(2020) - Steven Dougherty. Combinatorics and Finite Geometry

Springer Undergraduate Mathematics Series
Steven T. Dougherty
Combinatorics
and Finite
Geometry
Springer Undergraduate Mathematics
Series
Advisory Editors
Mark A. J. Chaplain, St. Andrews, UK
Angus Macintyre, Edinburgh, UK
Simon Scott, London, UK
Nicole Snashall, Leicester, UK
Endre Süli, Oxford, UK
Michael R. Tehranchi, Cambridge, UK
John F. Toland, Bath, UK
The Springer Undergraduate Mathematics Series (SUMS) is a series designed for
undergraduates in mathematics and the sciences worldwide. From core foundational
material to final year topics, SUMS books take a fresh and modern approach.
Textual explanations are supported by a wealth of examples, problems and
fully-worked solutions, with particular attention paid to universal areas of difficulty.
These practical and concise texts are designed for a one- or two-semester course but
the self-study approach makes them ideal for independent use.
More information about this series at http://www.springer.com/series/3423

Steven T. Dougherty
Combinatorics and Finite

Geometry
123
Steven T. Dougherty
Department of Mathematics
University of Scranton
Scranton, PA, USA
ISSN 1615-2085 ISSN 2197-4144 (electronic)

Springer Undergraduate Mathematics Series
ISBN 978-3-030-56394-3 ISBN 978-3-030-56395-0 (eBook)
https://doi.org/10.1007/978-3-030-56395-0
Mathematics Subject Classification: 05, 06, 51, 52, 94
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2020
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
This book is dedicated to my family.
Preface
The mathematics that will be discussed in this text falls into the branch of math-
ematics known as combinatorics. Combinatorics is a very old and very large branch
of mathematics encompassing numerous topics. One nice feature about combina-
torics is that you need to know very little mathematics to understand the open
questions. Unlike many branches of mathematics, there is very little technical
language and very few definitions standing in the way of understanding interesting
unsolved problems in the area. At present, there are numerous problems which no
one has answered yet, but that can be explained to almost anyone. In fact, amateurs
have, on occasion, made interesting discoveries in combinatorics. We shall talk
about one such example in the third chapter.
On the other hand, these open problems are often very, very, difficult to solve. In
a strange twist, it is often true that the easier it is to state a problem, the harder it is
to solve it. Many problems in mathematics or science seem hopelessly difficult
because it requires a good deal of study just to understand the language of the
problem. However, it is often true that the standard techniques of the field can be
applied to solve the problem. So even though the problems may seem very difficult,
they are not. When problems are easy to state and have been around for a while,
then you can be fairly certain that the standard techniques do not apply. These
problems are like infections for which known antibiotics simply do not work. To
solve them you need something new and original. In this same vein, they are often
very infectious, once you begin to think about them you crave to know their
solution. In a very real sense, the best in mathematics is like unrequited love, the
more the discipline refuses to reveal its secrets the more you desire them.
Combinatorics is also interesting because it has a wide overlap with other
branches of mathematics including abstract algebra, number theory, coding theory,
and topology. These and other branches of mathematics often have techniques that
are useful in solving the problems that arise in combinatorics and often problems in
these other areas have combinatorial solutions. Combinatorics also has a very wide
variety of applications in science. In fact, many parts of combinatorics, which were
purely abstract at their birth, were later vital to scientific applications. In fact, the
dawn of the electronic age has brought forth a great interest in combinatorial
matters because of their connection to computers and the communication of
information.
vii
viii Preface
While combinatorics is a very rich and diverse subject, we shall focus our
attention to some specific areas, for example, finite geometry and design theory.
Finite geometry, as the name suggests, is closely related to infinite geometry. They
share many of the same ideas and definitions. Geometry, along with number theory,
is one of the oldest branches of mathematics. In every culture that developed
mathematics, geometry developed first. The reason for this is clear. Namely,
geometry is probably the most intuitive branch of mathematics. The ideas behind
geometry are very natural to consider and have applications from the most ele-
mentary aspects of life to the most advanced science. In finite geometry, we study
these geometric ideas in a finite setting and relate these intuitive notions to com-
binatorial problems.
Geometry has often formed the basis for the development of the rest of math-
ematics. Modern mathematics was built on the base of ancient Greek mathematics,
which phrased every idea (even algebraic ones) in terms of geometry. Greek
mathematics greatly influenced the focus, notation, and tone of all modern math-
ematics. It was on this intuitive, geometric foundation that mathematics was built.
In the nineteenth century, mathematicians began to look at all mathematics
including geometry in a very abstract way and created models for different
geometries using algebraic methods. It was noticed that these same techniques
could be used to create geometries, which were finite in that they contained only
finitely many points and lines. These are geometries where many of the funda-
mental properties of standard geometry were true, but there were only a finite
number of points and lines. It was soon realized that finite geometry had numerous
connections with interesting questions from combinatorics. Design theory arose
from this connection. A design can be thought of as a geometric understanding of a
combinatorial problem.
Later, in the second half of the twentieth century, it was realized that these very
abstract ideas in finite geometry could be used in the very concrete problem of
electronic communication. Many of the ideas in this text have applications both in
the mathematics of coding theory (making sure that a message is received correctly)
and in cryptography (making sure that secret messages are not read by undesired
recipients). One small example of how finite geometry can be used in coding theory
will be shown in the text. Numerous other applications of combinatorics exist in
mathematics, statistics, and science. There has always been a healthy exchange of
ideas from those who study combinatorics as pure mathematics and others who
apply it. Namely, those who apply it give the researcher interesting problems to
think about, and the researcher gives solutions to the combinatorial questions raised
by others.
Many combinatorial problems, on the other hand, have their origins in recre-
ational mathematics and from questions arising in other branches of mathematics.
Often, some interesting questions in mathematics have arisen in very odd cir-
cumstances. For example, the Kirkman schoolgirl problem was raised as a recre-
ational problem as was the question of the existence of the mutually orthogonal
Latin squares. We shall give some examples, including these, in the text of how
Preface ix
some interesting recreational mathematics gave rise to some of the most

well-known and important questions in combinatorics.
Notations
We shall establish some notations that will be used throughout the text. We begin
with the natural numbers. We shall denote the natural numbers by N and we assume
that 0 is a natural number. Therefore, N ¼ f0; 1; 2; 3; . . .g: We denote the integers
by Z and Z ¼ f. . .; 3; 2; 1; 0; 1; 2; 3; . . .g: We denote the rationals by Q where
Q ¼ fba j a; b 2 Z; b 6¼ 0g: We denote the greatest integer function of x by bxc,
namely, bxc ¼ maxfn j n x; n 2 Z}.
Guide for Lecturers
An effort has been made to make each chapter independent and self-contained to
allow it to be read without having to refer back constantly to earlier chapters. The
topics in Sects. 1.1–1.4, 2.1, and 2.2 are really the only ones that must be under-
stood to read the remaining chapters. (It should be pointed out that while the major
results of Sect. 2.2, for example, the existence of finite fields of all prime power
orders, must be understood, it is not necessary to have a complete understanding
of the algebraic techniques used to obtain these results.) Given a knowledge
of these basic chapters, it should be possible for a student to learn the material
independently or to make an interesting course by choosing topics from the
remaining sections.
It is possible to make several different courses from this text. For students who
have not had a course in discrete mathematics, nor a bridge course, the author
suggests that Chaps. 1–6 make a nice introductory course on the subject. In this
scenario, it is probably best to do the first three sections of Chap. 6. It is also
possible to skip the topics from Sect. 2.3 as they are not used in the remainder of the
text, but it is probably best to choose one of these types of numbers to include. The
topics from Chap. 13 can be introduced throughout to allow students to get some
concrete experience with the combinatorial objects. In this case, one can make two
courses, where the second course consists of the remaining sections of Chap. 6
along with Chaps. 7–13. This configuration also works well if the first course is a
standard course and the second one is an independent study course.
For students who have had a course in discrete mathematics or a bridge course, it
is possible to skip Sects. 1.1–1.3, and 2.1. Given that these students are more likely
to be mathematically mature, one can then choose from topics in Chaps. 7 and 8 or
for students whose interest is in computer science one might wish to choose Chap.
11 or Chap. 12 (or both).
x Preface
For more advanced students, one can start with Chap. 3 and make a course from
Chaps. 3–10, choosing topics from the remaining 3 chapters based on student
interest.
Scranton, USA Steven T. Dougherty

Acknowledgements
The author is grateful to Dr. Cristina Fernández-Córdoba, Dr. Charles Redmond,

and Dr. Mercè Villanueva for their comments on earlier drafts of this text. The
author is particularly grateful to Dr. Esengül Saltürk who carefully read the entire
text and offered numerous suggestions.
xi
Contents
1 Foundational Combinatorial Structures . . . . . . . . . . . . . . . . . . . . . 1

1.1 Basic Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Multinomial Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Pigeonhole Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.1 Japanese Ladders . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5 Generating Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.5.1 Coin Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.5.2 Fibonacci Sequence . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 Foundational Algebraic Structures . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1 Modular Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1.1 Euclidean Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.2 Arithmetic Operations . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.3 Fermat’s Little Theorem and Euler’s
Generalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2 Finite Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3 Combinatorial Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.1 Geometric Numbers . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.2 Catalan Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.3.3 Stirling Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.3.4 Towers of Hanoi . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3 Mutually Orthogonal Latin Squares . . . . . . . . . . . . . . . . . . . . . . . . 75
3.1 36 Officers and Latin Squares . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2 Forming Latin Squares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3 Structure of Latin Squares . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4 Affine and Projective Planes . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 97
4.1 Introduction . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 97
4.2 Definitions . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 99
4.3 Planes from Fields . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 107
4.4 Connection Between Affine Planes and MOLS . . . . . . . . . . . . 111
4.5 Fundamental Question . . . . ............... . . . . . . . . . . . 113
xiii
xiv Contents
5 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.1 Königsberg Bridge Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.2 Simple Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.3 Colorings of Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.4 Directed Graphs and Relations . . . . . . . . . . . . . . . . . . . . . . . . 137
6 Higher Dimensional Finite Geometry . . . . . . . . . . . . . . . . . . . . . . . 141
6.1 Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.2 Affine Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.3 Projective Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.4 Desargues’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.5 The Bruck–Ryser Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.6 Arcs and Ovals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.7 Baer Subplanes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.8 Translation Planes and Non-Desarguesian Planes . . . . . . . . . . 177
7 Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 181
7.1 Designs . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 181
7.2 Biplanes . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 188
7.3 Symmetric Designs . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 190
7.4 Kirkman Schoolgirl Problem and Steiner Triple Systems . . . . . 193
7.5 Nets and Transversal Designs . . .................. . . . . 197
8 Combinatorial Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.1 Introduction to Hadamard Matrices . . . . . . . . . . . . . . . . . . . . 203
8.2 Hadamard Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.3 Generalized Hadamard Matrices . . . . . . . . . . . . . . . . . . . . . . . 212
8.4 Latin Hypercubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
8.5 Partially Ordered Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
8.6 Association Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
9 Discrete Probability—A Return to Counting . . . . . . . . . . . . . . . . . . 237
9.1 Definitions and Elementary Probability . . . . . . . . . . . . . . . . . . 237
9.2 Conditional Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
10 Automorphism Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
10.1 Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
10.2 Automorphisms of a Design . . . . . . . . . . . . . . . . . . . . . . . . . 253
10.3 Quasigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
10.4 Difference Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
11 Codes ..................... . . . . . . . . . . . . . . . . . . . . . . . . . . 263
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
11.2 Basics of Coding Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
11.3 Orthogonal Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
11.4 Syndrome Decoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
Contents xv
11.5 The Binary Hamming Code and the Projective Plane

of Order 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
11.6 Projective Geometry and Coding Theory . . . . . . . . . . . . . . . . 280
11.7 Sphere Packing Bound and Perfect Codes . . . . . . . . . . . . . . . 281
11.8 MDS Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
11.9 Weight Enumerators and the Assmus-Mattson Theorem . . . . . 286
11.10 Codes of Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
12 Cryptology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
12.1 Substitution Ciphers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
12.2 German Enigma Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
12.3 Public-Key Cryptography . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
12.3.1 RSA Cryptosystem . . . . . . . . . . . . . . . . . . . . . . . . . . 313
12.3.2 El Gamal Cryptosystem . . . . . . . . . . . . . . . . . . . . . . 315
12.3.3 Diffie-Hellman Key Exchange . . . . . . . . . . . . . . . . . . 316
12.4 McEliece Cryptographic System . . . . . . . . . . . . . . . . . . . . . . . 317
13 Games and Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
13.1 The Game . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
13.2 Weight Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
13.2.1 The Game on Planes . . . . . . . . . . . . . . . . . . . . . . . . . 326
13.2.2 The Game on Biplanes . . . . . . . . . . . . . . . . . . . . . . . 327
13.2.3 The Game on Symmetric Designs . . . . . . . . . . . . . . . 328
13.2.4 The Game on Hadamard Designs . . . . . . . . . . . . . . . 328
13.2.5 The Game on Nets . . . . . . . . . . . . . . . . . . . . . . . . . . 329
13.2.6 The Game on Steiner Triple Systems . . . . . . . . . . . . . 330
14 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Solutions to Selected Odd Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
Foundational Combinatorial
Structures 1
1.1 Basic Counting
Counting is at the core of combinatorics. In general, what we mean by this seemingly

simple term is that we wish to be able to count the number of objects when they are
described in an abstract way. For example, one might want to count the possible
outcomes in a lottery game in order to determine the probability of winning or one
might want to count the possible number of arrangements of symbols to make a
secret message.
To aid us in counting, we use the foundational objects and relations of mathemat-
ics, namely, sets and functions. A set is simply a collection of objects. In this text, we
are largely concerned with finite sets which simplify matters a great deal. A set can
be described by its elements, for example, {a1 , a2 , . . . , an } or it can be described
as {x | p(x)}, where p(x) is a statement about x. As examples, we can have the set
{1, 2, 3, 4} or we can have the set {x | x is an integer and 1 ≤ x ≤ 4}. These sets are
equal as they contain precisely the same elements.
We say that a ∈ A, if the element a is in A. In the previous example, we have
1 ∈ {1, 2, 3, 4}, but 5 ∈
/ A since 5 is not an element of A. The set B is a subset of
A, denoted B ⊆ A, if for every b ∈ B we have b ∈ A. It follows immediately that
A = B if and only if A ⊆ B and B ⊆ A.
A function f : A → B is a subset of A × B = {(a, b) | a ∈ A, b ∈ B} where for
each a ∈ A there exists a unique b ∈ B with (a, b) in the function; we write this
as f(a) = b. For example, if A is the set of people in the United States and B is
the set of social security numbers, then the function that maps each person to their
social security number can be thought of as the set of ordered pairs (a, b) where a
is a person and b is the social security number assigned to a. In the same way, we
can simply write f(a) = b. The key property of a function, meaning the one that
distinguishes it from a general relation, is that each element a has a unique b with
f(a) = b. Therefore, if we let A = B be the set of real numbers, and f be the relation
© The Editor(s) (if applicable) and The Author(s), under exclusive license 1
to Springer Nature Switzerland AG 2020
S. T. Dougherty, Combinatorics and Finite Geometry,
Springer Undergraduate Mathematics Series,
https://doi.org/10.1007/978-3-030-56395-0_1
2 1 Foundational Combinatorial Structures
√
that maps a to its square root, i.e., f(a) = a, then this is not a function since
22 = 4 and 2
√ (−2) = 4. It is possible to turn this into a function by changing it√to
f(a) = | a| which gives the positive square root for each number. When f(x) = x
is used in mathematics, it is generally assumed that it is the function as described
here.
We denote the cardinalityof a set A by |A|. Specifically |A| = n if and only if
there is a bijection between A and the set {1, 2, 3, . . . , n}. If A = ∅ then |A| = 0 and
if no n exists with such a bijection then the set is infinite.
There are three types of functions that are often used in counting arguments,
namely, injections, surjections, and bijections. A function is injective if f(a) = f(a )
implies a = a . Intuitively this means that no element in B is mapped more than once.
A function is surjective if for every b ∈ B there exists an a ∈ A with f(a) = b. In
other words, every element of B is the image of some element in A. A function
is bijective if it is both injective and surjective. A bijection makes a one-to-one
correspondence between the elements of A and B.
To make an inverse function, it is necessary for a function to be injective. If the
function were not injective then there would exist a1 , a2 with a1 = a2 such that
f(a1 ) = b and f(a2 ) = b. Then an inverse function could not be defined, since the
element b would have to be mapped to two distinct elements, violating the definition
of a function.
For an injective function f : A → B, define the inverse function f−1 : B → A by
−1
f (b) = a if and only if f(a) = b. If follows from the definition that if f : A → B
is a bijection then f−1 : B → A is also a bijection.
Lemma 1.1 Let f : A → B and g : B → C with f and g bijections. Then the com-
position g ◦ f : A → C is a bijection.
Proof Assume (g ◦ f)(a1 ) = (g ◦ f)(a2 ) then g(f(a1 )) = g(f(a2 )) which implies

that f(a1 ) = f(a2 ) since g is an injection. Then a1 = a2 , since f is an injection.
Therefore, g ◦ f is an injection.
Let c ∈ C. Then, since g is a surjection, there exists a b ∈ B with g(b) = c. Since
f is a surjection, we have that there exists an a ∈ A with f(a) = b. Then (g ◦ f)(a) =
g(f(a)) = g(b) = c and g ◦ f is a surjection. Therefore g ◦ f is a bijection and we
have our result.
A very important technique for comparing the cardinalities of sets is to construct

maps from one set to another that are either injective, surjective, or both and use the
results that follow.
Theorem 1.1 If A and B are finite sets and there exists a bijection f : A → B, then
|A| = |B|.
Proof Assume |B| = n, then there exists a bijection g : B → {1, 2, . . . , n}. Then
g ◦ f is a bijection from A to {1, 2, . . . , n}. Then, by definition, |A| = n giving that
|A| = |B|.
1.1 Basic Counting 3
In the next three theorems, functions are used to compare the relative sizes of
sets. The statements are true for infinite sets as well but we shall only prove them for
finite sets.
Theorem 1.2 If A and B are finite sets and there is an injection from A to B then
|A| ≤ |B|.
Proof Assume |A| > |B| and f is an injection from A to B with |A| = n, |B| = m. We
have n < m. We have a bijection from A to {1, 2, . . . , n}. Let ai be the element of A
that corresponds to i under this bijection. Then we can write A = {a1 , a2 , . . . , an }.
Let bi = f(ai ). Since f is an injection, f(am+1 ) = f(ai ) for 1 ≤ i ≤ m. But B −
{b1 , b2 , . . . , bm } is empty which gives a contradiction.
Theorem 1.3 If A and B are finite sets and there is a surjection from A to B then
|A| ≥ |B|.
Proof Assume |A| < |B| and f is a surjection from A to B with |A| = n, |B| = m.
We have n > m. Just as in the previous theorem, we have a bijection from A to
{1, 2, . . . , n} and we let ai be the element of A that corresponds to i under this
bijection, with bi = f(ai ). Then bm+1 is not the image of any element under the
function f, which contradicts that f is a surjection.
The following theorem is known as the Schröder-Bernstein Theorem. It is an

extremely important theorem in set theory and fairly difficult to prove for infinite
sets, where the proof requires some form of the Axiom of Choice. For finite sets the
proof is easy.
Theorem 1.4 If A and B are finite sets and there exists an injection f : A → B and
an injection g : B → A then |A| = |B|.
Proof The injection f gives |A| ≤ |B| by Theorem 1.2 and the injection f gives
|B| ≤ |A| by the same theorem. It follows that |A| = |B|.
The most basic counting principle is the multiplicative principle.

Multiplicative Counting Principle: If there are n ways of choosing an object
from a set A and there are m ways of choosing an object from a set B then there are
nm ways of choosing objects from A × B.
This is the principle stated in its most general form. Of course, it extends by
induction to the following.
Generalized Multiplicative Counting Principle: If there are ni ways of choos-

ing an object from the set Ai , then there are k
i=1 ni = n1 n2 . . . nk ways of
choosing elements from A1 × A2 × · · · × Ak .
Essentially, we are simply determining the cardinalityof
A1 × A2 × · · · × Ak = {(a1 , a2 , . . . , ak )|ai ∈ Ai }.
Example 1.1 As a simple example, consider the sets A = {1, 2, 3, 4} and B =

{a, b, c}. There are 4 · 3 = 12 pairs, namely,
{(1, a), (1, b), (1, c), (2, a), (2, b), (2, c), (3, a), (3, b), (3, c), (4, a), (4, b), (4, c)}.
Example 1.2 Assume someone has 5 pairs of pants, 8 shirts, and 3 pairs of shoes.
Then, this person has 5(8)(3) = 120 possible outfits by the multiplicative principle.
We shall now examine some other counting techniques.

Recall the recursive definition of n!, that is 0! = 1 and n! = n(n
− 1)!. It follows
immediately from this definition that for n = 0 we have that n! = n i=1 i.
Theorem 1.5 The number of ways of arranging n items is n!.
Proof Consider the n spaces into which the n items will be arranged. In the first
space there are n choices. In the second there are n − 1 choices since you can choose
any of the n objects except the one that was placed in the first space. At the k-th
space there are n − k + 1 choices since you can choose any of the n objects except
those used in the first k − 1 spaces. The generalized multiplicative counting principle
gives that there are
n

n(n − 1)(n − 2) · · · (2)1 = i = n!
i=1
ways of arranging the n items.
Example 1.3 Consider the ways of arranging the elements a, b, and c. They are as
follows:
abc
acb
bac
bca
cab
cba
Here 3! = 6 and there are 6 ways of arranging the elements.
The value of n! grows wildly in comparison to n. For example, when n is 10 we

have 10! = 3, 628, 800. As another example of how large this number is, consider a
standard set of playing cards. There are 52 cards in such a deck. Hence there are 52!
ways of arranging this deck. The number 52! is greater than 8 · 1067 . Consider that
there are 31, 557, 600 seconds in a year. If we assume the universe has existed at most
10 billion years, then there have been less than 1018 seconds since the beginning of
time. Assuming that at every second someone shuffled a million decks of cards, this
would still give only 1024 different arrangements of a deck that have ever occurred,
assuming of course that there were no repeated formations. Take a deck of cards and
shuffle it. Given this scenario, the chance that there has ever been a deck of cards
arranged in this manner is less than 10143 . You have a much better chance of winning
the lottery billions of times! The point here is that the number of ways of arranging
elements grows extremely quickly. This makes it quite difficult to use computers to
solve combinatorial problems, since the number of different arrangements to check
is often far too great for a computer.
Of course, there are more complicated ways of picking elements of a set. For
example, if you wanted to pick three people from a class to be president, secretary,
and treasurer, then the order in which they are picked makes quite a difference.
Whereas if you are simply picking three students to be a committee then the order
does not matter.
Let P(n, k) denote the number of ways of choosing k elements from n elements
where the order of the choice matters. Let C(n, k) denote the number of ways of
choosing k elements from n elements where the order of the choice does not matter.
We can determine both of these in the following theorem.
n! P(n,k)
Theorem 1.6 For 0 ≤ k ≤ n, P(n, k) = (n−k)!
and C(n, k) = k! =
n!
k!(n−k)!
.
Proof To choose k items from n items, consider the first space. There are n choices
for that space. Then there are n − 1 for the second as before. For the kth space
there are n − k + 1 choices and hence by the generalized multiplicative counting
n!
principle there are n(n − 1) · · · (n − k + 1) = (n−k)! ways of choosing k items
from n where the order of the choosing matters.
To determine the number of ways of choosing k items from n where order does
not matter, we know by Theorem 1.5 that each choice of k items can be arranged in
k! ways. So each choice in counting P(n, k) is in a set of size k! consisting of the
P(n,k) n!
same k elements permuted. Therefore, C(n, k) = k! = k!(n−k)! .
The number P(n, k) is often referred to as the number of permutations.

The
n
number C(n, k) is often called a combination and is sometimes denoted or
k
n Ck .
Example 1.4 A standard deck of cards consists of 52 cards with 4 suits each con-
taining 13 cards. A standard poker hand consists of 5 cards. There are C(52, 5) =
52!
5!47! = 2, 598, 960 possible poker hands. A flush is a set of 5 cards that are all in
the same suit. The number of poker hands that are a flush can be found by choos-
ing 5 cards from any suit and multiplying by the number of suits. Namely, there
are 4C(13, 5) = 4(1287) = 5148 hands that are a flush. A straight consists of 5
cards in a row from any suit. Any card can start a straight except for the face cards.
Namely, there are 40 cards that can start a straight. Given any of those cards, there
are C(4, 1) = 4 choices for the second, third, fourth, and fifth cards. Then the mul-
tiplicative principle gives that there are 40(4)(4)(4)(4) = 10, 240 poker hands that
are straights. Since there are fewer flushes than straights a flush beats a straight in
poker.
Example 1.5 A lottery game has 40 balls in a bin and 6 balls are pulled out. A
40!
winner is anyone who correctly picks all 6 numbers. There are C(40, 6) = 6!34! =
3, 838, 380 possible outcomes. Therefore, someone with a single ticket has a 1 out
of 3, 838, 380 chance of winning this game.
Theorem 1.7 For 0 ≤ k ≤ n, C(n, k) = C(n − 1, k) + C(n − 1, k − 1).
Proof First technique: Pick one element a out of a set A of cardinality n. Every
subset of size k of A either contains a or does not contain a. The number of those
that do not contain a is C(n − 1, k) since you are choosing k objects out of A − {a}.
The number of those that do contain a is C(n − 1, k − 1) since you are choosing
k − 1 objects out of A − {a} and adding the element a. Therefore, C(n, k) = C(n −
1, k) + C(n − 1, k − 1).
Second technique: We simply add C(n − 1, k) + C(n − 1, k − 1) using the for-
mula given in Theorem 1.6:
(n − 1)! (n − 1)!
C(n − 1, k) + C(n − 1, k − 1) = +
k!(n − 1 − k)! (k − 1)!(n − 1 − (k − 1))!
(n − 1)!(n − k) (n − 1)!k
= +
k!(n − 1 − k)!(n − k) k(k − 1)!(n − k)!
(n − 1)!(n − k) + (n − 1)!k
=
k!(n − k)!
(n − 1)!n
=
k!(n − k)!
n!
= = C(n, k).
k!(n − k)!
This recursion, together with the obvious fact that C(n, 0) = C(n, n) = 1, allows
us to compute any value of C(n, k). The values of C(n, k) are often written in the
following table, which is known as Pascal’s triangle. It was known prior to Pascal in
ancient China and to the Italian mathematician Tartaglia, so it is sometimes referred
to as Tartaglia’s triangle. The table is written so that the recursive nature of the
numbers is evident.
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1
1 7 21 35 35 21 7 1
1 8 28 56 70 56 28 8 1
.
.
.
The elements in each place are determined by adding the two numbers imme-
diately above them. This uses the result in Theorem 1.7. Notice that the rows and
columns are both indexed starting at 0 not 1. So the element in row 6 and column 2
is 15. Of course
6! 6(5)
C(6, 2) = 2! = = 15.
4! 2
The table is also symmetric; this follows from the next theorem.
Theorem 1.8 For 0 ≤ k ≤ n, C(n, k) = C(n, n − k).
Proof First technique: By choosing k objects out of n objects to be included in

a set, you are also choosing n − k out of n objects to be excluded from the set.
Therefore, C(n, k) = C(n, n − k).
Second technique: We can simply apply the formula given in Theorem 1.6 which
n! n!
gives C(n, k) = k!(n−k)! and C(n, n − k) = (n−k)!(k)! which are equal.
Theorem 1.9 (The Binomial Theorem) Let n ≥ 0, then

n

(x + y)n = C(n, k)xn−k yk .
k=0
Proof We shall prove the theorem by induction. For n = 0, we have (x + y)0 = 1

and 0k=0 C(n, k)xn−k yk = 1.

Now assume n k=0 C(n, k)x
n−k yk = (x + y)n . Then we have the following
(assuming that C(n, k) is 0 when k is negative):
n+1
n+1

C(n + 1, k)xn+1−k yk = (C(n, k) + C(n, k − 1))xn+1−k yk
k=0 k=0
n+1
n+1

= C(n, k)xn+1−k yk + C(n, k − 1)xn+1−k yk
k=0 k=0
n
n+1

= C(n, k)xn+1−k yk + C(n, k − 1)xn+1−k yk
k=0 k=1
n n
= C(n, k)xn+1−k yk + C(n, k)xn−k yk+1
k=0 k=0
n n

=x C(n, k)xn−k yk + y C(n, k)xn−k yk
k=0 k=0
= x(x + y)n + y(x + y)n = (x + y)n+1 .
Therefore by mathematical induction we have the result.
This theorem allows us to use Pascal’s triangle to determine (x + y)n . For exam-
ple, the coefficients of (x + y)5 come from the numbers in the row corresponding
to 5 in the table, namely
(x + y)5 = x5 + 5x4 y + 10x3 y2 + 10x2 y3 + 5xy4 + y5 .
Let S be a set. We define the power set of S to be
P(S) = {B |B ⊆ S}.
That is, the power set is a set containing all of the subsets of S as elements.
Theorem 1.10 Let S be a set with n elements, then |P(S)| = 2n .
Proof For each of the n elements there are 2 choices, namely, it is either in a subset
or not in a subset. Hence there are 2n possible subsets.
Example 1.6 Let S = {a, b, c}. Then
P(S) = {∅, {a}, {b}, {c}, {a, b}, {a, c}, {b, c}, {a, b, c}}.
So S has cardinality 3 and the power set has cardinality 23 = 8.

Notice that the sum of the elements in the nth row of Pascal’s n . We can
n triangle is 2n−k
prove this by applying the Binomial Theorem.That is, if k=0 C(n, k)x yk =
n n n
(x + y) then let x = y = 1 and we have k=0 C(n, k) = (2) . You can also
prove this using Theorem 1.10. Namely, if there are 2n subsets of a set of cardinality
n, then count the number
of subsets of each cardinality k with 0 ≤ k ≤ n, namely,
C(n, k). This gives n n
k=0 C(n, k) = 2 . Either proof gives the following corollary.
n
Corollary 1.1 For all n ≥ 0, we have that k=0 C(n, k) = 2n .
As an example, if n = 5, then 1 + 5 + 10 + 10 + 5 + 1 = 25 = 32.

The final counting scenario is if we are choosing k objects out of n where order
does not matter but we are allowed to replace elements.
Theorem 1.11 The number of ways of choosing k objects out of n, where order does
not matter, but repetitions are allowed is C(n + k − 1, k).
Proof Assume we have n objects and place them in a line as o1 , o2 , . . . , on . In

front of the oi you place a number of 1s indicating how many of object 1 you want.
Then you end this sequence of 1s with a 0 to indicate to move on to the next one.
Hence there are n + k − 1 places with a 1 or 0 in it, noting that there are precisely
k occurrences of 1. Therefore, the number of ways of choosing k objects out of n
where order does not matter but repetitions are allowed is C(n + k − 1, k).
Example 1.7 Assume there are 7 flavors of ice cream and you are allowed to take
3 scoops of any flavors that you want. How many possible combinations are there.
Here, there are 7 objects and 3 to choose from. Then by Theorem 1.11 the possible
9!
number of combinations is C(7 + 3 − 1, 3) = C(9, 3) = 3!6! = 84.
We summarize these counting techniques as follows.
Objects Chosen Repetitions Order Number of choices

n k Allowed Matters nk
n k Not Allowed Matters n!
P(n, k) = (n−k)!
(n+k−1)!
n k Allowed Does Not Matter C(n + k − 1, k) = (n−1)!k!
n k Not Allowed Does Not Matter n!

C(n, k) = (n−k)!k!
Exercises
1. Let An = {1, 2, . . . , n}. Prove that if n is even, then the number of even
numbers and odd numbers in An are equal and if n is odd, then the number
of even numbers and odd numbers in An are not equal. Prove which one is
greater in the second case.
2. Give examples of functions f from the set {1, 2, 3, . . . , n} to the set

{1, 2, 3, . . . , m} such that
a. f is injective but not surjective

b. f is surjective but not injective
c. f is bijective but not the identity
d. f is neither injective nor surjective
That is, find n, m, and f satisfying the conditions in each part.

3. Prove that if f is an injection from {1, 2, 3, . . . , n} to {1, 2, 3, . . . , n}, then f
must be a surjection and thus a bijection.
4. Prove that if n > m then there is no injection from {1, 2, 3, . . . , n} to
{1, 2, 3, . . . , m}.
5. Prove that if there exists an injection from A to A that is not a surjection then
A must be infinite.
6. Recall that the set of integers denoted by Z is the set
{. . . , −4, −3, −2, −1, 0, 1, 2, 3, 4, . . . }.
Give an example of a function from Z to Z that is an injection but not a

surjection. Give an example of a function from Z to Z that is a surjection but
not an injection.
7. Prove that if |A| = n and |A| = m then n = m, that is, prove that the cardinality
of a finite set is unique.
8. Let S = {a, b, c, d}. Find all 16 elements of P(S).
9. Prove that if p is a prime then C(p, i) is divisible by p for 0 < i < p.
10. Prove that n i=1 iC(n, i) = 2
n−1 n.
11. Prove that C(2n, n) is even for n > 0.

12. How many vectors of length n are there where the coordinates are chosen from
a set of order k.
13. Let a be an element in a set A of size n. Determine the number of subsets of
A of size k containing a.
14. A standard pair of dice consists of two cubes with the numbers 1 through 6
written on the faces. Determine the number of ways a player can roll the dice
and obtain a total of 2, 3, . . . , 12. Which is the most likely roll?
15. Assume a license plate has six places with a number between 1 and 9 or a letter
of the English alphabet in it. Determine the number of possible license plates.
16. Count the number of possible wins a player can make in a tic-tac-toe game
played on a 3 × 3 board, a 4 × 4 board, and a 3 × 3 × 3 board.
17. How many ways can you choose 6 books from 12 choices if repeats are allowed
and order does not matter?
18. On an 8 by 8 chessboard a rook attacks another rook if they share a row or
column. Determine how many rooks can be placed with no rook attacking
another rook on this board.
19. How many ways can you get 5 scoops of ice cream from 12 flavors?
20. On an 8 by 8 chessboard a bishop attacks another bishop if they share a diagonal.

Determine how many bishops can be placed with no bishop attacking another
on this board.
21. Determine how many tic-tac-toe games can be played on a 3 by 3 board.
Determine how many final configurations are there on this board.
22. Let A be a set with cardinality n and let B be a set with cardinality m.
Let f be the bijection between A and {1, 2, 3, . . . , n} and let g be the bijec-
tion between B and {1, 2, 3, . . . , m}. Prove that the function h : (A × B) →
{1, 2, 3, . . . , nm − 1, nm} defined by h(a, b) = f(a) + (g(b) − 1)n is a
bijection.
23. Assume |Ai | = ni . Prove by induction that |A1 × A2 × · · · × Ak | =
k
i=1 ni . Namely, prove that there exists a bijection from
A1 × A2 × · · · × Ak
to
k

{1, 2, 3, . . . , ni }
i=1
using the fact that A1 × A2 × · · · × Ak can be viewed as (A1 × A2 × · · · ×

Ak−1 ) × Ak .
24. Determine the number of poker hands that are a straight flush (a straight all in
the same suit), 4 of a kind, 3 of a kind, 2 of a kind, and a full house (3 of a
kind with 2 of a different kind).
25. The German army Enigma machine had three rotors which permuted letters and
a plugboard that interchanged six pairs of letters. Each rotor had 26 different
initial positions and the three worked in tandem. Count the number of possible
initial configurations for this machine.
26. A blackjack hand consists of two cards from a standard deck. Each card has
the value of its number, face cards have a value of 10 and an ace has the value
of either 1 or 11. Count the number of possible blackjack hands that sum to
12, 18, or 21.
27. In the classic Monty Hall problem there are three curtains. Behind one is a
prize. A player picks one curtain and then the host opens a curtain that does
not have a prize. The player is offered the choice of switching their choice
of curtain. Count the number of ways that switching gets the prize and count
the number of ways that switching does not win and deduce what strategy the
player should adopt.
28. The game of Morra is an Italian game that goes back to Roman times. The
game is played with two people who each put out a number of fingers on one
hand from 0 to 5. Simultaneously, the player calls out a guess for the sum of
the two numbers (saying Morra if the guess is 0). Count the number of possible
games. Then solve the harder problem of counting the number of games where
each player plays rationally, that is, a player only guesses a sum that is greater
than or equal to the number of fingers they put out. For example, a player puts
out a 4 then they would only guess sums from 4 to 9, since 0, 1, 2, 3 and 10
are impossible.
1.2 Multinomial Coefficients
We shall now give a natural generalization for binomial coefficients. Imagine we

have 12 objects to be placed in 3 boxes. The first box must have 4, the second box
must have 6 and the third box must have 2. There are C(12, 4) ways of choosing 4
objects for the first box. Then, since 4 objects have been removed, there are C(8, 6)
ways of choosing 6 objects for the second. Finally, there are 2 objects that remain, so
these two must be placed in the final box. The product of these three is the number
of ways of filling the boxes, which is
12! 8! 2! 12!
= .
4!8! 6!2! 2!0! 4!6!2!
Significant cancelation makes this a simpler problem than one might first have imag-
ined. We generalize this result in the following theorem.
Theorem 1.12 If n objects are to be placed in k boxes, where ki of them are placed
in the ith box, with ti=1 ki = n, then the number of ways of placing the n objects
in the t boxes is
n!
.
k1 !k2 !k3 ! · · · kt !
Proof The number of ways of placing objects in the boxes is
C(n, k1 )C(n − k1 , k2 )C(n − k1 − k2 , k3 ) · · · C(n − k1 − k2 − · · · − kt−1 , kt )

n! (n − k1 )! (n − k1 − k2 )! (n − t−1 i=1 kt )!
= ···
k1 !(n − k1 )! k2 !(n − k1 − k2 )! k3 !(n − k1 − k2 − k3 )! kt !0!
n!
= .
k1 !k2 !k3 ! · · · kt !
This gives the result.
n!
We call this number k1 !k2 !k3 !···kt ! the multinomial coefficient and we denote it
by

n
.
k1 , k2 , . . . , kt
1.2 Multinomial Coefficients 13
Example 1.8 Let n = 9, k1 = 2, k2 = 4, k3 = 3. Then

9 9!
= = 1260.
2, 4, 3 2!4!3!
Theorem 1.13 (The Multinomial Theorem) Let {x1 , x2 , . . . , xt } be a set of t inde-

terminates. Then
n

k k
n
(x1 + x2 + · · · + xt ) = x 1 x 2 · · · xk
t ,
t
(1.1)
k1 , k2 , k3 , . . . , kt 1 2
where the sum is done over all partitions of n.
Proof We prove this by induction on t. For t = 2, it is the binomial theorem proven

earlier in the chapter. Then assuming it is true for t = s − 1, consider (x1 + x2 +
· · · + xt−1 + xt ) as ((x1 + x2 + · · · + xt−1 ) + xt ) and apply the binomial theo-
rem again.
The details are left as Exercise 2. An alternate proof is asked for in
Exercise 4.
Example 1.9 We have the following for n = 3.

3 3 3 3
(x + y + z)3 = x3 + x2 y + x2 z + xy2
3, 0, 0 2, 1, 0 2, 0, 1 1, 2, 0

3 3 3 3
+ xyz + xz2 + y3 + y2 z
1, 1, 1 1, 0, 2 0, 3, 0 0, 2, 1

3 2 3
+ yz + z3
0, 1, 2 0, 0, 3
= x3 + 3x2 y + 3x2 z + 3xy2 + 6xyz + 3xz2 + y3 + 3y2 z + 3yz2 + z3 .
Corollary 1.2 Let n be a positive integer. Then
n

= tn ,
k1 , k2 , k3 , . . . , kt
where the sum is done over all possible partitions of n.
Proof Follows from Theorem 1.13, by letting xi = 1 for all i, 1 ≤ i ≤ t.
We shall now describe another use of the multinomial coefficient. Consider the
English word “syzygy”. The word consists of 6 letters but there are not 6! ways of
getting distinct arrangements of these letters, since some of the letters repeat. There
are 3 letters that are y, and 1 each of s, z, and g. Each of the 6! arrangements of the
letters occurs the number of ways they are arranging each of the letters within a set
of letters. Hence, for this word there are

6 6!
= = 120
3, 1, 1, 1 3!1!1!1!
distinct arrangements of the letters.

In the word “missing”, there are 2 letters that are s, 2 that are i, and 1 each of m,
n, and g. Therefore, the number of distinct arrangements of the letters of this word is

7 7!
= = 1260.
2, 2, 1, 1, 1 2!2!1!1!1!
Exercises
8
1. Compute 4,2,2 .
2. Give the details for the proof of Theorem 1.13.
3. Determine how many distinct ways there are for arranging the letters of Mis-
souri, Pennsylvania, and Mississippi.
4. Give an alternate proof of Theorem 1.13 by counting ways of producing the
k k
monomial x1 1 x2 2 · · · xkt
t .
1.3 Pigeonhole Principle
Theorem 1.2 states that if there is an injection from a set A to a set B then |A| ≤ |B|.
The contrapositive of this is that if |A| > |B|, then there can be no injections from
A to B. This is known as the Pigeonhole Principle, which we state now in its usual
form.
Pigeonhole Principle If n > m and n objects are placed in m boxes then at least
one box must contain two objects.
This seemingly simple statement has numerous powerful applications in counting.
We shall exhibit some of these now. In all applications of this principle, the key is
deciding what to call the objects and what to call the boxes.
Example 1.10 We shall show that given any 6 integers, there must be two of them
whose difference is a multiple of 5. Let the integers be the objects. Any number has
5 possible remainders when divided by 5, namely, 0, 1, 2, 3, or 4. Place the objects
in the box corresponding to its remainder when divided by 5. By the Pigeonhole
Principle, some box must have at least 2 elements. Assume they are in the box with
remainder r. Then these two numbers are of the form 5k + r and 5s + r for some
integers k and s. Then (5k + r) − (5s + r) = 5(k − s) which is divisible by 5.
1.3 Pigeonhole Principle 15
Notice that the Pigeonhole Principle does not determine exactly how many objects
are in each box. For example, in the previous example, you could pick 6 numbers all
of whom have remainder 2 when divided by 5. It does not guarantee that each box
is filled but rather that some box must have 2 elements.
Example 1.11 We will show that given any k + 1 numbers from 1 to 2k that one of
them will be a multiple of the other. Write each number from 1 to 2k as 2s m where
m is an odd number. If two numbers are written like this with the same odd part then
one must be a multiple of the other. There are k odd numbers between 1 and 2k. Let
the boxes be the odd parts and place a number 2s m in the box corresponding to m.
Therefore, by the Pigeonhole Principle there must be at least two numbers in some
box and so one is a multiple of the other.
Example 1.12 Given n people, there must be two people who know the names of
exactly the same number of other people in the room, provided that if person A
knows person B’s name then person B knows person A’s name, namely, the relation
is symmetric. The number of possible people, other than oneself, that you may know
their names, is from 0 to n − 1. This means that there are n boxes and n objects so a
simplistic application of the Pigeonhole Principle does not apply. First assume, that
everyone knows at least one person’s name. Then the number of people whose name
each of the n may know is 1, 2, . . . , n − 1. Hence, there are n people with n − 1
possibilities, so some 2 must know the same number of names. Next, assume that at
least one person knows no name. Then the number of people whose name each of the
n may know is 0, 1, 2, . . . , n − 2 since no one knows n − 1 people’s names (here is
where we must have that the relation is symmetric). Again there are n people with
n − 1 possibilities, so some 2 must know the same number of names.
Example 1.13 Assume we have 21 natural numbers greater than or equal to 1 and
less than or equal to 40. We shall show that some 2 of them must sum to 41. Consider
the sets {1, 40}, {2, 39}, {3, 38}, . . . , {19, 22}, {20, 21}. Since there are 20 of these
sets and 21 numbers, some 2 of them must be in the same set and so their sum is 41.
We can generalize the Pigeonhole Principle as follows.

Generalized Pigeonhole Principle If km + 1 objects are placed in m boxes then
at least one box must contain k + 1 objects.
Proof Assume no box contains k + 1 objects which means that each of the m boxes
contains at most k objects. This means that there are at most km objects which is a
contradiction. Therefore, some box must have at least k + 1 objects.
We can also use this technique for a proof by contradiction as in the following
example.
Example 1.14 Assume that 10 people are handed 40 dollar bills. We shall prove
that some two have the same number of dollar bills. Assume that they are all distinct,
then the smallest number that can be obtained is 0 + 1 + 2 + 3 + 4 + 5 + 6 + 7 +
8 + 9 = 45 which is greater than 40.
Of course, we see that the Generalized Pigeonhole Principle applies to all numbers
from km + 1 to (k + 1)m as well since there are at least km + 1 objects.
Example 1.15 We shall show that if you have a set of 37 = 3(12) + 1 people then
at least 4 of them will have birthdays in the same month. Let the twelve months be
the boxes and the people the objects. Place each person in the month of their birth.
Then, by the Generalized Pigeonhole Principle, one box must have at least 4 objects.
Therefore, at least 4 people share the same birth month.
Example 1.16 We shall show that if you have 37 non-zero two-digit numbers then 3
of them will have the same digit sum. There are 18 possible digital sums of two-digit
numbers, for example, consider the digit sums of the following: 10, 11, 12, 13, 14,
15, 16, 17, 18, 19, 92, 93, 94, 95, 96, 97, 98, 99. We have 2(18) + 1 numbers so
some three of them will have the same digit sum.
Exercises
1. Prove that if there are 8 colors to paint 27 cars, then at least 4 of them must be
painted the same color.
2. Determine how many people you need to have to ensure that at least 8 of them
will have birthdays on the same day of the week.
3. Determine how many numbers you need to have to ensure that at least 6 of
them will have the same remainder when divided by 6.
4. Prove that if you have 5 points in a square of size 1 then at least two of the
points must be within √1 of each other.
2
5. Prove that if you place 13 points on a line segment of length 4, then at least 4
of them are within distance 1 of each other.
6. Determine in what range the number of books must be with a page sum of
3585 to ensure that at least one of them has at least 513 pages.
7. Determine how many integers you need to ensure that at least 8 of them have
the same remainder when divided by 11.
1.4 Permutations
Permutations are one of the most interesting tools to use to understand a finite com-
binatorial object. Essentially, you can study the structure of objects by understanding
how they can be arranged. We shall develop only the most basic aspects of the theory
1.4 Permutations 17
of permutations, namely, only those parts that we shall use in other parts of the text.
The reader is directed to any abstract algebra text for a complete description.
We begin with the definition of a permutation. A permutation on a set A of size n
is a bijection σ : A → A. That is, each element of A gets mapped to a unique element
of A and each element is the image of a unique element. Recall that a bijection is a
map that is both surjective and injective.
The set of all permutations on a set of size n is called the symmetric group on n
letters and is denoted by Sn . It is immediate from the results in the basic counting
section that |Sn | = n!.
We note the following results about the permutations in Sn :
• Composition of functions is associative.

• The identity map is a permutation.
If σ is the identity than for a ∈ A, σ (a) = a so it is surjective. If σ (a) = σ (a )
then a = a and hence the map is injective.
• If σ and τ are both permutations then the composition σ ◦ τ is a permutation.
This was proven in Lemma 1.1.
• If σ is a permutation then the inverse map denoted by σ −1 is a permutation.
If a ∈ A then σ −1 (σ (a)) = a and the map is surjective. If σ −1 (a) = σ −1 (a )
then σ (σ −1 (a)) = σ (σ −1 (a )) and hence a = a and therefore the map is injec-
tive and it is a permutation.
These items prove that Sn is an object called a group. We shall define groups
later in the text in Sect. 10.1, but these properties that we have shown can be used
throughout the following sections.
As an example consider the following permutation on {1, 2, 3} :
1→2
2→3
3→1
We denote this permutation by (1, 2, 3) meaning that 1 is sent to 2, 2 is sent to 3

and 3 is sent to 1. The following permutation on {1, 2, 3, 4}
1→2
2→1
3→4
4→3
would be denoted (1, 2)(3, 4). In general the cycle (a1 , a2 , . . . , ak ) is the permu-
tation where ai is sent to ai+1 , for i from 1 to k − 1 and ak is sent to a1 .
Since a permutation is a function, then when we multiply permutations we are

simply applying the compositions of functions. Therefore when applying the product
of cycles the product is read from right to left.
For example, (1, 2)(2, 3) would be (2, 3, 1). If an element is not written in the
notation then it is assumed that it is mapped to itself.
A transposition is a permutation of two elements. For example, (1, 2) is the trans-
position that switches 1 and 2 and leaves the other elements fixed. Notice that every
transposition is its own inverse, that is (a, b)−1 = (a, b).
Theorem 1.14 Every permutation can be written as the product of transpositions.
Proof If (a1 , a2 , a3 , . . . , an ) is a permutation then
(a1 , a2 , a3 , . . . , an ) = (a1 , an )(a1 , an−1 )(a1 , an−2 ) . . . (a1 , a3 )(a1 , a2 ).
A permutation is said to be even if it can be written as the product of evenly many

transpositions and odd if it can be written as the product of oddly many transpositions.
Note that the identity which has 0 transpositions is an even permutation.
Theorem 1.15 The number of even permutations on the set {1, 2, . . . , n}, n > 1, is
equal to the number of odd permutations on the set {1, 2, . . . , n}.
Proof Let En denote the set of even permutations on the set {1, 2, . . . , n} and let
On denote the set of odd permutations on the set {1, 2, . . . , n}. Define the map
f : En → On by f(σ ) = (1, 2)σ. Clearly, this map sends even permutations to odd
permutations. Assume (1, 2)σ = (1, 2)σ then (1, 2)(1, 2)σ = (1, 2)(1, 2)σ which
gives σ = σ . Then f is an injection and so |En | ≤ |On |.
Then we let f map On to En in the same exact manner which gives |On | ≤ |En |.
This gives that |En | = |On | and we have the result.
It follows immediately that |En | = |On | = n!

2 . While the set of all permutations
is called the symmetric group, the set of even permutations is called the alternating
group.
1.4 Permutations 19
Theorem 1.16 If
σ = (a1 , a2 , a3 , . . . , an ) = (a1 , an )(a1 , an−1 )(a1 , an−2 ) · · · (a1 , a3 )(a1 , a2 )
then
σ −1 = (a1 , an , an−1 , . . . , a2 ) = (a1 , a2 )(a1 , a3 )(a1 , a4 ) · · · (a1 , an−1 )(a1 , an ).
Proof Simply multiply the permutations:
(a1 , an )(a1 , an−1 )(a1 , an−2 ) · · · (a1 , a3 )(a1 , a2 )(a1 , a2 )

(a1 , a3 )(a1 , a4 ) · · · (a1 , an−1 )(a1 , an ).
Using the fact that each transposition is its own inverse n − 1 times gives that this
product is the identity. Taking the multiplication in the opposite order gives the same
result.
Multiplication of permutations is, in general, not a commutative operation. For

example,
(1, 4, 3, 2)(2, 4, 3, 1) = (2, 3, 4)
but
(2, 4, 3, 1)(1, 4, 3, 2) = (1, 3, 4).
However, if the cycles are disjoint then permutations do commute. For example,
(1, 5, 3)(2, 4, 6) = (2, 4, 6)(1, 5, 3).
1.4.1 Japanese Ladders
We shall give a visual representation for permutations which are known as Japanese
ladders. Traditionally, a Japanese ladder (known in Japanese as Amidakuji) was used
to distribute objects or tasks to a group of people.
We begin by giving a graphical example before defining them rigorously.
1 2 3 4 5
. . . . .
.... .... .... .... ....
... ... ... ... ...
.. ... ... ... ...
... .
. .
. .
. ...
... ..
. ..
. ..
. ...
.... .... .... .... ..
....................................................................... .
. .......................................................................
.
.... ... .... .... ....
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ........................................................................ ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ........................................................................ ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ..
........................................................................ ... ..........................................................................
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
.. .. .. .. ..
3 2 5 4 1
The elements fall down the ladder and cross over any rung it encounters. The
element 1 falls down the first post and then moves to the second, then to the third,
then to the fourth, and finally to the fifth. The element 2 falls down the second post,
moves to the first, and then back to the second. The element 3 falls down the third
post, moves to the second, and finally to the first. The element 4 falls down the fourth
post, moves to the fifth, and back to the fourth. The element 5 falls down the fifth
post then moves to the fourth, and finally to the third.
Since 1 ends in the fifth place we say 1 → 5. Likewise we have 2 → 2, 3 → 1,
4 → 4, and 5 → 3. As a permutation, this ladder represents (1, 5, 3). Essentially, we
have written it as a product of transpositions (4, 5)(1, 2)(3, 4)(2, 3)(4, 5)(1, 2).
We can now define a Japanese ladder rigorously. A Japanese ladder is a represen-
tation of a permutation of the form:

(i, i + 1), (1.2)
i∈A
where A is an ordered list of elements of {1, 2, . . . , n}.

Here, each transposition (i, i + 1) corresponds to a rung of the ladder from the
ith post to the (i + 1)-st post.
There are several benefits to viewing a permutation in this graphical manner. The
first benefit of this representation is that the product of two permutations has a very
natural representation. Namely, the product of two ladders representing permutations
1.4 Permutations 21
in Sn can be represented by placing one ladder underneath the other. More precisely,
the ladder representing βα is formed by placing the ladder for β under the ladder for α.
Let α be the permutation: (1, 4, 3). It can be represented as follows:
1 2 3 4
.. .. .. ..
... ... ... ...
.. ... ... ..
... .
. ... ...
... ..
. ... ...
... ..
. ... ...
... ..
. ... ...
... ..
. ... ...
... ..
. ... ...
... ..
. ... ...
......................................................................... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... .. ...
α ... ......................................................................... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
........................................................................ ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
3 2 4 1
Let β be the permutation: (1, 2, 4, 3). It can be represented as follows:

1 2 3 4
... ... ... ...

... ... ... ...
.. ... ... ...
... .
. .
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... . ....................................................................... ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
β ...
...
..
.
..
.
..
.
..
.
...
...
... ..
. ..
. ...
... . .. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ..
.......................................................................... ...........................................................................
.. .. ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
.. .. .. ..
3 1 4 2
Placing α on the top of β (namely, α is performed first), we obtain the permutation

βα = (1, 3, 2, 4). We realize it as follows:
1 2 3 4
.. .. .. ..
... ... ... ...
.. ... ... ...
... .
. .
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
.... .... .... ...
........................................................................ ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ........................................................................ ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ..
..........................................................................
βα ........................................................................
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ........................................................................ ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
........................................................................ ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... .... ... ...
... .. ... ...
4 3 1 2
Placing β on the top of α (namely, β is performed first), we obtain the permutation

αβ = (1, 2, 3, 4). We realize it as follows:
1 2 3 4
... ... ... ...

... ... ... ...
.. .. .. ...
... .......................................................................... ...
... .
. .
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... ..
. ..
. ...
... .... .... ..
...................................................................... ......................................................................
.
.... .... .... ....
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
αβ ......................................................................
.... ....
.
.
....
...
...
... ... ... ...
... ... ... ...
... ... ... ...
... ........................................................................ ...
... ... ... ...
... ... ... ...
... ... ... ...
.... .... ... ...
...................................................................... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
... ... ... ...
.. .. .. ..
4 1 2 3
1.4 Permutations 23
The difference in these two ladders illustrates the non-commutativity of multipli-

cation, that is (1, 2, 3, 4) = αβ = βα = (1, 3, 2, 4).
We shall now show that every permutation can be realized as a Japanese lad-
der. We have already shown that every permutation can be written as a product of
transpositions. All that remains is to show that any transposition can be written as a
Japanese ladder. Take a < b, then the transposition (a, b) can be written as follows:
(a, b) = (a, a + 1)(a + 1, a + 2) · · · (b − 1, b)(b − 2, b − 1) · · · (a + 1, a + 2)(a, a + 1).
As an example we give the ladder for (1, 5):

1 2 3 4 5
.... .... .... .... ....

... ... ... ... ...
.. ... ... ... ...
... .
. .
. .
. ...
... ..
. ..
. ..
. ...
... ..
. ..
. ..
. ...
......................................................................... .... .... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... .... ... ...
... ....................................................................... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ........................................................................ ...
... ... .. .. ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ..
.........................................................................
(1, 5) ...
...
...
...
...
...
...
...
...
...
...
...
...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ........................................................................ ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... .... .... ... ...
... ...................................................................... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
........................................................................ ... ... ...
.. .. ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
... ... ... ... ...
5 2 3 4 1
Any permutation can have infinitely many representations as ladders. For example,
one can simply add two rungs at the bottom of any two adjacent posts in an existing
ladder. For example, if α is a Japanese ladder then (1, 2)(1, 2)α is a Japanese ladder
that represents the same permutation. Moreover, two ladders can represent the same
permutation in a non-trivial manner as in the following:
1 2 3 1 2 3
... ... ... ... ... ...

.... .... ... .... ... ...
... ... .... .. .... ....
.. ..
. .... .. ..
. ....
... ..
. ... ... ..
. ...
... .
.
. .... ... ...
. ..
........................................................................... .... ... ..........................................................................
.
. ...
.... .... ... ..
. ....
.... .... .... ... ...
. ....
.... .... .... ... ..... ....
... ... ... ... .
. ...
.... .... ... ... .
.
. ....
. .
.... ......................................................................... ........................................................................... ....
... ... ... ... ... ...
.... .... ... .... .... ....
.... .... ..... .... .... ....
... ... ... ... ... ..
......................................................................... .... .... ..........................................................................
.... .... .... .... .... ....
... ... ... ... ... ...
.... .... .... .... .... ....
... ... ... ... ... ...
.... .... .... .... .... ....
.. .. .. .. .. ..
3 2 1 3 2 1
Recall that even permutations can be written as the product of an even number of
transpositions, and odd permutations can be written as the product of an odd number
of transpositions. When a permutation is realized as a Japanese ladder, even permu-
tations have an even number of rungs and odd permutations have an odd number of
rungs. This visualization makes it obvious that the product of two even permutations
is even, two odd permutations is even, and the sum of an even permutation and an
odd permutation is odd.
Consider the permutation α = (a1 , a2 )(a3 , a4 )(a5 , a6 ) · · · (ak−1 , ak ) then
α −1 = (ak−1 , ak ) · · · (a5 , a6 )(a3 , a4 )(a1 , a2 ). When the permutation is real-
ized as a Japanese ladder, the inverse permutation is formed by flipping the ladder
upside down.
Recall the permutation α = (1, 4, 3) given above. Then we have that α −1 =
(1, 3, 4) be represented as follows:
1 2 3 4
.... .... .... ....

.... ..... ..... ....
... ... ... ...
... .
.
. ..... ...
.... ....
. .... ....
... ..
. ... ...
.... ..
.
. ..... ....
.... ..
.
. .... ....
.......................................................................... ... ...
. . ....
..... ..... ..... ....
... ... ...
..... ..... ..... ....
.... .... .... ....
... ... ... ...
..... ..... ..... ....
... ... .. ....
......................................................................... ...
α−1 .....
.... ..... ..... ....
....
... ... ...
..... ..... ..... ....
.... .... .... ....
... ... ... ...
..... ..... ..... ....
... .... .... ....
........................................................................... ... ...
.... .... ..... ....
... ... ... ....
..... .... ..... ....
.... ... .... ....
... .... ... ...
..... ..... ..... ....
.... .... .... ....
... ... ... ...
.
4 2 1 3
1.4 Permutations 25
It follows immediately from this visual interpretation that (αβ)−1 = β −1 α −1 .

Namely, if we have the ladder with β on top of α. Turning it upside down gives α
upside down on top of β upside down. Specifically, we place β above α to get αβ
then turning the whole new ladder upside down results in a ladder with the inverse
of α on top and the inverse of β on the bottom giving β −1 α −1 .
There is an algorithm for constructing a Japanese ladder for a given permutation,
see [30] for its relation to a game and [57] for its relation to computer searching.
Namely, you can comb the permutation into its proper order. Consider a sequence
of numbers on the bottom of the ladder: (a1 , a2 , . . . , an ). An inversion is any pair
ai , aj where i < j but ai > aj . The total number of inversions is the number of
minimal number of rungs needed to construct a Japanese ladder for the permutation.
Specifically, change each place where ai > ai+1 by placing a rung between the ith
and the (i + 1)-st rung.
For example, consider the following final state 7 3 4 2 1 5 6. The total number
of inversions is 6 + 2 + 2 + 1 = 11. This corresponds to the permutation (1, 5, 6, 7)
(2, 4, 3). We shall comb it in the manner described above to put it in proper order:
7342156
3742156
3472156
3427156
3421756
3421576
3421567
3241567
3214567
3124567
1324567
1234567
This corresponds to the permutation:
(1, 2)(2, 3)(3, 4)(4, 5)(5, 6)(6, 7)(2, 3)(3, 4)(2, 3)(1, 2)(2, 3). (1.3)
Notice that these transpositions are in the opposite way than they were given in the
algorithm. This is because the algorithm turns 7 3 4 2 1 5 6 into 1 2 3 4 5 6 7 and
the ladder is the inverse permutation of this one. Notice that there are precisely 11
rungs in this ladder.
Exercises
1. Multiply the following permutations:
a. (1, 3, 4)(2, 4, 5, 6)(1, 6, 5, 2)

b. (1, 2, 3)(1, 5, 2, 3)(3, 2, 4, 5)
c. (4, 5, 1, 3, 2)(2, 3, 4)(1, 6, 5, 3)
d. (1, 2, 3, 4, 5)(5, 4, 3, 2, 1)
e. (1, 2, 3, 4, 5)(1, 5, 4, 3, 2)
f. (2, 5)(1, 3)(4, 6)(7, 8)
2. Write the following permutations as a product of transpositions and determine

whether they are even or odd.
a. (1, 6, 3, 4, 5, 2)
b. (1, 4, 3, 2, 5, 6)
c. (2, 4, 6, 1, 3, 5)
d. (6, 2, 3, 1, 5, 4)
e. (3, 5, 7, 9, 1, 2, 4, 6, 8)
f. (4, 5, 3, 6, 2, 7, 8, 1, 9)
3. Prove that for two permutations σ and τ we have (σ τ )−1 = τ −1 σ −1 .

4. Find all 6 permutations in S3 . Then make the 6 × 6 multiplication table for these
permutations.
5. Prove that no permutation can be written both as evenly many transpositions and
as oddly many transpositions.
6. Prove that the product of two even permutations is even, the product of two odd
permutations is even, and the product of one odd and one even is odd.
7. Let F be a finite field. Let σa (x) = ax. Prove that for each a = 0 the map σa is a
permutation of the elements of F. Prove that σa (x) = σb (x) for all x ∈ F − {0}
if a = b.
8. Construct a Japanese ladder with a minimum number of rungs for each of the
following permutations:
a. (2, 4, 3, 5, 6, 7, 1)
b. (4, 2, 3, 1, 5, 6, 7)
c. (7, 6, 5, 4, 3, 2, 1)
d. (1, 3, 2, 4)(6, 5, 7)
e. (4, 2, 5)(1, 3, 6, 7)
f. (1, 7, 2, 6, 3, 5, 4)
9. A Caesar cipher is when there is a permutation on the set of 26 English letters

used to send a secret message. (Historically, Caesar simply cycled the letters by
3 positions). Determine how many possible ciphers can be used by this method.
1.4 Permutations 27
Next, count the number of ciphers that can be created by making a permutation
on the set of all 2 letter pairs. Then, count the number of ciphers that can be
created by making a permutation on the set of all 3 letter triples.
1.5 Generating Functions
1.5.1 Coin Counting
Generating functions are an excellent example of how to use algebra in counting.

Basically, you have a large collection of numbers you want to understand. Instead of
considering them as a large (infinite in fact) collection, we can consider them as a
single algebraic object, namely, an infinite series. We are not concerned about where
the series converges, as in calculus, but rather we consider them as a formal series,
meaning they are simply an algebraic object which we use to carry information for
us. One of the most popular descriptions of this can be found in Polya’s article [68].
We begin by recalling the standard example from calculus of a series, namely, the
geometric series:
∞

1
= x n = 1 + x + x2 + · · · . (1.4)
1−x
n=0
It is easy to see that the coefficient of xn in this series is the number of ways of
making n cents from pennies, namely, there is 1 way. If we change our pennies to
1
nickels then we can only get multiples of 5 for our sum. If we let P(x) = 1−x , then
we can have
∞

N(x) = P(x5 ) = x5n = 1 + x5 + x10 + x15 + · · · . (1.5)
n=0
Next, if we want to use both pennies and nickels we can get
1 1
NP(x) = N(x)P(x) =
1 − x 1 − x5
= 1 + x + x + x3 + x4 + 2x5 + 2x6 + 2x7 + 2x8 + 2x9 + 3x10 + · · · .
2
If we look at the coefficient of x6 , this should tell us how many ways are there
in getting 6 cents from pennies and nickels. Namely, the coefficient is 2 which
corresponds to 6 pennies and 1 nickel and 1 penny. Similarly, the coefficient of x10
is 3 corresponding to 10 pennies, 1 nickel and 5 pennies, and 2 nickels. This series
then counts all possible combinations and yet can be considered as a single algebraic
object.
We can use this algebraic structure to get other interesting relations. Notice that
1 1 1
NP(x) = 5
= P(x), (1.6)
1−x1−x 1 − x5
which gives
(1 − x5 )NP(x) = P(x)
NP(x) = x5 NP(x) + P(x).
Let NPn be the coefficient of xn in NP(x), and let Pn be the coefficient of xn

in P(x). Then from the previous equation, we have that
NPn = NPn−5 + Pn , (1.7)
which gives a recurrence relation to determine the exact value of NPn . Specifically,
the sequence of NPn , starting with n = 0 is
1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, . . . (1.8)
A closed form for NPn is immediate, namely,

n
NPn = + 1. (1.9)
5
We can now add dimes to the collection of coins we are allowed to use. Then
D(x) = P(x10 ) = 1 + x10 + x20 + x30 + · · · (1.10)
Then
1 1 1
DNP(x) = D(x)N(x)P(x) =
1 − x10 1 − x5 1 − x
= 1 + x + x2 + x3 + x4 + 2x5 + 2x6 + 2x7 + 2x8 + 2x9 + 4x10
+ 4x11 + 4x12 + 4x13 + 4x14 + 6x15 + 6x16 + 6x17 + 6x18 + 6x19 + 9x20 + · · ·
Now we have
1 1 1 1
DNP(x) = 5 10
= NP(x), (1.11)
1−x1−x 1−x 1 − x10
which gives
(1 − x10 )DNP(x) = NP(x)

DNP(x) = x10 DNP(x) + NP(x).
Then letting DNPn be the coefficient of xn in DNP(x) we have
DNPn = DNPn−10 + NPn , (1.12)

1.5 Generating Functions 29
1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 6, 6, 6, 6, 6, 9, 9, 9, 9, 9,
12, 12, 12, 12, 12, 16, 16, 16, 16, 16 . . .
Finally, we will add quarters to the collection of coins we are allowed to use. Then
Q(x) = P(x25 ) = 1 + x25 + x50 + x75 + · · · (1.13)
Then
1 1 1 1
QDNP(x) = N(x)P(x)D(x)Q(x) = . (1.14)
1 − x 1 − x 1 − x 1 − x25
5 10
Now we have
1 1 1 1 1
QDNP(x) = 5 10 25
= QDNP(x) (1.15)
1−x1−x 1−x 1−x 1 − x25
which gives
(1 − x25 )NP(x) = NP(x)

QDNP(x) = x25 QDNP(x) + DNP(x).
Then letting QDNPn be the coefficient of xn in QDNP(x) we have
QDNPn = QDNPn−25 + DNPn , (1.16)
1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 6, 6, 6, 6, 6, 9, 9, 9, 9, 9,
13, 13, 13, 13, 13, 18, 18, 18, 18, 18 . . .
We can use these ideas to solve finite problems. For example, assume we have
3 five dollar bills, 4 ten dollar bills, and 2 twenty dollar bills. We can determine
all values that we can obtain from these bills, and how many ways that we can
obtain these. For five dollar bills we have F(x) = 1 + x5 + x10 + x15 . For ten dollar
bills we have T (x) = 1 + x10 + x20 + x30 + x40 . For twenty dollar bills we have
W(x) = 1 + x20 + x40 .
Then we can multiply W(x)T (x)F(x) and get
1 + x5 + 2x10 + 2x15 + 3x20 + 3x25 + 4x30 + 4x35 + 5x40 + 5x45 + 5x50

+ 5x55 + 4x60 + 4x65 + 3x70 + 3x75 + 2x80 + 2x85 + x90 + x95 .
For example, we can see that there are 4 ways of making 35 dollars and 3 ways of
making 75 dollars. Hence, we have solved numerous problems all at once using a
simple algebraic computation.
We shall now show a technique for determining the coefficient for some frequently
appearing generating functions. We begin with a lemma.
Theorem 1.17 Let n be a positive integer. Then

∞

1
= (−1)k C(n + k − 1, k)xk (1.17)
(1 + x)n
k=0
and
∞

1
= C(n + k − 1, k)xk . (1.18)
(1 − x)n
k=0
Proof Let f(x) = (1 + x)−n . Then the kth derivative of f(x) is f(k) (x) = (−1)k
n(n + 1) · · · (n + k − 1)(1 + x)−(n+k) . Then f(k) (0) = (−1)k C(n + k − 1, k).
Then applying Taylor’s Theorem we have the first result.
Let g(x) = (1 − x)−n . Then the kth derivative of f(x) is f(k) (x) = n(n +
1) · · · (n + k − 1)(1 + x)−(n+k) . Then f(k) (0) = C(n + k − 1, k). Then apply-
ing Taylor’s Theorem we have the second result.
Example 1.17 Let n = 2, then
1
= 1 + 2x + 3x2 + 4x3 + 5x4 + · · · .
(1 − x)2
Then
x
1+ = 1 + x + 2x2 + 3x3 + 4x4 + 5x5 + · · · ,
(1 − x)2
which is the generating function counting the number of ways you can choose 1
object out of k objects.
Theorem 1.18 Let n be a positive integer. Then
1 − xn+1
= 1 + x + x2 + · · · + x n .
1−x
Proof We have
(1 − x)(1 + x + x2 + · · · + xn ) = 1 + x + x2 + · · · + xn − x − x2 − x3 − · · · − xn − xn+1 = 1 − xn+1 .

We shall show how to use these results to compute coefficients of generating

functions.
Example 1.18 We shall determine the coefficient of x14 in (x2 + x3 + x4 + · · · )4 .

We have
(x2 + x3 + x4 + · · · )4 = (x2 (1 + x + x2 + · · · ))4

1
= x8 (1 + x + x2 + · · · )4 = x8 .
(1 − x)4
Then the coefficient of x14 in (x2 + x3 + x4 + · · · )4 is the coefficient of x6 in

1
(1−x)4
which is C(4 + 6 − 1, 6) = C(9, 6) = 84.
1.5.2 Fibonacci Sequence
We shall now investigate one of the most famous and interesting sequences and use a
generating function to get a closed form for the sequence. Specifically, we are going
to study the Fibonacci sequence. The sequence first appeared in the early thirteenth
century text Liber Abaci by Leonardo di Pisa, whom we know better as Fibonacci.
We define the sequence recursively as follows:
F0 = 0, F1 = 1, Fn = Fn−1 + Fn−2 . (1.19)
Concretely, the sequence is
0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, . . . (1.20)
We can then think of the generating function for the series as
F(x) = 0 + x + x2 + 2x3 + 3x4 + 5x5 + 8x6 + 13x7 + 21x8 + 34x9 + 55x10 + 89x11 + · · ·
(1.21)
We would like to find a closed form for this sequence. Consider the following:
F(x) = 0 + x + x2 + 2x3 + 3x4 + 5x5 + 8x6 + · · ·

xF(x) = x2 + x3 + 2x4 + 3x5 + 5x6 + · · ·
x2 F(x) = x3 + x4 + 2x5 + 3x6 + · · ·
Using the relation Fn = Fn−1 + Fn−2 , we see that
F(x) − xF(x) − x2 F(x) = x

x
F(x) = .
1 − x − x2
In order to get this into the form we desire, we apply the standard technique of
partial fractions on this function. We begin by factoring the denominator. However,
we do not want to factor in terms of (x − α)(x − β) since this will not help us in
terms of series that we know. Rather, we want to factor in terms of (1 − αx)(1 − βx).
Therefore, we factor the denominator as
√ √
1+ 5 1− 5
1 − x − x2 = (1 − x)(1 − x).
2 2
√
We recognize the number 1+2 5 as the golden mean. We shall describe this number
in geometric terms. Assume a line segment is split into two segments of length L and
S, where the ratio of the longer side L to the smaller side S is the same as the ratio
of the entire line segment L + S is to the larger side L. Namely, we have
L L+S
= . (1.22)
S L
This gives
L2 = SL + S2
L2 − SL − S2 = 0
L L
( )2 − − 1 = 0.
S S
Solving this with the binomial equation gives that the ratio
√
L 1± 5
= .
S 2
√ √
1+ 5 1− 5
The standard notation is to have ϕ = 2 and ϕ = 2 . Then we can write
1 − x − x2 = (1 − ϕx)(1 − ϕx).
Applying the technique of partial fractions we have
√ 1 √ 1
x 5 5
= − . (1.23)
1 − x − x2 1 − ϕx 1 − ϕx
Since we have written these in terms of the standard geometric series, we now have
∞
1 n
F(x) = √ ( ϕ − ϕn) (1.24)
5 n=0
This gives a closed form for the nth term of the Fibonacci sequence, specifically
1
Fn = √ (ϕ n − ϕ n ). (1.25)
5
This formula was first published by Euler in [34].
Exercises
1. How many ways can you make 15 cents from pennies, nickels, and dimes?
2. How many ways can you make 1 dollar from pennies, nickels, dimes, and
quarters?
3. How many ways can you make 25 cents from pennies, nickels, dimes, and
quarters?
4. Use generating functions to see how many ways 50 dollars can be made using
5 dollar bills, 10 dollar bills, and 20 dollar bills.
5. Use generating functions to see how many ways you can make 50 cents from
dimes, quarters, and half-dollars.
6. Using the same recursion that is used for the Fibonacci sequence but changing
the first two values from 0 and 1 to 1 and 2, apply the technique of generating
functions to find a closed form for the new sequence.
F
7. Assuming that the limit exists, show that limn→∞ Fn+1 n
= ϕ.
8. Lucas numbers are defined to be the sequence defined by L0 = 2, L1 = 1,
Ln = Ln−1 + Ln−2 for n > 2. Write out the first 10 terms of the sequence.
Apply the technique of generating functions to find a closed form for the new
sequence.
Foundational Algebraic Structures
2
2.1 Modular Arithmetic
In this section, we shall develop a very small part of the theory of modular arith-
metic. The ideas behind it are actually very old, but it was first codified by Gauss
in Disquisitiones Arithmeticae [39]. In the opinion of the author, Gauss stands with
Euler and Archimedes as one of the three greatest mathematicians of all time. Others
may dispute this but certainly these three are always included in the top five. In any
serious study of discrete and combinatorial mathematics, the names of Euler and
Gauss will appear again and again.
We begin with the definition of the relation modulo on the integers. The integers
denoted by Z are the set {. . . , −3, −2, −1, 0, 1, 2, 3, . . . }. A relation on Z is simply
a subset of Z × Z.
Definition 2.1 Let a, b be integers and n a positive integer. We say that a ≡ b

(mod n) if and only if a − b = kn for some integer k.
As an example, we see that 7 ≡ 1 (mod 3) since 7 − 1 = 6 = 2(3). We also have

that −2 ≡ 1 (mod 3) since −2 − 1 = −3 = (−1)3.
We shall show that the relation modulo n is an equivalence relation. Namely, that
it is reflexive, symmetric, and transitive.
Any integer a has a − a = 0 = 0n so a ≡ a (mod n). Therefore the relation is
reflexive.
If a ≡ b (mod n) then a − b = kn so b − a = (−k)n and then b ≡ a (mod n)
and the relation is symmetric.
If a ≡ b (mod n) and b ≡ c (mod n) then a − b = kn and b − c = k n, then
a − c = a − b + b − c = kn + k n = (k + k )n and so a ≡ c (mod n). This shows
that the relation is transitive.
We have shown the following theorem.
36 2 Foundational Algebraic Structures
Theorem 2.1 The relation a ≡ b (mod n) is an equivalence relation on the inte-

gers.
The equivalence classes of any equivalence relation form a partition on the ambient
space. In other words, you can think of there being n bins where each integer is placed
in a unique bin. Instead of doing the arithmetic in the integers we can simply think
of doing the arithmetic with bins. We can denote the bins by 0, 1, . . . , n − 1 as we
shall now describe.
We see that if a, b are integers with a = b and 0 ≤ a, b ≤ n − 1 then a ≡ b
(mod n) since a − b = 0 and 0 < |a − b| < n and so a − b is not a multiple of
n. Moreover, any integer is equivalent modulo n to some number k with 0 ≤ k ≤
n − 1. This follows from the division algorithm, namely, an integer m = qn + r
with 0 ≤ r < n. That is we simply divide m by n and get remainder r. In this
situation m − r = qn and so m ≡ r (mod n).
What this implies is that we can always take an integer between 0 and n − 1 as
the representative of the equivalence class. That is, we can always divide a number
by n and then the remainder is what the number is equivalent to modulo n.
The ideas behind modular arithmetic are familiar to everyone. One example is
the way in which we tell time. If it is 11:00 then in 4 h it will be 3:00. What we have
done is simply taken 11 + 4 ≡ 3 (mod 12). Whenever we hit 12:00, we reset to 0.
The hours in a day are simply taken modulo 12. If it was 2:00 and someone asked
you what time it would be in 42 hours, you would simply calculate 42 + 2 (mod 12)
which is 6 + 2 = 8 and it would be 8:00. Another example is the days of the week.
We give a specific name (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday,
Saturday) to the seven representatives modulo 7. If it is Tuesday then we know in 6 d
it will be Monday. It is a modular computation we are all familiar making. Notice
that on a calendar that all numbers in a column are equivalent modulo 7.
2.1.1 Euclidean Algorithm
We shall use modular arithmetic to describe the Euclidean algorithm for finding the
greatest common divisor of two integers, since we shall need it in some of the proofs
that follow. The greatest common divisor of two integers a and b denoted gcd(a, b)
is the largest integer that divides both a and b, where an integer n divides an integer
m if and only if m = nk for some integer k.
Take any two integers a1 and a2 . Without loss of generality assume that a1 > a2 .
We shall say that a (mod b) = c if c is the unique number in {0, 1, . . . , b − 1} such
that a ≡ c (mod b). This can be thought of as viewing c as the remainder when
a is divided by b. Define ai for i > 2 to be ai−2 (mod ai−1 ). The sequence
a1 , a2 , a3 , . . . is a decreasing sequence of non-negative integers. Therefore, there
exists an integer s with as = 0 but as+1 = 0. That is, we have the following finite
sequence:
a1 , a2 , a3 , a4 , . . . , as−1 , as , 0.
2.1 Modular Arithmetic 37
As an example, consider the numbers 47 and 13. The sequence would be
47, 13, 8, 5, 3, 2, 1, 0.
Theorem 2.2 ( Euclidean algorithm) Let a1 , a2 be two positive integers with a1 >
a2 . Define ai for i > 2 to be ai−2 (mod ai−1 ). If as+1 = 0, then the integer as
is the greatest common divisor of a1 and a2 .
Proof Since as−1 (mod as ) = 0 we have that as−1 is a multiple of as . We know

ai−1 (mod ai ) = ai+1 . Therefore, there is an integer k with ai−1 − kai = ai+1 .
This gives that ai−1 = kai + ai+1 and it follows that if b is an integer that divides
ai and ai+1 then b divides ai−1 . Then, by induction, we have that as divides each
ai in the sequence.
Assume d is an integer that divides a1 and a2 . If d divides ai and ai−1 then
we know ai−1 (mod ai ) = ai+1 . Then, there is an integer h with ai−1 − hai =
ai+1 . This gives that d divides ai+1 and so d divides every integer ai in the sequence
by induction. In particular, d divides as .
We have shown that as divides a1 and a2 and that if a number divides both
numbers, then it must divide as and therefore be less than or equal to as . This gives
that as is the greatest common divisor of a1 and a2 .
Not only does the Euclidean algorithm give the greatest common divisor, it addi-
tionally provides a way to write the greatest common divisor as a linear combination
of the two numbers. Specifically, we have the following theorem.
Theorem 2.3 Let a1 and a2 be integers with d = gcd(a1 , a2 ). Then there exists
integers b1 , b2 with a1 b1 + a2 b2 = d.
Proof Apply the Euclidean algorithm to a1 and a2 to get the sequence
a1 , a2 , a3 , . . . , as , 0.
Let d = as .
We have already seen that ai−1 − ki (ai ) = ai+1 . Taking one step we get
ai−2 − ki−1 (ai−1 ) = ai . Then substituting for ai , we have ai−1 − ki (ai−2 −
ki−1 (ai−1 )) = ai+1 . Then, we have (1 + ki−1 )(ai−1 ) − ki (ai−2 ) = ai+1 . Thus,
if ai+1 can be written as a linear combination of ai−1 and ai , then it can be written
as a linear combination of ai−2 and ai−1 . Since as is a linear combination of as−1
and as−2 we can apply this repeatedly and write d = as as a linear combination of
a1 and a2 .
Example 2.1 We shall give an example based on our first use of the Euclidean
algorithm on 47 and 13. Recall that we had 47, 13, 8, 5, 3, 2, 1, 0 as our sequence.
Then
3−2 = 1
3 − (5 − 3) = 1
2(3) − 5 = 1
2(8 − 5) − 5 = 1
2(8) − 3(5) = 1
2(8) − 3(13 − 8) = 1
5(8) − 3(13) = 1
5(47 − 3(13)) − 3(13) = 1
5(47) − 18(13) = 1.
Thus we have the greatest common divisor, which is 1, written as 5(47) − 18(13).
That is 5 and −18 are the integers guaranteed by the theorem.
It is interesting to note that Euclid described all of this geometrically. Consider

two lengths of wood. Use the smaller to measure the larger and when the remaining
piece is smaller than the smaller of the original two pieces, cut the wood there. Then
with this piece and the smaller do the process again. Continue this until none is left
over. The piece you have is the greatest common divisor. This geometric approach
has other benefits. For example, Euclid uses this to show when two lengths are
incommensurate. Namely, if this process continues forever then there is no length
that measures both of them. Euclid√ applies this to the base and hypotenuse of a right
isosceles triangle showing that 2 is an irrational number.
2.1.2 Arithmetic Operations
We begin by showing that addition and multiplication are well defined in this setting.
Theorem 2.4 If a ≡ a and b ≡ b , then
a + b ≡ a + b (mod n) (2.1)
and
a ∗ b ≡ a ∗ b (mod n). (2.2)
Proof If a ≡ a and b ≡ b then a = a + kn and b = b + k n for some k, k ∈

Z. Then
a + b = a + kn + b + k n = (a + b) + (k + k )n ≡ a + b (mod n)
and
a b = (a + kn)(b + k n) = ab + bkn + ak n + kk n2
= (ab) + (bk + ak + kk n)n ≡ ab (mod n)
which gives the result.
The next theorem follows from the fact that addition and multiplication is com-
mutative on the integers.
Theorem 2.5 Let n be a positive integer, then
a+b≡b+a (mod n) (2.3)
and
a∗b≡b∗a (mod n). (2.4)
We can now define Zn as the set {0, 1, 2, . . . , n − 1} with addition and multi-
plication of the integers taken modulo n. In terms of abstract algebra, this object is
called a ring. It can have properties that are quite unlike the integers or the reals. For
example, in Z4 , 2 · 2 = 0 and 2x = 0 has two solutions 0 and 2, yet 2x = 1 has no
solutions.
Example 2.2 We give the addition and multiplication tables for Z4 .
+ 0123 ∗ 0123
0 0123 0 0000
1 1230 1 0123 (2.5)
2 2301 2 0202
3 3012 3 0321
Example 2.3 We give the addition and multiplication tables for Z5 .
+ 01234 ∗ 01234
0 01234 0 00000
1 12340 1 01234
(2.6)
2 23401 2 02413
3 34012 3 03142
4 40123 4 04321
The kind of odd occurrences we described for Z4 simply do not happen for Z5 .
Notice that linear equations have a unique solution and if the product of two numbers
is 0, then at least one of them must be 0. The reason for this is that p is prime. We
shall develop these results in the following theorems.
Theorem 2.6 The equation ax ≡ b (mod n) has a solution if and only if gcd(a, n)
divides b.
Proof If ax ≡ b (mod n) then ax − b = kn for some k. This gives that ax − kn =

b. If d divides a and n, with a = da and n = dn then da x − dkn = b which
implies d(a x − kn ) = b and d divides b.
If d = gcd(a, n) divides b, then write a = da and n = dn and b = db .
Consider the equation a x ≡ b (mod n ). We have that gcd(a , n ) = 1 so the
Euclidean algorithm gives that there exists integers e and f with a e + n f = 1.
This means that a e ≡ 1 (mod n ). Then we have
a x ≡ b (mod n )
ea x ≡ eb (mod n )
x ≡ eb (mod n ).
This gives that a (eb ) − b = kn for some k and then da (eb ) − db = dkn
which gives a(eb ) − b = kn. Hence, eb is a solution to ax ≡ b (mod n).
Example 2.4 We shall show some examples of solutions to modular equations.
• If 3x ≡ 4 (mod 6) then 3x − 4 = 6k for some k in the integers, which implies

3(x − 2k) = 4 which implies 3 divides 4, which is a contradiction. Therefore no
solution to this equation exists.
• If 5x ≡ 2 (mod 7) then using the Euclidean algorithm we get that 3(5) − 2(7) =
1. This gives that 3(5) ≡ 1 (mod 7). Then multiply both sides by 3 and we have
x ≡ 6 (mod 7).
• If 4x ≡ 8 (mod 12) then 4x − 8 = 12k for some k in the integers, then x − 2 =
3k and x ≡ 2 (mod 3). Hence the solutions are all of the values modulo 12 that
are 2 (mod 3), namely, 2, 5, 8, 11. The greatest common divisor of 12 and 4 is 4
and hence there are four solutions.
These three examples exhibit the possible situations for modular equations.
Theorem 2.7 If there do not exist a, b such that ab ≡ 0 (mod n), with neither a
nor b equivalent to 0 (mod n), then ab ≡ cb (mod n), b ≡ 0 (mod n) implies
a ≡ c (mod n).
Proof Assume ab ≡ cb (mod n). Then ab − cb ≡ 0 (mod n) which gives that

(a − c)b ≡ 0 (mod n). Since there are no two numbers which multiply to 0 (zero
divisors), we have that either a − c or b must be equivalent to 0, but we have assumed
that b is not. Therefore a − c ≡ 0 (mod n) giving that a ≡ c (mod n).
Definition 2.2 Let n be a positive integer.

• An element a ∈ Zn is a unit if there exists an element b such that ab ≡ 1

(mod n).
• A non-zero element c ∈ Zn is a zero divisor if there exists a d = 0 with cd ≡ 0
(mod n).
We begin by showing that the set of units is closed under multiplication.
Theorem 2.8 The product of two units in Zn is a unit.
Proof Let a and b be two units in Zn . This means there exist c and d with ac ≡
ca ≡ 1 (mod n) and bd ≡ db ≡ 1 (mod n). Then (ab)(dc) ≡ a(bd)c ≡ ac ≡
1 (mod n) This gives that ab is a unit.
Theorem 2.9 The product of a zero divisor and any element is a zero divisor in Zn .
Proof Let a be a zero divisor in Zn then there exists b = 0 with ba ≡ 0 (mod n).
Let c be any element of Zn . We have b(ac) ≡ (ba)c ≡ 0c ≡ 0 (mod n). Therefore
ac is a zero divisor in Zn .
Lemma 2.1 If ac ≡ 0 (mod n) for c ≡ 0 (mod n) then a does not have a multi-
plicative inverse.
Proof If a has a multiplicative inverse b then ba ≡ 1 (mod n). If ac ≡ 0 (mod n)

then bac ≡ b0 ≡ 0 (mod n) which implies c = 0 which is a contradiction.
As an example, 2 has no multiplicative inverse in Z6 since 2 ∗ 3 = 0.
Lemma 2.2 An element a ∈ Zn has a multiplicative inverse if and only if

gcd(a, n) = 1.
Proof Assume gcd(a, n) = k > 1. Then, let a = ka , n = n k with a , n ∈ Zn .

Consider an = a kn = a n = 0, with n = 0. By Lemma 2.1, we know that a
does not have a multiplicative inverse. Hence if gcd(a, n) = 1 then a does not have
a multiplicative inverse. This is equivalent to the contrapositive which says that if a
has a multiplicative inverse then gcd(a, n) = 1.
Assume gcd(a, n) = 1. Then by the Euclidean algorithm there exists integers
b, c with
ab + cn = 1.
Reading this equation modulo n we have ab ≡ 1 (mod n) and b is the multiplicative
inverse of a.
We let Un denote the set of units modulo n. That is Un ⊆ Zn − {0} and a ∈ Un

if and only if a is a unit.
Definition 2.3 The Euler φ function φ : N → N is defined to be φ(n) = |Un |.
We shall compute this function for a few values.
Theorem 2.10 Let p and q be distinct primes. Then φ(p) = p − 1 and φ(pq) =
(p − 1)(q − 1).
Proof If p is a prime then each number between 1 and p − 1 must be relatively

prime to p so φ(p) = p − 1.
If p and q are both primes, then the only numbers that are not relatively prime to
pq, that are less than pq, are the (p − 1) multiples of q and the (q − 1) multiples
of p. Since p and q are relatively prime then no multiple of p is also a multiple of
q. Hence
φ(pq) = pq − 1 − (q − 1) − (p − 1) = pq − q − p + 1 = (p − 1)(q − 1).
Theorem 2.11 If d = gcd(a, n) divides b, then ax ≡ b (mod n) has exactly d

solutions.
Proof If ax ≡ b (mod n) then ax − b = kn for some k. This gives that ax − kn =

b. If d divides a, b, and n, then let a = da , b = db , and n = dn . This gives that
da x − db = dkn . Hence a x ≡ b (mod n ). Since gcd(a , n ) = 1 there is a
solution by Theorem 2.6. Moreover, it is a unique solution by Lemma 2.2. Moreover,
this is easy to see by the Euclidean algorithm. If c is the inverse of a modulo n
then ca x ≡ x ≡ cb (mod n ) and there is a unique solution.
Let e = cb . The solutions to ax ≡ b (mod n) must all be solutions to a x ≡ b
(mod n ) as well. Therefore, the only possibilities are of the form e + αn . We
shall show that e, e + n , e + 2n , . . . , e + (d − 1)n are all solutions to ax ≡ b
(mod n). Notice that these are the only possibilities since e + dn ≡ e (mod n).
We have a(e + αn ) ≡ ae + da αn ≡ b + a αn ≡ b (mod n). Hence they
are all solutions and we have exactly d solutions.
2.1.3 Fermat’s Little Theorem and Euler’s Generalization
Exponentiation is also interesting modulo n and is used extensively in public-key

encryption. One of the most interesting results in this area is due to Fermat and is
known as Fermat’s Little Theorem. It is important to denote this theorem by the
adjective little since there are three very important Fermat’s Theorems. The first is
encountered by every calculus student in the beginning study of that subject. The
second is the little theorem and the third is Fermat’s last theorem which was the most
well known and highly sought after conjecture for centuries until it was proven by
Andrew Wiles in the late twentieth century.
Theorem 2.12 ( Fermat’s Little Theorem) If a ≡ 0 (mod p), with p a prime, then
ap−1 ≡ 1 (mod p).
Proof Consider the elements of Zp − {0} = {1, 2, . . . , p − 1}. All of the compu-
tations in the proof are done modulo p. Then the set {a, 2a, 3a, . . . , (p − 1)a} is
exactly this set permuted, since if ba = ca then b = c by Theorem 2.24. Hence we
have
p−1
p−1

i≡ (ai) (mod p)
i=1 i=1
p−1
p−1

p−1
i≡a i (mod p).
i=1 i=1
p−1
Then, since we know i=1 i is not 0, we can cancel, which gives ap−1 ≡ 1
(mod p).

Notice how useful this is, for example, 237113 ≡ 237112 237 ≡ (2374 )28 237 ≡
237 ≡ 2 (mod 5).
The computation was quite simple and did not require multiplying 237 by itself
113 times.
The next corollary follows immediately from the theorem and is often the way
Fermat’s Little Theorem is stated.
Corollary 2.1 Let a ≡ 0 (mod p), with p a prime, then ap ≡ a (mod p).
Proof Simply multiply each side of the congruence ap−1 ≡ 1 (mod p) by a.
It is possible to generalize Fermat’s Little Theorem using the Euler φ function.

Namely, Euler’s generalization of Fermat’s Little Theorem is the following:
Theorem 2.13 (Euler’s generalization of Fermat’s Little Theorem) If a is a num-

ber that is relatively prime to n then
aφ(n) ≡ 1 (mod n).
Proof Consider the set Un of units in Zn . If a, b, c ∈ Un then ab = ac then b = c

since a is a unit. Then let a ∈ Un and let b1 , . . . , bφ(n) be the elements of Un .
We have that
φ(n) φ(n)

bi ≡ abi (mod n)
i=1 i=1
φ(n) φ(n)

φ(n)
bi ≡ a bi (mod n)
i=1 i=1
1 ≡ aφ(n) (mod n),
φ(n)
since i=1 bi is a unit.
We can use this theorem to find the inverse of a unit in Zn .
Corollary 2.2 Let a be a unit in Zn . Then a−1 = aφ(n)−1 .
Proof We have
aaφ(n)−1 ≡ aφ(n) ≡ 1 (mod n)
This gives one of the most useful applications of Euler’s generalization.
Corollary 2.3 Let gcd(a, n) = 1. Then the solution to ax ≡ b (mod n) is x ≡

aφ(n)−1 b.
Proof By the previous corollary we have
ax ≡ b (mod n)
φ(n)−1
a ax ≡ aφ(n)−1 b (mod n)
φ(n)−1
x≡a b (mod n).
Example 2.5 Consider the modular equation 11x ≡ 7 (mod 23). We need to com-
pute 1121 mod 23. To illustrate the computational ease, we explicitly show how to
computer 1121 efficiently. We begin by repeatedly squaring 11 modulo 23, so that
we have
111 ≡ 11 (mod 23)

112 ≡ 6 (mod 23)
114 ≡ 13 (mod 23)
118 ≡ 8 (mod 23)
1116 ≡ 18 (mod 23).
Then using the base 2 representation of 21 we have
1121 ≡ 1116 114 111 ≡ (18)(13)(11) ≡ 21 (mod 23).
Then x ≡ 7(21) ≡ 9 (mod 23).
Given the importance of Euler’s generalization of Fermat’s Little Theorem, we

want to be able to compute φ(n) for any positive integer n. To accomplish this, we
need to use the Chinese Remainder Theorem. It is an extremely important theorem
which goes back to third century China and the mathematician Sunzi. In that form,
the theorem was stated in terms of remainders, but we shall state using the modern
notation of modular arithmetic.
Theorem 2.14 (Chinese Remainder Theorem) Let m, n be integers with gcd(m, n)

= 1. There is a unique simultaneous solution (mod mn) to the following congru-
ences:
x ≡ a (mod m)
x ≡ b (mod n).
Proof If x ≡ a (mod m) then x = a + km for some k ∈ Z. Then consider a +

km ≡ b (mod n). We get km ≡ b − a (mod n). Since gcd(m, n) = 1 there is a
unique solution for k (mod n). Call this solution c and let k = c + gn, for some
integer g. Then x = a + (c + gn)m = a + cm + g(nm). This gives x ≡ a + cm
(mod mn).
Example 2.6 To solve
x ≡ 2 (mod 5)
x ≡ 3 (mod 7),
let x = 2 + 5k. Then 2 + 5k ≡ 3 (mod 7) which gives 5k ≡ 1 (mod 7). The unique
solution to this congruence is k ≡ 3 (mod 7). Then k = 3 + 7g and x = 2 + 5k =
2 + 5(3 + 7g) = 17 + 35g which gives x ≡ 17 (mod 35).
Applying Theorem 2.14 repeatedly gives the following corollary.
n1 , n2 , . . . , ns be positive integers with gcd(ni , nj ) = 1 if

Corollary 2.4 Let
i = j. Let n = si=1 ni . There is a unique solution modulo n to the following
congruences:
x ≡ a1 (mod n1 )
x ≡ a2 (mod n2 )
..
.
x ≡ as (mod ns ).
Example 2.7 To solve the system of congruences:
x≡2 (mod 3)
x≡1 (mod 5)
x≡5 (mod 7)
x≡5 (mod 11),
we begin by solving the first two congruences to get x ≡ 11 (mod 15). Then we
solve this congruence simultaneously with x ≡ 5 (mod 7) to get x ≡ 26 (mod 105).
Solving this congruence with x ≡ 5 (mod 11), we have x ≡ 236 (mod 1155).
This theorem leads to the next theorem which is very useful in determining φ(n).
Theorem 2.15 Let n and m be relatively prime. Then φ(nm) = φ(n)φ(m).
Proof If a is a unit in Zmn , then a = 1 + kmn for some k ∈ Z. Then reading this
equation both (mod m) and (mod n), we have that a (mod m) and a (mod n)
must both be units. If e is a unit (mod m) and f is a unit (mod n) then there is a
unique solution (mod mn) to
x ≡ e (mod m)
x ≡ f (mod n).
Therefore, the units (mod mn) correspond to the ordered pairs (e, f) where e is a
unit (mod m) and f is a unit (mod n). This gives φ(nm) = φ(n)φ(m).
We shall now determine the value of φ(n) for prime powers.
Theorem 2.16 Let p be a prime, then φ(pe ) = (p − 1)pe−1 .

Proof There are pe − 1 numbers between 1 and pe − 1. The only numbers that are
not relatively prime are the multiples of p, of which there are pe−1 − 1 of them.
Then
φ(pe ) = pe − 1 − (pe−1 − 1) = pe − pe−1 = (p − 1)pe−1 .

Example 2.8 Let n = 27 = 33 , then φ(27) = (3 − 1)32 = 18.
The next theorem uses our previous results and gives a formula that will determine
φ(n) for any natural number n.
s ei
Theorem 2.17 Let n = i=1 pi where pi is a prime and pi = pj if i = j. Then
s
e −1
φ(n) = (pi − 1)pi i .
i=1
Proof The theorem follows directly from Theorems 2.15 and 2.16.
Example 2.9 Let n = 45 = 32 (5), then φ(45) = φ(32 )φ(5) = (3 − 1)3(5 − 1) =

24.
Given the fact that we can easily compute φ(n) using this theorem and that
we can use this result to solve modular equations by Corollary 2.3, it becomes a
straightforward matter to solve any linear modular equation.
Exercises
1. Use the Euclidean algorithm to find the greatest common divisor of the following:
a. 1601 and 543

b. 3412 and 5464
c. 234 and 99
d. 729 and 111
e. 512 and 248
2. Find the greatest common divisor of the following and write it as a linear com-
bination of these two numbers:
a. 64 and 14
b. 93 and 27
c. 12 and 96
d. 123 and 18
e. 29 and 37
3. Prove that if p is a prime then (a + b)p ≡ ap + bp (mod p).

4. Prove that if n is not a prime then there exists a, b such that ab ≡ 0 (mod n),
with neither a nor b equal to 0.
5. Prove that the sum of a unit and a unit is not necessarily a unit in Zn .
6. Prove that in Zn , if there are no zero divisors, then every non-zero element is a
unit.
7. Prove that if n is a prime then there do not exist a, b such that ab ≡ 0 (mod n),
with neither a nor b equal to 0.
8. Prove that (n − 1)! (mod n) is n − 1 if n is prime, 2 if n = 4 and 0 otherwise.
9. Find all solutions to x2 − 1 = 0 in Z4 , Z8 , and Z16 .
10. Determine φ(n) for 1 ≤ n ≤ 12.
11. Use Fermat’s Little Theorem to find the following:
a. 573112 (mod 7)
b. 1261432 (mod 11)
c. 1461203 (mod 13)
d. 21234 (mod 17)
e. 370 (mod 23)
12. Use Euler’s generalization to find the following:
a. 7123 (mod 15)

b. 6243 (mod 25)
c. 8111 (mod 27)
d. 3204 (mod 11)
e. 10403 (mod 35)
13. Solve the following:
a. 3x ≡ 7 (mod 19)
b. 2x ≡ 16 (mod 24)
c. 13x ≡ 27 (mod 71)
d. 11x ≡ 19 (mod 23)
e. 6x ≡ 9 (mod 15)
14. Prove that in Zn , every non-zero element is either a unit or a zero divisor. By
providing a counterexample, show that this is not the case for Z. Determine if
this is true in Q, R, and C.
2.2 Finite Fields 49
2.2 Finite Fields
The basic algebraic building block for developing Euclidean geometry is the set
of real numbers. Euclid and classical Greek geometers did not take an algebraic
approach to geometry. However, even for these geometers the axioms of the real
numbers were implicit in their work. For example, they would have taken it for
granted that between any two points on a line there is another point. It was not until
the early seventeenth century that a clear understanding of the relationship between
algebra and geometry would emerge. It is from this vantage point that we can take an
algebraic approach to geometry. Namely, we can view the Euclidean space in terms
of Cartesian geometry. In this way, we relate n-dimensional Euclidean space with
Rn .
The set of real numbers, together with the usual operations of addition and multi-
plication, is known in algebra as a field. In this section, we shall develop finite fields,
namely, structures which satisfy similar algebraic properties as the real numbers,
but with only finitely many elements. These objects will serve a similar purpose for
finite geometry as the set of real numbers served for infinite geometry. Many of the
analytic and topological properties of the real numbers will not be satisfied by these
objects, but the vital algebraic properties will still hold.
The study of fields is a very large and interesting branch of algebra. We shall
only touch on the most elementary properties of fields which are needed for the
construction of geometries. For an excellent introduction to the study of fields see
Fraleigh’s text [38]. For a classic text on the matter see Bourbaki’s text [13]. We
begin with the definition of a field.
Definition 2.4 A field (F, +, ∗) is a set F with two operations + and ∗ such that the
following hold
1. (additive closure) If a, b ∈ F then a + b ∈ F.

2. (additive associativity) For all a, b, c ∈ F we have (a + b) + c = a + (b + c).
3. (additive identity) There exists an element 0 ∈ F such that 0 + x = x + 0 = x
for all x ∈ F.
4. (additive inverses) For all a ∈ F there exists b ∈ F with a + b = b + a = 0.
5. (additive commutativity) For all a, b ∈ F we have a + b = b + a.
6. (multiplicative closure) If a, b ∈ F then a ∗ b ∈ F.
7. (multiplicative associativity) For all a, b, c ∈ F we have (a ∗ b) ∗ c = a ∗ (b ∗
c).
8. (multiplicative identity) There exists an element 1 ∈ F such that 1 ∗ x = x ∗ 1 =
x for all x ∈ F.
9. (multiplicative inverses) For all a ∈ F − {0} there exists b ∈ F with a ∗ b =
b ∗ a = 1.
10. (multiplicative commutativity) For all a, b ∈ F we have a ∗ b = b ∗ a.
11. (distributive property) For all a, b, c ∈ F we have a ∗ (b + c) = a ∗ b + a ∗ c
and (b + c) ∗ a = b ∗ a + c ∗ a.
We say that the order of the field is |F|. That is, it is the cardinality of the underlying
set which is the defining parameter of the algebraic object. It is possible for the order
to be infinite like the reals, but here we shall only be concerned with fields where the
order is a natural number. To avoid trivial cases we assume that the order is at least
2 for any finite field.
As is the usual convention we shall often use juxtaposition to indicate the multi-
plicative operation, that is, we write ab to mean a ∗ b.
Notice that the addition and multiplication tables given in (2.6) show that Z5 is a
field of order 5. Of course, Zn is not a field for all n, for example, Z4 is not a field
since 2 has no multiplicative inverse. The real numbers, R, the rational numbers Q,
and the complex numbers, C, are all fields with respect to their usual addition and
multiplication. The integers are not a field since, for example, the element 2 does not
have a multiplicative inverse in the integers.
A structure that satisfies all the axioms except possibly (9) and (10) is known as a
ring. Some texts will also allow axiom (8), the existence of an identity, to be violated
as well. Sometimes, an object without a multiplicative identity will be known as a
rng (that is, ring without the “i”) and those with a multiplicative identity will be
known as a ring with both words pronounced identically.
It is possible for a structure to satisfy all axioms except for item 10 (multiplica-
tive commutativity). The Quaternions are such an example, see [38] for a complete
description. It is not possible for a finite structure to satisfy all axioms except for
item 10 (multiplicative commutativity). That is, any finite ring that satisfies the first
nine axioms and the eleventh must satisfy all eleven.
Theorem 2.18 The ring Zn is a field if and only if n is a prime.
Proof By Exercise 3., we have that all of the items except multiplicative inverses
are satisfied. By Lemma 2.2, we know that an element has a multiplicative inverse
if and only if it is relatively prime to n. If n is prime then every non-zero element of
Zn is relatively prime to n. If n is not a prime then n = ab for some a, b. Then a
is not relatively prime to n and Zn is not a field.
Example 2.10 Consider n = 6. Here the elements 2, 3, and 4 do not have multi-
plicative inverses so Z6 cannot be a field.
Not every finite field is Zn for some n. For example, the following addition and
multiplication tables describe a finite field of order 4.
+ 0 1 ω ω2 ∗ 0 1 ω ω2
0 0 1 ω ω2 0 0 0 0 0
1 1 0 ω2 ω 1 0 1 ω ω2 (2.7)
ω ω ω2 0 1 ω 0 ω ω2 1
ω2 ω2 ω 1 0 ω2 0 ω2 1 ω
We shall give a very brief explanation of how fields of other orders are constructed.
Consider the following algebraic structure known as a polynomial ring. Let Zp [x] =
{a0 + a1 x + a2 x2 + · · · + ae−1 xe−1 | ai ∈ Zp , e ∈ N}. It can be easily verified
that this object is a ring. It is not, however, a field, since, for example, the element x
has no multiplicative inverse.
For any element p(x) ∈ Zp [x], let
p(x) = {a(x)p(x) | a(x) ∈ Zp [x]}. In terms
of ring theory, this object is known as an ideal. We can mod out by such an ideal in
a manner similar to the technique we used to perform modular arithmetic. Likewise,
it was the fact that we were moding out by a prime (namely, a number that was not
the non-trivial product of smaller numbers) that made Zp a field. Without giving a
detailed proof, we can state the following theorem which shows how to construct
fields of orders that are not prime.
Theorem 2.19 Let p(x) be an irreducible polynomial of degree e in Zp [x]. Then

Zp [x]/
p(x) is a field of order pe .
Example 2.11 Consider the polynomial x2 + 1 over Z3 . If it were to factor, it would

have to factor into linear terms which would mean that either 0, 1, or 2 would have
to be a root of the polynomial in Z3 . It is easy to verify that none of these is a root.
Therefore, x2 + 1 is an irreducible polynomial of degree 2 over Z3 . In practical
terms, this means that in Z3 [x]/
x2 + 1, we have that x2 + 1 = 0. This means that
every time we encounter x2 in a computation we can replace it with −1 which is 2
in Z3 . Therefore, the elements of this object are the set:
{0, 1, 2, x, x + 1, x + 2, 2x, 2x + 1, 2x + 2}.
The addition and multiplication are given as follows.
+ 0 1 2 x x + 1 x + 2 2x 2x + 1 2x + 2
0 0 1 2 x x + 1 x + 2 2x 2x + 1 2x + 2
1 1 2 0 x+1 x+2 x 2x + 1 2x + 2 2x
2 2 0 1 x+2 x x + 1 2x + 2 2x 2x + 1
x x x + 1 x + 2 2x 2x + 1 2x + 2 0 1 2
x+1 x+1 x+2 x 2x + 1 2x + 2 2x 1 2 0
x+2 x+2 x x + 1 2x + 2 2x 2x + 1 2 0 1
2x 2x 2x + 1 2x + 2 0 1 2 x x+1 x+2
2x + 1 2x + 1 2x + 2 2x 1 2 0 x+1 x+2 x
2x + 2 2x + 2 2x 2x + 1 2 0 1 x+2 x x+1
(2.8)
∗ 0 1 2 x x + 1 x + 2 2x 2x + 1 2x + 2
0 0 0 0 0 0 0 0 0 0
1 0 1 2 x x + 1 x + 2 2x 2x + 1 2x + 2
2 0 2 1 2x 2x + 2 2x + 1 x x+2 x+1
x 0 x 2x 2 x + 2 2x + 2 1 x + 1 2x + 1
(2.9)
x+1 0 x + 1 2x + 2 x + 2 2x 1 2x + 2 2 x
x+2 0 x + 2 2x + 1 2x + 2 1 x x + 1 2x 2
2x 0 2x x 1 2x + 2 x + 1 2 2x + 1 x + 2
2x + 1 0 2x + 1 x + 2 x + 1 2 2x 2x + 1 x 1
2x + 2 0 2x + 2 x + 1 2x + 1 x 2 x+2 1 2x
This object is a finite field of order 9.
We seek to not only construct finite fields but to classify them as well. We begin
with the necessary definitions. Let F and K be fields. A field homomorphism is a
map Φ : F → K such that Φ(ab) = Φ(a)Φ(b) and Φ(a + b) = Φ(a) + Φ(b), for
all a, b ∈ F. If a field homomorphism is also a bijection then it is said to be an
isomorphism.
We say that two fields F and K are isomorphic if there exists an isomorphism
Φ : F → K. Isomorphic fields are essentially identical in structure. For example, we
have the following easy results.
Theorem 2.20 Let F be a field with additive identity 0 and multiplicative identity
1 and let K be a field with additive identity 0 and multiplicative identity 1 . Let
Φ : F → K be a field isomorphism. Then Φ(0) = 0 and Φ(1) = 1 .
Proof For all a ∈ F we have a + 0 = 0 + a = a. Then Φ(a) = Φ(a + 0) = Φ(a)

+ Φ(0) and Φ(a) = Φ(0 + a) = Φ(0) + Φ(a). Since Φ is a bijection then every
element of K is of the form Φ(a) for some a ∈ F. Therefore, for all a ∈ K we have
a + Φ(0) = Φ(0) + a = a . Then Φ(0) is an additive identity in K. Since the
additive identity in a field is unique we have Φ(0) = 0 .
For all a ∈ F we have a1 = 1a = a. Then Φ(a) = Φ(a1) = Φ(a)Φ(1) and
Φ(a) = Φ(1a) = Φ(1)Φ(a). Since Φ is a bijection then every element of K is
of the form Φ(a) for some a ∈ F. Therefore, for all a ∈ K we have a Φ(1) =
Φ(1)a = a . Then Φ(1) is an additive identity in K. Since the additive identity in
a field is unique we have Φ(1) = 1 .
This leads naturally to the following theorem.
Theorem 2.21 Let Φ be a field isomorphism between F and K. If a ∈ F then

Φ(−a) = −Φ(a) and if a = 0 then Φ(a−1 ) = Φ(a)−1 .
Proof Let a ∈ F then a + (−a) = 0 which gives Φ(a + (−a)) = Φ(0). Then, we
have Φ(a) + Φ(−a) = 0 where 0 is the additive identity of K. Therefore Φ(−a) =
−Φ(a).
Let a ∈ F \ {0} then aa−1 = 1 which gives Φ(aa−1 ) = Φ(1). Then, we have
Φ(a)Φ(a−1 ) = 1 where 1 is the additive identity of K. Therefore Φ(a−1 ) =
Φ(a)−1 .
Example 2.12 Consider the isomorphism Φ : F4 → F4 , Φ(a) = a2 , where F4 is

the finite field given in Eq. 2.7. The isomorphism is realized as follows:
0→0
1→1
ω → ω2
ω2 → ω
We note that the additive identity is sent to the additive identity and the multiplicative
identity is sent to the multiplicative identity. The multiplicative inverse of ω is ω2 .
Then Φ sends ω to ω2 and its inverse ω2 to ω. Hence inverses go to inverses.
The following is a well-known theorem of algebra whose proof is beyond the

scope of this text. The proof can be found in [38].
Theorem 2.22 Finite fields exist if and only if their order is of the form pe where p
is a prime and e > 0. Any two finite fields of the same order are isomorphic.
With this theorem in mind, we can denote the finite field of order q, when q is a
prime power, by Fq since we know such a field exists and up to isomorphism there
is only one such field. This is not true for infinite fields. For example, the real and
the complex numbers are not isomorphic. This can be easily seen by the fact that if
there were an isomorphism Φ : C → R, then Φ(i) would need to satisfy x2 = −1
by the previous theorems. Since there is no solution to this equation in the reals, then
there cannot be an isomorphism.
We shall give a few more theorems to indicate the structure of a field.
Theorem 2.23 If ab = 0 in a field then either a = 0 or b = 0.
Proof If a is not 0 then there exists a multiplicative inverse of a denoted by a−1 .

Then we have a−1 ab = a−1 0 which implies b = 0. Hence either a or b must be 0.

This theorem says that there are no zero divisors in a field. For infinite structures
there are rings that have no zero divisors, such as the integers, that are not fields. It
will not occur for finite fields as we shall see in Theorem 2.25.
Theorem 2.24 In a field, if ab = cb, b = 0 then a = c and if ba = bc, b = 0 then

a = c.
Proof Assume ab = cb where b = 0. Then there exists an element b−1 with

bb−1 = 1. Then
ab = cb
abb−1 = cbb−1
a(1) = b(1)
a = b.
The second result follows from the first proof and the fact that multiplication is
commutative.
It is often a good idea, given Theorem 2.23, to show a ring has zero divisors when
showing that it is not a field. This indicates that not every non-zero element has a
multiplicative inverse.
An integral domain satisfies all field axioms except that it need not satisfy axiom
(9). It must also have no zero divisors, that is, if ab = 0 then either a = 0 or b = 0.
An infinite example would be the integers. The next theorem shows that there are no
finite examples.
Theorem 2.25 A finite integral domain is a field.
Proof Let D be a finite integral domain. All that we need to do is to show that every
non-zero element has a multiplicative inverse.
If a is non-zero and ac = ab then ac − ab = 0 and a(c − b) = 0. Then since
there are no zero divisors c must equal b. Consider the non-zero elements of D =
D − {0}. From the previous statement we know that each element of aD must be
distinct and so must coincide with the elements of D . Hence, there must be a b ∈ D
with ab = 1 since 1 ∈ D .

The characteristic of a finite field is a if ai=1 x = 0 for all x ∈ F, and a is the
smallest such number satisfying this property. That is, adding any number to itself
a times will result in 0. For example, the characteristic of Z2 and the field of order
4 already exhibited is 2 and the characteristic of the field of order 3 is 3.
Theorem 2.26 The characteristic of a finite field must be a prime.
Proof Assume the characteristic is a and a = bc with neither b nor c equal to 1. Then
b c
0= a i=1 1 = ( i=1 1)( i=1 1). This implies the existence of zero divisors and
we have a contradiction.
Example 2.13 The characteristic of the field of order 4 given in Eq. 2.7 is 2. The
characteristic of the field of order 9 given in Example 2.11 is 3. We note that both
characteristics are prime.
Often we shall use the space Fn which is shorthand for the cross product of F
with itself n times, F × F × · · · × F. Concretely
Fn = {(v1 , v2 , . . . , vn ) | vi ∈ F}.
For example,
F32 = {(0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0), (1, 1, 1)}.
The generalized multiplicative counting principle gives that there are |F|n ele-
ments in this space. We refer to it as a space rather than a set because it is naturally
a vector space. We shall describe vector spaces completely in Chap. 6. We shall use
these vector spaces extensively to construct combinatorial objects known as finite
geometries.
Exercises
1. Verify that the object with addition and multiplication in Eq. 2.7 is a field.
2. Find two solutions to x2 + x + 1 = 0 in the field of order 4.
3. Show that Zn satisfies all axioms of a field except for item 9 (multiplicative
inverses) for all n. That is, show that Zn is a commutative ring with identity.
4. Prove that the real numbers and the complex numbers are not isomorphic fields.
Hint: assume there is such an isomorphism from the complex numbers to the
real numbers and examine the image of the element i under that isomorphism.
5. Prove that in a field F of characteristic p that Φ(a) = ap is a homomorphism
from F to F. This map is not necessarily the identity as a novice might assume
from Fermat’s Little Theorem. For example, in the field of order 4 described
earlier, which has characteristic 2, we have ω2 = ω.
6. Determine if the set of n by n matrices with entries from a field is a field. If not
what axiom(s) does it violate?
7. Prove that for a field F, F × F, where (a, b) + (c, d) = (a + c, b + d) and
(a, b) ∗ (c, d) = (a ∗ c, b ∗ d), is not a field. What axiom does it violate?
8. Determine if

a −b
{ | a, b ∈ R}
b a
is a field with the usual matrix addition and multiplication.
9. Find the multiplicative inverses for the elements in Z11 .
10. Prove that a field isomorphism sends the additive identity to the additive identity
and the multiplicative identity to the multiplicative identity, that is, if f : F → E
with f an isomorphism then f(0F ) = 0E and f(1F ) = 1E .
11. Prove that in a field there is a unique solution to the equation ax + b = c for all
non-zero a.
12. Prove that in any finite field there must be an a satisfying the definition of the
characteristic.
13. Prove that no element can be the multiplicative inverse for two distinct elements
in a field.
14. Prove that in a field the equation x2 = a can have at most two solutions. Use
this to prove that in the finite field Zp , p a prime, precisely half of the non-zero
elements are squares. Determine which elements are squares for Z7 and Z11 .
15. Prove that if a is non-zero in a field F, and b is any element in F, then there exists
c ∈ F with ac = b.
16. Prove that the inverse of a field isomorphism is a field isomorphism.
17. Determine if x3 + x + 1 is irreducible over Z2 . If it is irreducible, then use it to
construct a field of order 8.
18. Let Φ be an isomorphism between two fields F and K. Assume α is a solution
to a0 + a1 x + a2 x2 + . . . an−1 xn−1 in F[x]. Prove that Φ(α) is a solution to
Φ(a0 ) + Φ(a1 )x + Φ(a2 )x2 + . . . Φ(an−1 )xn−1 in K[x].
19. Consider the set S = {0, 1, u, 1 + u} where all operations are done modulo 2.
There are four distinct possibilities, namely, u2 = 0, u2 = 1, u2 = u, and u2 =
1 + u which give algebraic structures. Determine which of these are a field and
which are not a field.
2.3 Combinatorial Numbers
2.3.1 Geometric Numbers
The first collection of combinatorial numbers correspond to geometric objects. We

shall examine these numbers as a sequence. A sequence is a function f : N → R. More
efficiently, we can often denote f(n) by an and write the sequence as a0 , a1 , a2 , . . . .
The square numbers are those numbers that can be described as a collection of
items placed in a square.
(2.10)
(2.11)
2.3 Combinatorial Numbers 57
(2.12)
(2.13)
This sequence can be described explicitly as f(n) = n2 . Recursively, we can
describe it as a0 = 0, an+1 = an + 2n + 1. We now prove that these two give the
same sequence.
Theorem 2.27 The sequence f(n) = n2 matches the sequence given recursively by
a0 = 0, an+1 = an + 2n + 1.
Proof We shall prove the result by induction. If n = 0, f(0) = 0 and a0 = 0.

Next assume that f(n) = an . Then f(n + 1) = (n + 1)2 = n2 + 2n + 1 =
f(n) + 2n + 1 = an+1 . Therefore, by the principle of mathematical induction we
have the result.
Concretely, the square numbers are
0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, . . .
Theorem 2.28 If p is prime that divides n2 , then p2 divides n2 .
Proof If p divides n2 = n · n, then either p divides n or p divides n by the definition

of a prime. Either case gives p divides n, giving n = pk for some integer k. Then
n2 = p2 k2 and p2 divides n2 .
As a consequence of this theorem, we note that any even square number must be
0 (mod 4). Also if a square number ends in 0 (divisible by 5 and 2) it must end in
00 (divisible by 52 22 ). More generally, we have that any square number that is 0
(mod p) must be 0 (mod ps ).
We note also that a square number written in base 10 must end in a square
(mod 10), namely, 0, 1, 4, 5, 6, or 9.
We can also get a formula for the sum of the first k positive square numbers.
Theorem 2.29 Let k be a positive integer. Then
k
k(k + 1)(2k + 1)
n2 = . (2.14)
6
n=1
Proof We shall prove the result by induction. If k = 1 the left side of the equation is
1(2)(3)
12 = 1 and the right side is 6 = 1 giving that the formula is correct for k = 1.
Assume the formula is correct for k and consider the sum of the first k + 1 squares:
k+1
k

2
n =( n2 ) + (k + 1)2
n=1 n=1
k(k + 1)(2k + 1) 6(k + 1)2
= +
6 6
k(2k + 1) + 6(k + 1)
= (k + 1)
6
2k2 + 7k + 6
= (k + 1)
6
(k + 1)(k + 2)(2k + 3)
= ,
6
which is the desired result. Therefore, by the principle of mathematical induction,
the formula is correct for all positive k.
The cubic numbers are those numbers that can be described as a collection of
items placed in a cube.
(2.15)
This sequence can be described explicitly as f(n) = n3 . Recursively, we can

describe it as a0 = 0, an+1 = an + 3n2 + 3n + 1. We now prove that these two
give the same sequence.
Theorem 2.30 The sequence f(n) = n3 matches the sequence given recursively by
a0 = 0, an+1 = an + 3n2 + 3n + 1.
Proof We shall prove the result by induction. If n = 0, f(0) = 0, and a0 = 0.

Next assume that f(n) = an . Then f(n + 1) = (n + 1)3 = n3 + 3n2 + 3n +
1 = f(n) + 3n2 + 3n + 1 = an+1 . Therefore, by the principle of mathematical
induction, we have the result.
Concretely, the cubic numbers are
0, 1, 8, 27, 64, 125, 216, 343, 512, . . .
Theorem 2.31 If p is prime that divides n3 , then p3 divides n3 .
Proof If p divides n3 = n · n · n, then either p divide n by the definition of a prime.

This gives p divides n, giving n = pk for some integer k. Then n3 = p3 k3 and p3
divides n3 .
This theorem says that if a cubic number is 0 (mod p) then it is 0 (mod p3 ) as

well. However, unlike square numbers, this does not reduce the possible final digit in
a cubic number in its base 10 representation since all numbers (mod 10) are cubes.
We can determine the sum of the first k cubic numbers.

k
k2 (k + 1)2
n3 = . (2.16)
4
n=1
Proof We shall prove the result by induction. If k = 1 the left side of the equation
1(4)
is 13 = 1 and the right side is 4 = 1 giving that the formula is correct for k = 1.
Assume the formula is correct for k and consider the sum of the first k + 1 squares:
k+1
k

n3 = ( n3 ) + (k + 1)3
n=1 n=1
k2 (k + 1)2
= + (k + 1)3
4
k2 + 4k + 4
= (k + 1)2
4
(k + 1)2 (k + 2)2
= ,
4
The triangular numbers are those numbers that can be described as a collection
of items placed in a triangle.
(2.17)
(2.18)
(2.19)
(2.20)
We note that the triangular numbers are the sum of the first n numbers. We can
determine a formula for this sum.
n n(n+1)
Theorem 2.33 Let n be a positive integer. Then j=1 j = 2 .
1(2)
Proof If n = 0, the left side of the equation is 1 and the right side is 2 = 1.
Therefore, the formula is correct for n = 1.
Assume the formula is correct for n and consider
n+1
n

j=( n) + (n + 1)
j=1 j=1
n(n + 1) 2(n + 1)
= +
2 2
n+2 (n + 1)(n + 2)
= (n + 1) = ,
2 2
the formula is correct for all positive n.
n
We note then that j=1 j = C(n + 1, 2). Additionally, we have the following
corollary.
Corollary 2.5 For any positive integer k, we have
k
k

3
n =( n)2 . (2.21)
n=1 n=1
Proof Follows directly from Theorem 2.32 and Theorem 2.33.
n(n+1)
This sequence can be described explicitly as f(n) = 2 . Recursively, we
can describe it as a0 = 0, an+1 = an + n + 1. We now prove that these two give
the same sequence.
n(n+1)
Theorem 2.34 The sequence f(n) = 2 matches the sequence given recur-
sively by a0 = 0, an+1 = an + n + 1.
Proof We shall prove the result by induction. If n = 0, f(0) = 0, and a0 = 0.

(n+1)(n+2) n(n+1) 2(n+1)
Next assume that f(n) = an . Then f(n + 1) = 2 = 2 + 2 =
an + (n + 1). Therefore, by the principle of mathematical induction, we have the
result.
Concretely, the triangular numbers are
0, 1, 3, 6, 10, 15, 21, 28, 36, 45, 55, . . .
Notice if we add each number to its successor in the sequence we get the following
sequence:
1, 4, 9, 16, 25, 36, 49, 64, 81, 100, . . .
which are the squares. We state this explicitly in the following theorem.
Theorem 2.35 The sum of two consecutive triangular numbers is a square.
Proof Let n be a positive integer. Then
n(n + 1) (n + 1)(n + 2) (n2 + n) + (n2 + 3n + 2)

+ =
2 2 2
2n2 + 4n + 2
= = n2 + 2n + 1 = (n + 1)2 .
2
We note also that this theorem says that any square number is the sum of two
triangular numbers.
We can also find the sum of the first k triangular numbers.

k
n(n + 1) n(n + 1)(n + 2)
= . (2.22)
2 6
n=1
Proof We shall prove the result by induction. If k = 1 the left side of the equation
1(2) 1(2)(3)
is 2 = 1 and the right side is 6 = 1 giving that the formula is correct for
k = 1.
Assume the formula is correct for k and consider the sum of the first k + 1
triangular numbers:
k+1
k
n(n + 1)
n(n + 1) (n + 1)(n + 2)
=( )+
2 2 2
n=1 n=1
n(n + 1)(n + 2) (n + 1)(n + 2)
= +
6 2
n(n + 2) + 3(n + 2)
= (n + 1)
6
(n + 1)(n + 2)(n + 3)
= ,
6

A perfect number is a number that is equal to the sum of its positive divisors that
are less than the number. For example, 6 = 1 + 2 + 3 and 28 = 1 + 2 + 4 + 7 + 14.
Therefore 6 and 28 are perfect numbers. It is not known if there exist odd perfect
numbers.
Even perfect numbers are related in a natural way to Mersenne primes. A Mersenne
prime is a prime number of the form 2p − 1 where p is a prime. We illustrate in the
following lemma why p must be prime.
Lemma 2.3 Let n be an integers with n > 1. If n is composite then 2n − 1 is

composite.
Proof Recall that xk − 1 = (x − 1)(1 + x + x2 + · · · + xk−1 ). If n = ab with

1 < a, b < n, then
2n − 1 = 2ab − 1 = (2a )b − 1.
Applying the factorization above we have
2n − 1 = (2a − 1)(1 + 2a + (2a )2 + · · · + (2a )b−1 ).
Since 1 < a, b < n, 1 < (2a − 1) < 2n − 1 and 1 < (1 + 2a + (2a )2 + · · · +

(2a )b−1 ) < 2n − 1, giving that 2n − 1 is composite.
The number 2p − 1 is not guaranteed to be prime when p is prime, but when it is

we have that (2p − 1)2p−1 is a perfect number. For example, if p = 2, 3, 5, 7, 13, 17,
s19, 31, then 2p − 1 is a prime. Generally, the largest known prime is a Mersenne
prime. This is because there is a much better algorithm for determining if 2p − 1 is
prime than there is for arbitrary n. This algorithm is known as the Lucas-Lehmer
test. As an example, the number 282,589,933 − 1 is known to be a prime number.
This number has millions of digits.
Lemma 2.4 If 2p − 1 is prime then (2p − 1)2p−1 is a perfect number.
Proof If 2p − 1 is prime then the divisors of (2p − 1)2p−1 are numbers of the form
(2p − 1)2i , 0 ≤ i ≤ p − 2 and 2j , 0 ≤ j ≤ p − 1. The sum of the first type is (2p −
1)(2p−1 − 1) and the sum of the second type is 2p − 1. Then (2p − 1)(2p−1 −
1) + (2p − 1) = (2p − 1)2p−1 and we have the result.
Example 2.14 Let p = 3, then 23 − 1 = 7 which is prime. Then (2p − 1)(2p−1 −

1) = 7(3) = 21 which is 7 + 14 and (2p − 1) = 7 which is 1 + 2 + 4. Therefore,
(2p − 1)2p−1 = 7(4) = 28 is a perfect number.
Theorem 2.37 The number (2p − 1)2p−1 is a triangular number.
n(n+1) (2p −1)2p

Proof Let n = 2p − 1. Then 2 = 2 = (2p − 1)2p−1 . Therefore,
(2p − 1)2p−1 is a triangular number.
Exercises
1. Determine a recursive formula for f(n) = n4 . Prove your formula is correct.

2. Prove that if p is a prime dividing ns , then ps divides ns .
3. Determine the possible final digit for a square number written in base 8.
4. Determine the possible final digit for a cubic number written in base 8.
2.3.2 Catalan Numbers
Catalan numbers were first described by Eugène Catalan in [22]. They have numerous
connections to a variety of combinatorial problems. We begin by defining the Catalan
numbers to be the numbers satisfying the following recursion:
n

Cn+1 = Ck Cn−k , C0 = 1. (2.23)
k=0
It follows that the first 8 Catalan numbers are the following.
n 0123 4 5 6 7 8
(2.24)
Cn 1 1 2 5 14 42 132 429 1430
We shall determine a closed formula for Cn as is given in Stanley’s text [81]. We

shall use generating functions to find the closed formula. First, we require a lemma
due to Newton.
Lemma 2.5 Let k ∈ R, n ∈ N. Then, for −1 < x < 1,

∞

(1 + x)k = C(k, n)xn , (2.25)
n=0
k(k−1)(k−2)···(k−n+1)
where C(k, n) = n! , and C(k, 0) = 1.
Proof The result is a direct consequence of Taylor’s Theorem.
We notice that this definition of C(k, n) matches the previous definition if k and
n are both natural numbers.
n.
Let p(x) = n≥0 Cn x Multiply the recurrence by xn and sum to obtain
n
Cn+1 xn = ( Ck Cn−k )xn . (2.26)
n≥0 n≥0 k=0
We note that

x Cn+1 xn = cn xn = p(x) − 1. (2.27)
n≥0 n≥1

Consider the series p(x)2 = ( n≥0 Cn x
n )2 . The coefficient of xn in this series
is
n

Ck Cn−k .
k=0
This follows simply from the definition of multiplication of series. Using this and
applying Eqs. 2.26 and 2.27 we have
p(x) − 1
= p(x)2 . (2.28)
x
Writing this as a binomial equation we have xp(x)2 − p(x) + 1 = 0.

Apply the quadratic equation to this equation with p(x) as the variable and we
have
1
1 ± (1 − 4x) 2
p(x) = .
2x
1
Now we can apply Lemma 2.5 to (1 − 4x) 2 (replacing x with 4x) for both the
choice of the plus sign and the minus sign. Using the plus sign gives p(x) = x1 −
1 − x + . . . which is not the series we are seeking. So we use the minus sign.
This gives that
1 1
p(x) = Cn xn = (1 − (1 − 4x) 2 )
2x
n≥0
1 1
= (1 − C( , n)(−4x)n )
2x 2
n≥0
1 1
= ( C( , n)(−1)n 4n xn )
2x 2
n≥1
1 1
= C( , n)(−1)n 4n xn−1
2 2
n≥1
(1(−1)(−3)(−5) · · · (−(2n − 3)) 1
= n
(−1)n 4n xn−1
n!2 2
n≥1
(1(1)(3)(5) · · · (2n − 3))

= 2n−1 xn−1
n!
n≥1
((1)(3)(5) · · · (2n − 1))
= 2n xn
(n + 1)!
n≥0
(2n)!
= xn .
n!(n + 1)!
n≥0
This gives the following result.
Theorem
2.38 Let n be a non-negative integer and let Cn satisfy the recursion
Cn+1 = n k=0 Ck Cn−k . Then
1
Cn = C(2n, n). (2.29)
n+1
Example 2.15 By the recursion in Eq. 2.23, we have
C8 = C0 C7 + C1 C6 + C2 C5 + C3 C4 + C4 C3 + C5 C2 + C6 C1 + C7 C0 .
Using the numbers in Eq. 2.24, we have
C8 = C0 C7 + C1 C6 + C2 C5 + C3 C4 + C4 C3 + C5 C2 + C6 C1 + C7 C0
= 1(429) + 1(132) + 2(42) + 5(14) + 14(5) + 42(2) + 132(1) + 429(1)
= 429 + 132 + 84 + 70 + 70 + 84 + 132 + 429
= 1430.
1
If n = 8, then n+1 C(2n, n) = 19 C(16, 8) = 19 (12870) = 1430.
Notice that we did not begin the discussion of Catalan numbers with a combina-
torial motivation, but rather we defined them via a recursive formula and from this
we obtained a closed formula for the Catalan numbers. Therefore, all we need to do
to show that these numbers solve a given combinatorial problem is to show that, that
problem can be solved with that recursion. In [81], Stanley gives 214 combinato-
rial applications of the Catalan numbers. Here we shall restrict ourselves to a much
smaller number of applications, pointing the interested reader to [81].
The first combinatorial application of the Catalan numbers is sometimes called a
ballot sequence. Consider a sequence ai of length 2n consisting of entries from the

set {1, −1} where each appears n times, with the property that k i=1 ai ≥ 0 for all
k with 1 ≤ k ≤ 2n.
For n = 1 there is one such sequence, 1, −1. For n = 2, there are two possible
sequences:
1, 1, −1, −1
1, −1, 1, −1.
For n = 3 there are five possible sequences:
1, 1, 1, −1, −1, −1
1, 1, −1, −1, 1, −1
1, 1, −1, 1, −1, −1
1, −1, 1, 1, −1, −1
1, −1, 1, −1, 1, −1
We see that these first few terms are the Catalan numbers. We prove this result.
Theorem 2.39 The number of sequences ai of length 2n consisting of entries from

the set {1, −1} where each appears n times, with the property that k
i=1 ai ≥ 0 for
all k, with 1 ≤ k ≤ 2n, is Cn .
Proof We begin by counting the number of sequences of length 2n with n elements

that are 1 and n elements that are −1. This number is the number of ways of choosing
(2n)!
n objects out of 2n objects which is C(2n, n) = n!n! .
Next, we need to count the number of such sequences that do not have non-negative
(2n)!
partial sums and subtract this number from n!n! .
Consider a sequence that has a partial sum that is negative. That is, si=1 ai < 0.
Let
tt be the smallest such s such that the partial sum is negative. Since t is minimal and
i=1 ai < 0, there must be an equal number of occurrences of 1 and −1 preceding
at which gives that t is odd. This means that there are t−1 2 occurrences of 1 before t
and n − t−1 2 and t−1
2 occurrences of −1 before t and n − t+1
2 since at must be −1.
If you form a new sequence bi of length 2n where bi = ai if i > t and bi = −ai
if i ≤ t, then the sum of this sequence must be 2 and have n + 1 occurrences of 1
(2n)!
and n − 1 occurrences of −1. The number of such sequences is (n+1)!(n−1)! .
Then we have
(2n)! (2n)! (2n)!(n + 1) (2n)!n
− = −
n!n! (n + 1)!(n − 1)! (n + 1)n!n! (n + 1)!(n − 1)!n
(2n!)
=
(n + 1)!n!
1
= C(2n, n) = Cn .
n+1

The next application involves triangulations. Consider the number of triangula-

tions of a convex (n + 2)-gon which was first studied by Euler. Equivalently, this
number is the number of n − 1 non-crossing diagonals of a convex polygon with
n + 2 sides. Consider the cases for n = 1, 2, 3 given in Fig. 2.1.
Fig. 2.1 Catalan numbers for n = 1, 2, and 3
This gives the number of such triangulations is 1 for n = 1, 2 for n = 2, and 5

for n = 3.
To find the five triangulations for n = 3, consider the one with n = 1. Opening
it and adjoining two new sides, each triangulated with the possibilities for n = 2,
and give the second and fifth triangulations for n = 3. The other three can be found
by opening the two for n = 2 and adjoining one new side with the triangulation for
n = 1. A straightforward generalization of this argument gives that the number of
triangulations of a convex (n + 2)-gon satisfies the desired recursion, which gives
the following.
Theorem 2.40 The number of triangulations of a convex (n + 2)-gon is Cn .

Exercises
1. Construct all 14 triangularization of hexagons.

2. Construct all ballot sequences for n = 4.
2.3.3 Stirling Numbers
Stirling numbers are named for James Stirling who wrote about them in [87].
Stirling numbers of the first kind counts the number
of ways to partition a set of
n
n things into k cycles. We denote this number by .
k
We can write a cycle as [1, 2, 3, 4, 5]. We notice that
[1, 2, 3, 4, 5] = [2, 3, 4, 5, 1] = [3, 4, 5, 1, 2] = [4, 5, 1, 2, 3] = [5, 1, 2, 3, 4].
We can always specify a single element from this equivalence class by distinguishing
a certain element to be the first in the representation of the cycle.
Example 2.16 Let n = 3 and k = 1. Then we shall count the number of ways that
the set {1, 2, 3} can be written in terms of one cycle. There are two ways, namely,
3
[1, 2, 3] and [1, 3, 2]. This gives that = 2.
1
Lemma 2.6 Let n be a positive integer, then

n
= (n − 1)!
1
and

n
= 1.
n
Proof Without loss of generality we can assume the symbols are 1, 2, 3, . . . , n.

Let k = 1. Since the elements are written in one cycle, every element is written
in that cycle. We can assume that we take the representation where 1 is the first
element in the cycle representation. Then every cycle is of the [1, σ] where σ is
form
n
a permutation of the remaining n − 1 elements. Therefore = (n − 1)!, since
1
there are (n − 1)! permutations on a set with n − 1 elements.
If k = n then they are written as n singleton cycles and there is only 1 way to do
this.
Theorem 2.41 Let n and k be positive integers. Then

n n−1 n−1
= (n − 1) + .
k k k−1
Proof Consider a set A with cardinality n + 1. Choose an element from the set,
call
n−1
it an+1 . Splitting into cycles puts an+1 in a cycle by itself, there such
k−1
There are n − 1 different
cycles. Otherwise an+1 is put into one of the k cycles.
n−1
places to insert an+1 . Hence, there are (n − 1) such cycles. This gives the
k
result.
Lemma 2.6 and Theorem 2.41 allow us to make a Pascal like triangle of the values
of the Stirling numbers of the first kind, where the rows are indexed by n and the
columns by k. We write it this way to make the recursion evident.
1
1 1
2 3 1
6 11 6 1
24 50 35 10 1
120 274 225 85 15 1
720 1764 1624 735 175 21 1
5040 13068 13132 6769 1960 322 28 1
40320 109584 118124 67284 22449 4536 546 36 1
.
.
.
Stirling numbers of the second kind counts the number of ways

topartition a set
n
of n things into k non-empty subsets. We denote this number by .
k
Example 2.17 Consider the set of cardinality 3, {1, 2, 3}. There is one partition into
3 sets, namely, {1}, {2}, {3}. There is one partition into 1 set, namely, {1, 2, 3}. There
are three partitions into 2 sets, namely,
{1, 2}, {3}

{1, 3}, {2}
{2, 3}, {1}.

3 3 3
This gives that = = 1 and = 3.
1 3 2
Lemma 2.7 Let n and k be non-negative integers, then

n n
= = 1. (2.30)
1 n
Proof There is only one way of partitioning a set of cardinality n into sets of cardi-
n
nality n, namely, the set itself, therefore = 1.
n
There is only one way of partitioning a set of cardinality n into sets of cardinality

n
1, namely each element of the set is in its own set of the partition, therefore = 1.
1

We now give the recursion for Stirling numbers of the second kind.
Theorem 2.42 Let n and k be non-negative integers, then

n n−1 n−1
=k + . (2.31)
k k k−1
Proof Consider a set A with cardinality n + 1. Chose an element from the set, call
it an+1 . Any partition of the set of cardinality n, A − {an+1 }, into k − 1 sets can
madea partition of A into k sets by adjoining {an+1 } to the partition. There are
be
n−1
such partitions. Any partition of A − {an+1 } into k sets can be made a
k−1
of A into k sets by placing an+1 in any of the k sets of the partition. There
partition
n−1
are k such ways of accomplishing this. This gives the result.
k
Lemma 2.7 and Theorem 2.42 allow us to make a Pascal like triangle of the values
of the Stirling numbers of the second kind, where the rows are indexed by n and the
columns by k. We write it this way to make the recursion evident.
1
1 1
1 3 1
1 7 6 1
1 15 25 10 1
1 31 90 65 15 1
1 63 301 350 140 21 1
1 127 966 1701 1050 266 28 1
1 255 3025 7770 6951 2646 462 36 1
.
.
.
Exercises

n n
1. Prove that = where k = n or k = n − 1. Determine the values in
k k
these cases.
2. Prove that triangular numbers are Stirling numbers of the second kind.
2.3.4 Towers of Hanoi
The problem of the Towers of Hanoi was first presented by Édouard Lucas in 1883
[60]. The problem involves a collection of disks of different sizes and three poles.
The disks are placed on the pole.
Objective: Move the entire stack of disks from one pole to another.
Rules:
1. A move consists of taking the top disk from one of the poles and putting it on the
top of another pole.
2. Only one disk can be moved at a time.
3. A larger disk can never be placed on a smaller disk.
Example 2.18 Consider the case with two disks, which we shall call 1 and 2, where
a larger number indicates a larger disk. We shall move the disks to the second pole.
1 1
−→ −→ −→
2 2 1 2 1 2
We see that the it takes 22 − 1 = 3 moves to complete the transfer.
Example 2.19 Consider the case with three disks, which we shall call 1, 2, and 3,
where a larger number indicates a larger disk. We shall move the disks to the second
pole.
1
2 −→ 2 −→ −→ 1 −→
3 3 1 3 1 2 3 2
1
1 −→ −→ 2 −→ 2
3 2 1 3 2 1 3 3
We see that it takes 23 − 1 = 7 moves to complete the transfer.
These two examples lead us to the following theorem.
Theorem 2.43 Let an be the number of moves required to move n disks from one
pole to another. Then, recursively
an+1 = 2an + 1, a1 = 1 (2.32)
and an = 2n − 1.
Proof First, we notice that a1 = 1 since it requires 1 move to move a single disk
from one pole to another.
Next, consider the disks labeled 1, 2, . . . , n. If they are on the first pole, it takes
an−1 moves to move 1, 2, . . . , n − 1 to the third pole. Then n can be moved to
the second pole. Then it takes an−1 moves 1, 2, . . . , n − 1 to the second pole. This
gives the recurrence an+1 = 2an + 1.
Finally, we show that this recurrence gives the closed formula. We note that if
n = 1, a1 = 1, and 21 − 1 = 1.
Then if an−1 = 2n−1 − 1, then an = 2an−1 + 1 = 2(2n−1 − 1) + 1 = 2n −
2 + 1 = 2n − 1. Therefore, by the principal of mathematical induction an = 2n − 1
for all n.
Exercises
1. Assume you have n disks and k poles, with k > n. Prove that the number of
moves to the disks is 2n − 1.
2. Write an algorithm to implement the solution to the Towers of Hanoi problem.
Mutually Orthogonal Latin Squares
3
3.1 36 Officers and Latin Squares
In 1782, the great mathematician Leonhard Euler asked the following question: Can
you arrange 36 officers of 6 ranks and 6 regiments in a 6 by 6 square so that each
row and column contains each rank and regiment exactly once?
Euler asked the question in his paper “Recherches sur une nouvelle espace de
quarré magique” [35]. This paper began the study of Latin squares and was one
of the most important papers in the history of combinatorics. It is fairly long by
mathematical standards but it is written in a style which has completely vanished
from mathematical publications. He wrote as if he were explaining the material to a
good friend who was sitting at his side. There is no intimidation here, no hopelessly
complicated language or notation. Euler’s writing, like his mathematics, is a delight.
While Euler’s work was the one that prompted the study of Latin squares, they had
been defined earlier by the Korean mathematician Choi Seok-Jeong and the French
mathematician Jacques Ozanam. The work of Choi Seok-Jeong was particularly
interesting and inventive, defining orthogonality and constructing magic squares. He
even gave practical applications for the ideas, but this work was largely forgotten. The
work of Jacques Ozanam related Latin squares to a problem about playing cards. See
[74], for a complete description of who deserves recognition for introducing Latin
squares.
This question seems remarkably easy to answer. At first glance most people seem
to think that they could be arranged in such a manner. Unfortunately, it proved to
be quite a difficult problem. It was not solved until 1901 when Col. Tarry, a colonel
in the French army in Algeria, and an amateur mathematician, showed that such an
arrangement was, in fact, impossible [90]. His solution was simply to go through
all possible arrangements by hand and see that none would work. While there are
some shortcuts which can be better understood after reading the explanation of the
problem which will follow, his proof was essentially an exhaustive search. In modern
76 3 Mutually Orthogonal Latin Squares
parlance, it would be known as a computer proof, although computers were not yet
in existence. To many in the pursuit of mathematical truth such proofs are highly
undesirable. This kind of inelegant proof does not explain why something is true.
A refinement of this proof was made by Fisher and Yates in 1934 [37], in which
they significantly reduced the number of Latin squares to consider by subdividing
the set of all Latin squares into classes. So as not to leave the reader in suspense,
there are proofs which solve the problem without any resort to an exhaustive search.
The interested reader can find them in [28,84]. Additionally, there is a letter from
the astronomer Heinrich Schumacher to Carl Gauss earlier than Tarry’s proof, that
his assistant Thomas Clausen had a solution to the problem, but this solution was
never published or made known. See [74] for a complete description.
In an attempt to solve the problem, Euler decided to denote the ranks and regiments
by Greek and Latin letters. Specifically, the ranks can be A, B, C, D, E, F and the
regiments can be α, β, γ, δ, , ζ. Listing just the ranks in a 6 by 6 square with each
row and column containing each rank exactly once produced a Latin square. For
example, the following is a Latin square:
⎛ ⎞
A B CD E F
⎜B A D C F E⎟
⎜ ⎟
⎜C D E F A B⎟
⎜ ⎟
⎜D C F E B A⎟.
⎜ ⎟
⎝ E F A B C D⎠
F E B ADC
To solve the problem you would need to find a corresponding Greek square such that
when overlapped with this square all possible pairs would appear.
Let us look at a smaller example. Can you arrange 9 officers of 3 ranks and
regiments in a 3 by 3 square so that each row and column contains each rank and
regiment exactly once? Now the ranks are A, B, C and the regiments are α, β, γ.
The following is a Latin square of order 3. Saying it is of order 3 means it is 3 by
3, which would mean that the above square was of order 6.
⎛ ⎞
ABC
⎝C A B⎠. (3.1)
BCA
The following is a Greek square of order 3.

⎛ ⎞
αβγ
⎝β γ α⎠. (3.2)
γαβ
Overlaying them gives the following Graeco-Latin square.

⎛ ⎞
Aα Bβ Cγ
⎝ Cβ Aγ Bα ⎠ . (3.3)
Bγ Cα Aβ
3.1 36 Officers and Latin Squares 77
Notice that each pair appears exactly once, so the answer for order 3 is that they
can be arranged. It may seem that the solution to order 3 was quite simple. The Latin
square was formed by listing the ranks as the first row and then cycling them to the
right. The Greek square was formed by listing the regiments as the first row and then
cycling them to the left.
We can try the same with 4. The Latin square would be
⎛ ⎞
A B CD
⎜D A B C ⎟
⎜ ⎟ (3.4)
⎝C D A B⎠
B CDA
and the Greek square would be

⎛ ⎞
αβγ δ
⎜β γ δ α⎟
⎜ ⎟ (3.5)
⎝γ δ α β⎠.
δαβγ
The Graeco-Latin square would be

⎛ ⎞
Aα Bβ Cγ Dδ
⎜ Dβ Aγ Bδ Cα ⎟
⎜ ⎟ (3.6)
⎝ Cγ Dδ Aα Bβ ⎠ .
Bδ Cα Dβ Aγ
Notice that Aα appears twice but Aβ never appears. Similar problems occur
throughout this square. This technique of cycling in opposite directions works if
the order is an odd number but it does not work if the order is an even number. In
particular, it does not help us to answer the 36 officer problem.
There is a way to arrange 16 officers. In fact you can do more. The three squares
that follow can be put together any two at a time to make a Graeco-Latin square (with
the provision of changing the numbers to either Latin or Greek letters). In general,
we simply use the digits between 0 and 1 less than the order of the square as elements
(or the numbers between 1 and the order) to simplify matters, but it is aesthetically
pleasing to use different letters.
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
A B C D αβγ δ 0123
⎜B AD C⎟ ⎜γ δ α β⎟ ⎜3 2 1 0⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝C DA B⎠ ⎝ δ γ β α⎠ ⎝1 0 3 2⎠
D C B A βαδ γ 2301
⎛ ⎞
Aα0 Bβ1 Cγ2 Dδ3
⎜ Bγ3 Aδ2 Dα1 Cβ0 ⎟
⎜ ⎟.
⎝ Cδ1 Dγ0 Aβ3 Bα2 ⎠
Dβ2 Cα3 Bδ0 Aγ1
When two Latin squares can be overlapped so that each pair appears exactly once
then we call the squares orthogonal. If we have a collection of squares so that any
two of them are orthogonal then we say that we have mutually orthogonal squares
or MOLS for short.
We shall now make rigorous definitions of these ideas.
Definition 3.1 Let A be an alphabet of size n. Then an n by n array L is a Latin

square of order n if each element of A appears exactly once in every row and every
column.
Example 3.1 Let A be the alphabet {a, b} then the following is a Latin square of
order 2:

ab
.
ba
Definition 3.2 If L = (ij ) is a Latin square of order n over an alphabet A and

M = (mij ) is a Latin square of order n over an alphabet B then L and M are
said to be orthogonal if every pair of A × B occurs exactly once in {(ij , mij )}.
If L1 , L2 , . . . , Lk are Latin squares of order n then they are a set of k Mutually
Orthogonal Latin Squares (MOLS) if Li and Lj are orthogonal for all i, j, i = j.
Hence, a Graeco-Latin square can be realized as 2 MOLS.

Latin squares where each row is determined by cycling each element one space
to the right (left), as was seen in the squares denoted given in (3.1), (3.2), (3.4), and
(3.5), are known as right (left) circulant Latin squares.
We shall show that cycling does not produce orthogonal squares when n is even.
Theorem 3.1 A left circulant Latin square and a right circulant Latin square are
not orthogonal when n is even.
Proof Without loss of generality let A = Zn = {0, 1, 2, . . . , n − 1} and use as the

first row 0 1 . . . (n − 1). Using this alphabet allows us to use the arithmetic we
have already defined for Zn . Let L be the left circulant square and R be the right
circulant square. It is easy to see that Li,j = i + j and Ri,j = j − i. Let n = 2k.
Consider the row Lk,j = k + j and Rk,j = j − k. Notice that in Z2k , we have that
k = −k. This gives that j + k = j − k and so Lk,j = Rk,j . Hence (a, a) appears
when overlapping the k-th row for each a and in the top row when i = 0. Since these
pairs appear twice, the two squares are not orthogonal.
Example 3.2 For n = 6 the left and right circulant Latin squares are as follows:
⎛ ⎞ ⎛ ⎞
012345 012345
⎜1 2 3 4 5 0⎟ ⎜5 0 1 2 3 4⎟
⎜ ⎟ ⎜ ⎟
⎜2 3 4 5 0 1⎟ ⎜4 5 0 1 2 3⎟
⎜ ⎟ ⎜ ⎟
⎜3 4 5 0 1 2⎟ ⎜3 4 5 0 1 2⎟.
⎜ ⎟ ⎜ ⎟
⎝4 5 0 1 2 3⎠ ⎝2 3 4 5 0 1⎠
501234 123450
Overlapping the squares gives us the following:

⎛ ⎞
00 11 22 33 44 55
⎜ 15 20 31 42 53 04 ⎟
⎜ ⎟
⎜ 24 35 40 51 02 13 ⎟
⎜ ⎟
⎜ 33 44 55 00 11 22 ⎟ .
⎜ ⎟
⎝ 42 53 04 15 20 31 ⎠
51 02 13 24 35 40
Notice that this is not a Graeco-Latin square as 00 appears twice but 01 does not
appear at all. In fact, every pair that appears, does so twice, and half of the possible
pairs do not appear at all.
Notice that when n = 2 the two squares are actually identical, that is,

01
L=R= .
10
Theorem 3.2 A left circulant Latin square and a right circulant Latin square with
the same first row are orthogonal if the order of the square is odd.
Proof Again, as in the previous theorem, without loss of generality let A = Zn =

{0, 1, 2, . . . , n − 1}, and we let L be the left circulant square and R be the right circu-
lant square. We have Li,j = i + j and Ri,j = j − i. If (i + j, j − i) = (i + j , j −
i ) then i + j = i + j and j − i = j − i . Adding these equations we get 2j = 2j .
Since n is odd we have that 2 is a unit in Zn , that is, 2 has a multiplicative inverse.
This gives that j = j . Using i + j = i + j we have that i = i . This gives that L
and R are orthogonal.
Example 3.3 For n = 5, the left and right circulant Latin squares are as follows:
⎛ ⎞ ⎛ ⎞
01234 01234
⎜1 2 3 4 0⎟ ⎜4 0 1 2 3⎟
⎜ ⎟ ⎜ ⎟
⎜2 3 4 0 1⎟ ⎜3 4 0 1 2⎟.
⎜ ⎟ ⎜ ⎟
⎝3 4 0 1 2⎠ ⎝2 3 4 0 1⎠
40123 12340
Overlapping the squares gives us the following:

⎛ ⎞
00 11 22 33 44
⎜ 14 20 31 42 03 ⎟
⎜ ⎟
⎜ 23 34 40 01 12 ⎟ .
⎜ ⎟
⎝ 32 43 04 10 21 ⎠
41 02 13 24 30
Notice that this is a Graeco-Latin square as each possible pair appears exactly once.
This shows that there is always at least 2 MOLS of order n where n is odd.
Theorem 3.3 The maximum number of MOLS of order n is n − 1.
Proof Assume that the symbols are 0, 1, . . . , n − 1 for each square. Without loss of
generality, we assume that the first row is
0 1 2 ... n − 1.
(If it is not, we simply rename the elements.) There are n − 1 choices for the first
element of the second row, since it cannot be 0. In each square, these must be distinct
since each element appears already with itself in the first row. This means we can
have at most n − 1 MOLS of order n.
For example, if the order is 5 then the first row of the squares can be made to be
0 1 2 3 4.
There can not be a 0 in the first coordinate of the second row for any square since
it already appears in that column. If any number, say 2, were to appear twice in that
place then (2, 2) would appear both in the first row and in the second when those
two squares were overlapped and the squares would not be orthogonal.
When there are n − 1 MOLS of order n we say that there is a complete set of
MOLS. In other words, we have the largest number of MOLS we can possibly have.
Example 3.4 Consider n = 3.

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
012 012 0 12
⎝1 2 0⎠ ⎝2 0 1⎠ ⎝ ⎠.
201 120
The first two squares are orthogonal. In the third square, there is no possible entry
to place in the box since 1 and 2 have already been used in the first two squares.
The fundamental question of Latin squares is the following.

Fundamental Question: What is the maximum number of Mutually Orthogonal
Latin Squares of order n, for any n?
Let us return to the original problem. The way Col. Tarry solved the problem was
to write out all possible Latin squares using 1, 2, 3, 4, 5, 6 realizing that the first row
could be
123456
Then he took each one and overlapped it with every other one until he had tested them
all in this manner. He then reported that none of the squares were orthogonal. Today,
of course, this could be done fairly easily with a computer. The first open order is
10, meaning the maximum number of MOLS of order 10 is unknown although we
do know of a pair of MOLS. The reader may ask why not just do it with a computer.
The answer is that there are far too many for any computer.
Euler had conjectured that there would be no Graeco-Latin square of order 6 and,
as usual, he was right. He also conjectured that there would be no Graeco-Latin
square of order for any even number that was not divisible by 4, that is,
2, 6, 10, 14, 18, 22, 26, . . . .
It is obviously true for 2 but Euler was uncharacteristically wrong for 10, 14, . . . .
This was not known until 1960 [11,12] when Bose, Shrikhande, and Parker showed
how to construct 2 MOLS of any desired order of this type greater than 6. In fact,
other than 2 and 6 there is a Graeco-Latin square of any order. We shall prove this
later.
Here is the Graeco-Latin square of order 10 that they found.
⎛ ⎞
αA θE ιB ηH κC δJ ζI D βG γF
⎜ ηI βB θF ιC αH κD J ζE γA δG ⎟
⎜ ⎟
⎜ ζJ αI γC θG ιD βH κE ηF δB A ⎟
⎜ ⎟
⎜ κF ηJ βI δD θA ιE γH αG C ζB ⎟
⎜ ⎟
⎜ δH κG αJ γI E θB ιF βA ζD ηC ⎟
⎜ ⎟. (3.7)
⎜ ιG H κA βJ δI ζF θC γB ηE αD ⎟
⎜ ⎟
⎜ θD ιA ζH κB γJ I ηG δC αF βE ⎟
⎜ ⎟
⎜ B ζC ηD αE βF γG δA θH ιI κJ ⎟
⎜ ⎟
⎝ βC γD δE F ζG ηA αB κI θJ ιH ⎠
γE δF G ζA ηB αC βD ιJ κH θI
At present, it is unknown if there exists a complete set of MOLS of order n where
n is not a prime power.
Theorem 3.4 A complete set of MOLS exists of order n if there is a field of order n.
Proof Let L be the addition table for a field F = {a0 , a1 , . . . , an−1 }, i.e. Li,j =
ai + aj . Then let L1 = L and let Lk be formed by replacing the h-th row of L with
the ak ah row of L. We know that L is a Latin square so permuting the rows does
not affect that it is a Latin square, specifically each row and column still has every
element exactly once.
Assume (Lb c b c
i,j , Li,j ) = (Li ,j , Li ,j ) where b = c. Then we have
ab ai + aj = ab ai + aj
ac ai + aj = ac ai + aj .
Subtracting these equations we get
(ab − ac )ai = (ab − ac )ai . (3.8)
Since b = c then ab − ac has a multiplicative inverse since we are in a field,

so ai = ai giving i = i . It follows immediately that aj = aj and hence the two
squares are orthogonal. This gives that L1 , L2 , . . . , Ln−1 are a set of n − 1 MOLS
of order n and hence are complete.
Example 3.5 We exhibit the six MOLS of order 7 using Theorem 3.4.
⎛ ⎞ ⎛ ⎞
0123456 0123456
⎜1 2 3 4 5 6 0⎟ ⎜2 3 4 5 6 0 1⎟
⎜ ⎟ ⎜ ⎟
⎜2 3 4 5 6 0 1⎟ ⎜4 5 6 0 1 2 3⎟
⎜ ⎟ ⎜ ⎟
L1 = ⎜ ⎟ 2 ⎜ ⎟
⎜3 4 5 6 0 1 2⎟ L = ⎜6 0 1 2 3 4 5⎟
⎜4 5 6 0 1 2 3⎟ ⎜1 2 3 4 5 6 0⎟
⎜ ⎟ ⎜ ⎟
⎝5 6 0 1 2 3 4⎠ ⎝3 4 5 6 0 1 2⎠
6012345 5601234
⎛ ⎞ ⎛ ⎞
0123456 0123456
⎜3 4 5 6 0 1 2⎟ ⎜4 5 6 0 1 2 3⎟
⎜ ⎟ ⎜ ⎟
⎜6 0 1 2 3 4 5⎟ ⎜1 2 3 4 5 6 0⎟
⎜ ⎟ ⎜ ⎟
L3 = ⎜ ⎟ 4 ⎜ ⎟
⎜2 3 4 5 6 0 1⎟ L = ⎜5 6 0 1 2 3 4⎟
⎜5 6 0 1 2 3 4⎟ ⎜2 3 4 5 6 0 1⎟
⎜ ⎟ ⎜ ⎟
⎝1 2 3 4 5 6 0⎠ ⎝6 0 1 2 3 4 5⎠
4560123 3456012
⎛ ⎞ ⎛ ⎞
0123456 0123456
⎜5 6 0 1 2 3 4⎟ ⎜6 0 1 2 3 4 5⎟
⎜ ⎟ ⎜ ⎟
⎜3 4 5 6 0 1 2⎟ ⎜5 6 0 1 2 3 4⎟
⎜ ⎟ ⎜ ⎟
L5 = ⎜ ⎟ 6 ⎜ ⎟
⎜1 2 3 4 5 6 0⎟ L = ⎜4 5 6 0 1 2 3⎟.
⎜6 0 1 2 3 4 5⎟ ⎜3 4 5 6 0 1 2⎟
⎜ ⎟ ⎜ ⎟
⎝4 5 6 0 1 2 3⎠ ⎝2 3 4 5 6 0 1⎠
2345601 1234560
These six Latin squares form a complete set of MOLS of order 7.
Definition 3.3 A magic square of order n is an n by n matrix where each element

of the set {0, 1, 2, . . . , n2 − 1} appears once and the sum of each row and column is
the same.
Example 3.6 The following is an example of a magic square of order 3:

⎛ ⎞
048
⎝5 6 1⎠.
723
Here the elements 0, 1, 2, . . . , 8 each appear once and the sum of every row and
column is 12.
Theorem 3.5 Let L and K be orthogonal Latin squares of order n over the alphabet
Zn . Then the square matrix M formed by
Mi,j = Li,j + nKi,j , (3.9)
where the operations are done in the integers, is a magic square of order n and the
(n−1)n(n+1)
sum of the elements of any row or column of M is 2 .
Proof In every row or column each element of Zn appears exactly once as a and
once as b in the set {a + bn}. Hence the sum of the row is
n−1
n−1
(n − 1)n (n − 1)n (n − 1)n
( i)n + ( i) = n+ = (n + 1).
2 2 2
i=1 i=1
Since the squares are orthogonal each pair (a, b) appears exactly once in their over-
lap, hence each number from 0 to n2 − 1 appears once in the set {a + bn}.
Example 3.7 As an example consider the following orthogonal Latin squares and
the magic square they form.
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
01234 01234 0 6 12 18 24
⎜1 2 3 4 0⎟ ⎜ 4 0 1 2 3 ⎟ ⎜ 21 2 8 14 15 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ 2 3 4 0 1 ⎟ + 5 ⎜ 3 4 0 1 2 ⎟ = ⎜ 17 23 4 5 11 ⎟ . (3.10)
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝3 4 0 1 2⎠ ⎝ 2 3 4 0 1 ⎠ ⎝ 13 19 20 1 7 ⎠
40123 12340 9 10 16 22 3
Each element from 0 to 24 appears once and the sum of each row and column is
4(5)(6)
2 = 60.
We shall now define how to take the direct product of two Latin squares. Let
A = (Aij ) and B = (Bij ) be Latin squares of order n and m, respectively. Define
⎛ ⎞
A0,0 B
A0,1 B . . . A0,n−1 B
⎜ A1,1 B . . . A1,n−1 B ⎟
A1,0 B
⎜ ⎟
A×B=⎜ .. ⎟, (3.11)
⎝ . ⎠
An−1,0 B An−1,1 B . . . An−1,n−1 B
where (i, j) is rewritten as im + j.

As an example, consider the following cross product of a Latin square of order 2
with a Latin square of order 3.
⎛ ⎞
012345
⎛ ⎞ ⎜2 0 1 5 3 4⎟
012 ⎜ ⎟
01 ⎜1 2 0 4 5 3⎟
⎝ ⎠
× 201 =⎜ ⎜ ⎟. (3.12)
10 ⎜ 3 4 5 0 1 2⎟
⎟
120 ⎝5 3 4 2 0 1⎠
453120
Theorem 3.6 If L and L are orthogonal Latin squares of order n and M and M
are orthogonal Latin squares of order m then L × M and L × M are orthogonal
Latin squares of order nm.
Proof The proof is straightforward and is left as an exercise. See Exercise 10.
This theorem is enough to tell us a great deal about the existence of a pair of
MOLS.
The following is a direct consequence of Theorem 2.19.
Lemma 3.1 There exist finite fields of every prime power order.
This lemma implies, by Theorem 3.4, that there is always a complete set of MOLS
for pe where p is a prime.
Theorem 3.7 If n ≡ 2 (mod 4) then there is a pair of MOLS of order n.

ei
Proof If n = pi where pi is a prime and pi = pj for i = j then we know there
exists MOLS of every order as long as the exponent of 2 is not 1, i.e., as long as n ≡ 2
(mod 4). Then we take the direct product of these MOLS to produce 2 MOLS of
order n by Theorem 3.6.
Given this theorem it is easy to see why Euler conjectured that there would be no
set of MOLS when n ≡ 2 (mod 4).
Example 3.8 Let n = 588 = 22 · 3 · 7. There exists a pair of MOLS of order 4, 3,

and 72 by Theorem 3.4. Then by applying Theorem 3.6 twice, we have that there is
a pair of MOLS of order 588.
We shall examine some ideas used in studying the construction of MOLS. Given
a Latin square L a transversal is a set of n coordinates in the square such that each
symbol appears once in the set and the set intersects each row and column exactly
once. The bold elements form a transversal in the following Latin square.
⎛ ⎞
12043
⎜2 1 3 0 4⎟
⎜ ⎟
⎜0 4 1 3 2⎟. (3.13)
⎜ ⎟
⎝4 3 2 1 0⎠
30421
An orthogonal mate is a set of n mutually disjoint transversals. To show a square
has no orthogonal mates, it is sufficient to show it has no transversals or even that it
has no transversals through a particular coordinate.
The following Latin square has no transversals.
⎛ ⎞
0123
⎜1 2 3 0⎟
⎜ ⎟ (3.14)
⎝2 3 0 1⎠.
3012
Theorem 3.8 Given a set of k-MOLS L1 , L2 , . . . , Lk , if there is a Latin square L

such that L is orthogonal to each Li , then each set of coordinates corresponding to
a symbol in L is a transversal to Li for each i, 1 ≤ i ≤ k.
Proof The set of coordinates of each symbol must be a transversal to Li , since Li is

orthogonal to L.
Given a Latin square, a k-transversal is a set of kn coordinates in the square

such that each symbol appears k times in the set and the set intersects each row
and column k times. (These objects can also be called a k-plex.) The union of k
transversals will form a k-transversal but there exist k transversals that are not the
union of transversals. For example, the previous Latin square has the following 2-
transversal.
⎛ ⎞
0123
⎜1 2 3 0⎟
⎜ ⎟ (3.15)
⎝2 3 0 1⎠.
3012
There are two important conjectures concerning transversals. They are as follows.
Conjecture 3.1 Every Latin square of odd order has a transversal.
Conjecture 3.2 Every Latin square of order n has n

2 disjoint 2-transversals.
An extremely large literature exists about Latin squares and their generalizations.
A good first reference would be the text by Denes and Keedwell, Latin Squares and
Applications [25] published first in 1974 (a second edition appeared in 2015). A
glance at the text indicates how much was known about Latin squares and how many
applications there were for Latin squares. Since the publication of [25], the number
of papers studying Latin squares has grown immensely. For a more recent text, see
Discrete Mathematics using Latin Squares, by Laywine and Mullen, [59].
Exercises
1. Prove that the addition table of a finite field is a Latin square.

2. Determine if the following Latin square has any 1-transversals or 2-transversals:
⎛ ⎞
12340
⎜2 1 0 3 4⎟
⎜ ⎟
⎜3 4 2 0 1⎟. (3.16)
⎜ ⎟
⎝4 0 1 2 3⎠
03412
3. Prove that the multiplication table of non-zero elements in a finite field is a Latin
square.
4. Prove that if there exists n − 2 MOLS of order n then there exists a unique Latin
square that completes the set to n − 1 MOLS.
5. Prove that any Latin square of order 3 has an orthogonal mate.
6. Construct a magic square from the orthogonal Latin squares of order 10.
7. Find four MOLS of order 5. The first is given below (3.17) and the other 3 can
be found by permuting the order of the rows of the first.
⎛ ⎞
01234
⎜1 2 3 4 0⎟
⎜ ⎟
⎜2 3 4 0 1⎟. (3.17)
⎜ ⎟
⎝3 4 0 1 2⎠
40123
8. Find all possible Latin squares of order 4 where the first row is 1, 2, 3, 4.
9. Show that over Zn , the square Lij = ai + j is a Latin square if and only if
gcd(a, n) = 1.
10. Prove that the cross product of Latin squares is a Latin square and prove Theo-
rem 3.6.
11. Let L and M be two orthogonal Latin squares of order 3. Compute L × L and
M × M and verify that the two Latin squares of order 9 are orthogonal.
3.2 Forming Latin Squares 87
12. Construct 2 MOLS of order 12 from the MOLS of order 3 and 4.

e e
13. Determine the minimum number of MOLS that can exist of order p11 p22 , where
p1 , p2 are distinct primes and ei > 0.
3.2 Forming Latin Squares
Our next task will be to find ways to construct Latin squares from existing Latin
squares and how to group them together.
Let L = (Lij ) be a Latin square of order n. Then the matrix formed by reversing
the roles of rows and columns is again a Latin square. This matrix is called the
transpose of the matrix and is denoted by Lt , namely, Lt = (Lji ).
Lemma 3.2 If L = (Lij ) is a Latin square of order n, then Lt = (Lji ) is a Latin

square of order n.
Proof Each row and each column of the transpose contains each element of the
alphabet exactly once since each column and each row contains each element exactly
once.
As an example, consider the following Latin square and its transpose:

⎛ ⎞ ⎛ ⎞
0123 0231
⎜2 0 3 1⎟ ⎜1 0 2 3⎟
L=⎜ ⎟ t ⎜ ⎟
⎝3 2 1 0⎠, L = ⎝2 3 1 0⎠. (3.18)
1302 3102
Notice that these Latin squares may not be distinct. The following is equal to its
transpose.
⎛ ⎞
0123
⎜1 0 3 2⎟
L=⎜ ⎝2 3 0 1⎠.
⎟ (3.19)
3210
Any Latin square that is equal to its transpose, L = Lt , is called a symmetric Latin
square.
Theorem 3.9 If L is a Latin square of order n with M orthogonal to L, then Mt is

orthogonal to Lt .
Proof If {(Li,j , Mi,j ) | 1 ≤ i, j ≤ n} gives all possible ordered pairs then it is imme-
diate that {(Lj,i , Mj,i ) | 1 ≤ i, j ≤ n} gives all possible ordered pairs since it is the
same set. This gives the result.
Example 3.9 The following two Latin squares are orthogonal:

⎛ ⎞ ⎛ ⎞
01234 01234
⎜2 3 4 0 1⎟ ⎜3 4 0 1 2⎟
⎜ ⎟ ⎜ ⎟
L = ⎜4 0 1 2 3⎟ M = ⎜
⎜ ⎟ ⎟
⎜1 2 3 4 0⎟.
⎝1 2 3 4 0⎠ ⎝4 0 1 2 3⎠
34012 23401
Their transposes are as follows and they are also orthogonal:

⎛ ⎞ ⎛ ⎞
02413 03142
⎜1 3 0 2 4⎟ ⎜1 4 2 0 3⎟
⎜ ⎟ ⎜ ⎟
Lt = ⎜
⎜ 2 4 1 3 0 ⎟ Mt = ⎜ 2 0 3 1 4 ⎟ .
⎟ ⎜ ⎟
⎝3 0 2 4 1⎠ ⎝3 1 4 2 0⎠
41302 42031
This exercise shows that if we are looking for squares that have an orthogonal mate
we need not consider the transpose after we have considered the original square. This
significantly reduces the number we have to consider (almost by half, remembering
that some squares are equal to their transpose). This is the type of reasoning that
Fisher and Yates used so well to reduce the computation in showing that there are
not two MOLS of order 6. We shall examine other ways of grouping squares together.
First, we shall show an alternate representation of a Latin square. Suppose we
have the first Latin square given in (3.18) where the rows and columns are indexed
by Z4 = {0, 1, 2, 3}. Then we can simply label the coordinates and the entry in those
coordinates. We use the first coordinate to label the row, the second to label the
column, and the third to label entry. For this square we would have
⎛ ⎞
R CS
⎜0 0 0⎟
⎜ ⎟
⎜0 1 1⎟
⎜ ⎟
⎜0 2 2⎟
⎜ ⎟
⎜0 3 3⎟
⎜ ⎟
⎜1 0 2⎟
⎜ ⎟
⎜1 1 0⎟
⎜ ⎟
⎜1 2 3⎟
⎜ ⎟
⎜1 3 1⎟ (3.20)
⎜ ⎟.
⎜2 0 3⎟
⎜ ⎟
⎜2 1 2⎟
⎜ ⎟
⎜2 2 1⎟
⎜ ⎟
⎜2 3 0⎟
⎜ ⎟
⎜3 0 1⎟
⎜ ⎟
⎜3 1 3⎟
⎜ ⎟
⎝3 2 0⎠
3 3 2
This is known as an orthogonal array. For a detailed survey of combinatorial

structures, which are equivalent to MOLS, see [53]. It is easy to see that we can
add additional columns for Latin squares orthogonal to the first and get a larger
orthogonal array but for right now we simply want to look at the representation of
one Latin square.
If we reverse the first and second columns we simply get the transpose of the
matrix. It should be clear that we can switch any two columns of this array and still
have a Latin square. If we switch the first and the third, that is, change the roles of
rows and symbols, then we call the resulting Latin square the row adjugate of the
original and denote it by Lr . If we switch the second and the third, that is, change
the roles of columns and symbols, then we call the resulting Latin square the column
adjugate of the original and denote it by Lc .
Lemma 3.3 If L is a Latin square of order n then Lc and Lr are Latin squares of
order n.
Proof See Exercise 5.
We have three operations described here which can be done in any order which
correspond to the 6 different ways of arranging the three columns of the orthogonal
array. Basically, we are permuting the set {R, C, S}. If we rename this set as {1, 2, 3}
we can use the notation previously given for permutations. Namely, we have the
following correspondence.
• The identity corresponds to the original Latin square.

• (1, 2) corresponds to the transpose.
• (1, 3) corresponds to the row adjugate.
• (2, 3) corresponds to the column adjugate.
• (1, 2, 3) corresponds to the transpose of the column adjugate.
• (1, 3, 2) corresponds to the transpose of the row adjugate.
Example 3.10 We shall examine all 6 squares formed by these operations on the
Latin square of order 4 given in (3.18).
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0123 0231 0132
⎜2 0 3 1⎟ t ⎜1 0 2 3⎟ r ⎜3 0 2 1⎟
L=⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝3 2 1 0⎠L = ⎝2 3 1 0⎠ L = ⎝1 2 0 3⎠. (3.21)
1302 3102 2310
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0123 0312 0132
⎜1 3 0 2⎟ r t ⎜1 0 2 3⎟ c t ⎜1 3 2 0⎟
Lc = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝ 3 2 1 0 ⎠ (L ) = ⎝ 3 2 0 1 ⎠ (L ) = ⎝ 2 0 1 3 ⎠ . (3.22)
2031 2130 3201
Theorem 3.10 Let L be a Latin square. Then
(Lt )r = (Lc )t
and
(Lt )c = (Lr )t .
Proof If we start with RCS, then taking the transpose gives CRS. Then the row
adjugate is SRC. If we take that column adjugate of RCS we have RSC, then taking
the transpose gives SRC. This gives the first equation.
If we start with RCS, then taking the transpose gives CRS. Then the column
adjugate is CSR. If we take that row adjugate of RCS we have SCR, then taking the
transpose gives CSR. This gives the second equation.
Given a pair of mutually orthogonal Latin squares we can construct an orthogonal

array with 4 columns by using the same first two columns and writing the symbols of
the second square in the fourth column. For example, given the two MOLS of order
3 given in (3.3) we construct the following orthogonal array:
⎛ ⎞
R C S1 S2
⎜0 0 0 0 ⎟
⎜ ⎟
⎜0 1 1 1 ⎟
⎜ ⎟
⎜0 2 2 2 ⎟
⎜ ⎟
⎜1 0 2 1 ⎟
⎜ ⎟ (3.23)
⎜ 1 1 0 2 ⎟.
⎜ ⎟
⎜1 2 1 0 ⎟
⎜ ⎟
⎜2 0 1 2 ⎟
⎜ ⎟
⎝2 1 2 0 ⎠
2 2 0 1
Lemma 3.4 If L is a Latin square of order n with an orthogonal mate M then Lr

and Lc have orthogonal mates.
Proof Consider the orthogonal array with 4 columns described above. If we take
the row adjugate of each we have S1 CR and S2 CR as the latin squares. Since each
pair appears in the original two squares when overlapped, this means that each pair
appears when overlapping the row adjugates. This is because no pair appears twice
in any two columns. This is the same proof for the column adjugates.
There are 4! = 24 ways of permuting the columns in (3.23). Any permutation of

the columns will produce a pair of MOLS.
There are more ways to produce Latin squares from an existing square. Consider
a Latin square L of order n. Any of the n! ways of permuting the rows results in
a Latin square and any of the n! ways of permuting the columns results in a Latin
square. The fact there are n! ways of doing this follows from Theorem 1.5.
Lemma 3.5 If L and M are orthogonal and L is formed by permuting the rows and
columns of L then the square M formed by performing the same permutations on
M is orthogonal to L .
Proof If every pair occurs exactly once when L and M are overlaid, then exactly
every pair occurs exactly once when L and M are overlaid. They are simply in a
different position.
For example, the following Latin squares are orthogonal.

⎛ ⎞ ⎛ ⎞
012 012
L = ⎝2 0 1⎠ M = ⎝1 2 0⎠. (3.24)
120 201
Now apply the following permutation to the rows:
R 0 → R1 ,
R 1 → R2 ,
R 2 → R0 .
and the following permutations to the columns:
C0 → C0 ,
C1 → C2 ,
C2 → C1 .
The resulting squares are

⎛ ⎞ ⎛ ⎞
102 210
L = ⎝ 0 2 1 ⎠ , and M = ⎝ 0 2 1 ⎠ , (3.25)
210 102
which are orthogonal.
Definition 3.4 If L is a Latin square, then any Latin square that can be obtained by any
combination of permuting the rows, permuting the columns, permuting the symbols,
taking the row adjugate, taking the column adjugate, and taking the transpose is said
to be in the main class of L.
Theorem 3.11 If L and L are Latin squares in the same main class then L has an
orthogonal mate if and only if L has an orthogonal mate.
Proof Follows from Theorem 3.9, Lemmas 3.4 and 3.5.
The importance of this theorem is that if you are looking for squares that have
orthogonal mates you need only look through the collection of main classes rather
than the entire set of Latin squares which can be quite large. This is essentially the
proof of Fisher and Yates for the 36 officer problem, given that they actually find
representatives of the main classes and check them for orthogonality.
To give an idea of just how large the set of Latin squares can be. The number of
Latin squares of order 9 (given in [6,66]) is
(9!)(8!)(377, 597, 570, 964, 258, 816) ≈ 1027 .
This should give the reader an idea of how hard it would be to search through them
all to find orthogonal mates, let alone trying to find 3 MOLS.
The reason the number is given in this manner is that they find all Latin squares
whose first row and first column is 1, 2, . . . , 9. Then by Theorem 1.5, there are 9!
ways of permuting the symbols in the first row and 8! ways of permuting the first
column, given that the first element was changed by the permutation of the rows and
reduces the number of possible permutations.
Similarly, in [64], it was shown that the number of Latin squares of order 10 is
7580721483160132811489280(10!)(9!) ≈ 9(1036 ),
and in [50], it was shown that the number of Latin squares of order 11 is
12216177315369229261482540(11!)(10!) ≈ 1039 .
It is easily seen that given the size of these numbers, searching for orthogonal mates
with an exhaustive search is not computationally feasible.
Exercises
1. Prove there exists a symmetric Latin square for each order n > 0.
2. Prove that the addition table of a finite field is a symmetric Latin square.
3. Prove that the multiplication table of the non-zero elements of a finite field is a
symmetric Latin square.
4. Produce the MOLS formed by allowing the permutation (1, 3, 2, 4) on the
columns of (3.23).
5. Prove Lemma 3.3.
3.3 Structure of Latin Squares 93
6. Find all 6 Latin squares formed from these operations applied to the Latin square
⎛ ⎞
0123
⎜3 2 1 0⎟
⎜ ⎟ (3.26)
⎝2 3 0 1⎠.
1032
7. Use Lemma 3.4 to find Latin squares that are orthogonal to L, Lt , Lr , Lc , (Lc )t
and (Lr )t if L is the Latin square
⎛ ⎞
012
⎝2 0 1⎠. (3.27)
120
3.3 Structure of Latin Squares
We shall now describe some aspects of Latin squares which helps to understand their
structure.
Definition 3.5 Let L be a Latin square of order n, indexed by the elements of Zn . If

there are subsets R, C ⊂ Zn with |R| = |C| = m, where Lij , i ∈ R, j ∈ C, is a Latin
square of order m, then this is said to be a subsquare of order m.
As an example, consider the following Latin square. The bold elements form a
subsquare of order 2.
⎛ ⎞
0123
⎜2 3 0 1⎟
⎜ ⎟ (3.28)
⎝3 2 1 0⎠.
1032
Theorem 3.12 The circulant Latin square of order n has a subsquare of order 2 if
and only if n is even.
Proof Let Lij = i + j where the matrix is indexed by Zn . If i1 , i2 are the rows and
j1 , j2 are the columns then we have that

Li1 ,j1 Li1 ,j2
Li2 ,j1 Li2 ,j2
is a Latin square. This gives that
Li1 ,j1 = Li2 ,j2

Li1 ,j2 = Li2 ,j1 .
That is,
i 1 + j1 = i2 + j2
i1 + j2 = i2 + j1 .
Then we have i1 − i2 = j2 − j1 = i2 − i1 which gives that 2i1 = 2i2 . If n is odd,

then 2 is a unit so i = 1 = i2 and there is no subsquare. If n is 2k then let i1 = 0
and i2 = k. Then solving j2 − j1 = k gives a Latin subsquare of size 2.
Notice that in the proof we actually can find many subsquares for different values
of i1 , i2 , j1 , and j2 .
While circulant Latin squares of odd order do not have a subsquare of order 2 a
circulant Latin square of odd order can have a subsquare. Consider the following,
the bold elements form a subsquare of order 3.
⎛ ⎞
012345678
⎜8 0 1 2 3 4 5 6 7⎟
⎜ ⎟
⎜7 8 0 1 2 3 4 5 6⎟
⎜ ⎟
⎜6 7 8 0 1 2 3 4 5⎟
⎜ ⎟
⎜5 6 7 8 0 1 2 3 4⎟. (3.29)
⎜ ⎟
⎜4 5 6 7 8 0 1 2 3⎟
⎜ ⎟
⎜3 4 5 6 7 8 0 1 2⎟
⎜ ⎟
⎝2 3 4 5 6 7 8 0 1⎠
123456780
We can generalize this idea in the following theorem.
Theorem 3.13 If s and n are relatively prime then there is no subsquare of order s
in the circulant Latin square of order n
Proof Let Li,j = i + j as before. Consider two rows in the subsquare:
Li1 ,j1 Li1 ,j2 Li1 ,j3 . . . Li1 ,js

Li2 ,j1 Li2 ,j2 Li2 ,j3 . . . Li2 ,js
These rows are permutations of the same sets of elements. This gives that their sums
must be identical, that is,
s
s

si1 + ja = si2 + ja .
a=1 a=1
This gives si1 = si2 . If gcd(s, n) = 1 then s is a unit and so i1 = i2 and there is
no subsquare.
3.3 Structure of Latin Squares 95
Definition 3.6 A partial Latin square of order n is an n by n array, where cells may
be empty or from a symbol set A of size n, such that each symbol occurs at most
once in any row or column.
Consider the extremely popular game of Sudoku. This game gives a partial Latin
square of order 9 with the property that each of 9 different 3 by 3 subsquares must
contain each of the elements 1, 2, . . . , 9. The aim of the game is to complete the
square to its unique completion.
As an example, the following is a partial Latin square of order 5.
⎛ ⎞
12345
⎜4 2 1⎟
⎜ ⎟
⎜3 1 2 4⎟. (3.30)
⎜ ⎟
⎝ 413 ⎠
2 4 3
Theorem 3.14 There exists a non-completable partial Latin square for all orders
n ≥ 2.
Proof Let L be any Latin square of order n using symbols 0, 1, 2, . . . , n − 1. Let M

be the partial Latin square:

L
M= .
n
In the open spaces of M no element from 0, 1, 2, . . . , n − 1 can be used since these
are represented in every row and column already since L is a Latin square. The open
spaces also cannot be filled with n since this already appears in the last row and last
column. Therefore, this partial Latin square is non-completable.
Example 3.11 The following partial Latin squares are not completable:
⎛ ⎞
⎛ ⎞ 012
01
0 ⎜1 2 0 ⎟
, ⎝1 0 ⎠, ⎜ ⎟
1 ⎝2 0 1 ⎠.
2
3
It is also possible for a partial Latin square to complete in two different ways. For
example, consider the partial Latin square:
⎛ ⎞
0
⎝ 0 ⎠.
0
This square can complete to either of the following:

⎛ ⎞ ⎛ ⎞
021 012
⎝1 0 2⎠ ⎝2 0 1⎠.
210 120
Exercises
1. Complete the partial Latin square in Eq. 3.30 to a Latin square. Determine if the
completion is unique.
2. Find all other subsquares in the Latin square in Eq. 3.28.
3. Prove that if gcd(s, n) = 1 then there is a subsquare of order gcd(n, s) in the
circulant Latin square of order n. Hint: We need to have si1 = si2 = si3 =
· · · = sis . Then how many solutions are there for sx ≡ si1 (mod n). Use these
to construct the subsquares.
4. Show that there are partial Latin squares that are not completable for all n > 2
using a different proof than the one given in Theorem 3.14.
Affine and Projective Planes
4
4.1 Introduction
The origins of all modern mathematics are in ancient Greek mathematics. For this
culture, mathematics was geometry and the most important textbook was Euclid’s
Elements [91]. This textbook would have been a necessary prerequisite for any study
of mathematics. It was the model for not only the content of mathematics but also how
mathematics should be done. The importance of this textbook to the development of
mathematics and logic cannot be overstated. It served as the model of mathematics
and of reasoning in general for a variety of cultures over the past 2300 years. When
modern mathematics was being developed this text not only provided the model for
the techniques of mathematics but also furnished some extremely interesting open
questions which fueled mathematical inquiry for millennia. Until very recently this
text was an integral part of most people’s education. In fact, in Plato’s academy, over
the door was written that no one destitute of mathematics should enter through the
portal. This means that education in any subject required knowledge of geometry
and the reasoning skills developed studying it.
The text began by making some basic definitions and then stating the basic axioms
of Euclidean geometry. An axiom is simply a mathematical statement that we accept
without proof. Euclid was willing to accept these statements as obvious and nec-
essary for the development of geometry. These axioms are the foundation of the
entire discipline. From these axioms, theorems are developed which are then used
to construct further theorems.
These axioms, in modern terminology, were as follows:
1. Any two points can be connected by a line.

2. Any line segment can be produced continuously in a straight line.
3. Given a center and a radius, a circle can be produced with that center and radius.
98 4 Affine and Projective Planes
Fig. 4.1 Axiom 5
4. All right angles are equal to one another. (Note that no mention is made of degrees
or radians. A right angle is simply defined as one half of the angle of a straight
line.)
5. If a straight line intersecting two lines made two interior angles on the same side
less than two right angles, then the two lines will intersect on that side.
From the perspective of finite geometry, we are not really concerned with Axioms
2, 3, and 4 since they largely pertain to infinite planes.
The fifth axiom was one of the most important and controversial questions in all
of intellectual history. The axiom can be restated into the following form: Given a
line and a point p not on that line there exists a unique line through p that is parallel
to . This is shown in Fig. 4.1.
In an infinite plane, it can be shown that there is at least one line parallel to . It was
widely believed that this axiom could be proven from the first four. It turns out, thanks
to some fantastic nineteenth-century mathematics done by Lobachevsky and Bolyai
that the fifth axiom is independent of the first four. This means that there are planes
where the first four axioms are true but the fifth axiom is false. In particular, there are
planes where there are infinitely many lines through p that are parallel to . These
geometries are called hyperbolic planes. Their discovery sent shock waves through
the mathematical and philosophical worlds and sparked a tremendous amount of
further study and discussion. In the final analysis, Euclid is vindicated since it was
absolutely necessary to state the fifth axiom.
If Axiom 2 is eliminated, a geometry can be produced that has no parallel lines.
These planes are called projective planes. A projective or hyperbolic plane is called
a non-Euclidean plane, since in these planes Euclid’s fifth axiom is false. A plane
where the fifth axiom holds is called an affine or Euclidean plane.
Many standard theorems in geometry require the fifth axiom. For example, one
of the first theorems of Euclidean geometry that students are exposed to is that the
sum of the angles of a triangle is a straight line. This follows from the following two
diagrams:
Namely, in Fig. 4.2, it is shown that alternate interior angles are equal if they are
formed by a transversal to two parallel lines. In Fig. 4.3, a line is drawn through the
top point of the triangle parallel to the base of the triangle. Notice that this requires
Euclid’s fifth axiom. Then using Fig. 4.2, we see that the angles marked α are equal
and the angles marked β are equal. Then since at the top of the triangle α, β, γ make
a straight line, then the sum of the angles in the triangle make a straight line.
The theorem that the sum of the angles in a triangle is a straight line is not true
for hyperbolic nor projective planes. In a hyperbolic plane the sum of the angles is
less than a straight line and for a projective plane the sum of the angles is more than
a straight line.
4.2 Definitions 99
Fig. 4.2 Alternate interior

angles
Fig. 4.3 The sum of the

angles in a triangle
It can be shown that a hyperbolic plane, that is, a plane that has more than one
line parallel to a given line through a point off that line, must be an infinite plane so
we restrict ourselves to affine and projective planes.
4.2 Definitions
We shall now apply these geometric ideas in a combinatorial setting. Namely, we

shall see what combinatorial objects we can construct that satisfy the geometric
axioms. We leave the terms points and lines undefined but recognize that a point can
be understood as the lines that are incident with it and a line can be understood as
the points that are incident with it. Euclid gave definitions for points and lines that
are not completely rigorous. He defined points as that which had no part and lines
as breadthless length. Of course, our intuitive idea of a point and line is very much
fixed. These intuitive ideas must be modified to study non-Euclidean geometry and
finite geometry. As an example, consider the surface of the earth. The lines on the
surface of the earth are the great circle routes, meaning those circles that lie on the
surface of the earth whose center coincides with the center of the earth. These are
the canonical lines on the earth as any airline traveler can tell you, but they may not
match our intuitive idea of a line. We shall show later that, in fact, these are lines in
a projective geometry.
We shall consider incidence structures which are a set of points P, a set of lines
L, and an incidence relation I ⊆ P × L, where (p, ) ∈ I means that p is incident
with . We say two lines are parallel if they have no point that is incident with both
of them or if they are identical.
We make the following definitions.
Definition 4.1 (Affine Plane) An affine plane is a set of points A, a set of lines M,
and an incidence relation J ⊆ A × M such that
Fig. 4.4 Trivial geometry
1. Through any two points there is a unique line incident with them. More precisely,
given any two points p, q ∈ A, there exists a unique line ∈ M with (p, ) ∈ J
and (q, ) ∈ J.
2. If p is a point not incident with then there exists a unique line through p parallel
to . More precisely, if p ∈ A, ∈ M with (p, ) ∈ / J, then there exists a unique
line m ∈ M with (p, m) ∈ J and parallel to m.
3. There exists at least three non-collinear points.
Definition 4.2 (Projective Plane) A projective plane is a set of points P and a set
of lines L and an incidence relation I ⊆ P × L such that
1. Through any two points there is a unique line incident with them. More precisely,
given any two points p, q ∈ P, there exists a unique line ∈ L with (p, ) ∈ I
and (q, ) ∈ I.
2. Any two lines meet in a unique point. More precisely, given any two lines , m ∈
L, there exists a unique point p ∈ P with (p, ) ∈ I and (p, m) ∈ I.
3. There exists at least 4 points, no three of which are collinear.
Notice that in a projective plane there are no parallel lines. In particular, Axiom
2 of affine planes is false for these planes.
The third axiom of the affine plane requires the existence of a triangle and the
third axiom of the projective plane requires the existence of a quadrangle. These
axioms are there simply to eliminate trivial cases which could satisfy the axioms.
For example, the diagram in Fig. 4.4 represents a system that would satisfy the axioms
trivially but is not an object we wish to study:
If Π = (P, L, I) is a projective plane, then the plane (L, P, I) is called the dual
plane. It follows from this that there is a duality in the axioms of the planes, see
Exercise 1. This means that for any theorem we prove about lines, there is a corre-
sponding theorem about points; and any theorem we prove about points, there is a
corresponding theorem about lines.
Let Π = (P, L, I) be a projective plane. A projective plane is said to have order
n when there are n + 1 points on a line. It will become apparent after we build a
projective plane why the order is n while there are n + 1 points on a line.
Lemma 4.1 On a projective plane of order n there are n + 1 points on each line
and n + 1 lines through each point.
Proof Let L be a line and p a point off L. Any line through p must meet L as in
the diagram below since any two lines must meet. Moreover, through any point on
L there exists a unique line through that point and p. Then since any line must meet
L we have the result.
We are now in a position to count the total number of points and lines (Fig. 4.5).
4.2 Definitions 101
Fig. 4.5 Number of points

on a line and lines through a
point
Theorem 4.1 Let Π = (P, L, I) be a projective plane of order n then
|P| = |L| = n2 + n + 1. (4.1)
Proof Consider a line L and the points p1 , p2 , . . . , pn+1 on L. Through each pi

there are n lines distinct from L. If M is such a line through pi and M is such a
line through pj then M = M since the only line through pi and pj is L. Now since
any line in the plane intersects L there are n(n + 1) lines distinct from L and so
n(n + 1) + 1 = n2 + n + 1 lines in the plane.
Of course, by the duality of the definitions, we have that the number of points
and lines are the same. However, we ask for a direct proof of the fact that |P| =
n2 + n + 1 in Exercise 3.
We shall consider the best-known example, namely, the projective plane of order
2. It has 7 points each with 3 lines through them and 7 lines each containing 3 points.
What follows is simply a representation of the plane, and it is important to remember
that each line consists only of three points. The novice may err and think that there
are points everywhere on the line as if it were a line in the Cartesian plane.
If the lines are listed L1 , . . . , L7 then we have the following correspondence
between points and lines:
L1 ↔ {A, B, C},
L2 ↔ {C, D, G},
L3 ↔ {A, F, G},
L4 ↔ {C, E, F},
L5 ↔ {A, D, E},
L6 ↔ {B, E, G},
L7 ↔ {B, D, F}.
Another way of looking at a representation of the plane is to look at its incidence

matrix. The incidence matrix is simply a matrix indexed by lines and points where
there is a 1 at location (L, p) if p is incident with L and a 0 otherwise. For this plane
the incidence matrix in Table 4.1.
Table 4.1 Incidence matrix for the projective plane of order 2

A B C D E F G
L1 1 1 1 0 0 0 0
L2 0 0 1 1 0 0 1
L3 1 0 0 0 0 1 1
L4 0 0 1 0 1 1 0
L5 1 0 0 1 1 0 0
L6 0 1 0 0 1 0 1
L7 0 1 0 1 0 1 0
Fig. 4.6 The projective

plane of order 2

plane of order 1
Notice that each row and column of the matrix has exactly 3 entries with a 1
in them. Later in this text, we shall show how this representation of the plane can
actually be used to communicate with deep space probes or along telephone lines
(Fig. 4.6).
The third axiom gives that this is actually the smallest case of a projective plane,
since there must be at least 4 points and the smallest number of the form n2 + n + 1
greater than 4 is 7, making n = 2. However, some people are willing to allow the
case for n = 1, that is a projective plane of order 1. It would consist of 3 points and
3 lines and would be represented as follows (Fig. 4.7).
If we had begun with the assumption that the number of lines and points were
equal to v and that there are n + 1 points on a line, then we could have counted the
number of points and lines as follows. We know that through any two points there
exists a unique line and there are C(v, 2) ways of picking two points. Also we know
4.2 Definitions 103
that there are C(n + 1, 2) ways of picking two points on a line, meaning that each
line is counted C(n + 1, 2) times in the C(v, 2) ways of counting lines. This gives
C(v, 2)
=v
C(n + 1, 2)
v(v − 1) (n + 1)(n)
/ =v
2 2
v(v − 1) = v(n + 1)(n)
v − 1 = n2 + n
v = n2 + n + 1.
We shall now give an example of a model of an infinite projective plane. Consider

a sphere S with center at the origin in three-dimensional Euclidean space. We will
only consider the top half of that sphere and only the points on the equator that fall
in the western hemisphere. The lines on this plane are the great circle routes. Now
it is easy to see that through any two points there is a line (simply slice the sphere
with the plane formed by the two points and the center of the sphere). Any two great
circle routes will in fact meet. It is easy to see now why we only considered the top
hemisphere since two great circle routes will meet twice on a sphere, once in each
hemisphere. Two lines that meet on the equator would meet on the opposite side
on the equator as well, which is why we eliminated half of the equator. The second
axiom of Euclid is, of course, not true since a line does not extend infinitely far in
both directions. These lines end at the equator.
We now turn our attention to affine planes. The first thing to consider in an affine
plane is the notion of parallelism. As a matter of definition we say that a line is
parallel to itself and two distinct lines are parallel if they have no points in common.
Theorem 4.2 Parallelism is an equivalence relation on the lines of an affine plane.
Proof Every line is parallel to itself by definition and hence the relation is reflexive.
If is parallel to m then either they are equal or disjoint. In either case m is
parallel to and the relation is symmetric.
Assume is parallel to m and m is parallel to k. If k and had a point of
intersection p then there would be two lines through p parallel to m contradicting
Axiom 2. Hence the relation is transitive and therefore an equivalence relation.
We shall do similar counts for the affine plane that we did earlier for the projective
plane. Let us consider an affine plane. Let π = (A, M, J) be an affine plane. Assume
that each line has n points incident with it. We say that such a plane has order n.
Let be a line and let m be a line that intersects at the point p. There are n − 1
points on m that are not on . Through each of these there must be a unique line
parallel to . Of course none of these could be the same since then through the point
of intersection there would be two lines parallel to . Since is parallel to itself, we
see that there are at least n lines in a parallel class. If there was another line parallel
to it would have to intersect m, but we have considered all such lines and hence
we have the following.
Lemma 4.2 In an affine plane of order n each parallel class has exactly n lines.
This lemma allows us to determine the cardinality of the point set.
Lemma 4.3 Let π = (A, M, J), then |A| = n2 .
Proof Let be a line. There are n lines in the parallel class of . Each of these has
n points incident with it and no two lines have any points in common. Hence, there
are n2 points on these lines. If there were another point in the plane not yet counted
there would be a line through it parallel to which is a contradiction. Therefore,
there are n2 points on the plane.
We can now count the number of parallel classes.
Lemma 4.4 There are n + 1 parallel classes in an affine plane of order n.
Proof Let p be a point in the plane. There are n + 1 lines through p. For any
parallel class, there must be a line from that class through p since the lines in any
parallel class cover all points in the plane. Hence each line comes from a different
parallel class since if two were in the same class then they would intersect and not be
parallel.
We can use these lemmas to determine the number of lines in an affine plane.
Theorem 4.3 In an affine plane of order n there are n2 + n lines.
Proof By Lemma 4.4, there are n + 1 parallel classes and by Lemma 4.2, each class
has n lines in it. Therefore, there are (n + 1)n = n2 + n lines in the plane.
We summarize all the counting results in the next theorem.
Theorem 4.4 Let π = (A, M, J) be an affine plane of order n. There are n points
on a line, n + 1 lines through a point, |A| = n2 , |M| = n2 + n, and parallelism
is an equivalence relation on the set of lines with n lines in each of n + 1 parallel
classes. Let Π = (P, L, I) be a projective plane of order n. There are n + 1 points
on each line, n + 1 lines through each point, and |P| = |L| = n2 + n + 1.
We give an example with the affine plane of order 2 in Fig. 4.8.
If the lines are listed as L1 , . . . , L6 , then we have the following correspondence

between points and lines:
4.2 Definitions 105
Fig. 4.8 Affine plane of

order 2
L1 ↔ {A, B},
L2 ↔ {C, D},
L3 ↔ {A, C},
L4 ↔ {B, D},
L5 ↔ {A, D},
L6 ↔ {B, C}.
There are three parallel classes here, namely, {L1 , L2 }, {L3 , L4 }, and {L5 , L6 }.
Note that each parallel class partitions the set of points.
Affine and projective planes are canonically connected via the next two theorems.
Theorem 4.5 Let π = (A, M, J) be an affine plane of order n. Let qi be a new

point corresponding to the i-th parallel class of π. Consider the incidence structure
with P = A ∪ {qi }, L = {m ∪ {qi } | m ∈ M, m in the i-th parallel class } ∪ {L∞ }
where L∞ is incident with a point p in P if and only if p = qi , and incidence is
induced by the incidence of π and qi is incident with each line in the i-th parallel
class. Then Π = (P, L, I) is a projective plane of order n.
Proof Let p and q be two distinct points. If they were both points in the affine plane
incident with the line in the i-th parallel class then the line L = ∪ {qi } is a line
incident with p and q. Since the only other possible line is L∞ which has only points
not in π on it we see that the line is unique. If neither p nor q are in π then L∞ is
the unique line through the two points. If p is a point of π and q = qi not a point of
π then the unique line through these two points is the unique line of the i-th parallel
class through p.
Consider two lines L and M. If L = ∪ {qi } and M = m ∪ {qj } then if i = j,
the lines meet at qi and if i = j then the lines meet at the unique point of
intersection of and m. If L = ∪ {qi } and M = L∞ then the unique point of
intersection is qi .
Basically, what we have done in this theorem is to make all lines meet at a point
that were parallel in the affine case. The language that is often used in this is that we
make parallel lines meet at a point at infinity and then all of these points at infinity
make a line at infinity. It is often said that the affine plane completes to a projective
plane.
Theorem 4.6 If Π = (P, L, I) is a projective plane of order n, then the incidence

structure formed by removing any line L and the points incident with L is an affine
plane of order n. Precisely, if A = P − {p | (p, L) ∈ I}, M = L − {L}, and inci-
dence J is induced by I then π = (A, M, J) is an affine plane.
The line that is removed is usually called the line at infinity. Of course, any line can
be removed and the construction works. Intuitively, this means that in the transition
from affine to projective and back, any line can be thought of as the line at infinity.
Consider the projective plane of order 2 given in Fig. 4.6. We shall take L7 to
be the line at infinity. This is the standard choice for this plane because somehow
it looks like a line at infinity. Remove L7 and the points B, D, F and the plane that
remains is given in the Fig. 4.9.
Notice that this plane is essentially the same as the previous description of the
affine plane of order 2. In fact, later we shall see that they are exactly the same.
This plane has the incidence matrix given in Table 4.2.
While the projective plane of order 1 seems perfectly natural, the same cannot be
said for what would be the affine plane of order 1. By removing a line and the points
on it from the projective plane of order 1 we would have 1 point with 2 lines through
it, which seems to be not at all what we wanted. From this point on when discussing
planes we always assume we have the third axiom, that is, the order of the plane is
always at least 2.
Let us consider our infinite example. Consider the usual Euclidean plane given
by z = 0 in standard three-dimensional Euclidean space R3 . Place the top half of
the sphere with half of the equator as described before so that the center of the
sphere is at (0, 0, 1). For any point (x, y) on the plane draw the unique line from that
Fig. 4.9 Affine plane formed

from the projective plane of
order 2
Table 4.2 Incidence matrix for the affine plane of order 2

A C E G
L1 1 1 0 0
L2 0 1 0 1
L3 1 0 0 1
L4 0 1 1 0
L5 1 0 1 0
L6 0 0 1 1
4.3 Planes from Fields 107
point to the point (0, 0, 1). This line intersects the sphere at a unique point. Hence,
there is a unique point on the half-sphere associated with each point on the plane,
except for those points on the equator. For each parallel class with slope α where
α ∈ R or α = ∞ for lines with no slope, associate the point on the equator that is the
intersection of the plane that is perpendicular to the plane z = 0 and intersects it at
the line y = αx or x = 0 if α = ∞. Then notice that each line on the plane traces a
great circle route on the half-sphere. The points on the plane correspond to the non-
equatorial points on the half-sphere, the equatorial points are the points at infinity,
the equator is the line at infinity, and the lines of the plane have a corresponding
line in the half-sphere, with the equator serving as L∞ . We see that this is precisely
the projective completion of the affine plane. Notice also that this finite half-sphere
contains all of the geometric information of the infinite plane, plus a bit more!
Exercises
1. Let Π = (P, L, I) be a projective plane. Show that (L, P, I) is also a projective

plane. That is, each line is now called a point and each point is now called a line.
Show also that reversing the roles of lines and points in an affine plane will not
produce another affine plane.
2. Assume there are n + 1 lines through each point in a projective plane and prove
that there must be exactly n + 1 points on each line.
3. Prove that in a projective plane of order n, we have that |P| = n2 + n + 1.
4. Determine the lines that are incident with each point in the projective plane of
order 2 given in Fig. 4.6.
5. Prove that in an affine plane there are n + 1 lines through a point.
6. Prove Theorem 4.3.
7. List the lines in the affine plane of order 2 through each point given in Fig. 4.8.
8. Show that the construction in Theorem 4.5 corresponds to the counting infor-
mation in Theorem 4.4.
10. Show that the construction in Theorem 4.6 corresponds to the counting infor-
mation in Theorem 4.4.
11. Construct an affine and projective plane of order 3.
4.3 Planes from Fields
The first affine plane that students encounter is the real affine plane. It is usually
described in terms of the Cartesian coordinate system with an x- and y-axis. The
points are given by (x, y) where x and y are real numbers and the lines are all of
the form y = mx + b or x = c. We can replace the real numbers with any field,
including finite fields, and we can still construct an affine plane. Throughout the
remainder of the section, we shall let F denote a field.
Let A = {(x, y) | x, y ∈ F}, M = {y = mx + b | m, b ∈ F} ∪ {x = c | c ∈ F}.
We say that the line y = mx + b contains all points (x, y) that satisfy the equation
and x = c contains all points (c, y) for any y. This is, of course, the natural way to
regard these structures.
First we shall show that it does, in fact, form a plane. Let (a, d), (a , d ) be two
distinct points. If a = a then the two points are on the line x = a. If they were both
on the same line of the form y = mx + b then we would have d = ma + b = d
and the two points would not be distinct. If a = a then there is no line of the form
x = c that both points are on. If both points are on y = mx + b then we have
d = ma + b and d = ma + b
⇒ d − ma = d − ma
⇒ d − d = m(a − a )
d − d
⇒m= .
a − a
Then m is uniquely determined and b = d − ma is uniquely determined as well.
Therefore, through any two points there exists a unique line. Notice how necessary it
was that we had a field. Namely, we needed to be able to divide by a − a when it is
non-zero, i.e., when a = a . If the algebraic structure were not a field, for example,
Z4 , we would not be able to guarantee that there is a line through any two points.
For example, if we were looking over Z4 there is not a line incident with the points
(3, 1) and (1, 2).
Any two lines of the form x = c are parallel. Given two lines y = mx + b and
x = c then their unique point of intersection is (c, cm + b). Given two lines y =
mx + b and y = m x + b . If m = m and (x, y) satisfies both equations then
y = mx + b and y = m x + b
⇒ mx + b = m x + b
⇒ b − b = (m − m)x
b − b
⇒x=
m −m
b−b
and x is uniquely determined. Then y = m m −m + b and there is a unique point
of intersection. Notice that again it was necessary to be able to divide by a non-zero
element.
Given two lines y = mx + b and y = mx + b , if there were a point of inter-
section, we would have b = y − mx = b and the lines would have to be identical.
Therefore, any two lines with identical m (slope) are parallel.
Let |F| = n. Since there are n choice for m there are n parallel lines in each class.
Also we note that there are n2 points in the plane. It should be evident now why the
order of the plane is given by n, namely, that the plane that comes from a field of
size n has order n. We have proven the following.
Theorem 4.7 Let F be a field of order n. Let A = {(x, y) | x, y ∈ F} and M = {y =

mx + b | m, b ∈ F} ∪ {x = c | c ∈ F} with the natural incidence relation J then
π = (A, M, J) is an affine plane of order n.
4.3 Planes from Fields 109

order 2
We shall do a simple example.

Let F = F2 = {0, 1} the field of order 2. The points are
(0, 0), (0, 1), (1, 0), (1, 1).
The lines are

x = 0, x = 1, y = 0, y = 1, y = x, y = x + 1.
We note that there are 4 = 22 points and 6 = 22 + 2 lines.
The following diagram describes this plane (Fig. 4.10).
The bottom horizontal line is y = 0 and consists of the points (0, 0) and (1, 0).
The top horizontal line is y = 1 and consists of the points (0, 1) and (1, 1). The left
vertical line is x = 0 and consists of the points (0, 0) and (0, 1). The right vertical
line is x = 1 and consists of the points (1, 0) and (1, 1). The ascending vertical line
is y = x and consists of the points (0, 0) and (1, 1). The descending vertical line is
y = x + 1 and consists of the points (0, 1) and (1, 0).
This plane may look different from the plane given in Fig. 4.9 but they are exactly
the same plane with a slightly different representation.
We shall now show how a projective plane can be constructed from a finite field.
Instead of looking at points as elements in F2 , we shall look at points as elements
in F3 . However, we only want a specified subset of these points.
We shall say that two elements of F3 , (a, b, c) and (a , b , c ) are equivalent,
written as
(a, b, c) ≡ (a , b , c ) (4.2)
if there exists a non-zero λ ∈ F with a = λa , b = λb , c = λc . For example, in F35
the elements that are equivalent to (1, 2, 3) are {(1, 2, 3), (2, 4, 1), (3, 1, 4), (4, 3, 2)}.
The points set P = (F3 − {(0, 0, 0)})/ ≡ . This simply means that the points are
the equivalence classes of F3 − {(0, 0, 0)} formed from the equivalence relation ≡ .
We usually just take a specific element of each class to represent the point.
As an example we shall show the points in (F33 − {(0, 0, 0)})/ ≡ . The points are
{(0, 0, 1), (0, 1, 0), (0, 1, 1), (0, 1, 2), (1, 0, 0), (1, 0, 1), (1, 0, 2),
(1, 1, 0), (1, 1, 1), (1, 1, 2), (1, 2, 0), (1, 2, 1), (1, 2, 2)}.
The lines are defined in the same way, that is, L = (F3 − {(0, 0, 0)})/ ≡ . In some
texts the lines are written as column vectors so that there is something to distinguish
lines and points. We shall distinguish them by writing points as the vector (a, b, c)
and lines as the vector [a, b, c].
Incidence for the plane is given by the following. The point (a, b, c) is on the line
[d, e, f] if and only if ad + be + cf = 0. It is clear why we had to eliminate (0, 0, 0)
since it would be incident with every line.
For example, for the plane of order 3, the point (1, 2, 2) is on the line [1, 1, 0]
since 1(1) + 2(1) + 2(0) = 1 + 2 = 0. In fact the point (1, 2, 2) is incident with the
four lines [0, 1, 2], [1, 0, 1], [1, 1, 0], and [1, 2, 2].
Theorem 4.8 If Π = (P, L, I) with P = L = (F3 − {(0, 0, 0)})/ ≡ and (a, b, c) is

incident with [d, e, f] if and only if ad + be + cf = 0, then Π is a projective plane
of order |F|.
Proof Take two distinct points (a, b, c) and (a , b , c ). We know from elementary
linear algebra that over any field the system of equations
ax + by + cz = 0
a x + b y + c z = 0
has a family of solutions with one degree of freedom provided that (a , b , c ) is

not a multiple of (a, b, c), which by the way the points are defined, we know this
one is not the multiple of another, since, in that case they would be the same point.
This means that we have at least one line that is incident with both points, all that
remains is to show that any two vectors [x, y, z] satisfying these two equations must
be multiples of each other. We know that [0, 0, 0] is a particular solution to this system
of equations so we know that the family of solutions is a vector space of dimension 1
and hence every vector in that space is of the form λ[d, e, f] for some vector [d, e, f].
This means that [d, e, f] is the unique line through these two points.
To show that any two lines meet in a unique point, simply reverse the words line
and point and the round parentheses with the square parentheses.
As an example, we shall construct the projective plane of order 2. We have
P = {(0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0), (1, 1, 1)}, (4.3)
L = {[0, 0, 1], [0, 1, 0], [0, 1, 1], [1, 0, 0], [1, 0, 1], [1, 1, 0], [1, 1, 1]}. (4.4)
4.4 Connection Between Affine Planes and MOLS 111
The incidence is as follows:
[0, 0, 1] ↔ {(1, 0, 0), (0, 1, 0), (1, 1, 0)};

[0, 1, 0] ↔ {(1, 0, 0), (0, 0, 1), (1, 0, 1)};
[0, 1, 1] ↔ {(1, 0, 0), (0, 1, 1), (1, 1, 1)};
[1, 0, 0] ↔ {(0, 1, 0), (0, 0, 1), (0, 1, 1)};
[1, 0, 1] ↔ {(0, 1, 0), (1, 0, 1), (1, 1, 1)};
[1, 1, 0] ↔ {(0, 0, 1), (1, 1, 0), (1, 1, 1)};
[1, 1, 1] ↔ {(1, 1, 0), (1, 0, 1), (0, 1, 1)}.
Recall that Lemma 3.1 states that finite fields exist for all orders pe where p is a
prime and e > 1. The technique we have just developed gives that if we have a field,
we can construct an affine and projective plane of the same order as the field. This
gives us the following important theorem.
Theorem 4.9 Affine and projective planes exist for all orders of the form pe where
p is a prime and e > 1.
Exercises
1. Verify all of the remaining counting information in Theorem 4.4 for the con-
struction given in Theorem 4.7. That is, count the number of lines, points on a
line, lines through a point, and parallel classes, and make sure they match what
is given in the theorem.
2. Find the points and lines of the affine plane of order 3 formed from the field of
order 3, F3 = {0, 1, 2}. Then make a graphic representation of the affine plane
of order 3.
3. Prove that ≡ defined in Eq. 4.2 is an equivalence relation on F3 .
4. Find all elements equivalent to (3, 2, 6) in F37 using the relation given in Eq. 4.2.
5. Prove that |P| = |(F3n − {(0, 0, 0)})/ ≡)| = n2 + n + 1.
6. Label the points and lines on Fig. 4.6 so that the incidence is correct.
7. Construct the projective plane of order 3 from the field of order 3. Namely, list
the 13 lines and the 4 points on each.
8. Construct an affine and projective plane of order 4 from the field of order 4.
4.4 Connection Between Affine Planes and MOLS
The two main problems we have studied up to this point are the existence of a
complete set of MOLS and the existence of finite affine and projective planes. They
may have seemed to be quite different and, in fact, their origins are quite different.
However, it turns out that they are really the same problem. We shall now show this
connection between a complete set of Mutually Orthogonal Latin Squares of order

n and an affine plane of order n.
Theorem 4.10 A complete set of MOLS of order n exist if and only if an affine plane
of order n exists.
Proof For both directions of the proof associate the n2 points of the plane with
the n2 coordinates of the Latin squares. We can describe the points as (a, b) where
a, b ∈ {0, 1, . . . , n − 1} = Zn .
Assume we have a complete set of MOLS of order n, that is, n − 1 MOLS of order
n on the alphabet {0, 1, . . . , n − 1}. The first two parallel classes have the lines that
are the horizontal and vertical lines of the grid. That is, the c-th line in the first parallel
class contains the n points (c, 0), (c, 1), . . . , (c, n − 1) and the c-th line in the second
parallel class contains the n points (0, c), (1, c), . . . , (n − 1, c). Let Lk = (Lij ) be
the k-th Latin square, then the (k + 2)-nd parallel class has lines corresponding to
the symbols of the square. That is, the c-th line is incident with the point (i, j) if
and only if Lij = c. By construction, these n lines are parallel. Moreover, since any
two lines corresponding to the symbols c and c from different parallel classes are
from orthogonal Latin squares, we know that the formed Graeco-Latin square has
the pair (c, c ) exactly once, so the lines meet exactly once. Through any point on
the grid there are n + 1 lines corresponding to the first two parallel classes and the
n − 1 Latin squares. This means that taking any point p there are n + 1 lines through
p each with n − 1 points other than p and so there are (n + 1)(n − 1) = n2 − 1
points on the plane that are connected to p with a line. Hence, there is a line through
any two points. Therefore, we have shown that we have constructed an affine plane
of order n.
If we assume we have an affine plane of order n then we use the first two parallel
classes to describe the grid. That is, the point (i, j) is the point of intersection of the
i-th line of the first parallel class and the j-th line of the second parallel class. Then
we reverse the construction, Lk i,j = c if and only if the c-th line of the (k + 2)-nd
parallel class is incident with the line (i, j). It is evident that these are Latin squares
since each line intersects each line in the first two parallel classes exactly once so
each symbol occurs in each row and column exactly once. Since any two lines from
different parallel classes intersect exactly once it shows that any two squares are
orthogonal since the pair (c, c ) appears exactly once.
As an example we show two MOLS of order 3 and the affine plane of order 3
(Fig. 4.11).
⎛ ⎞ ⎛ ⎞
123 123
⎝3 1 2⎠ ⎝2 3 1⎠
231 312
4.5 Fundamental Question 113

order 3
Notice that this theorem, and the knowledge that there is no solution to the 36
officer problem, gives that there is no affine plane of order 6. However, if one knew
that there were no plane of order 6, then this does not solve the 36 officer problem.
Exercises
1. Use the Latin square of order 2 to draw the affine plane of order 2.
2. Draw the affine plane of order 4 using the three MOLS of order 4. It may be
helpful to draw the third, fourth, and fifth parallel classes in different colors.
4.5 Fundamental Question
We make the following definition general to include both affine and projective planes.
The definition applies to any incidence structure with points, lines, and an incidence
relation between them.
Definition 4.3 Two incidence structures D = (P, L, I) and D = (P , L , I ) are

isomorphic if and only if there exists a bijection Φ : D → D , where Φ : P → P
and Φ : L → L , with (p, L) ∈ I if and only if (Φ(p), Φ(L)) ∈ I . In other words,
the isomorphism sends points to points and lines to lines bijectively with the property
that p is incident with L if and only if Φ(p) is incident with Φ(L).
We say that two planes are isomorphic if and only if there exists an isomorphism
between them. It is natural to think that two isomorphic planes are really the same
planes with different representations.
Theorem 4.11 Let Π and Σ be isomorphic planes (affine or projective), and let IΠ
and IΣ be their respective incidence matrices. Then IΠ can be transformed into IΣ
by permuting the rows and columns.
Proof Let Φ be the isomorphism between Π and Σ. Then if the rows correspond to
the points pi and the columns to Li then permute the rows and columns by Φ, that
is, the rows and columns are now
Φ(p1 ), Φ(p2 ), . . . , Φ(pn2 +n+1 )
and
Φ(L1 ), Φ(L2 ), . . . , Φ(Ln2 +n+1 )
which gives the incidence matrix of Σ.
We have seen that if Π = (P, L, I) is a projective plane then its dual ΠD =

(L, P, I) is a projective plane. There are examples of projective planes that are not
isomorphic to their dual, but if the plane is isomorphic to its dual then the incidence
matrix can have its rows and columns permuted to be the transpose of the incidence
matrix.
The fundamental question of finite planes is the following.
Fundamental Question: For which orders n do there exist finite affine and pro-
jective planes. For orders where planes exist, how many non-isomorphic planes are
there?
At present, no one has found a plane of non-prime power order. We do know that
there exist planes for all prime power orders as we can construct them from finite
fields. The first orders where there are planes which are not isomorphic to the plane
constructed from the finite field is at order 9. For order 9 there are four projective
planes. Since we know there are not two MOLS of order 6 we know that there is
no plane of order 6. An infinite number of orders are eliminated in the Bruck-Ryser
Theorem which we shall see later in Theorem 6.27.
The first order for which its existence could not be determined by the above
information was order 10. For many years, the existence of a plane of order 10 was a
large open question. The proof that there was no plane of order 10 was given in [58].
The proof relied heavily on a large (year long!) computer computation. Basically,
it was shown that if such a plane existed then there would have to exist a doubly
even self-dual binary code of length 112 with minimum weight 12 and no vectors
of weight 16. The computer computation basically showed that no such code could
exist by showing that certain configurations do not exist.
The two major conjectures for finite planes are the following.
Conjecture 4.1 There exists projective planes of order n if and only if n is a prime
power.
Conjecture 4.2 All projective planes of prime order are isomorphic.
For this first conjecture, we know that planes exist for all prime powers. Namely,
from any finite field we can construct a plane. Moreover, the non-Desarguesian planes
that we know, all have the same orders as Desarguesian planes. Some conjecture that
this case is always true, that is if a plane exists then a Desarguesian plane of the same
order exists. This conjecture is equivalent to the conjecture, discussed in a previous
section, that the only complete sets of MOLS occur for prime power orders.
4.5 Fundamental Question 115
The second conjecture arises from the fact that no non-Desarguesian plane of
prime order has yet to be discovered. It would correspond to the conjecture that the
only complete set of MOLS of prime order is isomorphic to the set constructed from
Fp .
Exercises
1. Show that any two affine planes of order 2 are isomorphic.
2. Show that any two projective planes of order 2 are isomorphic.
3. Show that any affine plane of order 3 must be Desarguesian by showing that
any pair of MOLS of order 3 must be equivalent to the ones arising from the
construction based on the field of order 3, F3 .
Graphs
5
5.1 Königsberg Bridge Problem
In an earlier chapter, we described how Euler began the study of Latin squares by
examining the 36 officer problem. Euler also began the study of graph theory by
examining the Königsberg bridge problem. In a short article [33], he started a branch
of combinatorics that has become one of the largest and most studied branches of
discrete mathematics. Moreover, it has found numerous applications in a variety of
fields. It has also been suggested that this paper began the study of topology as well.
We shall begin with an investigation of this problem and develop some of the more
elementary aspects of graph theory.
Unlike the 36 officer problem, which Euler did not solve, he did solve the Königs-
berg bridge problem quite easily and gave general solutions for all similar problems.
The problem was whether one could walk over each of the bridges of Königsberg,
Prussia exactly once. The city is now called Kalingrad and is in Russia. The bridges
at that time were arrayed as in the following diagram. Since then bridges have been
added so it no longer looks like this (Fig. 5.1).
There are seven bridges and four pieces of land (the two banks and the two islands).
He described the situation in Fig. 5.2 and asked whether a tour could be found over
each bridge exactly once.
The reasoning to solve this problem is quite simple. At each point (piece of land)
there are bridges that connect it to other points. Unless the point is the beginning
or end of the tour then there must be an even number of bridges at that point. This
is quite simple since if you arrive at a point, that is, neither the beginning nor the
end, then you must leave on a different bridge than you came. Hence, the number of
bridges at each of these middle points must be even. If the beginning is also the end
then the number of bridges at that point must be even as well. If the beginning is not
118 5 Graphs
Fig. 5.1 The bridges at Königsberg
Fig. 5.2 Graph of the

bridges at Königsberg
the end then there must be an odd number of bridges at those two points, since there
is exactly one unmatched bridge at those points.
Examining this diagram we see that at A there are three bridges, at B there are
three bridges, at C there are five bridges, and at D there are three bridges. Clearly
there can be no such tour across the bridges of Königsberg where each bridge is
crossed exactly once. We shall now generalize these ideas.
The kind of graph denoted here is called a multigraph, while a graph usually refers
to a graph with no repeated edges. It is also possible that the term graph can refer to
a family to which both of these objects, as well as others, belong. We can now give
the definition of a multigraph.
Definition 5.1 A multigraph is a structure G = (V, E) where V is a set of vertices,

E is a set of edges, and each edge is represented by ({a, b}, i) where a and b are
vertices connected by the edge and i is used to distinguish multiple edges. We say
that two vertices are adjacent if they are connected by an edge.
5.1 Königsberg Bridge Problem 119
The degree of a vertex is the number of edges incident with it. Notice that if a
vertex is connected to itself, called a loop, this edge contributes 2 to the degree of
the vertex. We denote the degree of a vertex v by deg(v).

Lemma 5.1 Let G = (V, E) be a multigraph, then 2|E| = v∈V deg(v).
Proof This follows from that fact that each edge contributes 1 to the degree of two
vertices.
The real power of graph theory is not in the rigorous definitions. For example, we
could have defined multigraph in a variety of ways. But rather it is that the diagram
form of the graph is highly intuitive. Thinking of a graph in terms of its representation
is often the best way to come to solutions for a host of combinatorial problems.
We make the following definition.
Definition 5.2 A tour in a graph G = (V, E) is an ordered k-tuple
(e1 , e2 , . . . , ek ),
ei = {vi , wi } ∈ E, where wi = vi+1 and ei = ej if i = j. An Euler tour is a tour

such that each edge of the graph appears in the tour exactly once. An Euler cycle
is an Euler tour where the starting vertex of the tour is also the ending vertex, i.e.,
v1 = wk . In this definition, we say that the tour is from the vertex v1 to wk .
The Königsberg bridge problem is then to find an Euler tour on the graph given in
Fig. 5.2. As we have seen this is impossible and we shall generalize this result with
two theorems.
We are not really interested in the existence of Euler tours if the graph is not
connected, that is, if the graph is made up of distinct parts. We make the rigorous
definition as follows.
Definition 5.3 A connected graph is a graph in which for any two vertices v and w
there is a tour from v to w. Otherwise, we say that the graph is disconnected.
It follows immediately from the definition that a disconnected graph is made up

of connected components.
Theorem 5.1 Let G = (V, E) be a connected graph. There exists an Euler cycle if
and only if every vertex has even degree.
Proof The fact that every vertex must have even degree is easy to see. Namely,
if a tour goes to a vertex w , i.e., ei = {v, w}, then it must leave that vertex, i.e.,
ei+1 = {w, x}. The starting vertex is also the ending vertex so the same applies
there.
120 5 Graphs
We shall show that this is a sufficient condition. Let G be a graph such that the
degree of every vertex is even. Pick any vertex v1 and choose an edge e1 that is of the
form {v1 , v2 }. Then choose an edge e2 of the form {v2 , v3 }. The process terminates
because the graph is finite and as long as the last edge is not of the form {vk , v1 } then
there will be additional choices since the degree of every vertex is even (intuitively,
that is, if the tour arrives at a vertex it can also leave). If this cycle is an Euler cycle
then we are done. If not then delete the edges of the tour and what remains is a graph
where every vertex has even degree. Choose one of the vertices that we have visited
that still has a non-zero degree (we can do this because the graph is connected).
Apply this technique again starting at this vertex. Continue this technique until an
Euler cycle is constructed.
What the proof of this theorem shows is that it is quite easy to construct an Euler
cycle simply by choosing edges nearly at random.
Theorem 5.2 Let G = (V, E) be a connected graph. There exists an Euler tour that
is not an Euler cycle if and only if there are exactly two vertices with odd degree.
Proof Add an edge between the two vertices of odd degree and then apply The-
orem 5.1. Then remove the added edge from the cycle to produce the Euler tour.
We note that in applying this theorem any Euler tour must begin at one of the
points with odd degree and end at the other with odd degree.
A variation on the idea of Euler tours and cycles is the notion of the Hamiltonian
tours and cycles, named after William Rowan Hamilton. See [45,46] for early work
on the subject. We now give their definitions.
Definition 5.4 A Hamiltonian tour is a tour that hits every vertex exactly once. A
Hamiltonian cycle hits every vertex exactly once except for the final vertex which
was also the beginning vertex.
Consider the following graph.

c d
a b
The tour {a, b}, {b, c}, {c, d} is a Hamiltonian tour and {a, b}, {b, c}, {c, d},
{d, a} is a Hamiltonian cycle. Notice that these do not involve every edge. The
tour {a, b}, {b, d}, {d, c} is a also a Hamiltonian tour and {a, b}, {b, d}, {d, c}, {c, a}
is also a Hamiltonian cycle. Notice that there are numerous Hamiltonian tours in this
graph.
5.1 Königsberg Bridge Problem 121
It is easy to see that any Hamiltonian cycle can be changed into a Hamiltonian
tour by removing one of its edges. However, a Hamiltonian tour can be changed into
a Hamiltonian cycle only when its endpoints are adjacent.
A graph is Hamiltonian-connected if for every pair of vertices there is a Hamil-
tonian tour between the two vertices.
There is not a complete characterization of Hamiltonian tours like there is for
Euler tours. However, Dirac [27] proved the following in 1952. We leave the proof
as an exercise.
Theorem 5.3 A graph with no multiple edges on n ≥ 3 vertices has a Hamiltonian

tour if the degree of every vertex is at least n
2.
The idea of a tour on a graph has also been used in a variety of popular notions.
For example, consider the set of all people on the earth. Let these people be the
vertices of the graph. Let there be an edge between two people if they know each
other. This leads to the popular notion of six degrees of separation. The idea is that
between any two people (vertices) there is a tour of distance less than or equal to 6.
For mathematicians, they are concerned with the collaboration graph. The ver-
tices are the people who have published a mathematics article. Two authors are
connected if they have ever coauthored a paper together. The most important tours
for mathematicians are tours that go from themselves to the great twentieth-century
mathematician Paul Erdös. The distance of this tour is known as an Erdös number.
The lower the number the closer you are to Erdös. The American Mathematical Soci-
ety’s webpage has a function that will calculate the distance of the tour between any
two authors. Of course, it is not true that all authors have such a tour, since there is a
significant number of authors who have never collaborated. However, an extremely
large portion of the graph is connected and the distance between two authors is often
surprisingly low.
Exercises
1. Use a modified algorithm from the proof of Theorem 5.1 to find an Euler tour
in the following graph. This exact problem is often a game played by children,
namely, to draw this figure without going over any of the lines twice.
122 5 Graphs
2. Consider the following three graphs:
a. Give the degree of each vertex in each graph.

b. Determine which (if any) of the three graphs has an Euler cycle.
c. Determine which (if any) of the three graphs has an Euler tour.
d. Determine which (if any) of the three graphs has a Hamiltonian cycle.
3. Construct a graph on six vertices that has neither an Euler tour nor an Euler cycle.
4. Let A and B be disjoint sets of vertices in a graph where every vertex of A is
connected exactly once to every vertex of B and no vertex of A is connected to
another vertex of A and no vertex of B is connected to another vertex of B. Prove
conditions on the cardinalities of A and B for Euler cycles and Euler tours to
exist.
5. Construct a connected graph on five vertices that has a Hamiltonian tour and a
graph on four vertices that does not have a Hamiltonian tour.
5.2 Simple Graphs 123
6. Prove that the number of Hamiltonian cycles on the complete graph, which is the
(n−1)!
graph where any two distinct edges are connected, with n vertices is 2 .
7. Determine for which complete graphs, which is the graph where any two distinct
vertices are connected, Euler tours and Euler cycles exist.
9. Prove that for all n > 1, there exists a graph on n vertices with no Euler cycle.
10. Find all Hamiltonian tours starting from a given vertex in the following cube
graph:
5.2 Simple Graphs
We shall now restrict ourselves to the most commonly studied family of graphs,
namely, simple graphs. A simple graph is a graph with no repeated edges and no
loops. Throughout the remainder of the text if we say graph it means simple graph.
Hence, we can define a graph as follows.
Definition 5.5 A simple graph is G = (V, E), where V is the set of vertices, and E
is the set of edges E, where {a, b} ∈ E, a = b, indicates that a and b are connected
by an edge.
We begin with a counting theorem.
n(n−1)
Theorem 5.4 The number of graphs on a set of n vertices is 2 2 .
n(n−1)
Proof There are C(n, 2) = 2 ways of picking two vertices and each pair is
either connected or not connected.
n(n−1)
Example 5.1 If n = 2 then 2 2 = 1. Graphs A and B are the two possible
graphs on two vertices.
A B
124 5 Graphs
Fig. 5.3 Graphs on three vertices
Fig. 5.4 Discrete graph on

five vertices
n(n−1)
Example 5.2 If n = 3 then 2 2 = 8. Hence, there are eight possible simple
graphs on three vertices and they are given in Fig. 5.3.
There are several important families of simple graphs. We shall describe some of
those families.
• The discrete graph on n vertices, Dn , is the graph consisting of n vertices and

no edges. The discrete graph on five vertices is given in Fig. 5.4.
• The complete graph on n vertices, Kn , is the graph for which any two distinct
vertices are connected by an edge. It is immediate that the degree of any vertex
in the complete graph is n − 1. The complete graph on five vertices is given in
Fig. 5.5.
• The linear graph on n vertices, Ln , is the connected graph on n vertices consisting
of a single Euler tour with no repeated vertices. We give an example of L7 in
Fig. 5.6.
• A cycle graph on n vertices, Cn , is the connected graph on n vertices consisting
of a single Euler cycle with no repeated vertices. Each vertex in this graph has
degree 2. We give the example of C6 in Fig. 5.7.
• The wheel graph, Wn+1 , is formed from the cycle graph Cn by adding an
additional vertex that is connected to every other vertex. The graph W7 is given
in Fig. 5.8.
Fig. 5.5 Complete graph on five vertices
Fig. 5.6 The linear graph on seven vertices
Fig. 5.7 The cycle graph on

six vertices
Fig. 5.8 The wheel graph

with seven vertices
Let A and B be subsets of the set of vertices V with A ∪ B = V and A ∩ B = ∅.

The sets A and B are said to be a partition of V. The graph G = (V, E) is said to
be a bipartite graph if v ∈ A implies w ∈ B, whenever {v, w} ∈ E. That is, if two
vertices are connected then they must be in different sets of the partition.
If each vertex in A is connected to a vertex in B, then it is said to be the complete
bipartite graph Km,n with m = |A|, n = |B|. We give the complete bipartite graph
K4,3 in Fig. 5.9. Here A = {a, b, c, d} and B = {e, f, g}.
We can generalize the notion of bipartite as follows. Let A1 , A2 , . . . , Ak be a
partition of the set of vertices V. The graph G = (V, E) is said to be a k-partite
126 5 Graphs
Fig. 5.9 The graph K4,3
Fig. 5.10 The graph K2,3,2
graph if v ∈ Ai implies w ∈ / Ai , whenever {v, w} ∈ E. That is, if two vertices are

connected then they must be in different sets of the partition.
If each vertex in Ai is connected to each vertex in Aj when i = j then this graph is
said to be the complete k-partite graph. We give the complete tripartite graph K2,3,2
in Fig. 5.10. Here A1 = {a, b}, A2 = {c, d, e}, and A3 = {f, g}.
A graph is said to be regular if the degree of each vertex is the same. For example,
the complete graph, Kn , is a regular graph since each vertex has degree n − 1. The
cycle graph, Cn , is regular since each vertex has degree 2. The wheel graph, Wn ,
is not, in general, since one vertex has degree n − 1 and the rest have degree 3.
However, W4 is regular since n − 1 = 3. The linear graph Ln , n > 2, is not regular
since each vertex has degree 2 except the first and the last which both have degree 1.
Example 5.3 A Latin square graph is constructed as follows. Given a Latin square
L of order n, the vertices are the n2 coordinates of L. Then two vertices (a, b) and
(c, d) are connected with an edge if a = c, b = d, or La,b = Lc,d . That is, they
are connected if they are in the same row, column, or share the same symbol. Each
vertex is connected to n − 1 vertices from its column, n − 1 vertices from its row,
and n − 1 vertices from its symbol. Therefore, a Latin square graph is a regular graph
of degree 3(n − 1).
Let G = (V, E) be a graph then G = (V , E ) is a subgraph of G if V ⊆ V and

E ⊆ E.
Fig. 5.11 Subgraph of K4,3
Example 5.4 The graph in Fig. 5.11 is a subgraph of the complete bipartite graph
K4,3 .
Let G = (V, E) be a graph. Let R be an equivalence relation on the set V. Then

let GR = (VR , ER ) be the graph formed where VR are the equivalence classes on V
under the relation R and {[v], [w]} ∈ ER if and only if there exists a ∈ [v], b ∈ [w]
with {a, b} ∈ E. The graph GR is called the quotient graph of G under the relation
R.
Example 5.5 Let G be the graph given below on the set of vertices V = {a, b, c, d,
e, f}. Let R be the equivalence relation on V induced by the partition {{a, b}, {c, d},
{e, f}}. Then GR is the quotient graph given in Fig. 5.12.
Two graphs G = (V, E) and G = (V , E ) are isomorphic graphs if there exists

a bijection f : V → V such that {v, w} is an edge in E if and only if {f(v), f(w)} is
an edge in E . In this case, the map f is called an isomorphism.
Example 5.6 The three graphs in Fig. 5.13 are isomorphic.
Theorem 5.5 Let G = (V, E) and G = (V , E ) be isomorphic graphs with isomor-

phism φ. If v ∈ V, then deg(v) = deg(φ(v)).
Proof Let d be the degree of v. If the complete list of vertices to which v is connected
is v1 , v2 , . . . , vd , then the complete list of vertices to which φ(v) is connected is
φ(v1 ), φ(v2 ), . . . , φ(vd ). The result follows.
The next corollary follows immediately from Theorem 5.5.

128 5 Graphs
Fig. 5.12 Quotient graph
Fig. 5.13 Isomorphic graphs
Corollary 5.1 Let G = (V, E) and G = (V , E ) be isomorphic graphs then the list
of degrees of the vertices of V is exactly the list of degrees of the vertices of V .
Proof The list

deg(v1 ), deg(v2 ), . . . , deg(vn )
is identical to the list
deg(φ(v1 )), deg(φ(v2 )), . . . , deg(φ(vn )).
The result follows.
We illustrate this with the following example.
Example 5.7 It follows from Corollary 5.1 that no two of the following three graphs
are isomorphic, since the degree lists are (2, 2, 2, 2), (1, 2, 2, 1), and (1, 3, 2, 2),
respectively. Notice that even though the first and the third have the same number of
vertices and edges they are not isomorphic.
Theorem 5.6 If G = (V, E) and G = (V , E ) are isomorphic graphs then G is a

bipartite graph if and only if G is a bipartite graph.
Proof Assume G is a bipartite graph and let f be the isomorphism from G to G . Let
A and B be the bipartite partition of the vertices of G. Let A = f(A) and B = f(B).
Then A and B partition the vertices of G . Assume {v , w } ∈ E , then v = f(v)
and w = f(w) for some v, w ∈ V. Then {v, w} ∈ E. Without loss of generality, this
implies that v ∈ A and w ∈ B since G is bipartite. Then f(v) ∈ A and f(w) ∈ B .
Hence, G is a bipartite graph.
The other direction follows from Exercise 10 using f−1 as the isomorphism.
Exercises
1. Prove that if G and G are isomorphic then G has an Euler tour if and only if G
has an Euler tour.
2. Determine the number of edges in the complete graph on n vertices.
3. Prove that the complete graph Kn has an Euler cycle if and only if n is odd. Find
such an Euler cycle on K5 . Prove that the only complete graph with an Euler
tour is K2 .
4. Prove that if G and G are isomorphic then G is connected if and only if G is
connected.
5. Determine the number of edges in the wheel graph Wn and the number of edges
in the cycle graph Cn .
6. Determine the number of distinct bipartite graphs on a set with n vertices.
7. If A and B partition the set of vertices with |A| = a and |B| = b then determines
the number of edges in the corresponding complete bipartite graph.
8. Prove that isomorphism is an equivalence relation on the set of all graphs.
9. We found eight simple graphs on three vertices in Example 5.2. Partition these
eight graphs into equivalence classes induced by isomorphism.
10. Prove that if f is an isomorphism from G to G then f−1 is an isomorphism from
G to G.
11. Let G and G be two bipartite graphs on the same points sets with the same
bipartite partition. Define GR and GR to be the quotient graphs formed by the
relation that two vertices are related if and only if they are in the same set of the
partition. Prove that GR and GR are isomorphic.
12. Extend Theorem 5.6 to k-partite graphs.
130 5 Graphs
13. Find two graphs with five vertices and an equal number of edges that are not
isomorphic.
14. Prove that any two graphs with the same number of vertices and with no edges
must be isomorphic. Determine the number of isomorphisms between them.
15. Prove that any two graphs with the same number of vertices and with a com-
plete set of edges must be isomorphic. Determine the number of isomorphisms
between them.
5.3 Colorings of Graphs
We now move to a highly intuitive aspect of graph theory, namely, the coloring of
graphs. It arises from some natural questions, namely, how many colors are needed
to color a political map so that no two adjacent states have identical colors. It leads
to one of the most interesting and controversial theorems of the twentieth century,
namely, the four color map theorem. We begin with some definitions.
Definition 5.6 A coloring of a graph G = (V, E) is a map from V to a set C such

that if {v, w} ∈ E then f(v) = f(w).
Intuitively it means that no two connected vertices have the same color.
Definition 5.7 The chromatic number of a graph is the minimum number of colors
needed to color a graph. This is denoted by χ(G).
The only graph with χ(G) = 1 would be a graph with no edges. It is also obvious
that assigning each vertex a different color will give a coloring. The complete graph
on n vertices requires n colors. These two results give the standard bounds, that is,
1 ≤ χ(G) ≤ n for graphs with n vertices. We shall consider a few examples.
• Discrete graph:
The discrete graph on n vertices, Dn , has chromatic number 1 since each vertex
can be colored the same.
• Complete graph:
The complete graph on n vertices, Kn , has chromatic number n since each vertex
is connected to every other vertex. Therefore, no two vertices can have the same
color.
• Linear graph:
The linear graph Ln , with n ≥ 2 has chromatic number 2 since the vertices are
alternatively colored.
5.3 Colorings of Graphs 131
• Bipartite graph:
A bipartite graph has chromatic number 2, that is, one color for each set in the
partition. Since no two vertices in the same set of the partition are connected then
they can all have the same color.
• Cycle graph:
The cycle graph, Cn , has chromatic number 2 if n is even and 3 if n is odd. If n is
even the colors are simply alternated around the cycle. If n is odd this technique
will not work since the last color would be the same as the first and these two
vertices are adjacent. Hence, you need an additional color. We give the examples
of C4 and C5 which illustrate the point.
G B
R B
B G B G
• Wheel graph
The wheel graph, Wn , has chromatic number 4 if n is even and 3 if n is odd. By
removing the center vertex from the wheel graph we obtain the cycle graph. If n
is even then we have a cycle graph with oddly many vertices which requires three
colors. The additional vertex is adjacent with ever other vertex and so requires a
color distinct from the rest. If n is odd then we have a cycle graph with evenly
many vertices which requires two colors. Then, the same argument gives that we
need three colors to color the graph. We illustrate it with the following example
of W5 and W6 .
G B
R B
R
Y
B G B G
We shall now show how to determine the chromatic number of a disconnected

graph by determining the chromatic number of its components.
Theorem 5.7 Let G be a graph with components G1 , G2 , . . . , Gs . Then χ(G) =

max{χ(Gi )}.
132 5 Graphs
Proof Since no vertex in Gi is connected with any vertex in Gj when i = j, then

the graph can be colored with as many colors that are needed to color the component
with the largest chromatic number.
Example 5.8 Consider the following disconnected graph.
The first component has chromatic number 2, the second component has chromatic
number 2, and the third component has chromatic number 4. Hence, the chromatic
number of the graph is 4.
A graph is said to be a planar graph if it can be drawn in R2 such that no two edges
(when viewed as curves in R2 ) intersect at a point that is not a vertex. Of course,
this does not mean that every rendering of a graph is planar but rather that there is
rendering that is planar.
For example, the following are isomorphic graphs where one is given without
intersecting edges and one is not.
One of the most famous problems of the twentieth century was the four color
map problem. The problem was brought to public attention in 1854, most probably
by Francis or Frederick Guthrie [41]. For an early description by Cayley, see [23].
Essentially, it asks whether a standard political map can be colored so that no neigh-
bors have the same color with only four colors. It is assumed that all countries are
connected, that is, that there is a tour on the map between any two countries. The
problem can easily be transformed into a graph theory question. That is, take each
country as a vertex and connect two countries with an edge if they share more than
a single point. The problem then becomes the following conjecture.
Conjecture 5.1 Any planar graph be colored with four colors.

Of course, not all planar graphs require four colors but the question simply asks
if four colors will always suffice.
The solution of the problem did not occur until 1976 when it was solved by Appel
and Haken assisted by Koch. A description of the solution can be found in [1–3].
One drawback of this solution is that an extremely large number of cases were
eliminated by computer—far more than could be examined by hand. This aspect of
the proof caused a bit of controversy over whether this solution should indeed be
considered a proof in the traditional sense. Since this early computer proof, numerous
problems have been solved with an extensive computer search. One of the most
famous cases would be the proof of the non-existence of the projective plane of
order 10, see [58]. The difficulty is that unlike standard proofs, the veracity of the
proof is difficult to verify. Lam has suggested that mathematicians take a cue from
physical scientists and only accept the proof with independent verification. The great
difficulty here is that few people would be willing to undertake such an arduous task
simply to prove someone else’s result correct.
We can now examine the number of ways a graph can be colored with a given set
of colors rather than how many colors are necessary to color it. Let G be a graph and
let PG (n) be the number of ways to make a coloring of G with n or fewer colors.
Certainly, this is a function, but it may not be as obvious that this is a polynomial for
all graphs G. We refer to this as the chromatic polynomial of the graph and denote
it by PG (x).
• Discrete graph:
The discrete graph on n vertices, Dn , has chromatic polynomial PG (x) = xn ,
since each vertex can be colored any of the x colors.
• Complete graph:
The complete graph on n vertices, Kn , has chromatic polynomial PG (x) = x(x −
1)(x − 2) · · · (x − n + 1) since each vertex that is colored lowers the number for
each other vertex by 1. For example, PK5 (x) = x(x − 1)(x − 2)(x − 3)(x − 4).
• Linear graph:
The linear graph Ln with n ≥ 2 has chromatic polynomial PG (x) = x(x − 1)n−1
since you have x choices for the first and then x − 1 for each of the following.
Example 5.9 Consider the complete bipartite graph K2,3 given in Fig. 5.14.
Assume there are x colors to color the graph. If the two on the left side are
colored the same then there are x choices for this, and each of those on the right
has x − 1 choices. If the two on the left side are colored differently then there are x
choices for the first and x − 1 for the second leaving x − 2 for each of those on the
right. This gives that the chromatic polynomial is x(x − 1)3 + x(x − 1)(x − 2)3 =
x(x − 1)(x3 − 5x2 + 10x − 7).
134 5 Graphs
Fig. 5.14 The complete

bipartite graph K2,3
Theorem 5.8 Let G be a graph with components G1 , G2 , . . . , Gs . Then

s

PG (x) = PGi (x). (5.1)
i=1
Proof Any coloring of Gi can be matched with any coloring of Gj with i = j since
no vertex of Gi is connected to any vertex in Gj . The result follows.
Example 5.10 Consider the following disconnected graph:
The first component has chromatic polynomial x(x − 1), the second has chromatic
polynomial x(x − 1)(x − 2), and the third has chromatic polynomial x(x − 1)3 .
Therefore, the chromatic polynomial of the graph is x3 (x − 1)5 (x − 2).
Theorem 5.9 Let G = (V, E) be a simple graph, let e = {a, b} ∈ E, and let Ge
be the graph (V, E − {e}). Then let R be the equivalence relation on V formed by
equating a and b. Then
PG (x) = PGe (x) − PGR (x). (5.2)

Proof Given a coloring of Ge , either a and b have the same color or they have
different colors. If they have different colors then it is a coloring of G. If they are of
the same color then it is a coloring of GR . This gives that PG (x) + PGR (x) = PGe (x)
Example 5.11 Consider the following graph, which is the complete graph on three
vertices.
b
a c
This graph has chromatic polynomial PG (x) = x(x − 1)(x − 2).
The graph Ge is
b
a c
and has chromatic polynomial PGe (x) = x(x − 1)2 .
The graph PGR is
a c
and has chromatic polynomial x(x − 1). Then
PGe (x) − PGR (x) = x(x − 1)2 − x(x − 1)

= x(x − 1)(x − 1 − 1)
= x(x − 1)(x − 2) = PG (x).
Exercises
1. Find the chromatic number of the following graph:

136 5 Graphs
2. Take a political map of one of the inhabited continents and construct its planar
graph G. Determine the chromatic number χ(G) of this map. Which continent(s)
have the highest chromatic number and which have the lowest chromatic number?
3. Construct a planar graph that requires four colors, that is, a graph with chromatic
number 4.
4. Determine the chromatic polynomial of the complete tripartite graph K2,3,2 .
5. Determine the chromatic polynomial of the complete bipartite graph K2,2 .
6. Draw the Latin square graph for the Latin square
⎛ ⎞
123
⎝2 3 1⎠.
312
Determine its chromatic number.

7. Let Δ(G) be the maximum degree of any vertex in the graph. Prove that χ(G) ≤
Δ(G) + 1. This can be done by ordering the vertices by degree and assigning an
available color to each vertex in this order. This is known as greedy coloring.
8. Prove that the complete bipartite graph K3,3 is not planar. This is a puzzle often
played by children. That is, they draw three houses and three utilities and try to
connect each utility to each house without crossing the connecting lines.
5.4 Directed Graphs and Relations 137
5.4 Directed Graphs and Relations
In addition to multigraphs and simple graphs, there are also directed graphs. In a
directed graph, the edges have a direction, so there is a difference between an edge
starting at vertex a and ending at vertex b and an edge starting at vertex b and ending
at vertex a. A directed graph G = (V, E) is a set of vertices V and a set of edges E
where (a, b) ∈ E if and only if there is an edge from a to b. Notice that the edges
here are not sets but ordered pairs since (a, b) and (b, a) are not the same edge. We
begin with an example of a directed graph on six vertices in Fig. 5.15.
For a directed graph, we need more general definitions of degree. For a vertex v, we
define degreein (v) = |{(w, v) | (w, v) ∈ E}| and degreeout (v) = |{(w, v) | (v, w)
∈ E}|.
The adjacency matrix M for a graph G = (V, E) is indexed by V and Ma,b = 1
if (a, b) ∈ E and 0 otherwise (Table 5.1).
Notice that the number of ones in the row corresponding to a vertex v is
degreeout (v) and the number of ones in the column corresponding to a vertex
v is degreein (v). Note also that the matrix is not symmetric.
Fig. 5.15 Directed graph on six vertices
Table 5.1 Adjacency matrix for the directed graph

A B C D E F
A 0 1 0 0 1 0
B 1 0 0 0 0 0
C 0 0 1 0 0 0
D 1 0 1 0 0 0
E 0 0 0 0 0 0
F 1 0 0 0 1 0
138 5 Graphs
Each edge adds to the degreein and degreeout of exactly one vertex. Thus, for
a directed graph G = (V, E), we have

2|E| = (degreein (v) + degreeout (v)).
v∈V
Recall that a relation on a set A is a subset of A × A.
Theorem 5.10 There is a bijective correspondence between the set of relations on

a set A of cardinality n and the set of directed graphs on a set of n-vertices.
Proof The correspondence is easy to describe. That is, the n vertices correspond to
the n elements of A. Then (a, b) is in the relation if and only if (a, b) is an edge in
the directed graph.
Example 5.12 Consider the relation on the set {a, b, c, d} given by
R = {(a, a), (a, c), (c, a), (d, d)}.
Then its corresponding graph is the following:
b d
a c
Likewise we have the following theorem.
Theorem 5.11 There is a bijective correspondence between the set of binary square
matrices of order n and the set of directed graphs on a set of n-vertices.
Proof The correspondence is easy to describe. That is, the n vertices correspond to
the n rows and columns of a matrix M. Then Ma,b = 1 if and only if (a, b) is an
edge in the directed graph.
Example 5.13 Given the graph in Example 5.12, its matrix M is indexed by a, b, c, d
and is
⎛ ⎞
1010
⎜0 0 0 0⎟
M=⎜ ⎝1 0 0 0⎠.
⎟
0001
5.4 Directed Graphs and Relations 139
We see from these two theorems that there is a bijective correspondence between
directed graphs, binary matrices, and relations on a set.
We shall describe some properties of relations and how they correspond to the
directed graph. A relation R on a set A is reflexive if (a, a) ∈ R for all a ∈ A. In
terms of the graph this means that at every vertex there is a loop. For the matrix, this
means that there is a 1 in every entry in the diagonal.
A relation is symmetric if (a, b) ∈ R implies (b, a) ∈ R. For the graph, this means
that if there is an edge going from a to b then there is an edge from b to a. For the
matrix, this means that the matrix is symmetric, that is, M = MT .
A relation is transitive if (a, b) and (b, c) are in R then (a, c) ∈ R. For the graph,
this means that if there is an edge from a to b and an edge from b to c then there is
an edge from a to c. For the matrix, this property is not as easily seen.
A tour in a directed graph G = (V, E) is an ordered k-tuple (e1 , e2 , . . . , ek ),
ei = (vi , wi ) ∈ E, where wi = vi+1 and ei = ej if i = j. In this definition, we say
that the tour is from the vertex v1 to wk .
We can mimic the definition of a Hamiltonian tour and cycle. The only difference
is that a tour must respect the direction of an edge. This means that a tour can contain
the edge from a to b only if there is an edge starting at a and ending at b. Namely,
in the tour you must follow the direction of the edge.
A complete directed graph is the graph such that for any vertex a and b, (a, b) is
an edge in the graph. Note that this graph has more than C(n, 2) edges, which is the
number of edges for the simple complete graph. It has 2C(n, 2) edges, since each
direction must have an edge as well. This is assuming that there are no loops, meaning
no edge connects a vertex to itself. If we allow loops then there are 2C(n, 2) + n
edges.
Exercises
1. Draw the directed graph for the relation
{(0, 1), (1, 2), (0, 0), (2, 4), (4, 3), (3, 4), (2, 2), (4, 1), (4, 2)}
on the set {0, 1, 2, 3, 4}. Give the in and out degrees of each vertex.
2. There are eight subsets of the set {reflexive, symmetric, transitive}, given
an example of a directed graph that corresponds exactly to each of these subsets.
Namely, the graph has the property if and only if it is in the subset.
3. Prove that in a directed graph corresponding to a transitive relation, that if there
is a tour from vertex v to vertex w, then (v, w) is an edge in the graph.
4. Mimic the proofs in Theorems 5.1 and 5.2 to determine when there exists
Euler tours and Euler cycles in a directed graph. Hint: Consider degreein and
degreeout for each vertex.
5. Prove that the number of Hamiltonian cycles on the complete directed graph
with n vertices is (n − 1)!.
Higher Dimensional Finite Geometry
6
We shall now turn our attention to higher dimensional geometries using the theory of
linear algebra as our setting. This approach can be used for finite or infinite geometry
although it is often not presented this way in most undergraduate geometry classes.
It is, however, a very powerful technique which reduces many difficult geometric
proofs to very simple algebraic proofs.
As an example of viewing geometry in algebraic terms, consider the real Euclidean
plane. Here, we can consider the points as elements (x, y) ∈ R2 . The lines on this
plane can be considered as one-dimensional subspaces (lines through the origin) and
their cosets (translates). A parallel class consists of all translates of a subspace. For
example, the subspace may consist of all (x, y) satisfying y = mx. Then we consider
all points in the space {(0, b) + (x, y)} where (x, y) satisfy y = mx. This new set
is all the points that satisfy the equation y = mx + b. Therefore, the lines are one-
dimensional vector spaces and their cosets and the usual algebraic manipulations can
be done to them to prove geometric results.
In three dimensions, the planes are two-dimensional subspaces and their translates.
This is usually described by taking the set of all (x, y, z) perpendicular to some
vector a, b, c, which describes a subspace of dimension 2 in this space, and then
specifying a point that it must pass through which determines which coset of this
space we want. If we are in n-dimensional space Rn , then the points are the elements
of the space, the lines are one-dimensional subspaces, their cosets and the planes are
two-dimensional subspaces and their cosets, and any k-dimensional subspace and
its cosets are k-dimensional geometries. We say that a n − 1-dimensional geometry
contained in n-dimensional space is a hyperplane.
A similar construction can be done to construct projective geometry, which we
shall describe in detail later. We shall begin our discussion by recalling some of the
necessary definitions and theorems from linear algebra that we shall need.
142 6 Higher Dimensional Finite Geometry
6.1 Linear Algebra
We begin by recalling the definition of a vector space over a field F. In this text, we
generally use a finite field but the same definition applies to any field.
Definition 6.1 A vector space V over a field F is a non-empty set V with addition
and scalar multiplication such that for all v, w, u ∈ V and all α, β ∈ F, we have
1. (Closure) v + w ∈ V;
2. (Commutativity) v + w = w + v;
3. (Associativity) v + (w + u) = (v + w) + u;
4. (Additive Identity) there exists 0 ∈ V with 0 + v = v + 0 = v;
5. (Inverses) there exists −v ∈ V with v + (−v) = (−v) + v = 0;
6. (Closure) αv ∈ V;
7. (Associativity) (αβ)v = α(βv);
8. (Distributive Property) (α + β)v = αv + βv;
9. (Scalar Identity) 1v = v;
10. (Distributive Property) α(v + w) = αv + αw.
Notice that we never used that F is a field in the definition. It is, however, vital in
the theory that we are going to pursue. If the field F were replaced with a ring then
these same axioms are used to define a module. This small change in the definition
makes enormous changes in the results. For example, consider the ring Z4 . The
space {0, 2} is a non-trivial module over this ring contained in the ring. This does not
happen for fields. There are no non-trivial subspaces of a field.
Definition 6.2 We say that W is a subspace of a vector space V if W is a vector

space and W ⊆ V.
Example 6.1 A line through the origin in a subspace of R2 and a plane through the
origin in a subspace of R3 .
Example 6.2 The space consisting of {(0, 0), (1, 1)} is a subspace of Fn
2.
Theorem 6.1 Let V be a vector space and let W ⊆ V. Then, if αv + βw ∈ W for

all v, w ∈ W and all α, β ∈ F, then W is a subspace of V.
Proof Assume αv + βw ∈ W for all v, w ∈ W and all α, β ∈ F. Axioms 2, 3, 7, 8, 9,

and 10 all follow from the fact that W ⊆ V. Axiom 1 follows by taking α = β = 1.
Axiom 4 follows by taking α = β = 0. Axiom 5 follows by taking α = −1 and
β = 0. Axiom 6 follows by taking β = 0. Therefore, W is a vector space.
6.1 Linear Algebra 143
Example 6.3 Let v ∈ V, V a vector space. Then the set W = {αv | α ∈ F} is a

subspace of V. We have αv + βv = (α + β)v ∈ W. Then by Theorem 6.1, W is a
subspace.
Definition 6.3 A set ofvectors v1 , . . . , vn is a spanning set of V if every vector in

V can be expressed as n i=1 αi vi .
We shall denote the set spanned by vectors v1 , . . . , vn by v1 , . . . , vn . Note that

a spanning set is not necessarily minimal. For example, R2 = (1, 0), (0, 1), (1, 1)
and hence {(1, 0), (0, 1), (1, 1)} is a spanning set for R2 ; however, {(1, 0), (0, 1)} is
also a spanning set for R2 .
Example 6.4 Let ei be the vector of length n, that is, 1 in the i-th coordinate and 0
elsewhere. The set e1 , e2 , . . . , en is a spanning set for Fn where F is any field.

Definition 6.4 A set of vectors {v1 , . . . , vn } is linearly independent if αi vi = 0
implies αi = 0 for all i.
Example 6.5 The set {(1, 2), (2, 1)} is not linearly independent over F3 since
1(1, 2) + 1(2, 1) = (0, 0). The set {(1, 0), (0, 1) is linearly independent over any
field since α(1, 0) + β(0, 1) = (0, 0) implies (α, β) = (0, 0) and then α = 0 and
β = 0.
Definition 6.5 If a set {v1 , . . . , vn } is a linearly independent spanning set for V then
{v1 , . . . , vn } is a basis for V. We then say that V has dimension n.
Example 6.6 The set e1 , e2 , . . . , en defined in Example 6.4 is a basis for Fn for
any field F.
We note that the basis in the previous example is not the only possible basis. For
example, {(1, 0), (1, 1)} is also a basis for F22 .
Theorem 6.2 A dimension k vector space over the field Fq has qk elements.
Proof Let v1 , v2 , . . . , vk be a basis for a k-dimensional vector space over Fq . Then

each element of the form k i=1 αi vi is distinct since the vectors are linearly inde-
pendent. For each αi there are q choices which gives qk vectors.
Example 6.7 Consider the two linearly independent vectors over F3 : (1, 1, 0),
(0, 1, 2). The space generated by these two vectors consists of all vectors of the form
α(1, 1, 0) + β(0, 1, 2) where α, β ∈ F3 . Hence, there are nine vectors in (1, 1, 0),
(0, 1, 2). Specifically, they are {(0, 0, 0), (1, 1, 0), (2, 2, 0), (0, 1, 2), (1, 2, 2), (2, 0,
2), (0, 2, 1), (1, 0, 1), (2, 1, 1)}.
We can now give a theorem that counts the number of subspaces of a given vector
space over a finite field.
Theorem 6.3 Let V be an n-dimensional vector space over Fq . The number of

subspaces of V of dimension k is
(qn − 1)(qn − q) · · · (qn − qk−1 )

. (6.1)
(qk − 1)(qk − q) · · · (qk − qk−1 )
Proof In V there are qn elements and so there are qn − 1 ways of picking a non-zero
element. That non-zero element generates a one-dimensional space with q elements.
This means there are qn − q ways of picking an element not in that space. These
two elements generate a space with q2 elements giving qn − q2 ways of picking
an element not in that space. Continuing, by induction, we see that the numerator
counts the number of ways of picking k linearly independent vectors from the space.
The bottom counts, in the same manner, the number of different bases that this
k-dimensional subspace has.
Example 6.8 Assume q = 2, n = 4 and k = 2. Then the number of two-dimensional

subspaces of a four-dimensional space over F2 is
(24 − 1)(24 − 2) 15(14)

2 2
= = 35.
(2 − 1)(2 − 2) (3)(2)
of k-dimensional subspaces of an n-dimensional space over Fq is

The number
n
denoted by . The notation is close to the notation for binomial coefficients and
k q
there are corresponding identities, such as the following theorem.
Theorem 6.4 Let n, k be integers with 0 ≤ k ≤ n, then

n n−1 k n−1
= +q . (6.2)
k q k−1 q k q
Proof Let V be a k-dimensional vector space over Fq . Let x be a vector in V. We

can write V as V = x ⊕ W where W is a k − 1-dimensional subspace of V. Every
k-dimensional subspace either contains x or does not contain it. If it does contain
x then it is of the form x ⊕ W where
W is a k − 1-dimensional subspace of
n−1
W. Hence, there are precisely such subspaces.
k−1 q
Let w1 , w2 , . . . , wk denote a basis for a k-dimensional subspace of W. There are
n−1
such subspaces. Let vi = wi + αi x. Then v1 , v2 , . . . , vk form a basis for
k q
a k-dimensional subspace of V that is not contained in W. Since there are q choices
for each αi there are qk such vector spaces. This accounts for all k-dimensional
subspaces of V which gives the identity.
Example 6.9 Let q = 5. Then

4 (54 − 1)(54 − 5)
= 2 = 806,
2 5 (5 − 1)(52 − 5)

3 (53 − 1)(53 − 5)
= 2 = 31,
2 5 (5 − 1)(52 − 5)

3 (53 − 1)
= = 31.
1 5 (5 − 1)
The recursion gives

4 3 2 3
= +5 = 31 + (25)(31) = 806. (6.3)
2 5 1 5 2 5
This recurrence leads to a Pascal type triangle. Label

the rows by n and the
n
columns by k, then we get the following tables for when q = 2 and q = 3.
k q

n
The values of when q = 2
k q
1
1 1
1 3 1
1 7 7 1 (6.4)
1 15 35 15 1
1 31 155 155 31 1
1 63 651 1395 651 63 1

n
The values of when q = 3
k q
1
1 1
1 4 1
1 13 13 1 (6.5)
1 40 130 40 1
1 121 1210 1210 121 1
1 364 11011 33880 11011 364 1
We can also do a strictly computational proof of Theorem 6.4. We have

n n−1 (qn − 1)(qn − q) · · · (qn − qk−1 ) n−1
− = −
k q k−1 q (qk − 1)(qk − q) · · · (qk − qk−1 ) k−1 q

(qn − 1)q(qn−1 − 1)q(qn−1 − q) · · · q(qn−1 − qk−2 ) n−1
= −
(qk − 1)q(qk−1 − 1)q(qk−1 − q) · · · q(qk−1 − qk−2 ) k−1 q

qn − 1 n − 1 n−1
= −
qk − 1 k − 1 q k−1 q
n
q −1 n−1
= ( k − 1)
q −1 k−1 q
n k
q −q n−1
= ( k )
q −1 k−1 q

qn−k − 1 n − 1
= qk ( k )
q −1 k−1 q

n−1
= qk .
k q
This gives that

n n−1 k n−1
− =q .
k q k−1 q k q
It follows that

n n−1 n−1
= + qk .
k q k−1 q k q
Examining the Pascal type triangle, we note that all of the elements were congruent
to 1 (mod q). We prove this fact as an easy consequence of the recursion.
Corollary 6.1 The number of k-dimensional subspaces of Fn

q , k ≤ n, is congruent
to 1 (mod q).
Proof The result follows by induction, noting that by the recursion, we have

n n−1 n−1
= + qk . (6.6)
k q k−1 q k q
It follows that

n n−1
≡ (mod q). (6.7)
k q k−1 q

n
Then we note that ≡ 1 (mod q), for n ≥ 1 and we have the result.
1 q
Example 6.10 Consider the values in Table 6.5. The values
1, 4, 13, 40, 130, 121, 121, 364, 11011, 33880
are all congruent to 1 (mod 3).
One of the most powerful tools in the study of linear algebra is the linear trans-
formation. A linear transformation T : V → W, where V and W are vector spaces
over a field F, is a map that satisfies, for v, w ∈ V, α ∈ F,
T (v + w) = T (v) + T (w);
T (αv) = αT (v).
The kernel of a transformation T is given by
Ker(T ) = {v | v ∈ V and T (v) = 0}. (6.8)
The image is
Im(T ) = {w | there exists v inV with T (v) = w}. (6.9)
We shall now give two important lemmas.
Lemma 6.1 If T is a linear transformation T : V → W then Ker(T ) is a subspace

of V.
Proof Let v, w ∈ Ker(T ), α, β ∈ F. Then T (αv + βw) = αT (v) + βT (w) = α0 +

β0 = 0 which gives that αv + βw ∈ Ker(T ) which gives that Ker(T ) is a subspace
of V.
Lemma 6.2 If T is a linear transformation T : V → W then Im(T ) is a subspace

of W.
Proof Let v, w ∈ Im(T ), α, β ∈ F. Then v = T (v ) and w = T (w ) for some v , w ∈

V. Then T (αv + βw ) = αT (v ) + βT (w ) = αv + βw ∈ Im(T ). Therefore, Im(T )
is a subspace of W.
Example 6.11 Let T : F32 → F32 , where

⎛ ⎞⎛ ⎞
110 a
T ((a, b, c)) = ⎝ 0 1 1 ⎠ ⎝ b ⎠ .
101 c
Then
Ker(T ) = {(0, 0, 0), (1, 1, 1)},
which is a subspace of F32 and
Im(T ) = {(0, 0, 0), (0, 1, 1), (1, 1, 0), (1, 0, 1)},
which is also a subspace of F32 .
This brings us to one of the most important theorems of linear algebra.
Theorem 6.5 Let T : V → W be a linear transformation. Then
dim(Ker(T )) + dim(Im(T )) = dim(V). (6.10)
Proof Let w1 , w2 , . . . , wk be a basis for the image, with vi ∈ V and T (vi ) = wi .

Let u1 , u2 , . . . , us be a basis for the kernel of T . We shall show that
{v1 , v2 , . . . , vk , u1 , u2 , . . . , us }
is a basis for V.
If v ∈ V then T (v) is in the image of T . This implies that T (v) = βi wi . Then
we have

T (v) = T ( βi vi )

T (v) − T ( βi vi ) = 0

T (v − βi vi ) = 0,

which implies that v − βi vi ∈ Ker(T ). This gives that v − βi vi = γi ui
and v is in the span of {v1 , v2 , . . . , vk , u1 , u2 , . . . , us }. Hence it spans V.
Next we shall show that the set is linearly independent. Assume

αi vi + βi ui = 0.
Then we have

T( αi vi + βi ui ) = 0

and αi vi + βi ui ∈ Ker(T ). This implies that αi = 0 for all i, sinceif it were
not then it would not be in the kernel. Then 0 = αi vi + βi ui = βi ui ∈
ker(T ) but we know that {u1 , u2 , . . . , us } are linearly independent so βi = 0 for all
i. Hence the set {w1 , w2 , . . . , wk , u1 , u2 , . . . , us } is a basis for V, which gives that
dim(V) = k + s = dim(Im(T )) + dim(Ker(T )).
Example 6.12 In Example 6.11, the ambient space had dimension 3, the kernel
had dimension 1, and the image had dimension 2. This illustrates the theorem as
3 = 1 + 2.
We shall define an inner product similar to the one used on the real numbers. Let
F be a field. On the space Fn
q define the inner product

[v, w] = vi wi . (6.11)
This inner product is called the Euclidean inner product.

If two vectors v, w have [v, w] = 0 then we say that v and w are orthogonal.
Let V be a subspace of Fn ⊥ n
q . Define V = {w | w ∈ Fq and [v, w] = 0 for all
v ∈ V}.
Lemma 6.3 Let u, v, w ∈ Fn q , α ∈ Fq , where Fq is the field of order q. Then [v +

u, w] = [v, w] + [u, w] and [αv, w] = α[v, w] = [v, αw].
Proof Let u, v, w ∈ Fn
q . Then

[v + u, w] = (vi + ui )wi

= vi wi + ui vi = vi wi + ui vi
= [v, w] + [u, w].
Next,

[αv, w] = αvi wi = α vi wi
= α[v, w].
Similarly,

[v, αw] = vi αwi = α vi wi
= α[v, w].
Lemma 6.4 Let V ⊆ Fn ⊥ n

q be a vector space, then V is a subspace of Fq .
Proof Let v, w ∈ V ⊥ . Then, let u ∈ V,
[αv + βw, u] = α[v, u] + β[w, u] = α0 + β0 = 0. (6.12)
Therefore, αv + βw ∈ V ⊥ and therefore V ⊥ is a subspace of Fn

q.
We use these lemmas to prove the following.

Theorem 6.6 Let V be a subspace of Fn q with basis {v1 , v2 , . . . , vk }. Then w ∈ V

⊥
if and only if [w, vi ] = 0 for all i.

Proof If v = αi vi ∈ V and w ∈ Fn
q , we have

[v, w] = [ αi vi , w]

= αi [vi , w].
Therefore, if [vi , w] = 0 for all i then w ∈ V ⊥ .

If w ∈ V ⊥ then [w, vi ] = 0 for all i since vi ∈ V.
Let {v1 , v2 , . . . , vk } be a basis for V, where V is a subspace of Fn q . Let M be

the k by n matrix whose i-th row is vi . We know from Exercise 16 that T (v) = Mv
is a linear transformation. Notice that T (w) = 0 if and only if [w, vi ] = 0 for all i.
This gives that Ker(T ) = V ⊥ . It is clear that w ∈ Im(T ) if and only if w ∈ V since
T (α1 , α2 , . . . , αn ) = αi vi .
Using Theorem 6.5, we have proven the following.
Theorem 6.7 Let V be a subspace of Fn ⊥

q , then dim(V) + dim(V ) = n.
We can use this result to prove the following theorem, which was evident when
noticing the symmetry in the Pascal type triangle.
Theorem 6.8 Let n and k be positive integers. Then

n n
= .
k q n−k q
Proof If V is a vector space in Fn ⊥

q of dimension k then V is a vector space in Fq
n
of dimension n − k. Therefore, there is a bijection between the set of k-dimensional

vector spaces in Fn n
q and the set of n − k-dimensional vector spaces in Fq given by
⊥ n
f(V) = V . It follows that the number of k-dimensional subspaces in Fq is equal to

n n
the number of n − k-dimensional subspaces in Fn q . Therefore, = .
k q n−k q

Again, we can give a purely computational proof. We have that

n (qn − 1)(qn − q) · · · (qn − qk−1 )
= .
k q (qk − 1)(qk − q) · · · (qk − qk−1 )
Factoring out all possible q from the numerator and denominator gives

n (qn − 1)(qn−1 − 1) · · · (qn−k+1 − 1)
= . (6.13)
k q (qk − 1)(qk−1 − 1) · · · (q − 1)
Similarly, we have

n (qn − 1)(qn − q) · · · (qn − qn−k−1 )
= .
n−k q (qn−k − 1)(qn−k − q) · · · (qn−k − qn−k−1 )
Factoring out all possible q from the numerator and denominator gives

n (qn − 1)(qn−1 − 1) · · · (qk+1 − 1)
= . (6.14)
n−k q (qn−k − 1)(qn−k−1 − 1) · · · (q − 1)
To show that the fraction

in Eq. 6.13 is equal to the fraction in Eq. 6.14, simply cross
multiply and get n i=1 (q i − 1). This gives the result.
Example 6.13 Let n = 5 and k = 2. Then

5 (35 − 1)(35 − 3) (35 − 1)(34 − 1)
= 2 2
=
2 3 (3 − 1)(3 − 3) (32 − 1)(3 − 1)
and

5 (35 − 1)(35 − 3)(35 − 32 ) (35 − 1)(34 − 1)(33 − 1)
= 3 = .
3 3 (3 − 1)(33 − 3)(33 − 32 ) (33 − 1)(32 − 1)(3 − 1)
It is clear that the two fractions are equal.
Let V be a subspace of Fnq then a coset of the subspace is w + V = {w + v | v ∈ V}.

We note that if w ∈ V then w + V = V. We now show that the coset of a vector space
is either the vector space itself or disjoint from the vector space.
Theorem 6.9 If w ∈
/ V then (w + V) ∩ V = ∅.
Proof Assume v ∈ w + V ∩ V. Then v = v + w for some v ∈ V, which gives w =

v − v and then w ∈ V which contradicts the assumption that w ∈
/ V.
We shall now show that the cardinalities of all cosets of a vector space have the
same cardinality as the vector space.
Theorem 6.10 Let V be a subspace of Fn n

q . Then, for w ∈ Fq , |V| = |w + V|.
Proof Consider the map f : V → w + V, where f(v) = w + v. We shall show that

this map is a bijection which shows that |V| = |w + V|. Let v1 , v2 be elements of V.
If f(v1 ) = f(v2 ) then w + v1 = w + v2 and therefore v1 = v2 . Therefore, f is an
injection. Given an element of w + V, it is of the form w + v by definition and so
f(v) = w + v and the map f is a surjection. Therefore, f is a bijection and we have
the result.
Finally, we have that the ambient space consists of the vector space and its cosets,
all of which have the same cardinality.
Corollary 6.2 The cosets of a vector space form a partition of the ambient space.
Proof The result follows from the fact that any two cosets are disjoint or identical
and that every element v is in v + V, where V is the vector space.
Example 6.14 A line through the origin is necessarily a vector subspace of a

Euclidean plane. Its cosets are the lines in a parallel class. These lines partition
the plane.
Example 6.15 Consider the subspace of F23 of dimension 1, V = {(0, 0), (1, 2), (2, 1)}.
There are two cosets of this vector space (1, 1) + V = {(1, 1), (2, 0), (0, 2)} and
(2, 2) + V = {(2, 2), (0, 1), (1, 0)}. These three sets partition the space F23 .
Exercises
1. Determine if the following set of vectors is linearly independent over F3 :
(1, 2, 1), (2, 0, 1), (2, 2, 0).
2. Prove that Fnq is a vector space for any field F.

3. Find a basis for Fnq for any field F.
4. Prove that any vector space has a basis.
5. Prove that {v1 , v2 , . . . , vk } is a linearly dependent set if and only if vi can be
written as a linear combination of vj , where j = i.
6. Prove that no basis of a vector space V over Fq is unique unless q = 2 and
dim(V) = 1.
7. Prove that any linearly independent set of vectors in a vector space V can be
extended to a basis for V.
8. Prove that any spanning set for a vector space V contains a basis for V.
9. Prove that the set V = {(x, y) | y = mx} is a subspace of F3q for any m ∈ F.
Determine the dimension of V.
6.2 Affine Geometry 153
10. Prove that the set V = {(x, y, z) | αx + βy + γz = 0} is a subspace of F3q for any
α, β, γ ∈ F. Determine the dimension of V. Determine what geometric object
this vector is.
11. Prove that the intersection of two vector spaces is a vector space.
12. Prove that any two bases for a vector space have the same number of vectors.
13. Prove that Fn q has dimension n.
14. Prove the identity in Theorem 6.4, by computing the right side and adding them
with a common denominator.
15. Prove that if T is a linear transformation T : V → W then T (0) = 0.
16. Let V = Fn q and let M be an r by n matrix with entries from a field F. Show that
T : Fn r
q → Fq defined by T (v) = Mv is a linear transformation.
6.2 Affine Geometry
We shall now describe classical Euclidean geometry in terms of the linear algebra
presented in the previous section. Euclidean refers to the fact that in these geometries
there exists parallel lines on a plane. The term affine is usually used instead of
Euclidean and we shall adopt this term.
While we can begin with any n-dimensional vector space over the field F, we
shall begin with the space Fn q for simplicity. We shall describe the n-dimensional
affine geometry over the field Fq which is denoted by AGn (Fq ).
Definition 6.6 If a set is either a k-dimensional subspace or a coset of a k-

dimensional vector space then we say that it is a k-flat.
Affine Geometry AGn (Fq )
• The points of AGn (Fq ) are the elements of Fn q.

• The lines of the space are the one-dimensional subspaces and their cosets, namely,
the 1-flats.
• The planes of the space are the two-dimensional subspaces and their cosets,
namely, the 2-flats.
• The k-dimensional geometries are the k-dimensional subspaces and their cosets,
namely, the k-flats.
• Hyperplanes are n − 1-dimensional subspaces and their cosets, namely, n − 1-
flats.
• A geometric object A is incident with a geometric object B if and only if A ⊂ B
or B ⊂ A. Namely, incidence is given by set-theoretic containment.
We note that by Corollary 6.2, the k-flats of a single subspace partition the ambient
space for any k.
As an example we shall take a look at the space AG3 (F2 ). There are 23 = 8
points in the space. They are
(0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0), (1, 1, 1).
By Theorem 6.7, we can find the two-dimensional subspaces by taking the orthog-
onals of the one-dimensional subspaces corresponding to the seven non-zero points.
By Theorem 6.2 we know that each two-dimensional subspace has four points in it.
Theorem 6.10 gives that the cardinality of a coset is the same as the cardinality of the
subspace. Therefore, we know there can be only one coset for each two-dimensional
subspace, specifically the complement of the subspace in the space F32 . We could
have counted the number of two-dimensional subspaces using Theorem 6.3 which,
with n = 3, k = 2 and q = 2, would give that there are
(23 − 1)(23 − 2) 7(6)

= =7
(22 − 1)(22 − 2) 3(2)
subspaces of dimension 2. This gives that there are 14 planes in AG3 (F2 ), 7 cor-
responding to the subspaces and 7 to their cosets. We list them by describing the
subspace as an orthogonal and then listing its complement.
P1 = (0, 0, 1)⊥ = {(1, 1, 0), (1, 0, 0), (0, 1, 0), (0, 0, 0)}
P2 = P1c = {(1, 1, 1), (1, 0, 1), (0, 1, 1), (0, 0, 1)}
P3 = (0, 1, 0)⊥ = {(1, 0, 0), (0, 0, 1), (1, 0, 1), (0, 0, 0)}
P4 = P3c = {(1, 1, 0), (0, 1, 1), (1, 1, 1), (0, 1, 0)}
P5 = (0, 1, 1)⊥ = {(1, 0, 0), (0, 1, 1), (1, 1, 1), (0, 0, 0)}
P6 = P5c = {(1, 1, 0), (0, 0, 1), (1, 0, 1), (0, 1, 0)}
P7 = (1, 0, 0)⊥ = {(0, 1, 1), (0, 1, 0), (0, 0, 1), (0, 0, 0)}
P8 = P7c = {(1, 1, 1), (1, 1, 0), (1, 0, 1), (1, 0, 0)}
P9 (1, 0, 1)⊥ = {(0, 1, 0), (1, 0, 1), (1, 1, 1), (0, 0, 0)}
P10 = P9c = {(1, 1, 0), (0, 0, 1), (0, 1, 1), (0, 1, 1)}
P11 = (1, 1, 0)⊥ = {(0, 0, 1), (1, 1, 0), (1, 1, 1), (0, 0, 0)}
c
P12 = P11 = {(1, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0)}
Table 6.1 Incidence matrix for the points and planes in AG3 (F2 )
(0, 0, 0) (0, 0, 1) (0, 1, 0) (0, 1, 1) (1, 0, 0) (1, 0, 1) (1, 1, 0) (1, 1, 1)
P1 1 0 1 0 1 0 1 0
P2 0 1 0 1 0 1 0 1
P3 1 1 0 0 1 1 0 0
P4 0 0 1 1 0 0 1 1
P5 1 0 0 1 1 0 0 1
P6 0 1 1 0 0 1 1 0
P7 1 1 1 1 0 0 0 0
P8 0 0 0 0 1 1 1 1
P9 1 0 1 0 0 1 0 1
P10 0 1 0 1 1 0 1 0
P11 1 1 0 0 0 0 1 1
P12 0 0 1 1 1 1 0 0
P13 1 0 0 1 0 1 1 0
P14 0 1 1 0 1 0 0 1
P13 = (1, 1, 1)⊥ = {(1, 1, 0), (0, 1, 1), (1, 0, 1), (0, 0, 0)}
c
P14 = P13 = {(1, 1, 1), (0, 1, 0), (1, 0, 0), (0, 0, 1)}.
Looking at the entries in Table 6.1 we see that there are 8 points, 14 planes and
each point is on exactly 7 planes and each plane has exactly 4 points on it.
The lines in the space are the one-dimensional subspaces and their cosets. A
one-dimensional subspace has two elements and so each subspace has four cosets
(including itself), since four mutually exclusive sets of size 2 would make eight
points in total. The number of one-dimensional subspaces is seven since there is one
corresponding to each non-zero point. Alternatively, we could have used Theorem 6.3
which, with n = 3, k = 1 and q = 2, would give that there are
(23 − 1) 7
1
= =7
(2 − 1) 1
subspaces of dimension 1. Hence, there are 28 lines in AG3 (F2 ).

A plane is a two-dimensional subspace or a coset of a two-dimensional sub-
space. Given two planes their intersection can either be two dimensional, if they
coincide, zero dimensional if they meet in a point or if they are parallel (or
skew in higher dimensions), or one dimensional if it is a line. For example, con-
sider the following two planes in AG5 (F2 ), π1 = (1, 0, 0, 0, 0), (0, 1, 0, 0, 0) and
π2 = (0, 0, 1, 0, 0), (0, 0, 0, 1, 0). It is clear that these two planes have only the
origin in common. Consider the coset of π2 that is π2 + (0, 0, 0, 0, 1). This plane
does not intersect π1 at all nor is it a coset of π1 .
Theorem 6.11 Let p1 , p2 be any two points in AGn (Fq ). Then there is a unique
line through p1 and p2 .
Proof Let v be the vector (p1 − p2 ) and let w be the vector (p1 − 0). Let V = v.
Then V is a one-dimensional subspace. Let L = w + V. Then p1 = w + 0v ∈ w + V
and p2 = w − v ∈ w + V. Therefore, p1 , p2 ∈ L, and L is a line.
If M is any other line containing p1 and p2 then M must be of the form u +
(p1 − p2 ). The cosets of V are disjoint, therefore since M and L are not disjoint,
we have that L = M.
Example 6.16 Consider the two points (0, 1, 2) and (2, 1, 1) in AG3 (F3 ). We have
that the difference of the two points gives the vector v = (2, 0, 2) and so the vec-
tor space V = (2, 0, 2) = {(0, 0, 0), (2, 0, 2), (1, 0, 1)}. The line is (0, 1, 2) + V =
{(0, 1, 2), (2, 1, 1), (1, 1, 0)}. Notice the line has three points and contains the two
points (0, 1, 2) and (2, 1, 1).
We can give the standard geometric definition of parallel and skew lines.
Definition 6.7 Two lines that are on the same plane that are disjoint or equal are
said to be parallel. Two lines that are not on the same plane that are disjoint are said
to be skew lines.
Theorem 6.12 Any three non-collinear points in AGn (Fq ) are incident with a
plane.
Proof Let p1 , p2 , and p3 be three non-collinear points. Let v be the vector (p1 −
p2 ), w be the vector (p1 − p3 ), and u be the vector (p1 − 0). Since the points are
not collinear we have that V = v, w is a two-dimensional subspace of Fn q . Then
let P = u + V. We have p1 = u + 0v, p2 = u − v and p3 = u − w. Therefore, all
three points are on P and P is the coset of a two-dimensional subspace. Hence, the
three points are on a plane.
Theorem 6.13 Any two lines in AGn (Fq ) either intersect at a point are parallel or
skew.
Proof Assume that the two lines are neither skew nor parallel. If the two lines share
two points p1 and p2 , then there is a unique line through p1 and p2 giving that the
lines are equal. Hence, they can only intersect in a point.
We shall now count the number of k-dimensional subspaces of affine space.

n
Theorem 6.14 The number of k-dimensional spaces in AGn (Fq ) is qn−k .
k q
Proof A k-dimensional space is either a k-dimensional subspace of Fnq or a coset

n
of such a space. There are such subspaces and each has qn−k distinct cosets.
k q
Example 6.17 In AGn (Fq ), we have the following counts of spaces:
• Let k = 0. The number of points is qn .

n −1
• Let k = 1. The number of lines is qq−1 qn−1 .
(qn −1)(qn −q) n−2
• Let k = 2. The number of planes is q .
(q2 −1)(q2−q)

n qn −1
• Let k = n − 1. The number of hyperplanes is qn−(n−1) = q−1 q.
k q
We note that, in general, the number of k-dimensional spaces is not equal to the
number of n − k-dimensional spaces.
The Game of Set
The game of Set™ is a well-known mathematics game. The game evolved from a
study of genetics rather than from mathematics. However, it is really an idea that we
have just discussed. It is played with a collection of cards with four attributes: shape
(ovals, squiggles, diamonds), color (red, purple, green), number (one, two, three),
and shading (solid, striped, outlined). A set consists of three cards in which each of
the cards’ features is the same on each card or is different on each card. To begin 12
cards are placed down and the first player that finds a set calls “set” and removes the
three cards in that set. These cards are then replaced and the play continues until the
entire deck is used. There are 81 cards in the deck.
Notice that given any two cards in a Set deck there is a unique card that completes
the set (prove this in Exercise 2). It follows that there are 1080 sets (see Exercise 4).
Consider each of the three choices for each attributes as 0, 1, or 2 assigned arbitrarily.
Then each card represents an element in F43 , which gives 81 cards. Therefore, we
with a point in AG4 (F3 ).
can associate each card
4
In F43 there are = 40 one-dimensional subspaces. Each of these has 27
1 3
cosets in F43 . Therefore, there are 1080 lines in AG4 (F3 ). This is, of course, the
exact number of sets in the game. Moreover, we know that any two cards complete
to a unique set and any two points complete to a unique line. Therefore, the game of
Set consists in trying to identify lines in AG4 (F3 ). For a complete description of the
mathematics of this game see “The joy of Set. The many mathematical dimensions
of a seemingly simple card game” [65].
Exercises
1. Give the incidence matrix for points and lines in AG2 (F3 ).
2. Prove that given any two cards in a Set deck there is a unique card that completes
the set.
3. Determine the number of points, lines, and planes in AG3 (F3 ).
4. Prove that there are 1080 sets in the game of Set.
5. Determine the number of points and hyperplanes in AG4 (F5 ).
6. Prove that the construction of AG2 (Fq ) given in this section is equivalent to the
construction of the affine plane given in the section on planes constructed from
fields.
7. Prove that every point in AGn (Fq ) is on a k-flat.
8. Give an example of skew lines in AG4 (F2 ).
9. Give an example of parallel planes in AG4 (F2 ).
6.3 Projective Geometry
We shall now describe projective geometry, which has a canonical description in

terms of linear algebra as well.
To describe n-dimensional projective space, PGn (Fq ), we use the ambient space
Fn+1
q , that is, n + 1-dimensional space over the field F. The k-dimensional projective
space will be described in terms of vectors spaces of this ambient space. In general,
the projective dimension of a geometric object is 1 less than the dimension as a vector
space of the associated space.
Projective Geometry PGn (Fq )
• The points in PGn (Fq ) are the one-dimensional subspaces of Fn+1
q .
• The lines in PGn (Fq ) are the two-dimensional subspaces of Fn+1
q .
• The planes in PGn (Fq ) are the three-dimensional subspaces of Fn+1
q .
• The hyperplanes in PGn (Fq ) are the n-dimensional subspaces of Fn+1 q .
• A projective geometric object A is incident with a projective object B if A is a
subspace of B or B is a subspace of A.
In general, an object with projective dimension s corresponds to an s + 1-

dimensional subspace. Notice that there is no mention of flats or cosets here. In
projective space, all of the objects are subspaces of Fn+1
q , that is, cosets are not
projective geometric objects.
Theorem 6.3 gives that the number of k-dimensional subspaces of a vector space
of dimension n is
(qn − 1)(qn − q) · · · (qn − qk−1 )
. (6.15)
(qk − 1)(qk − q) · · · (qk − qk−1 )
This gives the following theorem.
6.3 Projective Geometry 159
Theorem 6.15 In PGn (Fq ), the number of objects with projective dimension k is

n+1 (qn+1 − 1)(qn+1 − q) · · · (qn+1 − qk )
= . (6.16)
k+1 q (qk+1 − 1)(qk+1 − q) · · · (qk+1 − qk )
Proof We note that an object with projective dimension k is a subspace of dimension

k + 1 in an n + 1-dimensional space. Then Eq. 6.15 gives the result.
q3 −1
As an example, the number of points (k = 0) in PG2 (Fq ) is q−1 = q2 + q + 1
(q3 −1)(q3 −q)
and the number of lines (k = 1) is (q2 −1)(q2 −q) = q2 + q + 1. Consider the points
in PG2 (F3 ). They are the one-dimensional vector spaces in F33 . The number of points
is 32 + 3 + 1 and they are (0, 0, 1), (0, 1, 0), (0, 1, 1), (0, 1, 2), (1, 0, 0),
(1, 0, 1), (1, 0, 2), (1, 1, 0), (1, 1, 1), (1, 1, 2), (1, 2, 0), (1, 2, 1), and
(1, 2, 2).
Example 6.18 In PG4 (F3 ), we have the following counts of spaces:

5
• There are = 121 points.
1
3
5
• There are = 1210 lines.
2
3
5
• There are = 1210 planes.
3
3
5
• There are = 121 hyperplanes.
4 3
Incidence is given by set-theoretic containment, so that two spaces A and B are

incident if and only if A ⊂ B or B ⊂ A. We can now count the number of such
incidences.
If we fix the bigger space and count how many of the smaller spaces are on it we
can use the following slight adjustment of the previous theorem.
Theorem 6.16 In PGn (Fq ) the number of objects with projective dimension k in a
projective space of dimension s, with s > k, is
(qs+1 − 1)(qs+1 − q) . . . (qs+1 − qk )

. (6.17)
(qk+1 − 1)(qk+1 − q) . . . (qk+1 − qk )
Proof We are simply counting the number of k + 1-dimensional subspaces of an

s + 1-dimensional vector space.
If we fix the smaller space and count how many bigger spaces contain it then we use
the following theorem.
Theorem 6.17 If A has projective dimension k in PGn (Fq ), then the number of
s > k projective dimensional objects it is incident with is
(qn+1 − qk+1 )(qn+1 − qk+2 ) · · · (qn+1 − qs )

. (6.18)
(qs+1 − qk+1 )(qs+1 − qk+2 ) · · · (qs+1 − qs )
Proof We are looking for the number of s + 1-dimensional subspaces of Fn+1 q

containing a specific subspace of dimension k + 1. So the number of ways of picking
an element not in the k + 1-dimensional space is qn+1 − qk+1 . Then we pick an
element not in the space generated by the k + 1-dimensional space and the new
vector. We continue by induction until we have chosen (s + 1) − (k + 1) = s −
k vectors that are linearly independent over the k + 1-dimensional subspace. The
denominator counts the number of different basis for the larger space over the smaller
one.
As a simple example, we can count the number of lines that are incident with
3 −q
a point, i.e., s = 1, k = 0, n = 2. Equation 6.18 gives that the number is q
q2 −q
=
q + 1 as expected.
Theorem 6.18 Through any two points in PGn (Fq ) there exists a unique line.
Proof Let p, q be two points with p = v, q = w with p = q, i.e., v ∈

/ w, w ∈
/
v. Then the two-dimensional subspace v, w contains p and q. Any other line
through p and q must be two dimensional and contain v and w and hence must be
this line.
It is possible for two lines to not meet in PGn (Fq ). For example, in PG4 (F2 ),
the lines (1, 0, 0, 0, 0), (0, 1, 0, 0, 0) and (0, 0, 0, 1, 0), (0, 0, 0, 0, 1) do not share
a common one-dimensional subspace.
It is true, however, that in any plane the lines must meet. That is, given any 2 two-
dimensional spaces, they cannot have only a trivial intersection since together they
would generate a four-dimensional space contained in a three-dimensional space.
They cannot have a two-dimensional intersection since they would be the same line,
so their intersection must be a one-dimensional subspace, that is, a point. This idea
generalizes to the following.
Theorem 6.19 Two objects of projective dimensions s and s must intersect if they
are contained in a space of projective dimension s + s .
Proof Let M and M be s and s projective dimensional objects. Then M is an s + 1-

dimensional subspace and M is an s + 1-dimensional subspace. If they have only a
trivial intersection then together they generate an s + s + 2-dimensional space, but
they are contained in an s + s + 1-dimensional subspace which is a contradiction.
Hence, they must have a non-trivial intersection.
6.3 Projective Geometry 161
As an example, consider the space PG3 (F2 ). In this space, there are 15 points, 35
lines, and 15 planes. There are 5 lines through each point and 3 points on each line.
Each plane consists of 7 points and through each point there are 7 planes. Through
each line there are 3 planes and each plane contains 8 lines.
These seemingly very abstract ideas about affine and projective planes are actually
very useful in a practical engineering sense. In [4] (and elsewhere), it is described
how to use these spaces in the construction of error-correcting codes.
Theorem 6.20 The number of objects with projective dimension r is equal to the
number of objects with projective dimension n − r − 1 in PGn (Fq ).
Proof Each projective r-dimensional object corresponds to an r + 1-dimensional

vector space. The orthogonal of this vector space has dimension n + 1 − (r + 1) =
n − r. These n − r-dimensional subspaces correspond to projective n − r − 1-
dimensional objects. Then, since the map that sends a vector space to its orthogonal
is a bijection, we have that the number of objects with projective dimension r is equal
to the number of objects with projective dimension n − r − 1 in PGn (Fq ).
This theorem gives that the number of points is equal to the number of lines on
the projective plane PG2 (Fq ).
Example 6.19 In PG3 (F2 ), there are 15 points, 35 lines, and 15 planes. Each line
has 3 points on it and each plane has 7 points and 7 lines. Each line is incident with
3 planes.
Theorem 6.21 Any two lines in PG2 (Fq ) must intersect.
Proof Let L and M be two lines. These correspond to two-dimensional vector spaces
V and W. If the lines do not intersect then the dimension of the intersection of V
and W must be 0. However, in this case, V, W has dimension 4 but sits inside F3q
which is a contradiction. Therefore, L and M must intersect.
Theorem 6.22 Any three non-collinear distinct points are incident with a unique
plane in PGn (Fq ).
Proof Let u, v, w be three non-collinear points. Since they are distinct u, v
is a two-dimensional subspace. Then since they are collinear u, v, w cannot be
two dimensional so it must be three dimensional. Therefore, u, v, w is a plane
containing all three points.
Exercises
1. Give the incidence matrix of points and lines in PG2 (F3 ).

2. Give the incidence matrix of points and planes in PG3 (F2 ).
3. Count the number of lines in PG3 (F5 ).
4. Determine the number of points, lines, planes, and the number of each incidences
for PG3 (F3 ).
5. Prove that every point in PGn (Fq ) is contained in a k-dimensional object.
6. Determine the number of planes containing a given line in PG3 (Fq ).
7. Determine how large n must be there for two planes to exist that do not intersect
in PGn (Fq ).
8. Prove that the construction of PG2 (Fq ) given in this section is equivalent to the
construction of the projective plane given in the section on planes constructed
from fields.
6.4 Desargues’ Theorem
One of the most interesting and important configurational theorems for projective
planes is Desargues’ theorem. Desargues’ Theorem does not apply in all projec-
tive planes. Those planes in which it does hold are known as Desarguesian planes.
These are precisely the classical planes that we constructed from finite fields, that is,
PG2 (Fq ).
Theorem 6.23 (Desargues’ Theorem) Let p, q, r and p , q , r be two sets of distinct

non-collinear points in the projective plane PG2 (Fq ). That is, p, q, r and p , q , r
are triangles. Denote the line opposite a point by its capital letter. Let L1 denote
the line through p and p , L2 denote the line through q and q , and L3 denote the
line through r and r . The lines L1 , L2 and L3 meet in a point if and only if the
intersection points of P with P , Q with Q and R with R are incident with a line.
Proof We assume that all points and lines in the theorem are distinct to avoid the
trivial cases. The duality principle implies that we need only prove one direction of
the if and only if theorem.
We assume the lines L1 , L2 , and L3 meet in a point. That is, we assume L1 ∩ L2 ⊆
L3 . Recall that each line is a two-dimensional subspace of F3q . Let v ∈ R ∩ R , that
is, v is a vector in the one-dimensional subspace that is contained in both the space R
and R , where R is the line through p and q and R is the line through p and q . The
vector v can be expressed as vp + vq , where vp is a vector in the one-dimensional
space p and vq is a vector in the one-dimensional space q. This is because v is on the
line between p and q. In the same way, the vector v can be expressed as vp + vq
where vp is a vector in the one-dimensional space p and vq is a vector in the
6.4 Desargues’ Theorem 163
one-dimensional space q . Then we have
vp − vp = vq − vq ∈ p, p ∩ q, q = L1 ∩ L2 ⊆ L3 = r, r .
This follows since vp − vp ∈ L1 and vq − vq ∈ L2 and vp − vp = vq − vq so

the vector is in the intersection. We know the intersection is contained in L3 which
is generated by r and r . Thus, for some vr ∈ r, vr ∈ r we have
vp − vp = vq − vq = vr + vr .
We shall show that v = vp + vq = (vp − vr ) + (vq + vr ) ∈ (Q ∩ Q ), (P ∩ P ).

First, vp − vr = vp + vr and so vp − vr ∈ Q ∩ Q . Then, vq + vr = vq − vr
therefore vq + vr ∈ P ∩ P . We have (vp − vr ) + (vq + vr ) = vp + vq ∈ R ∩ R .
Therefore, the intersection of R and R lies on the line between the intersection of Q
with Q and the intersection of P with P and we have the result.
We note that although the theorem is geometric in its statement, its proof is highly
algebraic.
We can now show that Desargues’ theorem is true in any space which has at least
three dimensions. We give the proof as it was given in [43].
Theorem 6.24 (Multidimensional Desargues’ Theorem) Let p, q, r and p , q , r be

two sets of distinct non-collinear points, that is, p, q, r and p , q , r are triangles,
such that the triangles lie on two different planes. Denote the line opposite a point by
its capital letter. Let L1 denote the line through p and p , L2 denote the line through
q and q and L3 denote the line through r and r . Assume the lines L1 , L2 and L3
meet in a point t, then the intersection points of P with P , Q with Q and R with R
are incident with a line.
Proof The line between p and q and the line between p and q both lie in the plane
through t, p, and q and intersect in a point s1 . Similarly, the line between p and r
and the line between p and r intersect in a point s2 and the line between r and
q and the line between r and q intersect in a point s3 . The points s1 , s2 , and s3
must be incident both with the plane containing p, q, and r and the plane containing
p , q , and r and therefore on M which is the line of intersection of these two planes.
Therefore, the lines through s3 and s2 and through s1 and s2 and s1 and s3 coincide
with this intersection and so the points s1 , s2 , and s3 are collinear.
There is another theorem that is equivalent to Desargues’ theorem which is also

sometimes used to characterize those planes that come from a finite field. This the-
orem is known as Pappus’ theorem. We state the theorem.
Theorem 6.25 Let L and M be two distinct lines in the projective plane PG2 (Fq ),
where p1 , p2 , andp3 are distinct points on L and q1 , q2 , andq3 are distinct points
on M, where none of these six points are the point of intersection of L and M. Let
r1 be the point of intersection of the line through p1 and q2 and the line through q1
and p2 ; let r2 be the point of intersection of the line through p1 and q3 and the line
through q1 and p3 ; and let r3 be the point of intersection of the line through p2 and
q3 and the line through q2 and p3 . Then the points r1 , r2 , and r3 are collinear.
The configuration in the theorem is known as the Pappus configuration.

Any plane in which Desargues’ theorem holds is known as Desarguesian and any
plane in which Pappus’ theorem holds is known as Pappian.
The following can be found in Pickert’s 1955 text [67] or more recently in Steven-
son’s 1972 text [83].
Theorem 6.26 Every Pappian plane is Desarguesian.
Exercises
1. Verify that the unique projective plane of order 2 satisfies both Desargues’ theorem
and Pappus’ theorem.
2. Verify that the unique projective plane of order 3 satisfies both Desargues’ theorem
and Pappus’ theorem.
6.5 The Bruck–Ryser Theorem
There are many open problems in terms of the existence of finite projective planes.
Constructively, we know that we can create a projective plane of order n if n is
a power of a prime, but there is really only one strong theorem eliminating orders
as possible orders for projective planes, namely, the Bruck–Ryser theorem. This
theorem first appeared in [17]. We shall describe this proof in this section.
Rather than giving the original proof by Bruck and Ryser, we shall follow the
proof given in Hughes and Piper which is simpler and requires less knowledge of
number theory.
We state two lemmas that we will require. Their proofs can be found in any number
theory book.
Lemma 6.5 If n is a positive integer then n can be written as n = a2 + b2 + c2 +

d2 , where a, b, c, andd are integers.
Lemma 6.6 If n can be written as n = q2 + r2 where q, r are rational numbers

then n can be written as n = a2 + b2 where a, b are integers.
We can now state and prove the well-known Bruck–Ryser theorem first proven
in [17].
6.5 The Bruck–Ryser Theorem 165
Theorem 6.27 (Bruck–Ryser Theorem) Let n ≡ 1 or 2 (mod 4). If there exists a

projective plane of order n then n = a2 + b2 for some integers a, b.
Proof Let Π = (P, L, I) be a projective plane of order n with P = {p1 , . . . ,

pn2 +n+1 } and L = {L1 , . . . , Ln2 +n+1 }.
Let xi be an indeterminant and write

Λi = xj .
j,(pj ,Li )∈I
That is, we sum over all j for which pj is incident with the line Li . Hence, each Λi
is the sum of n + 1 distinct xj .
Then
n2
+n+1 n2
+n+1
Λ2k = (n + 1) x2k + 2 xk xj . (6.19)
k=1 k=1 k=j
This is easy to see since each xj appears in n + 1 different Λi . Hence, when the Λi
are squared there are n + 1 occurrences of the square. Then xi and xj occur in the
same Λi exactly once so when squaring you get xi xj twice in the summation.
2 +n+1 2
Then pulling out one n k=1 xk we get the following:
n2
+n+1 n2
+n+1 n2
+n+1
Λ2k = (n) x2k + ( xk )2 . (6.20)
k=1 k=1 k=1
Adding nx2n2 +n+2 to each side we get
n2
+n+1 n2
+n+2 n2
+n+1
Λ2k + nx2n2 +n+2 = (n) x2k +( xk )2 . (6.21)
k=1 k=1 k=1
For the remainder of the section, let n ≡ 1 or 2 (mod 4) then n2 + n + 1 ≡ 3

(mod 4).
Lemma 6.5 gives that there are integers a, b, c, d with n = a2 + b2 + c2 + d2 .
Let M be the following matrix:
⎛ ⎞
a b c d
⎜ −b a d −c ⎟
⎜ ⎟ (6.22)
⎝ −c −d a b ⎠ .
−d c −b a
We note that det(M) = n2 . The matrix M gives a linear transformation defined

by Φ(v) = vM.
We say that an integer t is represented by (w, x, y, z) if t = w2 + x2 + y2 + z2 .
The following is an easy computation to verify
(a2 + b2 + c2 + d2 )(w2 + x2 + y2 + z2 )
= (aw − bx − cy − dz)2 + (bw + ax − dy + cz)2
+ (cw + dx + ay − bz)2 + (dw − cx + by + az)2 .
Using this equation, we see that if t is represented by (w, x, y, z) then (w, x, y, z)M
represents tn.
Let (y1 , y2 , y3 , y4 ) be the image of (x1 , x2 , x3 , x4 ) under the linear transfor-
mation given by M, that is,
(y1 , y2 , y3 , y4 ) = (x1 , x2 , x3 , x4 )M.
We have
n(x21 + x22 + x23 + x24 ) = (y21 + y22 + y23 + y24 ). (6.23)
The determinant of M is non-zero and so there exists an inverse matrix M−1
which also gives a linear transformation. There is no guarantee that entries in M−1
are integers, but it is true that they must all be rational numbers. What we have is
that
(y1 , y2 , y3 , y4 )M−1 = (x1 , x2 , x3 , x4 ).
This gives that each xi is written as a linear combination of the yi where the coeffi-
cients in the linear combination are rational numbers.
Applying Eqs. 6.20 and 6.23 to the indeterminants
y1 , y2 , y3 , y4 , x5 , x6 , . . . , xn2 +n+2 ,
we get
n2 +n+1
n2 +n+2
n2 +n+1

Λ2 2
k + nx 2 = y2 2 2 2
1 + y2 + y3 + y4 + n x2
i +( xi )2 . (6.24)
n +n+2
k=1 k=5 k=1
Notice that n2 + n + 2 ≡ 0 (mod 4) so we can repeat this process and arrive at
n2
+n+1 n2
+n+2 n2
+n+1
Λ2k + nx2n2 +n+2 = y2i +( yi )2 , (6.25)
k=1 k=1 k=1
where the indeterminates in Λ are yi .

If the coefficient of y1 in Λ1 is −1 then put Λ1 = −y1 and otherwise put L1 = y1 .
This can be solved to get y1 as a linear expression in the remaining yi . Substituting
into Eq. 6.25 we get
n2
+n+1 n2
+n+2 n2
+n+1
Λ2k + nx2n2 +n+2 = y2i + ( yi )2 . (6.26)
k=2 k=2 k=1
6.6 Arcs and Ovals 167
Continuing this we have
n2
+n+1
nx2n2 +n+2 = y2n2 +n+2 +( yi )2 . (6.27)
k=1
This gives that there are rational numbers α and β with
nα2 = 1 + β2
1 β
n = 2 + ( )2 .
α α
Lemma 6.6 now implies that n can be written as the sum of two integer squares,
which gives the theorem.
This theorem eliminates many possible orders. For example, 14 is 2 (mod 4) and
14 is not the sum of two squares, and hence there is no plane of order 14. Notice that
a plane of order 10 is not eliminated by the theorem but there is no plane of order
10. Hence, the theorem is not a biconditional.
Exercises
1. Determine which orders less than 50 the Bruck–Ryser theorem eliminates.

2. Prove that the Bruck–Ryser theorem eliminates all planes of order 2k where k ≡ 3
(mod 4).
3. Verify that the determinant of the matrix M in (6.22) is n2 .
4. Prove that the Bruck–Ryser theorem cannot eliminate a plane if the order is pe ,
where p is a prime and e > 0.
6.6 Arcs and Ovals
In any finite plane, we are always interested in describing geometric configurations

which may occur in the plane. We shall now describe an interesting and important
collection of configurations which occur in planes. These are known as arcs and
ovals.
Definition 6.8 A k-arc is a set of k points in a plane such that no three are collinear.
An arc of maximal size is said to be an oval.
As an example, consider the projective plane of order 2. In this plane with seven
points, any collection of four points that are a compliment of a line has the property

plane of order 3
that no three of them are on a line. Hence, in this plane, there are exactly seven
ovals. We shall see shortly that this case is quite different from the case for planes
of odd order. Specifically, the maximum size of an arc in a plane of odd order is
different than the maximum size of a plane of even order. We exhibit this result in
the following theorem.
Theorem 6.28 The maximum size of an arc in a plane of order n is n + 2 if n is

even and n + 1 if n is odd.
Proof Let A be the set of points of an arc and let p be a point in the arc. For all q ∈ A
there is a line through p and q. These lines are distinct because no line intersects
A in more than two places. Since there are n + 1 lines through q it shows that
|A| ≤ n + 2.
If |A| = n + 2 then by the above argument no line can intersect A exactly once.
Namely, it must either be disjoint or intersect twice. Let q be a point not in A.
If there are α lines through q that intersect A twice, then consider the set Aq =
{(p, L) | p ∈ A and the line L through p and q intersects A twice }. Since there
are α lines through q that intersect the arc twice, then |Aq | = 2α. Counting in a
different way there are n + 2 points in A and each has a corresponding element in
Aq . Therefore, 2α = n + 2 which gives n = 2(α − 1) and n must be even.
We call an n + 1 arc in a plane of odd order an oval and an n + 2 arc in a plane

of even order a hyperoval.
While it is true that no point can be added to one of these arcs and still be an arc,
there are arcs of smaller size that no point can be added to them.
Hyperovals are very important in the study of finite planes for a variety of reasons.
As an example of this importance, we note that one of the most important steps in
proving the non-existence of a projective plane of order 10 was showing that such a
plane would have no hyperoval.
Consider the diagram of the projective plane of order 3 in Fig. 6.1.
Fig. 6.2 4-arc in the

projective plane of order 3
The 13 lines are
{a, b, c, j}, {d, e, f, j}, {g, h, i, j}

{a, d, g, }, {b, e, h, }, {c, f, i, }
{c, e, g, m}, {a, f, h, m}, {b, d, i, m}
{a, e, i, k}, {c, d, h, k}, {b, f, g, k}
{j, k, , m}.
The points a, c, g, i form a 4-arc illustrated in Fig. 6.2.
Definition 6.9 A conic in a projective plane is the set of points satisfying a non-
degenerate (not the product of two linear equations) homogeneous polynomial of
degree 2.
Theorem 6.29 A conic in PG2 (Fq ) has q + 1 points.
Proof Consider the three coordinates x1 , x2 , x3 in F3q . Given two independent vari-
ables in a non-degenerate homogeneous polynomial of degree 2, there are q2 solu-
tions in F3q . Eliminating the all-zero vector and then dividing by q − 1 for the non-
q2 −1
zero multiples we get q−1 = q + 1 projective points satisfying the equation.
Example 6.20 The points in PG2 (F3 ) are
{(0, 0, 1), (0, 1, 0), (0, 1, 1), (0, 1, 2), (1, 0, 0), (1, 0, 1), (1, 0, 2),
(1, 1, 0), (1, 1, 1), (1, 1, 2), (1, 2, 0), (1, 2, 1), (1, 2, 2)}.
• The conic x1 x2 + x2 x3 + x1 x3 =0 contains the points {(0, 0, 1), (0, 1, 0),

(1, 0, 0), (1, 1, 1)}.
• The conic x21 + x22 + x23 =0 contains the points {(1, 1, 1), (1, 1, 2), (1, 2, 1),
(1, 2, 2)}.
• The conic x21 + x2 x3 =0 contains the points {(0, 0, 1), (0, 1, 0), (1, 1, 2),
(1, 2, 1)}.
• The equation (x1 + x2 + x3 )(x1 − x2 − x3 ) = x21 − x22 − x23 + x2 x3 contains
the points
(0, 1, 2), (1, 0, 2), (1, 1, 1), (1, 2, 0), (1, 0, 1), (1, 1, 0), (1, 2, 2).
That is, it contains the points (0, 1, 2), (1, 0, 2), (1, 1, 1), (1, 2, 0) of the first line
and the points (0, 1, 2), (1, 0, 1), (1, 1, 0), (1, 2, 2) of the second line. Hence, it is
degenerate.
Theorem 6.30 Any non-degenerate conic in PG(Fq ), q odd, is an oval.
Proof Any conic that contains at least three points of a line contains the entire line
and so is degenerate. Therefore, any line meets the conic in either 0, 1, or 2 points.
Therefore, it is an oval.
We see that all non-degenerate conics are ovals. But it is not at all obvious that
ovals should be conics. In [52], it was conjectured by Järnefelt and Kustaanheimo
that all ovals in a desarguesian finite projective plane of odd order were in fact conics.
Some mathematicians were not convinced by this conjecture. In fact, the review for
this paper (MR0054979) written by Marshall Hall states the following:
It is conjectured that in a plane with p2 + p + 1 points a set of p + 1 points, no three on a

line, will form a quadric. The reviewer finds this conjecture implausible.
This conjecture was later proved by Segre in [77]. In the review of this paper,
(MR0071034) also written by Marshall Hall, he states the following.
If, when n is odd, we call n + 1 points, no three on a line, an oval, then it was conjectured
by Järnefelt and Kustaanheimo that in a Desarguesian plane of odd order n, an oval is
necessarily a conic. This conjecture is shown to be true in this paper. The method of proof
is ingenious.
He ends with the following:
The fact that this conjecture seemed implausible to the reviewer seems to have been at least
a partial incentive to the author to undertake this work. It would be very gratifying if further
expressions of doubt were as fruitful.
We shall reproduce the proof given by Segre with the notation changed to match
that of the text.
We begin with some lemmas.
Lemma 6.7 Let Fq be a finite field of order q, then the product of all non-zero
elements in Fq is −1.
Proof In a field, there are at most two solutions to a polynomial of degree 2. The
equation x2 − 1 = 0 has solutions 1 and −1. Therefore, no other element is its own
inverse except for 1 and −1. If the characteristic of the field is 2 then 1 = −1 is a
repeated root as x2 + 1 = (x + 1)2 in this case.
Each non-zero element of the field must also have a multiplicative inverse. There-
fore, in α∈F∗q α each element is multiplied by its multiplicative inverse except for

−1. This gives that α∈F∗q α = −1.
p−1
This lemma is a generalization of Wilson’s theorem which states that i=1 i ≡
−1 (mod p) when p is a prime.
We let Π = PG(2, Fq ) with q odd. Let O be an oval in Π, that is, O consists of
q + 1 points, no three of which are collinear. Take an arbitrary point p in Π. The oval
O has a tangent at this point (where a tangent is a line that meets the oval exactly
once). We know from Theorem 3 in [69] that no three tangents of O meet at a point.
Lemma 6.8 Let Π = PG(2, Fq ), q odd and let O be an oval in Π. Every inscribed
triangle of O and its circumscribed triangle are in perspective.
Proof As usual we denote the points of Π as {(x1 , x2 , x3 ) | xi ∈ Fq } − {(0, 0, 0)}/ ≡

where (x1 , x2 , x3 ) ≡ (x1 , x2 , x3 ) if and only if xi = λxi for some non-zero λ ∈ Fq .
Without loss of generality we identify the inscribed triangle with the triangle
consisting of points p1 = (1, 0, 0), p2 = (0, 1, 0), and p3 = (0, 0, 1). Denote the
tangents of O at pi by Li where L1 is given by x2 = k1 x3 , L2 is given by x3 = k2 x1 ,
and L3 is given by x1 = k3 x2 , where ki is a non-zero element of the field Fq .
Let c = (c1 , c2 , c3 ) be any of the remaining q − 2 points of the oval distinct
from p1 , p2 and p3 . Then we have that c1 c2 c3 = 0 since otherwise a line would
intersect the oval in more than two points. Additionally, the lines p1 c, p2 c, and p3 c
have equations of the form x2 = λ1 x3 , x3 = λ2 x1 , and x1 = λ3 x2 where the λi
are distinct from the ki and are non-zero. The coefficients are then determined by
λ1 = c2 c−1 −1 −1
3 , λ2 = c3 c1 , λ3 = c1 c2 .
Then we have
λ1 λ2 λ3 = c2 c−1 −1 −1
3 c3 c1 c1 c2 = 1. (6.28)
Conversely, if λ1 denotes any of the q − 2 non-zero elements of the field distinct
from k1 , the line x2 = λ1 x3 meets the oval at p1 and some other point denoted by
r which is distinct from the pi . Then, the coefficients λ2 and λ3 in the equations
x3 = λ2 x1 and x1 = λ3 x2 of the lines p2 r and p3 r are functions of λ1 connected
by the fact that λ1 λ2 λ3 = 1, which take each of the non-zero of values of the field
distinct from k2 and k3 , respectively. Multiplying the q − 2 equations obtained in
this manner, we have that

( α)3 = k1 k2 k3 .
α∈F∗q

Lemma 6.7 gives that ( α∈F∗q α) = −1. Therefore, we have that
k1 k2 k3 = −1. (6.29)
This gives that the point of intersection of L2 and L3 is (k3 , 1, k2 k3 ); the point of
intersection of L3 and L1 is (k3 k1 , k1 , 1); and the point of intersection of L1 and
L2 is (1, k1 k2 , k2 ). These three points are joined to p1 , p2 , and p3 , respectively,
by the lines x3 = k2 k3 x2 , x1 = k3 k1 x3 , and x2 = k1 k2 x1 . By Eq. 6.29, these
lines concur at the point (1, k1 k2 , −k2 ) which is the center of perspective of the
triangles.
We can now prove Segre’s famous theorem.
Theorem 6.31 (Segre’s Theorem) Every oval of PG(2, Fq ), with q odd, is a conic.
Proof We retain all of the notations given in Lemma 6.8. With reference to
Lemma 6.8, we can, without loss of generality, assume that the lines concur at the
point (1, 1, 1), that is, we assume k1 = k2 = k3 = −1.
We can denote the line Lb tangent to the oval at the point b by b1 x1 + b2 x2 +
b3 x3 = 0. This tangent line contains the point b but does not contain the points
p1 , p2 , or p3 . Then let
β1 = b1 − b2 − b3 ,
β2 = −b1 + b2 − b3 ,
β3 = −b1 − b2 + b3 .
It follows that
b1 c1 + b2 c2 + b3 c3 = 0 (6.30)
and
b1 b2 b3 β1 β2 β3 = 0. (6.31)
Lemma 6.8 gives that the triangles bp2 p3 and bp2 p3 are in perspective. This
gives that

c3 − c2 c1 + c3 −c1 − c2

b1 − b3 b2 0
= 0.

b1 − b2 0 b3
Then, we have b2 (c1 + c2 ) = b3 (c1 + c3 ). The inscribed triangles bp3 p1 ,

bp1 p2 and their circumscribed triangles give
b3 (c2 + c3 ) = b1 (c2 + c1 )
b1 (c3 + c1 ) = b2 (c3 + c2 ).
These last three equations, Eqs. 6.30 and 6.31, imply that
c2 c3 + c3 c1 + c1 c2 = 0.
Therefore, the q − 2 points lie on the conic 4x2 x3 x3 x1 + x1 x2 = 0, which also con-
tains the points p1 , p2 , and p3 . Therefore, there are q + 1 points on it and therefore
the oval must coincide with it.
Theorem 6.32 Assume there is a hyperoval in a projective plane of even order. Any
line in this projective plane of even order is either secant to the hyperoval or disjoint
from the hyperoval.
Proof Let p1 , p2 , . . . , pn+2 be the points on a hyperoval. Assume there is a line

L that meets the hyperoval only once. Without loss of generality assume that L is
incident with p1 . Through p1 and pj , with j = 1, there is a unique line, call it
Mj , that does not intersect any other point on the hyperoval. Then we have that
L, M2 , M3 , . . . , Mn+2 are lines through p1 and there are n + 2 lines through p1
which is a contradiction. Therefore, no line is tangent to the hyperoval.
Let π be a projective plane of even order. Fix a line to be the line at infinity. We
refer to those hyperovals that are secant to the line at infinity as hyperbolic hyperovals
and to those that are disjoint from the line at infinity as elliptic hyperovals.
Definition 6.10 A Singer cycle is a map ψ : Π → Π such that ψ maps points to

points, lines to lines and preserves incidence, that is, an automorphism, such that for
2
any two points p and p in Π there exists an i such that ψi (p) = p and ψn +n+1
is the identity.
A Singer cycle is known to exist in any Desarguesian plane. In other words, a

Singer cycle is a permutation of the points on the plane preserving incidence. This
was first shown in [80].
We can count the number of hyperbolic and elliptic hyperovals in a Desarguesian
plane under a Singer cycle.
Theorem 6.33 In a Singer cycle of a hyperoval in a Desarguesian projective plane

(n+1)(n+2) n(n−1)
of even order, there are 2 hyperovals that are hyperbolic and 2
hyperovals that are elliptic for any choice of L∞ .
Proof Let O be a hyperoval in Π, where Π is a Desarguesian plane of even order.

Let ψ be the Singer cycle. Then consider the set {ψi (O)}, which is the set of images
of a hyperoval in a Singer cycle with ψ0 (O) = O. Each point on the hyperoval O
will have (n + 1) values of i such that ψi maps it to a point on L∞ . Each of these
values of i actually sends two points of O to L∞ since any line is either disjoint or
secant to a hyperoval.
Fig. 6.3 Projective plane of

order 2
(n+1)(n+2)
Now we can count the hyperovals. There are 2 hyperovals that are
2 (n+1)(n+2) n(n−1)
hyperbolic in a Singer cycle, leaving n + n + 1 − 2 = 2 hyper-
ovals that are elliptic in a Singer cycle.
Consider the projective plane of order 2 in Fig. 6.3.

There are exactly seven hyperovals in the projective plane, namely,
{D, E, F, G}
{A, B, E, F}
{B, C, D, E}
{B, C, F, G}
{A, C, D, F}
{A, B, D, G}
{A, C, E, G}
Given any of these hyperovals the others are in the image of the Singer cycle
of that hyperoval. Take as L∞ the line {BDF} then the theorem gives that there are
(n+1)(n+2) n(n−1)
2 = 6 hyperovals that are hyperbolic and 2 = 1 hyperovals that
are elliptic. The unique elliptic hyperoval in this case is {A, C, E, G}.
The number of hyperovals increases greatly as the order of the plane increases.
For example, there are 168 hyperovals in the projective plane of order 4. Using any
line as L∞ of this projective plane, there are 120 hyperbolic hyperovals and 48 elliptic
hyperovals.
Exercises
1. Verify that no line meets the point set {a, c, g, i} three times in the projective
plane of order 3.
2. Prove the existence of a Singer cycle in a Desarguesian plane. Hint: View the
points as elements of F3q / ≡ as in the construction of Desarguesian planes. Take
a field of order |Fq |3 that is a Galois extension of Fq and as such can be viewed
as a vector space over Fq . Then consider the action formed by multiplication of
the elements of this field extension by a single element.
6.7 Baer Subplanes 175
3. Find another arc in the projective plane of order 3.

4. Use Segre’s theorem to find the conic for the oval given in Exercise 2.
6.7 Baer Subplanes
We shall now look at projective planes which are contained in other projective planes.
This topic is also an important topic in the study of infinite projective planes.
If Π = (P, L, I) is a projective plane of order n then Σ = (P , L , I ) is a sub-
plane of Π if P ⊆ P, L ⊆ L, I ⊆ I, and Σ is a projective plane.
The following theorem is due to Bruck (see [17]).
Theorem 6.34 If Σ is a subplane of order m of a projective plane Π of order n, then

n ≥ m2 + m or n = m2 .
Proof Let be a line in L . Notice that is also a line in L but it has more points in
Π that are incident with it. In Σ, has m + 1 points and hence there are n − m points
on in Π that are not on in Σ. Any two lines in Σ meet in Σ. Hence, for every line in
Σ there are n − m points that are not in P . Hence, there are (m2 + m + 1)(n − m)
points in P − P that have a line of Σ passing through them. Notice that there are
n2 + n + 1 points in P, m2 + m + 1 in P and so we have
n2 + n + 1 ≥ (m2 + m + 1) + (m2 + m + 1)(n − m)

n2 + n + 1 ≥ m2 + m + 1 + m2 n − m3 + mn − m2 + n − m
n2 ≥ m2 n − m3 + mn
0 ≥ −n2 + m2 n − m3 + mn
0 ≥ (m2 − n)(n − m).
We know that n > m and so this gives that n ≥ m2 .

We can have equality, that is, n = m2 and then each of the inequalities above
becomes an equation.
Assume that we do not have equality, that is, n > m2 . Then there exists a point
q ∈ P such that no line of Σ is incident with it. Hence, any line through q can have
at most one point from P on it, since if there were two then the line through these
two points would be a line from Σ passing through q. Now if r is a point of Σ then
there is a line through q and r in Π, meaning that the number of lines through q is
at least as large as the size of P which is m2 + m + 1. This gives that
n + 1 ≥ m2 + m + 1
n ≥ m2 + m,

In the case when n = m2 , the subplane is known as a Baer subplane.

The following proof can be found in [18].
Theorem 6.35 If Σ is a Baer subplane of Π then every point of the Π lies on a line
of the Baer subplane Σ and every line of the plane Π contains a point of the Baer
subplane Σ.
Proof Given the duality of points and lines in a projective plane, we only need to
prove one of the statements. We shall prove the first.
Let Π be a plane of order m2 and Σ be a Baer subplane of order m. Any line in
the plane Π contains at most m + 1 points that are in Σ, and therefore it contains at
least one point which is not in Σ. Let L be a line of Π and let p be a point on L that
is not in the subplane Σ. Since Σ is a subplane, there can be at most one line of Σ
incident with the point p (otherwise the two lines of Σ would meet in a point not in
Σ giving a contradiction).
Given any point in Σ, it must have a line of Π through it and p since Π is a
plane. Therefore, the m2 + m + 1 points of Σ are contained in the m2 + 1 lines of
Π through p. If the line L contained no point of Σ, then the lines of Π incident with
p could account for at most (m + 1) + (m2 − 1) = m2 + m points of the subplane
Σ. Namely, there is at most one point of Σ meeting it in m + 1 points, 1 meeting it
at 0 points, and m2 − 1 meeting it at 1 point. Since there are m2 + m + 1 points of
Σ this is a contradiction. This gives that the line L contains a point of Σ and gives
our result.
Definition 6.11 A blocking set B in a projective plane is a subset of the points of

the plane such that every line of Π contains at least one point in B and one point not
in B.
The proof of Theorem 6.35 gives the following corollary.
Corollary 6.3 If Σ is a Baer subplane of Π then Σ is a blocking set of Π.
Exercises
1. Let Π be a projective plane with subplane Σ. Prove that if there is an isomorphism

from Π to the plane Π then Π contains a subplane isomorphic to Σ.
2. Find the Baer subplane in the projective plane of order 4.
3. Prove that in the Desarguesian plane PG2 (Fq2 ) there is a Baer subplane.
4. Determine for which orders there exist subplanes of PG2 (Fq ).
6.8 Translation Planes and Non-Desarguesian Planes 177
6.8 Translation Planes and Non-Desarguesian Planes
We begin with a definition that is required to define translation planes.
Definition 6.12 Let Π be a finite projective plane. Let P be a point on the plane and
let L be a line. We define a central collineation with center P and axis L to be an
isomorphism that fixes the points of L and the lines through the point P.
We say that the collineation is an elation if the point p is incident with the line L.
For a description of groups see Chap. 10.
Theorem 6.36 The set of all central collineations with center P and axis L forms a
group with function composition.
Proof It is immediate that the identity map is a central collineation. We know that
function composition is associative.
To show closure, let σ and τ be two central collineations with center P and axis
L. Then, if Q is a point on L then σ(τ(Q)) = σ(Q) = Q and if M is a line through
P then σ(τ(M)) = σ(M) = M.
Let σ be a central collineation with center P and axis L. Then, if Q is a point on L
then σ(Q) = Q and then Q = σ−1 (σ(Q)) = σ−1 (Q). If M is a line through P then
σ(M) = M and then M = σ−1 (σ(M)) = σ−1 (M). Therefore, the inverse map of
a central collineation with center P and axis L is a central collineation with center P
and axis L. This gives that the set is a group.
Recall that a group G acts transitively on a set A if for every a, b ∈ A, there is a

group element σ ∈ G with σ(a) = b.
Definition 6.13 If the group of elations acts transitively on the points of the affine
plane π = ΠL then the line L is called a translation line and a projective plane with
such a line is said to be a translation plane.
Given a translation plane Π with translation line L, the affine plane π = ΠL is

said to be an affine translation plane.
We shall now describe some planes which do not satisfy Desargues’ theorem. We
have seen from Theorem 6.24 that any plane that can be embedded in three dimen-
sions is a Desarguesian plane. There are, however, planes that are not Desarguesian.
It is known that every plane of order less than or equal to 8 is a Desarguesian plane,
but there are three non-Desarguesian planes of order 9.
Definition 6.14 A finite ternary ring R is a set R together with two distinguished ele-
ments 0 and 1 and a ternary operation T (a, b, c) = ab + c satisfying the following:
1. For all a ∈ R, we have T (1, a, 0) = T (a, 1, 0) = a.

2. For all a, b ∈ R, we have T (a, 0, b) = T (0, a, b) = b.
3. For all a, b, c ∈ R, T (a, b, y) = c has a unique solution for the variable y.
4. For all a, b, c, d ∈ R, with a = c, the equations T (x, a, b) = T (x, c, d) have a
unique solution for x ∈ R.
If we relax the definition to include infinite ternary rings we require another axiom.
Theorem 6.37 If R is a ternary ring, then R gives rise to an affine plane with the
usual construction.
Proof The points of the plane are (x, y) where x and y are elements of R. For each,
x ∈ R there is a line consisting of all points (x, y) where y is any element of R. All
other lines are of the form y = T (x, a, b), which corresponds to the line y = xa + b.
It is easy to see that this forms an affine plane of order |R|.
This theorem leads to the following important theorem.
Theorem 6.38 If R is a ternary ring, then R gives rise to a projective plane.
Proof Theorem 6.37 gives that we can form an affine plane π of order |R| from R,
then simply make the projective completion to form a projective plane Π of order
|R|.
If the ternary ring is a finite field then the plane is Desarguesian. Moreover, it is
known that if the ternary ring is not a finite field then the plane is not Desarguesian.
We can now give a definition of a near field which will be a ternary ring which
we can construct more easily to form non-Desarguesian projective planes.
Definition 6.15 A near-field (Q, +, ∗) is a set Q with two operations + and ∗ such
that the following hold:
1. (additive closure) If a, b ∈ Q then a + b ∈ Q.

2. (additive associativity) For all a, b, c ∈ Q we have (a + b) + c = a + (b + c).
3. (additive identity) There exists an element 0 ∈ Q such that 0 + x = x + 0 = x
for all x ∈ Q.
4. (additive inverses) For all a ∈ Q there exists b ∈ Q with a + b = b + a = 0.
5. (additive commutativity) For all a, b ∈ Q we have a + b = b + a.
6. (multiplicative closure) If a, b ∈ Q then a ∗ b ∈ Q.
c).
8. (multiplicative identity) There exists an element 1 ∈ Q such that 1 ∗ x = x ∗
1 = x for all x ∈ Q.
9. (multiplicative inverse) For all a ∈ Q − {0}, there exists b with a ∗ b = 1.
6.8 Translation Planes and Non-Desarguesian Planes 179
10. (unique solution) For all a, b, c ∈ Q, a = b there is exactly one solution to

xa − xb = c.
11. (right distributive) For all a, b, c ∈ Q we have (b + c) ∗ a = b ∗ a + c ∗ a.
Notice that it is not necessarily a division ring since it only has right distribution
and it is not necessarily a field for the same reason and also because the multiplication
is not necessarily commutative. Of course, division rings and fields are examples of
near fields.
A near field is a ternary ring with function T (a, b, c) = ab + c. Therefore, if we
can construct a near field then we can construct a non-Desarguesian projective plane.
We shall now construct a near-field J9 with elements {0, 1, −1, i, −i, j, −j, k, −k}.
Defining equations for the near field are given by
ijk = −1,
i = j2 = k2 = −1,
2
j = 1 + i,
k = 1 − i.
The addition and multiplication tables for this near field are given in (6.32) and
(6.33). We note that the multiplication is non-commutative.
+ 0 1 −1 i −i j −j k −k
0 0 1 −1 i −i j −j k −k
1 1 −1 0 j k −k −i −j i
−1 −1 0 1 −k −j i k −i j
i i j −k −i 0 k −1 1 −j
(6.32)
−i −i k −j 0 i 1 −k j −1
j j −k i k 1 −j 0 −1 −i
−j −j −i k −1 −k 0 j i 1
k k −j −i 1 j −1 i −k 0
−k −k i j −j −1 −i 1 0 k
∗ 0 1 −1 i −i j −j k −k
0 0 0 0 0 0 0 0 0 0
1 0 1 −1 i −i j −j k −k
−1 0 −1 1 −i i −j j −k k
i 0 i −i −1 1 k −k −j j
(6.33)
−i 0 −i i 1 −1 −k k j −j
j 0 j −j −k k −1 1 i −i
−j 0 −j j k −k 1 −1 −i i
k 0 k −k j −j −i i −1 1
−k 0 −k k −j j i −i 1 −1
Corollary 6.4 The near-field J9 gives a non-Desarguesian projective plane of

order 9.
Proof Apply Theorem 6.38 to the near-field J9 .
This plane is known as the Hall plane. It was first described in [43]. We shall give
a geometric construction that also gives rise to this plane as well as many others.
Let Π be the Desarguesian projective plane PG2 (q2 ). We know that this plane
contains a Baer subplane. A Baer subline is any line of the Baer subplane contained
in a given line. Let L be a line in Π and let be a Baer subline of Π. Let π = ΠL be
the affine plane formed by treating L as the line at infinity.
We shall construct τ = (A, M, J). Let A consist of the points of π, that is, the
points of Π with the points of L removed. If m is a line of π such that the intersection
of m and L, where m is the projective completion of m, is not a point of then
m ∈ M. A set E of n + 1 points of L is a derivation set if every pair of distinct points
q1 and q2 of π which determines a line meeting L in a point of E, there is a Baer
subplane containing the two points and E. In this case, we say that a Baer subplane
satisfying this condition belongs to E. Any Baer subplane that belongs to E restricted
to the affine plane π is also in M. The incidence relation J is given in the canonical
manner. The structure τ is an affine plane and we call the plane τ the derived plane of
Π. Its projective completion T is also called the derived plane. For q > 2, the derived
planes are non-Desarguesian.
Given this construction we can state the following theorem.
Theorem 6.39 There exists non-Desarguesian projective planes for all orders q2 ,
where q is a prime power and q > 2.
For a detailed description of non-Dearguesian planes see [93].
Exercises
1. Verify that the structure given in (6.32) and (6.33) is a near field.
Designs
7
We shall now describe a more general incidence structure than a finite geometry,
which is called a design. Affine and projective planes are examples of designs, but
of course not all designs are planes, nor are they all finite geometries in the sense
defined in the previous chapter. Essentially, the idea is that we wish to study a class
of incidence structures, namely, points and sets of points, with some structure. The
origin of the term comes from the fact that there were originally of interest as designs
of experiments. In the enormously influential book by Ronald A. Fisher, The Design
of Experiments [36], Fisher described how designs could be used to make various
experiments, including the lady tasting tea experiment which sought to determine if
Muriel Bristol could determine, by taste, whether the tea or the milk was added first
to a cup. The book is foundational in the study of statistics and Fisher himself was
largely a statistician, but the book also contains a chapter on Latin squares. Designs
have many interesting applications inside and outside of mathematics. For example,
they are still used extensively in the designs of experiments and in the construction of
tournaments. Additionally, they also have many interesting connections to algebraic
coding theory, graph theory, and group theory.
7.1 Designs
We begin with the standard definition of a design.
Definition 7.1 The structure D = (P, B, I) is a t-(v, k, λ) design if |P| = v, every

block B ∈ B is incident with exactly k points, and every t distinct point is together
incident with exactly λ blocks.
182 7 Designs
Fig. 7.1 Biplane of order 4
A projective plane of order n is a 2-(n2 + n + 1, n + 1, 1) design and an affine

plane of order n is a 2-(n2 , n, 1) design.
Later in the chapter we shall study k-nets of order n, which are a 1 − (n2 , n, k)
design; balanced incomplete block design (BIBD) which is when t = 2; designs with
the same number of points and blocks, that is, v = b, which are called symmetric
designs; and Steiner triple systems which are when k = 3 and λ = 1.
As an example of a design that is not a plane, consider the design of points and
planes in AG3 (F2 ). Its incidence matrix is given in Table 6.1. Here the points of
the design are the points in the space and the blocks in the design are the planes in
the space. Therefore, we have that v = 8 and k = 4. We see that through any three
points there is exactly one plane (since no three points are collinear in this space), so
t = 3 and λ = 1. We see that there are C(8, 3) = 56 ways of choosing three points
and each plane is counted C(4, 3) = 4 times so there are 56 4 = 14 planes. Therefore,
the points and planes in AG3 (F2 ) make a 3-(8, 4, 1) design with 14 blocks.
As another example of a design, we shall describe a biplane on 16 points. Consider
the 16 points corresponding to the 16 places on a 4 by 4 grid. There are 16 blocks
β(i, j) where β(i, j) corresponds to the 6 points that are in the i-th row and the j-th
column distinct from the point (i, j). In Fig. 7.1, we show the blocks β(1, 1) and
β(3, 2). Notice that these two blocks meet in two places. This design in Fig. 7.1 is a
2-(16, 6, 2) design.
The parameters t-(v, k, λ) are not the only ones that describe the design. We
denote the number of blocks by b and let r be the number of blocks through a point.
There are, of course, restrictions on these values.
Theorem 7.1 In a t-(v, k, λ) design, we have
vr = bk.
Proof To prove this we simply count the size of the set I, that is, the incidence
relation, in two different ways. That is we count the number of incidences. There are
v points and r blocks through them so there are vr incidences. On the other hand,
there are b blocks and k points on each block so there are bk incidences. Hence,
vr = bk.
7.1 Designs 183
Theorem 7.2 In a 2-(v, k, λ) design, we have
r(k − 1) = λ(v − 1).
Proof To prove this, we fix a point p1 in the design, and hence there are v − 1 points
distinct from p1 . For any other points p, there are exactly λ blocks through p and p1
since t = 2. Each block is counted as many times as there are points on it other than
p1 , that is, k − 1 times. This gives that the number of the blocks through p must be
(v−1)λ
k−1 . Of course, the number of blocks through p is r by definition which gives
the result.
Consider the 3-(8, 4, 1) design formed by the points and planes in AG3 (F2 ). We
know b = 14. Using the equation in Theorem 7.1 we have 8(r) = 14(4) which gives
that r = 7. This gives that through each point there are seven planes.
4 −1
Let us consider the case of points and planes in PG3 (Fp ). There are pp−1 = p3 +
2 3
p + p + 1 points and an equal number of planes. Hence v = b = p + p + p + 1. 2
On any plane there are p2 + p + 1 points, giving that k = p2 + p + 1. Then we use

vr = bk which gives (p3 + p2 + p + 1)r = (p3 + p2 + p + 1)(p2 + p + 1) and
then r = p2 + p + 1. That is, through each point there are p2 + p + 1 planes. If
three points are collinear, then there are more planes through them than if there were
not collinear so it is not a 3-design. Through any two points there are p3 + p2 + p +
1 − (p + 1) = p3 + p2 ways of choosing a non-collinear point. These three points
give a plane since they generate a three-dimensional vector space. Doing this, each
plane is picked (p2 + p + 1) − (p + 1) = p2 ways. Hence, through any two points
there are p + 1 planes. Therefore, it is a 2-(p3 + p2 + p + 1, p2 + p + 1, p + 1)
design.
If a set (v, b, r, k, λ) satisfies the equations vr = bk and r(k − 1) = λ(v − 1)
then the parameters are said to be admissible. However, simply because they are
admissible does not mean that a design exists. For example, there is an analog of the
Bruck–Ryser theorem for designs which we state here without proof.
Theorem 7.3 (Bruck–Ryser–Chowla) If a (v, k, λ) design of order n exists then

either
1. v is even and n is square or

2. v is odd and there exists integers a, b, c with c2 = na2 + (−1)(v−1)/2 λb2 .
We denote by λs , 0 ≤ s ≤ t, the number of blocks incident with s points. It

follows that λt = λ.
Let D = (P, B, I) be a t-(v, k, λ) design. Let S be a set of s points in P, with
0 ≤ s < t.
We shall count the number of sets of size t that contain S and the blocks containing
these sets of size t. Specifically, consider the set
A = {(T, B) | S ⊆ T ⊆ B, |T | = t, B a block in the design}.

184 7 Designs

Through any t points there are λ blocks and there are v−s t−s ways of choosing
t − s points to add to the s points of S to make t points. Hence, we have that

v−s
|A| = λ . (7.1)
t−s
Now through the s points of S there are λs blocks. In each of these blocks, there
are k − s points to choose t − s from to make a set of size t, then we have that

k−s
|A| = λs . (7.2)
t−s
This gives the following.
Lemma 7.1 Let D be a t-(v, k, λ) design, with λs be the number of blocks through
s points with 0 ≤ s ≤ t. Then we have

v−s k−s
λ = λs . (7.3)
t−s t−s
Proof This follows from counting the set A in two ways given in Eqs. 7.1 and 7.2.

We can now find a recurrence relation for λs .
Theorem 7.4 Let D be a t-(v, k, λ) design, with λs the number of blocks through s
points with 0 ≤ s ≤ t. Then we have
v−s
λs = λs+1 . (7.4)
k−s
Proof The previous lemma implies
(v−s)!
λ( (t−s)!(v−s−(t−s))! ) (v − s)!(k − t)!
λs = (k−s)!
=λ
(k − s)!(v − t)!
(t−s)!(k−s−(t−s))!
(v − s)(v − s − 1) · · · (v − t + 1)
=λ .
(k − s)(k − s − 1) · · · (k − t + 1)
It then follows immediately that
(v − s)
λs = λs+1 .
(k − s)

7.1 Designs 185
Let I and J be sets of points with |I| = i, |J| = j, and i + j ≤ t.

j
We define λi as the number of blocks containing I and disjoint from J. We notice
from this definition that λ0i = λi .
Let p be a point not in I nor J. Then each block through I, disjoint from J, is
j
either incident with p or is not incident with p, and so is counted either in λi+1 if it
j+1
is incident with p or in λi . That is, either p is adjoined to I to make a set of size
i + 1 or p is adjoined to J to make a set of size j + 1.
Theorem 7.5 Let D be a design then

j j j+1
λi = λi+1 + λi . (7.5)
This recursion can be used to make a type of Pascal’s triangle. Notice that each of
j
the λi must be a non-negative integer for a design to exist given a set of parameters.
Corollary 7.1 Let D be a design then

v−i−j
j
λi = λ k−i
v−t
. (7.6)
k−t
Proof We only need to show that it satisfies both the initial condition and the recur-
sion.
First we examine the initial condition. If j = 0 then by Eq. 7.4 we have
v−i
t−i
λ0i = λi = λ k−i
t−i
(v−i)!
(t−i)!(v−t)!
=λ (k−i)!
(t−i)!(k−t)!
(v − i)!(k − t)!
=λ .
(v − t)!(k − i)!
By the formula we have

v−i−0
λ0i = λi = λ k−i
v−t
k−t
(v−i)!
(k−i)!(v−k)!
=λ (v−t)!
(v−k)!(k−t)!
(v − i)!(k − t)!
=λ .
(v − t)!(k − i)!
186 7 Designs
Next we have
v−i−1−j v−i−j−1 v−i−j
k−i−1 k−i
λ v−t +λ v−t = λ k−i
v−t
(7.7)
k−t k−t k−t
and we have the result.
j
We can now produce a Pascal type triangle for the values of λi . We write as
follows (note that occasionally the mirror image of ours is given):
λ00
λ10 λ01
λ20 λ11 λ02 (7.8)
λ30 λ21 λ12 λ03
..
.
The minimum weight vectors in the extended Golay code form a very nice 5-
(24, 8, 1) design by the Assmus–Mattson theorem, which will be described in detail
later in the text. It has the following triangle:
759
506 253
330 176 77
(7.9)
210 120 56 21
130 80 40 16 5
78 52 28 12 4 1
For the affine plane of order n, there are n2 + n lines and hence λ0 = n2 + n.
We know that λ2 = 1. Then, using Eq. 7.4, we get λ1 = n + 1. Of course, we knew
this since there are n + 1 lines through a point. Then using Eq. 7.5 we can fill in the
table for the affine plane of order n and we have
n2 + n
n2 −1 n+1 (7.10)
n2 − n − 1 n 1
The design in Exercise 7 has t = 5, v = 72, k = 16, and 249849 blocks and is
the design that would be formed by the minimum weight vectors of a [72, 36, 16]
Type II code. It is not known if such a code exists. If one were able to show that the
design above does not exist then the code could not exist.
Let D = (P, B, I) be a design. Let w be a block in D and define D = (P , B , I )
where P = P − {p| (p, v) ∈ I}, B = B − {w} and a point is incident with a block
in D if it was incident in D. The structure D is called the residual design. We shall
prove that it is a design in the following.
7.1 Designs 187
Theorem 7.6 Let D be a t-(v, k, λ) design. Then D is a t-(v − k, k, λ) design.
Proof The fact that there are v − k points in D follows from the fact that there are
k points on each block of D.
Take any t points in D . These are also points in D and in D there were exactly
λ blocks. Each of these blocks has a restriction as a block in D , since none of the
points on the removed block are in D . Thus, through these t points, there are λ
blocks. If there were another block in D through these points then its extension
would be a block through these points in D. Hence, D is a t-(v − k, k, λ) design.
A parallel class in a design is a subset of the set of blocks that partitions the point
set and has no points of intersection between any two blocks in the set. We say that
a design is resolvable if the set of blocks can be partitioned into parallel classes. As
an example an affine plane is a resolvable design.
Recall that a BIBD is a design with t = 2. This is perhaps the most widely stud-
ied subset of designs. The terms block and design are obvious. The reason for the
incompleteness is that it is not simply the trivial design consisting of all possible
subsets of size k and balanced refers to t = 2.
j
For this case, the triangle of λi is
b
b−r r (7.11)
b − 2r + λ r−λ λ
Theorem 7.7 If D = (P, B, I) is a (v, k, λ) BIBD then if B = {b | b is the com-

plement of a block in B} then D = (P, B , I) is a (v, v − k, b − 2r + λ) BIBD.
Proof It is immediate that the number of points in D is v and that the number of
points on a block is v − k. The number of complements of blocks containing two
distinct points is the number of blocks that is disjoint from two points, i.e., λ20 . This
can be read from the above table which gives that there are b − 2r + λ complements
of blocks through any two points.
The design given in this theorem is known as the complementary design.

Consider the complementary design of a finite projective plane. Any two blocks
in this design have n2 points. A point is in the intersection of two of these blocks if
it was on neither line. Hence, any two blocks meet in exactly n2 − n points.
In the next few sections, we shall give several examples of BIBDs. The reader
should consult [24] for an encyclopedic description of many of the designs we con-
sider here.
188 7 Designs
Exercises
1. Verify that the biplane on 16 points satisfies the equations in Theorems 7.1 and
7.2.
2. Verify that any two blocks meet exactly twice on the biplane on 16 points. Prove
that the number of blocks through a point and the number of points on a block is
6.
3. Verify that finite affine and projective planes satisfy the equations in Theorems 7.1
and 7.2.
j
4. Produce the triangle of λi for the biplane of order 4.
j
5. Produce the triangle of λi for the projective plane of order n.
6. Let P be a set with m elements and let B consist of all subsets of P of size k.
Show that (P, B, I) is a design and determine its parameters.
7. Assume there exists a design with t = 5, v = 72, k = 16, and 249849 blocks.
j
Construct the triangle of λi for this design. What is λ5 = λ for this design?
8. Determine the number of points in the intersection of two complements of lines
in affine plane of order n.
9. Determine the parameters of the complementary design for affine and projective
planes of order n.
7.2 Biplanes
We shall now examine a symmetric design similar to a projective plane, except that
the number of times two lines meet is now two and there are two lines through any
two points. This object is called a biplane. We begin with the definition.
Definition 7.2 (Biplane) A biplane is a set of points P, a set of lines L, and an

incidence relation I ⊆ P × L such that
1. Through any two points there are two lines incident with them. More precisely,
given any two points p, q ∈ P, there exists two lines , m ∈ L with (p, ) ∈ I,
(q, ) ∈ I, (p, m) ∈ I and (q, m) ∈ I.
2. Any two lines intersect in two points. More precisely, given any two lines
, m ∈ L, there exists two points p, q such that p, q ∈ P with (p, ) ∈ I,(p, m) ∈
I,(q, ) ∈ I,(q, m) ∈ I.
We shall now determine the number of points and lines in a biplane. Assume there
are n + 2 points on a line. We shall call n the order of the plane. This is justified
by the definition of order of a symmetric design since λ = 2. Given a point q off a
line there are two lines connecting that point with each of the n + 2 points on .
(n+2)(2)
However, each line intersects the line exactly twice, and hence there are 2
lines through the point q. This gives the following.
7.2 Biplanes 189
Lemma 7.2 In a biplane, there are n + 2 points on a line and n + 2 lines through
a point.
We can use this to get the number of points in a biplane.
n2 +3n+4
Theorem 7.8 Let Π = (P, L, I) be a biplane of order n, then |P| = |L| = 2 .
Proof Take a point q. There are n + 2 lines through q each of which has n + 1
points on it distinct from q. Each of these points is counted twice since any two lines
(n+2)(n+1)
through q intersect in a unique point distinct from q. Hence, there are 2
(n+2)(n+1) n2 +3n+4
points distinct from q. Hence, there are 2 +1= 2 points in the
biplane. By duality, the number of lines is the same.
2
Notice that n +3n+4
2 is an integer (see Exercise 1) for all natural numbers n,
and therefore no order is eliminated by this parameter not being an integer.
2
A biplane is a 2-( n +3n+4
2 , n + 2, 2) design. We have already given one example
of a biplane in the previous section, namely, in Fig. 7.1. This is a biplane of order 4.
We can give an example of a biplane of order 1 in the next example.
Example 7.1 Let P = {A, B, C, D} be a set of four points. Let the blocks be
{A, B, C}, {A, B, D}, {A, C, D}, {B, C, D}. It is easy to see that through any two
points there are exactly two blocks and that any two blocks have exactly two points
in common. This is the biplane of order 1. We can represent it as follows (Fig. 7.2).
12 +3(1)+4
We note that there are 2 = 4 points and four blocks in this design as
expected.
Example 7.2 We can now give an example of a biplane of order 2. Consider the pro-
jective plane of order 2. Recall that given the following diagram the set of hyperovals
were exactly (Fig. 7.3)

190 7 Designs
{D, E, F, G}
{A, B, E, F}
{B, C, D, E}
{B, C, F, G}
{A, C, D, F}
{A, B, D, G}
{A, C, E, G}
Notice that any two hyperovals intersect exactly twice and that there are seven
of them. This gives the following. If Π is the projective plane of order 2 then the
design is formed from the points of the plane and the blocks given by the hyperovals
22 +3(2)+4
in a biplane of order 2. We note that there are 2 = 7 points and blocks as
expected.
We have just given examples of biplanes of orders 1, 2, and 4. At present, the only
orders for which it is known that biplanes exist are 1, 2, 3, 4, 8, 9, 11. The reader
should be warned that sometimes in the literature the order of the biplane is given
by n + 2 where n is the way order is defined here.
A nice description of the biplane of order 3 can be found in [14] “The fabulous
(11, 5, 2) Biplane”. An interesting connection to coding theory is also presented. We
shall construct this biplane later, see Theorem 10.15.
Exercises
2
1. Prove that n +3n+4
2 is always an integer.
2. Determine for which orders less than 100 the Bruck–Ryser–Chowla theorem rules
out the existence of a biplane.
7.3 Symmetric Designs
One of the most interesting collections of designs is the symmetric designs. We have
seen two examples of them already, namely, the projective planes (λ = 1) and the
biplanes (λ = 2). We begin with the definition of a symmetric design.
Definition 7.3 A symmetric design is a 2-(v, k, λ) design where the number of points
equals the number of blocks and the axioms are symmetric for points and blocks.
As we did for projective planes and for biplanes, it is a simple counting problem
to determine the number of blocks and points in a symmetric design.
7.3 Symmetric Designs 191
Theorem 7.9 The number of blocks and the number of points in a symmetric 2-
(n+λ−1)(n+λ)
(v, k, λ) design of order n is λ + 1.
Proof Let there be n + λ points on a block L. Through each point there are n + λ − 1
blocks other than L. Each of these blocks in the (n + λ − 1)(n + λ) blocks is counted
λ times.
It is immediate from this that (n + λ − 1)(n + λ) must be divisible by λ for a

symmetric design to exist.
Theorem 7.10 If a symmetric 2-(v, k, λ) design of order n exists, then n2 ≡ n

(mod λ).
(n+λ−1)(n+λ)
Proof If the design exists then λ is an integer. Therefore,
(n + λ − 1)(n + λ) must be divisible by λ. This gives
(n + λ − 1)(n + λ) ≡ 0 (mod λ)
(n − 1)n ≡ 0 (mod λ)
n2 ≡ n (mod λ),
Example 7.3 If λ = 1, namely, projective planes, then n2 ≡ n (mod 1) for all n

and no orders are eliminated. If λ = 2, namely, biplanes, then n2 ≡ n (mod 2) for
all n and no orders are eliminated. If λ = 3, then n2 ≡ n (mod 3) implies that n
must be 0 or 1 (mod 3) which eliminates n ≡ 2 (mod 3).
Corollary 7.2 If λ is a prime and a symmetric 2-(v, k, λ) design of order n exists,

then n ≡ 0 or 1 (mod λ).
Proof If λ is a prime, then Zλ is a field and so there are only two solutions to
n2 − n ≡ 0 (mod λ), namely, 0 and 1. Then Theorem 7.10 gives the result.
We shall show an example of an interesting symmetric design. Recall that

PGm (Fq ) has points as one-dimensional subspaces of Fm+1
q and hyperplanes as m-
dimensional subspaces of Fm+1
q . We have seen before that the number of points and
m+1
hyperplanes is that same and is q q−1−1 . There is a natural correlation between these
two sets since the orthogonal of a one-dimensional subspace is an m-dimensional
m −1
subspace in this setting. The number of points on a hyperplane is qq−1 since it is
simply the number of one-dimensional subspaces of a dimension m space.
192 7 Designs
To determine the parameter λ, consider two distinct points. We shall determine

the number of blocks through these points. We need to count the number of m-
dimensional subspaces of Fm+1
q containing two dimension one-dimensional sub-
spaces.
Then we have
(qm+1 − q2 )(qm+1 − q3 ) · · · (qm+1 − qm−1 )

λ=
(qm − q2 )(qm − q3 ) · · · (qm − qm−1 )
(qm−1 − 1)(qm−1 − q) · · · (qm−1 − qm−3 )
=
(qm−2 − 1)(qm−2 − q) · · · (qm−2 − qm−3 )
qm−1 − 1
= .
q−1
It is easy to determine n since n + λ = k. This means that
qm − 1 qm−1 − 1 qm − qm−1
n= − = = qm−1 . (7.12)
q−1 q−1 q−1
These results give the following.
Theorem 7.11 The designs of points and hyperplanes in PGm (Fq ) is a
qm+1 − 1 qm − 1 qm−1 − 1
2−( , , )
q−1 q−1 q−1
symmetric design of order qm−1 .
We shall now consider the complement of a symmetric design and show that it is
also a symmetric design. Consider a symmetric design with n = λ. It has
(2n − 1)(2n)
v= = 4n − 1,
n
k = 2n,
λ = n.
Its complementary design has
v = 4n − 1,
k = 4n − 1 − 2n = 2n − 1,
λ = b − 2r + λ = 4n − 1 − 2(2n) + n = n − 1.
This gives the following.

7.4 Kirkman Schoolgirl Problem and Steiner Triple Systems 193
Theorem 7.12 The complement of a symmetric n-design of order n is a symmetric

(n − 1)-design of order n.
We have already seen an example of this theorem as the complementary design

of the biplane of order 2 is the projective plane of order 2.
Exercises
1. Verify that the formulas in Theorem 7.9 are the same as were previously obtained
for projective planes and biplanes.
2. Find all symmetric designs of orders 1, 2, and 3, by first finding all possible λ
that satisfy the divisibility condition implicit in Theorem 7.9. Considering com-
plementary designs should reduce the computation by half.
3. Verify that with the parameters in Theorem 7.11, we have that
(n + λ − 1)(n + λ)
v= + 1.
λ
7.4 Kirkman Schoolgirl Problem and Steiner Triple Systems
In [56], Kirkman asked the following question in a recreational publication:

Can 15 schoolgirls walk three abreast each day for a week such that each pair of
girls walk abreast exactly once?
Kirkman found seven possible solutions and there are, in fact, exactly seven
possible solutions to the problem up to isomorphism.
Essentially, the question asks to construct a resolvable design with v = 15 for the
15 girls, k = 3 since each group of girls walking abreast is a block, r = 7 since each
day gives another block through each point, and b = 35 since there are 35 blocks.
The value of λ is 1. The fact that the design is resolvable is vital since the girls are
walking abreast.
As an example, if the girls are denoted by
α, β, γ, δ, , φ, ζ, η, θ, ι, κ, λ, μ, ν, o
then a solution to the problem would be the following:
(α β γ), (δ φ), (ζ η θ), (ι κ λ), (μ ν o)

(α δ ζ), (β ι), (γ φ μ), (η κ ν), (θ λ o)
(α ν), (β δ o), (γ η λ), (φ θ κ), (ζ ι μ)
(α θ μ), (β ζ λ), (γ δ κ), ( η o), (φ ι ν)
(α η ι), (β κ μ), (γ θ), (δ λ ν), (φ ζ o)
194 7 Designs
(α φ λ), (β θ ν), (γ ι o), (δ η μ), ( ζ κ)

(α κ o), (β φ η), (γ ζ ν), (δ θ ι), ( λ μ).
This basic question generalizes to the idea of a Steiner triple system. A design
where k = 3, that is, the block size is 3 is called a triple system. If λ = 1 and
any two points are contained in a block, then the system is called a Steiner triple
system. The parameters must satisfy the usual conditions, namely, bk = vr and
(k − 1)r = λ(v − 1). Substituting k = 3 and λ = 1 gives
3b = vr
2r = v − 1.
It follows that
v(v − 1)
b=
6
v−1
r= .
2
It follows that
v(v − 1) ≡ 0 (mod 6)
v − 1 ≡ 0 (mod 2).
Therefore, a necessary condition for the existence of a Steiner triple system is that v
must be either 1 or 3 (mod 6). Steiner first noticed this as a necessary condition and
asked if it is also sufficient in [82]. It was proven that this was indeed the case by
Reiss in [73]. Neither of these two men were aware of Kirkman’s work years earlier.
Of course, a solution to the Kirkman problem actually requires more, namely, that the
blocks are resolvable. That is, they break up into classes which partition the space.
For v = 13, there are two non-isomorphic designs and for v = 15 there are 80.
See [44] for details.
We shall now prove that these conditions are in fact sufficient as well as necessary.
There are several ways of proceeding, we shall follow the path used by Hall in [44].
Theorem 7.13 If there exists a Steiner triple system with v points and a Steiner triple
system with v points then there is a Steiner triple system with vv points.
Proof Let D = (P, B, I ) be a Steiner triple systems with v points and let D =
(P, B, I ) be a Steiner triple systems with v points. Let P = P × P . Define the set
of blocks as B = {(a1 , b1 ), (a2 , b2 ), (a3 , b3 ) | (a1 = a2 = a3 and (b1 , b2 , b3 ) ∈
B ) or (b1 = b2 = b3 and (a1 , a2 , a3 ) ∈ B) or (a1 , a2 , a3 ) ∈ B and (b1 , b2 , b3 )
∈ B }. Take any two points (a1 , b1 ) and (a2 , b2 ). If a1 = a2 there is a unique block
7.4 Kirkman Schoolgirl Problem and Steiner Triple Systems 195
in B that contains b1 and b2 , say the third point is b3 then the two points are in the
block
{(a1 , b1 ), (a1 , b2 ), (a1 , b3 )}.
If b1 = b2 there is a unique block in B that contains a1 and a2 , say the third point
is a3 then the two points are in the block
{(a1 , b1 ), (a2 , b1 ), (a3 , b1 )}.
If a1 = a2 and b1 = b2 then let a3 be the third point on the block containing a1

and a2 in B and b3 be the third point on the block containing b1 and b2 in B , then
the block
{(a1 , b1 ), (a2 , b2 ), (a3 , b3 )}.
Then these blocks make a Steiner triple system on vv points.
Notice in the previous theorem by fixing the first or second coordinate there are
subsystems with v1 and v2 points, respectively.
vv (vv −1)
We shall verify that the number of points is 6 as it should be. The number
of points of the first type is the number of blocks in B times the number of points
v(v−1)v
in P which is 6 . The number of points of the second type is the number of
v (v −1)v
blocks in B times the number of points in P which is 6 . Let (a, b, c) be a
block in B and (a , b , c ) be a block in B . Then the following are blocks in B :
{(a, a ), (b, b ), (c, c )} , {(a, a ), (b, c ), (c, b )},

{(a, b ), (b, a ), (c, c )} , {(a, b ), (b, c ), (c, a )},
{(a, c ), (b, a ), (c, b )} , {(a, c ), (b, b ), (c, a )}.
Thus, the number of points of the third type is six times |B||B |, which is
v(v−1)v (v −1)
6 . Then we add the three terms to get
v(v − 1)v v (v − 1)v v(v − 1)v (v − 1) vv (v − 1 + v − 1 + (v − 1)(v − 1)

+ + =
6 6 6 6
vv (vv − 1)
=
6
as desired.
Theorem 7.14 If there exists a Steiner triple system with v1 points and a Steiner
triple system with v2 points, where the system with v2 points contains a subsystem
with v3 points, then there is a Steiner triple system with v3 + v1 (v2 − v3 ) points.
196 7 Designs
Proof Let D1 = (P1 , B1 , I1 ) be a Steiner triple system with v1 points, D2 =

(P2 , B2 , I2 ) be a Steiner triple system with v2 points, and D3 = (P3 , B3 , I3 )
be a subsystem of D2 with v3 points.
Define the point set P = P3 ∪ (P1 × (P2 − P3 )). Note that |P | = v + 3 +
v1 (v2 − v3 ).
List the points of P3 by p1 , p2 , . . . , pv3 . The block set B consists of the following
types:
• Blocks in B3 ;
• Triples of the form {a, b, c}, where a ∈ P3 , b, c ∈ P2 − P3 , with {a, b, c} a
block in B2 ; and
• Triples of the form {(b, pi ), (c, pj ), (d, pk ) | i + j + k ≡ 0 (mod v2 − v3 ) and
(b, c, d) ∈ B1 }.
If two points are in P3 , then the block containing the two points in B3 is the block.
If the point a ∈ P3 and another point (b, pi ) ∈ P1 × (P2 − P3 ), then we take the
block {a, b, c} in B2 .
If two points (b, pi ) and (c, pj ) are in P1 × (P2 − P3 ) then we take a block of
the third kind. Specifically, there is a unique d with (b, c, d) ∈ B1 and a unique k
with i + j + k ≡ 0 (mod v2 − v3 ). Then this is a Steiner triple system of the desired
size.
It can be shown by using Theorems 7.13 and 7.14 that if v ≡ 1 or 3 (mod 6) then
there exists a Steiner triple system with v points.
This does not answer the question of whether there exist a resolvable Steiner triple
system, which is referred to as a Kirkman triple system. If the design is resolvable
then v must be three (mod 6) since three must divide the number of points. Hence,
the question is when does there exist a (v, 3, 1) resolvable design. The question
remained open until 1970 when the following was shown by Ray-Chaudhuri and
Wilson [71].
Theorem 7.15 A resolvable (v, 3, 1)-design exists if and only if v ≡ 3 (mod 6).
Exercises
1. There is a unique Steiner triple system with v = 3 and a unique Steiner triple
system with v = 7. Produce these two, the first is trivial and the second will be
familiar to you.
2. Prove that the affine plane of order 3 is a Kirkman triple system with nine points.
7.5 Nets and Transversal Designs 197
7.5 Nets and Transversal Designs
In this section, we shall give another example of a resolvable design. Finite nets were
introduced by Bruck in 1951, see [16]. Infinite nets had already been introduced under
the German name gewebe, in 1928, by Reidemeister, see [72]. In some sense, one can
think of a finite net as an incomplete affine plane. However, not all nets are extendable
to an affine plane. In fact, it is a fundamental question of the theory, asking which
nets have extensions to affine planes. They were introduced to help understand the
structure of planes and to help determine for which orders planes exist. We shall see
that nets are also canonically isomorphic to mutually orthogonal Latin squares as
well, and are essentially a geometric way of looking at combinatorial structure.
Definition 7.4 A k-net of order n is an incidence structure consisting of n2 points

and nk lines satisfying the following four axioms:
1. Every line has n points.

2. Parallelism is an equivalence relation on lines, where two lines are said to be
parallel if they are disjoint or identical.
3. There are k parallel classes each consisting of n lines.
4. Any two non-parallel lines meet exactly once.
It is immediate that a (n + 1)-net of order n is an affine plane of order n.

One of the most important tools in the study of nets is transversals. This is because
transversals can be used to extend nets. We now give the definition of a transversal.
Definition 7.5 A transversal of a net is a set of n points having exactly one point in
common with each line of the net.
Example 7.4 Consider the diagram of a 4-net of order 5. We mark a transversal with
X. Notice that the transversal hits each line exactly once (Fig. 7.4).
Fig. 7.4 Transversal on a

4-net of order 5
198 7 Designs
A k-net of order n is extendable if it is a subnet of a (k + 1)-net of order n. An

extendable net must have n mutually exclusive transversals.
Proposition 7.1 An (n + 1)-net of order n has no transversals.
Proof Assume t is a transversal and let p and q be points on the transversal t.

We know that there is a line through p and q since the net is an affine plane. This
contradicts that t is incident with this line only once.
The following theorem is similar to Theorem 4.10.
Theorem 7.16 A set of (k − 2)-MOLS of order n exists if and only if a k-net of

order n exists.
Proof Associate the n2 points of the plane with the n2 coordinates of the Latin
squares. We can describe the points as (a, b) where a, b ∈ {0, 1, . . . , n − 1} = Zn .
Assume we have a set of k − 2 MOLS of order n on the alphabet {0, 1, . . . , n − 1}.
As with the affine plane the first two parallel classes have the lines that are the
horizontal and vertical lines of the grid. Let Ls = (Lij ) be the s-th Latin square, then
the (s + 2)-nd parallel class has lines corresponding to the symbols of the square.
That is, the c-th line is incident with the point (i, j) if and only if Lij = c. The proof
of Theorem 4.10 shows that this forms a k-net.
If we assume we have a k-net of order n then we use the first two parallel classes
to describe the grid. That is, the point (i, j) is the point of intersection of the i-th line
of the first parallel class and the j-th line of the second parallel class. Then we reverse
the construction, Lk i,j = c if and only if the c-th line of the (k + 2)-nd parallel class
is incident with the line (i, j). As in the proof of Theorem 4.10, it is easy to see that
these form k − 2 MOLS of order n.
Example 7.5 As an example, consider the diagram in Fig. 7.5 of a 4-net of order 5
formed from the following Latin squares:
01234 01234
12340 40123
23401 34012
34012 23401
40123 12340
Any Latin square of order n is equivalent to a 3-net of order n. For example, the
Latin square
⎛ ⎞
123
⎝3 1 2⎠
231
Fig. 7.5 4-net of order 5
Fig. 7.6 3-net of order 3
is equivalent to the 3-net in Fig. 7.6.

In terms of design parameters, a k-net of order n is a 1 − (n2 , n, k) design.
The following is the standard definition of the direct product of nets.
Definition 7.6 Let Nk , Nk be k-nets of order n and n , respectively. The direct
product of nets Nk × Nk is defined as follows:
1. the points Nk × Nk are ordered pairs (q, q ) with q a point of Nk and q a point
of Nk ;
2. the lines of Nk × Nk are ordered pairs (m, m ) with m, m from the i-th parallel
class of Nk and Nk , respectively; and
3. the point (q, q ) is incident with the line (m, m ) if and only if q is incident with
m and q is incident with m .
Example 7.6 The direct product of the 3-net of order 2 formed

⎛ by the⎞Latin square
123
12 ⎝
and the 3-net of order 3 formed by the Latin square 3 1 2 ⎠ is the 3-net
21
231
formed by the Latin square:
200 7 Designs
⎛ ⎞
A B C D E F
⎜C A B F D E⎟
⎜ ⎟
⎜B CA E F D⎟
⎜ ⎟.
⎜D E F A B C⎟
⎜ ⎟
⎝F D E CA B⎠
E F D B C A
The next lemma follows immediately from Exercise 2.
Lemma 7.3 If there exists an h-net of order n and an h-net of order n then there
exists an h-net of order nn .
e
Theorem 7.17 Let n = pi i , pi prime, with pi > 2 or if pi = 2 then ei > 1.
Then there exists a 4-net of order n, and therefore a Graeco-Latin square of order
n.
Proof We know that for any prime p > 2 there exists a (pe − 1)-net of order pe
since there exists an affine plane of any prime power order. If e > 1, then 2e − 1 > 2
and there exists a (2e − 1)-net of order 2e . By repeatedly applying Lemma 7.3, we
get that there exists at least a 4-net of order n.
This theorem means that the only orders for which there may not be a 4-net are 2
(mod 4). We have already stated that there is not a 4-net of order 2 and 6 and have
exhibited a 2-net for n = 10. It is known that there exists a pair of MOLS for orders
n ≡ 2 (mod 4) for n > 6, and hence a 4-net, see [11,12]. Hence, the only orders
for which there is no Graeco-Latin square are 2 and 6.
We shall now describe a design which is another way of examining MOLS which
is dual to a net.
Definition 7.7 A transversal design D of order n, block size k, is a set of points V

and blocks B such that
• V has kn points.
• V is partitioned into k classes, each of size n.
• Every pair of points is either contained in the same set of the partition or in a
block.
There are more general definitions but this is the one we shall use.
Theorem 7.18 The dual of a transversal design on kn points with block size k is a
k-net of order n.
Proof We notice that in the transversal design every pair of points is in the same
set or in a block and so every pair of blocks in the dual either intersects once or is
parallel. The number of blocks in the dual is kn and they are split into k parallel
classes by the second axiom of transversal designs.
Exercises
1. Prove that any 2-net of order n has n! transversals.

2. Prove that Nk × Nk is a net of order nn . Be sure to count the number of points,
lines, and parallel classes are all correct.
3. Determine the parameters of a transversal design that would be a solution to the
36 officer problem.
4. Take the direct product of a 4-net (affine plane) of order 3 with itself to produce
a Graeco-Latin square of order 9.
5. Determine the parameters of a transversal design that would be an affine plane.
6. Draw the net corresponding to 2 MOLS of order 10.
7. Prove that the 3-net corresponding to the Latin square:
⎛ ⎞
012
⎝1 2 0⎠
201
is isomorphic to the 3-net corresponding to the Latin square:

⎛ ⎞
012
⎝2 0 1⎠.
120
8. Find all possible k-nets of order n, up to isomorphism, for k ≤ n + 1 ≤ 4.

9. Prove that the 3-net formed from the following Latin square has no transversals:
⎛ ⎞
1234
⎜4 1 2 3⎟
⎜ ⎟
⎝3 4 1 2⎠.
2341
10. Prove that the number of blocks through a point in a transversal design is n and
use this to show there are n2 blocks.
Combinatorial Objects
8
8.1 Introduction to Hadamard Matrices
Hadamard matrices were studied by Hadamard [42] in 1893 as solutions to an

extremal problem in real analysis and they have numerous applications throughout
mathematics. Sylvester [89] had already studied them.
Definition 8.1 A square n by n matrix H with Hij ∈ {1, −1} is said to be a

Hadamard matrix if HHT = nIn , where In is the n by n identity matrix.
This implies that if v1 , v2 , . . . , vn are the rows of H then

0 if i = j
vi · vj = ,
n if i = j
where the operations are done in the integers.

Since HHT = nIn we have that HT H = nIn as well. This gives that any two
distinct columns have dot product 0 as well.
Theorem 8.1 If H is a Hadamard matrix of size n then HT is a Hadamard matrix

of size n.
Proof We have seen that if the dot product of any distinct rows is 0 and any row
with itself is n, then the same is true for the columns. This gives the result.
If n = 1 there are two matrices, namely, (1) and (−1). Notice that since there is
only 1 row these matrices satisfy the definition trivially. For n = 2, there are eight
different Hadamard matrices. They are
204 8 Combinatorial Objects

x x x x x −x −x x
, (8.1)
x −x −x x x x x x
where x = 1 and x = −1.
Theorem 8.2 Let H be an n by n Hadamard matrix with n > 2, then n = 4k for

some k.
Proof Take two rows in a n by n Hadamard matrix, let χi,j denote the number of
times the pair (i, j) occurs when overlapping the two rows. Then we have
χ1,1 + χ−1,−1 + χ1,−1 + χ−1,1 = n

2(χ1,1 + χ−1,−1 ) = n
since χ1,1 + χ−1,−1 = χ1,−1 + χ−1,1 because the dot product of the two rows
must be 0. This gives that n is an even integer.
This gives that n = 2m for some integer m and that any two rows of the matrix
must agree in exactly m places. Now assume that n is at least 4 and let r1 , r2 , r3
be three rows of the Hadamard matrix. If we assume that r1 and r2 agree in the
coordinates of the set A and disagree in the coordinate set D. Then, let B be the
subset of A where r1 agrees with r3 as well and E be the subset of D where r3
disagrees with r2 . Notice that |B| + |E| = m since this counts the number of places
where r3 agrees with r1 . Then, we have
|B| + |D − E| = m
|B| + |D| − |E| = m
|B| + m − |E| = m
|B| = |E|.
This gives that
|B| + |E| = m
|B| + |B| = m
2|B| = m
and m is even which makes n divisible by 4.
Given this theorem we see why usually one assumes that n is at least 4 when
discussing Hadamard matrices. From this point on when referring to a Hadamard
matrix of size 4n we shall say that n is the order of the matrix. It will become clear
why we do this after we show how to construct a design from a Hadamard matrix.
8.1 Introduction to Hadamard Matrices 205
Let H = (hij ) and M be Hadamard matrices of size n and k, respectively. Define

the Kronecker product of H and M to be the square matrix of size nk given by
⎛ ⎞
h11 M h12 M . . . h1n M
⎜ h21 M h22 M . . . h2n M ⎟
⎜ ⎟
H⊗M=⎜ .. ⎟. (8.2)
⎝ . ⎠
hn1 M hn2 M . . . hnn M
Theorem 8.3 If H and M are Hadamard matrices of size n and k, respectively, then
H ⊗ M is a Hadamard matrix of size nk.
Proof Let xi be the i-th row of M. We have
(ha1 xi , ha2 xi , . . . , han xi ) · (hb1 xj , hb2 xj , . . . , hbn xj )

n

= (xi · xj ) haα hbα .
α=1
If i = j and a = b then this value is nk. If i = j then xi · xj = 0 and so the value

is 0. If i = j and a = b then the second sum is 0. Hence, the dot product of a row
with itself is the size of the matrix and the dot product of any two distinct rows is 0.
Therefore, the Kronecker product is a Hadamard matrix.
As an example, let
⎛ ⎞
1 1 1 1
1 1 ⎜ −1 −1 1 1 ⎟
H= , M=⎜ ⎟
⎝ 1 −1 −1 1 ⎠ . (8.3)
1 −1
−1 1 −1 1
Then
⎛ ⎞
1 1 1 1 1 1 1 1
⎜ −1 −1 1 1 −1 −1 1 1 ⎟
⎜ ⎟
⎜ 1 −1 −1 1 1 −1 −1 1 ⎟
⎜ ⎟
M M ⎜ −1 1 −1 1 −1 1 −1 1 ⎟
H⊗M= ⎜
=⎜ ⎟. (8.4)
M −M ⎟
⎜ 1 1 1 1 −1 −1 −1 −1 ⎟
⎜ −1 −1 1 1 1 1 −1 −1 ⎟
⎜ ⎟
⎝ 1 −1 −1 1 −1 1 1 −1 ⎠
−1 1 −1 1 −1 1 −1 1
Corollary 8.1 There exists Hadamard matrices of arbitrarily large order.

Proof Let H be a Hadamard matrix of size 2. Then taking the repeated Kronecker
product with itself t times gives a Hadamard matrix of size 2t . This gives that there
exists a Hadamard matrix of all orders that are a power of 2. Hence, there exists
Hadamard matrices of arbitrarily large size.
This does not prove that there are Hadamard matrices of all possible orders. In
fact, it is an open question as to whether there exists Hadamard matrices of all
possible orders. The smallest unknown case is 4(167) = 668. See [55] for details
and a construction of the previous largest unknown case, which was 428. However,
we do have the following conjecture.
Conjecture 8.1 There exists Hadamard matrices for all sizes of the form 4n.
We shall describe another construction of Hadamard matrices. It is known as the

Payley construction. To give the construction we shall require some more information
about finite fields. If F is a field, then a is a square in F if there exists b ∈ F with
b2 = a. We usually consider 0 to be a special case so we shall only consider non-zero
elements in the next examples. Consider the field F3 then 1 = 12 = 22 and so 1 is
a square and 2 is a non-square. For F5 , 1 = 12 = 42 and 4 = 22 = 32 and so 1 and
4 are squares and 2 and 3 are non-squares.
Lemma 8.1 In a finite field of odd order, exactly half of the non-zero elements are
squares.
Proof For each x we have x2 = (−x)2 . Since, the order is odd, we have x = −x.
Therefore, the cardinality of the set of squares is at most half the cardinality of
the non-zero elements. Given a square, say a = b2 , the equation x2 − a = 0 fac-
tors into (x − b)(x + b) = 0. Since a finite field has no zero divisors, this equation
has at most two solutions. Therefore, precisely half of the non-zero elements are
squares.
Example 8.1 Consider the field of order 7. Here 12 = 62 = 1, 22 = 52 = 4, 32 =

42 = 2. Therefore, the squares are 1, 2, and 4 and the non-squares are 3, 5, and 6.
Example 8.2 For fields of even order Lemma 8.1 is not true. Consider the field
of order 4, {0, 1, ω, ω2 }. Then 12 = 1, ω2 = ω2 , (ω2 )2 = ω. Therefore, every
element is a square.
Define the following function χ : Fq → {0, 1, −1} :

⎧
⎨ 0 x = 0,
χ(x) = 1 x is a square, (8.5)
⎩
−1 x is not a square.
8.1 Introduction to Hadamard Matrices 207
Lemma 8.2 If a, b ∈ Fq then χ(ab) = χ(a)χ(b).
Proof If a or b is 0 then the result is trivial so we assume that neither a nor b is 0.

First we note that if a and b are both squares, then their product is a square, that is,
if a = c2 and b = d2 then ab = (cd)2 . Hence, multiplication on the set of squares
is closed. Let S be the set of squares then, if e is a non-square, the set {fe | fe ∈ S}
is the set of all non-squares since the number of non-squares is equal to the number
of squares and the map is injective. Then, if neither a nor b are squares then a = ef
and b = eg where f = k2 and g = h2 . Then ab = efeg = (ehk)2 .
Now consider the case if one of a and b is square and the other is not. Without
loss of generality assume a = c2 and b is not a square. If their product were m2
2
then c2 b = m2 and b = m c which is a contradiction.
The previous proof could be simplified a great deal if we had the theory of groups
which we discuss in the next chapter.
We shall let I denote the identity matrix, that is, the matrix that is 0 off the diagonal
and 1 along the diagonal and we shall let J denote the matrix where every element is
1. Let M be a matrix indexed by the elements of Fq with
Ma,b = χ(b − a). (8.6)
Notice that M has 0 on the diagonal.

We form the matrix
⎛ ⎞
1 1 1 ... 1
⎜1 ⎟
⎜ ⎟
⎜. ⎟
H=⎜ ⎜.
⎟.
⎟ (8.7)
⎜ M−I ⎟
⎝. ⎠
1
Theorem 8.4 If q ≡ 3 (mod 4) then the matrix H defined in Eq. 8.7 is a Hadamard
matrix of size q + 1.
Proof First we note that every element of H is either 1 or −1. We shall prove that
HHT = (q + 1)I.
Using Exercise 6 we see that
⎛ ⎞
1 1 1 ... 1
⎜1 ⎟
⎜ ⎟
⎜. ⎟
H =⎜
T
⎜.
⎟.
⎟
⎜ −M − I ⎟
⎝. ⎠
1
It is straightforward to see that

⎛ ⎞
q+1 0 0 ... 0
⎜ 0 ⎟
⎜ ⎟
⎜ . ⎟
HHT = ⎜
⎜ .
⎟.
⎟
⎜ J + (M − I)(−M − I) ⎟
⎝ . ⎠
0
Now we shall consider the value of (M − I)(−M − I) = −M2 + M − M + I =

−M2 + I. All that is needed now is to show that −M2 = MMT = qI − J. If the
a-th row in M is (m0 , m1 , . . . , mq−1 ), then ma = 0 and the a-th column of MT
is the a-th row of M and so we have MMTa,a = q − 1. This is as we expected, i.e.,
q − 1 down the diagonal. Now consider the a-th row of M and the b-th column of
MT which is the b-th row of M. The inner product of these two rows is

(MMT )a,b = χ(c − a)χ(c − b)
c∈Fq

= χ(d)χ(d + e) with d = c − a, e = a − b
d∈Fq

= χ(d)χ(df) with df = d + e notice f = 1, d = 0 drops out
d∈Fq

= χ(d)χ(d)χ(f)
d∈Fq

= χ(f) = −1.
f=1
As an example, consider the finite field of order 3. Then indexing by 0, 1, 2 we

have
⎛ ⎞
0 1 −1
M = ⎝ −1 0 1 ⎠ .
1 −1 0
Then the Hadamard matrix is
⎛ ⎞
1 1 1 1
⎜ 1 −1 1 −1 ⎟
H=⎜ ⎟
⎝ 1 −1 −1 1 ⎠ .
1 1 −1 −1
8.2 Hadamard Designs 209
Exercises
1. Show that multiplying a column or a row by −1 in a Hadamard matrix produces a

Hadamard matrix. Use this to show that each Hadamard matrix can be transformed
into a Hadamard matrix where the first row and column are all one.
2. Give a complete proof of Theorem 8.1.
3. Show that there are no Hadamard matrices with n = 3.
4. Prove that −1 is a square in the finite field Zq , q a prime, if and only if q ≡ 1
(mod 4).
5. Construct a 4 by 4 Hadamard matrix, i.e., a Hadamard matrix of order 1.
6. Use Exercise 4 and Lemma 8.2 to prove that if q ≡ 3 (mod 4) then Ma,b =
(−1)Mb,a .
7. Find all the squares and non-squares in the field of order 13.
8. Construct the Hadamard matrix from the Paley construction for the finite field of
order 7.
9. Determine where the proof of Theorem 8.4 fails when q ≡ 1 (mod 4). Hint:
consider MMT in this case.
8.2 Hadamard Designs
We shall show how Hadamard matrices relate to designs.

Let {p1 , p2 , . . . , p4n } = P be a set of points corresponding to the coordinates
of a Hadamard matrix Hi of size 4n. Denote by Hi the i-th row of H and let
Hi = {x1 , x2 , . . . , x4n }. For any j different from i with Hj = {y1 , y2 , . . . , y4n } let
Bj = {pα | yα = xα } (8.8)
and
Bj = P − Bj . (8.9)
Let the block set be
B = {Bj | j = i} ∪ {Bj | j = i}. (8.10)
The incidence between a point and a block is given by set-theoretic containment.
Lemma 8.3 Given a Hadamard matrix of size 4n the incidence structure defined
above is a 3 − (4n, 2n, n − 1) design.
Proof The fact that v = 4n is obvious. Given any two rows Hi and Hj , we know
that the number of coordinates in which they agree is equal to the number in which
they disagree since if they agree in q1 places and disagree in q2 places then the dot
product Hi · Hj = q1 − q2 . However, we know that the dot product must be 0 and
hence q1 = q2 = 2n since q1 + q2 = 4n. This gives that k = 2n.
Take any three points from P, that is, any three coordinates of the matrix H. It
was shown in the proof of Theorem 8.2 that any three rows of a Hadamard matrix of
size 4n coincide in exactly n places. Now we can assume that the row Hi is all ones
by multiplying coordinates by −1 if necessary. Hence, the number of blocks through
any three points is n − 1, corresponding to the n − 1 coordinates (not including the
coordinates in Hi ) where the three columns coincide. Hence, we have λ = n − 1
and t = 3.
It follows from the construction that there are 8n − 2 = 2(4n − 1) blocks

j
and hence we can compute the following triangle of λi for the Hadamard 3 −
(4n, 2n, n − 1) design:
8n − 2
4n − 1 4n − 1
(8.11)
2n − 1 2n 2n − 1
n−1 n n n−1
Theorem 8.5 Let D = (P, B) be a 3 − (4n, 2n, n − 1) design. Then if D is formed

by removing a point from P and the blocks not incident with it, we have that D is a
symmetric 2 − (4n − 1, 2n − 1, n − 1) design.
Proof The facts that v = 4n − 1 and k = 2n − 1 are obvious since we removed

the point p. Let q1 and q2 be two points. Through q1 , q2 , and p there were n − 1
blocks in D, and hence these n − 1 blocks are incident with q1 and q2 in D . If a
block in D was incident with q1 and q2 but not incident with p, then its restriction
would not be a block in D . Hence D is a 2 − (4n − 1, 2n − 1, n − 1) design. It
is a simple computation then to compute that the number of blocks is equal to the
number of points which is 4n − 1.
j
Its triangle of λi is given by
4n − 1
2n 2n − 1 (8.12)
n n n−1
Lemma 8.4 If there exists a 3 − (4n, 2n, n − 1) design, then there exists a Hadamard
matrix of size 4n.
Proof First we need to show that any two blocks are either disjoint or meet in n
places. Assume that two blocks do meet and take any point p in their intersection.
Using Theorem 8.5 we see that D formed from the 3 − (4n, 2n, n − 1) by removing
the point p is a symmetric 2 − (4n − 1, 2n − 1, n − 1) design. Hence, any two
blocks in D meet in n − 1 places. However, since they are blocks in D they also
intersect at p and hence meet at n places.
8.2 Hadamard Designs 211
Next we need to show that the complement of a block is indeed a block. Let B be
a block. Through each point on the block there are 4n − 2 blocks different from B
since there are 4n − 1 blocks through each point. There are 2n points on the block.
(4n−2)(2n)
Hence, the number of blocks that intersects B is n = 8n − 4. Each block
is counted n times, since it intersects B in n places. However, there are 8n − 2
blocks, which gives that there is B and 8n − 4 blocks that intersect B making 8n − 3
blocks which implies there must be a block that does not intersect B.
Use the 4n points of the design to make n coordinates for a matrix H. Then for
each pair of disjoint blocks B and B define a row v with vα = 1 if pα is incident
with B and vα = −1 if pα is incident with B. Then add a row of ones as the first
row of the matrix. This gives that every row (except the first) has n coordinates with
a 1 and n coordinates with a −1. If v is a row corresponding to the pair B, B and w
is a row corresponding to the pair B , B then B shares n coordinates with B and n
coordinates with B and the same is true for B. Hence, the dot product of v with w
is 0. Hence, the matrix H is a Hadamard matrix.
Theorem 8.6 A Hadamard matrix of size 4n exists if and only if a 3 − (4n, 2n, n −
1) design exists.
Proof Follows from Lemmas 8.3 and 8.4.
This design and the corresponding 2-design given in Theorem 8.5 are called
Hadamard designs.
As an example consider the following Hadamard matrix:
⎛ ⎞
1 1 1 1
⎜ 1 1 −1 −1 ⎟
⎜ ⎟
⎝ 1 −1 1 −1 ⎠ .
1 −1 −1 1
Denote the four points by p1 , p2 , p3 , p4 corresponding to the four coordinates and

let Hi in the construction be H1 . Then the blocks of the 3-design are
{p1 , p2 }, {p3 , p4 }, {p1 , p3 }, {p2 , p4 }, {p1 , p4 }, {p2 , p3 }.
Using p1 we construct a 2-design with blocks
{p2 }, {p3 }, {p4 }.
Then the incidence matrix of the 2-design is given by

⎛ ⎞
100
⎝0 1 0⎠.
001
Notice that if every 0 is replaced by −1 and a first row and column of all ones are
added then the incidence matrix produces the Hadamard matrix.
Exercises
1. Construct the 3-design and 2-design formed from the Hadamard matrix given by
the product of H and M in (8.3).
8.3 Generalized Hadamard Matrices
In this section, we shall give a generalization of Hadamard matrices. We begin with

the definition.
Definition 8.2 A generalized Hadamard matrix H(q, λ) = (hij ) of order n = qλ

over Fq is a qλ × qλ matrix with entries from Fq such that for every i, j, 1 ≤ i <
j ≤ qλ, each of the multisets {his − hjs : 1 ≤ s ≤ qλ} contains every element of
Fq exactly λ times.
Example 8.3 The following is an example of an H(3, 2) generalized Hadamard

matrix. Note that each element of F3 appears twice in the difference of any two
rows.
⎛ ⎞
000000
⎜0 0 1 1 2 2⎟
⎜ ⎟
⎜0 1 2 0 1 2⎟
H(3, 2) = ⎜ ⎟
⎜0 1 0 2 2 1⎟. (8.13)
⎜ ⎟
⎝0 2 1 2 1 0⎠
022101
Two generalized Hadamard matrices H1 and H2 of order n are said to be equiva-

lent if one can be obtained from the other by a permutation of the rows and columns
and a series of adding the same element of Fq to all the coordinates in a row or in
a column. It is always possible to change the first row and column of a generalized
Hadamard matrix into zeros. This generalized Hadamard matrix is called normalized.
Theorem 8.7 The multiplication table of Fq is an H(q, 1) generalized Hadamard

matrix.
Proof The α and β rows (α = β) of the multiplication table are of the form
αa1 , αa2 , . . . , αaq
and
βa1 , βa2 , . . . , βaq
8.3 Generalized Hadamard Matrices 213
where a1 , a2 , . . . , aq are the elements of Fq . Then the difference is the mul-

tiset {αai − βai | 1 ≤ i ≤ n}. If αai − βai = αaj − βaj , then (α − β)ai =
(α − β)aj . Since α = β we have ai = aj and i = j. Therefore, the n elements
in the multiset are distinct giving that the matrix is a generalized Hadamard matrix.
Example 8.4 Consider the following multiplication table of F5 :

⎛ ⎞
00000
⎜0 1 2 3 4⎟
⎜ ⎟
⎜0 2 4 1 3⎟. (8.14)
⎜ ⎟
⎝0 3 1 4 2⎠
04321
Here each element of F5 appears once in the difference between any two rows.
Therefore, λ = 1. Hence, this is a H(5, 1) generalized Hadamard matrix.
Theorem 8.8 If H(q, λ) is a generalized Hadamard matrix then H(q, λ)T is a gen-
eralized Hadamard matrix, where H(q, λ)T denotes the transpose of H(q, λ).
Proof It follows from the fact that since (Fq , +) is an abelian group then H(q, λ)T
is also a generalized Hadamard matrix.
Theorem 8.9 An ordinary Hadamard matrix of order 4λ exists if and only if a

generalized Hadamard matrix H(2, 2λ) over F2 exists.
Proof Let H be an ordinary Hadamard matrix of order 4λ. Any two rows of this
matrix consists of 1s and −1s such that any two rows have equally many occurrences
of (1, 1), (−1, −1) and (1, −1), (−1, 1). Form H by sending each 1 to 0 and each −1
to 1. Now in any two rows of H there are equally many occurrences of (0, 0), (1, 1)
and (0, 1), (1, 0). This gives that in the multiset formed from the differences of any
two rows there are an equal number (2λ in particular) of 0 and 1. Hence, H is a
generalized Hadamard matrix H(2, 2λ) over F2 .
The other direction is identical reversing the map.
We shall now give a standard method to construct generalized Hadamard matri-

ces from existing generalized Hadamard matrices. This technique is known as
the Kronecker sum construction. Let H(q, λ) = (hij ) be any qλ × qλ generalized
Hadamard matrix over Fq , and let B1 , B2 , . . . , Bqλ be any qμ × qμ generalized
Hadamard matrices over Fq .
The matrix in Eq. 8.15 gives a q2 λμ × q2 λμ generalized Hadamard matrix over
Fq , denoted by H ⊕ [B1 , B2 , . . . , Bn ], where n = qλ. If B1 = B2 = · · · = Bn =
B, then we write H ⊕ [B1 , B2 , . . . , Bn ] = H ⊕ B.
⎛ ⎞
h11 + B1 h12 + B1 · · · h1n + B1
⎜ h21 + B2 h22 + B2 · · · h2n + B2 ⎟
⎜ ⎟
H ⊕ [B1 , B2 , . . . , Bn ] = ⎜ .. .. .. .. ⎟. (8.15)
⎝ . . . . ⎠
hn1 + Bn hn2 + Bn · · · hnn + Bn
We leave the proof that the Kronecker sum construction gives a generalized
Hadamard matrix as Exercise 2
Let Sq be the normalized generalized Hadamard matrix H(q, 1) given by the
multiplicative table of Fq . As for ordinary Hadamard matrices over F2 , start-
ing from a generalized Hadamard matrix S1 = Sq , we can recursively define
ST as a generalized Hadamard matrix H(q, qt−1 ), constructed as ST = Sq ⊕
[St−1 , St−1 , . . . , St−1 ] = Sq ⊕ St−1 for t > 1, which is called a Sylvester gener-
alized Hadamard matrix.
Exercises
1. Verify that the matrix in Eq. 8.13 is a generalized Hadamard matrix.
2. Prove that the matrix given in Eq. 8.15 is a generalized Hadamard matrix.
8.4 Latin Hypercubes
In this section, we shall describe a generalization of the Latin squares called Latin
hypercubes. Essentially, these objects are multidimensional Latin squares. In other
words, the same condition on rows and columns that makes a matrix a Latin square is
applied to a k-dimensional matrix to make a Latin hypercube. The word hypercube
simply means a cube in (possibly) more than three dimensions.
To be specific we have the following definition. Note that we use the alphabet
{0, 1, . . . , n − 1} right from the beginning of the discussion for simplicity.
Definition 8.3 A Latin k-hypercube Lj1 ,...,jk of order n is a k-dimensional array

of size n, whose elements come from the set {0, 1, 2, . . . , n − 1}. If ji = ji for all
i = α and jα = jα then Lj1 ,j2 ,...,jk = Lj ,j ,...,j . That is, in any direction, each
1 2 k
“row” has each element from the set exactly once.
If L is a hypercube of order n then eliminating any set of dimensions will result in

a hypercube. Specifically, we have that the subhypercube of a Latin k-hypercube of
order n formed by fixing k − s coordinates is a Latin s-hypercube of order n. This
is because each line in the hypercube still contains each element exactly once.
Notice that a 1-hypercube, which we shall call a Latin line is simply a vector with
each of the elements from 0 to n − 1 listed once and 2-hypercube is a Latin square.
We can count the possible number of subhypercubes in the following theorem.
8.4 Latin Hypercubes 215

k
Theorem 8.10 For s ≤ k, a Latin k-hypercube of order n contains nk−s =
k−s

k
nk−s s-hypercubes of order n.
s

k
Proof There are exactly different ways of picking k − s coordinates out
k−s
of the k to fix. For each coordinate, there are n choices of which element to fix in
that coordinate.
We shall make a definition of orthogonality that mimics the definition of orthog-

onality for Latin squares.
Definition 8.4 Two Latin k-hypercubes are said to be orthogonal if each correspond-
ing pair of Latin subsquares is orthogonal. A set of s Latin k-hypercubes of order n
is said to be mutually orthogonal if each pair is orthogonal. In this case, we say we
have a set of s mutually orthogonal Latin k-hypercubes (MOLkC).
Theorem 8.11 There do not exist n mutually orthogonal Latin k-hypercubes of

order n.
Proof If there were n MOLkCs of order n then fixing any k − 2 dimension in every
hypercube we would get n Latin squares. These corresponding Latin squares would
be n MOLS of order n, which we have proven do not exist.
We shall now show how to construct a multidimensional magic hypercube in a

manner that is similar to the construction of a magic square.
Let L1 , . . . , Lk be a set of k-MOLkCs of order n. Define a k-dimensional hyper-
cube of order n as follows:
Mag(L1 , . . . , Lk )i1 ,i2 ,...,ik = (L1 )i1 ,i2 ,...,ik + (L2 )i1 ,i2 ,...,ik n
+ · · · + (Lk )i1 ,i2 ,...,ik nk−1
k

= (Lj )i1 ,i2 ,...,ik nj−1 .
j=1
We shall show that the sum along any line in this hypercube is a constant and that
elements in the hypercube each appear exactly once.
Theorem 8.12 Let L1 , L2 , . . . , Lk be a set of k- MOLkCs, then Mag(L1 , L2 , . . . ,

Lk ) is a k-hypercube of order n such that the numbers 0, 1, 2, . . . , nk − 1 can be
placed in the hypercube so that in each i-line the digits in any place of each k-digit
number written in base n are a permutation of 0, 1, . . . , n − 1.
Proof We leave the proof as an exercise, see Exercise 1
Theorem 8.13 Let Mag(L1 , L2 , . . . , Lk ) be a k-hypercube of order n formed from

n(nk −1)
MOLkCs L1 , . . . , Lk then the sum of every i-line is ( 2 ).
Proof Every number in {0, 1, 2, . . . , n − 1} occurs once in each digit place. When
summing the numbers in each digit’s place in an i-line we have (0 + 1 + 2 + · · · +
(n−1)n nk −1 n(nk −1)
n − 1)(1 + n + n2 + · · · + nk−1 ) = ( 2 )( n−1 ) =( 2 ).
Thus, in this hypercube, the sum in each i-line gives the same sum, namely,
n(nk −1)
( 2 ).
The [6, 3, 4] MDS code over F5 gives the following magic hypercube of order 5,
formed by stacking the following squares:
Using 3 MOLkCs of order 5 we can form the following magic hypercube. Simply
place the squares on top of each other.
⎛ ⎞ ⎛ ⎞
0 83 36 119 72 43 121 54 7 85
⎜ 64 17 95 28 106 ⎟ ⎜ 77 30 113 66 24 ⎟
⎜ ⎟ ⎜ ⎟
⎜ 123 51 9 87 40 ⎟ , ⎜ 11 94 47 100 58 ⎟
⎜ ⎟ ⎜ ⎟
⎝ 32 110 68 21 79 ⎠ ⎝ 70 3 81 39 117 ⎠
91 49 102 55 13 109 62 15 98 26
⎛ ⎞ ⎛ ⎞
56 14 92 45 103 99 27 105 63 16
⎜ 115 73 1 84 37 ⎟ ⎜ 8 86 44 122 50 ⎟
⎜ ⎟ ⎜ ⎟
⎜ 29 107 60 18 96 ⎟ , ⎜ 67 20 78 31 114 ⎟
⎜ ⎟ ⎜ ⎟
⎝ 88 41 124 52 5 ⎠ ⎝ 101 59 12 90 48 ⎠
22 75 33 111 69 35 118 71 4 82
⎛ ⎞
112 65 23 76 34
⎜ 46 104 57 10 93 ⎟
⎜ ⎟
⎜ 80 38 116 74 2 ⎟ .
⎜ ⎟
⎝ 19 97 25 108 61 ⎠
53 6 89 42 120
Exercises
1. Produce a proof of Theorem 8.12 modeling the proof done for Latin squares.
2. Produce a magic hypercube that is 3 by 3 by 3.
8.5 Partially Ordered Sets
Next, we shall study the combinatorial object partially ordered sets which has numer-
ous applications in mathematics and computer science. Recall, that a relation R on
8.5 Partially Ordered Sets 217
Fig. 8.1 Partially ordered set

with four elements
a set A is a subset of A × A. A relation R ⊆ A × A is reflexive if for all a ∈ A,

(a, a) ∈ R, it is antisymmetric if (a, b) ∈ R and (b, a) ∈ R implies a = b, and it is
transitive if (a, b) ∈ R and (b, c) ∈ R implies (a, c) ∈ R.
Definition 8.5 A partially ordered set is a set A together with a relation R on A such
that R is reflexive, antisymmetric, and transitive.
For example, (Z, ≤) is a partially ordered set since ≤ is reflexive, antisymmetric,

and transitive. In general, since R is reflexive, antisymmetric, and transitive, we often
denote the relation by ≤ when no confusion will arise. If there is a standard notation
for the relation we shall use the standard notation.
Example 8.5 The relation < on Z does not give a partially ordered set since < is
not reflexive. The relation ≥ on Z does give a partially ordered set since the relation
is reflexive, antisymmetric, and transitive.
As with any relation we can construct the directed graph corresponding to the relation.
Specifically, if (a, b) ∈ R then (a, b) is a directed edge in the corresponding graph.
We can simplify the graph since it is reflexive and transitive. Namely, since it is
reflexive, we can eliminate all loops on an element, since each element has such a
loop. Moreover, since it is transitive, we can take a minimal set of edges which gives
all edges by completing the transitivity. Put simply, if vertex a has a path going up to
vertex b then we have that a ≤ b. This graph is known as the Hasse diagram of the
partially ordered set. We begin with a simple example.
Example 8.6 Consider the partially ordered set (P({1, 2}), ⊆) shown in Fig. 8.1.
The set has four elements, namely, ∅, {1}, {2}, and {1, 2}.
We can generalize this result in the following theorem.
Theorem 8.14 Let A be a set, then (P(A), ⊆) is a partially ordered set.

Fig. 8.2 The poset

P({1, 2, 3})
Proof It is immediate that B ⊆ B for all sets B. Then if B ⊆ C and C ⊆ B then

C = B. Finally, if B ⊆ C ⊆ D, then every x ∈ B must be in C and then must be
in D giving that B ⊆ D. Therefore, the relation is reflexive, antisymmetric, and
transitive.
Example 8.7 Consider the partially ordered set P({1, 2, 3}) in Fig. 8.2. The set has
eight elements, namely, ∅, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}. Its Hasse dia-
gram is given as follows.
Definition 8.6 An element a in a partially ordered set (A, ≤) is maximal if there

exists no b with a < b. An element c in a partially ordered set (A, ≤) is minimal if
there exists no d with d < c.
Example 8.8 Consider the partially ordered set given in Example 8.6, then ∅ is a
minimal element and {1, 2} is a maximal element. Consider the partially ordered set
given in Example 8.7, then ∅ is a minimal set and {1, 2, 3} is a maximal element.
Example 8.9 Consider the partially ordered set in Fig. 8.3.

Here d, e, and f are all maximal elements and a and b are both minimal elements.
Notice that these elements do not have the property that they are greater than or less
than all elements in the partially ordered set. They simply have nothing greater than
or less than themselves.
For any relation R on a set A, that is, R ⊆ A × A, we can define the inverse
relation to be R−1 = {(b, a) | (a, b) ∈ R}.
Fig. 8.3 Poset on six points
Theorem 8.15 Let R be a relation on a set A. If (A, R) is a partially ordered set

then (A, R−1 ) is a partially ordered set.
Proof If a is any element in A, then (a, a) ∈ R which implies (a, a) ∈ R−1 . There-
fore, R−1 is reflexive.
If (a, b) ∈ R−1 and (b, a) ∈ R−1 then (b, a) ∈ R and (a, b) ∈ R which implies
a = b since R is antisymmetric. Therefore, R−1 is antisymmetric.
If (a, b) ∈ R−1 and (b, c) ∈ R−1 , then (b, a) ∈ R and (c, b) ∈ R which implies
(c, a) ∈ R since R is transitive. This gives that (a, c) ∈ R−1 which gives that R−1
is transitive. Therefore, (A, R−1 ) is a partially ordered set.
Example 8.10 We have that (Z, ≤) is a partially ordered set. Then by Theorem 8.15,
we have that (Z, ≥) is a partially ordered set.
Example 8.11 In Example 8.7, we considered the partially ordered set P({1, 2, 3})
with set containment as the relation. Using Theorem 8.15, we can use the same set
with reverse containment to get a partially ordered set with the Hasse diagram given
in Fig. 8.4. The set has eight elements, namely, ∅, {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3},
{1, 2, 3}.
We can now look at specific elements satisfying certain properties in a partially

ordered set.
Definition 8.7 An element I in a partially ordered set (A, ≤) is the greatest element
if for all a ∈ A, we have a ≤ I. An element 0 in a partially ordered set (A, ≤) is the
least element if for all a ∈ A, we have 0 ≤ a.
Example 8.12 In the partially ordered set (P(A), ⊆), the greatest element is A and
the least element is ∅.
We shall show the relationship between maximal elements and greatest elements
and between minimal elements and least elements.
Fig. 8.4 Reverse poset of

P({1, 2, 3})
Theorem 8.16 Let (A, ≤) be a partially ordered set. If a is the greatest element
then a is a maximal element. If b is the least element then b is a minimal element.
Proof Let a be a greatest element, then for any c in A we have c ≤ a by definition.

Therefore, there is no c with a ≤ c, c = a. Let b be a least element, then for any c
in A we have b ≤ c by definition. Therefore, there is no c with c ≤ b, c = b.
The converse of this theorem is not true as we have seen in Example 8.9.
For finite lattices we can say something more.
Theorem 8.17 Let (A, ≤) be a partially ordered set with |A| finite. Then A must
have a maximal and minimal element.
Proof Let a1 be an element in A. If there is no a2 with a1 < a2 then a1 is maximal.

Similarly, if there is no ai+1 with ai < ai+1 then ai is maximal. Given that the set
is finite, the sequence ai must be finite, that is, there is an an that is maximal.
A similar proof works for minimal elements by reversing the order of the
relation.
We note that the maximal and minimal elements guaranteed in the previous the-
orem are not necessarily unique. That is, a partially ordered set may have numerous
maximal and minimal elements. This is not true for the greatest and least elements
as we see in the following theorem.
Theorem 8.18 Let (A, ≤) be a partially ordered set, the greatest and least elements,
if they exist, are unique.
Fig. 8.5 Poset on 6 points
Proof Assume I and I are two greatest elements of the partially ordered set. Then
I ≤ I since I is greatest and I ≤ I since I is greatest. Then since the relation is
antisymmetric, we have I = I .
Assume 0 and 0 are two least elements of the partially ordered set. Then 0 ≤ 0
since 0 is least and 0 ≤ 0 since 0 is least. Then since the relation is antisymmetric,
we have 0 = 0 .
Theorem 8.19 Let I be the greatest element and let 0 be the least element in a
partially ordered set (A, R), then 0 be the greatest element and I is the least element
in a partially ordered set (A, R−1 ).
Proof If (a, I) ∈ R for all a then (I, a) ∈ R−1 and if (0, a) ∈ R for all a then (a, 0) ∈
R−1 . This gives the result.
We now want to examine specific kinds of partially ordered sets. We begin with
a necessary definition.
Definition 8.8 Let (A, ≤) be a partially ordered set and let B ⊆ A. The element
a ∈ A is a least upper bound of B if b ≤ a for all b ∈ B and if b ≤ c for all b ∈ B
then a ≤ c. The element d ∈ A is a greatest lower bound of B if d ≤ b for all b ∈ B
and if e ≤ b for all b ∈ B then e ≤ d.
Example 8.13 A subset of a partially ordered set need not have a least upper bound
nor a greatest lower bound. Consider the partially ordered set in Fig. 8.5.
The set {b, e} has neither a greatest lower bound nor a least upper bound.
For a set {a, b} we denote the least upper bound of the set {a, b} by l.u.b.(a, b)
and greatest lower bound of the set {a, b} by g.l.b.(a, b).
Definition 8.9 A partially ordered set is a lattice if and only if for every two elements
a and b in the partially ordered set, the least upper bound l.u.b.(a, b) and the greatest
lower bound g.l.b.(a, b) always exist.
In a lattice, we denote l.u.b.(a, b) as a ∨ b and g.l.b.(a, b) as a ∧ b. Given a

lattice with greatest element I and least element 0 the following rules are immediate
for all a in the lattice:
1. I ∧ a = a;
2. I ∨ a = I;
3. 0 ∧ a = 0;
4. 0 ∨ a = a.
Theorem 8.20 Let Dn be the set of positive integers dividing n. For a, b ∈ Dn , let
a ≤ b if and only if a divides b. Then Dn is a lattice.
Proof First, we note that if a is a divisor of n then a divides a so the relation is

reflexive. If a and b divide n where a divides b and b divides a, then a = b making
the relation antisymmetric. If a, b, and c are divisors of n, where a divides b and b
divides c, then b = ka for some k and c = gb for some b giving c = gk(a) and a
divides c making the relation transitive. Therefore, Dn is a partially ordered set.
Given a and b divisors of n, the g.l.b{a, b} is the largest number d such that d
divides a and d divides b, that is, the greatest common divisor of a and b. Then, if
n = ka, we have n = kda , where a = da so d is a divisor of n. Therefore, the
greatest lower bound exists. The l.u.b.{a, b} is the smallest number m such that a
divides m and b divides m, which is the least common multiple of a and b which
is ab
d = a b. Then a b divides n since b divides n and a divides n and they are
relatively prime. Therefore, m is a divisor of n. Therefore, the least upper bound
exists and Dn is a lattice.
The proof of this theorem gives the following corollary.
Corollary 8.2 Let Dn be the set of positive integers dividing n. Then a ∨ b =

lcm(a, b) and a ∧ b = gcd(a, b).
Theorem 8.21 The partially ordered set (P({1, 2, 3, . . . , k}), ⊆) is a lattice.
Proof Given any two subsets A and B of {1, 2, 3, . . . , k}, l.u.b.(A, B) is the smallest
set C such that A ⊆ C and B ⊆ C. This set C is A ∪ B by definition. Additionally,
g.l.b.(A, B) is the largest set D such that D ⊆ A and D ⊆ B. This set D is A ∩ B
by definition. Therefore, it is a lattice.
We can make the standard definition for the equivalence of partially ordered sets.
Definition 8.10 A partially ordered set (A, ≤) is linearly ordered if for all a, b ∈ A,
either a ≤ b or b < a.
Fig. 8.6 Linearly ordered set

with five points
Colloquially, this definition can be rephrased by saying that any pair of elements
is comparable.
Theorem 8.22 A linearly ordered partially ordered set (A, ≤) is a lattice.
Proof If a, b are any two elements in A, then either a ≤ b or b ≤ a. If a = b then

g.l.b.{a, b} = l.u.b.{a, b} = a. If a < b then g.l.b.{a, b} = a and l.u.b.{a, b} =
b. If b < a then g.l.b.{a, b} = b and l.u.b.{a, b} = a. Therefore, (A, ≤) is a
lattice.
Example 8.14 The Hasse diagram of a linearly ordered set is simply a vertical line
as shown in Fig. 8.6 for a linearly ordered set with five elements.
Definition 8.11 Let (A, ≤) and (A , ≤ ) be partially ordered sets. Then (A, ≤) and
(A , ≤ ) are isomorphic if there is a bijection Φ : A → A such that for a, b ∈ A,
a ≤ b if and only if Φ(a) ≤ Φ(b).
We give a theorem which shows two families of lattices are isomorphic.

k
Theorem 8.23 Let p1 , p2 , . . . , pk be distinct primes with n = i=1 pi . Then Dn
is isomorphic to P({1, 2, 3, . . . , k}).
Proof Make a map between the elements of Dn with P({1, 2, 3, . . . , k}) as follows:

Φ( pi ) = A. (8.16)
i∈A
Fig. 8.7 Poset D30
Since the pi are distinct and every divisor of n is a unique product of the pi by the
Fundamental Theorem of Arithmetic, we have that the map Φ is a bijection. Then
it is immediate that i∈A pi divides i∈B pi if and only if A ⊆ B. Therefore, the
partially ordered sets are isomorphic.
Example 8.15 Consider the partially ordered set D30 in Fig. 8.7. The set has eight
elements, namely, 1, 2, 3, 5, 6, 10, 15, 30.
Notice that the vertices and lines of this Hasse diagram are a two-dimensional
representation of a cube in three dimensions. It is clear from the picture that D30 is
isomorphic to P({1, 2, 3}) by the mapping:
1→∅
2 → {1}
3 → {2}
5 → {3}
6 → {1, 2}
10 → {1, 3}
15 → {2, 3}
30 → {1, 2, 3}
The following idea was first put forth by George Boole in [9]. We are only going to
study finite Boolean algebras. It is possible to also discuss infinite Boolean algebras,
but we are going to restrict ourselves to the finite case. As such, we take the definition
that is easiest and most natural for the finite case.
Definition 8.12 A Boolean algebra is a lattice that is isomorphic to (P(A), ⊆) for

some set A.
Theorem 8.24 Consider a Boolean algebra isomorphic to (P(A), ⊆) for some set
A. Then l.u.b.{B, C} = B ∪ C and g.l.b.{B, C} = B ∩ C.
Proof Any set that contains B and C must contain B ∪ C, moreover B ∪ C is the
smallest set containing both B and C by definition. Therefore, l.u.b.{B, C} = B ∪ C.
Any set that is contained in B and C must be in B ∩ C, moreover B ∩ C
is the largest set contained in both B and C by definition. Therefore, g.l.b.{B, C} =
B ∩ C.
This result shows why the notation is so natural, that is, in the language of lattices,
we have
B∧C=B∩C
and
B ∨ C = B ∪ C.
It naturally follows from this proof that any Boolean algebra is necessarily a
lattice.
Example 8.16 Consider the partially ordered set D210 given in Fig. 8.8. The set has
16 elements, namely, 1, 2, 3, 5, 6, 7, 10, 14, 15, 21, 30, 35, 42, 70, 105, 210.
Notice that the vertices and lines of this Hasse diagram are a two-dimensional
representation of a tesseract, which is a cube in four dimensions. It is clear from the
picture that D210 is isomorphic to P({1, 2, 3, 4}) by the mapping:
1→∅
2 → {1}
3 → {2}
5 → {3}
7 → {4}
6 → {1, 2}
10 → {1, 3}
14 → {1, 4}
15 → {2, 3}
21 → {2, 4}
35 → {3, 4}
30 → {1, 2, 3}
42 → {1, 2, 4}
Fig. 8.8 Poset D210
70 → {1, 3, 4}
105 → {2, 3, 4}
210 → {1, 2, 3, 4}
Therefore, this partially ordered set is a Boolean algebra.
Theorem 8.25 Let (A, ≤) be a finite Boolean algebra then |A| = 2n for some n.
Proof If (A, ≤) is a Boolean algebra then there is a bijection between A and P(B)
for some finite set B. Then P(B) has cardinality 2|B| . This gives the result.
Theorem 8.26 If (S, R) is a Boolean algebra, then (S, R−1 ) is a Boolean algebra.
Proof If (A, R) is a Boolean algebra then it is isomorphic to P(S) for some set S.
Define the map Φ : P(S) → P(S), by Φ(C) = C , where C is the complement of
the set C. Then we have C ⊂ D if and only if D ⊂ C which gives that Φ is a lattice
isomorphism. The lattice (S, R−1 ) is canonically isomorphic to (P(S), ⊇), which
gives the result.
It is immediate that the Hasse diagram for (S, R−1 ) is formed by inverting the
Hasse diagram for (S, R) as in Example 8.11.
Theorem 8.27 Let Dn be the set of positive integers dividing n. For a, b ∈ Dn , let
a ≤ b if and only if a divides b. If n is the product of distinct primes, then Dn is a
Boolean algebra.
s
Proof Assume n is the product of distinct primes, i.e., n = i=1 pi where pi = pj
if i = j. Define a map ψ : Dn → ({1, 2, 3, . . . , s}) by

ψ: pi = C,
i∈C
where C ⊆ {1, 2, 3, . . . , s}. It is immediate that if k, m ∈ Dn , then k divides m if

and only if ψ(k) ⊆ ψ(m).
The converse is left as an exercise, see Exercise 2

We can give an additional operation on a Boolean algebra for any element a,
namely, a , which is defined as the unique element such that a ∨ a = I and a ∧ a =
0 where I is the greatest element and 0 is the least element in the Boolean algebra.
We notice that this corresponds to the complement function for sets. Given the
canonical isomorphism between any Boolean algebra and P(A) the following rules
are immediate for all a, b in the Boolean algebra:
• a ∧ a = a;
• a ∨ a = a;
• a ∧ b = b ∧ a;
• a ∨ b = b ∨ a;
• (a ∧ b) ∧ c = a ∧ (b ∧ c);
• (a ∨ b) ∨ c = a ∨ (b ∨ c);
• a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c);
• a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c);
• (a ∧ b) = a ∨ b ;
• (a ∨ b) = a ∧ b ;
• I = 0;
• 0 = I;
• (a ) = a.
We shall show a canonical way of representing a Boolean algebra. Let A be a

finite set and let B be a subset of A. Define the following characteristic function:

1 a∈B
χB (a) =
0a∈/ B.
Then if |A| = n, list the elements as a1 , a2 , . . . , an . Then each subset of A can

be defined as a vector of length n, where the ith coordinate is given by χB (ai ).
Fig. 8.9 Boolean algebra on

{0, 1}2
In this sense, each Boolean algebra can be thought of as a partially ordered set
on {0, 1}n . The lattice operations can be done coordinate-wise with the following
functions:
∧01 ∨01
0 00 0 0 1.
1 01 1 11
Applying these functions coordinate-wise, we have for v, w ∈ {0, 1}n ,
l.u.b.{v, w} = v ∨ w
and
g.l.b{v, w} = v ∧ w.
For example, 0011 ∧ 1110 = 0010 and 0011 ∨ 1110 = 1111.
We note then that the compliment function is given as 0 = 1 and 1 = 0.
Example 8.17 The Hasse diagram for the Boolean algebra on {0, 1}2 is given in
Fig. 8.9:
The Hasse diagram for the Boolean algebra on {0, 1}3 is given in Fig. 8.10:
The Hasse diagram for the Boolean algebra on {0, 1}4 is given in Fig. 8.11:
Exercises
1. In D72 compute the following:
a. 8 ∨ 6
b. 8 ∧ 6
c. (2 ∨ 9) ∧ 4
d. (2 ∧ 9) ∨ 4
2. Prove that if n is not square-free then Dn is not a Boolean algebra.

Fig. 8.10 Boolean algebra on {0, 1}3
Fig. 8.11 Boolean algebra on {0, 1}4

3. Compute the following in {0, 1}5 :
a. 11011 ∧ 01011
b. 10011 ∨ 11100
c. (10100 ∧ 00111)
d. (00111 ∨ 11101)
4. Draw the Hasse diagram for Boolean algebras on {0, 1}n for n = 2, 3, 4 using
the inverse relation.
5. Prove that any two Boolean algebras with the same cardinality are isomorphic.
6. Give an example of the following:
a. A partially ordered set that has no greatest element.

b. A partially ordered set that has no least element.
c. A maximal element that is not a greatest element.
d. Two non-isomorphic partially ordered sets with four elements.
e. A partially ordered set with no maximal element.
f. A partially ordered set with no minimal element.
g. A partially ordered set that is not a lattice.
h. A lattice that is not a Boolean algebra.
7. Prove that if (A, R) is a lattice, then (A, R−1 ) is a lattice. Describe its Hasse
diagram.
8. Draw the Hasse diagram for D2310 .
9. Prove that the if n > 2, then a Boolean algebra with n elements is not linearly
ordered.
8.6 Association Schemes
Association schemes have their origin in statistics in terms of the design of exper-
iments. They have found numerous applications in combinatorics and algebra and
have found particular application in coding theory. Association schemes were intro-
duced in [10]. For further study, see [7]. We begin with the standard definition.
Definition 8.13 Let X be a finite set, with |X| = v. Let Ri be a subset of X × X, for
all i ∈ I = {i | i ∈ Z, 0 ≤ i ≤ d} with d > 0. We define = {Ri }i∈I . We say that
(X, ) is a d-class association scheme if the following properties are satisfied:

1. The relation R0 = (x, x) : x ∈ X is the identity relation.
2. For every x, y ∈ X, (x, y) ∈ Ri for exactly one i.
T T
3. every i ∈ I, there
For exists i ∈ I such that Ri = Ri , that is, we have Ri =
(x, y) | (y, x) ∈ Ri .
8.6 Association Schemes 231
4. If (x, y) ∈ Rk , the number of z ∈ X such that (x, z) ∈ Ri and (z, y) ∈ Rj is a

constant pkij .
The values pk ij are called the intersection numbers of the association scheme.
The elements x, y ∈ X are called i-th associates if (x, y) ∈ Ri . If i = i for all i,
namely, RTi = Ri for all i, then the association scheme is said to be symmetric. An
association scheme that does not satisfy this condition is said to be non-symmetric.
The association scheme (X, ) is said to be commutative if pk k
ij = pji , for all i, j, k ∈
I. Note that a symmetric association scheme is always commutative but that the
converse of this statement is not true.
Relations are often described by their adjacency matrices. Namely, given a relation
R on a set X, the elements of X are the coordinates of the matrix MR , and

1 (a, b) ∈ R
(MR )a,b =
0 (a, b) ∈
/ R.
In this setting, we shall denote the adjacency matrix for the relation Ri by Ai . Then
Ai is the v × v matrix whose rows and columns are labeled by the points of X and
defined by

1 if (x, y) ∈ Ri
(Ai )x,y =
0 otherwise.
Conditions 1 − −4 in the definition of (X, ) are equivalent to the following
conditions:
1. The matrix A0 = Iv , where Iv is the v by v identity matrix.

2. Summing the matrices, we have i∈I Ai = Jv , where Jv is the v by v matrix
where all entries are 1.
3. For all i ∈ I, there exists i ∈ I, such that Ai = ATi .
k
4. For all i, j ∈ I we have Ai Aj = pij Ak .
k∈I
If the association scheme is symmetric, then it follows that Ai = ATi , for all
i ∈ I. If the association scheme is commutative, then matrix multiplication on the Ai
commutes, namely, Ai Aj = Aj Ai , for all i, j ∈ I. The adjacency matrices generate
a (n + 1)-dimensional algebra A of symmetric matrices. This algebra is called the
Bose-Mesner algebra.
Given the description of association schemes it is easy to see that an association
scheme can be thought of as the complete graph on v vertices where an edge (a, b)
is marked by i if (a, b) ∈ Ri . Similarly, we can think of it as a v by v matrix where
the entry in (a, b) is i if (a, b) ∈ Ri .
We shall describe one of the most useful and interesting association schemes.
Consider the space Fn n
2 . It consists of 2 vectors of length n with entries either 0
or 1. Given two such vectors v and w ∈ Fn 2 , the Hamming distance is defined as
dH (v, w) = |{i | vi = wi }|.
Theorem 8.28 Let Ri be the relation on Fn 2 , where (v, w) ∈ Ri if and only if

dH (v, w) = i. Then (Fn
2 , {R 0 , R1 , . . . , Rn }) is an n-class association scheme.
Proof First, we have that a vector v has Hamming distance 0 from a vector w if and
only if v = w. Therefore, R0 is the identity relation.
Secondly, any two vectors v and w have a unique Hamming distance; therefore,
for all v, w ∈ Fn2 , we have that (v, w) ∈ Ri for exactly one i.
Thirdly, since dH (v, w) = dH (w, v) we have that each relation Ri is symmetric.
Fourthly, let (v, w) ∈ Rk , that is, dH (v, w) = k. We need to determine the number
of u such that dH (v, u) = i and dH (u, w) = j. We are changing i coordinates of the
vector v to change to get to u. Then we have j coordinates of u to change to get to
w. Noting that we must change in total k coordinates to get from v to w. We get that

C(k, i−j+k )C(n − k, i+j−k ) i + j − k ≡ 0 (mod 2)
pk
ij = 2 2
0 i + j − k ≡ 1 (mod 2).
We note that 1−j+k

2 + 1+j−k
2 = i which is the number of coordinates of v that are
changed.
Therefore, this is an association scheme.
This association scheme is known as the Hamming scheme. We note immediately

that the scheme is symmetric.
Example 8.18 Consider the Hamming scheme on F22 . It is a 2-class association

scheme with the following matrices:
A0 (0, 0) (0, 1) (1, 0) (1, 1)

(0, 0) 1 0 0 0
(0, 1) 0 1 0 0
(1, 0) 0 0 1 0
(1, 1) 0 0 0 1
A1 (0, 0) (0, 1) (1, 0) (1, 1)
(0, 0) 0 1 1 0
(0, 1) 1 0 0 1
(1, 0) 1 0 0 1
(1, 1) 0 1 1 0
A2 (0, 0) (0, 1) (1, 0) (1, 1)
(0, 0) 0 0 0 1
(0, 1) 0 0 1 0
(1, 0) 0 1 0 0
(1, 1) 1 0 0 0
We note that the sum of the matrices is the all one matrix J4 .
We can realize the association scheme in the following complete graph. Note that
since the scheme is symmetric, the edges do not need to be directed.
Finally, we can use a matrix to represent this graph as follows:
(0, 0) (0, 1) (1, 0) (1, 1)

(0, 0) 0 1 1 2
(0, 1) 1 0 2 1
(1, 0) 1 2 0 1
(1, 1) 2 1 1 0
Example 8.19 Geometrically, we can think of the elements in the Hamming scheme
as the vertices on an n-dimensional hypercube and the distance as the distance, in
edges, in that hypercube. For n = 2, we have the following 3-dimensional cube.
The matrix for this scheme is as follows:
(0, 0, 0) (0, 0, 1) (0, 1, 0) (0, 1, 1) (1, 0, 0) (1, 0, 1) (1, 1, 0) (1, 1, 1)

(0, 0, 0) 0 1 1 2 1 2 2 3
(0, 0, 1) 1 0 2 1 2 1 3 2
(0, 1, 0) 1 2 0 1 2 3 1 2
(0, 1, 1) 2 1 1 0 3 2 2 1
(1, 0, 0) 1 2 2 3 0 1 1 2
(1, 0, 1) 2 1 3 2 1 0 2 1
(1, 1, 0) 2 3 1 2 1 2 0 1
(1, 1, 1) 3 2 2 1 2 1 1 0
In each row and column, there are 3 ones, 3 twos, 1 three, and 1 zero. This
corresponds to the line in Pascal’s triangle 1 3 3 1 giving the binomial coefficients
for n = 3.
We shall now show how to make an association scheme from any finite group.
Groups are defined in Chap. 10 in Definition 10.1.
Theorem 8.29 Let G be a finite group with elements g0 , g1 , . . . , gn−1 where

g0 is the identity element of the group. Define Ri = {(x, y) | x = gi y}. Then
(G, {R0 , R1 , . . . , Rn−1 }) is an association scheme.
Proof Firstly, the relation R0 consists of all pairs (x, y) such that x = g0 y where g0
is the identity. Therefore, R0 is the identity relation by definition.
Secondly, given a pair (x, y), we have that if xg = y, then g = x−1 y. Hence,
there is a unique gi with (x, y) ∈ Ri .
Thirdly, if (x, y) ∈ Ri then x = gi y. Then gi−1 x = y, giving that (y, x) ∈ Ri
where gi = gi−1 . Therefore, RTi = Ri .
Fourthly, if (x, y) ∈ Rk this gives that xy−1 = gk . If (x, z) ∈ Ri and (z, y) ∈ Rj ,
then we have xz−1 = gi and zy−1 = gj . Then we have xz−1 zy−1 = xy−1 so we
we need gi gj = gk . Given any gi there is a unique gj satisfying this condition. For
this pair i, j, there is a unique z satisfying xz−1 = gi . Therefore, the fourth condition
is satisfied.
It is clear that the association scheme is commutative if and only if the group
G is commutative. The scheme is symmetric if and only if each element is its own
inverse.
Example 8.20 Consider the symmetric group of 3 letters, which consists of all per-
mutations of three elements. The group consists of the following elements: {g0 =
(1), g1 = (1, 2), g2 = (1, 3), g3 = (2, 3), g4 = (1, 2, 3), g5 = (1, 3, 2)}. We shall give
the matrix coming from the complete graph associated with this association scheme.
g0 g1 g2 g3 g4 g5
g0 0 1 2 3 4 5
g1 1 0 5 4 2 3
g2 2 4 0 5 3 1
g3 3 5 4 0 1 2
g4 5 2 3 1 0 4
g5 4 3 1 2 5 0
We note that in this case the matrix of the association scheme is a Latin square which
is not symmetric. Each matrix Ai for this scheme is actually a permutation matrix,
with precisely six ones in each Ai .
Exercises
1. Construct the matrix for the association scheme on the group (F2 × F2 , +). Prove
that this scheme is symmetric.
2. Let S be a set with n elements. For the Johnson scheme, the points are the C(n, k)
subsets of S with k elements. Two subsets x, y satisfy (x, y) ∈ Ri if and only if
|x ∩ y| = k − i. Prove that this is an association scheme.
3. Construct the Ai for the relation given in Example 8.20.
4. Draw the labeled complete graph for the relation given in Example 8.20.
5. Construct an infinite family of association schemes, based on groups, that is
symmetric.
Discrete Probability—A Return to
Counting 9
In this chapter, we return to counting problems, but we study them to understand the
probability of a given event. While most branches of mathematics have fairly lofty
origins, probability has its origins in gambling. Specifically, in determining when
a game of chance is fair to all involved. One of the earliest texts on the subject is
Christaan Huygens’ book [51].
The probability of an event is a number between 0 and 1 where 0 indicates that
event is impossible and 1 indicates that the event must occur. While probability has
been extended to infinite sets as well, we shall restrict ourselves to combinatorial
probability, meaning probability applied to finite sets. There are many important
differences between probability with finite sets as opposed to probability with infinite
sets. For example, with infinite sets a probability of 0 indicates something will almost
never happen and a probability of 1 indicates that an event will almost always happen.
For example, the probability of picking an irrational number out of the real numbers
is 1, even though rational numbers do exist. However, with finite sets the adjective
almost is done away with and we are talking about events never occurring or always
occurring.
9.1 Definitions and Elementary Probability
We begin with the standard definition of probability.
Definition 9.1 Given a sample space S and an event E ⊂ S, the probability of event
E is
|E|
P(E) = . (9.1)
|S|
238 9 Discrete Probability—A Return to Counting
Example 9.1 Assume S = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}. The probability of picking an

5 5
even number from this set is 10 , and the probability of picking an odd number is 10 .
9
The probability of picking a number greater than 1 is 10 . The probability of picking
17 is 0.
Example 9.2 In a well-known lottery game, a player is asked to pick a 3-digit number
ranging from 000 to 999. If the number chosen matches the number chosen for that
day, the state running the game will pay 500 dollars on a one dollar bet. There are 1000
1
possible outcomes and 1 that wins so the probability of winning is P(W) = 1000 .
Hence, the game is highly unfair since a payoff of 1000 dollars would be required to
make the game fair.
Example 9.3 In a common lottery game, a player picks six numbers from 1 to 40.
The state running the game also picks 6 for the day. If the player’s 6 matches the
state’s 6 then the player wins. There are C(40, 6) = 3, 838, 380 possible ways of
1
picking 6 from 40. Therefore, the probability of winning is P(W) = 3,838,380 . For
the game to be fair, a payoff of 3, 838, 380 dollars on a 1 dollar bet would be required.
In general, the payoff is determined by how many people have bet and when the last
winner was. In other words, if there is no winner on a play, the money collected in
the pot continues on to the next play. Given this scenario, if a player only wanted
to play a fair game, they should only play the game when the pot hits 3, 838, 380
dollars. Of course, even though the game would be fair, the probability of actually
winning would still be quite low.
Example 9.4 In American roulette, a small ball is placed on a wheel which is spun.
The ball can land in any of 38 positions corresponding to 00, 0, 1, 2, . . . , 36. Of
these 38 positions, 18 are red, 18 are black, and 2 (0 and 00) are green. A player
can choose a variety of ways of betting. For example, one can bet that the number
will be red, black, even, odd, one of four numbers (for certain groups), one of two
numbers (for certain pairs). If there are n ways a player can win, then the probability
n
of winning on that bet 38 . For example, the probability of winning on a bet of red is
18 9 1
P(R) = 38 = 19 , and the probability of winning on a bet of single number is 38 .
However, in the first case, the payoff on a bet of 1 dollar would be 2 dollars and in the
second case the payout on a bet of one dollar is 36 dollars (including the dollar bet
for both instances). In essence, the casino pays off as if there were 36 possibilities
rather than 38 possibilities. Therefore, the game is not fair.
Theorem 9.1 Let S be a finite sample space and let A and B be subsets of S.
1. 0 ≤ P(A) ≤ 1.
2. P(∅) = 0, P(S) = 1.
3. P(A ∪ B) = P(A) + P(B) − P(A ∩ B).
4. P(A) + P(A) = 1.
5. If A ⊆ B then P(A) ≤ P(B).
9.1 Definitions and Elementary Probability 239
Proof Statements 1, 2, and 5 are immediate from the definition.

For Statement 3, we know that |A ∪ B| = |A| + |B| − |A ∩ B|. Therefore
A ∪ B| |A| + |B| − |A ∩ B|
P(A ∪ B) = =
|S| |S|
|A| |B| |A ∩ B|
= + −
|S| |S| |S|
= P(A) + P(B) − P(A ∩ B).

For Statement 4, we have A = S \ A, so |A| = |S| − |A|. Then
|A| |S| − |A| |S|

P(A) + P(A) = + = = 1.
|S| |S| |S|
Definition 9.2 Two events A and B in a sample space S are mutually exclusive if
A ∩ B = ∅.
It follows from the definition and Statement 3 in Theorem 9.1 that if two events
A and B are mutually exclusive then P(A ∪ B) = P(A) + P(B).
Assume you have a standard deck of 52 cards. The probability of picking either
an ace or a spade is 16 4
52 = 13 since there are 13 spades one of which is an ace and 3
1
other aces. However, the probability of picking an ace is 13 , and the probability of
1 1 1 17 16
picking a spade is 4 and 13 + 4 = 52 = 52 , since these events are not mutually
exclusive.
Example 9.5 The probabilities for rolling a pair of dice are given below:
Dice sum Probability

1
2 36
2
3 36
3
4 36
4
5 36
5
6 36
6
7 36
5
8 36
4
9 36
3
10 36
2
11 36
1
12 36
These events are all mutually exclusive. As an example, the probability of getting a
2 3 6 1
3, 4, 7, or 12 is 36 + 36 + 36 + 36 = 12 1
36 = 3 .
Definition 9.3 Two events A and B in a sample space S are independent if P(A ∩
B) = P(A)P(B).
Intuitively, one thinks of this definition as A and B are independent if the occur-
rence of A has no bearing on the occurrence of B. For example, assume you pick a
ball out of a bin with a green ball, a red ball, and a yellow ball, then you roll a die.
There are 18 possible outcomes. The probability of picking a green ball followed
3
by rolling a 2, 4, or 6 is 18 = 16 as the three events are (green, 2), (green, 4), and
(green, 6). Now the probability of picking a green ball out of the bin is 13 , and the
probability of picking either a 2, 4, or 6 is 12 . Then the product of these probabilities
is 13 ( 12 ) = 16 . Therefore, the events are independent as we would suspect.
Example 9.6 The probability of flipping a coin ten times in a row and getting a
head each time is ( 12 )10 = 1024
1
since the events (each flip of the coin) are inde-
pendent. One might also think of this as there are 210 possible outcomes, and
1
HHHHHHHHHH is the only one with all heads. Therefore, the probability is 1024 .
Example 9.7 The probability that flower A is blooming is 0.9, and the probability
that flower B is blooming is 0.7. Then the probability that both flowers are blooming is
(0.9)(0.7) = 0.63. The probability that neither is blooming is (0.1)(0.3) = 0.03. The
probability that flower A is blooming and flower B is not is (0.9)(0.3) = 0.27, and the
probability that flower A is not blooming and flower B is blooming is (0.1)(0.7) =
0.07. We note that 0.63 + 0.03 + 0.27 + 0.07 = 1.
Theorem 9.2 Assume that the probability of an event occurring is P. The probability
that the event will occur k times in n trials is C(n, k)Pk (1 − P)n−k .
Proof There are C(n, k) ways that the event will occur k times in n trials. In each
of these, there are k locations for the event to occur and n − k locations for the event
not to occur. Therefore, the probability of this occurring is Pk (1 − P)n−k .
We notice that in this theorem, the value is the coefficient in the binomial theorem
with P and 1 − P substituted for x and y. Specifically
n

(x + y)n = C(n, k)xk yn−k
k=0
n
(P + (1 − P))n = C(n, k)Pk (1 − P)n−k
k=0
n
1= C(n, k)Pk (1 − P)n−k .
k=0
9.1 Definitions and Elementary Probability 241
This shows that the sum of the probabilities of the event occurring is 0, 1, 2, . . . , n
times is 1 as we would have expected.
Example 9.8 Assume there is a box with a red ball, a green ball, and a blue ball.
Assume a ball is extracted 4 times replacing it each time. Then, the probability that
a green ball is picked exactly twice is C(4, 2)( 13 )2 ( 23 )2 = 6( 19 94 ) = 24
81 .
Example 9.9 Assume a coin is flipped 7 times. To determine the probability that we
get at least 5 heads, we sum the probability that we get for 5 heads, 6 heads, and 7
heads. Namely, the probability is
1 1 1 1 1 1 29
C(7, 5)( )5 ( )2 + C(7, 6)( )6 ( )1 + C(7, 7)( )7 = (21 + 7 + 1)( )= .
2 2 2 2 2 128 128
Exercises
1. Determine the probability of rolling a pair of dice and getting a number greater
than or equal to 9.
2. Determine the probability of flipping a coin 10 times in a row and getting no
heads.
3. Determine the probability of flipping a coin 8 times and getting exactly 3 heads.
4. Determine the probability of flipping a coin 8 times and getting at least 6 heads.
5. Determine the probability of rolling a pair of dice twice and having a sum greater
than or equal to 6 for each throw.
6. Determine the probability of shuffling a deck of cards and having spades in the
first 13 spaces.
7. Determine the probability that the first and second numbers are different when
rolling a single die twice.
8. Determine the probability that the first and second numbers are equal when
rolling a single die twice.
9. Determine the probability that a coin is flipped 9 times and you get exactly 4
tails.
10. Determine the probability that a coin is flipped 7 times and you get exactly 5
heads.
11. Determine the probability that a coin is flipped 10 times and you get at least 9
heads.
12. Determine the probability that if six people, all with different ages, are arranged
in a row, that they will be arranged in descending order.
13. Assume there are six married couples in a room. Determine the probability that
if two people are chosen at random that they are married.
14. The hands in 5-card poker are two or kind, three of a kind, straight, flush, full
house, four of a kind, straight flush, royal flush. Determine the probabilities of
getting these hands and show that this is the proper order for the hands given that
a hand with a lesser probability should beat a hand with a greater probability.
15. Assume a standard deck of cards is shuffled. Determine the probability that it is
arranged in the same manner as a given deck.
16. Assume there are 10 questions on a multiple-choice test, where each question
has three possible answers and that a student answers the questions randomly.
Determine the probability that the student passes the test (where a pass is a 70%).
17. Assume p, q, and r are distinct primes. Determine the probability that a number
is chosen between 1 and pqr is relatively prime to pqr.
18. Assume an event has probability P. Prove that there are a number of trials such
that the probability of the event occurring at least once is greater than or equal
to 23 .
19. Determine the probability that three points on a projective plane of order n
chosen randomly are on a line.
20. Determine the probability that three points on an affine plane of order n chosen
randomly are on a line.
9.2 Conditional Probability
We shall now discuss conditional probability.
Definition 9.4 In a sample space S, with events A and B, the conditional probability
is defined as
P(A ∩ B)
P(A|B) = .
P(B)
We see immediately from the definition that if A and B are independent, then
P(A ∩ B) = P(A)P(B) and so P(A|B) = P(A) and P(B|A) = P(B).
Example 9.10 Given a standard deck of 52 cards. To determine the probability of

choosing a queen given that you have chosen a card, that is, a heart is P(A|B) =
1 1 1
52 / 4 = 13 . To determine the probability of choosing a heart given that the card
1 1
chosen is a queen is P(B|A) = 52 / 13 = 14 .
We shall now discuss one of the most interesting examples of conditional proba-
bility. It is known as the Monty Hall problem. It is called this because of its similarity
to events on an American television show hosted by Monty Hall. The scenario is as
follows. There are three curtains which hide prizes. Label these 1, 2, and 3. Behind
two of the curtains are worthless prizes, for the purpose of this description let us say
there is a can of corn behind these curtains. Behind one of the curtains is a valuable
prize, for the purpose of this description let us say there is a car behind this curtain.
The person playing the game wants to win the car. We assume that the prizes are
randomly assigned to the curtains.
9.2 Conditional Probability 243
Fig. 9.1 Monty Hall

problem
The host offers the player a choice of the three curtains. Let us suppose the player
picks curtain 1. The host then reveals one of the remaining two curtains as hiding the
can of corn. Note that we assume that the host chooses randomly if both contain a can
of corn. He then turns to the player and asks if the player would like to change their
answer. The question is should the player now change their answer to the remaining
curtain or stick with the curtain they originally chose. It is tempting to assume that
you should not change since the probability that the car is behind any curtain is 13 .
However, this is not the correct decision to make.
An intuitive way to think about the problem is that if a player does invoke the
switching strategy the only way they could lose is if they choose the correct curtain
initially, which has probability 13 . Therefore, the probability of winning with this
strategy is 1 − 13 = 23 .
A more rigorous description uses the following argument. The diagram in Fig. 9.1
indicates the probability of getting a car by initially choosing a door and then switch-
ing after the host raises a curtain that does not contain the car. Each of the first three
probabilities is 13 for the possibilities of the car being behind each curtain. The top
one is if the player chooses the curtain where the car is initially. Then the next two
probabilities are 12 for the possibilities for which curtain the host opens. In the bottom
two are the possibilities when the player chooses a curtain that does not contain the
car. Then the host has only one curtain which can be opened.
The probability for getting the car is ( 13 )(1) + ( 13 )(1) = 23 .
Finally, we can use conditional probability to determine the probability of success
using the switching technique. We take the probability of event A (ending on the right
curtain) given that a curtain unequal to the original choice is revealed and we have
1
3 2
1 1
= .
3 + 6
3
For a complete description of the Monty Hall problem and its implications, see
[76].
Theorem 9.3 (Bayes’ Theorem) In a sample space S, with events A, B,
P(B|A)P(A)
P(A|B) = . (9.2)
P(B)
Proof By the definition of conditional probability we have
P(A ∩ B)
P(A|B) =
P(B)
and
P(B ∩ A)
P(B|A) = .
P(A)
P(B|A)P(A)
This gives P(A)P(B|A) = P(B)P(A|B) and finally P(A|B) = P(B)
.
Theorem 9.4 Let S be a sample space with events A, B.
1. Events A and B are independent if and only if P(A) = P(A|B).

2. Events A and B are independent if and only if P(B) = P(B|A).
Proof We shall prove the first statement; the proof of the second is identical. Recall
by Definition 9.3 that A and B are independent if P(A ∩ B) = P(A)P(B). Then
P(A ∩ B)
P(A)P(B) = P(A ∩ B) ⇔ P(A) = = P(A|B).
P(B)
Exercises
1. Determine the probability that a card chosen from a standard deck is an ace given
that it is a spade.
2. The probability that a train departs on time is 0.9 and the probability that a train
both departs and arrives on time is 0.75. Given that the train departs on time
determine the probability that it arrives on time.
3. Determine the probability that a card chosen from a standard deck of cards is a
face card given that it is red.
4. Determine the probability that a card chosen from a standard deck of cards is a
king given that it is a spade.
Automorphism Groups
10
In this chapter, we shall use one of the most important structures in abstract algebra
as a tool to study finite incidence structures. The algebraic structure is a group. It
is generally the first structure one encounters in studying abstract algebra. We shall
begin with a very elementary study of finite groups, and then we shall study the
groups associated with various combinatorial structures.
10.1 Groups
The theory of groups is one of the largest and most important branches in all of
abstract algebra. It would be almost impossible to exhaust the study of groups in a
lifetime, let alone in a text. Here, we shall only require the most elementary facts
about groups in order to understand a bit about how to use them to study combinatorial
objects. The study of finite groups has many connections to the study of combinatorial
objects, and important results in both areas have come from this connection.
We begin with the definition of a group.
Definition 10.1 A group (G, ∗) is a set G with an operation ∗ such that
1. (Closure) If a, b ∈ G then a ∗ b ∈ G.
2. (Associativity) For all a, b, c ∈ G, (a ∗ b) ∗ c = a ∗ (b ∗ c).
3. (Identity) There exists an e ∈ G with e ∗ a = a ∗ e = a for all a ∈ G.
4. (Inverses) For all a ∈ G there exists b ∈ G with a ∗ b = e.
The order of the group (G, ∗) is |G|. If H ⊆ G and (H, ∗) is a group, then H is
said to be a subgroup of G.
246 10 Automorphism Groups
Example 10.1 The following are examples of groups: (Z, +), (Q, +), (R, +), (Q −
{0}, ∗), (R − {0}, ∗), (Zn , +). Notice that (Z − {0}, ∗) is not a group since, for exam-
ple, 2 has no multiplicative inverse in Z − {0}.
We note that all of the groups in the previous example have a commutative oper-
ation. These are given a special designation.
Definition 10.2 If (G, ∗) is a group such that a ∗ b = b ∗ a for all a, b ∈ G, then

the group is said to be an Abelian group.
The term Abelian comes from the name of the Norwegian mathematician Niels
Henrik Abel.
Example 10.2 Let Sn denote the set of all n! permutations on a set of size n. Let
the operation on this set be functional composition. Then (Sn , ◦) is a group of order
n! that is non-Abelian. This group is called the symmetric group on n letters. The
group Sn is an extremely important group since it can be shown that all groups can
be viewed as subgroups of this group.
Theorem 10.1 The operation table of a finite group is a Latin square.
Proof Let (G, ∗) be a finite group, and let L be a matrix indexed by the elements of
the finite group G = {g1 , g2 , . . . , gn } with Lgi ,gj = gi ∗ gj .
Consider the i-th row of L. This consists of the elements
gi ∗ g1 gi ∗ g2 . . . gi ∗ gn .
If gi ∗ gj = gi ∗ gk then if g−1
i is the inverse guaranteed by axiom 4 then
g−1 −1
i ∗ gi ∗ gj = gi ∗ gi ∗ gk
gj = gk .
Hence, each element of the group appears exactly once in each row. The j-th column
of L consists of the elements
g1 ∗ gj g2 ∗ gj . . . gn ∗ gj .
If gi ∗ gj = gk ∗ gj then if g−1
j is the inverse guaranteed by axiom 4 then
gi ∗ gj ∗ g−1
j = gk ∗ gj ∗ g−1
j
gj = gk .
Hence, each element of the group appears exactly once in each column.
10.1 Groups 247
Example 10.3 We give the operation table of the Klein-4 group, where e denotes
the identity.
∗ e a b c
e e a b c
a a e c b
b b c e a
c c b a e
We note that the 4 by 4 table gives a non-circulant Latin square of order 4. We can
also give the operation table of the group (Z4 , +).
+ 0123
0 0123
1 1230
2 2301
3 3012
Here, the 4 by 4 table gives a circulant Latin square of order 4.
For the rest of this section, we shall refer to a group (G, ∗) simply as G whenever
the operation is understood. Additionally, the operation will be denoted by juxtapo-
sition.
The order of an element g in a group G with identity e is the smallest natural
number n > 0 such that gn = e.
Let G and H be groups; then, the cross product of G and H is defined by
G × H = {(g, h) |g ∈ G, h ∈ H}, (10.1)
where
(g, h) ∗ (g , h ) = (g ∗G g , h ∗H h ) (10.2)
and ∗G is the operation in G and ∗H is the operation in H.
From this point, we can use juxtaposition to indicate the operation of the group.
Theorem 10.2 Let G and H be groups. Then G × H is a group.
Proof If (g, h), (g , h ) ∈ G × H, then gg ∈ G, hh ∈ H since they are groups

which gives (g, h)(g , h ) ∈ G × H. Therefore it is closed.
If (g, h), (g , h ), (g , h ) ∈ G × H then
((g, h)(g , h ))(g , h ) = (gg , hh )(g , h ) = ((gg )g , (hh )h ) =

(g(g g ), h(h h )) = (g, h)(g g , h h ) = (g, h)((g , h )(g , h )).
Therefore, the operation is associative.

Let eG and eH be the identity elements of G and H, respectively. Then (eG , eH )

(g, h) = (eg g, eH h) = (g, h) = (geg , heH ) = (g, h)(eG , eH ). Therefore, G × H
has an identity element.
Let g−1 be the inverse of g in G and let h−1 be the inverse of h in H. Then
(g, h)(g−1 , h−1 ) = (gg−1 , hh−1 ) = (eG , eH )

= (g−1 g, h−1 h) = (g−1 , h−1 )(g, h).
Therefore, every element has an inverse.
Example 10.4 Consider the group Z2 × Z2 . It has the following table:
+ (0, 0) (0, 1) (1, 0) (1, 1)

(0, 0) (0, 0) (0, 1) (1, 0) (1, 1)
(0, 1) (0, 1) (0, 0) (1, 1) (1, 0)
(1, 0) (1, 0) (1, 1) (0, 0) (0, 1)
(1, 1) (1, 1) (1, 0) (0, 1) (0, 0)
Let (G, ∗) be a group and H a subgroup of G, that is, H ⊆ G. A (left) coset of H

is aH = {a ∗ h |h ∈ H}.
Lemma 10.1 Let H be a subgroup of a finite group G then |aH| = |H| for all a ∈ G.
Proof Let ψ : H → aH by ψ(g) = ag. We shall show that ψ is a bijection. If

ψ(g) = ψ(g ) then ag = ag which implies a−1 ag = a−1 ag since we are in a
group and we know that a has an inverse with respect to the operation of G. This
gives that g = g and ψ is injective. Any element in aH is of the form ag for some
g ∈ H so ψ is naturally surjective and hence ψ is a bijection. Therefore, we have
that |aH| = |H|.
Example 10.5 Consider the group (Z8 , +). The group {0, 4} is a subgroup. The
cosets all have two elements, namely, 1 + {0, 4} = {1, 5}, 2 + {0, 4} = {2, 6}, and
3 + {0, 4} = {3, 7}.
Lemma 10.2 The cosets of a subgroup H partition G.
Proof We shall show that membership in a coset is an equivalence relation. That is,
a is related to b if b ∈ aH.
• Reflexive: Any element a is related to itself since ae ∈ aH where e is the identity

element of G.
10.1 Groups 249
• Symmetric: If b ∈ aH then b = ag for some g ∈ H. Then multiplying by g−1

on the right, which must also be in H since H is a group gives that a = bg−1 and
a ∈ bH.
• Transitive: If b ∈ aH and c ∈ bH then b = ag and c = bg for some g, g ∈ H.
Then c = a(gg ) ∈ aH and the relation is transitive.
The equivalence classes partition the group G.
Example 10.6 Consider the cosets given in Example 10.5. The set {0, 1, 2, 3, 4,
5, 6, 7} is partitioned by {0, 4}, {1, 5}, {2, 6}, and {3, 7}.
For a group G and a subgroup H, we define the index of H in G as the number of

cosets of H in G and denote it by [G : H].
Theorem 10.3 (Lagrange’s Theorem) For a finite group G and a subgroup H, we

have
|G| = |H|[G : H]. (10.3)
Proof Let G be a group and H a subgroup. Then by Lemma 10.2, we have
G = a1 H ∪ a2 H ∪ · · · ∪ a[G:H] H, (10.4)
where a1 = e and ai H ∩ aj H = ∅ when i

= j. Then since |ai H| = |H| by
Lemma 10.1 we have that |G| = |H|[G : H], which gives the result.
Corollary 10.1 Let G be a finite group G and let H be a subgroup of G. Then the
order of H must divide the order of G.
Proof By the previous theorem, we have |G| = |H|[G : H]. Then, since [G : H] is an
integer we have the result.
Example 10.7 Consider the group (Z10 , +), then the non-trivial subgroups are {0, 5}
and {0, 2, 4, 6, 8}. These are the only non-trivial subgroups since 2 and 5 are the only
integers dividing 10.
We shall now describe the action of a group on a set. We have seen an example
of this in Sect. 1.4 as Sn acted on any set with n elements. Let G be a group and
S a set. Then we say G acts on the set S if gs ∈ S for all g ∈ G and s ∈ S and
g(g s) = (gg )s for g, g ∈ G.
Define the orbit of an element a ∈ S as
Orb(a) = {b | there exits σ ∈ G with σ(a) = b}. (10.5)

Theorem 10.4 Let G be a group acting on a set S. For a ∈ S the set of elements of
G that fix a, i.e., H = {σ | σ ∈ G, σ(a) = a} is a subgroup of G.
Proof If σe is the identity of G, then σe a = a so a ∈ H and it is non-empty. If

σ, τ ∈ H, then (στ)(a) = σ(τ(a)) = σ(a) = a and H is closed. Also σ−1 σ(a) =
eσ (a) = a but σ−1 (σ(a)) = σ−1 (a) which gives σ−1 (a) = a and therefore H is
a subgroup of G.
Let G be a group acting on a set S. Let a ∈ S and let H be the subgroup of G of

elements that fix a. If τ ∈ σH then τ(a) = σ(a). This is easy to see since τ = σ ◦ σ
where σ fixes a so τ(a) = σ(σ (a)) = σ(a). Hence, the number of elements in the
orbit is precisely the number of cosets of the fixed subgroup in the ambient group.
Theorem 10.5 Let G be a group acting on a set S. For a ∈ S we have
|G|
|Orb(a)| = , (10.6)
|H(a)|
where H(a) is the subgroup of G that fixes a.
2πi
Consider the unit circle in the complex plane given in Fig. 10.1. Let ξn = e n
which corresponds to the complex number on the unit circle 2π n radians around the
circle in the counterclockwise direction. We have that ξn is a primitive root of unity,
meaning ξn i
n = 1 and ξn
= 1 for 1 ≤ i < n.
It is immediate that the set {1, ξn , ξ2n , . . . , ξn−1
n } forms a group of order n. This
group is the cyclic group of order n, denoted Cn . More abstractly, it can be defined
as
Cn = a | an = 1 = {ai | 0 ≤ i ≤ n − 1}. (10.7)
Theorem 10.6 The cyclic group of order n is an abelian group.
Proof If x, y ∈ Cn then x = ai and y = aj for some integers i and j. Then xy =

ai aj = ai+j = aj+i = aj ai = yx and so the group is abelian.
Consider the following triangle. We shall consider the group of symmetries of the
triangle on the plane. A rigid motion of the plane is an injective function from the
plane to itself that preserves distance. A symmetry of an object in the plane is a rigid
motion of the plane that maps the object to itself.
10.1 Groups 251
Fig. 10.1 Primitive root of unity
(10.8)
Consider rotating the triangle 2π
3 radians counterclockwise. This moves A to B’s
original location, B to C’s original location, and C to A’s original location. We shall
call this action α. Note that α2 is also a symmetry and α3 is the identity map, we
shall denote the identity by ι.
Consider a line from C to midpoint between A and B. Rotating the triangle π
around this line interchanges A and B and leaves C fixed. We shall call this symmetry
β. This symmetry is of order 2.
Consider the element αβ. This element is the same as rotating the triangle π around
the line from B to the midpoint of A and C. In fact, these two elements generate all
the symmetries of the triangle. The group of symmetries of an equilateral triangle
on the plane is a non-abelian group of order 6.
We note that αβ
= βα. We can build a multiplication table of the group generated
by α and β.
∗ ι α α2 β αβ βα
ι ι α α2 β αβ βα
α α α2 ι αβ βα β
α2 α2 ι α βα β αβ (10.9)
β β βα αβ ι α2 α
αβ αβ β βα α ι α2
βα βα αβ β α2 α ι
This group is the first non-trivial example of the dihedral group.
The dihedral group is defined as
D2t = a, b | bt = 1, a2 = 1, aba−1 = b−1 . (10.10)
This group can be seen as the automorphism group of a regular t sided figure in
a Euclidean plane. The element b corresponds to shifting the object 2π
t radians, and
the element a corresponds to flipping the object over.
Definition 10.3 We say that two groups (G, ∗) and (G , ∗ ) are isomorphic if there
is a bijection φ : G → G such that φ(g ∗ h) = φ(g) ∗ φ(h) for all g, h ∈ G.
Example 10.8 Consider the Klein-4 group given in Example 10.3 and the group
given in Example 10.4. The following bijection is an isomorphism:
e → (0, 0)
a → (0, 1)
b → (1, 0)
c → (1, 1).
Therefore, the two groups are isomorphic.
Exercises
1. Prove that (Zn , +) is a group.

2. Prove that the table of a subgroup (H, ∗) of the group (G, ∗) forms a Latin
subsquare of the square formed from the group (G, ∗).
3. Prove that the set of rational numbers with addition is a group.
4. Prove that the set of rational numbers with multiplication is not a group but that
the set of non-zero rational numbers with multiplication is a group.
5. Prove that (Zn − {0}, ∗), where ∗ is multiplication, is a group if and only if p
is prime.
6. Prove that the set of permutations on n letters Sn defined in Sect. 1.4 with the
operation of functional composition is a group.
10.2 Automorphisms of a Design 253
7. Prove that the identity in a group is unique.

8. Determine the group of symmetries of a square, D8 , in the plane. Write its
multiplication table.
9. Prove that the subset of the symmetries of a geometric object in the plane that
fixes a point on the object is a subgroup.
10. Determine the group of symmetries of a cube in three-dimensional Euclidean
space.
11. Prove that (Zn , +) is isomorphic to the cyclic group of order n.
12. Prove that the dihedral group D2t is non-commutative for t ≥ 3.
13. Prove that if G is a cyclic group of prime order then G has no non-trivial
subgroups.
14. Determine the group of symmetries of a regular tetrahedron.
10.2 Automorphisms of a Design
One of the most useful ways of examining the structure of any mathematical object
is to consider the group of automorphisms of an object.
Definition 10.4 An automorphism of an incidence structure D = (P, B, I) is a

bijection ψ : D → D, where ψ is a bijection from the set of points to the set of
points, a bijection from the set of blocks to the set of blocks and
(p, B) ∈ I ⇐⇒ (ψ(p), ψ(B)) ∈ I.
Notice that an automorphism is simply an isomorphism of a design with itself.

An automorphism is also called a collineation.
If φ and ψ are two automorphisms of a design D, then the composition of these
two functions is defined as φ ◦ ψ(x) = φ(ψ(x)).
Theorem 10.7 The set of automorphisms of a design D = (P, L, I) with functional

composition is a group.
Proof It is well known that the composition of bijections is a bijection. We

know (p, ) ∈ I if and only if (φ(p), φ()) ∈ I and (p, ) ∈ I if and only if
(ψ(p), ψ()) ∈ I. Therefore (p, ) ∈ I if and only if (φ(p), φ()) ∈ I if and only
if (ψ(φ(p)), ψ(φ())) ∈ I. Therefore, the composition of two automorphisms is an
automorphism and the set is closed under functional composition.
It is also well known that functional composition is associative.
The identity map serves as the identity of the group. Since if ι(x) = x for all
x ∈ D then ι ◦ φ = φ ◦ ι = φ.
If φ is an isomorphism then define the inverse φ−1 by φ(x) = y if and only if
φ (y) = x. Then φ ◦ φ−1 = ι and there are inverses. Hence, it is a group.
−1

order 2
Table 10.1 The points and p φ(p) L φ(L)

lines under the automorphism
φ A F L1 L3
B G L2 L5
C A L3 L7
D E L4 L1
E C L5 L4
F B L6 L2
G D L7 L6
Recall the following description of the projective plane of order 2 given in

Fig. 10.2.
List the lines L1 , . . . , L7 as follows:
L1 ↔ {A, B, C}
L2 ↔ {C, D, G}
L3 ↔ {A, F, G}
L4 ↔ {C, E, F}
L5 ↔ {A, D, E}
L6 ↔ {B, E, G}
L7 ↔ {B, D, F}
Consider the following automorphism φ (Table 10.1).

As an example, we can compute φ2 as well (Table 10.2).
Consider the automorphism φ as a permutation of the points. It can be written as
(A, F, B, G, D, E, C). In the same way, the automorphism can be viewed as a permu-
tation of the lines as (L1 , L3 , L7 , L6 , L2 , L5 , L4 ). It is clear that both permutations
have order 7, that is, applying them seven times results in the identity. This gives the
following.
Theorem 10.8 The group G7 = φ is a commutative group of order 7.
This group corresponds to a Singer cycle, that is, the points of the plane are viewed
as non-zero elements in the field of order 8, and the automorphisms correspond to
multiplication by a non-zero element.
10.2 Automorphisms of a Design 255
Table 10.2 The points and lines under the automorphism φ2

p φ2 (p) L φ2 (L)
A B L1 L7
B D L2 L4
C F L3 L6
D C L4 L3
E A L5 L1
F G L6 L5
G E L7 L2
Table 10.3 The points and lines under the automorphism ψ1

p ψ1 (p) L ψ1 (L)
A A L1 L5
B D L2 L6
C E L3 L3
D B L4 L4
E C L5 L1
F F L6 L2
G G L7 L7
Definition 10.5 A group G acts transitively on S if given a, b ∈ S there exists a

σ ∈ G with σ(a) = b.
Theorem 10.9 The group G7 acts transitively on the points and lines of the projec-
tive plane of order 2.
Proof This can be seen directly by examining the tables constructed in Exercise 3.
More elegantly, consider points p and q as elements in the field of order 8. Then the
automorphism corresponding to p−1 q sends p to q. In other words, it is a solution
to the equation px = q in the field.
The group G7 is not the full automorphism of the projective plane of order 2.
Notice that in each of the automorphisms φi that there are no fixed points. Let
p1 , . . . , p7 be the points of the plane of order 2. Construct an automorphism τ with
no fixed points by mapping p1 to p2 , p2 to p3 , and p3 to the third point on the line
through p1 and p2 and continuing in this manner. Construct another automorphism
by mapping p1 to p2 , p2 to p3 , and p3 to the third point on the line through p1 and
p3 and continuing in this manner. There are 48 automorphisms constructed in this
manner. We leave verification of this facts as an exercise, see Exercise 4.
Consider the following automorphism which we call ψ1 given in Table 10.3. We
construct this by fixing the points on line L1 , and the lines through point F.

p ψ2 (p) L ψ2 (L)
A A L1 L5
B E L2 L2
C D L3 L1
D G L4 L7
E F L5 L3
F B L6 L4
G G L7 L6

p φ(p) L φ(L)
A A L1 L5
B D L2 L4
C E L3 L3
D C L4 L6
E B L5 L1
F G L6 L7
G F L7 L2
The map ψ1 is an automorphism, and there are 21 automorphisms formed by

fixing a line and the remaining two lines through that point and then switching the
other four points in pairs. We leave this verification as an exercise, see Exercise 1.
construct this by fixing the point A and cycling the three points on line L2 .
There are 56 automorphisms constructed in the same manner as ψ2 . We leave
this verification as an exercise, see Exercise 6.
construct this by fixing a point and permuting the other two points on a line and
cycling the remaining four points.
There are 42 automorphisms constructed in the same manner as ψ3 . We leave
this verification as an exercise, see Exercise 4.
Combining the 21 automorphisms from Exercise 1, 48 automorphisms from Exer-
cise 4, the 56 automorphisms from Exercise 6, the 42 automorphisms from Exercise 8,
and the identity, we have 168 automorphisms. These automorphisms form the full
automorphism group of the projective plane of order 2. We state this as follows.
Theorem 10.10 The automorphism group of the plane of order 2 has order 168.
The automorphism group of the affine plane of order 2 is easier to construct. We

describe it in the following theorem.
10.3 Quasigroups 257
Theorem 10.11 The automorphism group of the affine plane of order 2 has 24
elements.
Proof There are four points in the affine plane of order 2. If σ is a permutation of
the points, then σ induces an automorphism since any two points form a line. That
is, if a, b are points then the line through a and b is mapped to the line through σ(a)
and σ(b). There are 4! = 24 automorphisms.
Exercises
1. Verify that ψ1 is an automorphism and prove that there are 21 automorphisms

formed by fixing a line and the remaining two lines through that point and then
switching the other four points in pairs.
2. Verify that φ is an automorphism.
3. Compute the automorphisms φi for i = 3, 4, 5, 6.
4. Verify that map τ is an automorphism, and that there are 48 automorphisms
constructed in this manner.
5. Verify that ψ2 is an automorphism.
6. Prove that there are 56 automorphisms constructed in the same manner as ψ2 .
7. Verify that ψ3 is an automorphism.
8. Prove that there are 42 automorphisms constructed in the same manner as ψ3 .
10.3 Quasigroups
We have seen that the multiplication table of a group is a Latin square. Yet, it is clear
that not all Latin squares can be seen as the multiplication table of a group nor are
they equivalent in some way to a multiplication table of a group. We shall examine
an algebraic structure which does correspond to a Latin square. We begin with the
definition.
Definition 10.6 A quasigroup (G, ∗) is a set G with an operation ∗ such that
1. (Closure) If a, b ∈ G then a ∗ b ∈ G.
2. (Left cancelation) For all x, y ∈ G, a ∗ x = a ∗ y implies x = y.
3. (Right cancelation) For all x, y ∈ G, x ∗ a = y ∗ a implies x = y.
For example, the following gives a quasigroup:
∗ A B C D E F
A A B C D E F
B C A B F D E
C B CA E F D
D D E F A B C
E F D E CA B
F E F D B C A
Note that there is no identity element nor is it associative. For example, B ∗ (D ∗
E) = B ∗ B = A, whereas (B ∗ D) ∗ E = F ∗ E = C. Also without an identity there
is no way to describe an inverse. Since a group has left and right cancelation, it is
easy to see that a group is a quasigroup but as we have shown that a quasigroup is
not necessarily a group.
Theorem 10.12 A table is a Latin square if and only if it is the multiplication table
of a quasigroup.
Proof Let (Q, ∗) be a quasigroup. Let Ta,b = a ∗ b. We shall show that T is a

Latin square. If Ta,b = Ta,c then a ∗ b = a ∗ c which implies that b = c by left
cancelation. Therefore, each row contains each element exactly once. If Ta,b = Tc,b
then a ∗ b = c ∗ b and by right cancelation a = c; hence, each element appears
exactly once in each column. Then, T is a Latin square.
Let L be a Latin square and define a ∗ b = Ta,b . If a ∗ x = b then Ta,x = b
and since each row has each element exactly once there is a unique x satisfying
this equation. If y ∗ a = b then Ty,a = b and since each column has each element
exactly once there is a unique y satisfying this equation. This means that the operation
∗ on the alphabet of L is a quasigroup.
Let (Q, ∗) be a quasigroup. A set R of Q is a subquasigroup of Q if R ⊆ Q and

R is a quasigroup. In the table given above, the set {A, B, C} is a subquasigroup.
Definition 10.7 A loop is a quasigroup that has an identity element. That is, there
exists an element e such that for all a we have e ∗ a = a ∗ e = a.
Theorem 10.13 Every Latin square is equivalent to a Latin square that is a loop.
Proof Let L be a Latin square and assume that the elements in the square are from
the set {0, 1, 2, . . . , n − 1}. Permute the columns so that the column’s first elements
are in standard order, i.e., 0, 1, 2, . . . , n − 1. Then permute the rows so that the first
elements are in standard order to form the Latin square L . Then 0 is the identity
element of the quasigroup corresponding to L .
10.4 Difference Sets 259
Example 10.9 Let L be the Latin square:

⎛ ⎞
2310
⎜1 0 2 3⎟
⎜ ⎟
⎝3 1 0 2⎠.
0231
Permute the columns to obtain

⎛ ⎞
0123
⎜3 2 1 0⎟
⎜ ⎟
⎝2 0 3 1⎠.
1302
Permute the rows to obtain the Latin square L :

⎛ ⎞
0123
⎜1 3 0 2⎟
⎜ ⎟
⎝2 0 3 1⎠.
3210
Then L is a loop with 0 as the identity.
It is immediate then that if a loop is associative then it is a group.
Exercises
1. Prove that in a loop each element has a left inverse.
2. Construct a loop that is not a group.
3. Let (G, ∗) and (G , ∗ ) be two quasigroups. Define the direct product of quasi-
groups to be (G × G , ∗ ) where (a, b) ∗ (c, d) = (a ∗ c, b ∗ d). Prove that
the direct product of quasigroups is a quasigroup.
10.4 Difference Sets
We shall describe a combinatorial object which uses groups and is useful in con-
structing designs.
Definition 10.8 Let G be a group, where the operation is denoted by + and the
inverse of g is denoted by −g. A difference set in G is a subset D of G such that each
non-identity element g of G can be written in exactly λ different ways of the form
x − y with x, y ∈ D.
Theorem 10.14 Let D be a difference set in a group G with |G| = v and |D| = k.
Let G be the point set P and define the set of blocks B by {D + g | g ∈ G}. Then
(P, B) is a 2 − (v, k, λ) design.
Proof Let a and b be two elements of G. Let c = a − b. We know c appears exactly

λ times in the difference set D. That is, a − b = di − dj for λ pairs of i, j and
di , dj ∈ D. This means that a − di = b − dj . Let gi = a − di . Then a and b are
both elements of D + gi . Since there are exactly λ elements gi , we have proven that
any two points are together incident with λ blocks.
Example 10.10 Let us consider the following example. Consider the group formed
by addition modulo 7. Let D = {1, 2, 4}.
Each non-identity element can be written as a difference from this set exactly
once, namely,
2−1 = 1
4−2 = 2
4−1 = 3
1−4 = 4
2−4 = 5
1 − 2 = 6.
Consider the sets D + g:
{1, 2, 4}, {2, 3, 5}, {3, 4, 6}, {4, 5, 0}, {5, 6, 1}, {6, 0, 2}, {0, 1, 3}.
This is a design we have seen before. Consider the diagram in Fig. 10.3.
Consider the group formed by addition modulo 11. Let D = {1, 3, 4, 5, 9}.
We invite the reader to verify that each non-identity element in D can be written as
a difference of elements of D exactly twice. For example 2 = 3 − 1 and 2 = 5 − 3.

order 2
10.4 Difference Sets 261
Table 10.6 Incidence matrix

0 1 2 3 4 5 6 7 8 9 10
B1 0 1 0 1 1 1 0 0 0 1 0
B2 0 0 1 0 1 1 1 0 0 0 1
B3 1 0 0 1 0 1 1 1 0 0 0
B4 0 1 0 0 1 0 1 1 1 0 0
B5 0 0 1 0 0 1 0 1 1 1 0
B6 0 0 0 1 0 0 1 0 1 1 1
B7 1 0 0 0 1 0 0 1 0 1 1
B8 1 1 0 0 0 1 0 0 1 0 1
B9 1 1 1 0 0 0 1 0 0 1 0
B10 0 1 1 1 0 0 0 1 0 0 1
B11 1 0 1 1 1 0 0 0 1 0 0
The cosets are
B1 = {1, 3, 4, 5, 9}
B2 = {2, 4, 5, 6, 10}
B3 = {3, 5, 6, 7, 0}
B4 = {4, 6, 7, 8, 1}
B5 = {5, 7, 8, 9, 2}
B6 = {6, 8, 9, 10, 3}
B7 = {7, 9, 10, 0, 4}
B8 = {8, 10, 0, 1, 5}
B9 = {9, 0, 1, 2, 6}
B10 = {10, 1, 2, 3, 7}
B11 = {0, 2, 3, 4, 8}
We shall show the incidence matrix in Table 10.6.

We notice that each block has five points, and each point is on five blocks and
that any two points are incident with exactly two blocks. Theorem 10.14 gives that
it is an (11, 5, 2) design. Therefore, we have the following.
Theorem 10.15 The design formed from the difference set {1, 3, 4, 5, 9} in the group
formed from addition modulo 11 is a biplane of order 3.
These two constructions generalize. We shall state the theorem but the proof
requires more number theory than we have assumed. First, we need a definition. A
non-zero element a of Zp is a square if there exists a b ∈ Zp , with a = b2 .
Theorem 10.16 Let p ≡ 3 (mod 4), with p = 4t − 1. Then the set of squares in Zp
is a difference set and produces a 2 − (4t − 1, 2t − 1, t − 2) design.
For proof of the theorem, see [44].
Exercises
1. Verify that any two elements are in exactly one set together in Example 10.10.
2. Verify that the diagram in Fig. 10.3 corresponds to the blocks given in Exam-
ple 10.10.
3. Use the incidence matrix to draw a diagram of the biplane of order 3.
Codes
11
11.1 Introduction
Algebraic coding theory arose in the last half of the twentieth century to answer
applications in the field of electronic communication. Specifically, the idea is to
transmit information over a channel such that, when it is received, whatever errors
that were made can be corrected with a high degree of certainty. Since its initial
study of this application, coding theory has grown into an interesting branch of
both pure and applied mathematics. From the very beginning of the study of coding
theory, numerous interesting connections between codes, combinatorics, and finite
geometry were found. Codes were used to study designs, and designs were used to
construct some of the best codes. We shall give some examples of the connection
between codes and combinatorics in this chapter. The reader interested in an in-
depth description of coding theory that can consult the standard texts of coding
theory, namely, MacWilliams and Sloane [62] or Huffman and Pless [49]. For more
elementary introductions to this interesting branch of mathematics, see the text by Hill
[48]. For an advanced description of algebraic coding theory over finite commutative
rings, see [29]. Coding theory is related to, but distinct from, cryptography which is
the study of secret communication, namely, how to communicate so that no one can
intercept your message. In coding theory, the idea is to communicate so that it can
be decoded correctly, whereas in cryptography the idea is to keep the information
secret.
From our point of view, we will not be overly concerned with applications of
coding theory to electronic communication but rather with the study of codes as
algebraic structures and their application to combinatorics and finite geometry. We
take the view of coding theory as a branch of pure mathematics alongside algebra,
number theory, and combinatorics. Much of the research done in coding theory
by mathematicians is far removed from its application. However, the application of
this theory benefits electronic communication greatly and moreover gives fascinating
264 11 Codes
open problems to mathematicians. In many ways, coding theory is in an ideal situation

as it is studied for its own sake and benefits from its close relationship to other
branches of mathematics and with its applications in electronic communication.
We shall begin with a description of the origin of coding theory and its basic
principles. For a machine, there are only two pieces of information: on or off, yes or
no, or from a mathematical point of view 1 or 0. All the information that is transmitted
over the phone, stored in your computer, or bounced off satellites is simply a set of
ones and zeros. It is in the way that the machines interpret these ones and zeros that
give them meaning to the users of these machines.
In the later half of the twentieth century, it was realized that techniques must be
developed to insure that the information could be communicated accurately. Shannon
wrote the landmark paper [78] in 1948 and gave birth to the subjects of Information
Theory and Coding Theory. The early work of Shannon and Hamming [47] set the
stage for a tremendous amount of research that followed. This research was very
important in the revolution of the information age that occurred in the late twentieth
century.
Consider the following situation. Someone is standing across a crowded room
and you want to convey the message “yes” to this person. If you simply said “yes”
once, there is a very good chance that he will not hear you and may assume that you
said no. There are two things that you will usually do in this situation. One is to say
yes a few times, so that he will assume you said the answer he heard most often.
In other words, if you say it four times and twice he distinctly heard yes, and once
he could not tell and once he thought you said no, he would still assume the correct
answer is yes. The other thing is that he would nod his head and say yes at the same
time. In that manner, if the person thought he heard no but saw your head nodding
in the affirmative he would know that it did not make sense and so motion to you to
repeat.
In a simplistic manner, these are two of the most commonly used methods to
insure correct information. The first is the repetition of information. The second is
a check to be sure that the message makes sense. The first technique hopes that the
correct information can be gleaned from the message and the second that a message
that is not received correctly can be discarded and new information will be requested.
Techniques not much more complicated than this were used at the beginning of
electronic communication. For example, instead of sending 1, the message 11111
might be sent. If it was received 11011 it would still be decoded correctly as 1. The
trouble with this technique is that it requires a good deal of information, in this case
5 times as much, to be sent over the communication channel.
As an example of the second technique, where some other information is sent as
a check, we will show how a parity check is used. Assume the desired message was
111; in this case, the message 1111 is sent and if 110 was to be sent then 1100 is sent.
In other words, you put a 1 at the end if there are oddly many ones and a 0 at the end
if there are evenly many ones. More mathematically, the last coordinate is simply
the sum of the other coordinates using the addition of the finite field of order 2. This
is known as a parity check. If the message 1110 is received, then you know that it
is not a possible message and so it is rejected. Although you can tell that an error
11.2 Basics of Coding Theory 265
has occurred, you cannot tell in which place the error occurred. Any of 1111, 1100,
1010, and 0110 could have been the desired message. Being able to tell if an error is
made is known as error detecting.
Amazingly, the techniques which are now in use not only can tell if an error
has been made but can determine where it was made and correct it! The mathemat-
ics is based on nineteenth-century linear algebra over finite fields and ideas from
combinatorics and finite geometry.
11.2 Basics of Coding Theory
We shall now give the basic definitions in a mathematical setting.

A code of length n over an alphabet A is simply a subset of
An = {(a1 , a2 , . . . , an ) | ai ∈ A}.
One can think of this as all possible words that can be spelled with the letters of
the alphabet. Most usually, the alphabet used is either a finite field or the finite ring
Zm . However, in the twenty-first century, the collection of acceptable alphabets has
grown significantly to include all finite Frobenius rings and even some infinite rings
as well. An element of a code is called a codeword. For most applications in the field
of electronic communications, the alphabet is the binary field F2 = {0, 1}.
To the ambient space An , we attach a distance. The Hamming distance between
any two codewords is defined to be the number of places where they differ, specifically
d(v, w) = |{i | vi = wi }|.
For example, the distance between (0, 0, 1, 1, 0) and (0, 0, 0, 0, 1) is 3 because they
differ in three places, namely, the last three coordinates.
Theorem 11.1 The Hamming distance is a metric on the space An . That is, it
satisfies the following three conditions:
1. We have d(v, w) = 0 if and only if v = w and d(v, w) ≥ 0 for all v and w.

2. For all v and w, d(v, w) = d(w, v).
3. For all u, v, w we have d(u, w) ≤ d(u, v) + d(v, w). This is known as the triangle
inequality.
Proof To prove the first statement, we simply notice that the function is determined
as a cardinality which is always greater than or equal to 0. If it is 0, then the vectors
agree in all coordinates and so are equal.
The second statement follows from the fact that {i | vi = wi } = {i | wi = vi }.
Consider the three vectors u, v, w. We can change u into v by changing d(u, v)
coordinates. We can change v into w by changing d(v, w) coordinates. Therefore,
the maximum number of coordinates in which u and w differ is d(u, v) + d(v, w).
This gives the third statement.

266 11 Codes
Given a code C ⊆ An , the minimum Hamming distance dC is the smallest dis-

tance between any two distinct vectors, that is,
dC = min{d(v, w) | v, w ∈ C, v = w}.
In general, we look at codes that have some sort of algebraic structure, rather
than those that are simply a collection of vectors. We define a linear code over the
alphabet F, where F is a field, to be a vector space over the field F. In other words, C
is a linear code if the following conditions hold:
• If v, w ∈ C then v + w ∈ C.
• If v ∈ C and α ∈ F then αv ∈ C.
Example 11.1 Let p be a prime and n a non-zero integer. Then the code of length
n consisting of the codewords
{(0, 0, . . . , 0), (1, 1, . . . , 1), . . . , (p − 1, p − 1, . . . , p − 1)}
is a linear code over Fp with p elements. This code is known as the repetition code.
Example 11.2 Consider the binary code C = {(0, 0, 0), (1, 1, 0), (0, 1, 0)}. This
code is not a linear code since (1, 1, 0) + (0, 1, 0) = (1, 0, 0) ∈
/ C.
The weight of a vector is the number of non-zero elements in a vector. The weight
of vector v is denoted by wt(v). The minimum weight of a code is the smallest of
all non-zero weights in a code.
Theorem 11.2 For a linear code, the minimum weight of the code is the minimum
distance.
Proof If v and w are vectors in a linear code C, then d(v, w) = d(v − w, 0) =

wt(v − w) where 0 is the all 0 vector. So if d is the minimum distance of a code
then there is a vector of weight d in the code. If there is a vector of weight d, then
the distance between that vector and the all-zero vector is d. Therefore, for a linear
code the minimum distance and the minimum weight are equal.

Lemma 11.1 Let V be a vector space of dimension k over the finite field Fq , then
|V| = qk .
Proof Let v1 , v2 , . . . , vk be the basis for V. Then since these vectors are linearly
independent, each of the following sums
α1 v1 + α2 v2 + . . . , αk vk
must be distinct. Therefore, there are q choices for each αi and hence qk vectors
in V.

11.2 Basics of Coding Theory 267
This leads immediately to the following.
Theorem 11.3 If C is a linear code over the field Fq , then |C| = qk for some k.
Proof By definition, a linear code is a vector space. Then by Lemma 11.1, we have
that |C| = qk where k is the dimension of C as a vector space.

Definition 11.1 Two codes C and C of length n over Fq are said to be permutation
equivalent if C can be obtained from C by a permutation of the coordinates.
Since any linear code is a vector space of dimension k in Fnq , it has a basis of
k vectors. Using the standard technique of Gaussian elimination, we have that any
linear code over a finite field Fq is permutation equivalent to a code that has a
generator matrix of the form:
(Ik | A),
where k is the dimension of the code and Ik is the k by k identity matrix.
Example 11.3 Consider the code generated by vectors (1, 1, 1, 1), (1, 1, 0, 0), and
(0, 0, 0, 1). Applying Gaussian elimination, we get the following matrix:
⎛ ⎞
1100
⎝0 0 1 0⎠.
0001
Then using the permutation (2, 4, 3) on the coordinates, we obtain

⎛ ⎞
1001
⎝0 1 0 0⎠,
0010
which is of the form (Ik | A). This code has cardinality 8.
If a code is of length n, has M vectors, and minimum distance d, then the code
is said to be an (n, M, d) code. If the code is linear and its dimension is k, then it
is said to be a [n, k, d] code. Of course, in this case we have that M = |Fq |k . If the
minimum weight is not known, then we may simply call it an [n, k] code.
The sphere of radius r around a vector v ∈ C, where C is a code, consisting of
all vectors in Fnq that are Hamming distance less than or equal to r. When a vector
w is received, it is decoded to the vector v ∈ C that is closest to it (if such a vector
exists). This notion of nearest neighbor decoding leads to the following theorem.
Theorem 11.4 Let C be an [n, k, d] code, then C can detect d − 1 errors. If d =

2t + 1 then C can correct t errors.
268 11 Codes
Proof Given a vector v ∈ C, if d − 1 or less errors are made, then the resulting
vector cannot be in C since the minimum distance of C is d. Hence, the code can
correct up to d − 1 errors.
If d = 2t + 1 and t errors are made to a vector v ∈ C, then by the triangle inequal-
ity, any other vector in C must have distance at least t + 1 from the resultant vector.
Therefore, there is a unique vector that is closest to this vector, namely, v, and the
errors are corrected.

Example 11.4 Consider the code C of length n over Fq generated by (1, 1, . . . , 1).
The minimum distance of this code is n; hence, it can detect n − 1 errors and correct
n−1
2 errors.
Exercises
1. Prove that there exists a linear [n, k] code for all k with 1 ≤ k ≤ n.
2. Produce the generating matrix in standard form by row reducing the following
over F2 :
⎛ ⎞
1101
⎜0 0 1 1⎟
⎜ ⎟
⎜1 1 1 1⎟.
⎜ ⎟
⎝1 0 0 0⎠
0110
3. Produce the generating matrix in standard form by row reducing the following
over F3 :
⎛ ⎞
1221
⎜2 1 1 1⎟
⎜ ⎟
⎝1 1 1 2⎠.
0101
4. Find all vectors in F42 that are at distance 2 or less from the vector (1, 0, 0, 1).
5. Produce a non-linear code whose minimum distance is not equal to its minimum
weight.
6. Produce a non-linear code over F2 that can correct two errors.
7. Produce a linear code over F5 that can correct two errors.
11.3 Orthogonal Codes
In this section, we shall investigate the dual of code with respect to the Euclidean
inner product. This dual code has numerous important properties and works in tandem
with the code in terms of correcting errors.
The space Fn q has an inner product attached to it. It is

[v, w] = vi wi , (11.1)
11.3 Orthogonal Codes 269
where the multiplication and addition are done in Fq .
Theorem 11.5 For u, v, w ∈ Fn

q and α, β ∈ Fq we have
1. [u, v] = [v, u];

2. [αu, βv] = αβ[u, v];
3. [u + v, w] = [u, w] + [v, w].

v] = ui vi = vi ui = [v, u].
Proof For the first statement, we have [u,
For the second statement, [αu, βv] = αu i βvi = αβ ui vi = αβ[u, v].
For the third statement, [u + v, w] = (ui + vi )wi = (ui wi + vi wi ) =
[u, w] + [v, w].

For any code C over Fq we can define the orthogonal of the code by
C⊥ = {w | [w, v] = 0 for all v ∈ C}. (11.2)
Theorem 11.6 If C is a code over Fq , then C⊥ is a linear code.
Proof If u is a vector in C, v and w are vectors in C⊥ and α ∈ Fq , then by Theo-

rem 11.5 we have
[u, v + w] = [u, v] + [u, w] = 0 + 0 = 0
and
[u, αv] = α[u, v] = α0 = 0.
So both v + w and αv are in C⊥ , and it is a linear vector space.

Note that we have not assumed that the code C is linear. What we have proven
is that the orthogonal code is linear even if C is not. We can also say the following
about the orthogonal of a non-linear code.
Theorem 11.7 If C is a non-linear code, then C⊥ = C⊥ . That is, the orthogonal
of a non-linear code is equal to the orthogonal of the linear code that it generates.
Proof Let v, u ∈ C. Then w ∈ C⊥ if and only if [w, v] = [w, u] = 0. This gives

that [w, αv + βu] = α[w, v] + β[w, u] = 0 + 0 = 0. This gives w ∈ C⊥ . There-
fore, C⊥ ⊆ C⊥ .
We have C ⊆ C which gives C⊥ ⊆ C⊥ . Then we have that C⊥ = C⊥ .
Theorem 11.8 If C is a linear [n, k] code over Fq , then C⊥ is a linear [n, n − k]

code.
270 11 Codes
Proof Consider the generator matrix of an [n, k] code of the form (Ik | A). To
construct a vector in C⊥ , make it anything on the n − k coordinate corresponding to
the matrix A. Call this length n − k vector v. Then to make it orthogonal, place in
the i-th coordinate, for 1 ≤ i ≤ k, the value −[v, Ai ], where Ai is the i-th row of A.
Then this vector is orthogonal to every row of the generator matrix and therefore in
C⊥ . Since there are n − k degrees of freedom in making the vector, the dimension
of C⊥ is n − k.

Example 11.5 Let C be the binary code of length n, generated by (1, 1, . . . , 1). The
code C has dimension 1 and minimum weight n. The dual code C⊥ has dimension
n − 1 and consists of all vectors of length n with even weight. This code is denoted
by En .
Theorem 11.9 If C is a code generated by (Ik | A), then C⊥ is generated by

(−AT | In−k ).
Proof By Theorem 11.8, we know that C⊥ must have dimension n − k. A code

generated by (−AT | In−k ) has dimension n − k. The only thing that must be
shown then is that the vectors in the code generated by (−AT | In−k ) are orthogonal
to the vectors in C. The inner product of the i-th row of (Ik | A) with the j-th row
of (−AT | In−k ) gives Aij − Aij = 0. Therefore, the two rows are orthogonal, and
(−AT | In−k ) generates C⊥ .

Example 11.6 Consider the code C over F5 generated by the matrix:

⎛ ⎞
10012
⎝0 1 0 3 4⎠.
00101
Then C⊥ has generator matrix:

−1 −3 0 1 0 42010
= .
−2 −4 −1 0 1 31401
The inner product of the first row of the first matrix with the second row of the second
matrix gives 2 − 2 = 0. Notice that A1,2 = 2.
Definition 11.2 We say that a code C is self-orthogonal if C ⊆ C⊥ and we say that

it is self-dual if C = C⊥ .
Example 11.7 The code C = {(0, 0), (1, 1)} satisfies C = C⊥ and is a self-dual
code. The code D = {(0, 0, 0), (1, 1, 0)} satisfies D ⊆ D⊥ and is a self-orthogonal
code.
Theorem 11.10 If C is a self-dual code over Fq of length n, then n must be even.

11.3 Orthogonal Codes 271
Proof If C is a self-dual code, then dim(C) + dim(C⊥ ) = n and dim(C) =

dim(C⊥ ) since C = C⊥ . This gives 2dim(C) = n and so n must be even.

We define C × D = {(v, w) | v ∈ C, w ∈ D}.
Lemma 11.2 If C is a self-dual code of length n and D is a self-dual code of length

m, then C × D is a self-dual code of length n + m.
n m
Proof We have C = C⊥ , D = D⊥ and |C| = q 2 , |D| = q 2 . Then |C × D| =
n+m
q 2 . Therefore, the code has the right cardinality. We only need to show self-
orthogonality. Given (v, w), (v , w ) ∈ C × D, then [(v, w), (v , w )] = [v, v ] +
[w, w ] = 0 + 0 = 0. Therefore, it is self-orthogonal and of the right cardinality so
the code is self-dual.

Theorem 11.11 If p ≡ 1 (mod 4) is a prime, then there exist self-dual codes of

length n for all even n over Fp .
Proof If p ≡ 1 (mod 4), then we know −1 is a square. That is, there exists α
such that α2 = −1. We have [(1, α), (1, α)] = 1 + α2 = 0. Therefore, the code C
generated by (1, α) is self-orthogonal. The code C has p elements, and so |C|2 =
p2 = |F2p | and so the code is self-dual.
Then, by Lemma 11.2, we can take the cross product of the code with itself t
times to get a self-dual code for all lengths of the form 2t.

Theorem 11.12 If p ≡ 3 (mod 4) is a prime, then there exist self-dual codes of

length n for all n ≡ 0 (mod 4) over Fp .
Proof If p ≡ 3 (mod 4), then we know that there exists α, β ∈ Fp with α2 + β2 =

−1. The code generated by the following matrix

10 α β
0 1 −β α
is a self-orthogonal [4, 2] code and is therefore self-dual.

Then by Lemma 11.2, we can take the cross product of the code with itself t times
to get a self-dual code for all lengths of the form 4t.

Exercises
1. Use the inner product to prove that the number of [n, k] codes over Fq is equal
to the number of [n, n − k] codes over Fq .
2. Take the code over F3 generated by the vectors (1, 2, 1, 2), (1, 1, 1, 1), and
(0, 0, 1, 2). Give the generator matrix of this code in standard form. Give the
generator matrix of its orthogonal.
272 11 Codes
3. Prove that if k > n2 then there is no self-orthogonal [n, k] code.

4. Let v be a vector in Fn
q . Determine the number of vectors w that are orthogonal
to v. Prove your answer.
5. Assume v and w are binary vectors where wt(v) ≡ wt(w) ≡ 0 (mod 4) with
[v, w] = 0. Prove that wt(v + w) ≡ 0 (mod 4).
6. Prove that there are no self-orthogonal vectors of length 1 over Fq .
11.4 Syndrome Decoding
In this section, we shall show how many of the ideas of finite geometry and linear
algebra are used to actually correct errors. Instead of starting with a generator matrix
for a code, we shall start with a generator matrix for its orthogonal. Assume H =
(In−k |B) generates the [n, n − k] code C⊥ . It follows immediately that v ∈ C if
and only if HvT = 0. This matrix H is known as the parity check matrix of C.
If you think of the linear transformation T : Fn n
q → Fq , given by T (v) = Hv ,
T
then the code C is the kernel of the linear transformation. The vectors in the code
C are precisely the vectors that carry information that we want to send across an
electronic channel. We define a coset of C to be
(C + w) = {v + w | v ∈ C}.
Lemma 11.3 Let C be a linear code in Fn n

q . Let w be any vector in Fq . Then |C| =
|(C + w)|.
Proof Define the function ψ : C → (C + w) by ψ(v) = v + w. We note that if

ψ(v) = ψ(u) then v + w = u + w which gives v = u. Therefore, the map is injec-
tive. The map is surjective by construction since every element in (C + w) is of the
form v + w for some v ∈ C. Then ψ is a bijection and |C| = |(C + w)|.

Lemma 11.4 Let C be a linear code in Fn n

q . Let w, u be vectors in Fq . Then either
(C + w) = (C + u) or (C + w) ∩ (C + u) = ∅.
Proof Assume there exists a vector z ∈ (C + w) ∩ (C + u). Then z = v + w and

z = v + u for some v, v ∈ C. This gives v + w = v + u and v − v = u − w.
Therefore, u − w = c ∈ C.
If x ∈ (C + w), then x = v + w for some v ∈ C. Then x = v + u − c =

(v − c) + u and therefore (C + w) ⊆ (C + u). Since by Lemma 11.3, we have
that their cardinalities are identical, this gives (C + w) = (C + u). Therefore, if the
intersection is not empty then they are identical.

11.4 Syndrome Decoding 273
Theorem 11.13 Let C be a linear [n, k] code in Fn

q . Then
qn−k

−1
Fn
q = (C + wi ),
i=0
where w0 = 0 and the wi are distinct vectors in Fn

q.
Proof By Lemma 11.4, the cosets are disjoint. Moreover, for any w ∈ Fn
q , we have
w ∈ (C + w). Therefore, the cosets form a partition of the space.

Lemmas 11.3, 11.4, and Theorem 11.13 are known collectively as LaGrange’s
theorem.
In this scenario, the wi are known as the coset leaders. We can choose any vectors
at all; however, in terms of the application of coding theory, we generally choose the
wi that have the smallest Hamming weights.
Example 11.8 Consider the binary code C with parity check matrix:

10101
.
01010
The code C is then a [5, 3] code. We shall write the ambient space as the code in the
first column, followed by the non-trivial cosets of the code in the next three columns
using the top vector in the column as the coset leader.
(0, 0, 0, 0, 0) (1, 0, 0, 0, 0) (0, 1, 0, 0, 0) (0, 0, 0, 1, 1)

(1, 0, 1, 0, 0) (0, 0, 1, 0, 0) (1, 1, 1, 0, 0) (1, 0, 1, 1, 1)
(0, 1, 0, 1, 0) (1, 1, 0, 1, 0) (0, 0, 0, 1, 0) (0, 1, 0, 0, 1)
(1, 0, 0, 0, 1) (0, 0, 0, 0, 1) (1, 1, 0, 0, 1) (1, 0, 0, 1, 0)
(0, 0, 1, 0, 1) (1, 0, 1, 0, 1) (0, 1, 1, 0, 1) (0, 0, 1, 1, 0)
(1, 1, 0, 1, 1) (0, 1, 0, 1, 1) (1, 0, 0, 1, 1) (1, 1, 0, 0, 0)
(0, 1, 1, 1, 1) (1, 1, 1, 1, 1) (0, 0, 1, 1, 1) (0, 1, 1, 0, 0)
(1, 1, 1, 1, 0) (0, 1, 1, 1, 0) (1, 0, 1, 1, 0) (1, 1, 1, 0, 1)
Notice for the first two cosets, we chose as coset leaders vectors with weight 1, the
smallest weight possible for a non-trivial coset leader. However, by the fourth coset
we had no more weight 1 vectors at our disposal, so we had to choose a vector with
higher weight.
Definition 11.3 Let C be a code in Fn

q with parity check matrix H. Then the syn-
drome of a vector v ∈ Fn
q is S(v) = Hv T.
Theorem 11.14 Let C be a code in Fn n

q with parity check matrix H and let v, w ∈ Fq .
We have S(v) = S(w) if and only if v and w are in the same coset of C.
274 11 Codes
Proof If S(v) = S(w), then HvT = HwT which implies H(vT − wT ) = 0. This
gives that v − w ∈ C. Therefore, v ∈ (C + w) and w ∈ (C + w); therefore, they are
in the same coset of C.
If u and x are both in (C + w), then u = c + w and x = c + w, where c, c ∈ C.
Then
S(u) = S(c + w) = H(cT + wT ) = HcT + HwT = 0 + HwT = S(w)
and
S(x) = S(c + w) = H((c )T + wT ) = H(c )T + HwT = 0 + HwT = S(w).

Assume a vector v is transmitted from person A to person B. Along the channel

errors are introduced into the vector, so that w = v + e is received, where e is the
error vector. Then person B computes S(w) which is equal to S(e). Then to decode
the vector, we take w − e to get v as was originally sent. The difficulty is that we may
not be sure that e represents the actual errors that occurred. In essence, we look at the
array of elements as was given in Example 11.8. We assume that the error vectors
are the coset leaders which appear at the top of each column. For example, assume
(0, 1, 1, 1, 0) is received, then it is located at the bottom of column 2. We assume
that the error vector is (1, 0, 0, 0, 0) and we decode the vector to (1, 1, 1, 1, 0).
In general, there is no reason to list all of the elements in the array. Instead, as
long as we can determine what the coset leaders are (noticing there is not a unique
choice), then we only need to compute the syndrome. In the next section, we shall
give a specific example of how this can be done.
Exercises
1. Find all vectors in the binary code with parity check matrix
⎛ ⎞
10010
⎝0 1 0 1 1⎠.
00001
Write the space F52 as the union of cosets of this code. Use these cosets to decode
the vector (1, 1, 1, 1, 1).
2. Using the array in Example 11.8, decode the following vectors: (1, 1, 1, 0, 0),
(0, 0, 1, 1, 1), and (1, 1, 1, 0, 1).
3. Prove that each possible syndrome occurs.
4. Let C be the binary code generated by (10110) and (01011). Produce the standard
array for C with error vectors: 00000, 10000, 01000, 00100, 00010, 00001, 11000,
and 10001. Use the array to decode the following vectors: 11110, 01111, 01110,
and 11010.
11.5 The Binary Hamming Code and the Projective Plane of Order 2 275
11.5 The Binary Hamming Code and the Projective Plane of

Order 2
In this section, we shall show a connection between a finite projective plane and
algebraic coding theory. That is, we shall examine the space F72 . There are 27 = 128
such vectors, since there are two choices for each coordinate.
The following matrix is the parity matrix for the code. The only acceptable
codewords are those that have an inner product of 0 with each of the three rows.
Recall the numbers 1 to 7 in base 2: 1, 10, 11, 100, 101, 110, 111. We notice that the
columns are formed by simply writing these numbers as vectors in F32 .
Therefore, we let
⎛ ⎞
0001111
H = ⎝0 1 1 0 0 1 1⎠. (11.3)
1010101
Consider a vector (1, 1, 1, 1, 1, 1, 0). The inner products are
[(1, 1, 1, 1, 1, 1, 0), (0, 0, 0, 1, 1, 1, 1)] = 1
[(1, 1, 1, 1, 1, 1, 0), (0, 1, 1, 0, 0, 1, 1)] = 1
[(1, 1, 1, 1, 1, 1, 0), (1, 0, 1, 0, 1, 0, 1)] = 1

so the vector is not an acceptable codeword.
Consider the vector (1, 1, 1, 0, 0, 0, 0). The inner products are
[(1, 1, 1, 0, 0, 0, 0), (0, 0, 0, 1, 1, 1, 1)] = 0
[(1, 1, 1, 0, 0, 0, 0), (0, 1, 1, 0, 0, 1, 1)] = 0
[(1, 1, 1, 0, 0, 0, 0), (1, 0, 1, 0, 1, 0, 1)] = 0

so the vector is an acceptable codeword.
We see that since the parity check matrix has dimension 3, this gives that the code
has dimension 7 − 3 = 4, and hence there are 24 = 16 vectors in the code. They are
the following:
(0, 0, 0, 0, 0, 0, 0) (1, 1, 1, 1, 1, 1, 1)
(1, 1, 1, 0, 0, 0, 0) (0, 0, 0, 1, 1, 1, 1)
(0, 0, 1, 1, 0, 0, 1) (1, 1, 0, 0, 1, 1, 0)
(1, 0, 0, 0, 0, 1, 1) (0, 1, 1, 1, 1, 0, 0)
276 11 Codes
(0, 1, 0, 1, 0, 1, 0) (1, 0, 1, 0, 1, 0, 1)
(0, 1, 0, 0, 1, 0, 1) (1, 0, 1, 1, 0, 1, 0)
(0, 0, 1, 0, 1, 1, 0) (1, 1, 0, 1, 0, 0, 1)
(1, 0, 0, 1, 1, 0, 0) (0, 1, 1, 0, 0, 1, 1).

Suppose you want to send the message 1110000. If it is received as 1110000 then
it is assumed to be correct. This is not because the recipient sees it is in the code but
rather because the message has inner product 0 with each of the rows of the parity
check matrix. In general, you do not want to search through an entire code, because
usually the codes contain an extremely large amount of vectors and searching takes
a large amount of computation. Checking the parity check matrix takes very little
computation. Assume that one coordinate was changed and it became 1110010. Let
us check the parity matrix for this vector.
[1110010, 0001111] = 1
[1110010, 0110011] = 1
[1110010, 1010101] = 0.
The vector 110 formed from these three operations is known as the syndrome.
Now we read this as a number in base 2. It is 6. So we know that the sixth place is
where the error was made and we change 1110010 to 1110000 which is read correctly.
The mathematical technique saw that an error was made, determined where it was
made and corrected it. To continue with the example, assume 1110000 was sent but
0110000 was read. The parity vector is
[0110000, 0001111] = 0
[0110000, 0110011] = 0
[0110000, 1010101] = 1.
The 001 is 1 so it is the first coordinate that was changed.
The reason it works is essentially geometry. Consider a floor that is made up of
square tiles. Put a mark in the center of each tile. Now pick a mark and stand on it. If
you move a distance that is less than the distance of half the diagonal of the square,
then you are still in that square. An observer can easily determine your original mark.
The reason the code works so nicely is for exactly the same reason. On the floor
no two squares overlap and a very similar thing happens for the code.
We have our seven-dimensional space which we already know has 27 = 128
elements (points). We have 16 points in our code. For each vector in the Hamming
code, there are seven vectors that are distance 1 from it (formed by changing each of
the seven coordinates) and 1 vector that is distance 0 from it (itself). So in each tile
there are eight vectors in it and the vector from the Hamming code is in the center.
There are 16 vectors in the code each with 8 in their tile so there are 16 · 8 = 128
vectors represented.
We have shown that each point in the space is either in the code or it is distance
from a unique point in the code. Such a code is known as a perfect code and is quite
rare. They will be discussed in the next section.
The Hamming code we have described has 16 possible messages and if zero
mistakes or one mistake is made the code will not only detect it but correct it. If two
mistakes are made we will not be so lucky. We will still be able to detect that there is
an error since it will not have all zeros when applying the inner product to the parity
check matrix but we will not be able to correct it. So we say that it can correct one
error and detect two errors. If you try to correct something that has two errors, then
you will correct it to the wrong message.
For example, say you want to send
1111111
and two errors are made sending

1111100.
[1111100, 0001111] = 0
[1111100, 0110011] = 0
[1111100, 1010101] = 1.
It corrects the message to 0111100, which is of course the wrong one. The code
will work remarkably well if we have a high probability that either no or at most one
mistake is made. If the probability is too great, then we need to build a bigger code
with a higher minimum distance between vectors.
It may be surprising that the Hamming code was partially discovered by gamblers
long before it was of use to electronic communication. Often gamblers play the
following kind of game. Pick seven Sunday football games that interest you and pick
the winner in each of the seven. Bet one dollar and if you get all seven right then you
will win 50 dollars and if you get only one wrong then you win 20 dollars. To the
person setting up this game, it seems like a fair bet from his standpoint. The gamblers
realized that they could pick their bets according to the vectors in the Hamming code
and be sure of either winning all 7 or at least 6 of 7. Betting 16 dollars would insure
at least a win of 20 dollars. For example, if there are seven games, with ai playing
bi in the i-th game, then if there is a 1 in the i-th coordinate then you say that ai will
and if there is a 0 you say that bi will win. Each vector represents a different 1 dollar
bet. Of course, soon enough the bookmakers understood what they were doing and
changed the rules.
278 11 Codes
We shall show how to visualize the Hamming code geometrically in a very dif-
ferent way than was described above. Look at the following representation of the
projective plane of order 2.
The lines of the plane are

{1, 2, 3}
{3, 4, 7}
{1, 6, 7}
{2, 4, 6}
{2, 5, 7}
{3, 5, 6}
{1, 4, 5}
The numbers on the points will also indicate a position in a coordinate. If you take
the first line and put a 1 for each position on that line you get 1110000 and for the
second you get 0011001. This is known as the characteristic function of a line, that
is, label the coordinates v1 , v2 , . . . , v7 and then the define the characteristic function
as

1 if (vi , L) ∈ I,
χL =
0 if (vi , L) ∈
/ I.
Notice that if you do it for each of the lines you will get the 7 vectors in the
Hamming code that have three ones in them. If you put a 0 in each position on the
line and a 1 elsewhere you get the 7 vectors that have 4 ones. Assume you want to
send the message 1110000 and a mistake is made so that you receive 1100000. All
you need do is find the closest line or complement to a line (meaning the four points
that are not on a specific line). That is, by changing one element you can always get
to a line or a complement to a line or to the all 0 vector or the all 1 vector.
Recall that the complements of the lines in the projective plane of order 2 are
the hyperovals of the plane. This means that the vectors in the code are the zero
vector, the all one vector, the characteristic functions of lines, and the characteristic
functions of hyperovals.
The geometry of the plane tells you exactly how to decode this code. Assume
you want to send the message 1110000 and a mistake is made so that you receive
1100000. All you need do is find the closest line or hyperoval. That is, by changing
one element you can always get to a line or a complement to a line. The geometry
of the plane and the algebra in the code are actually doing the exact same thing.
Another interesting connection is that if you take the subspace of Fn 2 that is
generated by the characteristic functions of the lines of the planes then you get the
space that consists of the 16 vectors of the Hamming code.
From a broad perspective, these two seemingly different entities are in reality the
same thing.
We shall describe why this works. The matrix H given in 11.3 generates a dimen-
sion 3 subspace of F72 , since the three rows are linearly independent. Its orthogonal,
the code C, then has dimension 4. We know then that C has 24 = 16 vectors and
we have displayed these vectors. Hence, H generates C⊥ . Consider the vectors ψi
defined by

1 if i = j,
(ψi )j = (11.4)
0 if i = j.
We know that none of the seven vectors ψi is in the code C.
Let w be a vector not in C. Then if x is in w + C we have x = w + c1 where
c1 ∈ C. Then H(w + c1 )T = HwT + HcT1 = HwT . Hence, for any vector in w + C
its product with H is the same.
Each coset of C has cardinality 16, and the cosets ψi + C are distinct for distinct
i. So the entire space F72 = ∪(ψi + C).
Now a vector c in the code is altered by changing the i-th coordinate then the new
vector is c + ψi . Then H(c + ψi ) is the i-th column of H which we have made the
binary representation of i. Hence, we know that if we get the i-th column of H by
taking the inner product with the three rows of H then we know that it was the i-th
coordinate of c that was changed.
Exercises
1. Produce the parity check matrix for the Hamming code of length 15. As a hint,
try writing out the numbers base 2 from 1 to 15. It might take a while to write
out all the vectors in this code, there are 2048 of them! But it is easy to find a few
vectors in the code and change a coordinate and find which one it is.
2. Use the parity check matrix from the previous exercise to decode the vectors:
111001101101101 and 001100110011010.
280 11 Codes
11.6 Projective Geometry and Coding Theory
There is a natural connection between codes and projective geometry. It can be used
to describe the generalized Hamming codes and several other important results. We
shall begin with a description of the generalized Hamming codes.
Recall that the points in PGn (Fq ) are the vectors in Fn+1q − {(0, 0, . . . , 0)}
moded out by the equivalence, where two vectors are equivalent if one is the scalar
n+1
multiple of the other. We saw that there were precisely q q−1−1 points in PGn (Fq ).
Lemma 11.5 Let H be the parity check matrix of a linear code C over Fq . A weight
s vector in C exists if and only if there are s linearly dependent columns in H.
Proof Assume v = (vi ) is a vector in the code C and H1 , H2 , . . . , Hn are the

columns of H.
Let v be a weight s vector. Since v is orthogonal to the rows of H then vi Hi =
0. Since exactly s of the vi are non-zero then the corresponding Hi are linearly
dependent.
If there are s linearly dependent columns of H, say Hj , j ∈ A where A is a subset
αj vj = 0. Let v be a
of {1, 2, . . . , n} of cardinality s, then there exists αj with
vector where vj = αj if j is in A and 0 otherwise. Then vi Hi = 0 and therefore
v is orthogonal to the rows of H which gives that v ∈ C.

Lemma 11.6 Let v1 , v2 , . . . , vs be distinct points of PGn (Fq ). The vectors
v1 , v2 , . . . , vs
are linearly independent as vectors in Fn+1

q if and only if the s points do not lie on
projective s − 2-dimensional space in PGn (Fq ).
Proof Assume v1 , v2 , . . . , vs are linearly independent vectors, then they form an

s-dimensional vector space in Fn+1 q which is an s − 1 projective dimensional space
in PGn (Fq ) which cannot lie on an s − 2 projective dimensional space.
If the points do not lie on an s − 2 projective space, then they must lie on a space
of projective dimension s − 1. Hence, as vectors they must generate at least an s-
dimensional subspace of Fn+1q . Since there are s vectors generating an s-dimensional
space, they must be linearly independent.

As an example, three vectors are linearly independent if they do not lie on a

projective line which is a one-dimensional projective space.
Theorem 11.15 An [n, k, d] code over Fq exists if and only if there exist n points in
PGn−k−1 (Fq ) such that no d − 1 points lie on a d − 3 projective space but some
d points lie on a projective d − 2 space.
11.7 Sphere Packing Bound and Perfect Codes 281
Proof Assume an [n, k, d] code C exists. Then there exists a generator matrix H
for C⊥ with n − k rows. View the columns of H as points in PGn−k−1 (Fq ). By
Lemma 11.6, no d − 1 points lie on a d − 3 projective space but some d points lie
on a projective d − 2 space.
Assume that there exists n points in PGn−k−1 (Fq ) such that no d − 1 points
lie on a projective d − 3 projective space but some d points lie on a projective
d − 2 space. Construct the matrix H with columns as these n points. Then, again by
Lemma 11.6 we have the result.

We can now construct the generalized Hamming codes. Let v1 , v2 , . . . , vn , n =

qr −1
q−1 , be the distinct points in PGr−1 (Fq ), that is, they are vectors of length r in
Frq . Let H be the r by n matrix where the columns of H are precisely the vectors
v1 , v2 , . . . , vn . Let Ham(r, q) be the code that has H as a parity check matrix, that
is, if H generates a code C then Ham(r, q) = C⊥ .
r r
Theorem 11.16 The code Ham(r, q) is a [ qq−1
−1 q −1
, q−1 − r, 3] code.
r
Proof The length is qq−1−1
because this is the number of points in the projective geom-
etry. The matrix H has r linearly independent rows so the dimension of Ham(r, q)
r −1
is qq−1 since the sum of the dimension of the code and its orthogonal equals the
length. Finally, since the points are distinct but there are three points on a line by
Lemma 11.6 the minimum distance is 3.

Exercises
1. Modify the decoding algorithm in the previous section to show how to correct a
single error using Ham(r, q). Hint: HvT may not be a column of H but perhaps
a multiple of a column. Determine which error produces that vector and the
decoding algorithm follows.
2. Determine the weights of the vectors in Ham(r, 2)⊥ .
3. Construct the parity check matrix for Ham(2, 5).
4. Construct the parity check matrix for Ham(2, 11).
11.7 Sphere Packing Bound and Perfect Codes
We shall describe an interesting bound on the size of the code. The bound depends
simply on counting principles. We use a few lemmas to prove it and to describe some
interesting aspects of the ambient space.
Lemma 11.7 If v is a vector in Fn s n

p then there are C(n, s)(p − 1) vectors in Fp
that have Hamming distance s from v.
282 11 Codes
Proof There are C(n, s) ways of choosing s coordinates from the n coordinates of
the ambient space in which to change the entry of v and there are p − 1 choices for
each coordinate to which the entry can be changed.

A sphere around a vector c of radius t is the set of all vectors whose Hamming
distance from v is less than or equal to t.
t
Lemma 11.8 If v is a vector in Fn
p , then there are s=0 C(n, s)(p − 1)
s in a
sphere of radius t around v.
Proof The possible Hamming distances are from 0 to t and we use Lemma 11.7 to
count them.

These lemmas can now produce the sphere packing bound.
Theorem 11.17 (Sphere Packing Bound) Let C be a code with minimum weight
2t + 1 then
t

|C|( C(n, s)(p − 1)s ) ≤ pn . (11.5)
s=0
Proof We know that all of the spheres are distinct since the spheres are radius t and
the minimum distance is 2t + 1. Hence, the number of vectors in all of the spheres
must be less than or equal to the number of vectors in the ambient space which is
pn .

Definition 11.4 A code with equality in the sphere packing bound is said to be a
perfect code.
It is clear why such codes would be thought to be perfect. In these codes, every
vector in the ambient space is of distance less than or equal to t with a unique vector
in the code. In this situation, the spheres of radius t around the codewords partition
the space.
We claimed in the previous section that the [7, 4, 3] Hamming code was perfect,
we can now prove it.
Corollary 11.1 The binary [7, 4, 3] Hamming code is a perfect code.

Proof The left side of the inequality 11.5 is |C| 1s=0 C(7, s)(2 − 1)2 = 16(1 +
7) = 128. The right side is 27 = 128. Hence, the code is perfect.

More generally we have the following.

11.8 MDS Codes 283
ps −1
Theorem 11.18 The generalized Hamming codes over Fp have length p−1 and
ps −1
have dimension p−1 − s and minimum weight 3. These codes are perfect codes.
ps −1
ps −1 p−1 −s
Proof We have n = p−1 and M = p . Then
ps −1
p−1 −s
ps − 1 ps −1
−s
p (1 + (p − 1)) = p p−1 (1 + ps − 1)
p−1
ps −1 ps −1
=p p−1 −s ps = p p−1 .
Therefore, the code is perfect.

Remarkably Golay came up with these codes in a single one-page paper early in
the history of coding theory [40]. The binary Golay code is closely related to the
Witt designs and the Mathieu groups.
Exercises
1. The ternary Golay code is a [11, 6, 5] code. Prove that it is perfect.

2. Find all vectors in F42 in a sphere of radius 2 around the vector (1, 0, 1, 0). Find
all vectors in F33 in a sphere of radius 1 around the vector (1, 2, 1).
3. The binary Golay code is a [23, 12, 7] code. Prove that it is perfect.
4. Prove that if the minimum distance of a code is 2t + 1 then two spheres around
codewords of radius less than or equal to t are distinct.
11.8 MDS Codes
One of the most important and interesting classes of codes is MDS codes. They also
have very interesting connections to combinatorics. In fact, one of the open problems
introduced in an earlier chapter can be phrased in terms of these codes.
We begin with a theorem that gives a bound on how big the minimum distance
can be for a code. We use this theorem to define MDS codes. Notice that the bound
depends only on combinatorial properties and does not use the underlying algebra
of the code.
Theorem 11.19 (Singleton) Let C be a code of length n over an alphabet A of size

q with minimum Hamming distance d and qk elements then
d ≤ n − k + 1. (11.6)
Proof Delete the first d − 1 coordinates of each code word. The vectors formed by
the remaining n − d + 1 coordinates must be distinct, since otherwise the distance of
284 11 Codes
their corresponding codewords would be less than d. Hence, qn−d+1 ≥ qk which

gives that d ≤ n − k + 1 and the result follows.

Notice that we did not need to assume that the alphabet A had any algebraic
structure. For example, we did not have to assume that it was a field. For the first
few connections, we need not assume anything about the algebraic structure; later,
however, we shall assume that A = Fq .
A code meeting the bound in Equation 11.6, that is, d = n − k + 1, is called a
Maximum Distance Separable (MDS) code. They are very important objects in the
study of algebraic coding theory.
There are three examples usually referred to as the trivial examples of MDS codes.
• The code Fn q has length n, dimension n, and minimum weight 1 and so is an

[n, n, 1] MDS code.
• The code generated by (1, 1, . . . , 1) has length n, dimension 1, and minimum
weight n and so is an [n, 1, n] MDS code.
• The code that is orthogonal to the second code, i.e., C = {v | [v, (1, 1, . . . , 1)] = 0}
has length n, dimension n − 1, minimum weight 2, and so is an [n, n − 1, 2] MDS
code.
It is not easy to know when MDS codes exist but the following is a well-known
conjecture.
Conjecture 11.1 If C is an [n, k, n − k + 1] MDS code over Fp then n ≤ p + 1.
We shall now give some interesting connections to some combinatorial objects

that we have already studied.
Theorem 11.20 A set of s MOLS of order q is equivalent to an [s + 2, q2 , s + 1]

MDS code.
Proof We shall prove one direction and leave the other direction which is basically
the reverse as an exercise. Let L1 , L2 , . . . , Ls be a set of MOLS of order q. Let a
vector beginning with (i, j) have in its (h + 2)-nd coordinate Lh i,j for all i, j ∈ Zq .
2
This gives q vectors of length s + 2.
If two vectors agree in the first or second coordinate, then they must disagree
in the (h + 2)-nd coordinate since no row or column of a Latin square can have
the same element twice. If two vectors disagree in the first and second coordinates,
then if they agree in the (h + 2)-nd coordinate and the (h + 2)-nd coordinate then
(Lh h h h h h
i,j , Li,j ) = (Li ,j , Li ,j ) with i = i and j = j contradicting that L and L are
orthogonal. Hence, any two vectors must disagree in at least s + 1 places. This is
also the largest the minimum weight that can be by the Singleton bound. Therefore,
the code is an [s + 2, 2, s + 1] code, and hence an MDS code.

11.8 MDS Codes 285
The next connection to MDS codes will only relate to codes over finite fields. So
from this point we shall assume that the alphabet is a finite field Fq of order q.
We shall first describe how we can define a code in terms of a parity check matrix.
Let H be a matrix with n columns and r linearly independent rows. Let C be a code
formed from all of the vectors orthogonal to each row. That is, let T (v) = HvT be a
linear transformation. Then the code C is defined by
C = {v | T (v) = 0} = Ker(T ). (11.7)
Then, C is a linear [n, n − r, d] code.

It is easy to see that the parity check matrix generates C⊥ . This gives the following.
Theorem 11.21 Let C be a linear [n, k, d] code, then C⊥ has length n and dimension
n − k.
We can now relate MDS codes and algebraic properties of their parity check
matrix.
Lemma 11.9 Let C be an [n, k, n − k + 1] MDS code with H its parity check matrix.
Then any n − k columns of H must be linearly independent.
Proof Any vector v in the code C represents a linear combination of wt(v) columns
of H summing to 0. Hence, the minimum weight of the code relates to the largest
number of linearly independent columns of H.

Lemma 11.10 If C is a linear MDS code over Fq , then C⊥ is a linear MDS code
over Fq .
Proof Let C be an [n, k, n − k + 1] MDS code, and let H be its parity check matrix.
Then H is a n − k by n matrix that also generates C⊥ with the property that any
n − k columns are linearly independent. Let v be a vector in C⊥ , with v = 0. This
gives that the maximum number of coordinates with a 0 in v is n − k − 1. This gives
that the minimum weight is k + 1 and C⊥ is an [n, n − k, k + 1] MDS code.

These lemmas lead naturally to the following theorem.
Theorem 11.22 Let C be an [n, k, d] code over Fq with parity check matrix H. Then
the following are equivalent:
• C is an MDS code;
• every n − k columns of the H are linearly independent;
• every k columns of the generator matrix are linearly independent.
Proof Follows from Lemma 11.9 and Lemma 11.10.

286 11 Codes
We can view the columns of H as points in projective space. Consider n points in

PGk−1 (Fq ) with the property such that no k points lie in a hyperplane PGk−2 (Fq ),
that is, any k are linearly independent as vectors. These n points are known as an
n-arc.
Theorem 11.23 An [n, k, n − k + 1] MDS code is equivalent to an n arc in

PGn−k−1 (Fq ).
Proof By Theorem 11.22, we know that an [n, k, n − k + 1] MDS code is equivalent

to a parity check matrix with any n − k linearly independent columns, so simply use
the n points as the columns of H.

Exercises
1. Construct the MDS code from two MOLS of order 3.

2. Prove the remaining direction of Theorem 11.20.
11.9 Weight Enumerators and the Assmus-Mattson Theorem
One of the most interesting and important connections between codes and designs
comes from the Assmus-Mattson theorem, which first appeared in [5]. This theorem
has been used both in constructing designs and in understanding the structure of
codes. In order to state the theorem, we shall first need a discussion of the weight
enumerator of a code. There are numerous weight enumerators that can be defined
for a code. For example, the complete weight enumerator, the symmetric weight
enumerator, the joint weight enumerator, and the higher weight enumerator are all
examples of weight enumerators. We shall define and use the simplest which is the
Hamming weight enumerator.
Let C be a code over the finite field Fq . We define the Hamming weight enumerator
of C as follows:

WC (x, y) = xn−wt(c) ywt(c) , (11.8)
c∈C
where wt(c) is the Hamming weight of the vector c. The reader should be aware that
in some texts the roles of x and y are reversed, for example, it is this way in [4]. It
follows immediately that the weight enumerator can also be seen as the following:
n

WC (x, y) = Ai xn−i yi , (11.9)
i=0
where there are Ai vectors in C of weight i.

As an example, the weight enumerator of the [7, 4, 3] Hamming code is
WC (x, y) = x7 + 7x4 y3 + 7x3 y4 + y7 . (11.10)

11.9 Weight Enumerators and the Assmus-Mattson Theorem 287
The weight enumerator of a code is related to the weight enumerator of the dual
code by the MacWilliams relations. These relations are some of the most interesting
and useful theorems in coding theory. These were first proved in the Ph.D. thesis of
Jesse MacWilliams [61].
We shall give a proof of the theorem in the binary case and state it for the more
general case. The proof for the general case is not that much more difficult but
requires a bit more algebra. We need a bit of machinery and a lemma.
Let f be a function with Fn
2 as its domain. The Hadamard transform f of f is
defined by

f(u) = (−1)u·w f(w), (11.11)
w∈Fn
2
where u · w is the usual dot product.
Lemma 11.11 If C is an [n, k] binary linear code then

1
f(w) = f(u). (11.12)
|C|
w∈C⊥ u∈C
Proof We have

f(u) = (−1)u·w f(w)
u∈C u∈C w∈Fn
2

= f(w) (−1)u·w .
w∈Fn
2
u∈C
If w ∈ C⊥ then (−1)v·w = 1 for all u ∈ C. If w ∈ / C⊥ then (−1)v·w is 1 and −1

equally often and hence cancels to 0 over the summation. This gives

f(u) = |C| f(w). (11.13)
u∈C w∈C⊥
Thus we have the result.

This brings us to one of the most important theorems in all of coding theory.
Theorem 11.24 (MacWilliams Relations) Let C be a linear code over F2 . Then
1
WC⊥ (x, y) = WC (x + y, x − y). (11.14)
|C|
288 11 Codes
Proof Apply Lemma 11.11 to the function f(v) = xn−wt(v) ywt(v) . Then we have

f(v) = (−1)v·w xn−wt(v) ywt(v)
w∈Fn
2
n

= (−1)v1 w1 +v2 w2 +···+vn wn x1−wi ywi
w∈Fn
2
i=1
1
1
1
n
= ... (−1)vi wi x1−wi ywi
w1 =0 w2 =0 wn =0 i=1
n
1
= (−1)vi α x1−α yα .
i=1 α=0
If vi = 0 then inner sum is x + y. If vi = 1 then the inner sum is x − y.

This gives that
f(v) = (x + y)n−wt(v) (x − y)wt(v) .

More generally, the theorem can be stated as follows.
Theorem 11.25 (MacWilliams Relations) Let C be a linear code over Fq . Then
1
WC⊥ (x, y) = WC (x + (q − 1)y, x − y). (11.15)
|C|

We often let x = 1 and simply write the weight enumerator as WC (y) = Ai yi .
This allows for the weight enumerator to be written in a simpler manner.
Given a vector v, the support of v consists of those coordinates where the vector
is non-zero. For example, if p1 , p2 , . . . , p8 are the coordinates of the space and
(3, 1, 0, 2, 0, 0, 1, 1) is a vector then its support is {p1 , p2 , p4 , p7 , p8 }. Viewing the
coordinates of Fn q as points we can construct designs using the supports as blocks.
We can now state the Assmus-Mattson theorem, without proof, as it appears in [4].
Theorem 11.26 (Assmus-Mattson Theorem) Let C be an [n, k, d] code over Fq

and C⊥ an [n, n − k, d⊥ ] code. Let w = n when q = 2 and if q is not 2 let w be the
largest integers satisfying
w+q−2
w−( ) < d.
q−1
Let w⊥ = n when q = 2 and if q is not 2 let w⊥ be the largest integers satisfying
w⊥ + q − 2
w⊥ − ( ) < d⊥ .
q−1
Suppose there is an integer t with 0 < t < d that satisfies the following: If WC⊥ =

Bi yi at most d − t of B1 , B2 , . . . , Bn−t are non-zero. Then for each i with
11.9 Weight Enumerators and the Assmus-Mattson Theorem 289
d ≤ i ≤ w the supports of the vectors of weight i in C, provided there are any, give
a t-design. Similarly, for each j with d⊥ < j ≤ min{w⊥ , n − t} the supports of the
vectors of weight j in C⊥ , provided there are any, give a t-design.
We shall apply the previous theorem to the binary self-dual [24, 12, 8] Golay code.
A self-dual code is a code that satisfies C = C⊥ . Here w = w⊥ = n and d = d⊥ = 8.
If t is 5 then d − t = 3. We see that of B1 , B2 , . . . , B19 only B8 , B12 and B16 are
non-zero. Hence, only d − t of B1 , B2 , . . . , Bn−t are non-zero. This proves that the
supports of the vectors of weight 8, weight 12 and weight 16, each form 5-designs.
j
The λi table for the 5 − (24, 8, 1) design formed by the weight eight vectors is given
in 11.16.
We shall consider the design formed by the weight 12 vectors. There are 24 points
and 2576 blocks. We have that λ0 = 2576. Using the relation that λs = k−s v−s
λs+1
we get that λ1 = 1288, λ2 = 616, λ3 = 280, λ4 = 120, and λ5 = 48.
j
Then we can construct the table for λi :
2576
1288 1288
616 672 616
(11.16)
280 336 336 280
120 160 176 160 120
48 72 88 88 72 48
We shall now examine a non-binary case. Consider the ternary self-dual [12, 6, 6]
code. It has weight enumerator:
WC (y) = 1 + 264y6 + 440y9 + 24y12 . (11.17)
We have d = d⊥ = 6. We must find the largest w with
w+q−2
w−( )<d
q−1
w+1
w−( )<6
2
w 1
− < 6.
2 2
This gives that w = w⊥ = 12. If t is 5 then d − t = 1. We see that of B1 , B2 , . . . , B7

only B6 is non-zero. Hence, only d − t of B1 , B2 , . . . , Bn−t are non-zero. This
proves that the supports of the vectors of weight 6, weight 9, and weight 12 each
form 5-designs.
290 11 Codes
Exercises
1. Find the weight enumerator of the orthogonal of the [7, 4, 3] Hamming code with
weight enumerator given in Equation 11.10.
2. Determine the parameters for the designs formed by the Assmus-Mattson theorem
for the ternary self-dual Golay code.
3. Find the weight enumerator for the code of length n generated by the all one vec-
tor. Use the MacWilliams relations to find the weight enumerator of its orthogonal.
Could this have been computed directly?
4. The self-dual [48, 24, 12] binary code has weight enumerator:
WC (y) = 1 + 17296 y12 + 535095 y16 + 3995376 y20 + 7681680 y24

+ 3995376 y28 + 535095 y32 + 17296 y36 + y48 .
• Use the Assmus-Mattson theorem to prove that the supports of any non-trivial
weight form a 5-design.
• Prove that the vectors of weight 12 form a 5 − (48, 12, 8) design.
• Produce the λji table for the 5 − (48, 12, 8) design.
11.10 Codes of Planes
Codes have often been very useful in understanding finite designs. In this section,
we shall describe some codes that come from designs, and we shall show how codes
can be used to give a short proof of part of the Bruck-Ryser theorem.
Recall that a finite projective plane Π of order n is a set of points P, a set of lines
L, and an incidence relation I between them, where |P| = |L| = n2 + n + 1, and
any two points are incident with a unique line, and any two lines are incident with a
unique point.
Let Π be a projective plane of order n, and let p be a prime dividing n.
The characteristic function of a line L is

1 if p is incident with L,
vL (p) = (11.18)
0 if p is not incident with L.
The code generated by the characteristic functions of lines over Fp is denoted by

Cp (Π). That is
Cp (Π) = vL | L ∈ L. (11.19)
We define the Hull to be
Hullp (Π) = Cp (Π) ∩ Cp (Π)⊥ . (11.20)

11.10 Codes of Planes 291
It is immediate that the Hull is a self-orthogonal code that is contained in Cp (Π).

In general, we always take the prime p to be a divisor of n. The interesting codes
are all over Fp where p divides n.
Lemma 11.12 Let Π be a projective plane of order n and let p be a prime dividing
n. Then

vL = j,
L∈L
where j is the all one vector.
Proof Through any point there are n + 1 lines and n + 1 ≡ 0 (mod p) when p
divides n which gives that in each coordinate of L∈L vL is a 1.

The following result can be found in [4].
Theorem 11.27 Let Π be a projective plane of order n, and let p be a prime dividing
n. We have that
Hullp (Π) = vL − vM | L and M lines of Π (11.21)
and Cp (Π) = Hullp (Π), j, where j is the all one vector.
Proof For any three lines L, M, and T of Π, we have [vT , vL − vM ] = 0, and there-
fore vL − vM is in Cp (Π)⊥ for any two lines vL , vM of Π. It is clear that these
vectors are in Cp (Π) and hence in Hullp (Π).
This gives that Hullp (Π) is at most of codimension 1 in Cp (Π) since Cp (Π) =
p (Π), vL for any line L. We note that j, the all one vector, is in Cp (Π) since
Hull
L∈L vL = j, and [j, vL ] = 1 for any line L, and therefore j is not in Hullp (Π)
giving that Cp (Π) = Hullp (Π), j.

The following lemma is well known, see [4].
Lemma 11.13 Let Π be a projective plane of order n. If p sharply divides n, that

2
is, p divides n but p2 does not, then dim (Cp (Π)) = n +n+2
2 .
It follows immediately from this lemma and by the previous theorem that
2 2
dim(Cp (Π)⊥ ) = n2 + n + 1 − n +n+2 2 = n +n+2
2 − 1 = dim(Hullp (Π))
which gives that Cp (Π) = Hullp (Π)⊥ .
The following is a standard theorem in coding theory, see [62] for a proof.
Recall that a self-dual code is a code C with C = C⊥ .
Lemma 11.14 Let p be a prime with p ≡ 3 (mod 4); if C is a self-dual code of

length m over Fp , then 4 divides m.
292 11 Codes
For the remainder we assume that p ≡ 3 (mod 4) and that p sharply divides n,
the order of the plane Π. We have that
Cp (Π) = Hullp (Π) ∪ (Hullp (Π) + j) ∪ (Hullp (Π) + 2j)

∪ · · · ∪ (Hullp (Π) + (p − 1)j),
since Cp (Π) = Hullp (Π), j, where (Hullp (Π) + ij) denotes the coset of
Hullp (Π) in Cp (Π) formed by adding ij to every vector in Hullp (Π). Let
Hi = (Hullp (Π) + ij), giving that Hi ∩ Hk = ∅ for i = k and
p−1

Hi = Cp (Π).
i=0
Let v ∈ Hi , w ∈ Hk then v = h + ij and w = h + kj where h, h ∈ Hullp (Π).

Then
[v, w] = [h + ij, h + kj] = [h, h ] + [h, kj] + [ij, h ] + [ij, kj] = ik. (11.22)

To the vectors in Hi adjoin some vector vi of length 3; and set C = (Hi , vi ). To
insure the linearity of this new code once we have chosen v1 and then vi is forced to be
iv1 . To insure that C is self-orthogonal we must have [v1 , v1 ] = −[H1 , H1 ] = −1,
where [Hi , Hk ] is the inner product of any vector in Hi with any vector in Hk .
Let v1 = (x, y, 0) where x2 + y2 = −1. It is well known that this has a solution
in Fp when p ≡ 3 (mod 4), with neither x nor y equal to 0. Then
[vi , vk ] = ikx2 + iky2 = ik(x2 + y2 ) = −ik = −[Hi , Hk ].
Let w be the vector of length n2 + n + 4 of the form (0, a, b, c). We want ax +

x
by = 0; hence, choose a non-zero a, and then we have b = −a( y ). Then, we want
2 2 2
a + b + c = 0 giving that
x2
a 2 + a2 + c2 = 0
y2
x2 2
c2 = −(1 + )a
y2
a2
c2 = .
y2
Then c = a
y . Note w is self-orthogonal, orthogonal to every vector in C and is
linearly independent over C . Let D = C , w, then D has length n2 + n + 4 and
2 2
has dimension n +n+22 + 1 = n +n+42 . This gives that D is a self-dual code. This
implies that n + n + 4 is divisible by 4, which implies n2 + n + 1 ≡ 1 (mod 4).
2
If n ≡ 1 or 2 (mod 4) then n2 + n + 1 is not 1 (mod 4) giving a special case of

the Bruck-Ryser theorem.
11.10 Codes of Planes 293
Theorem 11.28 If n ≡ 1 or 2 (mod 4) and p is a prime sharply dividing n with p

a prime and p ≡ 3 (mod 4), then there does not exist a projective plane of order n.
Exercises
1. Determine which order less than or equal to 30 are eliminated by

Theorem 11.28.
Cryptology
12
One of the most interesting and important modern applications of discrete combina-
torial mathematics is cryptology. Cryptology is the science of sending and receiving
information that is meant to be secret. It has been in use since the time of Julius Caesar
and has been played an important role in historical events including both world wars.
More recently, beginning in the late twentieth century, it has become foundational
in Internet security and commerce. Cryptology has never been so widely used as it
has been in the twenty-first century. Every time someone purchases an item on the
Internet, cryptology is used. It is often used in numerous electronic communications
every day.
Several terms are used to describe the various parts of this discipline. The term
cryptography refers to the science of creating a system to keep information secret,
whereas cryptanalysis refers to the science of breaking a system designed to keep
information secret. However, the terms cryptography and cryptology are often used
interchangeably without much confusion.
Cryptology has had a very long and interesting history. It is one branch of mathe-
matics that has had an immediate impact on world events. For example, during World
War I, the German Zimmerman made an overture to Mexico, via a secret message,
to attack the United States with German supplies and logistical help. They believed
this would keep America busy in the western hemisphere and keep them out of
the European conflict. The Americans intercepted and decoded the message which
helped push America toward involvement in the war. The system used to encrypt
the message was a straightforward combinatorial technique. During World War II,
every major nation involved developed their own cryptographic system, but perhaps
the most important was the German Enigma machine. This machine looked like a
typewriter and used a combinatorial system to encode the messages. Famously, the
British set up an organization at Bletchley park to decode the Enigma machine, and
many historians attribute to this decoding a shortening of the war by years. During
296 12 Cryptology
the sixteenth century, Mary Queen of Scots used a simple cryptographic system in the
Babington plot, which was decoded by Queen Elizabeth’s codebreaker. The results
of this codebreaking were used in her trial which ended with her execution in 1587.
For this text, we shall be primarily interested in the application of combinatorial
techniques to this study. From the very beginning, combinatorial techniques were
at the very core of the construction of cryptographic systems. In many ways, it is
the combinatorial explosion, namely, the rapidly increasing number of ways that an
event can happen that allows for secrecy.
We shall begin by describing various substitution ciphers and then move on to
more modern and sophisticated techniques.
12.1 Substitution Ciphers
Caesar Cipher
The most basic and possibly least effective cryptographic technique is a simple
substitution cipher. This has its origins in the Caesar cipher used by Julius Caesar
2000 years ago. Caesar took a message written in Latin and simply cycled the letters
of the message three spaces (assuming the standard ordering of letters a, b, c, d, …,
z). (We are using the ordering of the letters as is generally done in English, ignoring
the fact that the Romans did not use the letter j.) For example, the message
all of gaul is divided into three parts

becomes
DOO RI JDXO LV GLYLGHG LQWR WKUHH SDUWV.

Notice that the original message is written in small letters and the encoded mes-
sage, called the ciphertext, is written in capital letters. We shall continue using this
convention throughout this chapter.
As usual, for early cryptographic systems, this was used largely for military pur-
poses. This technique was very effective for Caesar because few people were even
literate in Latin among his enemies. So even this very simple technique was likely
to work. At that time, messages were carried by hand by an individual. This means
that intercepting the message meant intercepting the person carrying the message. It
was much more likely that the person carrying the message could destroy it if they
thought they were about to be captured, so even having the message intercepted was
less likely. However, anyone who knew that Caesar used this technique to encode
messages would have been able to instantly decode it. Given this fact, it was not a
very secure system. In this case, decoding this system is easy. Namely, if you know
that the writer is simply cycling the letters by some amount, then once you establish
one letter you can break the entire system. In a language like English, where there
are two prominent one letter words, namely, a and I, it is a simple matter to determine
one letter if the writer has made the mistake of indicating ends of words. Even if the
12.1 Substitution Ciphers 297
words are all written together without spaces, finding the letter corresponding to the
most common letter in English, namely, e, is again an easy matter. In fact, even an
ancient codebreaker would know the most common letter in any written language
they are reading.
Generalizing this technique is to simply make an arbitrary permutation of the 26
letters, rather than simply cycling them. We know there are 26! ≈ 4 × 1026 possible
permutations of the letters. It would seem at first glance that with so many possibilities
this would be an effective system. However, it is an extremely insecure system.
If you were sending random collections of letters, then the system would be much
more secure. However, generally a message is written in a human language like
English. In English, as in every language written with letters, certain letters are used
far more often than other languages. Frequency tables in most languages are very
well known and easily determined. This makes decrypting a secret message using a
simple cipher quite easy. As long as the message is long enough, all that is necessary
is to make a frequency table of the symbols in the encoded message and match them
with the frequency table for the language. Then using some simple reasoning in terms
of the language, decryption becomes quite simple.
For example, in modern terms, it is a simple matter to find an electronic text
written in the language of your choice, perhaps news stories or a book in electronic
form. Then, have the computer simply count the number of occurrences of each letter
to establish their frequency. Given a document that has used a substitution cipher,
you perform the same computation. Then, the probability is very high, if the message
is long enough, that the symbol used most often is e. The second is most likely t,
and the third is most likely a. It is possible that this is not true, but the probability
that the top three are e, t, and a in some order is very high. This technique is much
less likely to work to establish which symbol corresponds to x, z, and q, which occur
rather infrequently. However, given a correct decoding of the more common letters,
these letters will come from context.
It is a fairly simple exercise to write a computer program using these techniques to
decrypt messages encrypted with a simple cipher. Given the existence of a computer,
the simple cipher can be broken almost immediately by competent cryptographers.
We can make a more systematic technique for producing a substitution using
modular arithmetic. We can attach numerical values 0 to 25 based on the letters’
position in the standard ordering of the English alphabet. Then, the Caesar cipher
can be seen as simply φ(x) = x + 3 (mod 26). Then, an affine substitution cipher
can be made by defining any function of the form
ψ(x) = αx + β.
We need gcd(α, 26) to be 1; otherwise, the function is not injective and has no
inverse, which would be necessary to decode the message. To decode the message
you use the inverse function
ψ−1 (x) = α−1 (x − β).

298 12 Cryptology
Since this produces a simple substitution cipher, it is still quite an easy matter to
break the system.
By the nineteenth century, a simple substitution was no longer be used in any
serious situation. Next, we shall examine several most sophisticated techniques that
were used, which are based on substitution ciphers.
Vigenère Cipher
The Vigenère cipher goes back to the sixteenth century, and many thought it was
secure even up to the twentieth century. However, Babbage and Kasiski had already
shown how to attack it during the nineteenth century and Friedman developed further
attacks in the 1920s. This cipher uses the same sort of modular arithmetic as was
done in the affine substitution cipher but with a twist. Take a given message and turn
the text into numbers modulo 26. Then take a keyword, which will form the key for
the system. We shall take the word geometry. This word corresponds to the vector
(6, 4, 14, 12, 4, 19, 17, 24). We note that the length of this vector is 8.
Assume we want to send the following message:
donotgogentleintothatgoodnight.
Then we change the message into its corresponding numbers and make a sequence
of length 30. To each number with index i in the sequence modulo 8, we add the
number corresponding to the ith coordinate in the vector. For our example, we take
the following:
d o n o t g o g e n t l e i n
3 14 13 14 19 6 14 6 4 13 19 11 4 8 13
6 4 14 12 4 19 17 24 6 4 14 12 4 19 17
9 18 1 0 23 25 5 4 10 17 7 23 8 1 4
J S B A X Z F E K R H X I B E
t o t h a t g o o d n i g h t
19 14 19 7 0 19 6 14 14 3 13 8 6 7 19
24 6 4 14 12 4 19 17 24 6 4 14 12 4 19
17 20 23 21 12 23 25 5 12 9 17 22 18 11 12
R U X V M X Z F M J R W S L M
Then the ciphertext is
JSBAXZFEKRHXIBERUXVMXZFMJRWSLM.
While this is based on the substitution cipher which is highly insecure, there are two
more cryptographic techniques at play. First, it is a different substitution cipher for
each of the coordinates modulo the length of the key. Second, the length of the key
is a secret.
We can use probability and a geometric function to determine the length of the
key. Take the vector of the letters’ frequencies in English and call it v0 . That is,
(v0 )i = pi where pi is the probability that the letter i is used in a standard English
text. Let vj be the vector formed by shifting v0 cyclicly j times to the right.
It is a simple, but time-consuming exercise, to determine that the maximum value
of [vi , vj ], where this indicates the standard inner product, occurs when i = j. More-
over, this is intuitive, since the values are probabilities between 0 and 1. Therefore,
when taking the inner product of vi with vi , the larger values are multiplied by
larger numbers maximizing their effect of being large and the smaller numbers are
multiplied by smaller numbers minimizing their effect of being small.
First, we need to determine the length of the keyword. To do this write the cipher-
text on two sheets of paper and begin cycling the second one under the first and count
the number of occurrences where the same letter appears. Whichever shift has the
most such coincidences is probably the length of the key. This is a direct consequence
of the fact that the maximum value of the above inner product is [vi , vj ].
Once one knows the length of the key, one can use frequency analysis on the
coordinates that are the same modulo the length of the keyword. Alternatively, you
can take the frequency of letters in the jth coordinate modulo the length of the
keyword and divide each number by the total numbers counted. Call this vector w.
Then take [vi , w] for all i. Whichever i gives the largest inner product is most likely
the shift used for that substitution cipher which gives the value for each letter in the
keyword.
Playfair Cipher
The Playfair system was constructed by Charles Wheatstone in 1854 and named it
for Baron Playfair of St. Andrews, since he convinced the British government to use
it. It was used by the British military in both the Boer War and World War I. The
system begins by building a matrix which is used as a key.
Take a word, then write the letters used in the word to begin a 5 by 5 square,
discarding any letters that have already been used. Then complete the square using
the remaining letters in the alphabet (equating i and j). Using the word erudite, we
obtain the following matrix: e r ud i
t a b c f
g h k l m
n o p q s
v w x y z
Then split the message into groups of two letters and encode the message using
the following rules:
1. If a double letter appears, place an x between them. If there is an odd number of

letters, then place an x at the end.
2. When two letters are not in the same row nor the same column, replace each letter
by the letter that is in its row and in the other letter’s column.
3. When two letters are in the same row, replace each letter with the letter that is to
its right in the row (wrapping around the row if necessary).
4. When two letters are in the same column, replace each letter with the letter that
is below it in the column (wrapping around the column if necessary).
300 12 Cryptology
We shall use this to encode the following message: the eagle lands at midnight.
We begin by splitting the message into groups of two letters.
th ex ea gl el an ds at mi dn ig ht
Then apply the cipher to get the following.
AG UV RT HM DG TO IQ BA SF EQ EM GA
To decode a message, simply reverse the previous instructions.
For example, assume you received the message:
GD PO OH UI RD GD UW
Finding G and D in the table, you find the letters in their own rows and in the
others column and you get l and e. Finding P and O, we see that they are in the same
row, so we go backward in the row to get o and n. Continuing with this process, we
see that the original message is
leonhard euler.
In essence, this system is a glorified substitution cipher. Instead of substituting letters
one at a time, they are substituted two at a time. This type of system is easily defeated
since the frequencies of pairs of letters in English are well established as well.
Moreover, even though there are 262 = 676 pairs of letters in English, many of them
never occur or occur so infrequently that they can be discarded. For example, the
pair q followed by any letter other than u either never occurs or occurs in only
a few words which are unlikely to be used and of foreign origin so that one can
reasonably abandon them. It becomes a fairly simple matter to decode a message
using this system, given enough time. Moreover, if you can determine the keyword,
then decoding the messages becomes a trivial matter. The Germans during World
War I had a better system which we now describe.
ADFGX Cipher
To begin this system, you create a 5 by 5 matrix and fill it with the alphabet (equating
i and j as usual) and indexing the rows and columns by the letters A, D, F, G, and
X. These letters are chosen since the Morse codes of these letters were distinct
and not easily confused. Of course, at the time, one of the most used methods of
communications was the telegraph, and so this would have been of great importance.
We shall assume we have the following matrix:
A DF G X
A m u d c n
D f v w r p
F q b s e y
G t k g i x
X p l a z h
For each letter in the message, you replace it with the row and column index of
that letter. For example, the letter e would become FG. At this point, it is only a very
simple substitution cipher. However, the complexity is introduced in the next step.
Take a keyword, say Munich, and write the encoded message as a matrix with the
letters of the keyword indexing the columns. Then rearrange the columns by putting
the letters of the keyword in alphabetical order. The last step is to read the message
down the columns.
Assume our original message is: “strike up the band”. We first change it into its
representation in terms of A, D, F, G, and X. We get
FFGADGGGGDFGADDXGAXXFGFDXFAXAF.
Then, we write the message in the following matrix:
M UN I CH
F F GA DG
G GGD F G
A DDX GA
X X F G F D
X F A X A F
Next, we rearrange the matrix by putting the columns in alphabetical order:
CH I MNU
DGA F G F
F GD G GG
GA X A DD
F DG X F X
A F X X A F
Then the ciphertext is
DFGFAGGADFADXGXFGAXXGGDFAFGDXF.
To decode, the process is simply done in reverse.
We note that the security of this system lies in the fact that a frequency count is
not useful on the cipher text, since the pairs of letters corresponding to the original
message may have been scrabbled in the final ciphertext. Later, this technique was
replaced by the ADFGVX system, where all 26 letters and 10 digits could be used
in a message.
This technique is certainly more effective than the Playfair system, but it was
broken by Georges Painvin and the French Bureau du Chiffre using various tech-
niques. For example, if you can intercept two different ciphertexts sent around the
same time that agree in the first few characters. This is not as unlikely as it may seem,
since often military messages have similar beginnings. This was especially true for
302 12 Cryptology
the German military of the time. Hopefully, this means that the messages agree for
the first few words. This would mean that the top several entries of the columns are
identical.
Next search for other instances of identical occurrences. Hopefully, this indicates
the beginning of columns. If we are fortunate enough to be correct, this will indicate
the length of columns. If the columns are different lengths, the longer ones should be
in the beginning and the shorter ones at the end. Then try various arrangements of the
columns, which will give a straightforward substitution cipher. Applying frequency
analysis to each should decode one message and indicate the proper ordering of the
columns. While a computer could do this quite simply and quickly, by hand it is
no easy matter. However, the French were able to decode a significant number of
German transmissions.
Block Ciphers
The final substitution cipher that we shall discuss is the block cipher. We have seen
that substituting one letter at a time falls easily to frequency analysis. Even substi-
tuting two at a time, generally falls to frequency analysis as well, noting that the
frequency of pairs of letters is equally well known. One might then think to increase
the number of letters taken at a time to make frequency analysis less effective. This
can be done with a block cipher.
Take a message and break it into blocks of size n. Then write each letter as a
number modulo 26. This assumes that we are using the standard English alphabet.
Of course, one can use as large a modulus as you like, using as many symbols as you
like. To construct a substitution cipher for n letters at a time, construct an n by n
invertible matrix M. Then take a vector v of n symbols constructed from n letters.
Form the vector MvT = w, which is then the ciphertext. To decode the message, we
take M−1 w = M−1 MvT = vT .
Example 12.1 Take the message “send money” and split it into groups of size 3
and take the numerical value modulo 26. This gives (18, 4, 13)(3, 12, 14)(13, 4, 24).
Take the matrix
⎛ ⎞
312
M = ⎝2 1 4⎠.
533
Then M(18, 4, 13)T = (6, 14, 11)T , M(3, 12, 14)T = (23, 22, 15)T , and M(13,
4, 24)T = (13, 22, 19)T . Then the ciphertext is (6, 14, 11)(23, 22, 15)(13, 22, 19)
which gives
GOLXWPNWT.
The inverse matrix is
⎛ ⎞
15 21 14
M = ⎝ 20 19 22 ⎠ .
7 24 7
Then M−1 (6, 14, 11)T = (18, 4, 13)T , M−1 (23, 22, 15)T = (3, 12, 14), and M−1
(13, 22, 19) = (13, 4, 24), which gives
sendmoney.
Notice that the strength of the system is in the diffusion. For example, “sen” was
sent to “GOL”. If we take the word “ten”, then M(19, 4, 13)T = (9, 16, 16)T which
corresponds to “JQQ”. We note that “sen” and “ten” are sent to two very distant
elements.
Even with just a 3 by 3 matrix, this technique is far superior to a simple substitution
cipher. However, the computations can be quite lengthy even for a small message
without a computer. For example, it would be hard for someone during World War
I to compute these correctly for a long message even for a small matrix. With a
computer, you can use a very large matrix which will give a fair amount of security
using very simple linear algebra.
Other Techniques
We shall describe some other techniques that make substitution methods more effec-
tive.
The first technique is used to combat the effectiveness of frequency analysis. In
this system, instead of replacing the most common element “e” by a single element,
you replace it with a sequence of elements, 1 , 2 , 3 , . . . , κe , where i refers to
symbol in a set. That is, on the ith occurrence modulo κe , you replace “e” with
i . Replace “t” with τ1 , τ2 , τ3 , . . . , τκt and “a” with α1 , α2 , α3 , . . . , ακa . Then
continue, by replacing letters with a number of symbols that corresponds to its
frequency in the language. In other words, κ is chosen for each letter so that each
symbol has nearly the same frequency in a standard document.
While this technique is far superior to a substitution cipher, there are attacks
against it. However, they are much more sophisticated than standard frequency anal-
ysis.
Another technique is to construct a one time pad. This is a generalization of the
technique used in the Vigenère cipher. In this system, a prearranged sequence ai of
ones and zeros is made. The sequence is of length n. Then take a message and turn it
into a sequence of ones and zeros, mi . Then take si = ai + mi (mod 2) and send
it. The message is decoded by the same technique, namely, si + ai = mi (mod 2).
Essentially, there is a substitution cipher for every coordinate. It is as if you have a
Vigenère cipher keyword of length n.
The difficulty with this system is that it can only be used once. Each time the
system is used, it becomes less secure. Moreover, you need a very long sequence ai
to apply it to a meaningful message and you must exchange it ahead of time. It has
been rumored that this type of system was used between the president of the United
States and the premier of the Soviet Union during the cold war.
It is also possible to try and hide a secret message, which is called steganography.
This technique is often used to send secret photos. Think of a photo as a matrix of
zeros and ones, which are read to produce an image. In general, all computer files
304 12 Cryptology
can be thought of as being of this form. Two people agree before communicating on
which coordinates will be where the message is hidden. Then a much larger matrix
is also shared. For example, two people take a series of very high-definition pictures
which are viewed as n by n binary matrices M1 , M2 , . . . , Ms . They agree on which
k2 coordinates are where the secret photo will be hidden. Then they separate. The spy
can then take a photo with a resolution that fits into the k by k matrix. This matrix Si
is written on the pre-approved coordinates. Then the matrix Pi = Mi + Si (mod 2).
The picture can then be sent to headquarters or simply uploaded to the Internet. At
headquarters, they receive Pi and add Mi to it to obtain Pi + Mi = Si . Then they
have the secret photo taken by the spy.
One might believe that by adding the photo to Mi , this original photo will look
altered. However, if k is small enough compared to n it is almost impossible to tell
that the photo has been altered. Given also that millions of photos are uploaded to
the internet every day, it would be difficult, if not impossible, to test every photo if
there are some indications of this. To human sight, it is usually impossible to see any
difference between Mi and Pi .
Example 12.2 We shall show a very simple example to illustrate this steganographic
technique.
Assume the spy and headquarters agree on the following 22 coordinates in a 4 by
4 matrix:
⎛ ⎞
X
⎜X ⎟
⎜ ⎟.
⎝ X ⎠
X
Then assume the matrix M is given by the following:
⎛ ⎞
1111
⎜0 1 1 0⎟
⎜ ⎟
⎝0 0 1 1⎠.
1010
Headquarters receive the following matrix:

⎛ ⎞
1110
⎜1 1 1 0⎟
⎜ ⎟
⎝0 0 1 1⎠
1000
and adds it to M. The result is the following matrix:

⎛ ⎞
1
⎜1 ⎟
⎜ ⎟.
⎝ 0 ⎠
1
The information is read as (1, 1, 0, 1) which is interpreted as

11
.
01
Exercises
1. Use the word Britain to construct a Playfair matrix and encode the message:
“we shall meet under the apple tree” using this matrix.
2. Use the given ADFGX matrix and the word Berlin to encode the message:
“we shall meet under the apple tree”.
3. Determine the inverse of the affine cipher φ(x) = 3x + 9.
4. Use the affine cipher ψ(x) = 5x + 7 to encode the message: “this food is
delicious”. Find the inverse function and use it to decode the message.
5. Count the number of possible affine ciphers using (mod 26). Prove your
answer.
6. Try to encode the message: “this food is delicious” using the map ψ(x) =
13x + 2 to produce a convincing example as to why this does not work as a
cipher.
7. Prove that the matrix M in the block cipher must satisfy gcd(det(M), 26) = 1.
8. Use frequency analysis to decode the following possibly treacherous message
written in English:
MGXS HS LGX ZQKICX QY GKAES XJXSLC, HL WXZQAXC SXZX-
CCEIP YQI QSX BXQBRX LQ FHCCQRJX LGX BQRHLHZER WESFC
MGHZG GEJX ZQSSXZLXF LGXA MHLG ESQLGXI, ESF LQ ECCKAX
EAQSV LGX BQMXIC QY LGX XEILG, LGX CXBEIELX ESF XOKER
CLELHQS LQ MGHZG LGX REMC QY SELKIX ESF QY SELKIXÕC
VQF XSLHLRX LGXA, E FXZXSL IXCBXZL LQ LGX QBHSHQSC QY
AESTHSF IXOKHIXC LGEL LGXP CGQKRF FXZREIX LGX ZEKCXC
MGHZG HABXR LGXA LQ LGX CXBEIELHQS.
MX GQRF LGXCX LIKLGC LQ WX CXRY-XJHFXSL, LGEL ERR AXS
EIX ZIXELXF XOKER, LGEL LGXP EIX XSFQMXF WP LGXHI ZIX-
ELQI MHLG ZXILEHS KSERHXSEWRX IHVGLC, LGEL EAQSV
LGXCX EIX RHYX, RHWXILP ESF LGX BKICKHL QY GEBBHSXCC.-D
LGEL LQ CXZKIX LGXCX IHVGLC, VQJXISAXSLC EIX HSCLHLK-
LXF EAQSV AXS, FXIHJHSV LGXHI UKCL BQMXIC YIQA LGX
ZQSCXSL QY LGX VQJXISXF, LGEL MGXSXJXI ESP YQIA QY VQJX-
ISAXSL WXZQAXC FXCLIKZLHJX QY LGXCX XSFC, HL HC LGX
IHVGL QY LGX BXQBRX LQ ERLXI QI LQ EWQRHCG HL, ESF
LQ HSCLHLKLX SXM VQJXISAXSL, REPHSV HLC YQKSFELHQS
QS CKZG BIHSZHBRXC ESF QIVESHDHSV HLC BQMXIC HS CKZG
YQIA, EC LQ LGXA CGERR CXXA AQCL RHTXRP LQ XYYXZL
LGXHI CEYXLP ESF GEBBHSXCC.
306 12 Cryptology
12.2 German Enigma Machine
We now move to one of the most famous cases of breaking a cryptographic system,
namely, the German Enigma machine that was used in the Second World War. It
is one of the most important applications of combinatorial ideas (in terms of world
events) that has ever occurred. Successful breaking of the Enigma code is often said
to have reduced the Second World War by years. It was used in various forms by the
different branches in the German military throughout the duration of World War II.
We begin by describing the main components of the machine and determine the
number of possible initial configurations.
• Keyboard
The operator types on the keyboard exactly as one would do for a usual typewriter.
When one letter is typed, a different letter lights up. The one that the operator types
is the message and the one that lights up is the ciphertext. The operation is reversed
by the receiver, namely, they press the button that was sent as the ciphertext and
the original message lights up. This gives the reader an indication of the ease of
using this machine. The operators needed no special skills or intelligence, just the
ability to type and write down which letters were lit up. This simplicity allowed
it to be used throughout the German military.
• Plugboard
The plugboard uses six pairs of plugs that can be used to interchange six pairs of
letters. In other words, 6 pairs of letters are chosen from 26 letters and they are
exchanged. There are
C(26, 2)C(24, 2)C(22, 2)C(20, 2)C(18, 2)C(16, 2)/6! = 100, 391, 791, 500
ways of interchanging six pairs of letters.

The plugboard was not part of the original design but was added to significantly
increase the number of initial possible configurations.
• Static Rotor
The static rotor does nothing cryptographic. It performs a functional role in that
it simply turns mechanical action into electrical contacts.
• Rotors
The rotors are circular with letters written around the outside of the wheel. For
the Enigma used by the army, there are three possible rotors that can be used in
any order for the three rotor positions: right, middle, and left. Each has an inner
and outer contact. Hence, there C(5, 3) = 2!3!
5!
= 10 ways to pick three from five.
The German navy used a machine with more rotors which increased the number
of possibilities. We shall focus on the army version for our discussion. Each time
a letter is typed the rotors move one position. These act the way the hands of a
clock move. In other words, after a certain number of turns a rotor will cause the
next rotor to move one space over. So three rotors can act in a manner similar to a
second hand, minute hand, and hour hand. Each rotor can be placed with any of
12.2 German Enigma Machine 307
the 26 letters at the top. Hence, there are 263 = 17576 possible first positions for
the three rotors and 3! = 6 possible orderings of the three. Therefore, there are
C(5, 3)(6)(263 ) = 1, 054, 560
possible initial positions for the rotors.

• Reflector
There are two possible reflectors. The reflector must scramble the letters; other-
wise, it would send the letter back on the same path un-encrypted.
• Lampboard
The plugboard outputs the message to the lampboard. The operator records the
letter and sends it in the message.
Essentially, what the Enigma machine does is use a different substitution cipher
for each letter of the message. It takes advantage of the vast number of possibilities to
make it extremely difficult to decode the message. Moreover, the substitution cipher
used for each letter is different every day, and since the first thing operators did when
sending a message was to choose a setting using the standard setting of the day,
each cipher was different for each letter for each message. Therefore, all classical
techniques of decoding were useless against this machine.
Next, we shall count the number of possible initial configurations of the Enigma
machine.
Theorem 12.1 The number of possible initial configurations is
(1, 054, 560)(100, 391, 791, 500) = 105, 869, 167, 644, 240, 000.
Proof We saw that there were
C(26, 2)C(24, 2)C(22, 2)C(20, 2)C(18, 2)C(16, 2)/6! = 100, 391, 791, 500
possible ways or arranging the plugboard and
C(5, 3)(6)(263 ) = 1, 054, 560
possible ways of arranging the rotors.
This number is greater than 1018 , so it is clear that any type of straightforward
attempt of running through the possibilities would not be feasible, not even with a
high-speed computer. Even if we assume that a computer could try 109 possibilities,
it would still take 109 = 1, 000, 000, 000 seconds to go through them. However,
there are only 86, 400 in each day. It is precisely this combinatorial explosion that
gave this cryptosystem its high level of security.
In actual practice, each operator had a book for the month with the codes that
were going to be used for each day in that month. Then he put it on this initial setting
308 12 Cryptology
and chose three letters (say r, f, v) and then sent rfvrfv as his first message to a
receiver. This would indicate to the receiver that r, f, and v were the settings for the
three rotors. Of course, to do this the receiver must have the same codebook that the
sender has.
The reason that rfv is repeated is to ensure that if a mistake is made they will
realize it and ask for a retransmission. This prevents an entire message of garbled
information from being sent. However, this duplication of the key was the biggest
help to the cryptanalysts.
Let us assume the first six permutations that are performed by the Engima machine
are A, B, C, D, E, F. Then if xyzxyz gets sent to dmqvbn, it means that A inter-
changes x and d and D interchanges x and v. Therefore, the product DA (doing A
first) interchanges d and v. We have eliminated the unknown x. The same can be
done for permutations EB and FC.
Then, with enough data, we can write these permutations as products of cycles
like
(dvpfkxgzyo)(eijmunqlht)(bc)(rw)(a)(s). (12.1)
These rely only on the daily settings of the plugboard and the rotors, not the estab-
lished key that was found in the codebook. If the plugboard settings are changed
while the initial positions of the rotors remain the same, then the cycle lengths of
these permutations in Eq. 12.1 remain unchanged.
Polish mathematicians, Rejewski, Zygalski and Różycki, compiled a catalog of
all 105, 456 initial settings (this was when there were only three rotors possible)
along with the cycle lengths of the permutations formed from these before World
War II began. They smuggled this information to Britain.
Essentially, the codebreakers had to determine what the possible permutations
were based on the messages that were intercepted and the list of possible settings
with lengths of permutations. Each day this process had to be repeated to find the
daily keys. This attack was mechanized by building a “bombe”, which was an early
computer. The design is attributed to Alan Turing and is often considered to be one
of the first electronic computers. Given a modern computer, this search would be
quite simple, but for the technology of the day, this computation was quite daunting.
This codebreaking was done at Bletchley park in England. The British government
gathered together mathematicians, problem-solvers, and others to find help the initial
setting for each day, decode the messages, and translate them into English. The
activities at Bletchley Park were kept secret for decades after the end of the war.
Little credit was given to those who worked there nor to the Poles who had shown
them how to attack the Enigma years earlier.
Exercises
1. How many possible plugboard settings would there be if 10 pairs of letters could
be switched?
12.3 Public-Key Cryptography 309
2. How many possible rotor settings would there be if there were k rotors chosen
from n possible rotors, and each one could be placed in 26 different ways (meaning
a different letter at the top)?
3. Assume that the switchboard could interchange 13 pairs of letters, and that there
are five rotors chosen from eight possible rotors. How many initial configurations
would there be?
12.3 Public-Key Cryptography
The notion of public-key cryptography was first brought out by Diffie and Hellman
in [26]. The main goal of public-key cryptography is to construct a system which is
so secure that the whole world can know how the information is encoded and yet the
information can only be decoded by the person who constructs the system. This type
of system is necessary for things like Internet commerce. Consider a company with
a website that allows a customer to purchase items. In order to purchase the item, the
customer must enter information like their card number, their address, and their name.
This information must be kept secret. The company’s website must send information
to the computer of the user as to how this information is to be encrypted. Anyone
can purchase items on this website so anyone can figure out how the information is
encrypted. In other words, the company must tell the entire world how to send them
information. The system must be designed so that even though you know how the
information is encrypted, it is still impossible to decode the information. It is this
type of system which allows for such business to occur over the Internet. We begin
with some number-theoretic results which allow us to construct such systems.
We shall now prove the well-known Chinese remainder theorem, which we shall
use to prove important results needed for the RSA cryptosystem.
Theorem 12.2 Let n and m be natural numbers with gcd(n, m) = 1. Then
x ≡ a (mod m)
x ≡ b (mod n)
has a unique solution mod mn.
Proof If x ≡ a (mod m) then x = a + mk for some k ∈ Z. Then x ≡ b (mod n)

gives a + mk ≡ b (mod n) and then mk ≡ b − a (mod n). Then since m is rel-
atively prime to n there is a unique solution for k, call it s. So k = s + jn for
some j ∈ Z. Then x = a + m(s + jn) = a + ms + jmn giving that x ≡ a + ms
(mod mn) and we have the result.
Example 12.3 We shall solve the following system:
x ≡ 5 (mod 7)
x ≡ 6 (mod 11).
310 12 Cryptology
If x ≡ 5 (mod 7) we have that x = 5 + 7k for some k ∈ Z. Then we get 5 + 7k ≡ 6

(mod 11) which simplifies to 7k ≡ 1 (mod 11). Then multiplying both sides of the
equation by 8 we get k ≡ 8 (mod 11). This gives that k = 8 + 11s for some s ∈ Z.
Then
x = 5 + 7k = 5 + 7(8 + 11s) = 5 + 56 + 77 s = 61 + 77s.
This gives the answer that x ≡ 61 (mod 77).
It is imperative that the moduli are relatively prime to ensure that there is a unique
solution. For example, if one were to try to solve
x ≡ 1 (mod 4)
x ≡ 0 (mod 8),
we see immediately that there are no solutions to this system. Moreover, if one were
to try to solve
x ≡ 0 (mod 4)
x ≡ 0 (mod 6),
we would have as answers both 0 and 12 (mod 24).

Applying the Chinese remainder theorem inductively, we get the following corol-
lary.
Corollary 12.1 Let m1 , m2 , . . . , mt be natural numbers with gcd(mi , mj ) = 1 if

i = j. Then
x ≡ a1 (mod m1 )
x ≡ a2 (mod m2 )
..
.
x ≡ at (mod mt )
has a unique solution (mod m1 m2 · · · mt ).
Example 12.4 We shall solve the following system:
x ≡ 2 (mod 5)
x ≡ 5 (mod 9)
x ≡ 3 (mod 11).
We begin by solving the first two equations simultaneously. Starting with x = 2 + 5k,
we write 2 + 5k ≡ 5 (mod 9). This gives 5k ≡ 3 (mod 9). Multiplying both sides
by 2, we obtain k ≡ 6 (mod 9). This gives k = 6 + 9s and x = 2 + 5(6 + 9s) =

32 + 45 s. Therefore, the first two equations give that x ≡ 32 (mod 45). Then we
set 32 + 45 s ≡ 3 (mod 11), which gives s ≡ 4 (mod 11). Thus, s = 4 + 11t, and
x = 32 + 45(4 + 11t) = 212 + 495t. Finally, we have x ≡ 212 (mod 495).
For a much more general version of the Chinese remainder theorem and its appli-
cation to algebraic coding theory, see [29].
The following is a generalization of Theorem 2.10.
Theorem 12.3 If m and n are relatively prime, then φ(mn) = φ(m)φ(n).
Proof Let U r be the set of units in Zr . Then if a ∈ Umn then a is relatively prime
to m and relatively prime to n. Since m and n are relatively prime, there is a unique
solution to
x ≡ a (mod m)
x ≡ b (mod n)
for each a ∈ Um , b ∈ Un . Therefore, φ(mn) = φ(m)φ(n).
Example 12.5 Consider n = 4(3) = 12. Then numbers that are relatively prime to
4 are 1 and 3. By the previous theorem, we have φ(12) = φ(4)φ(3) = 2(2) = 4.
The numbers that are relatively prime to 3 are 1 and 2. Hence, writing numbers
as pairs modulo 4 and 3 and applying the Chinese remainder theorem, we have
CRT (1, 1) = 1, CRT (3, 1) = 7, CRT (1, 2) = 5, andCRT (3, 2) = 11 which are the
4 numbers relatively prime to 12.
Theorem 12.4 Let p be a prime. Then φ(pe ) = (p − 1)pe−1 .
Proof The only numbers that are not relatively prime to pe between 1 and pe − 1
are the pe−1 − 1 multiples of p. Thus, the number of elements relatively prime to
pe in this range is
pe − 1 − (pe−1 − 1) = pe − pe−1 = pe−1 (p − 1).
Example 12.6 Consider 27 = 33 , the numbers that are not relatively prime to 27
between 1 and 26 are 3, 6, 9, 12, 15, 18, 21, 24. Hence, the number that are relatively
prime is 26 − 8 = 18 = (3 − 1)32 .
Using the two previous theorems, we can now find the Euler φ function for any
positive integer.
312 12 Cryptology
e
Theorem 12.5 Let n = pi i , where pi is a prime with pi = pj if i = j. Then
e −1
φ(n) = (pi − 1)pi i .
e
Proof From Theorem 12.3, we have φ(n) = φ(pi i ). Then by Theorem 12.4,
e e −1
we have φ(pi i ) = pi i which gives the result.
We know by Fermat’s Little Theorem that if a ≡ 0 (mod p) then ap−1 ≡ 1

(mod p). But this does not mean that ai ≡ 1 for 0 < i < p − 1. For example, con-
sider any prime p > 3, then (p − 1)2 ≡ 1 (mod p) with 2 < p − 1. We are inter-
ested in those a for which this does not happen and we make the following definition.
Definition 12.1 Let p be a prime. An element a (mod p) is a primitive root

(mod p) if ap−1 ≡ 1 (mod p) but ai ≡ 1 (mod p) for 0 < i < p − 1.
Example 12.7 Consider the prime p. We shall give the smallest value for which ai ≡
1 (mod 7) for each of the non-zero values of a. We have 23 ≡ 1 (mod 7), 36 ≡ 1
(mod 7), 43 ≡ 1 (mod 7), 56 ≡ 1 (mod 7), and 62 ≡ 1 (mod 7). Therefore, 3 and
5 are primitive elements (mod 7).
Notice that in the previous example, there are two primitive roots and φ(7 − 1) =
2. We shall prove this result in general but first we need a lemma.
Lemma 12.1 Let a be a primitive root (mod p), then ai is a primitive root if and
only if gcd(i, p − 1) = 1.
Proof Assume gcd(i, p − 1) = 1 and (ai )j ≡ 1 (mod p). Then aij ≡ 1 (mod p)
giving that ij ≡ 0 (mod p − 1). Since gcd(i, p − 1) = 1 this implies that j ≡ 0
(mod p − 1). Then ai is a primitive root (mod p).
Assume gcd(i, p − 1) = 1 and (ai )j ≡ 1 (mod p). Then there exists k < p − 1
with ik ≡ 0 (mod p − 1). Then (ai )k ≡ aik ≡ 1. Then ai is not a primitive root
(mod p).
Theorem 12.6 The number of primitive elements (mod p) is φ(p − 1).
Proof First, we know that the multiplicative group of Fp is a cyclic group. This
means that the generator of the cyclic group a has the property that ap−1 ≡ 1
(mod p) and ah ≡ 1 (mod p) for 0 < h < p − 1.
By Lemma 12.1, we have that ai is a primitive root if and only if gcd(i, p − 1) =
1. Then the number of primitive roots is the number of elements less than p − 1 that
are relatively prime to p − 1, that is, φ(p − 1).
Example 12.8 Let p = 11, here 2 is a primitive root. We give the values of 2i for
all i.
i 0 1 2 3 4 5 6 7 8 9 10
2i 1 2 4 8 5 10 9 7 3 6 1
The numbers less than 10 that are relatively prime to 10 are 1, 3, 7, 9. Therefore, by
Lemma 12.1, we have that 2, 8, 7, and 6 are the four primitive roots (mod 11). We
note that φ(10) = 4.
Definition 12.2 Let a be a primitive root (mod p). Define La (b) = i where i is
the unique number with ai = b and 0 ≤ i < p − 1.
This function is defined for each prime but it is read (mod p − 1). In general, it
is very difficult to compute for large primes.
Example 12.9 Continuing Example 12.8, we have that L2 (3) = 8, L2 (5) = 10, and
L2 (9) = 6.
Assume a is a primitive root (mod p). Then the following rules apply. The proofs
are identical to the standard proofs for logarithms. Notice that these are all done
(mod p − 1) since φ(p) = p − 1.
• La (bc) = La (b) + La (c) (mod p − 1);

• La (bc−1 ) = La (b) − La (c) (mod p − 1);
• La (be ) = eLa (b) (mod p − 1).
12.3.1 RSA Cryptosystem
We shall now describe one of the most celebrated and profitable results in cryptogra-
phy. The discovery prompted a large industry using this technique to produce secure
Internet commerce. The RSA system was first given by Rivest, Shamir, and Adleman
in [75]. Most likely, every time you purchase something online, you are using the
RSA system. Virtually every time, very secure information is passed electronically,
you are using the RSA system. The strength of the system comes from the fact that it
is very difficult to factor large numbers, especially if they are the product of two very
large primes. While the theoretical foundations for these systems are, in fact, very
old, it would not have been possible to use this system before the advent of comput-
ers. This is because the computations that are used involve numbers with over 100
digits, so performing these computations by hand would have been impossible.
We begin by assuming we have a message that we wish to transmit. Any mes-
sage that one may want to send can be changed into a numeric message. There are
numerous techniques for doing this and they need not be cryptographically secure
to apply public-key encryption. Therefore, we shall assume that all of our messages
314 12 Cryptology
are numeric and we shall now show how to use this technique to send a numeric
message M.
Let p and q be primes. In practice, to make the system secure, these primes will
often be over 100 digits long. For added security, the primes can always be made to be
of longer length. It may seem that it is difficult to find primes that are over 100 digits
long, but in reality, it is quite easy. One can use a test like the Miller-Rabin primality
test, to find very large primes in a short period of time with very limited computer
time. It is not difficult to write a short program on a personal computer to find this
size of prime quite quickly. Then, we let n = pq. In practice, the number n will have
more than 200 digits. Again, this is no problem for any computer. It is immediate
from Theorems 12.3 and 12.4 that φ(n) = (p − 1)(q − 1). If we know p and q, this
is trivial to compute, but if you are handed a 200-digit number that is the product
of two large primes, then without knowing this factorization, it is computationally
infeasible to find φ(n). Next choose an integer e that is relatively prime to φ(n), of
course this is easy to find. There are practical considerations for choosing e but the
system will work for an e that is relatively prime to φ(n); however, certain values
of e make the system less secure.
Since e is relatively prime to φ(n), there exists a d with ed ≡ 1 (mod φ(n)) by
Lemma 2.2. That is, there exists some integer k with ed = φ(n)k + 1. In general,
this d is easy to find. One could use the Euclidean algorithm or one could simply use
Euler’s theorem, noting that eφ(φ(n)) ≡ 1 (mod φ(n)) so that eφ(φ(n))−1 e ≡ 1
(mod φ(n)). Then d = eφ(φ(n))−1 .
To send the message M, we calculate Me (mod n) and send this value. Note
that the value of e is told to anyone who wants it. If you want to send a message, all
you need to know is e and n and, of course, your message M.
The receiver, who knows d, takes
(Me )d ≡ Med (mod n)

≡ Mφ(n)k+1
≡ (Mφ (n))k M1
≡ 1k M ≡ M.
For the implementation of this technique, we allow e and n to be known by

anyone. Then, if you want to send M to the receiver, only the receiver knows p and
q, and therefore only they know φ(n) and d. Here, the key is public so that anyone
can send the information but only the receiver can decode it.
Public for RSA Private for RSA

n, e φ(n), d
The key to this system is that given the large number n = pq, in practice it is
at least (10100 )2 = 10200 , it is very hard to factor the number. If you were able to
factor n, then it is easy to compute φ(n) = (p − 1)(q − 1). It remains an extremely
difficult problem to factor large numbers even with a high-powered computer. One
possible way would be if a quantum computer were constructed, then Shor’s algo-
rithm would factor the numbers, see [79] for a complete description.
Example 12.10 We shall take a small example to illustrate the cryptosystem. We

shall use very small primes so that the reader can more clearly see the compu-
tations. Let p = 131 and q = 157. It is easy to verify that both of these num-
bers are prime. Compute n = 131(157) = 20567. Then φ(n) = (p − 1)(q − 1) =
130(156) = 20280. We choose e = 37. Then computing 37φ(φ(n))−1 = 4933, we
obtain 37(4933) ≡ 1 (mod 20280). Then d = 4933.
Let assume that the message we want to send is 22. Therefore, we send Me
(mod n) = 2237 (mod 20567) = 13524. When 13524 is received, the receiver
computes (Me )d (mod n) = 135244933 (mod 20567) = 22 and the correct mes-
sage is received.
12.3.2 El Gamal Cryptosystem
The next public-key cryptosystem that we shall discuss is the El Gamal Cryptosystem.
It is equally as powerful as RSA but is not used as much in actual practice, see [31].
While the security of RSA comes from the difficulty in factoring large numbers, the
security of El Gamal comes from the difficulty of taking a discrete logarithm.
We begin by letting p be a prime number. In practice, the prime p generally
has over 200 digits. As we stated before when discussing RSA it is not difficult
to find such primes. We recall that by Euler’s theorem or Fermat’s Little Theorem
that αp−1 ≡ 1 (mod p) for all α ≡ 0 (mod p). An element α is a primitive root
(mod p) if αk ≡ 1 (mod p) for all k with 1 ≤ k ≤ p − 2. We let α be a primitive
root (mod p).
As before, we can assume that any message can be seen as a numeric message;
therefore, let M be an integer that represents the message 0 ≤ M < p.
Let a be an integer which the receiver keeps secret. This integer must be between
1 and p − 1. Compute β = αa .
The sender chooses a secret integer k and computes r ≡ αk (mod p) and t ≡
β M (mod p). Then (r, t) is sent to the receiver.
k
The receiver computes
tr−a ≡ βk M(αk )−1 ≡ (αa )k Mα−ak ≡ M (mod M).
Then the message has been correctly decoded with perfect secrecy, since knowing
α and β, it is computationally impractical to compute the value of a.
Public for El Gamal Private for El Gamal

p, α, β a
Example 12.11 We shall illustrate this cryptosystem with a small example. Let p =
101. We choose α = 3 as our primitive root. We take a = 71 as our secret parameter.
316 12 Cryptology
Then β = 371 ≡ 51 (mod 101). We let p = 101, α = 3, β = 51 be made public.

Assume the message we want to send is 43, we pick our secret k = 13. Then r = αk
(mod p) = 313 (mod 101) = 38. Then t = βk M (mod p) = 5113 (mod 101) =
59.
The pair (38, 59) is sent. The receiver computes tr−a (mod p) = 59(38−71 )
(mod 101) = 59(3829 ) (mod 101) = 424, and the correct message is received.
12.3.3 Diffie-Hellman Key Exchange
Another technique which uses these ideas is the Diffie-Hellman key exchange. The
point of the Diffie-Hellman key exchange is not to exchange a large amount of
information but rather to establish a key that both parties can use for a system to
exchange information secretly. Assume Alice and Bob would like to establish a key.
They choose a prime p and an integer a which is a primitive root (mod p).
Alice chooses an exponent α and computes aα (mod p) and sends it to Bob. Bob
chooses an exponent β and computes aβ and sends it to Alice.
When Alice receives aβ she computes (aβ )α (mod p) ≡ aαβ (mod p). When
Bob receives aα he computes (aα )β (mod p) ≡ aαβ (mod p). Then both Alice
and Bob both have the number aαβ (mod p), and no one else could have this number
without knowing α and β and being able to compute a discrete logarithm. Notice
that Alice never knows what β is and Bob never knows what α is; however, they
both know aαβ which they can then use as a key.
Example 12.12 We shall show an example using small numbers. Let p = 13 and
let a = 2 be a primitive root in Z13 . Alice can take α = 5 and Bob can take β = 7.
Then Alice sends 25 = 6 and Bob sends 27 = 11. Alice receives 11 and computes
115 = 7. Bob receives 6 and computes 67 = 7.
Exercises
1. Solve the following systems of equations using the Chinese remainder theorem:
(a)
x ≡ 4 (mod 7)
x ≡ 5 (mod 11)
(b)
x ≡ 3 (mod 13)
x ≡ 7 (mod 17)
12.4 McEliece Cryptographic System 317
(c)
x ≡ 12 (mod 19)
x ≡ 15 (mod 22)
(d)
x ≡ 3 (mod 8)
x ≡ 2 (mod 13)
x ≡ 11 (mod 15)
2. Determine all primitive roots (mod 11) and (mod 13).

3. Determine L2 (a) for all values of a modulo 13.
4. Determine the number of primitive roots modulo 1601.
5. Determine φ(n) for all n with 2 ≤ n ≤ 25.
6. Using the prime 17, with a = 11, encode the message 9 using the El Gamal
cryptosystem. Determine r and t using k = 7, then decode the received message.
7. Using primes p = 11 and q = 17, with e = 7, encode the message M = 23 using
the RSA cryptosystem. Determine the parameter d and use it to decode the mes-
sage back to M.
8. Find a primitive root (mod 23) and use it in the Diffie-Hellman key exchange to
exchange the key 19 showing all steps.
12.4 McEliece Cryptographic System
In this section, we shall describe a cryptographic system that withstands quantum

attacks and is based on the theory of error-correcting codes. It was first described
by R. J. McEliece in [63]. This system uses the decoding of vectors using a given
code to uncover the message. It is certainly not used as much as the RSA or ElGamal
systems, but if a quantum computer were built then these two systems would become
immediately useless, as a quantum computer would defeat both of them. However,
at present, no quantum algorithm defeats this system so it would be a reliable system
should this occur.
We shall assume that the receiver chooses an [n, k, d] binary code, with d =
2t + 1, and finds a generator matrix for the code denoted by G. Of course, G has k
rows and n columns. We know that if d = 2t + 1 then the code can correct t errors
correctly using nearest neighbor decoding. This is a key step in implementing the
system. Then choose an invertible k by k matrix S and an n by n permutation matrix
P. Then take G = SGP which is again a k by n matrix. The matrix G is made
public.
318 12 Cryptology
The sender wants to send a message x, which is a vector of length k. The sender
picks a vector e with Hamming weight no more than t which acts as the error vector.
The sender forms y = xG + e and sends this as the message. If someone intercepts
this message, they must be able to determine what this error message is in order to
determine the sent message. It is obvious then that the system must be set up with
fairly large n and d to make the system effective.
The receiver performs the following computations:
1. Compute y = yP−1 = xSG + e , where e = eP. Note that e is also a weight t

vector since P is a permutation matrix.
2. Apply the decoding algorithm to y and correct it to x .
3. Compute the vector u such that uG = x .
4. Compute x = uS−1 .
It is immediate that the code we choose must have an efficient decoding algorithm
to make the system effectively computable. The security comes from the difficulty
of decoding y to x . In order to make this difficult, it is best to have a very large
d. As an example, the Goppa codes are often used, as their parameters are n = 2m ,
k = n − mt, and d = 2t + 1.
Example 12.13 The only codes that we have specified an easily computable decod-
ing algorithm in this text are the Hamming codes. Therefore, we shall use the [7, 4, 3]
Hamming code to illustrate the McEliece cryptographic system. Recall that the code
is given by the parity check matrix:
⎛ ⎞
0001111
H = ⎝0 1 1 0 0 1 1⎠. (12.2)
1010101
We shall use the matrix H in the decoding algorithm. We shall take as the generator
matrix the following matrix:
⎛ ⎞
1000011
⎜0 1 0 0 1 0 1⎟
G=⎜ ⎟
⎝0 0 1 0 1 1 0⎠. (12.3)
0001111
We note the first three rows correspond to lines in PG2 (F2 ) and the last row corre-
sponds to a hyperoval. Next, we choose a 4 by 4 invertible matrix S:
⎛ ⎞
1001
⎜0 1 1 1⎟
S=⎜ ⎟
⎝1 1 0 0⎠ (12.4)
1111
12.4 McEliece Cryptographic System 319
whose inverse is
⎛ ⎞
0101
⎜0 1 1 1⎟
S−1 =⎜ ⎟
⎝1 1 1 0⎠. (12.5)
1101
Next, we need a 7 by 7 permutation matrix:
⎛ ⎞
0000010
⎜0 0 0 1 0 0 0⎟
⎜ ⎟
⎜0 1 0 0 0 0 0⎟
⎜ ⎟
P=⎜ ⎟
⎜0 0 0 0 0 0 1⎟. (12.6)
⎜1 0 0 0 0 0 0⎟
⎜ ⎟
⎝0 0 1 0 0 0 0⎠
0000100
We note that there is precisely one 1 in each row and column. Next we compute
G = SGP, which is
⎛ ⎞
1000011
⎜1 1 0 1 0 0 1⎟
G = ⎜ ⎟
⎝1 0 1 1 0 1 0⎠. (12.7)
1111111
This matrix is made public.
Let us suppose we wish to send the message x = (1, 1, 1, 0). We compute xG
which is (1, 1, 1, 0, 0, 0, 0). Next we pick an error message with t = 1 error. We
choose e = (0, 0, 0, 0, 0, 1, 0). This gives y = (1, 1, 1, 0, 0, 1, 0) which is the sent
message.
The receiver takes this message and begins by computing y = yP−1 = (1, 0, 1, 0,
T
⎞ the decoding algorithm to thiss vector, that is, compute H(y ), which
⎛ Apply
1, 1, 0).
0
gives ⎝ 0 ⎠ . The error e is (1, 0, 0, 0, 0, 0, 0) since it occurs in the first coordinate.
1
Hence, the corrected vector is x = (0, 0, 1, 0, 1, 1, 0). Then u = (0, 0, 1, 0) since
uG = x . Finally, we compute uS−1 = (1, 1, 1, 0) which was the sent message.
Hence, we have secretly sent this message via this public-key encryption method.
Exercises
1. Prove that a binary k by k matrix is invertible if and only if the determinant is

1.
2. Construct an example of the McEliece cryptographic system using any binary
code.
3. Compute the determinant of the matrix S given in Eq. 12.4.
4. Verify that S and S−1 given in Eqs. 12.4 and 12.5 are inverse matrices.
5. Find the inverse of the matrix P given in Eq. 12.6.
Games and Designs
13
Games have often given rise to interesting combinatorial questions. For example,
determining whether there is a winning strategy in chess has proven to be an extremely
difficult mathematical question. Part of the reason for the difficulty is the combina-
torial explosion that we have noticed before. Chess is played on an 8 by 8 board with
32 pieces. These seemingly small numbers are large enough to make it so that the
number of possible games is far larger than any person or computer can handle. It is
estimated that the number of possible chess games is larger than 10120 which is far
larger than any computer could evaluate even if running for billions of years. More-
over, the game theory behind programming a computer to play chess well requires a
great deal of combinatorics as well as some very clever programming.
Games can also be useful in developing geometric and combinatorial intuition. In
this chapter, we shall describe a game using designs and a technique for determining
if the game has a winning strategy. The game we shall describe is a generalization
of a game played on finite planes that was introduced in the paper “Tic-Tac-Toe on
a Finite Plane” [19]. This game was further developed in [20,21].
We shall begin by describing some basics of combinatorial game theory. Then
we shall describe the specifics of the game. We introduce weight functions and use
them to examine the strategies of the game.
Some games that people play often have a probabilistic aspect to them. For exam-
ple, the play in poker or bridge is dependent upon the cards that each player receives
in the deal. These games require a different approach than a game that is purely
combinatorial. As an example, in a game of poker a player can attempt to bluff his
opponent by making him think that his hand is much better than it actually is. How-
ever, the same sort of bluffing is not possible in a game of checkers where all of the
information of the game is available to each player. In these games of perfect infor-
mation, there is no way to convince your opponent that your position is any different
than it actually is because nothing is hidden. There is a large literature devoted to
322 13 Games and Designs
probabilistic games and combinatorial games. There are also numerous applications
of these games in economics and political science. For a reference on game theory,
we suggest Rapoport’s book “Two-Person Game Theory” [70] or Straffin’s book
“Game Theory and Strategy” [88] and for a description of combinatorial games we
suggest “Winning Ways for your Combinatorial Plays” by Elwyn R. Berlekamp,
John Horton Conway, and Richard K. Guy [8].
We shall only consider games like chess or checkers that have no probabilistic
aspect and where there is perfect information. In other words, each player knows the
complete situation of the game, there is nothing hidden. Games can also be played
by various numbers of players. We consider only games played by two people.
We shall describe some of the basic terms of game theory. A strategy is an algo-
rithm for play. Specifically, given a specific state of the game a strategy will determine
what play the player should make at that point. It is called a winning strategy if it
assures that its application will give a win for the player, and it is called a drawing
strategy if it assures that a player can force a draw. For the positional games that we
will describe, there are two possibilities. The first is that the first player has a winning
strategy. The second is that the second player can force a draw. The reason for this is
what is called strategy stealing. If the second player could have a winning strategy,
then the first player could simply assume that a play has been made and then use the
strategy that the second player would have used as their own. While this does not
work for all games, it does apply for the positional games that we describe.
13.1 The Game
Most children have played the game Tic-Tac-Toe. The game is played on a 3 by
3 grid where the first player marks with an X and the second with an O, and they
continue until someone completes a horizontal or vertical line or a diagonal or until
each space is filled. Anyone who has played the game realizes after about 5 min. that
the second player can force a draw. We shall generalize this game to be played over
a design.
Let D = (P, B, I) be a t − (v, k, λ) design. The first player plays by marking a
point in the design with an X. Player 2 plays by marking a point with an O. A player
wins the game if they have marked all of the points on a block of the design. The
game is a draw if all points have been marked and no block has all points with the
same mark.
As an example consider the game played on the projective plane of order 2.
Throughout this section, we shall denote the i-th play of the first player by Xi and
the i-th play of the second player by Oi .
This is a win for the first player since X1 , X3 and X4 complete a line in the plane
(Fig. 13.1).
We shall now show that X has a winning strategy on the projective plane of order
2. The first player places an X on any of the seven points. Then O places a point
on any of the six remaining points. At this point the first player places an X on the
remaining point of the line through the point chosen by the first and second player.
13.1 The Game 323
Fig. 13.1 A win for X
Fig. 13.2 A win for X
This line now has two points with an X and one with an O. The second player now
places an O anywhere. The first player places an X on the line between the two
points containing an O. This prevents O from completing a line. Now there is a line
between X1 and X3 and a line between X2 and X3 . The second player can block only
one of these lines, and hence X4 will be a win for X. This is exactly what happened
in the previous example.
We shall now show that the second player has a drawing strategy on the biplane
of order 2. The second player simply places their first two points anywhere. After
the first players’ third move, there are two remaining points on the biplane. At least
one of them will not be on the line through O1 and O2 when viewing the biplane as
a projective plane of order 2. The second player places an O on that point. Then the
first player cannot complete a block since each block is the complement of a line.
Thus, the second player has a drawing strategy on the biplane of order 2. For the
biplane of order 1, it is easy to see that there is a drawing strategy since a block has
three points and the first player only has two plays.
Consider the game in Fig. 13.2 on the affine plane of order 3. Even though it has
the same number of points as the usual game of Tic-Tac-Toe, it is quite different.
The usual strategy of the second player fails to force a draw.
A draw is a configuration on the plane such that each point is marked X or O in
which no block has each point marked the same.
The fundamental question involved in this game is for which designs is there a
winning strategy and for which are there drawing strategies.
Exercises
1. Prove that the second player can force a draw for the standard game of Tic-Tac-
Toe.
2. Use the strategy exhibited in the previous to prove that the first player has a
winning strategy on the affine plane of order 3.
3. Prove that there are no draws in the projective plane of order 2 and the affine
planes of orders 2 and 3.
4. Prove that first player has a winning strategy on the affine plane of order 2.
13.2 Weight Functions
In this section, we shall describe a technique which will enable us to determine in

most cases whether there is a winning strategy or not. The technique uses weight
functions defined on a stage of the game. This is a standard technique used in the
theory of games. For example, in chess, a weight function can be developed to
decide what the next move should be. Namely, it gives a weight to the strengths and
weaknesses of a given move in terms of which pieces you can now attack and from
which you are now open to attack. Therefore, when teaching a computer to play
chess, the weight of the board for all possible moves can be examined and then the
move that gives the most favorable weight is chosen. We shall make a similar type
of weight function, which we shall use to determine where the next move should be.
We will be able to show that in many instances, playing by the results of this weight
function, we will be able to show that either the first player has a winning strategy
or the second player will have a drawing strategy.
The following result of Erdös and Selfridge [32] gives conditions for when the
second player can force a draw in many positional games, of which our version of
Tic-Tac-Toe is an example.
Theorem 13.1 Let P be a finite set of points. Let L = {A1 , A2 , . . . , Am } ⊆ 2P be a

collection of winning sets, where |Ai | = k for all 1 ≤ i ≤ m. Two players take turns
selecting points from P until either one player has all of the points of a winning set
or there is a draw. If m < 2k−1 , then the second player can force a draw.
The proof, given later, employs the following sets and weight functions. We shall
adopt the techniques to our own setting. Namely, we let D = (P, B, I) be a design.
Assume the current state of the game is [(X1 , . . . , Xi ), (O1 , . . . , Oi−1 )], that is,
Xi is the point marked by the first player in the i-th move and Oi is the point marked
by the second player in the i-th move. Let B represent the set of blocks and let
Bi = { − {X1 , . . . , Xi } | ∈ L, ∩ {O1 , . . . , Oi−1 } = ∅}. In other words,the set
Bi consists of all partial blocks that are not blocked by the second player as of the
i-th move of the second player. Recall that if there is an O placed on even one point
of the block, then it is blocked since the first player can not win by completing this
block.
13.2 Weight Functions 325
Therefore, Bi contains sets of points, s, where s ⊆ b for some block b ∈ B.

Additionally, we define B∞ to be the set when no more moves can be made in the
game. It is clear that if ∅ ∈ Bi , then the first player has won since there is a block
that has an X on each of its points.
We define Pi = P − {X1 , . . . , Xi , O1 , . . . , Oi−1 }. Namely, it is the set of
unmarked points at the i-th move of the second player.
For points p, q ∈ Pi , we define weight functions. What they are essentially giving
is the weight given that the position of play is Bi and then a player marks the point
q. They are defined by

w(Bi ) = 2−|s| , (13.1)
s∈Bi

w(q|Bi ) = 2−|s| , (13.2)
s∈Bi ,q∈s
and

w(p, q|Bi ) = 2−|s| . (13.3)
s∈Bi ,{p,q}⊆s
Theorem 13.2 We have that in any design,
w(Bi ) − w(Bi+1 ) = w(Oi |Bi ) − w(Xi+1 |Bi ) + w(Xi+1 , Oi |Bi ). (13.4)
Proof The left side of the equation gives the change in the weight from Bi to Bi+1 .
The right side represents a standard inclusion-exclusion counting technique. It adds
the weights of all partial blocks that contain Oi since those blocks are now blocked. It
subtracts the weights of the blocks through Xi+1 since these weights are now 2−|s+1|
in the new configuration. Then it adds the weight of the block that contained both
points (if such a block had not yet been blocked).
Notice that the closer the first player is to winning on a certain block, the higher
its corresponding weight is.
The relevant point is that if ∅ ∈ Bi , then w(Bi ) ≥ 1 since 20 = 1 and the first
player has won the game. If w(Bi ) ≤ 1, then ∅ ∈ / Bi and so the first player has not
won. This leads to a natural way for the second player to play the game. Namely,
the second player plays by minimizing the value of w(Bi ) after the second player
makes their i-th move. If it is possible for the second player to keep this weight less
than 1, then they have an algorithm to force a draw.
Consider Eq. 13.4. The second player can make their choice so that
w(Oi |Bi ) ≥ w(Xi+1 |Bi ).

In fact, this is how the second player chooses their play by maximizing this weight.
This shows that the second player can play so that the weights w(Bi ) are a non-
increasing sequence. Therefore, if the second player can ever get the weight of the
game below 1, they can be assured of forcing a draw.
We can now give a proof of Theorem 13.1. Although we have defined everything
in terms of designs, it is easy to see that it extends to the more general notation of
the theorem.
Proof of Theorem 13.1
Proof Consider the weight of the game after the first player has made a single move.
We notice that some blocks have an X on a single point and some do not. Hence, the
size of the sets in B1 are either k − 1 or k. But we know that 2−(k−1) > 2−k . This
gives the following:

w(B1 ) ≤ 2−(k−1)
b∈B
= |B|2−(k−1)
m
= k−1 .
2
m
Since we have assumed that m < 2k−1 , we know that 2k−1 < 1 and thus w(B1 ) < 1.
The above discussion now gives that the second player has a drawing strategy.
It is an immediate corollary that every design where the second player has a
drawing strategy can be marked so that each point has either an X or an O and
no block has all the same marking. In other words, we are able to show that a draw
exists. We are able to conclude this combinatorial information from a game-theoretic
argument.
We shall examine the situation for various designs that we have encountered in
the text.
13.2.1 The Game on Planes
The situation for planes was described in the paper [19]. We have already seen that
there is a winning strategy for the first player for the projective plane of order 2 and
the affine planes of orders 2 and 3.
Projective planes of order n have n2 + n + 1 winning sets and n + 1 points in
a winning set. For projective planes of order n > 4, we have n2 + n + 1 < 2n and
so there is a drawing strategy for the second player in these cases.
Affine planes of order n have n2 + n winning sets and n points in a winning set.
For projective planes of order n > 6, we have n2 + n + 1 < 2n and so there is a
drawing strategy for the second player in these cases. Of course, there is no plane of
order 6, but this tells us that there is a drawing strategy for all affine planes of order
n ≥ 7.
This leaves us with the cases of the affine planes of orders 4 and 5 and the projective
planes of orders 3 and 4.
We can examine the weight function more closely to examine the projective plane
of order 4. After the first player places an X, there are n2 lines with no markings and
n + 1 with a single X. Then
w(Bi ) = (n2 )2−(n+1) + (n + 1)2−n

= 16(2−5 ) + 5(2−4 )
13
= < 1.
16
This gives that there is a drawing strategy on this plane.
The affine plane of order 4 is perhaps the most interesting case. It has draws
yet the first player has a winning strategy. The fact that the first player has such a
strategy was shown by computer, independently by three different people (see [19],
for details). As far as actual play, this is probably the most interesting plane on which
to play.
We summarize the results in the following theorem.
Theorem 13.3 The second player can force a draw in all projective planes of order
n > 2, and the first player has a winning strategy on the projective plane of order 2.
The second player can force a draw in all affine planes of order n > 4, and the first
player has a winning strategy on all affine planes of order n ≤ 4.
13.2.2 The Game on Biplanes

2 +3n+4
Recall that a biplane of order n is a symmetric 2 − ( n 2 , n + 2, 2) design.
Theorem 13.4 There is a drawing strategy for all biplanes.
Proof We have shown that there is a drawing strategy for the biplane of order 1 and
for the biplane of order 2 (which will follow from the next argument as well). The
2
size of the blocks is n + 2, and the number of blocks (winning sets) is n +3n+4
2 . If
n ≥ 2 we have that
n2 + 3n + 4
< 2n+1
2
and so by Theorem 13.1 there is a drawing strategy.
13.2.3 The Game on Symmetric Designs
In this subsection, we shall only consider λ > 2 since we have already considered
the other cases.
(n+λ)(n+λ−1)
In a symmetric design, there are λ blocks and each block has n + λ
points.
(n+λ)(n+λ−1)
It is easy to see that for λ > 2 we have λ < 2n+λ−1 for all n. Then
applying Theorem 13.1 gives the following result.
Theorem 13.5 The second player has a drawing strategy on all symmetric designs
with λ > 2.
13.2.4 The Game on Hadamard Designs
We shall now consider the case of Hadamard 3-designs and 2-designs. The Hadamard
3-design of order 1 consists of four points with six blocks each containing two points.
In fact, every subset of size 2 of the four points is a block. It is easy to see that the
first player has a winning strategy since their second move guarantees a win. In fact,
there are no draws on this design. The Hadamard 2-design of order 1 consists of
three points and three blocks that consist of one point. Here, the first move gives a
victory and again there are no draws. The Hadamard 2-design is a (7, 3, 1) design,
and we have already seen that X has a winning strategy there.
We need to examine the Hadamard 3-design of order 2 which is a 3 − (8, 4, 1)
design with 14 blocks. We see that 14 is not less than 23 = 8, and so the theorem does
not apply. The design can be thought of as the completion of the Hadamard 2-design
which is a (7, 3, 1) design. That is, we add a single point, and the new blocks are the
old blocks adjoined with this new point and the compliment of old blocks. A win
for X would consist of four points on a block and X has exactly four plays. Hence,
each of the plays must be on a single block. We know that through any three points
there is one block. After the first player places his third X, the second player need
only marks the remaining point on this block if it has not already been marked with
an O. Thus, the second player has a drawing strategy in this case.
A Hadamard 3-design is a design with 8n − 2 blocks which are winning sets each
of size 2n. For n > 2, we have 8n − 2 < 22n−1 and so by Theorem 13.1 there is a
drawing strategy for the second player.
A Hadamard 2-design is a design with 4n − 1 blocks which are winning sets each
of size 2n − 1, for n > 2 we have 4n − 1 < 22n−2 and so by Theorem 13.1 there
is a drawing strategy for the second player.
Theorem 13.6 The first player has a winning strategy for the Hadamard 3-designs
of order 1, and the first player has a winning strategy for the Hadamard 2-designs of
orders 1 and 2. For n > 2, the second player has a drawing strategy for all Hadamard
2-designs and for n > 1 the second player has a drawing strategy for Hadamard
3-designs of order n.
13.2.5 The Game on Nets
We shall examine the situation for k-nets of order n.
Theorem 13.7 If a k-net of order n extends to a k -net of order n where there is a

drawing strategy, then there is a drawing strategy on the k-net.
Proof Since the point sets are the same, you can simply apply the same strategy.
Since a winning set in the net is also a winning set in the larger net, then the strategy
that forces a draw on the larger net will force a draw on the net as well.
To apply Theorem 13.1, we need
kn < 2n−1 .
We note that k is bounded above by n + 1 and so for n > 6 there is a drawing strategy
on all k-nets of order n.
We know for n = 6 the maximum value of k is 3. In this case 18 < 32 and so
there is a drawing strategy on all nets of order 6 by Theorem 13.1.
For n = 5 if k = 4 or 5, then we know the net completes to an affine plane. Then
by Theorem 13.7, the net must have a drawing strategy. For k = 3 we have 15 < 16
and so by Theorem 13.1 there is a drawing strategy. Theorem 13.7 gives that there
is a drawing strategy for k = 1 and k = 2. Therefore, there is a drawing strategy for
all k-nets of order 5.
Consider a 3-net of order 3. The first player places X1 arbitrarily and the same
for O1 . Then the first player places X2 anywhere that is on a line with X1 but not
on a line with O1 . Then O2 must be placed to block this line. The first player can
then place X3 on the intersection of the lines through X1 and X2 and through X1
and X3 . The second player can only block one of these lines, and the first player has
a winning strategy. Notice that this game has only one more winning set than the
usual game of Tic-Tac-Toe, but in reality it is quite different. In the usual game, the
center point is special because it has 4 winning sets through it. On a 3-net of order 3
no such point exists. Of course we know that there is a winning strategy on a 4-net
of order 3.
Consider a 3-net of order 4. After the first player places X1 arbitrarily, there are
three lines that have three unmarked points and nine that have no marked points.
This gives that
1 1 15
w(B1 ) = 3( ) + 9( ) = <1
8 16 16
and so the second player has a drawing strategy.
We have already seen that the first player has a winning strategy for a 5-net of
order 4.
Theorem 13.8 The second player can force a draw on all k-nets of order n with
n > 4 and for all k-nets with k ≤ 2 except for the 2-net of order 2 in which there is a
winning strategy. There is a winning strategy for all k-nets of order 3 if k > 2. There
is a drawing strategy for k-nets of order 4 for 1 ≤ k ≤ 3 and a winning strategy for
k = 5.
13.2.6 The Game on Steiner Triple Systems
For the Steiner triple system with three points, there is obviously a drawing strategy.
Wherever the second player places, their O blocks the only block in the system.
Steiner triple systems are at least a 2-design. Thus for v > 3 the same winning
strategy that the first player used on the projective plane of order 2 will work. That
is, the first two moves are arbitrary, then X2 is placed on the block with X1 and O1 .
After any placement of O2 , the first player places X3 on the block with O1 and O2 .
Then, the second player can only block either the block between X1 and X3 or the
block between X2 and X3 but not both. Then, the first player has a winning strategy.
We summarize this as follows.
Theorem 13.9 There is a drawing strategy for the second player for the Steiner
triple system with v = 3 and a winning strategy for the first player for all others.
Exercises
1. Change the definition of a win from marking all n + 1 points on a line on a finite
projective plane, to marking at least n points on a line of a finite projective plane.
Determine if the first player has a winning strategy or the second player has a
drawing strategy in this scenario for the projective planes of order 2, 3, and 4.
2. Change the definition of a win from marking all n points on a line on a finite
affine plane, to marking at least n − 1 points on a line of a finite projective plane.
Determine if the first player has a winning strategy or the second player has a
drawing strategy in this scenario for the affine planes of order 2, 3, and 4.
3. Prove that there is a drawing strategy on a 1-net of order 2 and a winning strategy
on all other nets of order 2.
4. Prove that there is a drawing strategy of k-nets of order 3 when 0 < k ≤ 2.
5. Prove that there is a drawing strategy on the projective plane of order 3. Proceed
as follows. Assume the first three plays are made arbitrarily. Then, the second
player makes his second play by placing an O on the line through X1 and X2
unless it has already been done in which case O2 is placed off this line. Then
X3 can be placed arbitrarily giving rise to four possible configurations. Compute
w(B3 ) in each of these cases and show that it is less than 1.
6. Construct a draw on the affine plane of order 4.

7. Using the same initial configurations in the previous, show that the second player
has a drawing strategy on the affine plane of order 5. The situation becomes much
more complicated, and it will probably be necessary to use a computer to evaluate
all possible configurations.
8. Describe the drawing strategy on the biplanes of order 3 without using the weight
functions, that is, describe it so that a person playing could use the technique to
force a draw.
9. Prove that draws exist for all affine planes of order n ≥ 3.
10. Show how to use a transversal to a k-net of order n, k ≤ n, to construct a draw
on the net.
Epilogue
14
Early in the text we encountered the following diagram:
Throughout, we have seen this image as a representation of many of the objects

studied in the text. Specifically, it represented the following:
• the projective plane of order 2,

• the biplane of order 2,
• the Hadamard 2-design coming from a Hadamard matrix of size 8,
• a Steiner triple system on seven points,
• the geometric representation of the [7, 4, 3] Hamming code, and
• a design formed from a difference set.
It is precisely this kind of interesting object which fuels the study of finite geometry
and combinatorics. We can see many different ways of approaching the object and
many different classes of objects for which this is the first example in the class. Often
in the study of mathematics in general and combinatorics, in particular, different paths
of inquiry will lead to the same ideas.
Numerous different questions in geometry, statistics, algebra, and information
theory have given rise to many interesting combinatorial structures which we have
attempted to illustrate in this text. Students wishing to further their study have a
variety of paths to follow.
334 14 Epilogue
1. For a classical text on combinatorics, see Hall’s text Combinatorial theory [44].
2. To learn more about Latin squares, see the book Latin squares and their applica-
tions by Keedwell and Dénes [54]. To learn about the connection between Latin
squares and many topics of discrete mathematics, see Discrete mathematics using
Latin squares by Laywine and Mullen [59].
3. To learn more about coding theory, see the text by Huffman and Pless, Funda-
mentals of error-correcting codes [49].
4. To learn about the connection between coding theory, finite geometry and designs,
see the text by Assmus and Key, Designs and their codes [4].
5. To learn more about cryptography, see the text Introduction to cryptography with
coding theory by Trappe [92], which is an introduction to many topics in cryp-
tography. To learn more in this area, see the advanced text Cryptography. Theory
and practice by Stinson [85].
6. To learn more about combinatorial games, see the landmark text by Berlekamp,
Conway and Guy, Winning ways for your mathematical plays [8].
7. To learn more about combinatorial designs, see Stinson’s text Combinatorial
designs. Constructions and analysis [86]. For an encyclopedic description of
designs, see the text edited by Colbourn and Dinitz The CRC handbook of com-
binatorial designs [24].
8. To learn more about matrices in combinatorics, see the text by Brualdi and Ryser
Combinatorial matrix theory [15].
At present, numerous open questions remain in almost every topic examined in

this text, and a large number of new applications are being introduced in information
theory, cryptography, computer science, and statistics every year. In general, math-
ematics often progresses in an attempt to solve major questions. It has often been
suggested that what a pure mathematician wants is a ridiculously difficult problem
on which incremental advances can be made.
The list of guiding problems that we have encountered have included the
following:
1. Determine the maximum number of MOLS of order n for all n.

2. Determine for which n there exists affine and projective planes of order n.
3. Determine the number of non-isomorphic planes for a given order n.
4. Determine when non-Desarguesian planes exist.
5. Determine precisely for which parameters there exists t-(v, k, λ) designs.
6. Determine for which n Hadamard matrices of order n exist.
Solutions to Selected Odd Problems
B.1 Problems of Chap. 1
Section 1.1
(1) If n is even, use the function from the odds to the evens defined by f(i) = i + 1.
This is a bijection and shows that the cardinalities of the two sets are the same. If
n is odd, then use f−1 . This function is not surjective, as it leaves n untouched.
Therefore, there are more odds than evens in this case.
(3) The image of f must have the same cardinality as the domain since the function
is injective. This gives that the cardinality of the image and codomain are the
same. Then, since the sets are finite, we have the result.
(5) If it were finite then Exercise 3 would give that the map must be surjective.
(7) If f is a bijection from A to {1, 2, . . . , n} and g is a bijection from A to
{1, 2, . . . , m}, then f ◦ g−1 is a bijection from {1, 2, . . . , m} to {1, 2, . . . , n}
p! p(p−1)!
(9) We have C(p, i) = (p−i)!i! = (p−i)!i! . If 0 < i < p, then the integer p does
not appear in the denominator, and hence there is nothing to cancel the p in the
numerator. Since we know this value must be an integer we have the result.
(11) We have
C(2n, n) = C(2n − 1, n) + C(2n − 1, n − 1)

(2n − 1)! (2n − 1)!
= +
(n − 1)!n! n!(n − 1)!
(2n − 1)!
=2 .
(n − 1)!n!
336 Solutions to Selected Odd Problems
(13) The number is 2k−1 .

(15) The total number is 356 = 1, 838, 265, 625.
(17) C(17, 6) = 12, 376.
(19) C(16, 5) = 4, 368.
(21) The number of games is 9! = 362, 880. The number of final configurations is
C(9, 5) = 126.
(23) Straightforward induction.
(25) 105, 869, 167, 644, 240, 000. See Chap. 12 for a complete description.
(27) There are 12 ways where switching works and 6 ways where switching does
not work. Therefore one should switch. See Chap. 9 for a complete description
of this problem.
Section 1.2
(1) 840.
(3) 10, 080, 3, 326, 400, 34, 650.
Section 1.3
(1) We have 3(8) + 1 = 25 and 27 > 25.

(3) 5(6) + 1 = 31.
(5) The boxes are 4-unit-length line segments forming the line. Then 3(4) + 1 = 13.
(7) 7(11) + 1 = 78.
Section 1.4
(1) (a) (1, 2, 3, 4, 5).

(b) (3, 1, 5, 2, 4).
(c) (1, 6).
(d) (1).
(e) (1).
(f) (2, 5)(1, 3)(4, 6)(7, 8).
(3) We have (στ)(τ−1 σ−1 ) = σ(ττ−1 )σ−1 = σσ−1 = (1).
(5) Prove it by contradiction. If it could be, then write it in both ways and equate
them. Then multiply both sides by the inverse of one of them. Then you would
have that the identity is written as an odd permutation which is a contradiction.
(7) To prove that it is injective, ax = ay implies x = y by multiplication on
both sides by a−1 , which is possible since it is a field. Surjection follows by
a(a−1 y)) = y. The second proof follows from the fact that σa and σb send
1 to a and b, respectively. Hence, they cannot be equal unless a and b are
equal.
(9) The counts are 26!, (262 )! and (263 )!.
Solutions to Selected Odd Problems 337
Section 1.5
(1) 6.
(3) 13.
(5) 8.
(7) Assume the limit is L. Then
Fn+1
lim =L
n→∞ Fn
Fn + Fn−1
lim =L
n→∞ Fn
Fn−1
lim 1 + =L
n→∞ Fn
1
1+ = L
L
L + 1 = L2 .
Applying the quadratic formula gives L = ϕ.
Section 2.1
(1) (a) 1 (b) 4 (c) 9 (d) 1 (e) 8.

(3) Apply the binomial theorem to get
p

p
(a + b) = C(p, k)xp−k yk .
k=0
p!
If 0 < k < p, then C(p, k) = k!(p−k)! is divisible by p, and hence is 0
(mod p). If k = 0 we get xp and if k = p we get yp . This gives the result.
(5) Consider 1 + 1 = 2 in Z12 .
(7) If ab ≡ 0 (mod n), n a prime, then if a ≡ 0 (mod n) there exists a−1 sat-
isfying a−1 a = 1. Multiplying both sides of the equation, we get a−1 ab = 0
which gives b = 0. Therefore, one of a or b must be 0.
(9) In Z4 the solutions are 1, 3. In Z8 the solutions are 1, 3, 5, 7. In Z16 the solutions
are 1, 7, 9, 15.
(11) (a) 1 (b) 3 (c) 1 (d) 4 (e) 12.
(13) (a) 15 (b) 8, 20 (c) 13 (d) 8 (e) 4, 9, 14.
Section 2.2
(1) Straightforward verification.

(5) It is immediate that Φ(ab) = (ab)p = ap bp = Φ(a)Φ(b). Exercise 3 in
Sect. 2.1 gives that Φ(a + b) = Φ(a) + Φ(b).
(7) The element (0, 1) is not (0, 0) and has no multiplicative inverse. Therefore, it
is not a field.
(9) We have: 1−1 = 1, 2−1 = 6, 3−1 = 4, 4−1 = 3, 5−1 = 9, 6−1 = 2, 7−1 = 8,
8−1 = 7, 9−1 = 5, 10−1 = 10.
(11) The unique solution is x = a−1 (c − b).
(13) If ba−1 = 1. Then multiplying by a on the right gives b = a.
(15) Let c = a−1 b.
(17) If x3 + x + 1 were reducible then either 1 or 0 would be a root since it would
factor into a degree 1 and a degree 2 polynomial. However, neither is a root. The
field of order 8 is then F2 [x]/x3 + x + 1 = {0, 1, x, x + 1, x2 , x2 + 1, x2 +
x, x2 + x + 1}.
(19) It is easy to see that the case u2 = 0 is not a field since there are zero divisors.
For the case u2 = u, it is not a field since u(1 + u) = 0 and there are zero
divisors. The case where u2 = 1 gives (1 + u)2 = 0, and we have zero divisors
and is therefore not a field. The case where u2 = 1 + u gives the field of order
4 realized as F2 [x]/x2 + x + 1.
Section 2.3.1
(1) The recursive definition is a0 = 0, an+1 = an + 4n3 + 6n2 + 4n + 1. The

proof is straightforward.
(3) The final digit must be 0, 1, or 4.
Section 2.3.2
(1) Straightforward.
Section 2.3.3

n n n n
(1) If k = n, = 1 and = 1. If k = n − 1, = n and = n.
n k n−1 k
Section 2.3.4
(1) Simply place each disk on a separate pole, placing the largest on the desired pole.
Then each disk is moved twice except the largest giving 2n − 1 moves.
Section 3.1
(1) Let Lij = i + j. Then Lij = Lij ⇒ i + j = i + j ⇒ j = j and Lij = Li j ⇒

i + j = i + j ⇒ i = i . Therefore, it is a Latin square.
(3) Let Lij = ij. Then Lij = Lij ⇒ ij = ij ⇒ j = j and Lij = Li j ⇒ ij =
i j ⇒ i = i . Therefore, it is a Latin square.
(5) Begin by showing that a Latin square of order 3 is equivalent to either
⎛ ⎞ ⎛ ⎞
012 012
⎝ 1 2 0 ⎠ or ⎝ 2 0 1 ⎠
201 120
and the result follows.

(7)
⎛ ⎞ ⎛ ⎞
01234 01234
⎜1 2 3 4 0⎟ ⎜2 3 4 0 1⎟
⎜ ⎟ ⎜ ⎟
⎜2 3 4 0 1⎟ ⎜4 0 1 2 3⎟
⎜ ⎟ ⎜ ⎟
⎝3 4 0 1 2⎠ ⎝1 2 3 4 0⎠
40123 34012
⎛ ⎞ ⎛ ⎞
01234 01234
⎜3 4 0 1 2⎟ ⎜4 0 1 2 3⎟
⎜ ⎟ ⎜ ⎟
⎜1 2 3 4 0⎟ ⎜3 4 0 1 2⎟.
⎜ ⎟ ⎜ ⎟
⎝4 0 1 2 3⎠ ⎝2 3 4 0 1⎠
23401 12340
(9) Let Lij = ai + j. Then Lij = Lij ⇒ ai + j = ai + j ⇒ j = j and Lij =
Li j ⇒ ai + j = ai + j ⇒ i = i since gcd(a, n) = 1. Therefore, it is a Latin
square. When gcd(a, n) = 1 and ai − ai = 0 gives a(i − i ) = 0. Let i − i
be b where ab = 0, and we have that it is not a Latin square in that case.
e e e e
(13) There exists p11 − 1 MOLS of order p11 and p22 − 1 MOLS of order p22 .
Then the minimum number is the minimum number of these two numbers by
the cross-product construction.
Section 3.2
(1) We have Lij = i + j = j + i = Lji .

(3) Lij = ij = ji = Lji .
(5) Fix any two columns. Each pair in these two columns must have a unique number
in the third column. The result follows from this fact.
(7) Let ⊥ indicate that two squares are orthogonal.

⎛ ⎞ ⎛ ⎞
012 012
L = ⎝2 0 1⎠ ⊥ ⎝1 2 0⎠
120 201
⎛ ⎞ ⎛ ⎞
021 012
LT = ⎝ 1 0 2 ⎠ ⊥ ⎝1 2 0⎠
210 201
⎛ ⎞ ⎛ ⎞
012 012
Lr = ⎝ 2 0 1 ⎠ ⊥ ⎝1 2 0⎠
120 201
⎛ ⎞ ⎛ ⎞
012 012
L = ⎝1 2 0⎠ ⊥
c ⎝2 0 1⎠
201 120
⎛ ⎞ ⎛ ⎞
012 021
(Lc )T = ⎝ 1 2 0 ⎠ ⊥ ⎝1 0 2⎠
201 210
⎛ ⎞ ⎛ ⎞
021 012
(Lr )T = ⎝ 1 0 2 ⎠ ⊥ ⎝1 2 0⎠.
210 201
Section 3.3
(1) The unique completed squares is

⎛ ⎞
12345
⎜4 3 2 5 1⎟
⎜ ⎟
⎜3 1 5 2 4⎟.
⎜ ⎟
⎝5 4 1 3 2⎠
25413
(3) Straightforward following the hint.
Section 4.1
(1) The first part follows trivially by replacing every occurrence of the word line
with point and vice versa in the proofs. For the second part, an affine plane of
order n has n2 points and n2 + n lines. Reversing the roles gives the wrong
number of points and lines.
(3) Given a point p, there are n + 1 lines through p. Each line has n points on it
distinct from p. Then the number of points is n(n + 1) + 1 = n2 + n + 1.
(5) Let p be a point and be a line off p. Through each of the n points on , there is
a line through p and that point. Then there is a unique line m through p parallel
to . This gives n + 1 lines through p.
(7) Through A are the lines {A, E}, {A, C} and {A, G}. Through G are the lines
{G, E}, {G, C} and {A, G}. Through E are the lines {E, G}, {C, E} and {A, E}.
Through C are the lines {A, C}, {C, E} and {G, C}.
(9) Through any two points the line between them is the restriction of the line
between them from the projective plane. Given two lines, if they meet at a
point on L (must be only one) they are now parallel; otherwise, they meet at
the same point they met in the projective plane.
(11) The affine plane:
π p1 p2 p3 p4 p5 p6 p7 p8 p9
1 1 1 1 0 0 0 0 0 0
2 0 0 0 1 1 1 0 0 0
3 0 0 0 0 0 0 1 1 1
4 1 0 0 1 0 0 1 0 0
5 0 1 0 0 1 0 0 1 0
6 0 0 1 0 0 1 0 0 1
7 1 0 0 0 1 0 0 0 1
8 0 1 0 0 0 1 1 0 0
9 0 0 1 1 0 0 0 1 0
10 0 0 1 0 1 0 1 0 0
11 1 0 0 0 0 1 0 1 0
12 0 1 0 1 0 0 0 0 1
The projective plane:
Π p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12 p13

L1 1 1 1 0 0 0 0 0 0 1 0 0 0
L2 0 0 0 1 1 1 0 0 0 1 0 0 0
L3 0 0 0 0 0 0 1 1 1 1 0 0 0
L4 1 0 0 1 0 0 1 0 0 0 1 0 0
L5 0 1 0 0 1 0 0 1 0 0 1 0 0
L6 0 0 1 0 0 1 0 0 1 0 1 0 0
L7 1 0 0 0 1 0 0 0 1 0 0 1 0
L8 0 1 0 0 0 1 1 0 0 0 0 1 0
L9 0 0 1 1 0 0 0 1 0 0 0 1 0
L10 0 0 1 0 1 0 1 0 0 0 0 0 1
L11 1 0 0 0 0 1 0 1 0 0 0 0 1
L12 0 1 0 1 0 0 0 0 1 0 0 0 1
L13 0 0 0 0 0 0 0 0 0 1 1 1 1
Section 4.2
(1) The number of points is |F × F| = n2 . The number of lines is the possibilities of x

for x = c which is n and the number of possibilities for m and b in y = mx + b
which is n2 . Therefore, the number of lines is n2 + n. The number of points on
a line is the number of elements (c, y) where c is fixed which is n or the number
of (x, y) satisfying y = mx + b. For a given m and b and x, there is a unique
y and there are n choices for x; therefore, there are n points on these lines.
Through any point (c, d), it is on the line x = c and fixing m there is a unique b
satisfying d = mc + b. Therefore, there are n + 1 lines through a point. There
is a parallel class for each m and one for the horizontal lines. Therefore, there
are n + 1 parallel classes.
(3) We have (a, b, c) = 1(a, b, c) so it is reflexive. If (a, b, c) = λ(d, e, f) then
(d, e, f) = λ1 (a, b, c) so it is symmetric. If (a, b, c) = λ(d, e, f) and (d, e, f) =
μ(g, h, i), then (a, b, c) = λμ(g, h, i) and so it is transitive. Therefore, it is an
equivalence relation.
3 −1
(5) We have |P| = nn−1 = n2 + n + 1.
(7) The points are (0, 0, 1), (0, 1, 0), (0, 1, 1), (0, 1, 2), (1, 0, 0), (1, 0, 1),
(1, 0, 2), (1, 1, 0), (1, 1, 1), (1, 1, 2), (1, 2, 0), (1, 2, 1), (1, 2, 2).
The lines are
L1 = [0, 0, 1] = {(0, 1, 0), (1, 0, 0), (1, 1, 0), (1, 2, 0)};

L2 = [0, 1, 0] = {(1, 0, 0), (0, 0, 1), (1, 0, 2), (1, 0, 1)};
L3 = [0, 1, 1] = {(0, 1, 2), (1, 0, 0), (1, 1, 2), (1, 2, 1)};
L4 = [0, 1, 2] = {(1, 0, 0), (0, 1, 1), (1, 1, 1), (1, 2, 2)};
L5 = [1, 0, 0] = {(0, 1, 1), (0, 1, 0), (0, 0, 1), (0, 1, 2)};
L6 = [1, 0, 1] = {(1, 0, 2), (0, 1, 0), (1, 1, 2), (1, 2, 2)};
L7 = [1, 0, 2] = {(1, 0, 1), (0, 1, 0), (1, 2, 1), (1, 1, 1)};
L8 = [1, 1, 0] = {(0, 0, 1), (1, 2, 0), (1, 2, 1), (1, 2, 2)};
L9 = [1, 1, 1] = {(1, 1, 1), (1, 2, 0), (1, 0, 2), (0, 1, 2)};
L10 = [1, 1, 2] = {(0, 1, 1), (1, 2, 0), (1, 0, 1), (1, 1, 2)};
L11 = [1, 2, 0] = {(0, 0, 1), (1, 1, 0), (1, 1, 1), (1, 1, 2)};
L12 = [1, 2, 1] = {(1, 1, 0), (1, 0, 2), (1, 2, 1), (0, 1, 1)};
L13 = [1, 2, 2] = {(0, 1, 2), (1, 2, 2), (1, 0, 1), (1, 1, 0)}.
Section 4.4
Section 4.5
(1) Once the grid is formed, there is a unique way to complete a plane.
(3) Follow the hint.
Section 5.1
(1) Start at the lower left and end at the lower right.
(3) The following graph has neither an Euler tour nor an Euler cycle.
(5) For five vertices use C5 . For four vertices use the following:
(7) A tour only exists if n = 2. Cycles exist if and only if n > 1 is odd.
(9) Prove it by induction by taking a graph that does not have an Euler cycle and
adjoin one vertex but only connect it to one vertex of the graph.
Section 5.2
(1) The tour v1 , v2 , . . . , vn is an Euler tour if and only if Φ(v1 ), Φ(v2 ), . . . ,

Φ(vn ) is an Euler tour.
(3) If n is odd then every vertex has even degree; therefore, it has an Euler cycle.
If n is even then every vertex has odd degree; therefore, it does not have an
Euler cycle. Only K2 has two odd degree vertices with the rest even.
(5) The wheel graph has 2(n − 1) edges and the cycle graph has n edges.
(7) The number of edges is ab.
(9) The partition is {A}, {B, C, D}, {E}, {F, G, H}.
(11) The quotient graph of both is either L2 or D2 .
(13) Use C5 and the following:
(15) Map the vertices by any bijection ψ. Then {v1 , v2 } is mapped to {ψv1 , ψ(v2 )}.
Section 5.2
(1) 3.
(3) Use K4 drawn as follows:
(5) x(x − 1)2 + x(x − 1)(x − 2)2 .

(7) Follow the hint.
Section 5.3
(1) The in and out degrees are as follows:
vertex in degree out degree

0 1 2
1 2 1
2 3 2
3 1 1
4 2 3
(3) If v1 , v2 , . . . , vn is a tour with (vi , vi+1 ) ∈ E. This implies (v1 , v3 ) ∈ E by

the transitivity of the relation. Therefore, v1 , v3 , . . . , vn is a tour. Then if
v1 , vi , vi+1 , . . . , vn is a tour then (v1 , vi+1 ) ∈ E by the transitivity of the rela-
tion. Then by induction (v1 , vn ) ∈ E.
(5) Number the vertices in any manner. From vertex 1 to the next vertex there are
n choices for a tour. Then after a tour including i vertices, then there are n − i
choices in continuing the tour. Then by induction the number of Hamilton tours
is (n − 1)!.
Section 6.1
(1) They are not linearly independent: (1, 2, 1) + 2(2, 0, 1) + 2(2, 2, 0) =

(0, 0, 0).
(3) The set (1,
0, 0, . . . , 0), (0, 1, 0, 0, . . . , 0), . . . (0, 0, . . . , 0, 1) is a basis of Fn
q.
(5) If vi = j=i αj vj then j=i αj vj − vi = 0 is a non-trivial linear combi-
nation summing to 0. If αi vi = 0, then choose an i with αi = 0 and
v = −( α1i ) j=i αj vj .
(7) If the span of v1 , . . . , vs is not all of V, then add any vector w not in the span.
Then v1 , . . . , vs , w is a linearly independent set.
(9) We have (x1 , mx1 ) + (x2 , mx2 ) = (x1 + x2 , m(x1 + x2 )) so it is closed
under addition. Then α(x, mx) = (αx, m(αx)) and it is closed under scalar
multiplication. Therefore, it is vector subspace of F2 . Its dimension is 1.
(11) If v, w ∈ V ∩ W, then αv + βw ∈ V since V is a vector space and αv + βw ∈
W since W is a vector space. Therefore, V ∩ W is a vector subspace of both
V and W.
(13) The set (1, 0, 0, . . . , 0), (0, 1, 0, 0, . . . , 0), . . . (0, 0, . . . , 0, 1) is a basis of Fnq
and it has n vectors in it.
(15) We have T (0) = T (v − v) = T (v) − T (v) = 0.
Section 6.2
(1)
L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L11 L12
P1 1 0 0 1 0 0 1 0 0 0 1 0
P2 1 0 0 0 1 0 0 1 0 0 0 1
P3 1 0 0 0 0 1 0 0 1 1 0 0
P4 0 1 0 1 0 0 0 0 1 0 0 1
P5 0 1 0 0 1 0 1 0 0 1 0 0
P6 0 1 0 0 0 1 0 1 0 0 1 0
P7 0 0 1 1 0 0 0 1 0 1 0 0
P8 0 0 1 0 1 0 0 0 1 0 1 0
P9 0 0 1 0 0 1 1 0 0 0 0 1
(3) The number of points is 27, then number of lines is 117, and the number of
planes is 39.
(5) The number of points is 625 and the number of hyperplanes is 780.
(7) Let p be a point. Take any k-dimensional subspace V and form the k flat p + V.
(9) Take the lines {(0, 0, 0, 0), (1, 1, 1, 1)} and {(1, 0, 1, 0), (0, 1, 0, 1)}. These are
parallel.
Section 6.3
(1)
L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L11 L12 L13
P1 1 0 0 1 0 0 1 0 0 0 1 0 0
P2 1 0 0 0 1 0 0 1 0 0 0 1 0
P3 1 0 0 0 0 1 0 0 1 1 0 0 0
P4 0 1 0 1 0 0 0 0 1 0 0 1 0
P5 0 1 0 0 1 0 1 0 0 1 0 0 0
P6 0 1 0 0 0 1 0 1 0 0 1 0 0
P7 0 0 1 1 0 0 0 1 0 1 0 0 0
P8 0 0 1 0 1 0 0 0 1 0 1 0 0
P9 0 0 1 0 0 1 1 0 0 0 0 1 0
P10 1 1 1 0 0 0 0 0 0 0 0 0 1
P11 0 0 0 1 1 1 0 0 0 0 0 0 1
P12 0 0 0 0 0 0 1 1 1 0 0 0 1
P13 0 0 0 0 0 0 0 0 0 1 1 1 1
(3) 806.
(5) Let v be a vector generating the one-dimensional subspace that is the point. Then
extend this to k + 1 linearly independent vectors (we can extend to a basis so
we can stop at k + 1). Then the point is on a k-dimensional object.
(7) 5.
Section 6.4
Section 6.5
(1) 6, 14, 21, 22, 30, 33, 38, 42, 46.

Section 6.6

(3) Another would be {h, b, d, f}.
Section 6.7
(1) The image of the subplane of Π is a subplane of Σ.

(3) Use the fact that Fq is a subfield of Fq2 . Construct PG2 (Fq ) and show that this
is a Baer subplane of PG2 (Fq2 ).
Section 6.8
Section 7.1

(5)
n2 + n + 1
n2 n+1 (B.1)
n2 − n n 1
(7) λ = 78.
(9) Projective: (n2 + n + 1, n2 , n2 − n); Affine: (n2 , n2 − n, n2 − n − 1).
Section 7.2
2 n(n+3)
(1) We have n +3n+4
2 = 2 + 2. If n is even then n
2 is an integer. If n is
n+3
odd, then n + 3 is even and 2 is an integer. Therefore, the number is always
an integer.
Section 7.3
(n+λ−1)(n+λ) n(n+1)
(1) If λ = 1, then λ +1= 1 + 1 = n2 + n + 1. If λ = 2,
(n+λ−1)(n+λ) (n+1)(n+2) 2
then λ +1= 2 + 1 = n +3n+4
2 .
Then
qm−1 −1 qm−1 −1
(n + λ − 1)(n + λ) (qm−1 + q−1 − 1)(qm−1 + q−1 )
+1 = +1
λ qm−1 −1
q−1
(qm −1 qm −1
q−1 − 1)( q−1 )
= +1
qm−1 −1
q−1
(q − q)(qm − 1)
m
= +1
(qm−1 − 1)(q − 1)
qm+1 − q q − 1
= +
q−1 q−1
qm+1 −1
= .
q−1
Section 7.4
(1) The system with v = 3 is the single block {α, β, γ}. The system with v = 7 is
the projective plane of order 2.
Section 7.5
(1) Start with the first horizontal line. On this line there are n choices for a point
on a transversal. Given that the first i horizontal lines have been marked, then
there are n − i choices to mark on the net horizontal line avoiding each marked
vertical line. Then there are n! choices and hence n! transversals.
(3) The parameters are v = 24, b = 36, k = 4, n = 6.
(5) The parameters are v = n2 + n, b = n2 , k = n + 1.
(7) Construct the isomorphism by fixing the parallel class corresponding to ver-
tical lines and by transposing the second and third horizontal lines. Then the
isomorphism sends line i in the third parallel class to line i.
(9) Start at each coordinate of the first line and show there are no transversals through
that point.
Section 8.1
(1) If v · w = 0 then multiplying any coordinate by −1 gives the same dot product
and v · −w = −0 = 0. Similarly, if v · v = n and vi = ±1 then multiplying by
−1 does not change this either. Since this applies to columns or rows we have
the first result. Then multiply each coordinate of the first row that has a −1 in it
by −1 to get a matrix where the first row is all 1. Then multiply each row that
begins with a −1 by −1 and we have a matrix that has first row and column
where all entries are 1.
(3) If there were one, then the first row and column would be made all 1 by the first
exercise. Then the matrix would be
⎛ ⎞
11 1
⎝1 a b⎠.
1cd
Then we have a + b = 0 and c + d = 0 giving b = −a and d = −c. Then the

orthogonality of the last two rows give ac + ac = 0 and 2ac = 0, which is not
possible for values of ±1. Hence, there is no Hadamard matrix with n = 3.
(5)
⎛ ⎞
1 1 1 1
⎜ 1 1 −1 −1 ⎟
⎜ ⎟
⎝ 1 −1 1 −1 ⎠ .
1 −1 −1 1
(7) The squares are 0, 1, 3, 4, 9, 10, 12. The non-squares are 2, 5, 6, 7, 8, 11.
(9) Simply follow the hint.
Section 8.2
Section 8.3
Section 8.4
(1) Each possible k-tuple occurs exactly once. Therefore, reading the k-tuples as
numbers in base n give each possible number between 0 and nk − 1 exactly
once.
Section 8.5
(1) (a) 24 (b) 2 (c) 2 (d) 4.

(3) (a) 01011 (b) 11111 (c) 11011 (d) 00000.
(5) If they are both Boolean algebras of cardinality 2n , then they are both iso-
morphic to P({1, 2, . . . , n}). Then the result follows from the composition of
isomorphisms.
(7) In (A, R−1 ) the role of l.u.b and g.l.b is reversed. Its Hasse diagram is the
original one upside down.
(9) If n > 2 then the Boolean algebra is isomorphic to P({1, 2, . . . , k}), where
k > 1. Then the elements corresponding to {1} and {2} are not comparable.
Section 8.6
(1)
00 01 10 11
00 0 1 2 3
01 1 0 3 2
10 2 3 0 1
11 3 2 1 0
It is symmetric since if a + b = c in this group, then a = c + b.
(3)
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
100000 010000 001000
⎜0 1 0 0 0 0⎟ ⎜1 0 0 0 0 0⎟ ⎜0 0 0 0 1 0⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 1 0 0 0⎟ ⎜0 0 0 0 0 1⎟ ⎜1 0 0 0 0 0⎟
A0 = ⎜ ⎟
⎜0 0 0 1 0 0⎟ A1 = ⎜ ⎟
⎜0 0 0 0 1 0⎟ A2 = ⎜ ⎟
⎜0 0 0 0 0 1⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝0 0 0 0 1 0⎠ ⎝0 0 0 1 0 0⎠ ⎝0 1 0 0 0 0⎠
000001 00100 000100
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
000100 000010 000001
⎜0 0 0 0 0 1⎟ ⎜0 0 0 1 0 0⎟ ⎜0 0 1 0 0 0⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 1 0⎟ ⎜0 1 0 0 0 0⎟ ⎜0 0 0 1 0 0⎟
⎜
A3 = ⎜ ⎟ ⎜
A4 = ⎜ ⎟ ⎜ ⎟
⎟ ⎟ A5 = ⎜ 0 1 0 0 0 0 ⎟ .
⎜1 0 0 0 0 0⎟ ⎜0 0 1 0 0 0⎟ ⎜ ⎟
⎝0 0 1 0 0 0⎠ ⎝0 0 0 0 0 1⎠ ⎝1 0 0 0 0 0⎠
010000 100000 000010
(5) Use the group that is the n fold product of (F2 , +).
Section 9.1
5
(1) 18 .
7
(3) 128 .
169
(5) 324 .
5
(7) 6.
63
(9) 256 .
11
(11) 1024 .
1
(13) 11 .
1
(15) ≈ 8(1067 )
.
(p−1)(q−1)(r−1)
(17) pqr .
n2 +n+1
(19) C(n2 +n+1,3)
.
Section 9.2
1
(1) 13 .
3
(3) 13 .
Section 10.1
(1) Closure is immediate, associativity follows from the associativity of addition

in the integers, the identity is 1, and the inverse of a is n − a.
(3) Closure follows from the fact that a c
b + d =
da+bc
bd and bd = 0 if neither b
nor d are 0. Associativity is straightforward. The identity is 01 and the inverse
of a −a
b is b .
(5) If n is a composite, then n = ab, then ab = 0 ∈ / Zn − {0}. This means that it
is not closed and so not a group. If n is prime then, for any a ∈ Zn − {0}, there
exists b, c with ab + cn = 1 since gcd(n, a) = 1. Then reading (mod n)
we have ab ≡ 1 (mod n). The remainder of the proof is straightforward.
(7) Assume e and e are both identities. Then we have e = ee = e .
(9) If σ and τ both fix a point p, the σ(τ(p)) = σ(p) = p so it is closed. If σ(p) = p
then taking σ−1 of both sides gives p = σ−1 (σ(p)) = σ−1 (p). Therefore, it
is a subgroup.
(11) The isomorphism sends ai to i (mod n).
(13) Any subgroup would have to have its order divide p but only 1 and p divide p
so the only subgroups are trivial.
Section 10.2
(1, 3, 5, 7) Straightforward verifications.

Section 10.3
(1) A loop must have an identity e. Then the map f(x) = xa must be injective by
the right cancelation property. This gives that there must be an element b with
ba = e.
(3) Similar to Theorem 10.2.
Section 10.4

Section 11.2
(1) Take any code with a generator matrix of the from (Ik |A), where Ik is the k by
⎛ and A⎞is any k by n − k matrix.
k identity matrix
1000
⎜0 1 0 0⎟
(3) It reduces to ⎜ ⎟
⎝0 0 1 0⎠.
0001
(5) The code {(1111), (1110)} has minimum distance 1 but minimum weight 3.
(7) The following code can correct two errors: {(00000), (11111), (22222),
(33333), (44444)}.
Section 11.3
1. Define a function from the set of k dimension codes to the set of n − k dimension
codes by Φ(C) = C⊥ . Then Φ(C) = Φ(D) implies C⊥ = D⊥ which gives
C = D and the map is injective. Then Φ(D⊥ ) = D for any code D. Therefore,
Φ is a bijection and the cardinality of the two sets is the same.
3. We have dim(C) ≤ dim(C⊥ ) and dim(C) + dim(C⊥ ) = n. Then if k > n 2
this gives a contradiction since the sum of the two dimensions must be greater
than n.
5. We have wt(v + w) = wt(v) + wt(w) − 2|v ∧ w|. Since [v, w] = 0 we have
|v ∧ w| ≡ 0 (mod 2). Then we have wt(v + w) ≡ 0 (mod 4).
Section 11.4
1. The code is {00000, 11110, 00100, 11010}.

The space is partitioned as follows:
00000 10000 01000 00010 00001 10001 00011 01001

11110 01110 10110 11100 11111 01111 11101 10111
00100 10100 01100 00110 00101 10101 00111 01101
11010 01010 10010 11000 11011 01011 11001 10011
The vector 11111 is decoded to 11110.

qn
3. The number of cosets of a k-dimensional code is qk
= qn−k . The number of
possible syndromes is qn−k . Then since every coset has a unique syndrome we
have the result.
Section 11.5
(1) The matrix is

⎛ ⎞
000000011111111
⎜0 0 0 1 1 1 1 0 0 0 0 1 1 1 1⎟
⎜ ⎟
⎝0 1 1 0 0 1 1 0 0 1 1 0 0 1 1⎠.
101010101010101
Section 11.6
(1) The syndrome is a scalar multiple of a column of the parity check matrix. Then
subtract that scalar from the position of that column to decode the vector.
(3) The matrix is

011111
.
101234
Section 11.7
(1) We have 36 (1 + 11(2) + 55(2)2 ) = 36 (35 ) = 311 .

(3) We have 212 (1 + 23 + 253 + 1771) = 212 (2048) = 212 211 = 223 .
Section 11.8
(1) We list the codewords as an array:
0000
0111
0222
1021
1102
1210
2012
2120
2201
Section 11.9
(1) The weight enumerator is x7 + 4x3 y4 .

(3) We have WC (x, y) = xn + yn . Then WC⊥ (x, y) = 12 ((x + y)n + (x − y)n )
n2
= i=1 C(n, 2i)xn−2i y2i . This could have been computed directly by rec-
ognizing C⊥ as the code consisting of all vectors of even weights.
Section 11.10
(1) The following are eliminated: 6, 14, 21, 30.
Section 12.1
(1) YCPLTMMGDYLEOFFNIALCRUSHLETCDY.
(3) φ−1 (x) = 9x + 23.
(5) φ(26)26 = 338.
(7) Applying the rules of determinants we have
det(MM−1 ) = det(M)det(M−1 ) = det(M)det(M)−1 = 1.
Then, since this is done modulo 26, the determinant of M must be relatively
prime to 26, since it is a unit modulo 26.
Section 12.2
(1) 546999052092292800000.
(3) 32755471104142671992819712000000.
Section 12.3
(1) (a) x ≡ 60 (mod 77) (b) x ≡ 211 (mod 221) (c) x ≡ 411 (mod 418), (d) x ≡
1211 (mod 1560).
(3) L2 (1) = 0, L2 (2) = 1, L2 (3) = 4, L2 (4) = 2, L2 (5) = 9, L2 (6) = 5, L2 (7) =
11, L2 (8) = 3, L2 (9) = 8, L2 (10) = 10, L2 (11) = 7, L2 (12) = 6.
(5) φ(2) = 1, φ(3) = 2, φ(4) = 2, φ(5) = 4, φ(6) = 2, φ(7) = 6, φ(8) = 4,
φ(9) = 6, φ(10) = 4, φ(11) = 10, φ(12) = 4, φ(13) = 12, φ(14) = 6, φ(15) =
8, φ(16) = 8, φ(17) = 16, φ(18) = 6, φ(19) = 18, φ(20) = 8, φ(21) = 12,
φ(22) = 10, φ(23) = 22, φ(24) = 8, φ(25) = 20.
(7) n = 187, M = 237 (mod 187) = 133, d = 23, 13323 (mod 187) = 23.
Section 12.4
(1) Applying the rules of determinants we have
det(MM−1 ) = det(M)det(M−1 )det(M)det(M)−1 = 1.
Therefore, det(M) = 1.
(3) ⎛
1. ⎞
0000100
⎜0 0 1 0 0 0 0⎟
⎜ ⎟
⎜0 0 0 0 0 1 0⎟
⎜ ⎟
(5) ⎜ ⎟
⎜0 1 0 0 0 0 0⎟.
⎜0 0 0 0 0 0 1⎟
⎜ ⎟
⎝1 0 0 0 0 0 0⎠
0001000
Section 13.1
(1) Let the first two moves be arbitrary, then show that no matter where the first
player puts his second move, the second player can counter the move by making
it impossible for the first player to have a winning strategy.
(3) Show that given the number of plays that the second player has, there is no way
to have a point on each of the lines in these planes.
Section 13.2
(1) For the plane of order 2, the first move can be made arbitrarily. Then given any
move by the second player, the next move by the first player gives two points
on some line and so it is a win for the first player. Therefore, there is a winning
strategy for the first player. For the plane of order 3, after the first three moves
by the first player, there are three lines with two marked points between these
points and the second player has only been able to block two. Therefore, the
first player has a winning strategy. For the plane of order 4, there is a drawing
strategy.
(3) For the 1-net, the second player can simply place points on the same line as the
first player giving a draw. For the other nets, the second move for the first player
can give a win.
Glossary
Affine plane An affine plane is a set of points A and a set of lines M and an incidence
relation J ⊆ A × M such that through any two points there is a unique line
incident with them. More precisely, given any two points p, q ∈ A there exists a
unique line ∈ M with (p, ) ∈ J and (q, ) ∈ I. If p is a point not incident with
then there exists a unique line through p parallel to . More precisely, if p ∈ A,
∈ M with (p, ) ∈ / J then there exists a unique line m ∈ M with (p, m) ∈ J
and parallel to m. There exists at least 3 non-collinear points. It is a 2-(n2 , n, 1)
design.
Arc A k-arc is a set of k points in a plane such that no three are collinear.
Association scheme Let X be a finite set, with |X| = v. Let Ri be a subset of X × X,
for all i ∈ I = {i | i ∈ Z, 0 ≤ i ≤ d} with d > 0. We define = {Ri }i∈I . We
say that (X, ) is a d-class association scheme if the following properties are
satisfied:

1. The relation R0 = (x, x) : x ∈ X is the identity relation.
2. For every x, y ∈ X, (x, y) ∈ Ri for exactly one i.
T t
For every i ∈ I, there
3. exists i ∈ I such that Ri = Ri , that is we have Ri =
(x, y) | (y, x) ∈ Ri .
4. If (x, y) ∈ Rk , the number of z ∈ X such that (x, z) ∈ Ri and (z, y) ∈ Rj is a
constant pk ij .
Automorphism An automorphism of an incidence structure D = (P, B, I) is a

bijection ψ : D → D, where ψ is a bijection from the set of points to the set of
points, a bijection from the set of blocks to the set of blocks and
(p, B) ∈ I ⇐⇒ (ψ(p), ψ(B)) ∈ I.
Baer subplane A Baer subplane is a subplane of order m of a projective plane of

order m2 .
n!
Binomial coefficient The binomial coefficient is defined as C(n, k) = k!(n−k)! .
358 Glossary
Bipartite graph Let G = (V, E) be a graph where A and B are subsets of the set of
vertices V, with A ∪ B = V and A ∩ B = ∅. The sets A and B are said to be a
partition of V. The graph G is said to be bipartite graph if v ∈ A implies w ∈ B,
whenever {v, w} ∈ E.
Biplane A biplane is a set of points P and a set of lines L and an incidence relation
I ⊆ P × L such that through any two points there are two lines incident with
them and any two lines intersect in two points.
Blocking set A blocking set B in a projective plane is a subset of the points of the
plane such that every line of Π contains at least one point in B and one point not
in B.
Catalan number Catalan numbers satisfy the recursion
n

Cn+1 = Ck Cn−k , C0 = 1.
k=0
1 (2n)!
Their closed form is Cn = n+1 n!n! .
Chromatic number The chromatic number of a graph is the least number of colors
needed to color the vertices such that no two vertices connected by an edge have
the same color.
Chromatic polynomial The chromatic polynomial of a graph is the number of pos-
sible ways to color a graph using n colors or less.
Code A code over an alphabet A of length n is a subset of An .
Coloring of a graph A coloring of a graph G is a map from V to a set C such that
if (v, w) ∈ E then f(v) = f(w).
Complete graph The complete graph on n vertices, Kn , is the graph for which any
two distinct vertices are connected by an edge.
Connected graph A connected graph is a graph in which for any two vertices v and
w there is a tour from v to w. Otherwise we say that the graph is disconnected.
Cubic number Cubic numbers satisfy the recursion an+1 = an + 3n2 + 3n + 1.
Their closed form is an = n3 .
Cycle graph A cycle graph on n vertices, Cn , is the connected graph on n vertices
consisting of a single Euler cycle with no repeated vertices.
Design A t-(v, k, λ) design is D = (P, B, I) where |P| = v, every block B ∈ B is
incident with exactly k points, and every t distinct points are together incident
with exactly λ blocks.
Degree of a vertex The degree of a vertex in a graph is the number of edges incident
with that vertex.
Difference Set A difference set in a group (G, +) is a subset D of G such that each
non-identity element g of G can be written in exactly λ different ways of the form
x − y with x, y ∈ D.
Directed graph A directed graph G = (V, E) is a set of vertices V and edges E,
where each edge is an ordered pair of the form (v1 , v2 ) where vi ∈ V.
Discrete graph The discrete graph on n vertices, Dn , is the graph consisting of n
vertices and no edges.
Glossary 359
Euler tour A tour in a graph G = (V, E) is an ordered k-tuple (e1 , e2 , . . . , ek ),

ei = {vi , wi } ∈ E, where wi = vi+1 and ei = ej if i = j. An Euler tour is a
tour such that each edge of the graph appears in the tour. An Euler cycle is an
Euler tour where the starting vertex of the tour is also the ending vertex, i.e.
v1 = wk .
Fibonacci sequence The sequence is defined as F0 = 0, F1 = 1, Fn = Fn−1 +
Fn−2 .
Field A finite field (F, +, ∗), is a set F with two operations + and ∗ such that (F, +) is
a commutative group; (F − {0}, ∗) is a commutative group; and for all a, b, c ∈ F,
we have a ∗ (b + c) = a ∗ b + a ∗ c and (b + c) ∗ a = b ∗ a + c ∗ a.
Generalized Hadamard matrix A generalized Hadamard (GH) matrix H(q, λ) =
(hij ) of order n = qλ over Fq is a qλ × qλ matrix with entries from Fq such
that for every i, j, 1 ≤ i < j ≤ qλ, each of the multisets {his − hjs : 1 ≤ s ≤ qλ}
contains every element of Fq exactly λ times.
Graph A graph G = (V, E) is a set of vertices V and edges E where each edge is of
the form {v1 , v2 } where vi ∈ V.
Group A group (G, ∗) is a set G with an operation ∗ such that if a, b ∈ G then
a ∗ b ∈ G; for all a, b, c ∈ G, (a ∗ b) ∗ c = a ∗ (b ∗ c); there exists an e ∈ G
with e ∗ a = a ∗ e = a for all a ∈ G; and for all a ∈ G there exists b ∈ G with
a ∗ b = e.
Hadamard matrix A Hadamard matrix is a square n by n matrix H with Hij ∈
{1, −1} such that HHt = nIn , where In is the n by n identity matrix.
Hadamard Design A Hadamard design is a 3-(4n, 2n, n − 1) design.
Hamiltonian cycle A Hamiltonian cycle hits every vertex exactly once except for
the final vertex which was also the beginning vertex.
Hamiltonian tour A Hamiltonian tour is a tour that hits every vertex exactly once.
Hyperoval A hyperoval is an arc of maximal order in a plane of even order, that is,
a set of n + 2 points such that no 3 are collinear.
Hyperplane A hyperplane is an n − 1 dimensional subspace of an n dimensional
space.
k-partite graph Let A1 , A2 , . . . , Ak be a partition of the set of vertices V. The
graph G is said to be a k-partite graph if v ∈ Ai implies w ∈ / Ai , whenever
{v, w} ∈ E. That is, if two vertices are connected then they must be in different
sets of the partition.
Latin hypercube A Latin k-hypercube Lj1 ,...,jk of order n is a k dimensional array
of size n, whose elements come from the set {0, 1, 2, . . . , n − 1}. If ji = ji for
all i = α and jα = jα then Lj1 ,j2 ,...,jk = Lj ,j ,...,j . That is, in any direction
1 2 k
each “row” has each element from the set exactly once.
Latin square A Latin square L is an n by n array such that each element of an
alphabet A with n elements, appears exactly once in every row and every column.
Latin subsquare Let L be a Latin square of order n, indexed by the elements of
Zn . If there are subsets R, C ⊂ Zn with |R| = |C| = m, where Lij , i ∈ R, j ∈ C,
is a Latin square of order m, this is said to be a subsquare of order m.
360 Glossary
Lattice A partially ordered set is a lattice if and only if for every two elements a and
b in the partially ordered set the least upper bound l.u.b.(a, b) and the greatest
lower bound g.l.b.(a, b) always exist.
Linear graph The linear graph on n vertices, Ln , is the connected graph on n
vertices consisting of a single Euler tour with no repeated vertices.
MDS (Maximum Distance Separable) code An [n, k, d] code is called MDS if
d = n − k + 1.
Main class If L is a Latin square, then any Latin square that can be obtained by
any combination of permuting the rows, permuting the columns, permuting the
symbols, taking the row adjugate, taking the column adjugate, and taking the
transpose is said to be in the main class of L.
Magic square A magic square of order n is an n by n matrix where each element
of the set {0, 1, 2, . . . , n2 − 1} appears once and the sum of each row and column
is the same.
Multinomial Coefficient The multinomial coefficient is defined as:

n n!
= .
k1 , k2 , . . . , kt k1 !k2 !k3 ! · · · kt !
Multigraph A multilgraph is a structure G = (V, E) where V is a set of vertices,

E is a set of edges, and each edge is represented by ({a, b}, i) where a and b are
vertices connected by the edge and i is used to distinguish multiple edges. We say
that two vertices are adjacent if they are connected by an edge.
Mutually Orthogonal Latin Squares (MOLS) If L = (ij ) is a Latin square of
order n over an alphabet A and M = (mij ) is a Latin square of order n over an
alphabet B then L and M are said to be orthogonal if every pair of A × B occurs
exactly once in {(ij , mij )}. If L1 , L2 , . . . , Lk are Latin squares of order n, then
they are a set of k Mutually Orthogonal Latin Squares if Li and Lj are orthogonal
for all i, j, i = j.
Near field A nearifield (Q, +, ∗), is a set Q, with two operations + and ∗, such that
the following hold:
1. (additive closure) If a, b ∈ Q then a + b ∈ Q.

2. (additive associativity) For all a, b, c ∈ Q we have (a + b) + c = a + (b + c).
3. (additive identity) There exists an element 0 ∈ Q such that 0 + x = x + 0 = x
for all x ∈ Q.
4. (additive inverses) For all a ∈ Q there exists b ∈ Q with a + b = b + a = 0.
5. (additive commutativity) For all a, b ∈ Q we have a + b = b + a.
6. (multiplicative closure) If a, b ∈ Q then a ∗ b ∈ Q.
c).
8. (multiplicative identity) There exists an element 1 ∈ Q such that 1 ∗ x = x ∗
1 = x for all x ∈ Q.
9. (multiplicative inverse) For all a ∈ Q − {0}, there exists b ∈ Q with a ∗ b = 1.
Glossary 361
10. (unique solution) For all a, b, c ∈ Q , a = b there is exactly one solution to

xa − xb = c.
11. (right distributive) For all a, b, c ∈ Q we have (b + c) ∗ a = b ∗ a + c ∗ a.
Nets A k-net of order n is an incidence structure consisting of n2 points and nk

lines satisfying the following four axioms: every line has n points; parallelism is
an equivalence relation on lines, where two lines are said to be parallel if they are
disjoint or identical; there are k parallel classes each consisting of n lines; any
two non-parallel lines meet exactly once. It is a 1-(n2 , n, k) design.
Oval An oval is an arc of maximal size in a plane of odd order, that is n + 1 points
in a plane of odd order such that no 3 points are collinear.
Parallel lines Two lines are parallel if they lie on a plane and are either equal or
disjoint.
Parity check matrix A matrix H is said to be a parity check matrix for the code C,
if C = {v | HvT = 0}.
Partial Latin square A partial Latin square of order n is an n by n array, where
cells may be empty or from a symbol set A of size n, such that each symbol
occurs at most once in any row or column.
Perfect code A perfect code is a code with equality in the sphere packing bound,
namely if C is a code
of length n over an alphabet of size p with minimum distance
2t + 1 then |C|( ts=0 C(n, s)(p − 1)s ) = pn .
Partially ordered set A partially ordered set is a set A together with a relation R
on A such that R is reflexive, antisymmetric, and transitive.
Permutation A permutation of set A is a bijection from A to A.
Permutations The number of permutations of k objects chosen from n objects is
n!
defined as P(n, k) = (n−k)! .
Planar graph A graph is said to be a planar graph if it can be drawn in R2 such
that no two edges (when viewed as curves in R2 ) intersect at a point that is not a
vertex.
Primitive root Let p be a prime. An element a (mod p) is a primitive root (mod p)
if ap−1 ≡ 1 (mod p) but ai ≡ 1 (mod p) for 0 < i < p − 1.
Probability Given a sample space S and an event E ⊂ S, the probability of event E
|E|
is P(E) = |S| .
Projective plane A projective plane is a set of points P and a set of lines L and an
incidence relation I ⊆ P × L such that through any two points there is a unique
line incident with them. More precisely, given any two points p, q ∈ P there exists
a unique line ∈ L with (p, ) ∈ I and (q, ) ∈ I. Any two lines meet in a unique
point. More precisely, given any two lines , m ∈ L there exists a unique point
p ∈ P with (p, ) ∈ I and (p, m) ∈ I. There exists at least four points no three
of which are collinear. It is a 2-(n2 + n + 1, n + 1, 1) design.
Quasigroup A quasigroup (Q, ∗) is a set G with an operation ∗ such that if a, b ∈ G
then a ∗ b ∈ G; for all a, b ∈ G there exists a unique x with a ∗ x = b; and for
all a, b ∈ G there exists a unique y with y ∗ a = b.
362 Glossary
Regular graph A graph is said to be regular if the degree of every vertex is the
same.
Residual design A residual design is D = (P , B , I ) where D = (P, B, I) is a
design and for a block v in D we define P = P − {p| (p, v) ∈ I}, B = B − {v}
and a point is incident with a block in D if it was incident in D.
Self-dual code A code C ⊆ Fn q is said to be self-dual if C = C .
⊥
Self-orthogonal code A code C ⊆ Fq is said to be self-orthogonal if C ⊆ C⊥ .

n
Singer cycle A singer cycle is a map ψ : Π → Π such that ψ maps points to points,
lines to lines and preserves incidence, that is an automorphism, such that for any
2
two points p and p in Π there exists an i such that ψi (p) = p and ψn +n+1 is
the identity.
Skew lines Two lines that do not lie on a plane but do not intersect are called skew
lines.
Sphere Packing Bound A code C of length n over an alphabet of size p with
minimum weight 2t + 1 satisfies |C|( ts=0 C(n, s)(p − 1)s ) ≤ pn .
Steiner Triple System A Steiner Triple System is a 2-(v, 3, 1) design.
Square numbers Square numbers satisfy the recursion an+1 = an + 2n + 1.
Their closed form is an = n2 .
Stirling numbers
of the firstkind Stirling
numbers of the first kind satisfy
the
n n−1 n−1 n n
recursion = (n − 1) + , = (n − 1)!, = 1.
k k k−1 1 n
Stirling numbers ofthe second
kind
Stirling
numbers
of the
second kind satisfy
n n−1 n−1 n n
the recursion =k + , = = 1.
k k k−1 1 n
Symmetric design A symmetric design is a 2-(v, k, λ) design where the number of
points equals the number of blocks and the axioms are symmetric for points and
blocks.
Syndrome Let C be a code in Fn q with parity check matrix H. Then the syndrome
of a vector v ∈ Fq is S(v) = HvT .
n
Transversal A transversal of a Latin square L is a set of n coordinates in the square

such that each symbol appears once in the set and the set intersects each row and
column exactly once.
Transversal design A transversal design D of order n, block size k, is a set of
points V and blocks B such that V has kn points; V is partitioned into k classes,
each of size n; and every pair of points is either contained in the same set of the
partition or in a block. It is a 1-(kn, k, n) design.
Triangular numbers Triangular numbers satisfy the recursion an+1 = an + n +
n(n+1)
1. Their closed form is an = 2 .
Unit Let n be a positive integer. An element a ∈ Zn is a unit if there exists an
element b ∈ Zn such that ab ≡ 1 (mod n).
Wheel graph The wheel graph, Wn+1 is formed from the cycle graph Cn by adding
an additional vertex that is connected to ever other vertex.
Zero divisor Let n be a positive integer. A non-zero element c ∈ Zn is a zero divisor
if there exists a d = 0 with cd ≡ 0 (mod n).
References
1. Appel, K., Haken, W.: Every planar map is four colorable. Part I. Discharging. Ill. J. Math.
21, 429–490 (1977)
2. Appel, K., Haken, W.: Every planar map is four colorable. Contemp. Math. 98, (1989)
3. Appel, K., Haken, W., Koch, J.: Every planar map is four colorable. Part II. Reducibility. Ill.
J. Math. 21, 491–567 (1977)
4. Assmus Jr., E.F., Key, J.D.: Designs and Their Codes. Cambridge University Press, Cambridge
(1992)
5. Assmus Jr., E.F., Mattson Jr., H.F.: New 5-designs. J. Comb. Theory 6, 122–151 (1969)
6. Bammel, S.E., Rothstein, J.: The number of 9 × 9 Latin squares. Discret. Math. 11, 93–95
(1975)
7. Bannai, E., Ito, T.: Algebraic Combinatorics I: Association Schemes. The Ben-
jamin/Cummings Publishing Co., Inc, Menlo Park (1984)
8. Berlekamp, E.R., Conway, J.H., Guy, R.K.: Winning Ways for Your Mathematical Plays.
Academic, New York (1982)
9. Boole, G.: An Investigation of the Laws of Thought. Prometheus Books, New York (1854)
10. Bose, R.C., Shimamoto, T.: Classification and analysis of partially balanced incomplete block
designs with two associate classes. J. Am. Stat. Assoc. 47, 151–184 (1952)
11. Bose, R.C., Shrikhande, S.S.: On the construction of sets of mutually orthogonal Latin squares
and the falsity of a conjecture of Euler. Trans. Am. Math. Soc. 95, 191–209 (1960)
12. Bose, R.C., Shrikhande, S.S., Parker, E.T.: Further results on the construction of mutually
orthogonal Latin squares and the falsity of Euler’s conjecture. Can. J. Math. 12, 189–203
(1960)
13. Bourbaki, N.: Corps communatifs, 12th edn. Hermann, Paris (1967)
14. Brown, E.: The fabulous (11, 5, 2) biplane. Math. Mag. 77, 2 (2004)
15. Brualdi, R.A., Ryser, H.J.: Combinatorial Matrix Theory. Encyclopedia of Mathematics and
Its Applications, vol. 39. Cambridge University Press, Cambridge (1991)
16. Bruck, R.H.: Finite nets I. Numerical invariants. Can. J. Math. 3, 94–107 (1951)
17. Bruck, R.H., Ryser, H.J.: The nonexistence of certain finite projective planes. Can. J. Math.
1, 88–93 (1949)
18. Bruen, A.: Baer subplanes and blocking sets. Bull. Am. Math. Soc. 76, 342–344 (1970)
19. Carroll, M.T., Dougherty, S.T.: Tic-Tac-Toe on a finite plane. Math. Mag. 260–274, (2004)
20. Carroll, M.T., Dougherty, S.T.: Fun and Games with Squares and Planes. MAA Press, Resource
for Teaching Discrete Mathematics (2009)
364 References
21. Carroll, M.T., Dougherty, S.T.: Tic-Tac-Toe on Affine Planes. The Mathematics of Various
Entertaining Subjects. Princeton University Press, Princeton (2016)
22. Catalan, E.: Note sur une equation aux différences finies. Journal de mathématiques pures et
appliquées 3, 508–516 (1838)
23. Cayley, A.: On the colourings of maps. Proc. R. Geogr. Soc. 1, 259–261 (1879)
24. Colbourn, C.J., Dinitz, J.H. (eds.): CRC Handbook of Combinatorial Designs. CRC Press,
Boca Raton (1996)
25. Denes, J., Keedwell, A.D.: Latin Squares and Applications. Academic, New York (1974)
26. Diffie, W., Hellman, M.E.: New directions in cryptography. IEEE Trans. Inf. Theory IT 22(6),
644–654 (1976)
27. Dirac, G.A.: Some theorems on abstract graphs. Proc. Lond. Math. Soc. 3rd Ser. 2, 69–81
(1952)
28. Dougherty, S.T.: A coding-theoretic solution to the 36 officer problem. Des. Codes Cryptogr.
4(1), 123–128 (1994)
29. Dougherty, S.T.: Algebraic Coding Theory over Finite Commutative Rings. SpringerBriefs in
Mathematics. Springer, Cham (2017)
30. Dougherty, S.T., Vasquez, J.: Amidakuji and games. J. Recreat. Math. 37(1), 46–56 (2008)
31. ElGamal, T.: A public-key cryptosystem and a signature scheme based on discrete logarithms.
IEEE Trans. Inf. Theory 31(4), 469–472 (1985)
32. Erdös, P., Selfridge, J.L.: On a combinatorial game. J. Comb. Theory 14, 298–301 (1973)
33. Euler, L.: From the problem of the seven bridges of Königsberg, Commentarii. St. Petersbg.
Acad. 8 (1736)
34. Euler, L.: Observationes Analyticae, Novi Commentarii Academiae Scientiarum Imperialis
Petropolitanae 11, 53–69 (1765). Reprinted in Leonardi Euleri Opera Omina Ser. I 15, 50–69
(1923), Teubner, Berlin-Leipzig
35. Euler, L.: Recherches sur une nouvelle espece des quarres magiques. Leonardi Euleri Opera
Omina Ser. I 7, 291–392 (1923), Tuebner, Berlin-Leipzig
36. Fisher, R.A.: The Design of Experiments. Macmillan, London (1935)
37. Fisher, R.A., Yates, F.: The 6 × 6 Latin squares. Proc. Camb. Philos. Soc. 30, 492–507 (1934)
38. Fraleigh, J.B.: A First Course in Abstract Algebra, 5th edn. Addison Wesley, Boston (1994)
39. Gauss, C.F.: Disquisitiones Arithmeticae (1801)
40. Golay, M.J.E.: Notes on digital coding. Proc. IEEE 37, 657
41. Guthrie, F.: Tinting maps. The Athenæum 1389, 726 (1854)
42. Hadamard, J.: Résolution d’une question relative aux déterminants. Bull. Sci. Math. 2, 240–
246 (1893)
43. Hall, M.: Projective planes. Trans. Am. Math. Soc. 54, 229–277 (1943)
44. Hall, M.: Combinatorial Theory. Ginn-Blaisdell, Waltham (1967)
45. Hamilton, W.R.: Memorandum respecting a new system of roots of unity. Philos. Mag. 12,
446 (1856)
46. Hamilton, W.R.: Account of the icosian calculus. Proc. R. Ir. Acad. 6, 415–416 (1858)
47. Hamming, R.W.: Error detecting and error correcting codes. Bell Syst. Tech. J. 29, 147–160
(1950)
48. Hill, R.: A First Course in Coding Theory. Oxford University Press, Oxford (1990)
49. Huffman, W.C., Pless, V.S.: Fundamentals of Error-Correcting Codes. Cambridge University
Press, Cambridge (2003)
50. Hulpke, A., Kaski, P., Östergard, P.: The number of Latin squares of order 11. Math. Comput.
80(274), 1197–1219 (2011)
51. Huygens, C.: De Ratiociniis in Ludo Aleae (The value of all chances in games of fortune)
(1657)
52. Järnefelt, G., Kustaanheimo, P.: An Observation on Finite Geometries. Den 11te Skandinaviske
Matematikerkongress, Trondheim, 166–182 (1949)
53. Jungnickel, D.: Latin Squares, Their Geometries and Their Groups. A Survey. IMA Volumes
in Mathematics and Its Applications, vol. 21. Springer, Berlin
References 365
54. Keedwell, A.D., Dénes, J.: Latin Squares and Their Applications, 2nd edn. Elsevier/North-
Holland, Amsterdam (2015)
55. Kharaghani, H., Tayfeh-Rezaie, B.: A Hadamard matrix of order 428. J. Comb. Des. 13(6),
435–440 (2005)
56. Kirkman, T.P.: On a problem in combinatorics. Camb. Dublin Math. J. 2, 191–204 (1847)
57. Knuth, D.E.: Axioms and Hulls. Lecture Notes in Computer Science, vol. 606. Springer, Berlin
(1992)
58. Lam, C.W.H., Thiel, L., Swiercz, S.: The non-existence of finite projective planes of order 10.
Can. J. Math. 41, 1117–1123 (1989)
59. Laywine, C.F., Mullen, G.L.: Discrete Mathematics Using Latin Squares. Wiley-Interscience
Series in Discrete Mathematics and Optimization. Wiley, New York (1998)
60. Lucas, É.: Récréations mathématiques, four volumes. Gauthier-Villars, Paris (1882–1894).
Reprinted by Blanchard, Paris (1959)
61. MacWilliams, F.J.: Combinatorial problems of elementary abelian groups, Ph.D. Dissertation,
Harvard University (1962)
62. MacWilliams, F.J., Sloane, N.J.A.: The Theory of Error-Correcting Codes. North-Holland,
Amsterdam (1977)
63. McEliece, R.J.: A public-key cryptosystem based on algebraic coding theory. DSN Prog. Rep.
44, 114–116 (1978)
64. McKay, B., Rogoyski, E.: Latin squares of order 10. Electron. J. Comb. 2, Note 3 (1995)
65. McMahon, L., Gordon, G., Gordon, H., Gordon, R.: The Joy of SET. The Many Mathemat-
ical Dimensions of a Seemingly Simple Card Game. Princeton University Press, Princeton;
National Museum of Mathematics, New York (2017)
66. Mullen, G.L., Purdy, D.: Some data concerning the number of Latin rectangles. J. Comb. Math.
Comb. Comput. 13, 161–165 (1993)
67. Pickert, G.: Projective Ebenen. Springer, Berlin (1955)
68. Póyla, G.: On picture writing. Am. Math. Mon. 63, 689–697 (1956)
69. Qvist, B.: Some remarks concerning curves of the second degree in a finite plane. Ann. Acad.
Sci. Fennicae (A,I) 134 (1952), 27 pp
70. Rapoport, A.: Two-Person Game Theory. University of Michigan Press, Ann Arbor (1966)
71. Ray-Chaudri, D.K., Wilson, R.M.: Solution of Kirkman’s schoolgirl problem. In: Combina-
torics (Proceedings of the Symposium in Pure Mathematics, vol. XIX, University of California,
Los Angeles, California, 1968), pp. 187–203. American Mathematical Society, Providence
(1971)
72. Reidemeister, K.: Topologische Fragen der Differentialgeometrie. Gewebe und Gruppen,
Math. Z. 29, 427–435 (1928)
73. Reiss, M.: Uber eine Steinersche Combinatoirische Aufgabe Welche in 45sten Bande Dieses
Journals, Siete 181, Gestellt Worden Ist, 0J. reine angew. Math. 56, 36–344 (1859)
74. Richardson, J.T.E.: Who introduced Western mathematicians to Latin squares? Br. J. Hist.
Math. 34(2), 95–103 (2019)
75. Rivest, R.L., Shamir, A., Adleman, L.: A method for obtaining digital signatures and public-
key cryptosystems. Commun. ACM 21(2), 120–126 (1978)
76. Rosenhouse, J.: The Monty Hall Problem. The Remarkable Story of Math’s Most Contentious
Brainteaser. Oxford University Press, Oxford (2009)
77. Segre, B.: Ovals in a finite projective plane. Can. J. Math. 7, 414–416 (1955)
78. Shannon, C.E.: A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423 and
623–656 (1948)
79. Shor, P.W.: Polynomial-time algorithms for prime factorization and discrete logarithms on a
quantum computer. SIAM J. Comput. 26(5), 1484–1509 (1997)
80. Singer, J.: A theorem in finite projective geometry and applications to number theory. Trans.
Am. Math. Soc. 43, 377–385 (1938)
81. Stanley, R.: Catalan Numbers. Cambridge University Press, Cambridge (2015)
82. Steiner, J.: Combinatorische Aufgabe. J. reine angew. Math. 45, 181–182 (1853)
83. Stevenson, F.: Projective Planes. Freeman, San Francisco (1972)
366 References
84. Stinson, D.R.: A short proof of the non-existence of a pair of orthogonal Latin squares of order
six. J. Comb. Theory A36, 373–376 (1984)
85. Stinson, D.R.: Cryptography. CRC Press Series on Discrete Mathematics and Its Applications,
Theory and Practice (1995)
86. Stinson, D.R.: Combinatorial Designs. Constructions and Analysis. Springer, New York (2004)
87. Stirling, J.: Methodus Differentialis. London (1730)
88. Straffin, P.D.: Game Theory and Strategy. The Mathematical Association of America, New
York (1993)
89. Sylvester, J.J.: Thoughts on inverse orthogonal matrices, simultaneous sign-succession, and
tessellated pavements in two or more colours, with applications to Newton’s rule, ornamental
tile-work, and the theory of numbers. Lond. Edinb. Dublin Philos. Mag. J. Sci. 34, 461–475
(1867)
90. Tarry, G.: Le probleme des 36 officers. Compte Rendu Ass. Franc. Pour l’avacement des
sciences 2, 170–203 (1901)
91. The Thirteen Books of Euclid’s Elements, translated by Thomas L. Heath. Cambridge Uni-
versity Press, Cambridge
92. Trappe, W.: Introduction to Cryptography with Coding Theory, 2nd edn. Pearson Prentice
Hall, Upper Saddle River (2006)
93. Weibel, C.: Survey of non-Desarguesian planes. Not. AMS 54(10), 1294–1303
Index
A Cross product, 84
Abelian, 246 Cubic number, 358
Affine geometry, 153 Cycle graph, 124, 358
Affine plane, 327, 357
Antisymmetric, 217 D
Arc, 357 Degree, 358
Archimedes, 35 Design, 181, 322, 358
Association scheme, 230, 357 Difference set, 358
Automorphism, 357 Directed graph, 358
Discrete graph, 124, 358
B Drawing strategy, 322
Baer subplane, 176, 357
Balanced incomplete block design, 182
Bayes’ Theorem, 244 E
Bijective, 2 Equivalence class, 36
Binomial coefficient, 357 Equivalence relation, 36
Binomial theorem, 240 Erdös, 121
Bipartite graph, 125, 358 Euclidean algorithm, 36, 37
Biplane, 188, 327, 358 Euclidean geometry, 153
Block, 181 Euler, 35, 117
Blocking set, 176, 358 Euler cycle, 119
Boolean algebra, 225 Euler’s generalization of Fermat’s Little
Bose-Mesner algebra, 231 Theorem, 43
Euler tour, 119, 359
C Euler φ function, 42, 311
Cardinality, 2, 4
Catalan number, 358 F
Chromatic number, 358 Factorial, 4
Chromatic polynomial, 358 Fermat’s Little Theorem, 43
Code, 265, 317, 358 Fibonacci, 31
Combination, 5 Fibonacci sequence, 31, 359
Combinatorics, vii Field, 359
Complete graph, 124, 358 Field homomorphism, 52
Connected graph, 358 Field isomorphism, 52
368 Index
Finite field, 52, 53 Lattice, 221, 359, 360

Finite set, 1 Least upper bound, 221
Function, 1 Liber Abaci, 31
Linear graph, 124, 360
G Loop, 258
Games, 321 Lucas numbers, 33
Gauss, 35
Generalized Hadamard matrix, 212, 359 M
Generalized multiplicative counting principle, Magic square, 83, 360
3 Main class, 91, 360
Generalized Pigeonhole Principle, 15 Maximal element, 220
Goppa code, 318 McEliece cryptographic system, 317
Graph, 359 MDS code, 360
Graph coloring, 130, 358 Minimal element, 220
Greatest lower bound, 221 Minimum weight, 266
Group, 234, 245, 359 Monty Hall problem, 11, 242
Morra, 11
H Multigraph, 118, 360
Hadamard design, 209, 328, 359 Multinomial coefficient, 360
Hadamard matrix, 203, 211, 359 Multiplicative counting principle, 3
Hall plane, 180 Mutually exclusive, 239
Hamilton cycle, 120, 359 Mutually Orthogonal Latin Squares (MOLS),
Hamilton tour, 120, 359 360
Hamming code, 275, 277, 282, 318
Hamming distance, 231, 265
N
Hamming scheme, 232
Near field, 178, 360
Hasse diagram, 217, 224
Net, 182, 329, 361
Hypercube, 359
Hyperoval, 279, 359 Niels Henrik Abel, 246
Hyperplane, 153, 359
O
I Oval, 361
Independent, 240
Injective, 2 P
Inverse function, 2 Parallel line, 156, 361
Isomorphic, 53 Parity check matrix, 275, 361
Isomorphic graphs, 127 Partial Latin square, 361
Partially ordered set, 216, 217, 361
J Pascal’s triangle, 6, 9
Johnson scheme, 235 Payley construction, 206
Perfect code, 361
K Permutation, 5, 16, 361
Klein-4 group, 247 Permutation matrix, 317, 319
Königsberg bridge problem, 117 Pigeonhole principle, 14
k-partite graph, 126, 359 Planar graph, 361
Kronecker product, 205 Primitive root, 312, 315, 361
Probability, 237, 361
L Projective geometry, 158, 191, 280
LaGrange’s Theorem, 249, 273 Projective plane, 182, 275, 326, 361
Latin hypercube, 214
Latin square, 246, 257, 359 Q
Latin square graph, 126 Quasigroup, 257, 361
Latin subsquare , 359 Quotient graph, 127
Index 369
R Symmetric, 35
Reflexive, 35, 217 Symmetric design, 190, 328, 362
Regular graph, 126, 361 Symmetric group, 246
Repetition code, 266 Syndrome, 362
Residual design, 362
Resolvable, 187 T
Resolvable design, 187 Tartaglia’s triangle, 6
RSA, 313 Ternary ring, 177
Tic-Tac-Toe, 322
S Tour, 119
Sample space, 237 Towers of Hanoi, 72
Schröder-Bernstein Theorem, 3 Transitive, 35, 217
Self-dual, 270 Translation line, 177
Self-dual code, 362 Translation plane, 177
Self-orthogonal, 270 Transversal, 362
Self-orthogonal code, 362 Transversal design, 362
Set, 1, 157 Triangular number, 362
Shor’s algorithm, 315 Tripartite, 126
Simple graph, 123
Singer cycle, 362 U
Singleton bound, 283 Unit, 362
Skew lines, 156, 362
Sphere packing bound, 362 W
Square number, 362 Weight function, 324
Steiner triple system, 330, 362 Wheel graph, 124, 131, 362
Stirling number, 69, 70, 362 Winning strategy, 322
Subgroup, 245
Subset, 1 Z
Surjective, 2 Zero divisor, 362

(2020) - Steven Dougherty. Combinatorics and Finite Geometry

Uploaded by

Copyright:

Available Formats

(2020) - Steven Dougherty. Combinatorics and Finite Geometry

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(2020) - Steven Dougherty. Combinatorics and Finite Geometry

Uploaded by

Copyright:

Available Formats

Springer Undergraduate Mathematics Series

More information about this series at http://www.springer.com/series/3423

Combinatorics and Finite

ISSN 1615-2085 ISSN 2197-4144 (electronic)

Mathematics Subject Classiﬁcation: 05, 06, 51, 52, 94

some interesting recreational mathematics gave rise to some of the most

Guide for Lecturers

Scranton, USA Steven T. Dougherty

The author is grateful to Dr. Cristina Fernández-Córdoba, Dr. Charles Redmond,

1 Foundational Combinatorial Structures . . . . . . . . . . . . . . . . . . . . . 1

11.5 The Binary Hamming Code and the Projective Plane

Solutions to Selected Odd Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

1.1 Basic Counting

Counting is at the core of combinatorics. In general, what we mean by this seemingly

Proof Assume (g ◦ f)(a1 ) = (g ◦ f)(a2 ) then g(f(a1 )) = g(f(a2 )) which implies

A very important technique for comparing the cardinalities of sets is to construct

The following theorem is known as the Schröder-Bernstein Theorem. It is an

The most basic counting principle is the multiplicative principle.

Essentially, we are simply determining the cardinalityof

Example 1.1 As a simple example, consider the sets A = {1, 2, 3, 4} and B =

We shall now examine some other counting techniques.

Theorem 1.5 The number of ways of arranging n items is n!.

ways of arranging the n items.

The value of n! grows wildly in comparison to n. For example, when n is 10 we

The number P(n, k) is often referred to as the number of permutations.

Theorem 1.7 For 0 ≤ k ≤ n, C(n, k) = C(n − 1, k) + C(n − 1, k − 1).

Theorem 1.8 For 0 ≤ k ≤ n, C(n, k) = C(n, n − k).

Proof First technique: By choosing k objects out of n objects to be included in

Theorem 1.9 (The Binomial Theorem) Let n ≥ 0, then

Proof We shall prove the theorem by induction. For n = 0, we have (x + y)0 = 1

(assuming that C(n, k) is 0 when k is negative):

= x(x + y)n + y(x + y)n = (x + y)n+1 .

Therefore by mathematical induction we have the result.

(x + y)5 = x5 + 5x4 y + 10x3 y2 + 10x2 y3 + 5xy4 + y5 .

Let S be a set. We define the power set of S to be

Theorem 1.10 Let S be a set with n elements, then |P(S)| = 2n .

Example 1.6 Let S = {a, b, c}. Then

So S has cardinality 3 and the power set has cardinality 23 = 8.

As an example, if n = 5, then 1 + 5 + 10 + 10 + 5 + 1 = 25 = 32.

Proof Assume we have n objects and place them in a line as o1 , o2 , . . . , on . In

We summarize these counting techniques as follows.

Objects Chosen Repetitions Order Number of choices

n k Not Allowed Does Not Matter n!

2. Give examples of functions f from the set {1, 2, 3, . . . , n} to the set

a. f is injective but not surjective

That is, find n, m, and f satisfying the conditions in each part.

{. . . , −4, −3, −2, −1, 0, 1, 2, 3, 4, . . . }.

Give an example of a function from Z to Z that is an injection but not a

11. Prove that C(2n, n) is even for n > 0.

20. On an 8 by 8 chessboard a bishop attacks another bishop if they share a diagonal.

using the fact that A1 × A2 × · · · × Ak can be viewed as (A1 × A2 × · · · ×

1.2 Multinomial Coefficients

We shall now give a natural generalization for binomial coefficients. Imagine we

Proof The number of ways of placing objects in the boxes is

C(n, k1 )C(n − k1 , k2 )C(n − k1 − k2 , k3 ) · · · C(n − k1 − k2 − · · · − kt−1 , kt )

This gives the result.

Example 1.8 Let n = 9, k1 = 2, k2 = 4, k3 = 3. Then

Theorem 1.13 (The Multinomial Theorem) Let {x1 , x2 , . . . , xt } be a set of t inde-

where the sum is done over all partitions of n.