikipedia, Édouard Lucas,
http://en.wikipedia.org/wiki/Edouard Lucas
[129] Wikipedia, Évariste Galois,
http://en.wikipedia.org/wiki/Evariste Galois
[130] Wikipedia, Howard Garns,
http://en.wikipedia.org/wiki/Howard Garns
[131] Wikipedia, Jacob Bernoulli,
http://en.wikipedia.org/wiki/Jacob Bernoulli
[132] Wikipedia, Joseph Kruskal,
http://en.wikipedia.org/wiki/Joseph Kruskal
[133] Wikipedia, Leonhard Euler,”
http://en.wikipedia.org/wiki/Leonhard Euler
[134] Wikipedia, Richard Hamming,
http://en.wikipedia.org/wiki/Richard Hamming
[135] Wikipedia, Isaac Newton,
http://en.wikipedia.org/wiki/Isaac Newton
[136] Wikipedia, Small World Experiment,
http://en.wikipedia.org/wiki/Small world experiment
[137] Wikipedia, Sudoku,
http://en.wikipedia.org/wiki/Sudoku
 References 415

[138] Wikipedia, William Rowan Hamilton,

http://en.wikipedia.org/wiki/William Rowan Hamilton
[139] H. Wilf, Generatingfunctionology, 3rd ed. (A. K. Peters, Wellesley,
2006).
[140] W. S. B. Woolhouse, Prize question 1733, Lady’s and Gentleman’s Diary
(1844), 84.
[141] B. Y. Wu and K.-M. Chao, Spanning Trees and Optimization Problems
(Chapman & Hall/CRC, Boca Raton, 2004).
[142] F. Yates, Incomplete randomized blocks, Ann. Eugenics 7 (1936), 121–
140.
416
Index

abelian group, 147 bipartite, 18

addition principle, 15, 87 bipartite adjacency matrix, 295
adjacency, 18 bipartite graph, 19
algorithm for block, 265
minimal spanning tree, 211 repeated, 268
antichain, 171 block code, 226–235
antisymmetric, 318 block design, 266
arrangement, 95 balance, 267
associativity, 308 covalency, 266
asymmetric, 318 dual, 273
atransitive, 318 frequency, 266
incidence matrix, 268
bachelor square, 248 linked, 273
balance, 267 multiple, 280
balanced incomplete block design, parameters, 266
265–279 replication number, 266
Bruck–Chowla–Ryser Theorem, simple, 268
273–274 treatment, 266
covalency, 266 trivial, 267
derived, 275–276, 280, 282, 283 variety, 266
dual, 273 bridge, 205
Fisher’s inequality, 271–272 Bruck–Chowla–Ryser Theorem, 273–
incidence matrix, 268 274
linked, 273 Burnside
multiple, 280 Cauchy–Frobenius–Burnside The-
parameters, 266 orem, 152
quasi-residual, 276
residual, 275–276, 280, 282, 283 cancellation law, 332
simple, 268 Cartesian product, 247, 311–312,
symmetric, 272–279, 282 330
trivial, 267, 275 of groups, 146
Bell number, 131–132 of Latin squares, 247
Bell triangle, 132 of matrices, 248, 330
Bernoulli process, 55 of posets, 168, 169
Bernoulli trial, 55–56 of sets, 14, 248
binary relation, 14, 18 Catalan number, 122–126
Binet’s formula, 110 Catalan, Eugène, 122
binomial coefficient, 29 Cauchy

417
418 Introduction to Combinatorics

Cauchy–Frobenius–Burnside The- D4 , 147, 148, 154, 156, 158, 160,

orem, 152 162
chain, 171 D6 (group), 149, 150
characteristic equation, 108 data compression, 235–237
of a matrix, 286 de Morgan’s Laws, 13, 91
check digit, 224 de Morgan’s laws, 308
Chinese Postman problem, 198 decoding, 223
circular relation, 25 degree, 20, 177–180
code, 223–240 degree sequence, 177
block, 226–235 derangement, 95–97, 119, 298
distance, 227 derived design, 275–276, 280, 282,
fixed-length, 235 283
Hamming, 232–235 determinant, 332–333
length, 226, 227 difference of matrices, 327
linear, 229–235 difference set, 276–279
parity-check, 227, 229 digraph, 18
perfect, 231–235 dihedral group, 147
single error-correcting, 228 dimensions of matrix, 325
variable-length, 235–237 directed graph, 18
Venn diagram, 224–235 disjoint sets, 308
codeword, 224 distributive law, 308
coloring, 143 lattices, 168
proper, 185 dot product, 328
combination, 27–39 down-set, 170
commutative, 147
commutativity, 308 eigenvalue, 286
comparable, 163 eigenvector, 286
complement, 19, 50 empty set, 307
complement of graph, 19 encoding, 223
complement of set, 13, 308 encoding matrix, 233
complete bipartite graph, 18 encryption, 225
complete design, 266 enumeration, 1, 15–17
complete graph, 18 equivalence class
complex numbers, 13, 305 enumerating, 151
conformable matrices, 329 equivalence relation, 14, 143, 318–
congruence, 15, 318 319
convolution identity, 119 Euler walks, 193–199
covalency, 266 mazes, 198–199
cover relation, 164 Euler’s Theorem, 194
cubic graph, 20 Euler, Leonhard, 122
cycle, 181–183 eulerian, 194
of permutation, 134 eulerization
cycle enumerator, 154 efficient, 197
cycle index, 154 eulerization number, 197
cyclic group, 148 Evans conjecture, 253
 Index 419

event, 48 graphical sequence, 177

expected value, 54 valid, 177
experiment, 48 greatest lower bound, 167
group, 144, 244
factorial abelian, 147
falling, 129, 132 action, 149
rising, 133 axioms, 144
falling factorial, 129, 132 Cartesian product, 146
Fibonacci number, 109, 112, 285 cycle notation, 145
Fisher’s inequality, 271–272 cyclic, 148
four color theorem, 185 dihedral, 147
F P (σ, 148 fixed points of, 148
frequency, 266 full symmetric, 148
Frobenius identity, 144
Cauchy–Frobenius–Burnside The- inverse, 144
orem, 152 isomorphism, 148
octic, 147
generating function, 103, 125 orbits of, 148
exponential, 113–115 stabilizer, 151
genetic code, 236–237 subgroup of, 148
geometric progression, 289 transitive, 148
geometric series, 104 two line notation, 145
Goodman’s Theorem, 72
Graeco-Latin square, 246 Hamilton cycles, 200–204, 206–208
graph, 11, 13, 18–20, 177–222 nearest neighbor algorithm, 202–
adjacency matrix, 287 204
bridge, 200, 205 sorted edges algorithm, 202–
circuit, 181 204
closed walk, 181 Hamming bound, 231
coloring, 185–187 Hamming code, 232–235
component, 182 Hasse diagram, 164
connected, 205 hat problem, 9–10, 234–235
connectedness, 182 height, 171
cubic, 177
cycle, 181–183 icosian game, 200
degree, 177–180 idempotence, 309
degree sequence, 177 idempotent Latin square, 251–252
distance, 182 identity matrix, 329
edge, 18 incidence, 266
hamiltonian, 200 incidence matrix, 268
path, 181–183 incomparable, 163
regular, 177 induced permutation, 149
simple walk, 181 induction, 15
vertex, 18 integers, 13, 305
walk, 180–183 intersection, 13, 50, 307
420 Introduction to Combinatorics

inverse matrix, 330–333 mathematical induction, 313–317

inverse relation, 318 matrix, 325–333
invertible, 331 bipartite adjacency, 295
irrational number, 306 cancellation law, 332
irreflexive, 318 Cartesian product, 248, 330
isolated vertex, 20 conformable, 329
isomorphic, 148, 168 determinant, 332–333
isomorphism, 148 diagonal, 286
dimensions, 325
join, 167
entries, 325
K (group), 148, 160, 161 identity, 329
Königsberg bridges, 10–11, 193–197 incidence, 268
Kruskal’s algorithm, 213 inverse, 330–333
invertible, 331
Latin rectangle, 253 non-singular, 331
Latin square, 4, 241–259 permanent, 294
and golf design, 257–259 scalar product, 326
bachelor, 248 square, 329
Cartesian product, 247 subtraction, 327
idempotent, 251–252 sum, 326
mutually orthogonal, 246 symmetric, 327
order, 241 zero, 326, 329
orthogonality, 245–251 zero-one, 325
partial, 253 maximal, 166
self-orthogonal, 250, 260 mean, 55
side, 241 meet, 167
standardized, 246 minimal, 166
symmetric, 244 minimum element, 166
lattice, 166, 167 monochromatic subgraph, 69
distributive, 168 Monty Hall problem, 53
linear code, 229–235 Morse code, 235–236
linked design, 273 multigraph, 18, 181, 291
loop, 244 multiple edges, 18
lower bound, 166 multiplication principle, 2, 15
greatest, 167
mutually exclusive, 51
Lucas, Edouard, 110
numbers, 110
natural numbers, 305
MacNeish’s Theorem, 248 nearest neighbor decoding, 228
map, 183–187 neighbor, 18
Maple, 115, 136–137, 158 network, 18
matching, 295 Newton, Isaac, 126
matchings in bipartites, 295 noise, 223
Mathematica, 115, 136–137, 157, non-singular, 331
293 null graph, 19
 Index 421

octic group, 147 positive integers, 305

orbit, 148 Prim’s algorithm, 211
order of set, 306 Principle of Inclusion and Exclu-
orthogonal array, 263 sion, 89
orthogonality, 245–251 principle of inclusion and exclusion,
outcome, 48 2, 97
probability, 38, 45–63
pancake problem, 2–4 probability model, 54
parity-check code, 227, 229 product of matrices, 328–333
partial Latin square, 253 proper fraction, 306
partial order, 163 proper subgraph, 19
partition, 166
passwords, 1–2 quasi-residual design, 276
path, 181–183 quasigroup, 243
perfect code, 231–235 Abelian, 244
permanent method, 294 idempotent, 244
permutation, 27–39, 134
algorithm to generate all, 38 Ramsey Multiplicity, 72–74
algorithm to generate random, Ramsey’s theorem, 69–81
39 general case, 81
fixed points of, 148 graphical case, 69–80
induced, 149 random event, 47
pigeonhole principle, 65–81 rational numbers, 13, 305
planarity, 184 real numbers, 13, 305
poker, 56–58 recurrence relation, 285
poset, 163–176 degree, 107
antichain, 171 homogeneous, 107
Cartesian product, 168, 169 linear, 107
chain, 171 reflexive, 14, 318
direct sum, 169 regular graph, 20
down-set, 170 relative complement, 307
greatest lower bound, 167 relative difference, 307
Hasse diagram, 164 repeated block, 268
height, 171 replication number, 266
isomorphism, 168 residual design, 275–276, 280, 282,
join, 167 283
lattice, 166, 167 residue class, 13, 306
least upper bound, 167 rising factorial, 133
lower bound, 166 rooted binary tree, 124
maximal element, 166 rule of sum, 88
maximum element, 166 Ryser’s Formula, 298
meet, 167
minimal element, 166 sample space, 48
upper bound, 166 scalar multiplication, 327
width, 171 scalar product, 326, 328
422 Introduction to Combinatorics

SDR, 21, 253, 296 surjection, 126

self-orthogonal Latin square, 250, symmetric, 14, 318
260 symmetric difference, 311–312
semigroup, 244 symmetric group, 148
sequence, 306 symmetric matrix, 327
set, 13–15, 305–319 systems of distinct representatives,
Cartesian product, 14, 248 20–21, 296
element, 305
finite, 305 Theon of Smyrna, 111
infinite, 305 total order, 163
member, 305 tower of Hanoi, 5–7
membership law, 305 transfer matrix method, 287
multiset, 307 transitive, 14, 318
multisubset, 307 transitive group, 148
null set, 307 transpose, 327
order, 306 transversal square, 251
power set, 307 treatment, 266
size, 306 tree, 204–214
subset, 306 Kruskal’s algorithm, 213
superset, 306 Prim’s algorithm, 211
set-theoretic difference, 307 rooted binary, 124
seven wargamers, 7–9 spanning tree, 208–214
shape of matrix, 325 tree diagram, 49–52
shortest path, 182–183 triangulation, 122
simple design, 268 trivial design, 267, 275
singular, 331
size of matrix, 325 union, 13, 50, 87, 307
size of set, 306 universal set, 308
skew, 318 upper bound, 166
square matrix, 329 least, 167
stabilizer, 151
star, 18, 23 valency, 20
state graph, 287 van der Waerden, B. L., 298
Stirling number variable-length code, 235–237
first kind, 132–135 variety, 266
second kind, 126–131 vector, 327–333
strong induction, 313 dot product, 328
subgraph, 19 scalar product, 328
subgroup, 148 vector addition, 327
subset, 13 Venn diagram, 13, 87, 309–311
subsquare, 254 Venn diagram code, 224–235
subtraction, 327 vertex, 18
Sudoku, 4, 253
sum of matrices, 326 walk, 180–183
sum-free sets, 74–77 closed, 181
 Index 423

Euler, 193–199
simple, 181
weak induction, 313
wheel, 25

zero matrix, 326, 329

zero-one matrix, 325