arXiv:2108.04902v1 [math.

HO] 10 Aug 2021

Counting Rocks!
An Introduction to Combinatorics
Colorado State University

Henry Adams, Kelly Emmrich, Maria Gillespie, Shannon Golden, and Rachel Pries

November 15, 2021

2
Preface

This textbook, Counting Rocks!, is the written component of an interactive introduction to combinatorics at
the undergraduate level. Throughout the text, we link to videos where we describe the material and provide
examples; see the Youtube playlist on the Colorado State University (CSU) Mathematics Youtube channel.
The major topics in this text are counting problems (Chapters 1-4), proof techniques (Chapter 5), re-
currence relations and generating functions (Chapters 6-7), and an introduction to graph theory (Chapters
8-12). The material and the problems we include are standard for an undergraduate combinatorics course.
In this text, one of our goals was to describe the mathematical structures underlying problems in combina-
torics. For example, we separate the description of sequences, permutations, sets and multisets in Chapter
3.
In addition to the videos, we would like to highlight some other features of this book. Most chapters
contain an investigation section, where students are led through a series of deeper problems on a topic. In
several sections, we show students how to use the free on-line computing software SAGE in order to solve
problems; this is especially useful for the problems on recurrence relations. We have included many helpful
figures throughout the text, and we end each chapter (and many of the sections) with a list of exercises of
varying difficulty.

Development of this textbook

For several years, faculty taught Math 301, Introduction to Combinatorics, at Colorado State University using
the textbook Discrete Mathematics: Elementary and Beyond by László Lovász, József Pelikán, and Katalin
Vesztergombi. We found this book very inspiring, but we needed to supplement it with notes on certain
topics. In 2019, Adams wrote an extensive set of lecture notes for the course, based on the aforementioned
book. In 2020, Gillespie and Pries decided that their students needed an interactive textbook on the material
covered in the CSU 301 combinatorics course in order to learn remotely during the COVID-19 pandemic.
This inspired the creation of an accessible, free online textbook along with short accompanying lecture
videos. Adams, Gillespie, and Pries applied for an Open Educational Resources (OER) grant at CSU to
help with the creation of the materials. The math department at CSU also provided some funding. These
resources also funded graduate student co-authors Kelly Emmrich and Shannon Golden. Our title, Counting
Rocks: An Introduction to Combinatorics, is in part an homage to the Rocky Mountains, near which we love
to work. It is possible that we may update this textbook or create more videos, as future versions of the
course are taught.
There are many textbooks written about undergraduate combinatorics, each with its own positive at-
tributes. This book and videos helped our students and we would like to share them with anyone else who
finds them useful.

Acknowledgments
The creation of these materials was supported by the CSU math department and the Open Educational
Resources grant provided by the CSU Libraries. We also thank the Fall 2020 Math 301 students at CSU for
their helpful feedback and suggestions.

3
4
Contents

1 What is combinatorics? 9
1.1 Introducing combinatorics with a handshake . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Three classical counting formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Triangular numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.2 Factorials and orderings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.3 Binomial coefficients and counting subsets . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Introduction to graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1 Walks on graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 When are two graphs the same? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 Introduction to SAGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Counting principles 19
2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 The addition and subtraction principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 The multiplication and division principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Combining the principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5 Sets, subsets, and the number of subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5.1 Basic definition of sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5.2 Set builder notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.3 Famous sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.4 Subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.5 The number of subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6 Addition and subtraction from the perspective of set theory . . . . . . . . . . . . . . . . . . . 29
2.6.1 Union, intersection, and the addition principle . . . . . . . . . . . . . . . . . . . . . . 29
2.6.2 Subsets, complements, and the subtraction principle . . . . . . . . . . . . . . . . . . . 30
2.7 Venn diagrams and the Principle of Inclusion-Exclusion . . . . . . . . . . . . . . . . . . . . . 31
2.8 The multiplication principle from the perspective of set theory . . . . . . . . . . . . . . . . . 33
2.8.1 Cartesian products and the multiplication principle . . . . . . . . . . . . . . . . . . . . 33
2.9 Set partitions, the division principle, and equivalence relations . . . . . . . . . . . . . . . . . 34
2.9.1 Set partitions and the division principle . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.9.2 Equivalence relations and modular arithmetic . . . . . . . . . . . . . . . . . . . . . . . 35
2.10 Additional Problems for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.11 Investigation: Divisors of a positive integer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.11.1 The number of divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.11.2 The sum of the divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.11.3 The Möbius function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.11.4 Multiplicative functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5
3 Counting combinations 41
3.1 The types of combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.1 Counting where order matters and repeats are allowed . . . . . . . . . . . . . . . . . . 43
3.3 Permutations and other sequences with distinct entries . . . . . . . . . . . . . . . . . . . . . . 45
3.3.1 Ordered subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Sets: When order doesn’t matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Multisets: sets with repeats allowed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.1 Definition of multiset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.2 Example of sticks and stones method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.3 Main result on multisets and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5.4 A variation on multiset problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.5 Using multisets to solve distribution problems . . . . . . . . . . . . . . . . . . . . . . . 51
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.7 Additional problems for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.8 Investigation: Counting problems in the game of poker . . . . . . . . . . . . . . . . . . . . . . 54

4 Pascal’s triangle and the Binomial Theorem 61

4.1 Pascal’s triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 The Binomial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 First identities in Pascal’s triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4 Additional identities in Pascal’s triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5 Counting anagrams with multinomial coefficients . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Ordered set partitions and the multinomial theorem . . . . . . . . . . . . . . . . . . . . . . . 72
4.7 Investigation: Combinatorial problems on a chessboard . . . . . . . . . . . . . . . . . . . . . . 75

5 Proof Techniques in Combinatorics 77

5.1 Why are proofs necessary? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.3 Counting in two ways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 Bijective proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.5 Proof by Contradiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.6 The Pigeonhole Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.7 Additional problems for Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.8 Investigation: Pascal’s triangle modulo 2 and Sierpinski’s triangle . . . . . . . . . . . . . . . . 94

6 Recurrence relations 97
6.1 Fibonacci numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.1.1 The golden ratio and a formula for the Fibonacci numbers . . . . . . . . . . . . . . . . 99
6.2 Linear recurrence relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2.1 Another recurrence relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2.2 SAGE commands for recurrence relations . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2.3 Solving linear recurrence relations of degree 2 . . . . . . . . . . . . . . . . . . . . . . . 102
6.2.4 Linear recurrence relations of higher order . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.3 The tower of Hanoi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.4 Regions of the plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.5 Derangements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.6 The Catalan numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

6
7 Generating Functions 113
7.1 Making change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.2 Generating functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.2.1 Are generating functions actually functions? . . . . . . . . . . . . . . . . . . . . . . . . 118
7.3 Operations on generating functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.3.1 Addition (and subtraction) of generating functions . . . . . . . . . . . . . . . . . . . . 119
7.3.2 Multiplication of generating functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3.3 Substituting monomials into generating functions . . . . . . . . . . . . . . . . . . . . . 121
7.3.4 Differentiation of generating functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.4 Solving recurrences with generating functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.4.1 The Tower of Hanoi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.4.2 Regions of the plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
7.5 Generating functions and linear recurrences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.6 The Catalan numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.7 Additional problems for Chapter 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.8 Investigation: Partitions and their generating functions . . . . . . . . . . . . . . . . . . . . . 136

8 Graph Theory Basics 137

8.1 Graphs in combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1.1 Graphs, vertices, edges, and adjacency . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.2 Everyday graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2.1 Complete graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2.2 Path and cycle graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2.3 Bipartite graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.3 The degree of a vertex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.4 Subgraphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.5 Walks and connected graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.6 Graph complements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
8.7 Storage structures for graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
8.8 Eulerian walks and Hamiltonian cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.9 Investigation: The number of walks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

9 Trees 161
9.1 The language of trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
9.2 The number of edges in a tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9.3 Labeled trees: Cayley’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.4 Binary trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
9.4.1 Tournaments and Catalan numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9.4.2 At-most binary trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
9.4.3 Investigation: Increasing binary trees and permutations . . . . . . . . . . . . . . . . . 170

10 Graph optimization 173

10.1 Minimal spanning trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.2 Traveling salesperson problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
10.3 Matchings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
10.3.1 Matchings of bipartite graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
10.4 Ramsey theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
10.4.1 Investigation: Ramsey numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

7
11 Planar Graphs 191
11.1 Planar graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.2 Euler’s formula for planar graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.3 Graphs that are not planar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
11.4 Euler’s formula for polyhedra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11.5 Investigation: Graphs on other surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

12 Graph Coloring 203

12.1 The chromatic number of a graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
12.2 What graphs are 2-colorable? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
12.3 Bounds on chromatic numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
12.4 Brooks’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
12.5 The chromatic polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
12.6 Coloring regions with two colors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
12.7 The four color theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

8
Chapter 1

What is combinatorics?

Combinatorics is the mathematical study of discrete mathematical objects and their combinations and
arrangements. Other kinds of mathematics (like calculus and analysis) use continuous mathematical objects,
which involve infinitesimally small increments. By contrast, discrete mathematics is the study of objects
that come in larger chunks, such as whole numbers, polygons, and networks.
If you have ever asked “I wonder how many different ways I could . . . ?”, then you have thought about
combinatorics. For example, if there are 10 classes you are interested in taking this semester, and you
want to enroll in either 3 or 4 of them, then there might be over 300 different schedules you could choose
from, depending on how many of the classes meet at conflicting times. How do you count the number of
possibilities, without listing them all by hand? Alternatively, how do you write a computer program to list
all of the possible schedules? Both of these approaches — either finding a clever way to count or constructing
a computer algorithm to do the work for you — rely heavily on the ideas and language of combinatorics and
discrete mathematics.
The history of combinatorics began in ancient times, in places such as Egypt, India, China, and Iran.
The work of scholars in ancient times led to the development of counting techinques during 1300-1700 by
individuals such as Fibonacci, Pascal, Leibniz, and De Moivre. In the 1700s and 1800s, mathematicians like
Euler and Cayley developed the study of graph theory.
The computer revolution greatly amplified the importance of combinatorics and graph theory. The most
obvious connections between this class and computer science involve networks and organization of data
structures. There are also applications of this class to other fields, like linguistics, physics, chemistry, and
biology. In general, from this class, you will gain experience solving problems about combinatorial structures
and algorithms; we hope you will continue to use the experience and ideas you gain from this class for the
benefit of many other people.
Here are some of the ideas we will study this semester.

1.1 Introducing combinatorics with a handshake

Suppose five students walk into their combinatorics class and each shakes hands with all the others. How
many handshakes took place?
There are several ways to approach this problem. One way is to reason that, because 5 students each
shake 4 hands, and each handshake involves 2 students, we counted 5 · 4 = 20 handshakes but counted each
handshake twice. So exactly 10 handshakes took place.
Another way to see this is to call the students a, b, c, d, e, and directly count the handshakes as follows.
Student a shakes hands with 4 others, then student b shakes hands with 3 others besides a, then student c
shakes 2 more hands besides a and b, then student d shakes 1 more hand (with e), and finally student e does
not need to shake any more hands. So the total number of handshakes is 1 + 2 + 3 + 4 = 10.
Finally, we can visually model this problem by drawing a dot to represent each student, and a line
segment connecting each pair of dots to represent the handshakes:

9
 a

b e

c d
With this visual tool, called a graph, we can count that there are 10 line segments (the handshakes)
between the labeled dots (the students). The graph above is called K5 , the “complete graph on 5 vertices.”
The dots representing the students are called vertices of the graph, and line segments between the vertices
are called edges. The intersection points of the line segments which are not labeled should be ignored; they
are not vertices. We say the graph is complete because every two vertices are connected by an edge.
In this class, Introduction to Combinatorics, we will learn about counting problems, mathematical proofs,
generating functions, and graph theory. Here are some examples of questions that we will answer. These
questions will become more complicated when we later increase the number of students beyond 5.

Chapters 2–4: Basic counting, sets, and binomial coefficients.

1. How many ways can the 5 students in the class line up?

2. How many ways can 3 students be chosen for a project from the class of 5 students?

3. If each student a, b, c, d, and e receives either 0 or 1 point for the quiz, how many sequences of five
numbers are possible as the list of their grades?

Chapters 5–7: Counting techniques using more theory, including bijections, recurrence rela-
tions, and generating functions.
1. How many ways can 10 pieces of halloween candy be distributed among the students?

2. How many problems on the group worksheet will be solved if the first student does one problem, each
other student does twice as many as the previous, and no problems are solved twice?

3. How many ways can a student make change for one dollar?

Chapters 8–10: Graph theory and optimization.

1. If student a has important news about the class, how many phone calls are needed to share it with the
other students?

2. How many cables are needed to connect the 40 buildings on campus with fiber optic internet?

3. Without lifting your pencil from the page, can you trace your pencil along the edges of the graph K5
so that each edge is drawn exactly once?

Chapters 11–12: Combinatorial geometry, including planar graphs and coloring problems.
1. Is it possible to redraw the edges of the graph K5 on the page so that they do not intersect?

2. On a piece of paper, 4 of the students draw a line. What is the largest number of regions the paper
can be divided into? What about for 5 students?

10
 3. Is it possible to color the regions in the previous problem with 3 colors so that any two regions sharing
a boundary edge have different colors? What about 2 colors?

Exercises
Before getting started, let’s discuss the questions above. At this stage, it is more important to develop good
communication patterns with your classmates than to answer the problems.1 A group works best if everyone
speaks for about the same amount of time and if questions are valued as much as answers. Make sure that
everyone in your group feels comfortable sharing their ideas and understands the material being discussed.
For each problem that your group can solve, write down an explanation of your answer and your thought
process. For each problem that your group cannot solve: decide if the question is clear or ambiguous; decide
if all the information needed to solve the problem is provided; and provide some ideas or a strategy for
solving the problem.

1. Introduce yourself to your group, including something about why you are taking this class, and some-
thing about your life outside of this class.
2. Try to solve the questions from Chapters 2-4.
3. Develop a strategy to approach the problems from Chapters 5-7.
4. Try to solve the problems from Chapters 8-10.
5. Generate ideas about the problems in Chapters 11-12.

1.2 Three classical counting formulas

As an introduction to the techniques in this class, we now discuss three classical problems in mathematics
that can be solved easily with combinatorial methods.

1.2.1 Triangular numbers

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=MJ3HzLuY9TY

We saw in the previous section that we could count the handshakes among 5 students by adding the
numbers 1 + 2 + 3 + 4. For 101 students shaking hands, we can count the number of handshakes by adding
the numbers from 1 to 100. One of the first signs of Gauss’ strength in mathematics was in elementary
school when he quickly computed that

1 + 2 + 3 + · · · + 99 + 100 = 5050.

More generally,
Lemma 1.2.1. The sum of the first n integers is

n(n + 1)
1 + 2 + ... + n = . (1.1)
2
One way to show this is by using a method of proof called induction, which will be covered in Section 5.2.
Here, we prove this identity directly in two more straight forward ways.
1 If you can already answer all these problems, for an arbitrary number of students, you will receive an automatic transfer

to the class on graduate combinatorics Math 501.

11
Proof. Method 1: This is an example of a proof by rearrangement. We can add the numbers in pairs,
starting with the first and last, then the second and second-to-last, and so on:

(1 + n) + (2 + (n − 1)) + (3 + (n − 2)) + · · · .

Each pair sums to n + 1. If n is even, then there are n2 pairs, with the last pair being n2 and n2 + 1. So the
total is the product of n + 1 (the sum of each pair) and n2 (the number of pairs) which equals (n + 1) n2 . If
n is odd, then there are n−1 n+1
2 pairs, and the entry 2 is not included in a pair. So the total is

n−1 n+1 n2 − 1 n + 1 n2 + n
(n + 1) + = + = .
2 2 2 2 2
In either case, the total is n(n+1)
2 .
Method 2: This is an example of a combinatorial proof, in which we count something in two different
ways. In this example, the first way will yield a count of 1+2+· · ·+n, and the second way will yield a count of
n(n+1)
2 . Since both are valid ways of counting the same object, we will have proven that 1+2+· · ·+n = n(n+1)
2 .
How might we do this?
Draw a rectangle of dots, with n rows and n + 1 columns. In the lower left half of the rectangle, color the
dots green, with 1 filled dot in the first row, 2 in the second row, 3 in the third row, up to n dots in the nth
row. The number of green dots is 1 + 2 + · · · + n, which is the left hand side of Equation 1.1. The rectangle
has n(n + 1) dots total and exactly half of them are green. So the number of green dots is n(n + 1)/2, which
is the right hand side of Equation 1.1. Hence 1 + 2 + · · · + n = n(n+1)
2 .

Due to the triangular nature of the picture above, the numbers

n(n + 1)
Tn =
2
are called triangular numbers. The first several triangular numbers are 1, 3, 6, 10, 15, . . .. The number of
handshakes among n + 1 people equals Tn .

1.2.2 Factorials and orderings

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=sG4jx p9PbA

Suppose the five students who walked into the class need to line up at the front desk to hand in their
homework. In how many different orders can they line up?
One way to solve this is to reason that there are 5 possibilities for the first student in line. Regardless of
which student is first, there are 4 possibilities for the next student. Then there are 3 choices for the third

12
student, then 2 choices for the fourth, and finally 1 choice for the last in line. So all together there are
5 · 4 · 3 · 2 · 1 = 120 possibilities.
This leads us to the important definition of a factorial.

Definition 1.2.2. Define 1! = 1, and if n > 1 is an integer, then define n! = n · (n − 1)!. We pronounce the
symbol n! as “n factorial.”

For instance, 2! = 2, 3! = 6, 4! = 24 and 5! = 120. For n ≥ 6, we can write

n! = n · (n − 1) · (n − 2) · · · 3 · 2 · 1.

Factorials are a helpful starting point for many similar problems about arranging things in order. For
instance, if only three of the five students line up at the front desk, then the number of different possible
line-ups is 5 · 4 · 3, which can be expressed as 5!/2!.

Example 1.2.3. Suppose there are 47 students in M301. If there are 47 labeled chairs, the number of ways
they can be seated is 47!. This is because there are 47 choices for which student is in chair 1, then 46 choices
for which student is in chair 2, etc, until there are 2 choices for which student is in chair 46, and only 1
choice for which student is in chair 47.

1.2.3 Binomial coefficients and counting subsets

Click on the icon at left or the URL below for this section’s short video lecture.
https://youtu.be/iXtxzRhhEjI

Binomial coefficients will play an important role throughout this class. They provide another example
of the importance of factorials in counting.

Definition 1.2.4. Let n be
a positive integer. Let {1, . . . , n} be the set of the smallest n positive integers.
The binomial coefficient nk , pronounced “n choose k”, is the number of ways to choose k elements from
{1, 2, . . . , n}.


Example 1.2.5. The binomial coefficient n2 is the number of ways to choose two numbers from the set


{1, . . . , n}. Then n2 = n(n−1)
2 , because there are n choices for the first element and n − 1 choices for the
second, and then we divide by 2 since the order of the two elements does not matter.

Lemma 1.2.6. The binomial coefficient has this formula:

 
n n!
= .
k k!(n − k)!

Proof. Method 1: There are n choices for the first element, then n − 1 choices for the second, and so on,
until there are n − k + 1 choices for the kth element. These choices are independent, so the number of ways
to select these k numbers is the product

n!
n(n − 1) · · · (n − k + 1) = .
(n − k)!

The same set of k numbers can be chosen in k! different ways. We divide by k! since the order of choosing
the numbers does not matter.
Method 2: There are n! ways to line up the numbers in {1, 2, . . . , n}. We then choose the first k
elements in line. We divide by k! because the order does not matter among the chosen numbers, and divide
by (n − k)! because the order does not matter among the numbers that are not chosen.

13
Example 1.2.7. Suppose there are 47 students in Math 301. If there are 50 labeled chairs, what is the
number of ways the 47 students can be seated?


Answer 1: 50 47 · 47!.


The reason is that there are 50 47 ways to choose 47 chairs to use and, once we have chosen 47 chairs,
there are 47! ways to assign the students to chairs.
50!
Answer 2: 3! .
To see this, add three ghost students to the class. There are 50! ways to assign the students and ghosts
to chairs. There are 3! ways to rearrange the 3 ghosts without changing the arrangement of the students.
We divide by 3! since we counted each arrangement of students 3! times.
Luckily, the two answers are the same because
 
50 50! 50!
· 47! = · 47! = .
47 47! 3! 3!

Exercises
1. What is the sum of the numbers from 10 to 100? That is, what is 10+11+12+...+100?

2. Ten people walk into a room and everyone shakes hands with everyone else. How many handshakes
occurred?

3. Compute the sum 3+6+9+12+...+99.

4. How many ways can you rearrange the letters in the word EIGHT?

5. In a betting game, several players keep out-bidding each other. The first player puts one chip into the
empty pot. The second puts two chips in, then the third puts three chips in, and so on. How many
players are needed to continue bidding in this fashion for the pot to contain at least 50 chips?

6. How many ways are there to choose 5 states to visit out of 50 states?

7. How many ways are there to rank the top 5 states out of 50 states that you would be most excited to
visit, in order from one through five?

8. How many ways are there to give 4 identical teddy bears to 7 kids, so that no kid receives more than
one teddy bear?



9. Verify that n2 = 2!(n−2)!
n!
is indeed equal to n(n−1)
2 , and that n+1
2 equals Tn .

10. Let sn be the sum of the first n odd integers, so s1 = 1 and s2 = 1 + 3 = 4.

(a) Find s3 and s4 .

(b) Make a conjecture about the value of sn .
(c) Try to prove your conjecture.

1.3 Introduction to graphs

Click on the icon at left or the URL below for this section’s short video lecture.
https://youtu.be/yHK59HOoaP0

14
 In Section 1.1, the graph of K5 helped us visualize the problem of 5 students shaking hands. More
generally, in Chapter 8, we will define Kn to be the graph
with n vertices, such that every pair of vertices
is connected by an edge. The number of edges in Kn is n2 .
Graphs are useful for other problems too. For example, suppose the 5 students meet 3 employers at a
job fair. Each employer shakes the hand of each student. How many handshakes are there? Draw a graph to
help you visualize these handshakes, using a vertex to represent each person and an edge to represent each
handshake. This is an example of a bipartite graph; this kind of graph is useful when studying connections
between two different types of things, like students and employers, or students and classes.

1.3.1 Walks on graphs

Here is the famous Königsberg bridge problem (now Kalingrad, Russia) — the drawing below has two islands
in the middle of the Pregel river, along with seven bridges connecting the islands and the river banks.2

Question 1.3.1. Can you plan a walk that crosses each bridge exactly once?
Answer. In 1736, Euler proved it is impossible. Try to explain to your group why it is impossible.

Question 1.3.2. The graph above is called the graph of Königsberg. Without lifting your pencil from the
page, can you trace your pencil along the edges of the graph above so that each edge is drawn exactly once?

1.3.2 When are two graphs the same?

When drawing a graph, the position of the vertices and the edges does not matter. The only information that
is relevant in a graph is which pairs of vertices are connected. In fact, the following two pictures illustrate
the same graph.

Here are some ways to tell if two graphs are different:

2 The last time Professor Pries taught this course, there were two students from Kalingrad in the class. They were very

excited to hear about this example and described how the bridges look to the class.

15
 1. The number of vertices for the two graphs is different.
2. The number of edges for the two graphs is different.

Question 1.3.3.
1. Find two graphs with the same number of vertices and the same number of edges which you are
convinced are still different.
2. Find another way of telling if two graphs are different. (There are many possible answers to this
question).
Question 1.3.4. Three of the following four graphs are the same, meaning that one of the following four
graphs is different from all of the others. Can you identify any two graphs which are the same? Can you
identify any two graphs that are different?

Exercises
1. How many vertices does the cube graph have?
2. How many edges does the cube graph have?
3. True or False: The graph of the cube is PLANAR, meaning it can be drawn on a piece of paper with
no edges crossing.
4. The graph of the cube is BIPARTITE, meaning its vertices can be colored green and gold so that
vertices of the same color are not connected by an edge.
5. What is the degree of each vertex in the cube graph?
6. Imagine a graph where the vertices represent apartment buildings and the edges represent streets. A
snow plow operator needs to plow each street, return home, and does not want to travel any more
distance than necessary.
(a) Find a way of removing edges from the graph of Königsberg to make the snow plow operator
happy.
(b) Find a way of adding edges to the graph of Königsberg to make the snow plow operator happy.
(c) Does the location of the plow operator’s house make a difference?
(d) In mathematical terms, describe what needs to be true about a graph for the snow plow operator
to be happy.
7. Again, let the vertices of a graph represent apartment buildings and the edges represent streets. The
mail carrier needs to deliver mail to each apartment building, return home, and does not want to travel
along any more roads than necessary.
(a) Draw several graphs where the mail carrier is happy and several where the mail carrier is unhappy.

16
 (b) Does the location of the mail carrier’s house make a difference?
(c) In mathematical terms, describe what needs to be true about a graph for the mail carrier to be
happy.
8. How is the graph of Königsberg related to the picture of the Königsberg bridge problem? How is
Question 1.3.2 related to Question 1.3.1?
9. Answer Question 1.3.3.
10. Answer Question 1.3.4.

1.4 Introduction to SAGE

Earlier in this chapter, we introduced triangular numbers, factorials, and binomial coefficients. Later in the
book, we will introduce other famous numbers like the Fibonacci numbers and Catalan numbers. It is often
nice to have an exact formula for these numbers. The process of computing them on a calculator can be
slow and leads to small errors.
A faster and more powerful method is to work with computing software. In this course, we will work
with SAGE, a free on-line open-source program, based on python.
To get started with Sage: go to the website
http://sagecell.sagemath.org/
Try typing easy commands like 2+3 or factor(2021) and then evaluating.
For larger jobs or if you want to save your work, you can start a free CoCalc account at:
http://www.sagemath.org/
The sage reference manual is at:
http://doc.sagemath.org/html/en/reference/
A more advanced overview of using sage for combinatorics can be found at
https://doc.sagemath.org/html/en/reference/combinat/sage/combinat/tutorial.html
Example 1.4.1. Which is bigger, 36 or 6!? One method is to compute:
sage: 3^6;
output 729
sage: factorial(6);
output 720
Another method is to write:
sage: 3^6 - factorial(6) > 0;
output: True


Example 1.4.2. For which values of k, is the binomial coefficient 11 bigger than 100? A slow method
11

k
is to compute k for each 0 ≤ k ≤ 11, and see if the answer is bigger than 100. Another method is to
compute them all at once:
sage: [binomial(11, k) for k in range(12)]
output: [1, 11, 55, 165, 330, 462, 462, 330, 165, 55, 11, 1]
11


Note that the command range(m) includes all integers 0 ≤ m ≤ m − 1. From this we see that k > 100
when 3 ≤ k ≤ 8. Here is another method that gives the values of k as the output.
sage: [k for k in range(12) if binomial(11,k) >100]
output: [3, 4, 5, 6, 7, 8]

17
Exercises
Use Sage to answer the following questions.

1. Factor 2022.
4
2. Compute 44 .
4
3. Which is bigger, (4!)! or 44 ?

4. Compute the product of the smallest positive fifty odd numbers (starting with 1).


5. What is the smallest value of k for which 13k is bigger than 10?


6. What is the smallest possible value of n + k for which nk is bigger than 100?

18
Chapter 2

Counting principles

In this chapter, we establish some of the basic principles of counting. In Sections 2.2 - 2.4, we use the basic
arithmetic principles of addition, subtraction, multiplication, and division to count things more efficiently.
In Section 2.5, we begin to reinterpret counting problems using set theory and count the number of subsets
of a finite set. In Sections 2.6 - 2.9, we use set theory to describe the four basic arithmetic principles and
the inclusion-exclusion principle.

2.1 Motivation
Students a, b, c, and d1 sat down in the library to start their combinatorics homework. They read the first
problem that they were assigned.

Question 2.1.1. How many ways are there to choose 6 numbers from the set {1, . . . , 20} so that at least
four of them are odd?

Student d watched as a, b, c each worked on the problem for several minutes, using paper, calculators,
and tablets. Finally, student d said ‘Hey, I don’t know how to get started with this.’
Student a: My thought is there are 10 odd numbers between 1 and 20. So there are 10 choices for the
first odd number, 9 choices for the second, 8 choices for the third, and 7 choices for the fourth. The last
two numbers can be odd or even and there are 16 numbers left. So that gives 16 choices for the fifth and 15
choices for the sixth. So the answer is

10 · 9 · 8 · 7 · 16 · 15 = 1, 209, 600.

Student b: You’re forgetting that the order doesn’t matter. So first we need to choose four numbers out
of the 10 odd numbers, then we need to choose 2 out of the remaining 16 numbers. So the answer is
   
10 16 10! 16!
· = · = 25, 200.
4 2 4!6! 2!14!


Student c: I have a different way of doing it. There are 206 ways to choose six integers out of the 20,
but the probability that a number is odd is 1/2. So the answer is
 
20 1 20! 1
· 4 = · 4 = 2422.5.
6 2 6!14! 2

Wait, that’s not an integer....

Student d: What if there are more than four odd numbers? Forget this, let’s try the next question.
1 To protect their privacy, students are identified by their first initial in this book.

19
Question 2.1.2. In a game of poker, you are dealt five cards from a standard 52 card deck. How many
ways can you be dealt a full house?
Student c: I’ve never played cards. What is a full house?2
Student a: It’s a triple of the same number together with a pair of a different number. Like 6 of hearts,
6 of clubs, and 6 of diamonds, together with 3 of hearts and 3 of spades. So I’m thinking for the first card,
we have 52 choices; no matter what number it is, to get a triple we need to choose two more out of the three
remaining cards of that number, and there are three ways to do that. Then there are 48 choices for the next
card; no matter what number it is, to get a pair we need to choose one more out of the three remaining
cards of that number, and there are three ways to do that. So the answer is:
52 · 3 · 48 · 3 = 22464.
Student b: That’s over counting again because it doesn’t matter which card is dealt first. Instead, let’s
choose two numbers out of 13, then choose three suits for the first number and then choose two suits for the
second number. So the answer is:
     
13 4 4 13 · 12
· · = · 4 · 6 = 1872.
2 3 2 2


Student d: On the web, it says that the probability of a full house is 0.001441 and there are 52 5 =
2, 598, 960 possible hands. If we multiply those together, then we get 3745.10136000000, so the answer is
probably 3745.
Student c: Maybe we need to read more of the book.

Exercises
1. Figure out something that student a said that was correct and something that was wrong.
2. Figure out something that student b said that was correct and something that was wrong.
3. Figure out something that student c said that was correct and something that was wrong.
4. Figure out something that student d said that was correct and something that was wrong.
5. Do you think the students are close to solving Question 2.1.1? What advice would you give them?
6. Do you think the students are close to solving Question 2.1.2? What advice would you give them?

2.2 The addition and subtraction principles

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=apKGi7NJeGM

The first way we learn to count is “by 1’s,” that is, add 1 to a running total for every item in a list. If
the number of things we are trying to count is finite, it is always possible to make a list of them and count
them one by one. However, this can take a very long time and it is easy to forget some things that should
be on the list, or to write some things twice. It speeds things up to use a computer to generate the list, but
even that can be slow if the list is very long.
One strategy is to split up the items into shorter lists. We can count the number on each shorter list
and then add those numbers together. For example, student E has climbed 5 tall mountains and 8 short
mountains so has climbed 5 + 8 = 13 mountains in total.
The more general principle is this:
2 If you are not familiar with cards or poker, there is a lot of helpful information in Section 3.8.

20
 Addition Principle: The sum a + b counts the total number of things in a collection formed
by adding a collection of b things to a collection of a things.
Example 2.2.1. Suppose we wish to count the number of two-digit positive integers that end in either 0
or 1. We can first count those that end in 0: there are 9 of them, one for each possible tens digit. There are
also, similarly, 9 of them that end in 1. So there are 9 + 9 = 18 such numbers in total.
Suppose now that we have 90 pieces of fruit (specifically apples, bananas, and oranges) in your fruit-
storing refrigerator. If we wanted to count how many apples and bananas we had, we could add up the
number of apples and the number of bananas. Let’s say there were a total of 87 apples and bananas - this
is a lot of fruit to count (and carry!). Is there an easier way to do this?
Subtraction Principle: If b things are removed from a collection of a things, then there are
a − b things left.
We can easily spot that there are 3 oranges in our refrigerator while the rest of the fruit are apples and
bananas. Thus, we have 90 − 3 = 87 apples and bananas.
Example 2.2.2. Suppose we wish to count the two-digit positive integers that end in either 0, 1, 2, 3, 4, 5, 6, 7,
or 8. We could do it by adding 9 to itself 9 times, but we could alternatively use the subtraction principle
by counting a larger set and subtracting the ones we don’t want!
In particular, there are 90 two-digit positive integers, namely the numbers from 10 up through 99. Exactly
9 of them end in the digit 9, and so we subtract these from the total of 90 to get 90 − 9 = 81 of them that
don’t end in a 9.

Exercises
1. Identify when students a, b, c, and d used (or should have used) the addition and subtraction principles
in Section 2.1.
2. Identify a place in Chapter 1 where the addition (resp. subtraction) principle was used.
3. Suppose you need to choose one more class to complete your schedule for next semester at CSU. You
need to choose exactly one course from the following list: math, science, and music theory. The math
class is offered at six different times, the science course is offered at two different times and the music
theory class is offered at three different times throughout the week. If none of the class times create a
conflict with your current schedule, how many options do you have for choosing a class?
4. How many two-digit numbers are not divisible by 5?
5. How many ways are there to choose four numbers from 1, . . . , 100, such that at most one is even?
6. How many numbers from 1 to 36 are not divisible by 3?
7. Suppose you want to make a bouquet of flowers native to Colorado. You have 53 flowers and 3 types:
poppymallow, harebell, and sunflowers. If you have 22 poppymallows and 17 harebells, how many total
flowers would be in a bouquet of all the sunflowers and poppymallows?

2.3 The multiplication and division principles

There are two other principles, called the multiplication and division principles. Section 2.2’s video includes
the multiplication principle, but the division principle is covered in the following video.

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=Q0gk4yBuVj0

Here is an example of the multiplication principle at work.

21
Example 2.3.1. Student k owns 15 shirts and 6 pairs of pants. How many possible outfits (consisting of
one shirt and one pair of pants) are possible? Well, to form an outfit, student k first chooses a shirt (in 15
ways) and then a pair of pants (in 6 ways). So there are 15 · 6 = 90 possible outfits.
The more general principle is this:

Multiplication Principle: The product a · b counts the number of ways to choose one thing in
a ways and then another in b ways.

Example 2.3.2. Suppose we wish to count the number of two-digit positive integers that end in either 0, 4,
or 8. There are 9 ways to choose the first digit, and 3 possibilities for the second digit, for a total of 9 · 3 = 27
possibilities.
Suppose now we want to sell our 90 pieces of fruit at the local farmer’s market. We can fit 10 pieces of
fruit in each fruit basket. Then we need 90/10 = 9 baskets to carry all the fruit.

Division Principle 1: If a things are grouped evenly into collections of size b, there are a/b
collections.

Let’s consider another scenario.

Example 2.3.3. Suppose 36 CSU students are traveling to Denver for a combinatorics conference. If each
university van seats 4 people, 36/4 = 9 vans are needed to transport the students.
Example 2.3.4. How many lotto tickets are there, if a lotto ticket is a collection of 5 different numbers
from {1, . . . , 90}, where order does not matter.
We already know one way to solve this problem! The number of lotto tickets is the number of ways to
choose 5 numbers out of 90. This equals
 
90 90! 90 · 89 · 88 · 87 · 86
= = = 43, 949, 268.
5 5! · 85! 5!
We can do this problem a different way using the division principle to gain a better understanding why
we divide by 5!. There are 90 · 89 · 88 · 87 · 86 ways to pick 5 different numbers from 1 to 90 one at a time.
We put two lotto tickets in the same pile if the numbers on them are the same, just written in a different
order. Then each pile has 5! = 120 lotto tickets in it. Since the order of the numbers on a lotto ticket does
not matter, we have counted each ticket 5! times. Hence the total number of lotto tickets is 90·89·88·87·86
5! .
An alternative formulation of the division principle is as follows:
Division Principle 2: If a things are grouped evenly into c collections of equal size, the number
of things in each collection is a/c.
For example, if we divide 90 pieces of fruit evenly into 9 baskets, then there are 10 pieces of fruit in each
basket. Similarly, if 36 CSU students driving to Denver divide themselves evenly into 9 cars, then there are
4 people in each car. We end this section with a more challenging example that can be solved with either
the division or multiplication principle.
Example 2.3.5. How many different ways can 10 people be grouped into 5 pairs of two?
One approach uses the multiplication principle: Find a predetermined way of ordering the people, such
as age). The youngest person has 9 possible partners, then the youngest remaining has 7 possible partners,
then the youngest remaining has 5 possible partners, then the youngest remaining has 3 possible partners,
and then the last two people will be partners. Hence the total number of ways is

9 · 7 · 5 · 3 · 1 = 945.

Another solution uses the division principle: There are 10! ways to organize the people in a line. We pair
the first and second people, the third and fourth people, etc. There are 5! ways to rearrange the ordering of

22
the pairs, which does not affect who is paired with whom. Also, there are 2 ways to switch the order of the
people in each of the five pairs, which also does not affect who is paired with whom. So the total number of
pairs is
10!
= 945.
5! · 25

Exercises
1. How many three-digit positive integers are even?
2. You have 6 distinct marbles and 2 identical buckets. How many different ways can you put 3 marbles
in each bucket?
3. In how many ways can you attach the letters of MATH in some order to a piece of string? (If you turn
the string over so that the order of the letters is reversed, it is not considered a different possibility.
So for instance, MATH and HTAM are considered the same ordering.)
4. There are 10 people in a dance class. How many ways can they pair off into 5 pairs of dance partners?
(Gender does not matter in making the pairings.)
5. There are 10 people in a dance class. In how many ways can the instructor choose 4 people and put
them into two pairs to demonstrate a dance move?
6. Explain why the sum of all the degrees of the vertices of any graph is even. (Hint: How many times is
each edge counted?)
7. There are 30 granola bars. Decide which sentence uses division principle 1 and which uses division
principle 2:
• If there are 5 granola bars in each box, then the number of boxes is 6.
• If there are 6 boxes of granola bars, then each box contains 5 granola bars.
8. How many ways are there to group 18 students into 9 teams of 2?
9. How many ways are there to group 18 students into 6 teams of 3?
10. Identify when students a, b, c, and d used (or should have used) the multiplication and division
principles in Section 2.1.
11. Identify a place in Chapter 1 when the multiplication (resp. division) principle was used.

2.4 Combining the principles

One difficulty in solving counting problems is that there are often several different ways to approach the
problems. Small errors in the set-up can make an answer completely wrong. In this section, we review the
four principles and explain how to distinguish between them and combine them.
We summarize the four basic principles in the following table. Take care that you’re using the right
principle at each step!

Expression Combinatorial meaning

a+b The total size of a collection formed by adding b things to a things.
a−b The number of elements left after removing b things from a things.
a·b The number of ways to choose one thing from a things and then one from b things.
a/b If a things can be sorted into collections of size b, then a/b is the number of
collections.

23
Remark 2.4.1. Here is another helpful tip for distinguishing addition and multiplication when counting.
Addition is used when making one choice or another. Multiplication is used when making one choice then
another (when the choices are taken in succession and do not depend on the earlier choices).
Example 2.4.2. Suppose there are 3 car models and 4 bike models. In how many ways can you buy one
vehicle? The answer is 3 + 4 = 7. In how many ways can you buy one car and one bike? The answer is
3 · 4 = 12.
Here is an example of a problem that requires combining the principles together.
Example 2.4.3. Bob has 5 brown shirts and 4 blue shirts, and 3 brown pants and 4 blue pants. He wants
to pick out an outfit in which his shirt and pants have the same color. How many ways can he do this?
We can split the problem into two cases: either he’ll wear a brown outfit or a blue outfit, and we will
add the results together by the addition principle.
For the brown case, he needs to pick one of 5 brown shirts and then one of 3 brown pants, so we
multiply these to get 5 · 3 = 15 possibilities.
For the blue case, he needs to pick one of 4 blue shirts and then one of 4 blue pants, so we multiply
these to get 4 · 4 = 16 possibilities.
Adding these together, we get 15 + 16 = 31 outfits in all.

Exercises
For the following exercises, make sure you identify each time you use one of the four principles.

1. Five people and five dogs meet in a park. The people all shake hands with each other and the people
all shake hands with the dogs, but the dogs do not shake each other’s hands. How many handshakes
took place?
2. The five people and five dogs meet in a park. How many ways can the people pair off with the dogs to
go for a walk?
3. How many three-digit numbers have two even digits and one odd digit?
4. How many rearrangements of the word MATH start with a consonant (M, T, or H)?
5. How many ways are there to group 100 students into 50 pairs of 2?
6. How many odd three-digit numbers are there that don’t end with 9?
7. A drama teacher has 30 student actors and actresses that make up a cast for the school play. Each
student has one role, where 10 students are in the chorus, 15 students are in supporting roles, and the
rest of the students are lead characters. There are 450 total costumes for the cast. If each cast member
wears the same number of costumes and all costumes are worn, how many total costumes are worn by
the lead characters?
8. A donut shop has 100 total donuts, where there are 38 vanilla frosted, 21 blueberry, 26 chocolate, and
the rest are jelly-filled. A very hungry customer wants all of your jelly-filled and blueberry donuts.
How many boxes do you need if each box holds six donuts? Note: it does not matter what donuts go
in what box.
9. You order dinner for a friend at a concession stand during a CSU soccer game. You notice both burgers
and burritos are being sold. To order a burger you must choose one option from each category:
• Bun: wheat or sesame seed
• Patty: beef, veggie, or turkey
• Sauce: ketchup, mustard, relish, mayonnaise, none

24
 To order a burrito, you must choose one option from each category:

• Tortilla: white, wheat, or corn

• Filling: shrimp, carnitas, carne asada, or fajita veggies
• Cheese: cheddar, pepper jack, none
• Rice: brown, jasmine

How many possible dinners could you order for your friend if you buy them either a burger or a burrito
(but not both)?

2.5 Sets, subsets, and the number of subsets

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=vwSn9UnHa1I

Sets are unordered collections, like a handful of marbles or a bag of fruit. Many counting problems can
best be phrased in terms of sets and subsets. In Sections 2.6 and 2.8, we will rephrase the four principles in
terms of sets.

2.5.1 Basic definition of sets

Definition 2.5.1 (Informal). A set is a collection of distinct objects (in no particular order). The objects
in a set are called elements. We write the objects in a set inside squiggly brackets: {}.3

For example, {a, b, c} and {1, 2, 3, 4} are both sets. The set {a, b, c} is the same set as {b, c, a} because
the order of the elements does not matter.

Notation 2.5.2. We write b ∈ A to mean that b is an element of the set A. Similarly, b 6∈ A if b is not an
element of A.

Example 2.5.3. For example, 4 ∈ {2, 4, 5} but 3 ∈

/ {2, 4, 5}.

It is sometimes helpful to draw a set in “blob notation”, consisting of a circle with points inside labeled
by the elements of the set. For instance, the figure below depicts the set {2, 4, 5}.

Definition 2.5.4. If A is a finite set, then its size (or cardinality) (written |A|) is the number of elements
in A.

For example, |{2, 4, 5}| = 3. The size of a set is found by counting the number of elements in the set.
For this reason, counting problems can often be phrased in terms of computing the cardinality of a set.
There is also a notion of cardinality for infinite sets, but we will not cover this topic in this document.
There are even different sizes of infinite sets!
3 Later in the book, we will consider multi-sets, where the objects in the set can be the same, for example {a, a, b}.

25
2.5.2 Set builder notation
Writing down sets of large or infinite size can be hard. One possible notation for the infinite set of positive
integers is {1, 2, 3, 4, . . .}, but this is a bit ambiguous. How do we know whether a number like 7 is included?
To describe larger or more complicated sets, without laboriously writing out every element, we use set builder
notation, as in the definition below.

Definition 2.5.5. The notation

{x ∈ A | x has property P }
represents the set of all elements x in the set A that satisfy property P . Sometimes we simply write
{x | x has property P } if the set A containing x is clear.

Example 2.5.6. {x ∈ {2, 4, 5} | x is even} = {2, 4}.

Example 2.5.7. Let V be the set of all integers whose last digit is zero. Then 30 ∈ V but 31 6∈ V . Another
description of V is
V = {10 · k | k is an integer}.

2.5.3 Famous sets

It is helpful to fix notation for a few important infinite sets that are used frequently in mathematics:

Definition 2.5.8. ∅ = { } is the empty set, the unique set containing no elements.
N = {0, 1, 2, 3, 4, . . .} is the set of nonnegative integers (or natural numbers).4
Z = {. . . , −2, −1, 0, 1, 2, . . .} is the set of integers.
Q = {a/b | a ∈ Z and 0 6= b ∈ Z} is the set of fractions (rational numbers).
R is the set of real numbers.
√
Example
√ 2.5.9. The number 1.73 ∈ Q is a good approximation for 3. In Section 5.5, we will prove that
3∈
/ Q.

Example 2.5.10. In the 1760s, Lambert proved that the real number π is not in Q.

Example 2.5.11. There are several ways to describe the set of all even natural numbers:

{x ∈ N | x is even} = {0, 2, 4, 6, 8, . . .} = {2 · k | k ∈ N}.

2.5.4 Subsets
Informally, a subset is a set contained in a bigger set.

Definition 2.5.12. A set A is a subset of a set B (written A ⊆ B) if every element of A is also an element
of B.

For instance, {2, 4} ⊆ {2, 3, 4}, but {2, 4, 5} 6⊆ {2, 3, 4}.

Example 2.5.13. Let A be any set. Then A ⊆ A because every element of A is an element of A. Also
∅ ⊆ A because there are no elements of ∅.

Example 2.5.14. For instance, all natural numbers are integers and all integers are rational numbers. All
rational numbers are real numbers. We can write this as a nested sequence of inclusions

∅ ⊆ N ⊆ Z ⊆ Q ⊆ R.
4 In this book, we include 0 ∈ N, but many people exclude 0 from being a natural number.

26
2.5.5 The number of subsets
We now address a famous counting question involving sets, which is to count the number of subsets of a set.
Here is an example of a concrete application of this problem.
Example 2.5.15. Four students say that they might go to office hours on Monday. How many different
groups might show up in office hours?
The above problem can be modeled by counting the possible subsets of the set {a, b, c, d} of four students.
To answer this question in general, we need to count the number of subsets in a set A of size n. Before
we dive into the theorem, let’s look at a few cases:
• n = 1: The subsets of A = {a} are ∅ and {a}, so a set with 1 element has 2 subsets.
• n = 2: The subsets of A = {a, b} are ∅, {a}, {b}, and {a, b}, so a set with 2 element has 4 subsets.
• n = 3: The subsets of A = {a, b, c} are ∅, {a}, {b}, {c}, {a, b}, {a, c}, {b, c}, and {a, b, c}, so a set with
3 element has 8 subsets.
In each case, our answer was a power of 2, with 21 = 2, 22 = 4, and 23 = 8. For a set {a, b, c, d} with
four elements, like in Example 2.5.15, we might guess that the number of subsets is 24 = 16.
The following theorem proves that this pattern holds for any set of finite size.
Theorem 2.5.16. If S is a set with n elements, then S has 2n subsets.
Proof of Theorem 2.5.16. Let’s label the elements of S as a1 , a2 , a3 , . . . , an , so that S = {a1 , a2 , a3 , . . . , an }.
Picking a subset of S is the same as
• First, choosing if a1 is in the subset or not (2 choices),
• Next, choosing if a2 is in the subset or not (2 choices),
..
.
• Finally, choosing if an is in the subset or not (2 choices).
At each step i, there are 2 choices (include ai or not). Since we make a decision n times, the total number
of subsets is 2| · 2 ·{z. . . · 2} = 2n by the multiplication principle.
n times

Remark 2.5.17. Here is a way to visualize this proof when n = 3. Let’s say S = {a1 , a2 , a3 }.

no a1 in subset? yes
a2 in subset? a2 in subset?

no yes no yes

a3 in subset? a3 in subset? a3 in subset? a3 in subset?

no yes no yes no yes no yes

∅ {a3 } {a2 } {a2 , a3 } {a1 } {a1 , a3 } {a1 , a2 } {a1 , a2 , a3 }

27
Remark 2.5.18. Theorem 2.5.16 even works when n = 0, since the empty set ∅ is the unique set with 0
elements, and there is 1 = 20 subset of ∅, namely ∅.

Remark 2.5.19. There is a more abstract way to think about Theorem 2.5.16. We can make another set
T , whose elements are the subsets of S, which are sets themselves. Then 2n is the cardinality of T .

Example 2.5.20. Students a, b, c say that they might show up to office hours. Among the 8 possible
outcomes, how many of them have an even number of students coming to office hours. The answer is 4,
namely
∅, {a, b}, {a, c}, {b, c}.

Exercises
1. Given the following sets, describe in words what the set contains. (There will be more than one correct
description)

(a) A = {2n | n ∈ Z}
(b) B = {x | 0 ≤ x < 2}
(c) C = {(x, y) | y = x2 }
(d) D = {y = mx + b | m = 3}

2. Given the following mathematical descriptions, write a set in set builder notation that matches the
description. (There will be more than one correct answer)

(a) The set containing all natural numbers greater than or equal to ten.
(b) The set containing all integers divisible by 3.
(c) The set containing nothing.
(d) The set of all perfect squares.

3. Let S = {1, . . . , 40}. List the elements in {x ∈ S | x is a power of a prime}.

4. You have 20 different books and you want to donate a subset of them to the library. The only
requirements are that you want to donate at least 1 book and you do not want to donate all 20 books.
How many different donations can you make?

5. This problem is about the subsets of {a, b, c, d}. First, list all the subsets.

(a) How many of them have odd order?

(b) How many of them contain {a}?
(c) How many have size 0, size 1, size 2, size 3, and size 4?

6. This problem is about the subsets of {a, b, c, d, e}. Without listing these subsets, make conjectures
about the following problems.

(a) What proportion of them have even order?

(b) What proportion of them contain a?
(c) How many have size 0, 1, 2, 3, 4, and 5?

7. What is the output in SAGE when you type S=Subsets([1,2,3,4,5,6]) in line 1 and S.cardinality() in
line 2?

8. If A is a set of size 4 and B is a set of size 7, what is the smallest possible size of the union of A and
B?

28
 9. You have 6 different books and you want to donate a subset of them to the library. The only require-
ments are that you want to donate at least 1 book and you do not want to donate all 6 books. How
many different donations can you make?
10. How many subsets of {1, 2, 3, 4, 5} contain the number 1?
11. If each student a, b, c, d, and e receives either 0 or 1 point for the quiz, how many sequences of five
numbers are possible as the list of their grades?

2.6 Addition and subtraction from the perspective of set theory

2.6.1 Union, intersection, and the addition principle
We now can define the set operations that allow us to phrase the addition principle in terms of sets.
Definition 2.6.1. The union of sets A and B is

A ∪ B = {x : x ∈ A or x ∈ B}.

Here “or” means x ∈ A or x ∈ B or both. (This is a different meaning of the word “or” than in “soup
or salad” or “paper or plastic”.)
Example 2.6.2. We see that {2, 4, 5} ∪ {2, 5, 7, 9} = {2, 4, 5, 7, 9}. The union consists of all the elements in
the region of the diagram covered by the overlapping circles.

Definition 2.6.3. The intersection of sets A and B is

A ∩ B = {x : x ∈ A and x ∈ B}.

Example 2.6.4. We have {2, 4, 5} ∩ {2, 5, 7, 9} = {2, 5}. In the diagram, it is the common region shared by
the overlapping circles.

Here is the set analogue of the addition principle.

Lemma 2.6.5 (The addition principle for sets). If A and B are finite sets with no elements in common
(that is, A ∩ B = ∅), then the union A ∪ B has size |A| + |B|.
Example 2.6.6. A student has climbed the set of big mountains A = {Longs peak, Pikes peak, Mt. Evans}
of size |A| = 3, and the set of small mountains B = {Arthur’s rock, Horsetooth} of size |B| = 2. Since there
is no overlap between the sets of large and small mountains, E has climbed |A| + |B| = 5 mountains.

29
2.6.2 Subsets, complements, and the subtraction principle
We now introduce the set operations needed to understand the subtraction principle in terms of sets. One
example of subsets can be constructed with the set difference operator as follows.
Definition 2.6.7. The difference of sets A and B is

A − B = {x ∈ A | x 6∈ B}.

Example 2.6.8. We see that {2, 4, 5} − {2, 5, 7, 9} = {4}, and {2, 5, 7, 9} − {2, 4, 5} = {7, 9}.
Definition 2.6.9. If B ⊆ A, then the complement of B in A is the difference A − B. If the set A is clear
from context, then this is written as B c , where the c stands for complement.

We can now state the set analogue of the subtraction principle.

Lemma 2.6.10 (The subtraction principle for sets). If A and B are finite sets and B ⊆ A, then |A − B| =
|A| − |B|.
Example 2.6.11. How many numbers in A = {1, . . . , 25} are not a multiple of 5?
To answer this, let B = {5, 10, 15, 20, 25}. Then B ⊆ A and B is exactly the subset of elements of A that
are a multiple of 5. So |A − B| = 25 − 5 = 20.

Exercises
1. Let A and B be sets with sizes |A| = 4 and |B| = 7.
(a) What are all the possible values of |A ∩ B|?
(b) What are all the possible values of |A ∪ B|?
(c) What are all the possible values of |B − A|?
2. See Question 2.1.1. Let A be the set of subsets of size 6 in {1, . . . , 20} which contain at least 4 odd
numbers.
(a) Write down several elements of A.
(b) Describe A as the union of three smaller subsets.
(c) Use the addition principle for sets to find |A|.
3. What is A − B if A ∩ B = ∅?
4. Find particular sets A, B, and C so that A − (B − C) 6= (A − B) − C.
5. Let A = {1, 2, 3, 4, 5} and B = {4, 5, 6, 7} be subsets of W = {x | 1 ≤ x ≤ 10, x ∈ Z}. Compute the
following:
(a) B c
(b) Ac
(c) A ∪ B

30
 (d) (A ∪ B)c
(e) Ac ∪ B c
(f) A ∩ B
(g) (A ∩ B)c
(h) Ac ∩ B c

Are any of the sets equal? If so, identify them.

2.7 Venn diagrams and the Principle of Inclusion-Exclusion

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=yhTBE8L3pro

In this section, we cover the Principle of Inclusion-Exclusion, which is a way to count the number of
elements in a union of sets when the sets intersect non-trivially. Drawing sets as overlapping circles to form
a Venn diagram is a helpful way to visualize the intersection of several sets.

Example 2.7.1. Students a, b, c, d, e, f , and g decide to walk to the local ice cream shop in Old Town after
their combinatorics class. As they walk, student d takes a quick poll on who likes mint ice cream and who
likes caramel ice cream. Students a and f love both mint and caramel. It would be a shame to choose just
one! So they vote for both. Tallying the votes, there are 5 votes for mint ice cream and 4 votes for caramel,
for a total of 9 votes. Student d quickly realizes that something is wrong and figures out the following Venn
diagram.

Here is another example of a problem that can be solved with a Venn diagram.

Example 2.7.2. How many numbers in the set S = {1, . . . , 210} have no factor in common with 21?
A number has a factor in common with 21 = 3 · 7 if it is a multiple of 3 or a multiple of 7 or both. Let
A be the set of multiples of 3 in S and let B be the set of multiples of 7 in S. The intersection A ∩ B is the
set of multiples of 21.
We want to count the elements that are not in A or B. This is the complement of A ∪ B in S. The
number of these is
|S − (A ∪ B)| = 210 − |A ∪ B|.
There are 70 multiples of 3 in S so |A| = 70. There are 30 multiples of 7 in S so |B| = 30. But if we
subtract both 70 and 30 from 210, we’ve removed each multiple of 21 twice, since it is divisible by both 3
and 7. There are 10 multiples of 21 that we removed twice, so we ad 10 to the count. So the answer is
210 − 70 − 30 + 10 = 120.

We state the Principle of Inclusion-Exclusion for two sets and then for three sets.

31
Theorem 2.7.3 (Principle of Inclusion-Exclusion for two sets). Suppose A and B are two finite sets. Then

|A ∪ B| = |A| + |B| − |A ∩ B|.

Remark 2.7.4. In Example 2.7.2, A and B are subsets of S = {1, . . . , 210}; A is the set of multiples of 3;
B is the set of multiples of 7. We calculated

|A ∪ B| = |A| + |B| − |A ∩ B| = 70 + 30 − 10 = 90.

The set of all numbers in S that are not divisible by 3 or 7 is the complement of A ∪ B in S, so it has size
|S − (A ∪ B)| = 210 − 90 = 120.

There is also a version of the inclusion-exclusion principle for three sets.

Theorem 2.7.5. Suppose A, B, C are three finite sets. Then

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

Stare at the following pictures for a while to understand why Theorem 2.7.5 is true!

Example 2.7.6. Continuing Example 2.7.1, two of the students a, b, c, d, e, f , and g also like strawberry
ice-cream. There is 1 student who likes mint and strawberry but not caramel. There are no students who
like strawberry and caramel but not mint. Show that there is exactly 1 student who likes all three flavors.
To solve this, we fill in the numbers in each part of the Venn diagram. We have the additional information
that there is no student who likes only strawberry, because all 7 students like either mint or caramel or both.
Among the 2 students who like strawberry, 1 also likes mint but not caramel and 0 like caramel but not
mint. This shows there is only one student who likes all three flavors.

Example 2.7.7. How many numbers in the set S = {1, . . . , 210} have no factor in common with 30?

32
 A number has a factor in common with 30 = 2 · 3 · 5 if it is a multiple of 2 or 3 or 5 or any combination
of these. Let A be the set of multiples of 2 in S. Let B be the set of multiples of 3 in S. Let C be the set
of multiples of 5 in S. Then
|A| = 105, |B| = 70, and |C| = 42.
It is useful that the numbers 2 and 3 and 5 are all relatively prime. So A ∩ B is the set of multiples of 6 in
S. Also A ∩ C is the set of multiples of 10 in S. Also B ∩ C is the set of multiples of 15 in S. So

|A ∩ B| = 35, |A ∩ C| = 21, and |B ∩ C| = 14.

Finally, A ∩ B ∩ C is the set of multiples of 30 in S so it has size 7. Using Theorem 2.7.5, we see that

|A ∪ B ∪ C| = 105 + 70 + 42 − 35 − 14 − 21 + 7 = 154.

We want to count the size of the complement of A ∪ B ∪ C in S since the complement contains the numbers
that have no factor in common with 30. It has size 210 − 154 = 56.

Exercises
1. What is |A ∪ B| if |A| = 10, |B| = 7, and |A ∩ B| = 3?

2. Suppose |A| = 10 and |B| = 7. What is the largest and smallest possible size for |A ∪ B|?

3. Suppose A and B are sets such that |A − B| = 17, |A ∩ B| = 5, and |A ∪ B| = 40. What is |B|?

4. Suppose A, B, and C are sets such that |A| = 10, |B| = 17, |A ∩ B| = 5, and |A ∪ B ∪ C| = 30. What
is the largest and smallest possible size of |C|?

5. Students a, b, and c compare their CSU Fall class schedules. Student a is registered for combinatorics,
art history, cooking, and physics. Student b is registered for combinatorics, health, astronomy, band,
and dance. Student c is registered for combinatorics, art history, dance, English literature, and African
American studies. Use the principle of inclusion-exclusion to compute how many classes they are taking
in total.

6. How many numbers in {1, . . . , 35} are relatively prime to 35 (i.e., have no factors in common with 35)?

7. How many integers in {1, 2, ...., 55} are not divisible by either 11 or 5?

8. How many integers in {1, 2, ...., 105} are not divisible by 3 or 5 or 7?

9. Suppose 50 socks lie in a drawer. Each one is either white or black, ankle-high or knee-high, and either
has a hole or doesn’t. 22 socks are white, four of these have a hole, and one of these four is knee-high.
Ten white socks are knee-high, ten black socks are knee-high, and five knee-high socks have a hole.
Exactly three ankle-high socks have a hole. Find the number of black, ankle-high socks, with no holes.

2.8 The multiplication principle from the perspective of set theory

We now introduce the set theory notation needed to describe the multiplication principle in terms of sets.

2.8.1 Cartesian products and the multiplication principle

If A and B are sets, then their Cartesian product, denoted A × B, is the set consisting of all ordered pairs
of elements from A and B:
A × B = {(a, b) | a ∈ A and b ∈ B}.

33
Example 2.8.1. The product {2, 3, 4} × {x, y} is the set of ordered pairs
{(2, x), (2, y), (3, x), (3, y), (4, x), (4, y)}.
Example 2.8.2. The product {2, 3, 4} × {2, 5} is the set of ordered pairs
{(2, 2), (2, 5), (3, 2), (3, 5), (4, 2), (4, 5)}.
Example 2.8.3. The product R × R can be thought of as all the points (x, y) in the usual coordinate plane
R2 that we use when graphing functions in calculus. This plane is often called the Cartesian plane, which is
where the term “Cartesian product” comes from.
We can now state the set analogue of the multiplication principle.
Lemma 2.8.4 (Multiplication principle for sets). If set A has |A| elements and set B has |B| elements then
set A × B has |A| · |B| elements.
Example 2.8.5. We arrange three boxes of granola bars in a line. In each box, five granola bars are stacked
up. We index each granola bar by a pair (i, j) where i ∈ A = {1, 2, 3} indexes the box and j ∈ B = {1, . . . , 5}
indexes the spot in the box. Then the number of granola bars is |A × B| = 3 · 5 = 15.
Example 2.8.6. Let A be the set of numbers in {1, . . . , 40} which are odd and not a multiple of 5. Suppose
we wish to find the size of A.
The numbers in A are exactly those whose last digit is in C = {1, 3, 7, 9} and whose first digit is in
B = {0, 1, 2, 3}. So A can be identified with B × C.5 So |A| = |B| · |C| = 4 · 4 = 16.

Exercises
1. If Student a owns 5 shirts and 2 pairs of pants, how many outfits consisting of one shirt and one pair
of pants can they wear? Justify your answer using sets.
2. See Question 2.1.2. Let A be the set of poker hands which are a full house (five cards including a triple
of one number and a pair of another number).
(a) Write down one example of a full house.
(b) With a classmate (or friend), play 20 questions until they guess which full house you wrote down.
For example, one question could be: do the cards in your triple have the number 7 on them?
(c) Describe A as the Cartesian product of several smaller sets. Hint: each smaller set tells you some
information, but not all information about the full house. For example, set B could index the
number on the cards in the triple.
(d) Use the multiplication principle for sets to find |A|.
3. Let W = {(1, 1), (2, 1), (2, 2), (3, 1), (1, 2), (3, 2)}. Find sets A and B such that A × B = W . verify the
multiplication principle with your sets.
4. Express R × R as a set. What does this represent?
5. Express [0, 1] × [0, 1] as a set. What does this represent?

2.9 Set partitions, the division principle, and equivalence relations

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=iqP5fyUL1MU

5 More precisely, there is a bijection from A to B × C where the number 10c + d maps to the pair (c, d). The topic of bijections

is covered in Section 5.4.

34
2.9.1 Set partitions and the division principle
The division principle is the most difficult to state in terms of set theory. To do this, we define a set partition,
which is a a helpful way to organize sets.

Definition 2.9.1. Let A be a finite set. A non-empty subset B of A is called a block. A set partition of
a finite set A is a set of blocks {B1 , B2 , . . . , Bk } of A such that:

1. the blocks are disjoint, that is, Bi ∩ Bj = ∅ for all i 6= j; and

2. the union of the blocks is A, that is B1 ∪ B2 ∪ · · · ∪ Bk = A.

Example 2.9.2. The following is a set partition of {1, 2, 3, 4, 5, 6, 7}:

{{1, 3, 4}, {2, 6}, {5, 7}}.

The division principle arises when we split a set into blocks of equal sizes. It can be stated as follows.

Lemma 2.9.3 (Division principle for sets). Suppose that {B1 , . . . , Bk } is a set partition of A such that each
of the blocks B1 , . . . , Bk has the same cardinality m. Then the number of blocks and the size of the blocks
satisfy these relationships: |A|/m = k, and |A|/k = m.

Example 2.9.4. How many ways can the letters in the word “COOL” be rearranged? If we pretend the
two O’s are different from each other, calling them O1 and O2 , then there are 4! = 4 · 3 · 2 · 1 = 24 ways
to order them, as discussed in section 1.2.2. We can partition these arrangements into blocks of size 2 that
have the letters in the same position except for the two O’s switched with each other.

{{CLO1 O2 , CLO2 O1 }, {CO1 LO2 , CO2 LO1 }, {CO1 O2 L, CO2 O1 L}, {LCO1 O2 , LCO2 O1 },
{LO1 CO2 , LO2 CO1 }, {LO1 O2 C, LO2 O1 C}, {O1 CLO2 , O2 CLO1 }, {O1 CO2 L, O2 CO1 L},
{O1 LCO2 , O2 LCO1 }, {O1 LO2 C, O2 L O1 C}, {O1 O2 CL, O2 O1 CL}, {O1 O2 LC, O2 O1 LC}}

Thus there are 24/2 = 12 blocks, or 12 distinct rearrangements of the letters in COOL. See Section 4.5 for
more examples of counting anagrams.

2.9.2 Equivalence relations and modular arithmetic

Equivalence relations are often used to divide a set into blocks. In fact, every equivalence relation on a set
makes a set partition.

Definition 2.9.5. A equivalence relation on a set S is a way of identifying elements x ∼ y with the three
following rules:

1. (reflexive) x ∼ x;

2. (symmetric) if x ∼ y, then y ∼ x;

3. (transitive) if x ∼ y and y ∼ z, then x ∼ z.

Example 2.9.6. How many ways can students A, B, C, D be seated at a rotating round table? There are
24 = 4! ways for the students to sit down at the table. We say that two seating arrangements are equivalent
if they look the same after a rotation. For example,

ABCD ∼ DABC ∼ CDAB ∼ BCDA.

Each seating arrangement is equivalent to three others. So we can divide the set of seating arrangements
into blocks, each of which has size 4. By the division principle, there are 6 blocks.

One important equivalence relation is called congruence modulo m.

35
Definition 2.9.7. Let m be a positive integer. Let x, by be integers. We say that x is congruent to y
modulo m when m divides y − x. This is written x ≡ y mod m.

Lemma 2.9.8. If r is the remainder when x is divided by m, then x ≡ r mod m. In particular, if m divides
x, then x ≡ 0 mod m.

Proof. If r is the remainder when a is divided by m, then a = km + r for some integer k. This means that
a−r = km, which is divisible by m. So a ≡ r mod m by definition. If m divides a = a−0, then a ≡ 0 mod m
by definition.

Example 2.9.9. The hours of a day are described with congruence modulo 12. For example, if it is 11
o’clock and you need to wait 2 hours before eating lunch, then you will eat lunch at 1 o’clock. We write
11 + 2 = 13 ≡ 1 mod 12.

Example 2.9.10. Every even integer is congruent to 0 modulo 2. Every odd integer is congruent to 1
modulo 2.

Example 2.9.11. The last digit of a number indicates its congruence modulo 10.

We will now give some examples of the division principle using blocks that are constructed using congru-
ence modulo m.

Example 2.9.12. How many numbers in the set S = {1, . . . , 40} are congruent to 1 modulo 5?
We list the numbers between 1 and 40 that are congruent to 1 modulo 5:

{1, 6, 11, 16, 21, 26, 31, 36}.

So the answer is 8. In this example, there is a set partition of S into 5 blocks of size 8, where two numbers
are in the same block exactly when they are congruent modulo 5.

This example works more generally.

Lemma 2.9.13. Suppose N is a multiple of m. Let S be the set S = {1, . . . , N }. Then S has a set partition
into m blocks, each of size N/m, where two numbers are in the same block exactly when they are congruent
modulo m.

Exercises
1. Let A = {a, b, c, d}. List all possible set partitions.

2. How many ways can the letters in “COLORADO” be rearranged.

3. How many ways can five students be seated at a rotating round table?

4. How many numbers in S = {1, . . . , 40} are congruent to 3 modulo 8? What are they?

5. Let S = 35 and m = 5. Find the set partition of S = {1, . . . , 35} from Lemma 2.9.13. Repeat this
question with m = 7.

6. Show that congruence is an equivalence relation.

7. If n is an integer, show n2 ≡ 0 mod 4 if n is even and n2 ≡ 1 mod 4 if n is odd.

8. Explain why Lemma 2.9.13 is true using Lemma 2.9.3.

36
2.10 Additional Problems for Chapter 2
1. Use the multiplication and division principles to find how many handshakes there are between seven
people, if every pair of person shakes hands.
2. Generalize your reasoning in the previous problem to give a proof using the multiplication and division
principles that the number of handshakes between n people, if everyone shakes hands, is n(n − 1)/2.
3. Bob has 4 pairs of socks, 2 pairs of shoes, 5 pairs of pants, 10 shirts, and one hat. How many ways
can he get dressed for the day (wearing one of each type of item of clothing)?
4. How many three-digit positive integers have only even digits?
5. Suppose A and B are sets such that |A| = 21, |B| = 15, and |A ∪ B| = 30. What is |A ∩ B|?
6. Using a Venn diagram, for any three sets A, B, C, explain why it is always true that
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C).

7. Using a Venn diagram, for any three sets A, B, C, explain why it is always true that
A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C).

8. How many numbers in {1, . . . , 105} are relatively prime to 105?

9. Let n = pq where p and q are distinct primes. Show that there are pq − p − q + 1 numbers in {1, . . . , n}
which are relatively prime to n.
10. (a) How many binary strings of length m are there?
(b) If s is a string of length m, let sc be the complement string where each 0 is replaced by a 1 and
each 1 is replaced by a zero. Write out the binary strings of length m = 3 and match each with
its complement.
(c) For each binary string s = s1 s2 s3 , plot the point (s1 , s2 , s3 ) in 3-dimensional space. What 3-
dimensional shape has these points as its corners and what is the relationship between the corners
plotted for s and its complement sc ?
11. Let S be the set of binary strings of length m.
(a) When m = 2, here are all the subsets of S of size k = 2:
{00, 01}, {00, 10}, {00, 11}, {01, 10}, {01, 11}, {10, 11}. (2.1)
Write out all the subsets of S of size k = 3.
(b) If T is a subset of S containing k strings, let T c be the subset of S containing their k complements.
When m = 2 and k = 2, match up each set in (2.1) with its complement. When m = 2 and k = 3,
match up each set for your answer to part (1) with its complement.
(c) Suppose that k is odd. If T is a subset of S of size k, prove that T and T c are not the same.
(d) Suppose k is even. If R is a subset of S of size k/2 and if T = R ∪ Rc , show that T is a subset of
S of size k and that T = T c .
(e) How many sequences of 5 letters can be made with the letters A,E,I,O,U,M,S, allowing repeats
of letters, for example, MMSSU?
(f) How many sequences of 5 letters can be made with the letters A,E,I,O,U,M,S, and have exactly
2 vowels, for example, MMSOU?
(g) How many ways are there to assign 5 jobs to 4 people, so that each person gets at least one job
(and every job is assigned to some person)?
(h) How many ways can the letters of PUPPY be rearranged (including the original spelling)?
(i) In the set {1, 2, ....., 24}, there are 3 integers which are 2 mod 8. What is their sum?

37
2.11 Investigation: Divisors of a positive integer
In this section, we study divisor functions, which were investigated by the famous mathematician Ramanujan
(1887-1920). Let N be a positive integer. A divisor function is a function defined on positive integers whose
value at N depends on the divisors of N .

2.11.1 The number of divisors

Definition 2.11.1. Let σ0 (N ) be the number of positive divisors of N , including 1 and N itself.

Example 2.11.2. • σ0 (22) = 4 because the positive divisors of 22 are 1, 2, 11, 22;

• σ0 (23) = 2 because the positive divisors of 23 are 1 and 23; and

• σ0 (24) = 8 because the positive divisors of 24 are 1, 2, 3, 4, 6, 8, 12, 24.

In the following questions, we will develop a formula for σ0 (N ) by working out cases of increasing difficulty.

Question 2.11.3. 1. Compute σ0 (N ) for N = 1, . . . , 20. Look for some patterns.

2. When is σ0 (N ) = 1? Explain why.

3. When is σ0 (N ) = 2? Explain why.

4. When is σ0 (N ) odd? Explain why.

Question 2.11.4. 1. Compute σ0 (N ) for N = 3, 9, 27, 81.

2. If N = pe for some prime p, find a formula for σ0 (N ).

3. Explain why the formula is true.

4. Give an example to show that the formula does not work when p is not prime.

Question 2.11.5. 1. Compute σ0 (N ) for N = 21, 26, 33, 35.

2. If N = p · q where p and q are distinct primes, find a formula for σ0 (N ).

3. Explain why the formula is true.

4. Let A be the set of multiples of p dividing N and B be the set of multiples of q dividing N . Draw a
Venn diagram for the divisors of N .

Question 2.11.6. 1. Compute σ0 (N ) for N = 30, 42, 70.

2. If N = p · q · r where p, q, r are distinct primes, find a formula for σ0 (N ).

3. Explain why the formula is true.

4. Let A and B be as in Question 2.11.5. Let C be the set of multiples of r dividing N . Draw a Venn
diagram for the divisors of N .

Question 2.11.7. Suppose N = p1 ·p2 · · · pn where p1 , . . . , pn are distinct primes. Find a formula for σ0 (N ).
Explain why the formula is true using Theorem 2.5.16.

Question 2.11.8. Suppose N = pe11 · pe22 · · · penn where p1 , . . . , pn are distinct primes. Find a formula for
σ0 (N ). Explain why the formula is true.

38
2.11.2 The sum of the divisors
Definition 2.11.9. Let σ1 (N ) be the sum of the positive divisors of N , including 1 and N itself.
Example 2.11.10. • σ1 (22) = 36 because 1 + 2 + 11 + 22 = 36;
• σ1 (23) = 24 because 1 + 23 = 24; and
• σ1 (24) = 60 because 1 + 2 + 3 + 4 + 6 + 8 + 12 + 24 = 60.
Question 2.11.11. Let p be a prime.
1. Find a formula for σ1 (p).
2. Find a formula for σ1 (pe ). Explain why it is true.
Question 2.11.12. Let p1 , . . . , pn be distinct primes.
1. Find a formula for σ1 (p1 · p2 ).
2. Find a formula for σ1 (p1 · p2 · · · pn ). Explain why the formula is true.
Question 2.11.13. Let p1 , . . . , pn be distinct primes. Find a formula for σ1 (pe11 · pe22 · · · penn ). Explain why
the formula is true.

2.11.3 The Möbius function

The Möbius function was defined by Möbius in 1832, but had been studied 30 years earlier by Gauss. It
plays a key role in the Möbius inversion formula and the Riemann hypothesis, which is a famous unsolved
problem in mathematics.
Definition 2.11.14. The Möbius function µ(N ) is defined as follows:
• If there is a prime p such that p2 divides N , let µ(N ) = 0;
• µ(1) = 1; and
• If N = p1 · p2 · · · pn where p1 , . . . , pn are distinct primes, let µ(N ) = (−1)n .
Example 2.11.15. Let N = 22. The divisors d of N are 1, 2, 11, 22. We compute µ(1) = 1, µ(2) = −1,
µ(11) = −1, and µ(22) = 1. We compute that µ(1) + µ(2) + µ(11) + µ(22) = 0.
P
Question 2.11.16. For the other values of N between 20 and 30, compute the value of d|N µ(N ), where
the sum is over all divisors of N including 1 and N . Make a conjecture about this.

2.11.4 Multiplicative functions

Let N and M be positive integers. We say that N and M are relatively prime if gcd(N, M ) = 1, meaning
that N and M have no prime factors in common.
Let f be a function on the positive integers. We say that f is multiplicative if f (N · M ) = f (N ) · f (M )
whenever N and M are relatively prime.
Question 2.11.17. 1. Show that σ0 is a multiplicative function.
2. Show that σ1 is a multiplicative function.
3. Show that µ is a multiplicative function.
4. For each of the functions σ0 , σ1 , µ, give an example that shows the multiplicative property fails when
M and N are not relatively prime.

39
40
Chapter 3

Counting combinations

We now arrive at the heart of combinatorics, and the origin of its name, which comes from counting combi-
nations.

3.1 The types of combinations

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=4Aw9vYMbDL8

How many ways can you choose a collection of k objects from a collection of n objects?
What we mean by this question might vary depending on the situation; in particular, whether order
matters and whether or not repeats are allowed. Consider the following examples of each setting.

Example 3.1.1. (Order does not matter, repeats not allowed.) Suppose you want to pick out two pieces
of fruit to bring to work from the refrigerator, which contains an apple,
a banana, an orange, a pear, and a
mango. In this case, order does not matter, so the number of ways is 52 = 2!·3!
5!
= 10.

Example 3.1.2. (Order matters, repeats not allowed.) On the other hand, suppose you want to eat one
fruit on each of Monday and Tuesday next week. In this case, order matters. Now there are 5 choices for
which fruit to eat on Monday, but then when you get to Tuesday there are only 4 choices left, so there are
a total of 5 · 4 = 20 choices.

Example 3.1.3. (Order matters, repeats allowed.) Suppose instead that your refrigerator is packed with
plenty of fruits of each of the five types! So for instance, you can eat an apple on both Monday and Tuesday
if you like - indeed, repeats are allowed. Now there are 5 choices for which type of fruit to eat each day, for
a total of 5 · 5 = 25 possibilities.

Example 3.1.4. (Order does not matter, repeats allowed) Finally, suppose repeats are allowed (there are
plenty of each type of fruit in the fridge) but order doesn’t matter (you are just grabbing two pieces of fruit
to put in your lunch bag for Monday). Now there are 5 ways to pick the two pieces of fruit to be the same
kind as each other, and if they’re different, there are 10 possibilities as we saw in Example 3.1.1. In total,
there are 10 + 5 = 15 possibilities.

The examples above show that the answer of “how many ways can I choose two fruits from five” can
either be 10, 20, 25, or 15 depending on whether order matters and whether repeats are allowed! We can
summarize our findings in the following table:

Number of ways to choose 2 pieces of fruit from 5 fruits

41
 Order does not matter Order matters
Repeats not allowed 10 20
Repeats allowed 15 25

In this chapter, we will study each of the four cases above in depth. The goal is to find a general formulas
for each entry of the table, by computing the number of ways to choose k objects from n objects in each
case. Throughout this chapter, we will assume k and n are natural numbers.

Exercises
1. Find the four entries in the table if another piece of fruit, a strawberry, is added to the refrigerator.

2. Find the four entries in the table if a third piece of fruit will be chosen (for Wednesday in Examples 3.1.2
and 3.1.3).

3. The Student Government of Combinations College consists of a president, vice president, secretary,
and treasurer. Seven people are running for student government. How many possible sets of four
people can make up the student government? (Ignore for now the choice of which of these four people
is president, vice president, etc.)

4. Using the setup of the previous problem, how many different possible choices of president, vice presi-
dent, secretary, and treasurer can there be, out of the seven people running?

5. How many ways can you pick two marshmallows from a large bag of colored marshmallows, where
there are 6 possibilities for the color? (For instance, you could pick two red marshmallow, or one red
and one blue, or any other combination of the 6 available colors. Assume the order in which you pick
them does not matter).

6. There are 6 different colors of marshmallows in a large bag of marshmallows. How many ways can you
stick 4 marshmallows on a toothpick that’s sticking out of the top of a birthday cake?

7. In the previous problem, how many possibilities are there if the four marshmallows all have distinct
colors?

3.2 Sequences

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=-FP3UgU9ZNA

Sequences allow us to order or list the elements in sets.

Definition 3.2.1. A sequence is a list of numbers or other symbols written in order.

A sequence can be either finite or infinite. We write a finite sequence as a list separated by commas and
enclosed by parentheses, as in:
(5, 2, 3, 6)
or
(2, 4, 6, 8, 10, . . .).
The former is a finite sequence, of length 4 since there are 4 numbers in the sequence. The latter is an
infinite sequence. This notation can be generalized using subscripts as follows.

42
Notation 3.2.2. We write
(a1 , a2 , . . . , an )
to denote a finite sequence of length n, whose first entry is a1 , second entry is a2 , etc, and ending with the
nth entry being an . We write
(a1 , a2 , a3 , a4 , . . .)
to denote the infinite sequence whose ith entry is ai for each integer i ≥ 1.
The set of numbers or symbols we use in a sequence is called the alphabet. Commonly used alphabets
are the binary alphabet {0, 1}, the letters in the English alphabet, ASCII, or the natural numbers.
The next remark explains some other variations on sequences.
Remark 3.2.3. 1. It is sometimes convenient to “zero-index”, by starting a sequence with a0 rather than
a1 , see Chapter 6.
2. It is sometimes convenient to write a sequence as a string or word, in which we drop the parentheses
and commas. For instance, the sequence (a, p, p, l, e) can be more easily written as the string of letters
apple, and the string of digits (1, 0, 1, 1, 0) can be written as the binary string 10110.
Warning: We do not want to write the sequence (1, 0, 10, 100, 1000) as the string 10101001000, because
we would lose information about the length of each entry. This topic is important in coding theory
where it is called the theory of uniquely decipherable codes.
3. A sequence of length n can also be thought of as a function that assigns a number (or symbol) to every
positive integer from 1 to n. An infinite sequence can also be thought of as a function that assigns a
number (or symbol) to every positive integer.

3.2.1 Counting where order matters and repeats are allowed

The type of counting seen in Example 3.1.3, where order matters and repeats are allowed, can be
restated in terms of counting sequences. Indeed, if we choose k objects from n objects in order, with repeats
allowed, we can write them as a sequence from an alphabet with n symbols.
We can count such sequences as follows.
Theorem 3.2.4. The number of sequences of length k composed from an alphabet with n symbols is nk .
More generally, there are nk combinations of k objects chosen from n objects, where order matters and repeats
are allowed.
Proof. There are n ways to choose the first symbol in the sequence, then n ways to choose the second, and
so on up to the kth. Therefore there are n · n · n · · · · · n = nk possibilities.
Using Theorem 3.2.4, we can complete the general formula for one entry of the table.
Number of ways to choose k objects from n objects

Order does not matter Order matters

Repeats not allowed

Repeats allowed nk

Let’s look at a few examples; for convenience we write these examples as strings.
Example 3.2.5. The number of binary strings (consisting of 0’s and 1’s) of length k is 2k (here n = 2). We
can verify this for k = 3 by listing all 8 = 23 possible binary strings of length three:
000, 001, 010, 011, 100, 101, 110, 111.

43
Example 3.2.6. The number of 3-digit strings of letters from the English alphabet is (26)3 = 17, 576. Some
examples are pxy or csu.

Example 3.2.7. The number of 3-digit strings of letters or numbers, such as p17, 2xy, or pqm, is (26+10)3 =
363 = 46, 656.

We can use the formula nk for any situation in which order matters and repeats are allowed, even if we
are not explicitly counting sequences or strings.

Example 3.2.8. At a buffet table, there are many samosas, chicken skewers, and carrots. The number of
ways to eat 10 snacks in a row is 310 .

The formula nk was derived using the multiplication principle, by considering how many choices there
are for each element of the sequence. In Theorem 3.2.4, there is one alphabet of size n that is used for all k
positions in the string. We can generalize this by using a different alphabet in each position.

Theorem 3.2.9. There are n1 · n2 · . . . · nk possible sequences of length k made by

• choosing one of n1 symbols as the 1-st entry,

• choosing one of n2 symbols as the 2-nd entry,

..
.

• choosing one of nk symbols as the k-th entry.

Remark 3.2.10. Theorem 3.2.4 is the special case of Theorem 3.2.9 where n1 = n2 = . . . = nn = n.

Example 3.2.11. How many 4-digit numbers are there that have no leading zeros? For example, 2347 is
one such number but 0023 is not.

Answer. There are 9 choices (1-9) for the first digit, and 10 choices (0-9) for the remaining three digits.
Hence the total number is 9 · 10 · 10 · 10 = 9000.

Example 3.2.12. How many 3-digit sequences of letters from the English alphabet in which the first and
last letters are consonants and the middle is a vowel (A, E, I, O, or U)? Since there are 5 vowels and 21
consonants, there are 21 · 5 · 21 = 2205 possibilities.

Exercises
1. (a) Find the number of all 4-digit strings of letters. String aabb is different from abab.
(b) Find the number of all 4-digit strings of letters in which no two consecutive letters are the same.
For example, strings xdwa and xdxd count but strings xdww and xddx do not.
(c) Find the number of all 4-digit strings of letters in which letters which differ by one slot are not
the same, and also letters which differ by two slots are not the same. For example, string xwzz
does not count since there are z’s which differ by one slot, string xzwz does not count since there
are z’s which differ by two slots, but string zxwz is okay.

2. How many ways are there to send one postcard to each of 10 different friends, from a large supply
of each of 4 different kinds of postcards (a ram, buffalo, fox, and goat print postcard). If Student A
receives a ram postcard and Student B receives a buffalo postcard, that is different than if Student A
receives a buffalo postcard and Student B receives a ram postcard.

3. How many 5-digit strings of digits 0-9 are there such that no two consecutive digits are the same? For
example, strings 01323 and 93572 are allowed, but 00232 and 93552 are not.

44
 4. Student d starts a job at a license plate manufacturer. Each license plate is of the form XXX ###,
where each entry marked X can be any capital letter from the English alphabet and each digit marked
# can be any digit 0-9. If Student d can make 33,000 license plates a day, how many days of work will
this give Student d?
5. Your neighbor works for your school’s accounting division, which receives scholarship donations from
all over the world. Each donor is assigned an account fund, labeled by a 5 digit string where the
first digit is ‘C’ (standing for Code), starting at CA000 through CA999, then CB000 through CB999,
etc. The available codes assigned via this system will run out after codes CZ000 through CZ999 are
assigned. About 700 new codes are used every year. At the beginning of the school year, the next code
to be assigned is CW182. Your neighbor plans to retire 5 years from now. Will the available codes run
out before she retires?

3.3 Permutations and other sequences with distinct entries

The seven students in the set P = {a, b, c, d, e, f, g} go out for ice cream. When they arrive at the ice cream
shop, they need to wait in line. Student a argues they should wait in alphabetical order, but student g has a
strong opposition to this ordering. This prompts the question: How many different ways are there for them
to wait in an ordered line?
To solve this kind of problem, we need the following definition.
Definition 3.3.1. A permutation of n objects is an ordering or arrangement of all n of them.
A permutation can be expressed as a sequence or string.
Example 3.3.2. For example, the set {a, b, c} has 6 permutations:

abc acb bac bca cab cba

Example 3.3.3. To write down a permutation of the set {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, we use sequence notation
rather than string notation, since 10 has two digits. One possible permutation is:

(3, 6, 7, 2, 1, 10, 4, 5, 9, 8).

Recall that n! = n · (n − 1) · (n − 2) · . . . · 3 · 2 · 1.
Theorem 3.3.4. The number of permutations of n objects is n!.
Proof. By Theorem 3.2.9, there are n choices for the first object, then n − 1 choices for the second object,
then n − 2 for the third, and so on until there is only one 1 choice remaining for the nth object. So there
are n · (n − 1) · (n − 2) · . . . · 3 · 2 · 1 = n! permutations in total.
Remark 3.3.5. If there are 0 objects, then there is only one way to arrange them (do nothing). This can
be thought of as the “empty sequence” or “empty string”. Therefore, we set 0! = 1.
Example 3.3.6. When seven students go out for ice cream, there are 7! = 5040 ways for them to line up,
since each way corresponds to a permutation of elements from P = {a, b, c, d, e, f, g}.
Example 3.3.7. Suppose there are 5 different fruits in your refrigerator: an apple, a banana, an orange, a
pear, and a mango. How many ways can you bring one fruit on each day of the work week? Let’s call the
fruits A, B, O, P , and M for short. Then a possible assignment of each of the fruits to a day of the work
week is a permutation of {A, B, O, P, M }. Therefore, the number of possibilities is 5! = 5 · 4 · 3 · 2 · 1 = 120.
The problems in this section can be rephrased as counting problems where order matters and repeats
are not allowed, in which we arrange all of the objects. In the next section, we consider the more general
problem of choosing k objects from n objects where order matters but repeats are not allowed. These can
be thought of as ordered subsets.

45
3.3.1 Ordered subsets
The ice cream adventurers P = {a, b, c, d, e, f, g} decide to wait in line for ice cream yet again. However,
to their collective horror, the ice cream shop is closing and can only serve 4 more customers. How many
ordered lines are there that contain 4 people from the set P ? To make such a line, we choose a subset S of
P of size 4 and then order the elements in S.
This problem is similar to the one seen in Example 3.1.2, where order matters and repeats are not
allowed. More generally, consider the following definition.
Definition 3.3.8. Suppose 0 ≤ k ≤ n. An ordered subset of size k from a set A is a string a1 , . . . , ak of
distinct elements of A.
Theorem 3.3.9. Suppose 0 ≤ k ≤ n. The number of ordered subsets of size k from a set with n elements is
n!
n · (n − 1) · (n − 2) · . . . · (n − k + 1) = . (3.1)
(n − k)!
n!
More generally, (n−k)! is the number of ways to choose k objects from n objects, where order matters and
repeats are not allowed.
Proof. An ordered subset of size k is a sequence of length k chosen from an alphabet of size n in which no
two entries are equal. There are n possibilities for the first entry. Once that entry is chosen, there are only
n − 1 remaining possibilities for the second one, then n − 2 for the third, and so on. For the final kth entry,
there are n − k + 1 possibilities. The decreasing number of possibilities for each entry guarantees that there
are no repeated entries. The formula in (3.1) follows from the multiplication principle.
Remark 3.3.10. Since n−k+1 = n−(k−1), there are k entries in the product n·(n−1)·(n−2)·. . .·(n−k+1).
It is helpful to think of these entries as being labeled from 0 to k − 1.
Example 3.3.11. There are 7 · 6 · 5 · 4 = 840 possible ways to form a line of four ice cream adventurers.
Using Theorem 3.3.9, we can complete the general formula for the upper right entry of the table.

Number of ways to choose k objects from n objects

Order doesn’t matter Order matters

n!
Repeats not allowed (n−k)!

Repeats allowed nk

Example 3.3.12. As in Example 3.1.2, the number of ways to choose one out of five fruits to eat on Monday
and then another to eat on Tuesday is
5! 5·4·3·2·1
= = 5 · 4 = 20.
3! 3·2·1
Tip: When dividing one factorial by another, expand both out as products and cancel as many terms as
5!
possible before multiplying. For example, the computation above was easier than computing 3! = 120
6 = 20.

Example 3.3.13. How many different arrangements are there for 47 students sitting in a classroom with
50 seats?
Answer. This is the number of ordered subsets of size 47 from a set of size 50. Hence the answer is
50 · 49 · . . . · 4 = 50!
3! .

Remark 3.3.14. In Section 1.2.3, we answered this question by adding 3 “ghosts” to the class. These
answers are really the same!

46
Exercises
1. How many different ways can you permute the 26 letters of the English alphabet?
2. How many ways are there to scramble the letters in ABSEINO, in the following situations?
(a) If each letter is used exactly once in a seven letter word.
(b) If each letter is used at most once in a five letter word.
(c) How many actual five letter English words, such as NOISE, can you make from these letters?
3. In how many ways can you visit 5 of the 50 state capitals? The trip
Sacramento → Dover → Baton Rouge → Bismarck → Denver
is different than the trip
Dover → Baton Rouge → Denver → Bismarck → Sacramento.

4. In how many ways can you rank your favorite 10 restaurants (from #1 to #10) out of a collection of
50 restaurants?
5. There are 128 NCAA DI football teams. How many possible top-25 ranked lists are there? One such
list might be:
1. CSU
2. Oklahoma
3. Washington
..
.
25. Alabama

3.4 Sets: When order doesn’t matter

After receiving numerous complaints about the length of its lines, the ice cream shop decided to reward their
customers with an Ice Cream for a Year contest. This contest allows 4 people to win free ice cream for a
year! The seven students in the set P = {a, b, c, d, e, f, g} eagerly enter, along with 100 other people. Out of
the 107 entrants, how many ways are there to choose 4 winners? All winners receive the same prize.
This problem is similar to the one seen in Example 3.1.1, where order does not matter and repeats are
not allowed. We can solve problems of this type with the binomial coefficient formula from Section 1.2.3.
Theorem 3.4.1. Suppose 0 ≤ k ≤ n. The number of subsets of size k in a set of size n is
 
n n!
= .
k k!(n − k)!


More generally, nk is the number of ways to choose k objects from n objects, where order does not matter
and repeats are not allowed.
Proof. We already proved this in Lemma 1.2.6. Here we repeat the proof from Method 1, using the vocabulary
n!
of sets. By Theorem 3.3.9, the number of ordered subsets of size k is (n−k)! . However, that counts each
unordered subset k! times, since k! is the number of permutations of k objects (Theorem 3.3.4). Hence, by
the division principle, the number of unordered subsets of size k is
n!  
(n−k)! n! n
= = .
k! k!(n − k)! k

47
 Using Theorem 3.4.1, we can complete the upper-left entry of the table.

Number of ways to choose k objects from n objects

Order does not matter Order matters

n

n!
Repeats not allowed k (n−k)!

Repeats allowed nk

Example 3.4.2. The number of ways to choose 2 fruits from 5 to bring to work on the same day is
 
5 5! 5·4·3·2·1
= = = 5 · 2 = 10.
2 2! · 3! 2·1·3·2·1

Example 3.4.3. How many ways are there to select the top 25 NCAA DI football teams (of which there
are 128) without ranking them?

This is the number of (unordered) subsets of size 25, giving 128 128! 128!
25 = 25!(128−25)! = 25!103!

Example 3.4.4.
The number of ways to pick 4 winners out of 107 entrants for the Ice Cream for a Year
contest is 107
4
107!
= 4!·103! = 5, 160, 610.

Choosing the four winners in Example 3.4.4 is the same as eliminating the 103 people who did not win.
So the number of ways to choose 103 people out of 107 gives the same answer. This leads us to the following
corollary, which is often used to simplify problems.



Corollary 3.4.5. These two binomial coefficients are equal: nk = n−k n
.
n


Surprisingly, the formula k can also be useful in counting certain sequences or strings.

Example 3.4.6. How many binary strings (from the alphabet {0, 1}) have exactly k zeroes and n − k ones?
To form such a binary string, we simply have to choose k positions for the 0’s, out of n possible positions.
This is the same as choosing an unordered
subset of k of the positions in the string (even though the string
itself is ordered). Therefore, there are nk such strings.

Exercises
1. You have 7 different rolls of wrapping paper and 4 identical teddy bears. How many different ways can
you wrap the teddy bears such that no two teddy bears are wrapped in the same wrapping paper?

2. Student a has six wristbands of different colors and student b has seven necklaces of different colors.
In how many ways can a trade two wristbands for three of b’s necklaces?

3. Suppose student d has 9 pencils (all different) and student e has 6 erasers (all different). In how many
different ways could student d trade 4 of their pencils for 2 of student e’s erasers?



4. Give an algebraic proof that n2 + n+1 2 = n2 . (Do not use induction, even if you know what that is.)

5. How many ways are there to send postcards to 10 different friends, from a large supply of each of 4
different kinds of postcards (a ram, buffalo, fox, and goat print postcard), in the following scenario?
You want to send either 1, 2, 3, or 4 postcards, such that no person gets two postcards of the same
animal? For example, student a might receive the ram, buffalo, and fox postcards, while student b
might receive only the ram postcard, and students a and b can receive the same set of postcards.

48
3.5 Multisets: sets with repeats allowed

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=8OSEZyL O7E

Student c was one of four lucky winners of the ice cream contest! To celebrate, the seven students decide
to get a large sundae in a bowl, filled with many different scoops of ice cream. There are 10 flavors of ice
cream and a large sundae has 15 scoops. Since it is a hot day in Colorado, all of the scoops will melt together,
so the order of the scoops does not matter. Each flavor can be used for as many of the 15 scoops as they
want. How many different sundaes can student c order?
This problem is similar to the one seen in Example 3.1.4. To solve problems of this type, we define a
multiset; intuitively, this is a set whose elements do not need to be different.

3.5.1 Definition of multiset

Example 3.5.1. The multiset {apple, apple, orange, pear, pear} denotes a collection of two identical apples,
one orange, and two identical pears.
Definition 3.5.2. A multiset is a set, together with a positive integer multiplicity assigned to each element.
Equivalently, it is a set S, together with a function f : S → N − {0}.
A multiset can be written down in several ways:
• by listing an element of the set m times if it has multiplicity m;
• by writing the multiplicity as a superscript on the element; (This method can sometimes lead to
confusion.)
• by writing down the function f .
Example 3.5.3. Here are several ways to write the multi-set with three 4’s, one 5, and two 6’s:
• {4, 4, 4, 5, 6, 6} or {4, 5, 4, 6, 4, 6} (the order does not matter).
• {4x3 , 5x1 , 6x2 } Warning! The 4x3 written here does not mean 4 raised to the power x3.
• the function f : {4, 5, 6} → N with f (4) = 3, f (5) = 1, and f (6) = 2.

3.5.2 Example of sticks and stones method

To count the number of multi-sets of a certain kind, we use what we call the “sticks and stones” method. (In
other sources, this is sometimes called the “stars and bars” method.) We illustrate this method first with
an example.
Example 3.5.4. Suppose we have an unlimited supply of 5 kinds of fruit: apples, bananas, mangoes,
oranges, and pears. Let’s write them as A, B, M, O, and P for short. How many ways can we make a bag
of 6 fruits?
Such a bag can be described as a multiset, for instance, {A, B, B, M, M, P }. An alternative way to write
this multiset is with a string of sticks (written | ) and stones (written •). The stones represent fruit and the
sticks are dividers between different types of fruit (say in alphabetical order A, B, M, O, P). So the multiset
{A, B, B, M, M, P } is written
• | • • | • • | | •.
The first stone indicates that there is one apple, then the next two stones indicates that there are two
bananas, and so on. There are no stones between the last two sticks because there are no oranges in the bag.

49
 Each bag can be written using six stones and four sticks. There are 4 sticks because there are 5 types of
fruit to separate. On the other hand, every string of six stones and four sticks describes a bag of fruit. For
example, the string
||•|• • ••|•
describes the bag {M, O, O, O, O, P }.
In this way, there is a 1-to-1 matching between the bags and the strings. So to count the number of bags
of six fruits (with five kinds of fruit), we can count the number of strings of 6 stones and 4 sticks This is
equal to the number of ways to choose 6 positions out of 10 possible positions for

the location of the stones.
The sticks are then placed in the remaining 4 positions. Thus the answer is 10 6 = 210.

Example 3.5.4 illustrates several facts. First, the number of multi-sets of size k whose elements come
from a fixed set of size n equals the number of combinations of k elements from a set of n elements where
order does not matter but repeats are allowed. Second, this number can be re-expressed as the number of
ways to write a string of k stones and n − 1 sticks. Third, this number equals the number of ways to choose
k locations for stones out of k + (n − 1) locations for sticks and stones. We state this more formally in the
next theorem.

3.5.3 Main result on multisets and examples

Theorem 3.5.5.
The number of multisets of size k in which every element is chosen from an alphabet of
size n is n+k−1 .
k

More generally, n+k−1
k is the number of ways to choose k objects from n objects, where order does not
matter and repeats are allowed.

Proof. We can write the multiset as a string of k stones and n − 1 sticks, where the n − 1 sticks separate
the n types of possible elements of the alphabet in some chosen “alphabetical order”. Then there are
n − 1 + k = n + k − 1 total sticks and stones in the string. We count the number of ways to choose

which k of the positions in the string contain a stone (as opposed to a stick). There are therefore n+k−1
k
possibilities.



Remark 3.5.6. By the symmetry of binominal coefficients, n+k−1 k = n+k−1
n−1 ; this corresponds to choosing
the location of the sticks instead.

Using Theorem 3.5.5, we can include the formula in the lower left portion of the table and thus complete
the table for counting combinations.

Number of ways to choose k objects from an n objects

Order doesn’t matter Order matters

n

n!
Repeats not allowed k (n−k)!

n+k−1


Repeats allowed k nk

Example 3.5.7. A store sells 6 colors of balloons. How many different ways can you buy 10 balloons?

Answer. We want to choose a collection

of k
= 10
balloons, out of an unlimited supply of n = 6 colors of
balloons, so the answer is n+k−1
n−1 = 6+10−1
6−1 = 15
5 .

Example 3.5.8. When Student C orders
the
large sundae, there are k = 15 scoops and n = 10 flavors, so
the number of possibilities is 10+15−1
10−1 = 24
9 = 1, 307, 504 different sundaes they can order — they will
have to order about 3,593 sundaes a day to make sure they try every one in a year!

50
3.5.4 A variation on multiset problems
Example 3.5.9. In Example 3.5.8, that is far too many sundaes for the students to eat every day! What
if instead they wanted to order their sundae so that there was at least one scoop of each flavor?
The first 10 scoops are determined (one of each flavor). So student
c now needs to choose k 0 = 5 more
scoops from among the n0 = 10 flavors. So the number of choices is 14 5 = 2, 002 sundaes to order - about 6
sundaes a day!

Corollary 3.5.10. Let m ≥ n. The number of ways to choose m elements from an n-element set where

order does not matter, repeats are allowed, and at least one of each of the n elements is chosen is m−1
n−1 .

Proof. We start by picking one of each element. Then we must pick m − n additional elements from the n
possibilities, where repeats are allowed and order does not matter. By Theorem 3.5.5, the number of ways
to do this is    
n + (m − n) − 1 m−1
= .
m−n m−n
m−1

m−1



By the symmetry of binomial coefficients, m−n = m−1−(m−n) = m−1
n−1 .

Example 3.5.11. A store sells 6 colors of balloons. How many different different bouquets of 10 balloons
are there that contain


at least one balloon of each color? In this case, m = 10 and n = 6. Hence the number
of ways is 10−1
6−1 = 5 .
9

3.5.5 Using multisets to solve distribution problems

n+k−1


Corollary 3.5.12. The number of ways to distribute k identical pennies to n people is k .

Proof. Label the people by the letters A1 , . . . , An . Suppose S is a multiset of size k, whose elements are the
letters A1 , . . . , An . Then we can distribute the pennies in this way: for 1 ≤ i ≤ n, the number of pennies
that person Ai receives equals the number of times that Ai is included in the multiset. Conversely, given a
way of distributing k pennies to the n people, we can create a multiset S of size k from {A1 , . . . , An } so that
the number of times Ai is included in S equals the number of pennies Ai received. So the number of ways
to distribute

k pennies to n people equals the number of multisets of size k from a set of size n; this equals
n+k−1
k by Theorem 3.5.5.

Example 3.5.13. When k = 9 and n = 4, for the multiset {A1 , A1 , A3 , A4 , A4 }, person A1 receives 2
pennies, person A2 receives 0 pennies, person A3 receives 1 penny, and person A4 receives 2 pennies.

Example 3.5.14. When k = 9 and n = 4, if person A2 receives all the pennies, then the multiset is
{A2 , A2 , A2 , A2 , A2 }.



Example 3.5.15. The number of ways to distribute k = 9 pennies to n = 4 people is n+k−1 = 4+9−1 =
12

k 9
9 = 220 ways.

Example 3.5.16. How many different ordered lists of 4 nonnegative integers have sum equal to 9? Examples
include (4, 0, 2, 3), (0, 4, 2, 3), and (0, 0, 0, 9).

Answer. The answer is the same as for Example 3.5.15, namely 220, because the number of these lists is the
same as the number of ways to distribute n = 9 identical pennies to k = 4 people:
1st number on list = number of pennies to person 1,
2nd number on list = number of pennies to person 2,
3rd number on list = number of pennies to person 3,
4th number on list = number of pennies to person 4.

Corollary 3.5.17. If m ≥ n, the number of ways
to distribute m identical pennies to n people, so that each
person must receive at least one penny, is m−1
n−1 .

51
Example 3.5.18. The number
of ways
to distribute m = 9 identical pennies to n = 4 people so that each
person gets at least one is 9−1
4−1 = 8
3 = 56.

Example 3.5.19. How many different ordered lists of 4 positive integers have sum equal to 9? Examples
include (3, 2, 1, 3), (1, 2, 3, 3), (2, 2, 2, 3), but not (4, 0, 2, 3) as 0 is not positive.

Answer. The answer is the same as for Example 3.5.19, namely 56. The reason is that the number of these
ordered lists is the same as the number of ways to distribute 9 identical pennies to 4 people where each
person gets at least one.

Exercises
1. How many ways can you spend $3 on sushi?

2. How many ways can you spend $10 on sushi?

3. How many ways can you spend $10 on sushi if you want at least one piece of each type?

4. How many ways can you distribute 14 pennies to 6 kids?

5. How many ways can you distribute 14 pennies to 6 kids if each kid gets at least 2?

6. You want to buy a bag of 15 lollipops and there are 5 different flavors to choose from. How many
different bags can you make? How many different bags can you make if you must have at least one of
each flavor?

7. You want to order a tray of 20 tacos and there are 4 different types to choose from. How many different
trays can you order? How many if the tray must contain at least one of each type of taco?

8. You want to make a bouquet of 50 balloons, and there are 10 different colors to choose from. All
balloons of the same color are the same, and there are an unlimited number of balloons of each color.
How many different bouquets can you make? How many if the bouquet must include at least one of
each color?

9. A store sells bouquets of 5 kinds of flowers. How many ways are there to make a bouquet of 12 flowers?
How many ways if the bouquet must include at least one of each kind of flower?

10. How many different ways can we place 10 identical balls into 4 labeled bins, if each bin must contain
at least one ball?

11. How many ways are there to distribute 10 identical pieces of paper to students a, b, and c? The only
requirement is that all 10 pieces of paper need to be handed out.

12. How many ways are there to paint 6 identical picture frames with 3 colors if each color must be used
at least once. Let G be green, Y be yellow and B be blue; then YYGGBB is the same as YGYGBB
since the frames are identical.

13. Explain why Corollary 3.5.17 is true.

3.6 Summary
The major results in this chapter are the counting formulas in this table.

52
 Number of ways to choose k objects from n objects

Order does not matter Order matters

n

n!
Repeats not allowed k (n−k)!

n+k−1


Repeats allowed k nk

3.7 Additional problems for Chapter 3

1. How many different strings are permutations of the letters in PILLOW? For example, LOPILW is one
such permutation.
2. Baseball team A has 7 pitchers, and baseball team B has 5 catchers. The two teams have agreed to
make a trade of the following form: 3 of team A’s pitchers will be traded for 2 of team B’s catchers.
The local newspaper decides to make a ranked list of their top 4 favorite possible trades of this kind.
How many different ranked lists can the newspaper make?
3. How many ordered strings with 10 symbols are there if exactly 7 symbols are numbers (0–9) and exactly
3 symbols are letters (a–z)? The string abc1222233 is different from 12c2a3b322.

4. Find the number of ways to distribute 20 candies to 5 students a, b, c, d, e if:

(a) the candies are identical.
(b) the candies are identical and every person receives at least one.
5. Find the number of ways to give 20 pillows to 12 students a through l if:

(a) the pillows are identical.

(b) the pillows are identical and every person receives at least one.
(c) the pillows are all different and there are no restrictions. For example, one student can receive all
20 pillows.
(d) How many different DNA sequences (whose alphabet consists of the letters A,C,G, and T) of
length 5 are there?
(e) How many different strings of four distinct letters from the English alphabet (which has 26 letters)
are there?
(f) How many binary strings of length 8 have the same number of 0’s as 1’s?
(g) How many strings of length 7 from the alphabet 0,1,2 have exactly three 0’s?
(h) How many strings of length 7 from the alphabet 0,1,2 have less than three 0’s?
6. How many ways can you roll a sum of 13 with three six-sided dice? A die is a cube with 6 faces where
each face contains one of the numbers 1 through 6.

53
3.8 Investigation: Counting problems in the game of poker
One day after Math 301, several students decide to play a friendly game of poker. During the game, one of
them asks “How many ways are there to get a full house?” Stumped by this question, the students turn to
their detailed notes from class to solve this problem.
Before we solve this problem, we first will introduce some vocabulary about playing cards. If you already
are familiar with the game of poker, you can skip the next subsection.

Facts about cards and poker

Poker is a card game played with a standard deck of 52 cards. The cards can be sorted into four suits,
labeled by the symbols ♥, ♠, ♦, and ♣, called hearts, spades, diamonds, and clubs, respectively. There are
13 cards in each suit (quick check: 13 · 4 = 52). The cards in each suit are labeled with values, from least
to greatest, 2,3,4,5,6,7,8,9,10, J=Jack, Q=Queen, K=King, A=Ace=1. The set of cards held by a player is
called a hand. Typically a hand contains five cards.

Example 3.8.1.
This hand contains the Ace of spades, the 10 of hearts, the 7 of diamonds, the 7 of clubs, and the Queen
of diamonds.

Another term used in many card games is called a run. A run is a set of three or more cards with
consecutive values where the suit does not matter: for example, the collection 3♥, 4♠, 5♠ is a run of three
cards, and 9♥, 10♠, J♣, Q♣, K♦ is a run of five cards. An important note is that an Ace can be either the
lowest or highest card in a run (but can not be in the middle of a run). An example where the Ace is “high”
is (Q♣, K♦, A♦) and an example where the Ace is “low” is (A♣, 2♥, 3♥). This set of three cards is not a
run: K♥, A♦, 2♣.

In this class, you do not need to know how to play poker. We just want to count the number of ways
to make different hands of five cards. In the following manual, we will describe each hand of poker, give an
example, and explain the method of how to count that hand.

Poker Hands
Royal flush
Definition of hand: A run consisting of an Ace, King, Queen, Jack, and 10, all of the same suit.
Example hand:

 
4
Number of hands: = 4.
1
Combinatorial proof: The values on the cards are determined. Therefore, we just need to choose one of four
suits.

54
Straight flush
Definition of hand: A straight is a run of five cards. A flush means all of the cards are of the same suit.
Example hand:

   
10 4
Number of hands: · − 4 = 36.
1 1
Combinatorial proof: We need to choose the value of the starting card and the suit of the run (the other four
values and suits are then determined). There are 10 ways to begin the run (Ace through 10) and four suits
to choose. However, if we start the run with a 10, we would have a royal flush, so we subtract the number of
royal flushes to get the answer. Note that if we tried to start a run with a Jack, we would not have enough
cards, since Aces end runs.

Four of a kind
Definition of hand: A four of a kind consists of four same-valued cards and any other card.
Example hand:

  
13 48
Number of hands: = 624.
1 1
Combinatorial proof: We first pick a value for the four cards. Then, we choose the other card in the hand;
since we chose 4 cards already, there are 52 − 4 = 48 cards left in the deck.

Full House
Definition of hand: This hand contains three cards of one value and two cards of a second value.
Example hand:

    
13 4 12 4
Number of hands: = 3, 744.
1 3 1 2
Combinatorial proof: First, choose the value of the card in the triple and choose 3 suits for those 3 cards.
Then, from the remaining 12 values left, choose the value for the pair, and choose 2 of the four suits for
those two cards.

55
Flush
Definition of hand: A hand is a flush if all five cards have the same suit. The only restriction is that the
hand cannot be a run of five cards (or else this would be a straight or royal flush).
Example hand:

  
4 13
Number of hands: − 40 = 5, 108.
1 5
Combinatorial proof: We first choose a suit for the hand. Then, we choose 5 values from the 13 possible
values. However, some of these hand combinations could be straight or royal flushes, so we must subtract
the total number of these two hands to get the answer.

Straight
Definition of hand: A straight is a run of five cards consisting of at least two suits. If they were all the same
suit this would be a straight or royal flush.
Example hand:

  5
10 4
Number of hands: − 40 = 10, 200.
1 1
Combinatorial proof: We need to pick the value of the starting card and a suit for each card. There are 10
ways to begin the run and four suits to choose from for each card. However, some of these hands are straight
and royal flushes, so we subtract off the total number of those.

Three of a kind
Definition of hand: A three of a kind consists of exactly three same-valued cards and two other cards with
distinct values. If the two other cards were the same value, this would be a full house.
Example hand:

   48
44

13 4 1 1
Number of hands: = 54, 912.
1 3 2!
Combinatorial proof: First, we choose the value for three of the cards and their three suits. We then choose
the remaining two cards from the deck: the first is chosen from the remaining 48 cards in the deck (since
we need to remove the possibility of a four of a kind); the second is chosen from the remaining 44 cards in

56
the deck (since we need to remove the possibility of a full house). However, notice that the order in which
we choose the last two cards does not affect the overall hand (for example, choosing 4♥ then 9♥ gives the
same hand as 9♥ then 4♥). So, we divide by 2!.

Two pairs
Definition of hand: This hand consists of three values: two pairs and a third card of a distinct value.
Example hand:

  2   
13 4 11 4
Number of hands: = 123, 552.
2 2 1 1
Combinatorial proof: First choose two values for the numbers that will be in pairs. Once these two values
are chosen (for example king and 10), one of them is larger. Choose two suits for the pair with the larger
number. Choose two suits for the pair with the smaller number. In order to avoid having three of a kind,
we choose the number for the last card from the remaining 11 values. Finally, we choose the suit of the last
card.

Pair
Definition of hand: Exactly two values are identical (different suits), all other cards are different values. For
example, consider having 4♥, 4♠, 6♦, K♣, J♦. If the other three cards were the same, this would be a full
house. If they contained an additional pair, it would be a two pair. If they contained the same value(s) as
the first pair, it would be a three or four of a kind.
Example hand:

   48
44
40

13 4 1 1 1
Number of hands: = 1, 098, 240
1 2 3!
Combinatorial proof: First, we choose the value and suits in our pair. Then, we choose the remaining 3 cards
— we start with 48 remaining cards in our deck (or else we would risk having three or four of a kind), then
44 cards (or else we would risk having two pairs), then 40 (again we would risk having two pairs). However,
notice the order in which we draw the cards does not matter, so we divide by 3! to get the final answer.

High Card
Definition of hand: This hand is none of the above hands.
Example hand:

57
Number of hands: 1, 302, 540
Combinatorial proof: Left as an exercise.

Exercises
1. (a) In your own words, explain how to count how many 4-of-a-kind poker hands there are. (Recall
that we consider the hands K♥, K♦, K♠, K♣, 2♦ and 2♦, K♠, K♣, K♦, K♥ to be the same.)
(b) What fraction of all possible poker hands are 4-of-a-kinds?

2. (a) In your own words, explain how to count how many different pair poker hands there are.
(b) What fraction of all possible poker hands are pair poker hands?

3. (a) In your own words, explain how to count how many different two pair poker hands there are. A
two pair consists of a pair of cards of one value, another pair of cards of a different value, and
then a fifth card of a distinct value.
(b) What fraction of all possible poker hands are two pair poker hands?

4. (a) In your own words, explain how to count how many different full house poker hands there are. A
full house consists of triple of cards of the same value, plus a pair of cards of the same value.
(b) What fraction of all possible poker hands are full houses?

5. Find another way to count the number of ways a player could obtain the following hands:

(a) Three of a kind

(b) Two pairs
(c) Pair

Check your answers numerically.

6. Use the Inclusion-Exclusion Principle to compute the number of ways to get a High Card, as follows.
Define the set A to be the set of all hands having at least two cards of the same number. Let B be the
set of all hands for which all cards are the same suit. Let C be the set of all straights. Then the set
of high cards is the complement of A ∪ B ∪ C.
Calculate the number of high card hands and check your answer numerically.

7. The students decide to play Poker with some revised rules. They now play with hands containing
4 cards instead of 5. Compute the number of ways a player can now have the following hands and
provide a combinatorial proof for each one:

(a) Four of a kind

(b) Three of a kind
(c) Two pairs
(d) Pair

58
8. For the following questions, compute the number of ways a player can now have (i) a straight flush,
(ii) a flush, and (iii) a straight, and provide a combinatorial proof for each one when:
(a) Aces only count as a high card (this eliminates A♥, 2♣, 3♦, 4♣, 5♠ from counting as a run).
(b) Aces only count as a low card (this eliminates 10♣, J♦, Q♣, K♠, A♦ from counting as a run).
(c) Aces can now be used in the middle of runs (this allows Q♣, K♠, A♦, 2♣, 3♦ to count as a run).
9. For the following questions, we will use a modified deck.
(a) Suppose all of the cards with value 8 have been removed. How many ways can you obtain a
straight? Keep in mind 5♣, 6♥, 7♦, 9♥, 10♣ is not a valid straight.
(b) Suppose one of the Joker cards has been added to the deck as a free card (a card that can take
the place of any card). Compute the number of the following hands and provide a combinatorial
proof:
i. Four of a kind
ii. Flush
iii. Full House

59
60
Chapter 4

Pascal’s triangle and the Binomial

Theorem


The binomial coefficient nk counts the number of ways to choose k objects from a set of n objects (where
order does not matter and repeats are not allowed). In Theorem 3.4.1 (and Lemma 1.2.6), we proved that
 
n n!
= .
k k!(n − k)!

Binomial coefficients are extremely useful in combinatorics and arise in many different areas of mathematics
and applications. In Section 4.1, we study Pascal’s triangle and Pascal’s recurrence for binomial coefficients,
and describe applications to counting lattice paths. In Section 4.2, we describe their role as coefficients in
the Binomial Theorem. Sections 4.3 and 4.4 contain identities involving binomial coefficients. In Sections 4.5
and 4.6, we generalize to study multinomial coefficients and their applications to anagrams.

4.1 Pascal’s triangle

Pascal’s triangle, named after the French mathematician/philosopher Blaise Pascal (1623–1662), provides a
way of quickly computing binomial coefficients for small values of n and k. The method is to
construct an
infinite triangular chart in which the entry in the n-th row and k-th (diagonal) column is nk , as follows.

k=0
 
0 k=1
n=0
  0  
1 1 k=2
n=1
  0   1  
2 2 2 k=3
n=2
  0   1   2  
3 3 3 3 k=4
n=3
  0   1   2   3  
4 4 4 4 4
n=4
0 1 2 3 4
. .. ..
.. . .

61
 The values in this triangle can be computed quickly using an identity known as Pascal’s Recurrence.
Theorem 4.1.1 (Pascal’s Recurrence). Suppose n is a natural number.
(a) Then    
n n
= 1 and = 1.
0 n
(b) If k is a natural number such that 0 ≤ k < n, then
     
n n n+1
+ = .
k k+1 k+1

Before proving Theorem 4.1.1, we provide some more explanation about it. Part (a) of Pascal’s Recurrence
says that the numbers on both the left and right sides of the triangle are all 1, as seen below:
n = 0: 1
n = 1: 1 1
n = 2: 1 1
n = 3: 1 1
n = 4: 1 1
n = 5: 1 1
n = 6: 1 1

Part (b) of Pascal’s Recurrence says that the other numbers in the triangle are the sum of the two
numbers just above them (to the left and right). For instance, to find the middle entry in the n = 2 row, we
compute 2 = 1 + 1.
n = 0: 1
n = 1: 1 1
n = 2: 1 2 1
n = 3: 1 1
n = 4: 1 1
n = 5: 1 1
n = 6: 1 1

We can then fill in each entry, for each row in succession, by adding the pair of numbers above it. For
example, the 3rd entry (k = 2) in the n = 6 row is 15 = 5 + 10. Here are the first six rows of Pascal’s
triangle.
n = 0: 1
n = 1: 1 1
n = 2: 1 2 1
n = 3: 1 3 3 1
n = 4: 1 4 6 4 1
n = 5: 1 5 10 10 5 1
n = 6: 1 6 15 20 15 6 1

62
 It is fast to compute the
first rows of entries in Pascal’s triangle recursively, which provides a useful
alternative to the formula nk = k!(n−k)!
n!
.
In the next lemma, we give another way to interpret Pascal’s recurrence. Imagine placing a ball at the
top of this grid. It can fall along the diagonal lines, moving down either to the left or to the right. It can
not fall up.


Lemma 4.1.2. The entry nk counts the number of possible ways the ball could have traveled to that location!

Proof. To reach the n-th

row, the ball needs to fall n steps down, either to the left or right. To reach the
spot with the entry nk , the ball needs to fall to the right k times. The route is given by a sequence of length
n, containing the letter
R (for right) k times and containing the letter L (for left) n − k times. The number
of such sequences is nk , since we need to choose which k entries of the sequence are the letter R.

Lemma 4.1.2 gives another way to interpret Pascal’s recurrence. Part (a) says that there is only one way
that the ball can fall to reach a spot along the outer diagonals. To reach a spot that is not on the outer
diagonals, the ball fell from the row above, coming either from the left or from the right. Part (b) says that
the number of ways the ball could have traveled to this spot is the sum of the ways it could have fallen to
those two spots in the row above.


Example 4.1.3. There is only one way for the ball to reach the entry 40 ; it is given by the route LLLL.


There are four ways for the ball to reach the entry 41 ; they are given by the routes LLLR, LLRL,
LRLL, and RLLL.

Example 4.1.4. The blocks north of campus are on a grid system. Starting at the corner of Howes and
Laurel, the students need to walk two blocks north and three blocks west to reach the ice cream store. How
many ways are there to do this (without traveling south or east)? After rotating Pascal’s triangle, we see
the answer is 10. We can write out these paths as

N N W W W, N W N W W, N W W N W, N W W W N, W N N W W,
W N W N W, W N W W N, W W N N W, W W N W N, W W W N N.

Finally, we will prove Pascal’s Recurrence algebraically. For a more elegant combinatorial proof, see
Example 5.3.3.


Proof of Theorem 4.1.1. To show (a), note that n0 = 0!·n! n! n!
= 1·n! = 1 since 0! = 1 by Remark 3.3.5.
n

n!
Similarly, n = n!·0! = 1.
For (b), we compute:
   
n n n! n!
+ = +
k k+1 k!(n − k)! (k + 1)!(n − k − 1)!
n! · (k + 1) n! · (n − k)
= +
(k + 1)!(n − k)! (k + 1)!(n − k)!
n!(k + 1 + n − k)
=
(k + 1)!(n − k)!
n!(n + 1)
=
(k + 1)!(n + 1 − (k + 1))!
(n + 1)!
=
(k + 1)!(n − 1 − (k + 1))!
 
n+1
= .
k+1

63
Exercises
1. Compute the rows n = 7 and n = 8 in Pascal’s triangle.

2. Which entries in Pascal’s triangle are odd?

n




3. Suppose k−1 = 120 and n+1 k = 330. What is nk ?
Hint: The best way to solve this problem does not require you to find n or k.


4. Find all the routes the ball could fall to reach the entry 42 in Pascal’s triangle.

5. Starting at the corner of Howes and Laurel, how many ways are there to walk three blocks north and
four blocks west (without traveling south or east)?

6. How many paths of length m + n are there from the lower left corner to the upper right corner in an
m × n grid? (Paths can only trace along edges of the squares of the grid.)


7. If p is a prime number and 1 ≤ k ≤ p − 1, explain why kp ≡ 0 mod p.

8. Starting at the corner of Howes and Laurel, how many ways are there to walk four blocks north and
two blocks west (without traveling south or east)?

9. You have 7 different hats and you want to loan some of these to a friend, who needs at least 2 hats.
How many ways are there to do this?

10. Remember how a ball can fall through the grid system of Pascal’s triangle. We suppose that the ball
has an equal probability of falling down to the left or to the right. Compute the probability that the
ball lands at the 3rd spot in the 5th row (with the outer diagonal being the case k = 0). Write your
answer as a decimal number between 0 and 1.

11. A ball falls through the grid system of Pascal’s triangle and lands on the n=10th row. Write a 2-3
sentence explanation of why the ball has a 50% probability of landing in a spot where k is even.

4.2 The Binomial Theorem

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=wFk-Z25-s7A

One of the most important applications of Pascal’s triangle and binomial coefficients is the expansion of
a power of a binomial. Starting with a binomial x + y, let us consider its first few powers:

(x + y)0 = 1,
(x + y)1 = x + y,
(x + y)2 = x2 + 2xy + y 2 ,
(x + y)3 = x3 + 3x2 y + 3xy 2 + y 3 ,
(x + y)4 = x4 + 4x3 y + 6x2 y 2 + 4xy 3 + y 4 .

The coefficients of these expansions are the numbers in Pascal’s triangle! The pattern becomes even more
clear if we align them this way:

64
 (x + y)0 : 1
1
(x + y) : x + y
(x + y)2 : x2 + 2xy + y2
(x + y)3 : x3 + 3x2 y + 3xy 2 + y3
(x + y)4 : x4 + 4x3 y + 6x2 y 2 + 4xy 3 + y4

Indeed, if we ignore the x’s and y’s in
the table above, the coefficients are precisely the numbers in Pascal’s
triangle. This is why the numbers nk are called binomial coefficients.

Remark 4.2.1. Some students are familiar with the ‘foil’ (first, outer, inner, last) method to compute the
case n = 2:
(x + y)2 = x · x + y · x + x · y + y · y.

Later in this section, we will generalize this, in a way that could be called a ‘giant foil method’.

We now can state the pattern of the coefficients as the Binomial Theorem, one of the oldest named
theorems in mathematics, first discovered by the Persian mathematician and engineer Al-Karaji ∼953–1029.

Theorem 4.2.2. (Binomial Theorem) Let x and y be variables. Then the expansion of (x + y)n is
         
n n n n n−1 1 n n−k k n 1 n−1 n n
(x + y) = x + x y + ... + x y + ... + x y + y .
0 1 k n−1 n

Proof. Note
(x + y)n = (x + y) · (x + y) · (x + y) · . . . · (x + y) · (x + y) .
| {z }
n times

In expanding this product, we need to take one entry from each term, multiply these entries together, and
then add all those products (‘the giant foil method’). For each of the n terms (x + y), we have the option to
choose either x or y. Since there are n terms, the product of these will be a monomial of the form xn−k y k
for some number k. To produce the monomial xn−k y k in the product, we must choose the
variable y for a
total of k times and then choose the variable x for the remaining

n − k times. There are nk ways to choose
n
k terms out of the n terms. So the coefficient of xn−k y k is k .

10 8 2
10

For example, if you expand (x + y) and combine like terms, then the coefficient in front of x y will be
2 = 45. This comes from the fact that out of the 10 terms x + y being multiplied together, we need to
choose the variable y from 2 of them (and the variable x from the remaining 8 of them) in order to produce
the monomial x8 y 2 .

Remark 4.2.3. By combining the Binomial Theorem with Pascal’s triangle, we can quickly expand powers
of the sum of two things! For instance, by reading row n = 6 from Pascal’s triangle, we can immediately
deduce that
(x + y)6 = x6 + 6x5 y + 15x4 y 2 + 20x3 y 3 + 15x2 y 4 + 6xy 5 + y 6 .

By reading row n = 3 from Pascal’s triangle, we can deduce that

(a2 + 5b)3 = (a2 )3 + 3(a2 )2 (5b) + 3(a2 )(5b)2 + (5b)3

= a6 + 15a4 + 75a2 b2 + 125b3 .

65
Exercises
1. Given a power of x + y and a monomial in its expansion, find the coefficient on that term using the
Binomial Theorem:

(a) (x + y)3 and x2 y

(b) (x + y)6 and x4 y 2
(c) (x + y)8 and x3 y 5

2. What is the coefficient of x5 y 3 in (x + y)8 ?

3. What is the coefficient of x2 in (x − 1)5 ?

4. What is the coefficient of x3 in (x − 1)(x − 2)(x − 3)(x − 4)?

5. What is the constant coefficient in (x − 1)(x − 2)(x − 3)(x − 4)?

6. Expand (x + 2)3 by substituting y = 2 into the binomial theorem.

7. Expand (x − 1)4 by substituting y = −1 into the binomial theorem.

8. Expand (a2 + 3b)4 using the binomial theorem.

9. What is the coefficient of a4 b2 in (a2 + 3b)4 ?

10. For a real number r, find the coefficients of (x − r)n , the polynomial whose only root is r.

11. The formulas of Vieta (1540-1603):

(a) Find formulas for S1 and S2 in terms of r1 and r2 if

(x − r1 ) · (x − r2 ) = x2 − S1 x + S2 .

(b) Find formulas for S1 , S2 , S3 in terms of r1 , r2 , r3 if

(x − r1 ) · (x − r2 ) · (x − r3 ) = x3 − S1 x2 + S2 x − S3 .

(c) Without expanding the product on the left side of following equation, make a conjecture about
the formulas for S1 , S2 , S3 , S4 in terms of r1 , r2 , r3 , r4 if

(x − r1 ) · (x − r2 ) · (x − r3 ) · (x − r4 ) = x4 − S1 x3 + S2 x2 − S3 x + S4 .

12. (a) Show that:

(x + y)2 ≡ x2 + y 2 mod 2.

(x + y)3 ≡ x3 + y 3 mod 3.

(x + y)5 ≡ x5 + y 5 mod 5.
p


(b) If p is a prime number and 1 ≤ k ≤ p − 1, then by Exercise 7, k ≡ 0 mod p. Use this to show
that (x + y)p ≡ xp + y p mod p.

66
4.3 First identities in Pascal’s triangle
Some surprising and beautiful patterns appear in Pascal’s triangle.

n = 0: 1
n = 1: 1 1
n = 2: 1 2 1
n = 3: 1 3 3 1
n = 4: 1 4 6 4 1
n = 5: 1 5 10 10 5 1
n = 6: 1 6 15 20 15 6 1

The first two patterns are easy to state.

• Symmetry. Recall that nk = n−k n
by Corollary 3.4.5. This is illustrated by the reflectional sym-
metry about the vertical axis (through 1, 2, 6, 20, . . .) in Pascal’s triangle.

• Sums of rows are powers of 2. For instance, the sum of the entries in the n = 4 row of Pascal’s
triangle is 1 + 4 + 6 + 4 + 1 = 16 = 24 . In general, the sum of the entries in row n is 2n .

We can prove the second pattern using the Binomial Theorem. We first re-state it using binomial
coefficient notation.

Proposition 4.3.1. (Sums of rows) If n is a natural number, then

       
n n n n
+ + + ... + = 2n .
0 1 2 n

Proof. By setting x = 1 and y = 1 in the binomial theorem, we see that

2n = (1 + 1)n
       
n n n n−1 1 n n−2 2 n n
= 1 + 1 1 + 1 1 + ... + 1
0 1 2 n
       
n n n n
= + + + ... + .
0 1 2 n

Building on Lemma 4.1.2 and Example 4.1.3, remember how a ball can fall through the grid system of
Pascal’s triangle. Unless stated otherwise, we suppose that the ball has an equal probability of falling down
to the left or to the right. Here is a nice interpretation of the second pattern in terms of the falling ball.

Lemma 4.3.2. Suppose that the probability that the ball falls to the left is the same as the probability that
the ball falls to

the right.

Then the probability that the ball passes through the location labeled by the binomial
coefficient nk is nk /2n .

Proof. If the probability that the ball falls to the left is the same as the probability that the ball falls to the
right, then the probability of falling each way is 1/2. After falling n rows, the probability of each route
is
1/2n . The probability that the ball passes through the location labeled by the binomial coefficient
nk is the
number of routes it could have taken to that location divided by 2n ; this gives a probability of nk /2n .

67
Remark 4.3.3. If the ball falls n steps, then it will land at one of the spots in the nth row. So the sum of
the probabilities of it landing at all the spots is 1. By multiplying each probability by 2n , we see that this
reproduces the pattern given in Proposition 4.3.1.


Example 4.3.4. Building on Example 4.1.3, there is only one path for the ball to reach the entry 40 ; and
the probability

that the ball passes through this location is 1/16; there are four paths for the ball to reach
the entry 41 and the probability that the ball passes through this location is 4/16 = 1/4.
Here is a third pattern in Pascal’s triangle.

• Alternating sums of the rows are 0. For instance, in the n = 4 row, the alternating sum of the
entries is 1 − 4 + 6 − 4 + 1 = 0. In general, the alternating sum of the entries in row n is zero.

The third pattern can also be proven using the Binomial Theorem. We first re-state it using binomial
coefficient notation.
Proposition 4.3.5. (Alternating sums of rows) If n is a positive natural number, then
         
n n n n n
− + − + ... ± = 0.
0 1 2 3 n
(The sign alternates between + and − in the sum. The meaning of the coefficient ± is that the last term is
positive if n is even and the last term is negative if n is odd.)
Proof. By setting x = 1 and y = −1 in the binomial theorem, we see that

0 = (1 + (−1))n
         
n n n n−1 n n−2 n n−3 n
= 1 + 1 (−1)1 + 1 (−1)2 + 1 (−1)3 + . . . + (−1)n
0 1 2 3 n
         
n n n n n
= − + − + ... ± .
0 1 2 3 n

We will give another proof of Proposition 4.3.5 in Proposition 5.4.8.

Exercises
1. You have 14 hats of different colors and you want to loan some of these to a friend. Use the results in
this section to answer the following problems.
(a) In how many ways can you do this if your friend needs at least two hats?
(b) In how many ways can you do this if your friend needs an even number (possibly 0) of hats?
2. For n = 4, 5, 6, find the sum of all the entries in the nth row of Pascal’s triangle, excluding the two
entries on the outer diagonals and the two entries adjacent to those. Find a formula for this sum which
works for all n ≥ 4.
3. For 0 ≤ k ≤ 5, compute the probability that the ball lands at the kth spot in the 5th row (with the
outer diagonal being the case k = 0).
4. Repeat the previous problem when the probability of falling down to the left is 1/3 and the probability
of falling down to the right is 2/3.
5. Label the entries of the nth row of Pascal’s triangle by the numbers k = 0 to k = n. Assume the
probability of the ball falling left or right is the same. For n ≥ 1, in this problem, we will figure out
the probability that the ball lands on an entry where k is even.

68
 (a) Compute this probability when n = 5 and when n = 6.
(b) Make a conjecture for what this probability is in general.
(c) Explain why your conjecture is true using the symmetry pattern and Proposition 4.3.1.

4.4 Additional identities in Pascal’s triangle

There are other patterns in Pascal’s triangle which are more subtle. For the next pattern, consider the sum
of the squares of the entries in each row:
n = 0: 12 =1
n = 1: 12 +12 =2
n = 2: 12 +22 +12 =6
n = 3: 12 +32 +32 +12 = 20
n = 4: 12 +42 +62 +42 +12 = 70

These sums (1, 2, 6, 20, 70, . . .) also appear in Pascal’s triangle, as the middle entry in every row where n is
even. This gives rise to pattern 4.

Proposition 4.4.1. (Sum of squares) If n is a natural number, then

 2  2  2  2  
n n n n 2n
+ + ... + + = .
0 1 n−1 n n

We will prove Theorem 4.4.1 in Proposition 5.3.4.

The fifth pattern is known as the hockey stick identity, because the set of binomial coefficients involved
in the pattern traces out the shape of a hockey stick in Pascal’s triangle.

Proposition 4.4.2. (Hockey stick identity) Suppose n is a positive integer and m is a non-negative integer.
Version 1:          
n n+1 n+2 n+m n+m+1
+ + + ··· + = .
0 1 2 m m
Version 2:          
n n+1 n+2 n+m n+m+1
+ + + ··· + = .
n n n n n+1
The second version of the hockey stick identity is obtained from the first by applying the first pattern,
symmetry, to each binomial coefficient. We will prove Theorem 4.4.2 in Proposition 5.2.10.

Example 4.4.3. We illustrate the first hockey stick identity.

• If n = 2, m = 0, then 1 = 1.

• If n = 2, m = 1, then 1 + 3 = 4.

• If n = 2, m = 2, then 1 + 3 + 6 = 10.

• If n = 2, m = 3, then 1 + 3 + 6 + 10 = 20.

• If n = 2, m = 4, then 1 + 3 + 6 + 10 + 15 = 35.

We can see version 1 of the hockey stick identity when n = 2 and m = 4 traced out in circles in the
following image:

69
 n = 0: 1
n = 1: 1 1
n = 2: 1 2 1

n = 3: 1 3 3 1

n = 4: 1 4 6 4 1

n = 5: 1 5 10 10 5 1

n = 6: 1 6 15 20 15 6 1

n = 7: 1 7 21 35 35 21 7 1

Exercises
1. Draw the example of version 2 of the hockey stick identity on Pascal’s triangle when n = 2 and m = 4.
2. When n = 2 and m = 4, explain why the hockey stick identity version 1
is true using Lemma 4.1.2.
Specifically, separate the 35 paths which end at the binomial coefficient 74 into sets of paths of sizes
15, 10, 6, 3, 1 in a natural way.
3. (Application to probability theory) The chi-square distribution of x0 , . . . , xn is Qn = x20 + · · · x2n .
(a) For n = 1, 2, 3, 4, 5, use Proposition 4.4.1 to compute the chi-square distribution Qn of the binomial
coefficients in the nth row of Pascal’s triangle.
(b) As n gets larger, what happens to the chi-square distribution Qn ?
(c) What is lim Qn ? Explain your answer.
n→∞

4.5 Counting anagrams with multinomial coefficients

The next two sections are about multinomial coefficients, a generalization of binomial coefficients.

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=YAO-1ugO3TI

We start with one application of multinomial coefficients, which is to count anagrams.

Definition 4.5.1. An anagram is a rearrangement of the letters in a word.
An anagram is different from a permutation, because some letters can occur in a word multiple times.
Example 4.5.2. For example, let’s consider the permutations of the word ZOO. Typically, the number of
permutations of three letters is 3! = 6. However, two of the letters in ZOO are the same. After we label the
two O’s as O1 and O2 , the six permutations are:
ZO1 O2 ZO2 O1
O1 ZO2 O2 ZO1
O1 O2 Z O2 O1 Z.
However, the number of anagrams is half of 6, since each anagram of ZOO appears 2 times on this list.
Recognizing that each row gives a different anagram, we can list the 6/2 = 3 anagrams of the word ZOO.
They are: ZOO, OZO, and OOZ.

70
Example 4.5.3. The word SASSY has 20 anagrams. Typically, the number of permutations of five letters
is 5! = 120. In the list of these permutations, each anagram is written 3! = 6 times, corresponding to the
rearrangement of the three S’s. So the number of anagrams is 5!/3! = 20.

Example 4.5.4. How many anagrams are there for FORTCOLLINS? One approach is to say that there are
11 letters in total, but two of these are L and two of these are O. So in the list of 11! permutations of the
letters, each word is written four times (the three extra times correspond to switching the locations of the
two L’s, switching the locations of the two O’s, and switching both the two L’s and the two O’s). Hence the
total number of anagrams is 11!/4.

Example 4.5.5. Let
W be a binary string of length n, having k zeros and n − k ones. Then the number of
anagrams of W is nk because the anagram is determined exactly by the location of the k zeros.

The same ideas from the previous examples lead to the following definition and theorem:

Definition 4.5.6. Let n be a natural number. Let k1 , . . . , km be m positive natural numbers such that

k1 + · · · + km = n.

Let W be a word with n letters of m different types, in which the letter of type i appears ki times, for
1 ≤ i ≤ m. The multinomial coefficient  
n
k1 , . . . , k m
is defined to be the number of anagrams of W .

Theorem 4.5.7. The multinomial coefficient has the following formula:

 
n n!
= . (4.1)
k1 , . . . , k m k1 ! · k2 ! · . . . · km !

We prove Theorem 4.5.7 in Section 4.6.

Example 4.5.8. How many anagrams are there for the word MISSISSIPPI? In this case n = 11, m = 4,
k1 = 4 (the I’s), k2 = 1 (the M), k3 = 2 (the P’s) and k4 = 4 (the S’s). So the answer is 11!/(4! · 1! · 2! · 4!).

Exercises
1. How many anagrams does the word BANANA have? For example, NABNAA and NANABA are two
anagrams of BANANA.

2. How many ways can you rearrange the letters in the word MOON?

3. How many anagrams does the word HORSETOOTH have?

4. How many anagrams does the word TENNESSEE have?

5. Consider all anagrams of the word SPEEDED. How many ways are there for you to make a ranked list
of your top 5 favorite anagrams, numbered #1 to #5, of SPEEDED?

6. How many paths are there from (0, 0, 0) to (3, 4, 5) in three dimensional space, using only unit steps
in directions (1, 0, 0), (0, 1, 0), and (0, 0, 1)?



7. Suppose m = 2. Show that k1n,k2 is the familiar binomial coefficient kn1 and show that the formula
n!
in (4.1) is the same as the formula k1 !(n−k 1 )!
found in Lemma 1.2.6 and Theorem 3.4.1.

71
4.6 Ordered set partitions and the multinomial theorem


Recall that the binomial coefficient nk counts the number of ways to choose a subset of size k from a set S
of size n. A different way to think of this is that we are separating the set S into two parts: the k elements
that are chosen, and the n − k elements that are not chosen. What happens if we want to separate the set
S into more than two parts?

Example 4.6.1. Suppose there are 7 different fruits in your refrigerator. You want to eat two fruits on
Monday, three on Tuesday (since that is your workout day), and two on Wednesday. In how many different
ways can you choose which fruits to eat on each day?
7

One way to solve this problem is to successively count possibilities for each day. On Monday,5
there are
2 ways to choose which two fruits you eat. On Tuesday, there are 5 fruits left, so there

are 3 ways to
choose which three fruits to eat. On Wednesday, there are 2 fruits left, so there is only 22 = 1 way to choose
the remaining fruit. So the total number of possibilities is
     
7 5 2 7! 5! 2!
· · = · · .
2 3 2 2! · 5! 3! · 2! 2! · 0!

Note that many of these factorials cancel, and we can express the result more simply as

7!
= 210.
2!3!2!
In the next definition, we generalize Definition 2.9.1, by replacing a set partition by an ordered set
partition.

Definition 4.6.2. An ordered set partition of a set S is a collection of non-empty subsets B1 , . . . , Bm

of S, such that the intersection of any two of them is trivial and such that the union of all of them is S.

In other words, the ordered set partition separates the elements of S into subsets B1 , . . . , Bm with no
overlap. The sets B1 , . . . , Bm are called the blocks of the ordered set partition.

Theorem 4.6.3. The number of ordered set partitions of an n-element set S into blocks of sizes k1 , k2 , . . . , km
is the multinomial coefficient  
n n!
= .
k1 , . . . , km k1 ! · k2 ! · · · km !
Proof. We count the number
of ordered set partitions of a set S of size n into blocks of sizes k1 , k2 , . . . , km
as follows. There are kn1 = k1 !(n−k
n!
1 )!
ways to choose the first block B1 . Out of the remaining n − k1

(n−k1 )!

elements, there are k2 = k2 !(n−k1 −k2 ) ways to choose the next block B2 . If m ≥ 3, there are n−kk13−k2 =
n−k1

(n−k1 −k2 )!
ways to choose the third block B3 , and so on. Multiplying these together, all the factorials
k3 !(n−k1 −k2 −k3 )!
cancel except for n! in the numerator and the terms k1 !, . . . , km ! in the denominator. The result follows.

Example 4.6.4. In Example 4.6.1, the set S is the set of fruits in your refrigerator, m = 3, and B1 , B2 , B3
are the sets of fruit you will eat on each
day (in order). Also n = 7, k1 = 2, k2 = 3, and k3 = 2. So the
7
number of ordered set partitions is 2,3,2 = 210.

Example 4.6.5. Suppose we have 11 different presents, and we would like to give 4 to person a, then 1 to
person b, then 2 to person c, then 4 to person d. The number of ways we can do this is
 
11 11!
= = 34650.
4, 1, 2, 4 4!1!2!4!

In the next remark, we compare anagrams with ordered set partitions. We will use this to prove Theo-
rem 4.5.7.

72
Remark 4.6.6. Suppose n = k1 + · · · + km . Suppose W is a word of length n having ki copies of the letter
ai , for 1 ≤ i ≤ m, where the letters are taken in alphabetical order. Anagrams of W are directly related to
ordered set partitions of the set S = {1, . . . , n}. Here is how that works.
Given an anagram of W , we can make an ordered block partition as follows: for 1 ≤ i ≤ m, find the
positions of the ith letter and put those numbers in the block Bi . For example, consider the anagram
IMSSSPIIPSI (of MISSISSIPPI): the I’s are in positions 1,7,8, and 11, the M is in position 2; the P’s are in
positions 5 and 9; and the S’s are in positions 3,4,5, and 10. From this, we make the ordered set partition
with blocks B1 = {1, 7, 8, 11}, B2 = {2}, B3 = {5, 9}, and B4 = {3, 4, 5, 10}. Conversely, given an ordered set
partition with blocks Bi of size ki for 1 ≤ i ≤ m, we can make an anagram of the word W , by putting the ith
letter in the positions numbered by the elements in the block Bi . For example, the blocks B1 = {2, 5, 8, 11},
B2 = {1}, B3 = {9, 10} and B4 = {3, 4, 6, 7} gives the original spelling of MISSISSIPPI.
Proof. (Proof of Theorem 4.5.7) By Remark 4.6.6, the number of anagrams of W is the same as the number
of ordered set partitions into blocks of sizes B1 , . . . , Bm . By Theorem 4.6.3, we obtain the desired formula
for the number of these.
The next example illustrates the difference between an ordered set partition and a set partition.
Example 4.6.7. Suppose there are 12 students in a class.
1. How many ways can they be separated into 4 groups of 3 who will be assigned to present homework
solutions on Monday, Tuesday, Wednesday, and Thursday?
2. How many ways can they be separated into 4 groups of 3?
Answer. 1. Each assignment is an ordered set partition of the 12 students into blocks B1 , . . . , B4 of size
three (speaking on Monday, Tuesday, Wednesday, Thursday respectively). So the answer is
 
12 12!
= .
3, 3, 3, 3 3! · 3! · 3! · 3!

2. This is a set partition since the blocks of students are not assigned to different days of the week. There
are 4! ways to order the 4 blocks. So we divide the answer to part (1) by 4!.
As a final application, we see what happens if we take
a power of a sum of more than two variables. The
n
answer to this question is why the expression k1 ,...,kn
is often called a multinomial coefficient.
Example 4.6.8. The third power of x + y + z is
(x + y + z)3 = x3 + y 3 + z 3 + 3x2 y + 3xy 2 + 3x2 z + 3xz 2 + 3y 2 z + 3yz 2 + 6xyz.
The coefficients of this expansion are all examples of multinomial coefficients.
Here is the Multinomial Theorem, which directly generalizes the Binomial Theorem from Theo-
rem 4.2.2.
Theorem 4.6.9. Consider the
expansion of (x1 + x2 + · · · + xm )n . The coefficient of xk11 · · · xkmm is the
n
multinomial coefficient k1 ,...,km
. In other words,
X  
n
n
(x1 + x2 + · · · + xm ) = xk1 · · · xkmm .
k1 , · · · , km 1
k1 +...+km =n

For example, the term 3x y in the above expansion of (x + y + z)3 can be written as 3x2 y 1 z 0 , which does
2

indeed agree with the multinomial coefficient

 
3
= 3.
2, 1, 0
The following proof is quite difficult and it is alright for you to come back to it after gaining more
experience with proofs.

73
Proof. We can count how many ways to obtain the monomial xk11 · · · xkmm by multiplying terms in the expan-
sion as follows. We label the factors of (x1 +x2 +· · ·+xm )n by F1 , . . . , Fn . (Each factor equals x1 +x2 +· · ·+xm
but we think of them as being different since one of them is first, second, etc). Let S = {F1 , . . . , Fn } be the
set of the n factors. To make the monomial xk11 · · · xkmm in the product, we need to choose the variable x1
from k1 of the factors, then choose the variable x2 from k2 of the factors, and so on. This forms an ordered
set partition of S with sizes k1 , k2 , . . . , km , and so the result follows by Theorem 4.6.3.

Exercises
1. Find the number of ways to give 20 pillows to 12 students a through l if the pillows are all different
and we give 2 pillows to each of students a, b, c, d, e, f, g, h and 1 pillow to each of students i, j, k, l.

2. How many ways are there to assign the 12 months (Jan-Dec) to three nurses a, b, and c so that each
worker gets four months of “on call” duty?
3. Five siblings decide that one of them needs to call their mother every day of the year. How many ways
are there to distribute the 365 = 5 · 73 days of the year among themselves so that each of them makes
73 calls? Write out the formula, but do not figure out the actual number of ways.
4. Nine friends are going to a math conference! The friends will share three hotel rooms, where the first
hotel room fits 4 people, the second fits 3 people, and the third fits 2 people. They are all staying
in the hotel for 5 nights and are allowed to rearrange rooms every night. (Repeating the same room
assignments for different nights is allowed). In how many different ways can the 9 friends create a
5-night schedule of room assignments? Write out the formula, but do not figure out the actual number
of ways.
5. You have eight different decorative Halloween stickers (a witch, a ghost, a pumpkin, a skeleton, a
vampire, a mummy, a bat, and a black cat) and four windows to decorate them with. In how many
ways can you put two stickers on each window?

6. How many ways can you put 16 differently-colored billiard balls into the six different pockets of a pool
table so that each corner pocket gets 3 balls and the two middle pockets get 2 each?

74
4.7 Investigation: Combinatorial problems on a chessboard
The problems in this section take place on a chess board. If you know the basics of chess pieces, you can
skip the next remark.

Remark 4.7.1. A chess board is an 8 × 8 board tiled in 1 × 1 squares, colored black and white, so that no
black square shares an edge with a white square (and vice-versa).
Chess is a game for two players which takes place on this board. Each player starts the game with pieces:
1 king, 1 queen, 2 rooks, 2 bishops, 2 knights, and 8 pawns. One player’s pieces are gold and the other’s are
silver. Any game piece can be placed on any square of the board, regardless of material and color.
Each type of piece can move in different ways. During a turn, a rook can move to any square in the
same row or the same column that it started the turn on. During a turn, a queen can move to any square
in the same row, column, or two diagonals that it started the turn on. Two rooks of different materials are
attacking each other if one of them can move to the square of the other in one move. Similarly, two queens
of different colors are attacking each other if one of them can move to the square of the other in one move.

In some of the problems below, we are working with rooks made from different materials, with an
unlimited supply of rooks made from each material. To make everything in the following problems precise,
it is also important to note that the chess board is glued to the table so it cannot be turned over or rotated.

1. How many ways are there to place two gold rooks and two silver rocks on the board (with no other
game pieces):

(a) so that no square has more than one rook on it?

(b) so that no square has more than one rook on it and no gold rook is attacking a silver rook?

2. What is the largest number of rooks (of the same material) that can be placed on the board so that
no two are in the same row or in the same column? How many ways can these rooks be placed on the
board?

3. Show that it is possible to put 16 rooks (four gold, four silver, four titanium, and four diamond) on
the board, so that no pair of rooks with different materials is attacking each other. How many ways
can this be done?

4. (The peaceful rooks problem) What is the largest number m such that m gold rooks and m silver rooks
can be placed on the board so that no gold rook is attacking a silver rook? How many ways can this
be done?

5. Suppose there is one queen on the board and no other pieces.

(a) Find a position for the queen where the number of squares it can move to is as small as possible.
Which position is this and how many squares can it move to?
(b) Find a position for the queen where the number of squares it can move to is as big as possible.
Which position is this and how many squares can it move to?

6. In this problem, we change the size of the board to n × n and put n gold queens on the board so that
no two (of the same material) are attacking.

(a) Show this is not possible if n = 2 or n = 3.

(b) Find a way to do this if n = 4.
(c) Find a way to do this if n = 5.
(d) Look up the 8 queens puzzle. Explain what it is and draw a solution.

75
7. In this problem, we change the size of the board to n × n. Let m be the maximum number such that m
gold queens and m silver queens can be placed on the n × n board so that no two of different materials
are attacking. For example, when n = 3, then m = 1, because the gold queen can be placed in the
corner and the silver queen can be placed in the middle of one of the opposite sides; however it is not
possible to place 2 gold queens and 2 silver queens on a 3 × 3 board without them attacking each other.

(a) When n = 4, show that m ≥ 2 by finding a way to put 2 gold queens and 2 silver queens on the
board.
(b) When n = 5, show that m ≥ 4 by finding a way to put 4 gold queens and 4 silver queens on the
board.
(c) Look up the peaceable queens problem. Find out the value of m when n = 8.

76
Chapter 5

Proof Techniques in Combinatorics

In this chapter, we explain how to prove statements in combinatorics. After explaining why proofs are
useful, we focus on proof techniques that are often useful in combinatorics: induction, counting in two ways,
bijective proofs, proof by contradiction, and the pigeonhole principle.

Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=TFNwxluopek

5.1 Why are proofs necessary?

Writing proofs is not easy. Sometimes students feel that professors assign proofs the same way that a drill
sergeant assigns push-ups: after 50 of these, you’ll be stronger!. At a basic level, proofs are needed to show
that a statement is true without a doubt. Beyond this, proofs are valuable because they explain the reasons
why statements are true.
Many people find proofs to be beautiful and powerful. Often mathematicians have favorite proofs,
which illustrate core topics in their research areas. Some mathematicians view proofs as the foundation of
mathematics, the solid base needed to build fantastic structures. Other mathematicians view proofs as a
living language, which connects the thoughts from ancient civilizations to the thoughts in modern times.
In this chapter, we will explain some proof techniques. We cannot promise that you will love proofs by
the end of the chapter, but hopefully writing a proof will seem more manageable. Maybe by the end of this
class, you will have a favorite proof too.
To get started, we will share some examples that show numerical data is sometimes misleading and share
some history about famous proofs in combinatorics and number theory.
Example 5.1.1. How many positive divisors does n! have? (This is denoted by σ0 (n!), see Section 2.11.1.)
To get started, let’s write out the first few cases. If n = 1, then n! is just 1, which only has one divisor
(1 itself). If n = 2, then 2! = 2, which has 2 divisors (1 and 2). If n = 3, then 3! = 6 which has 4 divisors
(1, 2, 3, and 6). With a little more effort, we find that 4! = 24 has 8 divisors and 5! = 125 has 16 divisors.
So the pattern of answers so far is 1, 2, 4, 8, 16. How many divisors do you think 6! has?
You might have guessed 32, since the answer seems to double at each step. But alas, 6! only has 30
divisors! This illustrates why we need mathematical proofs, to show that a pattern continues indefinitely.
We need to be sure that there is not some counterexample to the pattern that is so large that we simply
have not found it yet.
Example 5.1.2. Here are three numbers that are almost integers, meaning that they are so close to integers
that you might not realize they are transcendental numbers:
√
eπ 43
≈ 884736743.9997;

77
 √
eπ 67
≈ 147197952743.999998;
√
eπ 163
≈ 262537412640768743.9999999999992.
The last one is called Ramanujan’s constant after the famous mathematician Ramanujan
√
because, in April
1975, Gardner wrote in Scientific American that Ramanujan conjectured that eπ 163 is an integer; this was
an April fool’s hoax. See OEIS A060295.

Example 5.1.3. The Four Color Theorem states that for any map of countries drawn on a 2-dimensional
world map, it is always possible to use four colors to color the countries in the map, such that no two
countries sharing a border have the same color.
This was an open problem for a long time, and in 1976 it was finally proven by Appel and Haken using
a computer. They first used mathematical proof reasoning techniques to categorize the possible maps into
various types, and then used a computer to exhaustively check all of these cases. Computers can make a
proof much faster to complete when there are many cases to check.
a
Example 5.1.4. In the 1600s, Fermat studied numbers of the form Fa = 22 + 1, which are now called
Fermat numbers. The first few examples of Fermat numbers are

F0 = 3, F1 = 5, F2 = 17, F3 = 257, F4 = 65537.

After Fermat noticed that these five numbers are prime, he conjectured that Fa is always prime. But this
was found to be false! In 1732, Euler showed that

F5 = 4294967297 = 641 · 6700417.

Currently no one knows how many Fermat numbers are prime. Even with modern computing power,
no more examples of Fermat primes have been found. Fermat numbers are still useful because they can
have very large prime factors. In 2020, as part of the PrimeGrid collaboration, Brown, Reynolds, Penné &
Fougeron found the megaprime 13 · 25523860 + 1 as a factor of F5523858 .

Example 5.1.5. Fermat’s little theorem states: if p is a prime number, then p divides ap−1 − 1 for all a such
that gcd(a, p) = 1. For example, let p = 5. For a = 1, 2, 3, 4, then a4 − 1 equals 0, 15, 80, 255 (respectively)
and these numbers are all divisible by 5.
We could ask if the converse to Fermat’s little theorem is true: if n divides an−1 − 1 for all a such that
gcd(a, n) = 1, then is n prime? If you experimented with different choices of n, you might start to believe
that this is true. But there is a counterexample! The number n = 561 divides a560 − 1 for all a such that
gcd(a, 561) = 1. However, 561 is not prime, because it factors as 561 = 3 · 11 · 17. The number 561 is called
a Carmichael number; it is an example of a pseudo-prime (fake prime). This topic is useful in cryptography.

Example 5.1.6. A final famous problem named after Fermat is Fermat’s Last Theorem. In 1637, Fermat
conjectured that the equation
Xn + Y n = Zn
has no solutions when n ≥ 3 is an integer and when X, Y, Z are integers, unless XY Z = 0. This problem
motivated people to study number theory for hundreds of years and their work led to the development
of algebraic number theory and arithmetic geometry. In 1994, Wiles published a proof of Fermat’s Last
Theorem, building on work of Frey, Ribet, Serre, Taniyama and Shimura. Wiles received the Abel prize,
worth about $700,000, for the proof.

Exercises
1. Show that π 3 and sin(11) are almost integers.
√
2. The golden ratio is φ = (1 + 5)/2 = 1.618 . . .. Show that φ17 , φ18 , and φ19 are almost integers.

78
 3. Gelfond’s constant is eπ . Show that eπ − π is an almost integer. (Unlike the other examples in this
section, there is no known explanation for this, so it is viewed as a coincidence.)

4. Read the history of the proof of the 4-color theorem. What was unusual about the proof of Appel and
Haken?

5. Read the history of Tait’s conjecture in graph theory. What was important about Tutte’s work on this
conjecture?

6. Read the history of the proof of the Poincaré conjecture. How do you think Hamilton felt about
Perelman’s proof of this conjecture?

7. Read the history of the proof of the ABC-conjecture. Decide whether you think Mochizuki proved this
conjecture or not.

8. There is a conjecture about primes, called Vandiver’s Conjecture, which has been verified for all primes
less than 163 million. Yet many mathematicians do not believe it is true for all primes. What kind of
justification is needed to convince someone that the data is conclusive or inconclusive?

9. Explain the mistake in one of the false proofs at http://mathforum.org/dr.math/faq/faq.false.

proof.html or http://www.math.toronto.edu/mathnet/falseProofs/

5.2 Induction
Click on the icon at left or the URL below for this section’s short video lecture.
https://www.youtube.com/watch?v=cBff3lTYlBs

The natural numbers N = {0, 1, 2, 3, . . . , } are lined up like a row of dominoes.

The goal of this section is to see how to prove facts or formulas that are true for each natural number
n ∈ N. It is not possible to check one value of n at a time because there are infinity many of them. A good
analogy for an inductive proof is the idea of knocking over all the dominoes: first, you need to knock over
one of them (usually the first); then, you need a method to make sure that when each domino falls, it knocks
over the next. Thus, knocking over one domino sets off a chain reaction that causes all of the infinitely many
dominoes to fall over.
To state this mathematically, we consider a property or fact or formula called Pn , which could be true
or false for each n ∈ N. For example,

Pn : 1 + 2 + · · · n = n(n + 1)/2.

We already saw in Lemma 1.2.1 that this property Pn is true for all integers n ≥ 1. As another example:

Pn : The Fermat number Fn is prime.

We already saw in Example 5.1.4 that this property Pn is true when n = 1, 2, 3, 4 but false when n = 5. The
first is true for all n and we will see how to prove it using induction, but since the second is not true for all
n, we cannot prove it by any method.

Theorem 5.2.1. (The inductive principle) Suppose

• (Base case) the property Pn is true for n = 0, and

• (Inductive step) if property Pk is true for an arbitrary integer k ≥ 1, then so is property Pk+1 .

Then Pn is true for all integers n ∈ N.

79
 The above theorem is often called the Principle of Induction or just induction, and it is often thought of
as a method for proving that a statement Pn is true for all integers n ∈ N, as follows.

Method of Proof by Induction: To prove Pn is true for all natural numbers n ≥ 0:

1. Prove the base case: that Pn is true for n = 0. (This usually is an easy computation.)

2. Let k ∈ N be arbitrary, and assume Pk is true for this specific value k. (The statement Pk is called the
inductive hypothesis.)

3. Using the assumption that Pk is true, prove that Pk+1 is true.

Notice that steps 2 and 3 above together prove the inductive step, since to prove an “if... then” statement
in mathematics, one assumes that the hypothesis is true and then proves the conclusion holds under that
assumption.
In step 2, choosing k to be arbitrary means that we are showing that no matter what domino falls, it
will always knock over the next one. Combining this fact with the base case that the first domino falls, we
can conclude that every domino falls (the statement Pn is true for all n).

Remark 5.2.2. The inductive step can be replaced with: if property Pk−1 is true for an arbitrary integer
k ≥ 1, then so is property Pk . In other words, in step 2, we can let k ≥ 1 (rather than k ≥ 0) and assume
Pk−1 , and then in step 3, we prove Pk . This is a purely aesthetic choice, but sometimes makes the algebra
in a proof simpler to work with.

Example 5.2.3. The sum of the first n odd integers is n2 . In other words, the following property is true
for all integers n ≥ 1:
Pn : 1 + 3 + 5 + . . . + (2n − 1) = n2 .

Proof. We proceed by induction.

(Base case.) For n = 1, note that 1 = 12 .
(Inductive step.) Let k ≥ 1 be an arbitrary natural number, and suppose Pk is true, meaning that

1 + 3 + 5 + . . . + (2k − 1) = k 2 .

Adding 2k + 1 to both sides gives

1 + 3 + 5 + . . . + (2k − 1) + (2k + 1) = k 2 + 2k + 1 = (k + 1)2 ,

showing that Pk+1 is true as desired. Hence by induction, Pn is true for all n ≥ 1.

The next two lemmas can similarly be proven by induction (see the Exercises).

Lemma 5.2.4. (See Lemma 1.2.1) The following property is true for all integers n ≥ 1:

Pn : 1 + 2 + · · · n = n(n + 1)/2.

Lemma 5.2.5. The following property is true for all integers n ≥ 1:

1 + 2 + 4 + 8 + . . . + 2n−1 = 2n − 1.

Remark 5.2.6. Sometimes a statement Pn is false for small values of n. If there is a positive integer c for
which Pc is true and the inductive step works, then we can modify the method of induction to prove that
statement Pn is true for all n ≥ c (instead of n ≥ 0). To do this, we change the base case to prove Pc rather
than prove P0 , and also change the assumption on the inductive step to choose k ≥ c and assume Pk .

Example 5.2.7. Prove that n! ≥ 2n for all n ≥ 4.

80
Proof. In this case, the statement is false when n = 1, 2, 3 but true when n = 4. So we proceed by induction
on n, starting with the case n = 4.
(Base case.) The identity is true when n = 4 because 24 = 4! ≥ 24 = 16.
(Inductive step.) Let k ≥ 4 be arbitrary, and suppose Pk is true, meaning that k! ≥ 2k . Then

(k + 1)! = (k + 1)k! ≥ (k + 1)2k ≥ 5 · 2k ≥ 2k+1 ,

showing that Pk+1 is true as desired. Hence by induction, Pk is true for all n ≥ 4.
Remark 5.2.8. Suppose we want to show that two equations A and B are equal. One common mistake
that people make in writing proofs is to assume A = B and then simplify. Warning: this does not always
work! Here is an extreme example of a false claim and a false proof to illustrate this:
Claim. 3 = 5.
Proof. Note 3 = 5 implies 3 · 0 = 5 · 0 implies 0 = 0 which is true.
Here is an example of a false claim and a false proof by induction, that emphasizes the need to get both
the base case and the inductive hypothesis step correct.
Example 5.2.9. All horses are the same color.
Proof. We will prove by induction that all horses in a herd of size n are the same color for any n ≥ 1.
(Base case.) In any herd of 1 horse, all horses in that herd are the same color.
(Inductive step.) Let k ≥ 1 be arbitrary, and suppose all horses in a herd of size k are the same color.
Consider a herd of k + 1 horses.

....

same color same color

Removing the last horse, the inductive assumption says that the first horse is the same color as the horses
in the middle. Removing the first horse, the inductive assumption says that the last horse is the same color
as the horses in the middle. Hence all k + 1 horses in the herd are the same color. By induction, the horses
in a herd of size n are the same color for all n ≥ 1.
Clearly not all horses are the same color, so where did our proof above go wrong? Our mistake in
Example 5.2.9 is that the inductive step fails when going from k = 1 to k = 2, even though it works for
all larger values of k. When k = 1, then the middle herd is empty. This is why the bound on k in the
declaration “Let k ≥ 1 be arbitrary” is so important.
We will now give a proof by induction of the Hockey Stick Identity from Proposition 4.4.2. In this case,
there are two variables n and m in the formula, so we need to be careful to specify which one is increasing
in the inductive step.
Proposition 5.2.10. Suppose n is a positive integer and m is a non-negative integer. Then
         
n n+1 n+2 n+m n+m+1
+ + + ... + = .
0 1 2 m m
Proof. We proceed by induction on m for a fixed n.
(Base case.) The identity is true when m = 0 because
   
n n+0+1
=1= .
0 0

81
 (Inductive step.) Let m ∈ N be arbitrary, and assume that
         
n n+1 n+2 n+m n+m+1
+ + + ··· + = . (5.1)
0 1 2 m m
We must now show that
           
n n+1 n+2 n+m n+m+1 n+m+2
+ + + ··· + + = . (5.2)
0 1 2 m m+1 m+1

n+m+1

By substituting (5.1), we see that the left hand side of (5.2) simplifies to n+m+1 + m+1 which equals
n+m+2

m

m+1 by Pascal’s recurrence Theorem 4.1.1. Hence the statement is true for all non-negative integers m
by induction.
Remark 5.2.11. There is a variation of the inductive process, called Strong induction. In the inductive
step of this variant, we suppose that Pk is true for all 1 ≤ k ≤ n, and show that Pk is true when k = n + 1.

Exercises
1. Prove Lemma 5.2.4 using induction.
2. Prove Lemma 5.2.5 using induction.
3. Prove the geometric series formula using induction:
rn+1 − 1
1 + r + r2 + · · · + rn =
r−1

4. Prove using induction that the following identity is true for all integers n ≥ 1:
1 1 1 n
Pn : + + ... + = .
1·2 2·3 n(n + 1) n+1

5. Prove using induction that the following identity is true for all integers n ≥ 1:
n(n + 1)(2n + 1)
Pn : 12 + 22 + 32 + . . . + (n − 1)2 + n2 = .
6

6. Prove using induction that n2 ≥ 5n + 5 for all integers n ≥ 6.

7. Prove using induction that 2n ≥ 10n + 7 for all integers n ≥ 7.
8. Prove that every positive integer n ≥ 2 can be factored as a product of one or more prime numbers
using strong induction.
(Hint: In the induction step, consider two cases: if the number k+1 is prime then we are done;
otherwise it factors as a product of two smaller numbers, each of which factors into primes by the
induction hypothesis.)

5.3 Counting in two ways

Click on the icon at left or the URL below for the next two sections’ short video
lecture.
https://www.youtube.com/watch?v=C0g hGyeO30

We can prove some formulas from earlier in the book using the principle called counting in two ways.

82
Lemma 5.3.1. (Counting in Two Ways.) If a set has n elements and it also has m elements then n = m.

This sounds like a trivial principle, but it is incredibly useful in proving algebraic identities in combi-
natorics. Here is the way we can use this. Suppose we want to prove F = G where F and G are both
mathematical formulas. Then we simply have to find a collection of objects such that both F and G are the
answer to the question “how many are there in the collection”?

Example 5.3.2. Suppose a and b are positive integers. Here is a combinatorial proof that

a · b = b + b + ··· + b
| {z }
a summands

where the sum has a copies of b. By repeated application of the Addition Principle, we see that the right-
hand side, b + b + · · · + b, counts the number of students in a school that has a classes with b students each.
By the Multiplication Principle, a · b counts the number of ways to first walk into one of the a classrooms and
then single out one of the b students in that class. Thus a · b is also equal to the total number of students,
and so a · b = b + b + · · · + b.

We can also prove more intricate identities combinatorially.

Example 5.3.3. Here is a combinatorial proof of Pascal’s recurrence in Theorem 4.1.1 that
     
n+1 n n
= + .
k k−1 k

Proof. Let’s find the number of ways of choosing k people from a
classroom that has n students and 1
professor. Since there are n + 1 people in the room, the answer is n+1 k . On the other hand, we can separate
the ways of choosing the group

into (i) those that do not

contain the teacher and (ii) those that do contain
the teacher. There are nk choices in case (i) and k−1 n
in case (ii) (since we are choosing only k − 1 of the
students). By the addition principle, the number of ways is the sum of the answers for (i) and (ii) and the
result follows.

We now give combinatorial proofs of the “sum of squares” formula from Proposition 4.4.1.


n 2


n 2 n 2



n 2 2n


Proposition 5.3.4. If n is a natural number, then 0 + 1 + ... + n−1 + n = n .

We will prove this in two ways, neither of which is easy. The second proof proves an even harder identity,
Vandermonde’s identity, which reduces to Proposition 5.3.4 when we make a good choice for the variables.

Proof. Note that 2n n is the number of ways to choose n soccer players out of a team of 2n soccer players.
We will count that in another way by separating the 2n soccer players into two groups, by putting a gold
jersey on n of the soccer players and a green jersey on the other n soccer players. From the first group, we
choose k soccer players with gold jerseys. In order to choose n soccer players altogether, we need to choose
n − k soccer players with green jerseys. Symbolically,

1, 2, . . . , n, n + 1, n + 2, . . . , 2n .
| {z } | {z }
choose k choose n − k

So there are nk · n−k
n
ways to do that.
The choice of the number k was arbitrary and so we need to consider all the possible values of k, which
could be as small as 0 and as big as n. So we need to add up the number of ways to choose n soccer players

83
as k varies between 0 and n. This gives us:
  X n
2n
= (# choices of k players with gold jerseys) · (# choices of n − k players with green jerseys)
n
k=0
Xn    
n n
= ·
k n−k
k=0
Xn    
n n
= ·
k k
k=0
n
X n  2
=
k
k=0
 2  2  2  2
n n n n
= + + ... + + .
0 1 n−1 n

Here is another identity.

Proposition 5.3.5. (Vandermonde’s identity.) Suppose n, m, ` are non-negative integers such that 0 ≤ ` ≤
m + n.
  X̀   
n+m n m
= .
` k `−k
k=0



(In the summation above, we set nk = 0 if k > n and similarly set `−k
m
= 0 if ` − k > m.)
Proof. By analogy: in your kitchen cabinet, you have n granola bars (all different) and m bags of trail mix
(all different). The left hand side counts the number of ways to choose ` snacks out of n + m choices.

m
of ways to choose the snacks with exactly k granola bars and ` − k bags of trail
Let Dk be the number
mix. Then Dk = nk `−k because you are choosing k granola bars out of n and ` − k bags of trail mix out of
P`
m. We add up the number of ways for each possibility for k. This equals the right hand side k=0 Dk .
Using Vandermonde’s identity, we can find a second proof of Proposition 4.4.1.
Proof. In Proposition 5.3.5, substitute m = n = ` and simplify. This yields
  X n  2
2n n
= .
n i=0
i

Exercises
1. Find a counting in two ways proof for Gauss’s formula for the triangular numbers:
n(n + 1)
1 + 2 + 3 + ··· + n = .
2

2. Give a proof of the following identities by counting in two ways:

(a) nk = n−kn

n n−m




(b) m k = nk n−km




(c) n0 + n1 + · · · + nn = 2n

84
 n



(d) k · k = n · n−1
k−1
n





(e) 1 + 2 · n2 + 3 · n3 + · · · + n · n
n = n · 2n−1
3. Recall from Proposition 4.4.2 that the Hockey Stick identity (version 1) states that
       
n n+1 n+m n+m+1
+ + ··· + = .
0 1 m m

Find a counting in two ways proof of this identity.

5.4 Bijective proofs

Click on the icon at left or the URL below for this and the last section’s short
video lecture.
https://www.youtube.com/watch?v=C0g hGyeO30

Here is a sample problem that you will be able to solve at the end of this section.
Question 5.4.1. How many ways can you choose 4 numbers from {1, . . . , 100} so that none of them are
consecutive?


We already know that there are 1004 ways to choose 4 numbers from {1, . . . , 100} but it is not clear how
to deal with the condition that they are not consecutive. In this problem, it is not enough to count the size
of one set in two ways. Instead, we will use a bijection to compare the size of two different sets.
Let A and B be sets. A bijection is a matching or pairing of the elements of A with the elements of B.
Here is the formal definition.
Definition 5.4.2. Let A and B be sets. Let f : A → B be a map.
1. The map f is 1-to-1 if, whenever a1 6= a2 , then f (a1 ) 6= f (a2 ).
2. The map f is onto if for every b ∈ B, there is an a ∈ A such that f (a) = b.
3. The map f is a bijection if it is 1-to-1 and onto.
Intuitively, the onto condition means that the map f hits every element of B. Another way to say this is
that the range of f is all of B. The 1-to-1 condition means that no element of B is hit twice; it is equivalent
to saying, if f (a1 ) = f (a2 ), then a1 = a2 .
Example 5.4.3. Let Ω = {1, . . . , n}. Fix k such that 0 ≤ k ≤ n. Let A be the set of subsets of Ω of size k.
Let B be the set of subsets of Ω of size n − k. There is a bijection between A and B taking a subset S of
size k to the complement Ω − S which has size n − k.
Lemma 5.4.4. There is a bijection f : A → B if and only if f has an inverse map g : B → A.
Proof. Suppose f : A → B is a bijection. Here is how to define the map g : B → A. Given b ∈ B, since f is
surjective, there is an a ∈ A such that f (a) = b. Also a is unique since f is 1-to-1. Define g(b) = a. To show
that g is the inverse of a, we need to show that g ◦ f : A → A and f ◦ g : B → B are both the identity map.
That is true since (g ◦ f )(a) = g(f (a)) = g(b) = a and (f ◦ g)(b) = f (g(b)) = f (a) = b.
Conversely, if g : B → A is an inverse of f , we want to show that f is a bijection. To show that f is
1-to-1, suppose that f (a1 ) = f (a2 ). Then g(f (a1 )) = g(f (a2 )). So (g ◦ f )(a1 ) = (g ◦ f )(a2 ). Since g ◦ f is
the identity map, this shows that a1 = a2 , completing the proof that f is 1-to-1. Finally, to show that f is
onto, choose b ∈ B. Let a = g(b), which is in A. Then f (a) = f (g(b)) = (f ◦ g)(b). Since f ◦ g is the identity
map, this shows f (a) = b, completing the proof that f is onto.

85
Example 5.4.5. Let A be the set of positive integers whose last digit is 3. Let B be the set of positive
integers whose last digit is 7. There is a bijection f : A → B given by f (a) = a + 4. Its inverse map
g : B → A is given by g(b) = b − 4.

Example 5.4.6. Let A be the set of integers. Let B be the set of multiples of 7. There is a bijection
f : A → B given by f (a) = 7a. The inverse map is g : B → A given by g(b) = b/7.

Lemma 5.4.7. (Bijective proof.) If A and B are sets such that A has m elements, B has n elements, and
there is a bijection f : A → B, then m = n.

Again, this sounds like a very simple principle, but it can be tricky to execute in practice. Here, we
provide several harder examples.
Recall the following identity (alternating sum of binomial coefficients) from Proposition 4.3.5.
n

n



n n


Proposition 5.4.8. If n > 0, then 0 − 1 + 2 − ··· ± n = 0.

Another way to write Proposition 5.4.8 is to move the negative terms to the right hand side:
           
n n n n n n
+ + + ··· = + + + ···
0 2 4 1 3 5

For a set Ω of size n, we can interpret the left hand side as the number of subsets of Ω having even size, and
the right hand side as the number of subsets of Ω having odd size. Thus, to prove Proposition 5.4.8, we just
need to find a bijection between subsets of odd and even size, as in the following proposition.

Proposition 5.4.9. Let Ω = {1, . . . , n}. Let A (resp. B) be the set of subsets of Ω of even (resp. odd) size.
Then there is a bijection between A and B. In particular, the number of subsets of Ω of even size equals the
number of subsets of odd size.

Since there are 2n subsets of Ω, it follows that the number of subsets of even size is 2n /2 = 2n−1 .

Example 5.4.10. Let Ω = {1, 2, 3}. Its subsets can be paired as follows:

• ∅ ↔ {1, 2, 3}

• {1} ↔ {2, 3}

• {2} ↔ {1, 3}

• {3} ↔ {1, 2}

This example generalizes to a proof in the case that n is odd.

Proof of Proposition 5.4.9 when n is odd. Suppose n is odd. If S ⊂ Ω has size k, then the complement Ω − S
has size n − k. If k is even, then n − k is odd and vice-versa. Each subset has exactly one complement, so
we can pair each subset of even size with exactly one subset of odd size. In conclusion, when n is odd, there
is a bijection between the set A (resp. B) of subsets of Ω of even (resp. odd) size taking S to Ω − S.

The proof above does not work when n is even, so we will set up a different bijection. Here is a motivating
example when n = 4 to see how this bijection works.

86
Example 5.4.11. Let Ω = {1, 2, 3, 4}. Its subsets can be paired as follows:

{1, 2}
A

: {1} {1, 3}
A
{1, 2, 3}
>
z

∅ {2} {1, 4} {1, 2, 4} {1, 2, 3, 4}
A > <

~
{3} {2, 3} {1, 3, 4}
>

~ |
{4} {2, 4} {2, 3, 4}
~
{3, 4}

Proof of Proposition 5.4.9 for n even (and odd). Since Ω is nonempty, pick a distinguished element, say 1 ∈
Ω. Group the subsets of Ω into pairs where the two subsets in each pair differ only in whether they contain
1 or not. Every subset is in exactly one pair. Each pair contains one subset of even size and one subset of
odd size. So the number of subsets of odd size and of even size is the same.

As another example of a bijective proof, we prove the “sticks and stones” theorem (Theorem 3.5.5) again
here, using “M&M’s and toothpicks” instead. The logic we used in Section 3.5 to prove this result is really
a bijection in disguise.

Proposition 5.4.12. Suppose n is a positive integer and k is a non-negative integer such that 0 ≤ k ≤ n.
Then the number of ways to choose a set of size k from a set of size n (with repeats allowed) is
 
n+k−1
.
k

Proof. We first find a set A whose size is the number of ways to choose a set of size k from a set of size n. To
do this, let A be the set of cups filled with k M&M’s,

where there are n different possible colors of M&M’s.
Next, we find a set B whose size is n+k−1k . Let B be the set of binary sequences consisting of k zeroes
and n − 1 ones. Such a sequence has n + k
− 1 entries total and is determined by choosing which k positions
contain the 0’s. So the size of B is n+k−1
k .
We need to find a bijective map between A and B. Fix an ordering of colors, for example, rainbow order
with brown at the end. An element of the set A is a cup of k colored M&M’s. Given such a cup, spill the
M&M’s onto a plate and sort them by color, in a row. Now there are k M&M’s separated into n sections
(some of which may be empty). Put a toothpick between each section, using n − 1 toothpicks total. Finally,
put on color-filter goggles so that all the M&M’s look grey. What we see is a row of k grey M&M’s and
n − 1 toothpicks. Write a sequence of 0’s and 1’s by writing a 0 for each grey M&M and writing a 1 for each
toothpick. This results in a sequence of k zeroes and n − 1 ones, which is an element of the set B.
Now we will construct the inverse map from B to A. Starting from a row of 0’s and 1’s, we can interpret
this as a row of k grey M&M’s separated by n − 1 toothpicks, and then color the M&M’s according to the
specified ordering of colors and put them into a cup. Since these processes reverse each other, the two sets
are in bijection, as desired.

Example 5.4.13. Let’s finish the problem about choosing 4 numbers from {1, . . . , 100}, none of which are
consecutive. Call these numbers a1 , a2 , a3 , a4 , written in increasing order. We want a2 > a1 + 1, a3 > a2 + 1,
and a4 > a3 + 1. The number of ways to do this is the size of the set

A = {(a1 , a2 , a3 , a4 ) | 1 ≤ a1 < a2 − 1 < a3 − 2 < a4 − 3 ≤ 97.}

87
 Let’s define new numbers b1 , b2 , b3 , b4 so that these conditions are easier to state. Let b1 = a1 (the
conditions on a1 are easy, so there is no reason to change a1 ). Let b2 = a2 − 1; the condition a2 > a1 + 1
can be re-expressed as b2 > b1 . Next, let b3 = a3 − 2; the condition a3 > a2 + 1 can be re-expressed as
b3 + 2 > b2 + 2 or, more simply, b3 > b2 . Finally, let b4 = a4 − 3; the condition a4 > a3 + 1 can be re-expressed
as b4 > b3 . Notice that a4 ≤ 100 is equivalent to b4 ≤ 97.
Define a new set B with these conditions:

B = {(b1 , b2 , b3 , b4 ) | 1 ≤ b1 < b2 < b3 < b4 ≤ 97}.

There is a map f : A → B where f (a1 , a2 , a3 , a4 ) = (a1 , a2 − 1, a3 − 2, a4 − 3). It has an inverse map

g : B → A where g(b1 , b
2 , b3 , b4 ) = (b1 , b2 + 1, b3 + 2, b4 + 3). So f is a bijection.
The size of B is 97 4 because choosing an element of B is the same as choosing 4 distinct numbers in
{1, . . . , 97}. So that is the size of A also.

For additional facts about bijections and cardinality, see the following supplemental video.

Click on the icon at left or the URL below for more on bijections and cardinality.
https://www.youtube.com/watch?v=L5s0m5cjq2Q

Exercises
1. Follow the outline of Example 5.4.13 to find the number of ways to choose 5 numbers from {1, . . . , 80}
so that none of them are consecutive?

2. Follow the outline of Example 5.4.13 to find the number of ways to choose 4 numbers from {1, . . . , 100}
so that a1 ≤ a2 ≤ a3 ≤ a4 ≤ a5 .



3. Prove that nk = n−k n
using a bijection.

4. Let S be the set of binary sequences of length n. Explain why |S| = 2n . Find a bijection between S
and subsets of {1, 2, . . . , n}. Conclude that 2n is also the number of subsets of {1, 2, . . . , n}.

5.5 Proof by Contradiction

Here is the strategy for a proof by contradiction: assume the opposite (negation) of what you want to show;
then show that this implies something that is definitely false; this means that the initial assumption must
have been false.

Example 5.5.1. If I want to get on my flight which leaves Denver International Airport at 10 am, then I
need to leave the house before 7 am.

Proof. Assume I leave my house in Fort Collins at 7 (or later). Then I arrive at the parking lot by 8:15. After
taking the shuttle, I arrive at the airport at 8:30. After checking my bag, it is 8:45. After going through
security, it is 9:00. After taking the little train, it is 9:15. Then I run and run to my gate, which is the very
last one, and arrive there at 9:45. This is too late; the doors have closed.

In this section, we give some examples of proofs by contradiction that come from the subject of number
theory.
√
Proposition 5.5.2. The number 3 is irrational (not a fraction).

88
 √ √
Proof. Assume that 3 is rational (a fraction). Then we can write 3 = a/b, where a, b are integers with
b 6= 0. If gcd(a, b) 6= 1, then we can simplify the fraction. So, without loss of generality, we can suppose
that gcd(a, b) = 1. Squaring both sides, we see that 3 = a2 /b2 . So 3b2 = a2 . This means that 3 divides a2 .
Since 3 is prime, this shows that 3 divides a (since a has a unique factorization into primes). Write a = 3a1 ,
where a1 is an integer. Substituting, we see that 3b2 = (3a1 )2 = 9a21 . So b2 =√3a21 . Then 3 divides b2 , so 3
divides b. But then gcd(a, b) is a multiple of 3. So this is a contradiction. So 3 is not a fraction.
Below is Euclid’s proof that there are infinitely many primes, which he discovered around 300 B.C. The
reason to use a proof by contradiction is because no one has found a formula or algorithm to generate prime
numbers.
Proposition 5.5.3. There are infinitely many prime numbers.
Proof. Assume there are only finitely many primes; let r be theQrnumber of primes and let S = {p1 , . . . , pr }
be the set of all the primes. Consider the number N = 1 + i=1 pi . Then N is divisible by at least one
prime number q (since any positive integer has a unique prime factorization.) But N is not divisible by any
of the primes in S; this is because N ≡ 1 mod pi for each 1 ≤ i ≤ r. So q is a new prime which is not in S.
This contradicts the statement that S is the set of all the primes. So there are infinitely many primes.
• A combinatorial existence proof using contradiction (“Prove that there must have been two people who
shook hands the with the same number of people.”)
Remark 5.5.4. Sometimes proofs by contradiction are so fun that people use them more often than needed.
It is better to use a direct proof when you can.

Exercises
√
1. Prove that 5 is irrational.
√
3
2. Show that 2 is irrational.
Qr
3. Use the quantity N = 3 + (2 i=1 pi )2 to show there are infinitely many primes which are congruent
to 3 modulo 4.

5.6 The Pigeonhole Principle

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=8j1F KDijxc

The pigeonhole principle gives a way to show that certain objects need to be grouped together. This
principle relies on the strategy of a proof by contradiction and it has a lot of applications in combinatorics.
Example 5.6.1. In 2017, the population of Colorado was 5,612,000. Prove that there are at least 6 people
in Colorado with the same number of hairs on their heads. You may assume that each person’s head has
less than 1,000,000 hairs.
Answer. If there were
at most 5 people with 0 hairs,
at most 5 people with 1 hair,
at most 5 people with 2 hairs,
..
.
and at most 5 people with 999,999 hairs,
then the number of people in Colorado would be at most 5,000,000. Since 5, 612, 000 > 5, 000, 000, there
must be at least 6 people in Colorado with the same number of hairs on their heads.

89
Proposition 5.6.2. (Pigeonhole principle Version 1) If we place more than n objects into n boxes, then
some box contains 2 or more objects.

Proof. If each box contained at most 1 object, then the total number of objects would be at most n, which
is a contradiction.

Etymology. A pigeonhole is a place where birds “roost”, i.e., rest or sleep. This principle is often stated with
n + 1 pigeons roosting in n holes.

Remark 5.6.3. The pigeonhole principle does not say anything about how many of the boxes are empty.
It is possible that all of the pigeons are stuffed into the same hole.

Example 5.6.4. If 11 numbers are chosen from 1 to 100, then show that two of them have difference less
than 10.

Answer. Separate the numbers 1 − 100 into 10 “boxes”, namely 1 − 10, 11 − 20, 21 − 30, up to 91 − 100. Since
there are 11 numbers and only 10 boxes, two of the numbers must be in the same box. Then the difference
between those numbers is less than 10.

Example 5.6.5. A target is in the shape of an equilateral triangle, with side-length 2 meters. If we hit the
target with 5 arrows, prove that two arrows strike within 1 meter of each other.

To approach this using the pigeonhole principle, we divide the target into 4 parts, each of which is an
equilateral triangle of side-length 1, as drawn below. We will think of these four smaller triangles as the
“boxes” and the striking points of the arrows as the objects.

By the pigeonhole principle, one of the smaller triangles must contain the striking points of at least two
arrows. The distance between these striking points is at most the length of the side of the smaller triangle.
It follows that the striking points of those two arrows are within 1 meter of each other.

Example 5.6.6. There are n guests at a party, for some integer n ≥ 2. Some pairs of them shake hands;
(no one shakes hands with themself). Prove that at least two guests shook hands with the same number of
people.
Each guest shook between 0 and n − 1 hands. Let’s label the boxes by the numbers 0 through n − 1 and
put a person in box i if they shook hands with exactly i people. At first, it does not look like we can use
the pigeonhole principle because there are n people and n boxes.

90
 However, there is one piece of information we did not use. Either box 0 or box n − 1 is empty. The reason
is that if someone shook 0 hands, then no one shook hands with all n − 1 of the other guests; if someone
shook hands with all n − 1 of the other guests, then no one shook 0 hands. Hence we are assigning n people
to n − 1 boxes (labeled either by 0 through n − 2 or by 1 through n − 1). By the pigeonhole principle, two
people must be assigned to the same box, and hence shook the same number of hands.

Proposition 5.6.7. (Pigeonhole principle Version 2) Suppose k is a non-negative integer. If we place more
than kn objects into n boxes, then some box contains at least k + 1 objects

Remark 5.6.8. To obtain Version 1 from Version 2, substitute the value k = 1.

Proof. If each box contained at most k objects, then the total number of objects would be at most kn, which
is a contradiction.

To solve the problem about Colorado at the beginning of the section, we use Version 2 of the pigeonhole
principle as follows
Answer. Label the boxes by the numbers 1 through n = 1, 000, 000. People are assigned to the ith box if
they have exactly i hairs on their heads. Let k = 5. Since 5, 612, 000 > 5, 000, 000 = (5) · (1, 000, 000) = k · n,
there is a box with at least 6 = k + 1 people by the pigeonhole principle (Version 2). These 6 people all have
the same number of hairs on their heads.

Remark 5.6.9. For pigeonhole principle problems, our recommendation is to:

• Identify k + 1, and hence k.

• Define your n boxes (such that if enough objects land in a single box, then you have accomplished the
goal of the problem!)

• Make sure that the number of objects is bigger than k · n.

We provide some harder examples.

Example 5.6.10. If 17 people are seated in a row of 20 chairs, then prove some consecutive set of 5 chairs
are filled with people.

Brainstorm.

Answer. There are 3 empty chairs. These separate the row into n = 4 “boxes” which we can call the far
left, center left, center right, and far right. Each of these parts contains filled chairs, which are consecutively
filled since there are no other empty chairs. (It is possible for one part to be empty or to contain only one
chair.) Let k = 4, so that k + 1 = 5. Since 17 > 16 = 4 · 4 = k · n, Version 2 of the pigeonhole principle
says that some box contains at least k + 1 = 5 objects, which means there are at least 5 consecutively filled
chairs.

Example 5.6.11. If 17 people are seated in a circle of 20 chairs, then prove some consecutive set of 6 chairs
are filled with people.

Answer. The 3 empty chairs partition the circle into n = 3 “boxes” of consecutively filled chairs. Let
k + 1 = 6, so k = 5. Since 17 > 15 = 5 · 3 = k · n, Version 2 of the pigeonhole principle says that some box
contains at least k + 1 = 6 objects, which are consecutively filled chairs.
One of the most fun applications of the pigeonhole principle is to Ramsey theory. Here we give one
example of this.

91
Example 5.6.12. Consider the complete graph K6 with 6 vertices (every pair of its vertices is connected
by an edge). If you color the edges green and gold, show that there must be either a green triangle or a gold
triangle of edges.

Answer. Pick one vertex v. It has 5 edges connecting it to the 5 other vertices. Let n = 2 (for the 2 colors)
and let k = 2. Since 5 > 2 · 2, at least 3 of these 5 edges have the same color by Version 2 of the pigeonhole
principle. Let’s say that color is green; (if not, we can continue with the same reasoning by swapping the
colors).
These 3 green edges connect v to 3 other vertices, let’s call them w, x, y. If the edge between w and x
is green, then we are done because then v, w, x form a triangle with only green edges. Similarly, if the edge
between w and y is green, then we are done and if the edge between x and y is green then we are done. So
the only remaining case is if the three edges between w, x, y are all gold. But then they form a triangle with
only gold edges and we are done!

Exercises
1. If 5 people receive 11 letters, show that someone gets at least 3.

2. If a drawer contains 16 black and 16 white socks, how many do you need to pick in order to have a
matching pair?

3. The city of Phoenix has 1.5 million people. The number of hairs on a head is at most 250, 000. Show
that at least 7 people have the same number of hairs on their head.

4. More than 6,000 of the students attending a private university were born in one of the 50 US states.
Prove that at least 121 of these students were born in the same state.

5. In a set of 42 positive integers, prove that at least 5 of them have the same last digit.

6. In a set of 112 positive integers, prove that at least 12 of them have the same last digit.

7. Pick any 7 integers x1 , . . . , x7 . Show there are two whose sum or difference is a multiple of 10.

8. If 7 numbers are chosen from 1 to 100, show that two of them have a difference which is less than 17.

9. √
Prove that, among any 10 points inside a unit square, there must exist 3 of them that are all within
2/2 of each other.

10. (a) If 94 people are seated in a row of 100 chairs, then show that some consecutive set of 14 chairs is
filled with people.
(b) If 94 people are seated in a circle of 100 chairs, then show that some consecutive set of 16 chairs
is filled with people.

5.7 Additional problems for Chapter 5

1. Give a combinatorial proof (either by counting in two ways or using a bijection) for each of the following
formulas.

(a) n! = n · (n − 1) · (n − 2) · · · · · 3 · 2 · 1
(b) nk = n · n · · · · · n, the product of k copies of n
n!
(c) n · (n − 1) · (n − 2) · · · · (n − k + 1) = (n−k)!


(d) nk = k!(n−k)!
n!

2. Give a combinatorial proof of the following identities.

92
 Pn

n 2


(a) k=0 k k = n 2n−1
n−1
Pn

n k n−k
(b) k=0 k s t = (s + t)n (Here s, t, and n are positive integers.)
3. Give a combinatorial proof that the following identity holds for all positive integers n:
     
n n n
20 + 21 + ... + 2n = 3n .
0 1 n

4. If 30 numbers are chosen from 1 to 82, then show that there is a group of at least 5 of them such that
the difference between any two numbers in this group is less than 12.
5. Suppose that 64 chairs are arranged in a circle. Prove that if 57 people sit in this circle of 64 chairs,
then there are at least 9 consecutively filled chairs.
6. Suppose that 80 chairs are arranged in a row. Prove that if 75 people sit in this row of 80 chairs, then
there are at least 13 consecutively filled chairs.

7. An office needs to connect 15 computers to 10 printers, with the minimal number of cables. At any
moment, 10 computers might need printers.
(a) Show that r = 5 connections might not be enough for certain print jobs.
(b) Show that r = 6 connections is enough for every print job.

8. (a) Look up the statement of the Well-Ordering Principle. Explain how it is similar to and different
from induction.
(b) Prove the Well-Ordering principle using only strong induction and the axiom that 0 is smaller
than every positive integer, as follows.
i. Let P (n) be the statement that any subset of the natural numbers containing n has a least
element.
ii. Prove P (0).
iii. Prove that if P (0), P (1), . . . , P (n) are all true, then P (n + 1) is true.
9. Let p be a prime. Fermat’s Little Theorem states that ap ≡ a mod p, for any positive integer a. For
example, when p = 7: then 7 divides 17 − 1; 7 divides 27 − 2; 7 divides 37 − 3; etc.

(a) For any a ≥ 2, generalize Exercise 12b from Section 4.2 to show (x1 + · · · + xa )p ≡ xp1 + · · · +
xpa mod p.
(b) Substitute x1 = 1, . . . , xa = 1 to show that ap ≡ a mod p, which gives a proof of Fermat’s Little
Theorem.

10. In this problem, we will prove Fermat’s Little Theorem using a combinatorial argument.
(a) Let S be the set of sequences of length p whose entries are in the alphabet {1, . . . , a} such that
not all the entries of the sequence are the same. Explain why |S| = ap − a.
(b) We say that two sequences in S are equivalent if the entries of one can be rotated to be the entries
of the other. For example, when p = 5, then (1, 2, 3, 4, 5) is equivalent to (2, 3, 4, 5, 1). Explain
why each equivalence class in S has size p. Explain why it is important for p to be prime for this
to be true.
(c) Use the division principle to show that p divides ap − a.

93
5.8 Investigation: Pascal’s triangle modulo 2 and Sierpinski’s tri-
angle
Which elements of Pascal’s

triangle are even, and which are odd? In other words, what is the value of the
binomial coefficient nk mod 2? (See Section 2.9.2 for modular arithmetic.)

n = 0: 1
n = 1: 1 1
n = 2: 1 2 1
n = 3: 1 3 3 1
n = 4: 1 4 6 4 1
n = 5: 1 5 10 10 5 1
n = 6: 1 6 15 20 15 6 1

If an integer a is even, then the remainder when dividing a by 2 is 0, so a ≡ 0 mod 2; if a is odd, then
the remainder when dividing a by 2 is 1, so a ≡ 1 mod 2. We can therefore write down “Pascal’s triangle
mod 2” by replacing all even numbers with 0 and all odd numbers with 1:

n = 0: 1
n = 1: 1 1
n = 2: 1 0 1
n = 3: 1 1 1 1
n = 4: 1 0 0 0 1
n = 5: 1 1 0 0 1 1
n = 6: 1 0 1 0 1 0 1

Question 5.8.1. Continue writing down Pascal’s triangle mod 2 by copying the above table and then writing
the next 5 rows of 0’s and 1’s. Find three patterns for the arrangement of 0’s and 1’s.
Question 5.8.2. How can you determine the next row of Pascal’s triangle mod 2 from the previous, without
computing the original Pascal’s triangle? Explain your answer.
Question 5.8.3. After the row for n = 3, when is the next row that only contains 1’s? What does the row
after that one look like?
Proposition 5.8.4. If n = 2m is a power of 2, then the numbers in row n of Pascal’s triangle mod 2 are
all 0 except for the first and last entry.
The next three questions outline three different ways of proving the above proposition.
m

Question 5.8.5. Prove Proposition 5.8.4 with this method: Write 2k as (2m )!/(k!(2m − k)!). We want
to show there are more 2’s in the prime factorization of the numerator than in the denominator. Do this by
finding a way to pair up each factor of 2 in the denominator with a factor of 2 in the numerator. Show that
there are always more factors of 2 in the numerator.
Question 5.8.6. (An algebraic proof.) Prove Proposition 5.8.4 with this method:
m m m
1. Explain why it is enough to show that (x + y)2 ≡ x2 + y 2 mod 2. (Here we reduce the coefficients
of the polynomial mod 2 but not the exponents).

94
 2. Show that (x + y)2 ≡ x2 + y 2 mod 2.
m−1
3. Use induction on m to prove statement 1, where the inductive hypothesis is that (x + y)2 ≡
m−1 m−1
x2 + y2 mod 2. For the inductive step, square both sides of the equation.
Question 5.8.7. (A bijective/combinatorial proof.) Prove Proposition 5.8.4 with this method.
m

1. Explain why 2k is the number of ways to make a subset of S of size k. (For example, if m = 2 and


k = 2, then the number of subsets listed in (2.1) is 42 = 6.)
2. Because of part (1), we want to show the number of subsets of S of size k is even. Use part (3) of the
previous problem to explain why this is true if k is odd.
3. To handle the case when k is even, we will use induction on m; the inductive hypothesis is that the
m−1

statement is true for m − 1, in other words, that 2 k is even for all 0 < k < 2m−1 . Write down what
the base case is and verify that it is true.
4. If T is a subset, recall the definition of the complement

subset T c from Problem 11 in Section 2.10. For
2m
the inductive step, we want to show that k is even for all 0 < k < 2m . Suppose k is even (because
we are already done with the case that k is odd). Show that the number of subsets T such that T 6= T c
is even. Show that the number of subsets T such that T = T c is even using the inductive hypothesis
and part (4) of the previous problem.
Now we investigate a pattern in Pascal’s triangle modulo 2. Look at the numbers in the rows n = 0 to
n = 3. Compare them with the numbers on the left side (then right side) of rows n = 4 to n = 7. What do
you notice?
Question 5.8.8. 1. If you have a green ball in the n = 4 and k = 0 spot in Pascal’s triangle, and a gold
ball in the n = 4 and k = 4 of Pascal’s triangle, which is the closest spot of Pascal’s triangle that they
could both land?
2. Repeat the previous part when the green ball is at the n = 2m and k = 0 spot and the gold ball is at
the n = 2m and k = n spot.
3. Explain why there is an upside down triangle of zeros in the middle of the rows from n = 2m to
n = 2m+1 − 1.
Question 5.8.9. In Pascal’s triangle modulo 2, explain why the rows n = 0 to n = 2m − 1 repeat again on
the left hand and right hand sides of the rows n = 2m to n = 2m+1 − 1.
The next question is a more precise version of the previous two.
Question 5.8.10. 1. Convince yourself that the following claim produces the upside down triangle of
zeros described in Question 5.8.8: if n = 2m + `, where 0 ≤ ` ≤ 2m − 2, the kth entry of Pascal’s
triangle in the nth row is zero modulo 2 for ` + 1 ≤ k ≤ n − (` + 1).
2. Convince yourself that the following claim produces the self-similar pattern described in Question 5.8.9:
if 0 ≤ n ≤ 2m − 1 and 0 ≤ k ≤ n, then
   
n n + 2m
≡ (mod 2)
k k
and    
n n + 2m
≡ (mod 2).
k k + 2m
3. Show that (reducing coefficients but not exponents modulo 2)
m m m
(x + y)n+2 ≡ (x2 + y 2 ) · (x + y)n mod 2.

4. Use part (3) to prove the claims in parts (1) and (2).

95
Sierpinski’s triangle
Sierpinski’s triangle is a famous example of a fractal, which is a shape that exhibits self-similarity. In
particular, Sierpinski’s triangle is the unique 2-dimensional pattern that is made up of three shrunken copies
of itself, each half of the total size, and arranged in a triangle. Here is a sketch of a finite approximation of
Sierpinski’s triangle:

One can obtain this drawing by starting with the smallest triangle at the top, then copying it to its lower left
and lower right, and then repeating this process on the resulting shape several times. Sierpinski’s triangle
itself is actually the “limit” of this diagram as the number of operations go to infinity.
Notice that the pattern of Sierpinski’s triangle, formed by copying a triangle pattern down to the left
and down to the right at each step, is exactly the same as the pattern we discovered about Pascal’s triangle
mod 2 in Question 5.8.9. Indeed, the 1’s in Pascal’s triangle mod 2 outline a Sierpinski triangle shape!
Demonstrate this for the first 16 rows as follows:
Question 5.8.11. Use Sage to print out the first 16 rows of Pascal’s triangle mod 2. (It’s ok if it prints
as a sequence of lists which is left justified, rather than as an equilateral triangle.) Draw an approximation
of Sierpinski’s triangle on top of this printout by connecting the 1’s with lines going in the three triangular
directions.
There is another way to construct Sierpinski’s triangle. Let’s start with an equilateral triangle, colored
black. We divide it into 4 smaller equilateral triangles and change the color of the middle one to white.
Suppose the sides of the original triangle have length 1; then the sides of the smaller triangles have length
1/2. We then repeat this process for the 3 smaller black triangles. Repeating this process indefinitely
produces the Sierpinski triangle.

96
Chapter 6

Recurrence relations

Many interesting sequences of integers are defined using recurrence relations. In this chapter, we explore
fun examples of recurrence relations, including: the Fibonacci sequence, the tower of Hanoi, regions of the
plane, derangements, and Catalan numbers. For the most part, we avoid treating the subject systematically,
but in Section 6.2, we describe linear recurrence relations and their characteristic polynomials in general.
Definition 6.0.1. A recursive relation for an infinite sequence a1 , a2 , . . . is a formula for an in terms of the
previous values in the sequence.

6.1 Fibonacci numbers

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=VindfwTtkSA

The Fibonacci numbers form a well-known sequence of numbers

(1, 1, 2, 3, 5, 8, 13, 21, 34, 55, . . .).
They were first published in a book by Leonardo of Pisa in 1202 concerning the following question about
rabbits:
Question 6.1.1. Suppose that rabbits mature in one month, that an adult pair of rabbits produces exactly
one baby pair of rabbits each month, and that no rabbits ever die. Starting with one pair of baby rabbits in
month 1, how many pairs of rabbits will there be in month 17?
Answer. We compute the following table of the number Fn of pairs of rabbits in month n, starting with 1
rabbit.
n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Fn 1 1 2 3 5 8 13 21 34 55 89 144 233 377 610 987 1597
Definition 6.1.2. Set F1 = 1 and F2 = 1. For n ≥ 3, the n-th Fibonacci number Fn is defined by the
formula Fn = Fn−1 + Fn−2 .
While the question about rabbits is completely unrealistic, Fibonacci numbers show up in nature in many
unexpected places, including leaves on stems, pine cones, artichokes, pineapples, and family trees of bees1 .
There are also applications of Fibonacci numbers to many other scientific fields, including economics, logic,
optics, and pseudo random number generators.
Here is another concrete question whose answer involves the Fibonacci numbers.
1 Ask CSU math professor Dr. Patrick Shipman about some of his work on the Fibonacci numbers!

97
Question 6.1.3. A staircase has n steps. How many ways can you go up the stairs if you can take one or
two steps at a time and never go down?
For example, when n = 4, the number of ways is 5 and the different ways to go up can be represented by
(1, 1, 1, 1), (1, 1, 2), (1, 2, 1), (2, 1, 1), or (2, 2).
Let Sn be the number of ways to go up n steps in this way. We compute the following table.

n = # stairs 1 2 3 4 5
1 1,1 1,1,1 1,1,1,1 1,1,1,1,1
2 1,2 1,1,2 1,1,1,2
2,1 1,2,1 1,1,2,1
sequence of number of steps taken 2,1,1 1,2,1,1
2,2 1,2,2
2,1,1,1
2,1,2
2,2,1
Sn = # ways to climb 1 2 3 5 8

Notice, in order to climb 4 steps, you must first climb 2 steps at a time and then 2 more steps, or 1 step
and then 3 more steps as indicated in blue and red, respectively. Based on this example, we conjecture that
S4 = S3 + S2 . We will prove more generally that the stair climbing problem is in fact a recurrence relation
in Lemma 6.1.5, but for now let’s take a deeper look into the example of 4 stairs.
Example 6.1.4. Picture of S4 = S3 + S2 .

After the red step of size 1, there are S3 = 3 ways to proceed: 1, 1, 1 or 1, 2 or 2, 1.

After the blue step of size 2, there are S2 = 2 ways to proceed: 1, 1 or 2.
In total there are S4 = S3 + S2 ways:

1, 1, 1, 1 1, 1, 2 1, 2, 1 2, 1, 1 2, 2
| {z } | {z }
from S3 from S2

The values of Sn are all Fibonacci numbers, but they are shifted to the left compared to the values of Fn
in the table for Question 6.1.1. For example, S5 = F6 = 8.
Based on this, we write down the following lemma. We prove the lemma using strong induction as
described in Remark 5.2.11.
Lemma 6.1.5. The relationship between the number of ways to climb the stairs and the Fibonacci numbers
is that Sn = Fn+1 for all n ≥ 1.
Proof. In order for this to be true, we need to see whether the values of Sn satisfy the same recurrence
relation as the values of Fn . We claim that Sn = Sn−1 + Sn−2 for n ≥ 3.
If the claim is true, then we can use strong induction to show that Sn = Fn+1 for all n ≥ 1. Here is how
that works. The base cases n = 1 and n = 2 are true because S1 = 1 = F2 and S2 = 2 = F3 . Suppose that

98
n ≥ 3 and Si = Fi+1 for all 1 ≤ i ≤ n − 1. Using the claim, the inductive hypothesis, and the formula for
Fn+1 shows that
Sn = Sn−1 + Sn−2 = Fn + Fn−1 = Fn+1 .
So we need to check that Sn = Sn−1 + Sn−2 for n ≥ 3. Every sequence of steps up the n stairs starts
with either
• a size 1 step, after which there are Sn−1 ways to climb the remaining n − 1 stairs; or
• a size 2 step, after which there are Sn−2 ways to climb the remaining n − 2 stairs.
By the addition principle for sets, the number of sequences of steps up the n stairs is Sn−1 + Sn−2 , verifying
the claim.
Another example of the Fibonacci numbers occurring in graph theory can be found in Example 8.9.3.

6.1.1 The golden ratio and a formula for the Fibonacci numbers
One drawback with recurrence relations is that it is time-consuming to compute the values in the sequence.
For example, it would be nice to find the value of F100 , without needing to compute the value of Fn for
1 ≤ n ≤ 99. In this section, we describe a closed
√
form formula for the Fibonacci
√
numbers.
The golden ratio is the number α = 1+2 5 . Its conjugate is ᾱ = 1−2 5 . Note that α ≈ 1.61803 . . . and
ᾱ ≈ −0.61803 . . .. Using the quadratic formula, we see that α and ᾱ are the roots of the quadratic polynomial
f (x) = x2 − x − 1.
The golden ratio is an important number for the ratio of length to height in Greek architecture and
natural objects like sea shells.
Theorem 6.1.6. For all n ≥ 0, the Fibonacci number Fn is given by the formula
√ !n √ !n !
1 1+ 5 1− 5
Fn = √ − . (6.1)
5 2 2

Remark 6.1.7. This is a surprising statement! Initially, the right-hand side of (6.1) does not even look like
an integer! Also, it is not clear why there is a connection between the Fibonacci numbers and the golden
ratio.
Remark 6.1.8. Because |ᾱ| < 1, the powers of ᾱ approach 0. As a consequence, Fn ≈ αn for n large.
We will not prove Theorem 6.1.6 but there is some explanation of why it is true in Section 6.2, see
Example 6.2.14. It is also possible to prove it using induction.

Exercises
1. Consider the sequence defined by a0 = 2, a1 = 1, and for all n ≥ 1, an = an−1 + 2an−2 . Compute the
first several terms of the sequence. What is a5 ?
2. Let Dn be the number of ways that an n-by-2 board can be covered by 2-by-1 dominos, so that no
dominos overlap.
(a) Show that D1 = 1, D2 = 2, and D3 = 3.
(b) Show that Dn = Fn+1 for n ≥ 1.
3. In a subway car, there is a row of n seats. Find the number Sn of ways that the seats can be filled
by an arbitrary number of people, so that no two people sit next to each other. For example, S1 = 2
since the one seat can be empty or full. Letting e denote empty and f denote full, we see that S2 = 3,
because the choices are ee, f e, or ef . Show that Sn = Fn+2 for n ≥ 3.

99
 4. A fence is constructed from n posts. Prove that Fn+2 is equal to the number of ways to paint each
post either silver or navy, in such a way that no two silver posts are next to each other.
5. How many subsets of the set {1, 2, 3, ..., 9} contain no two consecutive integers?
F2 F3 F4 Fn Fn+1
6. What number do the successive quotients F1 , F2 , F3 , ..., Fn−1 , Fn , . . . approach as n gets larger
and larger?
7. Using (strong) induction, prove that Fn ≤ 2n−1 for all n ≥ 1.
8. Here is a table of the sum of the first n Fibonacci numbers.

n 1 2 3 4 5 6 7 8
F1 + F2 + · · · + Fn 1 2 4 7 12 20 33 54

For n ≥ 1, prove that F1 + F2 + · · · + Fn = Fn+2 − 1 by induction on n.

9. Let Fn be the n-th Fibonacci number. Show that for all n ≥ 1 we have

F1 + F3 + F5 + · · · + F2n−1 = F2n .

10. Let Fn be the n-th Fibonacci number. Prove that F12 + · · · + Fn2 = Fn Fn+1 for all n ≥ 1.
11. True or False: Starting with 1, 1, 2, the Fibonacci numbers rotate odd, odd, even, then odd, odd, even,
then odd, odd, even, etc.
12. What happens to the values of the Fibonacci sequence modulo 3?

6.2 Linear recurrence relations

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=8QMJNWQQ0lc

The examples in the last section were all related to the Fibonacci numbers. In this section, we explore
some of the many other recurrence relations. We explain how to implement recurrence relations in SAGE.
We define linear recurrence relations and describe the characteristic polynomial method of solving them.

6.2.1 Another recurrence relation

Here is a problem that introduces a new recurrence relation.
Example 6.2.1. How many ways can you go up a staircase with 9 steps if you can take 1, 2, or 3 steps at
a time?
Brainstorm.
# stairs 1 2 3 4
1 1,1 1,1,1 1,1,1,1
2 1,2 1,1,2
2,1 1,2,1
sequence of number of steps taken 3 1,3
2,1,1
2,2
3,1
# ways 1 2 4 7

100
Answer. Let Tn = # ways to climb a staircase with n steps. We claim that Tn = Tn−1 + Tn−2 + Tn−3 . The
reason is that every sequence of steps up the n stairs starts with either

• a size 1 step, after which there are Tn−1 ways to climb the remaining n − 1 stairs, or

• a size 2 step, after which there are Tn−2 ways to climb the remaining n − 2 stairs, or

• a size 3 step, after which there are Tn−3 ways to climb the remaining n − 3 stairs.

Using this recurrence relation and the base cases found by hand, we compute:
T1 T2 T3 T4 T5 T6 T7 T8 T9
1 2 4 7 13 24 44 81 149

Hence there are 149 ways to climb this staircase with 9 stairs.

Question 6.2.2. Find T10 , the number of ways to climb 10 stairs with these rules.

6.2.2 SAGE commands for recurrence relations

It is tedious to use a recurrence relation to compute numerous values in a sequence. Now we will see how to
use SAGE to do this work efficiently.

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/IstO1arARxI

Let’s first set-up the recursive definition of the sequence Tn . (Note that the indentation of these commands
is quite important in SAGE and may not be accurately reflected here.)
Sage commands:

def T(x):
if x ==1:
return 1;
if x ==2:
return 2;
if x==3:
return 4
else:
return T(x-1)+T(x-2)+T(x-3)

Now we can print the first several values of the sequence Tn .

[print(T(n)) for n in [1..10]]

Example 6.2.3. This is the recursive definition of the fibonacci sequence:

def fibo(x):
if x ==1:
return 1;
if x ==2:
return 1;
else:
return fibo(x-1)+fibo(x-2)

Use SAGE to print the first 18 Fibonacci numbers.

101
Example 6.2.4. A colony of wasps is building a nest in your roof. Let waspn be the number of worker
wasps (other than the queen) after n days. Suppose wasp1 = 1 and wasp2 = 1. For n ≥ 3, suppose that the
number of wasps increases by this recurrence relation:

waspn = waspn−1 + 20 · waspn−2 .

Question 6.2.5. Set-up the wasp recurrence relation in SAGE and use SAGE to find the number of wasps
on days 1-10.

Question 6.2.6. What happens if you try to find the 100th entry of any of the sequences above? Try to
compute T (100), fibo(100) or wasp(100) using SAGE.

Exercises
1. Let a0 , a1 , a2 , . . . be the sequence defined by a0 = 4,a1 = 5, and an = an−1 + 2an−2 . Then using the
roots of the characteristic polynomial, we know that an has the explicit formula b · 2n + c · (−1)n for
some values of b and c. Find b.

2. Consider the sequence defined by L0 = 0, L1 = 1, and Ln = 3Ln−1 − Ln−2 . Implement this recursion
in Sage in order to compute L20 .

6.2.3 Solving linear recurrence relations of degree 2

Suppose an is a sequence defined from a recurrence relation and some initial values. The computer is very
fast at computing the first several entries of an , but gets very slow at computing entries later in the sequence.
In this section, we explain how to find a closed-form formula for an when the recurrence relation is linear of
degree 2. In the next section, we cover similar material for linear recurrence relations of higher degree.

Definition 6.2.7. An linear recurrence relation of order 2 is an equation of the form

an = c1 an−1 + c2 an−2 . (6.2)

Here the values c1 and c2 are constants and c2 6= 0. The initial values of a linear recurrence relation of order
2 are a choice of the numbers a1 and a2 .

For example, the Fibonacci recurrence relation is linear of order 2 with constants c1 = c2 = 1, and initial
values a1 = a2 = 1. The wasp recurrence relation is linear of order 2 with constants c1 = 1 and c2 = 20 and
initial values a1 = a2 = 1.

Definition 6.2.8. The characteristic polynomial of (6.2) is the degree 2 polynomial

c(x) = x2 − c1 x − c2 .

By the quadratic formula, the roots of c(x) are

p
c1 ± c21 + 4c2
.
2
We call these roots r1 and r2 .

Theorem 6.2.9. Let r1 and r2 be the roots of the characteristic polynomial. Suppose that r1 6= r2 . Then
there are constants z1 and z2 such that, if n ≥ 1, then

an = z1 r1n + z2 r2n . (6.3)

Furthermore, there is only one choice of z1 and z2 such that (6.5) is true for all n.

102
 In general, the way to find z1 and z2 is to use (6.3) when n = 1 and n = 2. We solve for z1 and z2 , using
that we know a1 , a2 , r1 , r2 using these equations:

a1 = z1 r11 + z2 r21 and a2 = z1 r12 + z2 r22 .

2
For example, the characteristic polynomial
√ √ relation is c(x) = x − x − 1. Its
for the Fibonacci recurrence
roots are the golden ratio α = (1 + 5)/2 and its conjugate ᾱ = (1 − 5)/2.
We explain how to find the roots and the numbers z1 and z2 in SAGE. First, we define a ring of
polynomials, define the characteristic polynomial c(x) and use SAGE to find its roots.

R = PolynomialRing(RR, ’x’)
c=x^2-x-1
c.roots();

The output looks a little complicated. This notation tells us what the roots are and that each root
happens exactly once.

[(-1/2*sqrt(5) + 1/2, 1), (1/2*sqrt(5) + 1/2, 1)]

Let’s extract the roots from this complicated notation (switching their order to make things look better
at the end). Remembering that lists are indexed starting at 0, for the root r2 we take the 0th entry and
then the 0th entry of that; for the root r1 we take the 1st entry and then the 0th entry of that.

r2=c.roots()[0][0];
r1=c.roots()[1][0];

Now we need to find the constants z1 and z2 . To do this, we define them as variables and use (6.3) when
n = 1 and n = 2.

var(’z1,z2’)
eqnfibo = [z1*r1 + z2*r2==1, z1*r1^2 + z2*r2^2==1];
Efibo = solve(eqnfibo, z1,z2); Efibo
√ √ √ √
This shows us that z1 = (1/5) 5 = 1/ 5 and z2 = −(1/5) 5 = −1/ 5. We then define the closed form
formula, use it to check the case when n = 3 and then find the 100th Fibonacci number.

s=sqrt(5)
def Fibo(n): return expand((1/s)*(r1^n - r2^n));
Fibo(3);
Fibo(100);

Question 6.2.10. Find the characteristic polynomial of the wasp recurrence relation from Example 6.2.4.
Use SAGE to find its roots and then the constants z1 and z2 . Check your work by computing the third wasp
number. Find the 100th wasp number.

6.2.4 Linear recurrence relations of higher order

More generally, we can set up a recurrence relation as follows. Fix a positive integer d.

Definition 6.2.11. An linear recurrence relation of order d is an equation of the form

an = c1 an−1 + c2 an−2 + · · · + cd an−d . (6.4)

Here the values c1 , . . . , cd are constants and cd 6= 0. The initial values of a linear recurrence relation of order
d are a choice of the numbers a1 , . . . , ad−1 .

103
 Sometimes this is called an order d homogeneous linear recurrence with constant coefficients; what a
mouthful!
The stair climbing problem in this section is a linear recurrence relation of order d = 3 with constants
c1 = c2 = c3 = 1, and initial values a1 = 1, a2 = 2, a3 = 4.
Given a linear recurrence relation of order d and initial values a1 , . . . , ad−1 , we briefly describe how to
find a closed form formula for the values an for large n.

Definition 6.2.12. The characteristic polynomial of (6.4) is the degree d polynomial

c(x) = xd − c1 xd−1 − · · · − cd−1 x − cd .

Theorem 6.2.13. Let r1 , . . . , rd be the roots of the characteristic polynomial. Suppose that the d roots are
all different. Then there are constants z1 , . . . , zd such that, if n ≥ 1, then

an = z1 r1n + · · · zd rdn . (6.5)

Furthermore, there is only one choice of constants z1 , . . . , zn such that (6.5) is true for all n.

We won’t see the proof of this theorem until Chapter 7. Here is how to use the theorem to find an explicit
formula for the Fibonacci numbers without using SAGE.

Example 6.2.14. The characteristic polynomial √ of the Fibonacci recurrence

√ relation is c(x) = x2 −x−1. By
the quadratic formula, this has roots r1 = (1 + 5)/2 and ᾱ = (1 − 5)/2. So an = z1 α + z2 ᾱ. Substituting
n = 1 and n = 2, we obtain the following system of equations for z1 and z2 :
√ √
1+ 5 1− 5
1 = z1 + z2
2 2
√ !2 √ !2
1+ 5 1− 5
1 = z1 + z2
2 2
√ √
A long computation shows that z1 = 1/ 5 and z2 = −1/ 5. This shows where Theorem 6.1.6 comes from!

Exercises
1. Define the Lucas numbers by L0 = 2, L1 = 1, and Ln = Ln−1 + Ln−2 for n ≥ 2. Find L10 .

2. How many ways are there to climb a staircase with 9 steps if you can take either one or three steps at
a time?

(a) Find how many ways to climb n = 1, 2, 3, 4 steps if you can take either one or three stairs at a
time.
(b) Find a recurrence relation for this problem and explain why it is true.
(c) Compute what happens for n = 9 steps using the recurrence relation.

3. In how many ways can you cover a 3 × 10 grid of squares with identical dominoes, where each domino
is of size 3 × 1 (or 1 × 3 if you turn it sideways), and you must use exactly 10 dominoes?
Remark: Each domino covers 3 squares, and so the 10 dominoes cover the 30 squares exactly.

4. You have $9 dollars to spend. Each day at lunch you buy exactly one item: either an apple for $1, or a
yogurt for $2, or a sandwich for $4. You continue buying one item each day until you have exactly $0.
This could take anywhere from 9 days, if you buy 9 apples in a row, to 3 days, if you buy 2 sandwiches
in a row then an apple. Buying two sandwiches then an apple is different from buying an apple then
two sandwiches. In how many different ways can you spend your money?

104
 5. You have n dollars to spend. Each day you buy exactly one item: either a banana for $1 or a bagel
for $4, and you continue until you have exactly $0. Let Sn be the number of ways you can spend your
n dollars (buying a banana the first day and a bagel the second is different from a bagel the first day
and a banana the second). Find and explain a recurrence relation for Sn .
6. Solve the linear recurrence relation of order 2 given by an = 5an−1 − 6an−2 if a1 = 5 and a2 = 13.

6.3 The tower of Hanoi

The tower of Hanoi is a game with 3 pegs. On the left hand peg, there are disks of different sizes (each with
a hole in its center) stacked in order of size from largest on the bottom to smallest on the top. On each
move, you can transfer one of the disks to another peg, without ever putting a larger disk on top of a smaller
one. The object of the game is to move all the disks to another peg.

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=2HJqtgiD6Y8

Question 6.3.1. Can the tower of Hanoi be represented as a recurrence relation?

Brainstorm. Let Hn be the minimal number of moves needed to finish the game for n disks.
H1 = 1: if there is one disk, just move it to another peg.
H2 = 3: first move the small disk to the middle peg, then move the big disk to the right peg, then put
the small disk on top of the big disk on the right peg.
In the exercises, we will show that H3 = 7 and H4 = 15.
Lemma 6.3.2. The recurrence relation for the tower of Hanoi is H1 = 1 and Hn = 2Hn−1 + 1 for n ≥ 2.
Proof. By definition, it takes Hn−1 moves to transfer all the disks except the biggest one to another peg.
Then it takes 1 move to move the biggest disk to the empty peg. Then, by definition, it takes Hn−1 moves
to transfer all the other disks on top of the biggest one.
Lemma 6.3.3. A closed form formula for the minimal number of moves in the tower of Hanoi game is
Hn = 2n − 1.
Proof. The formula is true in the base case n = 1. The inductive hypothesis is that Hn−1 = 2n−1 − 1. By
Lemma 6.3.2, Hn = 2Hn−1 + 1. So Hn = 2(2n−1 − 1) + 1 = 2n − 2 + 1 = 2n − 1. So the result is true by
induction.

Exercises
1. How many steps does it take to solve the Tower of Hanoi problem with 5 disks?
2. True or false: the recursion for the Tower of Hanoi problem is a homogeneous linear recurrence that
can be solved with the methods of section 6.2.
3. When n = 3, label the disks d1 (smallest) to d3 (largest). Write out a sequence of 7 moves to show
that H3 = 7.
4. When n = 4, label the disks d1 (smallest) to d4 (largest). Write out a sequence of 15 moves to show
that H4 = 15.
5. In the sequences of moves found in problems 1 and 2, which disks does d1 rest on along the way?
6. When n = 5, label the disks d1 (smallest) to d5 (largest). In order to solve the tower of Hanoi in 31
steps, what rules do you need to follow about where the disks d1 and d2 can rest along the way?

105
 7. In order to solve the tower of Hanoi in the minimal number of steps, the moves must be chosen carefully.
Write an algorithm to solve the tower of Hanoi most efficiently.

6.4 Regions of the plane

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=CSUmdpc5dB8

On a piece of paper, draw non-parallel lines so that at most 2 lines intersect at a point. We define Pn to
be the number of regions formed if you draw n such lines.

Example 6.4.1.

(a) P1 = 2 (b) P2 = 4 = P1 + 2 (c) P3 = 7 = P2 + 3 (d) P4 = 11 = P3 + 4

Lemma 6.4.2. The recurrence relation for the number Pn of regions formed by drawing n non-parallel lines
so that at most two lines intersect in a point is P0 = 1 and Pn = Pn−1 + n for n ≥ 1.

Proof. If we have no lines, there is just one region, the entire plane, so P0 = 1. Then, suppose we start with
n − 1 lines, for which there are Pn−1 regions. The n-th line crosses n − 1 other lines, and hence divides n of
these regions in two. This shows that Pn = Pn−1 + n.

Example 6.4.3. Using the recurrence relation, we can compute more values of Pn .

P0 P1 P2 P3 P4 P5 P6 P7 P8 P9 P10
1 2 4 7 11 16 22 29 37 46 56

n2 +n+2
Proposition 6.4.4. A closed form formula for the number of regions is Pn = 2 for all n ≥ 1.

Proof. One way to prove this is by induction.

12 + 1 + 2
Base case. Note P1 = 2 = .
2
n2 + n + 2 (n + 1)2 + (n + 1) + 2 n2 + 3n + 4
Inductive step. Assume Pn = . To prove: Pn+1 = = .
2 2 2

106
 Note

Pn+1 = Pn + (n + 1) by the recurrence relation

2
n +n+2
= + (n + 1) by the inductive assumption
2
n2 + n + 2 2n + 2
= +
2 2
n2 + 3n + 4
= .
2
Hence we are done by induction.

Exercises
1. What is the largest number of regions that can be cut out using 11 lines in the plane?
2. Let’s draw n circles in the plane. We say that they are in general position if each pair of circles
intersects in exactly two points. Let an be the number of regions of the plane formed by n circles in
general position. For example, a1 = 2 because there are two regions (inside or outside the circle).

(a) Show that a2 = 4, a3 = 8, and a4 = 14.

(b) Explain why the nth circle intersects the other n − 1 circles in 2(n − 1) points.
(c) Label these points as P1 , . . . P2(n−1) so that they are arranged clockwise on the nth circle. Explain
why each of the arcs below separates a region formed by the n − 1 circles into 2 regions:

P1 P2 , P2 P3 , . . . , P2(n−1) P1 .

(d) Show that an = an−1 + 2(n − 1).

(e) Use induction to show that an = n2 − n + 2.

6.5 Derangements

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/zv0G3nCpkes

Example 6.5.1. How many ways are there to rearrange the letters in MAT, so that no letter is in its initial
position? The answer is only 2 (out of the 3! = 6 permutations): namely ATM and TMA.
Example 6.5.2. How many ways are there to rearrange the letters in MATH, so that no letter is in its
initial position? The answer is only 9 (out of the 4! = 24 permutations): namely:
AMHT, ATHM, AHMT, TMHA, THMA, THAM, HMAT, HTMA, HTAM.
It is possibly to check Examples 6.5.1 and 6.5.2 using the inclusion-exclusion principle, but it is not so
easy.
Definition 6.5.3. A derangement of a sequence a1 , a2 , . . . , an is a permutation such that no element ai
appears in its initial position (the ith spot). Let Dn be the number of derangements of a sequence of n
distinct letters.
We can compute that D1 = 0, D2 = 1, D3 = 2, and D4 = 9. To compute Dn when n is larger, it is
helpful to have a recurrence relation.

107
Proposition 6.5.4. For n ≥ 3, the number Dn of derangements of n objects satisfies the recurrence relation

Dn = (n − 1)(Dn−1 + Dn−2 ). (6.6)

Proof. Let’s work with derangements of the sequence 1, 2, 3, . . . , n. Given a derangement, let c be the number
of the spot where 1 appears. Since c 6= 1 (else this would not be a valid derangement), there are n − 1 choices
for c. Once c is fixed, we have two cases.
Case 1: the numbers 1 and c switched spots. For example, when n = 5 and c = 3, the derangement
34152 is in this case. Ignoring the first and the cth spot, the other n − 2 numbers are a derangement of their
initial positions, and there are Dn−2 ways to make that derangement.
Case 2: the numbers 1 and c did not switch spots. For example, when n = 5 and c = 3, the derangement
23154 is in this case. We are left with n − 1 numbers (2, 3, . . . , n) that need to fit into n − 1 spots (1, 2, . . . , n,
excluding c). Each number is restricted from exactly one spot: for the number c, this is because it cannot
be in the first spot in order to be in Case 2; for a number j other than c, it cannot be in the jth spot. So
the number of options is Dn−1 , the number of derangements of n − 1 objects.
There is no overlap between cases 1 and 2. So, by the addition principle, for a fixed choice of c, the
number of derangements is Dn−1 + Dn−2 . Since there are n − 1 choices for c, the number of derangements
is (n − 1)(Dn−1 + Dn−2 ) by the multiplication principle.

Using the recurrence relation, we can compute some more derangement values.

D1 D2 D3 D4 D5 D6 D7 D8 D9 D10
0 1 2 9 44 265 1,854 14,833 133,496 1,334,961

Here are two more formulas for derangements; one is recursive and the other is explicit. We will see
proofs of these facts in the exercises below.

Proposition 6.5.5. For n ≥ 2, the number Dn of derangements of n objects satisfies this other recurrence
relation
Dn = nDn−1 + (−1)n (6.7)

Theorem 6.5.6. Here is a closed form formula for the number of derangements:
 
1 1 1 (−1)n
Dn = n! 1 − + − + · · · + .
1! 2! 3! n!

Exercises
1. How many permutations of the word MONDAY have at least one letter in the correct spot?

2. What is the distance between D5 (the number of derangements of 5 objects) and 5!/e on the number
line, to the nearest 1/1000 as a decimal?

3. One cool fact about derangements is that Dn is the integer closest to n!/e. Check that this is true for
n = 1, . . . , 5.

4. Use the previous problem to estimate the probability that a permutation of n objects is a derangement
when n is large.

5. Prove Proposition 6.5.5 that Dn = nDn−1 + (−1)n by induction as follows:

(a) Show that (6.7) is true when n = 2.

(b) Write out the inductive hypothesis by substituting n − 1 for n into (6.7).
(c) Use part (2) and (6.6) to prove the inductive step.

108
 6. Prove Theorem 6.5.6 using Proposition 6.5.5.
7. Let Wk be the permutations of 1, . . . , n that fix the number k.
(a) For 1 ≤ k ≤ n, show that |Wk | = (n − 1)!.


(b) Show there are n2 ways to choose two sets Wk and W` with k 6= `; explain which permutations
are in Wk ∩ W` and find |Wk ∩ W` |.
(c) How many ways are there to choose i of the sets W1 , . . . , Wn ; what is the size of the intersection
of these sets?
(d) The inclusion-exclusion principle gives a way to find the size of the union of the sets W1 , . . . , Wn .
Pn n
Use that to find the size of the complement of this union and show that Dn = n! i=0 (−1) i! .

6.6 The Catalan numbers

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=NETfwiSbGEA

The Catalan numbers are one of the most celebrated sequences in combinatorics. They are defined by
the following recursion.
Definition 6.6.1. The Catalan numbers are the sequence of numbers C0 , C1 , C2 , . . . having initial value
C0 = 1 and satisfying the recurrence relation

Cn+1 = C0 Cn + C1 Cn−1 + C2 Cn−2 + · · · + Cn C0 .

Starting with n = 0, we can use the above recursion to compute that the first 10 Catalan numbers are

1, 1, 2, 5, 14, 42, 132, 429, 1430, 4862.

One of the most important combinatorial interpretations of the Catalan numbers is in terms of Dyck
paths.
Definition 6.6.2. A Dyck path of length 2n is a path on the integer lattice grid from (0, 0) to (n, n) that
stay on or above the main diagonal.
For example, here are the Dyck paths of length 6:

We can alternatively represent these paths as Dyck words, which are sequences of n U ’s and n R’s such
that as we read from right to left, the number of U ’s we have read is always at least equal to the number of
R’s we have read. Here are the six corresponding Dyck words of length 6:

UUURRR, UURURR, UURRUR, URUURR, URURUR.

Proposition 6.6.3. The number of Dyck paths of length 2n is equal to the nth Catalan number Cn .

109
Proof. Let Dn be the number of Dyck paths of length n. There is only one Dyck path of length 0, namely
the empty path. So D0 = 1 = C0 , and the initial value is satisfied.
To prove that Dn satisfies the recursion, consider the first time at which a Dyck path of length n + 1
returns to the diagonal after the first step. The height of this first return is some number i between 1 and
n + 1 inclusive. Let Dn+1,i be the number of Dyck paths of length n + 1 whose first return has height i.
Then by the addition principle we have

Dn+1 = Dn+1,1 + Dn+1,2 + · · · + Dn+1,n+1 . (6.8)

Now, to compute Dn+1,i , notice that since the first return to the diagonal is at height i, the first part of the
path (from point (0, 0) to (i, i)) is uniquely determined by drawing a Dyck path from point (0, 1) to (i − 1, i),
which gives Di−1 possibilities for the first part.
(n+1, n+1)

(i-1, i)
(i, i)

(1,0)

(0,0)

The remaining steps in the path simply form a Dyck path of height n + 1 − i, of which there are Dn+1−i .
Therefore, by the multiplication principle, we have

Dn+1,i = Di−1 Dn+1−i .

Thus in particular we have Dn+1,1 = D0 Dn , Dn+1,2 = D1 Dn−1 , and so on. Plugging these formulas into
Equation 6.8 above, we obtain

Dn+1 = D0 Dn + D1 Dn−1 + · · · + Dn D0

as desired. Therefore, Dn satisfies the same recursion as the Catalan numbers, so Dn = Cn for all n.

Theorem 6.6.4. A closed form formula for the nth Catalan number is
     
2n 2n 1 2n
Cn = − = . (6.9)
n n+1 n+1 n

We will not include a proof of Theorem 6.6.4 in this book. It is not hard to check that the two formulas
for Cn in (6.9) are the same.
Catalan numbers do not just count Dyck paths; in fact, there are 214 different known combinatorial
descriptions of the Catalan numbers2 ! Let’s consider some descriptions in the case n = 3.

Example 6.6.5. The fact that C3 = 5 is equivalent to there being 5 ways to:

1. put parentheses around 4 letters

((xy)z)w, (x(yz))w, (xy)(zw), x((yz)w), x(y(zw));

2 All of these descriptions can be found in the book “Catalan Numbers” by Richard Stanley.

110
2. divide a convex pentagon into triangles;

3. tile a stair of height 3 with 3 rectangles;

4. arrange the numbers {1, . . . , 6} in a 2 × 3 grid so that each row and each column is increasing;

1 2 3 1 2 4 1 2 5 1 3 4 1 3 5
, , , ,
4 5 6 3 5 6 3 4 6 2 5 6 2 4 6

5. pair the vertices of a hexagon so that the line segments joining paired vertices do not intersect;

6. draw a rooted binary tree with 4 leaves such that every vertex has either 0 or 2 children (we will see
more about trees in Chapter 9).

7. write down a sequence (a1 , a2 , . . . , a6 ) of 1’s and −1’s such that each partial sum a1 + a2 + · · · + aj is
nonnegative and the entire sum a1 + a2 + · · · + a6 is 0.

(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)

111
Exercises
1. How many lattice paths from (0,0) to (4,4) stay *on or below* the diagonal?
2. How many Dyck paths from (0,0) to (5,5) stay *strictly above* the diagonal, that is, they don’t touch
the diagonal at any point between (0,0) and (5,5)?

3. Suppose candidates Uppity and Rightley are running for class president. 13 of the students in the
class plan on voting; 6 of them will be voting for Uppity and 7 will be voting for Rightley. How many
sequences of ballots (which can be thought of as sequences of 6 U’s and 7 R’s in some order) have the
property that, as they are counted in order, Rightly is never ahead until the very last ballot is counted?

4. When n = 4, then C4 = 14. In these exercises, find the 14 ways to do each of the following.

(a) Draw a Dyck path;

(b) write a Dyck word;
(c) put parentheses around 5 letters xyzwv;
(d) divide a convex hexagon into triangles;
(e) tile a stair of height 4 with rectangles;
(f) arrange the numbers {1, . . . , 8} in a 2 × 4 grid so that each row and each column is increasing;
(g) pair the vertices of an octagon so that the line segments joining paired vertices do not intersect;
(h) draw a rooted binary tree with 5 leaves such that every vertex has either 0 or 2 children.
(i) write down a sequence (a1 , a2 , . . . , a8 ) of 1’s and −1’s such that each partial sum a1 + a2 + · · · + aj
is nonnegative and the entire sum a1 + a2 + · · · + a8 is 0.

112
Chapter 7

Generating Functions

This chapter centers on generating functions, which are a technique for using polynomials and series for
solving problems in combinatorics. Generating functions are the cornerstone of a field of mathematics called
algebraic combinatorics.

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/7Rw7pEMJneI

First we motivate why generating functions are useful by studying some problems about making change.

7.1 Making change

Example 7.1.1. Suppose you have 6 pennies and 2 nickels in your pocket. The pennies look identical and
the nickels look identical. How many different ways can you give some of the coins to your friend?
It is clear that the most you can give is 16 cents and the least you can give is 0 cents. There are two ways
that you can give 6 cents, either 1 penny and 1 nickel or 6 pennies. To be more systematic, the number of
cents you give will be
0, 1, 2, 3, 4, 5, or 6 in pennies plus 0, 5, or 10 in nickels.
We want to add the value of pennies to the value of nickels. The key insight is that we add exponents in a
product of polynomials. So we use the exponents above in two polynomials and then multiply them.

(x0 + x + x2 + x3 + x4 + x5 + x6 ) · (x0 + x5 + x10 ).

We expand this to get:

x0 + x + x2 + x3 + x4 + 2x5 + 2x6 + x7 + x8 + x9 + 2x10 + 2x11 + x12 + x13 + x14 + x15 + x16 .

The fact that the largest value you can give is 16 is represented by the term x16 = x6 · x10 , for 6 cents
in pennies and 10 cents in nickels. The fact that the smallest value you can give is 0 is represented by the
term x0 = x0 · x0 for 0 cents in pennies and 0 cents in nickels. The fact that there are two ways that you
can give 6 cents is represented by the term 2x6 = x6 · x0 + x1 · x5 , for either 6 cents in pennies and 0 cents
in nickels or 1 cent in pennies and 5 cents in nickels.
More generally, this polynomial encodes the number of ways to give any number of cents using 6 pennies
and 2 nickels! For example, the term 2x11 = x6 · x5 + x1 · x10 tells us that there are two ways to give 11
cents, which makes sense since you can give 6 cents in pennies and 5 cents in nickels or 1 cent in pennies and
10 cents in nickels. Just by reading the coefficients above, we see that there are 2 ways of giving 10 cents
but only one way of giving 7 cents. It even tells us that there are no ways of giving 17 cents!

113
 Let’s now try a somewhat more complicated example.

Example 7.1.2. Suppose you have 6 pennies ($0.01 each), 2 nickels ($0.05 each), 4 dimes ($0.10 each), and
3 quarters ($0.25 each) in your wallet. You can’t tell the difference between any two coins of the same type;
for instance, the pennies are indistinguishable from each other. In how many different ways can you make
change for a dollar?
Again, one way to approach this is to write out the possibilities. As a way of organizing it, we’ll start with
the maximum possible number of quarters, then the maximum possible number of dimes, and so on. So the
first possibility is using all 3 quarters, then the maximum of 2 dimes, and then a nickel. The next possibility
uses 3 quarters and 2 dimes but then 5 pennies instead of the nickel. Then we consider the possibilities with
3 quarters and only 1 dime, and so on. We get the following list:

• 3 quarters, 2 dimes, and a nickel

• 3 quarters, 2 dimes, and 5 pennies

• 3 quarters, 1 dime, 2 nickels, and 5 pennies

• 2 quarters, 4 dimes, and 2 nickels

• 2 quarters, 4 dimes, 1 nickel, and 5 pennies

We do not have enough smaller coins to use 2 quarters and only 3 dimes, or only 1 quarter. So there are
only 5 possibilities total.
Let’s now use polynomial multiplication to solve this instead. For each type of coin, if the coin is worth
k cents, we associate the polynomial 1 + xk + x2k + · · · + xmk where m is the number of that coin we have.
This way, the exponents in our polynomial encode the possible amounts of cents we can make using just
those coins. So we would get the polynomials:

• Pennies: 1 + x + x2 + x3 + x4 + x5 + x6

• Nickels: 1 + x5 + x10

• Dimes: 1 + x10 + x20 + x30 + x40

• Quarters: 1 + x25 + x50 + x75

We take the product of all four polynomials:

(1 + x + x2 + x3 + x4 + x5 + x6 ) · (1 + x5 + x10 ) · (1 + x10 + x20 + x30 + x40 ) · (1 + x25 + x50 + x75 )

Typing this product into Sage using the command expand, we find that the product expands as:

x131 + x130 + x129 + x128 + x127 + 2 x126 + 2 x125 + x124 + x123 + x122 + 3 x121 + 3 x120 + 2 x119 + 2 x118 +
117
2x + 3 x116 + 3 x115 + x114 + x113 + x112 + 3 x111 + 3 x110 + 2 x109 + 2 x108 + 2 x107 + 4 x106 + 4 x105 +
2 x + 2 x103 + 2 x102 + 5 x101 + 5 x100 + 3 x99 + 3 x98 + 3 x97 + 6 x96 + 6 x95 + 3 x94 + 3 x93 + 3 x92 + 6 x91 +
104

6 x90 + 3 x89 + 3 x88 + 3 x87 + 6 x86 + 6 x85 + 3 x84 + 3 x83 + 3 x82 + 6 x81 + 6 x80 + 3 x79 + 3 x78 + 3 x77 + 6 x76 +
6 x75 + 3 x74 + 3 x73 + 3 x72 + 6 x71 + 6 x70 + 3 x69 + 3 x68 + 3 x67 + 6 x66 + 6 x65 + 3 x64 + 3 x63 + 3 x62 +
6 x61 + 6 x60 + 3 x59 + 3 x58 + 3 x57 + 6 x56 + 6 x55 + 3 x54 + 3 x53 + 3 x52 + 6 x51 + 6 x50 + 3 x49 + 3 x48 +
3 x47 + 6 x46 + 6 x45 + 3 x44 + 3 x43 + 3 x42 + 6 x41 + 6 x40 + 3 x39 + 3 x38 + 3 x37 + 6 x36 + 6 x35 + 3 x34 +
3 x33 + 3 x32 + 5 x31 + 5 x30 + 2 x29 + 2 x28 + 2 x27 + 4 x26 + 4 x25 + 2 x24 + 2 x23 + 2 x22 + 3 x21 + 3 x20 +
x19 +x18 +x17 +3 x16 +3 x15 +2 x14 +2 x13 +2 x12 +3 x11 +3 x10 +x9 +x8 +x7 +2 x6 +2 x5 +x4 +x3 +x2 +x+1

Look closely at the x100 term above. Its coefficient is 5, which is exactly the number of ways of making
100 cents (one dollar) out of the coins! Indeed, each way of getting x100 from the product corresponds to a
choice of some number of the pennies, nickels, dimes, and quarters that add up to 100 cents.

114
Proposition 7.1.3. Suppose you have k1 pennies, k5 nickels, k10 dimes, and k25 quarters in your pocket.
Then the number of ways you can make change for n cents is the coefficient of xn in the polynomial

(1 + x + · · · + xk1 ) · (1 + x5 + · · · + x5k5 ) · (1 + x10 + · · · + x10k10 ) · (1 + x25 + · · · + x25k25 ).

Notice that the above reasoning works no matter what values the coins are. Here is an example using a
different kind of currency.

Example 7.1.4. In Harry Potter currency, a Knut is the analog of a penny, a Sickle is worth 29 Knuts, and
a Galleon is worth 17 Sickles (so a Galleon is 17 · 29 = 493 Knuts). If we have a wizarding money pouch
containing two Galleons, two Sickles, and three Knuts, the corresponding polynomials are:

• Knuts: 1 + x + x2 + x3

• Sickles: 1 + x29 + x58

• Galleons: 1 + x493 + x986

Multiplying these polynomials together and expanding in Sage, we find:

(1 + x + x2 + x3 )(1 + x29 + x58 )(1 + x493 + x986 )

= x1047 +x1046 +x1045 +x1044 +x1018 +x1017 +x1016 +x1015 +x989 +x988 +x987 +x986 +x554 +x553 +x552 +
x +x525 +x524 +x523 +x522 +x496 +x495 +x494 +x493 +x61 +x60 +x59 +x58 +x32 +x31 +x30 +x29 +x3 +x2 +x+1
551

Notice that all of the coefficients in this polynomial are 1. That means there is exactly one way to make
change for each of the exponents appearing: 1047, 1046, 1045, 1044, 1018, 1017, etc., and no ways to make
any other amount of wizarding money using the coins in that pouch.

We can summarize our findings in the following theorem.

Theorem 7.1.5. Suppose there are m types of coins of values k1 , k2 , . . . , km , and you have n1 , n2 , . . . , nm of
each type of coin respectively. The number of ways to make change for a total value of N using these coins
is the coefficient of xN in the product

(1 + xk1 + x2k1 + · · · + xn1 k1 )(1 + xk2 + x2k2 + · · · + xn2 k2 ) · · · (1 + xkm + x2km + · · · + xnm km ).

Proof. If we multiply together one term from each factor in the product, we get a product of the form

xa1 k1 xa2 k2 · · · xam km = xa1 k1 +a2 k2 +···+am km

where each ai ≤ ni . The exponent a1 k1 + a2 k2 + · · · + am km equals N exactly when adding ai copies of

ki for each i adds up to N , which corresponds to using ai coins of the i-th type to make change for N .
Therefore, the number of times xN occurs in the expansion of the product is exactly the number of ways to
make change for N .

Exercises
1. Suppose you have the same coins in your wallet as in the first example above, namely 6 pennies, 2
nickels, 4 dimes, and 3 quarters.

(a) How many ways can you give someone 70 cents?

(b) How many ways can you give someone 110 cents?
(c) What are the values m such that there is exactly one way to give someone m cents?

115
 2. The polynomial in Example 7.1.4 shows that there should be exactly one way of making exact change
for 553 Knut values using at most two Galleons, at most two Sickles, and at most three Knuts. Figure
out how to make change for 553 using these coins, and explain how it comes from multiplying terms
from the three factors in the product.
3. At the video arcade, the red token is worth 3 dollars, the green token is worth 4 dollars, and the black
token is worth 5 dollars. You have 8 red tokens, 6 green tokens, and 4 black tokens. The question is
how many ways you can spend 50 dollars in tokens.
(a) Find the polynomial f (x) that shows how many ways to spend every number of dollars in tokens.
(b) Expand f (x) in Sage using the command expand.
(c) How many ways can you spend 50 dollars in tokens?
4. The Haunted Game Cafe in Fort Collins sells buffalo wings in small or large baskets. The small basket
contains 9 wings and the large basket contains 15 wings. A group of friends wants to order some buffalo
wings, but there are only 6 small baskets and 5 large baskets left.
(a) Find a polynomial that shows how many ways they can order every possible number of buffalo
wings.
(b) How many ways can they order 60 buffalo wings?

7.2 Generating functions

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/evP2mK7l7I0

Suppose a cashier with an unlimited supply of each coin was trying to figure out how many ways they
can make change for a dollar. Then the polynomials from Section 7.1 would turn into infinite summations:
• Pennies: 1 + x + x2 + x3 + · · ·
• Nickels: 1 + x5 + x10 + x15 + · · ·
• Dimes: 1 + x10 + x20 + x30 + · · ·
• Quarters: 1 + x25 + x50 + x75 + · · ·
We therefore need to define a notion of addition and multiplication of these types of infinite series, which
we call generating functions.
Definition 7.2.1. The generating function of the sequence of numbers a0 , a1 , a2 , . . . is the series
∞
X
ai xi = a0 + a1 x + a2 x2 + a3 x3 + · · · .
i=0

(Here x is just a formal symbol, and the coefficients of the series are the important part!)
Remark 7.2.2. Any polynomial can be interpreted as a generating function of a sequence that is eventually
0; for instance, the generating function of the sequence

1, 1, 3, 0, 0, 0, 0, 0, 0, . . .

is
1 + x + 3x2 + 0x3 + 0x4 + 0x5 + · · · = 1 + x + 3x2 .

116
Example 7.2.3. The generating function of the sequence 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, . . . is the polynomial

1 + x + x2 + x3 + x4 + x5 + x6 ,

which is the polynomial we used to record the six pennies in Example 7.1.2. On the other hand, the
polynomial that recorded the two nickels was

1 + x5 + x10 ,

which is the generating function of the sequence

1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, . . .

having a 1 in the 0th, 5th, and 10th positions of the sequence and 0 everywhere else.
Example 7.2.4. The generating function of the sequence 1, 2, 3, 4, 5, . . . (in other words, the sequence
a0 , a1 , a2 , . . . defined by ai = i + 1) is

1 + 2x + 3x2 + 4x3 + · · · .

We will revisit this example in the next section.

Example 7.2.5. The generating function of the Fibonacci sequence 1, 1, 2, 3, 5, 8, . . . is

1 + x + 2x2 + 3x3 + 5x4 + 8x5 + · · · .

By the end of this chapter we will see how to use generating functions to find an explicit formula for the
Fibonacci numbers themselves.
Example 7.2.6.

n
(Revisiting

the
binomial theorem.) Fix a number n. The generating function of the
sequence nn , n−1 n
, n−2 , . . . , n0 is:
       
n n n 2 n n
+ ·x+ x + ··· + x .
n n−1 n−2 0

This looks very familiar! The right hand side is what we get if we substitute y = 1 into Theorem 4.2.2
and rearrange the terms:
       
n n n n n
(1 + x)n = + x+ x2 + · · · + x .
n n−1 n−2 0





In fact we can replace nn with n0 , replace n−1 n
with n1 , and so on by the symmetry of binomial
coefficients, to get the somewhat nicer looking identity
       
n n n 2 n n
(1 + x)n = + x+ x + ··· + x .
0 1 2 n

Here is a different way to think about this that is similar to the problem of making change. Let’s say there
are n people in a room and you need exactly k of them to complete a survey. You can ask each person to
hold up a card saying 0 if they did not complete the survey and saying 1 if they did. You need the sum of
the cards to be k.
Equivalently, from the polynomial 1 + x you can ask each person to choose the monomial 1 = x0 if they
did not complete the survey and to choose the monomial x = x1 if they did. Then you need the product of
the chosen monomials, which is the sum of the exponents, to equal xk . The conclusion is that the coefficient
of xk in (1 + x)n is the number
of ways to choose exactly

k people out of the n people to complete the survey,
and this number of ways is nk . In other words, nk is the coefficient of xk in (1 + x)n .

117
7.2.1 Are generating functions actually functions?
In this book, we will not be thinking of generating functions as functions. While it is common to use notation
like
A(x) = a0 + a1 x + a2 x2 + a3 x3 + · · ·
to denote a generating function, the notation A(x) is just a convenient shorthand. We will not be substituting
a numerical value of x into the formula for a generating function. Indeed, if we try to plug in certain values
of x into an infinite series, the sum may not even converge.
Instead, we like to think of x as a formal symbol. The reason we take powers of x is that the ways we
manipulate generating functions are very similar to the ways we manipulate polynomials and power series
in calculus. We will define properties of generating functions in the next section.
As Herbert Wilf says in his book Generatingfunctionology,

“A generating function is a clothesline on which we hang up a sequence of numbers up for display.”

In other words, it is the coefficients (the sequence a0 , a1 , a2 , . . . that we are hanging up for display) that we
are studying here, not any sort of output of the function. In combinatorics, we study the series itself to gain
information about its coefficients.

Remark 7.2.7. We can use any variable we want to write down the generating function of a sequence.
So far we have been considering generating functions in x, but for instance, the generating function of the
sequence 1, 2, 3, 4, 5, . . . in the variable y is

1 + 2y + 3y 2 + 4y 3 + · · ·

and its generating function in the variable , is

1 + 2, + 3,2 + 4,3 + · · ·

Exercises
1. Write down the generating function of the sequence of even numbers 0, 2, 4, 6, 8, . . . in the variable x,
in the variable y, and in the variable ,.

2. Write down the generating function (in x) of the sequence 1, −1, 1, −1, 1, −1, . . ..

3. What sequence gives rise to the following generating function?

1 + x2 + x4 + x6 + · · ·

7.3 Operations on generating functions

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=FFUH4IpS-V0

If the cashier has an unlimited supply of pennies, nickels, dimes, and quarters, we wish to “multiply”
the corresponding infinite generating functions just like we multiplied the polynomials in Example 7.1.2. We
therefore now need to define operations on generating functions. The four main operations we will consider
are addition, multiplication, substitution of a monomial, and differentiation.
P∞ P∞
In this section, let A(x) = n=0 an xn and B(x) = n=0 bn xn be two generating functions.

118
7.3.1 Addition (and subtraction) of generating functions
Addition is the simplest operation on generating functions; we simply add corresponding coefficients:
∞
X ∞
X ∞
X
A(x) + B(x) = a n xn + bn x n = (an + bn )xn . (7.1)
n=0 n=0 n=0

Writing out the summations, the first few terms look like:

(a0 + a1 x + a2 x2 + · · · ) + (b0 + b1 x + b2 x2 + · · · ) = (a0 + b0 ) + (a1 + b1 )x + (a2 + b2 )x2 + · · ·

Example 7.3.1. The sum of the generating functions

1 + x + x2 + x3 + x4 + x5 + x6 + · · ·

and
1 − x + x2 − x3 + x4 − x5 + x6 + · · ·
is
2 + 2x2 + 2x4 + 2x6 + · · · .
We can similarly subtract generating functions. In fact, any generating function minus itself equals 0
(the generating function of the sequence 0, 0, 0, 0, 0, . . .).

7.3.2 Multiplication of generating functions

Multiplication of generating functions works much like multiplication of polynomials. Let’s start with an
example.
Example 7.3.2. Consider the generating function of the sequence 1, 2, 3, 4, 5, 6, . . .. We will compute its
product with itself:

(1 + 2x + 3x2 + 4x3 + 5x4 + · · · )(1 + 2x + 3x2 + 4x3 + 5x4 + · · · )

Naively we would expect the expansion of this product to be performed by multiplying one term from the
left factor with one term from the right factor in all possible ways, and adding up the resulting terms. We
can organize these products in a table as follows:

× 1 2x 3x2 4x3 5x4 ···

1 1 2x 3x2 4x3 5x4
2x 2x 4x2 6x3 8x4 10x5
3x2 3x2 6x3 9x4 12x5 15x6
4x3 4x3 8x4 12x5 16x6 20x7
5x4 5x4 10x5 15x6 20x7 25x8
..
.

Finally, to compute the sum of all the entries of this table, we notice that the down-and-left diagonals all
have the same exponent of x, and we can combine the terms along each diagonal, obtaining the sum:

1 + (2x + 2x) + (3x2 + 4x2 + 3x2 ) + (4x3 + 6x3 + 6x3 + 4x3 ) + · · · = 1 + 4x + 10x2 + 20x3 + · · ·

We can generalize the above example as follows. To compute the product

A(x)B(x) = (a0 + a1 x + a2 x2 + · · · )(b0 + b1 x + b2 x2 + · · · ),

consider the table:

119
 × b0 b1 x b 2 x2 b3 x3 b4 x4 ···
a0 a0 b0 a0 b1 x a0 b2 x2 a0 b3 x3 a0 b4 x4
a1 x a1 b0 x a1 b1 x2 a1 b2 x3 a1 b3 x4 a1 b4 x5
a2 x2 a2 b0 x2 a2 b1 x3 a2 b2 x4 a2 b3 x5 a2 b4 x6
a3 x3 a3 b0 x3 a3 b1 x4 a3 b2 x5 a3 b3 x6 a3 b4 x7
a4 x4 a4 b0 x4 a4 b1 x5 a4 b2 x6 a4 b3 x7 a4 b4 x8
..
.

Again we add down the down-and-left diagonals to get the following summation:

A(x)B(x) = (a0 b0 ) + (a0 b1 + a1 b0 )x2 + (a0 b2 + a1 b1 + a2 b0 )x2 + · · · .

In general, we see that the coefficient of xn in the product above is a0 bn + a1 bn−1 + a2 bn−2 + · · · + an b0 . This
leads to the general definition of multiplication of generating functions.

Definition 7.3.3. We define the product of the generating functions A(x) and B(x) by

∞
! ∞
! ∞
X X X
n n
A(x)B(x) = an x bn x = (a0 bn + a1 bn−1 + a2 bn−2 + · · · + an b0 )xn
n=0 n=0 n=0
∞ n
!
X X
= ak bn−k xn .
n=0 k=0

Pn
The coefficient k=0 ak bn−k of xn is called the n-th convolution of the sequences a0 , a1 , . . . and b0 , b1 , . . ..

We can use the multiplication rule to prove one of the most important generating function identities,
known as a geometric series.

Theorem 7.3.4. We have the identity

1
1 + x + x2 + x3 + · · · = .
1−x

Proof. The meaning of this identity is that the product of the generating functions 1 + x + x2 + x3 + · · ·
and 1 − x is equal to 1 = 1 + 0x + 0x2 + 0x3 + · · · . Indeed, using the formula for multiplying generating
functions, we find that the coefficient of x0 is 1, and all the other coefficients are 1 − 1 = 0.

Remark 7.3.5. If you have a good memory, you might remember the formula in Theorem 7.3.4 from Calculus
II. In Calculus II, you learn that 1 + x + x2 + · · · is the Maclaurin series for the function f (x) = 1/(1 − x),
in other words the Taylor series for f (x) centered at x = 0; also you might have learned that the radius of
convergence for the Maclaurin series is R = 1.
It is important to clarify the difference between those statements in calculus and Theorem 7.3.4. In this
chapter, we are defining everything algebraically and only need the multiplication and addition rules for
series to prove Theorem 7.3.4. We will not focus on the radius of convergence since we are not evaluating
this series at any number.

A “closed form formula” for a generating function is a formula that

Pdoes not involve an infinite summation.
∞
In Theorem 7.3.4, we found the closed form formula 1/(1 − x) for i=0 xi . We will see in Section 7.4 that
finding a closed form formula for a generating function can often come in handy for computing its coefficients.

Example 7.3.6. A similar argument to Theorem 7.3.4 shows that the alternating series 1 − x + x2 − x3 +
x4 − x5 + · · · has the closed form formula 1/(1 + x).

120
 P∞
Example 7.3.7. Let’s compute the square of the geometric series i=0 xi . By the multiplication rule, we
have

(1 + x + x2 + x3 + · · · )(1 + x + x2 + x3 + · · · ) = 1 · 1 + (1 · 1 + 1 · 1)x + (1 · 1 + 1 · 1 + 1 · 1)x2 + · · ·

= 1 + 2x + 3x2 + 4x3 + · · ·

We can simplify the left hand side as 1/(1 − x)2 using Theorem 7.3.4, leading to the closed form generating
function identity
1
1 + 2x + 3x2 + 4x3 + · · · =
(1 − x)2
for the sequence 1, 2, 3, 4, . . ..
Example 7.3.8. Multiplying a generating function by x. We can think of x as the generating function
0 + x + 0x2 + 0x3 + · · · , and use the multiplication formula to find:

x(a0 + a1 x + a2 x2 + a3 x4 + · · · ) = (0 + 1 · x + 0x2 + 0x3 + · · · )(a0 + a1 x + a2 x2 + a3 x4 + · · · )

= (0 · a0 ) + (0 · a1 + 1 · a0 )x + (0 · a2 + 1 · a1 + 0 · a2 )x3 + · · ·
= 0 + a0 x + a1 x2 + a2 x3 + · · · .

This is indeed intuitively what we would expect if we simply “distributed” the x across the infinite summation.
In terms of coefficients, it also shows that multiplying a generating function by x shifts the coefficients;
precisely, when A(x) is the generating function of the sequence a0 , a1 , a2 , . . ., then the coefficient of xn in
xA(x) is an−1 . In summation notation,
∞
! ∞ ∞
X X X
n
x an x = an xn+1 = am−1 xm ,
n=0 n=0 m=1

where the last equality uses the substitution m = n + 1. Sometimes, we replace m by n in the last line and
write !
∞
X ∞
X
n
x an x = an−1 xn .
n=0 n=1

Example 7.3.9. For instance, we can multiply both sides of the geometric series formula by x to find
x
x + x2 + x3 + x4 + · · · = .
1−x
Similarly, we can multiply a generating function by xk to shift the coefficients by k spots, and we can
multiply through by a constant. For instance, we can multiply the geometric series formula by 2x2 to get
the identity
2x2
2x2 + 2x3 + 2x4 + 2x5 + · · · = .
1−x

7.3.3 Substituting monomials into generating functions

A monomial in x is any term of the form axn where a is a number and n is an integer called the degree. We
may substitute monomials of positive degree into generating function identities to produce more identities.
1
Example 7.3.10. Write G(x) = 1+x+x2 +x3 +x4 +· · · . Then we have the closed form formula G(x) = 1−x
by Theorem 7.3.4. We can substitute 2x for x to obtain

G(2x) = 1 + (2x) + (2x)2 + (2x)3 + · · ·

1
= 1 + 2x + 4x2 + 8x3 + · · · .
1 − 2x

121
Example 7.3.11. If we want to find a closed form formula for 1 + x2 + x4 + x6 + x8 + · · · , we notice that
1
it is the result of substituting x2 for x in the geometric series, to get G(x2 ) = 1−x 2.

Remark 7.3.12. Warning: do not try to substitute constants or formulas that are not monomials into
generating functions without caution! What goes wrong if we substitute y = 1 + x in the generating function
1 + y + y 2 + y 3 + y 4 + · · · ? What goes wrong if we substitute y = 1? More generally, the composition of
generating functions G(F (x)) makes sense only when F (x) has no constant term.

7.3.4 Differentiation of generating functions

We now define differentiation of generating functions, by extending the power rule for derivatives in calculus.

Definition 7.3.13. We define

∞
! ∞ ∞
d X X X
n n−1
an x = nan x = (n + 1)an+1 xn .
dx n=0 n=1 n=0

In the last equality, we re-indexed by replacing n with n + 1. More visually:

d

a0 + a1 x + a2 x2 + a3 x3 + · · · = a1 + 2a2 x + 3a3 x2 + 4a4 x3 + · · · .
dx

Remember that generating functions are formal power series that we are not thinking of as actual func-
tions. The amazing thing is that, despite this, all of the usual differentiation rules from calculus still apply
here. In particular, for any generating functions F (x) and G(x), the following identities hold:

d d d
• Sum Rule: (F (x) + G(x)) = F (x) + G(x)
dx dx dx

d d d
• Product Rule: (F (x)G(x)) = G(x) · F (x) + F (x) · G(x)
dx dx dx

• Quotient Rule: If G(x) has a nonzero constant term g0 x0 , then

  d d
d F (x) G(x) dx F (x) − F (x) dx G(x)
= 2
dx G(x) G(x)

• Chain Rule for monomials: Let H(x) = d

dx F (x) and let axn be any monomial in x. Then
d n n−1
dx (F (ax )) = ax · H(axn ).

Example 7.3.14. The derivative of the geometric series 1 + x + x2 + x3 + x4 + x5 + · · · is 1 + 2x + 3x2 +

4x3 + 5x4 + · · · . We therefore have

d
1 + 2x + 3x2 + 4x3 + 5x4 + · · · = (1 + x + x2 + x3 + x4 + · · · )
dx  
d 1
=
dx 1 − x
1
=
(1 − x)2

which gives another proof of the identity we found via multiplication in Example 7.3.7.

122
Exercises
1. Find a closed form formula for the infinite series 1 − x + x2 − x3 + · · · .
P∞
2. Prove that 1/(1 − nx) = k=0 nk xk for any fixed positive integer n.
1
3. For any positive integer m, show that 1 + xm + x2m + x3m + · · · = 1−xm .

4. Use the derivative to find the coefficients a0 , a1 , a2 , . . . that yield these closed form formulas:
(a)
1
= a0 + a1 x + a2 x2 + · · · .
(1 − x)3
(b)
1
= a0 + a1 x + a2 x2 + · · · .
(1 − x)4
(c)
1
= a0 + a1 x + a2 x2 + · · · .
(1 − x)N
5. Recall this fact about the sequence 1, 2, 3, . . . whose generating function has the closed form 1/(1 − x)2 .
Let ak = k + 1. Note that ak is the number of ways to buy k flags, from an infinite collection of gold
and green flags.
(a) Let a0 , a1 , . . . be the sequence found in part (a) of the previous problem, whose generating function
has the closed form 1/(1 − x)3 . Find a formula for ak . Show that ak is the number of ways to
buy k flags, from an infinite collection of gold, green, and pink flags.
(b) Repeat using the sequence found in part (b) of the previous problem, where the generating function
has the closed form 1/(1 − x)4 and there are 4 colors of flags.
(c) Repeat using the sequence found in part (c) of the previous problem, where the generating function
has the closed form 1/(1 − x)N and there are N colors of flags.
6. Find another way to verify the formulas you found for (a), (b), and (c) in the two previous problems
using the geometric series and multiplication of power series. Your answer should include the formula
for the number of ways to count n objects where order does not matter and repeats are allowed. Your
answer should include an explanation, like the ones in the section about making change, about the
meaning of the coefficient of xn .
1
7. We define E(x) to be the generating series for the sequence an = n! , whose first terms are 1, 1, 21 , 16 , . . ..
In other words
1 1
E(x) = 1 + x + x2 + x3 + · · · .
2! 3!
Use the process of taking a derivative of an infinite series to show that d(E(x))/dx = E(x). Explain
why this makes E(x) behave like the function ex .

7.4 Solving recurrences with generating functions

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=whqRcIVKLOw
Click on the icon at left or the URL below for another short video discussing the
second method utilized in this section.
https://youtu.be/9DCkPPY5WVA

123
 In this section, we will see how to find explicit formulas for recursively defined sequences using generating
functions. Then we use this to investigate some of the nonlinear recurrences that we defined in Chapter 6.
The steps to find explicit formulas for recursively defined sequences are as follows:

1. Define the generating function. If a0 , a1 , a2 , . . . is a recursively defined sequence, set A(x) =

a0 + a1 x + a2 x2 + · · · to be its generating function.

2. Solve for the generating function. Use the recursion that defines the sequence, along with the
generating function operations we have studied, to find an equation that A(x) satisfies. Solve this
equation to find a closed form formula for A(x) in terms of x.

3. Expand to compute the coefficients of the generating function. Expand the closed form
formula for A(x) found in the previous step as a generating function (this step will often involve the
geometric series formula or some other well-known identity). If this expansion has a nice formula for
the coefficient of xn , then that gives an explicit formula for an .

Example 7.4.1. Consider the sequence defined by a0 = 3 and for all n ≥ 1, an = 2an−1 . While it is not
hard to find an explicit formula for this sequence without using generating functions, we’ll start with this as
a simple example to demonstrate the three steps of the method.

1. Define the generating function: We define

∞
X
A(x) = an xn = a0 + a1 x + a2 x2 + a3 x3 + · · · .
n=0

2. Solve for the generating function: We would like to find a closed form formula for A(x). The
recursion formula an = 2an−1 tells us that each coefficient is twice the previous one. By Example 7.3.8,
the coefficient of xn in xA(x) is an−1 for every n ≥ 1. We can also multiply the result by 2 to get:
∞
X ∞
X
2xA(x) = 2an−1 xn = an xn
n=1 n=1

where the second equality above follows from the recursion an = 2an−1 . Now, the right hand side
above looks almost like A(x) — but wait! The summation starts at n = 1, not n = 0. Let’s fix this by
subtracting and adding the a0 term back in (which does not change the equality) to get an equation
for A(x):
∞
X ∞
X
2xA(x) = −a0 + a0 + an xn = −a0 + an xn = −a0 + A(x) = −3 + A(x).
n=1 n=0

We have nearly completed step 2 — solving the equation 2xA(x) = −3 + A(x) for A(x), we have
(2x − 1)A(x) = −3, so A(x) = 3/(1 − 2x).

3. Expand to compute the coefficients: By substituting 2x in for x in the geometric series formula,
and then multiplying by 3, we see that

X∞ X∞ X∞
3
A(x) = =3 (2x)n = 3 2 n xn = 3 · 2n xn .
1 − 2x n=0 n=0 n=0

P∞
Having completed
P∞ the three steps, we now have the big payoff: we know that A(x) = n=0 an xn , but also
A(x) = n=0 3 · 2n xn . Since these are the same generating function, we can conclude that an = 3 · 2n for
all n, and voila! We’ve found an explicit formula for an .

124
 Tip: It is often useful to get a more concrete handle on this generating function in steps 1 and 2 by using
the recursion to calculate the first few terms of the sequence. Starting with a0 = 3, we have a1 = 2a0 = 6,
a2 = 2a1 = 12, and a3 = 2a2 = 24. So
A(x) = 3+ 6x + 12x2 + 24x3 + · · ·
2xA(x) = 6x + 12x2 + 24x3 + . . .
By lining up the terms for A(x) and 2xA(x), we see that 2xA(x) − A(x) = 3.

7.4.1 The Tower of Hanoi

In Section 6.3, we studied the Tower of Hanoi game. We found that the recurrence relation for the number
Hn of steps it takes to solve satisfies H1 = 1 and Hn = 2Hn−1 + 1 for all n ≥ 2. Notice that if we set H0 = 0
we get a sequence that satisfies the recurrence for all n ≥ 1. Let’s use this recursion to try to solve for the
generating function of the sequence H0 , H1 , H2 , . . ., by going through the three steps again:
1. Define the generating function: First set H(x) to be the generating function:
H(x) = H0 + H1 x + H2 x2 + H3 x3 + · · · .

2. Solve for the generating function: We can use the recursion to replace the coefficients H1 , H2 , . . .
with 2H0 + 1, 2H1 + 1, 2H2 + 1 and so on respectively:
H(x) = H0 + (2H0 + 1)x + (2H1 + 1)x2 + (2H2 + 1)x3 + · · · .
We can set H0 = 0, and then by the addition rule for generating functions, we can split the remaining
terms up as a sum of generating functions:
H(x) = ((2H0 )x + (2H1 )x2 + (2H2 )x3 + · · · ) + (x + x2 + x3 + · · · ) (7.2)
2 2
H(x) = 2x(H0 + H1 x + H2 x + · · · ) + x(1 + x + x + · · · ) (7.3)
x
H(x) = 2xH(x) + (7.4)
1−x
where the last simplification is by the definition of H(x) and the geometric series formula. (Notice that
P the · · · notation by writing out the first few terms of each generating
we did this step using P function,
rather than using notation like in the previous example. Can you rewrite it using notation?)
We can now gather all the H(x) terms on the left hand side to solve for H(x):
x
H(x) − 2xH(x) = (7.5)
1−x
x
H(x)(1 − 2x) = (7.6)
1−x
x
H(x) = (7.7)
(1 − x)(1 − 2x)

3. Expand to compute the coefficients: We’d like to break up the fraction as a sum of two fractions
to make it look more like the geometric series formula. Luckily, there’s a tool from algebra that does
just that: the method of partial fractions. This states that there are numbers a and b such that
x a b
= + .
(1 − x)(1 − 2x) 1 − x 1 − 2x
There are a few ways to see that a = −1 and b = 1, for example, by finding a common denominator
for the right hand side. (Try adding the fractions to double check this yourself!). So we have
1 1
H(x) = − .
1 − 2x 1 − x

125
 Then we can expand both terms using the geometric series formula:
∞
X ∞
X ∞
X
H(x) = 2n xn − xn = (2n − 1)xn .
n=0 n=0 n=0

Therefore, the coefficient Hn is equal to 2n − 1 for all n.

7.4.2 Regions of the plane

In Section 6.4, we showed that the number Pn of regions in the plane cut out by drawing n lines (with no
two parallel and no three concurrent) satisfies the recurrence P0 = 1 and Pn = Pn−1 + n. We go through
the steps once again.

1. Define the generating function: We set

∞
X
P (x) = P n = P0 + P1 x + P2 x 2 + P3 x 3 + · · · .
n=0

2. Solve for the generating function: In order to make use of the recursion Pn = Pn−1 + n, we
multiply P (x) by x to obtain the generating function for Pn−1 :

∞
X
Pn−1 xn = xP (x)
n=1

and we also compute the generating function for n:

∞
X x
nxn = x + 2x2 + 3x3 + 4x4 + · · · = x(1 + 2x + 3x2 + 4x3 + · · · ) =
n=1
(1 − x)2

by Example 7.3.7. Adding these two generating functions, we have

∞
X x
(Pn−1 + n)xn = xP (x) +
n=1
(1 − x)2
∞
X x
Pn xn = xP (x) +
n=1
(1 − x)2
x
P (x) − P0 = xP (x) +
(1 − x)2
x
P (x) − xP (x) = P0 +
(1 − x)2
x
P (x)(1 − x) = 1 +
(1 − x)2
1 x
P (x) = +
1 − x (1 − x)3

1
3. Expand to compute the coefficients: The first term 1−x can be expanded using the geometric
x
series formula, but we haven’t seen the generating function (1−x)3 yet. For this, we can take another

126
 1
derivative of the generating function (1−x)2 , which we’ve already computed (Example 7.3.7):

X∞
1
= (n + 1)xn
(1 − x)2 n=0
∞
!
d
d X
(1 − x)−2 = (n + 1)x n
dx dx n=0
X∞
2
= n(n + 1)xn−1
(1 − x)3 n=1
∞
1 1X
= n(n + 1)xn−1
(1 − x)3 2 n=1
X∞
x n(n + 1) n
3
= x
(1 − x) n=1
2

where the last equality forms by multiplying both sides by x. Notice that since this sum has constant
P∞
term 0, we can also start the indexing of n at 0 to write it as n=0 n(n+1)
2 xn .
We can finally substitute to find
1 x
P (x) = +
1 − x (1 − x)3
X∞ X∞
n(n + 1) n
= xn + x
n=0 n=0
2
X∞  
n(n + 1)
= 1+ xn
n=0
2
X∞  2 
n +n+2
= xn
n=0
2

n2 +n+2
We can conclude that Pn = 2 for all n, as we found in Section 6.4.

Exercises
1. Find a closed form for the generating functions for the following recursively defined sequences:
(a) a0 = 1, a1 = 5, an = 4an−1 − 3an−2 for n ≥ 2.
(b) b0 = 1, b1 = 6, bn = 4bn−1 − 4bn−2 for n ≥ 2.
(c) c0 = 1, c1 = 1, c2 = 1, cn = cn−1 + cn−2 + cn−3 for n ≥ 3.
(d) d0 = 1, dn = 2dn−1 + n + 1 for n ≥ 1.
(e) e0 = 1, en = en−1 + en−2 + · · · + e2 + e1 + e0 for n ≥ 1.


(f) h0 = 1, hn = − n1 hn−1 for n ≥ 1.
2. In part (a) above, use partial fractions on the generating function to find an explicit formula for the
sequence an .
1
3. In part (b) above, use monomial substitution into the formula (1−x)2 = 1 + 2x + 3x2 + · · · to find an
explicit formula for bn .
4. Consider the double recursion defined by f0 = 1, f1 = 2, g0 = 1, g1 = 1, fn = fn−1 + gn−2 ,
gn = gn−1 + fn−1 . Find the generating functions F (x) and G(x) of the sequences fn and gn .

127
7.5 Generating functions and linear recurrences
We now show how to solve linear recurrences using generating functions.
Example 7.5.1. Suppose the sequence a0 , a1 , a2 , . . . is defined by a0 = 0, a1 = 1, and the recurrence relation
an = 5an−1 − 6an−2 for n ≥ 2. The first several terms of this sequence are 0, 1, 5, 19, 89, . . . , Then we define
A(x) to be the generating function for this sequence:
∞
X
A(x) = a0 + a1 x + a2 x2 + · · · = an xn .
n=0

Its first several terms are:

0 + x + 5x2 + 19x3 + 89x4 + · · · .
The next lemma shows how we can use generating functions to find a closed-form formula for an for all
n.
Lemma 7.5.2. If a0 = 0, a1 = 1, and an = 5an−1 − 6an−2 for n ≥ 2, then an = 3n − 2n for all n.
Proof. The first step is to use the recurrence relation to find a closed form formula for A(x). For all n ≥ 2,
we can substitute an = 5an−1 − 6an−2 , so we can write
∞
X
A(x) = a0 + a1 x + an xn
n=2
X∞
= a0 + a1 x + (5an−1 − 6an−2 )xn
n=2
∞
! ∞
!
X X
n n
= 0 + 1x + 5 an−1 x −6 an−2 x .
n=2 n=2

We re-index the first infinite sum in the line above, replacing n by n + 1, and the second infinite sum in the
line above by replacing n by n + 2. After that, we can factor out an x from each term in the first infinite
sum and an x2 in the second infinite sum. So
∞
! ∞
!
X X
n+1 n+2
A(x) = x + 5 an x −6 an x
n=1 n=0
∞
! ∞
!
X X
n 2 n
= x + 5x an x − 6x an x
n=1 n=0

Now the first infinite sum is the sum as A(x) without its constant term (which is zero anyway) and the
second infinite sum is the same as A(x). So

A(x) = x + 5xA(x) − 6x2 A(x)

= x + (5x − 6x2 )A(x)

We can now solve this equation for A(x) and then factor the denominator to find
x x
A(x) = 2
=
1 − 5x + 6x (1 − 2x)(1 − 3x)

We now use the method of partial fractions to write this as a sum of two fractions, of the form
r s
A(x) = + .
1 − 2x 1 − 3x

128
To solve for r and s, we combine the fractions to get r+s−(3r+2s)x
(1−2x)(1−3x) , so r + s − (3r + 2s)x = x. Therefore
r + s = 0 and 3r + 2s = −1. Solving, we find r = −1 and s = 1. Therefore
−1 1
A(x) = + ,
1 − 2x 1 − 3x
and we can now expand both fractions using the geometric series theorem. Indeed, we have
∞
X ∞
X ∞
X
A(x) = − 2 n xn + 3n xn = (3n − 2n )xn .
n=0 n=0 n=0
P∞
Since this is exactly the same generating function as n=0 an xn , we can conclude that an = 3n − 2n for all
n.

Remark 7.5.3. In the above example, here is a different way to solve for A(x). We recall that multiplying
A(x) by x shifts the coefficients by one index and multiplying by x2 shifts the coefficients by two indices. In
particular we have
∞
X
xA(x) = a0 x + a1 x2 + a2 x3 + · · · = an−1 xn
n=1

and
∞
X
x2 A(x) = a0 x2 + a1 x3 + a3 x4 + · · · = an−2 xn .
n=2
2
Therefore, if we add together A(x), −5xA(x), and 6x A(x), most of the coefficients will cancel since an −
5an−1 + 6an−2 = 0 for all n ≥ 2:

A(x) = a0 + a1 x + a2 x2 + a 3 x3 + ···
−5xA(x) = − 5a0 x − 5a1 x2 − 5a2 x3 + ···
+6x2 A(x) = 6a0 x2 + 6a1 x3 + ···
A(x) − 5xA(x) + 6x2 A(x) = a0 + (a1 − 5a0 )x + 0x2 + 0x3 + ···

Therefore, A(x)(1 − 5x + 6x2 ) = a0 + (a1 − 5a0 )x = x.

Notice in the previous calculation that we multiplied A(x) by something very close to the characteristic
polynomial of the recursion an − 5an−1 + 6an−2 = 0 (see Definition 6.2.12), which is x2 − 5x + 6. Indeed, we
multiply here by the reverse of the characteristic polynomial, defined by reversing the sequence of coefficients
to make it 1 − 5x + 6x2 instead.
We now apply this principle to solve another recursion.

Example 7.5.4. Let’s use generating functions to derive the explicit formula for the nth Fibonnacci number
Fn . To do this, it is easier to start with the index n = 0, so we write F0 = 0, F1 = 1, and Fn = Fn−1 + Fn−2
for all n ≥ 2. Consider the generating function
∞
X
G(x) = Fn xn .
n=0

As in the previous problem, we multiply by the reverse of the characteristic polynomial, 1 − x − x2 , to get
∞
X
G(x) − xG(x) − x2 G(x) = F0 + F1 x − F0 x + (Fn − Fn−1 − Fn−2 )xn
n=2
∞
X
= 0+1·x−0·x+ 0 · xn = x.
n=2

129
Therefore G(x) = x/(1 − x − x2 ).
√
We can factor
√
the denominator as (1 − ax)(1 − bx) where a + b = 1 and
ab = −1. Solving, we get a = 1+2 5 and b = 1−2 5 . Using partial fractions, we can write
A B
G(x) = + .
1 − ax 1 − bx
To find A and B, we add the fractions and compare coefficients: A(1 − √
bx) + B(1 − ax) = x, so A + B = 0
and Ab + Ba = −1. Thus A = −B and so A(b − a) = −1. Since a − b = 5 we have A = √15 and B = − √15 .
We now have √ √
1/ 5 1/ 5
G(x) = − .
1 − ax 1 − bx
Expanding each term as a geometric series, we find that
∞ ∞ ∞ √ !n √ !n !
X 1 n n X 1 n n X 1 1+ 5 1− 5
G(x) = √ a x − √ b x = √ − xn .
n=0
5 n=0
5 n=0
5 2 2

By the definition of equality of generating functions, it follows that

√ !n √ !n !
1 1+ 5 1− 5
Fn = √ − .
5 2 2

We can now generalize our methods above to prove the following theorem, which we stated before without
proof in Chapter 6.
Theorem 7.5.5. Let r1 , . . . , rd be the roots of the characteristic polynomial of the recurrence relation an =
c1 an−1 + c2 an−2 + · · · + cd an−d = 0. Suppose that the d roots are all different. Then there are constants
z1 , . . . , zd such that, if n ≥ 1, then
an = z1 r1n + · · · zd rdn . (7.8)
Furthermore, there is a unique choice of z1 , . . . , zd such that (7.8) is true.
Proof. The characteristic polynomial of the recurrence relation is c(x) = xd −c1 xd−1 −c2 xd−2 −· · ·−cd−1 x−cd .
Reversing its coefficients, we get the polynomial
r(x) = 1 − c1 x − c2 x2 − · · · − cd xd .
P∞
Now, let A(x) = n=0 an xn be the generating function of the recursively defined sequence a0 , a1 , . . .. Then
by using the generating function multiplication rule on r(x)A(x), we see that all the coefficients starting at
xd and beyond are 0. Thus r(x)A(x) = p(x) for some polynomial p(x), and so
p(x)
A(x) = .
r(x)
Now, we claim that r(x) factors as (1 − r1 x)(1 − r2 x) · · · (1 − rd x). Indeed, since r1 , . . . , rd are defined
to be the roots of c(x), we have c(x) = (x − r1 )(x − r2 ) · · · (x − rd ). But notice that r(x) = xd c(1/x) =
xd (1/x − r1 )(1/x − r2 ) · · · (1/x − rd ). Multiplying each factor through by one of the x’s from the xd factor,
we get r(x) = (1 − r1 x)(1 − r2 x) · · · (1 − rd x).
Putting this all together, we have
p(x) p(x)
A(x) = = .
r(x) (1 − r1 x)(1 − r2 x) · · · (1 − rd )x
Since the roots are assumed to be all different, this decomposes using partial fractions as a sum of fractions
of the form
z1 z2 zd
+ + ··· +
1 − r1 x 1 − r2 x 1 − rd x
for some constants z1 , . . . , zd . Expanding each of these terms using the geometric series formula and adding
the generating functions, we find that the n-th coefficient of the series is indeed z1 r1n + · · · + zd rdn .

130
7.6 The Catalan numbers
We end this chapter by deriving the generating function and explicit formula for the celebrated Catalan
numbers (see Section 6.6). Recall that the Catalan numbers are defined by the recursion

Cn+1 = C0 Cn + C1 Cn−1 + C2 Cn−2 + · · · + Cn C0 , (7.9)

with initial condition C0 = 1. In particular, for the first few values of n, this recursion says that

C1 = C0 C0 (7.10)
C2 = C0 C1 + C1 C0 (7.11)
C3 = C0 C2 + C1 C1 + C2 C0 (7.12)

and so on.

1. Define the generating function: We set

C(x) = C0 + C1 x + C2 x2 + C3 x3 + · · · .

2. Solve for the generating function: Notice that the right hand side of the Catalan recursion (7.9)
very closely resembles the coefficients we get when we multiply C(x) by itself. Indeed:

C(x)2 = C(x) · C(x) = C0 C0 + (C0 C1 + C1 C0 )x + (C0 C2 + C1 C1 + C2 C0 )x2 + · · ·

by the multiplication rule for generating functions. The expressions in parentheses on the right hand
side exactly match equations (7.10), (7.11), (7.12), and in general the coefficient of xn will match the
right hand side of equation (7.9). Substituting these in, we have

C(x)2 = C1 + C2 x + C3 x2 + C4 x3 + · · · .

We now want to re-express the right hand side in terms of C(x) to solve for C(x). In order to line up
the subscripts with the right powers of x, we can multiply both sides by x:

xC(x)2 = C1 x + C2 x2 + C3 x3 + C4 x4 + · · · .

The right hand side is almost C(x), but it is missing the C0 term at the start. So let’s add that to
both sides:
xC(x)2 + C0 = C0 + C1 x + C2 x2 + C3 x3 + C4 x4 + · · · .
Now the right hand side is C(x), and we can substitute C0 = 1 on the left, to get xC(x)2 + 1 = C(x),
or
xC(x)2 − C(x) + 1 = 0.
We can now use the quadratic formula to solve for C(x):
√
1±1 − 4x
C(x) =
2x
√
3. Expand to compute the coefficients: To interpret 1 − 4x as a generating function, we want to
think of it as “the generating function whose square is 1 − 4x”. To get an intuition for what this looks
like, we can start determining the coefficients one by one, starting with the equation:

(a0 + a1 x + a2 x2 + a3 x3 + · · · )2 = 1 − 4x;
a20 + (a0 a1 + a1 a0 )x + (a0 a2 + a1 a1 + a2 a0 )x2 + · · · = 1 − 4x.

131
 By comparing the coefficients of xn from the left and right sides of this equation, we can solve for
a0 , a1 , a2 , . . . one at a time. From the coefficients of x0 , we see that a20 = 1, and we take the positive
solution a0 = 1.1 From the coefficients of x1 , we see that a0 a1 + a1 a0 = −4, and substituting a0 = 1,
we find that a1 = −2. From the coefficients of x2 , we see that a0 a2 + a1 a1 + a2 a0 = 0, and substituting
a0 = 1, a1 = −2, we find that a2 = −2. Continuing in this manner, we claim that once the value of
a0 , a1 , . . . , an−1 are found, then there is a unique choice of an that makes the coefficients of xn from
the left and the right side of the equation equal. This provides a well-defined “square root” of 1 − 4x
as a generating function: √
1 − 4x = 1 − 2x − 2x2 − 4x3 − · · · .
√
This also means that in the formula C(x) = 1± 2x 1−4x
, we should use the negative term in the symbol
±, so that the constant term 1 cancels and all the terms in the numerator are divisible by 2x.
√
We’ve found the first several coefficients for the generating function of 1 − 4x but now we want to find
a formula for all of them. To do this, we will introduce a generalization of binomial numbers:

Definition 7.6.1 (Generalized Binomial Coefficients.). Let α be any rational number, and define
 
α α(α − 1)(α − 2) · · · (α − k + 1)
=
k k!


if k 6= 0 and α0 = 1.


If α is a positive integer n, then this definition is the same as the ordinary definition of nk .

Example 7.6.2. If we set α = 1/2 and k = 3 in the above formula, we have

 
1/2 (1/2)(−1/2)(−3/2) 3/8 1
= = = .
3 3! 6 16

Example 7.6.3. Let α = 1/2 and k = n + 1. Then

 
1/2 (1/2)(1/2 − 1)(1/2 − 2) · · · (1/2 − n)
=
n+1 (n + 1)!
(1/2)(−1/2)(−3/2)(−5/2) · · · ((1 − 2n)/2)
=
(n + 1)!
(1)(3)(5) · · · (2n − 1)(−1)n
=
2n+1 (n + 1)!
(1)(2)(3)(4)(5) · · · (2n − 1)(2n)(−1)n
=
(2)(4) · · · (2n)2n+1 (n + 1)!
(2n)!(−1)n
=
(2 · 1)(2 · 2)(2 · 3) · · · (2 · n)2n+1 (n + 1)!
(2n)!(−1)n
=
2 2n+1 n!(n + 1)!
n

√
To find a formula for the coefficients of 1 − 4x, we’ll use (without proof) this next theorem.

Theorem 7.6.4 (Generalized Binomial Theorem.). As a generating function,

∞  
X
α α k
(1 + x) = x .
k
k=0
1 This
√
is similar to the convention of setting 4 to be 2 rather than −2, even though both numbers have a square of 4.

132
Example 7.6.5. We set α = 1/2 and substitute −4x for x to see that:
X∞  
√ 1/2
1 − 4x = (−4x)k
k
k=0
X∞  
1/2
= (−1)k 4k xk
k
k=0

1 2n


We use this now to find the explicit formula Cn = n+1 n for the nth Catalan number.
1 2n


Corollary 7.6.6. The nth Catalan number is Cn = n+1 n .
√
Proof. The nth Catalan number Cn is the coefficient of xn in C(x) = 1± 2x 1−4x
. Using Example 7.6.5, we
see that
P

√ 1−
∞ 1/2 k k k
1 − 1 − 4x k=0 k (−1) 4 x
=
2x  2x 
P∞ 1/2

1 − 1 + k=1 k (−1)k 4k xk
=
2x
∞ 1/2


X n+1 (−1) n+1 n+1 n+1 n+1
2 2 x
= −
n=0
2x
X∞  
1/2
= (−1)n 2n 2n+1 xn
n=0
n + 1

In the third equality, we made the substitution n = k − 1.

Using Equation 7.6.3, we substitute for the binomial coefficients to see that the coefficient of xn is:
(2n)!(−1)n
Cn = (−1)n 2n 2n+1
2n 2n+1 n!(n+ 1)!
(2n)!
=
n!(n + 1)!
1 (2n)!
=
(n + 1) n!n!
 
1 2n
= .
n+1 n
This completes the proof.

7.7 Additional problems for Chapter 7

1. Suppose you have 8 pennies, 3 nickels, and 1 dime in your pocket and you want to give a subset of these
to your friend. For each of pennies, nickels, and dimes, write down a polynomial whose exponents are
the values of that coin you could give.
2. Define fp , fn , fd to be the polynomials for pennies, nickels, and dimes in Sage (sagecell.sagemath.org),
as in:
fd = 1 + x10 .
Use Sage to also define fp and fn , and expand the product F of these 3 polynomials and use the output
to answer the questions below.

133
 Note: If you just end the Sage cell with F rather than the F.coefficients line, it will return the entire
polynomial. The coefficients line will just return the coefficient of x17 and the number 17.

(a) What is the degree of the polynomial F? In other words, what is the largest power of x that
appears?
(b) How many ways can you give 17 cents?
(c) How many ways can you give 20 cents?
(d) If the number of dimes changes to 2, what does the degree of the polynomial change to?
(e) If the number of dimes changes to 2, what is the new number of ways you can give 20 cents?

Hint: These commands may be useful:

R=PolynomialRing(RR, x)
F=expand(fp*fn*fd);
F;
F.coefficients()[17]

3. What is the coefficient of x4 in the generating function whose closed form is 1/(1 − 3x)?

4. What is the coefficient of x4 in the generating function x2 /(1 − 3x)?

5. What is the coefficient of x4 in the generating function 1/(1 − 3x)2 ?

6. Let
F (x) = F0 + F1 x + F2 x2 + F3 x3 + · · ·
be the generating function of the Fibonacci numbers (with F0 = 0, F1 = 1, and Fn = Fn−1 + Fn−2 for
all n ≥ 2.) Compute the coefficient of x5 in the generating function 1/1 − xF (x).

7. Let F (x) be the generating function of the Fibonacci sequence as defined in the previous question.
What is the coefficient of x5 in d/dx F (x)?

8. Find a closed formula for the generating function of the recursively defined sequence defined by b0 =
5, b1 = 12, and bn = 5bn−1 + 6bn − 2 for all n ≥ 2.

9. Find a closed formula for the generating function of the sequence defined by e0 = 1 and en = en−1 +
en−2 + · · · + e0 for all n ≥ 1.

10. Prove, using generating functions, that

n
X  
n
k = n · 2n−1 .
k
k=0

P∞ x
11. We found that the generating function for the Fibonacci numbers is n=0 Fn xn = 1−x−x 2 . Write this
x 2
as 1−(x+x2 ) and use the geometric series formula to expand in terms of x + x to prove that

X
n−1
n−k−1

Fn = .
k
k=0

P∞ Q∞
12. Infinite products: Let G(x)Q= i=0 bi xi be a generating function. We say G(x) = n=0 Fn (x) if
m
the partial products Pm (x) = n=0 Fn (x) have the property that for all i, there exists N such that if
i
m > n, the coefficient of x in Pm equals bi . In other words, the Pm ’s coefficients eventually match
that of G.

134
(a) What generating function is equal to the infinite product

(1 + x)(1 + x2 )(1 + x4 )(1 + x8 )(1 + x16 ) · · ·?

(b) Let p(n) be the number of partitions of n, the number of ways of writing n as a sum of other
positive integers, where the order of the summands doesn’t matter. Show that
∞
X 1 1 1 1
p(n)xn = · · · ···
n=0
(1 − x) (1 − x2 ) (1 − x3 ) (1 − x4 )

(c) Prove that the number of partitions of n having distinct summands is equal to the number of
partitions of n having all its summands being odd. (Hint: can you factor the generating functions
of each of the two types of partitions?)

135
7.8 Investigation: Partitions and their generating functions
Generating functions can be used to enumerate many different combinatorial objects. One very famous
such object is called a partition. The following bonus video explores this topic.

Click on the icon at left or the URL below for this section’s video.
https://www.youtube.com/watch?v=Lghi2Otq5Q4

136
Chapter 8

Graph Theory Basics

Graphs are a convenient data structure for a wide variety of problems in mathematics and computer science.
Some graphs are valuable because they illustrate beautiful patterns and complicated symmetries. Other
graphs are valuable for their applications. If an optimization problem can be described in terms of a graph,
then diverse and powerful tools can be used to solve that problem.
In this chapter, we introduce the basic objects of graph theory, including vertices, edges, degrees, sub-
graphs, paths, cycles, and data structures. Later in the chapter, we describe results about walks on graphs,
including Euler cycles, Hamiltonian circuits, and the number of walks. In Chapter 9, we study a partic-
ular type of graph — trees — which are deceptively simple because they have no loops, but have deep
impacts because they can model hierarchical relationships. We solve optimization (maximization or mini-
mization) problems on graphs in Chapter 10, and consider graphs that can be drawn with no crossing edges
in Chapter 11.

Click on the icon at left or the URL below for a video covering sections 8.1–8.6.
https://www.youtube.com/watch?v=bfr4Y4PIXl4

8.1 Graphs in combinatorics

As Students C and D are finishing lunch at Corbett Hall, a classic Colorado hail storm arrives. Not wanting
to walk, they decide to take the bus. They first need to stop at the library to pick up a textbook. Next,
they need to attend their combinatorics lecture in Weber. Finally, after class, they plan on meeting Student
F at the Recreation Center. What bus routes should Students C and D take to avoid the hail?
Looking at the map can be pretty confusing and they do not want to get lost. We can instead simplify
this map to something called a graph1 .
In this class, a graph is a collection of vertices (dots) and edges (line segments) between the vertices.
The edges describe adjacency (a connection or relationship) between the vertices. In the graph associated
with the bus routes, the locations of the bus stops are represented by vertices and the routes between stops
are represented by edges. An example of two adjacent bus stops are Corbett Hall and Moby Arena.
As a second example of a graph, consider the social network in Figure 8.3. Each vertex corresponds to
a person2 , and an edge between two people represents that they have met before. The one intersection of
edges that does not have a dot means nothing and can be ignored. Suppose you are running for Student
Body President at CSU and want to collect votes. Which person would you want to contact first?
1 The word graph in mathematics can also mean the graph of a function: for example, the graph of the function y = x2 is a

parabola.
2 or at least a mammal

137
Figure 8.1: A map of CSU bus routes. CSU map from http://www.mappery.com/
Colorado-State-University-Map.

Figure 8.2: Bus routes near campus represented as a graph.

8.1.1 Graphs, vertices, edges, and adjacency

Definition 8.1.1. A graph G = (V, E) consists of a finite set V of vertices (or nodes) and a finite set E
of edges, where each edge e ∈ E is of the form e = {u, v} with u, v ∈ V .

The edge {u, v} is often written as uv. Unless mentioned otherwise, it is the same as the edge {v, u} = vu.

Remark 8.1.2. In this book, unless we say otherwise:

1. we do not allow any edge to be a self-loop (an edge uu that starts and ends at the same vertex u);

2. we do not allow more than one edge between any pair of vertices {u, v}.

138
 Figure 8.3: An example of a CSU social network.

Figure 8.4: A graph with 8 vertices and 8 edges.

Indeed, in Definition 8.1.1, the fact that each edge e is a set automatically implies that self-loops are not
allowed and the fact that E is a set automatically implies that multiple edges are not allowed. Later in the
book, we will also consider a more general class of graphs, by allowing multiple edges and self-loops.

Definition 8.1.3. Two vertices, u and v, are adjacent if there is an edge, uv, connecting them. If u and
v are adjacent, we write u ∼ v.

A graph is defined only by its set of vertices and its set of edges; the same graph may be drawn in different
ways. Indeed, the two graphs below are the same graph; they are just drawn differently in the plane.

A labeling of a graph is an assignment of numbers to each vertex. If a graph has n vertices, then then
we will always use the labels 0 through n − 1, where each such integer is used exactly once as a label.

Figure 8.5: We do not allow repeated edges (left) or self loops (right) in our graphs, unless stated otherwise.

139
Exercises
1. Let V be a set of 4 vertices. How many different graphs (V, E) are there if:

(a) the vertices are labeled as V = {v1 , v2 , v3 , v4 }? For example, in this case, the graph with the one
edge v1 v2 is considered different from the graph with the one edge v2 v3 .
(b) the vertices are not labeled? For example, in this case, all graphs having 4 vertices and one edge
are considered the same.

2. What is the largest number of edges that a graph with 5 vertices can have?

3. Let V be a set of 5 vertices. How many different graphs (V, E) with exactly three edges are there if:

(a) the vertices are labeled as V = {v1 , v2 , v3 , v4 , v5 }?

(b) the vertices are not labeled?

8.2 Everyday graphs

Some graphs are used frequently enough that they have names of their own. Here are some examples of
these. Let n ≥ 1.

8.2.1 Complete graphs

Definition 8.2.1. The complete graph Kn is the graph with n vertices having an edge between every
pair of distinct vertices.

Example 8.2.2. Below are the complete graphs, K1 , K2 , K3 and K4 , respectively.

Example 8.2.3. What is the largest number of edges that a graph with 9 vertices can have? In other words,
what is the maximal possible number of bus connections that can exist between the 9 stops in Figure 8.2,
assuming that any two bus stops have at most one bus route between them?

Brainstorm. How many possible edges can we draw on 9 vertices?

9

9


Answer. There are at most possible edges. The graph with all
2 2 edges is the complete graph K9 .


By a similar argument, we can show that Kn has n2 edges.

8.2.2 Path and cycle graphs

Definition 8.2.4. The path graph Pn is the graph with n vertices and n − 1 edges that can be drawn so
that all vertices and edges lie on a straight line.

Example 8.2.5. For instance, the path graph P5 has five vertices edges and four edges.

140
 The bus route in Figure 8.2 contains a path graph P6 connecting from Moby Arena, to Corbett Hall, to
the Library, to the CSU Transit Center, to Engineering, to the MAX Station.

Definition 8.2.6. If n ≥ 3, the cycle graph Cn is the graph with n edges and n vertices such that every
vertex is adjacent to exactly two others.

Example 8.2.7. Cycle graphs are usually arranged in the shape of a circle, as in the case of C6 below.

The green loop, or the Around the Horn bus, in Figure 8.2, connecting Weber to Engineering to the
MAX Station back to Weber, forms a cycle graph C3 inside the bus route map.

8.2.3 Bipartite graphs

The three most popular pizza parlors in Fort Collins are Pizza Casbah, Krazy Karl’s, and Cosmos’s Pizza.
Four student houses, A, B, C, and D often order pizza. Their patterns of pizza ordering can be modeled as
a graph. The graph contains seven vertices — three vertices for the three pizza parlors, and four vertices
for the four student houses. If a house ordered pizza from a particular parlor in the last month, then this
support is modeled by an edge in the graph. (Later, we might also choose to weight the edge by the number
of pizzas ordered from that parlor over the last month.)

A
Pizza Casbah
B
Krazy Karl’s
C
Cosmo’s Pizza
D
Figure 8.6: Bipartite graph with three vertices on the left corresponding to pizza parlors, and four vertices
on the right corresponding to houses. An edge represents an order between that house and that parlor in
the last month.

This graph has a particular structure: there are no edges between two pizza parlors, since a pizza parlor
never orders pizza from another parlor; similarly, there are no edges between two houses, since no house ever
orders pizza from another house!
Many problems in graph theory involve matching two different types of objects, like students with jobs,
or pets with owners.

Definition 8.2.8. A graph G = (V, E) is bipartite if V is the disjoint union of two sets L (left) and R
(right) and every edge connects a vertex in L with a vertex in R.

In other words, there are no edges between two vertices in L and no edges between two vertices in R.

141
Definition 8.2.9. The complete bipartite graph Bn,m is the bipartite graph with n vertices on the left
and m vertices on the right, such that every vertex on the left is adjacent to every vertex on the right.
In the example with three pizza parlors and four student houses, the complete bipartite graph B3,4 would
model the situation in which all four houses have ordered pizza from all three parlors within the last month.

Exercises
1. True or False: A graph with 30 edges has at least 9 vertices.
2. What is the largest n such that the graph in Figure 8.2 contains a cycle graph Cn (no edge can be
used twice)?
3. What is the largest n such that the graph in Figure 8.2 contains a path graph Pn (no edge can be used
twice)?
4. Explain why the complete bipartite graph B1,m is called a fan.
5. How many vertices does Bn,m have?
6. How many edges does Bn,m have?
7. Can you draw B3,3 so that none of the edges cross?

8.3 The degree of a vertex

The students walk around campus, ending at the bus stop at the MAX Station, where they plan to catch a
bus. How many different bus stops can they visit next? Looking at Figure 8.2, they decide they can visit
three stops next: Weber, Engineering, or Horsetooth. In other words, the degree of the CSU Transit Center
vertex is three. The students decide to take the bus to Horsetooth. After dipping their toes in the reservoir,
they realize that they have only one choice for their next stop, since their is only one route from Horsetooth.
In other words, the degree of the Horsetooth vertex is one.
Definition 8.3.1. If v is a vertex in a graph G, the degree of v, denoted deg(v), is the number of edges
adjacent to v.
Given a set of n non-negative numbers, we can ask whether this can be the set of degrees of the vertices
in a graph with n vertices. In the rest of this section, we give some examples and conditions about this
question.

142
Example 8.3.2. Is there a graph on 5 vertices with vertices of degrees 1, 2, 2, 2, 3?
Answer. Yes!

Example 8.3.3. Is there a graph on 9 vertices of degrees 1, 2, 2, 2, 3, 3, 3, 4, 4?

Answer. Yes, Figure 8.2 is one such example!
Example 8.3.4. Is there a graph on 5 vertices with vertices of degrees 0, 1, 2, 3, 4?
Answer. No. The vertex of degree 4 would be connected to every other vertex, and hence there cannot also
be a vertex of degree 0.

Example 8.3.5. Is there a graph on 5 vertices with vertices of degrees 2, 2, 3, 3, 3?

Brainstorm. Try to draw a picture of such a graph!
Answer. No, by the following theorem (since the sum of vertex degrees 2 + 2 + 3 + 3 + 3 = 13 is not even):
Theorem 8.3.6. The sum of the degrees of all vertices in a graph is twice the number of edges. In particular
the sum of all vertex degrees is even, and hence the number of vertices of odd degree must be even.
Proof of Theorem 8.3.6. In the sum of the degrees of all vertices, each edge {u, v} is counted exactly twice:
once in the degree of vertex u, and once in the degree of vertex v.
Example 8.3.7. Consider the following graph, which has 6 edges, and the sum of the vertex degrees is
2 · 6 = 12.
deg 2

deg 3 deg 2

deg 3 deg 2

Theorem 8.3.6 implies that in any graph, the number of vertices of odd degree must be even. We give a
second proof of this consequence. This proof is longer, but illustrates how induction can be used to prove
results in graph theory.
Proof that in any graph G, the number of vertices of odd degree must be even. This proof will use induction
on the number of edges in a graph.
The base case is a graph with no edges (only vertices), and so the number of vertices of odd degree is
zero, which is an even number.
In the inductive step, we will add one edge.3
3 There are a few ways that this is not as precise as it could be. First, we need to prove that we can always build a graph

by adding edges one at a time. Also, in an inductive proof, the number of edges would become arbitrarily large, but that is not
possible unless we allow multiple edges between vertices.

143
Adding an edge changes the degree of two vertices, as described by one of the following three cases:

• {even degree, even degree} → {odd degree, odd degree}.

• {even degree, odd degree} → {odd degree, even degree}.

• {odd degree, odd degree} → {even degree, even degree}.

In each of these cases, the number of vertices with odd degree stays even. Hence in any graph, the number
of vertices of odd degree must be even.

Exercises
1. Is there a graph on 6 vertices with vertices of degrees . . .

(a) 1, 1, 2, 2, 3, 3?
(b) 1, 1, 2, 2, 4, 6?
(c) 1, 1, 2, 3, 3, 5?
(d) 5, 5, 5, 5, 5, 5?
(e) 2, 2, 2, 2, 2, 4?

2. Is there a graph with:

(a) 6 vertices with vertices of degrees 0, 1, 1, 1, 2, 5?

(b) 7 vertices with vertices of degrees 0, 2, 2, 2, 4, 4, 6?

3. True or False: There exists a graph with 7 vertices such the total sum of all vertex degrees is 44.

4. True or False: If a graph has an odd number of vertices, then it has a vertex with even degree.

5. Re-express each of the following assertions as a theorem about degrees in graphs. The theorem should
be of the following form: In any graph G with at least 2 vertices, [some property about vertex degrees
holds]. Then explain why the theorem is true.

(a) At every party there are at least two people who know the same number of other people at the
party.
(b) If one person knows everyone at the party, then it is not possible for another person to know no
one.

6. In the complete bipartite graph Bn,m with n vertices on the left and m vertices on the right, what is
the degree of each vertex on the left? what is the degree of each vertex on the right?

7. Is there a graph on 6 vertices whose vertices have degrees 1,1,2,3,5,5?

8. Is there a graph on 6 vertices whose vertices have degrees 0,1,2,2,2,5?

9. Is there a graph on 6 vertices whose vertices have degrees 2,2,3,3,3,3?

144
8.4 Subgraphs
A recurring theme in mathematics is that whenever you have an interesting object, it is often fruitful to
also consider its subobjects. In this section, we consider subgraphs of a graph. Intuitively, a graph H is a
subgraph of G if each of its vertices is a vertex of G, and if each of its edges is a edge of G.
For example, consider the bus route graph in Figure 8.2. Suppose that the bus route from Moby Arena
to Corbett Hall closes due to construction. Then we obtain the subgraph in Figure 8.7(left), with a single
edge removed.

Figure 8.7: Two subgraphs of the bus route graph, as various edges and vertices get removed.

Suppose furthermore that repairs to the MAX Station need to be made. If the MAX Station closes,
then necessarily the three routes adjacent to the MAX Station close as well, and we obtain the subgraph in
Figure 8.7(right), with one vertex and three more edges removed.
We now formalize what it means for one graph to be a subgraph of another.

Definition 8.4.1. Let G = (V, E) be a graph. A graph G0 = (V 0 , E 0 ) is a subgraph of G if V 0 ⊆ V and

E 0 ⊆ E. A spanning subgraph of a graph G is a subgraph that contains all of the vertices of G.

Example 8.4.2. Let G be the graph drawn in (a). Then (b) is a subgraph of G. However, the object drawn
in (c) is not a subgraph of G because it is not even a graph (it is missing the vertex for one of its edges).
Also, the graph drawn in (d) is not a subgraph of G because it contains an edge that is not in G.

(a) (b) (c) (d)

In Figure 8.8, we draw a graph with three vertices and one edge, along with all of its subgraphs.
Note that a graph G is always a subgraph of itself. Similarly, the empty graph (which has no vertices or
edges) is a subgraph of any graph G.

Exercises
1. How many subgraphs does a labeled cycle C3 have? Assume the three vertices are labelled a, b, and
c. We consider the subgraph with two vertices a, b and a single edge {a, b} to be different from the
subgraph with two vertices b, c and a single edge {b, c}. Hint: count the number of subgraphs with 0
vertices, then the number of subgraphs with 1 vertex, then the number of subgraphs with 2 vertices,
etc.

145
Figure 8.8: A graph G (drawn in the top box) and all of its subgraphs. Each connection represents a
subgraph that is obtained from the one above it by removing a single vertex or edge.

2. How many subgraphs does a labeled cycle of length 4 have? Assume the four vertices are labelled a,
b, c, d. We consider the subgraph with two vertices a, b and a single edge {a, b} to be different from
the subgraph with two vertices b, c and a single edge {b, c}.

3. Explain why every graph with n vertices is a subgraph of Kn .

4. Explain why Pn−1 is a subgraph of Cn

5. Explain why C6 is a subgraph of B3,3 .

6. If Cn is a subgraph of a bipartite graph, explain why n is even.

7. True or False: A labeled graph with n vertices always has at least 2n subgraphs.

8.5 Walks and connected graphs

Click on the icon at left or the URL below for a video covering sections 8.1–8.6.
https://youtu.be/jWZbEoWIQgM

After class in Weber, some of the students need to go to their intramural water polo game. However,
Bus 2 in Figure 8.1 broke down while they were learning about derangements and now it is not available
for service. This means that the three red edges in Figure 8.2 disappear. The students are dismayed to find
that there is no longer a bus route to travel from Weber to Moby Arena.
We say that a graph is connected if you can travel from any vertex of the graph to any other vertex
by walking along edges. For example, when all of the busses are running, the CSU bus system is con-

146
nected. However, if the edge between the CSU transit center and engineering disappears, the graph becomes
disconnected. We make this notion more precise by first introducing the concept of a walk.

Definition 8.5.1. A walk in a graph is a sequence of vertices and edges v0 , e1 , v1 , e2 , v2 , . . . , vk−1 , ek ,

vk such that edge ei connects vertices vi−1 and vi , for 1 ≤ i ≤ k. For example, the sequence of vertices and
edges, v0 , e1 , v1 , e2 , v2 , e3 , v3 , is a walk on the following graph.

Definition 8.5.2. A graph G is connected if for any two vertices u and v of G, there is a walk in G from
u to v.

For example, the graph drawn below is connected. Indeed, no matter which two vertices you consider in
this graph, you can always find a walk between them.

By contrast, the graph drawn below is not connected, because there is no walk between a vertex on the
left side and a vertex on the right side.

It is often more convenient to work with connected graphs than with disconnected graphs. If a graph is
disconnected, it breaks up uniquely into its connected components, which are connected subgraphs that we
define below. It is then possible to work separately with each connected component of a graph.

Definition 8.5.3. A connected component H of a graph G is a maximal subgraph that is connected.

The word “maximal” in the above definition means that there are no connected subgraphs of G that
contain H and that are strictly larger than H. In other words, if H 0 is another connected subgraph of G,
and if H is a subgraph of H 0 , then H = H 0 .

Example 8.5.4. If G is the graph , then the connected components of G are and . Note
that is not a connected component of G because it is not a maximal connected subgraph.

Some of the exercises below illustrate that by deleting an edge from a connected graph, it is possible to
make it disconnected.

147
Exercises
1. What is the smallest number k such that it is possible to make the following graph disconnected by
removing k edges:

(a) K7 ?
(b) C7 ?
(c) P7 ?
(d) B3,4 ?

2. What is the smallest number t such that it is possible to divide the following graph into 3 connected
components by removing t edges:

(a) K7 ?
(b) C7 ?
(c) P7 ?
(d) B3,4 ?

3. By taking away 3 edges from the graph below, what is the maximal number of connected components
that you can make:

(a) K7 ?
(b) C7 ?
(c) P7 ?
(d) B3,4 ?

4. Let G be a graph. Let H1 = (V1 , E1 ) and H2 = (V2 , E2 ) be connected subgraphs with a vertex v
in common. Form their union H = (V, E), where V = V1 ∪ V2 and E = E1 ∪ E2 . Prove that H is
connected

5. Suppose that graph G is connected and contains a cycle. Prove that if an edge from this cycle is
removed from graph G, then the remaining graph is connected.

6. Let u and v be two vertices in a graph G that are not connected by an edge. Show that adding the
new edge uv creates a cycle if and only if u and v are in the same connected component of G.

8.6 Graph complements

Student A decides to host a picnic, with the goal of introducing new people to each other. There are 7
attendees for this picnic: Dr. Pries, Dr. Adams, Dr. Gillespie, Cam the Ram, Student A, Student B, and
Student C. These are the 7 people (or animals) represented by the vertices of the graph in Figure 8.3, or in
the image below on the left. Recall that an edge in this graph represents the fact that the two people at the
endpoints of this edge have met.
The goal of the picnic is to introduce new folks to each other, and so Student A decides to draw a new
graph in which each edge represents that two people have not yet met, and therefore need to be introduced to
each other. This new graph is drawn on the right below; it is the complement of the original social network
graph drawn on the left.

148
 Cam the Ram Cam the Ram
Dr. Pries Dr. Pries

Student C Student C

Student B Student B
Dr. Adams Dr. Adams
Dr. Gillespie

Dr. Gillespie

Student A Student A

Definition 8.6.1. Given a graph G, its complement G has the same vertex set as G, and has an edge
between vertices u and v precisely when G does not.

For example, the figure below has the path graphs P2 through P5 in the left column, and their complement
graphs P2 through P5 in the right column.

An equivalent definition of a graph complement is as follows. Let G = (V, E) be a graph with vertex set
V and edge set E. Let S denote the set of all pairs of vertices in V . Note that E is a subset of S. The
graph complement G can also be defined as G = (V, S − E), where S − E is the set complement as defined in
Definition 2.6.9. Explain why this agrees with the definition of graph complement given in Definition 8.6.1.

Exercises
1. Describe the complement of the following graphs:

(a) C5 ;
(b) B3,4 ;

2. Is the complement of C6 a connected graph?

3. If a vertex v has degree d in a graph G on n vertices, what is the degree of v in the complement of G?

149
 4. Prove that a graph complement G is a subgraph of the original graph G only if G is the complete
graph on its vertex set.
5. For which n, is the complement of Pn a path?
6. For which n, is the complement of Cn a cycle?
7. True or False: Let G and H be graphs. If G is the complement of H, then H is the complement of G.
8. Prove that the complement of Bn,m has two connected components, which are Kn and Km .

8.7 Storage structures for graphs

Given a graph, how do we store it in a computer? Once a graph has been stored, how do we edit it? We
give two possible graph representations to address these questions.
The first way to store a graph is as an adjacency matrix. Given a graph G with n vertices, label its
vertices from 0 to n − 1. The adjacency matrix encoding this graph will be an n × n matrix whose entries
are either 0 or 1. Entry (i, j) of the adjacency matrix is 1 when the edge {i, j} between vertices i and j is
in G, and 0 otherwise. Below is the graph C6 and its adjacency matrix.
0 1 2 3 4 5
2 1 0 0 1 0 0 0 1
1 1 0 1 0 0 0
3 0 2 0 1 0 1 0 0
3 0 0 1 0 1 0
4 0 0 0 1 0 1
4 5 5 1 0 0 0 1 0
There are n2 entries in this matrix that are each 0 or 1. The matrix is symmetric: entry (i, j) equals to
entry (j, i), because the edge {i, j} is the same as the edge {j, i}. Furthermore, each of the n entries along
2
the diagonal is 0, since self-loops are not allowed in graphs. Therefore, we only need to store the n 2−n entries
above the diagonal in order to fully represent the adjacency matrix.
The adjacency matrix depends on the labeling we chose for the vertices. Changing the labeling changes
the adjacency matrix by permuting the rows and the columns.
A second way to store a graph is a list of edges. We store n, the number of vertices, and then maintain
a list of size 2 × e, where e is the number of edges in our graph. Each column of this edge list stores the two
vertices incident to that edge. Below is the edge list for the graph C6 :
0 1 2 3 4 5
1 2 3 4 5 0
In this representation, we need to store 2e integers between 0 and n − 1.
An adjacency matrix has the advantage of being easily editable. Suppose we want to add (resp. remove)
the edge {i, j} from a graph. Then to update the storage structure, we only need to change the entries (i, j)
and (j, i) in the matrix from 0 to 1 (resp. from 1 to 0).
An edge list has the advantage of being a smaller storage method, especially if the graph is sparse,
meaning that the number of edges is small compared with the number of vertices. However, an edge list has
the disadvantage of being harder to edit. For example, to remove an edge, we need to search through the
entire list to find the location of the edge and then shuffle all of the later entries to the left.
For a graph that is a tree (connected graph with no cycles) and has a root, there is another disadvantage
of an edge list. For this kind of graph, it is useful to find all the ‘descendents’ of a vertex v, meaning all
the vertices that are further from the root than v is. When the graph is stored as an edge list, it is very
time-consuming to find all the descendants of a vertex, because you need to make repeated queries. For this
reason, in this context, an edge list is usually replaced by a materialized path. In the storage structure for
a materialized path, the idea is to store the entire walk from the root to each vertex.

150
Exercises
1. Write down the adjacency matrix and the edge list for the graphs below (for some labeling of the
vertices):
(a) K5 ;
(b) P5 ;
(c) C5 ;
(d) B2,3 .
2. Describe the structure of the adjacency matrix and the edge list for the graphs below, for a good choice
of labeling of the vertices:
(a) Kn ;
(b) Pn ;
(c) Cn ;
(d) Bn,m .
3. How many non-zero entries does the adjacency matrix for C4 have?

8.8 Eulerian walks and Hamiltonian cycles

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=oP2paQL8zmM

We are now prepared to return to one of our motivating problems from Chapter 1. Consider the following
map of seven bridges crossing the Pregel River in Königsberg, Prussia, which is today called Kaliningrad,
Russia. Is it possible to take a walk in which you cross each bridge exactly once? You do not need to return
to your starting location.

Euler had the insight to transform this question about bridge crossings into a question about graph theory.
What was truly innovative about this was that no one had ever thought about graph theory abstractly before
then. After this rephrasing, the Königsberg bridge problem can be transformed into the following question:
is there an “Eulerian walk” (which we will later define) in the following graph?

151
 (a) No Eulerian walks. (b) Eulerian walks exist! (c) Eulerian walks exist! (d) No Eulerian walks.
None of them are closed. All of them are closed.

Figure 8.9: Examples of graphs with or without (closed) Eulerian walks.

Note that here we are allowing multiple edges between two vertices in a graph. This happens because
there are multiple bridges between some of the land masses in the picture.
Recall from Section 8.5 that a walk in a graph is a sequence of vertices and edges v0 , e1 , v1 , e2 , v2 , . . . ,
vk−1 , ek , vk such that edge ei connects vertices vi−1 and vi for 1 ≤ i ≤ k.

We say that a walk is closed if it begins and ends at the same location, i.e. if v0 = vk .

Definition 8.8.1. An Eulerian walk through a connected graph G goes through every edge exactly once.

The following theorem is considered to be one of the earliest theorems in graph theory!

Theorem 8.8.2 (Euler, 1736). Let G be a connected graph.

(a) If G has more than two vertices with odd degree, then it has no Eulerian walks.

(b) If G has exactly two vertices of odd degree, then it has an Eulerian walk. Every Eulerian walk starts
and ends at these two vertices.

(c) If G has no vertices of odd degree, then it has an Eulerian walk. Every Eulerian walk is necessarily
closed.

Consider again the graph that results from the Königsberg bridge problem. This graph has four vertices
of odd degree. Since 4 is more than 2, by part (a) of Theorem 8.8.2 this graph has no Eulerian walks. It is
therefore not possible to take a walk in Königsberg that crosses each of the seven bridges exactly once!

Question 8.8.3. Why can’t we have one vertex of odd degree in a graph?

Answer. The sum of all vertex degrees is twice the number of edges (Section 8.3), and hence an even number.
It follows that in any graph, the number of vertices with odd degree is even.

152
 In the next several claims, we prove parts of Theorem 8.8.2.
Claim. If a vertex v has odd degree, then any Eulerian walk must start or end at v. This proves (a) and the
second sentence of (b).

Proof.

If the walk doesn’t start at v, then it:

• enters and leaves v (using up two edges),

• enters and leaves v (using up two edges),

• ...

This continues until there is one remaining edge incident to v (since v has odd degree). The walk then enters
along this edge and cannot leave. Hence if the walk doesn’t start at v, then it must end at v.

Claim. If a vertex v has even degree, then any Eulerian walk either starts and ends at v, or starts and ends
somewhere else. This proves the second sentence of (c).

Ideas towards the proof. Two example walks that start and end at v are drawn as follows:

Two example walks that neither start nor end at v are drawn as follows:

The complete proof is much like that of the previous claim.

153
 We omit the proofs that Eulerian walks exist in cases (b) and (c), although they are very interesting.
The main idea for (c) is that if a walk arrives at a vertex v by one edge, then it is possible to continue the
walk leaving by another edge, thus reducing the number of unused edges at v by 2. There is a clever proof
of (b) using (c); the main idea of this is to add an edge between the two vertices u and v with odd degree
in the graph, then find an Eulerian cycle using (c), and then to adjust the starting point of the cycle to be
u; deleting the edge between u and v produces an Eulerian walk that starts and ends at u and v.
Example 8.8.4. Consider the four connected graphs drawn below. If an Eulerian walk exists, then find
one, and say whether all such Eulerian walks are closed or not! If no Eulerian walk exists, then say why.

(a) (b) (c) (d)

Euler was able to solve the Königsberg bridge problem by transforming it into a question about graph
theory. Though there are efficient techniques for solving many questions in graph theory, it is important to
remark that not all graph theory questions have easy answers!
One such question is as follows: does there exist a walk in a graph that passes through each vertex exactly
once? Such a walk is called a Hamiltonian cycle.
Definition 8.8.5. A Hamiltonian cycle in a connected graph G is a closed walk that passes through each
vertex exactly once.
Here are two examples of Hamiltonian cycles in graphs, where the Hamiltonian cycle is drawn in red.

By contrast, the Petersen graph drawn below has no Hamiltonian cycle.

154
 In general, is very hard to determine whether a graph has a Hamiltonian cycle or not! Indeed, determining
whether a graph has a Hamiltonian cycle or not is NP-complete, which is a very difficult class of problems
in theoretical computer science.

Exercises
1. Does the complete graph K5 have an Eulerian walk?

2. Does the complete graph K5 have an Eulerian cycle (that is, an Eulerian walk that ends at the same
vertex it started at)?

3. The complete graph K6 have an Eulerian walk?

4. Does the complete bipartite graph B3,3 have an Eulerian walk?

5. Does the complete bipartite graph B3,3 have a Hamiltonian walk?

6. True or False: The complete graph K5 on 5 vertices contains a closed Eulerian cycle.

7. True or False: The complement of the path graph on 5 vertices has an Eulerian walk.

8. Solve Example 8.8.4.

9. True or False: There exists a connected graph G with vertices of degrees 2, 2, 2, 3, 3, 4, 4, 4, 6 that
contains an Eulerian walk starting at a vertex of degree 2 and ending at a vertex of degree 4.

10. True or False: There exists a graph G containing both closed and non-closed Eulerian walks.

11. True or False: There exists a connected graph G with vertices of degrees 3, 3, 4, 4, 4, 4, 4, 6 that
contains an Eulerian walk starting at a vertex of degree 4 and ending at a vertex of degree 4.

12. True or False: There exists a connected graph G with vertices of degrees 2, 2, 2, 2, 3, 4, 4, 6 that
contains a non-closed Eulerian walk starting at the vertex of degree 3.

13. Prove that a graph G with n vertices has a Hamiltonian cycle if and only if Cn is a subgraph of G.

14. For which pairs (n, m) does the complete bipartite graph Bn,m contain a Hamiltonian cycle?

8.9 Investigation: The number of walks

This section is about counting the number of walks between two vertices.

Example 8.9.1. Your hamster’s cage looks like this graph:

Let’s label the vertices of this graph by 1 (for the vertex of degree 1), 2 (for the vertex of degree 3, and then
3, 4, 5 around the cycle (in either direction).
Every minute, the hamster walks along an edge to an adjacent vertex (the hamster never remains put at
the same vertex and is always on the move). Pick a starting vertex i and an ending vertex j and a number
of minutes N . Our goal is to count the number of ways that the hamster can travel from vertex i to vertex
j in exactly N minutes; let’s call that number hi,j (N ).

155
 For example, let i = j = 1. Then h1,1 (N ) counts the number of ways the hamster can travel from vertex
1 back to vertex 1 in N minutes. When N = 1, then h1,1 (1) = 0 because after one minute, the hamster is
at vertex 2 and has not returned to vertex 1.
When N = 2, then the possible hamster routes are
1 7→ 2 7→ 1, 1 7→ 2 7→ 3, 1 7→ 2 7→ 5.
This shows that h1,1 (2) = 1 and h1,3 (2) = 1 and h1,5 (2) = 1. It also shows that h1,2 (2) = 0 and h1,4 (2) = 0.
When n = 3, the possible hamster routes starting at vertex 1 are:
1 7→ 2 7→ 3 7→ 4, 1 7→ 2 7→ 3 7→ 2, 1 7→ 2 7→ 5 7→ 4, 1 7→ 2 7→ 5 7→ 2, 1 7→ 2 7→ 1 7→ 2.
In particular, this shows that h1,1 (3) = 0, because the hamster is not at vertex 1 after 3 minutes.
In the hamster example above, there is a lot of information to keep track of: the starting position i, the
ending position j, and the number of minutes N . To keep track of all that information, for each positive
integer N , we will make a 5 × 5 matrix MN , where the entry in the ith row and jth column is hi,j (N ). For
example,  
0 1 0 0 0
 1 0 1 0 1 
 
M1 =  0 1 0 1 0 .

 0 0 1 0 1 
0 1 0 1 0
By definition, M1 is the adjacency matrix, because the i, jth entry is 1 exactly when the ith and jth vertices
are adjacent.
With a lot of work, we compute the matrix for N = 2:
 
1 0 1 0 1
 0 3 0 2 0 
 
M2 =   1 0 2 0 2 .

 0 2 0 2 0 
1 0 2 0 2
For example, the first row of M2 contains the values h1,j (2) that we computed before. Notice that the entries
in the diagonal of M2 are the degrees of the vertices. This makes sense, because the number of walks that
leave vertex i, go to another vertex j and then return to vertex i is exactly the degree of vertex i.
We could try to compute M3 by hand, but it would be very time-consuming. We would need to count
the number hi,v (2) of routes the hamster could take in two minutes, organized by their ending vertex v, and
then check if the hamster can travel from v to the endpoint j along an edge. For example, when i = 1 and
j = 4, we would compute the following table.
v 1 2 3 4 5
h1,v (2) 1 0 1 0 1
v adjacent to 3 no no yes no yes
We ignore the 2-minute routes ending at the vertices v = 1, 2, 4 and keep track of the 2-minute routes ending
at the vertices v = 3, 5, allowing us to compute:
h1,4 (3) = h1,1 (2) · 0 + h1,2 (2) · 0 + h1,3 (2) · 1 + h1,4 (2) · 0 + h1,5 (2) · 1
= 1 · 0 + 0 · 0 + 1 · 1 + 0 · 0 + 1 · 1 = 2.
As a sanity check, we can check that there are exactly two 3-minute routes starting at i = 1 and ending at
j = 4, namely:
1 7→ 2 7→ 3 7→ 4, 1 7→ 2 7→ 5 7→ 4.
So far, we have computed only one entry of the matrix M3 !
Luckily there is a faster way to compute the matrix M3 , by using matrix multiplication from linear
algebra. The first thing we do is to use Sage to check that M2 is the product of the matrix M1 with itself:

156
M1 = Matrix([[0,1,0,0,0], [1,0,1,0,1], [0,1,0,1,0], [0,0,1,0,1], [0,1,0,1,0]]);

M2 = Matrix([[1,0,1,0,1], [0,3,0,2,0], [1,0,2,0,2], [0,2,0,2,0], [1,0,2,0,2]]);

Then
M2==M1^2
gives an output of TRUE.
The cube of M3 can be computed on SAGE with the command M 3 = M 13 and it gives that
 
0 3 0 2 0
 3 0 5 0 5 
 
M3 = 
 0 5 0 4 0 .

 2 0 4 0 4 
0 5 0 4 0

It is time-consuming, but you can check that the i, jth entry of this matrix M 3 equals hi,j (3). For example,
the entry in the first row and fourth column is 2 and we computed that h1,4 (3) = 2.
The hamster example generalizes to this very important theorem.
Theorem 8.9.2. Let G be a labeled graph with n vertices. Let MN be the n × n matrix whose i, jth entry is
the number of walks in G of length N starting at vertex i and ending at vertex j. Then M1 is the adjacency
matrix of G and MN = M1N . In other words, the number hi,j (N ) of walks in G starting at vertex i and
ending at vertex j that have exact length N equals the entry in the ith row and jth column of the N th power
of the matrix M1 .
Proof. We provide only the main ideas in the proof. When N = 1, then the i, jth entry of M1 is the number
of 1-edge routes from vertex i to vertex j. This number equals 1 if vertices i and j are adjacent and equals
0 otherwise. That is the same as the i, jth entry of the adjacency matrix.
We will now prove the result by induction on N , with the case N = 1 being the previous paragraph. Let
N ≥ 2. The inductive hypothesis is that the i, vth entry of A = M1N −1 equals the number of number of
walks of length N − 1 starting at vertex i and ending at vertex v; this hypothesis holds for all n2 entries of
the matrix A, so all integer values of i and v such that 1 ≤ i ≤ n and 1 ≤ v ≤ n. Let B = M1 . We need to
show that the i, jth entry of the matrix multiplication M1N = A × B is the number of walks in G of length
N starting at vertex i and ending at vertex j.
This will rely on the following fact from linear algebra. Suppose A and B are both n × n matrices. By
matrix multiplication, if C = A × B, then C is also an n × n matrix. To find the entry ci,j in the ith row
and jth column of C, we take the dot product of the ith row of A with the jth column of B. In other words,
if the ith row of a is (ai,1 , . . . , ai,n ) and the jth column is (the transpose of) (b1,j , . . . , bn,j ), then

ci,j = ai,1 · b1,j + · · · + ai,n · bn,j .

So, fixing i and j, we need to see why this dot product equals the number of walks in G of length N
starting at vertex i and ending at vertex j. We separate these walks into groups, based on the last vertex
v in the walk which occurs before returning to the vertex j. If the last vertex of the walk (before returning
to j) is v, then this walk separates into (i) a walk in G of length N − 1 from i to v, and then (ii) a walk
of length 1 from v to j. The number of walks of type (i) is ai,v , the i, vth coefficient of A, by the inductive
hypothesis. The number of walks of type (ii) is 1 if v and j are adjacent in G and is 0 otherwise; this is bv,j ,
the v, jth entry in B. So the number of walks of length N from i to j such that v is the last vertex before j
is ai,v · bv,j , which equals ai,v if v and j are adjacent and which equals 0 if v and j are not adjacent. Since
1 ≤ v ≤ n, the total number of walks is
X n
ai,v · bv,j .
v=1

By the rules of matrix multiplication, this equals the entry ci,j of C = M1N .

157
 In fact, Theorem 8.9.2 is true even when the graph has a self-loop or directed edges (meaning edges that
can be traveled in only one direction). In the next example, we show a graph where the Fibonacci numbers
occur as the number of walks.
Example 8.9.3. Consider the directed graph drawn below.

u v

How many routes of length N are there from u to v?

Brainstorm. We brainstorm by making a table of the number of routes of length N from u to v, as n varies
from 0 to 5.
N 0 1 2 3 4 5
# ways 0 1 1 2 3 5

picture
We guess that the number of routes of length N from u to v is the N -th Fibonacci number, FN .
Answer. We give a proof to show that our above guess is correct.
Let PN be the number of routes of length N from u to v.
Note P0 = 0 and P1 = 1 (or, if you prefer, note P1 = 1 and P2 = 1).
Note PN +1 = PN + PN −1 , since every path of length N + 1 starts with either

• , after which there are PN continuing paths, or

• , after which there are PN −1 continuing paths.
Since PN satisfies the same recurrence relation as the Fibonacci numbers Fn , and since P0 = F0 and
P1 = F1 (or, if you prefer, P1 = F1 and P2 = F2 ), this shows PN = FN for all N .
Here is another way to solve this problem using Theorem 8.9.2. Let u be the first vertex and v be the
second vertex. The adjacency matrix for this graph is
 
1 1
M1 = .
1 0
(The self-loop from u to u leads to the non-zero entry in the diagonal of M1 ). By Theorem 8.9.2, the number
of routes from u to v of length N is the entry in the first row and second column of M1N . Let FN be the
N th Fibonacci number. For N ≥ 1, we claim that
 
N FN +1 FN
M1 = .
FN FN −1
We prove this by induction.

158
Proof. When N = 1, then the claim is true because F0 = 0, F1 = 1, and F2 = 1. The inductive hypothesis
is that  
N −1 FN FN −1
M1 = .
FN −1 FN −2
Then we compute M1N using matrix multiplication:

M1N = M1N −1 × M1
   
FN FN −1 1 1
= ×
FN −1 FN −2 1 0
 
FN + FN −1 FN
=
FN −1 + FN −2 FN −1
 
FN +1 FN
= .
FN FN −1

This ends the proof of the inductive step.

Exercises
1. Let G be the complete graph K4 , with labeled vertices.

(a) Find the adjacency matrix M1 of G.

(b) Using SAGE, compute Mn = M1n for n from 2 to 6 but do not write down the answer.
(c) Find the number of routes of length 6 that start and end at vertex 1.
(d) Find the number of routes of length 6 that start at vertex 1 and end at vertex 2.
(e) Explain why it makes sense that all the entries on the diagonal of Mn are the same number d and
all the entries that are not on the diagonal of Mn are the same number e.
(f) If n is even, what do you conjecture that d − e always equals?
(g) If n is odd, what do you conjecture that d − e always equals?

2. Let G be the cycle graph C4 , with labeled vertices.

(a) Find the adjacency matrix M1 of G.
(b) Using SAGE, compute Mn = M1n for n from 2 to 6 but do not write down the answer.
(c) Find the number of routes of length 6 that start and end at vertex 1.
(d) Find the number of routes of length 6 that start at vertex 1 and end at vertex 2.
(e) Half the entries of Mn are 0. Explain why this makes sense - your answer should split into two
cases depending on whether n is even or odd.
(f) Make a conjecture about which non-zero number appears in the other half of entries of Mn .
3. Let G be the path graph P4 , with labeled vertices.
(a) Find the adjacency matrix M1 of G.
(b) Using SAGE, compute Mn = M1n for n from 2 to 6 but do not write down the answer.
(c) Half the entries of Mn are 0. Explain why this makes sense - your answer should split into two
cases depending on whether n is even or odd.
(d) Explain what each of the non-zero entries in M6 means.
(e) Explain why it makes sense that the non-zero entries in the center of M6 are larger than the
others.

159
4. Show there are 149 routes from vertex 0 to vertex 9 in the following directed graph. Paths in this
directed graph are only allowed to cross an edge in the direction indicated by the arrow on that edge.

0 1 2 3 4 5 6 7 8 9

0 1 2 3 4 5 6 7 8 9

1 way 1 way 2 ways 4 ways 7 ways 13 ways 24 ways 44 ways 81 ways 149 ways

160
Chapter 9

Trees

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/cbfz9xeByok

Trees are in some sense the “simplest” of all possible graphs, but do not underestimate their usefulness!
Trees can be used to represent hierarchical structure, such as in a family tree of ancestors and descendants.
In computer science, trees are a powerful data structure, and they can be used to implement tasks such as
searching or sorting with faster times than naı̈ve implementations allow.
For example, suppose you are given a list of n numbers. You sort these numbers from smallest to largest,
and then you would like to store them in a way which

• does not require much space,

• allows you to search for a specific item in the list,

• makes it easy to add new items to the sorted list, and

• allows you to easily remove items from the list.

1 6

0 3 7

2 4 8

A binary search tree, such as the one drawn above, is a data structure that is very efficient for these tasks.
Note that each arrow pointing down and to the left points to a smaller number, and each arrow pointing
down and to the right points to a larger number. The amount of space a binary search tree requires is O(n),
where n is the number of vertices. Searching, adding, and deleting numbers are all operations of average
running time O(log n). Here O is the “big-O” notation used in computational complexity; these average
operation times are considered to be quite fast.
For example, to add the number 5.5 to the tree, we would start at the top vertex, the root, which in this
case is 5. Since 5.5 > 5, we proceed down and to the right. We get to the vertex 6, and since 5.5 < 6, we
insert a new vertex labeled 5.5 to the bottom left of vertex 6, along with a directed arrow from 6 to 5.5.

161
If we were later asked to verify whether 5.5 was present in our search tree or not, we would take the same
sequence of steps.
How was this search tree created? In particular, why have we chosen vertex 5 as the root of this tree?
This tree was created because we were handed the numbers 5, 1, 0, 6, 3, 2, 4, 7, 8, in that order, and asked
to store them as a binary search tree. Since we were handed 5 first, we made it the root (the top vertex). As
1 came next, and is less than 5, we added 1 to the bottom left of 5, connected by an arrow. Similarly, since
0 came next, and is less than 5 and less than 1, we added 0 to the bottom left of 1. Next comes 6, which is
more than 5, and so we added 6 to the bottom right of 5. Next comes 3, which is less than 5 and more than
1, and hence we add it to the bottom right of 1. We proeceed in this manner to create the entire tree.
In this chapter, we give a mathematical introduction to trees.

9.1 The language of trees

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/0J4Hc5Iy0T0

Trees are the simplest possible graphs.

Definition 9.1.1. A graph G = (V, E) is a tree if it is connected and contains no cycle as a subgraph.
Because a tree is connected, there is a walk between every pair of vertices. Because a tree has no cycles,
for every pair of vertices there is a unique walk between them that does not backtrack. This is one reason
why trees are the most simple graphs.
Below we show four examples of trees.

We also draw four non-examples of trees: the first three graphs are not trees because they contain cycles,
and the fourth graph is not a tree because it is not connected.

Definition 9.1.2. A leaf of a tree is a vertex of degree 1.

Let G = (V, E) be a connected graph. While keeping the vertex set V fixed, if we add an edge to G,
then it will still be connected. In contrast, while keeping the vertex set V fixed, if we remove an edge from
G then it might become disconnected. Trees can be described as the smallest possible connected graphs on
a given vertex set, as we make precise in the following theorem.

162
Theorem 9.1.3. A graph G is a tree if and only if it is connected, but deleting any edge makes it disconnected.
Proof. For the forward direction, let G be a tree. Then G is connected by definition. If removing some edge
uv produced a connected graph G0 , then the path from u to v in G0 along with the edge uv would create a
cycle in G, contradicting the fact that G is a tree. Hence removing any edge from G makes it disconnected.
For the reverse direction, to show G is a tree we must show it is cycle-free (we already know it is connected
by assumption). If G had a cycle, then removing any edge from this cycle would produce a connected graph,
which is a contradiction. Hence G is cycle-free, and thus a tree.
Similarly, if a graph G has no cycles, then the graph remains cycle-free if we remove edges. By contrast,
adding edges can certainly make a formerly cycle-free graph now have cycles.
Theorem 9.1.4. A graph G is a tree if and only if it contains no cycles, but adding any new edge creates
a cycle.
We omit the proof; it is included as one of the exercises in this section.
Therefore trees achieve a sweet-spot. Among all connected graphs, trees are the minimal ones, in the
sense that removing any edge from a tree makes it disconnected. Among all graphs without cycles, trees are
the maximal ones, in the sense that adding any edge to a tree creates a cycle.
If G is a connected graph with n vertices, then a spanning tree of G is a subgraph of G that is a tree
with n vertices. There can be many spanning trees in a graph.

Exercises
1. Show that every tree is bipartite.
2. Find an example of a bipartite graph that is not a tree.
3. How many edges do you need to remove from the complete bipartite graph B4,5 in order to make it
into a tree?
4. How many edges do you need to remove from P7 to divide it into 3 connected components?
5. Let T be a tree with 10 vertices. How many non-zero entries does the adjacency matrix for T have?
6. Let T be a tree with 10 vertices. To build an edge list for T, how many numbers between 1 and 10
need to be stored?
7. Show that both B1,6 and P7 are spanning trees inside K7 .
8. Show that both B1,n−1 and Pn are spanning trees inside Kn .
9. Show using strong induction that every connected graph contains a spanning tree.
10. Show that a graph G is a tree if and only if it contains no cycles, but adding any new edge creates a
cycle. This Theorem 9.1.4.

9.2 The number of edges in a tree

There are many different trees that one can draw with 14 vertices; some of them are drawn below.

163
Despite this wide variety, surprisingly, every tree with 14 vertices has exactly 13 edges. This is generalized
by the following theorem.

Theorem 9.2.1. Every tree with n vertices has n − 1 edges.

Proof. We give a proof using strong induction. As the base case, note that every tree with n = 1 vertex has
no edges.
For the inductive step, let n ≥ 2. Suppose that every tree with m < n vertices has m − 1 edges; we must
prove that every tree with n vertices has n − 1 edges.
Let G be an arbitrary tree with n vertices. Since G is connected with n ≥ 2, there must be at least one
edge e = v1 v2 in G. By Theorem 9.1.3, removing the edge e leaves a graph G0 that is disconnected, as shown
below. Also G0 has exactly two connected components because adding one edge can only create a connection
between two parts.

e
v1 v2

Let the connected component containing vertex v1 have m1 vertices, and let the connected component
containing vertex v2 have m2 vertices. (In the diagram above, m1 = 5 and m2 = 4.) Note m1 + m2 = n.
Since G has no cycles, the connected component containing vertex v1 has no cycles, and is therefore a tree.
Furthermore, since this tree has m1 < n edges, by our inductive assumption the connected component with
vertex v1 has m1 − 1 edges. Similarly, the connected component containing vertex v2 is a tree with m2 − 1
edges. Therefore, when we add edge e = v1 v2 back in, we see that the number of edges in tree G is

(m1 − 1) + (m2 − 1) + 1 = m1 + m2 − 1 = n − 1.

Hence we are done by strong induction.

Example 9.2.2. Each of the below trees have their number of edges equal to their number of vertices minus
one, as we know must be the case from Theorem 9.2.1.

6 vertices 7 vertices 8 vertices

5 edges 6 edges 7 edges

A forest is a graph whose connected components are all trees. The image below is of a single forest,
which consists of six trees. Note that one tree consists of only a single vertex, in the bottom right.

164
 The following theorem is a generalization of Theorem 9.2.1; let’s not miss the forest for the trees.

Theorem 9.2.3. Every forest with n vertices and m connected components has n − m edges.

Since every tree is a graph, the storage methods for graphs in Section 8.7 can also be used to store trees.
However, since the structure of trees is very specific, these storage structures for graphs are less efficient
when storing trees. Recall from Section 8.7 that a labeled graph with n vertices can be stored as an n × n
adjacency matrix with binary entries, either 0 or 1 depending on whether an edge is present or not. Such a
2
representation can store 2(n −n)/2 different labeled graphs with n vertices. But as we will learn in Cayley’s
theorem in Section 9.3, the number of labeled trees on n vertices is a lot smaller, only nn−2 . Similarly, recall
from Section 8.7 that a labeled graph with n vertices can be stored as a 2 × e list of edges. Since a tree
with n vertices has n − 1 edges, a tree can be stored as a 2 × (n − 1) list of edges, i.e. as 2(n − 1) entries in
{0, 1, . . . , n − 1}. This representation is closer to being efficient for trees, while still not perfectly efficient,
since we could store n2(n−1) different things, even though there are only nn−2 labeled trees with n vertices.
Binary search trees, which we described briefly at the very beginning of this chapter, are reasonably
efficient storage structure for labeled trees. Binary search trees are also easy to edit, say by adding or
removing nodes from the tree. Can you think of other good storage structures for trees?

Exercises
1. If G is a connected graph with n vertices and n − 1 edges, show that G is a tree.

2. How many edges need to be removed from K7 to create a spanning tree?

3. How many edges need to be removed from Kn to create a spanning tree?

4. How many edges need to be removed from B3,4 to create a spanning tree?

5. How many edges need to be removed from Bn,m to create a spanning tree?

6. True or False: There exists a tree with 6 vertices of degrees 1, 1, 2, 2, 3, 3.

7. For a tree with n vertices, what is the sum of the degrees of the vertices?

8. Prove Theorem 9.2.3 using the following strategy. Suppose G is a forest with n vertices and m connected
components. For i = 1, . . . , m, let connected component i have ni vertices. So n1 + . . . + nm = n.
Apply Theorem 9.2.1 on each connected component of G. How many edges do you get in total?

165
9.3 Labeled trees: Cayley’s theorem

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=Z15ofSdio3I

A labeling of a tree is an assignment of numbers to each vertex. If a tree has n vertices, then then we
will always use the labels 0 through n − 1, where each such integer is used exactly once as a label.
For example, the below two trees are different as labeled trees.
0 0

1 2 1 2

3 4 3 4

By contrast, the below two trees are the same as unlabeled trees!

How many different labeled trees with n are there, as a function of n? Let’s begin by doing some counts
for n small.
There is only a single labeled tree on n = 1 vertex — the single vertex, equipped with label 0.

There is only a single labeled tree on n = 2 vertices — an edge with its two vertices labeled 0 and 1.
Note that swapping these vertex labels still results in the same labeled tree!

0 1

There are three labeled trees on n = 3 vertices, drawn below.

0 0 0

1 2 1 2 1 2
With work, we can count the number of labeled
trees on n = 4 vertices. By Theorem 9.2.1 a tree on 4
vertices has 3 edges. As we see below, there are 42 = 6 possible locations for these 3 edges.

0 1

2 3
6


Hence there are 3 = 20 possible labeled graphs with 3 edges. Of these 20 labeled, all except for the
following 4 are trees.

166
 0 1 0 1 0 1 0 1

2 3 2 3 2 3 2 3

Hence there are 16 labeled trees on n = 4 vertices, as counted below.

Let L(n) be the number of labeled trees on n vertices. We computed that L(1) = 1, L(2) = 1, L(3) = 3,
and L(4) = 16. It is not easy to guess a formula for L(n) in general given the limited number of computations
we have done so far. But a formula is known, and it is included in Cayley’s theorem.
Theorem 9.3.1 (Cayley’s Theorem). The number of labeled trees on n vertices is L(n) = nn−2 .
The proof of Cayley’s theorem is much deeper than a simple count. To see this, try to expand the
methods we used to compute L(4) to compute L(5) or L(6).

Exercises
1. How many trees on 5 labeled vertices are there?
2. How many rooted trees on 5 labeled vertices are there?
3. How many doubly rooted trees on 5 labeled vertices are there?
4. Some of the problems below are about the NCAA basketball tournament. This is a single-elimination
tournament with 64 teams; it may be viewed as a rooted binary tree with 64 leaves. An unfilled NCAA
bracket is a partially labeled binary tree with 64 leaves, where the only labeled vertices are the 64 leaf
vertices in Round 0 (each labeled with one of the 64 teams in the tournament). To fill out a bracket
you must label the remaining vertices with which team you think will win each game.

(a) Explain why a filled out NCAA bracket is not a labeled tree.
(b) Show that the NCAA tournament has 63 games, as follows. Pretend each team brings one new
basketball to the tournament. One basketball is used in each match. The losing team departs
with the used ball, and the still-unused ball is taken by the match’s winner to the next round.
How many basketballs remain unused at the end of the tournament? How many basketballs must
have been used? Hence, how many games must have been played?
(c) How many ways are there to fill out the bracket? (The labels on the 64 leaf nodes are fixed; filling
out the bracket means choosing a winner for each game.)
(d) Draw all 16 labeled trees with 4 vertices.
(e) Read the proof of Cayley’s theorem at https://golem.ph.utexas.edu/category/2019/12/a_
visual_telling_of_joyals_pro.html

9.4 Binary trees

A binary tree is a recursive tree-like structure defined as follows.
Definition 9.4.1. A binary tree is either a single vertex with no edges, or a graph on more than one
vertex consisting of a root vertex v and two edges, L and R, each connecting v to the root vertex of a
smaller binary tree.
Using this definition, we can build up binary trees recursively as follows. The single vertex graph:
•
is a binary tree by the initial condition case of the recursive definition. Then, we can join two single vertex
graphs (which we label u and w) to a new root vertex v to obtain the following graph:

167
 Here, we draw edges L and R as going down-and-left and down-and-right from v, and we call u and w
the left child and right child of v, and v the parent of u and w. We can then build up more binary trees
by combining the ones we have constructed so far. For instance, we could have the left and right children of
the root both connect to the above three-node graph, or just one of the children being the above three-node
graph and the other being a single vertex. This gives us three more binary trees:

Because these trees are generated recursively and any new edge is marked as a left edge or a right edge
with one vertex being the “parent” and the other the “child”, the notion of left child and right child is well
defined for any vertex in a binary tree.
Using the notion of left and right children, a more informal alternate definition of a binary tree may be
stated as follows.

Definition 9.4.2 (Informal definition of binary tree). A binary tree is a graph formed by starting with
a root vertex at the top, and drawing edges and vertices downwards to the left or right in such a way that
every vertex either has two children (a left and right child) or no children.

9.4.1 Tournaments and Catalan numbers

One application of binary trees is as a way to model tournaments in which players face off in pairs, and
the winner advances to the next rounds. In the diagram below, we can think of the labels on the leaves as
the initial “seeding” of the tournament, where it is determined who will face off against one another first.

3 1 5

2 4

Suppose it is a chess tournament, so if a left child and right child of a node face off against each other,
we assign the left child to be the player who goes first, and the right to be the player who goes second (thus
making the distinction between left vs right important). Suppose the numbered labels are some measure of
skill, so that larger numbers will likely beat smaller numbers. Then in the first round of the tournament
drawn above, 2 and 4 face off against each other and 1 and 5 face off against each other, and 4 and 5 advance
to the appropriate parent nodes:

168
 5

4
3 1 5

2 4
Then, 4 defeats 3 and advances, and finally 5 defeats 4 in the last round and advances to the top node,
being declared the winner.

4 5

4
3 1 5

2 4
This was not the only tournament we could have seeded for 5 players! How many possible ways can we
set up such a tournament with n players? In terms of binary trees, this boils down to the following question.
Question 9.4.3. How many binary trees have exactly n leaves?
For now we will put aside the question of labeling the leaves with numbers, and just answer the question
above for unlabeled binary trees with n leaves.
Theorem 9.4.4. The number of binary trees with n leaves is the (n − 1)st Catalan number Cn−1 .
Proof. Let Bn be the number of binary trees with n leaves. We simply need to show that the num-
bers B1 , B2 , B3 , . . . satisfy the same recursion as the one that defines the sequence of Catalan numbers
C0 , C1 , C2 , . . ..
We first check the initial condition: we have B1 = 1 since there is only one binary tree with one leaf, and
C0 = 1, so B1 = C0 .
We now need to show the numbers Bn satisfy the Catalan recursion. To count the number of binary trees
on n + 1 leaves, note that any such tree has some number k ≥ 1 of leaves on the left subtree branching from
the root, and the other n + 1 − k on the right subtree. We break into cases based on the value of k: if k = 1
there are B1 Bn possibilities, if k = 2 there are B2 Bn−1 possibilities, and so on. Adding these together, we
have
Bn+1 = B1 Bn + B2 Bn−1 + B3 Bn−2 + · · · + Bn B1 .
Comparing this to the recursion

Cn = C0 Cn−1 + C1 Cn−2 + · · · + Cn−1 C0

we see that shifting all the indices up by 1 and replacing C by B does indeed make the recursions match.
Therefore Bn = Cn−1 for all n ≥ 1.
Let’s see how this applies to our tournaments. There are C4 binary trees with 5 leaves. Recall the explicit
formula for Catalan numbers that we derived in Section 6.6:
 
1 2n
Cn = .
n+1 n

169


Using this formula, we have C4 = 51 84 = 14. Finally, given a binary tree with 5 leaves, if we have 5 different
players, the number of ways we can choose who labels which leaf is just the number of ways of rearranging
those players, which is 5! = 120. Therefore there are 14 · 120 = 1680 different tournaments you can design
for the 5 players!

9.4.2 At-most binary trees

It is somewhat harder to count binary trees according to how many vertices they have, rather than how
many leaves. However, there is a nice answer if we modify the definition slightly.

Definition 9.4.5. An at-most binary tree is a tree constructed recursively starting from a root node at
the top, where every node has either a left child only, a right child only, both a left and right child, or neither.

Here are all of the at-most binary trees on 3 vertices:

Notice that there are 5 of them, which is a Catalan number. Indeed, the Catalan numbers arise here as well.

Theorem 9.4.6. The number of at-most binary trees on n vertices is the nth Catalan number Cn .

Proof. Let An be the number of at-most binary trees on n vertices. We will consider the empty tree as being
an at-most binary tree, so that A0 = 1. This is equal to C0 , so the initial condition of the Catalan recursion
is satisfied.
Now, we wish to show that the sequence of numbers An satisfies the same recursion as the Catalan
numbers. If n ≥ 1 there is a root vertex in any at-most binary tree on n vertices, and this vertex has either
a left child only, a right child only, both, or neither. In all cases, both the left and the right branches from
the root form a (possibly empty) at-most binary tree, where there are a total of n − 1 vertices in the left and
right trees.
If there are k vertices on the left, and n − 1 − k on the right, there are therefore Ak An−1−k possible trees
of this form. Summing over all k, we have

An = A0 An−1 + A1 An−2 + A2 An−3 + · · · + An−1 A0 ,

and so indeed the recursion is satisfied.

9.4.3 Investigation: Increasing binary trees and permutations

At-most binary trees, with appropriate vertex labelings, can give us a new way of thinking about permuta-
tions.

Definition 9.4.7. An increasing binary tree is an at-most binary tree along with a labeling of its vertices
by the numbers 1, 2, . . . , n such that the labels increase as one reads any downwards path in the tree.

Warning: An increasing binary tree may not be an actual binary tree, because it only needs to be an
at-most binary tree!
Here is an example of an increasing binary tree.

170
 1

3 2

6
4 5

7
This tree can be assigned to a permutation as follows. The 1 at the root corresponds to the 1 in the
permutation; then, the left subtree consists of the entries to the left of the 1 and the right subtree consists
of the entries to the right of the 1 in the permutation. So we know that the 1 has four numbers before it
and two after it in the permutation:

, , , , 1, ,
Then, we repeat this analysis on the right and left subtrees; the 3 is at the top of the left subtree, with
one entry to its left and two to its right, so in the permutation it is in the second position. Similarly the 2
is just after the 1:
, 3, , , 1, 2,
We then continue this process on the subtrees whose top nodes are 4, 6, and 5, to obtain the permutation

4, 3, 7, 6, 1, 2, 5

To show that this correspondence gives a bijection, we need to show that we can reverse it. We illustrate
this by continuing with this example. Starting from the permutation 4, 3, 7, 6, 1, 2, 5, can we reconstruct the
original tree? Indeed, we must place 1 at the top, and then its left and right children must be the smallest
element left of 1 and the smallest element right of 1 respectively. The left and right children of each successive
node is similarly determined.
We have shown:
Theorem 9.4.8. Increasing binary trees on n labeled vertices are in bijection with permutations of the set
{1, 2, 3, . . . , n}.
Since there are n! permutations, this immediately allows us to count the increasing binary trees.
Corollary 9.4.9. There are exactly n! increasing binary trees on n vertices.

Exercises
1. Draw all 14 binary trees having 5 leaves.
2. How many binary trees are there on 7 labeled leaves?
3. Draw all 14 at-most binary trees having 4 vertices.
4. How many unlabeled binary trees have exactly 4 leaves?
5. How many leaf-labeled binary trees are there on labeled leaves 1,2,3,4?
6. How many unlabeled at-most binary trees on 4 vertices are there?
7. How many at-most binary trees have 6 vertices?
8. Draw the permutation 2, 1, 5, 3, 4 as an increasing binary tree.

171
 9. Draw all 24 increasing binary trees on 4 vertices.
10. Define a weakly increasing binary tree to be an at-most binary tree with every vertex labeled by
a positive integer, such that for any vertex labeled v with left child w and/or right child u, we have
w > v and u ≥ v. In other words, we can now have repeated numbers, but ties can only go rightwards.
So the tree at left is a weakly increasing binary tree, but the one at right is not:

1 1

3 2 3 2

3 4 2 3 2 4

How many weakly increasing binary trees on 6 vertices use the numbers 1, 2, 2, 3, 3, 4 as their labels?
How many permutations are there of the sequence 1, 2, 2, 3, 3, 4? Can you find a bijection between
these two objects?

172
Chapter 10

Graph optimization

Consider the below graph, in which each vertex represents a ski lift hut, and in which each edge represents
a ski lift. Each ski lift edge is decorated with an arrow showing in which direction the lift moves up the
hill. Furthermore, each edge is labeled with the number of people that can be transported up the ski lift per
minute. What is the maximum number of people that can be moved from the bottom of the hill to the top
of the hill per minute?

Bottom of
mountain
30
20
30
30 20
40 30 40 40

20 40
30
50
30
50 30 10
Top of
mountain

It is clear that no more than 60 people can go up every minute, since at most 60 can leave the bottom of
the mountain. It turns out that it is possible to transport 60 people per minute from the bottom to the top
of the hill. Below, one strategy for how to transport 60 people per minute is drawn in red, with the number
indicating how many people ride a lift. Note that the number of passengers on a ski lift, as labeled below in
red, is always less than or equal to the capacity of that lift, as labeled above in black. To achieve this flow
of 60 people per minute, different skiers need to take different routes up the hill!

Bottom of
mountain
30
20
20
30 10
10 20

20 20
30
50
10
20 10 10
Top of
mountain

173
 In this chapter we study optimization problems on graphs. In an optimization problem, we are trying
to maximize or minimize something. We may want to maximize the number of people transported per
minute, or the number of pizzas we can deliver in an hour, or the amount of space between seats in a room.
Alternatively, we may want to minimize the cost we have to pay, the amount of trash we produce, or the
amount of time necessary to complete a task. Some optimization problems on graphs, such as finding the
maximum spanning tree in Section 10.1, can be solved efficiently via known algorithms. Other optimization
problems, such as the traveling salesperson problem in Section 10.2, are difficult to solve exactly, although
efficient algorithms to find approximately optimal solutions may exist.

10.1 Minimal spanning trees

Click on the icon at left or the URL below for this section’s short lecture video.
https://youtu.be/ld1tR8Oom5U

Suppose the state of Colorado wants to connect the following towns with fiber-optic cable, enabling fast
internet. The cost of installing fiber-optic cable along a single edge is indicated by the number label on that
edge. What is the cheapest way to connect these 6 cities with fiber-optic cable?

Laramie Laramie
7 Cheyenne 7 Cheyenne

6 4 6 4
5 5
Fort Fort
3 3
14 Collins Greeley 14 Collins Greeley

11 12 11 13 12
13

5.5 5.5
Boulder Denver Boulder Denver

Let us define the mathematical concept, minimal spanning trees, which underlies this fiber-optic cable
example.

Definition 10.1.1. A weighted graph is a graph equipped with a real number weight on each edge.

A sub-tree of a graph G is a subgraph of G which also happens to be a tree.

Definition 10.1.2. Given a weighted connected graph G, a minimal spanning tree is a sub-tree containing
all vertices of G whose total sum of edge weights is as small as possible.

Remark 10.1.3. If no two edges of G have the same weight, then a minimal spanning tree is unique.

Example 10.1.4. If some edges of G have the same weights, then a minimal spanning tree may or may not
be unique.

174
 Indeed, the weighted graph on the left above has a unique minimal spanning tree, whereas the weighted
graph on the right has two different minimal spanning trees.
A poor algorithm to compute a minimal spanning tree would to be to compute the cost of all possible
trees and then pick the cheapest. Let’s see that this is a poor algorithm when G is the complete labeled
graph on n vertices. In this case, there is a bijection between minimal spanning trees for G and labeled trees
on n vertices. By Cayley’s Theorem (Theorem 9.3.1), there are nn−2 possible labeled trees on n vertices to
check. For n = 50 vertices, we would have nn−2 = 5048 ≈ 3.6 × 1081 trees to check, which is on the order of
the number of atoms in the universe!
A much better algorithm for finding a minimal spanning tree is Kruskal’s algorithm.

Kruskal’s algorithm to produce a minimal spanning tree from a weighted connected graph G
1. Include all vertices of the graph G.
2. Include the cheapest edge (the one with the smallest weight). Remove that edge from consideration.
3. If the subgraph is not connected, add the cheapest remaining edge that connects two of its connected
components to the subgraph. Remove that edge from consideration.
4. Repeat step (3) until the graph is connected.
At each step of the algorithm beyond the first, one edge is added and the output is a subgraph of G with
no cycles. When the algorithm terminates, the output is a connected graph with no cycles, thus a tree, and
it contains all the vertices of G. Thus the output is a spanning tree of G. Later in the section, we will prove
that it is a minimal spanning tree.
If G has n vertices, then this algorithm will stop after adding exactly n − 1 edges by Theorem 9.2.1.
Kruskal’s algorithm is a greedy algorithm, which means that at each step you do the cheapest possible
thing to make progress. What is surprising is that this greedy algorithm also happens to give you the global
optimum!
If there are ties in Kruskal’s algorithm, i.e., if there is more than one cheapest remaining edge that does
not form a cycle, then these ties can be broken arbitrarily. Whichever way you proceed to break this tie will
still lead to a minimial spanning tree.
Question 10.1.5. If you and your friend each perform Kruskal’s algorithm on a weighted graph, will
you necessarily get the same minimal spanning trees as output? Will the minimal spanning trees you get
necessarily have the same costs?
Theorem 10.1.6. When performed on any weighted connected graph G, Kruskal’s algorithm will return a
minimal spanning tree.
Proof. Let T be a spanning tree found by Kruskal’s algorithm. Let G0 be any other spanning tree. Our goal
is to show cost(T ) ≤ cost(G0 ), which will then imply that T is a minimal spanning tree.
Let e be the first edge (when constructing tree T via Kruskal’s algorithm) that is not in G0 . Let G0 ∪ e
denote the graph that is obtained by adding the edge e to the graph G0 . By Theorem 9.1.4, adding any

175
new edge between pre-existing vertices in a tree creates a cycle, and therefore the graph G0 ∪ e has a cycle
C. This cycle C is not contained in the tree T , and therefore there is an edge f of the cycle C that is not
in the tree T .

Let G1 =(i)(G0 ∪ e) \ f be the graph obtained by removing edge f from G0 ∪ e. The weighting of the edge
e is less than or equal to the weighting of the edge f . The reason is that e was added to the graph of T
f at this stage of constructing T in Kruskal’s algorithm. So cost(G1 ) ≤ cost(G0 ). So the proof
rather than (ii)
will be complete if we can show the stronger statement that cost(T ) ≤ cost(G1 ).
We repeat(iii)this same process with the tree G1 in place of the tree G0 , obtaining a tree G2 . We note that G2
has one more edge from T added and one more edge not in T removed, and also satisfies cost(G1 ) ≥ cost(G2 ).
Since we keep(iv)adding edges in T while removing edges not in T , eventually we get Gk = T for some integer
k. We have
(v)
cost(G0 ) ≥ cost(G1 ) ≥ cost(G2 ) ≥ . . . ≥ cost(Gk ) = cost(T ),
e
as desired. Hence Kruskal’s algorithm always produces a minimal spanning tree.
f

c
Exercises
T
1. Consider the complete bipartite graph B3,3 , and label the vertices in the left set 1,2,3 and the vertices
in the right set 1,2,3. Then label each edge by the sum of the two vertices it connects; for instance,
G
the edge connecting left vertex 1 to right vertex 3 is labeled 4, and the edge connecting left vertex 2
to right vertex 2 is labeled 4 as well. What is the weight of a minimal spanning tree of this graph?
0

2. Answer the previous problem but for B4,4 labeled in the same way; label the vertices on the left and
1
right 1,2,3,4, and label each edge by the sum of its two vertices. What is the weight of a minimal
spanning tree?
2

3. A complete graph K4 with vertices labeled 1,2,3,4 has all its edges labeled by the same value 1. How
many different minimal spanning trees are there? (Hint: Cayley’s Theorem may come in handy.)

4. Find a minimal spanning tree on the weighted graph drawn below.

176
 5. Let G be a weighted connected graph with positive edge costs.

(a) Describe how to find a spanning tree for which the sum of the edge-costs is maximal.
Hint: Create a new weighted graph G0 by multiplying each edge weight of G by −1.
(b) Describe how to find a spanning tree for which the product of the edge-costs is minimal.
Hint: Create a new weighted graph G0 by editing the edge weights of G using logarithms.

6. Another way to measure the cost of a spanning tree is by the maximal cost of its edges; let’s call that
the max-edge cost. For example, the max-edge cost of the diagram at the beginning of this section is
11, for the edge between Boulder and Fort Collins. Decide whether the following statement is true or
false and explain: the spanning tree produced by Kruskal’s algorithm minimizes the max-edge cost. In
thinking about this, it might be helpful to sort the edges in order of cost so that the first edge has the
lowest cost and the last edge has the largest cost.

7. Draw a weighted connected graph that has at least three different minimal spanning trees.

8. Prove Remark 10.1.3, which says that if no two edges of a connected graph G have the same weight,
then there is a unique minimal spanning tree for G.

9. True or False: If two edges in a weighted connected graph have the same cost, then the graph has more
than one minimal spanning tree.

10.2 Traveling salesperson problem

Let G be a connected graph. A tour of G is a closed walk that is allowed to retrace its steps (cross an edge
more than once), and that visits every vertex of G.
Given a weighted connected graph G, can we find a tour of minimal cost?

177
Example 10.2.1. We could consider an arbitrary connected graph with arbitrary edge costs. An optimal
tour on this graph is drawn below in red.
1
2 2
3
3 17
2
1
Notice that an optimal tour on the above graph consists of utilizing an edge twice.
Example 10.2.2. We could consider the complete graph, where the edge costs are the Euclidean distances.
We have not drawn any of the edges in this complete graph, nor have we labeled the edge weights. But since
the edge costs are the distances on the page, you can tell that the tour drawn below is either optimal, or
otherwise very close to optimal.

In what contexts does the traveling salesperson problem arise? Consider a salesperson visiting cities and
then driving back home, while trying to use as little gas as possible. Or consider a manufacturing plant —
a drill bit needs to drill holes on a metal plate and then return to its starting position before the next metal
plate on the assembly line arrives. Below we show a metal plate with six drilled holes. The dotted lines
show the tour traveled by the drill bit. As this tour by the drill bit is repeated thousands of times per day
in the manufacturing plant, it is important to make the tour as short as possible.

In Section 10.1, we saw that finding a minimal spanning tree can be solved via a greedy algorithm, namely
Kruskal’s algorithm. By contrast, greedy approaches for finding an optimal traveling salesperson tour fail
miserably. Indeed, if you try to find a tour by first adding in the cheapest possible edges, you may be forced
later on to choose extremely expensive edges, leading to a suboptimal solution.
In general, it is very hard to find an optimal solution to the traveling salesperson problem. Finding an
optimal tour is an NP-hard problem, which roughly speaking means that the running time of a computer
program is likely super-polynomial in terms of the number of vertices in a graph.

Triangle inequality
A weighted complete graph G satisfies the triangle inequality if for any triangle (or subgraph C3 ) in the
graph G, the edge weights a, b, and c, of this triangle satisfy a ≤ b + c, b ≤ a + c, and c ≤ a + b.

178
In other words, the cost of traversing one side of a triangle is never more expensive than the cost of traversing
the other two sides of that triangle. Complete graphs drawn in the plane with edge weights given by Euclidean
distances always satisfy the triangle inequality.
In the tree shortcut algorithm, described below, we will find a way to approximate an optimal traveling
salesperson tour in a complete graph that satisfies the triangle inequality. While this tour need not be
optimal, we will show that it is optimal up to a factor of 2, so long as the weighted graph G is a complete
graph Kn and satisfies the triangle inequality.

Tree Shortcut Algorithm

(a) Find a minimal spanning tree for the complete weighted graph G.

(b) Consider the tour that wraps around the outside of the minimal spanning tree, crossing every edge in
this tree twice.

(c) When traversing the tour in (b), whenever possible skip visiting a vertex that has already been visited.

This algorithm is drawn below for the case of a complete graph on 7 vertices, with edge labels given by the
Euclidean distances between the points. The (arbitrary) starting edge we use to pass from step (b) to step
(c) is indicated by the red arrow.

We note that there is not a unique way to proceed from step (b) to step (c). Indeed, if you start your
shortcut tour by starting at a different edge (indicated by the red arrow), then you may end up with a
different shortcut tour, as drawn below.

179
Theorem 10.2.3. If the nonnegative costs in a weighted complete graph G satisfy the triangle inequality,
then the Tree Shortcut Algorithm finds a tour that costs at most twice as much as an optimal tour.

The proof of this theorem will involve two lemmas: the cost of an optimal tour is at least as much as the
cost of a minimal spanning tree, and twice the cost of a minimal spanning is at least as much as the cost of
the output of the Tree Shortcut Algorithm.

Lemma 10.2.4. In a connected graph with nonnegative edge weights, the cost of an optimal tour is at least
as much as the cost of a minimal spanning tree.

Proof. Let M be the cost of a minimal spanning tree in G. We claim

[cost of an optimal tour] ≥ [cost of an optimal tour minus any edge] ≥ M.

Indeed, the first inequality follows since removing an edge (with a nonnegative cost) cannot increase the total
cost. The second inequality follows since a tour minus an edge is a spanning tree (indeed it is a spanning
path), and hence must cost at least as much as M , the cost of a minimal spanning tree.

Lemma 10.2.5. If the edge costs in a weighted complete graph G satisfy the triangle inequality, then the
Tree Shortcut Algorithm finds a tour that costs at most twice as much as a minimal spanning tree.

Proof. Let M be the cost of a minimal spanning tree in G. Note that the tour produced by step (b) of the
Tree Shortcut Algorithm is a tour of cost 2M , twice the length of a minimal spanning tree. We claim

[cost of output of Tree Shortcut Algorithm] ≤ 2M.

This follows from the triangle inequality assumption: when progressing from step (b) of the Tree Shortcut
Algorithm (which is a tour of cost 2M ) to step (c), the cost of the tour can only decrease by the triangle
inequality assumption.

By combining these two lemmas, we obtain a proof of Theorem 10.2.3.

Proof of Theorem 10.2.3. Let M be the cost of a minimal spanning tree in G. We claim

[cost of output of Tree Shortcut Algorithm] ≤ 2M ≤ 2 · [cost of optimal tour].

Indeed, the first inequality follows from Lemma 10.2.5, and the second inequality follows from Lemma 10.2.4.

Question 10.2.6. Can you find an example to show that Theorem 10.2.3 may fail if the edge weights don’t
satisfy the triangle inequality?

Exercises
1. Describe a real-world example that interests you in which you might need to solve a traveling salesperson
type problem.

2. Let K4 be the complete graph on 4 vertices, with vertices labeled 0, 1, 2, and 3. How many different
tours of K4 are there that start and end at vertex 0, and that visit every other vertex exactly once?

3. Let K3,3 be the complete bipartite graph with 3 vertices on each side. Fix a single arbitrary vertex
v in K3,3 . How many different tours of K3,3 are there that start and end at vertex v, and that visit
every other vertex exactly once?

4. Let G be a weighted connected graph in which all edge weights are nonnegative. Prove that the cost
of a minimal spanning tree is less than or equal to the cost of an optimal tour (i.e., a tour solving the
Traveling Salesperson Problem).

180
 5. Let G be a weighted connected graph in which all edge weights are positive. Show that the cost of a
minimal spanning tree is strictly smaller (< instead of ≤) than the cost of an optimal tour (i.e., a tour
solving the Traveling Salesperson Problem).
6. Show by an example that if the triangle inequality is not true on a weighted complete graph, then a
tour found by the Tree Shortcut Algorithm can be longer than 1000 times an optimal tour.

10.3 Matchings

Click on the icon at left or the URL below for this section’s short lecture video.
https://www.youtube.com/watch?v=JgWQPKVOl3o

A matching in a graph G = (V, E) is a subset S of the edge set E such that no two of the edges in S
share a common vertex.
For instance, two matchings of the graph K5 are shown below, where the edges we chose to be in the
subset are shaded. The first matching contains only one edge and the second has two:

An important problem in combinatorial optimization is to find matchings of the largest possible size. In
other words, we would like as many vertices as possible to be an endpoint of an edge in the matching.
In the next two definitions, we will define maximal and maximum matchings. Even though the words
differ in only two letters — maximal vs. maximum — they mean different things! As you will see, every
maximum matching is also maximal, but not necessarily vice versa.
Definition 10.3.1. A matching is maximal if it is not contained in a larger matching.
For instance, the first matching of K5 drawn above is contained in the second matching. Thus it is
not maximal. However, the second drawing is maximal because there is only one vertex remaining, and
connecting it to any other vertex with a shaded edge will result in two bold edges sharing a vertex.
However, it is possible to have a maximal matching that does not include as many edges as possible.
Example 10.3.2. Consider the path graph P4 on 4 vertices. Here are two matchings:

Both are maximal, because we cannot add any more edges to make a larger matching. But the second
one includes more edges than the first one.

181
Definition 10.3.3. A maximum matching is a matching with the largest possible number of edges. A
perfect matching is a matching such that every vertex of V is an endpoint of (exactly one) edge in S.
In Example 10.3.2, the first matching is maximal but not maximum and the second matching is maximal,
maximum and perfect.
In K5 , every maximal matching has size 2, and this is the largest possible size of a matching, so it is
also a maximum matching. Exactly four vertices of K5 are contained in a maximum matching of K5 , so the
maximum matchings are not perfect.
One algorithm for finding a maximal matching is the greedy algorithm. Label the vertices 1, 2, 3, . . . , n
and choose the edge (i, j) where i is minimal and j is smallest among all edges from i. Then consider the
subgraph formed by deleting vertices i and j, and again choose the minimal vertex and edge out of it, and
so on until no more edges can be added to make a larger matching. By definition, this process results in a
maximal matching.
However, finding a maximum matching is in general much harder. The above greedy algorithm will not
necessarily result in a maximum matching. The Hopcroft-Karp algorithm is a general algorithm for finding
a maximum matching in a graph, and while we will not be describing the full algorithm here, we give a hint
of the strategy using the key idea of augmenting paths.
The idea is to start with a matching that is not maximum and replace it with a different matching that
contains one more edge.
Definition 10.3.4. Let G be a graph and let M be a matching of G. Suppose that G has (at least two)
vertices v and w that are not endpoints of edges in M . Given this data, an augmenting path is a path that
starts at v and ends at w that contains an odd number m of distinct edges e1 , e2 , . . . , em , whose endpoints
are all distinct, and such that the edges alternate between being in M and not in M :
e1 6∈ M, e2 ∈ M, e3 6∈ M, e4 ∈ M, . . . , em 6∈ M.
Example 10.3.5. Consider the complete bipartite graph B4,4 , where the vertices on the left are labeled
A, B, C, D and the vertices on the right are labeled 1, 2, 3, 4. Let G be the graph obtained from this by
removing the edge {4, A}. We start with the matching M of G consisting of the edges {1, B}, {2, C}, and
{3, D}. This is a maximal matching, because the edge {4, A} is missing so there is no way to increase the
number of matched vertices. On the other hand, it is not a maximum matching.
Here is an example of an augmenting path with edges:
{A, 1}, {1, B}, {B, 2}, {2, C}, {C, 3}, {3, D}, {D, 4}.
We illustrate it below, with the black edges being the edges in the path and the red overlined edges being
the ones in the matching M .
A 1

B 2

C 3

D 4
Suppose P is an augmenting path of a matching M . The set P − M contains all the edges of P that are
not in M . We can see that P − M is also a matching of G and the number of edges in P − M is one larger
than the number of edges in M .
For instance, in Example 10.3.5, P − M contains the edges
{A, 1}, {B, 2}, {C, 3}, {D, 4},
so P − M is a matching of size 4, which is a maximum matching (and a perfect matching).

182
Example 10.3.6. Let M be the set of red edges in the graph below. Find an augmenting path for M and
use it to find a larger matching than M ?

A 1

B 2

C 3

D 4

10.3.1 Matchings of bipartite graphs

In this section, we describe Hall’s Marriage Theorem for matchings in bipartite graphs. To illustrate it, we
start with an example.
Example 10.3.7. Suppose new CSU students a, b, c, and d all went to the same high school, and so they
would rather not be in the same dorm as each other so that they get more opportunities to meet new people.
They fill out their dorm preferences form, and unfortunately have quite similar preferences for dorms!
Student a checks only Parmelee, b only checks Braiden, c checks Academic Village, Braiden, and Corbett,
and d checks Braiden and Parmelee. Is there a way to match each of the students to a dorm so that none of
the four students are in the same dorm as each other?
To answer this, we can model it as a bipartite graph:

If we focus on students a, b, d, we notice that there is no way to do it; a needs to be in Parmelee to have
their preference satisfied, b needs to be in Braiden, and d needs to be in one or the other, forcing it to overlap
with a or b.
Example 10.3.8. Realizing the issue in the above example, the four students decide to change their pref-
erences form so that each checks exactly two preferred dorms. Student a checks Academic Village and
Parmelee, b checks Braiden and Corbett, c checks Academic Village and Corbett, and d checks Braiden and
Parmelee. Is there a way to match them to distinct dorms now?

183
 One way to do it is to match a to Academic Village, b to Braiden, c to Corbett, and d to Parmelee. But
there are other possibilities too; do you see another?

There are many situations in which is it important to find suitable matchings — for example when
matching medical residents to desired medical residencies, when matching pets to suitable homes, and when
matching prospective employees to appropriate jobs.
Notice that in the two examples above, we were trying to find a matching in a bipartite graph for which
every element on one side of the graph is matched with an element on the other side. We call such matchings
saturated, and any saturated matching are necessarily a maximum matching.

Definition 10.3.9. Let G be a bipartite graph with left set L, right set R, and |L| ≤ |R|. Then any
matching of G that matches every element of L with some element of R is called a saturated matching.

Hall’s Marriage Theorem tells us exactly when a saturated matching exists in a bipartite graph. To state
it, we need the concept of the neighborhood of a set of vertices in a graph.

Definition 10.3.10. Let G = (V, E) be a graph. The neighborhood of a subset S ⊆ V of the vertices,
denoted NG (S), is the set of all vertices v such that v is connected by an edge to some element of S.

For example, when S = {v} contains only one vertex v, then the neighborhood of S is the set of edges
adjacent to v and the size of the neighborhood is the degree of v.

Theorem 10.3.11. (Hall’s Marriage Theorem) Let G be a bipartite graph with left set L, right set R, and
|L| ≤ |R|. Then a saturated matching exists if and only if, for every subset X ⊆ L,

|X| ≤ |NG (X)|.

The proof of Hall’s Marriage Theorem is rather complicated and we will not provide the full details
here. But as intuition for one direction of the statement, suppose there is a subset X of the left set whose
neighborhood NG (X) is strictly smaller than X. Then the points in X cannot all be matched to distinct
points in R by the pigeonhole principle. So the inequality |X| ≤ |NG (X)| is necessary for a saturated
matching to exist. The interesting fact is that this condition is also sufficient.

Example 10.3.12. In the first examples of students and dormitories above, can you find a subset X of the
students for which the inequality |X| ≤ |NG (X)| is violated? In the second example, can you show that the
inequalities are all satisfied?

Exercises
1. The size of a maximum matching on the complete graph K6 ?

2. What is a maximum matching on the complete graph K7 ?

3. What is the size of a maximum matching on the path graph P6 ?

4. What is the smallest size of a maximal matching on the path graph P6 ?

5. What is the size of a maximum matching on the complete bipartite graph B20,21 ?

6. If G has a perfect matching, explain why the number of vertices of G is even.

7. In the path graph P5 , show that every maximal matching is maximum (but not perfect).

8. In the cycle graph C5 , show that every maximal matching is maximum (but not perfect).

9. For the path graphs P6 , . . . , P11 , find the size of a maximum matching and the smallest possible size
of a maximal matching.

184
10. For the cycle graphs C6 , . . . , C11 , find the size of a maximum matching and the smallest possible size
of a maximal matching.

11. Find the size of a maximum matching for the path graph Pn . Your answer should depend on whether
n is even or odd.

12. What is the smallest possible size of a maximal matching for Pn ? Your answer should depend on the
congruence of n modulo 3.

13. Find the size of a maximum matching for the cycle graph Cn . Your answer should depend on whether
n is even or odd.

14. What is the smallest possible size of a maximal matching for Cn ? Your answer should depend on the
congruence of n modulo 3.

15. Does the below bipartite graph have a saturated matching?

Each node on the left represents a pet, each node on the right represents a home, and tehre is an edge
when that home would be suitable for the pet. We would like to know if each pet can find a different
home.

16. Does the below bipartite graph have a saturated matching?

Each node on the left represents a person, each node on the right represents a job, and there is an
edge when a job would be suitable for that person. We would like to know if each of the jobs can be
simultaneously filled, with each person having exactly one job.

17. Does the below bipartite graph have a saturated matching?

185
 Each node on the left represents a person, each node on the right represents a job, and there is an
edge when a job would be suitable for that person. We would like to know if each of the jobs can be
simultaneously filled, with each person having exactly one job.

18. Fact: in a complete bipartite graph Bn,m with n ≤ m, a maximum matching has size n. Explain how
that fact follows from Hall’s Marriage Theorem.

19. Let G be a bipartite graph with the sets of left and right vertices having equal size n and every vertex
having the same degree k ≥ 1. Use Hall’s Marriage Theorem to show that G has a perfect matching.

10.4 Ramsey theory

We conclude this chapter with a famous topic in graph optimization known as Ramsey theory. We start with
a classical example.

Example 10.4.1. Six people walk into a room; some of the people know each other already and others do
not. Prove that either some three of them all mutually know each other, or some three of them all mutually
do not know each other.

To get a handle on the above problem, we can first model it as a graph theory problem; consider a graph
on 6 vertices that represent the people, and where each pair of people is connected either by a solid or dashed
edge according to whether they know each other or not, as shown:

1 2

6 3

5 4

In the above graph, notice that the vertices numbered 3, 4, 6 all mutually do not know each other. Their
induced subgraph is a smaller complete graph K3 , with all its edges the same “color” (where the two colors
are solid and dashed in this case, but any two colors would work). To state our problem more precisely in
terms of graphs, we make the following two definitions.

Definition 10.4.2. An m-clique in a graph is a subgraph isomorphic to Km .

Definition 10.4.3. Suppose a graph G has every edge assigned a color. A monochromatic m-clique in
G is an m-clique all of whose edges are the same color.

186
 For example, the graph above has a monochromatic 3-clique with vertices 3, 4, 6, where the edges between
them are all dashed.

Question 10.4.4. How many monochromatic 3-cliques can you find in the graph above?

We can now rephrase Example 10.4.1 in terms of graph theory:

Example 10.4.5. Prove that if each edge of K6 is colored with one of two colors, then there exists a
monochromatic 3-clique.

Proof of Example 10.4.5. To prove this, pick one of the vertices x, and consider all 5 edges attached to x.
Since each of these 5 edges is either solid or dashed, by the Pigeonhole Principle, at least three edges share
the same color. By symmetry, we can assume that this color is solid, so x is connected to at least three
vertices, say y, z, w, by solid edges.

x y

w
Now, consider the edges in the triangle formed by y, z, w. If any of these edges is solid, they form a solid
triangle with x, and we have a monochromatic (solid) 3-clique, as in the left first example below. But if none
of the three edges is solid, then the triangle y, z, w is a monochromatic (dashed) 3-clique, as in the right
example below.
x y x y

z z

w w
So in all cases we do indeed have a monochromatic 3-clique. This completes the proof.

We have shown that in a group of 6 people, either some 3 people know each other or some 3 people all
do not know each other. Interestingly, 6 people is the minimum number of people required for this fact to
be true; it is possible to gather 5 people in a room such that no three all know each other and no three all
do not know each other. (See Exercise 1 below.)

10.4.1 Investigation: Ramsey numbers

The example above can be generalized as follows. If we color the edges of the complete graph Kn with two
colors, is there guaranteed to be a monochromatic m-clique, for various values of n and m? And, for a given
m, what is the smallest value of n for which this is true?
We can further generalize this question using the notion of a Ramsey number.

Definition 10.4.6. The Ramsey number R(m1 , m2 ) is the smallest number n such that any coloring of
the edges of the complete graph Kn using 2 colors c1 and c2 must have one of the following:

• A monochromatic m1 -clique of color c1 , or

187
 • A monochromatic m2 -clique of color c2 .

Example 10.4.7. The number R(3, 3) is the smallest number n for which any coloring of the edges of Kn
in two colors contains a monochromatic 3-clique (of one color or the other). We found in Example 10.4.5
and Exercise 1 (below) that the smallest such n is 6, so R(3, 3) = 6.

Example 10.4.8. Let’s compute R(2, 2), which is the smallest number n such that any coloring of the edges
of Kn in two colors contains a monochromatic 2-clique (of one color or the other). A 2-clique is just an
edge, so we only need one edge of any color, and so n = 2 is the smallest possible number that satisfies the
condition. Therefore R(2, 2) = 2.

We can define Ramsey numbers not only for 2 colors, but also more generally for k colors.

Definition 10.4.9. The Ramsey number R(m1 , m2 , . . . , mk ) is the smallest number n such that any
coloring of the edges of the complete graph Kn using k colors c1 , c2 , . . . , ck must have one of the following:

• A monochromatic m1 -clique of color c1 , or

• A monochromatic m2 -clique of color c2 , or

• ···

• A monochromatic mk -clique of color ck .

Example 10.4.10. Let’s compute R(3, 3, 2). This is the smallest number n such that any coloring of the
edges of Kn using colors red, green, and blue has either a red 3-clique, a green 3-clique, or a blue 2-clique.
First note that since there is an edge-colored graph G on 5 vertices using only red and green colors that
does not have a red triangle or a green triangle (see Exercise 1 below), this graph G does not have a red
3-clique, a green 3-clique, or a blue 2-clique. So R(3, 3, 2) > 5.
We can now show that R(3, 3, 2) = 6. First, if there is any blue edge in a coloring of K6 , that is a blue
2-clique. On the other hand, if there are no blue edges, then we have a coloring of the edges either red or
green, and we know such a coloring either had a red 3-clique or green 3-clique since R(3, 3) = 6. Therefore
any coloring either has a red 3-clique, a green 3-clique, or a blue 2-clique, and so R(3, 3, 2) = 6.

Ramsey’s theorem says that the Ramsey number R(m1 , m2 , . . . , mk ) is always finite; that is, there
does exist some sufficiently large n for which the statement is true. But what is the minimum such number
n = R(m1 , m2 , . . . , mk )? This is in general an unsolved problem, and the study of computing the Ramsey
numbers is called Ramsey theory, and is one of a large class of optimization problems in graph theory
involving coloring graphs.

Exercises
1. Color the edges of K5 in two colors in a way that avoids creating a monochromatic 3-clique.

2. Let G be the complete graph K5 on 5 labeled vertices 1, 2, 3, 4, 5. How many different ways can you
color each edge of G either red or blue such that there exists a monochromatic 4-clique in the graph?

3. Let H be the complete graph K6 on 6 labeled vertices 1, 2, 3, 4, 5, 6. How many different 4-cliques are
subgraphs of G?

4. Show that R(n) = n for all positive integers n.

5. Show that R(2, n) = n for all positive integers n.

6. Compute R(2, 2, 2).

7. Explain why R(a, b, 2) = R(a, b) for any positive integers a and b.

188
 8. Prove that R(a, b) = R(b, a) for any positive integers a and b.
9. Prove that R(3, 4) = 9.
10. Look up online the value of R(4, 4), and explain what this value means.
11. Look up online what is known about R(5, 5), and explain what this means.

12. It is known that R(3, 3, 3) = 17. Investigate this as follows. Draw a complete graph on 6 vertices with
each edge colored either red, green, or blue that has no monochormatic 3-clique of any color. This
shows that R(3, 3, 3) > 6.
Then do the same for 7 vertices that shows that R(3, 3, 3) > 7, and so on. Can you show that
R(3, 3, 3) > 16?

189
190
Chapter 11

Planar Graphs

A graph is planar if it can be drawn on a piece of paper with no edges crossing. In this chapter we will
explore the many particularly nice properties of planar graphs and their applications.

11.1 Planar graphs

Click on the icon at left or the URL below for this chapter’s lecture video.
https://youtu.be/vbNBpvgeMdY

Recall that a graph is a set of vertices and a set of edges. There are many ways to set up the data for a
graph, for example an incidence matrix or an edge list. In this section, we focus on the more visual way of
describing a graph, by drawing it on paper. It is natural to draw a graph as simply and cleanly as possible.
For example, here are two ways to draw the graph K4 .

The graph on the right has the advantage that none of its edges intersect.

Definition 11.1.1. A planar graph is a connected graph drawn in the plane so that its edges do not cross.

The pictures above show K4 drawn as a non-planar graph and as a planar graph. In other words, the
property of being a planar graph depends not only on the graph but also on how it is drawn in the plane.
We say that a connected graph is planar if there is some way of drawing it in the plane as a planar graph.

Example 11.1.2. Is the below graph planar?

A
E B

D C

191
Answer. Yes. Two possible planar maps are drawn below.

A A
E B D C

D C B E

Example 11.1.3. Is the below graph planar? (Adding another edge, CD, only makes this harder.)

A
E B

D C
Answer. Yes. Two possible planar maps are drawn below.

A
A
D C
D B

C E B E

Remark 11.1.4. In designing computer chips, there is an advantage to having a small number of crossings
of electrical wires. This translates into a question on graph theory about minimizing the number of edge
crossings in a graph drawn in the plane.

In Section 11.2, we study properties of planar graphs. Then in later sections of this chapter, we use these
properties to tackle the problem of deciding whether a graph is a planar graph.

Remark 11.1.5. Fáry’s Theorem states that if a graph is planar, then it can be drawn as a planar graph
in which every edge is straight. See www.jasondavies.com/planarity for a fun game in which you drag
the vertices of a (planar) graph around until you obtain a planar embedding (with straight edges)!

Remark 11.1.6. It is not known if every planar graph can be drawn in the plane so that every edge is
straight with length being equal to an integer.

Exercises
1. Draw a 5-pointed star and then redraw it as a planar graph.

2. Draw B2,3 as a planar graph.

3. Draw B2,4 as a planar graph.

4. Draw B3,3 and then remove one of its edges. Redraw this graph as a planar graph.

192
11.2 Euler’s formula for planar graphs

Click on the icon at left or the URL below for this chapter’s lecture video.
https://www.youtube.com/watch?v=GJJOIKn-1PA

Euler’s formula gives a remarkable connection between the edges, vertices, and regions (or faces) of a
planar graph. Let’s make these ideas precise.
Definition 11.2.1. In a planar graph, let v be the number of vertices and let e be the number of edges.
The edges divide the plane into regions, which are called faces. Let f be the number of faces, including the
face(s) which touch the edge of the paper.

6 vertices 7 vertices 7 vertices

7 edges 9 edges 10 edges
3 faces 4 faces 5 faces
Definition 11.2.2. Define χ = v − e + f ; we call this quantity the Euler characteristic of the graph.
Example 11.2.3. For the planar graph K4 , v = 4 and e = 6 and f = 4. So χ = 4 − 6 + 4 = 2.
Example 11.2.4. For the graph Cn , drawn as a regular n-gon in the plane, v = n, e = n, and f = 2; (the
two faces are the inside and the outside regions). So χ = n − n + 2 = 2.
Example 11.2.5. For the graph Pn , drawn as a path in the plane, v = n, e = n − 1, and f = 1 (the path
makes a slit in the one face). So χ = n − (n − 1) + 1 = 2.
Example 11.2.6. Find a way to draw the complete bipartite graph B2,3 as a planar graph. Show v = 5,
e = 6, and f = 3. Show χ = 2.
At this point, you may be wondering why χ = 2 in all these examples. The reason is that the Euler
characteristic is 2 for every planar graph.
Theorem 11.2.7 (Euler’s formula). In a connected planar graph, let v be the number of vertices, let e be
the number of edges, and let f be the number of faces. Then χ = v − e + f = 2.
The proof uses this next lemma about cycles in graphs.
Lemma 11.2.8. Let G be a connected graph which is not a tree. If e is an edge that belongs to a cycle in
G, then the subgraph G0 = (G with edge e removed) is still connected.
Proof.

Graph G G'=(G with e removed)

193
The vertices of G and G0 are the same. The edges in G are all the edges of G0 together with the edge e. Let
u and w be arbitrary vertices in G0 . To show that G0 is connected, we must find a walk in G0 from u to w.
Since G is connected, there is a walk in G from u to w.

• If this walk does not use the edge e, then it is a walk in G0 and we are done!

• If this walk does use e, then there is a second walk in G from u to w which “goes the other way around
the cycle”. This second walk is also a walk in G0 .

The following illustrates the proof method above:

u u

v v

Walk in G Walk in G'

Here is another example that illustrates the proof method above:

u u

v v
Walk in G Walk in G'

We are now ready to prove Euler’s formula.

Proof. This proof will be by induction on the number of faces. For the base case, if f = 1, then the graph
has no cycles. So it is a connected tree. By Theorem 9.2.1, a tree with v vertices has v − 1 edges. So
χ = v − (v − 1) + 1 = 2. So Euler’s formula is true when f = 1.
Now suppose G is a connected graph with f ≥ 2 faces. We assume for induction that Euler’s formula is
true for any connected graph with 1 ≤ f 0 < f faces. Since f ≥ 2, the graph G is not a tree, so it has a cycle.
Pick an edge e in that cycle.
By Lemma 11.2.8, the subgraph G0 = (G with edge e removed) is still connected. The number of vertices
of G0 is v 0 = v. The number of edges is e0 = e − 1 since one edge was removed. The number of faces is
f 0 = f − 1.
By the inductive hypothesis, 2 = χ0 = v 0 − e0 + f 0 . Substitute that v 0 = v, e0 = e − 1 and f 0 = f − 1 to
see that 2 = v − (e − 1) + (f − 1) = v − e + f .

To visualize the inductive proof above, think of the edge e acting like a dam or levee that separates two
regions; once it is removed then water will flood those two regions simultaneously.

1
land
2
3 4
water
Remark 11.2.9. Euler’s formula is also true for connected planar graphs that allow loop edges (from a
vertex to itself) and multiple edges between vertices.

194
Exercises
1. Draw a regular n-gon. Add one vertex in the center and draw an edge between the center vertex and
every other vertex; your picture should resemble the spokes on a bicycle wheel. Compute v, e, and f
for this graph and check that χ = 2.
2. If a connected planar graph has 7 vertices and 11 edges, what is the number of faces? Draw a graph
with these values.
3. If a connected planar graph has 5 vertices and 5 faces, what is the number of edges? Draw a graph
with these values.
4. If a connected planar graph has 8 edges and 4 faces, what is the number of vertices? Draw a graph
with these values.
5. A planar graph has 6 vertices and 7 edges. How many faces does it have?
6. How many edges do you need to remove from K5 to make it a planar graph?
7. How many edges do you need to remove from B3,3 to make it a planar graph?
8. What is the biggest possible number of faces for a connected bipartite planar graph with 5 edges?

11.3 Graphs that are not planar

In this section, we will show that K5 , the complete graph on 5 vertices, is not planar. In other words, it is
not possible to draw K5 on a piece of paper without edges crossing.

2 5

3 4
For a planar graph, recall that v is the number of vertices, e is the number of edges, and f is the number
of faces.
3f
Lemma 11.3.1. If G is a planar graph, then e ≥ 2 .

Proof. Let E be the sum of the number of edges on every face. Every edge borders two faces. So 2e = E.
The graph has no multi-edges, so the number of edges on each face is greater than or equal to 3. So E ≥ 3f .
So 2e ≥ 3f .
3+4+5 3+3+3 3f
Example 11.3.2. For example, the below planar graph has e = 2 ≥ 2 = 2 .

5
4 3

3+3+5+7 3+3+3+3 3f
Example 11.3.3. The below planar graph has e = 2 ≥ 2 = 2 .

195
 7
3 3
5

Theorem 11.3.4. The complete graph K5 on 5 vertices is not planar.

Proof. Suppose for a contradiction that K5 can be drawn as a planar graph. Note that v = 5 and e = 52 =
10. By Euler’s formula, then v − e + f = 2. So 5 − 10 + f = 2. This implies that f = 7.
By Lemma 11.3.1, 10 = e ≥ 3f 3·7
2 = 2 = 10.5. This is a contradiction, and hence K5 cannot be planar.

Since K5 is not planar, any graph that contains K5 as a subgraph is not planar. One of the exercises is
to show that B3,3 is not planar; because of this, any graph that contains B3,3 as a subgraph is not planar.
The following theorem says that the reason any graph is non-planar is because either K5 or B3,3 is hidden
in the graph. In order to understand this theorem, we need to generalize the idea of subgraphs to a new
concept of minors. (In some texts, this theorem is expressed using subdivisions rather than minors.) We
include this theorem just for your general knowledge — we will often ask you to determine whether a graph
is planar or not without using Kuratowski’s Theorem!
Definition 11.3.5. Given a graph G, an edge contraction is a way to make a new graph by removing an
edge from G and merging the two vertices that were at its endpoints. A graph H is a minor of G if H can
be obtained from G by contracting some edges, deleting some edges, and deleting some isolated vertices.
In the graph below on the left, if we contract the red edge, then we obtain the graph below on the right.

Theorem 11.3.6 (Kuratowski’s Theorem). A graph G is planar if and only if it does not contain K5 or
K3,3 as a minor.
For example, the Petersen graph is not planar since it has K5 as a minor.

Indeed, one can contract all five edges connecting the “inner star” to the “outer pentagon” to form a copy
of K5 .

Problems
1. What is the smallest number of crossings needed for a graph of K5 drawn in the plane? Illustrate this.
2. Is it possible to find any set of nine edges from K8 such that if you remove those 9 edges then you
obtain a (connected) planar graph? If so, draw it. If not, explain why not.

196
 3. If G is a bipartite graph which is a planar graph, prove that e ≥ 2f .

4. Prove that the complete bipartite graph B3,3 is not planar. You are not allowed to use Kuratowski’s
Theorem.

House 1 Water

House 2 Electricity

House 3 Gas

(As a consequence, it is impossible to connect the above three houses to on the left to the water,
electricity and gas utilities on the right, without having some of the utility lines cross.)

5. Draw the graph B3,3 in the plane with the smallest number of crossings possible.

6. Show that the Petersen graph is not planar. (You are not allowed to use Kuratowski’s Theorem.) Use
the fact (which you don’t need to prove) that there are no cycles of length 3 or 4 in the Petersen graph.

7. Prove that a planar graph with v vertices (where v ≥ 3) has at most 3v − 6 edges. This result says
that a planar graph with a fixed number of vertices cannot have too many edges!

11.4 Euler’s formula for polyhedra

Click on the icon at left or the URL below for this chapter’s lecture video.
https://youtu.be/hgMS- xZxic

A polyhedron is a three-dimensional shape, whose faces are flat polygons and whose edges are straight
line segments intersecting at vertices. We restrict attention to convex polyhedra in three-dimensional space.
The word convex means that the polyhedron is an intersection of half-spaces, or equivalently, it means that
the polyhedron is the convex hull of its set of vertices, though we do not expect you to know what all of
these words mean.
Here are some prisms, which are examples of polyhedra.

Triangular prism: 6 vertices, 9 edges, 5 faces

Pentagonal prism: 10 vertices, 15 edges, 7 faces

The most famous polyhedra are the Platonic solids. These are the only regular polyhedra, where the word
regular means that all the faces are the same shape, all the edges are the same length, and all the interior
angles are the same.

197
 Tetrahedron: 4 vertices, 6 edges, 4 faces

Cube: 8 vertices, 12 edges, 6 faces

Octahedron: 6 vertices, 12 edges, 8 faces

Dodecahedron: 20 vertices, 30 edges, 12 faces

Icosahedron: 12 vertices, 30 edges, 20 faces

Remark 11.4.1. Each polyhedron P has a dual P D . Informally, P D is found by replacing each vertex with
a face and replacing each face with a vertex. Two vertices of P D are connected by an edge if and only if the
corresponding faces of P share an edge.
The cube and octrahedron are “dual” to each other; the dodecahedron and icosahedron are “dual” to
each other; and the tetrahedron is “self-dual”. Here is some evidence for this: the cube and the octahedron

198
have the same number of edges, but the number of vertices and faces have been flipped. Similarly, the
dodecahedron and icosahedron have the same number of edges, but the number of vertices and faces have
been flipped. For the tetrahedron, the number of vertices equals the number of faces.

Here are some other familiar examples of polyhedra (whose sides have been puffed out). All of these
images are licensed under the Wikipedia Creative Commons.

Soccer ball: 60 vertices, 90 edges, 32 faces (12 pentagons, 20 hexagons)

Basketball: 6 vertices, 12 edges, 8 faces

Football: 2 vertices, 4 edges, 4 faces

Volleyball: 32 vertices, 48 edges, 18 faces

You can check that all the polyhedra above have the same Euler characteristic χ = v − e + f = 2.

Theorem 11.4.2 (Euler). In any convex polyhedron we have χ = v − e + f = 2, where v is the number of
vertices, e is the number of edges, and f is the number of faces.

199
Proof. Remove one face of the polyhedron. Stretch it to lie flat in the plane. This gives a planar graph,
where the outside face of the planar map corresponds to the face we removed from the polyhedron.

This planar graph has the same values for v, e, and f . For any planar graph, χ = v − e + f = 2 by
Theorem 11.2.7. Hence the same is true for the polyhedron.

Remark 11.4.3. The above planar graph is called a Schlegel diagram of the polyhedron. Show this by
holding your iPhone flashlight up near the front face of a polyhedron formed by sticks and balls. The
projection of the edges of the polyhedron to the board then gives the Schlegel diagram!

Exercises
1. Suppose a polyhedron P is a triangulation of the sphere with 57 vertices and 84 edges. How many
faces does P have?

2. Draw planar graphs of the triangular and pentagonal prisms.

3. Draw planar graphs of the 5 platonic solids.

4. Draw several pictures of an orange with a graph on its peel.

(a) Make a table of the values of v, e, f for these graphs.

(b) Find the quantity χ = v − e + f .
(c) Make a conjecture about χ and try explain why it is true.

5. Which platonic solid is dual to the icosahedron?

6. Which platonic solid is dual to the tetrahedron?

7. For which platonic solid is the ratio e/v the biggest?

8. What is the Euler characteristic of the pentagonal prism?

9. What is the diameter of the cube graph?

11.5 Investigation: Graphs on other surfaces

In earlier sections in this chapter, we looked at connected graphs on the plane or on the sphere. For these
graphs, the Euler characteristic χ = v − e + f was always equal to χ = 2. We used this to show that K5 and
K3,3 are not planar graphs.
The Euler characteristic χ is an important concept in topology. In this section, we illustrate the value of
the Euler characteristic χ for graphs that are drawn on the torus.

200
 The torus is the surface of a doughnut. One way to make a torus is to take a piece of paper and tape the
top to the bottom (this makes a cylinder), and then tape the left side to the right.
Another way to make a torus is as a ‘pacman’ board. This is a flat piece of paper where each point on
the right edge is identified with the point on the left edge that has the same distance from the bottom, and
each point on the top edge is identified with the point on the bottom edge which has the same distance from
the left.
These descriptions of the torus are the same surface (in topology, we say that they are homeomorphic).
In the first construction, the tape forms two loops on the torus. By cutting the torus along the two loops,
it can be stretched back out to a flat piece of paper.
There is additional flexibility when drawing graphs on the torus because edges can loop around the hole
or through the hole. On the pacman board, this extra flexibility can be seen because an edge that goes off
the right side of the page comes in on the left, or an edge that goes off the top of the page comes in on the
bottom.
In general, the Euler characteristic of a graph on a surface is an integer that depends only on the surface,
and not on the graph, so long as the graph subdivides the surface into faces which are connected regions
with no non-contractible loops! It is a surprising fact that the Euler characteristic χ = v − e + f depends on
the surface, but not on the graph.
When studying graphs on the torus, it can be tricky to count the number of vertices, edges, and faces.
Here is an example graph on the pacman board.

In the above example on the pacman board, it looks like there are 5 vertices. However, there are only 2.
The reason is that the 4 corners are all the same point, because of the identifications of the edges! It looks
like there are 8 edges, the 4 radiating from the center and the 4 sides of the square. However, there are only
6, because the top and bottom edges are identified and the left and right edges are identified. It looks like
there are 4 faces and that is correct. The reason is that the top region is separated from the bottom region
by the edge along the top and the left region is separated from the right region by the edge along the left.
Thus v = 2, e = 6, and f = 4. Thus χ = v − e + f = 0.
Indeed, the Euler characteristic of any graph on the torus (in which the faces carved out by the graph are
each connected and contain no non-contractible loops) is χ = 0. This is true even if the number of vertices,
edges, and faces are quiet different than the above example — e.g. even if the number of vertices is v = 100
instead of v = 2.

201
 One interesting fact is that the complete graphs K5 , K6 , and K7 can be drawn on a torus without having
any edges cross, but K8 cannot.

Exercises
1. Draw several graphs on the torus. (These graphs need to have a cycle that goes around the hole and
a cycle that goes through the hole.) Make a table of the values of v, e, f for these graphs.
(a) Find the quantity χ = v − e + f .
(b) Make a conjecture about χ and try explain why it is true.

2. Draw some graphs on the pac man board. (These graphs need to have a cycle that goes from the
bottom to the top and a cycle that goes from the left to the right.) Make a table of the values of v, e, f
for these graphs.
(a) Find the quantity χ = v − e + f .
(b) Make a conjecture about χ and try explain why it is true.

3. Draw K5 on the surface of a doughnut, with no edges crossing.

Hint: it is easier to draw this on the pacman board.
4. Draw K3,3 on the surface of a doughnut with no edges crossing.
Hint: it is easier to draw this on the pacman board.

5. An IKEA furniture store box arrives with 32 attachment brackets, 64 metal rods that are 1 foot long,
and 32 panels measuring 1 square foot. Explain how to construct these into a 3 × 3 × 1 storage cabinet
with a window in the center. The storage cabinet must be fully enclosed but has no interior shelves.
6. Can you draw either K5 or K3,3 on the surface of a Klein bottle?

202
Chapter 12

Graph Coloring

12.1 The chromatic number of a graph

Imagine you are in charge of creating a new schedule for final exams at CSU, where students take at most
one final each day. You need to make sure that no student has two finals scheduled on the same day, since
the student cannot be in two places taking two exams at once. An easy way to solve this problem is to give
each course its own day; however, this is an impractical solution, since there are hundreds of courses offered
a semester and no one wants to turn the week of finals into the year of finals! Having all finals on the same
day also does not work, unless there is in no overlap of students between classes. Our goal for this scenario
is to minimize the number of days that the university offers finals, so that the students and faculty can have
the longest break possible between semesters.
One way to visualize this problem is with a graph. We can construct this graph to have n vertices
representing the n courses offered that semester, and an edge between two vertices if there is (at least)
one student who is in both of those courses. Now, all the finals on Monday of finals week will be assigned
the color blue. There will be no conflicts as long as any two adjacent vertices are not both blue, because
adjacency represents at least one student being in both classes. We can continue to introduce new colors
for each day needed (orange for Tuesday, yellow for Wednesday, . . .) until we have a graph where no two
vertices of the same color are adjacent. The smallest number of colors required is the minimum number of
days needed for final exams and is called the chromatic number of the graph.

Definition 12.1.1. Let G be a labeled graph. A vertex-coloring is a way of coloring the vertices of G so
that any two adjacent vertices have different colors.
A k-coloring of G is a vertex coloring of G that uses at most k colors.
A graph is k-colorable if it has a vertex-coloring with k colors.
The chromatic number C(G) of a graph G is the smallest integer k such that G is k-colorable.

For example, the following graph has chromatic number 3.

The following larger graph has the chromatic number 4.

203
Example 12.1.2. The chromatic number of Kn is C(Kn ) = n.

Proof. The graph Kn is n-colorable because each vertex can be a different color. The graph Kn is not
(n − 1)-colorable since by the pigeonhole principle, if we have only n − 1 colors then at least 2 (necessarily
adjacent) vertices will have the same color.

Example 12.1.3. The chromatic number of the empty graph with n ≥ 1 vertices (but no edges) is 1.

Example 12.1.4. The chromatic number of the path graph Pn with n vertices is C(Pn ) = 2 if n ≥ 2.

Example 12.1.5. The chromatic number of the cycle graph Cn is 2 when n is even and 3 when n ≥ 3 is
odd.

Remark 12.1.6. An edge-coloring of G is a way of coloring the edges of G such that all the edges ending
at a vertex v have different colors, for each vertex v of G. The edge-coloring number of G is the smallest
number k such that G has an edge-coloring with k colors.

In this book will consider vertex colorings more than edge colorings, and whenever we write chromatic
number we mean the minimum number of colors required for a vertex coloring (as defined above). Never-
theless, we also ask a few questions about edge colorings at the end of this section.

204
Exercises
1. Explain Examples 12.1.3, 12.1.4, and 12.1.5.
2. Find the chromatic number of the following graph.

D C
B

E G F

3. Find the chromatic number of the 5 platonic solids.

4. Find the chromatic number of the Petersen graph.
5. Show that the edge number of a graph is at least as big as the maximum degree of the vertices of G.
6. Find the edge coloring number of the following graphs:
(a) Pn ;
(b) Cn ;
(c) K4 ;
(d) K5 ;
(e) cube and octahedron.
7. Let G and H be two graphs that have no vertices (and hence no edges) in common. Let G ∪ H be the
graph that consists of all of the vertices and edges that live in either G or H. How is the chromatic
number C(G ∪ H) of the union related to the two chromatic numbers C(G) and C(H) of the individual
graphs?

12.2 What graphs are 2-colorable?

It turns out that we have already seen the graphs that can be colored with two colors. Indeed, they are the
bipartite graphs!
Proposition 12.2.1. Let G be a connected graph with at least 2 vertices. Then G is bipartite if and only if
C(G) = 2.
Proof. A graph is bipartite if and only if it is 2-colorable. The reason is that we can identify vertices on the
left side with the color gold and vertices on the right side with the color green. The condition for a 2-coloring
that adjacent vertices have different colors is the same as the condition for a bipartite graph that all edges
connect vertices on opposite sides. At least two colors are needed if the graph is connected and has at least
two vertices.

205
 The next theorem shows that 2-colorable graphs are those graphs that do not have any odd cycles.

Theorem 12.2.2. A graph G has C(G) ≤ 2 if and only if it contains no odd cycles.

Proof. The forward direction (⇒) follows since an odd cycle is not 2-colorable, and hence any graph con-
taining an odd cycle is not 2-colorable.
For the reverse direction (⇐), suppose that G contains no odd cycles. Pick a vertex “a” and color it red.
Color all neighbors of a blue. Then, color all neighbors of the new set of colored vertices red, etc. Continue,
alternating coloring the new neighbors red or blue, until all vertices are colored. The first three steps of this
process are shown here:

a a a

Suppose for contradiction that two adjacent vertices u and v have the same color in this coloring.
u

a w
v
Pick shortest paths from u to a and from v to a; these paths alternate colors. Let w be the first junction
point of these two paths. The paths from u to w and then reversed from w to v, along with edge uv, forms
an odd cycle; this is a contradiction. Hence G must be 2-colorable.

Remark 12.2.3. The proof of Theorem 12.2.2 also gives an algorithm for deciding if a graph is 2-colorable
(i.e., if it has no odd cycles).

Exercises
1. If G is a connected bipartite graph with at least one edge, then show that G has exactly two different
2-colorings.

2. Which platonic solids are 2-colorable?

3. Let n be a positive integer. Let G be the graph whose vertices are the set of binary strings of length
n, where two vertices are adjacent if and only if the strings differ in exactly one position. For example,
when n = 3, then G is the graph of the cube. Is G 2-colorable? Explain why or why not.

206
12.3 Bounds on chromatic numbers
Given a graph G with n vertices, what are the upper and lower bounds on the chromatic number C(G)?
For example, take a graph with three vertices. If there were no edges in our graph (the empty graph on
three vertices), we would have that C(G) = 1. If there was one edge (P2 and a vertex) or two edges (P3 ),
we would have C(G) = 2. Finally, if we had three edges, our graph (C3 = K3 ) would have C(G) = 3. Thus,
we have that a graph with n = 3 vertices has a lower bound of C(G) = 1 and upper bound of C(G) = 3.
We generalize this process in the following theorem.

Theorem 12.3.1. Let G be a graph with n vertices, where n ≥ 1. Then 1 ≤ C(G) ≤ n.

Proof. As shown in Example 12.1.2 and Example 12.1.3, the chromatic number of the complete graph Kn
is n and the chromatic number of the empty graph is 1. Now, consider a graph G1 that is not complete.
Then, there exists two vertices x1 and y1 that are not adjacent. We can therefore assign x1 and y1 to be
the same color and assign each of the remaining n − 2 vertices to be its own color. Then, we see that the
chromatic number of this graph is C(G1 ) ≤ n − 1 < n. Similarly, take another graph G2 that is not empty.
Then, there exists two vertices x2 and y2 that are adjacent. We must assign x2 and y2 to be two different
colors. Therefore, C(G2 ) ≥ 2 > 1.

In the beginning of this section, we showed that the chromatic number of K3 or C3 was 3. This means
that any graph that contains a triangle C3 = K3 is not 2-colorable. This can be generalized in two ways: by
looking at the cycle graph Cn when n ≥ 3 is odd; or by looking at the complete graph Kn when n ≥ 3.
The below graph contains a copy of K5 , the complete graph on 5 vertices. Since K5 is not 4-colorable,
this means that the entire graph is not 4-colorable.

Remark 12.3.2. Any graph containing a copy of Kn is not (n − 1)-colorable.

Recall also that any cycle graph with an odd number of vertices is not 2-colorable. It follows that the
following below graph, which contains a 7-cycle, is not 2-colorable.

Remark 12.3.3. Any graph containing a copy of Cn for n ≥ 3 odd is not 2-colorable.

207
 More generally, whenever H is a subgraph of G, then the chromatic number of H provides a lower bound
on the chromatic number of G

Proposition 12.3.4. If G is a graph and H is a subgraph of G, then C(H) ≤ C(G).

Proof. Let k < C(H). Suppose for a contradiction that G were k-colorable. If we simply removed all of the
vertices in G that are not in H (and also all of the edges in G that are not in H), then we would obtain a
valid coloring of H with at most k colors. This contradicts the fact that k < C(H).

Exercises
1. Show that the chromatic number of the following graph is at least 3.

2. Show that the chromatic number of the following graph is at least 3.

3. Show that the chromatic number of the following graph is at least 4.

4. Show that the chromatic number of the following graph is at least 5.

208
12.4 Brooks’ Theorem
What happens if we try to 3-color a graph in the same way we 2-colored a graph? Without loss of generality,
fix three of the vertices in the left-most triangle to be Red, Blue, and Green. How many ways can we try to
continue this coloring?

R or B

R or G ?

R = Red, B = Blue, G = Green

The number of options to try explodes!
Remark 12.4.1. One example application of coloring graphs is to scheduling. Suppose for example that
you are the math department chair, with a collection of math classes that need to be taught. The times for
these classes are already fixed. What is the minimum number of instructors you need to hire?
You can solve this problem by building a graph with a vertex for each math class, and by adding an edge
between two vertices if the corresponding classes overlap in time. The minimum number of colors needed to
color this graph is the same as the minimum possible number of instructors you need to hire.
See Problem #4 in Section 12.7 for another example scheduling application.
Remark 12.4.2. Deciding if a graph with n vertices is 3-colorable in general takes running time ≈ 2n/2 .
This is expensive. Nothing substantially better is known for k-colorings for any k ≥ 3.
Here’s a special case where k-colorings are possible:
Theorem 12.4.3 (Brooks’ Theorem). If every vertex in a graph G has degree at most d, then G is (d + 1)-
colorable.
Proof. Fix d. We proceed by induction on n, the number of vertices in G.
For the base case, if n ≤ d + 1, then clearly the graph is (d + 1)-colorable, since we can use a different
color for each vertex.
Now, assume any such graph with n vertices is (d + 1)-colorable. (We must show the same is true for
any such graph with n + 1 vertices.)
Given a graph G with n vertices, delete one vertex v. By the inductive assumption we can color what
remains. Since v has at most d neighbors, there is at least one free color out of the (d + 1)-colors we can use
to color v compatibly. Hence we are done by induction.
Below is a picture in the case when d = 3.

209
 G G\v G

Y Y
R B R B
P Y P
v
R = Red, B = Blue, Y = Yellow, P = Purple

Aside. Furthermore, a graph in which every vertex has degree at most d is d-colorable unless
• d = 2 and some connected component is an odd cycle, or
• d > 2 and some connected component is Kd+1 .
The proof of this aside is very hard!
Exercise. Show the following graph is not 3-colorable.
1

2 4 5
3

6 7

Remark 12.4.4. Sadly there is no K4 in this graph, because then we’d be done immediately.
Answer. Suppose for a contradiction that this graph were 3-colorable. Without loss of generality let vertex 1
be Red. Then vertices 2 and 3 are Blue and Green or Green and Blue, which implies 6 is Red. Also vertices
4 and 5 are Blue and Green or Green and Blue, which implies 7 is Red. This is a contradiction since 6 and
7 are adjacent! Hence this graph is not 3-colorable.
Exercise. Show the following graph is not 3-colorable.
A1

A2
E1 B1
E2 B2
P

D2 C2

D1 C1

210
 In the following solution, we will call the inner vertex P the central vertex, and the other vertices are in
five pairs of twins: A1 and A2 are twins, B1 and B2 are twins, and so on. In its pair, A1 is called the outer
twin and A2 is called the inner twin (and similarly for all other twin pairs).
Answer. Suppose for a contradiction this graph were 3-colorable. Without loss of generality let the central
vertex be Red. So all 5 inner vertices are Blue or Green. A way of coloring the vertices that agrees with this
choice is shown below (but naturally is not a 3-coloring - we will just pretend that it is a valid coloring as
an illustration).

A1

A2
E1 B1
E2 B2
P

D2 C2

D1 C1

We will show that, if this coloring is a valid 3-coloring, then we can use it to produce a valid 2-coloring of
the outer 5-cycle (but not of the entire graph) as follows: if any outer twin is colored Red, replace Red with
the color of its inner twin (Blue or Green). Then in this coloring of the outer cycle with Blue and Green, if
an outer twin v is changed from Red to Blue (for example), then both of v’s adjacent outer twins must be
Green. Indeed, neither of v’s adjacent outer twins could have been Blue since they are each connected to v’s
inner twin which is Blue. Also, neither of v’s adjacent outer twins could have been changed from Red (to
any other color) since v was originally Red.
But this is a contradiction—no odd cycle can be 2-colored. Hence the original graph must not be 3-
colorable.

Exercises
1. Brook’s Theorem guarantees that G is k-colorable for some k, but k might be larger than the chromatic
number C(G). Compare k and C(G) for each of the following graphs:
(a) the path graph Pn ;
(b) the cycle graph Cn ;
(c) each of the Platonic solids;
(d) the Petersen graph.

12.5 The chromatic polynomial

In this section, we think about the number k-colorings of a labeled graph G and we want to keep track of the
outcome when k is any positive integer. We define the chromatic polynomial of G as the polynomial whose
value at the input k is the number of k-colorings.
Here is an application to a real-world scenario. Suppose you have k employees and n jobs and some of
the jobs conflict with each other, then how many ways are there to assign the employees to the jobs? In this
scenario, let G be the labeled graph with n vertices, where the vertices are labeled by the n jobs, and an

211
edge is drawn between the vertices for two jobs exactly when those jobs conflict. Each employee is assigned
one of the k colors. Making an assignment of the employees to the jobs is the same as making a k-coloring
of G. If the number of employees is less than the chromatic number C(G), then there is no way to create
a proper vertex-coloring of a graph. However, for any k ≥ C(G), we want to find a way to represent the
number of different vertex-colorings. In fact, we can do such a thing with a polynomial. Now we want to
study the number of ways to k-color a labeled graph.

Theorem 12.5.1. If G is a labeled graph with n vertices, there is a polynomial PG (x) having the property
that PG (k) is the number of ways to k-color the vertices of G for any positive integer k.

What is amazing about this theorem is that the number of colors can be as large as we want, including
much larger than the number of coefficients of the polynomial.
The chromatic polynomial is not a generating function because the number of k-colorings is the value
PG (k) rather than its kth coefficient.

Example 12.5.2. If G is the labeled complete graph on n vertices, then PG (x) = x(x − 1) · · · (x − n + 1).

Example 12.5.3. If G is the empty labeled graph on n vertices, then PG (x) = xn .

Example 12.5.4. If G is the labeled path graph Pn , then PG (x) = x(x − 1)n−1 .

Theorem 12.5.5. If T is a labeled tree with n vertices, then PT (x) = x(x − 1)n−1 .

Proof. There are k choices for the color of the first labeled vertex. We now consider the vertices adjacent to
this first one. Since these vertices are adjacent to the first labeled vertex, the color used for the first labeled
vertex is not an option. So there are k − 1 choices for the color of any vertex adjacent to the first labeled
vertex. We continue in this fashion. We can color the starting vertex with any of the k colors. Then, each
branch can be colored with any of the (k − 1) colors remaining. The branches off those vertices are not
connected to the starting vertex anymore, so while one color is removed, the color of the starting vertex is
added back. Thus, there will be k − 1 colors left at all times for the n − 1 vertices remaining.

Theorem 12.5.6. Let G be a labeled graph and α be an edge of G. Let G α be the labeled graph obtained
by deleting α and G⊗α be the labeled graph obtained by contracting the edge α. Then

pG (x) = pG α (x) − pG⊗α (x).

Proof. Let v1 , v2 be the two vertices of G that are connected by the edge α. If we consider the graph G α ,
we have two disjoint methods of creating a k-coloring. The first method is a k-coloring where v1 and v2 are
the same color (since they are no longer adjacent), and the number of these k-coverings is pG⊗α (k). The
second method is a k-coloring where v1 and v2 are different colors, and the number of these k-colorings is
pG (k). Thus, pG α (k) = pG (k) + pG⊗α (k). Thus pG (k) = pG α (k) − pG⊗α (k).

Example 12.5.7. Let G be the labeled graph C4 . Then G α is a labeled P4 and G⊗α is a labeled C3 . The
chromatic polynomial for the labeled P4

pP4 (x) = x(x − 1)3 = x4 − 3x3 + 3x2 − x.

The chromatic polynomial for the labeled C3 is

pC3 (x) = x(x − 1)(x − 2) = x3 − 3x2 + 2x.

So
pC4 (x) = pP4 (x) − pC3 (x) = x4 − 4x3 + 6x2 − 3x = x(x − 1)(x2 − 3x + 3).

Example 12.5.8. Let G be the labeled cycle graph C5 . Then pG (x) = x(x − 1)(x − 2)(x2 − 2x + 2).

212
Proof. In this case, G α is a labeled path P5 and G⊗α is a labeled C4 . Thus,
pG (x) = x(x − 1)4 − (x4 − 4x3 + 6x2 − 3x) = x(x − 1)(x − 2)(x2 − 2x + 2)
as desired.
Since 0 and 1 and 2 are roots of pC5 but 3 is not, we see that χ(C5 ) = 3.
Corollary 12.5.9. If G is a labeled graph with n vertices, then there is a polynomial PG (x) such that PG (k)
is the number of k-colorings of G. Also, PG (x) has degree n and is monic (meaning that the coefficient of
xn in PG (x) is 1).

Exercises
1. Explain Example 12.5.2.
2. Explain Example 12.5.3.
3. Explain Example 12.5.4.
4. Find the chromatic polynomial of the labeled cycle C6 .

12.6 Coloring regions with two colors

Suppose we draw circles in the plane; these circles are allowed to intersect. The circles divide the plane into
regions, where the number of regions depends on the intersection pattern of the circles. We would like to
color these regions with as few colors as possible, so that adjacent regions have different colors. We say that
two regions are adjacent if they share a boundary arc of non-zero length (two regions which only intersect
at vertices are not considered to be adjacent). How many colors are needed to color these regions?

Theorem 12.6.1. The regions formed by circles in the plane can be colored with two colors so that regions
sharing a boundary arc are colored differently.
Proof. Let n be the number of circles drawn in the plane; we proceed by induction on n. The two colors we
will use are gray and white.
For the base case n = 1, note that we can draw the inside of our single circle gray, and the outside white.

For the inductive step, suppose the theorem is true for any n − 1 circles. Now consider n circles drawn
in the plane. Select any circle C and remove it.

213
 C

Figure 12.1: (Left) n circles. (Middle) With circle C removed, n − 1 circles remain. (Right) By the inductive
hypothesis, we can color the regions defined by n − 1 circles.

By the inductive hypothesis, we can color the picture on the right with n − 1 circles with two colors.
Now to color the picture with n circles, simply reverse the color of each region inside circle C.

Note this coloring works for arcs outside C, for arcs inside C, and for arcs which are part of C.

Remark 12.6.2. One can also color “elementary school drawings” such as the following with two colors.

214
Indeed, one can certainly color the underlying grid in a “checkerboard” fashion. Inductively add objects
such as the triangle or circle or quadrilateral above, repeating the inductive argument in the proof of Theo-
rem 12.6.1.

Remark 12.6.3. Consider the picture of n circles in the plane. The regions of the plane made by these
circles can be colored with 2 colors. Now, let’s make a graph by putting one vertex in each region and drawing
an edge between two vertices exactly when their regions are adjacent. Then this graph is 2-colorable, or
bipartite.

By Theorem 12.2.2, the graph drawn above has no odd cycles because it is 2-colorable. We can also
prove directly that this graph obtained from regions formed by circles in the plane contains no odd cycles.
The reason is that each edge of this graph crosses a single circle. But when walking around a cycle in this
graph, each circle C contributes an even number of edges to the length of this cycle (alternating in, out, in,
out, . . . of C). Hence every cycle in this graph has even length.

Exercises
1. Draw an “elementary school drawing” of your own design on top of a checkerboard, as in Remark 12.6.2.
Then 2-color the regions you have created.

2. Is it possible to color the below regions in the plane so that no two adjacent regions have the same
color?

3. Is it possible to color the below regions in the plane so that no two adjacent regions have the same
color?

215
 4. If G is a connected bipartite graph, then how many different ways are there to 2-color G?

5. If G is a bipartite graph with n different connected components, then how many different ways are
there to 2-color G?

12.7 The four color theorem

Imagine the life of an ambitious cartographer (a person who makes maps) in 1852. To properly distinguish
one region from another, it is crucial to utilize a different color for territories sharing a common border.
After making many maps for various royal affairs, you begin to question the coloring of your maps. How
many colors are needed to color planar maps in a satisfactory way? The below map can be colored with four
colors.

Coloring maps can be related to coloring planar graphs. Indeed, below we show how to turn this map into
a planar graph, and how a valid coloring of this map translates into a vertex coloring of the corresponding
planar graph. To form the graph, add a vertex for each region, and add an edge between two vertices if the
corresponding regions are adjacent. A valid coloring of the map, with no adjacent regions of the same color,
corresponds to a vertex coloring of the graph, meaning that no adjacent vertices have the same color.
Let us explain why this map requires at least 4 colors, in order for adjacent regions to have different
colors. Suppose for a contradiction 3 colors sufficed. Without loss of generality, color the region ‘N’ red,
color the region ‘C’ blue, and color the region ‘O’ green. Then necessarily the region ‘I’ must be blue, as it
is already adjacent to green and red regions, and we have only three colors at our disposal. And similarly,
the region ‘U’ must now necessarily be green. There is now no valid color for region ‘A’, as it is already
adjacent to regions labeled blue, red, and green. Hence at least 4 colors are needed to color this map!

216
 Figure 12.2: On the left is a planar map, and on the right is its corresponding planar graph.

O I

N U
C
A A

Below is a smaller map that requires 4 colors so that adjacent regions are not of the same color. Indeed,
the planar graph corresponding to this map is K4 , the complete graph on four vertices, which requires at
least 4 colors.

After successfully coloring many large maps with only four colors, our cartographer begins to wonder if
all planar maps can be colored with four colors. This is true and is now called the four-color theorem; it
states that any map drawn on the plane or the surface of a sphere can be colored using four colors in such
a way that areas with a common boundary are not the same color.

Theorem 12.7.1 (Four color theorem). Every planar map can be colored with only 4 colors, so that adjacent
regions (sharing a boundary of positive length) have different colors.

The history of the four color theorem is quite interesting. The question “Can every planar map be
4-colored” was first raised by Francis Guthrie in England in 1852. In 1879, Alfred Kempe published an
incorrect proof, which for a decade was regarded as correct before the error was found! Indeed, in 1886
the problem of showing that any planar map can be 4-colored was posed to the students of Clifton College,

217
with the (extremely difficult) instructions that “no solution may exceed 1 page of manuscript and 1 page
of diagrams.” In 1976, the first correct proof the four color theorem was given by Appel & Haken. This
proof relies on computer simulations, including 1,000 hours of computer time to check various cases. This
proof technique has raised interesting questions at the intersection of philosophy, mathematics, and computer
science (recall Example 5.1.3 in the chapter on proof techniques), such as:

• What is a proof?

• What does it mean for a proof to be verified or certified by a computer?

• What are the implications when a proof is too long for every step to be understood by a team of human
readers?

Computer assisted proofs have since grown in number, scope, and importance. The present-day proof of the
four color theorem has simplified greatly since 1976, but it still involves assistance from computers.
We next show, via a relatively short proof not requiring computers, that every planar map is 6-colorable.
One can give similar proofs that every planar map is 5-colorable, though they are a bit longer.
Proving that every planar map is 6-colorable is the same thing as showing that every planar graph is
6-colorable, since every planar map can be turned into a planar graph via the process shown in Figure 12.2.
To show that every planar map is 6-colorable, we will need the following two lemmas.

Lemma 12.7.2. Every planar graph has a vertex of degree at most 5.

Proof. We assume that every vertex has degree at least 6 and will find a contradiction. Let e be the number
of edges and v be the number of vertices. Then

2e = total sum of all vertex degrees by Theorem 8.3.6

≥ 6v,

since we assumed that every vertex has degree at least 6. Dividing by two, this gives e ≥ 3v. This contradicts
Exercise 7 in Section 11.3, which says a planar graph can have at most 3v − 6 edges.

Here is a generalization of Brook’s Theorem (Theorem 12.4.3).

Theorem 12.7.3. Let d be a positive integer. Suppose every subgraph of a graph G contains a vertex whose
degree is less than or equal to d. Then G is (d + 1)-colorable.

Proof. Let v be the number of vertices in the graph G. We will remove vertices (and all incident edges)
one-by-one to form a sequence of graphs G = Gv , Gv−1 , Gv−2 , . . . , G3 , G2 , G1 , where each Gi has exactly
i vertices.
By assumption, the graph G = Gv has a vertex of degree at most d; let’s call this vertex bv . We remove the
vertex bv and all its incident edges to form the subgraph Gv−1 . We remark that if Gv−1 is (d + 1)-colorable,
then so is Gv since the number of colors (d + 1) is greater than the number of vertices bv is adjacent to.
By assumption, the subgraph Gv−1 has a vertex of degree at most d; let’s call this vertex bv−1 . We
remove bv−1 and all its incident edges to form subgraph Gv−2 . Similarly, if Gv−2 is (d + 1)-colorable, then
so is Gv−1 since the number of colors is greater than the degree of the vertex bv−1 .
We continue in this way until we obtain a subgraph G1 with a single vertex. At each stage, we note that
if Gi−1 is (d + 1)-colorable, then so is Gi .
Note G1 is clearly (d + 1)-colorable because it is 1-colorable. Hence it follows that G2 is (d + 1)-colorable,
and therefore G3 is (d + 1)-colorable. We continue in this way until we observe that Gv−2 is (d + 1)-colorable,
and therefore Gv−1 is (d + 1)-colorable, and therefore Gv is (d + 1)-colorable.

Corollary 12.7.4. Any planar graph G, and hence any planar map, is 6-colorable.

Proof. Any subgraph of a planar graph G is planar and hence has a vertex of degree at most d = 5 by
Lemma 12.7.2. So Theorem 12.7.3 implies that G is colorable with d + 1 = 5 + 1 = 6 colors.

218
Figure 12.3: Figure for the proof of Theorem 12.7.3, showing how to collapse from the graph G = G7 down
to the graph G1 which is a single vertex, at each point removing a vertex of degree at most 3.

Exercises
1. Read the philosophical objections section on the Wikipedia page on computer-assisted proofs. Explain
your opinion on the validity of computer-assisted proofs.

2. Show that the graph drawn below is not 3-colorable but is 4-colorable.

3. Let G be a graph such that d + 1 vertices have an arbitrarily large degree and all of the other vertices
have degree at most d. Prove that G is (d + 1)-colorable.

4. Show that a planar graph G with 8 vertices and 13 edges cannot be 2-colored.
Hint: Show that in a planar map corresponding to G, one of the faces must be a triangle.

5. (a) A set of solar experiments is to be made at observatories. Each experiment begins on a given
day of the year and ends on a given day (each experiment is repeated for several years). An
observatory can perform only one experiment at a time.
Experiment A runs Sept 2 to Jan 3,
experiment B from Oct 15 to March 10,
experiment C from Nov 20 to Feb 17,
experiment D from Jan 23 to May 30,
experiment E from April 4 to July 28,
experiment F from April 30 to July 28, and
experiment G from June 24 to Sept 30.
The problem we’ll solve in (b) is: “What is the minimimum number of observatories required to
perform a given set of experiments annually?” Model this scheduling problem as a graph-coloring
problem, and draw the corresponding graph.
(b) Find a minimal coloring of the graph in (a) (i.e. the smallest k such that this graph is k-colorable),
and show that fewer colors will not suffice.

219
6. Show that a planar graph G with 10 vertices and 17 edges cannot be 2-colored.
7. True or False: A graph that is not d-colorable has a vertex of degree at least d.
8. True or False: A 2-colorable graph with n connected components can be 2-colored in exactly 2n different
ways.

220